diff options
Diffstat (limited to 'libavutil/x86')
-rw-r--r-- | libavutil/x86/Makefile | 4 | ||||
-rw-r--r-- | libavutil/x86/asm.h | 51 | ||||
-rw-r--r-- | libavutil/x86/bswap.h | 8 | ||||
-rw-r--r-- | libavutil/x86/cpu.c | 12 | ||||
-rw-r--r-- | libavutil/x86/cpu.h | 8 | ||||
-rw-r--r-- | libavutil/x86/cpuid.asm | 8 | ||||
-rw-r--r-- | libavutil/x86/emms.asm | 8 | ||||
-rw-r--r-- | libavutil/x86/emms.h | 12 | ||||
-rw-r--r-- | libavutil/x86/float_dsp.asm | 153 | ||||
-rw-r--r-- | libavutil/x86/float_dsp_init.c | 97 | ||||
-rw-r--r-- | libavutil/x86/intreadwrite.h | 8 | ||||
-rw-r--r-- | libavutil/x86/lls.asm | 11 | ||||
-rw-r--r-- | libavutil/x86/lls_init.c | 14 | ||||
-rw-r--r-- | libavutil/x86/pixelutils.asm | 165 | ||||
-rw-r--r-- | libavutil/x86/pixelutils.h | 26 | ||||
-rw-r--r-- | libavutil/x86/pixelutils_init.c | 58 | ||||
-rw-r--r-- | libavutil/x86/timer.h | 9 | ||||
-rw-r--r-- | libavutil/x86/w64xmmtest.h | 8 | ||||
-rw-r--r-- | libavutil/x86/x86inc.asm | 604 | ||||
-rw-r--r-- | libavutil/x86/x86util.asm | 141 |
20 files changed, 932 insertions, 473 deletions
diff --git a/libavutil/x86/Makefile b/libavutil/x86/Makefile index 1e19082233..ad3bdfc29d 100644 --- a/libavutil/x86/Makefile +++ b/libavutil/x86/Makefile @@ -2,7 +2,11 @@ OBJS += x86/cpu.o \ x86/float_dsp_init.o \ x86/lls_init.o \ +OBJS-$(CONFIG_PIXELUTILS) += x86/pixelutils_init.o \ + YASM-OBJS += x86/cpuid.o \ x86/emms.o \ x86/float_dsp.o \ x86/lls.o \ + +YASM-OBJS-$(CONFIG_PIXELUTILS) += x86/pixelutils.o \ diff --git a/libavutil/x86/asm.h b/libavutil/x86/asm.h index e30f5dbaf7..616ad6c96f 100644 --- a/libavutil/x86/asm.h +++ b/libavutil/x86/asm.h @@ -1,20 +1,20 @@ /* * copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -25,6 +25,7 @@ #include "config.h" typedef struct xmm_reg { uint64_t a, b; } xmm_reg; +typedef struct ymm_reg { uint64_t a, b, c, d; } ymm_reg; #if ARCH_X86_64 # define OPSIZE "q" @@ -107,6 +108,46 @@ typedef int x86_reg; # define LOCAL_MANGLE(a) #a #endif -#define MANGLE(a) EXTERN_PREFIX LOCAL_MANGLE(a) +#if HAVE_INLINE_ASM_DIRECT_SYMBOL_REFS +# define MANGLE(a) EXTERN_PREFIX LOCAL_MANGLE(a) +# define NAMED_CONSTRAINTS_ADD(...) +# define NAMED_CONSTRAINTS(...) +# define NAMED_CONSTRAINTS_ARRAY_ADD(...) +# define NAMED_CONSTRAINTS_ARRAY(...) +#else + /* When direct symbol references are used in code passed to a compiler that does not support them + * then these references need to be converted to named asm constraints instead. + * Instead of returning a direct symbol MANGLE now returns a named constraint for that specific symbol. + * In order for this to work there must also be a corresponding entry in the asm-interface. To add this + * entry use the macro NAMED_CONSTRAINTS() and pass in a list of each symbol reference used in the + * corresponding block of code. (e.g. NAMED_CONSTRAINTS(var1,var2,var3) where var1 is the first symbol etc. ). + * If there are already existing constraints then use NAMED_CONSTRAINTS_ADD to add to the existing constraint list. + */ +# define MANGLE(a) "%["#a"]" + // Intel/MSVC does not correctly expand va-args so we need a rather ugly hack in order to get it to work +# define FE_0(P,X) P(X) +# define FE_1(P,X,X1) P(X), FE_0(P,X1) +# define FE_2(P,X,X1,X2) P(X), FE_1(P,X1,X2) +# define FE_3(P,X,X1,X2,X3) P(X), FE_2(P,X1,X2,X3) +# define FE_4(P,X,X1,X2,X3,X4) P(X), FE_3(P,X1,X2,X3,X4) +# define FE_5(P,X,X1,X2,X3,X4,X5) P(X), FE_4(P,X1,X2,X3,X4,X5) +# define FE_6(P,X,X1,X2,X3,X4,X5,X6) P(X), FE_5(P,X1,X2,X3,X4,X5,X6) +# define FE_7(P,X,X1,X2,X3,X4,X5,X6,X7) P(X), FE_6(P,X1,X2,X3,X4,X5,X6,X7) +# define FE_8(P,X,X1,X2,X3,X4,X5,X6,X7,X8) P(X), FE_7(P,X1,X2,X3,X4,X5,X6,X7,X8) +# define FE_9(P,X,X1,X2,X3,X4,X5,X6,X7,X8,X9) P(X), FE_8(P,X1,X2,X3,X4,X5,X6,X7,X8,X9) +# define GET_FE_IMPL(_0,_1,_2,_3,_4,_5,_6,_7,_8,_9,NAME,...) NAME +# define GET_FE(A) GET_FE_IMPL A +# define GET_FE_GLUE(x, y) x y +# define FOR_EACH_VA(P,...) GET_FE_GLUE(GET_FE((__VA_ARGS__,FE_9,FE_8,FE_7,FE_6,FE_5,FE_4,FE_3,FE_2,FE_1,FE_0)), (P,__VA_ARGS__)) +# define NAME_CONSTRAINT(x) [x] "m"(x) + // Parameters are a list of each symbol reference required +# define NAMED_CONSTRAINTS_ADD(...) , FOR_EACH_VA(NAME_CONSTRAINT,__VA_ARGS__) + // Same but without comma for when there are no previously defined constraints +# define NAMED_CONSTRAINTS(...) FOR_EACH_VA(NAME_CONSTRAINT,__VA_ARGS__) + // Same as above NAMED_CONSTRAINTS except used for passing arrays/pointers instead of normal variables +# define NAME_CONSTRAINT_ARRAY(x) [x] "m"(*x) +# define NAMED_CONSTRAINTS_ARRAY_ADD(...) , FOR_EACH_VA(NAME_CONSTRAINT_ARRAY,__VA_ARGS__) +# define NAMED_CONSTRAINTS_ARRAY(...) FOR_EACH_VA(NAME_CONSTRAINT_ARRAY,__VA_ARGS__) +#endif #endif /* AVUTIL_X86_ASM_H */ diff --git a/libavutil/x86/bswap.h b/libavutil/x86/bswap.h index c73be9af81..08e2a62520 100644 --- a/libavutil/x86/bswap.h +++ b/libavutil/x86/bswap.h @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c index 8be6d94742..2b62e92479 100644 --- a/libavutil/x86/cpu.c +++ b/libavutil/x86/cpu.c @@ -3,20 +3,20 @@ * (c)1997-99 by H. Dietz and R. Fisher * Converted to C and improved by Fabrice Bellard. * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -45,7 +45,7 @@ "cpuid \n\t" \ "xchg %%"REG_b", %%"REG_S \ : "=a" (eax), "=S" (ebx), "=c" (ecx), "=d" (edx) \ - : "0" (index)) + : "0" (index), "2"(0)) #define xgetbv(index, eax, edx) \ __asm__ (".byte 0x0f, 0x01, 0xd0" : "=a"(eax), "=d"(edx) : "c" (index)) @@ -143,7 +143,7 @@ int ff_get_cpu_flags_x86(void) if (max_std_level >= 7) { cpuid(7, eax, ebx, ecx, edx); #if HAVE_AVX2 - if (ebx & 0x00000020) + if ((rval & AV_CPU_FLAG_AVX) && (ebx & 0x00000020)) rval |= AV_CPU_FLAG_AVX2; #endif /* HAVE_AVX2 */ /* BMI1/2 don't need OS support */ diff --git a/libavutil/x86/cpu.h b/libavutil/x86/cpu.h index 50da30e389..bc64b1b3bd 100644 --- a/libavutil/x86/cpu.h +++ b/libavutil/x86/cpu.h @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavutil/x86/cpuid.asm b/libavutil/x86/cpuid.asm index 1cb8e94ea3..c3f7866ec7 100644 --- a/libavutil/x86/cpuid.asm +++ b/libavutil/x86/cpuid.asm @@ -4,20 +4,20 @@ ;* Authors: Loren Merritt <lorenm@u.washington.edu> ;* Fiona Glaser <fiona@x264.com> ;* -;* This file is part of Libav. +;* This file is part of FFmpeg. ;* -;* Libav is free software; you can redistribute it and/or +;* 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. ;* -;* Libav is distributed in the hope that it will be useful, +;* 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 Libav; if not, write to the Free Software +;* License along with FFmpeg; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** diff --git a/libavutil/x86/emms.asm b/libavutil/x86/emms.asm index a6851acc99..0aad34af3f 100644 --- a/libavutil/x86/emms.asm +++ b/libavutil/x86/emms.asm @@ -1,20 +1,20 @@ ;***************************************************************************** ;* Copyright (C) 2013 Martin Storsjo ;* -;* This file is part of Libav. +;* This file is part of FFmpeg. ;* -;* Libav is free software; you can redistribute it and/or +;* 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. ;* -;* Libav is distributed in the hope that it will be useful, +;* 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 Libav; if not, write to the Free Software +;* License along with FFmpeg; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** diff --git a/libavutil/x86/emms.h b/libavutil/x86/emms.h index 2ed9e5d09d..a529b6bbbe 100644 --- a/libavutil/x86/emms.h +++ b/libavutil/x86/emms.h @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -21,6 +21,7 @@ #include "config.h" #include "libavutil/attributes.h" +#include "libavutil/cpu.h" void avpriv_emms_yasm(void); @@ -33,7 +34,8 @@ void avpriv_emms_yasm(void); */ static av_always_inline void emms_c(void) { - __asm__ volatile ("emms" ::: "memory"); + if(av_get_cpu_flags() & AV_CPU_FLAG_MMX) + __asm__ volatile ("emms" ::: "memory"); } #elif HAVE_MMX && HAVE_MM_EMPTY # include <mmintrin.h> diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm index d96249978a..ec3d22b230 100644 --- a/libavutil/x86/float_dsp.asm +++ b/libavutil/x86/float_dsp.asm @@ -1,20 +1,22 @@ ;***************************************************************************** ;* x86-optimized Float DSP functions ;* -;* This file is part of Libav. +;* Copyright 2006 Loren Merritt ;* -;* Libav is free software; you can redistribute it and/or +;* 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. ;* -;* Libav is distributed in the hope that it will be useful, +;* 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 Libav; if not, write to the Free Software +;* License along with FFmpeg; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** @@ -48,8 +50,10 @@ ALIGN 16 INIT_XMM sse VECTOR_FMUL +%if HAVE_AVX_EXTERNAL INIT_YMM avx VECTOR_FMUL +%endif ;------------------------------------------------------------------------------ ; void ff_vector_fmac_scalar(float *dst, const float *src, float mul, int len) @@ -57,33 +61,48 @@ VECTOR_FMUL %macro VECTOR_FMAC_SCALAR 0 %if UNIX64 -cglobal vector_fmac_scalar, 3,3,3, dst, src, len +cglobal vector_fmac_scalar, 3,3,5, dst, src, len %else -cglobal vector_fmac_scalar, 4,4,3, dst, src, mul, len +cglobal vector_fmac_scalar, 4,4,5, dst, src, mul, len %endif %if ARCH_X86_32 VBROADCASTSS m0, mulm %else %if WIN64 - mova xmm0, xmm2 + SWAP 0, 2 %endif - shufps xmm0, xmm0, 0 + shufps xm0, xm0, 0 %if cpuflag(avx) - vinsertf128 m0, m0, xmm0, 1 + vinsertf128 m0, m0, xm0, 1 %endif %endif lea lenq, [lend*4-64] .loop: -%assign a 0 -%rep 32/mmsize - mulps m1, m0, [srcq+lenq+(a+0)*mmsize] - mulps m2, m0, [srcq+lenq+(a+1)*mmsize] - addps m1, m1, [dstq+lenq+(a+0)*mmsize] - addps m2, m2, [dstq+lenq+(a+1)*mmsize] - mova [dstq+lenq+(a+0)*mmsize], m1 - mova [dstq+lenq+(a+1)*mmsize], m2 -%assign a a+2 -%endrep +%if cpuflag(fma3) + mova m1, [dstq+lenq] + mova m2, [dstq+lenq+1*mmsize] + fmaddps m1, m0, [srcq+lenq], m1 + fmaddps m2, m0, [srcq+lenq+1*mmsize], m2 +%else ; cpuflag + mulps m1, m0, [srcq+lenq] + mulps m2, m0, [srcq+lenq+1*mmsize] +%if mmsize < 32 + mulps m3, m0, [srcq+lenq+2*mmsize] + mulps m4, m0, [srcq+lenq+3*mmsize] +%endif ; mmsize + addps m1, m1, [dstq+lenq] + addps m2, m2, [dstq+lenq+1*mmsize] +%if mmsize < 32 + addps m3, m3, [dstq+lenq+2*mmsize] + addps m4, m4, [dstq+lenq+3*mmsize] +%endif ; mmsize +%endif ; cpuflag + mova [dstq+lenq], m1 + mova [dstq+lenq+1*mmsize], m2 +%if mmsize < 32 + mova [dstq+lenq+2*mmsize], m3 + mova [dstq+lenq+3*mmsize], m4 +%endif ; mmsize sub lenq, 64 jge .loop REP_RET @@ -91,8 +110,14 @@ cglobal vector_fmac_scalar, 4,4,3, dst, src, mul, len INIT_XMM sse VECTOR_FMAC_SCALAR +%if HAVE_AVX_EXTERNAL INIT_YMM avx VECTOR_FMAC_SCALAR +%endif +%if HAVE_FMA3_EXTERNAL +INIT_YMM fma3 +VECTOR_FMAC_SCALAR +%endif ;------------------------------------------------------------------------------ ; void ff_vector_fmul_scalar(float *dst, const float *src, float mul, int len) @@ -141,16 +166,11 @@ cglobal vector_dmul_scalar, 4,4,3, dst, src, mul, len VBROADCASTSD m0, mulm %else %if WIN64 - movlhps xmm2, xmm2 -%if cpuflag(avx) - vinsertf128 ymm2, ymm2, xmm2, 1 -%endif SWAP 0, 2 -%else - movlhps xmm0, xmm0 -%if cpuflag(avx) - vinsertf128 ymm0, ymm0, xmm0, 1 %endif + movlhps xm0, xm0 +%if cpuflag(avx) + vinsertf128 ym0, ym0, xm0, 1 %endif %endif lea lenq, [lend*8-2*mmsize] @@ -172,20 +192,85 @@ VECTOR_DMUL_SCALAR %endif ;----------------------------------------------------------------------------- +; vector_fmul_window(float *dst, const float *src0, +; const float *src1, const float *win, int len); +;----------------------------------------------------------------------------- +%macro VECTOR_FMUL_WINDOW 0 +cglobal vector_fmul_window, 5, 6, 6, dst, src0, src1, win, len, len1 + shl lend, 2 + lea len1q, [lenq - mmsize] + add src0q, lenq + add dstq, lenq + add winq, lenq + neg lenq +.loop + mova m0, [winq + lenq] + mova m4, [src0q + lenq] +%if cpuflag(sse) + mova m1, [winq + len1q] + mova m5, [src1q + len1q] + shufps m1, m1, 0x1b + shufps m5, m5, 0x1b + mova m2, m0 + mova m3, m1 + mulps m2, m4 + mulps m3, m5 + mulps m1, m4 + mulps m0, m5 + addps m2, m3 + subps m1, m0 + shufps m2, m2, 0x1b +%else + pswapd m1, [winq + len1q] + pswapd m5, [src1q + len1q] + mova m2, m0 + mova m3, m1 + pfmul m2, m4 + pfmul m3, m5 + pfmul m1, m4 + pfmul m0, m5 + pfadd m2, m3 + pfsub m1, m0 + pswapd m2, m2 +%endif + mova [dstq + lenq], m1 + mova [dstq + len1q], m2 + sub len1q, mmsize + add lenq, mmsize + jl .loop +%if mmsize == 8 + femms +%endif + REP_RET +%endmacro + +INIT_MMX 3dnowext +VECTOR_FMUL_WINDOW +INIT_XMM sse +VECTOR_FMUL_WINDOW + +;----------------------------------------------------------------------------- ; vector_fmul_add(float *dst, const float *src0, const float *src1, ; const float *src2, int len) ;----------------------------------------------------------------------------- %macro VECTOR_FMUL_ADD 0 -cglobal vector_fmul_add, 5,5,2, dst, src0, src1, src2, len +cglobal vector_fmul_add, 5,5,4, dst, src0, src1, src2, len lea lenq, [lend*4 - 2*mmsize] ALIGN 16 .loop: mova m0, [src0q + lenq] mova m1, [src0q + lenq + mmsize] +%if cpuflag(fma3) + mova m2, [src2q + lenq] + mova m3, [src2q + lenq + mmsize] + fmaddps m0, m0, [src1q + lenq], m2 + fmaddps m1, m1, [src1q + lenq + mmsize], m3 +%else mulps m0, m0, [src1q + lenq] mulps m1, m1, [src1q + lenq + mmsize] addps m0, m0, [src2q + lenq] addps m1, m1, [src2q + lenq + mmsize] +%endif mova [dstq + lenq], m0 mova [dstq + lenq + mmsize], m1 @@ -196,8 +281,14 @@ ALIGN 16 INIT_XMM sse VECTOR_FMUL_ADD +%if HAVE_AVX_EXTERNAL INIT_YMM avx VECTOR_FMUL_ADD +%endif +%if HAVE_FMA3_EXTERNAL +INIT_YMM fma3 +VECTOR_FMUL_ADD +%endif ;----------------------------------------------------------------------------- ; void vector_fmul_reverse(float *dst, const float *src0, const float *src1, @@ -233,8 +324,10 @@ ALIGN 16 INIT_XMM sse VECTOR_FMUL_REVERSE +%if HAVE_AVX_EXTERNAL INIT_YMM avx VECTOR_FMUL_REVERSE +%endif ; float scalarproduct_float_sse(const float *v1, const float *v2, int len) INIT_XMM sse @@ -272,8 +365,8 @@ cglobal butterflies_float, 3,3,3, src0, src1, len test lenq, lenq jz .end shl lenq, 2 - lea src0q, [src0q + lenq] - lea src1q, [src1q + lenq] + add src0q, lenq + add src1q, lenq neg lenq .loop: mova m0, [src0q + lenq] diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c index a04d91c923..64b3a4d6bd 100644 --- a/libavutil/x86/float_dsp_init.c +++ b/libavutil/x86/float_dsp_init.c @@ -1,18 +1,18 @@ /* - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -33,6 +33,8 @@ void ff_vector_fmac_scalar_sse(float *dst, const float *src, float mul, int len); void ff_vector_fmac_scalar_avx(float *dst, const float *src, float mul, int len); +void ff_vector_fmac_scalar_fma3(float *dst, const float *src, float mul, + int len); void ff_vector_fmul_scalar_sse(float *dst, const float *src, float mul, int len); @@ -42,10 +44,17 @@ void ff_vector_dmul_scalar_sse2(double *dst, const double *src, void ff_vector_dmul_scalar_avx(double *dst, const double *src, double mul, int len); +void ff_vector_fmul_window_3dnowext(float *dst, const float *src0, + const float *src1, const float *win, int len); +void ff_vector_fmul_window_sse(float *dst, const float *src0, + const float *src1, const float *win, int len); + void ff_vector_fmul_add_sse(float *dst, const float *src0, const float *src1, const float *src2, int len); void ff_vector_fmul_add_avx(float *dst, const float *src0, const float *src1, const float *src2, int len); +void ff_vector_fmul_add_fma3(float *dst, const float *src0, const float *src1, + const float *src2, int len); void ff_vector_fmul_reverse_sse(float *dst, const float *src0, const float *src1, int len); @@ -56,88 +65,18 @@ float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order); void ff_butterflies_float_sse(float *src0, float *src1, int len); -#if HAVE_6REGS && HAVE_INLINE_ASM -static void vector_fmul_window_3dnowext(float *dst, const float *src0, - const float *src1, const float *win, - int len) -{ - x86_reg i = -len * 4; - x86_reg j = len * 4 - 8; - __asm__ volatile ( - "1: \n" - "pswapd (%5, %1), %%mm1 \n" - "movq (%5, %0), %%mm0 \n" - "pswapd (%4, %1), %%mm5 \n" - "movq (%3, %0), %%mm4 \n" - "movq %%mm0, %%mm2 \n" - "movq %%mm1, %%mm3 \n" - "pfmul %%mm4, %%mm2 \n" // src0[len + i] * win[len + i] - "pfmul %%mm5, %%mm3 \n" // src1[j] * win[len + j] - "pfmul %%mm4, %%mm1 \n" // src0[len + i] * win[len + j] - "pfmul %%mm5, %%mm0 \n" // src1[j] * win[len + i] - "pfadd %%mm3, %%mm2 \n" - "pfsub %%mm0, %%mm1 \n" - "pswapd %%mm2, %%mm2 \n" - "movq %%mm1, (%2, %0) \n" - "movq %%mm2, (%2, %1) \n" - "sub $8, %1 \n" - "add $8, %0 \n" - "jl 1b \n" - "femms \n" - : "+r"(i), "+r"(j) - : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len) - ); -} - -static void vector_fmul_window_sse(float *dst, const float *src0, - const float *src1, const float *win, int len) -{ - x86_reg i = -len * 4; - x86_reg j = len * 4 - 16; - __asm__ volatile ( - "1: \n" - "movaps (%5, %1), %%xmm1 \n" - "movaps (%5, %0), %%xmm0 \n" - "movaps (%4, %1), %%xmm5 \n" - "movaps (%3, %0), %%xmm4 \n" - "shufps $0x1b, %%xmm1, %%xmm1 \n" - "shufps $0x1b, %%xmm5, %%xmm5 \n" - "movaps %%xmm0, %%xmm2 \n" - "movaps %%xmm1, %%xmm3 \n" - "mulps %%xmm4, %%xmm2 \n" // src0[len + i] * win[len + i] - "mulps %%xmm5, %%xmm3 \n" // src1[j] * win[len + j] - "mulps %%xmm4, %%xmm1 \n" // src0[len + i] * win[len + j] - "mulps %%xmm5, %%xmm0 \n" // src1[j] * win[len + i] - "addps %%xmm3, %%xmm2 \n" - "subps %%xmm0, %%xmm1 \n" - "shufps $0x1b, %%xmm2, %%xmm2 \n" - "movaps %%xmm1, (%2, %0) \n" - "movaps %%xmm2, (%2, %1) \n" - "sub $16, %1 \n" - "add $16, %0 \n" - "jl 1b \n" - : "+r"(i), "+r"(j) - : "r"(dst + len), "r"(src0 + len), "r"(src1), "r"(win + len) - ); -} -#endif /* HAVE_6REGS && HAVE_INLINE_ASM */ - av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) { int cpu_flags = av_get_cpu_flags(); -#if HAVE_6REGS && HAVE_INLINE_ASM - if (INLINE_AMD3DNOWEXT(cpu_flags)) { - fdsp->vector_fmul_window = vector_fmul_window_3dnowext; + if (EXTERNAL_AMD3DNOWEXT(cpu_flags)) { + fdsp->vector_fmul_window = ff_vector_fmul_window_3dnowext; } - if (INLINE_SSE(cpu_flags)) { - fdsp->vector_fmul_window = vector_fmul_window_sse; - } -#endif if (EXTERNAL_SSE(cpu_flags)) { fdsp->vector_fmul = ff_vector_fmul_sse; fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_sse; fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_sse; + fdsp->vector_fmul_window = ff_vector_fmul_window_sse; fdsp->vector_fmul_add = ff_vector_fmul_add_sse; fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_sse; fdsp->scalarproduct_float = ff_scalarproduct_float_sse; @@ -153,4 +92,8 @@ av_cold void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp) fdsp->vector_fmul_add = ff_vector_fmul_add_avx; fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_avx; } + if (EXTERNAL_FMA3(cpu_flags)) { + fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_fma3; + fdsp->vector_fmul_add = ff_vector_fmul_add_fma3; + } } diff --git a/libavutil/x86/intreadwrite.h b/libavutil/x86/intreadwrite.h index 635096e569..4061d19231 100644 --- a/libavutil/x86/intreadwrite.h +++ b/libavutil/x86/intreadwrite.h @@ -1,20 +1,20 @@ /* * Copyright (c) 2010 Alexander Strange <astrange@ithinksw.com> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavutil/x86/lls.asm b/libavutil/x86/lls.asm index eab85ed050..769befb769 100644 --- a/libavutil/x86/lls.asm +++ b/libavutil/x86/lls.asm @@ -3,20 +3,20 @@ ;* ;* Copyright (c) 2013 Loren Merritt ;* -;* This file is part of Libav. +;* This file is part of FFmpeg. ;* -;* Libav is free software; you can redistribute it and/or +;* 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. ;* -;* Libav is distributed in the hope that it will be useful, +;* 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 Libav; if not, write to the Free Software +;* License along with FFmpeg; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** @@ -125,6 +125,7 @@ cglobal update_lls, 2,5,8, ctx, var, i, j, covar2 .ret: REP_RET +%if HAVE_AVX_EXTERNAL INIT_YMM avx cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 %define covarq ctxq @@ -194,7 +195,7 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 jle .loop2x1 .ret: REP_RET - +%endif INIT_XMM sse2 cglobal evaluate_lls, 3,4,2, ctx, var, order, i diff --git a/libavutil/x86/lls_init.c b/libavutil/x86/lls_init.c index 888bc54a39..f53190488a 100644 --- a/libavutil/x86/lls_init.c +++ b/libavutil/x86/lls_init.c @@ -3,29 +3,29 @@ * * Copyright (c) 2013 Loren Merritt * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * 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/lls.h" #include "libavutil/x86/cpu.h" -void ff_update_lls_sse2(LLSModel *m, double *var); -void ff_update_lls_avx(LLSModel *m, double *var); -double ff_evaluate_lls_sse2(LLSModel *m, double *var, int order); +void ff_update_lls_sse2(LLSModel *m, const double *var); +void ff_update_lls_avx(LLSModel *m, const double *var); +double ff_evaluate_lls_sse2(LLSModel *m, const double *var, int order); av_cold void ff_init_lls_x86(LLSModel *m) { diff --git a/libavutil/x86/pixelutils.asm b/libavutil/x86/pixelutils.asm new file mode 100644 index 0000000000..7522f24a42 --- /dev/null +++ b/libavutil/x86/pixelutils.asm @@ -0,0 +1,165 @@ +;****************************************************************************** +;* Pixel utilities SIMD +;* +;* Copyright (C) 2002-2004 Michael Niedermayer <michaelni@gmx.at> +;* Copyright (C) 2014 Clément Bœsch <u pkh me> +;* +;* 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 "x86util.asm" + +SECTION_TEXT + +;------------------------------------------------------------------------------- +; int ff_pixelutils_sad_8x8_mmx(const uint8_t *src1, ptrdiff_t stride1, +; const uint8_t *src2, ptrdiff_t stride2); +;------------------------------------------------------------------------------- +INIT_MMX mmx +cglobal pixelutils_sad_8x8, 4,4,0, src1, stride1, src2, stride2 + pxor m7, m7 + pxor m6, m6 +%rep 4 + mova m0, [src1q] + mova m2, [src1q + stride1q] + mova m1, [src2q] + mova m3, [src2q + stride2q] + psubusb m4, m0, m1 + psubusb m5, m2, m3 + psubusb m1, m0 + psubusb m3, m2 + por m1, m4 + por m3, m5 + punpcklbw m0, m1, m7 + punpcklbw m2, m3, m7 + punpckhbw m1, m7 + punpckhbw m3, m7 + paddw m0, m1 + paddw m2, m3 + paddw m0, m2 + paddw m6, m0 + lea src1q, [src1q + 2*stride1q] + lea src2q, [src2q + 2*stride2q] +%endrep + psrlq m0, m6, 32 + paddw m6, m0 + psrlq m0, m6, 16 + paddw m6, m0 + movd eax, m6 + movzx eax, ax + RET + +;------------------------------------------------------------------------------- +; int ff_pixelutils_sad_8x8_mmxext(const uint8_t *src1, ptrdiff_t stride1, +; const uint8_t *src2, ptrdiff_t stride2); +;------------------------------------------------------------------------------- +INIT_MMX mmxext +cglobal pixelutils_sad_8x8, 4,4,0, src1, stride1, src2, stride2 + pxor m2, m2 +%rep 4 + mova m0, [src1q] + mova m1, [src1q + stride1q] + psadbw m0, [src2q] + psadbw m1, [src2q + stride2q] + paddw m2, m0 + paddw m2, m1 + lea src1q, [src1q + 2*stride1q] + lea src2q, [src2q + 2*stride2q] +%endrep + movd eax, m2 + RET + +;------------------------------------------------------------------------------- +; int ff_pixelutils_sad_16x16_mmxext(const uint8_t *src1, ptrdiff_t stride1, +; const uint8_t *src2, ptrdiff_t stride2); +;------------------------------------------------------------------------------- +INIT_MMX mmxext +cglobal pixelutils_sad_16x16, 4,4,0, src1, stride1, src2, stride2 + pxor m2, m2 +%rep 16 + mova m0, [src1q] + mova m1, [src1q + 8] + psadbw m0, [src2q] + psadbw m1, [src2q + 8] + paddw m2, m0 + paddw m2, m1 + add src1q, stride1q + add src2q, stride2q +%endrep + movd eax, m2 + RET + +;------------------------------------------------------------------------------- +; int ff_pixelutils_sad_16x16_sse(const uint8_t *src1, ptrdiff_t stride1, +; const uint8_t *src2, ptrdiff_t stride2); +;------------------------------------------------------------------------------- +INIT_XMM sse2 +cglobal pixelutils_sad_16x16, 4,4,5, src1, stride1, src2, stride2 + movu m4, [src1q] + movu m2, [src2q] + movu m1, [src1q + stride1q] + movu m3, [src2q + stride2q] + psadbw m4, m2 + psadbw m1, m3 + paddw m4, m1 +%rep 7 + lea src1q, [src1q + 2*stride1q] + lea src2q, [src2q + 2*stride2q] + movu m0, [src1q] + movu m2, [src2q] + movu m1, [src1q + stride1q] + movu m3, [src2q + stride2q] + psadbw m0, m2 + psadbw m1, m3 + paddw m4, m0 + paddw m4, m1 +%endrep + movhlps m0, m4 + paddw m4, m0 + movd eax, m4 + RET + +;------------------------------------------------------------------------------- +; int ff_pixelutils_sad_[au]_16x16_sse(const uint8_t *src1, ptrdiff_t stride1, +; const uint8_t *src2, ptrdiff_t stride2); +;------------------------------------------------------------------------------- +%macro SAD_XMM_16x16 1 +INIT_XMM sse2 +cglobal pixelutils_sad_%1_16x16, 4,4,3, src1, stride1, src2, stride2 + mov%1 m2, [src2q] + psadbw m2, [src1q] + mov%1 m1, [src2q + stride2q] + psadbw m1, [src1q + stride1q] + paddw m2, m1 +%rep 7 + lea src1q, [src1q + 2*stride1q] + lea src2q, [src2q + 2*stride2q] + mov%1 m0, [src2q] + psadbw m0, [src1q] + mov%1 m1, [src2q + stride2q] + psadbw m1, [src1q + stride1q] + paddw m2, m0 + paddw m2, m1 +%endrep + movhlps m0, m2 + paddw m2, m0 + movd eax, m2 + RET +%endmacro + +SAD_XMM_16x16 a +SAD_XMM_16x16 u diff --git a/libavutil/x86/pixelutils.h b/libavutil/x86/pixelutils.h new file mode 100644 index 0000000000..876cf46053 --- /dev/null +++ b/libavutil/x86/pixelutils.h @@ -0,0 +1,26 @@ +/* + * 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 + */ + +#ifndef AVUTIL_X86_PIXELUTILS_H +#define AVUTIL_X86_PIXELUTILS_H + +#include "libavutil/pixelutils.h" + +void ff_pixelutils_sad_init_x86(av_pixelutils_sad_fn *sad, int aligned); + +#endif /* AVUTIL_X86_PIXELUTILS_H */ diff --git a/libavutil/x86/pixelutils_init.c b/libavutil/x86/pixelutils_init.c new file mode 100644 index 0000000000..d60051067a --- /dev/null +++ b/libavutil/x86/pixelutils_init.c @@ -0,0 +1,58 @@ +/* + * 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 "config.h" + +#include "pixelutils.h" +#include "cpu.h" + +int ff_pixelutils_sad_8x8_mmx(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); +int ff_pixelutils_sad_8x8_mmxext(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); + +int ff_pixelutils_sad_16x16_mmxext(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); +int ff_pixelutils_sad_16x16_sse2(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); +int ff_pixelutils_sad_a_16x16_sse2(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); +int ff_pixelutils_sad_u_16x16_sse2(const uint8_t *src1, ptrdiff_t stride1, + const uint8_t *src2, ptrdiff_t stride2); + +void ff_pixelutils_sad_init_x86(av_pixelutils_sad_fn *sad, int aligned) +{ + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_MMX(cpu_flags)) { + sad[2] = ff_pixelutils_sad_8x8_mmx; + } + + if (EXTERNAL_MMXEXT(cpu_flags)) { + sad[2] = ff_pixelutils_sad_8x8_mmxext; + sad[3] = ff_pixelutils_sad_16x16_mmxext; + } + + if (EXTERNAL_SSE2(cpu_flags)) { + switch (aligned) { + case 0: sad[3] = ff_pixelutils_sad_16x16_sse2; break; // src1 unaligned, src2 unaligned + case 1: sad[3] = ff_pixelutils_sad_u_16x16_sse2; break; // src1 aligned, src2 unaligned + case 2: sad[3] = ff_pixelutils_sad_a_16x16_sse2; break; // src1 aligned, src2 aligned + } + } +} diff --git a/libavutil/x86/timer.h b/libavutil/x86/timer.h index cdd67ddbb5..5b24b511c1 100644 --- a/libavutil/x86/timer.h +++ b/libavutil/x86/timer.h @@ -1,20 +1,20 @@ /* * copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ @@ -25,6 +25,7 @@ #if HAVE_INLINE_ASM +#define FF_TIMER_UNITS "decicycles" #define AV_READ_TIME read_time static inline uint64_t read_time(void) diff --git a/libavutil/x86/w64xmmtest.h b/libavutil/x86/w64xmmtest.h index b4ce7d3daf..9df499f7f6 100644 --- a/libavutil/x86/w64xmmtest.h +++ b/libavutil/x86/w64xmmtest.h @@ -2,20 +2,20 @@ * check XMM registers for clobbers on Win64 * Copyright (c) 2008 Ramiro Polla <ramiro.polla@gmail.com> * - * This file is part of Libav. + * This file is part of FFmpeg. * - * Libav is free software; you can redistribute it and/or + * 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. * - * Libav is distributed in the hope that it will be useful, + * 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 Libav; if not, write to the Free Software + * License along with FFmpeg; if not, write to the Free Software * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA */ diff --git a/libavutil/x86/x86inc.asm b/libavutil/x86/x86inc.asm index e3ed3813e0..f1b4b85a87 100644 --- a/libavutil/x86/x86inc.asm +++ b/libavutil/x86/x86inc.asm @@ -871,7 +871,7 @@ SECTION .note.GNU-stack noalloc noexec nowrite progbits %assign %%i 0 %rep num_mmregs CAT_XDEFINE m, %%i, ymm %+ %%i - CAT_XDEFINE nymm, %%i, %%i + CAT_XDEFINE nnymm, %%i, %%i %assign %%i %%i+1 %endrep INIT_CPUFLAGS %1 @@ -1047,15 +1047,16 @@ INIT_XMM %endmacro ;%1 == instruction -;%2 == 1 if float, 0 if int -;%3 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise -;%4 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not -;%5+: operands -%macro RUN_AVX_INSTR 5-8+ - %ifnum sizeof%6 +;%2 == minimal instruction set +;%3 == 1 if float, 0 if int +;%4 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise +;%5 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not +;%6+: operands +%macro RUN_AVX_INSTR 6-9+ + %ifnum sizeof%7 + %assign __sizeofreg sizeof%7 + %elifnum sizeof%6 %assign __sizeofreg sizeof%6 - %elifnum sizeof%5 - %assign __sizeofreg sizeof%5 %else %assign __sizeofreg mmsize %endif @@ -1064,325 +1065,335 @@ INIT_XMM %xdefine __instr v%1 %else %xdefine __instr %1 - %if %0 >= 7+%3 + %if %0 >= 8+%4 %assign __emulate_avx 1 %endif %endif + %ifnidn %2, fnord + %ifdef cpuname + %if notcpuflag(%2) + %error use of ``%1'' %2 instruction in cpuname function: current_function + %elif cpuflags_%2 < cpuflags_sse && notcpuflag(sse2) && __sizeofreg > 8 + %error use of ``%1'' sse2 instruction in cpuname function: current_function + %endif + %endif + %endif %if __emulate_avx - %xdefine __src1 %6 - %xdefine __src2 %7 - %ifnidn %5, %6 - %if %0 >= 8 - CHECK_AVX_INSTR_EMU {%1 %5, %6, %7, %8}, %5, %7, %8 + %xdefine __src1 %7 + %xdefine __src2 %8 + %ifnidn %6, %7 + %if %0 >= 9 + CHECK_AVX_INSTR_EMU {%1 %6, %7, %8, %9}, %6, %8, %9 %else - CHECK_AVX_INSTR_EMU {%1 %5, %6, %7}, %5, %7 + CHECK_AVX_INSTR_EMU {%1 %6, %7, %8}, %6, %8 %endif - %if %4 && %3 == 0 - %ifnid %7 + %if %5 && %4 == 0 + %ifnid %8 ; 3-operand AVX instructions with a memory arg can only have it in src2, ; whereas SSE emulation prefers to have it in src1 (i.e. the mov). ; So, if the instruction is commutative with a memory arg, swap them. - %xdefine __src1 %7 - %xdefine __src2 %6 + %xdefine __src1 %8 + %xdefine __src2 %7 %endif %endif %if __sizeofreg == 8 - MOVQ %5, __src1 - %elif %2 - MOVAPS %5, __src1 + MOVQ %6, __src1 + %elif %3 + MOVAPS %6, __src1 %else - MOVDQA %5, __src1 + MOVDQA %6, __src1 %endif %endif - %if %0 >= 8 - %1 %5, __src2, %8 + %if %0 >= 9 + %1 %6, __src2, %9 %else - %1 %5, __src2 + %1 %6, __src2 %endif - %elif %0 >= 8 - __instr %5, %6, %7, %8 + %elif %0 >= 9 + __instr %6, %7, %8, %9 + %elif %0 == 8 + __instr %6, %7, %8 %elif %0 == 7 - __instr %5, %6, %7 - %elif %0 == 6 - __instr %5, %6 + __instr %6, %7 %else - __instr %5 + __instr %6 %endif %endmacro ;%1 == instruction -;%2 == 1 if float, 0 if int -;%3 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise -;%4 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not -%macro AVX_INSTR 1-4 0, 1, 0 - %macro %1 1-9 fnord, fnord, fnord, fnord, %1, %2, %3, %4 +;%2 == minimal instruction set +;%3 == 1 if float, 0 if int +;%4 == 1 if non-destructive or 4-operand (xmm, xmm, xmm, imm), 0 otherwise +;%5 == 1 if commutative (i.e. doesn't matter which src arg is which), 0 if not +%macro AVX_INSTR 1-5 fnord, 0, 1, 0 + %macro %1 1-10 fnord, fnord, fnord, fnord, %1, %2, %3, %4, %5 %ifidn %2, fnord - RUN_AVX_INSTR %6, %7, %8, %9, %1 + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1 %elifidn %3, fnord - RUN_AVX_INSTR %6, %7, %8, %9, %1, %2 + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2 %elifidn %4, fnord - RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3 + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3 %elifidn %5, fnord - RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3, %4 + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3, %4 %else - RUN_AVX_INSTR %6, %7, %8, %9, %1, %2, %3, %4, %5 + RUN_AVX_INSTR %6, %7, %8, %9, %10, %1, %2, %3, %4, %5 %endif %endmacro %endmacro ; Instructions with both VEX and non-VEX encodings ; Non-destructive instructions are written without parameters -AVX_INSTR addpd, 1, 0, 1 -AVX_INSTR addps, 1, 0, 1 -AVX_INSTR addsd, 1, 0, 1 -AVX_INSTR addss, 1, 0, 1 -AVX_INSTR addsubpd, 1, 0, 0 -AVX_INSTR addsubps, 1, 0, 0 -AVX_INSTR aesdec, 0, 0, 0 -AVX_INSTR aesdeclast, 0, 0, 0 -AVX_INSTR aesenc, 0, 0, 0 -AVX_INSTR aesenclast, 0, 0, 0 +AVX_INSTR addpd, sse2, 1, 0, 1 +AVX_INSTR addps, sse, 1, 0, 1 +AVX_INSTR addsd, sse2, 1, 0, 1 +AVX_INSTR addss, sse, 1, 0, 1 +AVX_INSTR addsubpd, sse3, 1, 0, 0 +AVX_INSTR addsubps, sse3, 1, 0, 0 +AVX_INSTR aesdec, fnord, 0, 0, 0 +AVX_INSTR aesdeclast, fnord, 0, 0, 0 +AVX_INSTR aesenc, fnord, 0, 0, 0 +AVX_INSTR aesenclast, fnord, 0, 0, 0 AVX_INSTR aesimc AVX_INSTR aeskeygenassist -AVX_INSTR andnpd, 1, 0, 0 -AVX_INSTR andnps, 1, 0, 0 -AVX_INSTR andpd, 1, 0, 1 -AVX_INSTR andps, 1, 0, 1 -AVX_INSTR blendpd, 1, 0, 0 -AVX_INSTR blendps, 1, 0, 0 -AVX_INSTR blendvpd, 1, 0, 0 -AVX_INSTR blendvps, 1, 0, 0 -AVX_INSTR cmppd, 1, 1, 0 -AVX_INSTR cmpps, 1, 1, 0 -AVX_INSTR cmpsd, 1, 1, 0 -AVX_INSTR cmpss, 1, 1, 0 -AVX_INSTR comisd -AVX_INSTR comiss -AVX_INSTR cvtdq2pd -AVX_INSTR cvtdq2ps -AVX_INSTR cvtpd2dq -AVX_INSTR cvtpd2ps -AVX_INSTR cvtps2dq -AVX_INSTR cvtps2pd -AVX_INSTR cvtsd2si -AVX_INSTR cvtsd2ss -AVX_INSTR cvtsi2sd -AVX_INSTR cvtsi2ss -AVX_INSTR cvtss2sd -AVX_INSTR cvtss2si -AVX_INSTR cvttpd2dq -AVX_INSTR cvttps2dq -AVX_INSTR cvttsd2si -AVX_INSTR cvttss2si -AVX_INSTR divpd, 1, 0, 0 -AVX_INSTR divps, 1, 0, 0 -AVX_INSTR divsd, 1, 0, 0 -AVX_INSTR divss, 1, 0, 0 -AVX_INSTR dppd, 1, 1, 0 -AVX_INSTR dpps, 1, 1, 0 -AVX_INSTR extractps -AVX_INSTR haddpd, 1, 0, 0 -AVX_INSTR haddps, 1, 0, 0 -AVX_INSTR hsubpd, 1, 0, 0 -AVX_INSTR hsubps, 1, 0, 0 -AVX_INSTR insertps, 1, 1, 0 -AVX_INSTR lddqu -AVX_INSTR ldmxcsr -AVX_INSTR maskmovdqu -AVX_INSTR maxpd, 1, 0, 1 -AVX_INSTR maxps, 1, 0, 1 -AVX_INSTR maxsd, 1, 0, 1 -AVX_INSTR maxss, 1, 0, 1 -AVX_INSTR minpd, 1, 0, 1 -AVX_INSTR minps, 1, 0, 1 -AVX_INSTR minsd, 1, 0, 1 -AVX_INSTR minss, 1, 0, 1 -AVX_INSTR movapd -AVX_INSTR movaps -AVX_INSTR movd -AVX_INSTR movddup -AVX_INSTR movdqa -AVX_INSTR movdqu -AVX_INSTR movhlps, 1, 0, 0 -AVX_INSTR movhpd, 1, 0, 0 -AVX_INSTR movhps, 1, 0, 0 -AVX_INSTR movlhps, 1, 0, 0 -AVX_INSTR movlpd, 1, 0, 0 -AVX_INSTR movlps, 1, 0, 0 -AVX_INSTR movmskpd -AVX_INSTR movmskps -AVX_INSTR movntdq -AVX_INSTR movntdqa -AVX_INSTR movntpd -AVX_INSTR movntps -AVX_INSTR movq -AVX_INSTR movsd, 1, 0, 0 -AVX_INSTR movshdup -AVX_INSTR movsldup -AVX_INSTR movss, 1, 0, 0 -AVX_INSTR movupd -AVX_INSTR movups -AVX_INSTR mpsadbw, 0, 1, 0 -AVX_INSTR mulpd, 1, 0, 1 -AVX_INSTR mulps, 1, 0, 1 -AVX_INSTR mulsd, 1, 0, 1 -AVX_INSTR mulss, 1, 0, 1 -AVX_INSTR orpd, 1, 0, 1 -AVX_INSTR orps, 1, 0, 1 -AVX_INSTR pabsb -AVX_INSTR pabsd -AVX_INSTR pabsw -AVX_INSTR packsswb, 0, 0, 0 -AVX_INSTR packssdw, 0, 0, 0 -AVX_INSTR packuswb, 0, 0, 0 -AVX_INSTR packusdw, 0, 0, 0 -AVX_INSTR paddb, 0, 0, 1 -AVX_INSTR paddw, 0, 0, 1 -AVX_INSTR paddd, 0, 0, 1 -AVX_INSTR paddq, 0, 0, 1 -AVX_INSTR paddsb, 0, 0, 1 -AVX_INSTR paddsw, 0, 0, 1 -AVX_INSTR paddusb, 0, 0, 1 -AVX_INSTR paddusw, 0, 0, 1 -AVX_INSTR palignr, 0, 1, 0 -AVX_INSTR pand, 0, 0, 1 -AVX_INSTR pandn, 0, 0, 0 -AVX_INSTR pavgb, 0, 0, 1 -AVX_INSTR pavgw, 0, 0, 1 -AVX_INSTR pblendvb, 0, 0, 0 -AVX_INSTR pblendw, 0, 1, 0 -AVX_INSTR pclmulqdq, 0, 1, 0 -AVX_INSTR pcmpestri -AVX_INSTR pcmpestrm -AVX_INSTR pcmpistri -AVX_INSTR pcmpistrm -AVX_INSTR pcmpeqb, 0, 0, 1 -AVX_INSTR pcmpeqw, 0, 0, 1 -AVX_INSTR pcmpeqd, 0, 0, 1 -AVX_INSTR pcmpeqq, 0, 0, 1 -AVX_INSTR pcmpgtb, 0, 0, 0 -AVX_INSTR pcmpgtw, 0, 0, 0 -AVX_INSTR pcmpgtd, 0, 0, 0 -AVX_INSTR pcmpgtq, 0, 0, 0 -AVX_INSTR pextrb -AVX_INSTR pextrd -AVX_INSTR pextrq -AVX_INSTR pextrw -AVX_INSTR phaddw, 0, 0, 0 -AVX_INSTR phaddd, 0, 0, 0 -AVX_INSTR phaddsw, 0, 0, 0 -AVX_INSTR phminposuw -AVX_INSTR phsubw, 0, 0, 0 -AVX_INSTR phsubd, 0, 0, 0 -AVX_INSTR phsubsw, 0, 0, 0 -AVX_INSTR pinsrb, 0, 1, 0 -AVX_INSTR pinsrd, 0, 1, 0 -AVX_INSTR pinsrq, 0, 1, 0 -AVX_INSTR pinsrw, 0, 1, 0 -AVX_INSTR pmaddwd, 0, 0, 1 -AVX_INSTR pmaddubsw, 0, 0, 0 -AVX_INSTR pmaxsb, 0, 0, 1 -AVX_INSTR pmaxsw, 0, 0, 1 -AVX_INSTR pmaxsd, 0, 0, 1 -AVX_INSTR pmaxub, 0, 0, 1 -AVX_INSTR pmaxuw, 0, 0, 1 -AVX_INSTR pmaxud, 0, 0, 1 -AVX_INSTR pminsb, 0, 0, 1 -AVX_INSTR pminsw, 0, 0, 1 -AVX_INSTR pminsd, 0, 0, 1 -AVX_INSTR pminub, 0, 0, 1 -AVX_INSTR pminuw, 0, 0, 1 -AVX_INSTR pminud, 0, 0, 1 -AVX_INSTR pmovmskb -AVX_INSTR pmovsxbw -AVX_INSTR pmovsxbd -AVX_INSTR pmovsxbq -AVX_INSTR pmovsxwd -AVX_INSTR pmovsxwq -AVX_INSTR pmovsxdq -AVX_INSTR pmovzxbw -AVX_INSTR pmovzxbd -AVX_INSTR pmovzxbq -AVX_INSTR pmovzxwd -AVX_INSTR pmovzxwq -AVX_INSTR pmovzxdq -AVX_INSTR pmuldq, 0, 0, 1 -AVX_INSTR pmulhrsw, 0, 0, 1 -AVX_INSTR pmulhuw, 0, 0, 1 -AVX_INSTR pmulhw, 0, 0, 1 -AVX_INSTR pmullw, 0, 0, 1 -AVX_INSTR pmulld, 0, 0, 1 -AVX_INSTR pmuludq, 0, 0, 1 -AVX_INSTR por, 0, 0, 1 -AVX_INSTR psadbw, 0, 0, 1 -AVX_INSTR pshufb, 0, 0, 0 -AVX_INSTR pshufd -AVX_INSTR pshufhw -AVX_INSTR pshuflw -AVX_INSTR psignb, 0, 0, 0 -AVX_INSTR psignw, 0, 0, 0 -AVX_INSTR psignd, 0, 0, 0 -AVX_INSTR psllw, 0, 0, 0 -AVX_INSTR pslld, 0, 0, 0 -AVX_INSTR psllq, 0, 0, 0 -AVX_INSTR pslldq, 0, 0, 0 -AVX_INSTR psraw, 0, 0, 0 -AVX_INSTR psrad, 0, 0, 0 -AVX_INSTR psrlw, 0, 0, 0 -AVX_INSTR psrld, 0, 0, 0 -AVX_INSTR psrlq, 0, 0, 0 -AVX_INSTR psrldq, 0, 0, 0 -AVX_INSTR psubb, 0, 0, 0 -AVX_INSTR psubw, 0, 0, 0 -AVX_INSTR psubd, 0, 0, 0 -AVX_INSTR psubq, 0, 0, 0 -AVX_INSTR psubsb, 0, 0, 0 -AVX_INSTR psubsw, 0, 0, 0 -AVX_INSTR psubusb, 0, 0, 0 -AVX_INSTR psubusw, 0, 0, 0 -AVX_INSTR ptest -AVX_INSTR punpckhbw, 0, 0, 0 -AVX_INSTR punpckhwd, 0, 0, 0 -AVX_INSTR punpckhdq, 0, 0, 0 -AVX_INSTR punpckhqdq, 0, 0, 0 -AVX_INSTR punpcklbw, 0, 0, 0 -AVX_INSTR punpcklwd, 0, 0, 0 -AVX_INSTR punpckldq, 0, 0, 0 -AVX_INSTR punpcklqdq, 0, 0, 0 -AVX_INSTR pxor, 0, 0, 1 -AVX_INSTR rcpps, 1, 0, 0 -AVX_INSTR rcpss, 1, 0, 0 -AVX_INSTR roundpd -AVX_INSTR roundps -AVX_INSTR roundsd -AVX_INSTR roundss -AVX_INSTR rsqrtps, 1, 0, 0 -AVX_INSTR rsqrtss, 1, 0, 0 -AVX_INSTR shufpd, 1, 1, 0 -AVX_INSTR shufps, 1, 1, 0 -AVX_INSTR sqrtpd, 1, 0, 0 -AVX_INSTR sqrtps, 1, 0, 0 -AVX_INSTR sqrtsd, 1, 0, 0 -AVX_INSTR sqrtss, 1, 0, 0 -AVX_INSTR stmxcsr -AVX_INSTR subpd, 1, 0, 0 -AVX_INSTR subps, 1, 0, 0 -AVX_INSTR subsd, 1, 0, 0 -AVX_INSTR subss, 1, 0, 0 -AVX_INSTR ucomisd -AVX_INSTR ucomiss -AVX_INSTR unpckhpd, 1, 0, 0 -AVX_INSTR unpckhps, 1, 0, 0 -AVX_INSTR unpcklpd, 1, 0, 0 -AVX_INSTR unpcklps, 1, 0, 0 -AVX_INSTR xorpd, 1, 0, 1 -AVX_INSTR xorps, 1, 0, 1 +AVX_INSTR andnpd, sse2, 1, 0, 0 +AVX_INSTR andnps, sse, 1, 0, 0 +AVX_INSTR andpd, sse2, 1, 0, 1 +AVX_INSTR andps, sse, 1, 0, 1 +AVX_INSTR blendpd, sse4, 1, 0, 0 +AVX_INSTR blendps, sse4, 1, 0, 0 +AVX_INSTR blendvpd, sse4, 1, 0, 0 +AVX_INSTR blendvps, sse4, 1, 0, 0 +AVX_INSTR cmppd, sse2, 1, 1, 0 +AVX_INSTR cmpps, sse, 1, 1, 0 +AVX_INSTR cmpsd, sse2, 1, 1, 0 +AVX_INSTR cmpss, sse, 1, 1, 0 +AVX_INSTR comisd, sse2 +AVX_INSTR comiss, sse +AVX_INSTR cvtdq2pd, sse2 +AVX_INSTR cvtdq2ps, sse2 +AVX_INSTR cvtpd2dq, sse2 +AVX_INSTR cvtpd2ps, sse2 +AVX_INSTR cvtps2dq, sse2 +AVX_INSTR cvtps2pd, sse2 +AVX_INSTR cvtsd2si, sse2 +AVX_INSTR cvtsd2ss, sse2 +AVX_INSTR cvtsi2sd, sse2 +AVX_INSTR cvtsi2ss, sse +AVX_INSTR cvtss2sd, sse2 +AVX_INSTR cvtss2si, sse +AVX_INSTR cvttpd2dq, sse2 +AVX_INSTR cvttps2dq, sse2 +AVX_INSTR cvttsd2si, sse2 +AVX_INSTR cvttss2si, sse +AVX_INSTR divpd, sse2, 1, 0, 0 +AVX_INSTR divps, sse, 1, 0, 0 +AVX_INSTR divsd, sse2, 1, 0, 0 +AVX_INSTR divss, sse, 1, 0, 0 +AVX_INSTR dppd, sse4, 1, 1, 0 +AVX_INSTR dpps, sse4, 1, 1, 0 +AVX_INSTR extractps, sse4 +AVX_INSTR haddpd, sse3, 1, 0, 0 +AVX_INSTR haddps, sse3, 1, 0, 0 +AVX_INSTR hsubpd, sse3, 1, 0, 0 +AVX_INSTR hsubps, sse3, 1, 0, 0 +AVX_INSTR insertps, sse4, 1, 1, 0 +AVX_INSTR lddqu, sse3 +AVX_INSTR ldmxcsr, sse +AVX_INSTR maskmovdqu, sse2 +AVX_INSTR maxpd, sse2, 1, 0, 1 +AVX_INSTR maxps, sse, 1, 0, 1 +AVX_INSTR maxsd, sse2, 1, 0, 1 +AVX_INSTR maxss, sse, 1, 0, 1 +AVX_INSTR minpd, sse2, 1, 0, 1 +AVX_INSTR minps, sse, 1, 0, 1 +AVX_INSTR minsd, sse2, 1, 0, 1 +AVX_INSTR minss, sse, 1, 0, 1 +AVX_INSTR movapd, sse2 +AVX_INSTR movaps, sse +AVX_INSTR movd, mmx +AVX_INSTR movddup, sse3 +AVX_INSTR movdqa, sse2 +AVX_INSTR movdqu, sse2 +AVX_INSTR movhlps, sse, 1, 0, 0 +AVX_INSTR movhpd, sse2, 1, 0, 0 +AVX_INSTR movhps, sse, 1, 0, 0 +AVX_INSTR movlhps, sse, 1, 0, 0 +AVX_INSTR movlpd, sse2, 1, 0, 0 +AVX_INSTR movlps, sse, 1, 0, 0 +AVX_INSTR movmskpd, sse2 +AVX_INSTR movmskps, sse +AVX_INSTR movntdq, sse2 +AVX_INSTR movntdqa, sse4 +AVX_INSTR movntpd, sse2 +AVX_INSTR movntps, sse +AVX_INSTR movq, mmx +AVX_INSTR movsd, sse2, 1, 0, 0 +AVX_INSTR movshdup, sse3 +AVX_INSTR movsldup, sse3 +AVX_INSTR movss, sse, 1, 0, 0 +AVX_INSTR movupd, sse2 +AVX_INSTR movups, sse +AVX_INSTR mpsadbw, sse4 +AVX_INSTR mulpd, sse2, 1, 0, 1 +AVX_INSTR mulps, sse, 1, 0, 1 +AVX_INSTR mulsd, sse2, 1, 0, 1 +AVX_INSTR mulss, sse, 1, 0, 1 +AVX_INSTR orpd, sse2, 1, 0, 1 +AVX_INSTR orps, sse, 1, 0, 1 +AVX_INSTR pabsb, ssse3 +AVX_INSTR pabsd, ssse3 +AVX_INSTR pabsw, ssse3 +AVX_INSTR packsswb, mmx, 0, 0, 0 +AVX_INSTR packssdw, mmx, 0, 0, 0 +AVX_INSTR packuswb, mmx, 0, 0, 0 +AVX_INSTR packusdw, sse4, 0, 0, 0 +AVX_INSTR paddb, mmx, 0, 0, 1 +AVX_INSTR paddw, mmx, 0, 0, 1 +AVX_INSTR paddd, mmx, 0, 0, 1 +AVX_INSTR paddq, sse2, 0, 0, 1 +AVX_INSTR paddsb, mmx, 0, 0, 1 +AVX_INSTR paddsw, mmx, 0, 0, 1 +AVX_INSTR paddusb, mmx, 0, 0, 1 +AVX_INSTR paddusw, mmx, 0, 0, 1 +AVX_INSTR palignr, ssse3 +AVX_INSTR pand, mmx, 0, 0, 1 +AVX_INSTR pandn, mmx, 0, 0, 0 +AVX_INSTR pavgb, mmx2, 0, 0, 1 +AVX_INSTR pavgw, mmx2, 0, 0, 1 +AVX_INSTR pblendvb, sse4, 0, 0, 0 +AVX_INSTR pblendw, sse4 +AVX_INSTR pclmulqdq +AVX_INSTR pcmpestri, sse42 +AVX_INSTR pcmpestrm, sse42 +AVX_INSTR pcmpistri, sse42 +AVX_INSTR pcmpistrm, sse42 +AVX_INSTR pcmpeqb, mmx, 0, 0, 1 +AVX_INSTR pcmpeqw, mmx, 0, 0, 1 +AVX_INSTR pcmpeqd, mmx, 0, 0, 1 +AVX_INSTR pcmpeqq, sse4, 0, 0, 1 +AVX_INSTR pcmpgtb, mmx, 0, 0, 0 +AVX_INSTR pcmpgtw, mmx, 0, 0, 0 +AVX_INSTR pcmpgtd, mmx, 0, 0, 0 +AVX_INSTR pcmpgtq, sse42, 0, 0, 0 +AVX_INSTR pextrb, sse4 +AVX_INSTR pextrd, sse4 +AVX_INSTR pextrq, sse4 +AVX_INSTR pextrw, mmx2 +AVX_INSTR phaddw, ssse3, 0, 0, 0 +AVX_INSTR phaddd, ssse3, 0, 0, 0 +AVX_INSTR phaddsw, ssse3, 0, 0, 0 +AVX_INSTR phminposuw, sse4 +AVX_INSTR phsubw, ssse3, 0, 0, 0 +AVX_INSTR phsubd, ssse3, 0, 0, 0 +AVX_INSTR phsubsw, ssse3, 0, 0, 0 +AVX_INSTR pinsrb, sse4 +AVX_INSTR pinsrd, sse4 +AVX_INSTR pinsrq, sse4 +AVX_INSTR pinsrw, mmx2 +AVX_INSTR pmaddwd, mmx, 0, 0, 1 +AVX_INSTR pmaddubsw, ssse3, 0, 0, 0 +AVX_INSTR pmaxsb, sse4, 0, 0, 1 +AVX_INSTR pmaxsw, mmx2, 0, 0, 1 +AVX_INSTR pmaxsd, sse4, 0, 0, 1 +AVX_INSTR pmaxub, mmx2, 0, 0, 1 +AVX_INSTR pmaxuw, sse4, 0, 0, 1 +AVX_INSTR pmaxud, sse4, 0, 0, 1 +AVX_INSTR pminsb, sse4, 0, 0, 1 +AVX_INSTR pminsw, mmx2, 0, 0, 1 +AVX_INSTR pminsd, sse4, 0, 0, 1 +AVX_INSTR pminub, mmx2, 0, 0, 1 +AVX_INSTR pminuw, sse4, 0, 0, 1 +AVX_INSTR pminud, sse4, 0, 0, 1 +AVX_INSTR pmovmskb, mmx2 +AVX_INSTR pmovsxbw, sse4 +AVX_INSTR pmovsxbd, sse4 +AVX_INSTR pmovsxbq, sse4 +AVX_INSTR pmovsxwd, sse4 +AVX_INSTR pmovsxwq, sse4 +AVX_INSTR pmovsxdq, sse4 +AVX_INSTR pmovzxbw, sse4 +AVX_INSTR pmovzxbd, sse4 +AVX_INSTR pmovzxbq, sse4 +AVX_INSTR pmovzxwd, sse4 +AVX_INSTR pmovzxwq, sse4 +AVX_INSTR pmovzxdq, sse4 +AVX_INSTR pmuldq, sse4, 0, 0, 1 +AVX_INSTR pmulhrsw, ssse3, 0, 0, 1 +AVX_INSTR pmulhuw, mmx2, 0, 0, 1 +AVX_INSTR pmulhw, mmx, 0, 0, 1 +AVX_INSTR pmullw, mmx, 0, 0, 1 +AVX_INSTR pmulld, sse4, 0, 0, 1 +AVX_INSTR pmuludq, sse2, 0, 0, 1 +AVX_INSTR por, mmx, 0, 0, 1 +AVX_INSTR psadbw, mmx2, 0, 0, 1 +AVX_INSTR pshufb, ssse3, 0, 0, 0 +AVX_INSTR pshufd, sse2 +AVX_INSTR pshufhw, sse2 +AVX_INSTR pshuflw, sse2 +AVX_INSTR psignb, ssse3, 0, 0, 0 +AVX_INSTR psignw, ssse3, 0, 0, 0 +AVX_INSTR psignd, ssse3, 0, 0, 0 +AVX_INSTR psllw, mmx, 0, 0, 0 +AVX_INSTR pslld, mmx, 0, 0, 0 +AVX_INSTR psllq, mmx, 0, 0, 0 +AVX_INSTR pslldq, sse2, 0, 0, 0 +AVX_INSTR psraw, mmx, 0, 0, 0 +AVX_INSTR psrad, mmx, 0, 0, 0 +AVX_INSTR psrlw, mmx, 0, 0, 0 +AVX_INSTR psrld, mmx, 0, 0, 0 +AVX_INSTR psrlq, mmx, 0, 0, 0 +AVX_INSTR psrldq, sse2, 0, 0, 0 +AVX_INSTR psubb, mmx, 0, 0, 0 +AVX_INSTR psubw, mmx, 0, 0, 0 +AVX_INSTR psubd, mmx, 0, 0, 0 +AVX_INSTR psubq, sse2, 0, 0, 0 +AVX_INSTR psubsb, mmx, 0, 0, 0 +AVX_INSTR psubsw, mmx, 0, 0, 0 +AVX_INSTR psubusb, mmx, 0, 0, 0 +AVX_INSTR psubusw, mmx, 0, 0, 0 +AVX_INSTR ptest, sse4 +AVX_INSTR punpckhbw, mmx, 0, 0, 0 +AVX_INSTR punpckhwd, mmx, 0, 0, 0 +AVX_INSTR punpckhdq, mmx, 0, 0, 0 +AVX_INSTR punpckhqdq, sse2, 0, 0, 0 +AVX_INSTR punpcklbw, mmx, 0, 0, 0 +AVX_INSTR punpcklwd, mmx, 0, 0, 0 +AVX_INSTR punpckldq, mmx, 0, 0, 0 +AVX_INSTR punpcklqdq, sse2, 0, 0, 0 +AVX_INSTR pxor, mmx, 0, 0, 1 +AVX_INSTR rcpps, sse, 1, 0, 0 +AVX_INSTR rcpss, sse, 1, 0, 0 +AVX_INSTR roundpd, sse4 +AVX_INSTR roundps, sse4 +AVX_INSTR roundsd, sse4 +AVX_INSTR roundss, sse4 +AVX_INSTR rsqrtps, sse, 1, 0, 0 +AVX_INSTR rsqrtss, sse, 1, 0, 0 +AVX_INSTR shufpd, sse2, 1, 1, 0 +AVX_INSTR shufps, sse, 1, 1, 0 +AVX_INSTR sqrtpd, sse2, 1, 0, 0 +AVX_INSTR sqrtps, sse, 1, 0, 0 +AVX_INSTR sqrtsd, sse2, 1, 0, 0 +AVX_INSTR sqrtss, sse, 1, 0, 0 +AVX_INSTR stmxcsr, sse +AVX_INSTR subpd, sse2, 1, 0, 0 +AVX_INSTR subps, sse, 1, 0, 0 +AVX_INSTR subsd, sse2, 1, 0, 0 +AVX_INSTR subss, sse, 1, 0, 0 +AVX_INSTR ucomisd, sse2 +AVX_INSTR ucomiss, sse +AVX_INSTR unpckhpd, sse2, 1, 0, 0 +AVX_INSTR unpckhps, sse, 1, 0, 0 +AVX_INSTR unpcklpd, sse2, 1, 0, 0 +AVX_INSTR unpcklps, sse, 1, 0, 0 +AVX_INSTR xorpd, sse2, 1, 0, 1 +AVX_INSTR xorps, sse, 1, 0, 1 ; 3DNow instructions, for sharing code between AVX, SSE and 3DN -AVX_INSTR pfadd, 1, 0, 1 -AVX_INSTR pfsub, 1, 0, 0 -AVX_INSTR pfmul, 1, 0, 1 +AVX_INSTR pfadd, 3dnow, 1, 0, 1 +AVX_INSTR pfsub, 3dnow, 1, 0, 0 +AVX_INSTR pfmul, 3dnow, 1, 0, 1 ; base-4 constants for shuffles %assign i 0 @@ -1402,21 +1413,6 @@ AVX_INSTR pfmul, 1, 0, 1 %undef i %undef j -%macro FMA_INSTR 3 - %macro %1 4-7 %1, %2, %3 - %if cpuflag(xop) - v%5 %1, %2, %3, %4 - %else - %6 %1, %2, %3 - %7 %1, %4 - %endif - %endmacro -%endmacro - -FMA_INSTR pmacsdd, pmulld, paddd -FMA_INSTR pmacsww, pmullw, paddw -FMA_INSTR pmadcswd, pmaddwd, paddd - ; tzcnt is equivalent to "rep bsf" and is backwards-compatible with bsf. ; This lets us use tzcnt without bumping the yasm version requirement yet. %define tzcnt rep bsf diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index 9f64dd13e1..d6702c1466 100644 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@ -6,20 +6,20 @@ ;* Authors: Loren Merritt <lorenm@u.washington.edu> ;* Holger Lubitz <holger@lubitz.org> ;* -;* This file is part of Libav. +;* This file is part of FFmpeg. ;* -;* Libav is free software; you can redistribute it and/or +;* 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. ;* -;* Libav is distributed in the hope that it will be useful, +;* 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 Libav; if not, write to the Free Software +;* License along with FFmpeg; if not, write to the Free Software ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA ;****************************************************************************** @@ -69,6 +69,15 @@ SWAP %2, %3 %endmacro +%macro TRANSPOSE2x4x4B 5 + SBUTTERFLY bw, %1, %2, %5 + SBUTTERFLY bw, %3, %4, %5 + SBUTTERFLY wd, %1, %3, %5 + SBUTTERFLY wd, %2, %4, %5 + SBUTTERFLY dq, %1, %2, %5 + SBUTTERFLY dq, %3, %4, %5 +%endmacro + %macro TRANSPOSE2x4x4W 5 SBUTTERFLY wd, %1, %2, %5 SBUTTERFLY wd, %3, %4, %5 @@ -98,6 +107,43 @@ SWAP %5, %2, %3 %endmacro +%macro TRANSPOSE8x4D 9-11 +%if ARCH_X86_64 + SBUTTERFLY dq, %1, %2, %9 + SBUTTERFLY dq, %3, %4, %9 + SBUTTERFLY dq, %5, %6, %9 + SBUTTERFLY dq, %7, %8, %9 + SBUTTERFLY qdq, %1, %3, %9 + SBUTTERFLY qdq, %2, %4, %9 + SBUTTERFLY qdq, %5, %7, %9 + SBUTTERFLY qdq, %6, %8, %9 + SWAP %2, %5 + SWAP %4, %7 +%else +; in: m0..m7 +; out: m0..m7, unless %11 in which case m2 is in %9 +; spills into %9 and %10 + movdqa %9, m%7 + SBUTTERFLY dq, %1, %2, %7 + movdqa %10, m%2 + movdqa m%7, %9 + SBUTTERFLY dq, %3, %4, %2 + SBUTTERFLY dq, %5, %6, %2 + SBUTTERFLY dq, %7, %8, %2 + SBUTTERFLY qdq, %1, %3, %2 + movdqa %9, m%3 + movdqa m%2, %10 + SBUTTERFLY qdq, %2, %4, %3 + SBUTTERFLY qdq, %5, %7, %3 + SBUTTERFLY qdq, %6, %8, %3 + SWAP %2, %5 + SWAP %4, %7 +%if %0<11 + movdqa m%3, %9 +%endif +%endif +%endmacro + %macro TRANSPOSE8x8W 9-11 %if ARCH_X86_64 SBUTTERFLY wd, %1, %2, %9 @@ -273,6 +319,44 @@ %endif %endmacro +%macro HADDD 2 ; sum junk +%if sizeof%1 == 32 +%define %2 xmm%2 + vextracti128 %2, %1, 1 +%define %1 xmm%1 + paddd %1, %2 +%endif +%if mmsize >= 16 +%if cpuflag(xop) && sizeof%1 == 16 + vphadddq %1, %1 +%endif + movhlps %2, %1 + paddd %1, %2 +%endif +%if notcpuflag(xop) || sizeof%1 != 16 +%if cpuflag(mmxext) + PSHUFLW %2, %1, q0032 +%else ; mmx + mova %2, %1 + psrlq %2, 32 +%endif + paddd %1, %2 +%endif +%undef %1 +%undef %2 +%endmacro + +%macro HADDW 2 ; reg, tmp +%if cpuflag(xop) && sizeof%1 == 16 + vphaddwq %1, %1 + movhlps %2, %1 + paddd %1, %2 +%else + pmaddwd %1, [pw_1] + HADDD %1, %2 +%endif +%endmacro + %macro PALIGNR 4-5 %if cpuflag(ssse3) %if %0==5 @@ -302,11 +386,19 @@ %endif %endmacro -%macro PAVGB 2 +%macro PAVGB 2-4 %if cpuflag(mmxext) pavgb %1, %2 %elif cpuflag(3dnow) pavgusb %1, %2 +%elif cpuflag(mmx) + movu %3, %2 + por %3, %1 + pxor %1, %2 + pand %1, %4 + psrlq %1, 1 + psubb %3, %1 + SWAP %1, %3 %endif %endmacro @@ -552,7 +644,9 @@ %endmacro %macro SPLATW 2-3 0 -%if mmsize == 16 +%if cpuflag(avx2) && %3 == 0 + vpbroadcastw %1, %2 +%elif mmsize == 16 pshuflw %1, %2, (%3)*0x55 punpcklqdq %1, %1 %elif cpuflag(mmxext) @@ -671,6 +765,25 @@ %endif %endmacro +%macro PMA_EMU 4 + %macro %1 5-8 %2, %3, %4 + %if cpuflag(xop) + v%6 %1, %2, %3, %4 + %elifidn %1, %4 + %7 %5, %2, %3 + %8 %1, %4, %5 + %else + %7 %1, %2, %3 + %8 %1, %4 + %endif + %endmacro +%endmacro + +PMA_EMU PMACSWW, pmacsww, pmullw, paddw +PMA_EMU PMACSDD, pmacsdd, pmulld, paddd ; sse4 emulation +PMA_EMU PMACSDQL, pmacsdql, pmuldq, paddq ; sse4 emulation +PMA_EMU PMADCSWD, pmadcswd, pmaddwd, paddd + ; Wrapper for non-FMA version of fmaddps %macro FMULADD_PS 5 %if cpuflag(fma3) || cpuflag(fma4) @@ -683,3 +796,19 @@ addps %1, %4 %endif %endmacro + +%macro LSHIFT 2 +%if mmsize > 8 + pslldq %1, %2 +%else + psllq %1, 8*(%2) +%endif +%endmacro + +%macro RSHIFT 2 +%if mmsize > 8 + psrldq %1, %2 +%else + psrlq %1, 8*(%2) +%endif +%endmacro |