diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile index 002144c56d..fa03f7ce26 100644 --- a/libavcodec/x86/Makefile +++ b/libavcodec/x86/Makefile @@ -2,6 +2,7 @@ OBJS += x86/constants.o \ x86/fmtconvert_init.o \ OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp_init.o +OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp_mmx.o OBJS-$(CONFIG_DCT) += x86/dct_init.o OBJS-$(CONFIG_DSPUTIL) += x86/dsputil_init.o OBJS-$(CONFIG_ENCODERS) += x86/dsputilenc_mmx.o \ @@ -53,7 +54,6 @@ OBJS-$(CONFIG_VP8_DECODER) += x86/vp8dsp_init.o OBJS-$(CONFIG_VP9_DECODER) += x86/vp9dsp_init.o OBJS-$(CONFIG_WEBP_DECODER) += x86/vp8dsp_init.o -MMX-OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp_mmx.o MMX-OBJS-$(CONFIG_DSPUTIL) += x86/dsputil_mmx.o \ x86/idct_mmx_xvid.o \ x86/idct_sse2_xvid.o \ @@ -69,6 +69,7 @@ YASM-OBJS += x86/deinterlace.o \ x86/fmtconvert.o \ YASM-OBJS-$(CONFIG_AC3DSP) += x86/ac3dsp.o +YASM-OBJS-$(CONFIG_BLOCKDSP) += x86/blockdsp.o YASM-OBJS-$(CONFIG_DCT) += x86/dct32.o YASM-OBJS-$(CONFIG_DIRAC_DECODER) += x86/diracdsp_mmx.o x86/diracdsp_yasm.o\ x86/dwt_yasm.o diff --git a/libavcodec/x86/blockdsp.asm b/libavcodec/x86/blockdsp.asm new file mode 100644 index 0000000000..247c750c74 --- /dev/null +++ b/libavcodec/x86/blockdsp.asm @@ -0,0 +1,84 @@ +;****************************************************************************** +;* +;* SIMD-optimized clear block functions +;* +;* This file is part of FFmpeg. +;* +;* FFmpeg is free software; you can redistribute it and/or +;* modify it under the terms of the GNU Lesser General Public +;* License as published by the Free Software Foundation; either +;* version 2.1 of the License, or (at your option) any later version. +;* +;* FFmpeg is distributed in the hope that it will be useful, +;* but WITHOUT ANY WARRANTY; without even the implied warranty of +;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU +;* Lesser General Public License for more details. +;* +;* You should have received a copy of the GNU Lesser General Public +;* License along with FFmpeg; if not, write to the Free Software +;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA +;****************************************************************************** + +%include "libavutil/x86/x86util.asm" + +SECTION_TEXT + +;---------------------------------------- +; void ff_clear_block(int16_t *blocks); +;---------------------------------------- +; %1 = number of xmm registers used +; %2 = number of inline store loops +%macro CLEAR_BLOCK 2 +cglobal clear_block, 1, 1, %1, blocks + ZERO m0, m0 +%assign %%i 0 +%rep %2 + mova [blocksq+mmsize*(0+%%i)], m0 + mova [blocksq+mmsize*(1+%%i)], m0 + mova [blocksq+mmsize*(2+%%i)], m0 + mova [blocksq+mmsize*(3+%%i)], m0 + mova [blocksq+mmsize*(4+%%i)], m0 + mova [blocksq+mmsize*(5+%%i)], m0 + mova [blocksq+mmsize*(6+%%i)], m0 + mova [blocksq+mmsize*(7+%%i)], m0 +%assign %%i %%i+8 +%endrep + RET +%endmacro + +INIT_MMX mmx +%define ZERO pxor +CLEAR_BLOCK 0, 2 +INIT_XMM sse +%define ZERO xorps +CLEAR_BLOCK 1, 1 + +;----------------------------------------- +; void ff_clear_blocks(int16_t *blocks); +;----------------------------------------- +; %1 = number of xmm registers used +%macro CLEAR_BLOCKS 1 +cglobal clear_blocks, 1, 2, %1, blocks, len + add blocksq, 768 + mov lenq, -768 + ZERO m0, m0 +.loop + mova [blocksq+lenq+mmsize*0], m0 + mova [blocksq+lenq+mmsize*1], m0 + mova [blocksq+lenq+mmsize*2], m0 + mova [blocksq+lenq+mmsize*3], m0 + mova [blocksq+lenq+mmsize*4], m0 + mova [blocksq+lenq+mmsize*5], m0 + mova [blocksq+lenq+mmsize*6], m0 + mova [blocksq+lenq+mmsize*7], m0 + add lenq, mmsize*8 + js .loop + RET +%endmacro + +INIT_MMX mmx +%define ZERO pxor +CLEAR_BLOCKS 0 +INIT_XMM sse +%define ZERO xorps +CLEAR_BLOCKS 1 diff --git a/libavcodec/x86/blockdsp_mmx.c b/libavcodec/x86/blockdsp_mmx.c index 53baad3b54..7780184af6 100644 --- a/libavcodec/x86/blockdsp_mmx.c +++ b/libavcodec/x86/blockdsp_mmx.c @@ -22,7 +22,6 @@ #include "libavutil/attributes.h" #include "libavutil/internal.h" #include "libavutil/cpu.h" -#include "libavutil/x86/asm.h" #include "libavutil/x86/cpu.h" #include "libavcodec/blockdsp.h" #include "libavcodec/version.h" @@ -43,7 +42,7 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c, unsigned high_bit_depth) int cpu_flags = av_get_cpu_flags(); if (!high_bit_depth) { - if (INLINE_MMX(cpu_flags)) { + if (EXTERNAL_MMX(cpu_flags)) { c->clear_block = ff_clear_block_mmx; c->clear_blocks = ff_clear_blocks_mmx; } @@ -52,7 +51,7 @@ av_cold void ff_blockdsp_init_x86(BlockDSPContext *c, unsigned high_bit_depth) if (CONFIG_XVMC && avctx->hwaccel && avctx->hwaccel->decode_mb) return; - if (INLINE_SSE(cpu_flags)) { + if (EXTERNAL_SSE(cpu_flags)) { c->clear_block = ff_clear_block_sse; c->clear_blocks = ff_clear_blocks_sse; } diff --git a/libavcodec/x86/dsputil.asm b/libavcodec/x86/dsputil.asm index 2209c52541..3bb5d9cbfe 100644 --- a/libavcodec/x86/dsputil.asm +++ b/libavcodec/x86/dsputil.asm @@ -248,66 +248,6 @@ BSWAP32_BUF INIT_XMM ssse3 BSWAP32_BUF -;---------------------------------------- -; void ff_clear_block(int16_t *blocks); -;---------------------------------------- -; %1 = number of xmm registers used -; %2 = number of inline store loops -%macro CLEAR_BLOCK 2 -cglobal clear_block, 1, 1, %1, blocks - ZERO m0, m0 -%assign %%i 0 -%rep %2 - mova [blocksq+mmsize*(0+%%i)], m0 - mova [blocksq+mmsize*(1+%%i)], m0 - mova [blocksq+mmsize*(2+%%i)], m0 - mova [blocksq+mmsize*(3+%%i)], m0 - mova [blocksq+mmsize*(4+%%i)], m0 - mova [blocksq+mmsize*(5+%%i)], m0 - mova [blocksq+mmsize*(6+%%i)], m0 - mova [blocksq+mmsize*(7+%%i)], m0 -%assign %%i %%i+8 -%endrep - RET -%endmacro - -INIT_MMX mmx -%define ZERO pxor -CLEAR_BLOCK 0, 2 -INIT_XMM sse -%define ZERO xorps -CLEAR_BLOCK 1, 1 - -;----------------------------------------- -; void ff_clear_blocks(int16_t *blocks); -;----------------------------------------- -; %1 = number of xmm registers used -%macro CLEAR_BLOCKS 1 -cglobal clear_blocks, 1, 2, %1, blocks, len - add blocksq, 768 - mov lenq, -768 - ZERO m0, m0 -.loop - mova [blocksq+lenq+mmsize*0], m0 - mova [blocksq+lenq+mmsize*1], m0 - mova [blocksq+lenq+mmsize*2], m0 - mova [blocksq+lenq+mmsize*3], m0 - mova [blocksq+lenq+mmsize*4], m0 - mova [blocksq+lenq+mmsize*5], m0 - mova [blocksq+lenq+mmsize*6], m0 - mova [blocksq+lenq+mmsize*7], m0 - add lenq, mmsize*8 - js .loop - RET -%endmacro - -INIT_MMX mmx -%define ZERO pxor -CLEAR_BLOCKS 0 -INIT_XMM sse -%define ZERO xorps -CLEAR_BLOCKS 1 - ;-------------------------------------------------------------------------- ;void ff_put_signed_pixels_clamped(const int16_t *block, uint8_t *pixels, ; int line_size)