diff options
| author | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2022-05-12 15:59:20 +0200 |
|---|---|---|
| committer | Allan Sandfeld Jensen <allan.jensen@qt.io> | 2022-05-25 06:57:22 +0000 |
| commit | f7eaed5286974984ba5f9e3189d8f49d03e99f81 (patch) | |
| tree | caed19b2af2024f35449fb0b781d0a25e09d4f8f /chromium/third_party/libvpx | |
| parent | 9729c4479fe23554eae6e6dd1f30ff488f470c84 (diff) | |
| download | qtwebengine-chromium-f7eaed5286974984ba5f9e3189d8f49d03e99f81.tar.gz | |
BASELINE: Update Chromium to 100.0.4896.167
Change-Id: I98cbeb5d7543d966ffe04d8cefded0c493a11333
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>
Diffstat (limited to 'chromium/third_party/libvpx')
103 files changed, 4267 insertions, 323 deletions
diff --git a/chromium/third_party/libvpx/README.chromium b/chromium/third_party/libvpx/README.chromium index a6f745012c8..eefa8d924ec 100644 --- a/chromium/third_party/libvpx/README.chromium +++ b/chromium/third_party/libvpx/README.chromium @@ -6,9 +6,9 @@ License: BSD License File: source/libvpx/LICENSE Security Critical: yes -Date: Tuesday December 07 2021 +Date: Wednesday February 09 2022 Branch: main -Commit: ab35ee100a38347433af24df05a5e1578172a2ae +Commit: df0d06de6d3b64e35b9e75ad72c571af061bc7b3 Description: Contains the sources used to compile libvpx binaries used by Google Chrome and diff --git a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.asm b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.asm index 0322157075e..736cf9a66f9 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.asm @@ -12,6 +12,8 @@ .set ARCH_X86_64 , 0 .set VPX_ARCH_PPC , 0 .set ARCH_PPC , 0 +.set VPX_ARCH_LOONGARCH , 0 +.set ARCH_LOONGARCH , 0 .set HAVE_NEON , 1 .set HAVE_NEON_ASM , 1 .set HAVE_MIPS32 , 0 @@ -29,6 +31,8 @@ .set HAVE_AVX512 , 0 .set HAVE_VSX , 0 .set HAVE_MMI , 0 +.set HAVE_LSX , 0 +.set HAVE_LASX , 0 .set HAVE_VPX_PORTS , 1 .set HAVE_PTHREAD_H , 1 .set HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.h b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.h index 969f1864607..eb04fc610fc 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 1 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_dsp_rtcd.h index 9c31497876c..cba0a393d4c 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/ios/arm-neon/vpx_dsp_rtcd.h @@ -812,11 +812,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.asm b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.asm index 806d43d4298..105518b9dd2 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.asm @@ -12,6 +12,8 @@ .set ARCH_X86_64 , 0 .set VPX_ARCH_PPC , 0 .set ARCH_PPC , 0 +.set VPX_ARCH_LOONGARCH , 0 +.set ARCH_LOONGARCH , 0 .set HAVE_NEON , 1 .set HAVE_NEON_ASM , 0 .set HAVE_MIPS32 , 0 @@ -29,6 +31,8 @@ .set HAVE_AVX512 , 0 .set HAVE_VSX , 0 .set HAVE_MMI , 0 +.set HAVE_LSX , 0 +.set HAVE_LASX , 0 .set HAVE_VPX_PORTS , 1 .set HAVE_PTHREAD_H , 1 .set HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.h b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.h index 2c1933eac37..0d6993f6572 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_dsp_rtcd.h index 9c31497876c..cba0a393d4c 100644 --- a/chromium/third_party/libvpx/source/config/ios/arm64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/ios/arm64/vpx_dsp_rtcd.h @@ -812,11 +812,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.asm index db796ea5984..f58a65019a7 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 1 .equ HAVE_NEON_ASM , 1 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.h index a5c309a5729..251793efc61 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 1 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_dsp_rtcd.h index 95df12a235f..93512dfba0d 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-cpu-detect/vpx_dsp_rtcd.h @@ -1042,15 +1042,15 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); RTCD_EXTERN int16_t (*vpx_int_pro_col)(const uint8_t* ref, const int width); -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -RTCD_EXTERN void (*vpx_int_pro_row)(int16_t* hbuf, +RTCD_EXTERN void (*vpx_int_pro_row)(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.asm index 88e5c40089f..e7963209f31 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 1 .equ HAVE_NEON_ASM , 1 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.h index 75556cceb20..e7dc716c7b8 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 1 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_dsp_rtcd.h index 71eb333a0bd..bf1619dfc79 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon-highbd/vpx_dsp_rtcd.h @@ -3512,11 +3512,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.asm index 54dae08f8cd..0e6ba6a4389 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 1 .equ HAVE_NEON_ASM , 1 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.h index 969f1864607..eb04fc610fc 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 1 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_dsp_rtcd.h index 9c31497876c..cba0a393d4c 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm-neon/vpx_dsp_rtcd.h @@ -812,11 +812,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.asm index 451803d457f..d5099be7e9f 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 0 .equ HAVE_NEON_ASM , 0 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.h index ee3225fc9d9..4ed1444715f 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm/vpx_dsp_rtcd.h index 792d50fbdef..ccb5eac407f 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm/vpx_dsp_rtcd.h @@ -535,7 +535,7 @@ void vpx_idct8x8_64_add_c(const tran_low_t* input, uint8_t* dest, int stride); int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_c -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.asm index ea77627738b..36713173bbb 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 1 .equ HAVE_NEON_ASM , 0 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.h index e0e39fa09fc..ad8b54b9091 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_dsp_rtcd.h index 71eb333a0bd..bf1619dfc79 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm64-highbd/vpx_dsp_rtcd.h @@ -3512,11 +3512,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.asm index b909ad61270..cfc72f2b3e6 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 1 .equ HAVE_NEON_ASM , 0 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.h index 2c1933eac37..0d6993f6572 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_dsp_rtcd.h index 9c31497876c..cba0a393d4c 100644 --- a/chromium/third_party/libvpx/source/config/linux/arm64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/arm64/vpx_dsp_rtcd.h @@ -812,11 +812,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.asm index 00712e52bbb..11500287265 100644 --- a/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.asm @@ -11,6 +11,8 @@ .equ ARCH_X86_64 , 0 .equ VPX_ARCH_PPC , 0 .equ ARCH_PPC , 0 +.equ VPX_ARCH_LOONGARCH , 0 +.equ ARCH_LOONGARCH , 0 .equ HAVE_NEON , 0 .equ HAVE_NEON_ASM , 0 .equ HAVE_MIPS32 , 0 @@ -28,6 +30,8 @@ .equ HAVE_AVX512 , 0 .equ HAVE_VSX , 0 .equ HAVE_MMI , 0 +.equ HAVE_LSX , 0 +.equ HAVE_LASX , 0 .equ HAVE_VPX_PORTS , 1 .equ HAVE_PTHREAD_H , 1 .equ HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.h index fddb76bd2f9..b7b528b92d6 100644 --- a/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/generic/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/generic/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/generic/vpx_dsp_rtcd.h index 8ba4d88055d..382fc401e44 100644 --- a/chromium/third_party/libvpx/source/config/linux/generic/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/generic/vpx_dsp_rtcd.h @@ -2820,7 +2820,7 @@ void vpx_idct8x8_64_add_c(const tran_low_t* input, uint8_t* dest, int stride); int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_c -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.asm index a5bda29c768..aa728c7a484 100644 --- a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 0 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 1 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.h index 11c14feef79..d123909e856 100644 --- a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_dsp_rtcd.h index 096e3a863e5..7df60970583 100644 --- a/chromium/third_party/libvpx/source/config/linux/ia32/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/ia32/vpx_dsp_rtcd.h @@ -5110,11 +5110,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_config.h index f79adefd882..09a17ef9d48 100644 --- a/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_dsp_rtcd.h index 63a0f75b874..5cd7f438036 100644 --- a/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/mips64el/vpx_dsp_rtcd.h @@ -535,7 +535,7 @@ void vpx_idct8x8_64_add_c(const tran_low_t* input, uint8_t* dest, int stride); int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_c -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_config.h index a0cdb81b4e9..17d362c27c2 100644 --- a/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 1 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_dsp_rtcd.h index 63a0f75b874..5cd7f438036 100644 --- a/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/mipsel/vpx_dsp_rtcd.h @@ -535,7 +535,7 @@ void vpx_idct8x8_64_add_c(const tran_low_t* input, uint8_t* dest, int stride); int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_c -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.asm b/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.asm index b9d78102159..982b2743511 100644 --- a/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 1 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 1 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.h b/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.h index 7c3c3388cf3..b3649442c2b 100644 --- a/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/linux/x64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 1 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/linux/x64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/linux/x64/vpx_dsp_rtcd.h index ed632331485..b0a87f4fa8b 100644 --- a/chromium/third_party/libvpx/source/config/linux/x64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/linux/x64/vpx_dsp_rtcd.h @@ -5187,11 +5187,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.asm b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.asm index a5bda29c768..aa728c7a484 100644 --- a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 0 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 1 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.h b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.h index 11c14feef79..d123909e856 100644 --- a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_dsp_rtcd.h index 096e3a863e5..7df60970583 100644 --- a/chromium/third_party/libvpx/source/config/mac/ia32/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/mac/ia32/vpx_dsp_rtcd.h @@ -5110,11 +5110,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.asm b/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.asm index b9d78102159..982b2743511 100644 --- a/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 1 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 1 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.h b/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.h index 7c3c3388cf3..b3649442c2b 100644 --- a/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/mac/x64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 1 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/mac/x64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/mac/x64/vpx_dsp_rtcd.h index ed632331485..b0a87f4fa8b 100644 --- a/chromium/third_party/libvpx/source/config/mac/x64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/mac/x64/vpx_dsp_rtcd.h @@ -5187,11 +5187,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/nacl/vpx_config.h b/chromium/third_party/libvpx/source/config/nacl/vpx_config.h index fddb76bd2f9..b7b528b92d6 100644 --- a/chromium/third_party/libvpx/source/config/nacl/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/nacl/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 1 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/nacl/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/nacl/vpx_dsp_rtcd.h index 8ba4d88055d..382fc401e44 100644 --- a/chromium/third_party/libvpx/source/config/nacl/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/nacl/vpx_dsp_rtcd.h @@ -2820,7 +2820,7 @@ void vpx_idct8x8_64_add_c(const tran_low_t* input, uint8_t* dest, int stride); int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_c -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/vpx_version.h b/chromium/third_party/libvpx/source/config/vpx_version.h index c78057a2721..9bba788a767 100644 --- a/chromium/third_party/libvpx/source/config/vpx_version.h +++ b/chromium/third_party/libvpx/source/config/vpx_version.h @@ -2,8 +2,8 @@ #define VERSION_MAJOR 1 #define VERSION_MINOR 11 #define VERSION_PATCH 0 -#define VERSION_EXTRA "53-gab35ee100" +#define VERSION_EXTRA "98-gdf0d06de6" #define VERSION_PACKED \ ((VERSION_MAJOR << 16) | (VERSION_MINOR << 8) | (VERSION_PATCH)) -#define VERSION_STRING_NOSP "v1.11.0-53-gab35ee100" -#define VERSION_STRING " v1.11.0-53-gab35ee100" +#define VERSION_STRING_NOSP "v1.11.0-98-gdf0d06de6" +#define VERSION_STRING " v1.11.0-98-gdf0d06de6" diff --git a/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.asm b/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.asm index 406fe8f0160..c0be5c1f4b0 100644 --- a/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.asm @@ -12,6 +12,8 @@ .set ARCH_X86_64 , 0 .set VPX_ARCH_PPC , 0 .set ARCH_PPC , 0 +.set VPX_ARCH_LOONGARCH , 0 +.set ARCH_LOONGARCH , 0 .set HAVE_NEON , 1 .set HAVE_NEON_ASM , 0 .set HAVE_MIPS32 , 0 @@ -29,6 +31,8 @@ .set HAVE_AVX512 , 0 .set HAVE_VSX , 0 .set HAVE_MMI , 0 +.set HAVE_LSX , 0 +.set HAVE_LASX , 0 .set HAVE_VPX_PORTS , 1 .set HAVE_PTHREAD_H , 0 .set HAVE_UNISTD_H , 0 diff --git a/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.h b/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.h index 08aecd33cb7..6a71201002b 100644 --- a/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/win/arm64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 1 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 0 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/win/arm64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/win/arm64/vpx_dsp_rtcd.h index 71eb333a0bd..bf1619dfc79 100644 --- a/chromium/third_party/libvpx/source/config/win/arm64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/win/arm64/vpx_dsp_rtcd.h @@ -3512,11 +3512,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_neon(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_neon -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_neon(int16_t* hbuf, +void vpx_int_pro_row_neon(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.asm b/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.asm index 2f5bd3fc130..3ccb43b8abf 100644 --- a/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 0 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 0 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.h b/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.h index cec887dd809..445e6f388a3 100644 --- a/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/win/ia32/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 0 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 0 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/win/ia32/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/win/ia32/vpx_dsp_rtcd.h index 096e3a863e5..7df60970583 100644 --- a/chromium/third_party/libvpx/source/config/win/ia32/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/win/ia32/vpx_dsp_rtcd.h @@ -5110,11 +5110,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/config/win/x64/vpx_config.asm b/chromium/third_party/libvpx/source/config/win/x64/vpx_config.asm index 507ab543f3e..5f505644c45 100644 --- a/chromium/third_party/libvpx/source/config/win/x64/vpx_config.asm +++ b/chromium/third_party/libvpx/source/config/win/x64/vpx_config.asm @@ -8,6 +8,8 @@ %define ARCH_X86_64 1 %define VPX_ARCH_PPC 0 %define ARCH_PPC 0 +%define VPX_ARCH_LOONGARCH 0 +%define ARCH_LOONGARCH 0 %define HAVE_NEON 0 %define HAVE_NEON_ASM 0 %define HAVE_MIPS32 0 @@ -25,6 +27,8 @@ %define HAVE_AVX512 0 %define HAVE_VSX 0 %define HAVE_MMI 0 +%define HAVE_LSX 0 +%define HAVE_LASX 0 %define HAVE_VPX_PORTS 1 %define HAVE_PTHREAD_H 0 %define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/win/x64/vpx_config.h b/chromium/third_party/libvpx/source/config/win/x64/vpx_config.h index 23a083b3854..125cffc4471 100644 --- a/chromium/third_party/libvpx/source/config/win/x64/vpx_config.h +++ b/chromium/third_party/libvpx/source/config/win/x64/vpx_config.h @@ -20,6 +20,8 @@ #define ARCH_X86_64 1 #define VPX_ARCH_PPC 0 #define ARCH_PPC 0 +#define VPX_ARCH_LOONGARCH 0 +#define ARCH_LOONGARCH 0 #define HAVE_NEON 0 #define HAVE_NEON_ASM 0 #define HAVE_MIPS32 0 @@ -37,6 +39,8 @@ #define HAVE_AVX512 0 #define HAVE_VSX 0 #define HAVE_MMI 0 +#define HAVE_LSX 0 +#define HAVE_LASX 0 #define HAVE_VPX_PORTS 1 #define HAVE_PTHREAD_H 0 #define HAVE_UNISTD_H 0 diff --git a/chromium/third_party/libvpx/source/config/win/x64/vpx_dsp_rtcd.h b/chromium/third_party/libvpx/source/config/win/x64/vpx_dsp_rtcd.h index ed632331485..b0a87f4fa8b 100644 --- a/chromium/third_party/libvpx/source/config/win/x64/vpx_dsp_rtcd.h +++ b/chromium/third_party/libvpx/source/config/win/x64/vpx_dsp_rtcd.h @@ -5187,11 +5187,11 @@ int16_t vpx_int_pro_col_c(const uint8_t* ref, const int width); int16_t vpx_int_pro_col_sse2(const uint8_t* ref, const int width); #define vpx_int_pro_col vpx_int_pro_col_sse2 -void vpx_int_pro_row_c(int16_t* hbuf, +void vpx_int_pro_row_c(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); -void vpx_int_pro_row_sse2(int16_t* hbuf, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t* ref, const int ref_stride, const int height); diff --git a/chromium/third_party/libvpx/source/libvpx/build/make/Makefile b/chromium/third_party/libvpx/source/libvpx/build/make/Makefile index 9ca97c8c640..b7a873cc812 100644 --- a/chromium/third_party/libvpx/source/libvpx/build/make/Makefile +++ b/chromium/third_party/libvpx/source/libvpx/build/make/Makefile @@ -151,6 +151,12 @@ $(BUILD_PFX)%_vsx.c.o: CFLAGS += -maltivec -mvsx $(BUILD_PFX)%_msa.c.d: CFLAGS += -mmsa $(BUILD_PFX)%_msa.c.o: CFLAGS += -mmsa +# LOONGARCH +$(BUILD_PFX)%_lsx.c.d: CFLAGS += -mlsx +$(BUILD_PFX)%_lsx.c.o: CFLAGS += -mlsx +$(BUILD_PFX)%_lasx.c.d: CFLAGS += -mlasx +$(BUILD_PFX)%_lasx.c.o: CFLAGS += -mlasx + $(BUILD_PFX)%.c.d: %.c $(if $(quiet),@echo " [DEP] $@") $(qexec)mkdir -p $(dir $@) diff --git a/chromium/third_party/libvpx/source/libvpx/build/make/configure.sh b/chromium/third_party/libvpx/source/libvpx/build/make/configure.sh index b24e79a0d21..581042e38e1 100644 --- a/chromium/third_party/libvpx/source/libvpx/build/make/configure.sh +++ b/chromium/third_party/libvpx/source/libvpx/build/make/configure.sh @@ -449,6 +449,17 @@ EOF fi } +check_inline_asm() { + log check_inline_asm "$@" + name="$1" + code="$2" + shift 2 + disable_feature $name + check_cc "$@" <<EOF && enable_feature $name +void foo(void) { __asm__ volatile($code); } +EOF +} + write_common_config_banner() { print_webm_license config.mk "##" "" echo '# This file automatically generated by configure. Do not edit!' >> config.mk @@ -766,6 +777,12 @@ process_common_toolchain() { *mips32el*) tgt_isa=mips32 ;; + loongarch32*) + tgt_isa=loongarch32 + ;; + loongarch64*) + tgt_isa=loongarch64 + ;; esac # detect tgt_os @@ -834,6 +851,11 @@ process_common_toolchain() { ppc*) enable_feature ppc ;; + loongarch*) + soft_enable lsx + soft_enable lasx + enable_feature loongarch + ;; esac # PIC is probably what we want when building shared libs @@ -1419,6 +1441,15 @@ EOF ;; esac ;; + loongarch*) + link_with_cc=gcc + setup_gnu_toolchain + + enabled lsx && check_inline_asm lsx '"vadd.b $vr0, $vr1, $vr1"' + enabled lsx && soft_enable runtime_cpu_detect + enabled lasx && check_inline_asm lasx '"xvadd.b $xr0, $xr1, $xr1"' + enabled lasx && soft_enable runtime_cpu_detect + ;; *-gcc|generic-gnu) link_with_cc=gcc enable_feature gcc @@ -1521,6 +1552,22 @@ EOF ;; esac + # only for LOONGARCH platforms + case ${toolchain} in + loongarch*) + if enabled big_endian; then + if enabled lsx; then + echo "lsx optimizations are available only for little endian platforms" + disable_feature lsx + fi + if enabled lasx; then + echo "lasx optimizations are available only for little endian platforms" + disable_feature lasx + fi + fi + ;; + esac + # glibc needs these if enabled linux; then add_cflags -D_LARGEFILE_SOURCE diff --git a/chromium/third_party/libvpx/source/libvpx/build/make/rtcd.pl b/chromium/third_party/libvpx/source/libvpx/build/make/rtcd.pl index acb9f6e4667..8ed776add81 100755 --- a/chromium/third_party/libvpx/source/libvpx/build/make/rtcd.pl +++ b/chromium/third_party/libvpx/source/libvpx/build/make/rtcd.pl @@ -387,6 +387,37 @@ EOF common_bottom; } +sub loongarch() { + determine_indirection("c", @ALL_ARCHS); + + # Assign the helper variable for each enabled extension + foreach my $opt (@ALL_ARCHS) { + my $opt_uc = uc $opt; + eval "\$have_${opt}=\"flags & HAS_${opt_uc}\""; + } + + common_top; + print <<EOF; +#include "vpx_config.h" + +#ifdef RTCD_C +#include "vpx_ports/loongarch.h" +static void setup_rtcd_internal(void) +{ + int flags = loongarch_cpu_caps(); + + (void)flags; +EOF + + set_function_pointers("c", @ALL_ARCHS); + + print <<EOF; +} +#endif +EOF + common_bottom; +} + sub unoptimized() { determine_indirection "c"; common_top; @@ -462,6 +493,22 @@ if ($opts{arch} eq 'x86') { } elsif ($opts{arch} =~ /^ppc/ ) { @ALL_ARCHS = filter(qw/vsx/); ppc; +} elsif ($opts{arch} =~ /loongarch/ ) { + @ALL_ARCHS = filter("$opts{arch}"); + open CONFIG_FILE, $opts{config} or + die "Error opening config file '$opts{config}': $!\n"; + while (<CONFIG_FILE>) { + if (/HAVE_LSX=yes/) { + @ALL_ARCHS = filter("$opts{arch}", qw/lsx/); + last; + } + if (/HAVE_LASX=yes/) { + @ALL_ARCHS = filter("$opts{arch}", qw/lasx/); + last; + } + } + close CONFIG_FILE; + loongarch; } else { unoptimized; } diff --git a/chromium/third_party/libvpx/source/libvpx/configure b/chromium/third_party/libvpx/source/libvpx/configure index b68f9fd7816..434ebbe366f 100755 --- a/chromium/third_party/libvpx/source/libvpx/configure +++ b/chromium/third_party/libvpx/source/libvpx/configure @@ -114,6 +114,8 @@ all_platforms="${all_platforms} armv7-win32-vs14" all_platforms="${all_platforms} armv7-win32-vs15" all_platforms="${all_platforms} armv7s-darwin-gcc" all_platforms="${all_platforms} armv8-linux-gcc" +all_platforms="${all_platforms} loongarch32-linux-gcc" +all_platforms="${all_platforms} loongarch64-linux-gcc" all_platforms="${all_platforms} mips32-linux-gcc" all_platforms="${all_platforms} mips64-linux-gcc" all_platforms="${all_platforms} ppc64le-linux-gcc" @@ -237,6 +239,7 @@ ARCH_LIST=" x86 x86_64 ppc + loongarch " ARCH_EXT_LIST_X86=" mmx @@ -252,6 +255,8 @@ ARCH_EXT_LIST_X86=" ARCH_EXT_LIST_LOONGSON=" mmi + lsx + lasx " ARCH_EXT_LIST=" diff --git a/chromium/third_party/libvpx/source/libvpx/examples/vp9_spatial_svc_encoder.c b/chromium/third_party/libvpx/source/libvpx/examples/vp9_spatial_svc_encoder.c index c37e608d172..455f6c9036e 100644 --- a/chromium/third_party/libvpx/source/libvpx/examples/vp9_spatial_svc_encoder.c +++ b/chromium/third_party/libvpx/source/libvpx/examples/vp9_spatial_svc_encoder.c @@ -579,7 +579,8 @@ static void set_frame_flags_bypass_mode_ex0( ref_frame_config->alt_fb_idx[sl] = 0; } else if (tl == 1) { ref_frame_config->lst_fb_idx[sl] = sl; - ref_frame_config->gld_fb_idx[sl] = num_spatial_layers + sl - 1; + ref_frame_config->gld_fb_idx[sl] = + (sl == 0) ? 0 : num_spatial_layers + sl - 1; ref_frame_config->alt_fb_idx[sl] = num_spatial_layers + sl; } // Set the reference and update flags. diff --git a/chromium/third_party/libvpx/source/libvpx/examples/vpx_temporal_svc_encoder.c b/chromium/third_party/libvpx/source/libvpx/examples/vpx_temporal_svc_encoder.c index bb761a4117b..a80027822ad 100644 --- a/chromium/third_party/libvpx/source/libvpx/examples/vpx_temporal_svc_encoder.c +++ b/chromium/third_party/libvpx/source/libvpx/examples/vpx_temporal_svc_encoder.c @@ -240,6 +240,38 @@ static void set_roi_map(const char *enc_name, vpx_codec_enc_cfg_t *cfg, } } } + +static void set_roi_skip_map(vpx_codec_enc_cfg_t *cfg, vpx_roi_map_t *roi, + int *skip_map, int *prev_mask_map, int frame_num) { + const int block_size = 8; + unsigned int i, j; + roi->rows = (cfg->g_h + block_size - 1) / block_size; + roi->cols = (cfg->g_w + block_size - 1) / block_size; + zero(roi->skip); + zero(roi->delta_q); + zero(roi->delta_lf); + memset(roi->ref_frame, -1, sizeof(roi->ref_frame)); + roi->ref_frame[1] = 1; + // Use segment 3 for skip. + roi->skip[3] = 1; + roi->roi_map = + (uint8_t *)calloc(roi->rows * roi->cols, sizeof(*roi->roi_map)); + for (i = 0; i < roi->rows; ++i) { + for (j = 0; j < roi->cols; ++j) { + const int idx = i * roi->cols + j; + // Use segment 3 for skip. + // prev_mask_map keeps track of blocks that have been stably on segment 3 + // for the past 10 frames. Only skip when the block is on segment 3 in + // both current map and prev_mask_map. + if (skip_map[idx] == 1 && prev_mask_map[idx] == 1) roi->roi_map[idx] = 3; + // Reset it every 10 frames so it doesn't propagate for too many frames. + if (frame_num % 10 == 0) + prev_mask_map[idx] = skip_map[idx]; + else if (prev_mask_map[idx] == 1 && skip_map[idx] == 0) + prev_mask_map[idx] = 0; + } + } +} #endif // Temporal scaling parameters: @@ -574,6 +606,23 @@ static void set_temporal_layer_pattern(int layering_mode, } } +#if ROI_MAP +static void read_mask(FILE *mask_file, int *seg_map) { + int mask_rows, mask_cols, i, j; + int *map_start = seg_map; + fscanf(mask_file, "%d %d\n", &mask_cols, &mask_rows); + for (i = 0; i < mask_rows; i++) { + for (j = 0; j < mask_cols; j++) { + fscanf(mask_file, "%d ", &seg_map[j]); + // reverse the bit + seg_map[j] = 1 - seg_map[j]; + } + seg_map += mask_cols; + } + seg_map = map_start; +} +#endif + int main(int argc, char **argv) { VpxVideoWriter *outfile[VPX_TS_MAX_LAYERS] = { NULL }; vpx_codec_ctx_t codec; @@ -613,7 +662,14 @@ int main(int argc, char **argv) { double sum_bitrate = 0.0; double sum_bitrate2 = 0.0; double framerate = 30.0; - +#if ROI_MAP + FILE *mask_file = NULL; + int block_size = 8; + int mask_rows = 0; + int mask_cols = 0; + int *mask_map; + int *prev_mask_map; +#endif zero(rc.layer_target_bitrate); memset(&layer_id, 0, sizeof(vpx_svc_layer_id_t)); memset(&input_ctx, 0, sizeof(input_ctx)); @@ -657,9 +713,15 @@ int main(int argc, char **argv) { die("Invalid layering mode (0..12) %s", argv[12]); } +#if ROI_MAP + if (argc != min_args + mode_to_num_layers[layering_mode] + 1) { + die("Invalid number of arguments"); + } +#else if (argc != min_args + mode_to_num_layers[layering_mode]) { die("Invalid number of arguments"); } +#endif input_ctx.filename = argv[1]; open_input_file(&input_ctx); @@ -817,6 +879,13 @@ int main(int argc, char **argv) { #endif // CONFIG_VP9_HIGHBITDEPTH die("Failed to initialize encoder"); +#if ROI_MAP + mask_rows = (cfg.g_h + block_size - 1) / block_size; + mask_cols = (cfg.g_w + block_size - 1) / block_size; + mask_map = (int *)calloc(mask_rows * mask_cols, sizeof(*mask_map)); + prev_mask_map = (int *)calloc(mask_rows * mask_cols, sizeof(*mask_map)); +#endif + if (strncmp(encoder->name, "vp8", 3) == 0) { vpx_codec_control(&codec, VP8E_SET_CPUUSED, -speed); vpx_codec_control(&codec, VP8E_SET_NOISE_SENSITIVITY, kVp8DenoiserOff); @@ -827,7 +896,6 @@ int main(int argc, char **argv) { if (vpx_codec_control(&codec, VP8E_SET_ROI_MAP, &roi)) die_codec(&codec, "Failed to set ROI map"); #endif - } else if (strncmp(encoder->name, "vp9", 3) == 0) { vpx_svc_extra_cfg_t svc_params; memset(&svc_params, 0, sizeof(svc_params)); @@ -843,12 +911,7 @@ int main(int argc, char **argv) { vpx_codec_control(&codec, VP9E_SET_TUNE_CONTENT, 0); vpx_codec_control(&codec, VP9E_SET_TILE_COLUMNS, get_msb(cfg.g_threads)); vpx_codec_control(&codec, VP9E_SET_DISABLE_LOOPFILTER, 0); -#if ROI_MAP - set_roi_map(encoder->name, &cfg, &roi); - if (vpx_codec_control(&codec, VP9E_SET_ROI_MAP, &roi)) - die_codec(&codec, "Failed to set ROI map"); - vpx_codec_control(&codec, VP9E_SET_AQ_MODE, 0); -#endif + if (cfg.g_threads > 1) vpx_codec_control(&codec, VP9E_SET_ROW_MT, 1); else @@ -881,6 +944,9 @@ int main(int argc, char **argv) { struct vpx_usec_timer timer; vpx_codec_iter_t iter = NULL; const vpx_codec_cx_pkt_t *pkt; +#if ROI_MAP + char mask_file_name[255]; +#endif // Update the temporal layer_id. No spatial layers in this test. layer_id.spatial_layer_id = 0; layer_id.temporal_layer_id = @@ -894,6 +960,19 @@ int main(int argc, char **argv) { } flags = layer_flags[frame_cnt % flag_periodicity]; if (layering_mode == 0) flags = 0; +#if ROI_MAP + snprintf(mask_file_name, sizeof(mask_file_name), "%s%05d.txt", + argv[argc - 1], frame_cnt); + mask_file = fopen(mask_file_name, "r"); + if (mask_file != NULL) { + read_mask(mask_file, mask_map); + fclose(mask_file); + // set_roi_map(encoder->name, &cfg, &roi); + set_roi_skip_map(&cfg, &roi, mask_map, prev_mask_map, frame_cnt); + if (vpx_codec_control(&codec, VP9E_SET_ROI_MAP, &roi)) + die_codec(&codec, "Failed to set ROI map"); + } +#endif frame_avail = read_frame(&input_ctx, &raw); if (frame_avail) ++rc.layer_input_frames[layer_id.temporal_layer_id]; vpx_usec_timer_start(&timer); @@ -963,6 +1042,10 @@ int main(int argc, char **argv) { ++frame_cnt; pts += frame_duration; } +#if ROI_MAP + free(mask_map); + free(prev_mask_map); +#endif close_input_file(&input_ctx); printout_rate_control_summary(&rc, &cfg, frame_cnt); printf("\n"); diff --git a/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/README.libvpx b/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/README.libvpx index 5cc0a837011..325604cc669 100644 --- a/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/README.libvpx +++ b/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/README.libvpx @@ -1,5 +1,5 @@ URL: https://chromium.googlesource.com/webm/libwebm -Version: 206d268d4d8066e5a37c49025325b80c95c771dd +Version: ee0bab576c338c9807249b99588e352b7268cb62 License: BSD License File: LICENSE.txt diff --git a/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/mkvmuxer/mkvmuxer.cc b/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/mkvmuxer/mkvmuxer.cc index 24c288863fc..ae365314392 100644 --- a/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/mkvmuxer/mkvmuxer.cc +++ b/chromium/third_party/libvpx/source/libvpx/third_party/libwebm/mkvmuxer/mkvmuxer.cc @@ -2622,8 +2622,7 @@ bool Cluster::Finalize(bool set_last_frame_duration, uint64_t duration) { uint64_t Cluster::Size() const { const uint64_t element_size = - EbmlMasterElementSize(static_cast<uint64_t>(libwebm::kMkvCluster), - uint64_t{0xFFFFFFFFFFFFFFFFU}) + + EbmlMasterElementSize(libwebm::kMkvCluster, 0xFFFFFFFFFFFFFFFFULL) + payload_size_; return element_size; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/generic/systemdependent.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/generic/systemdependent.c index cd1b02c9cc5..71529bdfd88 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/generic/systemdependent.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/generic/systemdependent.c @@ -18,6 +18,8 @@ #include "vpx_ports/ppc.h" #elif VPX_ARCH_MIPS #include "vpx_ports/mips.h" +#elif VPX_ARCH_LOONGARCH +#include "vpx_ports/loongarch.h" #endif #include "vp8/common/onyxc_int.h" #include "vp8/common/systemdependent.h" @@ -100,6 +102,8 @@ void vp8_machine_specific_config(VP8_COMMON *ctx) { ctx->cpu_caps = ppc_simd_caps(); #elif VPX_ARCH_MIPS ctx->cpu_caps = mips_cpu_caps(); +#elif VPX_ARCH_LOONGARCH + ctx->cpu_caps = loongarch_cpu_caps(); #else // generic-gnu targets. ctx->cpu_caps = 0; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/loopfilter_filters_lsx.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/loopfilter_filters_lsx.c new file mode 100644 index 00000000000..484b3d6ad0f --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/loopfilter_filters_lsx.c @@ -0,0 +1,393 @@ +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Lu Wang <wanglu@loongson.cn> + * + * 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 "./vp8_rtcd.h" +#include "vp8/common/loopfilter.h" +#include "vpx_util/loongson_intrinsics.h" + +#define VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev) \ + { \ + __m128i p2_m, p1_m, p0_m, q2_m, q1_m, q0_m; \ + __m128i u, filt, t1, t2, filt_sign, q0_sub_p0; \ + __m128i filt_r, filt_l; \ + __m128i temp0, temp1, temp2, temp3; \ + const __m128i cnst4b = __lsx_vldi(4); \ + const __m128i cnst3b = __lsx_vldi(3); \ + const __m128i cnst9h = __lsx_vldi(1033); \ + const __m128i cnst63h = __lsx_vldi(1087); \ + \ + p2_m = __lsx_vxori_b(p2, 0x80); \ + p1_m = __lsx_vxori_b(p1, 0x80); \ + p0_m = __lsx_vxori_b(p0, 0x80); \ + q0_m = __lsx_vxori_b(q0, 0x80); \ + q1_m = __lsx_vxori_b(q1, 0x80); \ + q2_m = __lsx_vxori_b(q2, 0x80); \ + \ + filt = __lsx_vssub_b(p1_m, q1_m); \ + q0_sub_p0 = __lsx_vssub_b(q0_m, p0_m); \ + filt = __lsx_vsadd_b(filt, q0_sub_p0); \ + filt = __lsx_vsadd_b(filt, q0_sub_p0); \ + filt = __lsx_vsadd_b(filt, q0_sub_p0); \ + filt = __lsx_vand_v(filt, mask); \ + \ + t2 = __lsx_vand_v(filt, hev); \ + hev = __lsx_vxori_b(hev, 0xff); \ + filt = __lsx_vand_v(hev, filt); \ + t1 = __lsx_vsadd_b(t2, cnst4b); \ + t1 = __lsx_vsra_b(t1, cnst3b); \ + t2 = __lsx_vsadd_b(t2, cnst3b); \ + t2 = __lsx_vsra_b(t2, cnst3b); \ + q0_m = __lsx_vssub_b(q0_m, t1); \ + p0_m = __lsx_vsadd_b(p0_m, t2); \ + filt_sign = __lsx_vslti_b(filt, 0); \ + filt_r = __lsx_vilvl_b(filt_sign, filt); \ + filt_l = __lsx_vilvh_b(filt_sign, filt); \ + temp0 = __lsx_vmul_h(filt_r, cnst9h); \ + temp1 = __lsx_vadd_h(temp0, cnst63h); \ + temp2 = __lsx_vmul_h(filt_l, cnst9h); \ + temp3 = __lsx_vadd_h(temp2, cnst63h); \ + \ + u = __lsx_vssrani_b_h(temp3, temp1, 7); \ + q2_m = __lsx_vssub_b(q2_m, u); \ + p2_m = __lsx_vsadd_b(p2_m, u); \ + q2 = __lsx_vxori_b(q2_m, 0x80); \ + p2 = __lsx_vxori_b(p2_m, 0x80); \ + \ + temp1 = __lsx_vadd_h(temp1, temp0); \ + temp3 = __lsx_vadd_h(temp3, temp2); \ + \ + u = __lsx_vssrani_b_h(temp3, temp1, 7); \ + q1_m = __lsx_vssub_b(q1_m, u); \ + p1_m = __lsx_vsadd_b(p1_m, u); \ + q1 = __lsx_vxori_b(q1_m, 0x80); \ + p1 = __lsx_vxori_b(p1_m, 0x80); \ + \ + temp1 = __lsx_vadd_h(temp1, temp0); \ + temp3 = __lsx_vadd_h(temp3, temp2); \ + \ + u = __lsx_vssrani_b_h(temp3, temp1, 7); \ + q0_m = __lsx_vssub_b(q0_m, u); \ + p0_m = __lsx_vsadd_b(p0_m, u); \ + q0 = __lsx_vxori_b(q0_m, 0x80); \ + p0 = __lsx_vxori_b(p0_m, 0x80); \ + } + +#define LPF_MASK_HEV(p3_in, p2_in, p1_in, p0_in, q0_in, q1_in, q2_in, q3_in, \ + limit_in, b_limit_in, thresh_in, hev_out, mask_out, \ + flat_out) \ + { \ + __m128i p3_asub_p2_m, p2_asub_p1_m, p1_asub_p0_m, q1_asub_q0_m; \ + __m128i p1_asub_q1_m, p0_asub_q0_m, q3_asub_q2_m, q2_asub_q1_m; \ + \ + p3_asub_p2_m = __lsx_vabsd_bu(p3_in, p2_in); \ + p2_asub_p1_m = __lsx_vabsd_bu(p2_in, p1_in); \ + p1_asub_p0_m = __lsx_vabsd_bu(p1_in, p0_in); \ + q1_asub_q0_m = __lsx_vabsd_bu(q1_in, q0_in); \ + q2_asub_q1_m = __lsx_vabsd_bu(q2_in, q1_in); \ + q3_asub_q2_m = __lsx_vabsd_bu(q3_in, q2_in); \ + p0_asub_q0_m = __lsx_vabsd_bu(p0_in, q0_in); \ + p1_asub_q1_m = __lsx_vabsd_bu(p1_in, q1_in); \ + flat_out = __lsx_vmax_bu(p1_asub_p0_m, q1_asub_q0_m); \ + hev_out = __lsx_vslt_bu(thresh_in, flat_out); \ + p0_asub_q0_m = __lsx_vsadd_bu(p0_asub_q0_m, p0_asub_q0_m); \ + p1_asub_q1_m = __lsx_vsrli_b(p1_asub_q1_m, 1); \ + p0_asub_q0_m = __lsx_vsadd_bu(p0_asub_q0_m, p1_asub_q1_m); \ + mask_out = __lsx_vslt_bu(b_limit_in, p0_asub_q0_m); \ + mask_out = __lsx_vmax_bu(flat_out, mask_out); \ + p3_asub_p2_m = __lsx_vmax_bu(p3_asub_p2_m, p2_asub_p1_m); \ + mask_out = __lsx_vmax_bu(p3_asub_p2_m, mask_out); \ + q2_asub_q1_m = __lsx_vmax_bu(q2_asub_q1_m, q3_asub_q2_m); \ + mask_out = __lsx_vmax_bu(q2_asub_q1_m, mask_out); \ + mask_out = __lsx_vslt_bu(limit_in, mask_out); \ + mask_out = __lsx_vxori_b(mask_out, 0xff); \ + } + +#define VP8_ST6x1_B(in0, in0_idx, in1, in1_idx, pdst, stride) \ + { \ + __lsx_vstelm_w(in0, pdst, 0, in0_idx); \ + __lsx_vstelm_h(in1, pdst + stride, 0, in1_idx); \ + } + +static inline void mbloop_filter_horizontal_edge_y_lsx( + uint8_t *src, int32_t pitch, const uint8_t b_limit_in, + const uint8_t limit_in, const uint8_t thresh_in) { + uint8_t *temp_src; + int32_t pitch_x2 = pitch << 1; + int32_t pitch_x3 = pitch_x2 + pitch; + int32_t pitch_x4 = pitch << 2; + + __m128i p3, p2, p1, p0, q3, q2, q1, q0; + __m128i mask, hev, flat, thresh, limit, b_limit; + + DUP2_ARG2(__lsx_vldrepl_b, &b_limit_in, 0, &limit_in, 0, b_limit, limit); + thresh = __lsx_vldrepl_b(&thresh_in, 0); + + temp_src = src - pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, p3, p2, p1, p0); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, q0, q1, q2, q3); + LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, + mask, flat); + VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev); + + temp_src = src - pitch_x3; + __lsx_vstx(p2, temp_src, 0); + __lsx_vstx(p1, temp_src, pitch); + __lsx_vstx(p0, temp_src, pitch_x2); + __lsx_vstx(q0, temp_src, pitch_x3); + temp_src += pitch_x4; + __lsx_vstx(q1, temp_src, 0); + __lsx_vstx(q2, temp_src, pitch); +} + +static inline void mbloop_filter_horizontal_edge_uv_lsx( + uint8_t *src_u, uint8_t *src_v, int32_t pitch, const uint8_t b_limit_in, + const uint8_t limit_in, const uint8_t thresh_in) { + uint8_t *temp_src; + int32_t pitch_x2 = pitch << 1; + int32_t pitch_x3 = pitch_x2 + pitch; + int32_t pitch_x4 = pitch << 2; + __m128i p3, p2, p1, p0, q3, q2, q1, q0; + __m128i mask, hev, flat, thresh, limit, b_limit; + __m128i p3_u, p2_u, p1_u, p0_u, q3_u, q2_u, q1_u, q0_u; + __m128i p3_v, p2_v, p1_v, p0_v, q3_v, q2_v, q1_v, q0_v; + + DUP2_ARG2(__lsx_vldrepl_b, &b_limit_in, 0, &limit_in, 0, b_limit, limit); + thresh = __lsx_vldrepl_b(&thresh_in, 0); + + temp_src = src_u - pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, p3_u, p2_u, p1_u, p0_u); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, q0_u, q1_u, q2_u, q3_u); + temp_src = src_v - pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, p3_v, p2_v, p1_v, p0_v); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, q0_v, q1_v, q2_v, q3_v); + + DUP4_ARG2(__lsx_vilvl_d, p3_v, p3_u, p2_v, p2_u, p1_v, p1_u, p0_v, p0_u, p3, + p2, p1, p0); + DUP4_ARG2(__lsx_vilvl_d, q0_v, q0_u, q1_v, q1_u, q2_v, q2_u, q3_v, q3_u, q0, + q1, q2, q3); + LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, + mask, flat); + VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev); + + src_u -= pitch_x3; + __lsx_vstelm_d(p2, src_u, 0, 0); + __lsx_vstelm_d(p1, src_u + pitch, 0, 0); + __lsx_vstelm_d(p0, src_u + pitch_x2, 0, 0); + __lsx_vstelm_d(q0, src_u + pitch_x3, 0, 0); + src_u += pitch_x4; + __lsx_vstelm_d(q1, src_u, 0, 0); + src_u += pitch; + __lsx_vstelm_d(q2, src_u, 0, 0); + + src_v -= pitch_x3; + __lsx_vstelm_d(p2, src_v, 0, 1); + __lsx_vstelm_d(p1, src_v + pitch, 0, 1); + __lsx_vstelm_d(p0, src_v + pitch_x2, 0, 1); + __lsx_vstelm_d(q0, src_v + pitch_x3, 0, 1); + src_v += pitch_x4; + __lsx_vstelm_d(q1, src_v, 0, 1); + src_v += pitch; + __lsx_vstelm_d(q2, src_v, 0, 1); +} + +static inline void mbloop_filter_vertical_edge_y_lsx(uint8_t *src, + int32_t pitch, + const uint8_t b_limit_in, + const uint8_t limit_in, + const uint8_t thresh_in) { + uint8_t *temp_src; + int32_t pitch_x2 = pitch << 1; + int32_t pitch_x3 = pitch_x2 + pitch; + int32_t pitch_x4 = pitch << 2; + + __m128i p3, p2, p1, p0, q3, q2, q1, q0; + __m128i mask, hev, flat, thresh, limit, b_limit; + __m128i row0, row1, row2, row3, row4, row5, row6, row7, row8; + __m128i row9, row10, row11, row12, row13, row14, row15; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + + DUP2_ARG2(__lsx_vldrepl_b, &b_limit_in, 0, &limit_in, 0, b_limit, limit); + thresh = __lsx_vldrepl_b(&thresh_in, 0); + temp_src = src - 4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, row0, row1, row2, row3); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, row4, row5, row6, row7); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, row8, row9, row10, row11); + temp_src += pitch_x4; + DUP4_ARG2(__lsx_vldx, temp_src, 0, temp_src, pitch, temp_src, pitch_x2, + temp_src, pitch_x3, row12, row13, row14, row15); + temp_src -= pitch_x4; + LSX_TRANSPOSE16x8_B(row0, row1, row2, row3, row4, row5, row6, row7, row8, + row9, row10, row11, row12, row13, row14, row15, p3, p2, + p1, p0, q0, q1, q2, q3); + + LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, + mask, flat); + VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev); + DUP2_ARG2(__lsx_vilvl_b, p1, p2, q0, p0, tmp0, tmp1); + tmp3 = __lsx_vilvl_h(tmp1, tmp0); + tmp4 = __lsx_vilvh_h(tmp1, tmp0); + DUP2_ARG2(__lsx_vilvh_b, p1, p2, q0, p0, tmp0, tmp1); + tmp6 = __lsx_vilvl_h(tmp1, tmp0); + tmp7 = __lsx_vilvh_h(tmp1, tmp0); + tmp2 = __lsx_vilvl_b(q2, q1); + tmp5 = __lsx_vilvh_b(q2, q1); + + temp_src = src - 3; + VP8_ST6x1_B(tmp3, 0, tmp2, 0, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp3, 1, tmp2, 1, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp3, 2, tmp2, 2, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp3, 3, tmp2, 3, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp4, 0, tmp2, 4, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp4, 1, tmp2, 5, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp4, 2, tmp2, 6, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp4, 3, tmp2, 7, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp6, 0, tmp5, 0, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp6, 1, tmp5, 1, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp6, 2, tmp5, 2, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp6, 3, tmp5, 3, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp7, 0, tmp5, 4, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp7, 1, tmp5, 5, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp7, 2, tmp5, 6, temp_src, 4); + temp_src += pitch; + VP8_ST6x1_B(tmp7, 3, tmp5, 7, temp_src, 4); +} + +static inline void mbloop_filter_vertical_edge_uv_lsx( + uint8_t *src_u, uint8_t *src_v, int32_t pitch, const uint8_t b_limit_in, + const uint8_t limit_in, const uint8_t thresh_in) { + int32_t pitch_x2 = pitch << 1; + int32_t pitch_x3 = pitch_x2 + pitch; + int32_t pitch_x4 = pitch << 2; + __m128i p3, p2, p1, p0, q3, q2, q1, q0; + __m128i mask, hev, flat, thresh, limit, b_limit; + __m128i row0, row1, row2, row3, row4, row5, row6, row7, row8; + __m128i row9, row10, row11, row12, row13, row14, row15; + __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + + DUP2_ARG2(__lsx_vldrepl_b, &b_limit_in, 0, &limit_in, 0, b_limit, limit); + thresh = __lsx_vldrepl_b(&thresh_in, 0); + + src_u -= 4; + DUP4_ARG2(__lsx_vldx, src_u, 0, src_u, pitch, src_u, pitch_x2, src_u, + pitch_x3, row0, row1, row2, row3); + src_u += pitch_x4; + DUP4_ARG2(__lsx_vldx, src_u, 0, src_u, pitch, src_u, pitch_x2, src_u, + pitch_x3, row4, row5, row6, row7); + src_v -= 4; + DUP4_ARG2(__lsx_vldx, src_v, 0, src_v, pitch, src_v, pitch_x2, src_v, + pitch_x3, row8, row9, row10, row11); + src_v += pitch_x4; + DUP4_ARG2(__lsx_vldx, src_v, 0, src_v, pitch, src_v, pitch_x2, src_v, + pitch_x3, row12, row13, row14, row15); + LSX_TRANSPOSE16x8_B(row0, row1, row2, row3, row4, row5, row6, row7, row8, + row9, row10, row11, row12, row13, row14, row15, p3, p2, + p1, p0, q0, q1, q2, q3); + + LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev, + mask, flat); + VP8_MBFILTER(p2, p1, p0, q0, q1, q2, mask, hev); + + DUP2_ARG2(__lsx_vilvl_b, p1, p2, q0, p0, tmp0, tmp1); + tmp3 = __lsx_vilvl_h(tmp1, tmp0); + tmp4 = __lsx_vilvh_h(tmp1, tmp0); + DUP2_ARG2(__lsx_vilvh_b, p1, p2, q0, p0, tmp0, tmp1); + tmp6 = __lsx_vilvl_h(tmp1, tmp0); + tmp7 = __lsx_vilvh_h(tmp1, tmp0); + tmp2 = __lsx_vilvl_b(q2, q1); + tmp5 = __lsx_vilvh_b(q2, q1); + + src_u += 1 - pitch_x4; + VP8_ST6x1_B(tmp3, 0, tmp2, 0, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp3, 1, tmp2, 1, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp3, 2, tmp2, 2, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp3, 3, tmp2, 3, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp4, 0, tmp2, 4, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp4, 1, tmp2, 5, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp4, 2, tmp2, 6, src_u, 4); + src_u += pitch; + VP8_ST6x1_B(tmp4, 3, tmp2, 7, src_u, 4); + + src_v += 1 - pitch_x4; + VP8_ST6x1_B(tmp6, 0, tmp5, 0, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp6, 1, tmp5, 1, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp6, 2, tmp5, 2, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp6, 3, tmp5, 3, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp7, 0, tmp5, 4, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp7, 1, tmp5, 5, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp7, 2, tmp5, 6, src_v, 4); + src_v += pitch; + VP8_ST6x1_B(tmp7, 3, tmp5, 7, src_v, 4); +} + +void vp8_loop_filter_mbh_lsx(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v, + int32_t pitch_y, int32_t pitch_u_v, + loop_filter_info *lpf_info_ptr) { + mbloop_filter_horizontal_edge_y_lsx(src_y, pitch_y, *lpf_info_ptr->mblim, + *lpf_info_ptr->lim, + *lpf_info_ptr->hev_thr); + if (src_u) { + mbloop_filter_horizontal_edge_uv_lsx( + src_u, src_v, pitch_u_v, *lpf_info_ptr->mblim, *lpf_info_ptr->lim, + *lpf_info_ptr->hev_thr); + } +} + +void vp8_loop_filter_mbv_lsx(uint8_t *src_y, uint8_t *src_u, uint8_t *src_v, + int32_t pitch_y, int32_t pitch_u_v, + loop_filter_info *lpf_info_ptr) { + mbloop_filter_vertical_edge_y_lsx(src_y, pitch_y, *lpf_info_ptr->mblim, + *lpf_info_ptr->lim, *lpf_info_ptr->hev_thr); + if (src_u) { + mbloop_filter_vertical_edge_uv_lsx(src_u, src_v, pitch_u_v, + *lpf_info_ptr->mblim, *lpf_info_ptr->lim, + *lpf_info_ptr->hev_thr); + } +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/sixtap_filter_lsx.c b/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/sixtap_filter_lsx.c new file mode 100644 index 00000000000..75fe533d988 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/loongarch/sixtap_filter_lsx.c @@ -0,0 +1,1164 @@ +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Lu Wang <wanglu@loongson.cn> + * + * 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 "./vp8_rtcd.h" +#include "vp8/common/filter.h" +#include "vpx_ports/mem.h" +#include "vpx_util/loongson_intrinsics.h" + +DECLARE_ALIGNED(16, static const int8_t, vp8_subpel_filters_lsx[7][8]) = { + { 0, -6, 123, 12, -1, 0, 0, 0 }, + { 2, -11, 108, 36, -8, 1, 0, 0 }, /* New 1/4 pel 6 tap filter */ + { 0, -9, 93, 50, -6, 0, 0, 0 }, + { 3, -16, 77, 77, -16, 3, 0, 0 }, /* New 1/2 pel 6 tap filter */ + { 0, -6, 50, 93, -9, 0, 0, 0 }, + { 1, -8, 36, 108, -11, 2, 0, 0 }, /* New 1/4 pel 6 tap filter */ + { 0, -1, 12, 123, -6, 0, 0, 0 }, +}; + +static const uint8_t vp8_mc_filt_mask_arr[16 * 3] = { + /* 8 width cases */ + 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, + /* 4 width cases */ + 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20, + /* 4 width cases */ + 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28 +}; + +#define DPADD_H3(in0, in1, in2, coeff0, coeff1, coeff2) \ + ({ \ + __m128i out0_m; \ + \ + out0_m = __lsx_vdp2_h_b(in0, coeff0); \ + out0_m = __lsx_vdp2add_h_b(out0_m, in1, coeff1); \ + out0_m = __lsx_vdp2add_h_b(out0_m, in2, coeff2); \ + \ + out0_m; \ + }) + +#define HORIZ_6TAP_FILT(src0, src1, mask0, mask1, mask2, filt_h0, filt_h1, \ + filt_h2) \ + ({ \ + __m128i vec0_m, vec1_m, vec2_m; \ + __m128i hz_out_m; \ + \ + DUP2_ARG3(__lsx_vshuf_b, src0, src1, mask0, src0, src1, mask1, vec0_m, \ + vec1_m); \ + vec2_m = __lsx_vshuf_b(src0, src1, mask2); \ + hz_out_m = DPADD_H3(vec0_m, vec1_m, vec2_m, filt_h0, filt_h1, filt_h2); \ + \ + hz_out_m = __lsx_vsrari_h(hz_out_m, VP8_FILTER_SHIFT); \ + hz_out_m = __lsx_vsat_h(hz_out_m, 7); \ + \ + hz_out_m; \ + }) + +#define HORIZ_6TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, \ + mask2, filt0, filt1, filt2, out0, out1, \ + out2, out3) \ + ({ \ + __m128i vec0_m, vec1_m, vec2_m, vec3_m, vec4_m, vec5_m, vec6_m, vec7_m; \ + \ + DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src1, src1, mask0, vec0_m, \ + vec1_m); \ + DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src3, src3, mask0, vec2_m, \ + vec3_m); \ + DUP4_ARG2(__lsx_vdp2_h_b, vec0_m, filt0, vec1_m, filt0, vec2_m, filt0, \ + vec3_m, filt0, out0, out1, out2, out3); \ + DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask1, src1, src1, mask1, vec0_m, \ + vec1_m); \ + DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask1, src3, src3, mask1, vec2_m, \ + vec3_m); \ + DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask2, src1, src1, mask2, vec4_m, \ + vec5_m); \ + DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask2, src3, src3, mask2, vec6_m, \ + vec7_m); \ + DUP4_ARG3(__lsx_vdp2add_h_b, out0, vec0_m, filt1, out1, vec1_m, filt1, \ + out2, vec2_m, filt1, out3, vec3_m, filt1, out0, out1, out2, \ + out3); \ + DUP4_ARG3(__lsx_vdp2add_h_b, out0, vec4_m, filt2, out1, vec5_m, filt2, \ + out2, vec6_m, filt2, out3, vec7_m, filt2, out0, out1, out2, \ + out3); \ + }) + +#define FILT_4TAP_DPADD_H(vec0, vec1, filt0, filt1) \ + ({ \ + __m128i tmp0; \ + \ + tmp0 = __lsx_vdp2_h_b(vec0, filt0); \ + tmp0 = __lsx_vdp2add_h_b(tmp0, vec1, filt1); \ + \ + tmp0; \ + }) + +#define HORIZ_4TAP_FILT(src0, src1, mask0, mask1, filt_h0, filt_h1) \ + ({ \ + __m128i vec0_m, vec1_m; \ + __m128i hz_out_m; \ + \ + DUP2_ARG3(__lsx_vshuf_b, src0, src1, mask0, src0, src1, mask1, vec0_m, \ + vec1_m); \ + hz_out_m = FILT_4TAP_DPADD_H(vec0_m, vec1_m, filt_h0, filt_h1); \ + hz_out_m = __lsx_vsrari_h(hz_out_m, VP8_FILTER_SHIFT); \ + hz_out_m = __lsx_vsat_h(hz_out_m, 7); \ + \ + hz_out_m; \ + }) + +#define HORIZ_4TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, \ + filt0, filt1, out0, out1, out2, out3) \ + ({ \ + __m128i vec0_m, vec1_m, vec2_m, vec3_m; \ + \ + DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src1, src1, mask0, vec0_m, \ + vec1_m); \ + DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src3, src3, mask0, vec2_m, \ + vec3_m); \ + DUP4_ARG2(__lsx_vdp2_h_b, vec0_m, filt0, vec1_m, filt0, vec2_m, filt0, \ + vec3_m, filt0, out0, out1, out2, out3); \ + DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask1, src1, src1, mask1, vec0_m, \ + vec1_m); \ + DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask1, src3, src3, mask1, vec2_m, \ + vec3_m); \ + DUP4_ARG3(__lsx_vdp2add_h_b, out0, vec0_m, filt1, out1, vec1_m, filt1, \ + out2, vec2_m, filt1, out3, vec3_m, filt1, out0, out1, out2, \ + out3); \ + }) + +static void common_hz_6t_8w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, filt0, filt1, filt2; + __m128i mask0, mask1, mask2, tmp0, tmp1; + __m128i filt, out0, out1, out2, out3; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 2; + + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + filt2 = __lsx_vreplvei_h(filt, 2); + DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2); + + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src += src_stride_x4; + HORIZ_6TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, filt0, + filt1, filt2, out0, out1, out2, out3); + DUP2_ARG3(__lsx_vssrarni_b_h, out1, out0, VP8_FILTER_SHIFT, out3, out2, + VP8_FILTER_SHIFT, tmp0, tmp1); + DUP2_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp0, tmp1); + __lsx_vstelm_d(tmp0, dst, 0, 0); + __lsx_vstelm_d(tmp0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(tmp1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(tmp1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + + for (loop_cnt = (height >> 2) - 1; loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src += src_stride_x4; + HORIZ_6TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, + filt0, filt1, filt2, out0, out1, out2, out3); + DUP2_ARG3(__lsx_vssrarni_b_h, out1, out0, VP8_FILTER_SHIFT, out3, out2, + VP8_FILTER_SHIFT, tmp0, tmp1); + DUP2_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp0, tmp1); + __lsx_vstelm_d(tmp0, dst, 0, 0); + __lsx_vstelm_d(tmp0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(tmp1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(tmp1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + } +} + +static void common_hz_6t_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, src7, filt0, filt1, filt2; + __m128i mask0, mask1, mask2, out; + __m128i filt, out0, out1, out2, out3, out4, out5, out6, out7; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 2; + + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + filt2 = __lsx_vreplvei_h(filt, 2); + DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src2, src4, src6); + src += 8; + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src1, src3, src5, src7); + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + DUP4_ARG2(__lsx_vxori_b, src4, 128, src5, 128, src6, 128, src7, 128, src4, + src5, src6, src7); + src += src_stride_x4 - 8; + + HORIZ_6TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, + filt0, filt1, filt2, out0, out1, out2, out3); + HORIZ_6TAP_8WID_4VECS_FILT(src4, src5, src6, src7, mask0, mask1, mask2, + filt0, filt1, filt2, out4, out5, out6, out7); + DUP4_ARG2(__lsx_vsrari_h, out0, VP8_FILTER_SHIFT, out1, VP8_FILTER_SHIFT, + out2, VP8_FILTER_SHIFT, out3, VP8_FILTER_SHIFT, out0, out1, out2, + out3); + DUP4_ARG2(__lsx_vsrari_h, out4, VP8_FILTER_SHIFT, out5, VP8_FILTER_SHIFT, + out6, VP8_FILTER_SHIFT, out7, VP8_FILTER_SHIFT, out4, out5, out6, + out7); + DUP4_ARG2(__lsx_vsat_h, out0, 7, out1, 7, out2, 7, out3, 7, out0, out1, + out2, out3); + DUP4_ARG2(__lsx_vsat_h, out4, 7, out5, 7, out6, 7, out7, 7, out4, out5, + out6, out7); + out = __lsx_vpickev_b(out1, out0); + out = __lsx_vxori_b(out, 128); + __lsx_vst(out, dst, 0); + out = __lsx_vpickev_b(out3, out2); + out = __lsx_vxori_b(out, 128); + __lsx_vstx(out, dst, dst_stride); + out = __lsx_vpickev_b(out5, out4); + out = __lsx_vxori_b(out, 128); + __lsx_vstx(out, dst, dst_stride_x2); + out = __lsx_vpickev_b(out7, out6); + out = __lsx_vxori_b(out, 128); + __lsx_vstx(out, dst, dst_stride_x3); + dst += dst_stride_x4; + } +} + +static void common_vt_6t_8w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src7, src8, src9, src10; + __m128i src10_r, src32_r, src76_r, src98_r, src21_r, src43_r, src87_r; + __m128i src109_r, filt0, filt1, filt2; + __m128i tmp0, tmp1; + __m128i filt, out0_r, out1_r, out2_r, out3_r; + + src -= src_stride_x2; + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + filt2 = __lsx_vreplvei_h(filt, 2); + + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + src += src_stride_x4; + src4 = __lsx_vld(src, 0); + src += src_stride; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src4 = __lsx_vxori_b(src4, 128); + DUP4_ARG2(__lsx_vilvl_b, src1, src0, src3, src2, src2, src1, src4, src3, + src10_r, src32_r, src21_r, src43_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src7, src8, src9, src10); + DUP4_ARG2(__lsx_vxori_b, src7, 128, src8, 128, src9, 128, src10, 128, src7, + src8, src9, src10); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vilvl_b, src7, src4, src8, src7, src9, src8, src10, src9, + src76_r, src87_r, src98_r, src109_r); + out0_r = DPADD_H3(src10_r, src32_r, src76_r, filt0, filt1, filt2); + out1_r = DPADD_H3(src21_r, src43_r, src87_r, filt0, filt1, filt2); + out2_r = DPADD_H3(src32_r, src76_r, src98_r, filt0, filt1, filt2); + out3_r = DPADD_H3(src43_r, src87_r, src109_r, filt0, filt1, filt2); + DUP2_ARG3(__lsx_vssrarni_b_h, out1_r, out0_r, VP8_FILTER_SHIFT, out3_r, + out2_r, VP8_FILTER_SHIFT, tmp0, tmp1); + DUP2_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp0, tmp1); + __lsx_vstelm_d(tmp0, dst, 0, 0); + __lsx_vstelm_d(tmp0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(tmp1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(tmp1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + + src10_r = src76_r; + src32_r = src98_r; + src21_r = src87_r; + src43_r = src109_r; + src4 = src10; + } +} + +static void common_vt_6t_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8; + __m128i src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r; + __m128i src87_r, src10_l, src32_l, src54_l, src76_l, src21_l, src43_l; + __m128i src65_l, src87_l, filt0, filt1, filt2; + __m128i tmp0, tmp1, tmp2, tmp3; + __m128i filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l; + + src -= src_stride_x2; + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + filt2 = __lsx_vreplvei_h(filt, 2); + + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + src += src_stride_x4; + src4 = __lsx_vldx(src, 0); + src += src_stride; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src4 = __lsx_vxori_b(src4, 128); + DUP4_ARG2(__lsx_vilvl_b, src1, src0, src3, src2, src4, src3, src2, src1, + src10_r, src32_r, src43_r, src21_r); + DUP4_ARG2(__lsx_vilvh_b, src1, src0, src3, src2, src4, src3, src2, src1, + src10_l, src32_l, src43_l, src21_l); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src5, src6, src7, src8); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src5, 128, src6, 128, src7, 128, src8, 128, src5, + src6, src7, src8); + DUP4_ARG2(__lsx_vilvl_b, src5, src4, src6, src5, src7, src6, src8, src7, + src54_r, src65_r, src76_r, src87_r); + DUP4_ARG2(__lsx_vilvh_b, src5, src4, src6, src5, src7, src6, src8, src7, + src54_l, src65_l, src76_l, src87_l); + out0_r = DPADD_H3(src10_r, src32_r, src54_r, filt0, filt1, filt2); + out1_r = DPADD_H3(src21_r, src43_r, src65_r, filt0, filt1, filt2); + out2_r = DPADD_H3(src32_r, src54_r, src76_r, filt0, filt1, filt2); + out3_r = DPADD_H3(src43_r, src65_r, src87_r, filt0, filt1, filt2); + out0_l = DPADD_H3(src10_l, src32_l, src54_l, filt0, filt1, filt2); + out1_l = DPADD_H3(src21_l, src43_l, src65_l, filt0, filt1, filt2); + out2_l = DPADD_H3(src32_l, src54_l, src76_l, filt0, filt1, filt2); + out3_l = DPADD_H3(src43_l, src65_l, src87_l, filt0, filt1, filt2); + DUP4_ARG3(__lsx_vssrarni_b_h, out0_l, out0_r, VP8_FILTER_SHIFT, out1_l, + out1_r, VP8_FILTER_SHIFT, out2_l, out2_r, VP8_FILTER_SHIFT, + out3_l, out3_r, VP8_FILTER_SHIFT, tmp0, tmp1, tmp2, tmp3); + DUP4_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp2, 128, tmp3, 128, tmp0, + tmp1, tmp2, tmp3); + __lsx_vstx(tmp0, dst, 0); + __lsx_vstx(tmp1, dst, dst_stride); + __lsx_vstx(tmp2, dst, dst_stride_x2); + __lsx_vstx(tmp3, dst, dst_stride_x3); + dst += dst_stride_x4; + + src10_r = src54_r; + src32_r = src76_r; + src21_r = src65_r; + src43_r = src87_r; + src10_l = src54_l; + src32_l = src76_l; + src21_l = src65_l; + src43_l = src87_l; + src4 = src8; + } +} + +static void common_hv_6ht_6vt_8w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter_horiz, + const int8_t *filter_vert, + int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8; + __m128i filt_hz0, filt_hz1, filt_hz2; + __m128i mask0, mask1, mask2, vec0, vec1; + __m128i filt, filt_vt0, filt_vt1, filt_vt2; + __m128i hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6; + __m128i hz_out7, hz_out8, out0, out1, out2, out3, out4, out5, out6, out7; + __m128i tmp0, tmp1, tmp2, tmp3; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= (2 + src_stride_x2); + + filt = __lsx_vld(filter_horiz, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_hz0, filt_hz1); + filt_hz2 = __lsx_vreplvei_h(filt, 2); + + DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2); + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + src += src_stride_x4; + src4 = __lsx_vldx(src, 0); + src += src_stride; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src4 = __lsx_vxori_b(src4, 128); + + hz_out0 = HORIZ_6TAP_FILT(src0, src0, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out1 = HORIZ_6TAP_FILT(src1, src1, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out2 = HORIZ_6TAP_FILT(src2, src2, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out3 = HORIZ_6TAP_FILT(src3, src3, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out4 = HORIZ_6TAP_FILT(src4, src4, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + filt = __lsx_vld(filter_vert, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_vt0, filt_vt1); + filt_vt2 = __lsx_vreplvei_h(filt, 2); + + DUP4_ARG2(__lsx_vpackev_b, hz_out1, hz_out0, hz_out3, hz_out2, hz_out2, + hz_out1, hz_out4, hz_out3, out0, out1, out3, out4); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src5, src6, src7, src8); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src5, 128, src6, 128, src7, 128, src8, 128, src5, + src6, src7, src8); + hz_out5 = HORIZ_6TAP_FILT(src5, src5, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + out2 = __lsx_vpackev_b(hz_out5, hz_out4); + tmp0 = DPADD_H3(out0, out1, out2, filt_vt0, filt_vt1, filt_vt2); + + hz_out6 = HORIZ_6TAP_FILT(src6, src6, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + out5 = __lsx_vpackev_b(hz_out6, hz_out5); + tmp1 = DPADD_H3(out3, out4, out5, filt_vt0, filt_vt1, filt_vt2); + + hz_out7 = HORIZ_6TAP_FILT(src7, src7, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + out7 = __lsx_vpackev_b(hz_out7, hz_out6); + tmp2 = DPADD_H3(out1, out2, out7, filt_vt0, filt_vt1, filt_vt2); + + hz_out8 = HORIZ_6TAP_FILT(src8, src8, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + out6 = __lsx_vpackev_b(hz_out8, hz_out7); + tmp3 = DPADD_H3(out4, out5, out6, filt_vt0, filt_vt1, filt_vt2); + + DUP2_ARG3(__lsx_vssrarni_b_h, tmp1, tmp0, VP8_FILTER_SHIFT, tmp3, tmp2, + VP8_FILTER_SHIFT, vec0, vec1); + DUP2_ARG2(__lsx_vxori_b, vec0, 128, vec1, 128, vec0, vec1); + + __lsx_vstelm_d(vec0, dst, 0, 0); + __lsx_vstelm_d(vec0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(vec1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(vec1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + + hz_out4 = hz_out8; + out0 = out2; + out1 = out7; + out3 = out5; + out4 = out6; + } +} + +static void common_hv_6ht_6vt_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter_horiz, + const int8_t *filter_vert, + int32_t height) { + common_hv_6ht_6vt_8w_lsx(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert, height); + common_hv_6ht_6vt_8w_lsx(src + 8, src_stride, dst + 8, dst_stride, + filter_horiz, filter_vert, height); +} + +static void common_hz_4t_8w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, filt0, filt1, mask0, mask1; + __m128i tmp0, tmp1; + __m128i filt, out0, out1, out2, out3; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 1; + + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + mask1 = __lsx_vaddi_bu(mask0, 2); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + HORIZ_4TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, filt0, + filt1, out0, out1, out2, out3); + DUP2_ARG3(__lsx_vssrarni_b_h, out1, out0, VP8_FILTER_SHIFT, out3, out2, + VP8_FILTER_SHIFT, tmp0, tmp1); + DUP2_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp0, tmp1); + __lsx_vstelm_d(tmp0, dst, 0, 0); + __lsx_vstelm_d(tmp0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(tmp1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(tmp1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + } +} + +static void common_hz_4t_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, src7; + __m128i filt0, filt1, mask0, mask1; + __m128i filt, out0, out1, out2, out3, out4, out5, out6, out7; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 1; + + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + mask1 = __lsx_vaddi_bu(mask0, 2); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src2, src4, src6); + src += 8; + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src1, src3, src5, src7); + src += src_stride_x4 - 8; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + DUP4_ARG2(__lsx_vxori_b, src4, 128, src5, 128, src6, 128, src7, 128, src4, + src5, src6, src7); + HORIZ_4TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, filt0, + filt1, out0, out1, out2, out3); + HORIZ_4TAP_8WID_4VECS_FILT(src4, src5, src6, src7, mask0, mask1, filt0, + filt1, out4, out5, out6, out7); + DUP4_ARG3(__lsx_vssrarni_b_h, out1, out0, VP8_FILTER_SHIFT, out3, out2, + VP8_FILTER_SHIFT, out5, out4, VP8_FILTER_SHIFT, out7, out6, + VP8_FILTER_SHIFT, out0, out1, out2, out3); + DUP4_ARG2(__lsx_vxori_b, out0, 128, out1, 128, out2, 128, out3, 128, out0, + out1, out2, out3); + __lsx_vstx(out0, dst, 0); + __lsx_vstx(out1, dst, dst_stride); + __lsx_vstx(out2, dst, dst_stride_x2); + __lsx_vstx(out3, dst, dst_stride_x3); + dst += dst_stride_x4; + } +} + +static void common_vt_4t_8w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src7, src8, src9, src10; + __m128i src10_r, src72_r, src98_r, src21_r, src87_r, src109_r, filt0, filt1; + __m128i tmp0, tmp1; + __m128i filt, out0_r, out1_r, out2_r, out3_r; + + src -= src_stride; + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + + DUP2_ARG2(__lsx_vldx, src, 0, src, src_stride, src0, src1); + src2 = __lsx_vldx(src, src_stride_x2); + src += src_stride_x3; + + DUP2_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src0, src1); + src2 = __lsx_vxori_b(src2, 128); + DUP2_ARG2(__lsx_vilvl_b, src1, src0, src2, src1, src10_r, src21_r); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src7, src8, src9, src10); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src7, 128, src8, 128, src9, 128, src10, 128, src7, + src8, src9, src10); + DUP4_ARG2(__lsx_vilvl_b, src7, src2, src8, src7, src9, src8, src10, src9, + src72_r, src87_r, src98_r, src109_r); + out0_r = FILT_4TAP_DPADD_H(src10_r, src72_r, filt0, filt1); + out1_r = FILT_4TAP_DPADD_H(src21_r, src87_r, filt0, filt1); + out2_r = FILT_4TAP_DPADD_H(src72_r, src98_r, filt0, filt1); + out3_r = FILT_4TAP_DPADD_H(src87_r, src109_r, filt0, filt1); + DUP2_ARG3(__lsx_vssrarni_b_h, out1_r, out0_r, VP8_FILTER_SHIFT, out3_r, + out2_r, VP8_FILTER_SHIFT, tmp0, tmp1); + DUP2_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp0, tmp1); + __lsx_vstelm_d(tmp0, dst, 0, 0); + __lsx_vstelm_d(tmp0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(tmp1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(tmp1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + + src10_r = src98_r; + src21_r = src109_r; + src2 = src10; + } +} + +static void common_vt_4t_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter, int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6; + __m128i src10_r, src32_r, src54_r, src21_r, src43_r, src65_r, src10_l; + __m128i src32_l, src54_l, src21_l, src43_l, src65_l, filt0, filt1; + __m128i tmp0, tmp1, tmp2, tmp3; + __m128i filt, out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l; + + src -= src_stride; + filt = __lsx_vld(filter, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt0, filt1); + + DUP2_ARG2(__lsx_vldx, src, 0, src, src_stride, src0, src1); + src2 = __lsx_vldx(src, src_stride_x2); + src += src_stride_x3; + + DUP2_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src0, src1); + src2 = __lsx_vxori_b(src2, 128); + DUP2_ARG2(__lsx_vilvl_b, src1, src0, src2, src1, src10_r, src21_r); + DUP2_ARG2(__lsx_vilvh_b, src1, src0, src2, src1, src10_l, src21_l); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src3, src4, src5, src6); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src3, 128, src4, 128, src5, 128, src6, 128, src3, + src4, src5, src6); + DUP4_ARG2(__lsx_vilvl_b, src3, src2, src4, src3, src5, src4, src6, src5, + src32_r, src43_r, src54_r, src65_r); + DUP4_ARG2(__lsx_vilvh_b, src3, src2, src4, src3, src5, src4, src6, src5, + src32_l, src43_l, src54_l, src65_l); + out0_r = FILT_4TAP_DPADD_H(src10_r, src32_r, filt0, filt1); + out1_r = FILT_4TAP_DPADD_H(src21_r, src43_r, filt0, filt1); + out2_r = FILT_4TAP_DPADD_H(src32_r, src54_r, filt0, filt1); + out3_r = FILT_4TAP_DPADD_H(src43_r, src65_r, filt0, filt1); + out0_l = FILT_4TAP_DPADD_H(src10_l, src32_l, filt0, filt1); + out1_l = FILT_4TAP_DPADD_H(src21_l, src43_l, filt0, filt1); + out2_l = FILT_4TAP_DPADD_H(src32_l, src54_l, filt0, filt1); + out3_l = FILT_4TAP_DPADD_H(src43_l, src65_l, filt0, filt1); + DUP4_ARG3(__lsx_vssrarni_b_h, out0_l, out0_r, VP8_FILTER_SHIFT, out1_l, + out1_r, VP8_FILTER_SHIFT, out2_l, out2_r, VP8_FILTER_SHIFT, + out3_l, out3_r, VP8_FILTER_SHIFT, tmp0, tmp1, tmp2, tmp3); + DUP4_ARG2(__lsx_vxori_b, tmp0, 128, tmp1, 128, tmp2, 128, tmp3, 128, tmp0, + tmp1, tmp2, tmp3); + __lsx_vstx(tmp0, dst, 0); + __lsx_vstx(tmp1, dst, dst_stride); + __lsx_vstx(tmp2, dst, dst_stride_x2); + __lsx_vstx(tmp3, dst, dst_stride_x3); + dst += dst_stride_x4; + + src10_r = src54_r; + src21_r = src65_r; + src10_l = src54_l; + src21_l = src65_l; + src2 = src6; + } +} + +static inline void common_hv_4ht_4vt_8w_lsx( + uint8_t *RESTRICT src, int32_t src_stride, uint8_t *RESTRICT dst, + int32_t dst_stride, const int8_t *filter_horiz, const int8_t *filter_vert, + int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, filt_hz0, filt_hz1; + __m128i mask0, mask1, out0, out1; + __m128i filt, filt_vt0, filt_vt1, tmp0, tmp1, tmp2, tmp3; + __m128i hz_out0, hz_out1, hz_out2, hz_out3; + __m128i vec0, vec1, vec2, vec3, vec4; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 1 + src_stride; + + filt = __lsx_vld(filter_horiz, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_hz0, filt_hz1); + mask1 = __lsx_vaddi_bu(mask0, 2); + + DUP2_ARG2(__lsx_vldx, src, 0, src, src_stride, src0, src1); + src2 = __lsx_vldx(src, src_stride_x2); + src += src_stride_x3; + + DUP2_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src0, src1); + src2 = __lsx_vxori_b(src2, 128); + hz_out0 = HORIZ_4TAP_FILT(src0, src0, mask0, mask1, filt_hz0, filt_hz1); + hz_out1 = HORIZ_4TAP_FILT(src1, src1, mask0, mask1, filt_hz0, filt_hz1); + hz_out2 = HORIZ_4TAP_FILT(src2, src2, mask0, mask1, filt_hz0, filt_hz1); + DUP2_ARG2(__lsx_vpackev_b, hz_out1, hz_out0, hz_out2, hz_out1, vec0, vec2); + + filt = __lsx_vld(filter_vert, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_vt0, filt_vt1); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src3, src4, src5, src6); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src3, 128, src4, 128, src5, 128, src6, 128, src3, + src4, src5, src6); + hz_out3 = HORIZ_4TAP_FILT(src3, src3, mask0, mask1, filt_hz0, filt_hz1); + vec1 = __lsx_vpackev_b(hz_out3, hz_out2); + tmp0 = FILT_4TAP_DPADD_H(vec0, vec1, filt_vt0, filt_vt1); + + hz_out0 = HORIZ_4TAP_FILT(src4, src4, mask0, mask1, filt_hz0, filt_hz1); + vec3 = __lsx_vpackev_b(hz_out0, hz_out3); + tmp1 = FILT_4TAP_DPADD_H(vec2, vec3, filt_vt0, filt_vt1); + + hz_out1 = HORIZ_4TAP_FILT(src5, src5, mask0, mask1, filt_hz0, filt_hz1); + vec4 = __lsx_vpackev_b(hz_out1, hz_out0); + tmp2 = FILT_4TAP_DPADD_H(vec1, vec4, filt_vt0, filt_vt1); + + hz_out2 = HORIZ_4TAP_FILT(src6, src6, mask0, mask1, filt_hz0, filt_hz1); + DUP2_ARG2(__lsx_vpackev_b, hz_out0, hz_out3, hz_out2, hz_out1, vec0, vec1); + tmp3 = FILT_4TAP_DPADD_H(vec0, vec1, filt_vt0, filt_vt1); + + DUP2_ARG3(__lsx_vssrarni_b_h, tmp1, tmp0, 7, tmp3, tmp2, 7, out0, out1); + DUP2_ARG2(__lsx_vxori_b, out0, 128, out1, 128, out0, out1); + __lsx_vstelm_d(out0, dst, 0, 0); + __lsx_vstelm_d(out0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(out1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(out1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + + vec0 = vec4; + vec2 = vec1; + } +} + +static void common_hv_4ht_4vt_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter_horiz, + const int8_t *filter_vert, + int32_t height) { + common_hv_4ht_4vt_8w_lsx(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert, height); + common_hv_4ht_4vt_8w_lsx(src + 8, src_stride, dst + 8, dst_stride, + filter_horiz, filter_vert, height); +} + +static inline void common_hv_6ht_4vt_8w_lsx( + uint8_t *RESTRICT src, int32_t src_stride, uint8_t *RESTRICT dst, + int32_t dst_stride, const int8_t *filter_horiz, const int8_t *filter_vert, + int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + + __m128i src0, src1, src2, src3, src4, src5, src6; + __m128i filt_hz0, filt_hz1, filt_hz2, mask0, mask1, mask2; + __m128i filt, filt_vt0, filt_vt1, hz_out0, hz_out1, hz_out2, hz_out3; + __m128i tmp0, tmp1, tmp2, tmp3, vec0, vec1, vec2, vec3; + __m128i out0, out1; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= (2 + src_stride); + + filt = __lsx_vld(filter_horiz, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_hz0, filt_hz1); + filt_hz2 = __lsx_vreplvei_h(filt, 2); + DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2); + + DUP2_ARG2(__lsx_vldx, src, 0, src, src_stride, src0, src1); + src2 = __lsx_vldx(src, src_stride_x2); + src += src_stride_x3; + + DUP2_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src0, src1); + src2 = __lsx_vxori_b(src2, 128); + hz_out0 = HORIZ_6TAP_FILT(src0, src0, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out1 = HORIZ_6TAP_FILT(src1, src1, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + hz_out2 = HORIZ_6TAP_FILT(src2, src2, mask0, mask1, mask2, filt_hz0, filt_hz1, + filt_hz2); + DUP2_ARG2(__lsx_vpackev_b, hz_out1, hz_out0, hz_out2, hz_out1, vec0, vec2); + + filt = __lsx_vld(filter_vert, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_vt0, filt_vt1); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src3, src4, src5, src6); + src += src_stride_x4; + DUP4_ARG2(__lsx_vxori_b, src3, 128, src4, 128, src5, 128, src6, 128, src3, + src4, src5, src6); + + hz_out3 = HORIZ_6TAP_FILT(src3, src3, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + vec1 = __lsx_vpackev_b(hz_out3, hz_out2); + tmp0 = FILT_4TAP_DPADD_H(vec0, vec1, filt_vt0, filt_vt1); + + hz_out0 = HORIZ_6TAP_FILT(src4, src4, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + vec3 = __lsx_vpackev_b(hz_out0, hz_out3); + tmp1 = FILT_4TAP_DPADD_H(vec2, vec3, filt_vt0, filt_vt1); + + hz_out1 = HORIZ_6TAP_FILT(src5, src5, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + vec0 = __lsx_vpackev_b(hz_out1, hz_out0); + tmp2 = FILT_4TAP_DPADD_H(vec1, vec0, filt_vt0, filt_vt1); + + hz_out2 = HORIZ_6TAP_FILT(src6, src6, mask0, mask1, mask2, filt_hz0, + filt_hz1, filt_hz2); + DUP2_ARG2(__lsx_vpackev_b, hz_out0, hz_out3, hz_out2, hz_out1, vec1, vec2); + tmp3 = FILT_4TAP_DPADD_H(vec1, vec2, filt_vt0, filt_vt1); + + DUP2_ARG3(__lsx_vssrarni_b_h, tmp1, tmp0, 7, tmp3, tmp2, 7, out0, out1); + DUP2_ARG2(__lsx_vxori_b, out0, 128, out1, 128, out0, out1); + __lsx_vstelm_d(out0, dst, 0, 0); + __lsx_vstelm_d(out0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(out1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(out1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + } +} + +static void common_hv_6ht_4vt_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter_horiz, + const int8_t *filter_vert, + int32_t height) { + common_hv_6ht_4vt_8w_lsx(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert, height); + common_hv_6ht_4vt_8w_lsx(src + 8, src_stride, dst + 8, dst_stride, + filter_horiz, filter_vert, height); +} + +static inline void common_hv_4ht_6vt_8w_lsx( + uint8_t *RESTRICT src, int32_t src_stride, uint8_t *RESTRICT dst, + int32_t dst_stride, const int8_t *filter_horiz, const int8_t *filter_vert, + int32_t height) { + uint32_t loop_cnt; + int32_t src_stride_x2 = src_stride << 1; + int32_t src_stride_x3 = src_stride_x2 + src_stride; + int32_t src_stride_x4 = src_stride << 2; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8; + __m128i filt_hz0, filt_hz1, mask0, mask1; + __m128i filt, filt_vt0, filt_vt1, filt_vt2, tmp0, tmp1, tmp2, tmp3; + __m128i hz_out0, hz_out1, hz_out2, hz_out3, hz_out4, hz_out5, hz_out6; + __m128i hz_out7, hz_out8, out0, out1, out2, out3, out4, out5, out6, out7; + __m128i vec0, vec1; + + mask0 = __lsx_vld(vp8_mc_filt_mask_arr, 0); + src -= 1 + src_stride_x2; + + filt = __lsx_vld(filter_horiz, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_hz0, filt_hz1); + mask1 = __lsx_vaddi_bu(mask0, 2); + + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src0, src1, src2, src3); + src += src_stride_x4; + src4 = __lsx_vld(src, 0); + src += src_stride; + + DUP4_ARG2(__lsx_vxori_b, src0, 128, src1, 128, src2, 128, src3, 128, src0, + src1, src2, src3); + src4 = __lsx_vxori_b(src4, 128); + hz_out0 = HORIZ_4TAP_FILT(src0, src0, mask0, mask1, filt_hz0, filt_hz1); + hz_out1 = HORIZ_4TAP_FILT(src1, src1, mask0, mask1, filt_hz0, filt_hz1); + hz_out2 = HORIZ_4TAP_FILT(src2, src2, mask0, mask1, filt_hz0, filt_hz1); + hz_out3 = HORIZ_4TAP_FILT(src3, src3, mask0, mask1, filt_hz0, filt_hz1); + hz_out4 = HORIZ_4TAP_FILT(src4, src4, mask0, mask1, filt_hz0, filt_hz1); + DUP2_ARG2(__lsx_vpackev_b, hz_out1, hz_out0, hz_out3, hz_out2, out0, out1); + DUP2_ARG2(__lsx_vpackev_b, hz_out2, hz_out1, hz_out4, hz_out3, out3, out4); + + filt = __lsx_vld(filter_vert, 0); + DUP2_ARG2(__lsx_vreplvei_h, filt, 0, filt, 1, filt_vt0, filt_vt1); + filt_vt2 = __lsx_vreplvei_h(filt, 2); + + for (loop_cnt = (height >> 2); loop_cnt--;) { + DUP4_ARG2(__lsx_vldx, src, 0, src, src_stride, src, src_stride_x2, src, + src_stride_x3, src5, src6, src7, src8); + src += src_stride_x4; + + DUP4_ARG2(__lsx_vxori_b, src5, 128, src6, 128, src7, 128, src8, 128, src5, + src6, src7, src8); + hz_out5 = HORIZ_4TAP_FILT(src5, src5, mask0, mask1, filt_hz0, filt_hz1); + out2 = __lsx_vpackev_b(hz_out5, hz_out4); + tmp0 = DPADD_H3(out0, out1, out2, filt_vt0, filt_vt1, filt_vt2); + + hz_out6 = HORIZ_4TAP_FILT(src6, src6, mask0, mask1, filt_hz0, filt_hz1); + out5 = __lsx_vpackev_b(hz_out6, hz_out5); + tmp1 = DPADD_H3(out3, out4, out5, filt_vt0, filt_vt1, filt_vt2); + + hz_out7 = HORIZ_4TAP_FILT(src7, src7, mask0, mask1, filt_hz0, filt_hz1); + out6 = __lsx_vpackev_b(hz_out7, hz_out6); + tmp2 = DPADD_H3(out1, out2, out6, filt_vt0, filt_vt1, filt_vt2); + + hz_out8 = HORIZ_4TAP_FILT(src8, src8, mask0, mask1, filt_hz0, filt_hz1); + out7 = __lsx_vpackev_b(hz_out8, hz_out7); + tmp3 = DPADD_H3(out4, out5, out7, filt_vt0, filt_vt1, filt_vt2); + DUP2_ARG3(__lsx_vssrarni_b_h, tmp1, tmp0, 7, tmp3, tmp2, 7, vec0, vec1); + DUP2_ARG2(__lsx_vxori_b, vec0, 128, vec1, 128, vec0, vec1); + __lsx_vstelm_d(vec0, dst, 0, 0); + __lsx_vstelm_d(vec0, dst + dst_stride, 0, 1); + __lsx_vstelm_d(vec1, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(vec1, dst + dst_stride_x3, 0, 1); + dst += dst_stride_x4; + hz_out4 = hz_out8; + out0 = out2; + out1 = out6; + out3 = out5; + out4 = out7; + } +} + +static void common_hv_4ht_6vt_16w_lsx(uint8_t *RESTRICT src, int32_t src_stride, + uint8_t *RESTRICT dst, int32_t dst_stride, + const int8_t *filter_horiz, + const int8_t *filter_vert, + int32_t height) { + common_hv_4ht_6vt_8w_lsx(src, src_stride, dst, dst_stride, filter_horiz, + filter_vert, height); + common_hv_4ht_6vt_8w_lsx(src + 8, src_stride, dst + 8, dst_stride, + filter_horiz, filter_vert, height); +} + +typedef void (*PVp8SixtapPredictFunc1)( + uint8_t *RESTRICT src, int32_t src_stride, uint8_t *RESTRICT dst, + int32_t dst_stride, const int8_t *filter_horiz, const int8_t *filter_vert, + int32_t height); + +typedef void (*PVp8SixtapPredictFunc2)(uint8_t *RESTRICT src, + int32_t src_stride, + uint8_t *RESTRICT dst, + int32_t dst_stride, const int8_t *filter, + int32_t height); + +void vp8_sixtap_predict8x8_lsx(uint8_t *RESTRICT src, int32_t src_stride, + int32_t xoffset, int32_t yoffset, + uint8_t *RESTRICT dst, int32_t dst_stride) { + const int8_t *h_filter = vp8_subpel_filters_lsx[xoffset - 1]; + const int8_t *v_filter = vp8_subpel_filters_lsx[yoffset - 1]; + + static PVp8SixtapPredictFunc1 Predict8x8Funcs1[4] = { + common_hv_6ht_6vt_8w_lsx, + common_hv_6ht_4vt_8w_lsx, + common_hv_4ht_6vt_8w_lsx, + common_hv_4ht_4vt_8w_lsx, + }; + + static PVp8SixtapPredictFunc2 Predict8x8Funcs2[4] = { common_vt_6t_8w_lsx, + common_vt_4t_8w_lsx, + common_hz_6t_8w_lsx, + common_hz_4t_8w_lsx }; + + if (yoffset < 8 && xoffset < 8) { + if (yoffset) { + if (xoffset) { + switch (xoffset & 1) { + case 0: + switch (yoffset & 1) { + case 0: + Predict8x8Funcs1[0](src, src_stride, dst, dst_stride, h_filter, + v_filter, 8); + break; + + case 1: + Predict8x8Funcs1[1](src, src_stride, dst, dst_stride, h_filter, + v_filter + 1, 8); + break; + } + break; + + case 1: + switch (yoffset & 1) { + case 0: + Predict8x8Funcs1[2](src, src_stride, dst, dst_stride, + h_filter + 1, v_filter, 8); + break; + + case 1: + Predict8x8Funcs1[3](src, src_stride, dst, dst_stride, + h_filter + 1, v_filter + 1, 8); + break; + } + break; + } + } else { + switch (yoffset & 1) { + case 0: + Predict8x8Funcs2[0](src, src_stride, dst, dst_stride, v_filter, 8); + break; + + case 1: + Predict8x8Funcs2[1](src, src_stride, dst, dst_stride, v_filter + 1, + 8); + break; + } + } + } else { + switch (xoffset & 1) { + case 1: + Predict8x8Funcs2[3](src, src_stride, dst, dst_stride, h_filter + 1, + 8); + break; + } + switch (xoffset) { + case 0: vp8_copy_mem8x8(src, src_stride, dst, dst_stride); break; + case 2: + case 4: + case 6: + Predict8x8Funcs2[2](src, src_stride, dst, dst_stride, h_filter, 8); + break; + } + } + } +} + +void vp8_sixtap_predict16x16_lsx(uint8_t *RESTRICT src, int32_t src_stride, + int32_t xoffset, int32_t yoffset, + uint8_t *RESTRICT dst, int32_t dst_stride) { + const int8_t *h_filter = vp8_subpel_filters_lsx[xoffset - 1]; + const int8_t *v_filter = vp8_subpel_filters_lsx[yoffset - 1]; + + static PVp8SixtapPredictFunc1 Predict16x16Funcs1[4] = { + common_hv_6ht_6vt_16w_lsx, + common_hv_6ht_4vt_16w_lsx, + common_hv_4ht_6vt_16w_lsx, + common_hv_4ht_4vt_16w_lsx, + }; + + static PVp8SixtapPredictFunc2 Predict16x16Funcs2[4] = { + common_vt_6t_16w_lsx, common_vt_4t_16w_lsx, common_hz_6t_16w_lsx, + common_hz_4t_16w_lsx + }; + + if (yoffset < 8 && xoffset < 8) { + if (yoffset) { + if (xoffset) { + switch (xoffset & 1) { + case 0: + switch (yoffset & 1) { + case 0: + Predict16x16Funcs1[0](src, src_stride, dst, dst_stride, + h_filter, v_filter, 16); + break; + + case 1: + Predict16x16Funcs1[1](src, src_stride, dst, dst_stride, + h_filter, v_filter + 1, 16); + break; + } + break; + + case 1: + switch (yoffset & 1) { + case 0: + Predict16x16Funcs1[2](src, src_stride, dst, dst_stride, + h_filter + 1, v_filter, 16); + break; + + case 1: + Predict16x16Funcs1[3](src, src_stride, dst, dst_stride, + h_filter, v_filter + 1, 16); + break; + } + break; + } + } else { + switch (yoffset & 1) { + case 0: + Predict16x16Funcs2[0](src, src_stride, dst, dst_stride, v_filter, + 16); + break; + + case 1: + Predict16x16Funcs2[1](src, src_stride, dst, dst_stride, + v_filter + 1, 16); + break; + } + } + } else { + switch (xoffset & 1) { + case 1: + Predict16x16Funcs2[3](src, src_stride, dst, dst_stride, h_filter + 1, + 16); + break; + } + switch (xoffset) { + case 0: vp8_copy_mem16x16(src, src_stride, dst, dst_stride); break; + case 2: + case 4: + case 6: + Predict16x16Funcs2[2](src, src_stride, dst, dst_stride, h_filter, 16); + break; + } + } + } +} diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl b/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl index 8452b5e8545..40117e3677d 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl +++ b/chromium/third_party/libvpx/source/libvpx/vp8/common/rtcd_defs.pl @@ -47,13 +47,13 @@ specialize qw/vp8_dequant_idct_add_uv_block sse2 neon dspr2 msa mmi/; # Loopfilter # add_proto qw/void vp8_loop_filter_mbv/, "unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, struct loop_filter_info *lfi"; -specialize qw/vp8_loop_filter_mbv sse2 neon dspr2 msa mmi/; +specialize qw/vp8_loop_filter_mbv sse2 neon dspr2 msa mmi lsx/; add_proto qw/void vp8_loop_filter_bv/, "unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, struct loop_filter_info *lfi"; specialize qw/vp8_loop_filter_bv sse2 neon dspr2 msa mmi/; add_proto qw/void vp8_loop_filter_mbh/, "unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, struct loop_filter_info *lfi"; -specialize qw/vp8_loop_filter_mbh sse2 neon dspr2 msa mmi/; +specialize qw/vp8_loop_filter_mbh sse2 neon dspr2 msa mmi lsx/; add_proto qw/void vp8_loop_filter_bh/, "unsigned char *y_ptr, unsigned char *u_ptr, unsigned char *v_ptr, int y_stride, int uv_stride, struct loop_filter_info *lfi"; specialize qw/vp8_loop_filter_bh sse2 neon dspr2 msa mmi/; @@ -146,10 +146,10 @@ if (vpx_config("CONFIG_POSTPROC") eq "yes") { # Subpixel # add_proto qw/void vp8_sixtap_predict16x16/, "unsigned char *src_ptr, int src_pixels_per_line, int xoffset, int yoffset, unsigned char *dst_ptr, int dst_pitch"; -specialize qw/vp8_sixtap_predict16x16 sse2 ssse3 neon dspr2 msa mmi/; +specialize qw/vp8_sixtap_predict16x16 sse2 ssse3 neon dspr2 msa mmi lsx/; add_proto qw/void vp8_sixtap_predict8x8/, "unsigned char *src_ptr, int src_pixels_per_line, int xoffset, int yoffset, unsigned char *dst_ptr, int dst_pitch"; -specialize qw/vp8_sixtap_predict8x8 sse2 ssse3 neon dspr2 msa mmi/; +specialize qw/vp8_sixtap_predict8x8 sse2 ssse3 neon dspr2 msa mmi lsx/; add_proto qw/void vp8_sixtap_predict8x4/, "unsigned char *src_ptr, int src_pixels_per_line, int xoffset, int yoffset, unsigned char *dst_ptr, int dst_pitch"; specialize qw/vp8_sixtap_predict8x4 sse2 ssse3 neon dspr2 msa mmi/; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c index 59bce951e00..f09177c7f51 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/encoder/onyx_if.c @@ -36,6 +36,7 @@ #include "vp8/common/swapyv12buffer.h" #include "vp8/common/threading.h" #include "vpx_ports/system_state.h" +#include "vpx_ports/vpx_once.h" #include "vpx_ports/vpx_timer.h" #include "vpx_util/vpx_write_yuv_frame.h" #if VPX_ARCH_ARM @@ -394,16 +395,13 @@ static void setup_features(VP8_COMP *cpi) { static void dealloc_raw_frame_buffers(VP8_COMP *cpi); -void vp8_initialize_enc(void) { - static volatile int init_done = 0; - - if (!init_done) { - vpx_dsp_rtcd(); - vp8_init_intra_predictors(); - init_done = 1; - } +static void initialize_enc(void) { + vpx_dsp_rtcd(); + vp8_init_intra_predictors(); } +void vp8_initialize_enc(void) { once(initialize_enc); } + static void dealloc_compressor_data(VP8_COMP *cpi) { vpx_free(cpi->tplist); cpi->tplist = NULL; diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk index 286a93a0565..909924ce8dd 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_common.mk @@ -124,6 +124,10 @@ ifeq ($(CONFIG_POSTPROC),yes) VP8_COMMON_SRCS-$(HAVE_MSA) += common/mips/msa/mfqe_msa.c endif +# common (loongarch LSX intrinsics) +VP8_COMMON_SRCS-$(HAVE_LSX) += common/loongarch/loopfilter_filters_lsx.c +VP8_COMMON_SRCS-$(HAVE_LSX) += common/loongarch/sixtap_filter_lsx.c + # common (neon intrinsics) VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/loopfilter_arm.c VP8_COMMON_SRCS-$(HAVE_NEON) += common/arm/loopfilter_arm.h diff --git a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c index ab954c46f22..21fed0e8ed7 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c +++ b/chromium/third_party/libvpx/source/libvpx/vp8/vp8_cx_iface.c @@ -18,7 +18,6 @@ #include "vpx_mem/vpx_mem.h" #include "vpx_ports/static_assert.h" #include "vpx_ports/system_state.h" -#include "vpx_ports/vpx_once.h" #include "vpx_util/vpx_timestamp.h" #include "vp8/encoder/onyx_int.h" #include "vpx/vp8cx.h" @@ -694,7 +693,7 @@ static vpx_codec_err_t vp8e_init(vpx_codec_ctx_t *ctx, ctx->priv->enc.total_encoders = 1; } - once(vp8_initialize_enc); + vp8_initialize_enc(); res = validate_config(priv, &priv->cfg, &priv->vp8_cfg, 0); diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_rtcd.c b/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_rtcd.c index d8c870aa3f6..37762ca15a6 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_rtcd.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_rtcd.c @@ -12,8 +12,4 @@ #include "./vp9_rtcd.h" #include "vpx_ports/vpx_once.h" -void vp9_rtcd() { - // TODO(JBB): Remove this once, by insuring that both the encoder and - // decoder setup functions are protected by once(); - once(setup_rtcd_internal); -} +void vp9_rtcd() { once(setup_rtcd_internal); } diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_seg_common.h b/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_seg_common.h index b63e4f49990..5e71c2fca5b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_seg_common.h +++ b/chromium/third_party/libvpx/source/libvpx/vp9/common/vp9_seg_common.h @@ -25,6 +25,11 @@ extern "C" { #define PREDICTION_PROBS 3 +// Segment ID used to skip background encoding +#define BACKGROUND_SEG_SKIP_ID 3 +// Number of frames that don't skip after a key frame +#define FRAMES_NO_SKIPPING_AFTER_KEY 20 + // Segment level features. typedef enum { SEG_LVL_ALT_Q = 0, // Use alternate Quantizer .... diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_aq_cyclicrefresh.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_aq_cyclicrefresh.c index f06fe472683..e336179e907 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_aq_cyclicrefresh.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_aq_cyclicrefresh.c @@ -497,7 +497,9 @@ void vp9_cyclic_refresh_update_parameters(VP9_COMP *const cpi) { rc->avg_frame_low_motion < thresh_low_motion && rc->frames_since_key > 40) || (!cpi->use_svc && rc->avg_frame_qindex[INTER_FRAME] > qp_max_thresh && - rc->frames_since_key > 20)) { + rc->frames_since_key > 20) || + (cpi->roi.enabled && cpi->roi.skip[BACKGROUND_SEG_SKIP_ID] && + rc->frames_since_key > FRAMES_NO_SKIPPING_AFTER_KEY)) { cr->apply_cyclic_refresh = 0; return; } diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_bitstream.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_bitstream.c index c23e150a454..99cc2ee8312 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_bitstream.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_bitstream.c @@ -236,11 +236,11 @@ static void write_ref_frames(const VP9_COMMON *cm, const MACROBLOCKD *const xd, } } -static void pack_inter_mode_mvs( - VP9_COMP *cpi, const MACROBLOCKD *const xd, - const MB_MODE_INFO_EXT *const mbmi_ext, vpx_writer *w, - unsigned int *const max_mv_magnitude, - int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE]) { +static void pack_inter_mode_mvs(VP9_COMP *cpi, const MACROBLOCKD *const xd, + const MB_MODE_INFO_EXT *const mbmi_ext, + vpx_writer *w, + unsigned int *const max_mv_magnitude, + int interp_filter_selected[][SWITCHABLE]) { VP9_COMMON *const cm = &cpi->common; const nmv_context *nmvc = &cm->fc->nmvc; const struct segmentation *const seg = &cm->seg; @@ -373,11 +373,12 @@ static void write_mb_modes_kf(const VP9_COMMON *cm, const MACROBLOCKD *xd, write_intra_mode(w, mi->uv_mode, vp9_kf_uv_mode_prob[mi->mode]); } -static void write_modes_b( - VP9_COMP *cpi, MACROBLOCKD *const xd, const TileInfo *const tile, - vpx_writer *w, TOKENEXTRA **tok, const TOKENEXTRA *const tok_end, - int mi_row, int mi_col, unsigned int *const max_mv_magnitude, - int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE]) { +static void write_modes_b(VP9_COMP *cpi, MACROBLOCKD *const xd, + const TileInfo *const tile, vpx_writer *w, + TOKENEXTRA **tok, const TOKENEXTRA *const tok_end, + int mi_row, int mi_col, + unsigned int *const max_mv_magnitude, + int interp_filter_selected[][SWITCHABLE]) { const VP9_COMMON *const cm = &cpi->common; const MB_MODE_INFO_EXT *const mbmi_ext = cpi->td.mb.mbmi_ext_base + (mi_row * cm->mi_cols + mi_col); @@ -422,12 +423,12 @@ static void write_partition(const VP9_COMMON *const cm, } } -static void write_modes_sb( - VP9_COMP *cpi, MACROBLOCKD *const xd, const TileInfo *const tile, - vpx_writer *w, TOKENEXTRA **tok, const TOKENEXTRA *const tok_end, - int mi_row, int mi_col, BLOCK_SIZE bsize, - unsigned int *const max_mv_magnitude, - int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE]) { +static void write_modes_sb(VP9_COMP *cpi, MACROBLOCKD *const xd, + const TileInfo *const tile, vpx_writer *w, + TOKENEXTRA **tok, const TOKENEXTRA *const tok_end, + int mi_row, int mi_col, BLOCK_SIZE bsize, + unsigned int *const max_mv_magnitude, + int interp_filter_selected[][SWITCHABLE]) { const VP9_COMMON *const cm = &cpi->common; const int bsl = b_width_log2_lookup[bsize]; const int bs = (1 << bsl) / 4; @@ -485,11 +486,10 @@ static void write_modes_sb( update_partition_context(xd, mi_row, mi_col, subsize, bsize); } -static void write_modes( - VP9_COMP *cpi, MACROBLOCKD *const xd, const TileInfo *const tile, - vpx_writer *w, int tile_row, int tile_col, - unsigned int *const max_mv_magnitude, - int interp_filter_selected[MAX_REF_FRAMES][SWITCHABLE]) { +static void write_modes(VP9_COMP *cpi, MACROBLOCKD *const xd, + const TileInfo *const tile, vpx_writer *w, int tile_row, + int tile_col, unsigned int *const max_mv_magnitude, + int interp_filter_selected[][SWITCHABLE]) { const VP9_COMMON *const cm = &cpi->common; int mi_row, mi_col, tile_sb_row; TOKENEXTRA *tok = NULL; @@ -563,7 +563,7 @@ static void update_coef_probs_common(vpx_writer *const bc, VP9_COMP *cpi, for (t = 0; t < entropy_nodes_update; ++t) { vpx_prob newp = new_coef_probs[i][j][k][l][t]; const vpx_prob oldp = old_coef_probs[i][j][k][l][t]; - int s; + int64_t s; int u = 0; if (t == PIVOT_NODE) s = vp9_prob_diff_update_savings_search_model( @@ -600,7 +600,7 @@ static void update_coef_probs_common(vpx_writer *const bc, VP9_COMP *cpi, vpx_prob newp = new_coef_probs[i][j][k][l][t]; vpx_prob *oldp = old_coef_probs[i][j][k][l] + t; const vpx_prob upd = DIFF_UPDATE_PROB; - int s; + int64_t s; int u = 0; if (t == PIVOT_NODE) s = vp9_prob_diff_update_savings_search_model( @@ -636,7 +636,7 @@ static void update_coef_probs_common(vpx_writer *const bc, VP9_COMP *cpi, for (t = 0; t < entropy_nodes_update; ++t) { vpx_prob newp = new_coef_probs[i][j][k][l][t]; vpx_prob *oldp = old_coef_probs[i][j][k][l] + t; - int s; + int64_t s; int u = 0; if (t == PIVOT_NODE) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_cost.h b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_cost.h index 638d72a9161..ee0033fa315 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_cost.h +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_cost.h @@ -29,9 +29,8 @@ extern const uint16_t vp9_prob_cost[256]; #define vp9_cost_bit(prob, bit) vp9_cost_zero((bit) ? 256 - (prob) : (prob)) -static INLINE unsigned int cost_branch256(const unsigned int ct[2], - vpx_prob p) { - return ct[0] * vp9_cost_zero(p) + ct[1] * vp9_cost_one(p); +static INLINE uint64_t cost_branch256(const unsigned int ct[2], vpx_prob p) { + return (uint64_t)ct[0] * vp9_cost_zero(p) + (uint64_t)ct[1] * vp9_cost_one(p); } static INLINE int treed_cost(vpx_tree tree, const vpx_prob *probs, int bits, diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encodeframe.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encodeframe.c index 131c4887f27..fc4089865dc 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encodeframe.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encodeframe.c @@ -5513,16 +5513,6 @@ static void encode_nonrd_sb_row(VP9_COMP *cpi, ThreadData *td, x->arf_frame_usage = 0; x->lastgolden_frame_usage = 0; - if (seg->enabled) { - const uint8_t *const map = - seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map; - int segment_id = get_segment_id(cm, map, BLOCK_64X64, mi_row, mi_col); - seg_skip = segfeature_active(seg, segment_id, SEG_LVL_SKIP); - if (seg_skip) { - partition_search_type = FIXED_PARTITION; - } - } - if (cpi->compute_source_sad_onepass && cpi->sf.use_source_sad) { int shift = cpi->Source->y_stride * (mi_row << 3) + (mi_col << 3); int sb_offset2 = ((cm->mi_cols + 7) >> 3) * (mi_row >> 3) + (mi_col >> 3); @@ -5534,6 +5524,38 @@ static void encode_nonrd_sb_row(VP9_COMP *cpi, ThreadData *td, partition_search_type = REFERENCE_PARTITION; } + if (seg->enabled) { + const uint8_t *const map = + seg->update_map ? cpi->segmentation_map : cm->last_frame_seg_map; + int segment_id = get_segment_id(cm, map, BLOCK_64X64, mi_row, mi_col); + seg_skip = segfeature_active(seg, segment_id, SEG_LVL_SKIP); + + if (cpi->roi.enabled && cpi->roi.skip[BACKGROUND_SEG_SKIP_ID] && + cpi->rc.frames_since_key > FRAMES_NO_SKIPPING_AFTER_KEY && + x->content_state_sb > kLowSadLowSumdiff) { + // For ROI with skip, force segment = 0 (no skip) over whole + // superblock to avoid artifacts if temporal change in source_sad is + // not 0. + int xi, yi; + const int bw = num_8x8_blocks_wide_lookup[BLOCK_64X64]; + const int bh = num_8x8_blocks_high_lookup[BLOCK_64X64]; + const int xmis = VPXMIN(cm->mi_cols - mi_col, bw); + const int ymis = VPXMIN(cm->mi_rows - mi_row, bh); + const int block_index = mi_row * cm->mi_cols + mi_col; + set_mode_info_offsets(cm, x, xd, mi_row, mi_col); + for (yi = 0; yi < ymis; yi++) + for (xi = 0; xi < xmis; xi++) { + int map_offset = block_index + yi * cm->mi_cols + xi; + cpi->segmentation_map[map_offset] = 0; + } + set_segment_index(cpi, x, mi_row, mi_col, BLOCK_64X64, 0); + seg_skip = 0; + } + if (seg_skip) { + partition_search_type = FIXED_PARTITION; + } + } + // Set the partition type of the 64X64 block switch (partition_search_type) { case VAR_BASED_PARTITION: diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.c index 8fdd86916fd..4609a6bb26c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.c @@ -25,6 +25,7 @@ #endif #include "vpx_ports/mem.h" #include "vpx_ports/system_state.h" +#include "vpx_ports/vpx_once.h" #include "vpx_ports/vpx_timer.h" #if CONFIG_BITSTREAM_DEBUG || CONFIG_MISMATCH_DEBUG #include "vpx_util/vpx_debug_util.h" @@ -585,8 +586,6 @@ static void apply_roi_map(VP9_COMP *cpi) { int ref_frame[8]; int internal_delta_q[MAX_SEGMENTS]; int i; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; // TODO(jianj): Investigate why ROI not working in speed < 5 or in non // realtime mode. @@ -618,7 +617,7 @@ static void apply_roi_map(VP9_COMP *cpi) { } if (skip[i] != 0) { vp9_enable_segfeature(seg, i, SEG_LVL_SKIP); - vp9_set_segdata(seg, i, SEG_LVL_SKIP, skip[i]); + vp9_set_segdata(seg, i, SEG_LVL_SKIP, 0); } if (ref_frame[i] >= 0) { int valid_ref = 1; @@ -627,7 +626,7 @@ static void apply_roi_map(VP9_COMP *cpi) { valid_ref = 0; // If GOLDEN is selected, make sure it's set as reference. if (ref_frame[i] == GOLDEN_FRAME && - !(cpi->ref_frame_flags & flag_list[ref_frame[i]])) { + !(cpi->ref_frame_flags & ref_frame_to_flag(ref_frame[i]))) { valid_ref = 0; } // GOLDEN was updated in previous encoded frame, so GOLDEN and LAST are @@ -929,24 +928,21 @@ static void vp9_swap_mi_and_prev_mi(VP9_COMMON *cm) { cm->prev_mi_grid_visible = cm->prev_mi_grid_base + cm->mi_stride + 1; } -void vp9_initialize_enc(void) { - static volatile int init_done = 0; - - if (!init_done) { - vp9_rtcd(); - vpx_dsp_rtcd(); - vpx_scale_rtcd(); - vp9_init_intra_predictors(); - vp9_init_me_luts(); - vp9_rc_init_minq_luts(); - vp9_entropy_mv_init(); +static void initialize_enc(void) { + vp9_rtcd(); + vpx_dsp_rtcd(); + vpx_scale_rtcd(); + vp9_init_intra_predictors(); + vp9_init_me_luts(); + vp9_rc_init_minq_luts(); + vp9_entropy_mv_init(); #if !CONFIG_REALTIME_ONLY - vp9_temporal_filter_init(); + vp9_temporal_filter_init(); #endif - init_done = 1; - } } +void vp9_initialize_enc(void) { once(initialize_enc); } + static void dealloc_compressor_data(VP9_COMP *cpi) { VP9_COMMON *const cm = &cpi->common; int i; @@ -2223,6 +2219,7 @@ static INLINE void vpx_img_chroma_subsampling(vpx_img_fmt_t fmt, switch (fmt) { case VPX_IMG_FMT_I420: case VPX_IMG_FMT_YV12: + case VPX_IMG_FMT_NV12: case VPX_IMG_FMT_I422: case VPX_IMG_FMT_I42016: case VPX_IMG_FMT_I42216: *subsampling_x = 1; break; @@ -2233,6 +2230,7 @@ static INLINE void vpx_img_chroma_subsampling(vpx_img_fmt_t fmt, case VPX_IMG_FMT_I420: case VPX_IMG_FMT_I440: case VPX_IMG_FMT_YV12: + case VPX_IMG_FMT_NV12: case VPX_IMG_FMT_I42016: case VPX_IMG_FMT_I44016: *subsampling_y = 1; break; default: *subsampling_y = 0; break; @@ -4118,11 +4116,22 @@ static int encode_without_recode_loop(VP9_COMP *cpi, size_t *size, vp9_alt_ref_aq_setup_map(cpi->alt_ref_aq, cpi); } else { #endif - if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) { + // If ROI is enabled and skip feature is used for segmentation, apply cyclic + // refresh but not apply ROI for skip for the first 20 frames (defined by + // FRAMES_NO_SKIPPING_AFTER_KEY) after key frame to improve quality. + if (cpi->roi.enabled && !frame_is_intra_only(cm)) { + if (cpi->roi.skip[BACKGROUND_SEG_SKIP_ID]) { + if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) + vp9_cyclic_refresh_setup(cpi); + if (cpi->rc.frames_since_key > FRAMES_NO_SKIPPING_AFTER_KEY) + apply_roi_map(cpi); + } else { + apply_roi_map(cpi); + } + } else if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) { vp9_cyclic_refresh_setup(cpi); - } else if (cpi->roi.enabled && !frame_is_intra_only(cm)) { - apply_roi_map(cpi); } + #if !CONFIG_REALTIME_ONLY } #endif diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.h b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.h index 9774a64ccfc..1d589452507 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.h +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_encoder.h @@ -1196,6 +1196,13 @@ static INLINE int frame_is_kf_gf_arf(const VP9_COMP *cpi) { (cpi->refresh_golden_frame && !cpi->rc.is_src_frame_alt_ref); } +static INLINE int ref_frame_to_flag(int8_t ref_frame) { + static const int kVp9RefFlagList[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, + VP9_ALT_FLAG }; + assert(ref_frame >= LAST_FRAME && ref_frame <= ALTREF_FRAME); + return kVp9RefFlagList[ref_frame]; +} + static INLINE int get_ref_frame_map_idx(const VP9_COMP *cpi, MV_REFERENCE_FRAME ref_frame) { if (ref_frame == LAST_FRAME) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_pickmode.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_pickmode.c index 695fd484fc6..c8e167f25bd 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_pickmode.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_pickmode.c @@ -1247,7 +1247,7 @@ static INLINE void find_predictors( VP9_COMP *cpi, MACROBLOCK *x, MV_REFERENCE_FRAME ref_frame, int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES], int const_motion[MAX_REF_FRAMES], int *ref_frame_skip_mask, - const int flag_list[4], TileDataEnc *tile_data, int mi_row, int mi_col, + TileDataEnc *tile_data, int mi_row, int mi_col, struct buf_2d yv12_mb[4][MAX_MB_PLANE], BLOCK_SIZE bsize, int force_skip_low_temp_var, int comp_pred_allowed) { VP9_COMMON *const cm = &cpi->common; @@ -1259,7 +1259,7 @@ static INLINE void find_predictors( frame_mv[NEWMV][ref_frame].as_int = INVALID_MV; frame_mv[ZEROMV][ref_frame].as_int = 0; // this needs various further optimizations. to be continued.. - if ((cpi->ref_frame_flags & flag_list[ref_frame]) && (yv12 != NULL)) { + if ((cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)) && (yv12 != NULL)) { int_mv *const candidates = x->mbmi_ext->ref_mvs[ref_frame]; const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf; vp9_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, sf); @@ -1690,8 +1690,6 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES]; uint8_t mode_checked[MB_MODE_COUNT][MAX_REF_FRAMES]; struct buf_2d yv12_mb[4][MAX_MB_PLANE]; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; RD_COST this_rdc, best_rdc; // var_y and sse_y are saved to be used in skipping checking unsigned int var_y = UINT_MAX; @@ -1925,14 +1923,14 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, // constrain the inter mode to only test zero motion. if (cpi->use_svc && svc->force_zero_mode_spatial_ref && svc->spatial_layer_id > 0 && !gf_temporal_ref) { - if (cpi->ref_frame_flags & flag_list[LAST_FRAME]) { + if (cpi->ref_frame_flags & VP9_LAST_FLAG) { struct scale_factors *const sf = &cm->frame_refs[LAST_FRAME - 1].sf; if (vp9_is_scaled(sf)) { svc_force_zero_mode[LAST_FRAME - 1] = 1; inter_layer_ref = LAST_FRAME; } } - if (cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) { + if (cpi->ref_frame_flags & VP9_GOLD_FLAG) { struct scale_factors *const sf = &cm->frame_refs[GOLDEN_FRAME - 1].sf; if (vp9_is_scaled(sf)) { svc_force_zero_mode[GOLDEN_FRAME - 1] = 1; @@ -1957,7 +1955,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, cpi->rc.avg_frame_low_motion < 60)) usable_ref_frame = LAST_FRAME; - if (!((cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) && + if (!((cpi->ref_frame_flags & VP9_GOLD_FLAG) && !svc_force_zero_mode[GOLDEN_FRAME - 1] && !force_skip_low_temp_var)) use_golden_nonzeromv = 0; @@ -1985,12 +1983,11 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, // Skip find_predictor if the reference frame is not in the // ref_frame_flags (i.e., not used as a reference for this frame). skip_ref_find_pred[ref_frame] = - !(cpi->ref_frame_flags & flag_list[ref_frame]); + !(cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)); if (!skip_ref_find_pred[ref_frame]) { find_predictors(cpi, x, ref_frame, frame_mv, const_motion, - &ref_frame_skip_mask, flag_list, tile_data, mi_row, - mi_col, yv12_mb, bsize, force_skip_low_temp_var, - comp_modes > 0); + &ref_frame_skip_mask, tile_data, mi_row, mi_col, yv12_mb, + bsize, force_skip_low_temp_var, comp_modes > 0); } } @@ -2014,7 +2011,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, // than current layer: force check of GF-ZEROMV before early exit // due to skip flag. if (svc->spatial_layer_id > 0 && no_scaling && - (cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) && + (cpi->ref_frame_flags & VP9_GOLD_FLAG) && cm->base_qindex > svc->lower_layer_qindex + 10) force_test_gf_zeromv = 1; @@ -2094,7 +2091,8 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, if (comp_pred) { if (!cpi->allow_comp_inter_inter) continue; // Skip compound inter modes if ARF is not available. - if (!(cpi->ref_frame_flags & flag_list[second_ref_frame])) continue; + if (!(cpi->ref_frame_flags & ref_frame_to_flag(second_ref_frame))) + continue; // Do not allow compound prediction if the segment level reference frame // feature is in use as in this case there can only be one reference. if (segfeature_active(seg, mi->segment_id, SEG_LVL_REF_FRAME)) continue; @@ -2107,7 +2105,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, (!cpi->use_svc && sse_zeromv_normalized < thresh_skip_golden))) continue; - if (!(cpi->ref_frame_flags & flag_list[ref_frame])) continue; + if (!(cpi->ref_frame_flags & ref_frame_to_flag(ref_frame))) continue; // For screen content. If zero_temp_sad source is computed: skip // non-zero motion check for stationary blocks. If the superblock is @@ -2190,7 +2188,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, if (usable_ref_frame < ALTREF_FRAME) { if (!force_skip_low_temp_var && usable_ref_frame > LAST_FRAME) { i = (ref_frame == LAST_FRAME) ? GOLDEN_FRAME : LAST_FRAME; - if ((cpi->ref_frame_flags & flag_list[i])) + if ((cpi->ref_frame_flags & ref_frame_to_flag(i))) if (x->pred_mv_sad[ref_frame] > (x->pred_mv_sad[i] << 1)) ref_frame_skip_mask |= (1 << ref_frame); } @@ -2199,9 +2197,9 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, ref_frame == ALTREF_FRAME)) { int ref1 = (ref_frame == GOLDEN_FRAME) ? LAST_FRAME : GOLDEN_FRAME; int ref2 = (ref_frame == ALTREF_FRAME) ? LAST_FRAME : ALTREF_FRAME; - if (((cpi->ref_frame_flags & flag_list[ref1]) && + if (((cpi->ref_frame_flags & ref_frame_to_flag(ref1)) && (x->pred_mv_sad[ref_frame] > (x->pred_mv_sad[ref1] << 1))) || - ((cpi->ref_frame_flags & flag_list[ref2]) && + ((cpi->ref_frame_flags & ref_frame_to_flag(ref2)) && (x->pred_mv_sad[ref_frame] > (x->pred_mv_sad[ref2] << 1)))) ref_frame_skip_mask |= (1 << ref_frame); } @@ -2488,7 +2486,7 @@ void vp9_pick_inter_mode(VP9_COMP *cpi, MACROBLOCK *x, TileDataEnc *tile_data, perform_intra_pred = svc->temporal_layer_id == 0 || svc->layer_context[svc->temporal_layer_id].is_key_frame || - !(cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) || + !(cpi->ref_frame_flags & VP9_GOLD_FLAG) || (!svc->layer_context[svc->temporal_layer_id].is_key_frame && svc_force_zero_mode[best_pickmode.best_ref_frame - 1]); inter_mode_thresh = (inter_mode_thresh << 1) + inter_mode_thresh; @@ -2747,8 +2745,6 @@ void vp9_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x, int mi_row, MV_REFERENCE_FRAME best_ref_frame = NONE; unsigned char segment_id = mi->segment_id; struct buf_2d yv12_mb[4][MAX_MB_PLANE]; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; int64_t best_rd = INT64_MAX; b_mode_info bsi[MAX_REF_FRAMES][4]; int ref_frame_skip_mask = 0; @@ -2764,7 +2760,8 @@ void vp9_pick_inter_mode_sub8x8(VP9_COMP *cpi, MACROBLOCK *x, int mi_row, int_mv dummy_mv[2]; x->pred_mv_sad[ref_frame] = INT_MAX; - if ((cpi->ref_frame_flags & flag_list[ref_frame]) && (yv12 != NULL)) { + if ((cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)) && + (yv12 != NULL)) { int_mv *const candidates = mbmi_ext->ref_mvs[ref_frame]; const struct scale_factors *const sf = &cm->frame_refs[ref_frame - 1].sf; vp9_setup_pred_block(xd, yv12_mb[ref_frame], yv12, mi_row, mi_col, sf, diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_ratectrl.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_ratectrl.c index ac346115fb2..08529739149 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_ratectrl.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_ratectrl.c @@ -2214,7 +2214,6 @@ static void set_intra_only_frame(VP9_COMP *cpi) { // only 3 reference buffers can be updated, but for temporal layers > 1 // we generally need to use buffer slots 4 and 5. if ((cm->current_video_frame == 0 && svc->number_temporal_layers > 1) || - svc->temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS || svc->number_spatial_layers > 3 || svc->number_temporal_layers > 3 || svc->number_spatial_layers == 1) return; @@ -2235,11 +2234,15 @@ static void set_intra_only_frame(VP9_COMP *cpi) { cpi->lst_fb_idx = -1; cpi->gld_fb_idx = -1; cpi->alt_fb_idx = -1; + svc->update_buffer_slot[0] = 0; // For intra-only frame we need to refresh all slots that were // being used for the base layer (fb_idx_base[i] == 1). // Start with assigning last first, then golden and then alt. for (i = 0; i < REF_FRAMES; ++i) { - if (svc->fb_idx_base[i] == 1) count++; + if (svc->fb_idx_base[i] == 1) { + svc->update_buffer_slot[0] |= 1 << i; + count++; + } if (count == 1 && cpi->lst_fb_idx == -1) cpi->lst_fb_idx = i; if (count == 2 && cpi->gld_fb_idx == -1) cpi->gld_fb_idx = i; if (count == 3 && cpi->alt_fb_idx == -1) cpi->alt_fb_idx = i; @@ -2248,6 +2251,12 @@ static void set_intra_only_frame(VP9_COMP *cpi) { // to the lst_fb_idx. if (cpi->gld_fb_idx == -1) cpi->gld_fb_idx = cpi->lst_fb_idx; if (cpi->alt_fb_idx == -1) cpi->alt_fb_idx = cpi->lst_fb_idx; + if (svc->temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) { + cpi->ext_refresh_last_frame = 0; + cpi->ext_refresh_golden_frame = 0; + cpi->ext_refresh_alt_ref_frame = 0; + cpi->ref_frame_flags = 0; + } } } @@ -2390,6 +2399,9 @@ void vp9_rc_get_svc_params(VP9_COMP *cpi) { set_intra_only_frame(cpi); target = vp9_calc_iframe_target_size_one_pass_cbr(cpi); } + // Overlay frame predicts from LAST (intra-only) + if (svc->previous_frame_is_intra_only) cpi->ref_frame_flags |= VP9_LAST_FLAG; + // Any update/change of global cyclic refresh parameters (amount/delta-qp) // should be done here, before the frame qp is selected. if (cpi->oxcf.aq_mode == CYCLIC_REFRESH_AQ) diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_rdopt.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_rdopt.c index a1687dcf461..0171a057207 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_rdopt.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_rdopt.c @@ -3315,8 +3315,6 @@ void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, TileDataEnc *tile_data, int_mv single_newmv[MAX_REF_FRAMES] = { { 0 } }; INTERP_FILTER single_inter_filter[MB_MODE_COUNT][MAX_REF_FRAMES]; int single_skippable[MB_MODE_COUNT][MAX_REF_FRAMES]; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; int64_t best_rd = best_rd_so_far; int64_t best_pred_diff[REFERENCE_MODES]; int64_t best_pred_rd[REFERENCE_MODES]; @@ -3392,7 +3390,7 @@ void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, TileDataEnc *tile_data, for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { x->pred_mv_sad[ref_frame] = INT_MAX; - if ((cpi->ref_frame_flags & flag_list[ref_frame]) && + if ((cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)) && !(is_rect_partition && (ctx->skip_ref_frame_mask & (1 << ref_frame)))) { assert(get_ref_frame_buffer(cpi, ref_frame) != NULL); setup_buffer_inter(cpi, x, ref_frame, bsize, mi_row, mi_col, @@ -3403,7 +3401,7 @@ void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, TileDataEnc *tile_data, } for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { - if (!(cpi->ref_frame_flags & flag_list[ref_frame])) { + if (!(cpi->ref_frame_flags & ref_frame_to_flag(ref_frame))) { // Skip checking missing references in both single and compound reference // modes. Note that a mode will be skipped if both reference frames // are masked out. @@ -3609,7 +3607,8 @@ void vp9_rd_pick_inter_mode_sb(VP9_COMP *cpi, TileDataEnc *tile_data, continue; // Skip compound inter modes if ARF is not available. - if (!(cpi->ref_frame_flags & flag_list[second_ref_frame])) continue; + if (!(cpi->ref_frame_flags & ref_frame_to_flag(second_ref_frame))) + continue; // Do not allow compound prediction if the segment level reference frame // feature is in use as in this case there can only be one reference. @@ -4140,8 +4139,6 @@ void vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, TileDataEnc *tile_data, int comp_pred, i; int_mv frame_mv[MB_MODE_COUNT][MAX_REF_FRAMES]; struct buf_2d yv12_mb[4][MAX_MB_PLANE]; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; int64_t best_rd = best_rd_so_far; int64_t best_yrd = best_rd_so_far; // FIXME(rbultje) more precise int64_t best_pred_diff[REFERENCE_MODES]; @@ -4191,7 +4188,7 @@ void vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, TileDataEnc *tile_data, rd_cost->rate = INT_MAX; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) { - if (cpi->ref_frame_flags & flag_list[ref_frame]) { + if (cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)) { setup_buffer_inter(cpi, x, ref_frame, bsize, mi_row, mi_col, frame_mv[NEARESTMV], frame_mv[NEARMV], yv12_mb); } else { @@ -4276,7 +4273,8 @@ void vp9_rd_pick_inter_mode_sub8x8(VP9_COMP *cpi, TileDataEnc *tile_data, cm->ref_frame_sign_bias[second_ref_frame]) continue; - if (!(cpi->ref_frame_flags & flag_list[second_ref_frame])) continue; + if (!(cpi->ref_frame_flags & ref_frame_to_flag(second_ref_frame))) + continue; // Do not allow compound prediction if the segment level reference frame // feature is in use as in this case there can only be one reference. if (segfeature_active(seg, segment_id, SEG_LVL_REF_FRAME)) continue; diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_speed_features.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_speed_features.c index 81695e9156d..7d7b2c3fb46 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_speed_features.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_speed_features.c @@ -495,11 +495,10 @@ static void set_rt_speed_feature_framesize_independent( (cpi->external_resize == 1 || cpi->oxcf.resize_mode == RESIZE_DYNAMIC)) { MV_REFERENCE_FRAME ref_frame; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame); - if (yv12 != NULL && (cpi->ref_frame_flags & flag_list[ref_frame])) { + if (yv12 != NULL && + (cpi->ref_frame_flags & ref_frame_to_flag(ref_frame))) { const struct scale_factors *const scale_fac = &cm->frame_refs[ref_frame - 1].sf; if (vp9_is_scaled(scale_fac)) sf->reference_masking = 0; diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.c index 19bbd5373fd..3953253dbbe 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.c @@ -114,19 +114,20 @@ void vp9_write_prob_diff_update(vpx_writer *w, vpx_prob newp, vpx_prob oldp) { encode_term_subexp(w, delp); } -int vp9_prob_diff_update_savings_search(const unsigned int *ct, vpx_prob oldp, - vpx_prob *bestp, vpx_prob upd) { - const int old_b = cost_branch256(ct, oldp); - int bestsavings = 0; +int64_t vp9_prob_diff_update_savings_search(const unsigned int *ct, + vpx_prob oldp, vpx_prob *bestp, + vpx_prob upd) { + const int64_t old_b = cost_branch256(ct, oldp); + int64_t bestsavings = 0; vpx_prob newp, bestnewp = oldp; const int step = *bestp > oldp ? -1 : 1; const int upd_cost = vp9_cost_one(upd) - vp9_cost_zero(upd); if (old_b > upd_cost + (MIN_DELP_BITS << VP9_PROB_COST_SHIFT)) { for (newp = *bestp; newp != oldp; newp += step) { - const int new_b = cost_branch256(ct, newp); - const int update_b = prob_diff_update_cost(newp, oldp) + upd_cost; - const int savings = old_b - new_b - update_b; + const int64_t new_b = cost_branch256(ct, newp); + const int64_t update_b = prob_diff_update_cost(newp, oldp) + upd_cost; + const int64_t savings = old_b - new_b - update_b; if (savings > bestsavings) { bestsavings = savings; bestnewp = newp; @@ -137,15 +138,15 @@ int vp9_prob_diff_update_savings_search(const unsigned int *ct, vpx_prob oldp, return bestsavings; } -int vp9_prob_diff_update_savings_search_model(const unsigned int *ct, - const vpx_prob oldp, - vpx_prob *bestp, vpx_prob upd, - int stepsize) { - int i, old_b, new_b, update_b, savings, bestsavings; - int newp; - const int step_sign = *bestp > oldp ? -1 : 1; - const int step = stepsize * step_sign; - const int upd_cost = vp9_cost_one(upd) - vp9_cost_zero(upd); +int64_t vp9_prob_diff_update_savings_search_model(const unsigned int *ct, + const vpx_prob oldp, + vpx_prob *bestp, vpx_prob upd, + int stepsize) { + int64_t i, old_b, new_b, update_b, savings, bestsavings; + int64_t newp; + const int64_t step_sign = *bestp > oldp ? -1 : 1; + const int64_t step = stepsize * step_sign; + const int64_t upd_cost = vp9_cost_one(upd) - vp9_cost_zero(upd); const vpx_prob *newplist, *oldplist; vpx_prob bestnewp; oldplist = vp9_pareto8_full[oldp - 1]; @@ -162,14 +163,14 @@ int vp9_prob_diff_update_savings_search_model(const unsigned int *ct, for (newp = *bestp; (newp - oldp) * step_sign < 0; newp += step) { if (newp < 1 || newp > 255) continue; newplist = vp9_pareto8_full[newp - 1]; - new_b = cost_branch256(ct + 2 * PIVOT_NODE, newp); + new_b = cost_branch256(ct + 2 * PIVOT_NODE, (vpx_prob)newp); for (i = UNCONSTRAINED_NODES; i < ENTROPY_NODES; ++i) new_b += cost_branch256(ct + 2 * i, newplist[i - UNCONSTRAINED_NODES]); - update_b = prob_diff_update_cost(newp, oldp) + upd_cost; + update_b = prob_diff_update_cost((vpx_prob)newp, oldp) + upd_cost; savings = old_b - new_b - update_b; if (savings > bestsavings) { bestsavings = savings; - bestnewp = newp; + bestnewp = (vpx_prob)newp; } } } @@ -182,7 +183,7 @@ void vp9_cond_prob_diff_update(vpx_writer *w, vpx_prob *oldp, const unsigned int ct[2]) { const vpx_prob upd = DIFF_UPDATE_PROB; vpx_prob newp = get_binary_prob(ct[0], ct[1]); - const int savings = + const int64_t savings = vp9_prob_diff_update_savings_search(ct, *oldp, &newp, upd); assert(newp >= 1); if (savings > 0) { diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.h b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.h index f0d544b5270..2d016d24c5e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.h +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_subexp.h @@ -25,13 +25,14 @@ void vp9_write_prob_diff_update(struct vpx_writer *w, vpx_prob newp, void vp9_cond_prob_diff_update(struct vpx_writer *w, vpx_prob *oldp, const unsigned int ct[2]); -int vp9_prob_diff_update_savings_search(const unsigned int *ct, vpx_prob oldp, - vpx_prob *bestp, vpx_prob upd); - -int vp9_prob_diff_update_savings_search_model(const unsigned int *ct, - const vpx_prob oldp, - vpx_prob *bestp, vpx_prob upd, - int stepsize); +int64_t vp9_prob_diff_update_savings_search(const unsigned int *ct, + vpx_prob oldp, vpx_prob *bestp, + vpx_prob upd); + +int64_t vp9_prob_diff_update_savings_search_model(const unsigned int *ct, + const vpx_prob oldp, + vpx_prob *bestp, vpx_prob upd, + int stepsize); #ifdef __cplusplus } // extern "C" diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_svc_layercontext.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_svc_layercontext.c index ad3a8f7afac..a57a70ab16c 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_svc_layercontext.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/vp9_svc_layercontext.c @@ -462,30 +462,27 @@ static void reset_fb_idx_unused(VP9_COMP *const cpi) { // fb_idx for that reference to the first one used/referenced. // This is to avoid setting fb_idx for a reference to a slot that is not // used/needed (i.e., since that reference is not referenced or refreshed). - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; MV_REFERENCE_FRAME ref_frame; MV_REFERENCE_FRAME first_ref = 0; int first_fb_idx = 0; int fb_idx[3] = { cpi->lst_fb_idx, cpi->gld_fb_idx, cpi->alt_fb_idx }; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ref_frame++) { - if (cpi->ref_frame_flags & flag_list[ref_frame]) { + if (cpi->ref_frame_flags & ref_frame_to_flag(ref_frame)) { first_ref = ref_frame; first_fb_idx = fb_idx[ref_frame - 1]; break; } } if (first_ref > 0) { - if (first_ref != LAST_FRAME && - !(cpi->ref_frame_flags & flag_list[LAST_FRAME]) && + if (first_ref != LAST_FRAME && !(cpi->ref_frame_flags & VP9_LAST_FLAG) && !cpi->ext_refresh_last_frame) cpi->lst_fb_idx = first_fb_idx; else if (first_ref != GOLDEN_FRAME && - !(cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]) && + !(cpi->ref_frame_flags & VP9_GOLD_FLAG) && !cpi->ext_refresh_golden_frame) cpi->gld_fb_idx = first_fb_idx; else if (first_ref != ALTREF_FRAME && - !(cpi->ref_frame_flags & flag_list[ALTREF_FRAME]) && + !(cpi->ref_frame_flags & VP9_ALT_FLAG) && !cpi->ext_refresh_alt_ref_frame) cpi->alt_fb_idx = first_fb_idx; } @@ -730,8 +727,6 @@ static void set_flags_and_fb_idx_bypass_via_set_ref_frame_config( void vp9_copy_flags_ref_update_idx(VP9_COMP *const cpi) { SVC *const svc = &cpi->svc; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; int sl = svc->spatial_layer_id; svc->lst_fb_idx[sl] = cpi->lst_fb_idx; svc->gld_fb_idx[sl] = cpi->gld_fb_idx; @@ -754,12 +749,9 @@ void vp9_copy_flags_ref_update_idx(VP9_COMP *const cpi) { svc->update_golden[sl] = (uint8_t)cpi->refresh_golden_frame; svc->update_altref[sl] = (uint8_t)cpi->refresh_alt_ref_frame; - svc->reference_last[sl] = - (uint8_t)(cpi->ref_frame_flags & flag_list[LAST_FRAME]); - svc->reference_golden[sl] = - (uint8_t)(cpi->ref_frame_flags & flag_list[GOLDEN_FRAME]); - svc->reference_altref[sl] = - (uint8_t)(cpi->ref_frame_flags & flag_list[ALTREF_FRAME]); + svc->reference_last[sl] = (uint8_t)(cpi->ref_frame_flags & VP9_LAST_FLAG); + svc->reference_golden[sl] = (uint8_t)(cpi->ref_frame_flags & VP9_GOLD_FLAG); + svc->reference_altref[sl] = (uint8_t)(cpi->ref_frame_flags & VP9_ALT_FLAG); } int vp9_one_pass_cbr_svc_start_layer(VP9_COMP *const cpi) { @@ -1080,15 +1072,14 @@ void vp9_svc_constrain_inter_layer_pred(VP9_COMP *const cpi) { svc->disable_inter_layer_pred == INTER_LAYER_PRED_OFF || svc->drop_spatial_layer[sl - 1]) { MV_REFERENCE_FRAME ref_frame; - static const int flag_list[4] = { 0, VP9_LAST_FLAG, VP9_GOLD_FLAG, - VP9_ALT_FLAG }; for (ref_frame = LAST_FRAME; ref_frame <= ALTREF_FRAME; ++ref_frame) { const YV12_BUFFER_CONFIG *yv12 = get_ref_frame_buffer(cpi, ref_frame); - if (yv12 != NULL && (cpi->ref_frame_flags & flag_list[ref_frame])) { + if (yv12 != NULL && + (cpi->ref_frame_flags & ref_frame_to_flag(ref_frame))) { const struct scale_factors *const scale_fac = &cm->frame_refs[ref_frame - 1].sf; if (vp9_is_scaled(scale_fac)) { - cpi->ref_frame_flags &= (~flag_list[ref_frame]); + cpi->ref_frame_flags &= (~ref_frame_to_flag(ref_frame)); // Point golden/altref frame buffer index to last. if (!svc->simulcast_mode) { if (ref_frame == GOLDEN_FRAME) @@ -1243,6 +1234,7 @@ void vp9_svc_check_spatial_layer_sync(VP9_COMP *const cpi) { void vp9_svc_update_ref_frame_buffer_idx(VP9_COMP *const cpi) { SVC *const svc = &cpi->svc; + int i = 0; // Update the usage of frame buffer index for base spatial layers. if (svc->spatial_layer_id == 0) { if ((cpi->ref_frame_flags & VP9_LAST_FLAG) || cpi->refresh_last_frame) @@ -1251,6 +1243,11 @@ void vp9_svc_update_ref_frame_buffer_idx(VP9_COMP *const cpi) { svc->fb_idx_base[cpi->gld_fb_idx] = 1; if ((cpi->ref_frame_flags & VP9_ALT_FLAG) || cpi->refresh_alt_ref_frame) svc->fb_idx_base[cpi->alt_fb_idx] = 1; + // For bypass/flexible mode: check for refresh slots. + if (svc->temporal_layering_mode == VP9E_TEMPORAL_LAYERING_MODE_BYPASS) { + for (i = 0; i < REF_FRAMES; ++i) + if (svc->update_buffer_slot[0] & (1 << i)) svc->fb_idx_base[i] = 1; + } } } diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_diamond_search_sad_avx.c b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_diamond_search_sad_avx.c index 4be6a5ea02e..fcf50eb2a77 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_diamond_search_sad_avx.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/encoder/x86/vp9_diamond_search_sad_avx.c @@ -282,7 +282,14 @@ int vp9_diamond_search_sad_avx(const MACROBLOCK *x, // Update the global minimum if the local minimum is smaller if (LIKELY(local_best_sad < best_sad)) { +#if defined(__GNUC__) && __GNUC__ >= 4 && !defined(__clang__) +#pragma GCC diagnostic push +#pragma GCC diagnostic ignored "-Wmaybe-uninitialized" +#endif new_bmv = ((const int_mv *)&v_these_mv_w)[local_best_idx]; +#if defined(__GNUC__) && __GNUC__ >= 4 && !defined(__clang__) +#pragma GCC diagnostic pop +#endif new_best_address = ((const uint8_t **)v_blocka)[local_best_idx]; best_sad = local_best_sad; diff --git a/chromium/third_party/libvpx/source/libvpx/vp9/vp9_cx_iface.c b/chromium/third_party/libvpx/source/libvpx/vp9/vp9_cx_iface.c index cc4081c4f56..76274437c6f 100644 --- a/chromium/third_party/libvpx/source/libvpx/vp9/vp9_cx_iface.c +++ b/chromium/third_party/libvpx/source/libvpx/vp9/vp9_cx_iface.c @@ -15,7 +15,6 @@ #include "vpx/vpx_encoder.h" #include "vpx/vpx_ext_ratectrl.h" #include "vpx_dsp/psnr.h" -#include "vpx_ports/vpx_once.h" #include "vpx_ports/static_assert.h" #include "vpx_ports/system_state.h" #include "vpx_util/vpx_timestamp.h" @@ -381,8 +380,8 @@ static vpx_codec_err_t validate_img(vpx_codec_alg_priv_t *ctx, case VPX_IMG_FMT_I440: if (ctx->cfg.g_profile != (unsigned int)PROFILE_1) { ERROR( - "Invalid image format. I422, I444, I440, NV12 images are " - "not supported in profile."); + "Invalid image format. I422, I444, I440 images are not supported " + "in profile."); } break; case VPX_IMG_FMT_I42216: @@ -397,8 +396,8 @@ static vpx_codec_err_t validate_img(vpx_codec_alg_priv_t *ctx, break; default: ERROR( - "Invalid image format. Only YV12, I420, I422, I444 images are " - "supported."); + "Invalid image format. Only YV12, I420, I422, I444, I440, NV12 " + "images are supported."); break; } @@ -1096,7 +1095,7 @@ static vpx_codec_err_t encoder_init(vpx_codec_ctx_t *ctx, } priv->extra_cfg = default_extra_cfg; - once(vp9_initialize_enc); + vp9_initialize_enc(); res = validate_config(priv, &priv->cfg, &priv->extra_cfg); diff --git a/chromium/third_party/libvpx/source/libvpx/vpx/internal/vpx_ratectrl_rtc.h b/chromium/third_party/libvpx/source/libvpx/vpx/internal/vpx_ratectrl_rtc.h index 0474e0a85b0..65398c654d4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx/internal/vpx_ratectrl_rtc.h +++ b/chromium/third_party/libvpx/source/libvpx/vpx/internal/vpx_ratectrl_rtc.h @@ -8,8 +8,8 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef VPX_VPX_RATECTRL_RTC_H_ -#define VPX_VPX_RATECTRL_RTC_H_ +#ifndef VPX_VPX_INTERNAL_VPX_RATECTRL_RTC_H_ +#define VPX_VPX_INTERNAL_VPX_RATECTRL_RTC_H_ #include "vpx/vpx_encoder.h" @@ -59,4 +59,4 @@ struct VpxRateControlRtcConfig { int aq_mode; }; } // namespace libvpx -#endif +#endif // VPX_VPX_INTERNAL_VPX_RATECTRL_RTC_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vpx/vp8cx.h b/chromium/third_party/libvpx/source/libvpx/vpx/vp8cx.h index 47c38d3b5ee..6b02aa8657e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx/vp8cx.h +++ b/chromium/third_party/libvpx/source/libvpx/vpx/vp8cx.h @@ -952,28 +952,12 @@ typedef struct vpx_svc_spatial_layer_sync { * */ -VPX_CTRL_USE_TYPE(VP8E_SET_FRAME_FLAGS, int) -#define VPX_CTRL_VP8E_SET_FRAME_FLAGS -VPX_CTRL_USE_TYPE(VP8E_SET_TEMPORAL_LAYER_ID, int) -#define VPX_CTRL_VP8E_SET_TEMPORAL_LAYER_ID VPX_CTRL_USE_TYPE(VP8E_SET_ROI_MAP, vpx_roi_map_t *) #define VPX_CTRL_VP8E_SET_ROI_MAP -VPX_CTRL_USE_TYPE(VP9E_SET_ROI_MAP, vpx_roi_map_t *) -#define VPX_CTRL_VP9E_SET_ROI_MAP VPX_CTRL_USE_TYPE(VP8E_SET_ACTIVEMAP, vpx_active_map_t *) #define VPX_CTRL_VP8E_SET_ACTIVEMAP VPX_CTRL_USE_TYPE(VP8E_SET_SCALEMODE, vpx_scaling_mode_t *) #define VPX_CTRL_VP8E_SET_SCALEMODE - -VPX_CTRL_USE_TYPE(VP9E_SET_SVC, int) -#define VPX_CTRL_VP9E_SET_SVC -VPX_CTRL_USE_TYPE(VP9E_SET_SVC_PARAMETERS, void *) -#define VPX_CTRL_VP9E_SET_SVC_PARAMETERS -VPX_CTRL_USE_TYPE(VP9E_REGISTER_CX_CALLBACK, void *) -#define VPX_CTRL_VP9E_REGISTER_CX_CALLBACK -VPX_CTRL_USE_TYPE(VP9E_SET_SVC_LAYER_ID, vpx_svc_layer_id_t *) -#define VPX_CTRL_VP9E_SET_SVC_LAYER_ID - VPX_CTRL_USE_TYPE(VP8E_SET_CPUUSED, int) #define VPX_CTRL_VP8E_SET_CPUUSED VPX_CTRL_USE_TYPE(VP8E_SET_ENABLEAUTOALTREF, unsigned int) @@ -986,7 +970,10 @@ VPX_CTRL_USE_TYPE(VP8E_SET_STATIC_THRESHOLD, unsigned int) #define VPX_CTRL_VP8E_SET_STATIC_THRESHOLD VPX_CTRL_USE_TYPE(VP8E_SET_TOKEN_PARTITIONS, int) /* vp8e_token_partitions */ #define VPX_CTRL_VP8E_SET_TOKEN_PARTITIONS - +VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER, int *) +#define VPX_CTRL_VP8E_GET_LAST_QUANTIZER +VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER_64, int *) +#define VPX_CTRL_VP8E_GET_LAST_QUANTIZER_64 VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_MAXFRAMES, unsigned int) #define VPX_CTRL_VP8E_SET_ARNR_MAXFRAMES VPX_CTRL_USE_TYPE(VP8E_SET_ARNR_STRENGTH, unsigned int) @@ -997,133 +984,107 @@ VPX_CTRL_USE_TYPE(VP8E_SET_TUNING, int) /* vp8e_tuning */ #define VPX_CTRL_VP8E_SET_TUNING VPX_CTRL_USE_TYPE(VP8E_SET_CQ_LEVEL, unsigned int) #define VPX_CTRL_VP8E_SET_CQ_LEVEL - -VPX_CTRL_USE_TYPE(VP9E_SET_TILE_COLUMNS, int) -#define VPX_CTRL_VP9E_SET_TILE_COLUMNS -VPX_CTRL_USE_TYPE(VP9E_SET_TILE_ROWS, int) -#define VPX_CTRL_VP9E_SET_TILE_ROWS - -VPX_CTRL_USE_TYPE(VP9E_SET_TPL, int) -#define VPX_CTRL_VP9E_SET_TPL - -VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER, int *) -#define VPX_CTRL_VP8E_GET_LAST_QUANTIZER -VPX_CTRL_USE_TYPE(VP8E_GET_LAST_QUANTIZER_64, int *) -#define VPX_CTRL_VP8E_GET_LAST_QUANTIZER_64 -VPX_CTRL_USE_TYPE(VP9E_GET_LAST_QUANTIZER_SVC_LAYERS, int *) -#define VPX_CTRL_VP9E_GET_LAST_QUANTIZER_SVC_LAYERS - -VPX_CTRL_USE_TYPE(VP9E_GET_SVC_LAYER_ID, vpx_svc_layer_id_t *) -#define VPX_CTRL_VP9E_GET_SVC_LAYER_ID - VPX_CTRL_USE_TYPE(VP8E_SET_MAX_INTRA_BITRATE_PCT, unsigned int) #define VPX_CTRL_VP8E_SET_MAX_INTRA_BITRATE_PCT +VPX_CTRL_USE_TYPE(VP8E_SET_FRAME_FLAGS, int) +#define VPX_CTRL_VP8E_SET_FRAME_FLAGS VPX_CTRL_USE_TYPE(VP9E_SET_MAX_INTER_BITRATE_PCT, unsigned int) #define VPX_CTRL_VP9E_SET_MAX_INTER_BITRATE_PCT - -VPX_CTRL_USE_TYPE(VP8E_SET_GF_CBR_BOOST_PCT, unsigned int) -#define VPX_CTRL_VP8E_SET_GF_CBR_BOOST_PCT - -VPX_CTRL_USE_TYPE(VP8E_SET_SCREEN_CONTENT_MODE, unsigned int) -#define VPX_CTRL_VP8E_SET_SCREEN_CONTENT_MODE - VPX_CTRL_USE_TYPE(VP9E_SET_GF_CBR_BOOST_PCT, unsigned int) #define VPX_CTRL_VP9E_SET_GF_CBR_BOOST_PCT - +VPX_CTRL_USE_TYPE(VP8E_SET_TEMPORAL_LAYER_ID, int) +#define VPX_CTRL_VP8E_SET_TEMPORAL_LAYER_ID +VPX_CTRL_USE_TYPE(VP8E_SET_SCREEN_CONTENT_MODE, unsigned int) +#define VPX_CTRL_VP8E_SET_SCREEN_CONTENT_MODE VPX_CTRL_USE_TYPE(VP9E_SET_LOSSLESS, unsigned int) #define VPX_CTRL_VP9E_SET_LOSSLESS - +VPX_CTRL_USE_TYPE(VP9E_SET_TILE_COLUMNS, int) +#define VPX_CTRL_VP9E_SET_TILE_COLUMNS +VPX_CTRL_USE_TYPE(VP9E_SET_TILE_ROWS, int) +#define VPX_CTRL_VP9E_SET_TILE_ROWS VPX_CTRL_USE_TYPE(VP9E_SET_FRAME_PARALLEL_DECODING, unsigned int) #define VPX_CTRL_VP9E_SET_FRAME_PARALLEL_DECODING - VPX_CTRL_USE_TYPE(VP9E_SET_AQ_MODE, unsigned int) #define VPX_CTRL_VP9E_SET_AQ_MODE - -VPX_CTRL_USE_TYPE(VP9E_SET_ALT_REF_AQ, int) -#define VPX_CTRL_VP9E_SET_ALT_REF_AQ - VPX_CTRL_USE_TYPE(VP9E_SET_FRAME_PERIODIC_BOOST, unsigned int) #define VPX_CTRL_VP9E_SET_FRAME_PERIODIC_BOOST - VPX_CTRL_USE_TYPE(VP9E_SET_NOISE_SENSITIVITY, unsigned int) #define VPX_CTRL_VP9E_SET_NOISE_SENSITIVITY - +VPX_CTRL_USE_TYPE(VP9E_SET_SVC, int) +#define VPX_CTRL_VP9E_SET_SVC +VPX_CTRL_USE_TYPE(VP9E_SET_ROI_MAP, vpx_roi_map_t *) +#define VPX_CTRL_VP9E_SET_ROI_MAP +VPX_CTRL_USE_TYPE(VP9E_SET_SVC_PARAMETERS, void *) +#define VPX_CTRL_VP9E_SET_SVC_PARAMETERS +VPX_CTRL_USE_TYPE(VP9E_SET_SVC_LAYER_ID, vpx_svc_layer_id_t *) +#define VPX_CTRL_VP9E_SET_SVC_LAYER_ID VPX_CTRL_USE_TYPE(VP9E_SET_TUNE_CONTENT, int) /* vp9e_tune_content */ #define VPX_CTRL_VP9E_SET_TUNE_CONTENT - +VPX_CTRL_USE_TYPE(VP9E_GET_SVC_LAYER_ID, vpx_svc_layer_id_t *) +#define VPX_CTRL_VP9E_GET_SVC_LAYER_ID +VPX_CTRL_USE_TYPE(VP9E_REGISTER_CX_CALLBACK, void *) +#define VPX_CTRL_VP9E_REGISTER_CX_CALLBACK VPX_CTRL_USE_TYPE(VP9E_SET_COLOR_SPACE, int) #define VPX_CTRL_VP9E_SET_COLOR_SPACE - +VPX_CTRL_USE_TYPE(VP9E_SET_TEMPORAL_LAYERING_MODE, + int) /* VP9E_TEMPORAL_LAYERING_MODE */ +#define VPX_CTRL_VP9E_SET_TEMPORAL_LAYERING_MODE VPX_CTRL_USE_TYPE(VP9E_SET_MIN_GF_INTERVAL, unsigned int) #define VPX_CTRL_VP9E_SET_MIN_GF_INTERVAL - VPX_CTRL_USE_TYPE(VP9E_SET_MAX_GF_INTERVAL, unsigned int) #define VPX_CTRL_VP9E_SET_MAX_GF_INTERVAL - VPX_CTRL_USE_TYPE(VP9E_GET_ACTIVEMAP, vpx_active_map_t *) #define VPX_CTRL_VP9E_GET_ACTIVEMAP - VPX_CTRL_USE_TYPE(VP9E_SET_COLOR_RANGE, int) #define VPX_CTRL_VP9E_SET_COLOR_RANGE - VPX_CTRL_USE_TYPE(VP9E_SET_SVC_REF_FRAME_CONFIG, vpx_svc_ref_frame_config_t *) #define VPX_CTRL_VP9E_SET_SVC_REF_FRAME_CONFIG - VPX_CTRL_USE_TYPE(VP9E_SET_RENDER_SIZE, int *) #define VPX_CTRL_VP9E_SET_RENDER_SIZE - VPX_CTRL_USE_TYPE(VP9E_SET_TARGET_LEVEL, unsigned int) #define VPX_CTRL_VP9E_SET_TARGET_LEVEL - VPX_CTRL_USE_TYPE(VP9E_SET_ROW_MT, unsigned int) #define VPX_CTRL_VP9E_SET_ROW_MT - VPX_CTRL_USE_TYPE(VP9E_GET_LEVEL, int *) #define VPX_CTRL_VP9E_GET_LEVEL - -VPX_CTRL_USE_TYPE(VP9E_GET_LOOPFILTER_LEVEL, int *) -#define VPX_CTRL_VP9E_GET_LOOPFILTER_LEVEL - +VPX_CTRL_USE_TYPE(VP9E_SET_ALT_REF_AQ, int) +#define VPX_CTRL_VP9E_SET_ALT_REF_AQ +VPX_CTRL_USE_TYPE(VP8E_SET_GF_CBR_BOOST_PCT, unsigned int) +#define VPX_CTRL_VP8E_SET_GF_CBR_BOOST_PCT VPX_CTRL_USE_TYPE(VP9E_ENABLE_MOTION_VECTOR_UNIT_TEST, unsigned int) #define VPX_CTRL_VP9E_ENABLE_MOTION_VECTOR_UNIT_TEST - VPX_CTRL_USE_TYPE(VP9E_SET_SVC_INTER_LAYER_PRED, unsigned int) #define VPX_CTRL_VP9E_SET_SVC_INTER_LAYER_PRED - VPX_CTRL_USE_TYPE(VP9E_SET_SVC_FRAME_DROP_LAYER, vpx_svc_frame_drop_t *) #define VPX_CTRL_VP9E_SET_SVC_FRAME_DROP_LAYER - VPX_CTRL_USE_TYPE(VP9E_GET_SVC_REF_FRAME_CONFIG, vpx_svc_ref_frame_config_t *) #define VPX_CTRL_VP9E_GET_SVC_REF_FRAME_CONFIG - VPX_CTRL_USE_TYPE(VP9E_SET_SVC_GF_TEMPORAL_REF, unsigned int) #define VPX_CTRL_VP9E_SET_SVC_GF_TEMPORAL_REF - VPX_CTRL_USE_TYPE(VP9E_SET_SVC_SPATIAL_LAYER_SYNC, vpx_svc_spatial_layer_sync_t *) #define VPX_CTRL_VP9E_SET_SVC_SPATIAL_LAYER_SYNC - +VPX_CTRL_USE_TYPE(VP9E_SET_TPL, int) +#define VPX_CTRL_VP9E_SET_TPL VPX_CTRL_USE_TYPE(VP9E_SET_POSTENCODE_DROP, unsigned int) #define VPX_CTRL_VP9E_SET_POSTENCODE_DROP - VPX_CTRL_USE_TYPE(VP9E_SET_DELTA_Q_UV, int) #define VPX_CTRL_VP9E_SET_DELTA_Q_UV - VPX_CTRL_USE_TYPE(VP9E_SET_DISABLE_OVERSHOOT_MAXQ_CBR, int) #define VPX_CTRL_VP9E_SET_DISABLE_OVERSHOOT_MAXQ_CBR - VPX_CTRL_USE_TYPE(VP9E_SET_DISABLE_LOOPFILTER, int) #define VPX_CTRL_VP9E_SET_DISABLE_LOOPFILTER - +VPX_CTRL_USE_TYPE(VP9E_SET_EXTERNAL_RATE_CONTROL, vpx_rc_funcs_t *) +#define VPX_CTRL_VP9E_SET_EXTERNAL_RATE_CONTROL VPX_CTRL_USE_TYPE(VP9E_SET_RTC_EXTERNAL_RATECTRL, int) #define VPX_CTRL_VP9E_SET_RTC_EXTERNAL_RATECTRL - +VPX_CTRL_USE_TYPE(VP9E_GET_LOOPFILTER_LEVEL, int *) +#define VPX_CTRL_VP9E_GET_LOOPFILTER_LEVEL +VPX_CTRL_USE_TYPE(VP9E_GET_LAST_QUANTIZER_SVC_LAYERS, int *) +#define VPX_CTRL_VP9E_GET_LAST_QUANTIZER_SVC_LAYERS VPX_CTRL_USE_TYPE(VP8E_SET_RTC_EXTERNAL_RATECTRL, int) #define VPX_CTRL_VP8E_SET_RTC_EXTERNAL_RATECTRL -VPX_CTRL_USE_TYPE(VP9E_SET_EXTERNAL_RATE_CONTROL, vpx_rc_funcs_t *) -#define VPX_CTRL_VP9E_SET_EXTERNAL_RATE_CONTROL - /*!\endcond */ /*! @} - end defgroup vp8_encoder */ #ifdef __cplusplus diff --git a/chromium/third_party/libvpx/source/libvpx/vpx/vp8dx.h b/chromium/third_party/libvpx/source/libvpx/vpx/vp8dx.h index af92f21ae30..506a8936be9 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx/vp8dx.h +++ b/chromium/third_party/libvpx/source/libvpx/vpx/vp8dx.h @@ -177,28 +177,30 @@ VPX_CTRL_USE_TYPE(VP8D_GET_FRAME_CORRUPTED, int *) #define VPX_CTRL_VP8D_GET_FRAME_CORRUPTED VPX_CTRL_USE_TYPE(VP8D_GET_LAST_REF_USED, int *) #define VPX_CTRL_VP8D_GET_LAST_REF_USED -VPX_CTRL_USE_TYPE(VPXD_GET_LAST_QUANTIZER, int *) -#define VPX_CTRL_VPXD_GET_LAST_QUANTIZER VPX_CTRL_USE_TYPE(VPXD_SET_DECRYPTOR, vpx_decrypt_init *) #define VPX_CTRL_VPXD_SET_DECRYPTOR VPX_CTRL_USE_TYPE(VP8D_SET_DECRYPTOR, vpx_decrypt_init *) #define VPX_CTRL_VP8D_SET_DECRYPTOR +VPX_CTRL_USE_TYPE(VP9D_GET_FRAME_SIZE, int *) +#define VPX_CTRL_VP9D_GET_FRAME_SIZE VPX_CTRL_USE_TYPE(VP9D_GET_DISPLAY_SIZE, int *) #define VPX_CTRL_VP9D_GET_DISPLAY_SIZE VPX_CTRL_USE_TYPE(VP9D_GET_BIT_DEPTH, unsigned int *) #define VPX_CTRL_VP9D_GET_BIT_DEPTH -VPX_CTRL_USE_TYPE(VP9D_GET_FRAME_SIZE, int *) -#define VPX_CTRL_VP9D_GET_FRAME_SIZE +VPX_CTRL_USE_TYPE(VP9_SET_BYTE_ALIGNMENT, int) +#define VPX_CTRL_VP9_SET_BYTE_ALIGNMENT VPX_CTRL_USE_TYPE(VP9_INVERT_TILE_DECODE_ORDER, int) #define VPX_CTRL_VP9_INVERT_TILE_DECODE_ORDER -#define VPX_CTRL_VP9_DECODE_SVC_SPATIAL_LAYER -VPX_CTRL_USE_TYPE(VP9_DECODE_SVC_SPATIAL_LAYER, int) -#define VPX_CTRL_VP9_SET_SKIP_LOOP_FILTER VPX_CTRL_USE_TYPE(VP9_SET_SKIP_LOOP_FILTER, int) -#define VPX_CTRL_VP9_DECODE_SET_ROW_MT +#define VPX_CTRL_VP9_SET_SKIP_LOOP_FILTER +VPX_CTRL_USE_TYPE(VP9_DECODE_SVC_SPATIAL_LAYER, int) +#define VPX_CTRL_VP9_DECODE_SVC_SPATIAL_LAYER +VPX_CTRL_USE_TYPE(VPXD_GET_LAST_QUANTIZER, int *) +#define VPX_CTRL_VPXD_GET_LAST_QUANTIZER VPX_CTRL_USE_TYPE(VP9D_SET_ROW_MT, int) -#define VPX_CTRL_VP9_SET_LOOP_FILTER_OPT +#define VPX_CTRL_VP9_DECODE_SET_ROW_MT VPX_CTRL_USE_TYPE(VP9D_SET_LOOP_FILTER_OPT, int) +#define VPX_CTRL_VP9_SET_LOOP_FILTER_OPT /*!\endcond */ /*! @} - end defgroup vp8_decoder */ diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/loongarch/intrapred_lsx.c b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/loongarch/intrapred_lsx.c new file mode 100644 index 00000000000..f9902117919 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/loongarch/intrapred_lsx.c @@ -0,0 +1,98 @@ +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Lu Wang <wanglu@loongson.cn> + * + * 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 "./vpx_dsp_rtcd.h" +#include "vpx_util/loongson_intrinsics.h" + +static inline void intra_predict_dc_8x8_lsx(const uint8_t *src_top, + const uint8_t *src_left, + uint8_t *dst, int32_t dst_stride) { + uint64_t val0, val1; + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i store, sum_h, sum_w, sum_d; + __m128i src = { 0 }; + + val0 = *(const uint64_t *)src_top; + val1 = *(const uint64_t *)src_left; + DUP2_ARG3(__lsx_vinsgr2vr_d, src, val0, 0, src, val1, 1, src, src); + sum_h = __lsx_vhaddw_hu_bu(src, src); + sum_w = __lsx_vhaddw_wu_hu(sum_h, sum_h); + sum_d = __lsx_vhaddw_du_wu(sum_w, sum_w); + sum_w = __lsx_vpickev_w(sum_d, sum_d); + sum_d = __lsx_vhaddw_du_wu(sum_w, sum_w); + sum_w = __lsx_vsrari_w(sum_d, 4); + store = __lsx_vreplvei_b(sum_w, 0); + + __lsx_vstelm_d(store, dst, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride_x3, 0, 0); + dst += dst_stride_x4; + __lsx_vstelm_d(store, dst, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride_x2, 0, 0); + __lsx_vstelm_d(store, dst + dst_stride_x3, 0, 0); +} + +static inline void intra_predict_dc_16x16_lsx(const uint8_t *src_top, + const uint8_t *src_left, + uint8_t *dst, + int32_t dst_stride) { + int32_t dst_stride_x2 = dst_stride << 1; + int32_t dst_stride_x3 = dst_stride_x2 + dst_stride; + int32_t dst_stride_x4 = dst_stride << 2; + __m128i top, left, out; + __m128i sum_h, sum_top, sum_left; + __m128i sum_w; + __m128i sum_d; + + DUP2_ARG2(__lsx_vld, src_top, 0, src_left, 0, top, left); + DUP2_ARG2(__lsx_vhaddw_hu_bu, top, top, left, left, sum_top, sum_left); + sum_h = __lsx_vadd_h(sum_top, sum_left); + sum_w = __lsx_vhaddw_wu_hu(sum_h, sum_h); + sum_d = __lsx_vhaddw_du_wu(sum_w, sum_w); + sum_w = __lsx_vpickev_w(sum_d, sum_d); + sum_d = __lsx_vhaddw_du_wu(sum_w, sum_w); + sum_w = __lsx_vsrari_w(sum_d, 5); + out = __lsx_vreplvei_b(sum_w, 0); + + __lsx_vstx(out, dst, 0); + __lsx_vstx(out, dst, dst_stride); + __lsx_vstx(out, dst, dst_stride_x2); + __lsx_vstx(out, dst, dst_stride_x3); + dst += dst_stride_x4; + __lsx_vstx(out, dst, 0); + __lsx_vstx(out, dst, dst_stride); + __lsx_vstx(out, dst, dst_stride_x2); + __lsx_vstx(out, dst, dst_stride_x3); + dst += dst_stride_x4; + __lsx_vstx(out, dst, 0); + __lsx_vstx(out, dst, dst_stride); + __lsx_vstx(out, dst, dst_stride_x2); + __lsx_vstx(out, dst, dst_stride_x3); + dst += dst_stride_x4; + __lsx_vstx(out, dst, 0); + __lsx_vstx(out, dst, dst_stride); + __lsx_vstx(out, dst, dst_stride_x2); + __lsx_vstx(out, dst, dst_stride_x3); +} + +void vpx_dc_predictor_8x8_lsx(uint8_t *dst, ptrdiff_t y_stride, + const uint8_t *above, const uint8_t *left) { + intra_predict_dc_8x8_lsx(above, left, dst, y_stride); +} + +void vpx_dc_predictor_16x16_lsx(uint8_t *dst, ptrdiff_t y_stride, + const uint8_t *above, const uint8_t *left) { + intra_predict_dc_16x16_lsx(above, left, dst, y_stride); +} diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp.mk b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp.mk index 01653102ef3..c948e12a39b 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp.mk +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp.mk @@ -74,6 +74,7 @@ endif # CONFIG_POSTPROC DSP_SRCS-$(HAVE_NEON_ASM) += arm/intrapred_neon_asm$(ASM) DSP_SRCS-$(HAVE_NEON) += arm/intrapred_neon.c DSP_SRCS-$(HAVE_MSA) += mips/intrapred_msa.c +DSP_SRCS-$(HAVE_LSX) += loongarch/intrapred_lsx.c DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred4_dspr2.c DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred8_dspr2.c DSP_SRCS-$(HAVE_DSPR2) += mips/intrapred16_dspr2.c diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp_rtcd_defs.pl b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp_rtcd_defs.pl index fd7eefdad05..c721e190b03 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp_rtcd_defs.pl +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/vpx_dsp_rtcd_defs.pl @@ -117,7 +117,7 @@ specialize qw/vpx_tm_predictor_8x8 neon dspr2 msa sse2/; add_proto qw/void vpx_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left"; # TODO(crbug.com/webm/1522): Re-enable vsx implementation. -specialize qw/vpx_dc_predictor_8x8 dspr2 neon msa sse2/; +specialize qw/vpx_dc_predictor_8x8 dspr2 neon msa sse2 lsx/; add_proto qw/void vpx_dc_top_predictor_8x8/, "uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left"; specialize qw/vpx_dc_top_predictor_8x8 neon msa sse2/; @@ -155,7 +155,7 @@ add_proto qw/void vpx_tm_predictor_16x16/, "uint8_t *dst, ptrdiff_t stride, cons specialize qw/vpx_tm_predictor_16x16 neon msa sse2 vsx/; add_proto qw/void vpx_dc_predictor_16x16/, "uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left"; -specialize qw/vpx_dc_predictor_16x16 dspr2 neon msa sse2 vsx/; +specialize qw/vpx_dc_predictor_16x16 dspr2 neon msa sse2 vsx lsx/; add_proto qw/void vpx_dc_top_predictor_16x16/, "uint8_t *dst, ptrdiff_t stride, const uint8_t *above, const uint8_t *left"; specialize qw/vpx_dc_top_predictor_16x16 neon msa sse2 vsx/; @@ -824,7 +824,7 @@ if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") { specialize qw/vpx_satd avx2 sse2 neon msa/; } - add_proto qw/void vpx_int_pro_row/, "int16_t *hbuf, const uint8_t *ref, const int ref_stride, const int height"; + add_proto qw/void vpx_int_pro_row/, "int16_t hbuf[16], const uint8_t *ref, const int ref_stride, const int height"; specialize qw/vpx_int_pro_row sse2 neon msa/; add_proto qw/int16_t vpx_int_pro_col/, "const uint8_t *ref, const int width"; diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/avg_intrin_sse2.c b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/avg_intrin_sse2.c index 3cba258f619..9da2f34c9bd 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/avg_intrin_sse2.c +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/avg_intrin_sse2.c @@ -464,7 +464,7 @@ int vpx_satd_sse2(const tran_low_t *coeff, int length) { return _mm_cvtsi128_si32(accum); } -void vpx_int_pro_row_sse2(int16_t *hbuf, const uint8_t *ref, +void vpx_int_pro_row_sse2(int16_t hbuf[16], const uint8_t *ref, const int ref_stride, const int height) { int idx; __m128i zero = _mm_setzero_si128(); diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx2.c b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx2.c index a5c4f8c5371..9dd06669188 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx2.c +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx2.c @@ -22,8 +22,8 @@ static INLINE void calc_final_4(const __m256i *const sums /*[4]*/, } void vpx_sad32x32x4d_avx2(const uint8_t *src_ptr, int src_stride, - const uint8_t *const ref_array[4], int ref_stride, - uint32_t sad_array[4]) { + const uint8_t *const ref_array[/*4*/], int ref_stride, + uint32_t *sad_array /*[4]*/) { int i; const uint8_t *refs[4]; __m256i sums[4]; @@ -127,8 +127,8 @@ void vpx_sad32x32x8_avx2(const uint8_t *src_ptr, int src_stride, } void vpx_sad64x64x4d_avx2(const uint8_t *src_ptr, int src_stride, - const uint8_t *const ref_array[4], int ref_stride, - uint32_t sad_array[4]) { + const uint8_t *const ref_array[/*4*/], int ref_stride, + uint32_t *sad_array /*[4]*/) { __m256i sums[4]; int i; const uint8_t *refs[4]; diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx512.c b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx512.c index 4c5d70464de..2fa9108718e 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx512.c +++ b/chromium/third_party/libvpx/source/libvpx/vpx_dsp/x86/sad4d_avx512.c @@ -12,8 +12,8 @@ #include "vpx/vpx_integer.h" void vpx_sad64x64x4d_avx512(const uint8_t *src_ptr, int src_stride, - const uint8_t *const ref_array[4], int ref_stride, - uint32_t res[4]) { + const uint8_t *const ref_array[/*4*/], + int ref_stride, uint32_t *res /*[4]*/) { __m512i src_reg, ref0_reg, ref1_reg, ref2_reg, ref3_reg; __m512i sum_ref0, sum_ref1, sum_ref2, sum_ref3; __m512i sum_mlow, sum_mhigh; diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch.h b/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch.h new file mode 100644 index 00000000000..d93ff9f5f0f --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch.h @@ -0,0 +1,29 @@ +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Jin Bo <jinbo@loongson.cn> + * Contributed by Lu Wang <wanglu@loongson.cn> + * + * 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 VPX_VPX_PORTS_LOONGARCH_H_ +#define VPX_VPX_PORTS_LOONGARCH_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#define HAS_LSX 0x01 +#define HAS_LASX 0x02 + +int loongarch_cpu_caps(void); + +#ifdef __cplusplus +} // extern "C" +#endif + +#endif // VPX_VPX_PORTS_LOONGARCH_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch_cpudetect.c b/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch_cpudetect.c new file mode 100644 index 00000000000..7b4322d35e1 --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vpx_ports/loongarch_cpudetect.c @@ -0,0 +1,40 @@ +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * Contributed by Jin Bo <jinbo@loongson.cn> + * Contributed by Lu Wang <wanglu@loongson.cn> + * + * 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 "./vpx_config.h" +#include "vpx_ports/loongarch.h" + +#define LOONGARCH_CFG2 0x02 +#define LOONGARCH_CFG2_LSX (1 << 6) +#define LOONGARCH_CFG2_LASX (1 << 7) + +#if CONFIG_RUNTIME_CPU_DETECT +#if defined(__loongarch__) && defined(__linux__) +int loongarch_cpu_caps(void) { + int reg = 0; + int flag = 0; + + __asm__ volatile("cpucfg %0, %1 \n\t" : "+&r"(reg) : "r"(LOONGARCH_CFG2)); + if (reg & LOONGARCH_CFG2_LSX) flag |= HAS_LSX; + + if (reg & LOONGARCH_CFG2_LASX) flag |= HAS_LASX; + + return flag; +} +#else /* end __loongarch__ && __linux__ */ +#error \ + "--enable-runtime-cpu-detect selected, but no CPU detection method " \ +"available for your platform. Reconfigure with --disable-runtime-cpu-detect." +#endif +#else /* end CONFIG_RUNTIME_CPU_DETECT */ +int loongarch_cpu_caps(void) { return 0; } +#endif diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_ports/mips.h b/chromium/third_party/libvpx/source/libvpx/vpx_ports/mips.h index bdc7525f7be..439de754fd4 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_ports/mips.h +++ b/chromium/third_party/libvpx/source/libvpx/vpx_ports/mips.h @@ -8,8 +8,8 @@ * be found in the AUTHORS file in the root of the source tree. */ -#ifndef VPX_PORTS_MIPS_H_ -#define VPX_PORTS_MIPS_H_ +#ifndef VPX_VPX_PORTS_MIPS_H_ +#define VPX_VPX_PORTS_MIPS_H_ #ifdef __cplusplus extern "C" { @@ -24,4 +24,4 @@ int mips_cpu_caps(void); } // extern "C" #endif -#endif // VPX_PORTS_MIPS_H_ +#endif // VPX_VPX_PORTS_MIPS_H_ diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_once.h b/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_once.h index 4eb592b87e5..d8a8ed89fef 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_once.h +++ b/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_once.h @@ -95,7 +95,7 @@ static void once(void (*func)(void)) { #define INCL_DOS #include <os2.h> static void once(void (*func)(void)) { - static int done; + static volatile int done; /* If the initialization is complete, return early. */ if (done) return; @@ -128,7 +128,7 @@ static void once(void (*func)(void)) { */ static void once(void (*func)(void)) { - static int done; + static volatile int done; if (!done) { func(); diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_ports.mk b/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_ports.mk index e5001be4967..e30e87cefb0 100644 --- a/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_ports.mk +++ b/chromium/third_party/libvpx/source/libvpx/vpx_ports/vpx_ports.mk @@ -45,6 +45,9 @@ PORTS_SRCS-$(VPX_ARCH_PPC) += ppc.h PORTS_SRCS-$(VPX_ARCH_MIPS) += mips_cpudetect.c PORTS_SRCS-$(VPX_ARCH_MIPS) += mips.h +PORTS_SRCS-$(VPX_ARCH_LOONGARCH) += loongarch_cpudetect.c +PORTS_SRCS-$(VPX_ARCH_LOONGARCH) += loongarch.h + ifeq ($(VPX_ARCH_MIPS), yes) PORTS_SRCS-yes += asmdefs_mmi.h endif diff --git a/chromium/third_party/libvpx/source/libvpx/vpx_util/loongson_intrinsics.h b/chromium/third_party/libvpx/source/libvpx/vpx_util/loongson_intrinsics.h new file mode 100644 index 00000000000..a34b6e8b44b --- /dev/null +++ b/chromium/third_party/libvpx/source/libvpx/vpx_util/loongson_intrinsics.h @@ -0,0 +1,1869 @@ +/* + * Copyright (c) 2022 The WebM 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 VPX_VPX_UTIL_LOONGSON_INTRINSICS_H_ +#define VPX_VPX_UTIL_LOONGSON_INTRINSICS_H_ + +/* + * Copyright (c) 2021 Loongson Technology Corporation Limited + * 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. + * + * Contributed by Shiyou Yin <yinshiyou-hf@loongson.cn> + * Xiwei Gu <guxiwei-hf@loongson.cn> + * Lu Wang <wanglu@loongson.cn> + * + * This file is a header file for loongarch builtin extension. + * + */ + +#ifndef LOONGSON_INTRINSICS_H +#define LOONGSON_INTRINSICS_H + +/** + * MAJOR version: Macro usage changes. + * MINOR version: Add new functions, or bug fixes. + * MICRO version: Comment changes or implementation changes. + */ +#define LSOM_VERSION_MAJOR 1 +#define LSOM_VERSION_MINOR 0 +#define LSOM_VERSION_MICRO 5 + +#define DUP2_ARG1(_INS, _IN0, _IN1, _OUT0, _OUT1) \ + { \ + _OUT0 = _INS(_IN0); \ + _OUT1 = _INS(_IN1); \ + } + +#define DUP2_ARG2(_INS, _IN0, _IN1, _IN2, _IN3, _OUT0, _OUT1) \ + { \ + _OUT0 = _INS(_IN0, _IN1); \ + _OUT1 = _INS(_IN2, _IN3); \ + } + +#define DUP2_ARG3(_INS, _IN0, _IN1, _IN2, _IN3, _IN4, _IN5, _OUT0, _OUT1) \ + { \ + _OUT0 = _INS(_IN0, _IN1, _IN2); \ + _OUT1 = _INS(_IN3, _IN4, _IN5); \ + } + +#define DUP4_ARG1(_INS, _IN0, _IN1, _IN2, _IN3, _OUT0, _OUT1, _OUT2, _OUT3) \ + { \ + DUP2_ARG1(_INS, _IN0, _IN1, _OUT0, _OUT1); \ + DUP2_ARG1(_INS, _IN2, _IN3, _OUT2, _OUT3); \ + } + +#define DUP4_ARG2(_INS, _IN0, _IN1, _IN2, _IN3, _IN4, _IN5, _IN6, _IN7, _OUT0, \ + _OUT1, _OUT2, _OUT3) \ + { \ + DUP2_ARG2(_INS, _IN0, _IN1, _IN2, _IN3, _OUT0, _OUT1); \ + DUP2_ARG2(_INS, _IN4, _IN5, _IN6, _IN7, _OUT2, _OUT3); \ + } + +#define DUP4_ARG3(_INS, _IN0, _IN1, _IN2, _IN3, _IN4, _IN5, _IN6, _IN7, _IN8, \ + _IN9, _IN10, _IN11, _OUT0, _OUT1, _OUT2, _OUT3) \ + { \ + DUP2_ARG3(_INS, _IN0, _IN1, _IN2, _IN3, _IN4, _IN5, _OUT0, _OUT1); \ + DUP2_ARG3(_INS, _IN6, _IN7, _IN8, _IN9, _IN10, _IN11, _OUT2, _OUT3); \ + } + +#ifdef __loongarch_sx +#include <lsxintrin.h> +/* + * ============================================================================= + * Description : Dot product & addition of byte vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Signed byte elements from in_h are multiplied by + * signed byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Then the results plus to signed half-word elements from in_c. + * Example : out = __lsx_vdp2add_h_b(in_c, in_h, in_l) + * in_c : 1,2,3,4, 1,2,3,4 + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1 + * out : 23,40,41,26, 23,40,41,26 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2add_h_b(__m128i in_c, __m128i in_h, + __m128i in_l) { + __m128i out; + + out = __lsx_vmaddwev_h_b(in_c, in_h, in_l); + out = __lsx_vmaddwod_h_b(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product & addition of byte vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Unsigned byte elements from in_h are multiplied by + * unsigned byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * The results plus to signed half-word elements from in_c. + * Example : out = __lsx_vdp2add_h_b(in_c, in_h, in_l) + * in_c : 1,2,3,4, 1,2,3,4 + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1 + * out : 23,40,41,26, 23,40,41,26 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2add_h_bu(__m128i in_c, __m128i in_h, + __m128i in_l) { + __m128i out; + + out = __lsx_vmaddwev_h_bu(in_c, in_h, in_l); + out = __lsx_vmaddwod_h_bu(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product & addition of half-word vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Outputs - out + * Return Type - __m128i + * Details : Signed half-word elements from in_h are multiplied by + * signed half-word elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Then the results plus to signed word elements from in_c. + * Example : out = __lsx_vdp2add_h_b(in_c, in_h, in_l) + * in_c : 1,2,3,4 + * in_h : 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1 + * out : 23,40,41,26 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2add_w_h(__m128i in_c, __m128i in_h, + __m128i in_l) { + __m128i out; + + out = __lsx_vmaddwev_w_h(in_c, in_h, in_l); + out = __lsx_vmaddwod_w_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Signed byte elements from in_h are multiplied by + * signed byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Example : out = __lsx_vdp2_h_b(in_h, in_l) + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1 + * out : 22,38,38,22, 22,38,38,22 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2_h_b(__m128i in_h, __m128i in_l) { + __m128i out; + + out = __lsx_vmulwev_h_b(in_h, in_l); + out = __lsx_vmaddwod_h_b(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Unsigned byte elements from in_h are multiplied by + * unsigned byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Example : out = __lsx_vdp2_h_bu(in_h, in_l) + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1 + * out : 22,38,38,22, 22,38,38,22 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2_h_bu(__m128i in_h, __m128i in_l) { + __m128i out; + + out = __lsx_vmulwev_h_bu(in_h, in_l); + out = __lsx_vmaddwod_h_bu(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Unsigned byte elements from in_h are multiplied by + * signed byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Example : out = __lsx_vdp2_h_bu_b(in_h, in_l) + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,-1 + * out : 22,38,38,22, 22,38,38,6 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2_h_bu_b(__m128i in_h, __m128i in_l) { + __m128i out; + + out = __lsx_vmulwev_h_bu_b(in_h, in_l); + out = __lsx_vmaddwod_h_bu_b(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Outputs - out + * Return Type - halfword + * Details : Signed byte elements from in_h are multiplied by + * signed byte elements from in_l, and then added adjacent to + * each other to get results with the twice size of input. + * Example : out = __lsx_vdp2_w_h(in_h, in_l) + * in_h : 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1 + * out : 22,38,38,22 + * ============================================================================= + */ +static inline __m128i __lsx_vdp2_w_h(__m128i in_h, __m128i in_l) { + __m128i out; + + out = __lsx_vmulwev_w_h(in_h, in_l); + out = __lsx_vmaddwod_w_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Clip all halfword elements of input vector between min & max + * out = ((_in) < (min)) ? (min) : (((_in) > (max)) ? (max) : + * (_in)) + * Arguments : Inputs - _in (input vector) + * - min (min threshold) + * - max (max threshold) + * Outputs - out (output vector with clipped elements) + * Return Type - signed halfword + * Example : out = __lsx_vclip_h(_in) + * _in : -8,2,280,249, -8,255,280,249 + * min : 1,1,1,1, 1,1,1,1 + * max : 9,9,9,9, 9,9,9,9 + * out : 1,2,9,9, 1,9,9,9 + * ============================================================================= + */ +static inline __m128i __lsx_vclip_h(__m128i _in, __m128i min, __m128i max) { + __m128i out; + + out = __lsx_vmax_h(min, _in); + out = __lsx_vmin_h(max, out); + return out; +} + +/* + * ============================================================================= + * Description : Set each element of vector between 0 and 255 + * Arguments : Inputs - _in + * Outputs - out + * Return Type - halfword + * Details : Signed byte elements from _in are clamped between 0 and 255. + * Example : out = __lsx_vclip255_h(_in) + * _in : -8,255,280,249, -8,255,280,249 + * out : 0,255,255,249, 0,255,255,249 + * ============================================================================= + */ +static inline __m128i __lsx_vclip255_h(__m128i _in) { + __m128i out; + + out = __lsx_vmaxi_h(_in, 0); + out = __lsx_vsat_hu(out, 7); + return out; +} + +/* + * ============================================================================= + * Description : Set each element of vector between 0 and 255 + * Arguments : Inputs - _in + * Outputs - out + * Return Type - word + * Details : Signed byte elements from _in are clamped between 0 and 255. + * Example : out = __lsx_vclip255_w(_in) + * _in : -8,255,280,249 + * out : 0,255,255,249 + * ============================================================================= + */ +static inline __m128i __lsx_vclip255_w(__m128i _in) { + __m128i out; + + out = __lsx_vmaxi_w(_in, 0); + out = __lsx_vsat_wu(out, 7); + return out; +} + +/* + * ============================================================================= + * Description : Swap two variables + * Arguments : Inputs - _in0, _in1 + * Outputs - _in0, _in1 (in-place) + * Details : Swapping of two input variables using xor + * Example : LSX_SWAP(_in0, _in1) + * _in0 : 1,2,3,4 + * _in1 : 5,6,7,8 + * _in0(out) : 5,6,7,8 + * _in1(out) : 1,2,3,4 + * ============================================================================= + */ +#define LSX_SWAP(_in0, _in1) \ + { \ + _in0 = __lsx_vxor_v(_in0, _in1); \ + _in1 = __lsx_vxor_v(_in0, _in1); \ + _in0 = __lsx_vxor_v(_in0, _in1); \ + } + +/* + * ============================================================================= + * Description : Transpose 4x4 block with word elements in vectors + * Arguments : Inputs - in0, in1, in2, in3 + * Outputs - out0, out1, out2, out3 + * Details : + * Example : + * 1, 2, 3, 4 1, 5, 9,13 + * 5, 6, 7, 8 to 2, 6,10,14 + * 9,10,11,12 =====> 3, 7,11,15 + * 13,14,15,16 4, 8,12,16 + * ============================================================================= + */ +#define LSX_TRANSPOSE4x4_W(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + __m128i _t0, _t1, _t2, _t3; \ + \ + _t0 = __lsx_vilvl_w(_in1, _in0); \ + _t1 = __lsx_vilvh_w(_in1, _in0); \ + _t2 = __lsx_vilvl_w(_in3, _in2); \ + _t3 = __lsx_vilvh_w(_in3, _in2); \ + _out0 = __lsx_vilvl_d(_t2, _t0); \ + _out1 = __lsx_vilvh_d(_t2, _t0); \ + _out2 = __lsx_vilvl_d(_t3, _t1); \ + _out3 = __lsx_vilvh_d(_t3, _t1); \ + } + +/* + * ============================================================================= + * Description : Transpose 8x8 block with byte elements in vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7 + * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6, + * _out7 + * Details : The rows of the matrix become columns, and the columns + * become rows. + * Example : LSX_TRANSPOSE8x8_B + * _in0 : 00,01,02,03,04,05,06,07, 00,00,00,00,00,00,00,00 + * _in1 : 10,11,12,13,14,15,16,17, 00,00,00,00,00,00,00,00 + * _in2 : 20,21,22,23,24,25,26,27, 00,00,00,00,00,00,00,00 + * _in3 : 30,31,32,33,34,35,36,37, 00,00,00,00,00,00,00,00 + * _in4 : 40,41,42,43,44,45,46,47, 00,00,00,00,00,00,00,00 + * _in5 : 50,51,52,53,54,55,56,57, 00,00,00,00,00,00,00,00 + * _in6 : 60,61,62,63,64,65,66,67, 00,00,00,00,00,00,00,00 + * _in7 : 70,71,72,73,74,75,76,77, 00,00,00,00,00,00,00,00 + * + * _ out0 : 00,10,20,30,40,50,60,70, 00,00,00,00,00,00,00,00 + * _ out1 : 01,11,21,31,41,51,61,71, 00,00,00,00,00,00,00,00 + * _ out2 : 02,12,22,32,42,52,62,72, 00,00,00,00,00,00,00,00 + * _ out3 : 03,13,23,33,43,53,63,73, 00,00,00,00,00,00,00,00 + * _ out4 : 04,14,24,34,44,54,64,74, 00,00,00,00,00,00,00,00 + * _ out5 : 05,15,25,35,45,55,65,75, 00,00,00,00,00,00,00,00 + * _ out6 : 06,16,26,36,46,56,66,76, 00,00,00,00,00,00,00,00 + * _ out7 : 07,17,27,37,47,57,67,77, 00,00,00,00,00,00,00,00 + * ============================================================================= + */ +#define LSX_TRANSPOSE8x8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + __m128i zero = { 0 }; \ + __m128i shuf8 = { 0x0F0E0D0C0B0A0908, 0x1716151413121110 }; \ + __m128i _t0, _t1, _t2, _t3, _t4, _t5, _t6, _t7; \ + \ + _t0 = __lsx_vilvl_b(_in2, _in0); \ + _t1 = __lsx_vilvl_b(_in3, _in1); \ + _t2 = __lsx_vilvl_b(_in6, _in4); \ + _t3 = __lsx_vilvl_b(_in7, _in5); \ + _t4 = __lsx_vilvl_b(_t1, _t0); \ + _t5 = __lsx_vilvh_b(_t1, _t0); \ + _t6 = __lsx_vilvl_b(_t3, _t2); \ + _t7 = __lsx_vilvh_b(_t3, _t2); \ + _out0 = __lsx_vilvl_w(_t6, _t4); \ + _out2 = __lsx_vilvh_w(_t6, _t4); \ + _out4 = __lsx_vilvl_w(_t7, _t5); \ + _out6 = __lsx_vilvh_w(_t7, _t5); \ + _out1 = __lsx_vshuf_b(zero, _out0, shuf8); \ + _out3 = __lsx_vshuf_b(zero, _out2, shuf8); \ + _out5 = __lsx_vshuf_b(zero, _out4, shuf8); \ + _out7 = __lsx_vshuf_b(zero, _out6, shuf8); \ + } + +/* + * ============================================================================= + * Description : Transpose 8x8 block with half-word elements in vectors + * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7 + * Outputs - out0, out1, out2, out3, out4, out5, out6, out7 + * Details : + * Example : + * 00,01,02,03,04,05,06,07 00,10,20,30,40,50,60,70 + * 10,11,12,13,14,15,16,17 01,11,21,31,41,51,61,71 + * 20,21,22,23,24,25,26,27 02,12,22,32,42,52,62,72 + * 30,31,32,33,34,35,36,37 to 03,13,23,33,43,53,63,73 + * 40,41,42,43,44,45,46,47 ======> 04,14,24,34,44,54,64,74 + * 50,51,52,53,54,55,56,57 05,15,25,35,45,55,65,75 + * 60,61,62,63,64,65,66,67 06,16,26,36,46,56,66,76 + * 70,71,72,73,74,75,76,77 07,17,27,37,47,57,67,77 + * ============================================================================= + */ +#define LSX_TRANSPOSE8x8_H(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + __m128i _s0, _s1, _t0, _t1, _t2, _t3, _t4, _t5, _t6, _t7; \ + \ + _s0 = __lsx_vilvl_h(_in6, _in4); \ + _s1 = __lsx_vilvl_h(_in7, _in5); \ + _t0 = __lsx_vilvl_h(_s1, _s0); \ + _t1 = __lsx_vilvh_h(_s1, _s0); \ + _s0 = __lsx_vilvh_h(_in6, _in4); \ + _s1 = __lsx_vilvh_h(_in7, _in5); \ + _t2 = __lsx_vilvl_h(_s1, _s0); \ + _t3 = __lsx_vilvh_h(_s1, _s0); \ + _s0 = __lsx_vilvl_h(_in2, _in0); \ + _s1 = __lsx_vilvl_h(_in3, _in1); \ + _t4 = __lsx_vilvl_h(_s1, _s0); \ + _t5 = __lsx_vilvh_h(_s1, _s0); \ + _s0 = __lsx_vilvh_h(_in2, _in0); \ + _s1 = __lsx_vilvh_h(_in3, _in1); \ + _t6 = __lsx_vilvl_h(_s1, _s0); \ + _t7 = __lsx_vilvh_h(_s1, _s0); \ + \ + _out0 = __lsx_vpickev_d(_t0, _t4); \ + _out2 = __lsx_vpickev_d(_t1, _t5); \ + _out4 = __lsx_vpickev_d(_t2, _t6); \ + _out6 = __lsx_vpickev_d(_t3, _t7); \ + _out1 = __lsx_vpickod_d(_t0, _t4); \ + _out3 = __lsx_vpickod_d(_t1, _t5); \ + _out5 = __lsx_vpickod_d(_t2, _t6); \ + _out7 = __lsx_vpickod_d(_t3, _t7); \ + } + +/* + * ============================================================================= + * Description : Transpose input 8x4 byte block into 4x8 + * Arguments : Inputs - _in0, _in1, _in2, _in3 (input 8x4 byte block) + * Outputs - _out0, _out1, _out2, _out3 (output 4x8 byte block) + * Return Type - as per RTYPE + * Details : The rows of the matrix become columns, and the columns become + * rows. + * Example : LSX_TRANSPOSE8x4_B + * _in0 : 00,01,02,03,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in1 : 10,11,12,13,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in2 : 20,21,22,23,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in3 : 30,31,32,33,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in4 : 40,41,42,43,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in5 : 50,51,52,53,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in6 : 60,61,62,63,00,00,00,00, 00,00,00,00,00,00,00,00 + * _in7 : 70,71,72,73,00,00,00,00, 00,00,00,00,00,00,00,00 + * + * _out0 : 00,10,20,30,40,50,60,70, 00,00,00,00,00,00,00,00 + * _out1 : 01,11,21,31,41,51,61,71, 00,00,00,00,00,00,00,00 + * _out2 : 02,12,22,32,42,52,62,72, 00,00,00,00,00,00,00,00 + * _out3 : 03,13,23,33,43,53,63,73, 00,00,00,00,00,00,00,00 + * ============================================================================= + */ +#define LSX_TRANSPOSE8x4_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3) \ + { \ + __m128i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + \ + _tmp0_m = __lsx_vpackev_w(_in4, _in0); \ + _tmp1_m = __lsx_vpackev_w(_in5, _in1); \ + _tmp2_m = __lsx_vilvl_b(_tmp1_m, _tmp0_m); \ + _tmp0_m = __lsx_vpackev_w(_in6, _in2); \ + _tmp1_m = __lsx_vpackev_w(_in7, _in3); \ + \ + _tmp3_m = __lsx_vilvl_b(_tmp1_m, _tmp0_m); \ + _tmp0_m = __lsx_vilvl_h(_tmp3_m, _tmp2_m); \ + _tmp1_m = __lsx_vilvh_h(_tmp3_m, _tmp2_m); \ + \ + _out0 = __lsx_vilvl_w(_tmp1_m, _tmp0_m); \ + _out2 = __lsx_vilvh_w(_tmp1_m, _tmp0_m); \ + _out1 = __lsx_vilvh_d(_out2, _out0); \ + _out3 = __lsx_vilvh_d(_out0, _out2); \ + } + +/* + * ============================================================================= + * Description : Transpose 16x8 block with byte elements in vectors + * Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, in8 + * in9, in10, in11, in12, in13, in14, in15 + * Outputs - out0, out1, out2, out3, out4, out5, out6, out7 + * Details : + * Example : + * 000,001,002,003,004,005,006,007 + * 008,009,010,011,012,013,014,015 + * 016,017,018,019,020,021,022,023 + * 024,025,026,027,028,029,030,031 + * 032,033,034,035,036,037,038,039 + * 040,041,042,043,044,045,046,047 000,008,...,112,120 + * 048,049,050,051,052,053,054,055 001,009,...,113,121 + * 056,057,058,059,060,061,062,063 to 002,010,...,114,122 + * 064,068,066,067,068,069,070,071 =====> 003,011,...,115,123 + * 072,073,074,075,076,077,078,079 004,012,...,116,124 + * 080,081,082,083,084,085,086,087 005,013,...,117,125 + * 088,089,090,091,092,093,094,095 006,014,...,118,126 + * 096,097,098,099,100,101,102,103 007,015,...,119,127 + * 104,105,106,107,108,109,110,111 + * 112,113,114,115,116,117,118,119 + * 120,121,122,123,124,125,126,127 + * ============================================================================= + */ +#define LSX_TRANSPOSE16x8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _in8, _in9, _in10, _in11, _in12, _in13, _in14, \ + _in15, _out0, _out1, _out2, _out3, _out4, _out5, \ + _out6, _out7) \ + { \ + __m128i _tmp0, _tmp1, _tmp2, _tmp3, _tmp4, _tmp5, _tmp6, _tmp7; \ + __m128i _t0, _t1, _t2, _t3, _t4, _t5, _t6, _t7; \ + DUP4_ARG2(__lsx_vilvl_b, _in2, _in0, _in3, _in1, _in6, _in4, _in7, _in5, \ + _tmp0, _tmp1, _tmp2, _tmp3); \ + DUP4_ARG2(__lsx_vilvl_b, _in10, _in8, _in11, _in9, _in14, _in12, _in15, \ + _in13, _tmp4, _tmp5, _tmp6, _tmp7); \ + DUP2_ARG2(__lsx_vilvl_b, _tmp1, _tmp0, _tmp3, _tmp2, _t0, _t2); \ + DUP2_ARG2(__lsx_vilvh_b, _tmp1, _tmp0, _tmp3, _tmp2, _t1, _t3); \ + DUP2_ARG2(__lsx_vilvl_b, _tmp5, _tmp4, _tmp7, _tmp6, _t4, _t6); \ + DUP2_ARG2(__lsx_vilvh_b, _tmp5, _tmp4, _tmp7, _tmp6, _t5, _t7); \ + DUP2_ARG2(__lsx_vilvl_w, _t2, _t0, _t3, _t1, _tmp0, _tmp4); \ + DUP2_ARG2(__lsx_vilvh_w, _t2, _t0, _t3, _t1, _tmp2, _tmp6); \ + DUP2_ARG2(__lsx_vilvl_w, _t6, _t4, _t7, _t5, _tmp1, _tmp5); \ + DUP2_ARG2(__lsx_vilvh_w, _t6, _t4, _t7, _t5, _tmp3, _tmp7); \ + DUP2_ARG2(__lsx_vilvl_d, _tmp1, _tmp0, _tmp3, _tmp2, _out0, _out2); \ + DUP2_ARG2(__lsx_vilvh_d, _tmp1, _tmp0, _tmp3, _tmp2, _out1, _out3); \ + DUP2_ARG2(__lsx_vilvl_d, _tmp5, _tmp4, _tmp7, _tmp6, _out4, _out6); \ + DUP2_ARG2(__lsx_vilvh_d, _tmp5, _tmp4, _tmp7, _tmp6, _out5, _out7); \ + } + +/* + * ============================================================================= + * Description : Butterfly of 4 input vectors + * Arguments : Inputs - in0, in1, in2, in3 + * Outputs - out0, out1, out2, out3 + * Details : Butterfly operation + * Example : + * out0 = in0 + in3; + * out1 = in1 + in2; + * out2 = in1 - in2; + * out3 = in0 - in3; + * ============================================================================= + */ +#define LSX_BUTTERFLY_4_B(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lsx_vadd_b(_in0, _in3); \ + _out1 = __lsx_vadd_b(_in1, _in2); \ + _out2 = __lsx_vsub_b(_in1, _in2); \ + _out3 = __lsx_vsub_b(_in0, _in3); \ + } +#define LSX_BUTTERFLY_4_H(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lsx_vadd_h(_in0, _in3); \ + _out1 = __lsx_vadd_h(_in1, _in2); \ + _out2 = __lsx_vsub_h(_in1, _in2); \ + _out3 = __lsx_vsub_h(_in0, _in3); \ + } +#define LSX_BUTTERFLY_4_W(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lsx_vadd_w(_in0, _in3); \ + _out1 = __lsx_vadd_w(_in1, _in2); \ + _out2 = __lsx_vsub_w(_in1, _in2); \ + _out3 = __lsx_vsub_w(_in0, _in3); \ + } +#define LSX_BUTTERFLY_4_D(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lsx_vadd_d(_in0, _in3); \ + _out1 = __lsx_vadd_d(_in1, _in2); \ + _out2 = __lsx_vsub_d(_in1, _in2); \ + _out3 = __lsx_vsub_d(_in0, _in3); \ + } + +/* + * ============================================================================= + * Description : Butterfly of 8 input vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3, ~ + * Outputs - _out0, _out1, _out2, _out3, ~ + * Details : Butterfly operation + * Example : + * _out0 = _in0 + _in7; + * _out1 = _in1 + _in6; + * _out2 = _in2 + _in5; + * _out3 = _in3 + _in4; + * _out4 = _in3 - _in4; + * _out5 = _in2 - _in5; + * _out6 = _in1 - _in6; + * _out7 = _in0 - _in7; + * ============================================================================= + */ +#define LSX_BUTTERFLY_8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lsx_vadd_b(_in0, _in7); \ + _out1 = __lsx_vadd_b(_in1, _in6); \ + _out2 = __lsx_vadd_b(_in2, _in5); \ + _out3 = __lsx_vadd_b(_in3, _in4); \ + _out4 = __lsx_vsub_b(_in3, _in4); \ + _out5 = __lsx_vsub_b(_in2, _in5); \ + _out6 = __lsx_vsub_b(_in1, _in6); \ + _out7 = __lsx_vsub_b(_in0, _in7); \ + } + +#define LSX_BUTTERFLY_8_H(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lsx_vadd_h(_in0, _in7); \ + _out1 = __lsx_vadd_h(_in1, _in6); \ + _out2 = __lsx_vadd_h(_in2, _in5); \ + _out3 = __lsx_vadd_h(_in3, _in4); \ + _out4 = __lsx_vsub_h(_in3, _in4); \ + _out5 = __lsx_vsub_h(_in2, _in5); \ + _out6 = __lsx_vsub_h(_in1, _in6); \ + _out7 = __lsx_vsub_h(_in0, _in7); \ + } + +#define LSX_BUTTERFLY_8_W(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lsx_vadd_w(_in0, _in7); \ + _out1 = __lsx_vadd_w(_in1, _in6); \ + _out2 = __lsx_vadd_w(_in2, _in5); \ + _out3 = __lsx_vadd_w(_in3, _in4); \ + _out4 = __lsx_vsub_w(_in3, _in4); \ + _out5 = __lsx_vsub_w(_in2, _in5); \ + _out6 = __lsx_vsub_w(_in1, _in6); \ + _out7 = __lsx_vsub_w(_in0, _in7); \ + } + +#define LSX_BUTTERFLY_8_D(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lsx_vadd_d(_in0, _in7); \ + _out1 = __lsx_vadd_d(_in1, _in6); \ + _out2 = __lsx_vadd_d(_in2, _in5); \ + _out3 = __lsx_vadd_d(_in3, _in4); \ + _out4 = __lsx_vsub_d(_in3, _in4); \ + _out5 = __lsx_vsub_d(_in2, _in5); \ + _out6 = __lsx_vsub_d(_in1, _in6); \ + _out7 = __lsx_vsub_d(_in0, _in7); \ + } + +#endif // LSX + +#ifdef __loongarch_asx +#include <lasxintrin.h> +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed halfword + * Details : Unsigned byte elements from in_h are multiplied with + * unsigned byte elements from in_l producing a result + * twice the size of input i.e. signed halfword. + * Then this multiplied results of adjacent odd-even elements + * are added to the out vector + * Example : See out = __lasx_xvdp2_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2_h_bu(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_h_bu(in_h, in_l); + out = __lasx_xvmaddwod_h_bu(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of byte vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed halfword + * Details : Signed byte elements from in_h are multiplied with + * signed byte elements from in_l producing a result + * twice the size of input i.e. signed halfword. + * Then this multiplication results of adjacent odd-even elements + * are added to the out vector + * Example : See out = __lasx_xvdp2_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2_h_b(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_h_b(in_h, in_l); + out = __lasx_xvmaddwod_h_b(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed word + * Details : Signed halfword elements from in_h are multiplied with + * signed halfword elements from in_l producing a result + * twice the size of input i.e. signed word. + * Then this multiplied results of adjacent odd-even elements + * are added to the out vector. + * Example : out = __lasx_xvdp2_w_h(in_h, in_l) + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1 + * out : 22,38,38,22, 22,38,38,22 + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2_w_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_w_h(in_h, in_l); + out = __lasx_xvmaddwod_w_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of word vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed double + * Details : Signed word elements from in_h are multiplied with + * signed word elements from in_l producing a result + * twice the size of input i.e. signed double-word. + * Then this multiplied results of adjacent odd-even elements + * are added to the out vector. + * Example : See out = __lasx_xvdp2_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2_d_w(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_d_w(in_h, in_l); + out = __lasx_xvmaddwod_d_w(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed word + * Details : Unsigned halfword elements from in_h are multiplied with + * signed halfword elements from in_l producing a result + * twice the size of input i.e. unsigned word. + * Multiplication result of adjacent odd-even elements + * are added to the out vector + * Example : See out = __lasx_xvdp2_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2_w_hu_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_w_hu_h(in_h, in_l); + out = __lasx_xvmaddwod_w_hu_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product & addition of byte vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - halfword + * Details : Signed byte elements from in_h are multiplied with + * signed byte elements from in_l producing a result + * twice the size of input i.e. signed halfword. + * Then this multiplied results of adjacent odd-even elements + * are added to the in_c vector. + * Example : See out = __lasx_xvdp2add_w_h(in_c, in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2add_h_b(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmaddwev_h_b(in_c, in_h, in_l); + out = __lasx_xvmaddwod_h_b(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Return Type - per RTYPE + * Details : Signed halfword elements from in_h are multiplied with + * signed halfword elements from in_l producing a result + * twice the size of input i.e. signed word. + * Multiplication result of adjacent odd-even elements + * are added to the in_c vector. + * Example : out = __lasx_xvdp2add_w_h(in_c, in_h, in_l) + * in_c : 1,2,3,4, 1,2,3,4 + * in_h : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8, + * in_l : 8,7,6,5, 4,3,2,1, 8,7,6,5, 4,3,2,1, + * out : 23,40,41,26, 23,40,41,26 + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2add_w_h(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmaddwev_w_h(in_c, in_h, in_l); + out = __lasx_xvmaddwod_w_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Return Type - signed word + * Details : Unsigned halfword elements from in_h are multiplied with + * unsigned halfword elements from in_l producing a result + * twice the size of input i.e. signed word. + * Multiplication result of adjacent odd-even elements + * are added to the in_c vector. + * Example : See out = __lasx_xvdp2add_w_h(in_c, in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2add_w_hu(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmaddwev_w_hu(in_c, in_h, in_l); + out = __lasx_xvmaddwod_w_hu(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Return Type - signed word + * Details : Unsigned halfword elements from in_h are multiplied with + * signed halfword elements from in_l producing a result + * twice the size of input i.e. signed word. + * Multiplication result of adjacent odd-even elements + * are added to the in_c vector + * Example : See out = __lasx_xvdp2add_w_h(in_c, in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2add_w_hu_h(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmaddwev_w_hu_h(in_c, in_h, in_l); + out = __lasx_xvmaddwod_w_hu_h(out, in_h, in_l); + return out; +} + +/* + * ============================================================================= + * Description : Vector Unsigned Dot Product and Subtract + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Return Type - signed halfword + * Details : Unsigned byte elements from in_h are multiplied with + * unsigned byte elements from in_l producing a result + * twice the size of input i.e. signed halfword. + * Multiplication result of adjacent odd-even elements + * are added together and subtracted from double width elements + * in_c vector. + * Example : See out = __lasx_xvdp2sub_w_h(in_c, in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2sub_h_bu(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_h_bu(in_h, in_l); + out = __lasx_xvmaddwod_h_bu(out, in_h, in_l); + out = __lasx_xvsub_h(in_c, out); + return out; +} + +/* + * ============================================================================= + * Description : Vector Signed Dot Product and Subtract + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Return Type - signed word + * Details : Signed halfword elements from in_h are multiplied with + * Signed halfword elements from in_l producing a result + * twice the size of input i.e. signed word. + * Multiplication result of adjacent odd-even elements + * are added together and subtracted from double width elements + * in_c vector. + * Example : out = __lasx_xvdp2sub_w_h(in_c, in_h, in_l) + * in_c : 0,0,0,0, 0,0,0,0 + * in_h : 3,1,3,0, 0,0,0,1, 0,0,1,1, 0,0,0,1 + * in_l : 2,1,1,0, 1,0,0,0, 0,0,1,0, 1,0,0,1 + * out : -7,-3,0,0, 0,-1,0,-1 + * ============================================================================= + */ +static inline __m256i __lasx_xvdp2sub_w_h(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_w_h(in_h, in_l); + out = __lasx_xvmaddwod_w_h(out, in_h, in_l); + out = __lasx_xvsub_w(in_c, out); + return out; +} + +/* + * ============================================================================= + * Description : Dot product of halfword vector elements + * Arguments : Inputs - in_h, in_l + * Output - out + * Return Type - signed word + * Details : Signed halfword elements from in_h are multiplied with + * signed halfword elements from in_l producing a result + * four times the size of input i.e. signed doubleword. + * Then this multiplication results of four adjacent elements + * are added together and stored to the out vector. + * Example : out = __lasx_xvdp4_d_h(in_h, in_l) + * in_h : 3,1,3,0, 0,0,0,1, 0,0,1,-1, 0,0,0,1 + * in_l : -2,1,1,0, 1,0,0,0, 0,0,1, 0, 1,0,0,1 + * out : -2,0,1,1 + * ============================================================================= + */ +static inline __m256i __lasx_xvdp4_d_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvmulwev_w_h(in_h, in_l); + out = __lasx_xvmaddwod_w_h(out, in_h, in_l); + out = __lasx_xvhaddw_d_w(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The high half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are added after the + * higher half of the two-fold sign extension (signed byte + * to signed halfword) and stored to the out vector. + * Example : See out = __lasx_xvaddwh_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvaddwh_h_b(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvilvh_b(in_h, in_l); + out = __lasx_xvhaddw_h_b(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The high half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are added after the + * higher half of the two-fold sign extension (signed halfword + * to signed word) and stored to the out vector. + * Example : out = __lasx_xvaddwh_w_h(in_h, in_l) + * in_h : 3, 0,3,0, 0,0,0,-1, 0,0,1,-1, 0,0,0,1 + * in_l : 2,-1,1,2, 1,0,0, 0, 1,0,1, 0, 1,0,0,1 + * out : 1,0,0,-1, 1,0,0, 2 + * ============================================================================= + */ +static inline __m256i __lasx_xvaddwh_w_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvilvh_h(in_h, in_l); + out = __lasx_xvhaddw_w_h(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are added after the + * lower half of the two-fold sign extension (signed byte + * to signed halfword) and stored to the out vector. + * Example : See out = __lasx_xvaddwl_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvaddwl_h_b(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvilvl_b(in_h, in_l); + out = __lasx_xvhaddw_h_b(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are added after the + * lower half of the two-fold sign extension (signed halfword + * to signed word) and stored to the out vector. + * Example : out = __lasx_xvaddwl_w_h(in_h, in_l) + * in_h : 3, 0,3,0, 0,0,0,-1, 0,0,1,-1, 0,0,0,1 + * in_l : 2,-1,1,2, 1,0,0, 0, 1,0,1, 0, 1,0,0,1 + * out : 5,-1,4,2, 1,0,2,-1 + * ============================================================================= + */ +static inline __m256i __lasx_xvaddwl_w_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvilvl_h(in_h, in_l); + out = __lasx_xvhaddw_w_h(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The out vector and the out vector are added after the + * lower half of the two-fold zero extension (unsigned byte + * to unsigned halfword) and stored to the out vector. + * Example : See out = __lasx_xvaddwl_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvaddwl_h_bu(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvilvl_b(in_h, in_l); + out = __lasx_xvhaddw_hu_bu(out, out); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_l vector after double zero extension (unsigned byte to + * signed halfword)，added to the in_h vector. + * Example : See out = __lasx_xvaddw_w_w_h(in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvaddw_h_h_bu(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvsllwil_hu_bu(in_l, 0); + out = __lasx_xvadd_h(in_h, out); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are expanded and + * added after being doubled. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_l vector after double sign extension (signed halfword to + * signed word), added to the in_h vector. + * Example : out = __lasx_xvaddw_w_w_h(in_h, in_l) + * in_h : 0, 1,0,0, -1,0,0,1, + * in_l : 2,-1,1,2, 1,0,0,0, 0,0,1,0, 1,0,0,1, + * out : 2, 0,1,2, -1,0,1,1, + * ============================================================================= + */ +static inline __m256i __lasx_xvaddw_w_w_h(__m256i in_h, __m256i in_l) { + __m256i out; + + out = __lasx_xvsllwil_w_h(in_l, 0); + out = __lasx_xvadd_w(in_h, out); + return out; +} + +/* + * ============================================================================= + * Description : Multiplication and addition calculation after expansion + * of the lower half of the vector. + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are multiplied after + * the lower half of the two-fold sign extension (signed halfword + * to signed word), and the result is added to the vector in_c, + * then stored to the out vector. + * Example : out = __lasx_xvmaddwl_w_h(in_c, in_h, in_l) + * in_c : 1,2,3,4, 5,6,7,8 + * in_h : 1,2,3,4, 1,2,3,4, 5,6,7,8, 5,6,7,8 + * in_l : 200, 300, 400, 500, 2000, 3000, 4000, 5000, + * -200,-300,-400,-500, -2000,-3000,-4000,-5000 + * out : 201, 602,1203,2004, -995, -1794,-2793,-3992 + * ============================================================================= + */ +static inline __m256i __lasx_xvmaddwl_w_h(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i tmp0, tmp1, out; + + tmp0 = __lasx_xvsllwil_w_h(in_h, 0); + tmp1 = __lasx_xvsllwil_w_h(in_l, 0); + tmp0 = __lasx_xvmul_w(tmp0, tmp1); + out = __lasx_xvadd_w(tmp0, in_c); + return out; +} + +/* + * ============================================================================= + * Description : Multiplication and addition calculation after expansion + * of the higher half of the vector. + * Arguments : Inputs - in_c, in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are multiplied after + * the higher half of the two-fold sign extension (signed + * halfword to signed word), and the result is added to + * the vector in_c, then stored to the out vector. + * Example : See out = __lasx_xvmaddwl_w_h(in_c, in_h, in_l) + * ============================================================================= + */ +static inline __m256i __lasx_xvmaddwh_w_h(__m256i in_c, __m256i in_h, + __m256i in_l) { + __m256i tmp0, tmp1, out; + + tmp0 = __lasx_xvilvh_h(in_h, in_h); + tmp1 = __lasx_xvilvh_h(in_l, in_l); + tmp0 = __lasx_xvmulwev_w_h(tmp0, tmp1); + out = __lasx_xvadd_w(tmp0, in_c); + return out; +} + +/* + * ============================================================================= + * Description : Multiplication calculation after expansion of the lower + * half of the vector. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are multiplied after + * the lower half of the two-fold sign extension (signed + * halfword to signed word), then stored to the out vector. + * Example : out = __lasx_xvmulwl_w_h(in_h, in_l) + * in_h : 3,-1,3,0, 0,0,0,-1, 0,0,1,-1, 0,0,0,1 + * in_l : 2,-1,1,2, 1,0,0, 0, 0,0,1, 0, 1,0,0,1 + * out : 6,1,3,0, 0,0,1,0 + * ============================================================================= + */ +static inline __m256i __lasx_xvmulwl_w_h(__m256i in_h, __m256i in_l) { + __m256i tmp0, tmp1, out; + + tmp0 = __lasx_xvsllwil_w_h(in_h, 0); + tmp1 = __lasx_xvsllwil_w_h(in_l, 0); + out = __lasx_xvmul_w(tmp0, tmp1); + return out; +} + +/* + * ============================================================================= + * Description : Multiplication calculation after expansion of the lower + * half of the vector. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector and the in_l vector are multiplied after + * the lower half of the two-fold sign extension (signed + * halfword to signed word), then stored to the out vector. + * Example : out = __lasx_xvmulwh_w_h(in_h, in_l) + * in_h : 3,-1,3,0, 0,0,0,-1, 0,0,1,-1, 0,0,0,1 + * in_l : 2,-1,1,2, 1,0,0, 0, 0,0,1, 0, 1,0,0,1 + * out : 0,0,0,0, 0,0,0,1 + * ============================================================================= + */ +static inline __m256i __lasx_xvmulwh_w_h(__m256i in_h, __m256i in_l) { + __m256i tmp0, tmp1, out; + + tmp0 = __lasx_xvilvh_h(in_h, in_h); + tmp1 = __lasx_xvilvh_h(in_l, in_l); + out = __lasx_xvmulwev_w_h(tmp0, tmp1); + return out; +} + +/* + * ============================================================================= + * Description : The low half of the vector elements are added to the high half + * after being doubled, then saturated. + * Arguments : Inputs - in_h, in_l + * Output - out + * Details : The in_h vector adds the in_l vector after the lower half of + * the two-fold zero extension (unsigned byte to unsigned + * halfword) and then saturated. The results are stored to the out + * vector. + * Example : out = __lasx_xvsaddw_hu_hu_bu(in_h, in_l) + * in_h : 2,65532,1,2, 1,0,0,0, 0,0,1,0, 1,0,0,1 + * in_l : 3,6,3,0, 0,0,0,1, 0,0,1,1, 0,0,0,1, 3,18,3,0, 0,0,0,1, 0,0,1,1, + * 0,0,0,1 + * out : 5,65535,4,2, 1,0,0,1, 3,18,4,0, 1,0,0,2, + * ============================================================================= + */ +static inline __m256i __lasx_xvsaddw_hu_hu_bu(__m256i in_h, __m256i in_l) { + __m256i tmp1, out; + __m256i zero = { 0 }; + + tmp1 = __lasx_xvilvl_b(zero, in_l); + out = __lasx_xvsadd_hu(in_h, tmp1); + return out; +} + +/* + * ============================================================================= + * Description : Clip all halfword elements of input vector between min & max + * out = ((in) < (min)) ? (min) : (((in) > (max)) ? (max) : (in)) + * Arguments : Inputs - in (input vector) + * - min (min threshold) + * - max (max threshold) + * Outputs - in (output vector with clipped elements) + * Return Type - signed halfword + * Example : out = __lasx_xvclip_h(in, min, max) + * in : -8,2,280,249, -8,255,280,249, 4,4,4,4, 5,5,5,5 + * min : 1,1,1,1, 1,1,1,1, 1,1,1,1, 1,1,1,1 + * max : 9,9,9,9, 9,9,9,9, 9,9,9,9, 9,9,9,9 + * out : 1,2,9,9, 1,9,9,9, 4,4,4,4, 5,5,5,5 + * ============================================================================= + */ +static inline __m256i __lasx_xvclip_h(__m256i in, __m256i min, __m256i max) { + __m256i out; + + out = __lasx_xvmax_h(min, in); + out = __lasx_xvmin_h(max, out); + return out; +} + +/* + * ============================================================================= + * Description : Clip all signed halfword elements of input vector + * between 0 & 255 + * Arguments : Inputs - in (input vector) + * Outputs - out (output vector with clipped elements) + * Return Type - signed halfword + * Example : See out = __lasx_xvclip255_w(in) + * ============================================================================= + */ +static inline __m256i __lasx_xvclip255_h(__m256i in) { + __m256i out; + + out = __lasx_xvmaxi_h(in, 0); + out = __lasx_xvsat_hu(out, 7); + return out; +} + +/* + * ============================================================================= + * Description : Clip all signed word elements of input vector + * between 0 & 255 + * Arguments : Inputs - in (input vector) + * Output - out (output vector with clipped elements) + * Return Type - signed word + * Example : out = __lasx_xvclip255_w(in) + * in : -8,255,280,249, -8,255,280,249 + * out : 0,255,255,249, 0,255,255,249 + * ============================================================================= + */ +static inline __m256i __lasx_xvclip255_w(__m256i in) { + __m256i out; + + out = __lasx_xvmaxi_w(in, 0); + out = __lasx_xvsat_wu(out, 7); + return out; +} + +/* + * ============================================================================= + * Description : Indexed halfword element values are replicated to all + * elements in output vector. If 'idx < 8' use xvsplati_l_*, + * if 'idx >= 8' use xvsplati_h_*. + * Arguments : Inputs - in, idx + * Output - out + * Details : Idx element value from in vector is replicated to all + * elements in out vector. + * Valid index range for halfword operation is 0-7 + * Example : out = __lasx_xvsplati_l_h(in, idx) + * in : 20,10,11,12, 13,14,15,16, 0,0,2,0, 0,0,0,0 + * idx : 0x02 + * out : 11,11,11,11, 11,11,11,11, 11,11,11,11, 11,11,11,11 + * ============================================================================= + */ +static inline __m256i __lasx_xvsplati_l_h(__m256i in, int idx) { + __m256i out; + + out = __lasx_xvpermi_q(in, in, 0x02); + out = __lasx_xvreplve_h(out, idx); + return out; +} + +/* + * ============================================================================= + * Description : Indexed halfword element values are replicated to all + * elements in output vector. If 'idx < 8' use xvsplati_l_*, + * if 'idx >= 8' use xvsplati_h_*. + * Arguments : Inputs - in, idx + * Output - out + * Details : Idx element value from in vector is replicated to all + * elements in out vector. + * Valid index range for halfword operation is 0-7 + * Example : out = __lasx_xvsplati_h_h(in, idx) + * in : 20,10,11,12, 13,14,15,16, 0,2,0,0, 0,0,0,0 + * idx : 0x09 + * out : 2,2,2,2, 2,2,2,2, 2,2,2,2, 2,2,2,2 + * ============================================================================= + */ +static inline __m256i __lasx_xvsplati_h_h(__m256i in, int idx) { + __m256i out; + + out = __lasx_xvpermi_q(in, in, 0x13); + out = __lasx_xvreplve_h(out, idx); + return out; +} + +/* + * ============================================================================= + * Description : Transpose 4x4 block with double-word elements in vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3 + * Outputs - _out0, _out1, _out2, _out3 + * Example : LASX_TRANSPOSE4x4_D + * _in0 : 1,2,3,4 + * _in1 : 1,2,3,4 + * _in2 : 1,2,3,4 + * _in3 : 1,2,3,4 + * + * _out0 : 1,1,1,1 + * _out1 : 2,2,2,2 + * _out2 : 3,3,3,3 + * _out3 : 4,4,4,4 + * ============================================================================= + */ +#define LASX_TRANSPOSE4x4_D(_in0, _in1, _in2, _in3, _out0, _out1, _out2, \ + _out3) \ + { \ + __m256i _tmp0, _tmp1, _tmp2, _tmp3; \ + _tmp0 = __lasx_xvilvl_d(_in1, _in0); \ + _tmp1 = __lasx_xvilvh_d(_in1, _in0); \ + _tmp2 = __lasx_xvilvl_d(_in3, _in2); \ + _tmp3 = __lasx_xvilvh_d(_in3, _in2); \ + _out0 = __lasx_xvpermi_q(_tmp2, _tmp0, 0x20); \ + _out2 = __lasx_xvpermi_q(_tmp2, _tmp0, 0x31); \ + _out1 = __lasx_xvpermi_q(_tmp3, _tmp1, 0x20); \ + _out3 = __lasx_xvpermi_q(_tmp3, _tmp1, 0x31); \ + } + +/* + * ============================================================================= + * Description : Transpose 8x8 block with word elements in vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7 + * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6, + * _out7 + * Example : LASX_TRANSPOSE8x8_W + * _in0 : 1,2,3,4,5,6,7,8 + * _in1 : 2,2,3,4,5,6,7,8 + * _in2 : 3,2,3,4,5,6,7,8 + * _in3 : 4,2,3,4,5,6,7,8 + * _in4 : 5,2,3,4,5,6,7,8 + * _in5 : 6,2,3,4,5,6,7,8 + * _in6 : 7,2,3,4,5,6,7,8 + * _in7 : 8,2,3,4,5,6,7,8 + * + * _out0 : 1,2,3,4,5,6,7,8 + * _out1 : 2,2,2,2,2,2,2,2 + * _out2 : 3,3,3,3,3,3,3,3 + * _out3 : 4,4,4,4,4,4,4,4 + * _out4 : 5,5,5,5,5,5,5,5 + * _out5 : 6,6,6,6,6,6,6,6 + * _out6 : 7,7,7,7,7,7,7,7 + * _out7 : 8,8,8,8,8,8,8,8 + * ============================================================================= + */ +#define LASX_TRANSPOSE8x8_W(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + __m256i _s0_m, _s1_m; \ + __m256i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + __m256i _tmp4_m, _tmp5_m, _tmp6_m, _tmp7_m; \ + \ + _s0_m = __lasx_xvilvl_w(_in2, _in0); \ + _s1_m = __lasx_xvilvl_w(_in3, _in1); \ + _tmp0_m = __lasx_xvilvl_w(_s1_m, _s0_m); \ + _tmp1_m = __lasx_xvilvh_w(_s1_m, _s0_m); \ + _s0_m = __lasx_xvilvh_w(_in2, _in0); \ + _s1_m = __lasx_xvilvh_w(_in3, _in1); \ + _tmp2_m = __lasx_xvilvl_w(_s1_m, _s0_m); \ + _tmp3_m = __lasx_xvilvh_w(_s1_m, _s0_m); \ + _s0_m = __lasx_xvilvl_w(_in6, _in4); \ + _s1_m = __lasx_xvilvl_w(_in7, _in5); \ + _tmp4_m = __lasx_xvilvl_w(_s1_m, _s0_m); \ + _tmp5_m = __lasx_xvilvh_w(_s1_m, _s0_m); \ + _s0_m = __lasx_xvilvh_w(_in6, _in4); \ + _s1_m = __lasx_xvilvh_w(_in7, _in5); \ + _tmp6_m = __lasx_xvilvl_w(_s1_m, _s0_m); \ + _tmp7_m = __lasx_xvilvh_w(_s1_m, _s0_m); \ + _out0 = __lasx_xvpermi_q(_tmp4_m, _tmp0_m, 0x20); \ + _out1 = __lasx_xvpermi_q(_tmp5_m, _tmp1_m, 0x20); \ + _out2 = __lasx_xvpermi_q(_tmp6_m, _tmp2_m, 0x20); \ + _out3 = __lasx_xvpermi_q(_tmp7_m, _tmp3_m, 0x20); \ + _out4 = __lasx_xvpermi_q(_tmp4_m, _tmp0_m, 0x31); \ + _out5 = __lasx_xvpermi_q(_tmp5_m, _tmp1_m, 0x31); \ + _out6 = __lasx_xvpermi_q(_tmp6_m, _tmp2_m, 0x31); \ + _out7 = __lasx_xvpermi_q(_tmp7_m, _tmp3_m, 0x31); \ + } + +/* + * ============================================================================= + * Description : Transpose input 16x8 byte block + * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, + * _in8, _in9, _in10, _in11, _in12, _in13, _in14, _in15 + * (input 16x8 byte block) + * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6, + * _out7 (output 8x16 byte block) + * Details : The rows of the matrix become columns, and the columns become + * rows. + * Example : See LASX_TRANSPOSE16x8_H + * ============================================================================= + */ +#define LASX_TRANSPOSE16x8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _in8, _in9, _in10, _in11, _in12, _in13, _in14, \ + _in15, _out0, _out1, _out2, _out3, _out4, _out5, \ + _out6, _out7) \ + { \ + __m256i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + __m256i _tmp4_m, _tmp5_m, _tmp6_m, _tmp7_m; \ + \ + _tmp0_m = __lasx_xvilvl_b(_in2, _in0); \ + _tmp1_m = __lasx_xvilvl_b(_in3, _in1); \ + _tmp2_m = __lasx_xvilvl_b(_in6, _in4); \ + _tmp3_m = __lasx_xvilvl_b(_in7, _in5); \ + _tmp4_m = __lasx_xvilvl_b(_in10, _in8); \ + _tmp5_m = __lasx_xvilvl_b(_in11, _in9); \ + _tmp6_m = __lasx_xvilvl_b(_in14, _in12); \ + _tmp7_m = __lasx_xvilvl_b(_in15, _in13); \ + _out0 = __lasx_xvilvl_b(_tmp1_m, _tmp0_m); \ + _out1 = __lasx_xvilvh_b(_tmp1_m, _tmp0_m); \ + _out2 = __lasx_xvilvl_b(_tmp3_m, _tmp2_m); \ + _out3 = __lasx_xvilvh_b(_tmp3_m, _tmp2_m); \ + _out4 = __lasx_xvilvl_b(_tmp5_m, _tmp4_m); \ + _out5 = __lasx_xvilvh_b(_tmp5_m, _tmp4_m); \ + _out6 = __lasx_xvilvl_b(_tmp7_m, _tmp6_m); \ + _out7 = __lasx_xvilvh_b(_tmp7_m, _tmp6_m); \ + _tmp0_m = __lasx_xvilvl_w(_out2, _out0); \ + _tmp2_m = __lasx_xvilvh_w(_out2, _out0); \ + _tmp4_m = __lasx_xvilvl_w(_out3, _out1); \ + _tmp6_m = __lasx_xvilvh_w(_out3, _out1); \ + _tmp1_m = __lasx_xvilvl_w(_out6, _out4); \ + _tmp3_m = __lasx_xvilvh_w(_out6, _out4); \ + _tmp5_m = __lasx_xvilvl_w(_out7, _out5); \ + _tmp7_m = __lasx_xvilvh_w(_out7, _out5); \ + _out0 = __lasx_xvilvl_d(_tmp1_m, _tmp0_m); \ + _out1 = __lasx_xvilvh_d(_tmp1_m, _tmp0_m); \ + _out2 = __lasx_xvilvl_d(_tmp3_m, _tmp2_m); \ + _out3 = __lasx_xvilvh_d(_tmp3_m, _tmp2_m); \ + _out4 = __lasx_xvilvl_d(_tmp5_m, _tmp4_m); \ + _out5 = __lasx_xvilvh_d(_tmp5_m, _tmp4_m); \ + _out6 = __lasx_xvilvl_d(_tmp7_m, _tmp6_m); \ + _out7 = __lasx_xvilvh_d(_tmp7_m, _tmp6_m); \ + } + +/* + * ============================================================================= + * Description : Transpose input 16x8 byte block + * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, + * _in8, _in9, _in10, _in11, _in12, _in13, _in14, _in15 + * (input 16x8 byte block) + * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6, + * _out7 (output 8x16 byte block) + * Details : The rows of the matrix become columns, and the columns become + * rows. + * Example : LASX_TRANSPOSE16x8_H + * _in0 : 1,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in1 : 2,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in2 : 3,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in3 : 4,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in4 : 5,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in5 : 6,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in6 : 7,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in7 : 8,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in8 : 9,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in9 : 1,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in10 : 0,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in11 : 2,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in12 : 3,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in13 : 7,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in14 : 5,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * _in15 : 6,2,3,4,5,6,7,8,0,0,0,0,0,0,0,0 + * + * _out0 : 1,2,3,4,5,6,7,8,9,1,0,2,3,7,5,6 + * _out1 : 2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2 + * _out2 : 3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3 + * _out3 : 4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4 + * _out4 : 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5 + * _out5 : 6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6 + * _out6 : 7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7 + * _out7 : 8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8 + * ============================================================================= + */ +#define LASX_TRANSPOSE16x8_H(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _in8, _in9, _in10, _in11, _in12, _in13, _in14, \ + _in15, _out0, _out1, _out2, _out3, _out4, _out5, \ + _out6, _out7) \ + { \ + __m256i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + __m256i _tmp4_m, _tmp5_m, _tmp6_m, _tmp7_m; \ + __m256i _t0, _t1, _t2, _t3, _t4, _t5, _t6, _t7; \ + \ + _tmp0_m = __lasx_xvilvl_h(_in2, _in0); \ + _tmp1_m = __lasx_xvilvl_h(_in3, _in1); \ + _tmp2_m = __lasx_xvilvl_h(_in6, _in4); \ + _tmp3_m = __lasx_xvilvl_h(_in7, _in5); \ + _tmp4_m = __lasx_xvilvl_h(_in10, _in8); \ + _tmp5_m = __lasx_xvilvl_h(_in11, _in9); \ + _tmp6_m = __lasx_xvilvl_h(_in14, _in12); \ + _tmp7_m = __lasx_xvilvl_h(_in15, _in13); \ + _t0 = __lasx_xvilvl_h(_tmp1_m, _tmp0_m); \ + _t1 = __lasx_xvilvh_h(_tmp1_m, _tmp0_m); \ + _t2 = __lasx_xvilvl_h(_tmp3_m, _tmp2_m); \ + _t3 = __lasx_xvilvh_h(_tmp3_m, _tmp2_m); \ + _t4 = __lasx_xvilvl_h(_tmp5_m, _tmp4_m); \ + _t5 = __lasx_xvilvh_h(_tmp5_m, _tmp4_m); \ + _t6 = __lasx_xvilvl_h(_tmp7_m, _tmp6_m); \ + _t7 = __lasx_xvilvh_h(_tmp7_m, _tmp6_m); \ + _tmp0_m = __lasx_xvilvl_d(_t2, _t0); \ + _tmp2_m = __lasx_xvilvh_d(_t2, _t0); \ + _tmp4_m = __lasx_xvilvl_d(_t3, _t1); \ + _tmp6_m = __lasx_xvilvh_d(_t3, _t1); \ + _tmp1_m = __lasx_xvilvl_d(_t6, _t4); \ + _tmp3_m = __lasx_xvilvh_d(_t6, _t4); \ + _tmp5_m = __lasx_xvilvl_d(_t7, _t5); \ + _tmp7_m = __lasx_xvilvh_d(_t7, _t5); \ + _out0 = __lasx_xvpermi_q(_tmp1_m, _tmp0_m, 0x20); \ + _out1 = __lasx_xvpermi_q(_tmp3_m, _tmp2_m, 0x20); \ + _out2 = __lasx_xvpermi_q(_tmp5_m, _tmp4_m, 0x20); \ + _out3 = __lasx_xvpermi_q(_tmp7_m, _tmp6_m, 0x20); \ + \ + _tmp0_m = __lasx_xvilvh_h(_in2, _in0); \ + _tmp1_m = __lasx_xvilvh_h(_in3, _in1); \ + _tmp2_m = __lasx_xvilvh_h(_in6, _in4); \ + _tmp3_m = __lasx_xvilvh_h(_in7, _in5); \ + _tmp4_m = __lasx_xvilvh_h(_in10, _in8); \ + _tmp5_m = __lasx_xvilvh_h(_in11, _in9); \ + _tmp6_m = __lasx_xvilvh_h(_in14, _in12); \ + _tmp7_m = __lasx_xvilvh_h(_in15, _in13); \ + _t0 = __lasx_xvilvl_h(_tmp1_m, _tmp0_m); \ + _t1 = __lasx_xvilvh_h(_tmp1_m, _tmp0_m); \ + _t2 = __lasx_xvilvl_h(_tmp3_m, _tmp2_m); \ + _t3 = __lasx_xvilvh_h(_tmp3_m, _tmp2_m); \ + _t4 = __lasx_xvilvl_h(_tmp5_m, _tmp4_m); \ + _t5 = __lasx_xvilvh_h(_tmp5_m, _tmp4_m); \ + _t6 = __lasx_xvilvl_h(_tmp7_m, _tmp6_m); \ + _t7 = __lasx_xvilvh_h(_tmp7_m, _tmp6_m); \ + _tmp0_m = __lasx_xvilvl_d(_t2, _t0); \ + _tmp2_m = __lasx_xvilvh_d(_t2, _t0); \ + _tmp4_m = __lasx_xvilvl_d(_t3, _t1); \ + _tmp6_m = __lasx_xvilvh_d(_t3, _t1); \ + _tmp1_m = __lasx_xvilvl_d(_t6, _t4); \ + _tmp3_m = __lasx_xvilvh_d(_t6, _t4); \ + _tmp5_m = __lasx_xvilvl_d(_t7, _t5); \ + _tmp7_m = __lasx_xvilvh_d(_t7, _t5); \ + _out4 = __lasx_xvpermi_q(_tmp1_m, _tmp0_m, 0x20); \ + _out5 = __lasx_xvpermi_q(_tmp3_m, _tmp2_m, 0x20); \ + _out6 = __lasx_xvpermi_q(_tmp5_m, _tmp4_m, 0x20); \ + _out7 = __lasx_xvpermi_q(_tmp7_m, _tmp6_m, 0x20); \ + } + +/* + * ============================================================================= + * Description : Transpose 4x4 block with halfword elements in vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3 + * Outputs - _out0, _out1, _out2, _out3 + * Return Type - signed halfword + * Details : The rows of the matrix become columns, and the columns become + * rows. + * Example : See LASX_TRANSPOSE8x8_H + * ============================================================================= + */ +#define LASX_TRANSPOSE4x4_H(_in0, _in1, _in2, _in3, _out0, _out1, _out2, \ + _out3) \ + { \ + __m256i _s0_m, _s1_m; \ + \ + _s0_m = __lasx_xvilvl_h(_in1, _in0); \ + _s1_m = __lasx_xvilvl_h(_in3, _in2); \ + _out0 = __lasx_xvilvl_w(_s1_m, _s0_m); \ + _out2 = __lasx_xvilvh_w(_s1_m, _s0_m); \ + _out1 = __lasx_xvilvh_d(_out0, _out0); \ + _out3 = __lasx_xvilvh_d(_out2, _out2); \ + } + +/* + * ============================================================================= + * Description : Transpose input 8x8 byte block + * Arguments : Inputs - _in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7 + * (input 8x8 byte block) + * Outputs - _out0, _out1, _out2, _out3, _out4, _out5, _out6, + * _out7 (output 8x8 byte block) + * Example : See LASX_TRANSPOSE8x8_H + * ============================================================================= + */ +#define LASX_TRANSPOSE8x8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + __m256i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + __m256i _tmp4_m, _tmp5_m, _tmp6_m, _tmp7_m; \ + _tmp0_m = __lasx_xvilvl_b(_in2, _in0); \ + _tmp1_m = __lasx_xvilvl_b(_in3, _in1); \ + _tmp2_m = __lasx_xvilvl_b(_in6, _in4); \ + _tmp3_m = __lasx_xvilvl_b(_in7, _in5); \ + _tmp4_m = __lasx_xvilvl_b(_tmp1_m, _tmp0_m); \ + _tmp5_m = __lasx_xvilvh_b(_tmp1_m, _tmp0_m); \ + _tmp6_m = __lasx_xvilvl_b(_tmp3_m, _tmp2_m); \ + _tmp7_m = __lasx_xvilvh_b(_tmp3_m, _tmp2_m); \ + _out0 = __lasx_xvilvl_w(_tmp6_m, _tmp4_m); \ + _out2 = __lasx_xvilvh_w(_tmp6_m, _tmp4_m); \ + _out4 = __lasx_xvilvl_w(_tmp7_m, _tmp5_m); \ + _out6 = __lasx_xvilvh_w(_tmp7_m, _tmp5_m); \ + _out1 = __lasx_xvbsrl_v(_out0, 8); \ + _out3 = __lasx_xvbsrl_v(_out2, 8); \ + _out5 = __lasx_xvbsrl_v(_out4, 8); \ + _out7 = __lasx_xvbsrl_v(_out6, 8); \ + } + +/* + * ============================================================================= + * Description : Transpose 8x8 block with halfword elements in vectors. + * Arguments : Inputs - _in0, _in1, ~ + * Outputs - _out0, _out1, ~ + * Details : The rows of the matrix become columns, and the columns become + * rows. + * Example : LASX_TRANSPOSE8x8_H + * _in0 : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * _in1 : 8,2,3,4, 5,6,7,8, 8,2,3,4, 5,6,7,8 + * _in2 : 8,2,3,4, 5,6,7,8, 8,2,3,4, 5,6,7,8 + * _in3 : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * _in4 : 9,2,3,4, 5,6,7,8, 9,2,3,4, 5,6,7,8 + * _in5 : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * _in6 : 1,2,3,4, 5,6,7,8, 1,2,3,4, 5,6,7,8 + * _in7 : 9,2,3,4, 5,6,7,8, 9,2,3,4, 5,6,7,8 + * + * _out0 : 1,8,8,1, 9,1,1,9, 1,8,8,1, 9,1,1,9 + * _out1 : 2,2,2,2, 2,2,2,2, 2,2,2,2, 2,2,2,2 + * _out2 : 3,3,3,3, 3,3,3,3, 3,3,3,3, 3,3,3,3 + * _out3 : 4,4,4,4, 4,4,4,4, 4,4,4,4, 4,4,4,4 + * _out4 : 5,5,5,5, 5,5,5,5, 5,5,5,5, 5,5,5,5 + * _out5 : 6,6,6,6, 6,6,6,6, 6,6,6,6, 6,6,6,6 + * _out6 : 7,7,7,7, 7,7,7,7, 7,7,7,7, 7,7,7,7 + * _out7 : 8,8,8,8, 8,8,8,8, 8,8,8,8, 8,8,8,8 + * ============================================================================= + */ +#define LASX_TRANSPOSE8x8_H(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + __m256i _s0_m, _s1_m; \ + __m256i _tmp0_m, _tmp1_m, _tmp2_m, _tmp3_m; \ + __m256i _tmp4_m, _tmp5_m, _tmp6_m, _tmp7_m; \ + \ + _s0_m = __lasx_xvilvl_h(_in6, _in4); \ + _s1_m = __lasx_xvilvl_h(_in7, _in5); \ + _tmp0_m = __lasx_xvilvl_h(_s1_m, _s0_m); \ + _tmp1_m = __lasx_xvilvh_h(_s1_m, _s0_m); \ + _s0_m = __lasx_xvilvh_h(_in6, _in4); \ + _s1_m = __lasx_xvilvh_h(_in7, _in5); \ + _tmp2_m = __lasx_xvilvl_h(_s1_m, _s0_m); \ + _tmp3_m = __lasx_xvilvh_h(_s1_m, _s0_m); \ + \ + _s0_m = __lasx_xvilvl_h(_in2, _in0); \ + _s1_m = __lasx_xvilvl_h(_in3, _in1); \ + _tmp4_m = __lasx_xvilvl_h(_s1_m, _s0_m); \ + _tmp5_m = __lasx_xvilvh_h(_s1_m, _s0_m); \ + _s0_m = __lasx_xvilvh_h(_in2, _in0); \ + _s1_m = __lasx_xvilvh_h(_in3, _in1); \ + _tmp6_m = __lasx_xvilvl_h(_s1_m, _s0_m); \ + _tmp7_m = __lasx_xvilvh_h(_s1_m, _s0_m); \ + \ + _out0 = __lasx_xvpickev_d(_tmp0_m, _tmp4_m); \ + _out2 = __lasx_xvpickev_d(_tmp1_m, _tmp5_m); \ + _out4 = __lasx_xvpickev_d(_tmp2_m, _tmp6_m); \ + _out6 = __lasx_xvpickev_d(_tmp3_m, _tmp7_m); \ + _out1 = __lasx_xvpickod_d(_tmp0_m, _tmp4_m); \ + _out3 = __lasx_xvpickod_d(_tmp1_m, _tmp5_m); \ + _out5 = __lasx_xvpickod_d(_tmp2_m, _tmp6_m); \ + _out7 = __lasx_xvpickod_d(_tmp3_m, _tmp7_m); \ + } + +/* + * ============================================================================= + * Description : Butterfly of 4 input vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3 + * Outputs - _out0, _out1, _out2, _out3 + * Details : Butterfly operation + * Example : LASX_BUTTERFLY_4 + * _out0 = _in0 + _in3; + * _out1 = _in1 + _in2; + * _out2 = _in1 - _in2; + * _out3 = _in0 - _in3; + * ============================================================================= + */ +#define LASX_BUTTERFLY_4_B(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lasx_xvadd_b(_in0, _in3); \ + _out1 = __lasx_xvadd_b(_in1, _in2); \ + _out2 = __lasx_xvsub_b(_in1, _in2); \ + _out3 = __lasx_xvsub_b(_in0, _in3); \ + } +#define LASX_BUTTERFLY_4_H(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lasx_xvadd_h(_in0, _in3); \ + _out1 = __lasx_xvadd_h(_in1, _in2); \ + _out2 = __lasx_xvsub_h(_in1, _in2); \ + _out3 = __lasx_xvsub_h(_in0, _in3); \ + } +#define LASX_BUTTERFLY_4_W(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lasx_xvadd_w(_in0, _in3); \ + _out1 = __lasx_xvadd_w(_in1, _in2); \ + _out2 = __lasx_xvsub_w(_in1, _in2); \ + _out3 = __lasx_xvsub_w(_in0, _in3); \ + } +#define LASX_BUTTERFLY_4_D(_in0, _in1, _in2, _in3, _out0, _out1, _out2, _out3) \ + { \ + _out0 = __lasx_xvadd_d(_in0, _in3); \ + _out1 = __lasx_xvadd_d(_in1, _in2); \ + _out2 = __lasx_xvsub_d(_in1, _in2); \ + _out3 = __lasx_xvsub_d(_in0, _in3); \ + } + +/* + * ============================================================================= + * Description : Butterfly of 8 input vectors + * Arguments : Inputs - _in0, _in1, _in2, _in3, ~ + * Outputs - _out0, _out1, _out2, _out3, ~ + * Details : Butterfly operation + * Example : LASX_BUTTERFLY_8 + * _out0 = _in0 + _in7; + * _out1 = _in1 + _in6; + * _out2 = _in2 + _in5; + * _out3 = _in3 + _in4; + * _out4 = _in3 - _in4; + * _out5 = _in2 - _in5; + * _out6 = _in1 - _in6; + * _out7 = _in0 - _in7; + * ============================================================================= + */ +#define LASX_BUTTERFLY_8_B(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lasx_xvadd_b(_in0, _in7); \ + _out1 = __lasx_xvadd_b(_in1, _in6); \ + _out2 = __lasx_xvadd_b(_in2, _in5); \ + _out3 = __lasx_xvadd_b(_in3, _in4); \ + _out4 = __lasx_xvsub_b(_in3, _in4); \ + _out5 = __lasx_xvsub_b(_in2, _in5); \ + _out6 = __lasx_xvsub_b(_in1, _in6); \ + _out7 = __lasx_xvsub_b(_in0, _in7); \ + } + +#define LASX_BUTTERFLY_8_H(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lasx_xvadd_h(_in0, _in7); \ + _out1 = __lasx_xvadd_h(_in1, _in6); \ + _out2 = __lasx_xvadd_h(_in2, _in5); \ + _out3 = __lasx_xvadd_h(_in3, _in4); \ + _out4 = __lasx_xvsub_h(_in3, _in4); \ + _out5 = __lasx_xvsub_h(_in2, _in5); \ + _out6 = __lasx_xvsub_h(_in1, _in6); \ + _out7 = __lasx_xvsub_h(_in0, _in7); \ + } + +#define LASX_BUTTERFLY_8_W(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lasx_xvadd_w(_in0, _in7); \ + _out1 = __lasx_xvadd_w(_in1, _in6); \ + _out2 = __lasx_xvadd_w(_in2, _in5); \ + _out3 = __lasx_xvadd_w(_in3, _in4); \ + _out4 = __lasx_xvsub_w(_in3, _in4); \ + _out5 = __lasx_xvsub_w(_in2, _in5); \ + _out6 = __lasx_xvsub_w(_in1, _in6); \ + _out7 = __lasx_xvsub_w(_in0, _in7); \ + } + +#define LASX_BUTTERFLY_8_D(_in0, _in1, _in2, _in3, _in4, _in5, _in6, _in7, \ + _out0, _out1, _out2, _out3, _out4, _out5, _out6, \ + _out7) \ + { \ + _out0 = __lasx_xvadd_d(_in0, _in7); \ + _out1 = __lasx_xvadd_d(_in1, _in6); \ + _out2 = __lasx_xvadd_d(_in2, _in5); \ + _out3 = __lasx_xvadd_d(_in3, _in4); \ + _out4 = __lasx_xvsub_d(_in3, _in4); \ + _out5 = __lasx_xvsub_d(_in2, _in5); \ + _out6 = __lasx_xvsub_d(_in1, _in6); \ + _out7 = __lasx_xvsub_d(_in0, _in7); \ + } + +#endif // LASX + +/* + * ============================================================================= + * Description : Print out elements in vector. + * Arguments : Inputs - RTYPE, _element_num, _in0, _enter + * Outputs - + * Details : Print out '_element_num' elements in 'RTYPE' vector '_in0', if + * '_enter' is TRUE, prefix "\nVP:" will be added first. + * Example : VECT_PRINT(v4i32,4,in0,1); // in0: 1,2,3,4 + * VP:1,2,3,4, + * ============================================================================= + */ +#define VECT_PRINT(RTYPE, element_num, in0, enter) \ + { \ + RTYPE _tmp0 = (RTYPE)in0; \ + int _i = 0; \ + if (enter) printf("\nVP:"); \ + for (_i = 0; _i < element_num; _i++) printf("%d,", _tmp0[_i]); \ + } + +#endif /* LOONGSON_INTRINSICS_H */ +#endif /* VPX_VPX_UTIL_LOONGSON_INTRINSICS_H_ */ |