sysmocom
/
pjproject
11
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Re #2004: Update libyuv version from libyuv git master repo dated 27 July 2017, the compile errors on old gcc versions issue persists though. 

git-svn-id: https://svn.pjsip.org/repos/pjproject/trunk@5633 74dad513-b988-da41-8d7b-12977e46ad98
This commit is contained in:
Nanang Izzuddin 2017-07-28 02:51:44 +00:00
parent e33d5ff658
commit 4b6c5064c3
69 changed files with 26282 additions and 16359 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
third_party/build/yuv/Makefile vendored
View File

@ -48,14 +48,14 @@ export YUV_OBJS = \
rotate.o \
rotate_common.o \
rotate_gcc.o \
rotate_mips.o \
rotate_dspr2.o \
rotate_neon64.o \
rotate_neon.o \
rotate_win.o \
row_any.o \
row_common.o \
row_gcc.o \
row_mips.o \
row_dspr2.o \
row_neon64.o \
row_neon.o \
row_win.o \
@ -64,7 +64,7 @@ export YUV_OBJS = \
scale.o \
scale_common.o \
scale_gcc.o \
scale_mips.o \
scale_dspr2.o \
scale_neon64.o \
scale_neon.o \
scale_win.o \

19
third_party/build/yuv/Notes.txt vendored
View File

@ -1,5 +1,7 @@
Notes:
* Source code for libyuv from https://chromium.googlesource.com/libyuv/libyuv/ dated 23 June 2016.
* Source code for libyuv from https://chromium.googlesource.com/libyuv/libyuv/ dated 27 July 2017.
* All code is compilable, except for compare_win.cc
- Use older version (https://chromium.googlesource.com/libyuv/libyuv/+/baf6a3c1bd385e7ffe6b7634560e71fb49e4f589%5E%21/)
Since there's a compiler error on:
@ -15,7 +17,14 @@ Notes:
__declspec(naked)
--------------------------------------------------------------------------------------
* Disable some compiler warning which apear alot:
- warning C4100: unreferenced formal parameter
- warning C4127: conditional expression is constant
- warning C4244: '=' : conversion from 'uint32' to 'uint8', possible loss of data
* Added these lines to file include/libyuv/basic_types.h:
--
#if _MSC_VER==1400
# include <stdint.h> // for uint8_t
#endif
...
#if defined(_MSC_VER)
# pragma warning(disable:4996) // This function or variable may be unsafe.
#endif
--

4
third_party/yuv/include/libyuv.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_H_
#define INCLUDE_LIBYUV_H_
#include "libyuv/basic_types.h"
@ -29,4 +29,4 @@
#include "libyuv/version.h"
#include "libyuv/video_common.h"
#endif // INCLUDE_LIBYUV_H_ NOLINT
#endif // INCLUDE_LIBYUV_H_

56
third_party/yuv/include/libyuv/basic_types.h vendored
View File

@ -8,17 +8,24 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_BASIC_TYPES_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_BASIC_TYPES_H_
#define INCLUDE_LIBYUV_BASIC_TYPES_H_
#include <stddef.h> // for NULL, size_t
#if defined(__ANDROID__) || (defined(_MSC_VER) && (_MSC_VER < 1600))
#if defined(_MSC_VER) && (_MSC_VER < 1600)
#if _MSC_VER==1400
# include <stdint.h> // for uint8_t
#endif
#include <sys/types.h> // for uintptr_t on x86
#else
#include <stdint.h> // for uintptr_t
#endif
#if defined(_MSC_VER)
# pragma warning(disable:4996) // This function or variable may be unsafe.
#endif
#ifndef GG_LONGLONG
#ifndef INT_TYPES_DEFINED
#define INT_TYPES_DEFINED
@ -26,31 +33,31 @@
typedef unsigned __int64 uint64;
typedef __int64 int64;
#ifndef INT64_C
#define INT64_C(x) x ## I64
#define INT64_C(x) x##I64
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## UI64
#define UINT64_C(x) x##UI64
#endif
#define INT64_F "I64"
#else // COMPILER_MSVC
#if defined(__LP64__) && !defined(__OpenBSD__) && !defined(__APPLE__)
typedef unsigned long uint64; // NOLINT
typedef long int64; // NOLINT
typedef long int64; // NOLINT
#ifndef INT64_C
#define INT64_C(x) x ## L
#define INT64_C(x) x##L
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## UL
#define UINT64_C(x) x##UL
#endif
#define INT64_F "l"
#else // defined(__LP64__) && !defined(__OpenBSD__) && !defined(__APPLE__)
typedef unsigned long long uint64; // NOLINT
typedef long long int64; // NOLINT
typedef long long int64; // NOLINT
#ifndef INT64_C
#define INT64_C(x) x ## LL
#define INT64_C(x) x##LL
#endif
#ifndef UINT64_C
#define UINT64_C(x) x ## ULL
#define UINT64_C(x) x##ULL
#endif
#define INT64_F "ll"
#endif // __LP64__
@ -58,15 +65,15 @@ typedef long long int64; // NOLINT
typedef unsigned int uint32;
typedef int int32;
typedef unsigned short uint16; // NOLINT
typedef short int16; // NOLINT
typedef short int16; // NOLINT
typedef unsigned char uint8;
typedef signed char int8;
#endif // INT_TYPES_DEFINED
#endif // GG_LONGLONG
// Detect compiler is for x86 or x64.
#if defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86)
#if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
defined(_M_IX86)
#define CPU_X86 1
#endif
// Detect compiler is for ARM.
@ -76,12 +83,12 @@ typedef signed char int8;
#ifndef ALIGNP
#ifdef __cplusplus
#define ALIGNP(p, t) \
(reinterpret_cast<uint8*>(((reinterpret_cast<uintptr_t>(p) + \
((t) - 1)) & ~((t) - 1))))
#define ALIGNP(p, t) \
reinterpret_cast<uint8*>( \
((reinterpret_cast<uintptr_t>(p) + ((t)-1)) & ~((t)-1)))
#else
#define ALIGNP(p, t) \
((uint8*)((((uintptr_t)(p) + ((t) - 1)) & ~((t) - 1)))) /* NOLINT */
(uint8*)((((uintptr_t)(p) + ((t)-1)) & ~((t)-1))) /* NOLINT */
#endif
#endif
@ -95,9 +102,9 @@ typedef signed char int8;
#define LIBYUV_API
#endif // LIBYUV_BUILDING_SHARED_LIBRARY
#elif defined(__GNUC__) && (__GNUC__ >= 4) && !defined(__APPLE__) && \
(defined(LIBYUV_BUILDING_SHARED_LIBRARY) || \
defined(LIBYUV_USING_SHARED_LIBRARY))
#define LIBYUV_API __attribute__ ((visibility ("default")))
(defined(LIBYUV_BUILDING_SHARED_LIBRARY) || \
defined(LIBYUV_USING_SHARED_LIBRARY))
#define LIBYUV_API __attribute__((visibility("default")))
#else
#define LIBYUV_API
#endif // __GNUC__
@ -108,11 +115,10 @@ typedef signed char int8;
#define LIBYUV_TRUE 1
// Visual C x86 or GCC little endian.
#if defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86) || \
defined(__arm__) || defined(_M_ARM) || \
(defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
#if defined(__x86_64__) || defined(_M_X64) || defined(__i386__) || \
defined(_M_IX86) || defined(__arm__) || defined(_M_ARM) || \
(defined(__BYTE_ORDER__) && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__)
#define LIBYUV_LITTLE_ENDIAN
#endif
#endif // INCLUDE_LIBYUV_BASIC_TYPES_H_ NOLINT
#endif // INCLUDE_LIBYUV_BASIC_TYPES_H_

82
third_party/yuv/include/libyuv/compare.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_COMPARE_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_COMPARE_H_
#define INCLUDE_LIBYUV_COMPARE_H_
#include "libyuv/basic_types.h"
@ -22,6 +22,12 @@ extern "C" {
LIBYUV_API
uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed);
// Hamming Distance
LIBYUV_API
uint64 ComputeHammingDistance(const uint8* src_a,
const uint8* src_b,
int count);
// Scan an opaque argb image and return fourcc based on alpha offset.
// Returns FOURCC_ARGB, FOURCC_BGRA, or 0 if unknown.
LIBYUV_API
@ -29,13 +35,15 @@ uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height);
// Sum Square Error - used to compute Mean Square Error or PSNR.
LIBYUV_API
uint64 ComputeSumSquareError(const uint8* src_a,
const uint8* src_b, int count);
uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b, int count);
LIBYUV_API
uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
uint64 ComputeSumSquareErrorPlane(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height);
static const int kMaxPsnr = 128;
@ -43,36 +51,56 @@ LIBYUV_API
double SumSquareErrorToPsnr(uint64 sse, uint64 count);
LIBYUV_API
double CalcFramePsnr(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
double CalcFramePsnr(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height);
LIBYUV_API
double I420Psnr(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height);
double I420Psnr(const uint8* src_y_a,
int stride_y_a,
const uint8* src_u_a,
int stride_u_a,
const uint8* src_v_a,
int stride_v_a,
const uint8* src_y_b,
int stride_y_b,
const uint8* src_u_b,
int stride_u_b,
const uint8* src_v_b,
int stride_v_b,
int width,
int height);
LIBYUV_API
double CalcFrameSsim(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height);
double CalcFrameSsim(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height);
LIBYUV_API
double I420Ssim(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height);
double I420Ssim(const uint8* src_y_a,
int stride_y_a,
const uint8* src_u_a,
int stride_u_a,
const uint8* src_v_a,
int stride_v_a,
const uint8* src_y_b,
int stride_y_b,
const uint8* src_u_b,
int stride_u_b,
const uint8* src_v_b,
int stride_v_b,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_COMPARE_H_ NOLINT
#endif // INCLUDE_LIBYUV_COMPARE_H_

20
third_party/yuv/include/libyuv/compare_row.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_COMPARE_ROW_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_COMPARE_ROW_H_
#define INCLUDE_LIBYUV_COMPARE_ROW_H_
#include "libyuv/basic_types.h"
@ -30,8 +30,8 @@ extern "C" {
#endif
// Visual C 2012 required for AVX2.
#if defined(_M_IX86) && !defined(__clang__) && \
defined(_MSC_VER) && _MSC_VER >= 1700
#if defined(_M_IX86) && !defined(__clang__) && defined(_MSC_VER) && \
_MSC_VER >= 1700
#define VISUALC_HAS_AVX2 1
#endif // VisualStudio >= 2012
@ -42,16 +42,17 @@ extern "C" {
#endif // clang >= 3.4
#endif // __clang__
#if !defined(LIBYUV_DISABLE_X86) && \
defined(_M_IX86) && (defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && \
(defined(VISUALC_HAS_AVX2) || defined(CLANG_HAS_AVX2))
#define HAS_HASHDJB2_AVX2
#endif
// The following are available for Visual C and GCC:
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) || defined(_M_IX86)))
(defined(__x86_64__) || defined(__i386__) || defined(_M_IX86))
#define HAS_HASHDJB2_SSE41
#define HAS_SUMSQUAREERROR_SSE2
#define HAS_HAMMINGDISTANCE_X86
#endif
// The following are available for Visual C and clangcl 32 bit:
@ -65,8 +66,13 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
#define HAS_SUMSQUAREERROR_NEON
#define HAS_HAMMINGDISTANCE_NEON
#endif
uint32 HammingDistance_C(const uint8* src_a, const uint8* src_b, int count);
uint32 HammingDistance_X86(const uint8* src_a, const uint8* src_b, int count);
uint32 HammingDistance_NEON(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_C(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count);
uint32 SumSquareError_AVX2(const uint8* src_a, const uint8* src_b, int count);
@ -81,4 +87,4 @@ uint32 HashDjb2_AVX2(const uint8* src, int count, uint32 seed);
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_COMPARE_ROW_H_ NOLINT
#endif // INCLUDE_LIBYUV_COMPARE_ROW_H_

373
third_party/yuv/include/libyuv/convert.h vendored
View File

@ -8,13 +8,18 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_CONVERT_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_CONVERT_H_
#define INCLUDE_LIBYUV_CONVERT_H_
#include "libyuv/basic_types.h"
#include "libyuv/rotate.h" // For enum RotationMode.
// TODO(fbarchard): fix WebRTC source to include following libyuv headers:
#include "libyuv/convert_argb.h" // For WebRTC I420ToARGB. b/620
#include "libyuv/convert_from.h" // For WebRTC ConvertFromI420. b/620
#include "libyuv/planar_functions.h" // For WebRTC I420Rect, CopyPlane. b/618
#ifdef __cplusplus
namespace libyuv {
extern "C" {
@ -22,184 +27,295 @@ extern "C" {
// Convert I444 to I420.
LIBYUV_API
int I444ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I444ToI420(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert I422 to I420.
LIBYUV_API
int I422ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
// Convert I411 to I420.
LIBYUV_API
int I411ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I422ToI420(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Copy I420 to I420.
#define I420ToI420 I420Copy
LIBYUV_API
int I420Copy(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420Copy(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert I400 (grey) to I420.
LIBYUV_API
int I400ToI420(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I400ToI420(const uint8* src_y,
int src_stride_y,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
#define J400ToJ420 I400ToI420
// Convert NV12 to I420.
LIBYUV_API
int NV12ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int NV12ToI420(const uint8* src_y,
int src_stride_y,
const uint8* src_uv,
int src_stride_uv,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert NV21 to I420.
LIBYUV_API
int NV21ToI420(const uint8* src_y, int src_stride_y,
const uint8* src_vu, int src_stride_vu,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int NV21ToI420(const uint8* src_y,
int src_stride_y,
const uint8* src_vu,
int src_stride_vu,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert YUY2 to I420.
LIBYUV_API
int YUY2ToI420(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int YUY2ToI420(const uint8* src_yuy2,
int src_stride_yuy2,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert UYVY to I420.
LIBYUV_API
int UYVYToI420(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int UYVYToI420(const uint8* src_uyvy,
int src_stride_uyvy,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert M420 to I420.
LIBYUV_API
int M420ToI420(const uint8* src_m420, int src_stride_m420,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int M420ToI420(const uint8* src_m420,
int src_stride_m420,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert Android420 to I420.
LIBYUV_API
int Android420ToI420(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
int pixel_stride_uv,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// ARGB little endian (bgra in memory) to I420.
LIBYUV_API
int ARGBToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// BGRA little endian (argb in memory) to I420.
LIBYUV_API
int BGRAToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int BGRAToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// ABGR little endian (rgba in memory) to I420.
LIBYUV_API
int ABGRToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ABGRToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGBA little endian (abgr in memory) to I420.
LIBYUV_API
int RGBAToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGBAToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGB little endian (bgr in memory) to I420.
LIBYUV_API
int RGB24ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGB24ToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGB big endian (rgb in memory) to I420.
LIBYUV_API
int RAWToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RAWToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGB16 (RGBP fourcc) little endian to I420.
LIBYUV_API
int RGB565ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int RGB565ToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGB15 (RGBO fourcc) little endian to I420.
LIBYUV_API
int ARGB1555ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGB1555ToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// RGB12 (R444 fourcc) little endian to I420.
LIBYUV_API
int ARGB4444ToI420(const uint8* src_frame, int src_stride_frame,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGB4444ToI420(const uint8* src_frame,
int src_stride_frame,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
#ifdef HAVE_JPEG
// src_width/height provided by capture.
// dst_width/height for clipping determine final size.
LIBYUV_API
int MJPGToI420(const uint8* sample, size_t sample_size,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height,
int dst_width, int dst_height);
int MJPGToI420(const uint8* sample,
size_t sample_size,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int src_width,
int src_height,
int dst_width,
int dst_height);
// Query size of MJPG in pixels.
LIBYUV_API
int MJPGSize(const uint8* sample, size_t sample_size,
int* width, int* height);
int MJPGSize(const uint8* sample, size_t sample_size, int* width, int* height);
#endif
// Convert camera sample to I420 with cropping, rotation and vertical flip.
@ -225,13 +341,20 @@ int MJPGSize(const uint8* sample, size_t sample_size,
// "format" is a fourcc. ie 'I420', 'YUY2'
// Returns 0 for successful; -1 for invalid parameter. Non-zero for failure.
LIBYUV_API
int ConvertToI420(const uint8* src_frame, size_t src_size,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToI420(const uint8* src_frame,
size_t src_size,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 format);
@ -240,4 +363,4 @@ int ConvertToI420(const uint8* src_frame, size_t src_size,
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_CONVERT_H_ NOLINT
#endif // INCLUDE_LIBYUV_CONVERT_H_

435
third_party/yuv/include/libyuv/convert_argb.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_CONVERT_ARGB_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_CONVERT_ARGB_H_
#define INCLUDE_LIBYUV_CONVERT_ARGB_H_
#include "libyuv/basic_types.h"
@ -30,246 +30,384 @@ extern "C" {
// Copy ARGB to ARGB.
LIBYUV_API
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopy(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I420 to ARGB.
LIBYUV_API
int I420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Duplicate prototype for function in convert_from.h for remoting.
LIBYUV_API
int I420ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I422 to ARGB.
LIBYUV_API
int I422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I422ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I444 to ARGB.
LIBYUV_API
int I444ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I444ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J444 to ARGB.
LIBYUV_API
int J444ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J444ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I444 to ABGR.
LIBYUV_API
int I444ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
// Convert I411 to ARGB.
LIBYUV_API
int I411ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I444ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert I420 with Alpha to preattenuated ARGB.
LIBYUV_API
int I420AlphaToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
const uint8* src_a, int src_stride_a,
uint8* dst_argb, int dst_stride_argb,
int width, int height, int attenuate);
int I420AlphaToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
const uint8* src_a,
int src_stride_a,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height,
int attenuate);
// Convert I420 with Alpha to preattenuated ABGR.
LIBYUV_API
int I420AlphaToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
const uint8* src_a, int src_stride_a,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height, int attenuate);
int I420AlphaToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
const uint8* src_a,
int src_stride_a,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height,
int attenuate);
// Convert I400 (grey) to ARGB. Reverse of ARGBToI400.
LIBYUV_API
int I400ToARGB(const uint8* src_y, int src_stride_y,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I400ToARGB(const uint8* src_y,
int src_stride_y,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J400 (jpeg grey) to ARGB.
LIBYUV_API
int J400ToARGB(const uint8* src_y, int src_stride_y,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J400ToARGB(const uint8* src_y,
int src_stride_y,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Alias.
#define YToARGB I400ToARGB
// Convert NV12 to ARGB.
LIBYUV_API
int NV12ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int NV12ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_uv,
int src_stride_uv,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert NV21 to ARGB.
LIBYUV_API
int NV21ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_vu, int src_stride_vu,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int NV21ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_vu,
int src_stride_vu,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert M420 to ARGB.
LIBYUV_API
int M420ToARGB(const uint8* src_m420, int src_stride_m420,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int M420ToARGB(const uint8* src_m420,
int src_stride_m420,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert YUY2 to ARGB.
LIBYUV_API
int YUY2ToARGB(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int YUY2ToARGB(const uint8* src_yuy2,
int src_stride_yuy2,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert UYVY to ARGB.
LIBYUV_API
int UYVYToARGB(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int UYVYToARGB(const uint8* src_uyvy,
int src_stride_uyvy,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J420 to ARGB.
LIBYUV_API
int J420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J420ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J422 to ARGB.
LIBYUV_API
int J422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int J422ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert J420 to ABGR.
LIBYUV_API
int J420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int J420ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert J422 to ABGR.
LIBYUV_API
int J422ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int J422ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H420 to ARGB.
LIBYUV_API
int H420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int H420ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert H422 to ARGB.
LIBYUV_API
int H422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int H422ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert H420 to ABGR.
LIBYUV_API
int H420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int H420ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert H422 to ABGR.
LIBYUV_API
int H422ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int H422ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// BGRA little endian (argb in memory) to ARGB.
LIBYUV_API
int BGRAToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int BGRAToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// ABGR little endian (rgba in memory) to ARGB.
LIBYUV_API
int ABGRToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ABGRToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGBA little endian (abgr in memory) to ARGB.
LIBYUV_API
int RGBAToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGBAToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Deprecated function name.
#define BG24ToARGB RGB24ToARGB
// RGB little endian (bgr in memory) to ARGB.
LIBYUV_API
int RGB24ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGB24ToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB big endian (rgb in memory) to ARGB.
LIBYUV_API
int RAWToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RAWToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB16 (RGBP fourcc) little endian to ARGB.
LIBYUV_API
int RGB565ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int RGB565ToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB15 (RGBO fourcc) little endian to ARGB.
LIBYUV_API
int ARGB1555ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGB1555ToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// RGB12 (R444 fourcc) little endian to ARGB.
LIBYUV_API
int ARGB4444ToARGB(const uint8* src_frame, int src_stride_frame,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGB4444ToARGB(const uint8* src_frame,
int src_stride_frame,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
#ifdef HAVE_JPEG
// src_width/height provided by capture
// dst_width/height for clipping determine final size.
LIBYUV_API
int MJPGToARGB(const uint8* sample, size_t sample_size,
uint8* dst_argb, int dst_stride_argb,
int src_width, int src_height,
int dst_width, int dst_height);
int MJPGToARGB(const uint8* sample,
size_t sample_size,
uint8* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
int dst_width,
int dst_height);
#endif
// Convert camera sample to ARGB with cropping, rotation and vertical flip.
@ -295,11 +433,16 @@ int MJPGToARGB(const uint8* sample, size_t sample_size,
// "format" is a fourcc. ie 'I420', 'YUY2'
// Returns 0 for successful; -1 for invalid parameter. Non-zero for failure.
LIBYUV_API
int ConvertToARGB(const uint8* src_frame, size_t src_size,
uint8* dst_argb, int dst_stride_argb,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToARGB(const uint8* src_frame,
size_t src_size,
uint8* dst_argb,
int dst_stride_argb,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 format);
@ -308,4 +451,4 @@ int ConvertToARGB(const uint8* src_frame, size_t src_size,
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_CONVERT_ARGB_H_ NOLINT
#endif // INCLUDE_LIBYUV_CONVERT_ARGB_H_

308
third_party/yuv/include/libyuv/convert_from.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_CONVERT_FROM_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_CONVERT_FROM_H_
#define INCLUDE_LIBYUV_CONVERT_FROM_H_
#include "libyuv/basic_types.h"
@ -24,151 +24,249 @@ extern "C" {
// I420Copy in convert to I420ToI420.
LIBYUV_API
int I420ToI422(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420ToI422(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
LIBYUV_API
int I420ToI444(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
LIBYUV_API
int I420ToI411(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420ToI444(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Copy to I400. Source can be I420, I422, I444, I400, NV12 or NV21.
LIBYUV_API
int I400Copy(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
int width, int height);
int I400Copy(const uint8* src_y,
int src_stride_y,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
LIBYUV_API
int I420ToNV12(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int I420ToNV12(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_uv,
int dst_stride_uv,
int width,
int height);
LIBYUV_API
int I420ToNV21(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int I420ToNV21(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_vu,
int dst_stride_vu,
int width,
int height);
LIBYUV_API
int I420ToYUY2(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToYUY2(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I420ToUYVY(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToUYVY(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I420ToARGB(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToARGB(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
LIBYUV_API
int I420ToBGRA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToBGRA(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
LIBYUV_API
int I420ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int I420ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
LIBYUV_API
int I420ToRGBA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_rgba, int dst_stride_rgba,
int width, int height);
int I420ToRGBA(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_rgba,
int dst_stride_rgba,
int width,
int height);
LIBYUV_API
int I420ToRGB24(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToRGB24(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I420ToRAW(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToRAW(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I420ToRGB565(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToRGB565(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I422ToRGB565(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
// Convert I420 To RGB565 with 4x4 dither matrix (16 bytes).
// Values in dither matrix from 0 to 7 recommended.
// The order of the dither matrix is first byte is upper left.
LIBYUV_API
int I420ToRGB565Dither(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
const uint8* dither4x4, int width, int height);
int I420ToRGB565Dither(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
const uint8* dither4x4,
int width,
int height);
LIBYUV_API
int I420ToARGB1555(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToARGB1555(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
LIBYUV_API
int I420ToARGB4444(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I420ToARGB4444(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
// Convert I420 to specified format.
// "dst_sample_stride" is bytes in a row for the destination. Pass 0 if the
// buffer has contiguous rows. Can be negative. A multiple of 16 is optimal.
LIBYUV_API
int ConvertFromI420(const uint8* y, int y_stride,
const uint8* u, int u_stride,
const uint8* v, int v_stride,
uint8* dst_sample, int dst_sample_stride,
int width, int height,
int ConvertFromI420(const uint8* y,
int y_stride,
const uint8* u,
int u_stride,
const uint8* v,
int v_stride,
uint8* dst_sample,
int dst_sample_stride,
int width,
int height,
uint32 format);
#ifdef __cplusplus
@ -176,4 +274,4 @@ int ConvertFromI420(const uint8* y, int y_stride,
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_CONVERT_FROM_H_ NOLINT
#endif // INCLUDE_LIBYUV_CONVERT_FROM_H_

259
third_party/yuv/include/libyuv/convert_from_argb.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_CONVERT_FROM_ARGB_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_CONVERT_FROM_ARGB_H_
#define INCLUDE_LIBYUV_CONVERT_FROM_ARGB_H_
#include "libyuv/basic_types.h"
@ -21,45 +21,66 @@ extern "C" {
// Copy ARGB to ARGB.
#define ARGBToARGB ARGBCopy
LIBYUV_API
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopy(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert ARGB To BGRA.
LIBYUV_API
int ARGBToBGRA(const uint8* src_argb, int src_stride_argb,
uint8* dst_bgra, int dst_stride_bgra,
int width, int height);
int ARGBToBGRA(const uint8* src_argb,
int src_stride_argb,
uint8* dst_bgra,
int dst_stride_bgra,
int width,
int height);
// Convert ARGB To ABGR.
LIBYUV_API
int ARGBToABGR(const uint8* src_argb, int src_stride_argb,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int ARGBToABGR(const uint8* src_argb,
int src_stride_argb,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert ARGB To RGBA.
LIBYUV_API
int ARGBToRGBA(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgba, int dst_stride_rgba,
int width, int height);
int ARGBToRGBA(const uint8* src_argb,
int src_stride_argb,
uint8* dst_rgba,
int dst_stride_rgba,
int width,
int height);
// Convert ARGB To RGB24.
LIBYUV_API
int ARGBToRGB24(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb24, int dst_stride_rgb24,
int width, int height);
int ARGBToRGB24(const uint8* src_argb,
int src_stride_argb,
uint8* dst_rgb24,
int dst_stride_rgb24,
int width,
int height);
// Convert ARGB To RAW.
LIBYUV_API
int ARGBToRAW(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb, int dst_stride_rgb,
int width, int height);
int ARGBToRAW(const uint8* src_argb,
int src_stride_argb,
uint8* dst_rgb,
int dst_stride_rgb,
int width,
int height);
// Convert ARGB To RGB565.
LIBYUV_API
int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb565, int dst_stride_rgb565,
int width, int height);
int ARGBToRGB565(const uint8* src_argb,
int src_stride_argb,
uint8* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
// Convert ARGB To RGB565 with 4x4 dither matrix (16 bytes).
// Values in dither matrix from 0 to 7 recommended.
@ -67,124 +88,178 @@ int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
// TODO(fbarchard): Consider pointer to 2d array for dither4x4.
// const uint8(*dither)[4][4];
LIBYUV_API
int ARGBToRGB565Dither(const uint8* src_argb, int src_stride_argb,
uint8* dst_rgb565, int dst_stride_rgb565,
const uint8* dither4x4, int width, int height);
int ARGBToRGB565Dither(const uint8* src_argb,
int src_stride_argb,
uint8* dst_rgb565,
int dst_stride_rgb565,
const uint8* dither4x4,
int width,
int height);
// Convert ARGB To ARGB1555.
LIBYUV_API
int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb1555, int dst_stride_argb1555,
int width, int height);
int ARGBToARGB1555(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb1555,
int dst_stride_argb1555,
int width,
int height);
// Convert ARGB To ARGB4444.
LIBYUV_API
int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb4444, int dst_stride_argb4444,
int width, int height);
int ARGBToARGB4444(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb4444,
int dst_stride_argb4444,
int width,
int height);
// Convert ARGB To I444.
LIBYUV_API
int ARGBToI444(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI444(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB To I422.
LIBYUV_API
int ARGBToI422(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI422(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB To I420. (also in convert.h)
LIBYUV_API
int ARGBToI420(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToI420(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB to J420. (JPeg full range I420).
LIBYUV_API
int ARGBToJ420(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToJ420(const uint8* src_argb,
int src_stride_argb,
uint8* dst_yj,
int dst_stride_yj,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB to J422.
LIBYUV_API
int ARGBToJ422(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
// Convert ARGB To I411.
LIBYUV_API
int ARGBToI411(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int ARGBToJ422(const uint8* src_argb,
int src_stride_argb,
uint8* dst_yj,
int dst_stride_yj,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert ARGB to J400. (JPeg full range).
LIBYUV_API
int ARGBToJ400(const uint8* src_argb, int src_stride_argb,
uint8* dst_yj, int dst_stride_yj,
int width, int height);
int ARGBToJ400(const uint8* src_argb,
int src_stride_argb,
uint8* dst_yj,
int dst_stride_yj,
int width,
int height);
// Convert ARGB to I400.
LIBYUV_API
int ARGBToI400(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
int width, int height);
int ARGBToI400(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Convert ARGB to G. (Reverse of J400toARGB, which replicates G back to ARGB)
LIBYUV_API
int ARGBToG(const uint8* src_argb, int src_stride_argb,
uint8* dst_g, int dst_stride_g,
int width, int height);
int ARGBToG(const uint8* src_argb,
int src_stride_argb,
uint8* dst_g,
int dst_stride_g,
int width,
int height);
// Convert ARGB To NV12.
LIBYUV_API
int ARGBToNV12(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int ARGBToNV12(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_uv,
int dst_stride_uv,
int width,
int height);
// Convert ARGB To NV21.
LIBYUV_API
int ARGBToNV21(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int ARGBToNV21(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_vu,
int dst_stride_vu,
int width,
int height);
// Convert ARGB To NV21.
LIBYUV_API
int ARGBToNV21(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
uint8* dst_vu, int dst_stride_vu,
int width, int height);
int ARGBToNV21(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
uint8* dst_vu,
int dst_stride_vu,
int width,
int height);
// Convert ARGB To YUY2.
LIBYUV_API
int ARGBToYUY2(const uint8* src_argb, int src_stride_argb,
uint8* dst_yuy2, int dst_stride_yuy2,
int width, int height);
int ARGBToYUY2(const uint8* src_argb,
int src_stride_argb,
uint8* dst_yuy2,
int dst_stride_yuy2,
int width,
int height);
// Convert ARGB To UYVY.
LIBYUV_API
int ARGBToUYVY(const uint8* src_argb, int src_stride_argb,
uint8* dst_uyvy, int dst_stride_uyvy,
int width, int height);
int ARGBToUYVY(const uint8* src_argb,
int src_stride_argb,
uint8* dst_uyvy,
int dst_stride_uyvy,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_CONVERT_FROM_ARGB_H_ NOLINT
#endif // INCLUDE_LIBYUV_CONVERT_FROM_ARGB_H_

41
third_party/yuv/include/libyuv/cpu_id.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_CPU_ID_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_CPU_ID_H_
#define INCLUDE_LIBYUV_CPU_ID_H_
#include "libyuv/basic_types.h"
@ -31,50 +31,61 @@ static const int kCpuHasX86 = 0x10;
static const int kCpuHasSSE2 = 0x20;
static const int kCpuHasSSSE3 = 0x40;
static const int kCpuHasSSE41 = 0x80;
static const int kCpuHasSSE42 = 0x100;
static const int kCpuHasSSE42 = 0x100; // unused at this time.
static const int kCpuHasAVX = 0x200;
static const int kCpuHasAVX2 = 0x400;
static const int kCpuHasERMS = 0x800;
static const int kCpuHasFMA3 = 0x1000;
static const int kCpuHasAVX3 = 0x2000;
// 0x2000, 0x4000, 0x8000 reserved for future X86 flags.
static const int kCpuHasF16C = 0x4000;
// 0x8000 reserved for future X86 flags.
// These flags are only valid on MIPS processors.
static const int kCpuHasMIPS = 0x10000;
static const int kCpuHasDSPR2 = 0x20000;
static const int kCpuHasMSA = 0x40000;
// Internal function used to auto-init.
// Optional init function. TestCpuFlag does an auto-init.
// Returns cpu_info flags.
LIBYUV_API
int InitCpuFlags(void);
// Detect CPU has SSE2 etc.
// Test_flag parameter should be one of kCpuHas constants above.
// Returns non-zero if instruction set is detected
static __inline int TestCpuFlag(int test_flag) {
LIBYUV_API extern int cpu_info_;
#ifdef __ATOMIC_RELAXED
int cpu_info = __atomic_load_n(&cpu_info_, __ATOMIC_RELAXED);
#else
int cpu_info = cpu_info_;
#endif
return (!cpu_info ? InitCpuFlags() : cpu_info) & test_flag;
}
// Internal function for parsing /proc/cpuinfo.
LIBYUV_API
int ArmCpuCaps(const char* cpuinfo_name);
// Detect CPU has SSE2 etc.
// Test_flag parameter should be one of kCpuHas constants above.
// returns non-zero if instruction set is detected
static __inline int TestCpuFlag(int test_flag) {
LIBYUV_API extern int cpu_info_;
return (!cpu_info_ ? InitCpuFlags() : cpu_info_) & test_flag;
}
// For testing, allow CPU flags to be disabled.
// ie MaskCpuFlags(~kCpuHasSSSE3) to disable SSSE3.
// MaskCpuFlags(-1) to enable all cpu specific optimizations.
// MaskCpuFlags(1) to disable all cpu specific optimizations.
// MaskCpuFlags(0) to reset state so next call will auto init.
// Returns cpu_info flags.
LIBYUV_API
void MaskCpuFlags(int enable_flags);
int MaskCpuFlags(int enable_flags);
// Low level cpuid for X86. Returns zeros on other CPUs.
// eax is the info type that you want.
// ecx is typically the cpu number, and should normally be zero.
LIBYUV_API
void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info);
void CpuId(int eax, int ecx, int* cpu_info);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_CPU_ID_H_ NOLINT
#endif // INCLUDE_LIBYUV_CPU_ID_H_

233
third_party/yuv/include/libyuv/macros_msa.h vendored Normal file
View File

@ -0,0 +1,233 @@
/*
* Copyright 2016 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_MACROS_MSA_H_
#define INCLUDE_LIBYUV_MACROS_MSA_H_
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include <msa.h>
#include <stdint.h>
#if (__mips_isa_rev >= 6)
#define LW(psrc) \
({ \
uint8* psrc_lw_m = (uint8*)(psrc); /* NOLINT */ \
uint32 val_m; \
asm volatile("lw %[val_m], %[psrc_lw_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_lw_m] "m"(*psrc_lw_m)); \
val_m; \
})
#if (__mips == 64)
#define LD(psrc) \
({ \
uint8* psrc_ld_m = (uint8*)(psrc); /* NOLINT */ \
uint64 val_m = 0; \
asm volatile("ld %[val_m], %[psrc_ld_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_ld_m] "m"(*psrc_ld_m)); \
val_m; \
})
#else // !(__mips == 64)
#define LD(psrc) \
({ \
uint8* psrc_ld_m = (uint8*)(psrc); /* NOLINT */ \
uint32 val0_m, val1_m; \
uint64 val_m = 0; \
val0_m = LW(psrc_ld_m); \
val1_m = LW(psrc_ld_m + 4); \
val_m = (uint64)(val1_m); /* NOLINT */ \
val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \
val_m; \
})
#endif // (__mips == 64)
#define SW(val, pdst) \
({ \
uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val_m = (val); \
asm volatile("sw %[val_m], %[pdst_sw_m] \n" \
: [pdst_sw_m] "=m"(*pdst_sw_m) \
: [val_m] "r"(val_m)); \
})
#if (__mips == 64)
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint64_t val_m = (val); \
asm volatile("sd %[val_m], %[pdst_sd_m] \n" \
: [pdst_sd_m] "=m"(*pdst_sd_m) \
: [val_m] "r"(val_m)); \
})
#else // !(__mips == 64)
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val0_m, val1_m; \
val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
SW(val0_m, pdst_sd_m); \
SW(val1_m, pdst_sd_m + 4); \
})
#endif // !(__mips == 64)
#else // !(__mips_isa_rev >= 6)
#define LW(psrc) \
({ \
uint8* psrc_lw_m = (uint8*)(psrc); /* NOLINT */ \
uint32 val_m; \
asm volatile("ulw %[val_m], %[psrc_lw_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_lw_m] "m"(*psrc_lw_m)); \
val_m; \
})
#if (__mips == 64)
#define LD(psrc) \
({ \
uint8* psrc_ld_m = (uint8*)(psrc); /* NOLINT */ \
uint64 val_m = 0; \
asm volatile("uld %[val_m], %[psrc_ld_m] \n" \
: [val_m] "=r"(val_m) \
: [psrc_ld_m] "m"(*psrc_ld_m)); \
val_m; \
})
#else // !(__mips == 64)
#define LD(psrc) \
({ \
uint8* psrc_ld_m = (uint8*)(psrc); /* NOLINT */ \
uint32 val0_m, val1_m; \
uint64 val_m = 0; \
val0_m = LW(psrc_ld_m); \
val1_m = LW(psrc_ld_m + 4); \
val_m = (uint64)(val1_m); /* NOLINT */ \
val_m = (uint64)((val_m << 32) & 0xFFFFFFFF00000000); /* NOLINT */ \
val_m = (uint64)(val_m | (uint64)val0_m); /* NOLINT */ \
val_m; \
})
#endif // (__mips == 64)
#define SW(val, pdst) \
({ \
uint8_t* pdst_sw_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val_m = (val); \
asm volatile("usw %[val_m], %[pdst_sw_m] \n" \
: [pdst_sw_m] "=m"(*pdst_sw_m) \
: [val_m] "r"(val_m)); \
})
#define SD(val, pdst) \
({ \
uint8_t* pdst_sd_m = (uint8_t*)(pdst); /* NOLINT */ \
uint32_t val0_m, val1_m; \
val0_m = (uint32_t)((val)&0x00000000FFFFFFFF); \
val1_m = (uint32_t)(((val) >> 32) & 0x00000000FFFFFFFF); \
SW(val0_m, pdst_sd_m); \
SW(val1_m, pdst_sd_m + 4); \
})
#endif // (__mips_isa_rev >= 6)
// TODO(fbarchard): Consider removing __VAR_ARGS versions.
#define LD_B(RTYPE, psrc) *((RTYPE*)(psrc)) /* NOLINT */
#define LD_UB(...) LD_B(v16u8, __VA_ARGS__)
#define ST_B(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UB(...) ST_B(v16u8, __VA_ARGS__)
#define ST_H(RTYPE, in, pdst) *((RTYPE*)(pdst)) = (in) /* NOLINT */
#define ST_UH(...) ST_H(v8u16, __VA_ARGS__)
/* Description : Load two vectors with 16 'byte' sized elements
Arguments : Inputs - psrc, stride
Outputs - out0, out1
Return Type - as per RTYPE
Details : Load 16 byte elements in 'out0' from (psrc)
Load 16 byte elements in 'out1' from (psrc + stride)
*/
#define LD_B2(RTYPE, psrc, stride, out0, out1) \
{ \
out0 = LD_B(RTYPE, (psrc)); \
out1 = LD_B(RTYPE, (psrc) + stride); \
}
#define LD_UB2(...) LD_B2(v16u8, __VA_ARGS__)
#define LD_B4(RTYPE, psrc, stride, out0, out1, out2, out3) \
{ \
LD_B2(RTYPE, (psrc), stride, out0, out1); \
LD_B2(RTYPE, (psrc) + 2 * stride, stride, out2, out3); \
}
#define LD_UB4(...) LD_B4(v16u8, __VA_ARGS__)
/* Description : Store two vectors with stride each having 16 'byte' sized
elements
Arguments : Inputs - in0, in1, pdst, stride
Details : Store 16 byte elements from 'in0' to (pdst)
Store 16 byte elements from 'in1' to (pdst + stride)
*/
#define ST_B2(RTYPE, in0, in1, pdst, stride) \
{ \
ST_B(RTYPE, in0, (pdst)); \
ST_B(RTYPE, in1, (pdst) + stride); \
}
#define ST_UB2(...) ST_B2(v16u8, __VA_ARGS__)
#define ST_B4(RTYPE, in0, in1, in2, in3, pdst, stride) \
{ \
ST_B2(RTYPE, in0, in1, (pdst), stride); \
ST_B2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
}
#define ST_UB4(...) ST_B4(v16u8, __VA_ARGS__)
/* Description : Store vectors of 8 halfword elements with stride
Arguments : Inputs - in0, in1, pdst, stride
Details : Store 8 halfword elements from 'in0' to (pdst)
Store 8 halfword elements from 'in1' to (pdst + stride)
*/
#define ST_H2(RTYPE, in0, in1, pdst, stride) \
{ \
ST_H(RTYPE, in0, (pdst)); \
ST_H(RTYPE, in1, (pdst) + stride); \
}
#define ST_UH2(...) ST_H2(v8u16, __VA_ARGS__)
// TODO(fbarchard): Consider using __msa_vshf_b and __msa_ilvr_b directly.
/* Description : Shuffle byte vector elements as per mask vector
Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
Outputs - out0, out1
Return Type - as per RTYPE
Details : Byte elements from 'in0' & 'in1' are copied selectively to
'out0' as per control vector 'mask0'
*/
#define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) \
{ \
out0 = (RTYPE)__msa_vshf_b((v16i8)mask0, (v16i8)in1, (v16i8)in0); \
out1 = (RTYPE)__msa_vshf_b((v16i8)mask1, (v16i8)in3, (v16i8)in2); \
}
#define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
/* Description : Interleave both left and right half of input vectors
Arguments : Inputs - in0, in1
Outputs - out0, out1
Return Type - as per RTYPE
Details : Right half of byte elements from 'in0' and 'in1' are
interleaved and written to 'out0'
*/
#define ILVRL_B2(RTYPE, in0, in1, out0, out1) \
{ \
out0 = (RTYPE)__msa_ilvr_b((v16i8)in0, (v16i8)in1); \
out1 = (RTYPE)__msa_ilvl_b((v16i8)in0, (v16i8)in1); \
}
#define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
#endif /* !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa) */
#endif // INCLUDE_LIBYUV_MACROS_MSA_H_

15
third_party/yuv/include/libyuv/mjpeg_decoder.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_MJPEG_DECODER_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_MJPEG_DECODER_H_
#define INCLUDE_LIBYUV_MJPEG_DECODER_H_
#include "libyuv/basic_types.h"
@ -37,7 +37,6 @@ static const uint32 kUnknownDataSize = 0xFFFFFFFF;
enum JpegSubsamplingType {
kJpegYuv420,
kJpegYuv422,
kJpegYuv411,
kJpegYuv444,
kJpegYuv400,
kJpegUnknown
@ -145,12 +144,16 @@ class LIBYUV_API MJpegDecoder {
// callback function. Each call will get the data for a whole number of
// image scanlines.
// TODO(fbarchard): Add dst_x, dst_y to allow specific rect to be decoded.
LIBYUV_BOOL DecodeToCallback(CallbackFunction fn, void* opaque,
int dst_width, int dst_height);
LIBYUV_BOOL DecodeToCallback(CallbackFunction fn,
void* opaque,
int dst_width,
int dst_height);
// The helper function which recognizes the jpeg sub-sampling type.
static JpegSubsamplingType JpegSubsamplingTypeHelper(
int* subsample_x, int* subsample_y, int number_of_components);
int* subsample_x,
int* subsample_y,
int number_of_components);
private:
void AllocOutputBuffers(int num_outbufs);
@ -189,4 +192,4 @@ class LIBYUV_API MJpegDecoder {
} // namespace libyuv
#endif // __cplusplus
#endif // INCLUDE_LIBYUV_MJPEG_DECODER_H_ NOLINT
#endif // INCLUDE_LIBYUV_MJPEG_DECODER_H_

735
third_party/yuv/include/libyuv/planar_functions.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_
#define INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_
#include "libyuv/basic_types.h"
@ -24,86 +24,164 @@ extern "C" {
// Copy a plane of data.
LIBYUV_API
void CopyPlane(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
int width, int height);
void CopyPlane(const uint8* src_y,
int src_stride_y,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
LIBYUV_API
void CopyPlane_16(const uint16* src_y, int src_stride_y,
uint16* dst_y, int dst_stride_y,
int width, int height);
void CopyPlane_16(const uint16* src_y,
int src_stride_y,
uint16* dst_y,
int dst_stride_y,
int width,
int height);
// Set a plane of data to a 32 bit value.
LIBYUV_API
void SetPlane(uint8* dst_y, int dst_stride_y,
int width, int height,
void SetPlane(uint8* dst_y,
int dst_stride_y,
int width,
int height,
uint32 value);
// Split interleaved UV plane into separate U and V planes.
LIBYUV_API
void SplitUVPlane(const uint8* src_uv,
int src_stride_uv,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Merge separate U and V planes into one interleaved UV plane.
LIBYUV_API
void MergeUVPlane(const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_uv,
int dst_stride_uv,
int width,
int height);
// Copy I400. Supports inverting.
LIBYUV_API
int I400ToI400(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
int width, int height);
int I400ToI400(const uint8* src_y,
int src_stride_y,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
#define J400ToJ400 I400ToI400
// Copy I422 to I422.
#define I422ToI422 I422Copy
LIBYUV_API
int I422Copy(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I422Copy(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Copy I444 to I444.
#define I444ToI444 I444Copy
LIBYUV_API
int I444Copy(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I444Copy(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert YUY2 to I422.
LIBYUV_API
int YUY2ToI422(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int YUY2ToI422(const uint8* src_yuy2,
int src_stride_yuy2,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Convert UYVY to I422.
LIBYUV_API
int UYVYToI422(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int UYVYToI422(const uint8* src_uyvy,
int src_stride_uyvy,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
LIBYUV_API
int YUY2ToNV12(const uint8* src_yuy2, int src_stride_yuy2,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int YUY2ToNV12(const uint8* src_yuy2,
int src_stride_yuy2,
uint8* dst_y,
int dst_stride_y,
uint8* dst_uv,
int dst_stride_uv,
int width,
int height);
LIBYUV_API
int UYVYToNV12(const uint8* src_uyvy, int src_stride_uyvy,
uint8* dst_y, int dst_stride_y,
uint8* dst_uv, int dst_stride_uv,
int width, int height);
int UYVYToNV12(const uint8* src_uyvy,
int src_stride_uyvy,
uint8* dst_y,
int dst_stride_y,
uint8* dst_uv,
int dst_stride_uv,
int width,
int height);
LIBYUV_API
int YUY2ToY(const uint8* src_yuy2,
int src_stride_yuy2,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Convert I420 to I400. (calls CopyPlane ignoring u/v).
LIBYUV_API
int I420ToI400(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
int width, int height);
int I420ToI400(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Alias
#define J420ToJ400 I420ToI400
@ -111,13 +189,20 @@ int I420ToI400(const uint8* src_y, int src_stride_y,
// I420 mirror.
LIBYUV_API
int I420Mirror(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420Mirror(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Alias
#define I400ToI400Mirror I400Mirror
@ -125,87 +210,139 @@ int I420Mirror(const uint8* src_y, int src_stride_y,
// I400 mirror. A single plane is mirrored horizontally.
// Pass negative height to achieve 180 degree rotation.
LIBYUV_API
int I400Mirror(const uint8* src_y, int src_stride_y,
uint8* dst_y, int dst_stride_y,
int width, int height);
int I400Mirror(const uint8* src_y,
int src_stride_y,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Alias
#define ARGBToARGBMirror ARGBMirror
// ARGB mirror.
LIBYUV_API
int ARGBMirror(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBMirror(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert NV12 to RGB565.
LIBYUV_API
int NV12ToRGB565(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_rgb565, int dst_stride_rgb565,
int width, int height);
int NV12ToRGB565(const uint8* src_y,
int src_stride_y,
const uint8* src_uv,
int src_stride_uv,
uint8* dst_rgb565,
int dst_stride_rgb565,
int width,
int height);
// I422ToARGB is in convert_argb.h
// Convert I422 to BGRA.
LIBYUV_API
int I422ToBGRA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_bgra, int dst_stride_bgra,
int width, int height);
int I422ToBGRA(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_bgra,
int dst_stride_bgra,
int width,
int height);
// Convert I422 to ABGR.
LIBYUV_API
int I422ToABGR(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_abgr, int dst_stride_abgr,
int width, int height);
int I422ToABGR(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_abgr,
int dst_stride_abgr,
int width,
int height);
// Convert I422 to RGBA.
LIBYUV_API
int I422ToRGBA(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_rgba, int dst_stride_rgba,
int width, int height);
int I422ToRGBA(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_rgba,
int dst_stride_rgba,
int width,
int height);
// Alias
#define RGB24ToRAW RAWToRGB24
LIBYUV_API
int RAWToRGB24(const uint8* src_raw, int src_stride_raw,
uint8* dst_rgb24, int dst_stride_rgb24,
int width, int height);
int RAWToRGB24(const uint8* src_raw,
int src_stride_raw,
uint8* dst_rgb24,
int dst_stride_rgb24,
int width,
int height);
// Draw a rectangle into I420.
LIBYUV_API
int I420Rect(uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int x, int y, int width, int height,
int value_y, int value_u, int value_v);
int I420Rect(uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int x,
int y,
int width,
int height,
int value_y,
int value_u,
int value_v);
// Draw a rectangle into ARGB.
LIBYUV_API
int ARGBRect(uint8* dst_argb, int dst_stride_argb,
int x, int y, int width, int height, uint32 value);
int ARGBRect(uint8* dst_argb,
int dst_stride_argb,
int x,
int y,
int width,
int height,
uint32 value);
// Convert ARGB to gray scale ARGB.
LIBYUV_API
int ARGBGrayTo(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBGrayTo(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Make a rectangle of ARGB gray scale.
LIBYUV_API
int ARGBGray(uint8* dst_argb, int dst_stride_argb,
int x, int y, int width, int height);
int ARGBGray(uint8* dst_argb,
int dst_stride_argb,
int x,
int y,
int width,
int height);
// Make a rectangle of ARGB Sepia tone.
LIBYUV_API
int ARGBSepia(uint8* dst_argb, int dst_stride_argb,
int x, int y, int width, int height);
int ARGBSepia(uint8* dst_argb,
int dst_stride_argb,
int x,
int y,
int width,
int height);
// Apply a matrix rotation to each ARGB pixel.
// matrix_argb is 4 signed ARGB values. -128 to 127 representing -2 to 2.
@ -214,10 +351,13 @@ int ARGBSepia(uint8* dst_argb, int dst_stride_argb,
// The next 4 coefficients apply to B, G, R, A and produce R of the output.
// The last 4 coefficients apply to B, G, R, A and produce A of the output.
LIBYUV_API
int ARGBColorMatrix(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int ARGBColorMatrix(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
const int8* matrix_argb,
int width, int height);
int width,
int height);
// Deprecated. Use ARGBColorMatrix instead.
// Apply a matrix rotation to each ARGB pixel.
@ -226,32 +366,47 @@ int ARGBColorMatrix(const uint8* src_argb, int src_stride_argb,
// The next 4 coefficients apply to B, G, R, A and produce G of the output.
// The last 4 coefficients apply to B, G, R, A and produce R of the output.
LIBYUV_API
int RGBColorMatrix(uint8* dst_argb, int dst_stride_argb,
int RGBColorMatrix(uint8* dst_argb,
int dst_stride_argb,
const int8* matrix_rgb,
int x, int y, int width, int height);
int x,
int y,
int width,
int height);
// Apply a color table each ARGB pixel.
// Table contains 256 ARGB values.
LIBYUV_API
int ARGBColorTable(uint8* dst_argb, int dst_stride_argb,
int ARGBColorTable(uint8* dst_argb,
int dst_stride_argb,
const uint8* table_argb,
int x, int y, int width, int height);
int x,
int y,
int width,
int height);
// Apply a color table each ARGB pixel but preserve destination alpha.
// Table contains 256 ARGB values.
LIBYUV_API
int RGBColorTable(uint8* dst_argb, int dst_stride_argb,
int RGBColorTable(uint8* dst_argb,
int dst_stride_argb,
const uint8* table_argb,
int x, int y, int width, int height);
int x,
int y,
int width,
int height);
// Apply a luma/color table each ARGB pixel but preserve destination alpha.
// Table contains 32768 values indexed by [Y][C] where 7 it 7 bit luma from
// RGB (YJ style) and C is an 8 bit color component (R, G or B).
LIBYUV_API
int ARGBLumaColorTable(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int ARGBLumaColorTable(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
const uint8* luma_rgb_table,
int width, int height);
int width,
int height);
// Apply a 3 term polynomial to ARGB values.
// poly points to a 4x4 matrix. The first row is constants. The 2nd row is
@ -262,46 +417,80 @@ int ARGBLumaColorTable(const uint8* src_argb, int src_stride_argb,
// A polynomial approximation can be dirived using software such as 'R'.
LIBYUV_API
int ARGBPolynomial(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int ARGBPolynomial(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
const float* poly,
int width, int height);
int width,
int height);
// Convert plane of 16 bit shorts to half floats.
// Source values are multiplied by scale before storing as half float.
LIBYUV_API
int HalfFloatPlane(const uint16* src_y,
int src_stride_y,
uint16* dst_y,
int dst_stride_y,
float scale,
int width,
int height);
// Quantize a rectangle of ARGB. Alpha unaffected.
// scale is a 16 bit fractional fixed point scaler between 0 and 65535.
// interval_size should be a value between 1 and 255.
// interval_offset should be a value between 0 and 255.
LIBYUV_API
int ARGBQuantize(uint8* dst_argb, int dst_stride_argb,
int scale, int interval_size, int interval_offset,
int x, int y, int width, int height);
int ARGBQuantize(uint8* dst_argb,
int dst_stride_argb,
int scale,
int interval_size,
int interval_offset,
int x,
int y,
int width,
int height);
// Copy ARGB to ARGB.
LIBYUV_API
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopy(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Copy Alpha channel of ARGB to alpha of ARGB.
LIBYUV_API
int ARGBCopyAlpha(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopyAlpha(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Extract the alpha channel from ARGB.
LIBYUV_API
int ARGBExtractAlpha(const uint8* src_argb, int src_stride_argb,
uint8* dst_a, int dst_stride_a,
int width, int height);
int ARGBExtractAlpha(const uint8* src_argb,
int src_stride_argb,
uint8* dst_a,
int dst_stride_a,
int width,
int height);
// Copy Y channel to Alpha of ARGB.
LIBYUV_API
int ARGBCopyYToAlpha(const uint8* src_y, int src_stride_y,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBCopyYToAlpha(const uint8* src_y,
int src_stride_y,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
typedef void (*ARGBBlendRow)(const uint8* src_argb0, const uint8* src_argb1,
uint8* dst_argb, int width);
typedef void (*ARGBBlendRow)(const uint8* src_argb0,
const uint8* src_argb1,
uint8* dst_argb,
int width);
// Get function to Alpha Blend ARGB pixels and store to destination.
LIBYUV_API
@ -311,92 +500,143 @@ ARGBBlendRow GetARGBBlend();
// Source is pre-multiplied by alpha using ARGBAttenuate.
// Alpha of destination is set to 255.
LIBYUV_API
int ARGBBlend(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBBlend(const uint8* src_argb0,
int src_stride_argb0,
const uint8* src_argb1,
int src_stride_argb1,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Alpha Blend plane and store to destination.
// Source is not pre-multiplied by alpha.
LIBYUV_API
int BlendPlane(const uint8* src_y0, int src_stride_y0,
const uint8* src_y1, int src_stride_y1,
const uint8* alpha, int alpha_stride,
uint8* dst_y, int dst_stride_y,
int width, int height);
int BlendPlane(const uint8* src_y0,
int src_stride_y0,
const uint8* src_y1,
int src_stride_y1,
const uint8* alpha,
int alpha_stride,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Alpha Blend YUV images and store to destination.
// Source is not pre-multiplied by alpha.
// Alpha is full width x height and subsampled to half size to apply to UV.
LIBYUV_API
int I420Blend(const uint8* src_y0, int src_stride_y0,
const uint8* src_u0, int src_stride_u0,
const uint8* src_v0, int src_stride_v0,
const uint8* src_y1, int src_stride_y1,
const uint8* src_u1, int src_stride_u1,
const uint8* src_v1, int src_stride_v1,
const uint8* alpha, int alpha_stride,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height);
int I420Blend(const uint8* src_y0,
int src_stride_y0,
const uint8* src_u0,
int src_stride_u0,
const uint8* src_v0,
int src_stride_v0,
const uint8* src_y1,
int src_stride_y1,
const uint8* src_u1,
int src_stride_u1,
const uint8* src_v1,
int src_stride_v1,
const uint8* alpha,
int alpha_stride,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height);
// Multiply ARGB image by ARGB image. Shifted down by 8. Saturates to 255.
LIBYUV_API
int ARGBMultiply(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBMultiply(const uint8* src_argb0,
int src_stride_argb0,
const uint8* src_argb1,
int src_stride_argb1,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Add ARGB image with ARGB image. Saturates to 255.
LIBYUV_API
int ARGBAdd(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBAdd(const uint8* src_argb0,
int src_stride_argb0,
const uint8* src_argb1,
int src_stride_argb1,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Subtract ARGB image (argb1) from ARGB image (argb0). Saturates to 0.
LIBYUV_API
int ARGBSubtract(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBSubtract(const uint8* src_argb0,
int src_stride_argb0,
const uint8* src_argb1,
int src_stride_argb1,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert I422 to YUY2.
LIBYUV_API
int I422ToYUY2(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I422ToYUY2(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
// Convert I422 to UYVY.
LIBYUV_API
int I422ToUYVY(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_frame, int dst_stride_frame,
int width, int height);
int I422ToUYVY(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_frame,
int dst_stride_frame,
int width,
int height);
// Convert unattentuated ARGB to preattenuated ARGB.
LIBYUV_API
int ARGBAttenuate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBAttenuate(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Convert preattentuated ARGB to unattenuated ARGB.
LIBYUV_API
int ARGBUnattenuate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBUnattenuate(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Internal function - do not call directly.
// Computes table of cumulative sum for image where the value is the sum
// of all values above and to the left of the entry. Used by ARGBBlur.
LIBYUV_API
int ARGBComputeCumulativeSum(const uint8* src_argb, int src_stride_argb,
int32* dst_cumsum, int dst_stride32_cumsum,
int width, int height);
int ARGBComputeCumulativeSum(const uint8* src_argb,
int src_stride_argb,
int32* dst_cumsum,
int dst_stride32_cumsum,
int width,
int height);
// Blur ARGB image.
// dst_cumsum table of width * (height + 1) * 16 bytes aligned to
@ -405,49 +645,79 @@ int ARGBComputeCumulativeSum(const uint8* src_argb, int src_stride_argb,
// radius is number of pixels around the center. e.g. 1 = 3x3. 2=5x5.
// Blur is optimized for radius of 5 (11x11) or less.
LIBYUV_API
int ARGBBlur(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int32* dst_cumsum, int dst_stride32_cumsum,
int width, int height, int radius);
int ARGBBlur(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int32* dst_cumsum,
int dst_stride32_cumsum,
int width,
int height,
int radius);
// Multiply ARGB image by ARGB value.
LIBYUV_API
int ARGBShade(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height, uint32 value);
int ARGBShade(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height,
uint32 value);
// Interpolate between two images using specified amount of interpolation
// (0 to 255) and store to destination.
// 'interpolation' is specified as 8 bit fraction where 0 means 100% src0
// and 255 means 1% src0 and 99% src1.
LIBYUV_API
int InterpolatePlane(const uint8* src0, int src_stride0,
const uint8* src1, int src_stride1,
uint8* dst, int dst_stride,
int width, int height, int interpolation);
int InterpolatePlane(const uint8* src0,
int src_stride0,
const uint8* src1,
int src_stride1,
uint8* dst,
int dst_stride,
int width,
int height,
int interpolation);
// Interpolate between two ARGB images using specified amount of interpolation
// Internally calls InterpolatePlane with width * 4 (bpp).
LIBYUV_API
int ARGBInterpolate(const uint8* src_argb0, int src_stride_argb0,
const uint8* src_argb1, int src_stride_argb1,
uint8* dst_argb, int dst_stride_argb,
int width, int height, int interpolation);
int ARGBInterpolate(const uint8* src_argb0,
int src_stride_argb0,
const uint8* src_argb1,
int src_stride_argb1,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height,
int interpolation);
// Interpolate between two YUV images using specified amount of interpolation
// Internally calls InterpolatePlane on each plane where the U and V planes
// are half width and half height.
LIBYUV_API
int I420Interpolate(const uint8* src0_y, int src0_stride_y,
const uint8* src0_u, int src0_stride_u,
const uint8* src0_v, int src0_stride_v,
const uint8* src1_y, int src1_stride_y,
const uint8* src1_u, int src1_stride_u,
const uint8* src1_v, int src1_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height, int interpolation);
int I420Interpolate(const uint8* src0_y,
int src0_stride_y,
const uint8* src0_u,
int src0_stride_u,
const uint8* src0_v,
int src0_stride_v,
const uint8* src1_y,
int src1_stride_y,
const uint8* src1_u,
int src1_stride_u,
const uint8* src1_v,
int src1_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height,
int interpolation);
#if defined(__pnacl__) || defined(__CLR_VER) || \
(defined(__i386__) && !defined(__SSE2__))
@ -468,40 +738,59 @@ int I420Interpolate(const uint8* src0_y, int src0_stride_y,
// Row function for copying pixels from a source with a slope to a row
// of destination. Useful for scaling, rotation, mirror, texture mapping.
LIBYUV_API
void ARGBAffineRow_C(const uint8* src_argb, int src_argb_stride,
uint8* dst_argb, const float* uv_dudv, int width);
void ARGBAffineRow_C(const uint8* src_argb,
int src_argb_stride,
uint8* dst_argb,
const float* uv_dudv,
int width);
LIBYUV_API
void ARGBAffineRow_SSE2(const uint8* src_argb, int src_argb_stride,
uint8* dst_argb, const float* uv_dudv, int width);
void ARGBAffineRow_SSE2(const uint8* src_argb,
int src_argb_stride,
uint8* dst_argb,
const float* uv_dudv,
int width);
// Shuffle ARGB channel order. e.g. BGRA to ARGB.
// shuffler is 16 bytes and must be aligned.
LIBYUV_API
int ARGBShuffle(const uint8* src_bgra, int src_stride_bgra,
uint8* dst_argb, int dst_stride_argb,
const uint8* shuffler, int width, int height);
int ARGBShuffle(const uint8* src_bgra,
int src_stride_bgra,
uint8* dst_argb,
int dst_stride_argb,
const uint8* shuffler,
int width,
int height);
// Sobel ARGB effect with planar output.
LIBYUV_API
int ARGBSobelToPlane(const uint8* src_argb, int src_stride_argb,
uint8* dst_y, int dst_stride_y,
int width, int height);
int ARGBSobelToPlane(const uint8* src_argb,
int src_stride_argb,
uint8* dst_y,
int dst_stride_y,
int width,
int height);
// Sobel ARGB effect.
LIBYUV_API
int ARGBSobel(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBSobel(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
// Sobel ARGB effect w/ Sobel X, Sobel, Sobel Y in ARGB.
LIBYUV_API
int ARGBSobelXY(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
int ARGBSobelXY(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_ NOLINT
#endif // INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_

143
third_party/yuv/include/libyuv/rotate.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_ROTATE_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_ROTATE_H_
#define INCLUDE_LIBYUV_ROTATE_H_
#include "libyuv/basic_types.h"
@ -20,8 +20,8 @@ extern "C" {
// Supported rotation.
typedef enum RotationMode {
kRotate0 = 0, // No rotation.
kRotate90 = 90, // Rotate 90 degrees clockwise.
kRotate0 = 0, // No rotation.
kRotate90 = 90, // Rotate 90 degrees clockwise.
kRotate180 = 180, // Rotate 180 degrees.
kRotate270 = 270, // Rotate 270 degrees clockwise.
@ -33,85 +33,132 @@ typedef enum RotationMode {
// Rotate I420 frame.
LIBYUV_API
int I420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height, enum RotationMode mode);
int I420Rotate(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int src_width,
int src_height,
enum RotationMode mode);
// Rotate NV12 input and store in I420.
LIBYUV_API
int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int src_width, int src_height, enum RotationMode mode);
int NV12ToI420Rotate(const uint8* src_y,
int src_stride_y,
const uint8* src_uv,
int src_stride_uv,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int src_width,
int src_height,
enum RotationMode mode);
// Rotate a plane by 0, 90, 180, or 270.
LIBYUV_API
int RotatePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int src_width, int src_height, enum RotationMode mode);
int RotatePlane(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int src_width,
int src_height,
enum RotationMode mode);
// Rotate planes by 90, 180, 270. Deprecated.
LIBYUV_API
void RotatePlane90(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane90(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotatePlane180(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane180(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotatePlane270(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void RotatePlane270(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void RotateUV90(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV90(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height);
// Rotations for when U and V are interleaved.
// These functions take one input pointer and
// split the data into two buffers while
// rotating them. Deprecated.
LIBYUV_API
void RotateUV180(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV180(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height);
LIBYUV_API
void RotateUV270(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void RotateUV270(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height);
// The 90 and 270 functions are based on transposes.
// Doing a transpose with reversing the read/write
// order will result in a rotation by +- 90 degrees.
// Deprecated.
LIBYUV_API
void TransposePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height);
void TransposePlane(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height);
LIBYUV_API
void TransposeUV(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void TransposeUV(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_ROTATE_H_ NOLINT
#endif // INCLUDE_LIBYUV_ROTATE_H_

14
third_party/yuv/include/libyuv/rotate_argb.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_ROTATE_ARGB_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_ROTATE_ARGB_H_
#define INCLUDE_LIBYUV_ROTATE_ARGB_H_
#include "libyuv/basic_types.h"
@ -21,13 +21,17 @@ extern "C" {
// Rotate ARGB frame
LIBYUV_API
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int src_width, int src_height, enum RotationMode mode);
int ARGBRotate(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int src_width,
int src_height,
enum RotationMode mode);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_ROTATE_ARGB_H_ NOLINT
#endif // INCLUDE_LIBYUV_ROTATE_ARGB_H_

213
third_party/yuv/include/libyuv/rotate_row.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_ROTATE_ROW_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_ROTATE_ROW_H_
#define INCLUDE_LIBYUV_ROTATE_ROW_H_
#include "libyuv/basic_types.h"
@ -36,7 +36,8 @@ extern "C" {
// The following are available for GCC 32 or 64 bit but not NaCL for 64 bit:
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__i386__) || (defined(__x86_64__) && !defined(__native_client__)))
(defined(__i386__) || \
(defined(__x86_64__) && !defined(__native_client__)))
#define HAS_TRANSPOSEWX8_SSSE3
#endif
@ -53,69 +54,175 @@ extern "C" {
#define HAS_TRANSPOSEUVWX8_NEON
#endif
#if !defined(LIBYUV_DISABLE_MIPS) && !defined(__native_client__) && \
defined(__mips__) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#if !defined(LIBYUV_DISABLE_DSPR2) && !defined(__native_client__) && \
defined(__mips__) && defined(__mips_dsp) && (__mips_dsp_rev >= 2)
#define HAS_TRANSPOSEWX8_DSPR2
#define HAS_TRANSPOSEUVWX8_DSPR2
#endif // defined(__mips__)
void TransposeWxH_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height);
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#define HAS_TRANSPOSEWX16_MSA
#define HAS_TRANSPOSEUVWX16_MSA
#endif
void TransposeWx8_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWxH_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height);
void TransposeWx8_Any_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Any_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Fast_Any_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_Any_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width);
void TransposeWx8_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx16_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_NEON(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_Fast_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_DSPR2(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_Fast_DSPR2(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx16_MSA(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeUVWxH_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height);
void TransposeWx8_Any_NEON(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_Any_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_Fast_Any_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx8_Any_DSPR2(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeWx16_Any_MSA(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width);
void TransposeUVWx8_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWxH_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height);
void TransposeUVWx8_Any_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_Any_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_Any_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width);
void TransposeUVWx8_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_SSE2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_NEON(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_DSPR2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_MSA(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_SSE2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_NEON(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx8_Any_DSPR2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
void TransposeUVWx16_Any_MSA(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_ROTATE_ROW_H_ NOLINT
#endif // INCLUDE_LIBYUV_ROTATE_ROW_H_

2301
third_party/yuv/include/libyuv/row.h vendored
View File

File diff suppressed because it is too large Load Diff

113
third_party/yuv/include/libyuv/scale.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_SCALE_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_SCALE_H_
#define INCLUDE_LIBYUV_SCALE_H_
#include "libyuv/basic_types.h"
@ -20,25 +20,33 @@ extern "C" {
// Supported filtering.
typedef enum FilterMode {
kFilterNone = 0, // Point sample; Fastest.
kFilterLinear = 1, // Filter horizontally only.
kFilterNone = 0, // Point sample; Fastest.
kFilterLinear = 1, // Filter horizontally only.
kFilterBilinear = 2, // Faster than box, but lower quality scaling down.
kFilterBox = 3 // Highest quality.
kFilterBox = 3 // Highest quality.
} FilterModeEnum;
// Scale a YUV plane.
LIBYUV_API
void ScalePlane(const uint8* src, int src_stride,
int src_width, int src_height,
uint8* dst, int dst_stride,
int dst_width, int dst_height,
void ScalePlane(const uint8* src,
int src_stride,
int src_width,
int src_height,
uint8* dst,
int dst_stride,
int dst_width,
int dst_height,
enum FilterMode filtering);
LIBYUV_API
void ScalePlane_16(const uint16* src, int src_stride,
int src_width, int src_height,
uint16* dst, int dst_stride,
int dst_width, int dst_height,
void ScalePlane_16(const uint16* src,
int src_stride,
int src_width,
int src_height,
uint16* dst,
int dst_stride,
int dst_width,
int dst_height,
enum FilterMode filtering);
// Scales a YUV 4:2:0 image from the src width and height to the
@ -52,42 +60,73 @@ void ScalePlane_16(const uint16* src, int src_stride,
// Returns 0 if successful.
LIBYUV_API
int I420Scale(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
int src_width, int src_height,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int dst_width, int dst_height,
int I420Scale(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
int src_width,
int src_height,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int dst_width,
int dst_height,
enum FilterMode filtering);
LIBYUV_API
int I420Scale_16(const uint16* src_y, int src_stride_y,
const uint16* src_u, int src_stride_u,
const uint16* src_v, int src_stride_v,
int src_width, int src_height,
uint16* dst_y, int dst_stride_y,
uint16* dst_u, int dst_stride_u,
uint16* dst_v, int dst_stride_v,
int dst_width, int dst_height,
int I420Scale_16(const uint16* src_y,
int src_stride_y,
const uint16* src_u,
int src_stride_u,
const uint16* src_v,
int src_stride_v,
int src_width,
int src_height,
uint16* dst_y,
int dst_stride_y,
uint16* dst_u,
int dst_stride_u,
uint16* dst_v,
int dst_stride_v,
int dst_width,
int dst_height,
enum FilterMode filtering);
#ifdef __cplusplus
// Legacy API. Deprecated.
LIBYUV_API
int Scale(const uint8* src_y, const uint8* src_u, const uint8* src_v,
int src_stride_y, int src_stride_u, int src_stride_v,
int src_width, int src_height,
uint8* dst_y, uint8* dst_u, uint8* dst_v,
int dst_stride_y, int dst_stride_u, int dst_stride_v,
int dst_width, int dst_height,
int Scale(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
int src_stride_y,
int src_stride_u,
int src_stride_v,
int src_width,
int src_height,
uint8* dst_y,
uint8* dst_u,
uint8* dst_v,
int dst_stride_y,
int dst_stride_u,
int dst_stride_v,
int dst_width,
int dst_height,
LIBYUV_BOOL interpolate);
// Legacy API. Deprecated.
LIBYUV_API
int ScaleOffset(const uint8* src_i420, int src_width, int src_height,
uint8* dst_i420, int dst_width, int dst_height, int dst_yoffset,
int ScaleOffset(const uint8* src_i420,
int src_width,
int src_height,
uint8* dst_i420,
int dst_width,
int dst_height,
int dst_yoffset,
LIBYUV_BOOL interpolate);
// For testing, allow disabling of specialized scalers.
@ -100,4 +139,4 @@ void SetUseReferenceImpl(LIBYUV_BOOL use);
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_SCALE_H_ NOLINT
#endif // INCLUDE_LIBYUV_SCALE_H_

56
third_party/yuv/include/libyuv/scale_argb.h vendored
View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_SCALE_ARGB_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_SCALE_ARGB_H_
#define INCLUDE_LIBYUV_SCALE_ARGB_H_
#include "libyuv/basic_types.h"
@ -20,32 +20,52 @@ extern "C" {
#endif
LIBYUV_API
int ARGBScale(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int ARGBScale(const uint8* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
enum FilterMode filtering);
// Clipped scale takes destination rectangle coordinates for clip values.
LIBYUV_API
int ARGBScaleClip(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int ARGBScaleClip(const uint8* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering);
// Scale with YUV conversion to ARGB and clipping.
LIBYUV_API
int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
int YUVToARGBScaleClip(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint32 src_fourcc,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
uint32 dst_fourcc,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering);
#ifdef __cplusplus
@ -53,4 +73,4 @@ int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y,
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_SCALE_ARGB_H_ NOLINT
#endif // INCLUDE_LIBYUV_SCALE_ARGB_H_

987
third_party/yuv/include/libyuv/scale_row.h vendored
View File

File diff suppressed because it is too large Load Diff

6
third_party/yuv/include/libyuv/version.h vendored
View File

@ -8,9 +8,9 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1598
#define LIBYUV_VERSION 1662
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
#endif // INCLUDE_LIBYUV_VERSION_H_

28
third_party/yuv/include/libyuv/video_common.h vendored
View File

@ -10,7 +10,7 @@
// Common definitions for video, including fourcc and VideoFormat.
#ifndef INCLUDE_LIBYUV_VIDEO_COMMON_H_ // NOLINT
#ifndef INCLUDE_LIBYUV_VIDEO_COMMON_H_
#define INCLUDE_LIBYUV_VIDEO_COMMON_H_
#include "libyuv/basic_types.h"
@ -28,13 +28,13 @@ extern "C" {
// Needs to be a macro otherwise the OS X compiler complains when the kFormat*
// constants are used in a switch.
#ifdef __cplusplus
#define FOURCC(a, b, c, d) ( \
(static_cast<uint32>(a)) | (static_cast<uint32>(b) << 8) | \
(static_cast<uint32>(c) << 16) | (static_cast<uint32>(d) << 24))
#define FOURCC(a, b, c, d) \
((static_cast<uint32>(a)) | (static_cast<uint32>(b) << 8) | \
(static_cast<uint32>(c) << 16) | (static_cast<uint32>(d) << 24))
#else
#define FOURCC(a, b, c, d) ( \
((uint32)(a)) | ((uint32)(b) << 8) | /* NOLINT */ \
((uint32)(c) << 16) | ((uint32)(d) << 24)) /* NOLINT */
#define FOURCC(a, b, c, d) \
(((uint32)(a)) | ((uint32)(b) << 8) | /* NOLINT */ \
((uint32)(c) << 16) | ((uint32)(d) << 24)) /* NOLINT */
#endif
// Some pages discussing FourCC codes:
@ -49,18 +49,18 @@ extern "C" {
// Secondary formats are converted in 2 steps.
// Auxilliary formats call primary converters.
enum FourCC {
// 9 Primary YUV formats: 5 planar, 2 biplanar, 2 packed.
// 8 Primary YUV formats: 5 planar, 2 biplanar, 2 packed.
FOURCC_I420 = FOURCC('I', '4', '2', '0'),
FOURCC_I422 = FOURCC('I', '4', '2', '2'),
FOURCC_I444 = FOURCC('I', '4', '4', '4'),
FOURCC_I411 = FOURCC('I', '4', '1', '1'),
FOURCC_I411 = FOURCC('I', '4', '1', '1'), // deprecated.
FOURCC_I400 = FOURCC('I', '4', '0', '0'),
FOURCC_NV21 = FOURCC('N', 'V', '2', '1'),
FOURCC_NV12 = FOURCC('N', 'V', '1', '2'),
FOURCC_YUY2 = FOURCC('Y', 'U', 'Y', '2'),
FOURCC_UYVY = FOURCC('U', 'Y', 'V', 'Y'),
// 2 Secondary YUV formats: row biplanar.
// 1 Secondary YUV format: row biplanar.
FOURCC_M420 = FOURCC('M', '4', '2', '0'),
FOURCC_Q420 = FOURCC('Q', '4', '2', '0'), // deprecated.
@ -69,7 +69,7 @@ enum FourCC {
FOURCC_BGRA = FOURCC('B', 'G', 'R', 'A'),
FOURCC_ABGR = FOURCC('A', 'B', 'G', 'R'),
FOURCC_24BG = FOURCC('2', '4', 'B', 'G'),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RAW = FOURCC('r', 'a', 'w', ' '),
FOURCC_RGBA = FOURCC('R', 'G', 'B', 'A'),
FOURCC_RGBP = FOURCC('R', 'G', 'B', 'P'), // rgb565 LE.
FOURCC_RGBO = FOURCC('R', 'G', 'B', 'O'), // argb1555 LE.
@ -137,7 +137,7 @@ enum FourCCBpp {
FOURCC_BPP_ABGR = 32,
FOURCC_BPP_RGBA = 32,
FOURCC_BPP_24BG = 24,
FOURCC_BPP_RAW = 24,
FOURCC_BPP_RAW = 24,
FOURCC_BPP_RGBP = 16,
FOURCC_BPP_RGBO = 16,
FOURCC_BPP_R444 = 16,
@ -170,7 +170,7 @@ enum FourCCBpp {
FOURCC_BPP_CM24 = 24,
// Match any fourcc.
FOURCC_BPP_ANY = 0, // 0 means unknown.
FOURCC_BPP_ANY = 0, // 0 means unknown.
};
// Converts fourcc aliases into canonical ones.
@ -181,4 +181,4 @@ LIBYUV_API uint32 CanonicalFourCC(uint32 fourcc);
} // namespace libyuv
#endif
#endif // INCLUDE_LIBYUV_VIDEO_COMMON_H_ NOLINT
#endif // INCLUDE_LIBYUV_VIDEO_COMMON_H_

185
third_party/yuv/source/compare.cc vendored
View File

@ -32,8 +32,7 @@ LIBYUV_API
uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed) {
const int kBlockSize = 1 << 15; // 32768;
int remainder;
uint32 (*HashDjb2_SSE)(const uint8* src, int count, uint32 seed) =
HashDjb2_C;
uint32 (*HashDjb2_SSE)(const uint8* src, int count, uint32 seed) = HashDjb2_C;
#if defined(HAS_HASHDJB2_SSE41)
if (TestCpuFlag(kCpuHasSSE41)) {
HashDjb2_SSE = HashDjb2_SSE41;
@ -50,13 +49,13 @@ uint32 HashDjb2(const uint8* src, uint64 count, uint32 seed) {
src += kBlockSize;
count -= kBlockSize;
}
remainder = (int)(count) & ~15;
remainder = (int)count & ~15;
if (remainder) {
seed = HashDjb2_SSE(src, remainder, seed);
src += remainder;
count -= remainder;
}
remainder = (int)(count) & 15;
remainder = (int)count & 15;
if (remainder) {
seed = HashDjb2_C(src, remainder, seed);
}
@ -111,9 +110,55 @@ uint32 ARGBDetect(const uint8* argb, int stride_argb, int width, int height) {
return fourcc;
}
LIBYUV_API
uint64 ComputeHammingDistance(const uint8* src_a,
const uint8* src_b,
int count) {
const int kBlockSize = 65536;
int remainder = count & (kBlockSize - 1) & ~31;
uint64 diff = 0;
int i;
uint32 (*HammingDistance)(const uint8* src_a, const uint8* src_b, int count) =
HammingDistance_C;
#if defined(HAS_HAMMINGDISTANCE_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
HammingDistance = HammingDistance_NEON;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_X86)
if (TestCpuFlag(kCpuHasX86)) {
HammingDistance = HammingDistance_X86;
}
#endif
#if defined(HAS_HAMMINGDISTANCE_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
HammingDistance = HammingDistance_AVX2;
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+ : diff)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
diff += HammingDistance(src_a + i, src_b + i, kBlockSize);
}
src_a += count & ~(kBlockSize - 1);
src_b += count & ~(kBlockSize - 1);
if (remainder) {
diff += HammingDistance(src_a, src_b, remainder);
src_a += remainder;
src_b += remainder;
}
remainder = count & 31;
if (remainder) {
diff += HammingDistance_C(src_a, src_b, remainder);
}
return diff;
}
// TODO(fbarchard): Refactor into row function.
LIBYUV_API
uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
uint64 ComputeSumSquareError(const uint8* src_a,
const uint8* src_b,
int count) {
// SumSquareError returns values 0 to 65535 for each squared difference.
// Up to 65536 of those can be summed and remain within a uint32.
@ -142,7 +187,7 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
}
#endif
#ifdef _OPENMP
#pragma omp parallel for reduction(+: sse)
#pragma omp parallel for reduction(+ : sse)
#endif
for (i = 0; i < (count - (kBlockSize - 1)); i += kBlockSize) {
sse += SumSquareError(src_a + i, src_b + i, kBlockSize);
@ -162,14 +207,16 @@ uint64 ComputeSumSquareError(const uint8* src_a, const uint8* src_b,
}
LIBYUV_API
uint64 ComputeSumSquareErrorPlane(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
uint64 ComputeSumSquareErrorPlane(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height) {
uint64 sse = 0;
int h;
// Coalesce rows.
if (stride_a == width &&
stride_b == width) {
if (stride_a == width && stride_b == width) {
width *= height;
height = 1;
stride_a = stride_b = 0;
@ -186,10 +233,10 @@ LIBYUV_API
double SumSquareErrorToPsnr(uint64 sse, uint64 count) {
double psnr;
if (sse > 0) {
double mse = (double)(count) / (double)(sse);
double mse = (double)count / (double)sse;
psnr = 10.0 * log10(255.0 * 255.0 * mse);
} else {
psnr = kMaxPsnr; // Limit to prevent divide by 0
psnr = kMaxPsnr; // Limit to prevent divide by 0
}
if (psnr > kMaxPsnr)
@ -199,45 +246,53 @@ double SumSquareErrorToPsnr(uint64 sse, uint64 count) {
}
LIBYUV_API
double CalcFramePsnr(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
double CalcFramePsnr(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height) {
const uint64 samples = width * height;
const uint64 sse = ComputeSumSquareErrorPlane(src_a, stride_a,
src_b, stride_b,
width, height);
const uint64 sse = ComputeSumSquareErrorPlane(src_a, stride_a, src_b,
stride_b, width, height);
return SumSquareErrorToPsnr(sse, samples);
}
LIBYUV_API
double I420Psnr(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height) {
const uint64 sse_y = ComputeSumSquareErrorPlane(src_y_a, stride_y_a,
src_y_b, stride_y_b,
width, height);
double I420Psnr(const uint8* src_y_a,
int stride_y_a,
const uint8* src_u_a,
int stride_u_a,
const uint8* src_v_a,
int stride_v_a,
const uint8* src_y_b,
int stride_y_b,
const uint8* src_u_b,
int stride_u_b,
const uint8* src_v_b,
int stride_v_b,
int width,
int height) {
const uint64 sse_y = ComputeSumSquareErrorPlane(src_y_a, stride_y_a, src_y_b,
stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const uint64 sse_u = ComputeSumSquareErrorPlane(src_u_a, stride_u_a,
src_u_b, stride_u_b,
width_uv, height_uv);
const uint64 sse_v = ComputeSumSquareErrorPlane(src_v_a, stride_v_a,
src_v_b, stride_v_b,
width_uv, height_uv);
const uint64 sse_u = ComputeSumSquareErrorPlane(
src_u_a, stride_u_a, src_u_b, stride_u_b, width_uv, height_uv);
const uint64 sse_v = ComputeSumSquareErrorPlane(
src_v_a, stride_v_a, src_v_b, stride_v_b, width_uv, height_uv);
const uint64 samples = width * height + 2 * (width_uv * height_uv);
const uint64 sse = sse_y + sse_u + sse_v;
return SumSquareErrorToPsnr(sse, samples);
}
static const int64 cc1 = 26634; // (64^2*(.01*255)^2
static const int64 cc1 = 26634; // (64^2*(.01*255)^2
static const int64 cc2 = 239708; // (64^2*(.03*255)^2
static double Ssim8x8_C(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b) {
static double Ssim8x8_C(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b) {
int64 sum_a = 0;
int64 sum_b = 0;
int64 sum_sq_a = 0;
@ -270,12 +325,12 @@ static double Ssim8x8_C(const uint8* src_a, int stride_a,
const int64 ssim_n = (2 * sum_a_x_sum_b + c1) *
(2 * count * sum_axb - 2 * sum_a_x_sum_b + c2);
const int64 sum_a_sq = sum_a*sum_a;
const int64 sum_b_sq = sum_b*sum_b;
const int64 sum_a_sq = sum_a * sum_a;
const int64 sum_b_sq = sum_b * sum_b;
const int64 ssim_d = (sum_a_sq + sum_b_sq + c1) *
(count * sum_sq_a - sum_a_sq +
count * sum_sq_b - sum_b_sq + c2);
const int64 ssim_d =
(sum_a_sq + sum_b_sq + c1) *
(count * sum_sq_a - sum_a_sq + count * sum_sq_b - sum_b_sq + c2);
if (ssim_d == 0.0) {
return DBL_MAX;
@ -288,13 +343,16 @@ static double Ssim8x8_C(const uint8* src_a, int stride_a,
// on the 4x4 pixel grid. Such arrangement allows the windows to overlap
// block boundaries to penalize blocking artifacts.
LIBYUV_API
double CalcFrameSsim(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b,
int width, int height) {
double CalcFrameSsim(const uint8* src_a,
int stride_a,
const uint8* src_b,
int stride_b,
int width,
int height) {
int samples = 0;
double ssim_total = 0;
double (*Ssim8x8)(const uint8* src_a, int stride_a,
const uint8* src_b, int stride_b) = Ssim8x8_C;
double (*Ssim8x8)(const uint8* src_a, int stride_a, const uint8* src_b,
int stride_b) = Ssim8x8_C;
// sample point start with each 4x4 location
int i;
@ -314,22 +372,27 @@ double CalcFrameSsim(const uint8* src_a, int stride_a,
}
LIBYUV_API
double I420Ssim(const uint8* src_y_a, int stride_y_a,
const uint8* src_u_a, int stride_u_a,
const uint8* src_v_a, int stride_v_a,
const uint8* src_y_b, int stride_y_b,
const uint8* src_u_b, int stride_u_b,
const uint8* src_v_b, int stride_v_b,
int width, int height) {
const double ssim_y = CalcFrameSsim(src_y_a, stride_y_a,
src_y_b, stride_y_b, width, height);
double I420Ssim(const uint8* src_y_a,
int stride_y_a,
const uint8* src_u_a,
int stride_u_a,
const uint8* src_v_a,
int stride_v_a,
const uint8* src_y_b,
int stride_y_b,
const uint8* src_u_b,
int stride_u_b,
const uint8* src_v_b,
int stride_v_b,
int width,
int height) {
const double ssim_y =
CalcFrameSsim(src_y_a, stride_y_a, src_y_b, stride_y_b, width, height);
const int width_uv = (width + 1) >> 1;
const int height_uv = (height + 1) >> 1;
const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a,
src_u_b, stride_u_b,
const double ssim_u = CalcFrameSsim(src_u_a, stride_u_a, src_u_b, stride_u_b,
width_uv, height_uv);
const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a,
src_v_b, stride_v_b,
const double ssim_v = CalcFrameSsim(src_v_a, stride_v_a, src_v_b, stride_v_b,
width_uv, height_uv);
return ssim_y * 0.8 + 0.1 * (ssim_u + ssim_v);
}

44
third_party/yuv/source/compare_common.cc vendored
View File

@ -17,6 +17,50 @@ namespace libyuv {
extern "C" {
#endif
#if ORIGINAL_OPT
uint32 HammingDistance_C(const uint8* src_a, const uint8* src_b, int count) {
uint32 diff = 0u;
int i;
for (i = 0; i < count; ++i) {
int x = src_a[i] ^ src_b[i];
if (x & 1)
++diff;
if (x & 2)
++diff;
if (x & 4)
++diff;
if (x & 8)
++diff;
if (x & 16)
++diff;
if (x & 32)
++diff;
if (x & 64)
++diff;
if (x & 128)
++diff;
}
return diff;
}
#endif
// Hakmem method for hamming distance.
uint32 HammingDistance_C(const uint8* src_a, const uint8* src_b, int count) {
uint32 diff = 0u;
int i;
for (i = 0; i < count - 3; i += 4) {
uint32 x = *((uint32*)src_a) ^ *((uint32*)src_b);
uint32 u = x - ((x >> 1) & 0x55555555);
u = ((u >> 2) & 0x33333333) + (u & 0x33333333);
diff += ((((u + (u >> 4)) & 0x0f0f0f0f) * 0x01010101) >> 24);
src_a += 4;
src_b += 4;
}
return diff;
}
uint32 SumSquareError_C(const uint8* src_a, const uint8* src_b, int count) {
uint32 sse = 0u;
int i;

48
third_party/yuv/source/compare_gcc.cc vendored
View File

@ -22,6 +22,19 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
uint32 HammingDistance_X86(const uint8* src_a, const uint8* src_b, int count) {
uint32 diff = 0u;
int i;
for (i = 0; i < count - 7; i += 8) {
uint64 x = *((uint64*)src_a) ^ *((uint64*)src_b);
src_a += 8;
src_b += 8;
diff += __builtin_popcountll(x);
}
return diff;
}
uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
uint32 sse;
asm volatile (
@ -62,30 +75,30 @@ uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
return sse;
}
static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 }; // 33 ^ 16
static uvec32 kHash16x33 = {0x92d9e201, 0, 0, 0}; // 33 ^ 16
static uvec32 kHashMul0 = {
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
0x0c3525e1, // 33 ^ 15
0xa3476dc1, // 33 ^ 14
0x3b4039a1, // 33 ^ 13
0x4f5f0981, // 33 ^ 12
};
static uvec32 kHashMul1 = {
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
0x30f35d61, // 33 ^ 11
0x855cb541, // 33 ^ 10
0x040a9121, // 33 ^ 9
0x747c7101, // 33 ^ 8
};
static uvec32 kHashMul2 = {
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
0xec41d4e1, // 33 ^ 7
0x4cfa3cc1, // 33 ^ 6
0x025528a1, // 33 ^ 5
0x00121881, // 33 ^ 4
};
static uvec32 kHashMul3 = {
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
0x00008c61, // 33 ^ 3
0x00000441, // 33 ^ 2
0x00000021, // 33 ^ 1
0x00000001, // 33 ^ 0
};
uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
@ -148,4 +161,3 @@ uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
} // extern "C"
} // namespace libyuv
#endif

38
third_party/yuv/source/compare_neon.cc vendored
View File

@ -21,8 +21,42 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32 HammingDistance_NEON(const uint8* src_a, const uint8* src_b, int count) {
uint32 diff;
asm volatile (
"vmov.u16 q4, #0 \n" // accumulator
"1: \n"
"vld1.8 {q0, q1}, [%0]! \n"
"vld1.8 {q2, q3}, [%1]! \n"
"veor.32 q0, q0, q2 \n"
"veor.32 q1, q1, q3 \n"
"vcnt.i8 q0, q0 \n"
"vcnt.i8 q1, q1 \n"
"subs %2, %2, #32 \n"
"vadd.u8 q0, q0, q1 \n" // 16 byte counts
"vpadal.u8 q4, q0 \n" // 8 shorts
"bgt 1b \n"
"vpaddl.u16 q0, q4 \n" // 4 ints
"vpadd.u32 d0, d0, d1 \n"
"vpadd.u32 d0, d0, d0 \n"
"vmov.32 %3, d0[0] \n"
: "+r"(src_a),
"+r"(src_b),
"+r"(count),
"=r"(diff)
:
: "cc", "q0", "q1", "q2", "q3", "q4");
return diff;
}
uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
volatile uint32 sse;
uint32 sse;
asm volatile (
"vmov.u8 q8, #0 \n"
"vmov.u8 q10, #0 \n"
@ -30,9 +64,7 @@ uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
"vmov.u8 q11, #0 \n"
"1: \n"
MEMACCESS(0)
"vld1.8 {q0}, [%0]! \n"
MEMACCESS(1)
"vld1.8 {q1}, [%1]! \n"
"subs %2, %2, #16 \n"
"vsubl.u8 q2, d0, d2 \n"

34
third_party/yuv/source/compare_neon64.cc vendored
View File

@ -20,8 +20,38 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
// 256 bits at a time
// uses short accumulator which restricts count to 131 KB
uint32 HammingDistance_NEON(const uint8* src_a, const uint8* src_b, int count) {
uint32 diff;
asm volatile (
"movi v4.8h, #0 \n"
"1: \n"
"ld1 {v0.16b, v1.16b}, [%0], #32 \n"
"ld1 {v2.16b, v3.16b}, [%1], #32 \n"
"eor v0.16b, v0.16b, v2.16b \n"
"eor v1.16b, v1.16b, v3.16b \n"
"cnt v0.16b, v0.16b \n"
"cnt v1.16b, v1.16b \n"
"subs %w2, %w2, #32 \n"
"add v0.16b, v0.16b, v1.16b \n"
"uadalp v4.8h, v0.16b \n"
"b.gt 1b \n"
"uaddlv s4, v4.8h \n"
"fmov %w3, s4 \n"
: "+r"(src_a),
"+r"(src_b),
"+r"(count),
"=r"(diff)
:
: "cc", "v0", "v1", "v2", "v3", "v4");
return diff;
}
uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
volatile uint32 sse;
uint32 sse;
asm volatile (
"eor v16.16b, v16.16b, v16.16b \n"
"eor v18.16b, v18.16b, v18.16b \n"
@ -29,9 +59,7 @@ uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) {
"eor v19.16b, v19.16b, v19.16b \n"
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b}, [%0], #16 \n"
MEMACCESS(1)
"ld1 {v1.16b}, [%1], #16 \n"
"subs %w2, %w2, #16 \n"
"usubl v2.8h, v0.8b, v1.8b \n"

969
third_party/yuv/source/convert.cc vendored
View File

File diff suppressed because it is too large Load Diff

1132
third_party/yuv/source/convert_argb.cc vendored
View File

File diff suppressed because it is too large Load Diff

990
third_party/yuv/source/convert_from.cc vendored
View File

File diff suppressed because it is too large Load Diff

710
third_party/yuv/source/convert_from_argb.cc vendored
View File

File diff suppressed because it is too large Load Diff

183
third_party/yuv/source/convert_jpeg.cc vendored
View File

@ -37,13 +37,9 @@ static void JpegCopyI420(void* opaque,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I420Copy(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I420Copy(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@ -55,13 +51,9 @@ static void JpegI422ToI420(void* opaque,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I422ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I422ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@ -73,31 +65,9 @@ static void JpegI444ToI420(void* opaque,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I444ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
dest->h -= rows;
}
static void JpegI411ToI420(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I411ToI420(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I444ToI420(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->y, dest->y_stride, dest->u, dest->u_stride, dest->v,
dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@ -109,11 +79,8 @@ static void JpegI400ToI420(void* opaque,
const int* strides,
int rows) {
I420Buffers* dest = (I420Buffers*)(opaque);
I400ToI420(data[0], strides[0],
dest->y, dest->y_stride,
dest->u, dest->u_stride,
dest->v, dest->v_stride,
dest->w, rows);
I400ToI420(data[0], strides[0], dest->y, dest->y_stride, dest->u,
dest->u_stride, dest->v, dest->v_stride, dest->w, rows);
dest->y += rows * dest->y_stride;
dest->u += ((rows + 1) >> 1) * dest->u_stride;
dest->v += ((rows + 1) >> 1) * dest->v_stride;
@ -122,8 +89,7 @@ static void JpegI400ToI420(void* opaque,
// Query size of MJPG in pixels.
LIBYUV_API
int MJPGSize(const uint8* sample, size_t sample_size,
int* width, int* height) {
int MJPGSize(const uint8* sample, size_t sample_size, int* width, int* height) {
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
if (ret) {
@ -139,11 +105,16 @@ int MJPGSize(const uint8* sample, size_t sample_size,
LIBYUV_API
int MJPGToI420(const uint8* sample,
size_t sample_size,
uint8* y, int y_stride,
uint8* u, int u_stride,
uint8* v, int v_stride,
int w, int h,
int dw, int dh) {
uint8* y,
int y_stride,
uint8* u,
int u_stride,
uint8* v,
int v_stride,
int w,
int h,
int dw,
int dh) {
if (sample_size == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
@ -152,17 +123,16 @@ int MJPGToI420(const uint8* sample,
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
if (ret && (mjpeg_decoder.GetWidth() != w ||
mjpeg_decoder.GetHeight() != h)) {
if (ret &&
(mjpeg_decoder.GetWidth() != w || mjpeg_decoder.GetHeight() != h)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
I420Buffers bufs = { y, y_stride, u, u_stride, v, v_stride, dw, dh };
I420Buffers bufs = {y, y_stride, u, u_stride, v, v_stride, dw, dh};
// YUV420
if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
@ -171,7 +141,7 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegCopyI420, &bufs, dw, dh);
// YUV422
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@ -182,7 +152,7 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToI420, &bufs, dw, dh);
// YUV444
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@ -193,18 +163,7 @@ int MJPGToI420(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToI420, &bufs, dw, dh);
// YUV411
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 4 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI411ToI420, &bufs, dw, dh);
// YUV400
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
@ -213,7 +172,7 @@ int MJPGToI420(const uint8* sample,
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToI420, &bufs, dw, dh);
} else {
// TODO(fbarchard): Implement conversion for any other colorspace/sample
// factors that occur in practice. 411 is supported by libjpeg
// factors that occur in practice.
// ERROR: Unable to convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;
@ -231,15 +190,12 @@ struct ARGBBuffers {
};
static void JpegI420ToARGB(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
const uint8* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I420ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I420ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
@ -249,11 +205,8 @@ static void JpegI422ToARGB(void* opaque,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I422ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I422ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
@ -263,25 +216,8 @@ static void JpegI444ToARGB(void* opaque,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I444ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
static void JpegI411ToARGB(void* opaque,
const uint8* const* data,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I411ToARGB(data[0], strides[0],
data[1], strides[1],
data[2], strides[2],
dest->argb, dest->argb_stride,
dest->w, rows);
I444ToARGB(data[0], strides[0], data[1], strides[1], data[2], strides[2],
dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
@ -291,9 +227,7 @@ static void JpegI400ToARGB(void* opaque,
const int* strides,
int rows) {
ARGBBuffers* dest = (ARGBBuffers*)(opaque);
I400ToARGB(data[0], strides[0],
dest->argb, dest->argb_stride,
dest->w, rows);
I400ToARGB(data[0], strides[0], dest->argb, dest->argb_stride, dest->w, rows);
dest->argb += rows * dest->argb_stride;
dest->h -= rows;
}
@ -303,9 +237,12 @@ static void JpegI400ToARGB(void* opaque,
LIBYUV_API
int MJPGToARGB(const uint8* sample,
size_t sample_size,
uint8* argb, int argb_stride,
int w, int h,
int dw, int dh) {
uint8* argb,
int argb_stride,
int w,
int h,
int dw,
int dh) {
if (sample_size == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown
return -1;
@ -314,17 +251,16 @@ int MJPGToARGB(const uint8* sample,
// TODO(fbarchard): Port MJpeg to C.
MJpegDecoder mjpeg_decoder;
LIBYUV_BOOL ret = mjpeg_decoder.LoadFrame(sample, sample_size);
if (ret && (mjpeg_decoder.GetWidth() != w ||
mjpeg_decoder.GetHeight() != h)) {
if (ret &&
(mjpeg_decoder.GetWidth() != w || mjpeg_decoder.GetHeight() != h)) {
// ERROR: MJPEG frame has unexpected dimensions
mjpeg_decoder.UnloadFrame();
return 1; // runtime failure
}
if (ret) {
ARGBBuffers bufs = { argb, argb_stride, dw, dh };
ARGBBuffers bufs = {argb, argb_stride, dw, dh};
// YUV420
if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
if (mjpeg_decoder.GetColorSpace() == MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 2 &&
mjpeg_decoder.GetHorizSampFactor(0) == 2 &&
@ -333,7 +269,7 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI420ToARGB, &bufs, dw, dh);
// YUV422
// YUV422
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@ -344,7 +280,7 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI422ToARGB, &bufs, dw, dh);
// YUV444
// YUV444
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
@ -355,18 +291,7 @@ int MJPGToARGB(const uint8* sample,
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI444ToARGB, &bufs, dw, dh);
// YUV411
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceYCbCr &&
mjpeg_decoder.GetNumComponents() == 3 &&
mjpeg_decoder.GetVertSampFactor(0) == 1 &&
mjpeg_decoder.GetHorizSampFactor(0) == 4 &&
mjpeg_decoder.GetVertSampFactor(1) == 1 &&
mjpeg_decoder.GetHorizSampFactor(1) == 1 &&
mjpeg_decoder.GetVertSampFactor(2) == 1 &&
mjpeg_decoder.GetHorizSampFactor(2) == 1) {
ret = mjpeg_decoder.DecodeToCallback(&JpegI411ToARGB, &bufs, dw, dh);
// YUV400
// YUV400
} else if (mjpeg_decoder.GetColorSpace() ==
MJpegDecoder::kColorSpaceGrayscale &&
mjpeg_decoder.GetNumComponents() == 1 &&
@ -375,7 +300,7 @@ int MJPGToARGB(const uint8* sample,
ret = mjpeg_decoder.DecodeToCallback(&JpegI400ToARGB, &bufs, dw, dh);
} else {
// TODO(fbarchard): Implement conversion for any other colorspace/sample
// factors that occur in practice. 411 is supported by libjpeg
// factors that occur in practice.
// ERROR: Unable to convert MJPEG frame because format is not supported
mjpeg_decoder.UnloadFrame();
return 1;

185
third_party/yuv/source/convert_to_argb.cc vendored
View File

@ -29,11 +29,16 @@ extern "C" {
// sample_size is measured in bytes and is the size of the frame.
// With MJPEG it is the compressed size of the frame.
LIBYUV_API
int ConvertToARGB(const uint8* sample, size_t sample_size,
uint8* crop_argb, int argb_stride,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
int ConvertToARGB(const uint8* sample,
size_t sample_size,
uint8* crop_argb,
int argb_stride,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 fourcc) {
uint32 format = CanonicalFourCC(fourcc);
@ -45,20 +50,19 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
int r = 0;
// One pass rotation is available for some formats. For the rest, convert
// to I420 (with optional vertical flipping) into a temporary I420 buffer,
// and then rotate the I420 to the final destination buffer.
// to ARGB (with optional vertical flipping) into a temporary ARGB buffer,
// and then rotate the ARGB to the final destination buffer.
// For in-place conversion, if destination crop_argb is same as source sample,
// also enable temporary buffer.
LIBYUV_BOOL need_buf = (rotation && format != FOURCC_ARGB) ||
crop_argb == sample;
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_ARGB) || crop_argb == sample;
uint8* dest_argb = crop_argb;
int dest_argb_stride = argb_stride;
uint8* rotate_buffer = NULL;
int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
if (crop_argb == NULL || sample == NULL ||
src_width <= 0 || crop_width <= 0 ||
src_height == 0 || crop_height == 0) {
if (crop_argb == NULL || sample == NULL || src_width <= 0 ||
crop_width <= 0 || src_height == 0 || crop_height == 0) {
return -1;
}
if (src_height < 0) {
@ -67,7 +71,7 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
if (need_buf) {
int argb_size = crop_width * 4 * abs_crop_height;
rotate_buffer = (uint8*)malloc(argb_size);
rotate_buffer = (uint8*)malloc(argb_size); /* NOLINT */
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
@ -79,100 +83,85 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
// Single plane formats
case FOURCC_YUY2:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToARGB(src, aligned_src_width * 2,
crop_argb, argb_stride,
r = YUY2ToARGB(src, aligned_src_width * 2, crop_argb, argb_stride,
crop_width, inv_crop_height);
break;
case FOURCC_UYVY:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToARGB(src, aligned_src_width * 2,
crop_argb, argb_stride,
r = UYVYToARGB(src, aligned_src_width * 2, crop_argb, argb_stride,
crop_width, inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToARGB(src, src_width * 3,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RGB24ToARGB(src, src_width * 3, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToARGB(src, src_width * 3,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RAWToARGB(src, src_width * 3, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_ARGB:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
if (!need_buf && !rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToARGB(src, src_width * 4, crop_argb, argb_stride, crop_width,
inv_crop_height);
}
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = BGRAToARGB(src, src_width * 4, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = ABGRToARGB(src, src_width * 4, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToARGB(src, src_width * 4,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RGBAToARGB(src, src_width * 4, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToARGB(src, src_width * 2,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = RGB565ToARGB(src, src_width * 2, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToARGB(src, src_width * 2,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = ARGB1555ToARGB(src, src_width * 2, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToARGB(src, src_width * 2,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = ARGB4444ToARGB(src, src_width * 2, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToARGB(src, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I400ToARGB(src, src_width, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
r = NV12ToARGB(src, src_width,
src_uv, aligned_src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = NV12ToARGB(src, src_width, src_uv, aligned_src_width, crop_argb,
argb_stride, crop_width, inv_crop_height);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + aligned_src_width * (src_height + crop_y / 2) + crop_x;
// Call NV12 but with u and v parameters swapped.
r = NV21ToARGB(src, src_width,
src_uv, aligned_src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = NV21ToARGB(src, src_width, src_uv, aligned_src_width, crop_argb,
argb_stride, crop_width, inv_crop_height);
break;
case FOURCC_M420:
src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
r = M420ToARGB(src, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = M420ToARGB(src, src_width, crop_argb, argb_stride, crop_width,
inv_crop_height);
break;
// Triplanar formats
case FOURCC_I420:
@ -184,20 +173,17 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
}
r = I420ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
crop_argb, argb_stride, crop_width, inv_crop_height);
break;
}
@ -208,14 +194,11 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
int halfwidth = (src_width + 1) / 2;
int halfheight = (abs_src_height + 1) / 2;
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = J420ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
r = J420ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
crop_argb, argb_stride, crop_width, inv_crop_height);
break;
}
@ -226,21 +209,18 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
const uint8* src_v;
int halfwidth = (src_width + 1) / 2;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
}
r = I422ToARGB(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I422ToARGB(src_y, src_width, src_u, halfwidth, src_v, halfwidth,
crop_argb, argb_stride, crop_width, inv_crop_height);
break;
}
case FOURCC_I444:
@ -255,32 +235,14 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToARGB(src_y, src_width,
src_u, src_width,
src_v, src_width,
crop_argb, argb_stride,
crop_width, inv_crop_height);
break;
}
case FOURCC_I411: {
int quarterwidth = (src_width + 3) / 4;
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u = sample + src_width * abs_src_height +
quarterwidth * crop_y + crop_x / 4;
const uint8* src_v = sample + src_width * abs_src_height +
quarterwidth * (abs_src_height + crop_y) + crop_x / 4;
r = I411ToARGB(src_y, src_width,
src_u, quarterwidth,
src_v, quarterwidth,
crop_argb, argb_stride,
crop_width, inv_crop_height);
r = I444ToARGB(src_y, src_width, src_u, src_width, src_v, src_width,
crop_argb, argb_stride, crop_width, inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToARGB(sample, sample_size,
crop_argb, argb_stride,
src_width, abs_src_height, crop_width, inv_crop_height);
r = MJPGToARGB(sample, sample_size, crop_argb, argb_stride, src_width,
abs_src_height, crop_width, inv_crop_height);
break;
#endif
default:
@ -289,11 +251,14 @@ int ConvertToARGB(const uint8* sample, size_t sample_size,
if (need_buf) {
if (!r) {
r = ARGBRotate(crop_argb, argb_stride,
dest_argb, dest_argb_stride,
r = ARGBRotate(crop_argb, argb_stride, dest_argb, dest_argb_stride,
crop_width, abs_crop_height, rotation);
}
free(rotate_buffer);
} else if (rotation) {
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBRotate(src, src_width * 4, crop_argb, argb_stride, crop_width,
inv_crop_height, rotation);
}
return r;

208
third_party/yuv/source/convert_to_i420.cc vendored
View File

@ -27,12 +27,18 @@ extern "C" {
LIBYUV_API
int ConvertToI420(const uint8* sample,
size_t sample_size,
uint8* y, int y_stride,
uint8* u, int u_stride,
uint8* v, int v_stride,
int crop_x, int crop_y,
int src_width, int src_height,
int crop_width, int crop_height,
uint8* y,
int y_stride,
uint8* u,
int u_stride,
uint8* v,
int v_stride,
int crop_x,
int crop_y,
int src_width,
int src_height,
int crop_width,
int crop_height,
enum RotationMode rotation,
uint32 fourcc) {
uint32 format = CanonicalFourCC(fourcc);
@ -43,9 +49,10 @@ int ConvertToI420(const uint8* sample,
// TODO(nisse): Why allow crop_height < 0?
const int abs_crop_height = (crop_height < 0) ? -crop_height : crop_height;
int r = 0;
LIBYUV_BOOL need_buf = (rotation && format != FOURCC_I420 &&
format != FOURCC_NV12 && format != FOURCC_NV21 &&
format != FOURCC_YV12) || y == sample;
LIBYUV_BOOL need_buf =
(rotation && format != FOURCC_I420 && format != FOURCC_NV12 &&
format != FOURCC_NV21 && format != FOURCC_YV12) ||
y == sample;
uint8* tmp_y = y;
uint8* tmp_u = u;
uint8* tmp_v = v;
@ -56,8 +63,7 @@ int ConvertToI420(const uint8* sample,
const int inv_crop_height =
(src_height < 0) ? -abs_crop_height : abs_crop_height;
if (!y || !u || !v || !sample ||
src_width <= 0 || crop_width <= 0 ||
if (!y || !u || !v || !sample || src_width <= 0 || crop_width <= 0 ||
src_height == 0 || crop_height == 0) {
return -1;
}
@ -70,7 +76,7 @@ int ConvertToI420(const uint8* sample,
if (need_buf) {
int y_size = crop_width * abs_crop_height;
int uv_size = ((crop_width + 1) / 2) * ((abs_crop_height + 1) / 2);
rotate_buffer = (uint8*)malloc(y_size + uv_size * 2);
rotate_buffer = (uint8*)malloc(y_size + uv_size * 2); /* NOLINT */
if (!rotate_buffer) {
return 1; // Out of memory runtime error.
}
@ -85,130 +91,85 @@ int ConvertToI420(const uint8* sample,
// Single plane formats
case FOURCC_YUY2:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = YUY2ToI420(src, aligned_src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = YUY2ToI420(src, aligned_src_width * 2, y, y_stride, u, u_stride, v,
v_stride, crop_width, inv_crop_height);
break;
case FOURCC_UYVY:
src = sample + (aligned_src_width * crop_y + crop_x) * 2;
r = UYVYToI420(src, aligned_src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = UYVYToI420(src, aligned_src_width * 2, y, y_stride, u, u_stride, v,
v_stride, crop_width, inv_crop_height);
break;
case FOURCC_RGBP:
src = sample + (src_width * crop_y + crop_x) * 2;
r = RGB565ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = RGB565ToI420(src, src_width * 2, y, y_stride, u, u_stride, v,
v_stride, crop_width, inv_crop_height);
break;
case FOURCC_RGBO:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB1555ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ARGB1555ToI420(src, src_width * 2, y, y_stride, u, u_stride, v,
v_stride, crop_width, inv_crop_height);
break;
case FOURCC_R444:
src = sample + (src_width * crop_y + crop_x) * 2;
r = ARGB4444ToI420(src, src_width * 2,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = ARGB4444ToI420(src, src_width * 2, y, y_stride, u, u_stride, v,
v_stride, crop_width, inv_crop_height);
break;
case FOURCC_24BG:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RGB24ToI420(src, src_width * 3,
y, y_stride,
u, u_stride,
v, v_stride,
r = RGB24ToI420(src, src_width * 3, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_RAW:
src = sample + (src_width * crop_y + crop_x) * 3;
r = RAWToI420(src, src_width * 3,
y, y_stride,
u, u_stride,
v, v_stride,
r = RAWToI420(src, src_width * 3, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_ARGB:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ARGBToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
r = ARGBToI420(src, src_width * 4, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_BGRA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = BGRAToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
r = BGRAToI420(src, src_width * 4, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_ABGR:
src = sample + (src_width * crop_y + crop_x) * 4;
r = ABGRToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
r = ABGRToI420(src, src_width * 4, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_RGBA:
src = sample + (src_width * crop_y + crop_x) * 4;
r = RGBAToI420(src, src_width * 4,
y, y_stride,
u, u_stride,
v, v_stride,
r = RGBAToI420(src, src_width * 4, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
case FOURCC_I400:
src = sample + src_width * crop_y + crop_x;
r = I400ToI420(src, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
r = I400ToI420(src, src_width, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
// Biplanar formats
case FOURCC_NV12:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
r = NV12ToI420Rotate(src, src_width,
src_uv, aligned_src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height, rotation);
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, y,
y_stride, u, u_stride, v, v_stride, crop_width,
inv_crop_height, rotation);
break;
case FOURCC_NV21:
src = sample + (src_width * crop_y + crop_x);
src_uv = sample + (src_width * src_height) +
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
((crop_y / 2) * aligned_src_width) + ((crop_x / 2) * 2);
// Call NV12 but with u and v parameters swapped.
r = NV12ToI420Rotate(src, src_width,
src_uv, aligned_src_width,
y, y_stride,
v, v_stride,
u, u_stride,
crop_width, inv_crop_height, rotation);
r = NV12ToI420Rotate(src, src_width, src_uv, aligned_src_width, y,
y_stride, v, v_stride, u, u_stride, crop_width,
inv_crop_height, rotation);
break;
case FOURCC_M420:
src = sample + (src_width * crop_y) * 12 / 8 + crop_x;
r = M420ToI420(src, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
r = M420ToI420(src, src_width, y, y_stride, u, u_stride, v, v_stride,
crop_width, inv_crop_height);
break;
// Triplanar formats
@ -221,22 +182,18 @@ int ConvertToI420(const uint8* sample,
int halfheight = (abs_src_height + 1) / 2;
if (format == FOURCC_YV12) {
src_v = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
(halfwidth * crop_y + crop_x) / 2;
(halfwidth * crop_y + crop_x) / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
halfwidth * (halfheight + crop_y / 2) + crop_x / 2;
}
r = I420Rotate(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height, rotation);
r = I420Rotate(src_y, src_width, src_u, halfwidth, src_v, halfwidth, y,
y_stride, u, u_stride, v, v_stride, crop_width,
inv_crop_height, rotation);
break;
}
case FOURCC_I422:
@ -246,23 +203,19 @@ int ConvertToI420(const uint8* sample,
const uint8* src_v;
int halfwidth = (src_width + 1) / 2;
if (format == FOURCC_YV16) {
src_v = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_v = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_u = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
} else {
src_u = sample + src_width * abs_src_height +
halfwidth * crop_y + crop_x / 2;
src_u = sample + src_width * abs_src_height + halfwidth * crop_y +
crop_x / 2;
src_v = sample + src_width * abs_src_height +
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
halfwidth * (abs_src_height + crop_y) + crop_x / 2;
}
r = I422ToI420(src_y, src_width,
src_u, halfwidth,
src_v, halfwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = I422ToI420(src_y, src_width, src_u, halfwidth, src_v, halfwidth, y,
y_stride, u, u_stride, v, v_stride, crop_width,
inv_crop_height);
break;
}
case FOURCC_I444:
@ -277,37 +230,14 @@ int ConvertToI420(const uint8* sample,
src_u = sample + src_width * (abs_src_height + crop_y) + crop_x;
src_v = sample + src_width * (abs_src_height * 2 + crop_y) + crop_x;
}
r = I444ToI420(src_y, src_width,
src_u, src_width,
src_v, src_width,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
break;
}
case FOURCC_I411: {
int quarterwidth = (src_width + 3) / 4;
const uint8* src_y = sample + src_width * crop_y + crop_x;
const uint8* src_u = sample + src_width * abs_src_height +
quarterwidth * crop_y + crop_x / 4;
const uint8* src_v = sample + src_width * abs_src_height +
quarterwidth * (abs_src_height + crop_y) + crop_x / 4;
r = I411ToI420(src_y, src_width,
src_u, quarterwidth,
src_v, quarterwidth,
y, y_stride,
u, u_stride,
v, v_stride,
crop_width, inv_crop_height);
r = I444ToI420(src_y, src_width, src_u, src_width, src_v, src_width, y,
y_stride, u, u_stride, v, v_stride, crop_width,
inv_crop_height);
break;
}
#ifdef HAVE_JPEG
case FOURCC_MJPG:
r = MJPGToI420(sample, sample_size,
y, y_stride,
u, u_stride,
v, v_stride,
r = MJPGToI420(sample, sample_size, y, y_stride, u, u_stride, v, v_stride,
src_width, abs_src_height, crop_width, inv_crop_height);
break;
#endif
@ -317,13 +247,9 @@ int ConvertToI420(const uint8* sample,
if (need_buf) {
if (!r) {
r = I420Rotate(y, y_stride,
u, u_stride,
v, v_stride,
tmp_y, tmp_y_stride,
tmp_u, tmp_u_stride,
tmp_v, tmp_v_stride,
crop_width, abs_crop_height, rotation);
r = I420Rotate(y, y_stride, u, u_stride, v, v_stride, tmp_y, tmp_y_stride,
tmp_u, tmp_u_stride, tmp_v, tmp_v_stride, crop_width,
abs_crop_height, rotation);
}
free(rotate_buffer);
}

148
third_party/yuv/source/cpu_id.cc vendored
View File

@ -13,7 +13,7 @@
#if defined(_MSC_VER)
#include <intrin.h> // For __cpuidex()
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
!defined(__native_client__) && (defined(_M_IX86) || defined(_M_X64)) && \
defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
#include <immintrin.h> // For _xgetbv()
@ -43,16 +43,20 @@ extern "C" {
#define SAFEBUFFERS
#endif
// cpu_info_ variable for SIMD instruction sets detected.
LIBYUV_API int cpu_info_ = 0;
// TODO(fbarchard): Consider using int for cpuid so casting is not needed.
// Low level cpuid for X86.
#if (defined(_M_IX86) || defined(_M_X64) || \
defined(__i386__) || defined(__x86_64__)) && \
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER)
LIBYUV_API
void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
void CpuId(int info_eax, int info_ecx, int* cpu_info) {
#if defined(_MSC_VER)
// Visual C version uses intrinsic or inline x86 assembly.
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
__cpuidex((int*)(cpu_info), info_eax, info_ecx);
__cpuidex(cpu_info, info_eax, info_ecx);
#elif defined(_M_IX86)
__asm {
mov eax, info_eax
@ -66,26 +70,26 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
}
#else // Visual C but not x86
if (info_ecx == 0) {
__cpuid((int*)(cpu_info), info_eax);
__cpuid(cpu_info, info_eax);
} else {
cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0;
cpu_info[3] = cpu_info[2] = cpu_info[1] = cpu_info[0] = 0u;
}
#endif
// GCC version uses inline x86 assembly.
#else // defined(_MSC_VER)
uint32 info_ebx, info_edx;
asm volatile (
#if defined( __i386__) && defined(__PIC__)
// Preserve ebx for fpic 32 bit.
"mov %%ebx, %%edi \n"
"cpuid \n"
"xchg %%edi, %%ebx \n"
: "=D" (info_ebx),
int info_ebx, info_edx;
asm volatile(
#if defined(__i386__) && defined(__PIC__)
// Preserve ebx for fpic 32 bit.
"mov %%ebx, %%edi \n"
"cpuid \n"
"xchg %%edi, %%ebx \n"
: "=D"(info_ebx),
#else
"cpuid \n"
: "=b" (info_ebx),
"cpuid \n"
: "=b"(info_ebx),
#endif // defined( __i386__) && defined(__PIC__)
"+a" (info_eax), "+c" (info_ecx), "=d" (info_edx));
"+a"(info_eax), "+c"(info_ecx), "=d"(info_edx));
cpu_info[0] = info_eax;
cpu_info[1] = info_ebx;
cpu_info[2] = info_ecx;
@ -94,7 +98,9 @@ void CpuId(uint32 info_eax, uint32 info_ecx, uint32* cpu_info) {
}
#else // (defined(_M_IX86) || defined(_M_X64) ...
LIBYUV_API
void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
void CpuId(int eax, int ecx, int* cpu_info) {
(void)eax;
(void)ecx;
cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
}
#endif
@ -111,20 +117,22 @@ void CpuId(uint32 eax, uint32 ecx, uint32* cpu_info) {
#if defined(_M_IX86) && (_MSC_VER < 1900)
#pragma optimize("g", off)
#endif
#if (defined(_M_IX86) || defined(_M_X64) || \
defined(__i386__) || defined(__x86_64__)) && \
#if (defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || \
defined(__x86_64__)) && \
!defined(__pnacl__) && !defined(__CLR_VER) && !defined(__native_client__)
#define HAS_XGETBV
// X86 CPUs have xgetbv to detect OS saves high parts of ymm registers.
int GetXCR0() {
uint32 xcr0 = 0u;
int xcr0 = 0;
#if defined(_MSC_FULL_VER) && (_MSC_FULL_VER >= 160040219)
xcr0 = (uint32)(_xgetbv(0)); // VS2010 SP1 required.
xcr0 = _xgetbv(0); // VS2010 SP1 required.
#elif defined(__i386__) || defined(__x86_64__)
asm(".byte 0x0f, 0x01, 0xd0" : "=a" (xcr0) : "c" (0) : "%edx");
asm(".byte 0x0f, 0x01, 0xd0" : "=a"(xcr0) : "c"(0) : "%edx");
#endif // defined(__i386__) || defined(__x86_64__)
return xcr0;
}
#else
// xgetbv unavailable to query for OSSave support. Return 0.
#define GetXCR0() 0
#endif // defined(_M_IX86) || defined(_M_X64) ..
// Return optimization to previous setting.
#if defined(_M_IX86) && (_MSC_VER < 1900)
@ -133,8 +141,7 @@ int GetXCR0() {
// based on libvpx arm_cpudetect.c
// For Arm, but public to allow testing on any CPU
LIBYUV_API SAFEBUFFERS
int ArmCpuCaps(const char* cpuinfo_name) {
LIBYUV_API SAFEBUFFERS int ArmCpuCaps(const char* cpuinfo_name) {
char cpuinfo_line[512];
FILE* f = fopen(cpuinfo_name, "r");
if (!f) {
@ -151,7 +158,7 @@ int ArmCpuCaps(const char* cpuinfo_name) {
}
// aarch64 uses asimd for Neon.
p = strstr(cpuinfo_line, " asimd");
if (p && (p[6] == ' ' || p[6] == '\n')) {
if (p) {
fclose(f);
return kCpuHasNEON;
}
@ -161,13 +168,38 @@ int ArmCpuCaps(const char* cpuinfo_name) {
return 0;
}
// CPU detect function for SIMD instruction sets.
LIBYUV_API
int cpu_info_ = 0; // cpu_info is not initialized yet.
// TODO(fbarchard): Consider read_msa_ir().
// TODO(fbarchard): Add unittest.
LIBYUV_API SAFEBUFFERS int MipsCpuCaps(const char* cpuinfo_name,
const char ase[]) {
char cpuinfo_line[512];
FILE* f = fopen(cpuinfo_name, "r");
if (!f) {
// ase enabled if /proc/cpuinfo is unavailable.
if (strcmp(ase, " msa") == 0) {
return kCpuHasMSA;
}
return kCpuHasDSPR2;
}
while (fgets(cpuinfo_line, sizeof(cpuinfo_line) - 1, f)) {
if (memcmp(cpuinfo_line, "ASEs implemented", 16) == 0) {
char* p = strstr(cpuinfo_line, ase);
if (p) {
fclose(f);
if (strcmp(ase, " msa") == 0) {
return kCpuHasMSA;
}
return kCpuHasDSPR2;
}
}
}
fclose(f);
return 0;
}
// Test environment variable for disabling CPU features. Any non-zero value
// to disable. Zero ignored to make it easy to set the variable on/off.
#if !defined(__native_client__) && !defined(_M_ARM) && !defined(_MSC_VER)
#if !defined(__native_client__) && !defined(_M_ARM)
static LIBYUV_BOOL TestEnv(const char* name) {
const char* var = getenv(name);
@ -184,39 +216,35 @@ static LIBYUV_BOOL TestEnv(const char*) {
}
#endif
LIBYUV_API SAFEBUFFERS
int InitCpuFlags(void) {
// TODO(fbarchard): swap kCpuInit logic so 0 means uninitialized.
static SAFEBUFFERS int GetCpuFlags(void) {
int cpu_info = 0;
#if !defined(__pnacl__) && !defined(__CLR_VER) && defined(CPU_X86)
uint32 cpu_info0[4] = { 0, 0, 0, 0 };
uint32 cpu_info1[4] = { 0, 0, 0, 0 };
uint32 cpu_info7[4] = { 0, 0, 0, 0 };
int cpu_info0[4] = {0, 0, 0, 0};
int cpu_info1[4] = {0, 0, 0, 0};
int cpu_info7[4] = {0, 0, 0, 0};
CpuId(0, 0, cpu_info0);
CpuId(1, 0, cpu_info1);
if (cpu_info0[0] >= 7) {
CpuId(7, 0, cpu_info7);
}
cpu_info = ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
cpu_info = kCpuHasX86 | ((cpu_info1[3] & 0x04000000) ? kCpuHasSSE2 : 0) |
((cpu_info1[2] & 0x00000200) ? kCpuHasSSSE3 : 0) |
((cpu_info1[2] & 0x00080000) ? kCpuHasSSE41 : 0) |
((cpu_info1[2] & 0x00100000) ? kCpuHasSSE42 : 0) |
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0) |
((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
kCpuHasX86;
((cpu_info7[1] & 0x00000200) ? kCpuHasERMS : 0);
#ifdef HAS_XGETBV
// AVX requires CPU has AVX, XSAVE and OSXSave for xgetbv
// AVX requires OS saves YMM registers.
if (((cpu_info1[2] & 0x1c000000) == 0x1c000000) && // AVX and OSXSave
((GetXCR0() & 6) == 6)) { // Test OS saves YMM registers
cpu_info |= ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) | kCpuHasAVX;
cpu_info |= kCpuHasAVX | ((cpu_info7[1] & 0x00000020) ? kCpuHasAVX2 : 0) |
((cpu_info1[2] & 0x00001000) ? kCpuHasFMA3 : 0) |
((cpu_info1[2] & 0x20000000) ? kCpuHasF16C : 0);
// Detect AVX512bw
if ((GetXCR0() & 0xe0) == 0xe0) {
cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX3 : 0;
}
}
#endif
// Environment variable overrides for testing.
if (TestEnv("LIBYUV_DISABLE_X86")) {
@ -249,15 +277,25 @@ int InitCpuFlags(void) {
if (TestEnv("LIBYUV_DISABLE_AVX3")) {
cpu_info &= ~kCpuHasAVX3;
}
if (TestEnv("LIBYUV_DISABLE_F16C")) {
cpu_info &= ~kCpuHasF16C;
}
#endif
#if defined(__mips__) && defined(__linux__)
#if defined(__mips_dspr2)
cpu_info |= kCpuHasDSPR2;
#endif
#if defined(__mips_msa)
cpu_info = MipsCpuCaps("/proc/cpuinfo", " msa");
#endif
cpu_info |= kCpuHasMIPS;
if (getenv("LIBYUV_DISABLE_DSPR2")) {
cpu_info &= ~kCpuHasDSPR2;
}
if (getenv("LIBYUV_DISABLE_MSA")) {
cpu_info &= ~kCpuHasMSA;
}
#endif
#if defined(__arm__) || defined(__aarch64__)
// gcc -mfpu=neon defines __ARM_NEON__
@ -283,15 +321,25 @@ int InitCpuFlags(void) {
if (TestEnv("LIBYUV_DISABLE_ASM")) {
cpu_info = 0;
}
cpu_info |= kCpuInitialized;
cpu_info_ = cpu_info;
cpu_info |= kCpuInitialized;
return cpu_info;
}
// Note that use of this function is not thread safe.
LIBYUV_API
void MaskCpuFlags(int enable_flags) {
cpu_info_ = InitCpuFlags() & enable_flags;
int MaskCpuFlags(int enable_flags) {
int cpu_info = GetCpuFlags() & enable_flags;
#ifdef __ATOMIC_RELAXED
__atomic_store_n(&cpu_info_, cpu_info, __ATOMIC_RELAXED);
#else
cpu_info_ = cpu_info;
#endif
return cpu_info;
}
LIBYUV_API
int InitCpuFlags(void) {
return MaskCpuFlags(-1);
}
#ifdef __cplusplus

125
third_party/yuv/source/mjpeg_decoder.cc vendored
View File

@ -13,6 +13,10 @@
#ifdef HAVE_JPEG
#include <assert.h>
#ifdef __cplusplus
#include <new>
#endif
#if !defined(__pnacl__) && !defined(__CLR_VER) && \
!defined(COVERAGE_ENABLED) && !defined(TARGET_IPHONE_SIMULATOR)
// Must be included before jpeglib.
@ -21,7 +25,7 @@
#if defined(_MSC_VER)
// disable warning 4324: structure was padded due to __declspec(align())
#pragma warning(disable:4324)
#pragma warning(disable : 4324)
#endif
#endif
@ -62,6 +66,7 @@ void init_source(jpeg_decompress_struct* cinfo);
void skip_input_data(jpeg_decompress_struct* cinfo, long num_bytes); // NOLINT
void term_source(jpeg_decompress_struct* cinfo);
void ErrorHandler(jpeg_common_struct* cinfo);
void OutputHandler(jpeg_common_struct* cinfo);
MJpegDecoder::MJpegDecoder()
: has_scanline_padding_(LIBYUV_FALSE),
@ -77,6 +82,7 @@ MJpegDecoder::MJpegDecoder()
decompress_struct_->err = jpeg_std_error(&error_mgr_->base);
// Override standard exit()-based error handler.
error_mgr_->base.error_exit = &ErrorHandler;
error_mgr_->base.output_message = &OutputHandler;
#endif
decompress_struct_->client_data = NULL;
source_mgr_->init_source = &init_source;
@ -127,7 +133,7 @@ LIBYUV_BOOL MJpegDecoder::LoadFrame(const uint8* src, size_t src_len) {
if (scanlines_[i]) {
delete scanlines_[i];
}
scanlines_[i] = new uint8* [scanlines_size];
scanlines_[i] = new uint8*[scanlines_size];
scanlines_sizes_[i] = scanlines_size;
}
@ -193,13 +199,11 @@ int MJpegDecoder::GetVertSampFactor(int component) {
}
int MJpegDecoder::GetHorizSubSampFactor(int component) {
return decompress_struct_->max_h_samp_factor /
GetHorizSampFactor(component);
return decompress_struct_->max_h_samp_factor / GetHorizSampFactor(component);
}
int MJpegDecoder::GetVertSubSampFactor(int component) {
return decompress_struct_->max_v_samp_factor /
GetVertSampFactor(component);
return decompress_struct_->max_v_samp_factor / GetVertSampFactor(component);
}
int MJpegDecoder::GetImageScanlinesPerImcuRow() {
@ -243,10 +247,10 @@ LIBYUV_BOOL MJpegDecoder::UnloadFrame() {
}
// TODO(fbarchard): Allow rectangle to be specified: x, y, width, height.
LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
uint8** planes, int dst_width, int dst_height) {
if (dst_width != GetWidth() ||
dst_height > GetHeight()) {
LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(uint8** planes,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
@ -287,14 +291,13 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
for (int i = 0; i < num_outbufs_; ++i) {
// TODO(fbarchard): Compute skip to avoid this
assert(skip % GetVertSubSampFactor(i) == 0);
int rows_to_skip =
DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int scanlines_to_copy = GetComponentScanlinesPerImcuRow(i) -
rows_to_skip;
int rows_to_skip = DivideAndRoundDown(skip, GetVertSubSampFactor(i));
int scanlines_to_copy =
GetComponentScanlinesPerImcuRow(i) - rows_to_skip;
int data_to_skip = rows_to_skip * GetComponentStride(i);
CopyPlane(databuf_[i] + data_to_skip, GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i] + data_to_skip, GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i),
scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
lines_left -= (GetImageScanlinesPerImcuRow() - skip);
@ -303,16 +306,15 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
// Read full MCUs but cropped horizontally
for (; lines_left > GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
}
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy = GetComponentScanlinesPerImcuRow(i);
CopyPlane(databuf_[i], GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
@ -326,19 +328,19 @@ LIBYUV_BOOL MJpegDecoder::DecodeToBuffers(
for (int i = 0; i < num_outbufs_; ++i) {
int scanlines_to_copy =
DivideAndRoundUp(lines_left, GetVertSubSampFactor(i));
CopyPlane(databuf_[i], GetComponentStride(i),
planes[i], GetComponentWidth(i),
GetComponentWidth(i), scanlines_to_copy);
CopyPlane(databuf_[i], GetComponentStride(i), planes[i],
GetComponentWidth(i), GetComponentWidth(i), scanlines_to_copy);
planes[i] += scanlines_to_copy * GetComponentWidth(i);
}
}
return FinishDecode();
}
LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn, void* opaque,
int dst_width, int dst_height) {
if (dst_width != GetWidth() ||
dst_height > GetHeight()) {
LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn,
void* opaque,
int dst_width,
int dst_height) {
if (dst_width != GetWidth() || dst_height > GetHeight()) {
// ERROR: Bad dimensions
return LIBYUV_FALSE;
}
@ -393,7 +395,7 @@ LIBYUV_BOOL MJpegDecoder::DecodeToCallback(CallbackFunction fn, void* opaque,
}
// Read full MCUs until we get to the crop point.
for (; lines_left >= GetImageScanlinesPerImcuRow();
lines_left -= GetImageScanlinesPerImcuRow()) {
lines_left -= GetImageScanlinesPerImcuRow()) {
if (!DecodeImcuRow()) {
FinishDecode();
return LIBYUV_FALSE;
@ -433,22 +435,22 @@ void skip_input_data(j_decompress_ptr cinfo, long num_bytes) { // NOLINT
}
void term_source(j_decompress_ptr cinfo) {
// Nothing to do.
(void)cinfo; // Nothing to do.
}
#ifdef HAVE_SETJMP
void ErrorHandler(j_common_ptr cinfo) {
// This is called when a jpeglib command experiences an error. Unfortunately
// jpeglib's error handling model is not very flexible, because it expects the
// error handler to not return--i.e., it wants the program to terminate. To
// recover from errors we use setjmp() as shown in their example. setjmp() is
// C's implementation for the "call with current continuation" functionality
// seen in some functional programming languages.
// A formatted message can be output, but is unsafe for release.
// This is called when a jpeglib command experiences an error. Unfortunately
// jpeglib's error handling model is not very flexible, because it expects the
// error handler to not return--i.e., it wants the program to terminate. To
// recover from errors we use setjmp() as shown in their example. setjmp() is
// C's implementation for the "call with current continuation" functionality
// seen in some functional programming languages.
// A formatted message can be output, but is unsafe for release.
#ifdef DEBUG
char buf[JMSG_LENGTH_MAX];
(*cinfo->err->format_message)(cinfo, buf);
// ERROR: Error in jpeglib: buf
// ERROR: Error in jpeglib: buf
#endif
SetJmpErrorMgr* mgr = reinterpret_cast<SetJmpErrorMgr*>(cinfo->err);
@ -456,7 +458,13 @@ void ErrorHandler(j_common_ptr cinfo) {
// and causes it to return (for a second time) with value 1.
longjmp(mgr->setjmp_buffer, 1);
}
#endif
// Suppress fprintf warnings.
void OutputHandler(j_common_ptr cinfo) {
(void)cinfo;
}
#endif // HAVE_SETJMP
void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
if (num_outbufs != num_outbufs_) {
@ -465,9 +473,9 @@ void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
// it.
DestroyOutputBuffers();
scanlines_ = new uint8** [num_outbufs];
scanlines_ = new uint8**[num_outbufs];
scanlines_sizes_ = new int[num_outbufs];
databuf_ = new uint8* [num_outbufs];
databuf_ = new uint8*[num_outbufs];
databuf_strides_ = new int[num_outbufs];
for (int i = 0; i < num_outbufs; ++i) {
@ -483,13 +491,13 @@ void MJpegDecoder::AllocOutputBuffers(int num_outbufs) {
void MJpegDecoder::DestroyOutputBuffers() {
for (int i = 0; i < num_outbufs_; ++i) {
delete [] scanlines_[i];
delete [] databuf_[i];
delete[] scanlines_[i];
delete[] databuf_[i];
}
delete [] scanlines_;
delete [] databuf_;
delete [] scanlines_sizes_;
delete [] databuf_strides_;
delete[] scanlines_;
delete[] databuf_;
delete[] scanlines_sizes_;
delete[] databuf_strides_;
scanlines_ = NULL;
databuf_ = NULL;
scanlines_sizes_ = NULL;
@ -535,26 +543,26 @@ void MJpegDecoder::SetScanlinePointers(uint8** data) {
inline LIBYUV_BOOL MJpegDecoder::DecodeImcuRow() {
return (unsigned int)(GetImageScanlinesPerImcuRow()) ==
jpeg_read_raw_data(decompress_struct_,
scanlines_,
GetImageScanlinesPerImcuRow());
jpeg_read_raw_data(decompress_struct_, scanlines_,
GetImageScanlinesPerImcuRow());
}
// The helper function which recognizes the jpeg sub-sampling type.
JpegSubsamplingType MJpegDecoder::JpegSubsamplingTypeHelper(
int* subsample_x, int* subsample_y, int number_of_components) {
int* subsample_x,
int* subsample_y,
int number_of_components) {
if (number_of_components == 3) { // Color images.
if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 2 && subsample_y[1] == 2 &&
subsample_x[2] == 2 && subsample_y[2] == 2) {
if (subsample_x[0] == 1 && subsample_y[0] == 1 && subsample_x[1] == 2 &&
subsample_y[1] == 2 && subsample_x[2] == 2 && subsample_y[2] == 2) {
return kJpegYuv420;
} else if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 2 && subsample_y[1] == 1 &&
subsample_x[2] == 2 && subsample_y[2] == 1) {
subsample_x[1] == 2 && subsample_y[1] == 1 &&
subsample_x[2] == 2 && subsample_y[2] == 1) {
return kJpegYuv422;
} else if (subsample_x[0] == 1 && subsample_y[0] == 1 &&
subsample_x[1] == 1 && subsample_y[1] == 1 &&
subsample_x[2] == 1 && subsample_y[2] == 1) {
subsample_x[1] == 1 && subsample_y[1] == 1 &&
subsample_x[2] == 1 && subsample_y[2] == 1) {
return kJpegYuv444;
}
} else if (number_of_components == 1) { // Grey-scale images.
@ -567,4 +575,3 @@ JpegSubsamplingType MJpegDecoder::JpegSubsamplingTypeHelper(
} // namespace libyuv
#endif // HAVE_JPEG

3
third_party/yuv/source/mjpeg_validate.cc vendored
View File

@ -24,7 +24,7 @@ static LIBYUV_BOOL ScanEOI(const uint8* sample, size_t sample_size) {
const uint8* it = sample;
while (it < end) {
// TODO(fbarchard): scan for 0xd9 instead.
it = static_cast<const uint8 *>(memchr(it, 0xff, end - it));
it = static_cast<const uint8*>(memchr(it, 0xff, end - it));
if (it == NULL) {
break;
}
@ -68,4 +68,3 @@ LIBYUV_BOOL ValidateJpeg(const uint8* sample, size_t sample_size) {
} // extern "C"
} // namespace libyuv
#endif

1506
third_party/yuv/source/planar_functions.cc vendored
View File

File diff suppressed because it is too large Load Diff

361
third_party/yuv/source/rotate.cc vendored
View File

@ -10,8 +10,8 @@
#include "libyuv/rotate.h"
#include "libyuv/cpu_id.h"
#include "libyuv/convert.h"
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
@ -22,12 +22,20 @@ extern "C" {
#endif
LIBYUV_API
void TransposePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void TransposePlane(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
int i = height;
void (*TransposeWx8)(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) = TransposeWx8_C;
#if defined(HAS_TRANSPOSEWX16_MSA)
void (*TransposeWx16)(const uint8* src, int src_stride, uint8* dst,
int dst_stride, int width) = TransposeWx16_C;
#else
void (*TransposeWx8)(const uint8* src, int src_stride, uint8* dst,
int dst_stride, int width) = TransposeWx8_C;
#endif
#if defined(HAS_TRANSPOSEWX8_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
TransposeWx8 = TransposeWx8_NEON;
@ -51,22 +59,40 @@ void TransposePlane(const uint8* src, int src_stride,
#endif
#if defined(HAS_TRANSPOSEWX8_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2)) {
if (IS_ALIGNED(width, 4) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
if (IS_ALIGNED(width, 4) && IS_ALIGNED(src, 4) &&
IS_ALIGNED(src_stride, 4)) {
TransposeWx8 = TransposeWx8_Fast_DSPR2;
} else {
TransposeWx8 = TransposeWx8_DSPR2;
}
}
#endif
#if defined(HAS_TRANSPOSEWX16_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
TransposeWx16 = TransposeWx16_Any_MSA;
if (IS_ALIGNED(width, 16)) {
TransposeWx16 = TransposeWx16_MSA;
}
}
#endif
#if defined(HAS_TRANSPOSEWX16_MSA)
// Work across the source in 16x16 tiles
while (i >= 16) {
TransposeWx16(src, src_stride, dst, dst_stride, width);
src += 16 * src_stride; // Go down 16 rows.
dst += 16; // Move over 16 columns.
i -= 16;
}
#else
// Work across the source in 8x8 tiles
while (i >= 8) {
TransposeWx8(src, src_stride, dst, dst_stride, width);
src += 8 * src_stride; // Go down 8 rows.
dst += 8; // Move over 8 columns.
src += 8 * src_stride; // Go down 8 rows.
dst += 8; // Move over 8 columns.
i -= 8;
}
#endif
if (i > 0) {
TransposeWxH_C(src, src_stride, dst, dst_stride, width, i);
@ -74,9 +100,12 @@ void TransposePlane(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane90(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane90(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Rotate by 90 is a transpose with the source read
// from bottom to top. So set the source pointer to the end
// of the buffer and flip the sign of the source stride.
@ -86,9 +115,12 @@ void RotatePlane90(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane270(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane270(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Rotate by 270 is a transpose with the destination written
// from bottom to top. So set the destination pointer to the end
// of the buffer and flip the sign of the destination stride.
@ -98,9 +130,12 @@ void RotatePlane270(const uint8* src, int src_stride,
}
LIBYUV_API
void RotatePlane180(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void RotatePlane180(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Swap first and last row and mirror the content. Uses a temporary row.
align_buffer_64(row, width);
const uint8* src_bot = src + src_stride * (height - 1);
@ -135,12 +170,20 @@ void RotatePlane180(const uint8* src, int src_stride,
#endif
// TODO(fbarchard): Mirror on mips handle unaligned memory.
#if defined(HAS_MIRRORROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4) &&
IS_ALIGNED(dst, 4) && IS_ALIGNED(dst_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src, 4) &&
IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst, 4) &&
IS_ALIGNED(dst_stride, 4)) {
MirrorRow = MirrorRow_DSPR2;
}
#endif
#if defined(HAS_MIRRORROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
MirrorRow = MirrorRow_Any_MSA;
if (IS_ALIGNED(width, 64)) {
MirrorRow = MirrorRow_MSA;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
CopyRow = IS_ALIGNED(width, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
@ -181,15 +224,24 @@ void RotatePlane180(const uint8* src, int src_stride,
}
LIBYUV_API
void TransposeUV(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void TransposeUV(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height) {
int i = height;
void (*TransposeUVWx8)(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
#if defined(HAS_TRANSPOSEUVWX16_MSA)
void (*TransposeUVWx16)(const uint8* src, int src_stride, uint8* dst_a,
int dst_stride_a, uint8* dst_b, int dst_stride_b,
int width) = TransposeUVWx16_C;
#else
void (*TransposeUVWx8)(const uint8* src, int src_stride, uint8* dst_a,
int dst_stride_a, uint8* dst_b, int dst_stride_b,
int width) = TransposeUVWx8_C;
#endif
#if defined(HAS_TRANSPOSEUVWX8_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
TransposeUVWx8 = TransposeUVWx8_NEON;
@ -204,68 +256,92 @@ void TransposeUV(const uint8* src, int src_stride,
}
#endif
#if defined(HAS_TRANSPOSEUVWX8_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(width, 2) && IS_ALIGNED(src, 4) &&
IS_ALIGNED(src_stride, 4)) {
TransposeUVWx8 = TransposeUVWx8_DSPR2;
}
#endif
#if defined(HAS_TRANSPOSEUVWX16_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
TransposeUVWx16 = TransposeUVWx16_Any_MSA;
if (IS_ALIGNED(width, 8)) {
TransposeUVWx16 = TransposeUVWx16_MSA;
}
}
#endif
#if defined(HAS_TRANSPOSEUVWX16_MSA)
// Work through the source in 8x8 tiles.
while (i >= 16) {
TransposeUVWx16(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
src += 16 * src_stride; // Go down 16 rows.
dst_a += 16; // Move over 8 columns.
dst_b += 16; // Move over 8 columns.
i -= 16;
}
#else
// Work through the source in 8x8 tiles.
while (i >= 8) {
TransposeUVWx8(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
TransposeUVWx8(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
src += 8 * src_stride; // Go down 8 rows.
dst_a += 8; // Move over 8 columns.
dst_b += 8; // Move over 8 columns.
src += 8 * src_stride; // Go down 8 rows.
dst_a += 8; // Move over 8 columns.
dst_b += 8; // Move over 8 columns.
i -= 8;
}
#endif
if (i > 0) {
TransposeUVWxH_C(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
TransposeUVWxH_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width, i);
}
}
LIBYUV_API
void RotateUV90(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV90(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height) {
src += src_stride * (height - 1);
src_stride = -src_stride;
TransposeUV(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
width, height);
TransposeUV(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, width,
height);
}
LIBYUV_API
void RotateUV270(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV270(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height) {
dst_a += dst_stride_a * (width - 1);
dst_b += dst_stride_b * (width - 1);
dst_stride_a = -dst_stride_a;
dst_stride_b = -dst_stride_b;
TransposeUV(src, src_stride,
dst_a, dst_stride_a,
dst_b, dst_stride_b,
width, height);
TransposeUV(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, width,
height);
}
// Rotate 180 is a horizontal and vertical flip.
LIBYUV_API
void RotateUV180(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void RotateUV180(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height) {
int i;
void (*MirrorUVRow)(const uint8* src, uint8* dst_u, uint8* dst_v, int width) =
MirrorUVRow_C;
@ -280,8 +356,8 @@ void RotateUV180(const uint8* src, int src_stride,
}
#endif
#if defined(HAS_MIRRORUVROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src, 4) && IS_ALIGNED(src_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src, 4) &&
IS_ALIGNED(src_stride, 4)) {
MirrorUVRow = MirrorUVRow_DSPR2;
}
#endif
@ -298,9 +374,12 @@ void RotateUV180(const uint8* src, int src_stride,
}
LIBYUV_API
int RotatePlane(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height,
int RotatePlane(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height,
enum RotationMode mode) {
if (!src || width <= 0 || height == 0 || !dst) {
return -1;
@ -316,24 +395,16 @@ int RotatePlane(const uint8* src, int src_stride,
switch (mode) {
case kRotate0:
// copy frame
CopyPlane(src, src_stride,
dst, dst_stride,
width, height);
CopyPlane(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate90:
RotatePlane90(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane90(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate270:
RotatePlane270(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane270(src, src_stride, dst, dst_stride, width, height);
return 0;
case kRotate180:
RotatePlane180(src, src_stride,
dst, dst_stride,
width, height);
RotatePlane180(src, src_stride, dst, dst_stride, width, height);
return 0;
default:
break;
@ -342,18 +413,25 @@ int RotatePlane(const uint8* src, int src_stride,
}
LIBYUV_API
int I420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
int I420Rotate(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height,
enum RotationMode mode) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if (!src_y || !src_u || !src_v || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
if (!src_y || !src_u || !src_v || width <= 0 || height == 0 || !dst_y ||
!dst_u || !dst_v) {
return -1;
}
@ -372,45 +450,29 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
switch (mode) {
case kRotate0:
// copy frame
return I420Copy(src_y, src_stride_y,
src_u, src_stride_u,
src_v, src_stride_v,
dst_y, dst_stride_y,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
width, height);
return I420Copy(src_y, src_stride_y, src_u, src_stride_u, src_v,
src_stride_v, dst_y, dst_stride_y, dst_u, dst_stride_u,
dst_v, dst_stride_v, width, height);
case kRotate90:
RotatePlane90(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane90(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane90(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane90(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane90(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane90(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
case kRotate270:
RotatePlane270(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane270(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane270(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane270(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane270(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane270(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
case kRotate180:
RotatePlane180(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotatePlane180(src_u, src_stride_u,
dst_u, dst_stride_u,
halfwidth, halfheight);
RotatePlane180(src_v, src_stride_v,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane180(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotatePlane180(src_u, src_stride_u, dst_u, dst_stride_u, halfwidth,
halfheight);
RotatePlane180(src_v, src_stride_v, dst_v, dst_stride_v, halfwidth,
halfheight);
return 0;
default:
break;
@ -419,17 +481,23 @@ int I420Rotate(const uint8* src_y, int src_stride_y,
}
LIBYUV_API
int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv,
uint8* dst_y, int dst_stride_y,
uint8* dst_u, int dst_stride_u,
uint8* dst_v, int dst_stride_v,
int width, int height,
int NV12ToI420Rotate(const uint8* src_y,
int src_stride_y,
const uint8* src_uv,
int src_stride_uv,
uint8* dst_y,
int dst_stride_y,
uint8* dst_u,
int dst_stride_u,
uint8* dst_v,
int dst_stride_v,
int width,
int height,
enum RotationMode mode) {
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if (!src_y || !src_uv || width <= 0 || height == 0 ||
!dst_y || !dst_u || !dst_v) {
if (!src_y || !src_uv || width <= 0 || height == 0 || !dst_y || !dst_u ||
!dst_v) {
return -1;
}
@ -446,38 +514,23 @@ int NV12ToI420Rotate(const uint8* src_y, int src_stride_y,
switch (mode) {
case kRotate0:
// copy frame
return NV12ToI420(src_y, src_stride_y,
src_uv, src_stride_uv,
dst_y, dst_stride_y,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
return NV12ToI420(src_y, src_stride_y, src_uv, src_stride_uv, dst_y,
dst_stride_y, dst_u, dst_stride_u, dst_v, dst_stride_v,
width, height);
case kRotate90:
RotatePlane90(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV90(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane90(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV90(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
case kRotate270:
RotatePlane270(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV270(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane270(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV270(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
case kRotate180:
RotatePlane180(src_y, src_stride_y,
dst_y, dst_stride_y,
width, height);
RotateUV180(src_uv, src_stride_uv,
dst_u, dst_stride_u,
dst_v, dst_stride_v,
halfwidth, halfheight);
RotatePlane180(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
RotateUV180(src_uv, src_stride_uv, dst_u, dst_stride_u, dst_v,
dst_stride_v, halfwidth, halfheight);
return 0;
default:
break;

54
third_party/yuv/source/rotate_any.cc vendored
View File

@ -18,16 +18,16 @@ namespace libyuv {
extern "C" {
#endif
#define TANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8* src, int src_stride, \
uint8* dst, int dst_stride, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst, dst_stride, n); \
} \
TransposeWx8_C(src + n, src_stride, dst + n * dst_stride, dst_stride, r);\
}
#define TANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8* src, int src_stride, uint8* dst, int dst_stride, \
int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst, dst_stride, n); \
} \
TransposeWx8_C(src + n, src_stride, dst + n * dst_stride, dst_stride, r); \
}
#ifdef HAS_TRANSPOSEWX8_NEON
TANY(TransposeWx8_Any_NEON, TransposeWx8_NEON, 7)
@ -41,22 +41,22 @@ TANY(TransposeWx8_Fast_Any_SSSE3, TransposeWx8_Fast_SSSE3, 15)
#ifdef HAS_TRANSPOSEWX8_DSPR2
TANY(TransposeWx8_Any_DSPR2, TransposeWx8_DSPR2, 7)
#endif
#ifdef HAS_TRANSPOSEWX16_MSA
TANY(TransposeWx16_Any_MSA, TransposeWx16_MSA, 15)
#endif
#undef TANY
#define TUVANY(NAMEANY, TPOS_SIMD, MASK) \
void NAMEANY(const uint8* src, int src_stride, \
uint8* dst_a, int dst_stride_a, \
uint8* dst_b, int dst_stride_b, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, \
n); \
} \
TransposeUVWx8_C(src + n * 2, src_stride, \
dst_a + n * dst_stride_a, dst_stride_a, \
dst_b + n * dst_stride_b, dst_stride_b, r); \
}
void NAMEANY(const uint8* src, int src_stride, uint8* dst_a, \
int dst_stride_a, uint8* dst_b, int dst_stride_b, int width) { \
int r = width & MASK; \
int n = width - r; \
if (n > 0) { \
TPOS_SIMD(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b, n); \
} \
TransposeUVWx8_C(src + n * 2, src_stride, dst_a + n * dst_stride_a, \
dst_stride_a, dst_b + n * dst_stride_b, dst_stride_b, r); \
}
#ifdef HAS_TRANSPOSEUVWX8_NEON
TUVANY(TransposeUVWx8_Any_NEON, TransposeUVWx8_NEON, 7)
@ -67,14 +67,12 @@ TUVANY(TransposeUVWx8_Any_SSE2, TransposeUVWx8_SSE2, 7)
#ifdef HAS_TRANSPOSEUVWX8_DSPR2
TUVANY(TransposeUVWx8_Any_DSPR2, TransposeUVWx8_DSPR2, 7)
#endif
#ifdef HAS_TRANSPOSEUVWX16_MSA
TUVANY(TransposeUVWx16_Any_MSA, TransposeUVWx16_MSA, 7)
#endif
#undef TUVANY
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

103
third_party/yuv/source/rotate_argb.cc vendored
View File

@ -10,8 +10,8 @@
#include "libyuv/rotate.h"
#include "libyuv/cpu_id.h"
#include "libyuv/convert.h"
#include "libyuv/cpu_id.h"
#include "libyuv/planar_functions.h"
#include "libyuv/row.h"
@ -22,29 +22,44 @@ extern "C" {
// ARGBScale has a function to copy pixels to a row, striding each source
// pixel by a constant.
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(_M_IX86) || \
(defined(__x86_64__) && !defined(__native_client__)) || defined(__i386__))
#if !defined(LIBYUV_DISABLE_X86) && \
(defined(_M_IX86) || \
(defined(__x86_64__) && !defined(__native_client__)) || \
defined(__i386__))
#define HAS_SCALEARGBROWDOWNEVEN_SSE2
void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr, int src_stride,
int src_stepx, uint8* dst_ptr, int dst_width);
void ScaleARGBRowDownEven_SSE2(const uint8* src_ptr,
int src_stride,
int src_stepx,
uint8* dst_ptr,
int dst_width);
#endif
#if !defined(LIBYUV_DISABLE_NEON) && !defined(__native_client__) && \
(defined(__ARM_NEON__) || defined(LIBYUV_NEON) || defined(__aarch64__))
#define HAS_SCALEARGBROWDOWNEVEN_NEON
void ScaleARGBRowDownEven_NEON(const uint8* src_ptr, int src_stride,
int src_stepx, uint8* dst_ptr, int dst_width);
void ScaleARGBRowDownEven_NEON(const uint8* src_ptr,
int src_stride,
int src_stepx,
uint8* dst_ptr,
int dst_width);
#endif
void ScaleARGBRowDownEven_C(const uint8* src_ptr, int,
int src_stepx, uint8* dst_ptr, int dst_width);
void ScaleARGBRowDownEven_C(const uint8* src_ptr,
int,
int src_stepx,
uint8* dst_ptr,
int dst_width);
static void ARGBTranspose(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
static void ARGBTranspose(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
int i;
int src_pixel_step = src_stride >> 2;
void (*ScaleARGBRowDownEven)(const uint8* src_ptr, int src_stride,
int src_step, uint8* dst_ptr, int dst_width) = ScaleARGBRowDownEven_C;
int src_step, uint8* dst_ptr, int dst_width) =
ScaleARGBRowDownEven_C;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(height, 4)) { // Width of dest.
ScaleARGBRowDownEven = ScaleARGBRowDownEven_SSE2;
@ -63,8 +78,12 @@ static void ARGBTranspose(const uint8* src, int src_stride,
}
}
void ARGBRotate90(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
void ARGBRotate90(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Rotate by 90 is a ARGBTranspose with the source read
// from bottom to top. So set the source pointer to the end
// of the buffer and flip the sign of the source stride.
@ -73,8 +92,12 @@ void ARGBRotate90(const uint8* src, int src_stride,
ARGBTranspose(src, src_stride, dst, dst_stride, width, height);
}
void ARGBRotate270(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
void ARGBRotate270(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Rotate by 270 is a ARGBTranspose with the destination written
// from bottom to top. So set the destination pointer to the end
// of the buffer and flip the sign of the destination stride.
@ -83,8 +106,12 @@ void ARGBRotate270(const uint8* src, int src_stride,
ARGBTranspose(src, src_stride, dst, dst_stride, width, height);
}
void ARGBRotate180(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width, int height) {
void ARGBRotate180(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
// Swap first and last row and mirror the content. Uses a temporary row.
align_buffer_64(row, width * 4);
const uint8* src_bot = src + src_stride * (height - 1);
@ -118,6 +145,14 @@ void ARGBRotate180(const uint8* src, int src_stride,
}
}
#endif
#if defined(HAS_ARGBMIRRORROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ARGBMirrorRow = ARGBMirrorRow_Any_MSA;
if (IS_ALIGNED(width, 16)) {
ARGBMirrorRow = ARGBMirrorRow_MSA;
}
}
#endif
#if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
CopyRow = IS_ALIGNED(width * 4, 32) ? CopyRow_SSE2 : CopyRow_Any_SSE2;
@ -146,9 +181,9 @@ void ARGBRotate180(const uint8* src, int src_stride,
// Odd height will harmlessly mirror the middle row twice.
for (y = 0; y < half_height; ++y) {
ARGBMirrorRow(src, row, width); // Mirror first row into a buffer
ARGBMirrorRow(src, row, width); // Mirror first row into a buffer
ARGBMirrorRow(src_bot, dst, width); // Mirror last row into first row
CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last
CopyRow(row, dst_bot, width * 4); // Copy first mirrored row into last
src += src_stride;
dst += dst_stride;
src_bot -= src_stride;
@ -158,8 +193,12 @@ void ARGBRotate180(const uint8* src, int src_stride,
}
LIBYUV_API
int ARGBRotate(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb, int width, int height,
int ARGBRotate(const uint8* src_argb,
int src_stride_argb,
uint8* dst_argb,
int dst_stride_argb,
int width,
int height,
enum RotationMode mode) {
if (!src_argb || width <= 0 || height == 0 || !dst_argb) {
return -1;
@ -175,23 +214,19 @@ int ARGBRotate(const uint8* src_argb, int src_stride_argb,
switch (mode) {
case kRotate0:
// copy frame
return ARGBCopy(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
return ARGBCopy(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width, height);
case kRotate90:
ARGBRotate90(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate90(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
case kRotate270:
ARGBRotate270(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate270(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
case kRotate180:
ARGBRotate180(src_argb, src_stride_argb,
dst_argb, dst_stride_argb,
width, height);
ARGBRotate180(src_argb, src_stride_argb, dst_argb, dst_stride_argb, width,
height);
return 0;
default:
break;

40
third_party/yuv/source/rotate_common.cc vendored
View File

@ -8,16 +8,19 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
void TransposeWx8_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
void TransposeWx8_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst[0] = src[0 * src_stride];
@ -33,9 +36,13 @@ void TransposeWx8_C(const uint8* src, int src_stride,
}
}
void TransposeUVWx8_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width) {
void TransposeUVWx8_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
int i;
for (i = 0; i < width; ++i) {
dst_a[0] = src[0 * src_stride + 0];
@ -60,9 +67,12 @@ void TransposeUVWx8_C(const uint8* src, int src_stride,
}
}
void TransposeWxH_C(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
int width, int height) {
void TransposeWxH_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width,
int height) {
int i;
for (i = 0; i < width; ++i) {
int j;
@ -72,10 +82,14 @@ void TransposeWxH_C(const uint8* src, int src_stride,
}
}
void TransposeUVWxH_C(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int width, int height) {
void TransposeUVWxH_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width,
int height) {
int i;
for (i = 0; i < width * 2; i += 2) {
int j;

307
third_party/yuv/source/rotate_mips.cc → third_party/yuv/source/rotate_dspr2.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
@ -18,18 +18,20 @@ namespace libyuv {
extern "C" {
#endif
#if !defined(LIBYUV_DISABLE_MIPS) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
#if !defined(LIBYUV_DISABLE_DSPR2) && defined(__mips_dsp) && \
(__mips_dsp_rev >= 2) && (_MIPS_SIM == _MIPS_SIM_ABI32)
void TransposeWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__asm__ __volatile__ (
void TransposeWx8_DSPR2(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"addu $t3, $t2, %[src_stride] \n"
"addu $t5, $t4, %[src_stride] \n"
"addu $t6, $t2, $t4 \n"
@ -38,8 +40,8 @@ void TransposeWx8_DSPR2(const uint8* src, int src_stride,
"or $t0, $t0, $t1 \n"
"bnez $t0, 11f \n"
" subu $t7, $t9, %[src_stride] \n"
//dst + dst_stride word aligned
"1: \n"
// dst + dst_stride word aligned
"1: \n"
"lbu $t0, 0(%[src]) \n"
"lbux $t1, %[src_stride](%[src]) \n"
"lbux $t8, $t2(%[src]) \n"
@ -65,8 +67,8 @@ void TransposeWx8_DSPR2(const uint8* src, int src_stride,
"bnez %[width], 1b \n"
" addu %[dst], %[dst], %[dst_stride] \n"
"b 2f \n"
//dst + dst_stride unaligned
"11: \n"
// dst + dst_stride unaligned
"11: \n"
"lbu $t0, 0(%[src]) \n"
"lbux $t1, %[src_stride](%[src]) \n"
"lbux $t8, $t2(%[src]) \n"
@ -92,23 +94,20 @@ void TransposeWx8_DSPR2(const uint8* src, int src_stride,
"swr $s1, 4(%[dst]) \n"
"swl $s1, 7(%[dst]) \n"
"bnez %[width], 11b \n"
"addu %[dst], %[dst], %[dst_stride] \n"
"2: \n"
"addu %[dst], %[dst], %[dst_stride] \n"
"2: \n"
".set pop \n"
:[src] "+r" (src),
[dst] "+r" (dst),
[width] "+r" (width)
:[src_stride] "r" (src_stride),
[dst_stride] "r" (dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1"
);
: [src] "+r"(src), [dst] "+r"(dst), [width] "+r"(width)
: [src_stride] "r"(src_stride), [dst_stride] "r"(dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9", "s0", "s1");
}
void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__asm__ __volatile__ (
void TransposeWx8_Fast_DSPR2(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
__asm__ __volatile__(
".set noat \n"
".set push \n"
".set noreorder \n"
@ -126,67 +125,67 @@ void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
"or $t0, $t0, $t1 \n"
"bnez $t0, 11f \n"
" subu $t7, $t9, %[src_stride] \n"
//dst + dst_stride word aligned
// dst + dst_stride word aligned
"1: \n"
"lw $t0, 0(%[src]) \n"
"lwx $t1, %[src_stride](%[src]) \n"
"lwx $t8, $t2(%[src]) \n"
"lwx $t9, $t3(%[src]) \n"
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
"precr.qb.ph $s4, $s1, $s0 \n"
"precrq.qb.ph $s5, $s1, $s0 \n"
"precr.qb.ph $s6, $s3, $s2 \n"
"precrq.qb.ph $s7, $s3, $s2 \n"
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
"lwx $t0, $t4(%[src]) \n"
"lwx $t1, $t5(%[src]) \n"
"lwx $t8, $t6(%[src]) \n"
"lwx $t9, $t7(%[src]) \n"
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
"precr.qb.ph $t0, $s1, $s0 \n"
"precrq.qb.ph $t1, $s1, $s0 \n"
"precr.qb.ph $t8, $s3, $s2 \n"
"precrq.qb.ph $t9, $s3, $s2 \n"
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
"addu $s0, %[dst], %[dst_stride] \n"
"addu $s1, $s0, %[dst_stride] \n"
@ -207,67 +206,67 @@ void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
"bnez $AT, 1b \n"
" addu %[dst], $s2, %[dst_stride] \n"
"b 2f \n"
//dst + dst_stride unaligned
// dst + dst_stride unaligned
"11: \n"
"lw $t0, 0(%[src]) \n"
"lwx $t1, %[src_stride](%[src]) \n"
"lwx $t8, $t2(%[src]) \n"
"lwx $t9, $t3(%[src]) \n"
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
// t0 = | 30 | 20 | 10 | 00 |
// t1 = | 31 | 21 | 11 | 01 |
// t8 = | 32 | 22 | 12 | 02 |
// t9 = | 33 | 23 | 13 | 03 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
// s0 = | 21 | 01 | 20 | 00 |
// s1 = | 23 | 03 | 22 | 02 |
// s2 = | 31 | 11 | 30 | 10 |
// s3 = | 33 | 13 | 32 | 12 |
"precr.qb.ph $s4, $s1, $s0 \n"
"precrq.qb.ph $s5, $s1, $s0 \n"
"precr.qb.ph $s6, $s3, $s2 \n"
"precrq.qb.ph $s7, $s3, $s2 \n"
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
// s4 = | 03 | 02 | 01 | 00 |
// s5 = | 23 | 22 | 21 | 20 |
// s6 = | 13 | 12 | 11 | 10 |
// s7 = | 33 | 32 | 31 | 30 |
"lwx $t0, $t4(%[src]) \n"
"lwx $t1, $t5(%[src]) \n"
"lwx $t8, $t6(%[src]) \n"
"lwx $t9, $t7(%[src]) \n"
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
// t0 = | 34 | 24 | 14 | 04 |
// t1 = | 35 | 25 | 15 | 05 |
// t8 = | 36 | 26 | 16 | 06 |
// t9 = | 37 | 27 | 17 | 07 |
"precr.qb.ph $s0, $t1, $t0 \n"
"precr.qb.ph $s1, $t9, $t8 \n"
"precrq.qb.ph $s2, $t1, $t0 \n"
"precrq.qb.ph $s3, $t9, $t8 \n"
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
// s0 = | 25 | 05 | 24 | 04 |
// s1 = | 27 | 07 | 26 | 06 |
// s2 = | 35 | 15 | 34 | 14 |
// s3 = | 37 | 17 | 36 | 16 |
"precr.qb.ph $t0, $s1, $s0 \n"
"precrq.qb.ph $t1, $s1, $s0 \n"
"precr.qb.ph $t8, $s3, $s2 \n"
"precrq.qb.ph $t9, $s3, $s2 \n"
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
// t0 = | 07 | 06 | 05 | 04 |
// t1 = | 27 | 26 | 25 | 24 |
// t8 = | 17 | 16 | 15 | 14 |
// t9 = | 37 | 36 | 35 | 34 |
"addu $s0, %[dst], %[dst_stride] \n"
"addu $s1, $s0, %[dst_stride] \n"
@ -298,34 +297,33 @@ void TransposeWx8_Fast_DSPR2(const uint8* src, int src_stride,
"2: \n"
".set pop \n"
".set at \n"
:[src] "+r" (src),
[dst] "+r" (dst),
[width] "+r" (width)
:[src_stride] "r" (src_stride),
[dst_stride] "r" (dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7"
);
: [src] "+r"(src), [dst] "+r"(dst), [width] "+r"(width)
: [src_stride] "r"(src_stride), [dst_stride] "r"(dst_stride)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9", "s0", "s1",
"s2", "s3", "s4", "s5", "s6", "s7");
}
void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
void TransposeUVWx8_DSPR2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
__asm__ __volatile__ (
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"beqz %[width], 2f \n"
" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
"addu $t3, $t2, %[src_stride] \n"
"addu $t5, $t4, %[src_stride] \n"
"addu $t6, $t2, $t4 \n"
"subu $t7, $t9, %[src_stride] \n"
"srl $t1, %[width], 1 \n"
// check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
// check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
"andi $t0, %[dst_a], 0x3 \n"
"andi $t8, %[dst_b], 0x3 \n"
"or $t0, $t0, $t8 \n"
@ -335,52 +333,52 @@ void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
"or $t0, $t0, $t8 \n"
"bnez $t0, 11f \n"
" nop \n"
// dst + dst_stride word aligned (both, a & b dst addresses)
"1: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
// dst + dst_stride word aligned (both, a & b dst addresses)
"1: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
"addu $s5, %[dst_a], %[dst_stride_a] \n"
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"addu $s6, %[dst_b], %[dst_stride_b] \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"sw $s3, 0($s5) \n"
"sw $s4, 0($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"sw $s3, 0(%[dst_a]) \n"
"sw $s4, 0(%[dst_b]) \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"sw $s3, 4($s5) \n"
"sw $s4, 4($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"addiu %[src], 4 \n"
"addiu $t1, -1 \n"
@ -394,59 +392,59 @@ void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
"b 2f \n"
" nop \n"
// dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
"11: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
// dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
"11: \n"
"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
"addu $s5, %[dst_a], %[dst_stride_a] \n"
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
"addu $s6, %[dst_b], %[dst_stride_b] \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
"swr $s3, 0($s5) \n"
"swl $s3, 3($s5) \n"
"swr $s4, 0($s6) \n"
"swl $s4, 3($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
"swr $s3, 0(%[dst_a]) \n"
"swl $s3, 3(%[dst_a]) \n"
"swr $s4, 0(%[dst_b]) \n"
"swl $s4, 3(%[dst_b]) \n"
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
"sll $t0, $t0, 16 \n"
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
"sll $t9, $t9, 16 \n"
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
"swr $s3, 4($s5) \n"
"swl $s3, 7($s5) \n"
"swr $s4, 4($s6) \n"
"swl $s4, 7($s6) \n"
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
"addiu %[src], 4 \n"
"addiu $t1, -1 \n"
@ -462,18 +460,11 @@ void TransposeUVWx8_DSPR2(const uint8* src, int src_stride,
"2: \n"
".set pop \n"
: [src] "+r" (src),
[dst_a] "+r" (dst_a),
[dst_b] "+r" (dst_b),
[width] "+r" (width),
[src_stride] "+r" (src_stride)
: [dst_stride_a] "r" (dst_stride_a),
[dst_stride_b] "r" (dst_stride_b)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3",
"s4", "s5", "s6"
);
: [src] "+r"(src), [dst_a] "+r"(dst_a), [dst_b] "+r"(dst_b),
[width] "+r"(width), [src_stride] "+r"(src_stride)
: [dst_stride_a] "r"(dst_stride_a), [dst_stride_b] "r"(dst_stride_b)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9", "s0", "s1",
"s2", "s3", "s4", "s5", "s6");
}
#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)

660
third_party/yuv/source/rotate_gcc.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
@ -22,342 +22,348 @@ extern "C" {
// Transpose 8x8. 32 or 64 bit, but not NaCL for 64 bit.
#if defined(HAS_TRANSPOSEWX8_SSSE3)
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movq (%0),%%xmm0 \n"
"movq (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"movq (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"movq (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movq (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"movq (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movq (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"lea 0x8(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"sub $0x8,%2 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
);
void TransposeWx8_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movq (%0),%%xmm0 \n"
"movq (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"movq (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"movq (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movq (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"movq (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movq (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"lea 0x8(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"sub $0x8,%2 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7");
}
#endif // defined(HAS_TRANSPOSEWX8_SSSE3)
// Transpose 16x8. 64 bit
#if defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
void TransposeWx8_Fast_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqu (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"movdqa %%xmm8,%%xmm9 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"palignr $0x8,%%xmm9,%%xmm9 \n"
"movdqu (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm2,%%xmm10 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm10 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movdqa %%xmm10,%%xmm11 \n"
"movdqu (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"movdqu (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm4,%%xmm12 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm12 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movdqa %%xmm12,%%xmm13 \n"
"movdqu (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movdqu (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm6,%%xmm14 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"punpckhbw %%xmm7,%%xmm14 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"movdqa %%xmm14,%%xmm15 \n"
"lea 0x10(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"punpcklwd %%xmm10,%%xmm8 \n"
"punpcklwd %%xmm11,%%xmm9 \n"
"movdqa %%xmm8,%%xmm10 \n"
"movdqa %%xmm9,%%xmm11 \n"
"palignr $0x8,%%xmm10,%%xmm10 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"punpcklwd %%xmm14,%%xmm12 \n"
"punpcklwd %%xmm15,%%xmm13 \n"
"movdqa %%xmm12,%%xmm14 \n"
"movdqa %%xmm13,%%xmm15 \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm12,%%xmm8 \n"
"movq %%xmm8,(%1) \n"
"movdqa %%xmm8,%%xmm12 \n"
"palignr $0x8,%%xmm12,%%xmm12 \n"
"movq %%xmm12,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm14,%%xmm10 \n"
"movdqa %%xmm10,%%xmm14 \n"
"movq %%xmm10,(%1) \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"punpckldq %%xmm13,%%xmm9 \n"
"movq %%xmm14,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm9,%%xmm13 \n"
"movq %%xmm9,(%1) \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movq %%xmm13,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm15,%%xmm11 \n"
"movq %%xmm11,(%1) \n"
"movdqa %%xmm11,%%xmm15 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"sub $0x10,%2 \n"
"movq %%xmm15,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14", "xmm15"
);
void TransposeWx8_Fast_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%3),%%xmm1 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqu (%0),%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"movdqa %%xmm8,%%xmm9 \n"
"palignr $0x8,%%xmm1,%%xmm1 \n"
"palignr $0x8,%%xmm9,%%xmm9 \n"
"movdqu (%0,%3),%%xmm3 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm2,%%xmm10 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm10 \n"
"movdqa %%xmm2,%%xmm3 \n"
"movdqa %%xmm10,%%xmm11 \n"
"movdqu (%0),%%xmm4 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"movdqu (%0,%3),%%xmm5 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm4,%%xmm12 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm12 \n"
"movdqa %%xmm4,%%xmm5 \n"
"movdqa %%xmm12,%%xmm13 \n"
"movdqu (%0),%%xmm6 \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movdqu (%0,%3),%%xmm7 \n"
"lea (%0,%3,2),%0 \n"
"movdqa %%xmm6,%%xmm14 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"punpckhbw %%xmm7,%%xmm14 \n"
"neg %3 \n"
"movdqa %%xmm6,%%xmm7 \n"
"movdqa %%xmm14,%%xmm15 \n"
"lea 0x10(%0,%3,8),%0 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"neg %3 \n"
// Second round of bit swap.
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,%%xmm3 \n"
"palignr $0x8,%%xmm2,%%xmm2 \n"
"palignr $0x8,%%xmm3,%%xmm3 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm4,%%xmm6 \n"
"movdqa %%xmm5,%%xmm7 \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"punpcklwd %%xmm10,%%xmm8 \n"
"punpcklwd %%xmm11,%%xmm9 \n"
"movdqa %%xmm8,%%xmm10 \n"
"movdqa %%xmm9,%%xmm11 \n"
"palignr $0x8,%%xmm10,%%xmm10 \n"
"palignr $0x8,%%xmm11,%%xmm11 \n"
"punpcklwd %%xmm14,%%xmm12 \n"
"punpcklwd %%xmm15,%%xmm13 \n"
"movdqa %%xmm12,%%xmm14 \n"
"movdqa %%xmm13,%%xmm15 \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
// Third round of bit swap.
// Write to the destination pointer.
"punpckldq %%xmm4,%%xmm0 \n"
"movq %%xmm0,(%1) \n"
"movdqa %%xmm0,%%xmm4 \n"
"palignr $0x8,%%xmm4,%%xmm4 \n"
"movq %%xmm4,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movdqa %%xmm2,%%xmm6 \n"
"movq %%xmm2,(%1) \n"
"palignr $0x8,%%xmm6,%%xmm6 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movq %%xmm6,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm1,%%xmm5 \n"
"movq %%xmm1,(%1) \n"
"palignr $0x8,%%xmm5,%%xmm5 \n"
"movq %%xmm5,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movq %%xmm3,(%1) \n"
"movdqa %%xmm3,%%xmm7 \n"
"palignr $0x8,%%xmm7,%%xmm7 \n"
"movq %%xmm7,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm12,%%xmm8 \n"
"movq %%xmm8,(%1) \n"
"movdqa %%xmm8,%%xmm12 \n"
"palignr $0x8,%%xmm12,%%xmm12 \n"
"movq %%xmm12,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm14,%%xmm10 \n"
"movdqa %%xmm10,%%xmm14 \n"
"movq %%xmm10,(%1) \n"
"palignr $0x8,%%xmm14,%%xmm14 \n"
"punpckldq %%xmm13,%%xmm9 \n"
"movq %%xmm14,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"movdqa %%xmm9,%%xmm13 \n"
"movq %%xmm9,(%1) \n"
"palignr $0x8,%%xmm13,%%xmm13 \n"
"movq %%xmm13,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"punpckldq %%xmm15,%%xmm11 \n"
"movq %%xmm11,(%1) \n"
"movdqa %%xmm11,%%xmm15 \n"
"palignr $0x8,%%xmm15,%%xmm15 \n"
"sub $0x10,%2 \n"
"movq %%xmm15,(%1,%4) \n"
"lea (%1,%4,2),%1 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst), // %1
"+r"(width) // %2
: "r"((intptr_t)(src_stride)), // %3
"r"((intptr_t)(dst_stride)) // %4
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9", "xmm10", "xmm11", "xmm12", "xmm13", "xmm14",
"xmm15");
}
#endif // defined(HAS_TRANSPOSEWX8_FAST_SSSE3)
// Transpose UV 8x8. 64 bit.
#if defined(HAS_TRANSPOSEUVWX8_SSE2)
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b, int width) {
asm volatile (
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%4),%%xmm1 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqa %%xmm8,%%xmm1 \n"
"movdqu (%0),%%xmm2 \n"
"movdqu (%0,%4),%%xmm3 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm8 \n"
"movdqa %%xmm8,%%xmm3 \n"
"movdqu (%0),%%xmm4 \n"
"movdqu (%0,%4),%%xmm5 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm4,%%xmm8 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm8 \n"
"movdqa %%xmm8,%%xmm5 \n"
"movdqu (%0),%%xmm6 \n"
"movdqu (%0,%4),%%xmm7 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm6,%%xmm8 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %4 \n"
"lea 0x10(%0,%4,8),%0 \n"
"punpckhbw %%xmm7,%%xmm8 \n"
"movdqa %%xmm8,%%xmm7 \n"
"neg %4 \n"
// Second round of bit swap.
"movdqa %%xmm0,%%xmm8 \n"
"movdqa %%xmm1,%%xmm9 \n"
"punpckhwd %%xmm2,%%xmm8 \n"
"punpckhwd %%xmm3,%%xmm9 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm8,%%xmm2 \n"
"movdqa %%xmm9,%%xmm3 \n"
"movdqa %%xmm4,%%xmm8 \n"
"movdqa %%xmm5,%%xmm9 \n"
"punpckhwd %%xmm6,%%xmm8 \n"
"punpckhwd %%xmm7,%%xmm9 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm8,%%xmm6 \n"
"movdqa %%xmm9,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"movdqa %%xmm0,%%xmm8 \n"
"punpckldq %%xmm4,%%xmm0 \n"
"movlpd %%xmm0,(%1) \n" // Write back U channel
"movhpd %%xmm0,(%2) \n" // Write back V channel
"punpckhdq %%xmm4,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movlpd %%xmm2,(%1) \n"
"movhpd %%xmm2,(%2) \n"
"punpckhdq %%xmm6,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm1,%%xmm8 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movlpd %%xmm1,(%1) \n"
"movhpd %%xmm1,(%2) \n"
"punpckhdq %%xmm5,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm3,%%xmm8 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movlpd %%xmm3,(%1) \n"
"movhpd %%xmm3,(%2) \n"
"punpckhdq %%xmm7,%%xmm8 \n"
"sub $0x8,%3 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst_a), // %1
"+r"(dst_b), // %2
"+r"(width) // %3
: "r"((intptr_t)(src_stride)), // %4
"r"((intptr_t)(dst_stride_a)), // %5
"r"((intptr_t)(dst_stride_b)) // %6
: "memory", "cc",
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7",
"xmm8", "xmm9"
);
void TransposeUVWx8_SSE2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
asm volatile(
// Read in the data from the source pointer.
// First round of bit swap.
LABELALIGN
"1: \n"
"movdqu (%0),%%xmm0 \n"
"movdqu (%0,%4),%%xmm1 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm0,%%xmm8 \n"
"punpcklbw %%xmm1,%%xmm0 \n"
"punpckhbw %%xmm1,%%xmm8 \n"
"movdqa %%xmm8,%%xmm1 \n"
"movdqu (%0),%%xmm2 \n"
"movdqu (%0,%4),%%xmm3 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"punpckhbw %%xmm3,%%xmm8 \n"
"movdqa %%xmm8,%%xmm3 \n"
"movdqu (%0),%%xmm4 \n"
"movdqu (%0,%4),%%xmm5 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm4,%%xmm8 \n"
"punpcklbw %%xmm5,%%xmm4 \n"
"punpckhbw %%xmm5,%%xmm8 \n"
"movdqa %%xmm8,%%xmm5 \n"
"movdqu (%0),%%xmm6 \n"
"movdqu (%0,%4),%%xmm7 \n"
"lea (%0,%4,2),%0 \n"
"movdqa %%xmm6,%%xmm8 \n"
"punpcklbw %%xmm7,%%xmm6 \n"
"neg %4 \n"
"lea 0x10(%0,%4,8),%0 \n"
"punpckhbw %%xmm7,%%xmm8 \n"
"movdqa %%xmm8,%%xmm7 \n"
"neg %4 \n"
// Second round of bit swap.
"movdqa %%xmm0,%%xmm8 \n"
"movdqa %%xmm1,%%xmm9 \n"
"punpckhwd %%xmm2,%%xmm8 \n"
"punpckhwd %%xmm3,%%xmm9 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpcklwd %%xmm3,%%xmm1 \n"
"movdqa %%xmm8,%%xmm2 \n"
"movdqa %%xmm9,%%xmm3 \n"
"movdqa %%xmm4,%%xmm8 \n"
"movdqa %%xmm5,%%xmm9 \n"
"punpckhwd %%xmm6,%%xmm8 \n"
"punpckhwd %%xmm7,%%xmm9 \n"
"punpcklwd %%xmm6,%%xmm4 \n"
"punpcklwd %%xmm7,%%xmm5 \n"
"movdqa %%xmm8,%%xmm6 \n"
"movdqa %%xmm9,%%xmm7 \n"
// Third round of bit swap.
// Write to the destination pointer.
"movdqa %%xmm0,%%xmm8 \n"
"punpckldq %%xmm4,%%xmm0 \n"
"movlpd %%xmm0,(%1) \n" // Write back U channel
"movhpd %%xmm0,(%2) \n" // Write back V channel
"punpckhdq %%xmm4,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm2,%%xmm8 \n"
"punpckldq %%xmm6,%%xmm2 \n"
"movlpd %%xmm2,(%1) \n"
"movhpd %%xmm2,(%2) \n"
"punpckhdq %%xmm6,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm1,%%xmm8 \n"
"punpckldq %%xmm5,%%xmm1 \n"
"movlpd %%xmm1,(%1) \n"
"movhpd %%xmm1,(%2) \n"
"punpckhdq %%xmm5,%%xmm8 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"movdqa %%xmm3,%%xmm8 \n"
"punpckldq %%xmm7,%%xmm3 \n"
"movlpd %%xmm3,(%1) \n"
"movhpd %%xmm3,(%2) \n"
"punpckhdq %%xmm7,%%xmm8 \n"
"sub $0x8,%3 \n"
"movlpd %%xmm8,(%1,%5) \n"
"lea (%1,%5,2),%1 \n"
"movhpd %%xmm8,(%2,%6) \n"
"lea (%2,%6,2),%2 \n"
"jg 1b \n"
: "+r"(src), // %0
"+r"(dst_a), // %1
"+r"(dst_b), // %2
"+r"(width) // %3
: "r"((intptr_t)(src_stride)), // %4
"r"((intptr_t)(dst_stride_a)), // %5
"r"((intptr_t)(dst_stride_b)) // %6
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6",
"xmm7", "xmm8", "xmm9");
}
#endif // defined(HAS_TRANSPOSEUVWX8_SSE2)
#endif // defined(__x86_64__) || defined(__i386__)

250
third_party/yuv/source/rotate_msa.cc vendored Normal file
View File

@ -0,0 +1,250 @@
/*
* Copyright 2016 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/rotate_row.h"
// This module is for GCC MSA
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include "libyuv/macros_msa.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
#define ILVRL_B(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_b((v16i8)in1, (v16i8)in0); \
out1 = (v16u8)__msa_ilvl_b((v16i8)in1, (v16i8)in0); \
out2 = (v16u8)__msa_ilvr_b((v16i8)in3, (v16i8)in2); \
out3 = (v16u8)__msa_ilvl_b((v16i8)in3, (v16i8)in2); \
}
#define ILVRL_H(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_h((v8i16)in1, (v8i16)in0); \
out1 = (v16u8)__msa_ilvl_h((v8i16)in1, (v8i16)in0); \
out2 = (v16u8)__msa_ilvr_h((v8i16)in3, (v8i16)in2); \
out3 = (v16u8)__msa_ilvl_h((v8i16)in3, (v8i16)in2); \
}
#define ILVRL_W(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_w((v4i32)in1, (v4i32)in0); \
out1 = (v16u8)__msa_ilvl_w((v4i32)in1, (v4i32)in0); \
out2 = (v16u8)__msa_ilvr_w((v4i32)in3, (v4i32)in2); \
out3 = (v16u8)__msa_ilvl_w((v4i32)in3, (v4i32)in2); \
}
#define ILVRL_D(in0, in1, in2, in3, out0, out1, out2, out3) \
{ \
out0 = (v16u8)__msa_ilvr_d((v2i64)in1, (v2i64)in0); \
out1 = (v16u8)__msa_ilvl_d((v2i64)in1, (v2i64)in0); \
out2 = (v16u8)__msa_ilvr_d((v2i64)in3, (v2i64)in2); \
out3 = (v16u8)__msa_ilvl_d((v2i64)in3, (v2i64)in2); \
}
void TransposeWx16_C(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
TransposeWx8_C(src, src_stride, dst, dst_stride, width);
TransposeWx8_C((src + 8 * src_stride), src_stride, (dst + 8), dst_stride,
width);
}
void TransposeUVWx16_C(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
TransposeUVWx8_C(src, src_stride, dst_a, dst_stride_a, dst_b, dst_stride_b,
width);
TransposeUVWx8_C((src + 8 * src_stride), src_stride, (dst_a + 8),
dst_stride_a, (dst_b + 8), dst_stride_b, width);
}
void TransposeWx16_MSA(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
int x;
const uint8* s;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < width; x += 16) {
s = src;
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
dst += dst_stride * 4;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
ST_UB4(dst0, dst1, dst2, dst3, dst, dst_stride);
src += 16;
dst += dst_stride * 4;
}
}
void TransposeUVWx16_MSA(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
int x;
const uint8* s;
v16u8 src0, src1, src2, src3, dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3;
v16u8 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
v16u8 res0, res1, res2, res3, res4, res5, res6, res7, res8, res9;
for (x = 0; x < width; x += 8) {
s = src;
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
ILVRL_W(reg0, reg4, reg1, reg5, res0, res1, res2, res3);
ILVRL_W(reg2, reg6, reg3, reg7, res4, res5, res6, res7);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg0, reg1, reg2, reg3);
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src1 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src2 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
src3 = (v16u8)__msa_ld_b((v16i8*)s, 0);
s += src_stride;
ILVRL_B(src0, src1, src2, src3, vec0, vec1, vec2, vec3);
ILVRL_H(vec0, vec2, vec1, vec3, reg4, reg5, reg6, reg7);
res8 = (v16u8)__msa_ilvr_w((v4i32)reg4, (v4i32)reg0);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg4, (v4i32)reg0);
ILVRL_D(res0, res8, res1, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg5, (v4i32)reg1);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg5, (v4i32)reg1);
ILVRL_D(res2, res8, res3, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg6, (v4i32)reg2);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg6, (v4i32)reg2);
ILVRL_D(res4, res8, res5, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
res8 = (v16u8)__msa_ilvr_w((v4i32)reg7, (v4i32)reg3);
res9 = (v16u8)__msa_ilvl_w((v4i32)reg7, (v4i32)reg3);
ILVRL_D(res6, res8, res7, res9, dst0, dst1, dst2, dst3);
ST_UB2(dst0, dst2, dst_a, dst_stride_a);
ST_UB2(dst1, dst3, dst_b, dst_stride_b);
src += 16;
dst_a += dst_stride_a * 2;
dst_b += dst_stride_b * 2;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)

563
third_party/yuv/source/rotate_neon.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
@ -21,38 +21,32 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) && \
!defined(__aarch64__)
static uvec8 kVTbl4x4Transpose =
{ 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
static uvec8 kVTbl4x4Transpose = {0, 4, 8, 12, 1, 5, 9, 13,
2, 6, 10, 14, 3, 7, 11, 15};
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride,
void TransposeWx8_NEON(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
const uint8* src_temp;
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %5, #8 \n"
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %5, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.8 {d0}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d1}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d2}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d3}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d4}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d5}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d6}, [%0], %2 \n"
MEMACCESS(0)
"vld1.8 {d7}, [%0] \n"
"vtrn.8 d1, d0 \n"
@ -77,21 +71,13 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.8 {d1}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d3}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d2}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d5}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d4}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d7}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d6}, [%0] \n"
"add %1, #8 \n" // src += 8
@ -99,180 +85,138 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"subs %5, #8 \n" // w -= 8
"bge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %5, #8 \n"
"beq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %5, #8 \n"
"beq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %5, #2 \n"
"blt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %5, #2 \n"
"blt 3f \n"
"cmp %5, #4 \n"
"blt 2f \n"
"cmp %5, #4 \n"
"blt 2f \n"
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.32 {d0[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d0[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d1[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d1[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d2[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d2[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d3[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.32 {d3[1]}, [%0] \n"
// 4x8 block
"mov %0, %1 \n"
"vld1.32 {d0[0]}, [%0], %2 \n"
"vld1.32 {d0[1]}, [%0], %2 \n"
"vld1.32 {d1[0]}, [%0], %2 \n"
"vld1.32 {d1[1]}, [%0], %2 \n"
"vld1.32 {d2[0]}, [%0], %2 \n"
"vld1.32 {d2[1]}, [%0], %2 \n"
"vld1.32 {d3[0]}, [%0], %2 \n"
"vld1.32 {d3[1]}, [%0] \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(6)
"vld1.8 {q3}, [%6] \n"
"vld1.8 {q3}, [%6] \n"
"vtbl.8 d4, {d0, d1}, d6 \n"
"vtbl.8 d5, {d0, d1}, d7 \n"
"vtbl.8 d0, {d2, d3}, d6 \n"
"vtbl.8 d1, {d2, d3}, d7 \n"
"vtbl.8 d4, {d0, d1}, d6 \n"
"vtbl.8 d5, {d0, d1}, d7 \n"
"vtbl.8 d0, {d2, d3}, d6 \n"
"vtbl.8 d1, {d2, d3}, d7 \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
MEMACCESS(0)
"vst1.32 {d4[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d4[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d5[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d5[1]}, [%0] \n"
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
"vst1.32 {d4[0]}, [%0], %4 \n"
"vst1.32 {d4[1]}, [%0], %4 \n"
"vst1.32 {d5[0]}, [%0], %4 \n"
"vst1.32 {d5[1]}, [%0] \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"vst1.32 {d0[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d0[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d1[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d1[1]}, [%0] \n"
"add %0, %3, #4 \n"
"vst1.32 {d0[0]}, [%0], %4 \n"
"vst1.32 {d0[1]}, [%0], %4 \n"
"vst1.32 {d1[0]}, [%0], %4 \n"
"vst1.32 {d1[1]}, [%0] \n"
"add %1, #4 \n" // src += 4
"add %3, %3, %4, lsl #2 \n" // dst += 4 * dst_stride
"subs %5, #4 \n" // w -= 4
"beq 4f \n"
"add %1, #4 \n" // src += 4
"add %3, %3, %4, lsl #2 \n" // dst += 4 * dst_stride
"subs %5, #4 \n" // w -= 4
"beq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %5, #2 \n"
"blt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %5, #2 \n"
"blt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.16 {d0[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d0[3]}, [%0], %2 \n"
MEMACCESS(0)
"vld1.16 {d1[3]}, [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"vld1.16 {d0[0]}, [%0], %2 \n"
"vld1.16 {d1[0]}, [%0], %2 \n"
"vld1.16 {d0[1]}, [%0], %2 \n"
"vld1.16 {d1[1]}, [%0], %2 \n"
"vld1.16 {d0[2]}, [%0], %2 \n"
"vld1.16 {d1[2]}, [%0], %2 \n"
"vld1.16 {d0[3]}, [%0], %2 \n"
"vld1.16 {d1[3]}, [%0] \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d0, d1 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.64 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.64 {d1}, [%0] \n"
"vst1.64 {d0}, [%0], %4 \n"
"vst1.64 {d1}, [%0] \n"
"add %1, #2 \n" // src += 2
"add %3, %3, %4, lsl #1 \n" // dst += 2 * dst_stride
"subs %5, #2 \n" // w -= 2
"beq 4f \n"
"add %1, #2 \n" // src += 2
"add %3, %3, %4, lsl #1 \n" // dst += 2 * dst_stride
"subs %5, #2 \n" // w -= 2
"beq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"vld1.8 {d0[0]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[1]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[2]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[3]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[4]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[5]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[6]}, [%1], %2 \n"
MEMACCESS(1)
"vld1.8 {d0[7]}, [%1] \n"
// 1x8 block
"3: \n"
"vld1.8 {d0[0]}, [%1], %2 \n"
"vld1.8 {d0[1]}, [%1], %2 \n"
"vld1.8 {d0[2]}, [%1], %2 \n"
"vld1.8 {d0[3]}, [%1], %2 \n"
"vld1.8 {d0[4]}, [%1], %2 \n"
"vld1.8 {d0[5]}, [%1], %2 \n"
"vld1.8 {d0[6]}, [%1], %2 \n"
"vld1.8 {d0[7]}, [%1] \n"
MEMACCESS(3)
"vst1.64 {d0}, [%3] \n"
"vst1.64 {d0}, [%3] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst), // %3
"+r"(dst_stride), // %4
"+r"(width) // %5
: "r"(&kVTbl4x4Transpose) // %6
: "memory", "cc", "q0", "q1", "q2", "q3"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst), // %3
"+r"(dst_stride), // %4
"+r"(width) // %5
: "r"(&kVTbl4x4Transpose) // %6
: "memory", "cc", "q0", "q1", "q2", "q3");
}
static uvec8 kVTbl4x4TransposeDi =
{ 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
static uvec8 kVTbl4x4TransposeDi = {0, 8, 1, 9, 2, 10, 3, 11,
4, 12, 5, 13, 6, 14, 7, 15};
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
void TransposeUVWx8_NEON(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
const uint8* src_temp;
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %7, #8 \n"
asm volatile(
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %7, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld2.8 {d0, d1}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d2, d3}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d4, d5}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d6, d7}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d16, d17}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d18, d19}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d20, d21}, [%0], %2 \n"
MEMACCESS(0)
"vld2.8 {d22, d23}, [%0] \n"
"vtrn.8 q1, q0 \n"
@ -301,40 +245,24 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.8 {d2}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d6}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d4}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d18}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d16}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d22}, [%0], %4 \n"
MEMACCESS(0)
"vst1.8 {d20}, [%0] \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.8 {d3}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d1}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d7}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d5}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d19}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d17}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d23}, [%0], %6 \n"
MEMACCESS(0)
"vst1.8 {d21}, [%0] \n"
"add %1, #8*2 \n" // src += 8*2
@ -343,187 +271,142 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"subs %7, #8 \n" // w -= 8
"bge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %7, #8 \n"
"beq 4f \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %7, #8 \n"
"beq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %7, #2 \n"
"blt 3f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %7, #2 \n"
"blt 3f \n"
"cmp %7, #4 \n"
"blt 2f \n"
"cmp %7, #4 \n"
"blt 2f \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"vld1.64 {d0}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d1}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d2}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d3}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d4}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d5}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d6}, [%0], %2 \n"
MEMACCESS(0)
"vld1.64 {d7}, [%0] \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
"vld1.64 {d0}, [%0], %2 \n"
"vld1.64 {d1}, [%0], %2 \n"
"vld1.64 {d2}, [%0], %2 \n"
"vld1.64 {d3}, [%0], %2 \n"
"vld1.64 {d4}, [%0], %2 \n"
"vld1.64 {d5}, [%0], %2 \n"
"vld1.64 {d6}, [%0], %2 \n"
"vld1.64 {d7}, [%0] \n"
MEMACCESS(8)
"vld1.8 {q15}, [%8] \n"
"vld1.8 {q15}, [%8] \n"
"vtrn.8 q0, q1 \n"
"vtrn.8 q2, q3 \n"
"vtrn.8 q0, q1 \n"
"vtrn.8 q2, q3 \n"
"vtbl.8 d16, {d0, d1}, d30 \n"
"vtbl.8 d17, {d0, d1}, d31 \n"
"vtbl.8 d18, {d2, d3}, d30 \n"
"vtbl.8 d19, {d2, d3}, d31 \n"
"vtbl.8 d20, {d4, d5}, d30 \n"
"vtbl.8 d21, {d4, d5}, d31 \n"
"vtbl.8 d22, {d6, d7}, d30 \n"
"vtbl.8 d23, {d6, d7}, d31 \n"
"vtbl.8 d16, {d0, d1}, d30 \n"
"vtbl.8 d17, {d0, d1}, d31 \n"
"vtbl.8 d18, {d2, d3}, d30 \n"
"vtbl.8 d19, {d2, d3}, d31 \n"
"vtbl.8 d20, {d4, d5}, d30 \n"
"vtbl.8 d21, {d4, d5}, d31 \n"
"vtbl.8 d22, {d6, d7}, d30 \n"
"vtbl.8 d23, {d6, d7}, d31 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.32 {d16[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d16[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d17[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d17[1]}, [%0], %4 \n"
"vst1.32 {d16[0]}, [%0], %4 \n"
"vst1.32 {d16[1]}, [%0], %4 \n"
"vst1.32 {d17[0]}, [%0], %4 \n"
"vst1.32 {d17[1]}, [%0], %4 \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"vst1.32 {d20[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d20[1]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d21[0]}, [%0], %4 \n"
MEMACCESS(0)
"vst1.32 {d21[1]}, [%0] \n"
"add %0, %3, #4 \n"
"vst1.32 {d20[0]}, [%0], %4 \n"
"vst1.32 {d20[1]}, [%0], %4 \n"
"vst1.32 {d21[0]}, [%0], %4 \n"
"vst1.32 {d21[1]}, [%0] \n"
"mov %0, %5 \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.32 {d18[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d18[1]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d19[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d19[1]}, [%0], %6 \n"
"vst1.32 {d18[0]}, [%0], %6 \n"
"vst1.32 {d18[1]}, [%0], %6 \n"
"vst1.32 {d19[0]}, [%0], %6 \n"
"vst1.32 {d19[1]}, [%0], %6 \n"
"add %0, %5, #4 \n"
MEMACCESS(0)
"vst1.32 {d22[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d22[1]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d23[0]}, [%0], %6 \n"
MEMACCESS(0)
"vst1.32 {d23[1]}, [%0] \n"
"add %0, %5, #4 \n"
"vst1.32 {d22[0]}, [%0], %6 \n"
"vst1.32 {d22[1]}, [%0], %6 \n"
"vst1.32 {d23[0]}, [%0], %6 \n"
"vst1.32 {d23[1]}, [%0] \n"
"add %1, #4*2 \n" // src += 4 * 2
"add %3, %3, %4, lsl #2 \n" // dst_a += 4 * dst_stride_a
"add %5, %5, %6, lsl #2 \n" // dst_b += 4 * dst_stride_b
"subs %7, #4 \n" // w -= 4
"beq 4f \n"
"add %1, #4*2 \n" // src += 4 * 2
"add %3, %3, %4, lsl #2 \n" // dst_a += 4 *
// dst_stride_a
"add %5, %5, %6, lsl #2 \n" // dst_b += 4 *
// dst_stride_b
"subs %7, #4 \n" // w -= 4
"beq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %7, #2 \n"
"blt 3f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %7, #2 \n"
"blt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"vld2.16 {d0[0], d2[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[0], d3[0]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[1], d2[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[1], d3[1]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[2], d2[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[2], d3[2]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d0[3], d2[3]}, [%0], %2 \n"
MEMACCESS(0)
"vld2.16 {d1[3], d3[3]}, [%0] \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
"vld2.16 {d0[0], d2[0]}, [%0], %2 \n"
"vld2.16 {d1[0], d3[0]}, [%0], %2 \n"
"vld2.16 {d0[1], d2[1]}, [%0], %2 \n"
"vld2.16 {d1[1], d3[1]}, [%0], %2 \n"
"vld2.16 {d0[2], d2[2]}, [%0], %2 \n"
"vld2.16 {d1[2], d3[2]}, [%0], %2 \n"
"vld2.16 {d0[3], d2[3]}, [%0], %2 \n"
"vld2.16 {d1[3], d3[3]}, [%0] \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d2, d3 \n"
"vtrn.8 d0, d1 \n"
"vtrn.8 d2, d3 \n"
"mov %0, %3 \n"
"mov %0, %3 \n"
MEMACCESS(0)
"vst1.64 {d0}, [%0], %4 \n"
MEMACCESS(0)
"vst1.64 {d2}, [%0] \n"
"vst1.64 {d0}, [%0], %4 \n"
"vst1.64 {d2}, [%0] \n"
"mov %0, %5 \n"
"mov %0, %5 \n"
MEMACCESS(0)
"vst1.64 {d1}, [%0], %6 \n"
MEMACCESS(0)
"vst1.64 {d3}, [%0] \n"
"vst1.64 {d1}, [%0], %6 \n"
"vst1.64 {d3}, [%0] \n"
"add %1, #2*2 \n" // src += 2 * 2
"add %3, %3, %4, lsl #1 \n" // dst_a += 2 * dst_stride_a
"add %5, %5, %6, lsl #1 \n" // dst_b += 2 * dst_stride_b
"subs %7, #2 \n" // w -= 2
"beq 4f \n"
"add %1, #2*2 \n" // src += 2 * 2
"add %3, %3, %4, lsl #1 \n" // dst_a += 2 *
// dst_stride_a
"add %5, %5, %6, lsl #1 \n" // dst_b += 2 *
// dst_stride_b
"subs %7, #2 \n" // w -= 2
"beq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"vld2.8 {d0[0], d1[0]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[1], d1[1]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[2], d1[2]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[3], d1[3]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[4], d1[4]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[5], d1[5]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[6], d1[6]}, [%1], %2 \n"
MEMACCESS(1)
"vld2.8 {d0[7], d1[7]}, [%1] \n"
// 1x8 block
"3: \n"
"vld2.8 {d0[0], d1[0]}, [%1], %2 \n"
"vld2.8 {d0[1], d1[1]}, [%1], %2 \n"
"vld2.8 {d0[2], d1[2]}, [%1], %2 \n"
"vld2.8 {d0[3], d1[3]}, [%1], %2 \n"
"vld2.8 {d0[4], d1[4]}, [%1], %2 \n"
"vld2.8 {d0[5], d1[5]}, [%1], %2 \n"
"vld2.8 {d0[6], d1[6]}, [%1], %2 \n"
"vld2.8 {d0[7], d1[7]}, [%1] \n"
MEMACCESS(3)
"vst1.64 {d0}, [%3] \n"
MEMACCESS(5)
"vst1.64 {d1}, [%5] \n"
"vst1.64 {d0}, [%3] \n"
"vst1.64 {d1}, [%5] \n"
"4: \n"
"4: \n"
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst_a), // %3
"+r"(dst_stride_a), // %4
"+r"(dst_b), // %5
"+r"(dst_stride_b), // %6
"+r"(width) // %7
: "r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc",
"q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
);
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(src_stride), // %2
"+r"(dst_a), // %3
"+r"(dst_stride_a), // %4
"+r"(dst_b), // %5
"+r"(dst_stride_b), // %6
"+r"(width) // %7
: "r"(&kVTbl4x4TransposeDi) // %8
: "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11");
}
#endif // defined(__ARM_NEON__) && !defined(__aarch64__)

189
third_party/yuv/source/rotate_neon64.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#include "libyuv/basic_types.h"
@ -21,38 +21,32 @@ extern "C" {
// This module is for GCC Neon armv8 64 bit.
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
static uvec8 kVTbl4x4Transpose =
{ 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
static uvec8 kVTbl4x4Transpose = {0, 4, 8, 12, 1, 5, 9, 13,
2, 6, 10, 14, 3, 7, 11, 15};
void TransposeWx8_NEON(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
void TransposeWx8_NEON(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
const uint8* src_temp;
int64 width64 = (int64) width; // Work around clang 3.4 warning.
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %3, %3, #8 \n"
"sub %w3, %w3, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.8b}, [%0] \n"
"trn2 v16.8b, v0.8b, v1.8b \n"
@ -84,62 +78,45 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v17.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v21.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v20.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v23.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v22.8b}, [%0] \n"
"add %1, %1, #8 \n" // src += 8
"add %2, %2, %6, lsl #3 \n" // dst += 8 * dst_stride
"subs %3, %3, #8 \n" // w -= 8
"subs %w3, %w3, #8 \n" // w -= 8
"b.ge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %3, %3, #8 \n"
"adds %w3, %w3, #8 \n"
"b.eq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %3, #2 \n"
"cmp %w3, #2 \n"
"b.lt 3f \n"
"cmp %3, #4 \n"
"cmp %w3, #4 \n"
"b.lt 2f \n"
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.s}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.s}[3], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.s}[3], [%0] \n"
"mov %0, %2 \n"
MEMACCESS(4)
"ld1 {v2.16b}, [%4] \n"
"tbl v3.16b, {v0.16b}, v2.16b \n"
@ -147,53 +124,37 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
// TODO(frkoenig): Rework shuffle above to
// write out with 4 instead of 8 writes.
MEMACCESS(0)
"st1 {v3.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v3.s}[3], [%0] \n"
"add %0, %2, #4 \n"
MEMACCESS(0)
"st1 {v0.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v0.s}[3], [%0] \n"
"add %1, %1, #4 \n" // src += 4
"add %2, %2, %6, lsl #2 \n" // dst += 4 * dst_stride
"subs %3, %3, #4 \n" // w -= 4
"subs %w3, %w3, #4 \n" // w -= 4
"b.eq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %3, #2 \n"
"cmp %w3, #2 \n"
"b.lt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v0.h}[3], [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.h}[3], [%0] \n"
"trn2 v2.8b, v0.8b, v1.8b \n"
@ -201,36 +162,25 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v3.8b}, [%0], %6 \n"
MEMACCESS(0)
"st1 {v2.8b}, [%0] \n"
"add %1, %1, #2 \n" // src += 2
"add %2, %2, %6, lsl #1 \n" // dst += 2 * dst_stride
"subs %3, %3, #2 \n" // w -= 2
"subs %w3, %w3, #2 \n" // w -= 2
"b.eq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"ld1 {v0.b}[0], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[1], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[2], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[3], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[4], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[5], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[6], [%1], %5 \n"
MEMACCESS(1)
"ld1 {v0.b}[7], [%1] \n"
MEMACCESS(2)
"st1 {v0.8b}, [%2] \n"
"4: \n"
@ -238,7 +188,7 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
: "=&r"(src_temp), // %0
"+r"(src), // %1
"+r"(dst), // %2
"+r"(width64) // %3
"+r"(width) // %3
: "r"(&kVTbl4x4Transpose), // %4
"r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride)) // %6
@ -247,41 +197,35 @@ void TransposeWx8_NEON(const uint8* src, int src_stride,
);
}
static uint8 kVTbl4x4TransposeDi[32] =
{ 0, 16, 32, 48, 2, 18, 34, 50, 4, 20, 36, 52, 6, 22, 38, 54,
1, 17, 33, 49, 3, 19, 35, 51, 5, 21, 37, 53, 7, 23, 39, 55};
static uint8 kVTbl4x4TransposeDi[32] = {
0, 16, 32, 48, 2, 18, 34, 50, 4, 20, 36, 52, 6, 22, 38, 54,
1, 17, 33, 49, 3, 19, 35, 51, 5, 21, 37, 53, 7, 23, 39, 55};
void TransposeUVWx8_NEON(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
void TransposeUVWx8_NEON(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int width) {
const uint8* src_temp;
int64 width64 = (int64) width; // Work around clang 3.4 warning.
asm volatile (
// loops are on blocks of 8. loop will stop when
// counter gets to or below 0. starting the counter
// at w-8 allow for this
"sub %4, %4, #8 \n"
"sub %w4, %w4, #8 \n"
// handle 8x8 blocks. this should be the majority of the plane
"1: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.16b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.16b}, [%0] \n"
"trn1 v16.16b, v0.16b, v1.16b \n"
@ -313,81 +257,56 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v16.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v17.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v17.d}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v19.d}[1], [%0] \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v20.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v22.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v21.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v23.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v20.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v22.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v21.d}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v23.d}[1], [%0] \n"
"add %1, %1, #16 \n" // src += 8*2
"add %2, %2, %6, lsl #3 \n" // dst_a += 8 * dst_stride_a
"add %3, %3, %7, lsl #3 \n" // dst_b += 8 * dst_stride_b
"subs %4, %4, #8 \n" // w -= 8
"subs %w4, %w4, #8 \n" // w -= 8
"b.ge 1b \n"
// add 8 back to counter. if the result is 0 there are
// no residuals.
"adds %4, %4, #8 \n"
"adds %w4, %w4, #8 \n"
"b.eq 4f \n"
// some residual, so between 1 and 7 lines left to transpose
"cmp %4, #2 \n"
"cmp %w4, #2 \n"
"b.lt 3f \n"
"cmp %4, #4 \n"
"cmp %w4, #4 \n"
"b.lt 2f \n"
// TODO(frkoenig): Clean this up
// 4x8 block
"mov %0, %1 \n"
MEMACCESS(0)
"ld1 {v0.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v1.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v2.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v3.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v4.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v5.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v6.8b}, [%0], %5 \n"
MEMACCESS(0)
"ld1 {v7.8b}, [%0] \n"
MEMACCESS(8)
"ld1 {v30.16b}, [%8], #16 \n"
"ld1 {v31.16b}, [%8] \n"
@ -398,75 +317,51 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v16.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v16.s}[3], [%0], %6 \n"
"add %0, %2, #4 \n"
MEMACCESS(0)
"st1 {v18.s}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[1], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[2], [%0], %6 \n"
MEMACCESS(0)
"st1 {v18.s}[3], [%0] \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v17.s}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[2], [%0], %7 \n"
MEMACCESS(0)
"st1 {v17.s}[3], [%0], %7 \n"
"add %0, %3, #4 \n"
MEMACCESS(0)
"st1 {v19.s}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[1], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[2], [%0], %7 \n"
MEMACCESS(0)
"st1 {v19.s}[3], [%0] \n"
"add %1, %1, #8 \n" // src += 4 * 2
"add %2, %2, %6, lsl #2 \n" // dst_a += 4 * dst_stride_a
"add %3, %3, %7, lsl #2 \n" // dst_b += 4 * dst_stride_b
"subs %4, %4, #4 \n" // w -= 4
"subs %w4, %w4, #4 \n" // w -= 4
"b.eq 4f \n"
// some residual, check to see if it includes a 2x8 block,
// or less
"cmp %4, #2 \n"
"cmp %w4, #2 \n"
"b.lt 3f \n"
// 2x8 block
"2: \n"
"mov %0, %1 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[0], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[1], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[2], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v0.h, v1.h}[3], [%0], %5 \n"
MEMACCESS(0)
"ld2 {v2.h, v3.h}[3], [%0] \n"
"trn1 v4.8b, v0.8b, v2.8b \n"
@ -476,46 +371,32 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"mov %0, %2 \n"
MEMACCESS(0)
"st1 {v4.d}[0], [%0], %6 \n"
MEMACCESS(0)
"st1 {v6.d}[0], [%0] \n"
"mov %0, %3 \n"
MEMACCESS(0)
"st1 {v5.d}[0], [%0], %7 \n"
MEMACCESS(0)
"st1 {v7.d}[0], [%0] \n"
"add %1, %1, #4 \n" // src += 2 * 2
"add %2, %2, %6, lsl #1 \n" // dst_a += 2 * dst_stride_a
"add %3, %3, %7, lsl #1 \n" // dst_b += 2 * dst_stride_b
"subs %4, %4, #2 \n" // w -= 2
"subs %w4, %w4, #2 \n" // w -= 2
"b.eq 4f \n"
// 1x8 block
"3: \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[0], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[1], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[2], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[3], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[4], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[5], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[6], [%1], %5 \n"
MEMACCESS(1)
"ld2 {v0.b, v1.b}[7], [%1] \n"
MEMACCESS(2)
"st1 {v0.d}[0], [%2] \n"
MEMACCESS(3)
"st1 {v1.d}[0], [%3] \n"
"4: \n"
@ -524,7 +405,7 @@ void TransposeUVWx8_NEON(const uint8* src, int src_stride,
"+r"(src), // %1
"+r"(dst_a), // %2
"+r"(dst_b), // %3
"+r"(width64) // %4
"+r"(width) // %4
: "r"(static_cast<ptrdiff_t>(src_stride)), // %5
"r"(static_cast<ptrdiff_t>(dst_stride_a)), // %6
"r"(static_cast<ptrdiff_t>(dst_stride_b)), // %7

43
third_party/yuv/source/rotate_win.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#include "libyuv/rotate_row.h"
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
@ -19,15 +19,17 @@ extern "C" {
// This module is for 32 bit Visual C x86 and clangcl
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86)
__declspec(naked)
void TransposeWx8_SSSE3(const uint8* src, int src_stride,
uint8* dst, int dst_stride, int width) {
__declspec(naked) void TransposeWx8_SSSE3(const uint8* src,
int src_stride,
uint8* dst,
int dst_stride,
int width) {
__asm {
push edi
push esi
push ebp
mov eax, [esp + 12 + 4] // src
mov edi, [esp + 12 + 8] // src_stride
mov eax, [esp + 12 + 4] // src
mov edi, [esp + 12 + 8] // src_stride
mov edx, [esp + 12 + 12] // dst
mov esi, [esp + 12 + 16] // dst_stride
mov ecx, [esp + 12 + 20] // width
@ -110,18 +112,20 @@ void TransposeWx8_SSSE3(const uint8* src, int src_stride,
}
}
__declspec(naked)
void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
uint8* dst_a, int dst_stride_a,
uint8* dst_b, int dst_stride_b,
int w) {
__declspec(naked) void TransposeUVWx8_SSE2(const uint8* src,
int src_stride,
uint8* dst_a,
int dst_stride_a,
uint8* dst_b,
int dst_stride_b,
int w) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 16 + 4] // src
mov edi, [esp + 16 + 8] // src_stride
mov eax, [esp + 16 + 4] // src
mov edi, [esp + 16 + 8] // src_stride
mov edx, [esp + 16 + 12] // dst_a
mov esi, [esp + 16 + 16] // dst_stride_a
mov ebx, [esp + 16 + 20] // dst_b
@ -133,9 +137,9 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
mov ecx, [ecx + 16 + 28] // w
align 4
convertloop:
// Read in the data from the source pointer.
// First round of bit swap.
convertloop:
movdqu xmm0, [eax]
movdqu xmm1, [eax + edi]
lea eax, [eax + 2 * edi]
@ -162,7 +166,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea eax, [eax + 2 * edi]
movdqu [esp], xmm5 // backup xmm5
neg edi
movdqa xmm5, xmm6 // use xmm5 as temp register.
movdqa xmm5, xmm6 // use xmm5 as temp register.
punpcklbw xmm6, xmm7
punpckhbw xmm5, xmm7
movdqa xmm7, xmm5
@ -183,10 +187,11 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
movdqa xmm6, xmm5
movdqu xmm5, [esp] // restore xmm5
movdqu [esp], xmm6 // backup xmm6
movdqa xmm6, xmm5 // use xmm6 as temp register.
movdqa xmm6, xmm5 // use xmm6 as temp register.
punpcklwd xmm5, xmm7
punpckhwd xmm6, xmm7
movdqa xmm7, xmm6
// Third round of bit swap.
// Write to the destination pointer.
movdqa xmm6, xmm0
@ -200,7 +205,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm4
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm2 // use xmm0 as the temp register.
movdqa xmm0, xmm2 // use xmm0 as the temp register.
punpckldq xmm2, xmm6
movlpd qword ptr [edx], xmm2
movhpd qword ptr [ebx], xmm2
@ -209,7 +214,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm1 // use xmm0 as the temp register.
movdqa xmm0, xmm1 // use xmm0 as the temp register.
punpckldq xmm1, xmm5
movlpd qword ptr [edx], xmm1
movhpd qword ptr [ebx], xmm1
@ -218,7 +223,7 @@ void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
lea edx, [edx + 2 * esi]
movhpd qword ptr [ebx + ebp], xmm0
lea ebx, [ebx + 2 * ebp]
movdqa xmm0, xmm3 // use xmm0 as the temp register.
movdqa xmm0, xmm3 // use xmm0 as the temp register.
punpckldq xmm3, xmm7
movlpd qword ptr [edx], xmm3
movhpd qword ptr [ebx], xmm3

768
third_party/yuv/source/row_any.cc vendored
View File

File diff suppressed because it is too large Load Diff

1106
third_party/yuv/source/row_common.cc vendored
View File

File diff suppressed because it is too large Load Diff

1721
third_party/yuv/source/row_dspr2.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

1823
third_party/yuv/source/row_gcc.cc vendored
View File

File diff suppressed because it is too large Load Diff

782
third_party/yuv/source/row_mips.cc vendored
View File

@ -1,782 +0,0 @@
/*
* Copyright (c) 2012 The LibYuv project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/row.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
// The following are available on Mips platforms:
#if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips__) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
#ifdef HAS_COPYROW_MIPS
void CopyRow_MIPS(const uint8* src, uint8* dst, int count) {
__asm__ __volatile__ (
".set noreorder \n"
".set noat \n"
"slti $at, %[count], 8 \n"
"bne $at ,$zero, $last8 \n"
"xor $t8, %[src], %[dst] \n"
"andi $t8, $t8, 0x3 \n"
"bne $t8, $zero, unaligned \n"
"negu $a3, %[dst] \n"
// make dst/src aligned
"andi $a3, $a3, 0x3 \n"
"beq $a3, $zero, $chk16w \n"
// word-aligned now count is the remining bytes count
"subu %[count], %[count], $a3 \n"
"lwr $t8, 0(%[src]) \n"
"addu %[src], %[src], $a3 \n"
"swr $t8, 0(%[dst]) \n"
"addu %[dst], %[dst], $a3 \n"
// Now the dst/src are mutually word-aligned with word-aligned addresses
"$chk16w: \n"
"andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
// t8 is the byte count after 64-byte chunks
"beq %[count], $t8, chk8w \n"
// There will be at most 1 32-byte chunk after it
"subu $a3, %[count], $t8 \n" // the reminder
// Here a3 counts bytes in 16w chunks
"addu $a3, %[dst], $a3 \n"
// Now a3 is the final dst after 64-byte chunks
"addu $t0, %[dst], %[count] \n"
// t0 is the "past the end" address
// When in the loop we exercise "pref 30,x(a1)", the a1+x should not be past
// the "t0-32" address
// This means: for x=128 the last "safe" a1 address is "t0-160"
// Alternatively, for x=64 the last "safe" a1 address is "t0-96"
// we will use "pref 30,128(a1)", so "t0-160" is the limit
"subu $t9, $t0, 160 \n"
// t9 is the "last safe pref 30,128(a1)" address
"pref 0, 0(%[src]) \n" // first line of src
"pref 0, 32(%[src]) \n" // second line of src
"pref 0, 64(%[src]) \n"
"pref 30, 32(%[dst]) \n"
// In case the a1 > t9 don't use "pref 30" at all
"sgtu $v1, %[dst], $t9 \n"
"bgtz $v1, $loop16w \n"
"nop \n"
// otherwise, start with using pref30
"pref 30, 64(%[dst]) \n"
"$loop16w: \n"
"pref 0, 96(%[src]) \n"
"lw $t0, 0(%[src]) \n"
"bgtz $v1, $skip_pref30_96 \n" // skip
"lw $t1, 4(%[src]) \n"
"pref 30, 96(%[dst]) \n" // continue
"$skip_pref30_96: \n"
"lw $t2, 8(%[src]) \n"
"lw $t3, 12(%[src]) \n"
"lw $t4, 16(%[src]) \n"
"lw $t5, 20(%[src]) \n"
"lw $t6, 24(%[src]) \n"
"lw $t7, 28(%[src]) \n"
"pref 0, 128(%[src]) \n"
// bring the next lines of src, addr 128
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"lw $t0, 32(%[src]) \n"
"bgtz $v1, $skip_pref30_128 \n" // skip pref 30,128(a1)
"lw $t1, 36(%[src]) \n"
"pref 30, 128(%[dst]) \n" // set dest, addr 128
"$skip_pref30_128: \n"
"lw $t2, 40(%[src]) \n"
"lw $t3, 44(%[src]) \n"
"lw $t4, 48(%[src]) \n"
"lw $t5, 52(%[src]) \n"
"lw $t6, 56(%[src]) \n"
"lw $t7, 60(%[src]) \n"
"pref 0, 160(%[src]) \n"
// bring the next lines of src, addr 160
"sw $t0, 32(%[dst]) \n"
"sw $t1, 36(%[dst]) \n"
"sw $t2, 40(%[dst]) \n"
"sw $t3, 44(%[dst]) \n"
"sw $t4, 48(%[dst]) \n"
"sw $t5, 52(%[dst]) \n"
"sw $t6, 56(%[dst]) \n"
"sw $t7, 60(%[dst]) \n"
"addiu %[dst], %[dst], 64 \n" // adding 64 to dest
"sgtu $v1, %[dst], $t9 \n"
"bne %[dst], $a3, $loop16w \n"
" addiu %[src], %[src], 64 \n" // adding 64 to src
"move %[count], $t8 \n"
// Here we have src and dest word-aligned but less than 64-bytes to go
"chk8w: \n"
"pref 0, 0x0(%[src]) \n"
"andi $t8, %[count], 0x1f \n" // 32-byte chunk?
// the t8 is the reminder count past 32-bytes
"beq %[count], $t8, chk1w \n"
// count=t8,no 32-byte chunk
" nop \n"
"lw $t0, 0(%[src]) \n"
"lw $t1, 4(%[src]) \n"
"lw $t2, 8(%[src]) \n"
"lw $t3, 12(%[src]) \n"
"lw $t4, 16(%[src]) \n"
"lw $t5, 20(%[src]) \n"
"lw $t6, 24(%[src]) \n"
"lw $t7, 28(%[src]) \n"
"addiu %[src], %[src], 32 \n"
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"addiu %[dst], %[dst], 32 \n"
"chk1w: \n"
"andi %[count], $t8, 0x3 \n"
// now count is the reminder past 1w chunks
"beq %[count], $t8, $last8 \n"
" subu $a3, $t8, %[count] \n"
// a3 is count of bytes in 1w chunks
"addu $a3, %[dst], $a3 \n"
// now a3 is the dst address past the 1w chunks
// copying in words (4-byte chunks)
"$wordCopy_loop: \n"
"lw $t3, 0(%[src]) \n"
// the first t3 may be equal t0 ... optimize?
"addiu %[src], %[src],4 \n"
"addiu %[dst], %[dst],4 \n"
"bne %[dst], $a3,$wordCopy_loop \n"
" sw $t3, -4(%[dst]) \n"
// For the last (<8) bytes
"$last8: \n"
"blez %[count], leave \n"
" addu $a3, %[dst], %[count] \n" // a3 -last dst address
"$last8loop: \n"
"lb $v1, 0(%[src]) \n"
"addiu %[src], %[src], 1 \n"
"addiu %[dst], %[dst], 1 \n"
"bne %[dst], $a3, $last8loop \n"
" sb $v1, -1(%[dst]) \n"
"leave: \n"
" j $ra \n"
" nop \n"
//
// UNALIGNED case
//
"unaligned: \n"
// got here with a3="negu a1"
"andi $a3, $a3, 0x3 \n" // a1 is word aligned?
"beqz $a3, $ua_chk16w \n"
" subu %[count], %[count], $a3 \n"
// bytes left after initial a3 bytes
"lwr $v1, 0(%[src]) \n"
"lwl $v1, 3(%[src]) \n"
"addu %[src], %[src], $a3 \n" // a3 may be 1, 2 or 3
"swr $v1, 0(%[dst]) \n"
"addu %[dst], %[dst], $a3 \n"
// below the dst will be word aligned (NOTE1)
"$ua_chk16w: \n"
"andi $t8, %[count], 0x3f \n" // whole 64-B chunks?
// t8 is the byte count after 64-byte chunks
"beq %[count], $t8, ua_chk8w \n"
// if a2==t8, no 64-byte chunks
// There will be at most 1 32-byte chunk after it
"subu $a3, %[count], $t8 \n" // the reminder
// Here a3 counts bytes in 16w chunks
"addu $a3, %[dst], $a3 \n"
// Now a3 is the final dst after 64-byte chunks
"addu $t0, %[dst], %[count] \n" // t0 "past the end"
"subu $t9, $t0, 160 \n"
// t9 is the "last safe pref 30,128(a1)" address
"pref 0, 0(%[src]) \n" // first line of src
"pref 0, 32(%[src]) \n" // second line addr 32
"pref 0, 64(%[src]) \n"
"pref 30, 32(%[dst]) \n"
// safe, as we have at least 64 bytes ahead
// In case the a1 > t9 don't use "pref 30" at all
"sgtu $v1, %[dst], $t9 \n"
"bgtz $v1, $ua_loop16w \n"
// skip "pref 30,64(a1)" for too short arrays
" nop \n"
// otherwise, start with using pref30
"pref 30, 64(%[dst]) \n"
"$ua_loop16w: \n"
"pref 0, 96(%[src]) \n"
"lwr $t0, 0(%[src]) \n"
"lwl $t0, 3(%[src]) \n"
"lwr $t1, 4(%[src]) \n"
"bgtz $v1, $ua_skip_pref30_96 \n"
" lwl $t1, 7(%[src]) \n"
"pref 30, 96(%[dst]) \n"
// continue setting up the dest, addr 96
"$ua_skip_pref30_96: \n"
"lwr $t2, 8(%[src]) \n"
"lwl $t2, 11(%[src]) \n"
"lwr $t3, 12(%[src]) \n"
"lwl $t3, 15(%[src]) \n"
"lwr $t4, 16(%[src]) \n"
"lwl $t4, 19(%[src]) \n"
"lwr $t5, 20(%[src]) \n"
"lwl $t5, 23(%[src]) \n"
"lwr $t6, 24(%[src]) \n"
"lwl $t6, 27(%[src]) \n"
"lwr $t7, 28(%[src]) \n"
"lwl $t7, 31(%[src]) \n"
"pref 0, 128(%[src]) \n"
// bring the next lines of src, addr 128
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"lwr $t0, 32(%[src]) \n"
"lwl $t0, 35(%[src]) \n"
"lwr $t1, 36(%[src]) \n"
"bgtz $v1, ua_skip_pref30_128 \n"
" lwl $t1, 39(%[src]) \n"
"pref 30, 128(%[dst]) \n"
// continue setting up the dest, addr 128
"ua_skip_pref30_128: \n"
"lwr $t2, 40(%[src]) \n"
"lwl $t2, 43(%[src]) \n"
"lwr $t3, 44(%[src]) \n"
"lwl $t3, 47(%[src]) \n"
"lwr $t4, 48(%[src]) \n"
"lwl $t4, 51(%[src]) \n"
"lwr $t5, 52(%[src]) \n"
"lwl $t5, 55(%[src]) \n"
"lwr $t6, 56(%[src]) \n"
"lwl $t6, 59(%[src]) \n"
"lwr $t7, 60(%[src]) \n"
"lwl $t7, 63(%[src]) \n"
"pref 0, 160(%[src]) \n"
// bring the next lines of src, addr 160
"sw $t0, 32(%[dst]) \n"
"sw $t1, 36(%[dst]) \n"
"sw $t2, 40(%[dst]) \n"
"sw $t3, 44(%[dst]) \n"
"sw $t4, 48(%[dst]) \n"
"sw $t5, 52(%[dst]) \n"
"sw $t6, 56(%[dst]) \n"
"sw $t7, 60(%[dst]) \n"
"addiu %[dst],%[dst],64 \n" // adding 64 to dest
"sgtu $v1,%[dst],$t9 \n"
"bne %[dst],$a3,$ua_loop16w \n"
" addiu %[src],%[src],64 \n" // adding 64 to src
"move %[count],$t8 \n"
// Here we have src and dest word-aligned but less than 64-bytes to go
"ua_chk8w: \n"
"pref 0, 0x0(%[src]) \n"
"andi $t8, %[count], 0x1f \n" // 32-byte chunk?
// the t8 is the reminder count
"beq %[count], $t8, $ua_chk1w \n"
// when count==t8, no 32-byte chunk
"lwr $t0, 0(%[src]) \n"
"lwl $t0, 3(%[src]) \n"
"lwr $t1, 4(%[src]) \n"
"lwl $t1, 7(%[src]) \n"
"lwr $t2, 8(%[src]) \n"
"lwl $t2, 11(%[src]) \n"
"lwr $t3, 12(%[src]) \n"
"lwl $t3, 15(%[src]) \n"
"lwr $t4, 16(%[src]) \n"
"lwl $t4, 19(%[src]) \n"
"lwr $t5, 20(%[src]) \n"
"lwl $t5, 23(%[src]) \n"
"lwr $t6, 24(%[src]) \n"
"lwl $t6, 27(%[src]) \n"
"lwr $t7, 28(%[src]) \n"
"lwl $t7, 31(%[src]) \n"
"addiu %[src], %[src], 32 \n"
"sw $t0, 0(%[dst]) \n"
"sw $t1, 4(%[dst]) \n"
"sw $t2, 8(%[dst]) \n"
"sw $t3, 12(%[dst]) \n"
"sw $t4, 16(%[dst]) \n"
"sw $t5, 20(%[dst]) \n"
"sw $t6, 24(%[dst]) \n"
"sw $t7, 28(%[dst]) \n"
"addiu %[dst], %[dst], 32 \n"
"$ua_chk1w: \n"
"andi %[count], $t8, 0x3 \n"
// now count is the reminder past 1w chunks
"beq %[count], $t8, ua_smallCopy \n"
"subu $a3, $t8, %[count] \n"
// a3 is count of bytes in 1w chunks
"addu $a3, %[dst], $a3 \n"
// now a3 is the dst address past the 1w chunks
// copying in words (4-byte chunks)
"$ua_wordCopy_loop: \n"
"lwr $v1, 0(%[src]) \n"
"lwl $v1, 3(%[src]) \n"
"addiu %[src], %[src], 4 \n"
"addiu %[dst], %[dst], 4 \n"
// note: dst=a1 is word aligned here, see NOTE1
"bne %[dst], $a3, $ua_wordCopy_loop \n"
" sw $v1,-4(%[dst]) \n"
// Now less than 4 bytes (value in count) left to copy
"ua_smallCopy: \n"
"beqz %[count], leave \n"
" addu $a3, %[dst], %[count] \n" // a3 = last dst address
"$ua_smallCopy_loop: \n"
"lb $v1, 0(%[src]) \n"
"addiu %[src], %[src], 1 \n"
"addiu %[dst], %[dst], 1 \n"
"bne %[dst],$a3,$ua_smallCopy_loop \n"
" sb $v1, -1(%[dst]) \n"
"j $ra \n"
" nop \n"
".set at \n"
".set reorder \n"
: [dst] "+r" (dst), [src] "+r" (src)
: [count] "r" (count)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7",
"t8", "t9", "a3", "v1", "at"
);
}
#endif // HAS_COPYROW_MIPS
// DSPR2 functions
#if !defined(LIBYUV_DISABLE_MIPS) && defined(__mips_dsp) && \
(__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32) && (__mips_isa_rev < 6)
void SplitUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t4, %[width], 4 \n" // multiplies of 16
"blez $t4, 2f \n"
" andi %[width], %[width], 0xf \n" // residual
"1: \n"
"addiu $t4, $t4, -1 \n"
"lw $t0, 0(%[src_uv]) \n" // V1 | U1 | V0 | U0
"lw $t1, 4(%[src_uv]) \n" // V3 | U3 | V2 | U2
"lw $t2, 8(%[src_uv]) \n" // V5 | U5 | V4 | U4
"lw $t3, 12(%[src_uv]) \n" // V7 | U7 | V6 | U6
"lw $t5, 16(%[src_uv]) \n" // V9 | U9 | V8 | U8
"lw $t6, 20(%[src_uv]) \n" // V11 | U11 | V10 | U10
"lw $t7, 24(%[src_uv]) \n" // V13 | U13 | V12 | U12
"lw $t8, 28(%[src_uv]) \n" // V15 | U15 | V14 | U14
"addiu %[src_uv], %[src_uv], 32 \n"
"precrq.qb.ph $t9, $t1, $t0 \n" // V3 | V2 | V1 | V0
"precr.qb.ph $t0, $t1, $t0 \n" // U3 | U2 | U1 | U0
"precrq.qb.ph $t1, $t3, $t2 \n" // V7 | V6 | V5 | V4
"precr.qb.ph $t2, $t3, $t2 \n" // U7 | U6 | U5 | U4
"precrq.qb.ph $t3, $t6, $t5 \n" // V11 | V10 | V9 | V8
"precr.qb.ph $t5, $t6, $t5 \n" // U11 | U10 | U9 | U8
"precrq.qb.ph $t6, $t8, $t7 \n" // V15 | V14 | V13 | V12
"precr.qb.ph $t7, $t8, $t7 \n" // U15 | U14 | U13 | U12
"sw $t9, 0(%[dst_v]) \n"
"sw $t0, 0(%[dst_u]) \n"
"sw $t1, 4(%[dst_v]) \n"
"sw $t2, 4(%[dst_u]) \n"
"sw $t3, 8(%[dst_v]) \n"
"sw $t5, 8(%[dst_u]) \n"
"sw $t6, 12(%[dst_v]) \n"
"sw $t7, 12(%[dst_u]) \n"
"addiu %[dst_v], %[dst_v], 16 \n"
"bgtz $t4, 1b \n"
" addiu %[dst_u], %[dst_u], 16 \n"
"beqz %[width], 3f \n"
" nop \n"
"2: \n"
"lbu $t0, 0(%[src_uv]) \n"
"lbu $t1, 1(%[src_uv]) \n"
"addiu %[src_uv], %[src_uv], 2 \n"
"addiu %[width], %[width], -1 \n"
"sb $t0, 0(%[dst_u]) \n"
"sb $t1, 0(%[dst_v]) \n"
"addiu %[dst_u], %[dst_u], 1 \n"
"bgtz %[width], 2b \n"
" addiu %[dst_v], %[dst_v], 1 \n"
"3: \n"
".set pop \n"
: [src_uv] "+r" (src_uv),
[width] "+r" (width),
[dst_u] "+r" (dst_u),
[dst_v] "+r" (dst_v)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6", "t7", "t8", "t9"
);
}
void MirrorRow_DSPR2(const uint8* src, uint8* dst, int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"srl $t4, %[width], 4 \n" // multiplies of 16
"andi $t5, %[width], 0xf \n"
"blez $t4, 2f \n"
" addu %[src], %[src], %[width] \n" // src += width
"1: \n"
"lw $t0, -16(%[src]) \n" // |3|2|1|0|
"lw $t1, -12(%[src]) \n" // |7|6|5|4|
"lw $t2, -8(%[src]) \n" // |11|10|9|8|
"lw $t3, -4(%[src]) \n" // |15|14|13|12|
"wsbh $t0, $t0 \n" // |2|3|0|1|
"wsbh $t1, $t1 \n" // |6|7|4|5|
"wsbh $t2, $t2 \n" // |10|11|8|9|
"wsbh $t3, $t3 \n" // |14|15|12|13|
"rotr $t0, $t0, 16 \n" // |0|1|2|3|
"rotr $t1, $t1, 16 \n" // |4|5|6|7|
"rotr $t2, $t2, 16 \n" // |8|9|10|11|
"rotr $t3, $t3, 16 \n" // |12|13|14|15|
"addiu %[src], %[src], -16 \n"
"addiu $t4, $t4, -1 \n"
"sw $t3, 0(%[dst]) \n" // |15|14|13|12|
"sw $t2, 4(%[dst]) \n" // |11|10|9|8|
"sw $t1, 8(%[dst]) \n" // |7|6|5|4|
"sw $t0, 12(%[dst]) \n" // |3|2|1|0|
"bgtz $t4, 1b \n"
" addiu %[dst], %[dst], 16 \n"
"beqz $t5, 3f \n"
" nop \n"
"2: \n"
"lbu $t0, -1(%[src]) \n"
"addiu $t5, $t5, -1 \n"
"addiu %[src], %[src], -1 \n"
"sb $t0, 0(%[dst]) \n"
"bgez $t5, 2b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
".set pop \n"
: [src] "+r" (src), [dst] "+r" (dst)
: [width] "r" (width)
: "t0", "t1", "t2", "t3", "t4", "t5"
);
}
void MirrorUVRow_DSPR2(const uint8* src_uv, uint8* dst_u, uint8* dst_v,
int width) {
int x;
int y;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"addu $t4, %[width], %[width] \n"
"srl %[x], %[width], 4 \n"
"andi %[y], %[width], 0xf \n"
"blez %[x], 2f \n"
" addu %[src_uv], %[src_uv], $t4 \n"
"1: \n"
"lw $t0, -32(%[src_uv]) \n" // |3|2|1|0|
"lw $t1, -28(%[src_uv]) \n" // |7|6|5|4|
"lw $t2, -24(%[src_uv]) \n" // |11|10|9|8|
"lw $t3, -20(%[src_uv]) \n" // |15|14|13|12|
"lw $t4, -16(%[src_uv]) \n" // |19|18|17|16|
"lw $t6, -12(%[src_uv]) \n" // |23|22|21|20|
"lw $t7, -8(%[src_uv]) \n" // |27|26|25|24|
"lw $t8, -4(%[src_uv]) \n" // |31|30|29|28|
"rotr $t0, $t0, 16 \n" // |1|0|3|2|
"rotr $t1, $t1, 16 \n" // |5|4|7|6|
"rotr $t2, $t2, 16 \n" // |9|8|11|10|
"rotr $t3, $t3, 16 \n" // |13|12|15|14|
"rotr $t4, $t4, 16 \n" // |17|16|19|18|
"rotr $t6, $t6, 16 \n" // |21|20|23|22|
"rotr $t7, $t7, 16 \n" // |25|24|27|26|
"rotr $t8, $t8, 16 \n" // |29|28|31|30|
"precr.qb.ph $t9, $t0, $t1 \n" // |0|2|4|6|
"precrq.qb.ph $t5, $t0, $t1 \n" // |1|3|5|7|
"precr.qb.ph $t0, $t2, $t3 \n" // |8|10|12|14|
"precrq.qb.ph $t1, $t2, $t3 \n" // |9|11|13|15|
"precr.qb.ph $t2, $t4, $t6 \n" // |16|18|20|22|
"precrq.qb.ph $t3, $t4, $t6 \n" // |17|19|21|23|
"precr.qb.ph $t4, $t7, $t8 \n" // |24|26|28|30|
"precrq.qb.ph $t6, $t7, $t8 \n" // |25|27|29|31|
"addiu %[src_uv], %[src_uv], -32 \n"
"addiu %[x], %[x], -1 \n"
"swr $t4, 0(%[dst_u]) \n"
"swl $t4, 3(%[dst_u]) \n" // |30|28|26|24|
"swr $t6, 0(%[dst_v]) \n"
"swl $t6, 3(%[dst_v]) \n" // |31|29|27|25|
"swr $t2, 4(%[dst_u]) \n"
"swl $t2, 7(%[dst_u]) \n" // |22|20|18|16|
"swr $t3, 4(%[dst_v]) \n"
"swl $t3, 7(%[dst_v]) \n" // |23|21|19|17|
"swr $t0, 8(%[dst_u]) \n"
"swl $t0, 11(%[dst_u]) \n" // |14|12|10|8|
"swr $t1, 8(%[dst_v]) \n"
"swl $t1, 11(%[dst_v]) \n" // |15|13|11|9|
"swr $t9, 12(%[dst_u]) \n"
"swl $t9, 15(%[dst_u]) \n" // |6|4|2|0|
"swr $t5, 12(%[dst_v]) \n"
"swl $t5, 15(%[dst_v]) \n" // |7|5|3|1|
"addiu %[dst_v], %[dst_v], 16 \n"
"bgtz %[x], 1b \n"
" addiu %[dst_u], %[dst_u], 16 \n"
"beqz %[y], 3f \n"
" nop \n"
"b 2f \n"
" nop \n"
"2: \n"
"lbu $t0, -2(%[src_uv]) \n"
"lbu $t1, -1(%[src_uv]) \n"
"addiu %[src_uv], %[src_uv], -2 \n"
"addiu %[y], %[y], -1 \n"
"sb $t0, 0(%[dst_u]) \n"
"sb $t1, 0(%[dst_v]) \n"
"addiu %[dst_u], %[dst_u], 1 \n"
"bgtz %[y], 2b \n"
" addiu %[dst_v], %[dst_v], 1 \n"
"3: \n"
".set pop \n"
: [src_uv] "+r" (src_uv),
[dst_u] "+r" (dst_u),
[dst_v] "+r" (dst_v),
[x] "=&r" (x),
[y] "=&r" (y)
: [width] "r" (width)
: "t0", "t1", "t2", "t3", "t4",
"t5", "t7", "t8", "t9"
);
}
// Convert (4 Y and 2 VU) I422 and arrange RGB values into
// t5 = | 0 | B0 | 0 | b0 |
// t4 = | 0 | B1 | 0 | b1 |
// t9 = | 0 | G0 | 0 | g0 |
// t8 = | 0 | G1 | 0 | g1 |
// t2 = | 0 | R0 | 0 | r0 |
// t1 = | 0 | R1 | 0 | r1 |
#define YUVTORGB \
"lw $t0, 0(%[y_buf]) \n" \
"lhu $t1, 0(%[u_buf]) \n" \
"lhu $t2, 0(%[v_buf]) \n" \
"preceu.ph.qbr $t1, $t1 \n" \
"preceu.ph.qbr $t2, $t2 \n" \
"preceu.ph.qbra $t3, $t0 \n" \
"preceu.ph.qbla $t0, $t0 \n" \
"subu.ph $t1, $t1, $s5 \n" \
"subu.ph $t2, $t2, $s5 \n" \
"subu.ph $t3, $t3, $s4 \n" \
"subu.ph $t0, $t0, $s4 \n" \
"mul.ph $t3, $t3, $s0 \n" \
"mul.ph $t0, $t0, $s0 \n" \
"shll.ph $t4, $t1, 0x7 \n" \
"subu.ph $t4, $t4, $t1 \n" \
"mul.ph $t6, $t1, $s1 \n" \
"mul.ph $t1, $t2, $s2 \n" \
"addq_s.ph $t5, $t4, $t3 \n" \
"addq_s.ph $t4, $t4, $t0 \n" \
"shra.ph $t5, $t5, 6 \n" \
"shra.ph $t4, $t4, 6 \n" \
"addiu %[u_buf], 2 \n" \
"addiu %[v_buf], 2 \n" \
"addu.ph $t6, $t6, $t1 \n" \
"mul.ph $t1, $t2, $s3 \n" \
"addu.ph $t9, $t6, $t3 \n" \
"addu.ph $t8, $t6, $t0 \n" \
"shra.ph $t9, $t9, 6 \n" \
"shra.ph $t8, $t8, 6 \n" \
"addu.ph $t2, $t1, $t3 \n" \
"addu.ph $t1, $t1, $t0 \n" \
"shra.ph $t2, $t2, 6 \n" \
"shra.ph $t1, $t1, 6 \n" \
"subu.ph $t5, $t5, $s5 \n" \
"subu.ph $t4, $t4, $s5 \n" \
"subu.ph $t9, $t9, $s5 \n" \
"subu.ph $t8, $t8, $s5 \n" \
"subu.ph $t2, $t2, $s5 \n" \
"subu.ph $t1, $t1, $s5 \n" \
"shll_s.ph $t5, $t5, 8 \n" \
"shll_s.ph $t4, $t4, 8 \n" \
"shll_s.ph $t9, $t9, 8 \n" \
"shll_s.ph $t8, $t8, 8 \n" \
"shll_s.ph $t2, $t2, 8 \n" \
"shll_s.ph $t1, $t1, 8 \n" \
"shra.ph $t5, $t5, 8 \n" \
"shra.ph $t4, $t4, 8 \n" \
"shra.ph $t9, $t9, 8 \n" \
"shra.ph $t8, $t8, 8 \n" \
"shra.ph $t2, $t2, 8 \n" \
"shra.ph $t1, $t1, 8 \n" \
"addu.ph $t5, $t5, $s5 \n" \
"addu.ph $t4, $t4, $s5 \n" \
"addu.ph $t9, $t9, $s5 \n" \
"addu.ph $t8, $t8, $s5 \n" \
"addu.ph $t2, $t2, $s5 \n" \
"addu.ph $t1, $t1, $s5 \n"
// TODO(fbarchard): accept yuv conversion constants.
void I422ToARGBRow_DSPR2(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
const struct YuvConstants* yuvconstants,
int width) {
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"beqz %[width], 2f \n"
" repl.ph $s0, 74 \n" // |YG|YG| = |74|74|
"repl.ph $s1, -25 \n" // |UG|UG| = |-25|-25|
"repl.ph $s2, -52 \n" // |VG|VG| = |-52|-52|
"repl.ph $s3, 102 \n" // |VR|VR| = |102|102|
"repl.ph $s4, 16 \n" // |0|16|0|16|
"repl.ph $s5, 128 \n" // |128|128| // clipping
"lui $s6, 0xff00 \n"
"ori $s6, 0xff00 \n" // |ff|00|ff|00|ff|
"1: \n"
YUVTORGB
// Arranging into argb format
"precr.qb.ph $t4, $t8, $t4 \n" // |G1|g1|B1|b1|
"precr.qb.ph $t5, $t9, $t5 \n" // |G0|g0|B0|b0|
"addiu %[width], -4 \n"
"precrq.qb.ph $t8, $t4, $t5 \n" // |G1|B1|G0|B0|
"precr.qb.ph $t9, $t4, $t5 \n" // |g1|b1|g0|b0|
"precr.qb.ph $t2, $t1, $t2 \n" // |R1|r1|R0|r0|
"addiu %[y_buf], 4 \n"
"preceu.ph.qbla $t1, $t2 \n" // |0 |R1|0 |R0|
"preceu.ph.qbra $t2, $t2 \n" // |0 |r1|0 |r0|
"or $t1, $t1, $s6 \n" // |ff|R1|ff|R0|
"or $t2, $t2, $s6 \n" // |ff|r1|ff|r0|
"precrq.ph.w $t0, $t2, $t9 \n" // |ff|r1|g1|b1|
"precrq.ph.w $t3, $t1, $t8 \n" // |ff|R1|G1|B1|
"sll $t9, $t9, 16 \n"
"sll $t8, $t8, 16 \n"
"packrl.ph $t2, $t2, $t9 \n" // |ff|r0|g0|b0|
"packrl.ph $t1, $t1, $t8 \n" // |ff|R0|G0|B0|
// Store results.
"sw $t2, 0(%[rgb_buf]) \n"
"sw $t0, 4(%[rgb_buf]) \n"
"sw $t1, 8(%[rgb_buf]) \n"
"sw $t3, 12(%[rgb_buf]) \n"
"bnez %[width], 1b \n"
" addiu %[rgb_buf], 16 \n"
"2: \n"
".set pop \n"
:[y_buf] "+r" (y_buf),
[u_buf] "+r" (u_buf),
[v_buf] "+r" (v_buf),
[width] "+r" (width),
[rgb_buf] "+r" (rgb_buf)
:
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9",
"s0", "s1", "s2", "s3",
"s4", "s5", "s6"
);
}
// Bilinear filter 8x2 -> 8x1
void InterpolateRow_DSPR2(uint8* dst_ptr, const uint8* src_ptr,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) {
int y0_fraction = 256 - source_y_fraction;
const uint8* src_ptr1 = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
"replv.ph $t0, %[y0_fraction] \n"
"replv.ph $t1, %[source_y_fraction] \n"
"1: \n"
"lw $t2, 0(%[src_ptr]) \n"
"lw $t3, 0(%[src_ptr1]) \n"
"lw $t4, 4(%[src_ptr]) \n"
"lw $t5, 4(%[src_ptr1]) \n"
"muleu_s.ph.qbl $t6, $t2, $t0 \n"
"muleu_s.ph.qbr $t7, $t2, $t0 \n"
"muleu_s.ph.qbl $t8, $t3, $t1 \n"
"muleu_s.ph.qbr $t9, $t3, $t1 \n"
"muleu_s.ph.qbl $t2, $t4, $t0 \n"
"muleu_s.ph.qbr $t3, $t4, $t0 \n"
"muleu_s.ph.qbl $t4, $t5, $t1 \n"
"muleu_s.ph.qbr $t5, $t5, $t1 \n"
"addq.ph $t6, $t6, $t8 \n"
"addq.ph $t7, $t7, $t9 \n"
"addq.ph $t2, $t2, $t4 \n"
"addq.ph $t3, $t3, $t5 \n"
"shra.ph $t6, $t6, 8 \n"
"shra.ph $t7, $t7, 8 \n"
"shra.ph $t2, $t2, 8 \n"
"shra.ph $t3, $t3, 8 \n"
"precr.qb.ph $t6, $t6, $t7 \n"
"precr.qb.ph $t2, $t2, $t3 \n"
"addiu %[src_ptr], %[src_ptr], 8 \n"
"addiu %[src_ptr1], %[src_ptr1], 8 \n"
"addiu %[dst_width], %[dst_width], -8 \n"
"sw $t6, 0(%[dst_ptr]) \n"
"sw $t2, 4(%[dst_ptr]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[dst_ptr], %[dst_ptr], 8 \n"
".set pop \n"
: [dst_ptr] "+r" (dst_ptr),
[src_ptr1] "+r" (src_ptr1),
[src_ptr] "+r" (src_ptr),
[dst_width] "+r" (dst_width)
: [source_y_fraction] "r" (source_y_fraction),
[y0_fraction] "r" (y0_fraction),
[src_stride] "r" (src_stride)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
}
#endif // __mips_dsp_rev >= 2
#endif // defined(__mips__)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

2977
third_party/yuv/source/row_msa.cc vendored Normal file
View File

File diff suppressed because it is too large Load Diff

4035
third_party/yuv/source/row_neon.cc vendored
View File

File diff suppressed because it is too large Load Diff

3679
third_party/yuv/source/row_neon64.cc vendored
View File

File diff suppressed because it is too large Load Diff

3708
third_party/yuv/source/row_win.cc vendored
View File

File diff suppressed because it is too large Load Diff

781
third_party/yuv/source/scale.cc vendored
View File

File diff suppressed because it is too large Load Diff

465
third_party/yuv/source/scale_any.cc vendored
View File

@ -19,16 +19,16 @@ extern "C" {
#endif
// Definition for ScaleFilterCols, ScaleARGBCols and ScaleARGBFilterCols
#define CANY(NAMEANY, TERP_SIMD, TERP_C, BPP, MASK) \
void NAMEANY(uint8* dst_ptr, const uint8* src_ptr, \
int dst_width, int x, int dx) { \
int n = dst_width & ~MASK; \
if (n > 0) { \
TERP_SIMD(dst_ptr, src_ptr, n, x, dx); \
} \
TERP_C(dst_ptr + n * BPP, src_ptr, \
dst_width & MASK, x + n * dx, dx); \
}
#define CANY(NAMEANY, TERP_SIMD, TERP_C, BPP, MASK) \
void NAMEANY(uint8* dst_ptr, const uint8* src_ptr, int dst_width, int x, \
int dx) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
TERP_SIMD(dst_ptr, src_ptr, n, x, dx); \
} \
TERP_C(dst_ptr + n * BPP, src_ptr, r, x + n * dx, dx); \
}
#ifdef HAS_SCALEFILTERCOLS_NEON
CANY(ScaleFilterCols_Any_NEON, ScaleFilterCols_NEON, ScaleFilterCols_C, 1, 7)
@ -37,167 +37,379 @@ CANY(ScaleFilterCols_Any_NEON, ScaleFilterCols_NEON, ScaleFilterCols_C, 1, 7)
CANY(ScaleARGBCols_Any_NEON, ScaleARGBCols_NEON, ScaleARGBCols_C, 4, 7)
#endif
#ifdef HAS_SCALEARGBFILTERCOLS_NEON
CANY(ScaleARGBFilterCols_Any_NEON, ScaleARGBFilterCols_NEON,
ScaleARGBFilterCols_C, 4, 3)
CANY(ScaleARGBFilterCols_Any_NEON,
ScaleARGBFilterCols_NEON,
ScaleARGBFilterCols_C,
4,
3)
#endif
#undef CANY
// Fixed scale down.
#define SDANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
// Mask may be non-power of 2, so use MOD
#define SDANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, uint8* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); /* NOLINT */ \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
// Fixed scale down for odd source width. Used by I420Blend subsampling.
// Since dst_width is (width + 1) / 2, this function scales one less pixel
// and copies the last pixel.
#define SDODD(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)(dst_width - 1) % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r); \
}
#define SDODD(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, FACTOR, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, uint8* dst_ptr, \
int dst_width) { \
int r = (int)((unsigned int)(dst_width - 1) % (MASK + 1)); /* NOLINT */ \
int n = (dst_width - 1) - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * FACTOR) * BPP, src_stride, \
dst_ptr + n * BPP, r + 1); \
}
#ifdef HAS_SCALEROWDOWN2_SSSE3
SDANY(ScaleRowDown2_Any_SSSE3, ScaleRowDown2_SSSE3, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_SSSE3, ScaleRowDown2Linear_SSSE3,
ScaleRowDown2Linear_C, 2, 1, 15)
SDANY(ScaleRowDown2Box_Any_SSSE3, ScaleRowDown2Box_SSSE3, ScaleRowDown2Box_C,
2, 1, 15)
SDODD(ScaleRowDown2Box_Odd_SSSE3, ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_Odd_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_SSSE3,
ScaleRowDown2Linear_SSSE3,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_SSSE3,
ScaleRowDown2Box_SSSE3,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN2_AVX2
SDANY(ScaleRowDown2_Any_AVX2, ScaleRowDown2_AVX2, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_AVX2, ScaleRowDown2Linear_AVX2,
ScaleRowDown2Linear_C, 2, 1, 31)
SDANY(ScaleRowDown2Box_Any_AVX2, ScaleRowDown2Box_AVX2, ScaleRowDown2Box_C,
2, 1, 31)
SDODD(ScaleRowDown2Box_Odd_AVX2, ScaleRowDown2Box_AVX2, ScaleRowDown2Box_Odd_C,
2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_AVX2,
ScaleRowDown2Linear_AVX2,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_C,
2,
1,
31)
SDODD(ScaleRowDown2Box_Odd_AVX2,
ScaleRowDown2Box_AVX2,
ScaleRowDown2Box_Odd_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN2_NEON
SDANY(ScaleRowDown2_Any_NEON, ScaleRowDown2_NEON, ScaleRowDown2_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_NEON, ScaleRowDown2Linear_NEON,
ScaleRowDown2Linear_C, 2, 1, 15)
SDANY(ScaleRowDown2Box_Any_NEON, ScaleRowDown2Box_NEON,
ScaleRowDown2Box_C, 2, 1, 15)
SDODD(ScaleRowDown2Box_Odd_NEON, ScaleRowDown2Box_NEON,
ScaleRowDown2Box_Odd_C, 2, 1, 15)
SDANY(ScaleRowDown2Linear_Any_NEON,
ScaleRowDown2Linear_NEON,
ScaleRowDown2Linear_C,
2,
1,
15)
SDANY(ScaleRowDown2Box_Any_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_C,
2,
1,
15)
SDODD(ScaleRowDown2Box_Odd_NEON,
ScaleRowDown2Box_NEON,
ScaleRowDown2Box_Odd_C,
2,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN2_MSA
SDANY(ScaleRowDown2_Any_MSA, ScaleRowDown2_MSA, ScaleRowDown2_C, 2, 1, 31)
SDANY(ScaleRowDown2Linear_Any_MSA,
ScaleRowDown2Linear_MSA,
ScaleRowDown2Linear_C,
2,
1,
31)
SDANY(ScaleRowDown2Box_Any_MSA,
ScaleRowDown2Box_MSA,
ScaleRowDown2Box_C,
2,
1,
31)
#endif
#ifdef HAS_SCALEROWDOWN4_SSSE3
SDANY(ScaleRowDown4_Any_SSSE3, ScaleRowDown4_SSSE3, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_SSSE3, ScaleRowDown4Box_SSSE3, ScaleRowDown4Box_C,
4, 1, 7)
SDANY(ScaleRowDown4Box_Any_SSSE3,
ScaleRowDown4Box_SSSE3,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_AVX2
SDANY(ScaleRowDown4_Any_AVX2, ScaleRowDown4_AVX2, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_AVX2, ScaleRowDown4Box_AVX2, ScaleRowDown4Box_C,
4, 1, 15)
SDANY(ScaleRowDown4Box_Any_AVX2,
ScaleRowDown4Box_AVX2,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN4_NEON
SDANY(ScaleRowDown4_Any_NEON, ScaleRowDown4_NEON, ScaleRowDown4_C, 4, 1, 7)
SDANY(ScaleRowDown4Box_Any_NEON, ScaleRowDown4Box_NEON, ScaleRowDown4Box_C,
4, 1, 7)
SDANY(ScaleRowDown4Box_Any_NEON,
ScaleRowDown4Box_NEON,
ScaleRowDown4Box_C,
4,
1,
7)
#endif
#ifdef HAS_SCALEROWDOWN4_MSA
SDANY(ScaleRowDown4_Any_MSA, ScaleRowDown4_MSA, ScaleRowDown4_C, 4, 1, 15)
SDANY(ScaleRowDown4Box_Any_MSA,
ScaleRowDown4Box_MSA,
ScaleRowDown4Box_C,
4,
1,
15)
#endif
#ifdef HAS_SCALEROWDOWN34_SSSE3
SDANY(ScaleRowDown34_Any_SSSE3, ScaleRowDown34_SSSE3,
ScaleRowDown34_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_0_Box_Any_SSSE3, ScaleRowDown34_0_Box_SSSE3,
ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_1_Box_Any_SSSE3, ScaleRowDown34_1_Box_SSSE3,
ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_Any_SSSE3,
ScaleRowDown34_SSSE3,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_SSSE3,
ScaleRowDown34_0_Box_SSSE3,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_SSSE3,
ScaleRowDown34_1_Box_SSSE3,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#ifdef HAS_SCALEROWDOWN34_NEON
SDANY(ScaleRowDown34_Any_NEON, ScaleRowDown34_NEON,
ScaleRowDown34_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_0_Box_Any_NEON, ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_1_Box_Any_NEON, ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C, 4 / 3, 1, 23)
SDANY(ScaleRowDown34_Any_NEON,
ScaleRowDown34_NEON,
ScaleRowDown34_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_0_Box_Any_NEON,
ScaleRowDown34_0_Box_NEON,
ScaleRowDown34_0_Box_C,
4 / 3,
1,
23)
SDANY(ScaleRowDown34_1_Box_Any_NEON,
ScaleRowDown34_1_Box_NEON,
ScaleRowDown34_1_Box_C,
4 / 3,
1,
23)
#endif
#ifdef HAS_SCALEROWDOWN38_SSSE3
SDANY(ScaleRowDown38_Any_SSSE3, ScaleRowDown38_SSSE3,
ScaleRowDown38_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_3_Box_Any_SSSE3, ScaleRowDown38_3_Box_SSSE3,
ScaleRowDown38_3_Box_C, 8 / 3, 1, 5)
SDANY(ScaleRowDown38_2_Box_Any_SSSE3, ScaleRowDown38_2_Box_SSSE3,
ScaleRowDown38_2_Box_C, 8 / 3, 1, 5)
SDANY(ScaleRowDown38_Any_SSSE3,
ScaleRowDown38_SSSE3,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_SSSE3,
ScaleRowDown38_3_Box_SSSE3,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
5)
SDANY(ScaleRowDown38_2_Box_Any_SSSE3,
ScaleRowDown38_2_Box_SSSE3,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
5)
#endif
#ifdef HAS_SCALEROWDOWN38_NEON
SDANY(ScaleRowDown38_Any_NEON, ScaleRowDown38_NEON,
ScaleRowDown38_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_3_Box_Any_NEON, ScaleRowDown38_3_Box_NEON,
ScaleRowDown38_3_Box_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_2_Box_Any_NEON, ScaleRowDown38_2_Box_NEON,
ScaleRowDown38_2_Box_C, 8 / 3, 1, 11)
SDANY(ScaleRowDown38_Any_NEON,
ScaleRowDown38_NEON,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_NEON,
ScaleRowDown38_3_Box_NEON,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_NEON,
ScaleRowDown38_2_Box_NEON,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEROWDOWN38_MSA
SDANY(ScaleRowDown38_Any_MSA,
ScaleRowDown38_MSA,
ScaleRowDown38_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_3_Box_Any_MSA,
ScaleRowDown38_3_Box_MSA,
ScaleRowDown38_3_Box_C,
8 / 3,
1,
11)
SDANY(ScaleRowDown38_2_Box_Any_MSA,
ScaleRowDown38_2_Box_MSA,
ScaleRowDown38_2_Box_C,
8 / 3,
1,
11)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_SSE2
SDANY(ScaleARGBRowDown2_Any_SSE2, ScaleARGBRowDown2_SSE2,
ScaleARGBRowDown2_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2Linear_Any_SSE2, ScaleARGBRowDown2Linear_SSE2,
ScaleARGBRowDown2Linear_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2Box_Any_SSE2, ScaleARGBRowDown2Box_SSE2,
ScaleARGBRowDown2Box_C, 2, 4, 3)
SDANY(ScaleARGBRowDown2_Any_SSE2,
ScaleARGBRowDown2_SSE2,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_SSE2,
ScaleARGBRowDown2Linear_SSE2,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_SSE2,
ScaleARGBRowDown2Box_SSE2,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_NEON
SDANY(ScaleARGBRowDown2_Any_NEON, ScaleARGBRowDown2_NEON,
ScaleARGBRowDown2_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2Linear_Any_NEON, ScaleARGBRowDown2Linear_NEON,
ScaleARGBRowDown2Linear_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2Box_Any_NEON, ScaleARGBRowDown2Box_NEON,
ScaleARGBRowDown2Box_C, 2, 4, 7)
SDANY(ScaleARGBRowDown2_Any_NEON,
ScaleARGBRowDown2_NEON,
ScaleARGBRowDown2_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Linear_Any_NEON,
ScaleARGBRowDown2Linear_NEON,
ScaleARGBRowDown2Linear_C,
2,
4,
7)
SDANY(ScaleARGBRowDown2Box_Any_NEON,
ScaleARGBRowDown2Box_NEON,
ScaleARGBRowDown2Box_C,
2,
4,
7)
#endif
#ifdef HAS_SCALEARGBROWDOWN2_MSA
SDANY(ScaleARGBRowDown2_Any_MSA,
ScaleARGBRowDown2_MSA,
ScaleARGBRowDown2_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Linear_Any_MSA,
ScaleARGBRowDown2Linear_MSA,
ScaleARGBRowDown2Linear_C,
2,
4,
3)
SDANY(ScaleARGBRowDown2Box_Any_MSA,
ScaleARGBRowDown2Box_MSA,
ScaleARGBRowDown2Box_C,
2,
4,
3)
#endif
#undef SDANY
// Scale down by even scale factor.
#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, int src_stepx, \
uint8* dst_ptr, int dst_width) { \
int r = (int)((unsigned int)dst_width % (MASK + 1)); \
int n = dst_width - r; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride, \
src_stepx, dst_ptr + n * BPP, r); \
}
#define SDAANY(NAMEANY, SCALEROWDOWN_SIMD, SCALEROWDOWN_C, BPP, MASK) \
void NAMEANY(const uint8* src_ptr, ptrdiff_t src_stride, int src_stepx, \
uint8* dst_ptr, int dst_width) { \
int r = dst_width & MASK; \
int n = dst_width & ~MASK; \
if (n > 0) { \
SCALEROWDOWN_SIMD(src_ptr, src_stride, src_stepx, dst_ptr, n); \
} \
SCALEROWDOWN_C(src_ptr + (n * src_stepx) * BPP, src_stride, src_stepx, \
dst_ptr + n * BPP, r); \
}
#ifdef HAS_SCALEARGBROWDOWNEVEN_SSE2
SDAANY(ScaleARGBRowDownEven_Any_SSE2, ScaleARGBRowDownEven_SSE2,
ScaleARGBRowDownEven_C, 4, 3)
SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2, ScaleARGBRowDownEvenBox_SSE2,
ScaleARGBRowDownEvenBox_C, 4, 3)
SDAANY(ScaleARGBRowDownEven_Any_SSE2,
ScaleARGBRowDownEven_SSE2,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_SSE2,
ScaleARGBRowDownEvenBox_SSE2,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_NEON
SDAANY(ScaleARGBRowDownEven_Any_NEON, ScaleARGBRowDownEven_NEON,
ScaleARGBRowDownEven_C, 4, 3)
SDAANY(ScaleARGBRowDownEvenBox_Any_NEON, ScaleARGBRowDownEvenBox_NEON,
ScaleARGBRowDownEvenBox_C, 4, 3)
SDAANY(ScaleARGBRowDownEven_Any_NEON,
ScaleARGBRowDownEven_NEON,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_NEON,
ScaleARGBRowDownEvenBox_NEON,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
#ifdef HAS_SCALEARGBROWDOWNEVEN_MSA
SDAANY(ScaleARGBRowDownEven_Any_MSA,
ScaleARGBRowDownEven_MSA,
ScaleARGBRowDownEven_C,
4,
3)
SDAANY(ScaleARGBRowDownEvenBox_Any_MSA,
ScaleARGBRowDownEvenBox_MSA,
ScaleARGBRowDownEvenBox_C,
4,
3)
#endif
// Add rows box filter scale down.
#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK) \
void NAMEANY(const uint8* src_ptr, uint16* dst_ptr, int src_width) { \
int n = src_width & ~MASK; \
if (n > 0) { \
SCALEADDROW_SIMD(src_ptr, dst_ptr, n); \
} \
SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK); \
}
#define SAANY(NAMEANY, SCALEADDROW_SIMD, SCALEADDROW_C, MASK) \
void NAMEANY(const uint8* src_ptr, uint16* dst_ptr, int src_width) { \
int n = src_width & ~MASK; \
if (n > 0) { \
SCALEADDROW_SIMD(src_ptr, dst_ptr, n); \
} \
SCALEADDROW_C(src_ptr + n, dst_ptr + n, src_width & MASK); \
}
#ifdef HAS_SCALEADDROW_SSE2
SAANY(ScaleAddRow_Any_SSE2, ScaleAddRow_SSE2, ScaleAddRow_C, 15)
@ -208,14 +420,15 @@ SAANY(ScaleAddRow_Any_AVX2, ScaleAddRow_AVX2, ScaleAddRow_C, 31)
#ifdef HAS_SCALEADDROW_NEON
SAANY(ScaleAddRow_Any_NEON, ScaleAddRow_NEON, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_MSA
SAANY(ScaleAddRow_Any_MSA, ScaleAddRow_MSA, ScaleAddRow_C, 15)
#endif
#ifdef HAS_SCALEADDROW_DSPR2
SAANY(ScaleAddRow_Any_DSPR2, ScaleAddRow_DSPR2, ScaleAddRow_C, 15)
#endif
#undef SAANY
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif

481
third_party/yuv/source/scale_argb.cc vendored
View File

@ -30,20 +30,31 @@ static __inline int Abs(int v) {
// ScaleARGB ARGB, 1/2
// This is an optimized version for scaling down a ARGB to 1/2 of
// its original size.
static void ScaleARGBDown2(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBDown2(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
int row_stride = src_stride * (dy >> 16);
void (*ScaleARGBRowDown2)(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) =
filtering == kFilterNone ? ScaleARGBRowDown2_C :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_C :
ScaleARGBRowDown2Box_C);
assert(dx == 65536 * 2); // Test scale factor of 2.
filtering == kFilterNone
? ScaleARGBRowDown2_C
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_C
: ScaleARGBRowDown2Box_C);
(void)src_width;
(void)src_height;
(void)dx;
assert(dx == 65536 * 2); // Test scale factor of 2.
assert((dy & 0x1ffff) == 0); // Test vertical scale is multiple of 2.
// Advance to odd row, even column.
if (filtering == kFilterBilinear) {
@ -54,25 +65,49 @@ static void ScaleARGBDown2(int src_width, int src_height,
#if defined(HAS_SCALEARGBROWDOWN2_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_Any_SSE2 :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_SSE2 :
ScaleARGBRowDown2Box_Any_SSE2);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_SSE2
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_SSE2
: ScaleARGBRowDown2Box_Any_SSE2);
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_SSE2 :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2 :
ScaleARGBRowDown2Box_SSE2);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_SSE2
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_SSE2
: ScaleARGBRowDown2Box_SSE2);
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_Any_NEON :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_NEON :
ScaleARGBRowDown2Box_Any_NEON);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_NEON
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_NEON
: ScaleARGBRowDown2Box_Any_NEON);
if (IS_ALIGNED(dst_width, 8)) {
ScaleARGBRowDown2 = filtering == kFilterNone ? ScaleARGBRowDown2_NEON :
(filtering == kFilterLinear ? ScaleARGBRowDown2Linear_NEON :
ScaleARGBRowDown2Box_NEON);
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_NEON
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_NEON
: ScaleARGBRowDown2Box_NEON);
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWN2_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_Any_MSA
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_Any_MSA
: ScaleARGBRowDown2Box_Any_MSA);
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDown2 =
filtering == kFilterNone
? ScaleARGBRowDown2_MSA
: (filtering == kFilterLinear ? ScaleARGBRowDown2Linear_MSA
: ScaleARGBRowDown2Box_MSA);
}
}
#endif
@ -90,21 +125,32 @@ static void ScaleARGBDown2(int src_width, int src_height,
// ScaleARGB ARGB, 1/4
// This is an optimized version for scaling down a ARGB to 1/4 of
// its original size.
static void ScaleARGBDown4Box(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy) {
static void ScaleARGBDown4Box(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy) {
int j;
// Allocate 2 rows of ARGB.
const int kRowSize = (dst_width * 2 * 4 + 31) & ~31;
align_buffer_64(row, kRowSize * 2);
int row_stride = src_stride * (dy >> 16);
void (*ScaleARGBRowDown2)(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) = ScaleARGBRowDown2Box_C;
uint8* dst_argb, int dst_width) =
ScaleARGBRowDown2Box_C;
// Advance to odd row, even column.
src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
assert(dx == 65536 * 4); // Test scale factor of 4.
(void)src_width;
(void)src_height;
(void)dx;
assert(dx == 65536 * 4); // Test scale factor of 4.
assert((dy & 0x3ffff) == 0); // Test vertical scale is multiple of 4.
#if defined(HAS_SCALEARGBROWDOWN2_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
@ -125,8 +171,8 @@ static void ScaleARGBDown4Box(int src_width, int src_height,
for (j = 0; j < dst_height; ++j) {
ScaleARGBRowDown2(src_argb, src_stride, row, dst_width * 2);
ScaleARGBRowDown2(src_argb + src_stride * 2, src_stride,
row + kRowSize, dst_width * 2);
ScaleARGBRowDown2(src_argb + src_stride * 2, src_stride, row + kRowSize,
dst_width * 2);
ScaleARGBRowDown2(row, kRowSize, dst_argb, dst_width);
src_argb += row_stride;
dst_argb += dst_stride;
@ -137,11 +183,18 @@ static void ScaleARGBDown4Box(int src_width, int src_height,
// ScaleARGB ARGB Even
// This is an optimized version for scaling down a ARGB to even
// multiple of its original size.
static void ScaleARGBDownEven(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBDownEven(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
int col_step = dx >> 16;
@ -149,26 +202,38 @@ static void ScaleARGBDownEven(int src_width, int src_height,
void (*ScaleARGBRowDownEven)(const uint8* src_argb, ptrdiff_t src_stride,
int src_step, uint8* dst_argb, int dst_width) =
filtering ? ScaleARGBRowDownEvenBox_C : ScaleARGBRowDownEven_C;
(void)src_width;
(void)src_height;
assert(IS_ALIGNED(src_width, 2));
assert(IS_ALIGNED(src_height, 2));
src_argb += (y >> 16) * src_stride + (x >> 16) * 4;
#if defined(HAS_SCALEARGBROWDOWNEVEN_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_SSE2 :
ScaleARGBRowDownEven_Any_SSE2;
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_SSE2
: ScaleARGBRowDownEven_Any_SSE2;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_SSE2 :
ScaleARGBRowDownEven_SSE2;
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_SSE2 : ScaleARGBRowDownEven_SSE2;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_NEON :
ScaleARGBRowDownEven_Any_NEON;
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_NEON
: ScaleARGBRowDownEven_Any_NEON;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_NEON :
ScaleARGBRowDownEven_NEON;
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_NEON : ScaleARGBRowDownEven_NEON;
}
}
#endif
#if defined(HAS_SCALEARGBROWDOWNEVEN_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
ScaleARGBRowDownEven = filtering ? ScaleARGBRowDownEvenBox_Any_MSA
: ScaleARGBRowDownEven_Any_MSA;
if (IS_ALIGNED(dst_width, 4)) {
ScaleARGBRowDownEven =
filtering ? ScaleARGBRowDownEvenBox_MSA : ScaleARGBRowDownEven_MSA;
}
}
#endif
@ -184,25 +249,32 @@ static void ScaleARGBDownEven(int src_width, int src_height,
}
// Scale ARGB down with bilinear interpolation.
static void ScaleARGBBilinearDown(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBBilinearDown(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*InterpolateRow)(uint8 * dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8 * dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
(src_width >= 32768) ? ScaleARGBFilterCols64_C : ScaleARGBFilterCols_C;
int64 xlast = x + (int64)(dst_width - 1) * dx;
int64 xl = (dx >= 0) ? x : xlast;
int64 xr = (dx >= 0) ? xlast : x;
int clip_src_width;
xl = (xl >> 16) & ~3; // Left edge aligned.
xr = (xr >> 16) + 1; // Right most pixel used. Bilinear uses 2 pixels.
xl = (xl >> 16) & ~3; // Left edge aligned.
xr = (xr >> 16) + 1; // Right most pixel used. Bilinear uses 2 pixels.
xr = (xr + 1 + 3) & ~3; // 1 beyond 4 pixel aligned right most pixel.
if (xr > src_width) {
xr = src_width;
@ -235,14 +307,22 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) &&
IS_ALIGNED(src_stride, 4)) {
InterpolateRow = InterpolateRow_Any_DSPR2;
if (IS_ALIGNED(clip_src_width, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
}
}
#endif
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(clip_src_width, 32)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
ScaleARGBFilterCols = ScaleARGBFilterCols_SSSE3;
@ -286,18 +366,25 @@ static void ScaleARGBBilinearDown(int src_width, int src_height,
}
// Scale ARGB up with bilinear interpolation.
static void ScaleARGBBilinearUp(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy,
static void ScaleARGBBilinearUp(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*InterpolateRow)(uint8 * dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
void (*ScaleARGBFilterCols)(uint8 * dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
filtering ? ScaleARGBFilterCols_C : ScaleARGBCols_C;
const int max_y = (src_height - 1) << 16;
#if defined(HAS_INTERPOLATEROW_SSSE3)
@ -325,14 +412,22 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(dst_argb, 4) &&
IS_ALIGNED(dst_stride, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
}
#endif
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
if (src_width >= 32768) {
ScaleARGBFilterCols = filtering ?
ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
ScaleARGBFilterCols =
filtering ? ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
}
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
if (filtering && TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
@ -423,8 +518,10 @@ static void ScaleARGBBilinearUp(int src_width, int src_height,
#ifdef YUVSCALEUP
// Scale YUV to ARGB up with bilinear interpolation.
static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
int dst_width, int dst_height,
static void ScaleYUVToARGBBilinearUp(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride_y,
int src_stride_u,
int src_stride_v,
@ -433,14 +530,15 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb,
int x, int dx, int y, int dy,
int x,
int dx,
int y,
int dy,
enum FilterMode filtering) {
int j;
void (*I422ToARGBRow)(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
int width) = I422ToARGBRow_C;
void (*I422ToARGBRow)(const uint8* y_buf, const uint8* u_buf,
const uint8* v_buf, uint8* rgb_buf, int width) =
I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
@ -474,10 +572,18 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
I422ToARGBRow = I422ToARGBRow_DSPR2;
}
#endif
#if defined(HAS_I422TOARGBROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
I422ToARGBRow = I422ToARGBRow_Any_MSA;
if (IS_ALIGNED(src_width, 8)) {
I422ToARGBRow = I422ToARGBRow_MSA;
}
}
#endif
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*InterpolateRow)(uint8 * dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
#if defined(HAS_INTERPOLATEROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3)) {
InterpolateRow = InterpolateRow_Any_SSSE3;
@ -503,18 +609,26 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride_argb, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(dst_argb, 4) &&
IS_ALIGNED(dst_stride_argb, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
}
#endif
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(dst_width, 8)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
void (*ScaleARGBFilterCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*ScaleARGBFilterCols)(uint8 * dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
filtering ? ScaleARGBFilterCols_C : ScaleARGBCols_C;
if (src_width >= 32768) {
ScaleARGBFilterCols = filtering ?
ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
ScaleARGBFilterCols =
filtering ? ScaleARGBFilterCols64_C : ScaleARGBCols64_C;
}
#if defined(HAS_SCALEARGBFILTERCOLS_SSSE3)
if (filtering && TestCpuFlag(kCpuHasSSSE3) && src_width < 32768) {
@ -635,15 +749,23 @@ static void ScaleYUVToARGBBilinearUp(int src_width, int src_height,
// of x and dx is the integer part of the source position and
// the lower 16 bits are the fixed decimal part.
static void ScaleARGBSimple(int src_width, int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int dx, int y, int dy) {
static void ScaleARGBSimple(int src_width,
int src_height,
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int dx,
int y,
int dy) {
int j;
void (*ScaleARGBCols)(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) =
void (*ScaleARGBCols)(uint8 * dst_argb, const uint8* src_argb, int dst_width,
int x, int dx) =
(src_width >= 32768) ? ScaleARGBCols64_C : ScaleARGBCols_C;
(void)src_height;
#if defined(HAS_SCALEARGBCOLS_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && src_width < 32768) {
ScaleARGBCols = ScaleARGBCols_SSE2;
@ -667,8 +789,8 @@ static void ScaleARGBSimple(int src_width, int src_height,
}
for (j = 0; j < dst_height; ++j) {
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * src_stride,
dst_width, x, dx);
ScaleARGBCols(dst_argb, src_argb + (y >> 16) * src_stride, dst_width, x,
dx);
dst_argb += dst_stride;
y += dy;
}
@ -677,11 +799,18 @@ static void ScaleARGBSimple(int src_width, int src_height,
// ScaleARGB a ARGB.
// This function in turn calls a scaling function
// suitable for handling the desired resolutions.
static void ScaleARGB(const uint8* src, int src_stride,
int src_width, int src_height,
uint8* dst, int dst_stride,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
static void ScaleARGB(const uint8* src,
int src_stride,
int src_width,
int src_height,
uint8* dst,
int dst_stride,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
// Initial source x/y coordinate and step values as 16.16 fixed point.
int x = 0;
@ -690,8 +819,7 @@ static void ScaleARGB(const uint8* src, int src_stride,
int dy = 0;
// ARGB does not support box filter yet, but allow the user to pass it.
// Simplify filtering when possible.
filtering = ScaleFilterReduce(src_width, src_height,
dst_width, dst_height,
filtering = ScaleFilterReduce(src_width, src_height, dst_width, dst_height,
filtering);
// Negative src_height means invert the image.
@ -700,17 +828,17 @@ static void ScaleARGB(const uint8* src, int src_stride,
src = src + (src_height - 1) * src_stride;
src_stride = -src_stride;
}
ScaleSlope(src_width, src_height, dst_width, dst_height, filtering,
&x, &y, &dx, &dy);
ScaleSlope(src_width, src_height, dst_width, dst_height, filtering, &x, &y,
&dx, &dy);
src_width = Abs(src_width);
if (clip_x) {
int64 clipf = (int64)(clip_x) * dx;
int64 clipf = (int64)(clip_x)*dx;
x += (clipf & 0xffff);
src += (clipf >> 16) * 4;
dst += clip_x * 4;
}
if (clip_y) {
int64 clipf = (int64)(clip_y) * dy;
int64 clipf = (int64)(clip_y)*dy;
y += (clipf & 0xffff);
src += (clipf >> 16) * src_stride;
dst += clip_y * dst_stride;
@ -725,24 +853,20 @@ static void ScaleARGB(const uint8* src, int src_stride,
if (!(dx & 0x10000) && !(dy & 0x10000)) {
if (dx == 0x20000) {
// Optimized 1/2 downsample.
ScaleARGBDown2(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBDown2(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
if (dx == 0x40000 && filtering == kFilterBox) {
// Optimized 1/4 box downsample.
ScaleARGBDown4Box(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy);
ScaleARGBDown4Box(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy);
return;
}
ScaleARGBDownEven(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBDownEven(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
// Optimized odd scale down. ie 3, 5, 7, 9x.
@ -759,96 +883,105 @@ static void ScaleARGB(const uint8* src, int src_stride,
}
if (dx == 0x10000 && (x & 0xffff) == 0) {
// Arbitrary scale vertically, but unscaled vertically.
ScalePlaneVertical(src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, y, dy, 4, filtering);
ScalePlaneVertical(src_height, clip_width, clip_height, src_stride,
dst_stride, src, dst, x, y, dy, 4, filtering);
return;
}
if (filtering && dy < 65536) {
ScaleARGBBilinearUp(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBBilinearUp(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
if (filtering) {
ScaleARGBBilinearDown(src_width, src_height,
clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy, filtering);
ScaleARGBBilinearDown(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst, x, dx, y, dy,
filtering);
return;
}
ScaleARGBSimple(src_width, src_height, clip_width, clip_height,
src_stride, dst_stride, src, dst,
x, dx, y, dy);
ScaleARGBSimple(src_width, src_height, clip_width, clip_height, src_stride,
dst_stride, src, dst, x, dx, y, dy);
}
LIBYUV_API
int ARGBScaleClip(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int ARGBScaleClip(const uint8* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
if (!src_argb || src_width == 0 || src_height == 0 ||
!dst_argb || dst_width <= 0 || dst_height <= 0 ||
clip_x < 0 || clip_y < 0 ||
if (!src_argb || src_width == 0 || src_height == 0 || !dst_argb ||
dst_width <= 0 || dst_height <= 0 || clip_x < 0 || clip_y < 0 ||
clip_width > 32768 || clip_height > 32768 ||
(clip_x + clip_width) > dst_width ||
(clip_y + clip_height) > dst_height) {
return -1;
}
ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
dst_argb, dst_stride_argb, dst_width, dst_height,
clip_x, clip_y, clip_width, clip_height, filtering);
ScaleARGB(src_argb, src_stride_argb, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, clip_x, clip_y, clip_width,
clip_height, filtering);
return 0;
}
// Scale an ARGB image.
LIBYUV_API
int ARGBScale(const uint8* src_argb, int src_stride_argb,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int dst_width, int dst_height,
int ARGBScale(const uint8* src_argb,
int src_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
int dst_width,
int dst_height,
enum FilterMode filtering) {
if (!src_argb || src_width == 0 || src_height == 0 ||
src_width > 32768 || src_height > 32768 ||
!dst_argb || dst_width <= 0 || dst_height <= 0) {
if (!src_argb || src_width == 0 || src_height == 0 || src_width > 32768 ||
src_height > 32768 || !dst_argb || dst_width <= 0 || dst_height <= 0) {
return -1;
}
ScaleARGB(src_argb, src_stride_argb, src_width, src_height,
dst_argb, dst_stride_argb, dst_width, dst_height,
0, 0, dst_width, dst_height, filtering);
ScaleARGB(src_argb, src_stride_argb, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, 0, 0, dst_width, dst_height,
filtering);
return 0;
}
// Scale with YUV conversion to ARGB and clipping.
LIBYUV_API
int YUVToARGBScaleClip(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v,
int YUVToARGBScaleClip(const uint8* src_y,
int src_stride_y,
const uint8* src_u,
int src_stride_u,
const uint8* src_v,
int src_stride_v,
uint32 src_fourcc,
int src_width, int src_height,
uint8* dst_argb, int dst_stride_argb,
int src_width,
int src_height,
uint8* dst_argb,
int dst_stride_argb,
uint32 dst_fourcc,
int dst_width, int dst_height,
int clip_x, int clip_y, int clip_width, int clip_height,
int dst_width,
int dst_height,
int clip_x,
int clip_y,
int clip_width,
int clip_height,
enum FilterMode filtering) {
uint8* argb_buffer = (uint8*)malloc(src_width * src_height * 4);
int r;
I420ToARGB(src_y, src_stride_y,
src_u, src_stride_u,
src_v, src_stride_v,
argb_buffer, src_width * 4,
src_width, src_height);
(void)src_fourcc; // TODO(fbarchard): implement and/or assert.
(void)dst_fourcc;
I420ToARGB(src_y, src_stride_y, src_u, src_stride_u, src_v, src_stride_v,
argb_buffer, src_width * 4, src_width, src_height);
r = ARGBScaleClip(argb_buffer, src_width * 4,
src_width, src_height,
dst_argb, dst_stride_argb,
dst_width, dst_height,
clip_x, clip_y, clip_width, clip_height,
filtering);
r = ARGBScaleClip(argb_buffer, src_width * 4, src_width, src_height, dst_argb,
dst_stride_argb, dst_width, dst_height, clip_x, clip_y,
clip_width, clip_height, filtering);
free(argb_buffer);
return r;
}

615
third_party/yuv/source/scale_common.cc vendored
View File

@ -28,9 +28,12 @@ static __inline int Abs(int v) {
}
// CPU agnostic row functions
void ScaleRowDown2_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src_ptr[1];
dst[1] = src_ptr[3];
@ -42,9 +45,12 @@ void ScaleRowDown2_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown2_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown2_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src_ptr[1];
dst[1] = src_ptr[3];
@ -56,10 +62,13 @@ void ScaleRowDown2_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown2Linear_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Linear_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
const uint8* s = src_ptr;
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = (s[0] + s[1] + 1) >> 1;
dst[1] = (s[2] + s[3] + 1) >> 1;
@ -71,10 +80,13 @@ void ScaleRowDown2Linear_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown2Linear_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown2Linear_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
const uint16* s = src_ptr;
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = (s[0] + s[1] + 1) >> 1;
dst[1] = (s[2] + s[3] + 1) >> 1;
@ -86,8 +98,10 @@ void ScaleRowDown2Linear_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown2Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
const uint8* s = src_ptr;
const uint8* t = src_ptr + src_stride;
int x;
@ -103,8 +117,10 @@ void ScaleRowDown2Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown2Box_Odd_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Box_Odd_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
const uint8* s = src_ptr;
const uint8* t = src_ptr + src_stride;
int x;
@ -125,8 +141,10 @@ void ScaleRowDown2Box_Odd_C(const uint8* src_ptr, ptrdiff_t src_stride,
dst[0] = (s[0] + t[0] + 1) >> 1;
}
void ScaleRowDown2Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown2Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
const uint16* s = src_ptr;
const uint16* t = src_ptr + src_stride;
int x;
@ -142,9 +160,12 @@ void ScaleRowDown2Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown4_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown4_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src_ptr[2];
dst[1] = src_ptr[6];
@ -156,9 +177,12 @@ void ScaleRowDown4_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown4_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown4_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src_ptr[2];
dst[1] = src_ptr[6];
@ -170,81 +194,88 @@ void ScaleRowDown4_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown4Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown4Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
intptr_t stride = src_stride;
int x;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[3] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride + 3] +
src_ptr[stride * 2 + 0] + src_ptr[stride * 2 + 1] +
src_ptr[stride * 2 + 2] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 3 + 0] + src_ptr[stride * 3 + 1] +
src_ptr[stride * 3 + 2] + src_ptr[stride * 3 + 3] +
8) >> 4;
src_ptr[stride + 0] + src_ptr[stride + 1] + src_ptr[stride + 2] +
src_ptr[stride + 3] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2] +
src_ptr[stride * 2 + 3] + src_ptr[stride * 3 + 0] +
src_ptr[stride * 3 + 1] + src_ptr[stride * 3 + 2] +
src_ptr[stride * 3 + 3] + 8) >>
4;
dst[1] = (src_ptr[4] + src_ptr[5] + src_ptr[6] + src_ptr[7] +
src_ptr[stride + 4] + src_ptr[stride + 5] +
src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7] +
src_ptr[stride * 3 + 4] + src_ptr[stride * 3 + 5] +
src_ptr[stride * 3 + 6] + src_ptr[stride * 3 + 7] +
8) >> 4;
src_ptr[stride + 4] + src_ptr[stride + 5] + src_ptr[stride + 6] +
src_ptr[stride + 7] + src_ptr[stride * 2 + 4] +
src_ptr[stride * 2 + 5] + src_ptr[stride * 2 + 6] +
src_ptr[stride * 2 + 7] + src_ptr[stride * 3 + 4] +
src_ptr[stride * 3 + 5] + src_ptr[stride * 3 + 6] +
src_ptr[stride * 3 + 7] + 8) >>
4;
dst += 2;
src_ptr += 8;
}
if (dst_width & 1) {
dst[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[3] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride + 3] +
src_ptr[stride * 2 + 0] + src_ptr[stride * 2 + 1] +
src_ptr[stride * 2 + 2] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 3 + 0] + src_ptr[stride * 3 + 1] +
src_ptr[stride * 3 + 2] + src_ptr[stride * 3 + 3] +
8) >> 4;
src_ptr[stride + 0] + src_ptr[stride + 1] + src_ptr[stride + 2] +
src_ptr[stride + 3] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2] +
src_ptr[stride * 2 + 3] + src_ptr[stride * 3 + 0] +
src_ptr[stride * 3 + 1] + src_ptr[stride * 3 + 2] +
src_ptr[stride * 3 + 3] + 8) >>
4;
}
}
void ScaleRowDown4Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown4Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
intptr_t stride = src_stride;
int x;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[3] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride + 3] +
src_ptr[stride * 2 + 0] + src_ptr[stride * 2 + 1] +
src_ptr[stride * 2 + 2] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 3 + 0] + src_ptr[stride * 3 + 1] +
src_ptr[stride * 3 + 2] + src_ptr[stride * 3 + 3] +
8) >> 4;
src_ptr[stride + 0] + src_ptr[stride + 1] + src_ptr[stride + 2] +
src_ptr[stride + 3] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2] +
src_ptr[stride * 2 + 3] + src_ptr[stride * 3 + 0] +
src_ptr[stride * 3 + 1] + src_ptr[stride * 3 + 2] +
src_ptr[stride * 3 + 3] + 8) >>
4;
dst[1] = (src_ptr[4] + src_ptr[5] + src_ptr[6] + src_ptr[7] +
src_ptr[stride + 4] + src_ptr[stride + 5] +
src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7] +
src_ptr[stride * 3 + 4] + src_ptr[stride * 3 + 5] +
src_ptr[stride * 3 + 6] + src_ptr[stride * 3 + 7] +
8) >> 4;
src_ptr[stride + 4] + src_ptr[stride + 5] + src_ptr[stride + 6] +
src_ptr[stride + 7] + src_ptr[stride * 2 + 4] +
src_ptr[stride * 2 + 5] + src_ptr[stride * 2 + 6] +
src_ptr[stride * 2 + 7] + src_ptr[stride * 3 + 4] +
src_ptr[stride * 3 + 5] + src_ptr[stride * 3 + 6] +
src_ptr[stride * 3 + 7] + 8) >>
4;
dst += 2;
src_ptr += 8;
}
if (dst_width & 1) {
dst[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[3] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride + 3] +
src_ptr[stride * 2 + 0] + src_ptr[stride * 2 + 1] +
src_ptr[stride * 2 + 2] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 3 + 0] + src_ptr[stride * 3 + 1] +
src_ptr[stride * 3 + 2] + src_ptr[stride * 3 + 3] +
8) >> 4;
src_ptr[stride + 0] + src_ptr[stride + 1] + src_ptr[stride + 2] +
src_ptr[stride + 3] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2] +
src_ptr[stride * 2 + 3] + src_ptr[stride * 3 + 0] +
src_ptr[stride * 3 + 1] + src_ptr[stride * 3 + 2] +
src_ptr[stride * 3 + 3] + 8) >>
4;
}
}
void ScaleRowDown34_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown34_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
int x;
(void)src_stride;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (x = 0; x < dst_width; x += 3) {
dst[0] = src_ptr[0];
@ -255,9 +286,12 @@ void ScaleRowDown34_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown34_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown34_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
int x;
(void)src_stride;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (x = 0; x < dst_width; x += 3) {
dst[0] = src_ptr[0];
@ -269,8 +303,10 @@ void ScaleRowDown34_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
// Filter rows 0 and 1 together, 3 : 1
void ScaleRowDown34_0_Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
void ScaleRowDown34_0_Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* d,
int dst_width) {
const uint8* s = src_ptr;
const uint8* t = src_ptr + src_stride;
int x;
@ -291,8 +327,10 @@ void ScaleRowDown34_0_Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown34_0_Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* d, int dst_width) {
void ScaleRowDown34_0_Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* d,
int dst_width) {
const uint16* s = src_ptr;
const uint16* t = src_ptr + src_stride;
int x;
@ -314,8 +352,10 @@ void ScaleRowDown34_0_Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
// Filter rows 1 and 2 together, 1 : 1
void ScaleRowDown34_1_Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
void ScaleRowDown34_1_Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* d,
int dst_width) {
const uint8* s = src_ptr;
const uint8* t = src_ptr + src_stride;
int x;
@ -336,8 +376,10 @@ void ScaleRowDown34_1_Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown34_1_Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* d, int dst_width) {
void ScaleRowDown34_1_Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* d,
int dst_width) {
const uint16* s = src_ptr;
const uint16* t = src_ptr + src_stride;
int x;
@ -359,8 +401,11 @@ void ScaleRowDown34_1_Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
}
// Scales a single row of pixels using point sampling.
void ScaleCols_C(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
void ScaleCols_C(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst_ptr[0] = src_ptr[x >> 16];
@ -374,8 +419,11 @@ void ScaleCols_C(uint8* dst_ptr, const uint8* src_ptr,
}
}
void ScaleCols_16_C(uint16* dst_ptr, const uint16* src_ptr,
int dst_width, int x, int dx) {
void ScaleCols_16_C(uint16* dst_ptr,
const uint16* src_ptr,
int dst_width,
int x,
int dx) {
int j;
for (j = 0; j < dst_width - 1; j += 2) {
dst_ptr[0] = src_ptr[x >> 16];
@ -390,9 +438,14 @@ void ScaleCols_16_C(uint16* dst_ptr, const uint16* src_ptr,
}
// Scales a single row of pixels up by 2x using point sampling.
void ScaleColsUp2_C(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
void ScaleColsUp2_C(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
int j;
(void)x;
(void)dx;
for (j = 0; j < dst_width - 1; j += 2) {
dst_ptr[1] = dst_ptr[0] = src_ptr[0];
src_ptr += 1;
@ -403,9 +456,14 @@ void ScaleColsUp2_C(uint8* dst_ptr, const uint8* src_ptr,
}
}
void ScaleColsUp2_16_C(uint16* dst_ptr, const uint16* src_ptr,
int dst_width, int x, int dx) {
void ScaleColsUp2_16_C(uint16* dst_ptr,
const uint16* src_ptr,
int dst_width,
int x,
int dx) {
int j;
(void)x;
(void)dx;
for (j = 0; j < dst_width - 1; j += 2) {
dst_ptr[1] = dst_ptr[0] = src_ptr[0];
src_ptr += 1;
@ -417,11 +475,20 @@ void ScaleColsUp2_16_C(uint16* dst_ptr, const uint16* src_ptr,
}
// (1-f)a + fb can be replaced with a + f(b-a)
#define BLENDER(a, b, f) (uint8)((int)(a) + \
((int)(f) * ((int)(b) - (int)(a)) >> 16))
#if defined(__arm__) || defined(__aarch64__)
#define BLENDER(a, b, f) \
(uint8)((int)(a) + ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
#else
// Intel uses 7 bit math with rounding.
#define BLENDER(a, b, f) \
(uint8)((int)(a) + (((int)((f) >> 9) * ((int)(b) - (int)(a)) + 0x40) >> 7))
#endif
void ScaleFilterCols_C(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
void ScaleFilterCols_C(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int xi = x >> 16;
@ -444,8 +511,11 @@ void ScaleFilterCols_C(uint8* dst_ptr, const uint8* src_ptr,
}
}
void ScaleFilterCols64_C(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x32, int dx) {
void ScaleFilterCols64_C(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x32,
int dx) {
int64 x = (int64)(x32);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
@ -470,11 +540,15 @@ void ScaleFilterCols64_C(uint8* dst_ptr, const uint8* src_ptr,
}
#undef BLENDER
#define BLENDER(a, b, f) (uint16)((int)(a) + \
((int)(f) * ((int)(b) - (int)(a)) >> 16))
// Same as 8 bit arm blender but return is cast to uint16
#define BLENDER(a, b, f) \
(uint16)((int)(a) + ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
void ScaleFilterCols_16_C(uint16* dst_ptr, const uint16* src_ptr,
int dst_width, int x, int dx) {
void ScaleFilterCols_16_C(uint16* dst_ptr,
const uint16* src_ptr,
int dst_width,
int x,
int dx) {
int j;
for (j = 0; j < dst_width - 1; j += 2) {
int xi = x >> 16;
@ -497,8 +571,11 @@ void ScaleFilterCols_16_C(uint16* dst_ptr, const uint16* src_ptr,
}
}
void ScaleFilterCols64_16_C(uint16* dst_ptr, const uint16* src_ptr,
int dst_width, int x32, int dx) {
void ScaleFilterCols64_16_C(uint16* dst_ptr,
const uint16* src_ptr,
int dst_width,
int x32,
int dx) {
int64 x = (int64)(x32);
int j;
for (j = 0; j < dst_width - 1; j += 2) {
@ -523,9 +600,12 @@ void ScaleFilterCols64_16_C(uint16* dst_ptr, const uint16* src_ptr,
}
#undef BLENDER
void ScaleRowDown38_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown38_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
int x;
(void)src_stride;
assert(dst_width % 3 == 0);
for (x = 0; x < dst_width; x += 3) {
dst[0] = src_ptr[0];
@ -536,9 +616,12 @@ void ScaleRowDown38_C(const uint8* src_ptr, ptrdiff_t src_stride,
}
}
void ScaleRowDown38_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst, int dst_width) {
void ScaleRowDown38_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst,
int dst_width) {
int x;
(void)src_stride;
assert(dst_width % 3 == 0);
for (x = 0; x < dst_width; x += 3) {
dst[0] = src_ptr[0];
@ -552,25 +635,29 @@ void ScaleRowDown38_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
// 8x3 -> 3x1
void ScaleRowDown38_3_Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
int i;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (i = 0; i < dst_width; i += 3) {
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2]) *
(65536 / 9) >> 16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] +
src_ptr[stride + 3] + src_ptr[stride + 4] +
src_ptr[stride + 5] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5]) *
(65536 / 9) >> 16;
dst_ptr[2] = (src_ptr[6] + src_ptr[7] +
src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7]) *
(65536 / 6) >> 16;
dst_ptr[0] =
(src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[stride + 0] +
src_ptr[stride + 1] + src_ptr[stride + 2] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2]) *
(65536 / 9) >>
16;
dst_ptr[1] =
(src_ptr[3] + src_ptr[4] + src_ptr[5] + src_ptr[stride + 3] +
src_ptr[stride + 4] + src_ptr[stride + 5] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5]) *
(65536 / 9) >>
16;
dst_ptr[2] =
(src_ptr[6] + src_ptr[7] + src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7]) *
(65536 / 6) >>
16;
src_ptr += 8;
dst_ptr += 3;
}
@ -578,66 +665,80 @@ void ScaleRowDown38_3_Box_C(const uint8* src_ptr,
void ScaleRowDown38_3_Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst_ptr, int dst_width) {
uint16* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
int i;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (i = 0; i < dst_width; i += 3) {
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2]) *
(65536 / 9) >> 16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] +
src_ptr[stride + 3] + src_ptr[stride + 4] +
src_ptr[stride + 5] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5]) *
(65536 / 9) >> 16;
dst_ptr[2] = (src_ptr[6] + src_ptr[7] +
src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7]) *
(65536 / 6) >> 16;
dst_ptr[0] =
(src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[stride + 0] +
src_ptr[stride + 1] + src_ptr[stride + 2] + src_ptr[stride * 2 + 0] +
src_ptr[stride * 2 + 1] + src_ptr[stride * 2 + 2]) *
(65536 / 9) >>
16;
dst_ptr[1] =
(src_ptr[3] + src_ptr[4] + src_ptr[5] + src_ptr[stride + 3] +
src_ptr[stride + 4] + src_ptr[stride + 5] + src_ptr[stride * 2 + 3] +
src_ptr[stride * 2 + 4] + src_ptr[stride * 2 + 5]) *
(65536 / 9) >>
16;
dst_ptr[2] =
(src_ptr[6] + src_ptr[7] + src_ptr[stride + 6] + src_ptr[stride + 7] +
src_ptr[stride * 2 + 6] + src_ptr[stride * 2 + 7]) *
(65536 / 6) >>
16;
src_ptr += 8;
dst_ptr += 3;
}
}
// 8x2 -> 3x1
void ScaleRowDown38_2_Box_C(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown38_2_Box_C(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
int i;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (i = 0; i < dst_width; i += 3) {
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2]) * (65536 / 6) >> 16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] +
src_ptr[stride + 3] + src_ptr[stride + 4] +
src_ptr[stride + 5]) * (65536 / 6) >> 16;
dst_ptr[2] = (src_ptr[6] + src_ptr[7] +
src_ptr[stride + 6] + src_ptr[stride + 7]) *
(65536 / 4) >> 16;
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[stride + 0] +
src_ptr[stride + 1] + src_ptr[stride + 2]) *
(65536 / 6) >>
16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] + src_ptr[stride + 3] +
src_ptr[stride + 4] + src_ptr[stride + 5]) *
(65536 / 6) >>
16;
dst_ptr[2] =
(src_ptr[6] + src_ptr[7] + src_ptr[stride + 6] + src_ptr[stride + 7]) *
(65536 / 4) >>
16;
src_ptr += 8;
dst_ptr += 3;
}
}
void ScaleRowDown38_2_Box_16_C(const uint16* src_ptr, ptrdiff_t src_stride,
uint16* dst_ptr, int dst_width) {
void ScaleRowDown38_2_Box_16_C(const uint16* src_ptr,
ptrdiff_t src_stride,
uint16* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
int i;
assert((dst_width % 3 == 0) && (dst_width > 0));
for (i = 0; i < dst_width; i += 3) {
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] +
src_ptr[stride + 0] + src_ptr[stride + 1] +
src_ptr[stride + 2]) * (65536 / 6) >> 16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] +
src_ptr[stride + 3] + src_ptr[stride + 4] +
src_ptr[stride + 5]) * (65536 / 6) >> 16;
dst_ptr[2] = (src_ptr[6] + src_ptr[7] +
src_ptr[stride + 6] + src_ptr[stride + 7]) *
(65536 / 4) >> 16;
dst_ptr[0] = (src_ptr[0] + src_ptr[1] + src_ptr[2] + src_ptr[stride + 0] +
src_ptr[stride + 1] + src_ptr[stride + 2]) *
(65536 / 6) >>
16;
dst_ptr[1] = (src_ptr[3] + src_ptr[4] + src_ptr[5] + src_ptr[stride + 3] +
src_ptr[stride + 4] + src_ptr[stride + 5]) *
(65536 / 6) >>
16;
dst_ptr[2] =
(src_ptr[6] + src_ptr[7] + src_ptr[stride + 6] + src_ptr[stride + 7]) *
(65536 / 4) >>
16;
src_ptr += 8;
dst_ptr += 3;
}
@ -673,11 +774,12 @@ void ScaleAddRow_16_C(const uint16* src_ptr, uint32* dst_ptr, int src_width) {
void ScaleARGBRowDown2_C(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
int x;
(void)src_stride;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src[1];
dst[1] = src[3];
@ -691,8 +793,10 @@ void ScaleARGBRowDown2_C(const uint8* src_argb,
void ScaleARGBRowDown2Linear_C(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
int x;
(void)src_stride;
for (x = 0; x < dst_width; ++x) {
dst_argb[0] = (src_argb[0] + src_argb[4] + 1) >> 1;
dst_argb[1] = (src_argb[1] + src_argb[5] + 1) >> 1;
@ -703,29 +807,37 @@ void ScaleARGBRowDown2Linear_C(const uint8* src_argb,
}
}
void ScaleARGBRowDown2Box_C(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
void ScaleARGBRowDown2Box_C(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb,
int dst_width) {
int x;
for (x = 0; x < dst_width; ++x) {
dst_argb[0] = (src_argb[0] + src_argb[4] +
src_argb[src_stride] + src_argb[src_stride + 4] + 2) >> 2;
dst_argb[1] = (src_argb[1] + src_argb[5] +
src_argb[src_stride + 1] + src_argb[src_stride + 5] + 2) >> 2;
dst_argb[2] = (src_argb[2] + src_argb[6] +
src_argb[src_stride + 2] + src_argb[src_stride + 6] + 2) >> 2;
dst_argb[3] = (src_argb[3] + src_argb[7] +
src_argb[src_stride + 3] + src_argb[src_stride + 7] + 2) >> 2;
dst_argb[0] = (src_argb[0] + src_argb[4] + src_argb[src_stride] +
src_argb[src_stride + 4] + 2) >>
2;
dst_argb[1] = (src_argb[1] + src_argb[5] + src_argb[src_stride + 1] +
src_argb[src_stride + 5] + 2) >>
2;
dst_argb[2] = (src_argb[2] + src_argb[6] + src_argb[src_stride + 2] +
src_argb[src_stride + 6] + 2) >>
2;
dst_argb[3] = (src_argb[3] + src_argb[7] + src_argb[src_stride + 3] +
src_argb[src_stride + 7] + 2) >>
2;
src_argb += 8;
dst_argb += 4;
}
}
void ScaleARGBRowDownEven_C(const uint8* src_argb, ptrdiff_t src_stride,
void ScaleARGBRowDownEven_C(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
(void)src_stride;
int x;
for (x = 0; x < dst_width - 1; x += 2) {
dst[0] = src[0];
@ -741,25 +853,33 @@ void ScaleARGBRowDownEven_C(const uint8* src_argb, ptrdiff_t src_stride,
void ScaleARGBRowDownEvenBox_C(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
int x;
for (x = 0; x < dst_width; ++x) {
dst_argb[0] = (src_argb[0] + src_argb[4] +
src_argb[src_stride] + src_argb[src_stride + 4] + 2) >> 2;
dst_argb[1] = (src_argb[1] + src_argb[5] +
src_argb[src_stride + 1] + src_argb[src_stride + 5] + 2) >> 2;
dst_argb[2] = (src_argb[2] + src_argb[6] +
src_argb[src_stride + 2] + src_argb[src_stride + 6] + 2) >> 2;
dst_argb[3] = (src_argb[3] + src_argb[7] +
src_argb[src_stride + 3] + src_argb[src_stride + 7] + 2) >> 2;
dst_argb[0] = (src_argb[0] + src_argb[4] + src_argb[src_stride] +
src_argb[src_stride + 4] + 2) >>
2;
dst_argb[1] = (src_argb[1] + src_argb[5] + src_argb[src_stride + 1] +
src_argb[src_stride + 5] + 2) >>
2;
dst_argb[2] = (src_argb[2] + src_argb[6] + src_argb[src_stride + 2] +
src_argb[src_stride + 6] + 2) >>
2;
dst_argb[3] = (src_argb[3] + src_argb[7] + src_argb[src_stride + 3] +
src_argb[src_stride + 7] + 2) >>
2;
src_argb += src_stepx * 4;
dst_argb += 4;
}
}
// Scales a single row of pixels using point sampling.
void ScaleARGBCols_C(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBCols_C(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
int j;
@ -775,8 +895,11 @@ void ScaleARGBCols_C(uint8* dst_argb, const uint8* src_argb,
}
}
void ScaleARGBCols64_C(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x32, int dx) {
void ScaleARGBCols64_C(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x32,
int dx) {
int64 x = (int64)(x32);
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
@ -794,11 +917,16 @@ void ScaleARGBCols64_C(uint8* dst_argb, const uint8* src_argb,
}
// Scales a single row of pixels up by 2x using point sampling.
void ScaleARGBColsUp2_C(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBColsUp2_C(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
int j;
(void)x;
(void)dx;
for (j = 0; j < dst_width - 1; j += 2) {
dst[1] = dst[0] = src[0];
src += 1;
@ -809,16 +937,20 @@ void ScaleARGBColsUp2_C(uint8* dst_argb, const uint8* src_argb,
}
}
// TODO(fbarchard): Replace 0x7f ^ f with 128-f. bug=607.
// Mimics SSSE3 blender
#define BLENDER1(a, b, f) ((a) * (0x7f ^ f) + (b) * f) >> 7
#define BLENDERC(a, b, f, s) (uint32)( \
BLENDER1(((a) >> s) & 255, ((b) >> s) & 255, f) << s)
#define BLENDER(a, b, f) \
BLENDERC(a, b, f, 24) | BLENDERC(a, b, f, 16) | \
BLENDERC(a, b, f, 8) | BLENDERC(a, b, f, 0)
#define BLENDER1(a, b, f) ((a) * (0x7f ^ f) + (b)*f) >> 7
#define BLENDERC(a, b, f, s) \
(uint32)(BLENDER1(((a) >> s) & 255, ((b) >> s) & 255, f) << s)
#define BLENDER(a, b, f) \
BLENDERC(a, b, f, 24) | BLENDERC(a, b, f, 16) | BLENDERC(a, b, f, 8) | \
BLENDERC(a, b, f, 0)
void ScaleARGBFilterCols_C(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBFilterCols_C(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
int j;
@ -846,8 +978,11 @@ void ScaleARGBFilterCols_C(uint8* dst_argb, const uint8* src_argb,
}
}
void ScaleARGBFilterCols64_C(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x32, int dx) {
void ScaleARGBFilterCols64_C(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x32,
int dx) {
int64 x = (int64)(x32);
const uint32* src = (const uint32*)(src_argb);
uint32* dst = (uint32*)(dst_argb);
@ -881,16 +1016,22 @@ void ScaleARGBFilterCols64_C(uint8* dst_argb, const uint8* src_argb,
// Scale plane vertically with bilinear interpolation.
void ScalePlaneVertical(int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint8* src_argb, uint8* dst_argb,
int x, int y, int dy,
int bpp, enum FilterMode filtering) {
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint8* src_argb,
uint8* dst_argb,
int x,
int y,
int dy,
int bpp,
enum FilterMode filtering) {
// TODO(fbarchard): Allow higher bpp.
int dst_width_bytes = dst_width * bpp;
void (*InterpolateRow)(uint8* dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_C;
void (*InterpolateRow)(uint8 * dst_argb, const uint8* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_C;
const int max_y = (src_height > 1) ? ((src_height - 1) << 16) - 1 : 0;
int j;
assert(bpp >= 1 && bpp <= 4);
@ -923,14 +1064,22 @@ void ScalePlaneVertical(int src_height,
}
#endif
#if defined(HAS_INTERPOLATEROW_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) &&
IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_argb, 4) &&
IS_ALIGNED(dst_stride, 4)) {
InterpolateRow = InterpolateRow_Any_DSPR2;
if (IS_ALIGNED(dst_width_bytes, 4)) {
InterpolateRow = InterpolateRow_DSPR2;
}
}
#endif
#if defined(HAS_INTERPOLATEROW_MSA)
if (TestCpuFlag(kCpuHasMSA)) {
InterpolateRow = InterpolateRow_Any_MSA;
if (IS_ALIGNED(dst_width_bytes, 32)) {
InterpolateRow = InterpolateRow_MSA;
}
}
#endif
for (j = 0; j < dst_height; ++j) {
int yi;
@ -940,23 +1089,29 @@ void ScalePlaneVertical(int src_height,
}
yi = y >> 16;
yf = filtering ? ((y >> 8) & 255) : 0;
InterpolateRow(dst_argb, src_argb + yi * src_stride,
src_stride, dst_width_bytes, yf);
InterpolateRow(dst_argb, src_argb + yi * src_stride, src_stride,
dst_width_bytes, yf);
dst_argb += dst_stride;
y += dy;
}
}
void ScalePlaneVertical_16(int src_height,
int dst_width, int dst_height,
int src_stride, int dst_stride,
const uint16* src_argb, uint16* dst_argb,
int x, int y, int dy,
int wpp, enum FilterMode filtering) {
int dst_width,
int dst_height,
int src_stride,
int dst_stride,
const uint16* src_argb,
uint16* dst_argb,
int x,
int y,
int dy,
int wpp,
enum FilterMode filtering) {
// TODO(fbarchard): Allow higher wpp.
int dst_width_words = dst_width * wpp;
void (*InterpolateRow)(uint16* dst_argb, const uint16* src_argb,
ptrdiff_t src_stride, int dst_width, int source_y_fraction) =
InterpolateRow_16_C;
void (*InterpolateRow)(uint16 * dst_argb, const uint16* src_argb,
ptrdiff_t src_stride, int dst_width,
int source_y_fraction) = InterpolateRow_16_C;
const int max_y = (src_height > 1) ? ((src_height - 1) << 16) - 1 : 0;
int j;
assert(wpp >= 1 && wpp <= 2);
@ -997,9 +1152,9 @@ void ScalePlaneVertical_16(int src_height,
}
#endif
#if defined(HAS_INTERPOLATEROW_16_DSPR2)
if (TestCpuFlag(kCpuHasDSPR2) &&
IS_ALIGNED(src_argb, 4) && IS_ALIGNED(src_stride, 4) &&
IS_ALIGNED(dst_argb, 4) && IS_ALIGNED(dst_stride, 4)) {
if (TestCpuFlag(kCpuHasDSPR2) && IS_ALIGNED(src_argb, 4) &&
IS_ALIGNED(src_stride, 4) && IS_ALIGNED(dst_argb, 4) &&
IS_ALIGNED(dst_stride, 4)) {
InterpolateRow = InterpolateRow_Any_16_DSPR2;
if (IS_ALIGNED(dst_width_bytes, 4)) {
InterpolateRow = InterpolateRow_16_DSPR2;
@ -1014,16 +1169,18 @@ void ScalePlaneVertical_16(int src_height,
}
yi = y >> 16;
yf = filtering ? ((y >> 8) & 255) : 0;
InterpolateRow(dst_argb, src_argb + yi * src_stride,
src_stride, dst_width_words, yf);
InterpolateRow(dst_argb, src_argb + yi * src_stride, src_stride,
dst_width_words, yf);
dst_argb += dst_stride;
y += dy;
}
}
// Simplify the filtering based on scale factors.
enum FilterMode ScaleFilterReduce(int src_width, int src_height,
int dst_width, int dst_height,
enum FilterMode ScaleFilterReduce(int src_width,
int src_height,
int dst_width,
int dst_height,
enum FilterMode filtering) {
if (src_width < 0) {
src_width = -src_width;
@ -1070,17 +1227,21 @@ int FixedDiv_C(int num, int div) {
// Divide num by div and return as 16.16 fixed point result.
int FixedDiv1_C(int num, int div) {
return (int)((((int64)(num) << 16) - 0x00010001) /
(div - 1));
return (int)((((int64)(num) << 16) - 0x00010001) / (div - 1));
}
#define CENTERSTART(dx, s) (dx < 0) ? -((-dx >> 1) + s) : ((dx >> 1) + s)
// Compute slope values for stepping.
void ScaleSlope(int src_width, int src_height,
int dst_width, int dst_height,
void ScaleSlope(int src_width,
int src_height,
int dst_width,
int dst_height,
enum FilterMode filtering,
int* x, int* y, int* dx, int* dy) {
int* x,
int* y,
int* dx,
int* dy) {
assert(x != NULL);
assert(y != NULL);
assert(dx != NULL);
@ -1112,7 +1273,7 @@ void ScaleSlope(int src_width, int src_height,
*x = 0;
}
if (dst_height <= src_height) {
*dy = FixedDiv(src_height, dst_height);
*dy = FixedDiv(src_height, dst_height);
*y = CENTERSTART(*dy, -32768); // Subtract 0.5 (32768) to center filter.
} else if (dst_height > 1) {
*dy = FixedDiv1(src_height, dst_height);

498
third_party/yuv/source/scale_mips.cc → third_party/yuv/source/scale_dspr2.cc vendored
View File

@ -17,168 +17,167 @@ extern "C" {
#endif
// This module is for GCC MIPS DSPR2
#if !defined(LIBYUV_DISABLE_MIPS) && \
defined(__mips_dsp) && (__mips_dsp_rev >= 2) && \
(_MIPS_SIM == _MIPS_SIM_ABI32)
#if !defined(LIBYUV_DISABLE_DSPR2) && defined(__mips_dsp) && \
(__mips_dsp_rev >= 2) && (_MIPS_SIM == _MIPS_SIM_ABI32)
void ScaleRowDown2_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 4 \n" // iterations -> by 16
"beqz $t9, 2f \n"
" nop \n"
"srl $t9, %[dst_width], 4 \n" // iterations -> by 16
"beqz $t9, 2f \n"
" nop \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
// TODO(fbarchard): Use odd pixels instead of even.
"precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0|
"precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8|
"precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16|
"precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu $t9, $t9, -1 \n"
"sw $t8, 0(%[dst]) \n"
"sw $t0, 4(%[dst]) \n"
"sw $t1, 8(%[dst]) \n"
"sw $t2, 12(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 16 \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
// TODO(fbarchard): Use odd pixels instead of even.
"precrq.qb.ph $t8, $t1, $t0 \n" // |7|5|3|1|
"precrq.qb.ph $t0, $t3, $t2 \n" // |15|13|11|9|
"precrq.qb.ph $t1, $t5, $t4 \n" // |23|21|19|17|
"precrq.qb.ph $t2, $t7, $t6 \n" // |31|29|27|25|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu $t9, $t9, -1 \n"
"sw $t8, 0(%[dst]) \n"
"sw $t0, 4(%[dst]) \n"
"sw $t1, 8(%[dst]) \n"
"sw $t2, 12(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 16 \n"
"2: \n"
"andi $t9, %[dst_width], 0xf \n" // residue
"beqz $t9, 3f \n"
" nop \n"
"2: \n"
"andi $t9, %[dst_width], 0xf \n" // residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lbu $t0, 0(%[src_ptr]) \n"
"addiu %[src_ptr], %[src_ptr], 2 \n"
"addiu $t9, $t9, -1 \n"
"sb $t0, 0(%[dst]) \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 1 \n"
"21: \n"
"lbu $t0, 1(%[src_ptr]) \n"
"addiu %[src_ptr], %[src_ptr], 2 \n"
"addiu $t9, $t9, -1 \n"
"sb $t0, 0(%[dst]) \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
"3: \n"
".set pop \n"
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst)
: [dst_width] "r"(dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9");
}
void ScaleRowDown2Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
const uint8* t = src_ptr + src_stride;
__asm__ __volatile__ (
".set push \n"
".set noreorder \n"
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 3 \n" // iterations -> step 8
"bltz $t9, 2f \n"
" nop \n"
"srl $t9, %[dst_width], 3 \n" // iterations -> step 8
"bltz $t9, 2f \n"
" nop \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 0(%[t]) \n" // |19|18|17|16|
"lw $t5, 4(%[t]) \n" // |23|22|21|20|
"lw $t6, 8(%[t]) \n" // |27|26|25|24|
"lw $t7, 12(%[t]) \n" // |31|30|29|28|
"addiu $t9, $t9, -1 \n"
"srl $t8, $t0, 16 \n" // |X|X|3|2|
"ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
"ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
"raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
"raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
"shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
"shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
"srl $t8, $t1, 16 \n" // |X|X|7|6|
"ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
"ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
"raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
"raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
"shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
"shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
"srl $t8, $t2, 16 \n" // |X|X|11|10|
"ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
"ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
"raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
"raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
"shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
"shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
"srl $t8, $t3, 16 \n" // |X|X|15|14|
"ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
"ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
"raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
"raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
"shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
"shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
"addiu %[src_ptr], %[src_ptr], 16 \n"
"addiu %[t], %[t], 16 \n"
"sb $t0, 0(%[dst]) \n"
"sb $t4, 1(%[dst]) \n"
"sb $t1, 2(%[dst]) \n"
"sb $t5, 3(%[dst]) \n"
"sb $t2, 4(%[dst]) \n"
"sb $t6, 5(%[dst]) \n"
"sb $t3, 6(%[dst]) \n"
"sb $t7, 7(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 8 \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
"lw $t4, 0(%[t]) \n" // |19|18|17|16|
"lw $t5, 4(%[t]) \n" // |23|22|21|20|
"lw $t6, 8(%[t]) \n" // |27|26|25|24|
"lw $t7, 12(%[t]) \n" // |31|30|29|28|
"addiu $t9, $t9, -1 \n"
"srl $t8, $t0, 16 \n" // |X|X|3|2|
"ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
"ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
"raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
"raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
"shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
"shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
"srl $t8, $t1, 16 \n" // |X|X|7|6|
"ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
"ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
"raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
"raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
"shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
"shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
"srl $t8, $t2, 16 \n" // |X|X|11|10|
"ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
"ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
"raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
"raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
"shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
"shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
"srl $t8, $t3, 16 \n" // |X|X|15|14|
"ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
"ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
"raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
"raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
"shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
"shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
"addiu %[src_ptr], %[src_ptr], 16 \n"
"addiu %[t], %[t], 16 \n"
"sb $t0, 0(%[dst]) \n"
"sb $t4, 1(%[dst]) \n"
"sb $t1, 2(%[dst]) \n"
"sb $t5, 3(%[dst]) \n"
"sb $t2, 4(%[dst]) \n"
"sb $t6, 5(%[dst]) \n"
"sb $t3, 6(%[dst]) \n"
"sb $t7, 7(%[dst]) \n"
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 8 \n"
"2: \n"
"andi $t9, %[dst_width], 0x7 \n" // x = residue
"beqz $t9, 3f \n"
" nop \n"
"2: \n"
"andi $t9, %[dst_width], 0x7 \n" // x = residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lwr $t1, 0(%[src_ptr]) \n"
"lwl $t1, 3(%[src_ptr]) \n"
"lwr $t2, 0(%[t]) \n"
"lwl $t2, 3(%[t]) \n"
"srl $t8, $t1, 16 \n"
"ins $t1, $t2, 16, 16 \n"
"ins $t2, $t8, 0, 16 \n"
"raddu.w.qb $t1, $t1 \n"
"raddu.w.qb $t2, $t2 \n"
"shra_r.w $t1, $t1, 2 \n"
"shra_r.w $t2, $t2, 2 \n"
"sb $t1, 0(%[dst]) \n"
"sb $t2, 1(%[dst]) \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"addiu $t9, $t9, -2 \n"
"addiu %[t], %[t], 4 \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 2 \n"
"21: \n"
"lwr $t1, 0(%[src_ptr]) \n"
"lwl $t1, 3(%[src_ptr]) \n"
"lwr $t2, 0(%[t]) \n"
"lwl $t2, 3(%[t]) \n"
"srl $t8, $t1, 16 \n"
"ins $t1, $t2, 16, 16 \n"
"ins $t2, $t8, 0, 16 \n"
"raddu.w.qb $t1, $t1 \n"
"raddu.w.qb $t2, $t2 \n"
"shra_r.w $t1, $t1, 2 \n"
"shra_r.w $t2, $t2, 2 \n"
"sb $t1, 0(%[dst]) \n"
"sb $t2, 1(%[dst]) \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"addiu $t9, $t9, -2 \n"
"addiu %[t], %[t], 4 \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 2 \n"
"3: \n"
".set pop \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst), [t] "+r" (t)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst), [t] "+r"(t)
: [dst_width] "r"(dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9");
}
void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
void ScaleRowDown4_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
@ -186,7 +185,7 @@ void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"beqz $t9, 2f \n"
" nop \n"
"1: \n"
"1: \n"
"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
@ -199,8 +198,8 @@ void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8|
"precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16|
"precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24|
"precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0|
"precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16|
"precrq.qb.ph $t1, $t2, $t1 \n" // |14|10|6|2|
"precrq.qb.ph $t5, $t6, $t5 \n" // |30|26|22|18|
"addiu %[src_ptr], %[src_ptr], 32 \n"
"addiu $t9, $t9, -1 \n"
"sw $t1, 0(%[dst]) \n"
@ -208,44 +207,43 @@ void ScaleRowDown4_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"bgtz $t9, 1b \n"
" addiu %[dst], %[dst], 8 \n"
"2: \n"
"2: \n"
"andi $t9, %[dst_width], 7 \n" // residue
"beqz $t9, 3f \n"
" nop \n"
"21: \n"
"lbu $t1, 0(%[src_ptr]) \n"
"21: \n"
"lbu $t1, 2(%[src_ptr]) \n"
"addiu %[src_ptr], %[src_ptr], 4 \n"
"addiu $t9, $t9, -1 \n"
"sb $t1, 0(%[dst]) \n"
"bgtz $t9, 21b \n"
" addiu %[dst], %[dst], 1 \n"
"3: \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst)
: [dst_width] "r" (dst_width)
: "t1", "t2", "t3", "t4", "t5",
"t6", "t7", "t8", "t9"
);
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst)
: [dst_width] "r"(dst_width)
: "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9");
}
void ScaleRowDown4Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown4Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
intptr_t stride = src_stride;
const uint8* s1 = src_ptr + stride;
const uint8* s2 = s1 + stride;
const uint8* s3 = s2 + stride;
__asm__ __volatile__ (
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"srl $t9, %[dst_width], 1 \n"
"andi $t8, %[dst_width], 1 \n"
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 0(%[s1]) \n" // |7|6|5|4|
"lw $t2, 0(%[s2]) \n" // |11|10|9|8|
@ -299,23 +297,20 @@ void ScaleRowDown4Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"2: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[s1] "+r" (s1),
[s2] "+r" (s2),
[s3] "+r" (s3)
: [dst_width] "r" (dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6","t7", "t8", "t9"
);
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst), [s1] "+r"(s1), [s2] "+r"(s2),
[s3] "+r"(s3)
: [dst_width] "r"(dst_width)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9");
}
void ScaleRowDown34_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
void ScaleRowDown34_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"1: \n"
"1: \n"
"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
@ -347,23 +342,21 @@ void ScaleRowDown34_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"bnez %[dst_width], 1b \n"
" addiu %[dst], %[dst], 24 \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[dst_width] "+r" (dst_width)
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst), [dst_width] "+r"(dst_width)
:
: "t0", "t1", "t2", "t3", "t4", "t5",
"t6","t7", "t8", "t9"
);
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8", "t9");
}
void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
__asm__ __volatile__ (
void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* d,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"repl.ph $t3, 3 \n" // 0x00030003
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
"rotr $t2, $t0, 8 \n" // |S0|S3|S2|S1|
@ -400,26 +393,24 @@ void ScaleRowDown34_0_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"sb $t6, 2(%[d]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[d], %[d], 3 \n"
"3: \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[src_stride] "+r" (src_stride),
[d] "+r" (d),
[dst_width] "+r" (dst_width)
: [src_ptr] "+r"(src_ptr), [src_stride] "+r"(src_stride), [d] "+r"(d),
[dst_width] "+r"(dst_width)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6"
);
: "t0", "t1", "t2", "t3", "t4", "t5", "t6");
}
void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* d, int dst_width) {
__asm__ __volatile__ (
void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* d,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"repl.ph $t2, 3 \n" // 0x00030003
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
"rotr $t4, $t0, 8 \n" // |S0|S3|S2|S1|
@ -452,25 +443,23 @@ void ScaleRowDown34_1_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"sb $t6, 2(%[d]) \n"
"bgtz %[dst_width], 1b \n"
" addiu %[d], %[d], 3 \n"
"3: \n"
"3: \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[src_stride] "+r" (src_stride),
[d] "+r" (d),
[dst_width] "+r" (dst_width)
: [src_ptr] "+r"(src_ptr), [src_stride] "+r"(src_stride), [d] "+r"(d),
[dst_width] "+r"(dst_width)
:
: "t0", "t1", "t2", "t3",
"t4", "t5", "t6"
);
: "t0", "t1", "t2", "t3", "t4", "t5", "t6");
}
void ScaleRowDown38_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
__asm__ __volatile__ (
void ScaleRowDown38_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
@ -501,26 +490,24 @@ void ScaleRowDown38_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"bgez $t8, 1b \n"
" addiu %[dst], %[dst], 12 \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst] "+r" (dst),
[dst_width] "+r" (dst_width)
: [src_ptr] "+r"(src_ptr), [dst] "+r"(dst), [dst_width] "+r"(dst_width)
:
: "t0", "t1", "t2", "t3", "t4",
"t5", "t6", "t7", "t8"
);
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8");
}
void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
const uint8* t = src_ptr + stride;
const int c = 0x2AAA;
__asm__ __volatile__ (
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
"lw $t2, 0(%[t]) \n" // |T3|T2|T1|T0|
@ -554,18 +541,16 @@ void ScaleRowDown38_2_Box_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
"bgtz %[dst_width], 1b \n"
" sb $t0, -3(%[dst_ptr]) \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst_ptr] "+r" (dst_ptr),
[t] "+r" (t),
[dst_width] "+r" (dst_width)
: [c] "r" (c)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6"
);
: [src_ptr] "+r"(src_ptr), [dst_ptr] "+r"(dst_ptr), [t] "+r"(t),
[dst_width] "+r"(dst_width)
: [c] "r"(c)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6");
}
void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
intptr_t stride = src_stride;
const uint8* s1 = src_ptr + stride;
stride += stride;
@ -573,11 +558,11 @@ void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
const int c1 = 0x1C71;
const int c2 = 0x2AAA;
__asm__ __volatile__ (
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"1: \n"
"1: \n"
"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
"lw $t2, 0(%[s1]) \n" // |T3|T2|T1|T0|
@ -624,15 +609,55 @@ void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
"bgtz %[dst_width], 1b \n"
" sb $t0, -3(%[dst_ptr]) \n"
".set pop \n"
: [src_ptr] "+r" (src_ptr),
[dst_ptr] "+r" (dst_ptr),
[s1] "+r" (s1),
[s2] "+r" (s2),
[dst_width] "+r" (dst_width)
: [c1] "r" (c1), [c2] "r" (c2)
: "t0", "t1", "t2", "t3", "t4",
"t5", "t6", "t7", "t8"
);
: [src_ptr] "+r"(src_ptr), [dst_ptr] "+r"(dst_ptr), [s1] "+r"(s1),
[s2] "+r"(s2), [dst_width] "+r"(dst_width)
: [c1] "r"(c1), [c2] "r"(c2)
: "t0", "t1", "t2", "t3", "t4", "t5", "t6", "t7", "t8");
}
void ScaleAddRow_DSPR2(const uint8* src_ptr, uint16* dst_ptr, int src_width) {
int x;
for (x = 0; x < ((src_width - 1)); x += 8) {
uint32 tmp_t1, tmp_t2, tmp_t3, tmp_t4;
uint32 tmp_t5, tmp_t6, tmp_t7, tmp_t8;
__asm__ __volatile__(
".set push \n"
".set noreorder \n"
"lw %[tmp_t5], 0(%[src_ptr]) \n"
"lw %[tmp_t6], 4(%[src_ptr]) \n"
"lw %[tmp_t1], 0(%[dst_ptr]) \n"
"lw %[tmp_t2], 4(%[dst_ptr]) \n"
"lw %[tmp_t3], 8(%[dst_ptr]) \n"
"lw %[tmp_t4], 12(%[dst_ptr]) \n"
"preceu.ph.qbr %[tmp_t7], %[tmp_t5] \n"
"preceu.ph.qbl %[tmp_t8], %[tmp_t5] \n"
"addu.ph %[tmp_t1], %[tmp_t1], %[tmp_t7] \n"
"addu.ph %[tmp_t2], %[tmp_t2], %[tmp_t8] \n"
"preceu.ph.qbr %[tmp_t7], %[tmp_t6] \n"
"preceu.ph.qbl %[tmp_t8], %[tmp_t6] \n"
"addu.ph %[tmp_t3], %[tmp_t3], %[tmp_t7] \n"
"addu.ph %[tmp_t4], %[tmp_t4], %[tmp_t8] \n"
"sw %[tmp_t1], 0(%[dst_ptr]) \n"
"sw %[tmp_t2], 4(%[dst_ptr]) \n"
"sw %[tmp_t3], 8(%[dst_ptr]) \n"
"sw %[tmp_t4], 12(%[dst_ptr]) \n"
".set pop \n"
:
[tmp_t1] "=&r"(tmp_t1), [tmp_t2] "=&r"(tmp_t2), [tmp_t3] "=&r"(tmp_t3),
[tmp_t4] "=&r"(tmp_t4), [tmp_t5] "=&r"(tmp_t5), [tmp_t6] "=&r"(tmp_t6),
[tmp_t7] "=&r"(tmp_t7), [tmp_t8] "=&r"(tmp_t8), [src_ptr] "+r"(src_ptr)
: [dst_ptr] "r"(dst_ptr));
src_ptr += 8;
dst_ptr += 8;
}
if ((src_width)&7) {
for (x = 0; x < ((src_width - 1) & 7); x += 1) {
dst_ptr[0] += src_ptr[0];
src_ptr += 1;
dst_ptr += 1;
}
}
}
#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
@ -641,4 +666,3 @@ void ScaleRowDown38_3_Box_DSPR2(const uint8* src_ptr,
} // extern "C"
} // namespace libyuv
#endif

479
third_party/yuv/source/scale_gcc.cc vendored
View File

@ -21,85 +21,82 @@ extern "C" {
(defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
// Offsets for source bytes 0 to 9
static uvec8 kShuf0 =
{ 0, 1, 3, 4, 5, 7, 8, 9, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShuf0 = {0, 1, 3, 4, 5, 7, 8, 9,
128, 128, 128, 128, 128, 128, 128, 128};
// Offsets for source bytes 11 to 20 with 8 subtracted = 3 to 12.
static uvec8 kShuf1 =
{ 3, 4, 5, 7, 8, 9, 11, 12, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShuf1 = {3, 4, 5, 7, 8, 9, 11, 12,
128, 128, 128, 128, 128, 128, 128, 128};
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
static uvec8 kShuf2 =
{ 5, 7, 8, 9, 11, 12, 13, 15, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShuf2 = {5, 7, 8, 9, 11, 12, 13, 15,
128, 128, 128, 128, 128, 128, 128, 128};
// Offsets for source bytes 0 to 10
static uvec8 kShuf01 =
{ 0, 1, 1, 2, 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10 };
static uvec8 kShuf01 = {0, 1, 1, 2, 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10};
// Offsets for source bytes 10 to 21 with 8 subtracted = 3 to 13.
static uvec8 kShuf11 =
{ 2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13 };
static uvec8 kShuf11 = {2, 3, 4, 5, 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13};
// Offsets for source bytes 21 to 31 with 16 subtracted = 5 to 31.
static uvec8 kShuf21 =
{ 5, 6, 6, 7, 8, 9, 9, 10, 10, 11, 12, 13, 13, 14, 14, 15 };
static uvec8 kShuf21 = {5, 6, 6, 7, 8, 9, 9, 10,
10, 11, 12, 13, 13, 14, 14, 15};
// Coefficients for source bytes 0 to 10
static uvec8 kMadd01 =
{ 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2 };
static uvec8 kMadd01 = {3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2};
// Coefficients for source bytes 10 to 21
static uvec8 kMadd11 =
{ 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1 };
static uvec8 kMadd11 = {1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1};
// Coefficients for source bytes 21 to 31
static uvec8 kMadd21 =
{ 2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3 };
static uvec8 kMadd21 = {2, 2, 1, 3, 3, 1, 2, 2, 1, 3, 3, 1, 2, 2, 1, 3};
// Coefficients for source bytes 21 to 31
static vec16 kRound34 =
{ 2, 2, 2, 2, 2, 2, 2, 2 };
static vec16 kRound34 = {2, 2, 2, 2, 2, 2, 2, 2};
static uvec8 kShuf38a =
{ 0, 3, 6, 8, 11, 14, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShuf38a = {0, 3, 6, 8, 11, 14, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128};
static uvec8 kShuf38b =
{ 128, 128, 128, 128, 128, 128, 0, 3, 6, 8, 11, 14, 128, 128, 128, 128 };
static uvec8 kShuf38b = {128, 128, 128, 128, 128, 128, 0, 3,
6, 8, 11, 14, 128, 128, 128, 128};
// Arrange words 0,3,6 into 0,1,2
static uvec8 kShufAc =
{ 0, 1, 6, 7, 12, 13, 128, 128, 128, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShufAc = {0, 1, 6, 7, 12, 13, 128, 128,
128, 128, 128, 128, 128, 128, 128, 128};
// Arrange words 0,3,6 into 3,4,5
static uvec8 kShufAc3 =
{ 128, 128, 128, 128, 128, 128, 0, 1, 6, 7, 12, 13, 128, 128, 128, 128 };
static uvec8 kShufAc3 = {128, 128, 128, 128, 128, 128, 0, 1,
6, 7, 12, 13, 128, 128, 128, 128};
// Scaling values for boxes of 3x3 and 2x3
static uvec16 kScaleAc33 =
{ 65536 / 9, 65536 / 9, 65536 / 6, 65536 / 9, 65536 / 9, 65536 / 6, 0, 0 };
static uvec16 kScaleAc33 = {65536 / 9, 65536 / 9, 65536 / 6, 65536 / 9,
65536 / 9, 65536 / 6, 0, 0};
// Arrange first value for pixels 0,1,2,3,4,5
static uvec8 kShufAb0 =
{ 0, 128, 3, 128, 6, 128, 8, 128, 11, 128, 14, 128, 128, 128, 128, 128 };
static uvec8 kShufAb0 = {0, 128, 3, 128, 6, 128, 8, 128,
11, 128, 14, 128, 128, 128, 128, 128};
// Arrange second value for pixels 0,1,2,3,4,5
static uvec8 kShufAb1 =
{ 1, 128, 4, 128, 7, 128, 9, 128, 12, 128, 15, 128, 128, 128, 128, 128 };
static uvec8 kShufAb1 = {1, 128, 4, 128, 7, 128, 9, 128,
12, 128, 15, 128, 128, 128, 128, 128};
// Arrange third value for pixels 0,1,2,3,4,5
static uvec8 kShufAb2 =
{ 2, 128, 5, 128, 128, 128, 10, 128, 13, 128, 128, 128, 128, 128, 128, 128 };
static uvec8 kShufAb2 = {2, 128, 5, 128, 128, 128, 10, 128,
13, 128, 128, 128, 128, 128, 128, 128};
// Scaling values for boxes of 3x2 and 2x2
static uvec16 kScaleAb2 =
{ 65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3, 65536 / 3, 65536 / 2, 0, 0 };
static uvec16 kScaleAb2 = {65536 / 3, 65536 / 3, 65536 / 2, 65536 / 3,
65536 / 3, 65536 / 2, 0, 0};
// GCC versions of row functions are verbatim conversions from Visual C.
// Generated using gcc disassembly on Visual C object file:
// objdump -D yuvscaler.obj >yuvscaler.txt
void ScaleRowDown2_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
LABELALIGN
"1: \n"
@ -120,8 +117,11 @@ void ScaleRowDown2_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown2Linear_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2Linear_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"pcmpeqb %%xmm4,%%xmm4 \n"
"psrlw $0xf,%%xmm4 \n"
@ -149,8 +149,10 @@ void ScaleRowDown2Linear_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown2Box_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
asm volatile (
"pcmpeqb %%xmm4,%%xmm4 \n"
"psrlw $0xf,%%xmm4 \n"
@ -189,8 +191,11 @@ void ScaleRowDown2Box_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
}
#ifdef HAS_SCALEROWDOWN2_AVX2
void ScaleRowDown2_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2_AVX2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
LABELALIGN
"1: \n"
@ -213,8 +218,11 @@ void ScaleRowDown2_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown2Linear_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2Linear_AVX2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"vpcmpeqb %%ymm4,%%ymm4,%%ymm4 \n"
"vpsrlw $0xf,%%ymm4,%%ymm4 \n"
@ -244,8 +252,10 @@ void ScaleRowDown2Linear_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown2Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown2Box_AVX2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
asm volatile (
"vpcmpeqb %%ymm4,%%ymm4,%%ymm4 \n"
"vpsrlw $0xf,%%ymm4,%%ymm4 \n"
@ -286,8 +296,11 @@ void ScaleRowDown2Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
}
#endif // HAS_SCALEROWDOWN2_AVX2
void ScaleRowDown4_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"pcmpeqb %%xmm5,%%xmm5 \n"
"psrld $0x18,%%xmm5 \n"
@ -314,8 +327,10 @@ void ScaleRowDown4_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown4Box_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
intptr_t stridex3;
asm volatile (
"pcmpeqb %%xmm4,%%xmm4 \n"
@ -368,10 +383,12 @@ void ScaleRowDown4Box_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
#ifdef HAS_SCALEROWDOWN4_AVX2
void ScaleRowDown4_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4_AVX2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"vpcmpeqb %%ymm5,%%ymm5,%%ymm5 \n"
"vpsrld $0x18,%%ymm5,%%ymm5 \n"
@ -400,8 +417,10 @@ void ScaleRowDown4_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown4Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4Box_AVX2(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
asm volatile (
"vpcmpeqb %%ymm4,%%ymm4,%%ymm4 \n"
"vpsrlw $0xf,%%ymm4,%%ymm4 \n"
@ -455,17 +474,20 @@ void ScaleRowDown4Box_AVX2(const uint8* src_ptr, ptrdiff_t src_stride,
}
#endif // HAS_SCALEROWDOWN4_AVX2
void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"movdqa %0,%%xmm3 \n"
"movdqa %1,%%xmm4 \n"
"movdqa %2,%%xmm5 \n"
:
: "m"(kShuf0), // %0
"m"(kShuf1), // %1
"m"(kShuf2) // %2
);
void ScaleRowDown34_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile(
"movdqa %0,%%xmm3 \n"
"movdqa %1,%%xmm4 \n"
"movdqa %2,%%xmm5 \n"
:
: "m"(kShuf0), // %0
"m"(kShuf1), // %1
"m"(kShuf2) // %2
);
asm volatile (
LABELALIGN
"1: \n"
@ -492,25 +514,26 @@ void ScaleRowDown34_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
void ScaleRowDown34_1_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"movdqa %0,%%xmm2 \n" // kShuf01
"movdqa %1,%%xmm3 \n" // kShuf11
"movdqa %2,%%xmm4 \n" // kShuf21
:
: "m"(kShuf01), // %0
"m"(kShuf11), // %1
"m"(kShuf21) // %2
);
asm volatile (
"movdqa %0,%%xmm5 \n" // kMadd01
"movdqa %1,%%xmm0 \n" // kMadd11
"movdqa %2,%%xmm1 \n" // kRound34
:
: "m"(kMadd01), // %0
"m"(kMadd11), // %1
"m"(kRound34) // %2
);
uint8* dst_ptr,
int dst_width) {
asm volatile(
"movdqa %0,%%xmm2 \n" // kShuf01
"movdqa %1,%%xmm3 \n" // kShuf11
"movdqa %2,%%xmm4 \n" // kShuf21
:
: "m"(kShuf01), // %0
"m"(kShuf11), // %1
"m"(kShuf21) // %2
);
asm volatile(
"movdqa %0,%%xmm5 \n" // kMadd01
"movdqa %1,%%xmm0 \n" // kMadd11
"movdqa %2,%%xmm1 \n" // kRound34
:
: "m"(kMadd01), // %0
"m"(kMadd11), // %1
"m"(kRound34) // %2
);
asm volatile (
LABELALIGN
"1: \n"
@ -557,25 +580,26 @@ void ScaleRowDown34_1_Box_SSSE3(const uint8* src_ptr,
void ScaleRowDown34_0_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"movdqa %0,%%xmm2 \n" // kShuf01
"movdqa %1,%%xmm3 \n" // kShuf11
"movdqa %2,%%xmm4 \n" // kShuf21
:
: "m"(kShuf01), // %0
"m"(kShuf11), // %1
"m"(kShuf21) // %2
);
asm volatile (
"movdqa %0,%%xmm5 \n" // kMadd01
"movdqa %1,%%xmm0 \n" // kMadd11
"movdqa %2,%%xmm1 \n" // kRound34
:
: "m"(kMadd01), // %0
"m"(kMadd11), // %1
"m"(kRound34) // %2
);
uint8* dst_ptr,
int dst_width) {
asm volatile(
"movdqa %0,%%xmm2 \n" // kShuf01
"movdqa %1,%%xmm3 \n" // kShuf11
"movdqa %2,%%xmm4 \n" // kShuf21
:
: "m"(kShuf01), // %0
"m"(kShuf11), // %1
"m"(kShuf21) // %2
);
asm volatile(
"movdqa %0,%%xmm5 \n" // kMadd01
"movdqa %1,%%xmm0 \n" // kMadd11
"movdqa %2,%%xmm1 \n" // kRound34
:
: "m"(kMadd01), // %0
"m"(kMadd11), // %1
"m"(kRound34) // %2
);
asm volatile (
LABELALIGN
@ -624,8 +648,11 @@ void ScaleRowDown34_0_Box_SSSE3(const uint8* src_ptr,
);
}
void ScaleRowDown38_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown38_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"movdqa %3,%%xmm4 \n"
"movdqa %4,%%xmm5 \n"
@ -655,18 +682,19 @@ void ScaleRowDown38_SSSE3(const uint8* src_ptr, ptrdiff_t src_stride,
void ScaleRowDown38_2_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"movdqa %0,%%xmm2 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm4 \n"
"movdqa %3,%%xmm5 \n"
:
: "m"(kShufAb0), // %0
"m"(kShufAb1), // %1
"m"(kShufAb2), // %2
"m"(kScaleAb2) // %3
);
uint8* dst_ptr,
int dst_width) {
asm volatile(
"movdqa %0,%%xmm2 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm4 \n"
"movdqa %3,%%xmm5 \n"
:
: "m"(kShufAb0), // %0
"m"(kShufAb1), // %1
"m"(kShufAb2), // %2
"m"(kScaleAb2) // %3
);
asm volatile (
LABELALIGN
"1: \n"
@ -700,17 +728,18 @@ void ScaleRowDown38_2_Box_SSSE3(const uint8* src_ptr,
void ScaleRowDown38_3_Box_SSSE3(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
asm volatile (
"movdqa %0,%%xmm2 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm4 \n"
"pxor %%xmm5,%%xmm5 \n"
:
: "m"(kShufAc), // %0
"m"(kShufAc3), // %1
"m"(kScaleAc33) // %2
);
uint8* dst_ptr,
int dst_width) {
asm volatile(
"movdqa %0,%%xmm2 \n"
"movdqa %1,%%xmm3 \n"
"movdqa %2,%%xmm4 \n"
"pxor %%xmm5,%%xmm5 \n"
:
: "m"(kShufAc), // %0
"m"(kShufAc3), // %1
"m"(kScaleAc33) // %2
);
asm volatile (
LABELALIGN
"1: \n"
@ -790,7 +819,6 @@ void ScaleAddRow_SSE2(const uint8* src_ptr, uint16* dst_ptr, int src_width) {
);
}
#ifdef HAS_SCALEADDROW_AVX2
// Reads 32 bytes and accumulates to 32 shorts at a time.
void ScaleAddRow_AVX2(const uint8* src_ptr, uint16* dst_ptr, int src_width) {
@ -821,9 +849,21 @@ void ScaleAddRow_AVX2(const uint8* src_ptr, uint16* dst_ptr, int src_width) {
}
#endif // HAS_SCALEADDROW_AVX2
// Constant for making pixels signed to avoid pmaddubsw
// saturation.
static uvec8 kFsub80 = {0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80,
0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80};
// Constant for making pixels unsigned and adding .5 for rounding.
static uvec16 kFadd40 = {0x4040, 0x4040, 0x4040, 0x4040,
0x4040, 0x4040, 0x4040, 0x4040};
// Bilinear column filtering. SSSE3 version.
void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
void ScaleFilterCols_SSSE3(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
intptr_t x0, x1, temp_pixel;
asm volatile (
"movd %6,%%xmm2 \n"
@ -831,7 +871,10 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
"movl $0x04040000,%k2 \n"
"movd %k2,%%xmm5 \n"
"pcmpeqb %%xmm6,%%xmm6 \n"
"psrlw $0x9,%%xmm6 \n"
"psrlw $0x9,%%xmm6 \n" // 0x007f007f
"pcmpeqb %%xmm7,%%xmm7 \n"
"psrlw $15,%%xmm7 \n" // 0x00010001
"pextrw $0x1,%%xmm2,%k3 \n"
"subl $0x2,%5 \n"
"jl 29f \n"
@ -853,16 +896,19 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
"movd %k2,%%xmm4 \n"
"pshufb %%xmm5,%%xmm1 \n"
"punpcklwd %%xmm4,%%xmm0 \n"
"pxor %%xmm6,%%xmm1 \n"
"pmaddubsw %%xmm1,%%xmm0 \n"
"psubb %8,%%xmm0 \n" // make pixels signed.
"pxor %%xmm6,%%xmm1 \n" // 128 - f = (f ^ 127 ) + 1
"paddusb %%xmm7,%%xmm1 \n"
"pmaddubsw %%xmm0,%%xmm1 \n"
"pextrw $0x1,%%xmm2,%k3 \n"
"pextrw $0x3,%%xmm2,%k4 \n"
"psrlw $0x7,%%xmm0 \n"
"packuswb %%xmm0,%%xmm0 \n"
"movd %%xmm0,%k2 \n"
"paddw %9,%%xmm1 \n" // make pixels unsigned.
"psrlw $0x7,%%xmm1 \n"
"packuswb %%xmm1,%%xmm1 \n"
"movd %%xmm1,%k2 \n"
"mov %w2," MEMACCESS(0) " \n"
"lea " MEMLEA(0x2,0) ",%0 \n"
"sub $0x2,%5 \n"
"subl $0x2,%5 \n"
"jge 2b \n"
LABELALIGN
@ -873,11 +919,14 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
"movd %k2,%%xmm0 \n"
"psrlw $0x9,%%xmm2 \n"
"pshufb %%xmm5,%%xmm2 \n"
"psubb %8,%%xmm0 \n" // make pixels signed.
"pxor %%xmm6,%%xmm2 \n"
"pmaddubsw %%xmm2,%%xmm0 \n"
"psrlw $0x7,%%xmm0 \n"
"packuswb %%xmm0,%%xmm0 \n"
"movd %%xmm0,%k2 \n"
"paddusb %%xmm7,%%xmm2 \n"
"pmaddubsw %%xmm0,%%xmm2 \n"
"paddw %9,%%xmm2 \n" // make pixels unsigned.
"psrlw $0x7,%%xmm2 \n"
"packuswb %%xmm2,%%xmm2 \n"
"movd %%xmm2,%k2 \n"
"mov %b2," MEMACCESS(0) " \n"
"99: \n"
: "+r"(dst_ptr), // %0
@ -885,18 +934,34 @@ void ScaleFilterCols_SSSE3(uint8* dst_ptr, const uint8* src_ptr,
"=&a"(temp_pixel), // %2
"=&r"(x0), // %3
"=&r"(x1), // %4
#if defined(__x86_64__)
"+rm"(dst_width) // %5
#else
"+m"(dst_width) // %5
#endif
: "rm"(x), // %6
"rm"(dx) // %7
"rm"(dx), // %7
#if defined(__x86_64__)
"x"(kFsub80), // %8
"x"(kFadd40) // %9
#else
"m"(kFsub80), // %8
"m"(kFadd40) // %9
#endif
: "memory", "cc", NACL_R14
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
"xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
);
}
// Reads 4 pixels, duplicates them and writes 8 pixels.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
void ScaleColsUp2_SSE2(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
void ScaleColsUp2_SSE2(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
(void)x;
(void)dx;
asm volatile (
LABELALIGN
"1: \n"
@ -920,7 +985,9 @@ void ScaleColsUp2_SSE2(uint8* dst_ptr, const uint8* src_ptr,
void ScaleARGBRowDown2_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile (
LABELALIGN
"1: \n"
@ -941,7 +1008,9 @@ void ScaleARGBRowDown2_SSE2(const uint8* src_argb,
void ScaleARGBRowDown2Linear_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile (
LABELALIGN
"1: \n"
@ -965,7 +1034,8 @@ void ScaleARGBRowDown2Linear_SSE2(const uint8* src_argb,
void ScaleARGBRowDown2Box_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
asm volatile (
LABELALIGN
"1: \n"
@ -995,10 +1065,14 @@ void ScaleARGBRowDown2Box_SSE2(const uint8* src_argb,
// Reads 4 pixels at a time.
// Alignment requirement: dst_argb 16 byte aligned.
void ScaleARGBRowDownEven_SSE2(const uint8* src_argb, ptrdiff_t src_stride,
int src_stepx, uint8* dst_argb, int dst_width) {
void ScaleARGBRowDownEven_SSE2(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb,
int dst_width) {
intptr_t src_stepx_x4 = (intptr_t)(src_stepx);
intptr_t src_stepx_x12;
(void)src_stride;
asm volatile (
"lea " MEMLEA3(0x00,1,4) ",%1 \n"
"lea " MEMLEA4(0x00,1,1,2) ",%4 \n"
@ -1029,8 +1103,10 @@ void ScaleARGBRowDownEven_SSE2(const uint8* src_argb, ptrdiff_t src_stride,
// Blends four 2x2 to 4x1.
// Alignment requirement: dst_argb 16 byte aligned.
void ScaleARGBRowDownEvenBox_SSE2(const uint8* src_argb,
ptrdiff_t src_stride, int src_stepx,
uint8* dst_argb, int dst_width) {
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb,
int dst_width) {
intptr_t src_stepx_x4 = (intptr_t)(src_stepx);
intptr_t src_stepx_x12;
intptr_t row1 = (intptr_t)(src_stride);
@ -1072,8 +1148,11 @@ void ScaleARGBRowDownEvenBox_SSE2(const uint8* src_argb,
);
}
void ScaleARGBCols_SSE2(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBCols_SSE2(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
intptr_t x0, x1;
asm volatile (
"movd %5,%%xmm2 \n"
@ -1141,8 +1220,13 @@ void ScaleARGBCols_SSE2(uint8* dst_argb, const uint8* src_argb,
// Reads 4 pixels, duplicates them and writes 8 pixels.
// Alignment requirement: src_argb 16 byte aligned, dst_argb 16 byte aligned.
void ScaleARGBColsUp2_SSE2(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBColsUp2_SSE2(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
(void)x;
(void)dx;
asm volatile (
LABELALIGN
"1: \n"
@ -1167,26 +1251,29 @@ void ScaleARGBColsUp2_SSE2(uint8* dst_argb, const uint8* src_argb,
// Shuffle table for arranging 2 pixels into pairs for pmaddubsw
static uvec8 kShuffleColARGB = {
0u, 4u, 1u, 5u, 2u, 6u, 3u, 7u, // bbggrraa 1st pixel
8u, 12u, 9u, 13u, 10u, 14u, 11u, 15u // bbggrraa 2nd pixel
0u, 4u, 1u, 5u, 2u, 6u, 3u, 7u, // bbggrraa 1st pixel
8u, 12u, 9u, 13u, 10u, 14u, 11u, 15u // bbggrraa 2nd pixel
};
// Shuffle table for duplicating 2 fractions into 8 bytes each
static uvec8 kShuffleFractions = {
0u, 0u, 0u, 0u, 0u, 0u, 0u, 0u, 4u, 4u, 4u, 4u, 4u, 4u, 4u, 4u,
0u, 0u, 0u, 0u, 0u, 0u, 0u, 0u, 4u, 4u, 4u, 4u, 4u, 4u, 4u, 4u,
};
// Bilinear row filtering combines 4x2 -> 4x1. SSSE3 version
void ScaleARGBFilterCols_SSSE3(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBFilterCols_SSSE3(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
intptr_t x0, x1;
asm volatile (
"movdqa %0,%%xmm4 \n"
"movdqa %1,%%xmm5 \n"
:
: "m"(kShuffleColARGB), // %0
"m"(kShuffleFractions) // %1
);
asm volatile(
"movdqa %0,%%xmm4 \n"
"movdqa %1,%%xmm5 \n"
:
: "m"(kShuffleColARGB), // %0
"m"(kShuffleFractions) // %1
);
asm volatile (
"movd %5,%%xmm2 \n"
@ -1253,34 +1340,32 @@ void ScaleARGBFilterCols_SSSE3(uint8* dst_argb, const uint8* src_argb,
// Divide num by div and return as 16.16 fixed point result.
int FixedDiv_X86(int num, int div) {
asm volatile (
"cdq \n"
"shld $0x10,%%eax,%%edx \n"
"shl $0x10,%%eax \n"
"idiv %1 \n"
"mov %0, %%eax \n"
: "+a"(num) // %0
: "c"(div) // %1
: "memory", "cc", "edx"
);
asm volatile(
"cdq \n"
"shld $0x10,%%eax,%%edx \n"
"shl $0x10,%%eax \n"
"idiv %1 \n"
"mov %0, %%eax \n"
: "+a"(num) // %0
: "c"(div) // %1
: "memory", "cc", "edx");
return num;
}
// Divide num - 1 by div - 1 and return as 16.16 fixed point result.
int FixedDiv1_X86(int num, int div) {
asm volatile (
"cdq \n"
"shld $0x10,%%eax,%%edx \n"
"shl $0x10,%%eax \n"
"sub $0x10001,%%eax \n"
"sbb $0x0,%%edx \n"
"sub $0x1,%1 \n"
"idiv %1 \n"
"mov %0, %%eax \n"
: "+a"(num) // %0
: "c"(div) // %1
: "memory", "cc", "edx"
);
asm volatile(
"cdq \n"
"shld $0x10,%%eax,%%edx \n"
"shl $0x10,%%eax \n"
"sub $0x10001,%%eax \n"
"sbb $0x0,%%edx \n"
"sub $0x1,%1 \n"
"idiv %1 \n"
"mov %0, %%eax \n"
: "+a"(num) // %0
: "c"(div) // %1
: "memory", "cc", "edx");
return num;
}

553
third_party/yuv/source/scale_msa.cc vendored Normal file
View File

@ -0,0 +1,553 @@
/*
* Copyright 2016 The LibYuv Project Authors. All rights reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "libyuv/scale_row.h"
// This module is for GCC MSA
#if !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)
#include "libyuv/macros_msa.h"
#ifdef __cplusplus
namespace libyuv {
extern "C" {
#endif
void ScaleARGBRowDown2_MSA(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
v16u8 src0, src1, dst0;
(void)src_stride;
for (x = 0; x < dst_width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
dst0 = (v16u8)__msa_pckod_w((v4i32)src1, (v4i32)src0);
ST_UB(dst0, dst_argb);
src_argb += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDown2Linear_MSA(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
v16u8 src0, src1, vec0, vec1, dst0;
(void)src_stride;
for (x = 0; x < dst_width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_argb, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_argb, 16);
vec0 = (v16u8)__msa_pckev_w((v4i32)src1, (v4i32)src0);
vec1 = (v16u8)__msa_pckod_w((v4i32)src1, (v4i32)src0);
dst0 = (v16u8)__msa_aver_u_b((v16u8)vec0, (v16u8)vec1);
ST_UB(dst0, dst_argb);
src_argb += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDown2Box_MSA(const uint8_t* src_argb,
ptrdiff_t src_stride,
uint8_t* dst_argb,
int dst_width) {
int x;
const uint8_t* s = src_argb;
const uint8_t* t = src_argb + src_stride;
v16u8 src0, src1, src2, src3, vec0, vec1, vec2, vec3, dst0;
v8u16 reg0, reg1, reg2, reg3;
v16i8 shuffler = {0, 4, 1, 5, 2, 6, 3, 7, 8, 12, 9, 13, 10, 14, 11, 15};
for (x = 0; x < dst_width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)t, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)t, 16);
vec0 = (v16u8)__msa_vshf_b(shuffler, (v16i8)src0, (v16i8)src0);
vec1 = (v16u8)__msa_vshf_b(shuffler, (v16i8)src1, (v16i8)src1);
vec2 = (v16u8)__msa_vshf_b(shuffler, (v16i8)src2, (v16i8)src2);
vec3 = (v16u8)__msa_vshf_b(shuffler, (v16i8)src3, (v16i8)src3);
reg0 = __msa_hadd_u_h(vec0, vec0);
reg1 = __msa_hadd_u_h(vec1, vec1);
reg2 = __msa_hadd_u_h(vec2, vec2);
reg3 = __msa_hadd_u_h(vec3, vec3);
reg0 += reg2;
reg1 += reg3;
reg0 = (v8u16)__msa_srari_h((v8i16)reg0, 2);
reg1 = (v8u16)__msa_srari_h((v8i16)reg1, 2);
dst0 = (v16u8)__msa_pckev_b((v16i8)reg1, (v16i8)reg0);
ST_UB(dst0, dst_argb);
s += 32;
t += 32;
dst_argb += 16;
}
}
void ScaleARGBRowDownEven_MSA(const uint8_t* src_argb,
ptrdiff_t src_stride,
int32_t src_stepx,
uint8_t* dst_argb,
int dst_width) {
int x;
int32_t stepx = src_stepx * 4;
int32_t data0, data1, data2, data3;
(void)src_stride;
for (x = 0; x < dst_width; x += 4) {
data0 = LW(src_argb);
data1 = LW(src_argb + stepx);
data2 = LW(src_argb + stepx * 2);
data3 = LW(src_argb + stepx * 3);
SW(data0, dst_argb);
SW(data1, dst_argb + 4);
SW(data2, dst_argb + 8);
SW(data3, dst_argb + 12);
src_argb += stepx * 4;
dst_argb += 16;
}
}
void ScaleARGBRowDownEvenBox_MSA(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb,
int dst_width) {
int x;
const uint8* nxt_argb = src_argb + src_stride;
int32_t stepx = src_stepx * 4;
int64_t data0, data1, data2, data3;
v16u8 src0 = {0}, src1 = {0}, src2 = {0}, src3 = {0};
v16u8 vec0, vec1, vec2, vec3;
v8u16 reg0, reg1, reg2, reg3, reg4, reg5, reg6, reg7;
v16u8 dst0;
for (x = 0; x < dst_width; x += 4) {
data0 = LD(src_argb);
data1 = LD(src_argb + stepx);
data2 = LD(src_argb + stepx * 2);
data3 = LD(src_argb + stepx * 3);
src0 = (v16u8)__msa_insert_d((v2i64)src0, 0, data0);
src0 = (v16u8)__msa_insert_d((v2i64)src0, 1, data1);
src1 = (v16u8)__msa_insert_d((v2i64)src1, 0, data2);
src1 = (v16u8)__msa_insert_d((v2i64)src1, 1, data3);
data0 = LD(nxt_argb);
data1 = LD(nxt_argb + stepx);
data2 = LD(nxt_argb + stepx * 2);
data3 = LD(nxt_argb + stepx * 3);
src2 = (v16u8)__msa_insert_d((v2i64)src2, 0, data0);
src2 = (v16u8)__msa_insert_d((v2i64)src2, 1, data1);
src3 = (v16u8)__msa_insert_d((v2i64)src3, 0, data2);
src3 = (v16u8)__msa_insert_d((v2i64)src3, 1, data3);
vec0 = (v16u8)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
vec1 = (v16u8)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
vec2 = (v16u8)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
vec3 = (v16u8)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
reg0 = __msa_hadd_u_h(vec0, vec0);
reg1 = __msa_hadd_u_h(vec1, vec1);
reg2 = __msa_hadd_u_h(vec2, vec2);
reg3 = __msa_hadd_u_h(vec3, vec3);
reg4 = (v8u16)__msa_pckev_d((v2i64)reg2, (v2i64)reg0);
reg5 = (v8u16)__msa_pckev_d((v2i64)reg3, (v2i64)reg1);
reg6 = (v8u16)__msa_pckod_d((v2i64)reg2, (v2i64)reg0);
reg7 = (v8u16)__msa_pckod_d((v2i64)reg3, (v2i64)reg1);
reg4 += reg6;
reg5 += reg7;
reg4 = (v8u16)__msa_srari_h((v8i16)reg4, 2);
reg5 = (v8u16)__msa_srari_h((v8i16)reg5, 2);
dst0 = (v16u8)__msa_pckev_b((v16i8)reg5, (v16i8)reg4);
ST_UB(dst0, dst_argb);
src_argb += stepx * 4;
nxt_argb += stepx * 4;
dst_argb += 16;
}
}
void ScaleRowDown2_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
v16u8 src0, src1, src2, src3, dst0, dst1;
(void)src_stride;
for (x = 0; x < dst_width; x += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 48);
dst0 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
dst1 = (v16u8)__msa_pckod_b((v16i8)src3, (v16i8)src2);
ST_UB2(dst0, dst1, dst, 16);
src_ptr += 64;
dst += 32;
}
}
void ScaleRowDown2Linear_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
v16u8 src0, src1, src2, src3, vec0, vec1, vec2, vec3, dst0, dst1;
(void)src_stride;
for (x = 0; x < dst_width; x += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 48);
vec0 = (v16u8)__msa_pckev_b((v16i8)src1, (v16i8)src0);
vec2 = (v16u8)__msa_pckev_b((v16i8)src3, (v16i8)src2);
vec1 = (v16u8)__msa_pckod_b((v16i8)src1, (v16i8)src0);
vec3 = (v16u8)__msa_pckod_b((v16i8)src3, (v16i8)src2);
dst0 = __msa_aver_u_b(vec1, vec0);
dst1 = __msa_aver_u_b(vec3, vec2);
ST_UB2(dst0, dst1, dst, 16);
src_ptr += 64;
dst += 32;
}
}
void ScaleRowDown2Box_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
const uint8_t* s = src_ptr;
const uint8_t* t = src_ptr + src_stride;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7, dst0, dst1;
v8u16 vec0, vec1, vec2, vec3;
for (x = 0; x < dst_width; x += 32) {
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)s, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)s, 48);
src4 = (v16u8)__msa_ld_b((v16i8*)t, 0);
src5 = (v16u8)__msa_ld_b((v16i8*)t, 16);
src6 = (v16u8)__msa_ld_b((v16i8*)t, 32);
src7 = (v16u8)__msa_ld_b((v16i8*)t, 48);
vec0 = __msa_hadd_u_h(src0, src0);
vec1 = __msa_hadd_u_h(src1, src1);
vec2 = __msa_hadd_u_h(src2, src2);
vec3 = __msa_hadd_u_h(src3, src3);
vec0 += __msa_hadd_u_h(src4, src4);
vec1 += __msa_hadd_u_h(src5, src5);
vec2 += __msa_hadd_u_h(src6, src6);
vec3 += __msa_hadd_u_h(src7, src7);
vec0 = (v8u16)__msa_srari_h((v8i16)vec0, 2);
vec1 = (v8u16)__msa_srari_h((v8i16)vec1, 2);
vec2 = (v8u16)__msa_srari_h((v8i16)vec2, 2);
vec3 = (v8u16)__msa_srari_h((v8i16)vec3, 2);
dst0 = (v16u8)__msa_pckev_b((v16i8)vec1, (v16i8)vec0);
dst1 = (v16u8)__msa_pckev_b((v16i8)vec3, (v16i8)vec2);
ST_UB2(dst0, dst1, dst, 16);
s += 64;
t += 64;
dst += 32;
}
}
void ScaleRowDown4_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
v16u8 src0, src1, src2, src3, vec0, vec1, dst0;
(void)src_stride;
for (x = 0; x < dst_width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 48);
vec0 = (v16u8)__msa_pckev_b((v16i8)src1, (v16i8)src0);
vec1 = (v16u8)__msa_pckev_b((v16i8)src3, (v16i8)src2);
dst0 = (v16u8)__msa_pckod_b((v16i8)vec1, (v16i8)vec0);
ST_UB(dst0, dst);
src_ptr += 64;
dst += 16;
}
}
void ScaleRowDown4Box_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x;
const uint8_t* s = src_ptr;
const uint8_t* t0 = s + src_stride;
const uint8_t* t1 = s + src_stride * 2;
const uint8_t* t2 = s + src_stride * 3;
v16u8 src0, src1, src2, src3, src4, src5, src6, src7, dst0;
v8u16 vec0, vec1, vec2, vec3;
v4u32 reg0, reg1, reg2, reg3;
for (x = 0; x < dst_width; x += 16) {
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)s, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)s, 48);
src4 = (v16u8)__msa_ld_b((v16i8*)t0, 0);
src5 = (v16u8)__msa_ld_b((v16i8*)t0, 16);
src6 = (v16u8)__msa_ld_b((v16i8*)t0, 32);
src7 = (v16u8)__msa_ld_b((v16i8*)t0, 48);
vec0 = __msa_hadd_u_h(src0, src0);
vec1 = __msa_hadd_u_h(src1, src1);
vec2 = __msa_hadd_u_h(src2, src2);
vec3 = __msa_hadd_u_h(src3, src3);
vec0 += __msa_hadd_u_h(src4, src4);
vec1 += __msa_hadd_u_h(src5, src5);
vec2 += __msa_hadd_u_h(src6, src6);
vec3 += __msa_hadd_u_h(src7, src7);
src0 = (v16u8)__msa_ld_b((v16i8*)t1, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)t1, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)t1, 32);
src3 = (v16u8)__msa_ld_b((v16i8*)t1, 48);
src4 = (v16u8)__msa_ld_b((v16i8*)t2, 0);
src5 = (v16u8)__msa_ld_b((v16i8*)t2, 16);
src6 = (v16u8)__msa_ld_b((v16i8*)t2, 32);
src7 = (v16u8)__msa_ld_b((v16i8*)t2, 48);
vec0 += __msa_hadd_u_h(src0, src0);
vec1 += __msa_hadd_u_h(src1, src1);
vec2 += __msa_hadd_u_h(src2, src2);
vec3 += __msa_hadd_u_h(src3, src3);
vec0 += __msa_hadd_u_h(src4, src4);
vec1 += __msa_hadd_u_h(src5, src5);
vec2 += __msa_hadd_u_h(src6, src6);
vec3 += __msa_hadd_u_h(src7, src7);
reg0 = __msa_hadd_u_w(vec0, vec0);
reg1 = __msa_hadd_u_w(vec1, vec1);
reg2 = __msa_hadd_u_w(vec2, vec2);
reg3 = __msa_hadd_u_w(vec3, vec3);
reg0 = (v4u32)__msa_srari_w((v4i32)reg0, 4);
reg1 = (v4u32)__msa_srari_w((v4i32)reg1, 4);
reg2 = (v4u32)__msa_srari_w((v4i32)reg2, 4);
reg3 = (v4u32)__msa_srari_w((v4i32)reg3, 4);
vec0 = (v8u16)__msa_pckev_h((v8i16)reg1, (v8i16)reg0);
vec1 = (v8u16)__msa_pckev_h((v8i16)reg3, (v8i16)reg2);
dst0 = (v16u8)__msa_pckev_b((v16i8)vec1, (v16i8)vec0);
ST_UB(dst0, dst);
s += 64;
t0 += 64;
t1 += 64;
t2 += 64;
dst += 16;
}
}
void ScaleRowDown38_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst,
int dst_width) {
int x, width;
uint64_t dst0;
uint32_t dst1;
v16u8 src0, src1, vec0;
v16i8 mask = {0, 3, 6, 8, 11, 14, 16, 19, 22, 24, 27, 30, 0, 0, 0, 0};
(void)src_stride;
assert(dst_width % 3 == 0);
width = dst_width / 3;
for (x = 0; x < width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)src_ptr, 16);
vec0 = (v16u8)__msa_vshf_b(mask, (v16i8)src1, (v16i8)src0);
dst0 = __msa_copy_u_d((v2i64)vec0, 0);
dst1 = __msa_copy_u_w((v4i32)vec0, 2);
SD(dst0, dst);
SW(dst1, dst + 8);
src_ptr += 32;
dst += 12;
}
}
void ScaleRowDown38_2_Box_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst_ptr,
int dst_width) {
int x, width;
const uint8_t* s = src_ptr;
const uint8_t* t = src_ptr + src_stride;
uint64_t dst0;
uint32_t dst1;
v16u8 src0, src1, src2, src3, out;
v8u16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v4u32 tmp0, tmp1, tmp2, tmp3, tmp4;
v8i16 zero = {0};
v8i16 mask = {0, 1, 2, 8, 3, 4, 5, 9};
v16i8 dst_mask = {0, 2, 16, 4, 6, 18, 8, 10, 20, 12, 14, 22, 0, 0, 0, 0};
v4u32 const_0x2AAA = (v4u32)__msa_fill_w(0x2AAA);
v4u32 const_0x4000 = (v4u32)__msa_fill_w(0x4000);
assert((dst_width % 3 == 0) && (dst_width > 0));
width = dst_width / 3;
for (x = 0; x < width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)t, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)t, 16);
vec0 = (v8u16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
vec1 = (v8u16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
vec2 = (v8u16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
vec3 = (v8u16)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
vec0 = __msa_hadd_u_h((v16u8)vec0, (v16u8)vec0);
vec1 = __msa_hadd_u_h((v16u8)vec1, (v16u8)vec1);
vec2 = __msa_hadd_u_h((v16u8)vec2, (v16u8)vec2);
vec3 = __msa_hadd_u_h((v16u8)vec3, (v16u8)vec3);
vec4 = (v8u16)__msa_vshf_h(mask, zero, (v8i16)vec0);
vec5 = (v8u16)__msa_vshf_h(mask, zero, (v8i16)vec1);
vec6 = (v8u16)__msa_vshf_h(mask, zero, (v8i16)vec2);
vec7 = (v8u16)__msa_vshf_h(mask, zero, (v8i16)vec3);
vec0 = (v8u16)__msa_pckod_w((v4i32)vec1, (v4i32)vec0);
vec1 = (v8u16)__msa_pckod_w((v4i32)vec3, (v4i32)vec2);
vec0 = (v8u16)__msa_pckod_w((v4i32)vec1, (v4i32)vec0);
tmp0 = __msa_hadd_u_w(vec4, vec4);
tmp1 = __msa_hadd_u_w(vec5, vec5);
tmp2 = __msa_hadd_u_w(vec6, vec6);
tmp3 = __msa_hadd_u_w(vec7, vec7);
tmp4 = __msa_hadd_u_w(vec0, vec0);
vec0 = (v8u16)__msa_pckev_h((v8i16)tmp1, (v8i16)tmp0);
vec1 = (v8u16)__msa_pckev_h((v8i16)tmp3, (v8i16)tmp2);
tmp0 = __msa_hadd_u_w(vec0, vec0);
tmp1 = __msa_hadd_u_w(vec1, vec1);
tmp0 *= const_0x2AAA;
tmp1 *= const_0x2AAA;
tmp4 *= const_0x4000;
tmp0 = (v4u32)__msa_srai_w((v4i32)tmp0, 16);
tmp1 = (v4u32)__msa_srai_w((v4i32)tmp1, 16);
tmp4 = (v4u32)__msa_srai_w((v4i32)tmp4, 16);
vec0 = (v8u16)__msa_pckev_h((v8i16)tmp1, (v8i16)tmp0);
vec1 = (v8u16)__msa_pckev_h((v8i16)tmp4, (v8i16)tmp4);
out = (v16u8)__msa_vshf_b(dst_mask, (v16i8)vec1, (v16i8)vec0);
dst0 = __msa_copy_u_d((v2i64)out, 0);
dst1 = __msa_copy_u_w((v4i32)out, 2);
SD(dst0, dst_ptr);
SW(dst1, dst_ptr + 8);
s += 32;
t += 32;
dst_ptr += 12;
}
}
void ScaleRowDown38_3_Box_MSA(const uint8_t* src_ptr,
ptrdiff_t src_stride,
uint8_t* dst_ptr,
int dst_width) {
int x, width;
const uint8_t* s = src_ptr;
const uint8_t* t0 = s + src_stride;
const uint8_t* t1 = s + src_stride * 2;
uint64_t dst0;
uint32_t dst1;
v16u8 src0, src1, src2, src3, src4, src5, out;
v8u16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
v4u32 tmp0, tmp1, tmp2, tmp3, tmp4;
v8u16 zero = {0};
v8i16 mask = {0, 1, 2, 8, 3, 4, 5, 9};
v16i8 dst_mask = {0, 2, 16, 4, 6, 18, 8, 10, 20, 12, 14, 22, 0, 0, 0, 0};
v4u32 const_0x1C71 = (v4u32)__msa_fill_w(0x1C71);
v4u32 const_0x2AAA = (v4u32)__msa_fill_w(0x2AAA);
assert((dst_width % 3 == 0) && (dst_width > 0));
width = dst_width / 3;
for (x = 0; x < width; x += 4) {
src0 = (v16u8)__msa_ld_b((v16i8*)s, 0);
src1 = (v16u8)__msa_ld_b((v16i8*)s, 16);
src2 = (v16u8)__msa_ld_b((v16i8*)t0, 0);
src3 = (v16u8)__msa_ld_b((v16i8*)t0, 16);
src4 = (v16u8)__msa_ld_b((v16i8*)t1, 0);
src5 = (v16u8)__msa_ld_b((v16i8*)t1, 16);
vec0 = (v8u16)__msa_ilvr_b((v16i8)src2, (v16i8)src0);
vec1 = (v8u16)__msa_ilvl_b((v16i8)src2, (v16i8)src0);
vec2 = (v8u16)__msa_ilvr_b((v16i8)src3, (v16i8)src1);
vec3 = (v8u16)__msa_ilvl_b((v16i8)src3, (v16i8)src1);
vec4 = (v8u16)__msa_ilvr_b((v16i8)zero, (v16i8)src4);
vec5 = (v8u16)__msa_ilvl_b((v16i8)zero, (v16i8)src4);
vec6 = (v8u16)__msa_ilvr_b((v16i8)zero, (v16i8)src5);
vec7 = (v8u16)__msa_ilvl_b((v16i8)zero, (v16i8)src5);
vec0 = __msa_hadd_u_h((v16u8)vec0, (v16u8)vec0);
vec1 = __msa_hadd_u_h((v16u8)vec1, (v16u8)vec1);
vec2 = __msa_hadd_u_h((v16u8)vec2, (v16u8)vec2);
vec3 = __msa_hadd_u_h((v16u8)vec3, (v16u8)vec3);
vec0 += __msa_hadd_u_h((v16u8)vec4, (v16u8)vec4);
vec1 += __msa_hadd_u_h((v16u8)vec5, (v16u8)vec5);
vec2 += __msa_hadd_u_h((v16u8)vec6, (v16u8)vec6);
vec3 += __msa_hadd_u_h((v16u8)vec7, (v16u8)vec7);
vec4 = (v8u16)__msa_vshf_h(mask, (v8i16)zero, (v8i16)vec0);
vec5 = (v8u16)__msa_vshf_h(mask, (v8i16)zero, (v8i16)vec1);
vec6 = (v8u16)__msa_vshf_h(mask, (v8i16)zero, (v8i16)vec2);
vec7 = (v8u16)__msa_vshf_h(mask, (v8i16)zero, (v8i16)vec3);
vec0 = (v8u16)__msa_pckod_w((v4i32)vec1, (v4i32)vec0);
vec1 = (v8u16)__msa_pckod_w((v4i32)vec3, (v4i32)vec2);
vec0 = (v8u16)__msa_pckod_w((v4i32)vec1, (v4i32)vec0);
tmp0 = __msa_hadd_u_w(vec4, vec4);
tmp1 = __msa_hadd_u_w(vec5, vec5);
tmp2 = __msa_hadd_u_w(vec6, vec6);
tmp3 = __msa_hadd_u_w(vec7, vec7);
tmp4 = __msa_hadd_u_w(vec0, vec0);
vec0 = (v8u16)__msa_pckev_h((v8i16)tmp1, (v8i16)tmp0);
vec1 = (v8u16)__msa_pckev_h((v8i16)tmp3, (v8i16)tmp2);
tmp0 = __msa_hadd_u_w(vec0, vec0);
tmp1 = __msa_hadd_u_w(vec1, vec1);
tmp0 *= const_0x1C71;
tmp1 *= const_0x1C71;
tmp4 *= const_0x2AAA;
tmp0 = (v4u32)__msa_srai_w((v4i32)tmp0, 16);
tmp1 = (v4u32)__msa_srai_w((v4i32)tmp1, 16);
tmp4 = (v4u32)__msa_srai_w((v4i32)tmp4, 16);
vec0 = (v8u16)__msa_pckev_h((v8i16)tmp1, (v8i16)tmp0);
vec1 = (v8u16)__msa_pckev_h((v8i16)tmp4, (v8i16)tmp4);
out = (v16u8)__msa_vshf_b(dst_mask, (v16i8)vec1, (v16i8)vec0);
dst0 = __msa_copy_u_d((v2i64)out, 0);
dst1 = __msa_copy_u_w((v4i32)out, 2);
SD(dst0, dst_ptr);
SW(dst1, dst_ptr + 8);
s += 32;
t0 += 32;
t1 += 32;
dst_ptr += 12;
}
}
void ScaleAddRow_MSA(const uint8_t* src_ptr, uint16_t* dst_ptr, int src_width) {
int x;
v16u8 src0;
v8u16 dst0, dst1;
v16i8 zero = {0};
assert(src_width > 0);
for (x = 0; x < src_width; x += 16) {
src0 = LD_UB(src_ptr);
dst0 = (v8u16)__msa_ld_h((v8i16*)dst_ptr, 0);
dst1 = (v8u16)__msa_ld_h((v8i16*)dst_ptr, 16);
dst0 += (v8u16)__msa_ilvr_b(zero, (v16i8)src0);
dst1 += (v8u16)__msa_ilvl_b(zero, (v16i8)src0);
ST_UH2(dst0, dst1, dst_ptr, 8);
src_ptr += 16;
dst_ptr += 16;
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
#endif
#endif // !defined(LIBYUV_DISABLE_MSA) && defined(__mips_msa)

1423
third_party/yuv/source/scale_neon.cc vendored
View File

File diff suppressed because it is too large Load Diff

329
third_party/yuv/source/scale_neon64.cc vendored
View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/scale.h"
#include "libyuv/row.h"
#include "libyuv/scale.h"
#include "libyuv/scale_row.h"
#ifdef __cplusplus
@ -21,15 +21,16 @@ extern "C" {
#if !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
// Read 32x1 throw away even pixels, and write 16x1.
void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
// load even pixels into v0, odd into v1
MEMACCESS(0)
"ld2 {v0.16b,v1.16b}, [%0], #32 \n"
"subs %w2, %w2, #16 \n" // 16 processed per loop
MEMACCESS(1)
"st1 {v1.16b}, [%1], #16 \n" // store odd pixels
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -41,18 +42,19 @@ void ScaleRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
}
// Read 32x1 average down and write 16x1.
void ScaleRowDown2Linear_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Linear_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b,v1.16b}, [%0], #32 \n" // load pixels and post inc
"subs %w2, %w2, #16 \n" // 16 processed per loop
"uaddlp v0.8h, v0.16b \n" // add adjacent
"uaddlp v1.8h, v1.16b \n"
"rshrn v0.8b, v0.8h, #1 \n" // downshift, round and pack
"rshrn2 v0.16b, v1.8h, #1 \n"
MEMACCESS(1)
"st1 {v0.16b}, [%1], #16 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -64,15 +66,15 @@ void ScaleRowDown2Linear_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
}
// Read 32x2 average down and write 16x1.
void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleRowDown2Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
asm volatile (
// change the stride to row 2 pointer
"add %1, %1, %0 \n"
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b,v1.16b}, [%0], #32 \n" // load row 1 and post inc
MEMACCESS(1)
"ld1 {v0.16b, v1.16b}, [%0], #32 \n" // load row 1 and post inc
"ld1 {v2.16b, v3.16b}, [%1], #32 \n" // load row 2 and post inc
"subs %w3, %w3, #16 \n" // 16 processed per loop
"uaddlp v0.8h, v0.16b \n" // row 1 add adjacent
@ -81,7 +83,6 @@ void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
"uadalp v1.8h, v3.16b \n"
"rshrn v0.8b, v0.8h, #2 \n" // downshift, round and pack
"rshrn2 v0.16b, v1.8h, #2 \n"
MEMACCESS(2)
"st1 {v0.16b}, [%2], #16 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -93,14 +94,15 @@ void ScaleRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown4_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
"subs %w2, %w2, #8 \n" // 8 processed per loop
MEMACCESS(1)
"st1 {v2.8b}, [%1], #8 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -111,20 +113,18 @@ void ScaleRowDown4_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleRowDown4Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
void ScaleRowDown4Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr,
int dst_width) {
const uint8* src_ptr1 = src_ptr + src_stride;
const uint8* src_ptr2 = src_ptr + src_stride * 2;
const uint8* src_ptr3 = src_ptr + src_stride * 3;
asm volatile (
asm volatile (
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b}, [%0], #16 \n" // load up 16x4
MEMACCESS(3)
"ld1 {v1.16b}, [%2], #16 \n"
MEMACCESS(4)
"ld1 {v2.16b}, [%3], #16 \n"
MEMACCESS(5)
"ld1 {v3.16b}, [%4], #16 \n"
"subs %w5, %w5, #4 \n"
"uaddlp v0.8h, v0.16b \n"
@ -133,7 +133,6 @@ asm volatile (
"uadalp v0.8h, v3.16b \n"
"addp v0.8h, v0.8h, v0.8h \n"
"rshrn v0.8b, v0.8h, #4 \n" // divide by 16 w/rounding
MEMACCESS(1)
"st1 {v0.s}[0], [%1], #4 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -152,15 +151,15 @@ asm volatile (
// Point samples 32 pixels to 24 pixels.
void ScaleRowDown34_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
"subs %w2, %w2, #24 \n"
"orr v2.16b, v3.16b, v3.16b \n" // order v0, v1, v2
MEMACCESS(1)
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
@ -172,15 +171,14 @@ void ScaleRowDown34_NEON(const uint8* src_ptr,
void ScaleRowDown34_0_Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
asm volatile (
"movi v20.8b, #3 \n"
"add %3, %3, %0 \n"
"1: \n"
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
MEMACCESS(3)
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%3], #32 \n" // src line 1
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%3], #32 \n" // src line 1
"subs %w2, %w2, #24 \n"
// filter src line 0 with src line 1
@ -216,8 +214,7 @@ void ScaleRowDown34_0_Box_NEON(const uint8* src_ptr,
"umlal v16.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v16.8h, #2 \n"
MEMACCESS(1)
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -232,15 +229,14 @@ void ScaleRowDown34_0_Box_NEON(const uint8* src_ptr,
void ScaleRowDown34_1_Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
asm volatile (
"movi v20.8b, #3 \n"
"add %3, %3, %0 \n"
"1: \n"
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
MEMACCESS(3)
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%3], #32 \n" // src line 1
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n" // src line 0
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%3], #32 \n" // src line 1
"subs %w2, %w2, #24 \n"
// average src line 0 with src line 1
"urhadd v0.8b, v0.8b, v4.8b \n"
@ -261,8 +257,7 @@ void ScaleRowDown34_1_Box_NEON(const uint8* src_ptr,
"umlal v4.8h, v3.8b, v20.8b \n"
"uqrshrn v2.8b, v4.8h, #2 \n"
MEMACCESS(1)
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"st3 {v0.8b,v1.8b,v2.8b}, [%1], #24 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1
@ -273,32 +268,27 @@ void ScaleRowDown34_1_Box_NEON(const uint8* src_ptr,
);
}
static uvec8 kShuf38 =
{ 0, 3, 6, 8, 11, 14, 16, 19, 22, 24, 27, 30, 0, 0, 0, 0 };
static uvec8 kShuf38_2 =
{ 0, 16, 32, 2, 18, 33, 4, 20, 34, 6, 22, 35, 0, 0, 0, 0 };
static vec16 kMult38_Div6 =
{ 65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12,
65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12 };
static vec16 kMult38_Div9 =
{ 65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18,
65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18 };
static uvec8 kShuf38 = {0, 3, 6, 8, 11, 14, 16, 19, 22, 24, 27, 30, 0, 0, 0, 0};
static uvec8 kShuf38_2 = {0, 16, 32, 2, 18, 33, 4, 20,
34, 6, 22, 35, 0, 0, 0, 0};
static vec16 kMult38_Div6 = {65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12,
65536 / 12, 65536 / 12, 65536 / 12, 65536 / 12};
static vec16 kMult38_Div9 = {65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18,
65536 / 18, 65536 / 18, 65536 / 18, 65536 / 18};
// 32 -> 12
void ScaleRowDown38_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
(void)src_stride;
asm volatile (
MEMACCESS(3)
"ld1 {v3.16b}, [%3] \n"
"1: \n"
MEMACCESS(0)
"ld1 {v0.16b,v1.16b}, [%0], #32 \n"
"ld1 {v0.16b,v1.16b}, [%0], #32 \n"
"subs %w2, %w2, #12 \n"
"tbl v2.16b, {v0.16b,v1.16b}, v3.16b \n"
MEMACCESS(1)
"tbl v2.16b, {v0.16b,v1.16b}, v3.16b \n"
"st1 {v2.8b}, [%1], #8 \n"
MEMACCESS(1)
"st1 {v2.s}[2], [%1], #4 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -312,16 +302,14 @@ void ScaleRowDown38_NEON(const uint8* src_ptr,
// 32x3 -> 12x1
void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
const uint8* src_ptr1 = src_ptr + src_stride * 2;
ptrdiff_t tmp_src_stride = src_stride;
asm volatile (
MEMACCESS(5)
"ld1 {v29.8h}, [%5] \n"
MEMACCESS(6)
"ld1 {v30.16b}, [%6] \n"
MEMACCESS(7)
"ld1 {v31.8h}, [%7] \n"
"add %2, %2, %0 \n"
"1: \n"
@ -330,12 +318,9 @@ void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
// 10 50 11 51 12 52 13 53
// 20 60 21 61 22 62 23 63
// 30 70 31 71 32 72 33 73
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"
MEMACCESS(3)
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%2], #32 \n"
MEMACCESS(4)
"ld4 {v16.8b,v17.8b,v18.8b,v19.8b}, [%3], #32 \n"
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%2], #32 \n"
"ld4 {v16.8b,v17.8b,v18.8b,v19.8b}, [%3], #32 \n"
"subs %w4, %w4, #12 \n"
// Shuffle the input data around to get align the data
@ -419,9 +404,7 @@ void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
// be adjacent
"tbl v3.16b, {v0.16b, v1.16b, v2.16b}, v30.16b \n"
MEMACCESS(1)
"st1 {v3.8b}, [%1], #8 \n"
MEMACCESS(1)
"st1 {v3.s}[2], [%1], #4 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -441,13 +424,12 @@ void OMITFP ScaleRowDown38_3_Box_NEON(const uint8* src_ptr,
// 32x2 -> 12x1
void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst_ptr, int dst_width) {
uint8* dst_ptr,
int dst_width) {
// TODO(fbarchard): use src_stride directly for clang 3.5+.
ptrdiff_t tmp_src_stride = src_stride;
asm volatile (
MEMACCESS(4)
"ld1 {v30.8h}, [%4] \n"
MEMACCESS(5)
"ld1 {v31.16b}, [%5] \n"
"add %2, %2, %0 \n"
"1: \n"
@ -456,10 +438,8 @@ void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
// 10 50 11 51 12 52 13 53
// 20 60 21 61 22 62 23 63
// 30 70 31 71 32 72 33 73
MEMACCESS(0)
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"
MEMACCESS(3)
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%2], #32 \n"
"ld4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%0], #32 \n"
"ld4 {v4.8b,v5.8b,v6.8b,v7.8b}, [%2], #32 \n"
"subs %w3, %w3, #12 \n"
// Shuffle the input data around to get align the data
@ -529,9 +509,7 @@ void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
"tbl v3.16b, {v0.16b, v1.16b, v2.16b}, v31.16b \n"
MEMACCESS(1)
"st1 {v3.8b}, [%1], #8 \n"
MEMACCESS(1)
"st1 {v3.s}[2], [%1], #4 \n"
"b.gt 1b \n"
: "+r"(src_ptr), // %0
@ -545,8 +523,11 @@ void ScaleRowDown38_2_Box_NEON(const uint8* src_ptr,
);
}
void ScaleAddRows_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint16* dst_ptr, int src_width, int src_height) {
void ScaleAddRows_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint16* dst_ptr,
int src_width,
int src_height) {
const uint8* src_tmp;
asm volatile (
"1: \n"
@ -556,13 +537,11 @@ void ScaleAddRows_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
"eor v3.16b, v3.16b, v3.16b \n"
"2: \n"
// load 16 pixels into q0
MEMACCESS(0)
"ld1 {v0.16b}, [%0], %3 \n"
"uaddw2 v3.8h, v3.8h, v0.16b \n"
"uaddw v2.8h, v2.8h, v0.8b \n"
"subs w12, w12, #1 \n"
"b.gt 2b \n"
MEMACCESS(2)
"st1 {v2.8h, v3.8h}, [%2], #32 \n" // store pixels
"add %1, %1, #16 \n"
"subs %w4, %w4, #16 \n" // 16 processed per loop
@ -578,23 +557,30 @@ void ScaleAddRows_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
// clang-format off
// TODO(Yang Zhang): Investigate less load instructions for
// the x/dx stepping
#define LOAD2_DATA8_LANE(n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5 \n" \
"add %3, %3, %4 \n" \
MEMACCESS(6) \
"ld2 {v4.b, v5.b}["#n"], [%6] \n"
#define LOAD2_DATA8_LANE(n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5 \n" \
"add %3, %3, %4 \n" \
"ld2 {v4.b, v5.b}[" #n "], [%6] \n"
// clang-format on
void ScaleFilterCols_NEON(uint8* dst_ptr, const uint8* src_ptr,
int dst_width, int x, int dx) {
// The NEON version mimics this formula (from row_common.cc):
// #define BLENDER(a, b, f) (uint8)((int)(a) +
// ((((int)((f)) * ((int)(b) - (int)(a))) + 0x8000) >> 16))
void ScaleFilterCols_NEON(uint8* dst_ptr,
const uint8* src_ptr,
int dst_width,
int x,
int dx) {
int dx_offset[4] = {0, 1, 2, 3};
int* tmp = dx_offset;
const uint8* src_tmp = src_ptr;
int64 dst_width64 = (int64) dst_width; // Work around ios 64 bit warning.
int64 x64 = (int64) x;
int64 dx64 = (int64) dx;
int64 x64 = (int64)x;
int64 dx64 = (int64)dx;
asm volatile (
"dup v0.4s, %w3 \n" // x
"dup v1.4s, %w4 \n" // dx
@ -626,12 +612,11 @@ void ScaleFilterCols_NEON(uint8* dst_ptr, const uint8* src_ptr,
"ushll2 v6.4s, v6.8h, #0 \n"
"mul v16.4s, v16.4s, v7.4s \n"
"mul v17.4s, v17.4s, v6.4s \n"
"shrn v6.4h, v16.4s, #16 \n"
"shrn2 v6.8h, v17.4s, #16 \n"
"rshrn v6.4h, v16.4s, #16 \n"
"rshrn2 v6.8h, v17.4s, #16 \n"
"add v4.8h, v4.8h, v6.8h \n"
"xtn v4.8b, v4.8h \n"
MEMACCESS(0)
"st1 {v4.8b}, [%0], #8 \n" // store pixels
"add v1.4s, v1.4s, v0.4s \n"
"add v2.4s, v2.4s, v0.4s \n"
@ -639,7 +624,7 @@ void ScaleFilterCols_NEON(uint8* dst_ptr, const uint8* src_ptr,
"b.gt 1b \n"
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
"+r"(dst_width64), // %2
"+r"(dst_width), // %2
"+r"(x64), // %3
"+r"(dx64), // %4
"+r"(tmp), // %5
@ -654,9 +639,11 @@ void ScaleFilterCols_NEON(uint8* dst_ptr, const uint8* src_ptr,
// 16x2 -> 16x1
void ScaleFilterRows_NEON(uint8* dst_ptr,
const uint8* src_ptr, ptrdiff_t src_stride,
int dst_width, int source_y_fraction) {
int y_fraction = 256 - source_y_fraction;
const uint8* src_ptr,
ptrdiff_t src_stride,
int dst_width,
int source_y_fraction) {
int y_fraction = 256 - source_y_fraction;
asm volatile (
"cmp %w4, #0 \n"
"b.eq 100f \n"
@ -672,9 +659,7 @@ void ScaleFilterRows_NEON(uint8* dst_ptr,
"dup v4.8b, %w5 \n"
// General purpose row blend.
"1: \n"
MEMACCESS(1)
"ld1 {v0.16b}, [%1], #16 \n"
MEMACCESS(2)
"ld1 {v1.16b}, [%2], #16 \n"
"subs %w3, %w3, #16 \n"
"umull v6.8h, v0.8b, v4.8b \n"
@ -683,63 +668,50 @@ void ScaleFilterRows_NEON(uint8* dst_ptr,
"umlal2 v7.8h, v1.16b, v5.16b \n"
"rshrn v0.8b, v6.8h, #8 \n"
"rshrn2 v0.16b, v7.8h, #8 \n"
MEMACCESS(0)
"st1 {v0.16b}, [%0], #16 \n"
"b.gt 1b \n"
"b 99f \n"
// Blend 25 / 75.
"25: \n"
MEMACCESS(1)
"ld1 {v0.16b}, [%1], #16 \n"
MEMACCESS(2)
"ld1 {v1.16b}, [%2], #16 \n"
"subs %w3, %w3, #16 \n"
"urhadd v0.16b, v0.16b, v1.16b \n"
"urhadd v0.16b, v0.16b, v1.16b \n"
MEMACCESS(0)
"st1 {v0.16b}, [%0], #16 \n"
"b.gt 25b \n"
"b 99f \n"
// Blend 50 / 50.
"50: \n"
MEMACCESS(1)
"ld1 {v0.16b}, [%1], #16 \n"
MEMACCESS(2)
"ld1 {v1.16b}, [%2], #16 \n"
"subs %w3, %w3, #16 \n"
"urhadd v0.16b, v0.16b, v1.16b \n"
MEMACCESS(0)
"st1 {v0.16b}, [%0], #16 \n"
"b.gt 50b \n"
"b 99f \n"
// Blend 75 / 25.
"75: \n"
MEMACCESS(1)
"ld1 {v1.16b}, [%1], #16 \n"
MEMACCESS(2)
"ld1 {v0.16b}, [%2], #16 \n"
"subs %w3, %w3, #16 \n"
"urhadd v0.16b, v0.16b, v1.16b \n"
"urhadd v0.16b, v0.16b, v1.16b \n"
MEMACCESS(0)
"st1 {v0.16b}, [%0], #16 \n"
"b.gt 75b \n"
"b 99f \n"
// Blend 100 / 0 - Copy row unchanged.
"100: \n"
MEMACCESS(1)
"ld1 {v0.16b}, [%1], #16 \n"
"subs %w3, %w3, #16 \n"
MEMACCESS(0)
"st1 {v0.16b}, [%0], #16 \n"
"b.gt 100b \n"
"99: \n"
MEMACCESS(0)
"st1 {v0.b}[15], [%0] \n"
: "+r"(dst_ptr), // %0
"+r"(src_ptr), // %1
@ -752,19 +724,18 @@ void ScaleFilterRows_NEON(uint8* dst_ptr,
);
}
void ScaleARGBRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleARGBRowDown2_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
// load even pixels into q0, odd into q1
MEMACCESS (0)
"ld2 {v0.4s, v1.4s}, [%0], #32 \n"
MEMACCESS (0)
"ld2 {v2.4s, v3.4s}, [%0], #32 \n"
"subs %w2, %w2, #8 \n" // 8 processed per loop
MEMACCESS (1)
"st1 {v1.16b}, [%1], #16 \n" // store odd pixels
MEMACCESS (1)
"st1 {v3.16b}, [%1], #16 \n"
"b.gt 1b \n"
: "+r" (src_ptr), // %0
@ -775,11 +746,13 @@ void ScaleARGBRowDown2_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
);
}
void ScaleARGBRowDown2Linear_NEON(const uint8* src_argb, ptrdiff_t src_stride,
uint8* dst_argb, int dst_width) {
void ScaleARGBRowDown2Linear_NEON(const uint8* src_argb,
ptrdiff_t src_stride,
uint8* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
MEMACCESS (0)
// load 8 ARGB pixels.
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n"
"subs %w2, %w2, #8 \n" // 8 processed per loop.
@ -791,7 +764,6 @@ void ScaleARGBRowDown2Linear_NEON(const uint8* src_argb, ptrdiff_t src_stride,
"rshrn v1.8b, v1.8h, #1 \n"
"rshrn v2.8b, v2.8h, #1 \n"
"rshrn v3.8b, v3.8h, #1 \n"
MEMACCESS (1)
"st4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%1], #32 \n"
"b.gt 1b \n"
: "+r"(src_argb), // %0
@ -802,20 +774,20 @@ void ScaleARGBRowDown2Linear_NEON(const uint8* src_argb, ptrdiff_t src_stride,
);
}
void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
uint8* dst, int dst_width) {
void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr,
ptrdiff_t src_stride,
uint8* dst,
int dst_width) {
asm volatile (
// change the stride to row 2 pointer
"add %1, %1, %0 \n"
"1: \n"
MEMACCESS (0)
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 8 ARGB pixels.
"subs %w3, %w3, #8 \n" // 8 processed per loop.
"uaddlp v0.8h, v0.16b \n" // B 16 bytes -> 8 shorts.
"uaddlp v1.8h, v1.16b \n" // G 16 bytes -> 8 shorts.
"uaddlp v2.8h, v2.16b \n" // R 16 bytes -> 8 shorts.
"uaddlp v3.8h, v3.16b \n" // A 16 bytes -> 8 shorts.
MEMACCESS (1)
"ld4 {v16.16b,v17.16b,v18.16b,v19.16b}, [%1], #64 \n" // load 8 more ARGB pixels.
"uadalp v0.8h, v16.16b \n" // B 16 bytes -> 8 shorts.
"uadalp v1.8h, v17.16b \n" // G 16 bytes -> 8 shorts.
@ -825,7 +797,6 @@ void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
"rshrn v1.8b, v1.8h, #2 \n"
"rshrn v2.8b, v2.8h, #2 \n"
"rshrn v3.8b, v3.8h, #2 \n"
MEMACCESS (2)
"st4 {v0.8b,v1.8b,v2.8b,v3.8b}, [%2], #32 \n"
"b.gt 1b \n"
: "+r" (src_ptr), // %0
@ -839,20 +810,19 @@ void ScaleARGBRowDown2Box_NEON(const uint8* src_ptr, ptrdiff_t src_stride,
// Reads 4 pixels at a time.
// Alignment requirement: src_argb 4 byte aligned.
void ScaleARGBRowDownEven_NEON(const uint8* src_argb, ptrdiff_t src_stride,
int src_stepx, uint8* dst_argb, int dst_width) {
void ScaleARGBRowDownEven_NEON(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb,
int dst_width) {
(void)src_stride;
asm volatile (
"1: \n"
MEMACCESS(0)
"ld1 {v0.s}[0], [%0], %3 \n"
MEMACCESS(0)
"ld1 {v0.s}[1], [%0], %3 \n"
MEMACCESS(0)
"ld1 {v0.s}[2], [%0], %3 \n"
MEMACCESS(0)
"ld1 {v0.s}[3], [%0], %3 \n"
"subs %w2, %w2, #4 \n" // 4 pixels per loop.
MEMACCESS(1)
"st1 {v0.16b}, [%1], #16 \n"
"b.gt 1b \n"
: "+r"(src_argb), // %0
@ -867,27 +837,21 @@ void ScaleARGBRowDownEven_NEON(const uint8* src_argb, ptrdiff_t src_stride,
// Alignment requirement: src_argb 4 byte aligned.
// TODO(Yang Zhang): Might be worth another optimization pass in future.
// It could be upgraded to 8 pixels at a time to start with.
void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb, ptrdiff_t src_stride,
void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb,
ptrdiff_t src_stride,
int src_stepx,
uint8* dst_argb, int dst_width) {
uint8* dst_argb,
int dst_width) {
asm volatile (
"add %1, %1, %0 \n"
"1: \n"
MEMACCESS(0)
"ld1 {v0.8b}, [%0], %4 \n" // Read 4 2x2 blocks -> 2x1
MEMACCESS(1)
"ld1 {v1.8b}, [%1], %4 \n"
MEMACCESS(0)
"ld1 {v2.8b}, [%0], %4 \n"
MEMACCESS(1)
"ld1 {v3.8b}, [%1], %4 \n"
MEMACCESS(0)
"ld1 {v4.8b}, [%0], %4 \n"
MEMACCESS(1)
"ld1 {v5.8b}, [%1], %4 \n"
MEMACCESS(0)
"ld1 {v6.8b}, [%0], %4 \n"
MEMACCESS(1)
"ld1 {v7.8b}, [%1], %4 \n"
"uaddl v0.8h, v0.8b, v1.8b \n"
"uaddl v2.8h, v2.8b, v3.8b \n"
@ -904,7 +868,6 @@ void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb, ptrdiff_t src_stride,
"rshrn v0.8b, v0.8h, #2 \n" // first 2 pixels.
"rshrn2 v0.16b, v4.8h, #2 \n" // next 2 pixels.
"subs %w3, %w3, #4 \n" // 4 pixels per loop.
MEMACCESS(2)
"st1 {v0.16b}, [%2], #16 \n"
"b.gt 1b \n"
: "+r"(src_argb), // %0
@ -916,21 +879,24 @@ void ScaleARGBRowDownEvenBox_NEON(const uint8* src_argb, ptrdiff_t src_stride,
);
}
// clang-format off
// TODO(Yang Zhang): Investigate less load instructions for
// the x/dx stepping
#define LOAD1_DATA32_LANE(vn, n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5, lsl #2 \n" \
"add %3, %3, %4 \n" \
MEMACCESS(6) \
"ld1 {"#vn".s}["#n"], [%6] \n"
#define LOAD1_DATA32_LANE(vn, n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5, lsl #2 \n" \
"add %3, %3, %4 \n" \
"ld1 {" #vn ".s}[" #n "], [%6] \n"
// clang-format on
void ScaleARGBCols_NEON(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBCols_NEON(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
const uint8* src_tmp = src_argb;
int64 dst_width64 = (int64) dst_width; // Work around ios 64 bit warning.
int64 x64 = (int64) x;
int64 dx64 = (int64) dx;
int64 x64 = (int64)x;
int64 dx64 = (int64)dx;
int64 tmp64;
asm volatile (
"1: \n"
@ -943,13 +909,12 @@ void ScaleARGBCols_NEON(uint8* dst_argb, const uint8* src_argb,
LOAD1_DATA32_LANE(v1, 2)
LOAD1_DATA32_LANE(v1, 3)
MEMACCESS(0)
"st1 {v0.4s, v1.4s}, [%0], #32 \n" // store pixels
"subs %w2, %w2, #8 \n" // 8 processed per loop
"b.gt 1b \n"
"b.gt 1b \n"
: "+r"(dst_argb), // %0
"+r"(src_argb), // %1
"+r"(dst_width64), // %2
"+r"(dst_width), // %2
"+r"(x64), // %3
"+r"(dx64), // %4
"=&r"(tmp64), // %5
@ -961,23 +926,26 @@ void ScaleARGBCols_NEON(uint8* dst_argb, const uint8* src_argb,
#undef LOAD1_DATA32_LANE
// clang-format off
// TODO(Yang Zhang): Investigate less load instructions for
// the x/dx stepping
#define LOAD2_DATA32_LANE(vn1, vn2, n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5, lsl #2 \n" \
"add %3, %3, %4 \n" \
MEMACCESS(6) \
"ld2 {"#vn1".s, "#vn2".s}["#n"], [%6] \n"
#define LOAD2_DATA32_LANE(vn1, vn2, n) \
"lsr %5, %3, #16 \n" \
"add %6, %1, %5, lsl #2 \n" \
"add %3, %3, %4 \n" \
"ld2 {" #vn1 ".s, " #vn2 ".s}[" #n "], [%6] \n"
// clang-format on
void ScaleARGBFilterCols_NEON(uint8* dst_argb, const uint8* src_argb,
int dst_width, int x, int dx) {
void ScaleARGBFilterCols_NEON(uint8* dst_argb,
const uint8* src_argb,
int dst_width,
int x,
int dx) {
int dx_offset[4] = {0, 1, 2, 3};
int* tmp = dx_offset;
const uint8* src_tmp = src_argb;
int64 dst_width64 = (int64) dst_width; // Work around ios 64 bit warning.
int64 x64 = (int64) x;
int64 dx64 = (int64) dx;
int64 x64 = (int64)x;
int64 dx64 = (int64)dx;
asm volatile (
"dup v0.4s, %w3 \n" // x
"dup v1.4s, %w4 \n" // dx
@ -1014,14 +982,13 @@ void ScaleARGBFilterCols_NEON(uint8* dst_argb, const uint8* src_argb,
"shrn v0.8b, v16.8h, #7 \n"
"shrn2 v0.16b, v17.8h, #7 \n"
MEMACCESS(0)
"st1 {v0.4s}, [%0], #16 \n" // store pixels
"add v5.4s, v5.4s, v6.4s \n"
"subs %w2, %w2, #4 \n" // 4 processed per loop
"b.gt 1b \n"
: "+r"(dst_argb), // %0
"+r"(src_argb), // %1
"+r"(dst_width64), // %2
"+r"(dst_width), // %2
"+r"(x64), // %3
"+r"(dx64), // %4
"+r"(tmp), // %5

855
third_party/yuv/source/scale_win.cc vendored
View File

File diff suppressed because it is too large Load Diff

38
third_party/yuv/source/video_common.cc vendored
View File

@ -8,7 +8,6 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "libyuv/video_common.h"
#ifdef __cplusplus
@ -24,24 +23,24 @@ struct FourCCAliasEntry {
};
static const struct FourCCAliasEntry kFourCCAliases[] = {
{FOURCC_IYUV, FOURCC_I420},
{FOURCC_YU12, FOURCC_I420},
{FOURCC_YU16, FOURCC_I422},
{FOURCC_YU24, FOURCC_I444},
{FOURCC_YUYV, FOURCC_YUY2},
{FOURCC_YUVS, FOURCC_YUY2}, // kCMPixelFormat_422YpCbCr8_yuvs
{FOURCC_HDYC, FOURCC_UYVY},
{FOURCC_2VUY, FOURCC_UYVY}, // kCMPixelFormat_422YpCbCr8
{FOURCC_JPEG, FOURCC_MJPG}, // Note: JPEG has DHT while MJPG does not.
{FOURCC_DMB1, FOURCC_MJPG},
{FOURCC_BA81, FOURCC_BGGR}, // deprecated.
{FOURCC_RGB3, FOURCC_RAW },
{FOURCC_BGR3, FOURCC_24BG},
{FOURCC_CM32, FOURCC_BGRA}, // kCMPixelFormat_32ARGB
{FOURCC_CM24, FOURCC_RAW }, // kCMPixelFormat_24RGB
{FOURCC_L555, FOURCC_RGBO}, // kCMPixelFormat_16LE555
{FOURCC_L565, FOURCC_RGBP}, // kCMPixelFormat_16LE565
{FOURCC_5551, FOURCC_RGBO}, // kCMPixelFormat_16LE5551
{FOURCC_IYUV, FOURCC_I420},
{FOURCC_YU12, FOURCC_I420},
{FOURCC_YU16, FOURCC_I422},
{FOURCC_YU24, FOURCC_I444},
{FOURCC_YUYV, FOURCC_YUY2},
{FOURCC_YUVS, FOURCC_YUY2}, // kCMPixelFormat_422YpCbCr8_yuvs
{FOURCC_HDYC, FOURCC_UYVY},
{FOURCC_2VUY, FOURCC_UYVY}, // kCMPixelFormat_422YpCbCr8
{FOURCC_JPEG, FOURCC_MJPG}, // Note: JPEG has DHT while MJPG does not.
{FOURCC_DMB1, FOURCC_MJPG},
{FOURCC_BA81, FOURCC_BGGR}, // deprecated.
{FOURCC_RGB3, FOURCC_RAW},
{FOURCC_BGR3, FOURCC_24BG},
{FOURCC_CM32, FOURCC_BGRA}, // kCMPixelFormat_32ARGB
{FOURCC_CM24, FOURCC_RAW}, // kCMPixelFormat_24RGB
{FOURCC_L555, FOURCC_RGBO}, // kCMPixelFormat_16LE555
{FOURCC_L565, FOURCC_RGBP}, // kCMPixelFormat_16LE565
{FOURCC_5551, FOURCC_RGBO}, // kCMPixelFormat_16LE5551
};
// TODO(fbarchard): Consider mapping kCMPixelFormat_32BGRA to FOURCC_ARGB.
// {FOURCC_BGRA, FOURCC_ARGB}, // kCMPixelFormat_32BGRA
@ -62,4 +61,3 @@ uint32 CanonicalFourCC(uint32 fourcc) {
} // extern "C"
} // namespace libyuv
#endif