ffmpeg | branch: master | James Almer <[email protected]> | Sat Oct 21 12:07:16 2017 -0300| [2904db90458a1253e4aea6844ba9a59ac11923b6] | committer: James Almer
Merge commit '994c4bc10751e39c7ed9f67ffd0c0dea5223daf2' * commit '994c4bc10751e39c7ed9f67ffd0c0dea5223daf2': x86util: Port all macros to cpuflags See d5f8a642f6eb1c6e305c41dabddd0fd36ffb3f77 Merged-by: James Almer <[email protected]> > http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=2904db90458a1253e4aea6844ba9a59ac11923b6 --- libavcodec/x86/audiodsp.asm | 5 +---- libavfilter/x86/yadif-16.asm | 24 ------------------------ libavutil/x86/x86util.asm | 42 ++++++++++++++++++++++++++---------------- libswscale/x86/scale.asm | 10 +--------- 4 files changed, 28 insertions(+), 53 deletions(-) diff --git a/libavcodec/x86/audiodsp.asm b/libavcodec/x86/audiodsp.asm index 3973808ca5..de395e5fa8 100644 --- a/libavcodec/x86/audiodsp.asm +++ b/libavcodec/x86/audiodsp.asm @@ -62,7 +62,7 @@ SCALARPRODUCT ; %1 = number of xmm registers used ; %2 = number of inline load/process/store loops per asm loop ; %3 = process 4*mmsize (%3=0) or 8*mmsize (%3=1) bytes per loop -; %4 = CLIPD function takes min/max as float instead of int (CLIPD_SSE2) +; %4 = CLIPD function takes min/max as float instead of int (SSE2 version) ; %5 = suffix %macro VECTOR_CLIP_INT32 4-5 cglobal vector_clip_int32%5, 5,5,%1, dst, src, min, max, len @@ -118,14 +118,11 @@ cglobal vector_clip_int32%5, 5,5,%1, dst, src, min, max, len %endmacro INIT_MMX mmx -%define CLIPD CLIPD_MMX VECTOR_CLIP_INT32 0, 1, 0, 0 INIT_XMM sse2 VECTOR_CLIP_INT32 6, 1, 0, 0, _int -%define CLIPD CLIPD_SSE2 VECTOR_CLIP_INT32 6, 2, 0, 1 INIT_XMM sse4 -%define CLIPD CLIPD_SSE41 %ifdef m8 VECTOR_CLIP_INT32 11, 1, 1, 0 %else diff --git a/libavfilter/x86/yadif-16.asm b/libavfilter/x86/yadif-16.asm index 79d127dfaa..9053b378a5 100644 --- a/libavfilter/x86/yadif-16.asm +++ b/libavfilter/x86/yadif-16.asm @@ -54,30 +54,6 @@ SECTION .text %endif %endmacro -%macro PMINSD 3 -%if cpuflag(sse4) - pminsd %1, %2 -%else - mova %3, %2 - pcmpgtd %3, %1 - pand %1, %3 - pandn %3, %2 - por %1, %3 -%endif -%endmacro - -%macro PMAXSD 3 -%if cpuflag(sse4) - pmaxsd %1, %2 -%else - mova %3, %1 - pcmpgtd %3, %2 - pand %1, %3 - pandn %3, %2 - por %1, %3 -%endif -%endmacro - %macro PMAXUW 2 %if cpuflag(sse4) pmaxuw %1, %2 diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index e1220dfc1a..21419125d5 100644 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@ -357,7 +357,7 @@ %endif %endmacro -%macro ABSB 2 ; source mmreg, temp mmreg (unused for ssse3) +%macro ABSB 2 ; source mmreg, temp mmreg (unused for SSSE3) %if cpuflag(ssse3) pabsb %1, %1 %else @@ -381,7 +381,7 @@ %endif %endmacro -%macro ABSD2_MMX 4 +%macro ABSD2 4 pxor %3, %3 pxor %4, %4 pcmpgtd %3, %1 @@ -475,7 +475,7 @@ %else palignr %1, %2, %3 %endif -%elif cpuflag(mmx) ; [dst,] src1, src2, imm, tmp +%else ; [dst,] src1, src2, imm, tmp %define %%dst %1 %if %0==5 %ifnidn %1, %2 @@ -799,37 +799,47 @@ pminsw %1, %3 %endmacro -%macro PMINSD_MMX 3 ; dst, src, tmp +%macro PMINSD 3 ; dst, src, tmp/unused +%if cpuflag(sse4) + pminsd %1, %2 +%elif cpuflag(sse2) + cvtdq2ps %1, %1 + minps %1, %2 + cvtps2dq %1, %1 +%else mova %3, %2 pcmpgtd %3, %1 pxor %1, %2 pand %1, %3 pxor %1, %2 +%endif %endmacro -%macro PMAXSD_MMX 3 ; dst, src, tmp +%macro PMAXSD 3 ; dst, src, tmp/unused +%if cpuflag(sse4) + pmaxsd %1, %2 +%else mova %3, %1 pcmpgtd %3, %2 pand %1, %3 pandn %3, %2 por %1, %3 +%endif %endmacro -%macro CLIPD_MMX 3-4 ; src/dst, min, max, tmp - PMINSD_MMX %1, %3, %4 - PMAXSD_MMX %1, %2, %4 -%endmacro - -%macro CLIPD_SSE2 3-4 ; src/dst, min (float), max (float), unused +%macro CLIPD 3-4 +%if cpuflag(sse4); src/dst, min, max, unused + pminsd %1, %3 + pmaxsd %1, %2 +%elif cpuflag(sse2) ; src/dst, min (float), max (float), unused cvtdq2ps %1, %1 minps %1, %3 maxps %1, %2 cvtps2dq %1, %1 -%endmacro - -%macro CLIPD_SSE41 3-4 ; src/dst, min, max, unused - pminsd %1, %3 - pmaxsd %1, %2 +%else ; src/dst, min, max, tmp + PMINSD %1, %3, %4 + PMAXSD %1, %2, %4 +%endif %endmacro %macro VBROADCASTSS 2 ; dst xmm/ymm, src m32/xmm diff --git a/libswscale/x86/scale.asm b/libswscale/x86/scale.asm index f9781703a9..83cabff722 100644 --- a/libswscale/x86/scale.asm +++ b/libswscale/x86/scale.asm @@ -364,15 +364,7 @@ cglobal hscale%1to%2_%4, %5, 10, %6, pos0, dst, w, srcmem, filter, fltpos, fltsi movd [dstq+wq*2], m0 %endif ; %3 ==/!= X %else ; %2 == 19 -%if mmsize == 8 - PMINSD_MMX m0, m2, m4 -%elif cpuflag(sse4) - pminsd m0, m2 -%else ; sse2/ssse3 - cvtdq2ps m0, m0 - minps m0, m2 - cvtps2dq m0, m0 -%endif ; mmx/sse2/ssse3/sse4 + PMINSD m0, m2, m4 %ifnidn %3, X mova [dstq+wq*(4>>wshr)], m0 %else ; %3 == X ====================================================================== diff --cc libavfilter/x86/yadif-16.asm index 79d127dfaa,0000000000..9053b378a5 mode 100644,000000..100644 --- a/libavfilter/x86/yadif-16.asm +++ b/libavfilter/x86/yadif-16.asm @@@ -1,317 -1,0 +1,293 @@@ +;***************************************************************************** +;* x86-optimized functions for yadif filter +;* +;* Copyright (C) 2006 Michael Niedermayer <[email protected]> +;* Copyright (c) 2013 Daniel Kang <[email protected]> +;* Copyright (c) 2011-2013 James Darnley <[email protected]> +;* +;* 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_RODATA + +pw_1: times 8 dw 1 +pw_8000: times 8 dw 0x8000 +pd_1: times 4 dd 1 +pd_8000: times 4 dd 0x8000 + +SECTION .text + +%macro PABS 2 +%if cpuflag(ssse3) + pabsd %1, %1 +%else + pxor %2, %2 + pcmpgtd %2, %1 + pxor %1, %2 + psubd %1, %2 +%endif +%endmacro + +%macro PACK 1 +%if cpuflag(sse4) + packusdw %1, %1 +%else + psubd %1, [pd_8000] + packssdw %1, %1 + paddw %1, [pw_8000] +%endif +%endmacro + - %macro PMINSD 3 - %if cpuflag(sse4) - pminsd %1, %2 - %else - mova %3, %2 - pcmpgtd %3, %1 - pand %1, %3 - pandn %3, %2 - por %1, %3 - %endif - %endmacro - - %macro PMAXSD 3 - %if cpuflag(sse4) - pmaxsd %1, %2 - %else - mova %3, %1 - pcmpgtd %3, %2 - pand %1, %3 - pandn %3, %2 - por %1, %3 - %endif - %endmacro - +%macro PMAXUW 2 +%if cpuflag(sse4) + pmaxuw %1, %2 +%else + psubusw %1, %2 + paddusw %1, %2 +%endif +%endmacro + +%macro CHECK 2 + movu m2, [curq+t1+%1*2] + movu m3, [curq+t0+%2*2] + mova m4, m2 + mova m5, m2 + pxor m4, m3 + pavgw m5, m3 + pand m4, [pw_1] + psubusw m5, m4 + RSHIFT m5, 2 + punpcklwd m5, m7 + mova m4, m2 + psubusw m2, m3 + psubusw m3, m4 + PMAXUW m2, m3 + mova m3, m2 + mova m4, m2 + RSHIFT m3, 2 + RSHIFT m4, 4 + punpcklwd m2, m7 + punpcklwd m3, m7 + punpcklwd m4, m7 + paddd m2, m3 + paddd m2, m4 +%endmacro + +%macro CHECK1 0 + mova m3, m0 + pcmpgtd m3, m2 + PMINSD m0, m2, m6 + mova m6, m3 + pand m5, m3 + pandn m3, m1 + por m3, m5 + mova m1, m3 +%endmacro + +%macro CHECK2 0 + paddd m6, [pd_1] + pslld m6, 30 + paddd m2, m6 + mova m3, m0 + pcmpgtd m3, m2 + PMINSD m0, m2, m4 + pand m5, m3 + pandn m3, m1 + por m3, m5 + mova m1, m3 +%endmacro + +; This version of CHECK2 has 3 fewer instructions on sets older than SSE4 but I +; am not sure whether it is any faster. A rewrite or refactor of the filter +; code should make it possible to eliminate the move instruction at the end. It +; exists to satisfy the expectation that the "score" values are in m1. + +; %macro CHECK2 0 +; mova m3, m0 +; pcmpgtd m0, m2 +; pand m0, m6 +; mova m6, m0 +; pand m5, m6 +; pand m2, m0 +; pandn m6, m1 +; pandn m0, m3 +; por m6, m5 +; por m0, m2 +; mova m1, m6 +; %endmacro + +%macro LOAD 2 + movh %1, %2 + punpcklwd %1, m7 +%endmacro + +%macro FILTER 3 +.loop%1: + pxor m7, m7 + LOAD m0, [curq+t1] + LOAD m1, [curq+t0] + LOAD m2, [%2] + LOAD m3, [%3] + mova m4, m3 + paddd m3, m2 + psrad m3, 1 + mova [rsp+ 0], m0 + mova [rsp+16], m3 + mova [rsp+32], m1 + psubd m2, m4 + PABS m2, m4 + LOAD m3, [prevq+t1] + LOAD m4, [prevq+t0] + psubd m3, m0 + psubd m4, m1 + PABS m3, m5 + PABS m4, m5 + paddd m3, m4 + psrld m2, 1 + psrld m3, 1 + PMAXSD m2, m3, m6 + LOAD m3, [nextq+t1] + LOAD m4, [nextq+t0] + psubd m3, m0 + psubd m4, m1 + PABS m3, m5 + PABS m4, m5 + paddd m3, m4 + psrld m3, 1 + PMAXSD m2, m3, m6 + mova [rsp+48], m2 + + paddd m1, m0 + paddd m0, m0 + psubd m0, m1 + psrld m1, 1 + PABS m0, m2 + + movu m2, [curq+t1-1*2] + movu m3, [curq+t0-1*2] + mova m4, m2 + psubusw m2, m3 + psubusw m3, m4 + PMAXUW m2, m3 + mova m3, m2 + RSHIFT m3, 4 + punpcklwd m2, m7 + punpcklwd m3, m7 + paddd m0, m2 + paddd m0, m3 + psubd m0, [pd_1] + + CHECK -2, 0 + CHECK1 + CHECK -3, 1 + CHECK2 + CHECK 0, -2 + CHECK1 + CHECK 1, -3 + CHECK2 + + mova m6, [rsp+48] + cmp DWORD r8m, 2 + jge .end%1 + LOAD m2, [%2+t1*2] + LOAD m4, [%3+t1*2] + LOAD m3, [%2+t0*2] + LOAD m5, [%3+t0*2] + paddd m2, m4 + paddd m3, m5 + psrld m2, 1 + psrld m3, 1 + mova m4, [rsp+ 0] + mova m5, [rsp+16] + mova m7, [rsp+32] + psubd m2, m4 + psubd m3, m7 + mova m0, m5 + psubd m5, m4 + psubd m0, m7 + mova m4, m2 + PMINSD m2, m3, m7 + PMAXSD m3, m4, m7 + PMAXSD m2, m5, m7 + PMINSD m3, m5, m7 + PMAXSD m2, m0, m7 + PMINSD m3, m0, m7 + pxor m4, m4 + PMAXSD m6, m3, m7 + psubd m4, m2 + PMAXSD m6, m4, m7 + +.end%1: + mova m2, [rsp+16] + mova m3, m2 + psubd m2, m6 + paddd m3, m6 + PMAXSD m1, m2, m7 + PMINSD m1, m3, m7 + PACK m1 + + movh [dstq], m1 + add dstq, mmsize/2 + add prevq, mmsize/2 + add curq, mmsize/2 + add nextq, mmsize/2 + sub DWORD r4m, mmsize/4 + jg .loop%1 +%endmacro + +%macro YADIF 0 +%if ARCH_X86_32 +cglobal yadif_filter_line_16bit, 4, 6, 8, 80, dst, prev, cur, next, w, \ + prefs, mrefs, parity, mode +%else +cglobal yadif_filter_line_16bit, 4, 7, 8, 80, dst, prev, cur, next, w, \ + prefs, mrefs, parity, mode +%endif +%if ARCH_X86_32 + mov r4, r5mp + mov r5, r6mp + DECLARE_REG_TMP 4,5 +%else + movsxd r5, DWORD r5m + movsxd r6, DWORD r6m + DECLARE_REG_TMP 5,6 +%endif + + cmp DWORD paritym, 0 + je .parity0 + FILTER 1, prevq, curq + jmp .ret + +.parity0: + FILTER 0, curq, nextq + +.ret: + RET +%endmacro + +INIT_XMM sse4 +YADIF +INIT_XMM ssse3 +YADIF +INIT_XMM sse2 +YADIF +%if ARCH_X86_32 +INIT_MMX mmxext +YADIF +%endif diff --cc libavutil/x86/x86util.asm index e1220dfc1a,81a3ada6e6..21419125d5 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@@ -825,32 -643,18 +836,31 @@@ minps %1, %3 maxps %1, %2 cvtps2dq %1, %1 - %endmacro - - %macro CLIPD_SSE41 3-4 ; src/dst, min, max, unused - pminsd %1, %3 - pmaxsd %1, %2 + %else ; src/dst, min, max, tmp + PMINSD %1, %3, %4 + PMAXSD %1, %2, %4 + %endif %endmacro -%macro VBROADCASTSS 2 ; dst xmm/ymm, src m32 -%if cpuflag(avx) - vbroadcastss %1, %2 -%else ; sse - movss %1, %2 - shufps %1, %1, 0 +%macro VBROADCASTSS 2 ; dst xmm/ymm, src m32/xmm +%if cpuflag(avx2) + vbroadcastss %1, %2 +%elif cpuflag(avx) + %ifnum sizeof%2 ; avx1 register + shufps xmm%1, xmm%2, xmm%2, q0000 + %if sizeof%1 >= 32 ; mmsize>=32 + vinsertf128 %1, %1, xmm%1, 1 + %endif + %else ; avx1 memory + vbroadcastss %1, %2 + %endif +%else + %ifnum sizeof%2 ; sse register + shufps %1, %2, %2, q0000 + %else ; sse memory + movss %1, %2 + shufps %1, %1, 0 + %endif %endif %endmacro _______________________________________________ ffmpeg-cvslog mailing list [email protected] http://ffmpeg.org/mailman/listinfo/ffmpeg-cvslog
