From c2d337429c7c87ee559efe54dbc0f84f2a25c3a4 Mon Sep 17 00:00:00 2001 From: "Ronald S. Bultje" Date: Fri, 21 Oct 2011 00:00:39 -0700 Subject: H264: change weight/biweight functions to take a height argument. Neon parts by Mans Rullgard . --- libavcodec/x86/h264_weight.asm | 210 ++++++++++++++++------------------------- 1 file changed, 79 insertions(+), 131 deletions(-) (limited to 'libavcodec/x86/h264_weight.asm') diff --git a/libavcodec/x86/h264_weight.asm b/libavcodec/x86/h264_weight.asm index d80ca32583..bc8bfd686e 100644 --- a/libavcodec/x86/h264_weight.asm +++ b/libavcodec/x86/h264_weight.asm @@ -28,21 +28,20 @@ SECTION .text ;----------------------------------------------------------------------------- ; biweight pred: ; -; void h264_biweight_16x16_sse2(uint8_t *dst, uint8_t *src, int stride, -; int log2_denom, int weightd, int weights, -; int offset); +; void h264_biweight_16_sse2(uint8_t *dst, uint8_t *src, int stride, +; int height, int log2_denom, int weightd, +; int weights, int offset); ; and -; void h264_weight_16x16_sse2(uint8_t *dst, int stride, -; int log2_denom, int weight, -; int offset); +; void h264_weight_16_sse2(uint8_t *dst, int stride, int height, +; int log2_denom, int weight, int offset); ;----------------------------------------------------------------------------- %macro WEIGHT_SETUP 0 - add r4, r4 - inc r4 - movd m3, r3d - movd m5, r4d - movd m6, r2d + add r5, r5 + inc r5 + movd m3, r4d + movd m5, r5d + movd m6, r3d pslld m5, m6 psrld m5, 1 %if mmsize == 16 @@ -71,60 +70,41 @@ SECTION .text packuswb m0, m1 %endmacro -%macro WEIGHT_FUNC_DBL_MM 1 -cglobal h264_weight_16x%1_mmx2, 5, 5, 0 +INIT_MMX +cglobal h264_weight_16_mmx2, 6, 6, 0 WEIGHT_SETUP - mov r2, %1 -%if %1 == 16 .nextrow WEIGHT_OP 0, 4 mova [r0 ], m0 WEIGHT_OP 8, 12 mova [r0+8], m0 add r0, r1 - dec r2 + dec r2d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_weight_16x16_mmx2.nextrow) -%endif -%endmacro -INIT_MMX -WEIGHT_FUNC_DBL_MM 16 -WEIGHT_FUNC_DBL_MM 8 - -%macro WEIGHT_FUNC_MM 4 -cglobal h264_weight_%1x%2_%4, 7, 7, %3 +%macro WEIGHT_FUNC_MM 3 +cglobal h264_weight_%1_%3, 6, 6, %2 WEIGHT_SETUP - mov r2, %2 -%if %2 == 16 .nextrow WEIGHT_OP 0, mmsize/2 mova [r0], m0 add r0, r1 - dec r2 + dec r2d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_weight_%1x16_%4.nextrow) -%endif %endmacro INIT_MMX -WEIGHT_FUNC_MM 8, 16, 0, mmx2 -WEIGHT_FUNC_MM 8, 8, 0, mmx2 -WEIGHT_FUNC_MM 8, 4, 0, mmx2 +WEIGHT_FUNC_MM 8, 0, mmx2 INIT_XMM -WEIGHT_FUNC_MM 16, 16, 8, sse2 -WEIGHT_FUNC_MM 16, 8, 8, sse2 +WEIGHT_FUNC_MM 16, 8, sse2 -%macro WEIGHT_FUNC_HALF_MM 5 -cglobal h264_weight_%1x%2_%5, 5, 5, %4 +%macro WEIGHT_FUNC_HALF_MM 3 +cglobal h264_weight_%1_%3, 6, 6, %2 WEIGHT_SETUP - mov r2, %2/2 + sar r2d, 1 lea r3, [r1*2] -%if %2 == mmsize .nextrow WEIGHT_OP 0, r1 movh [r0], m0 @@ -135,31 +115,34 @@ cglobal h264_weight_%1x%2_%5, 5, 5, %4 movh [r0+r1], m0 %endif add r0, r3 - dec r2 + dec r2d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_weight_%1x%3_%5.nextrow) -%endif %endmacro INIT_MMX -WEIGHT_FUNC_HALF_MM 4, 8, 8, 0, mmx2 -WEIGHT_FUNC_HALF_MM 4, 4, 8, 0, mmx2 -WEIGHT_FUNC_HALF_MM 4, 2, 8, 0, mmx2 +WEIGHT_FUNC_HALF_MM 4, 0, mmx2 +WEIGHT_FUNC_HALF_MM 4, 0, mmx2 +WEIGHT_FUNC_HALF_MM 4, 0, mmx2 INIT_XMM -WEIGHT_FUNC_HALF_MM 8, 16, 16, 8, sse2 -WEIGHT_FUNC_HALF_MM 8, 8, 16, 8, sse2 -WEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2 +WEIGHT_FUNC_HALF_MM 8, 8, sse2 +WEIGHT_FUNC_HALF_MM 8, 8, sse2 +WEIGHT_FUNC_HALF_MM 8, 8, sse2 %macro BIWEIGHT_SETUP 0 - add r6, 1 - or r6, 1 - add r3, 1 - movd m3, r4d - movd m4, r5d - movd m5, r6d - movd m6, r3d +%ifdef ARCH_X86_64 +%define off_regd r11d +%else +%define off_regd r3d +%endif + mov off_regd, r7m + add off_regd, 1 + or off_regd, 1 + add r4, 1 + movd m3, r5d + movd m4, r6d + movd m5, off_regd + movd m6, r4d pslld m5, m6 psrld m5, 1 %if mmsize == 16 @@ -195,11 +178,10 @@ WEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2 packuswb m0, m1 %endmacro -%macro BIWEIGHT_FUNC_DBL_MM 1 -cglobal h264_biweight_16x%1_mmx2, 7, 7, 0 +INIT_MMX +cglobal h264_biweight_16_mmx2, 7, 7, 0 BIWEIGHT_SETUP - mov r3, %1 -%if %1 == 16 + movifnidn r3d, r3m .nextrow BIWEIGHT_STEPA 0, 1, 0 BIWEIGHT_STEPA 1, 2, 4 @@ -211,23 +193,14 @@ cglobal h264_biweight_16x%1_mmx2, 7, 7, 0 mova [r0+8], m0 add r0, r2 add r1, r2 - dec r3 + dec r3d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_biweight_16x16_mmx2.nextrow) -%endif -%endmacro -INIT_MMX -BIWEIGHT_FUNC_DBL_MM 16 -BIWEIGHT_FUNC_DBL_MM 8 - -%macro BIWEIGHT_FUNC_MM 4 -cglobal h264_biweight_%1x%2_%4, 7, 7, %3 +%macro BIWEIGHT_FUNC_MM 3 +cglobal h264_biweight_%1_%3, 7, 7, %2 BIWEIGHT_SETUP - mov r3, %2 -%if %2 == 16 + movifnidn r3d, r3m .nextrow BIWEIGHT_STEPA 0, 1, 0 BIWEIGHT_STEPA 1, 2, mmsize/2 @@ -235,28 +208,22 @@ cglobal h264_biweight_%1x%2_%4, 7, 7, %3 mova [r0], m0 add r0, r2 add r1, r2 - dec r3 + dec r3d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_biweight_%1x16_%4.nextrow) -%endif %endmacro INIT_MMX -BIWEIGHT_FUNC_MM 8, 16, 0, mmx2 -BIWEIGHT_FUNC_MM 8, 8, 0, mmx2 -BIWEIGHT_FUNC_MM 8, 4, 0, mmx2 +BIWEIGHT_FUNC_MM 8, 0, mmx2 INIT_XMM -BIWEIGHT_FUNC_MM 16, 16, 8, sse2 -BIWEIGHT_FUNC_MM 16, 8, 8, sse2 +BIWEIGHT_FUNC_MM 16, 8, sse2 -%macro BIWEIGHT_FUNC_HALF_MM 5 -cglobal h264_biweight_%1x%2_%5, 7, 7, %4 +%macro BIWEIGHT_FUNC_HALF_MM 3 +cglobal h264_biweight_%1_%3, 7, 7, %2 BIWEIGHT_SETUP - mov r3, %2/2 + movifnidn r3d, r3m + sar r3, 1 lea r4, [r2*2] -%if %2 == mmsize .nextrow BIWEIGHT_STEPA 0, 1, 0 BIWEIGHT_STEPA 1, 2, r2 @@ -270,31 +237,30 @@ cglobal h264_biweight_%1x%2_%5, 7, 7, %4 %endif add r0, r4 add r1, r4 - dec r3 + dec r3d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_biweight_%1x%3_%5.nextrow) -%endif %endmacro INIT_MMX -BIWEIGHT_FUNC_HALF_MM 4, 8, 8, 0, mmx2 -BIWEIGHT_FUNC_HALF_MM 4, 4, 8, 0, mmx2 -BIWEIGHT_FUNC_HALF_MM 4, 2, 8, 0, mmx2 +BIWEIGHT_FUNC_HALF_MM 4, 0, mmx2 INIT_XMM -BIWEIGHT_FUNC_HALF_MM 8, 16, 16, 8, sse2 -BIWEIGHT_FUNC_HALF_MM 8, 8, 16, 8, sse2 -BIWEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2 +BIWEIGHT_FUNC_HALF_MM 8, 8, sse2 %macro BIWEIGHT_SSSE3_SETUP 0 - add r6, 1 - or r6, 1 - add r3, 1 - movd m4, r4d - movd m0, r5d - movd m5, r6d - movd m6, r3d +%ifdef ARCH_X86_64 +%define off_regd r11d +%else +%define off_regd r3d +%endif + mov off_regd, r7m + add off_regd, 1 + or off_regd, 1 + add r4, 1 + movd m4, r5d + movd m0, r6d + movd m5, off_regd + movd m6, r4d pslld m5, m6 psrld m5, 1 punpcklbw m4, m0 @@ -314,12 +280,11 @@ BIWEIGHT_FUNC_HALF_MM 8, 4, 16, 8, sse2 packuswb m0, m2 %endmacro -%macro BIWEIGHT_SSSE3_16 1 -cglobal h264_biweight_16x%1_ssse3, 7, 7, 8 +INIT_XMM +cglobal h264_biweight_16_ssse3, 7, 7, 8 BIWEIGHT_SSSE3_SETUP - mov r3, %1 + movifnidn r3d, r3m -%if %1 == 16 .nextrow movh m0, [r0] movh m2, [r0+8] @@ -330,25 +295,17 @@ cglobal h264_biweight_16x%1_ssse3, 7, 7, 8 mova [r0], m0 add r0, r2 add r1, r2 - dec r3 + dec r3d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_biweight_16x16_ssse3.nextrow) -%endif -%endmacro INIT_XMM -BIWEIGHT_SSSE3_16 16 -BIWEIGHT_SSSE3_16 8 - -%macro BIWEIGHT_SSSE3_8 1 -cglobal h264_biweight_8x%1_ssse3, 7, 7, 8 +cglobal h264_biweight_8_ssse3, 7, 7, 8 BIWEIGHT_SSSE3_SETUP - mov r3, %1/2 + movifnidn r3d, r3m + sar r3, 1 lea r4, [r2*2] -%if %1 == 16 .nextrow movh m0, [r0] movh m1, [r1] @@ -361,15 +318,6 @@ cglobal h264_biweight_8x%1_ssse3, 7, 7, 8 movhps [r0+r2], m0 add r0, r4 add r1, r4 - dec r3 + dec r3d jnz .nextrow REP_RET -%else - jmp mangle(ff_h264_biweight_8x16_ssse3.nextrow) -%endif -%endmacro - -INIT_XMM -BIWEIGHT_SSSE3_8 16 -BIWEIGHT_SSSE3_8 8 -BIWEIGHT_SSSE3_8 4 -- cgit v1.2.3