diff options
Diffstat (limited to 'libavutil/x86')
| -rw-r--r-- | libavutil/x86/Makefile | 10 | ||||
| -rw-r--r-- | libavutil/x86/asm.h | 56 | ||||
| -rw-r--r-- | libavutil/x86/bswap.h | 44 | ||||
| -rw-r--r-- | libavutil/x86/cpu.c | 14 | ||||
| -rw-r--r-- | libavutil/x86/cpu.h | 11 | ||||
| -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/fixed_dsp.asm | 48 | ||||
| -rw-r--r-- | libavutil/x86/fixed_dsp_init.c | 35 | ||||
| -rw-r--r-- | libavutil/x86/float_dsp.asm | 167 | ||||
| -rw-r--r-- | libavutil/x86/float_dsp_init.c | 97 | ||||
| -rw-r--r-- | libavutil/x86/intmath.h | 136 | ||||
| -rw-r--r-- | libavutil/x86/intreadwrite.h | 8 | ||||
| -rw-r--r-- | libavutil/x86/lls.asm | 66 | ||||
| -rw-r--r-- | libavutil/x86/lls_init.c | 18 | ||||
| -rw-r--r-- | libavutil/x86/pixelutils.asm | 165 | ||||
| -rw-r--r-- | libavutil/x86/pixelutils.h | 26 | ||||
| -rw-r--r-- | libavutil/x86/pixelutils_init.c | 64 | ||||
| -rw-r--r-- | libavutil/x86/timer.h | 9 | ||||
| -rw-r--r-- | libavutil/x86/w64xmmtest.h | 13 | ||||
| -rw-r--r-- | libavutil/x86/x86inc.asm | 273 | ||||
| -rw-r--r-- | libavutil/x86/x86util.asm | 132 |
23 files changed, 1104 insertions, 316 deletions
diff --git a/libavutil/x86/Makefile b/libavutil/x86/Makefile index 1e19082233..94d8832062 100644 --- a/libavutil/x86/Makefile +++ b/libavutil/x86/Makefile @@ -1,8 +1,16 @@ OBJS += x86/cpu.o \ + x86/fixed_dsp_init.o \ x86/float_dsp_init.o \ x86/lls_init.o \ +OBJS-$(CONFIG_PIXELUTILS) += x86/pixelutils_init.o \ + +EMMS_OBJS_$(HAVE_MMX_INLINE)_$(HAVE_MMX_EXTERNAL)_$(HAVE_MM_EMPTY) = x86/emms.o + YASM-OBJS += x86/cpuid.o \ - x86/emms.o \ + $(EMMS_OBJS__yes_) \ + x86/fixed_dsp.o \ x86/float_dsp.o \ x86/lls.o \ + +YASM-OBJS-$(CONFIG_PIXELUTILS) += x86/pixelutils.o \ diff --git a/libavutil/x86/asm.h b/libavutil/x86/asm.h index e30f5dbaf7..109b65e542 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" @@ -37,7 +38,8 @@ typedef struct xmm_reg { uint64_t a, b; } xmm_reg; # define PTR_SIZE "8" typedef int64_t x86_reg; -# define REG_SP "rsp" +/* REG_SP is defined in Solaris sys headers, so use REG_sp */ +# define REG_sp "rsp" # define REG_BP "rbp" # define REGBP rbp # define REGa rax @@ -58,7 +60,7 @@ typedef int64_t x86_reg; # define PTR_SIZE "4" typedef int32_t x86_reg; -# define REG_SP "esp" +# define REG_sp "esp" # define REG_BP "ebp" # define REGBP ebp # define REGa eax @@ -107,6 +109,46 @@ 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..ffa59e4c82 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 */ @@ -25,21 +25,47 @@ #define AVUTIL_X86_BSWAP_H #include <stdint.h> +#if defined(_MSC_VER) +#include <intrin.h> +#endif #include "config.h" #include "libavutil/attributes.h" -#if HAVE_INLINE_ASM +#if defined(_MSC_VER) + +#define av_bswap16 av_bswap16 +static av_always_inline av_const uint16_t av_bswap16(uint16_t x) +{ + return _rotr16(x, 8); +} + +#define av_bswap32 av_bswap32 +static av_always_inline av_const uint32_t av_bswap32(uint32_t x) +{ + return _byteswap_ulong(x); +} + +#if ARCH_X86_64 +#define av_bswap64 av_bswap64 +static inline uint64_t av_const av_bswap64(uint64_t x) +{ + return _byteswap_uint64(x); +} +#endif + + +#elif HAVE_INLINE_ASM -#if !AV_GCC_VERSION_AT_LEAST(4,1) +#if AV_GCC_VERSION_AT_MOST(4,0) #define av_bswap16 av_bswap16 static av_always_inline av_const unsigned av_bswap16(unsigned x) { __asm__("rorw $8, %w0" : "+r"(x)); return x; } -#endif /* !AV_GCC_VERSION_AT_LEAST(4,1) */ +#endif /* AV_GCC_VERSION_AT_MOST(4,0) */ -#if !AV_GCC_VERSION_AT_LEAST(4,5) +#if AV_GCC_VERSION_AT_MOST(4,4) || defined(__INTEL_COMPILER) #define av_bswap32 av_bswap32 static av_always_inline av_const uint32_t av_bswap32(uint32_t x) { @@ -55,7 +81,7 @@ static inline uint64_t av_const av_bswap64(uint64_t x) return x; } #endif -#endif /* !AV_GCC_VERSION_AT_LEAST(4,5) */ +#endif /* AV_GCC_VERSION_AT_MOST(4,4) */ #endif /* HAVE_INLINE_ASM */ #endif /* AVUTIL_X86_BSWAP_H */ diff --git a/libavutil/x86/cpu.c b/libavutil/x86/cpu.c index 098ccf7004..f57d72d0a7 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)) @@ -126,6 +126,8 @@ int ff_get_cpu_flags_x86(void) rval |= AV_CPU_FLAG_SSE4; if (ecx & 0x00100000 ) rval |= AV_CPU_FLAG_SSE42; + if (ecx & 0x01000000 ) + rval |= AV_CPU_FLAG_AESNI; #if HAVE_AVX /* Check OXSAVE and AVX bits */ if ((ecx & 0x18000000) == 0x18000000) { @@ -143,7 +145,7 @@ 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 0695436548..dc102c6015 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 */ @@ -47,6 +47,7 @@ #define X86_FMA3(flags) CPUEXT(flags, FMA3) #define X86_FMA4(flags) CPUEXT(flags, FMA4) #define X86_AVX2(flags) CPUEXT(flags, AVX2) +#define X86_AESNI(flags) CPUEXT(flags, AESNI) #define EXTERNAL_AMD3DNOW(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOW) #define EXTERNAL_AMD3DNOWEXT(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AMD3DNOWEXT) @@ -69,6 +70,7 @@ #define EXTERNAL_FMA3(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, FMA3) #define EXTERNAL_FMA4(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, FMA4) #define EXTERNAL_AVX2(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AVX2) +#define EXTERNAL_AESNI(flags) CPUEXT_SUFFIX(flags, _EXTERNAL, AESNI) #define INLINE_AMD3DNOW(flags) CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOW) #define INLINE_AMD3DNOWEXT(flags) CPUEXT_SUFFIX(flags, _INLINE, AMD3DNOWEXT) @@ -91,6 +93,7 @@ #define INLINE_FMA3(flags) CPUEXT_SUFFIX(flags, _INLINE, FMA3) #define INLINE_FMA4(flags) CPUEXT_SUFFIX(flags, _INLINE, FMA4) #define INLINE_AVX2(flags) CPUEXT_SUFFIX(flags, _INLINE, AVX2) +#define INLINE_AESNI(flags) CPUEXT_SUFFIX(flags, _INLINE, AESNI) void ff_cpu_cpuid(int index, int *eax, int *ebx, int *ecx, int *edx); void ff_cpu_xgetbv(int op, int *eax, int *edx); 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/fixed_dsp.asm b/libavutil/x86/fixed_dsp.asm new file mode 100644 index 0000000000..979dd5c334 --- /dev/null +++ b/libavutil/x86/fixed_dsp.asm @@ -0,0 +1,48 @@ +;***************************************************************************** +;* x86-optimized Float DSP functions +;* +;* Copyright 2016 James Almer +;* +;* 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 + +;----------------------------------------------------------------------------- +; void ff_butterflies_fixed(float *src0, float *src1, int len); +;----------------------------------------------------------------------------- +INIT_XMM sse2 +cglobal butterflies_fixed, 3,3,3, src0, src1, len + shl lend, 2 + add src0q, lenq + add src1q, lenq + neg lenq + +align 16 +.loop: + mova m0, [src0q + lenq] + mova m1, [src1q + lenq] + mova m2, m0 + paddd m0, m1 + psubd m2, m1 + mova [src0q + lenq], m0 + mova [src1q + lenq], m2 + add lenq, mmsize + jl .loop + RET diff --git a/libavutil/x86/fixed_dsp_init.c b/libavutil/x86/fixed_dsp_init.c new file mode 100644 index 0000000000..303a2eb922 --- /dev/null +++ b/libavutil/x86/fixed_dsp_init.c @@ -0,0 +1,35 @@ +/* + * 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 "libavutil/attributes.h" +#include "libavutil/cpu.h" +#include "libavutil/fixed_dsp.h" +#include "cpu.h" + +void ff_butterflies_fixed_sse2(int *src0, int *src1, int len); + +av_cold void ff_fixed_dsp_init_x86(AVFixedDSPContext *fdsp) +{ + int cpu_flags = av_get_cpu_flags(); + + if (EXTERNAL_SSE2(cpu_flags)) { + fdsp->butterflies_fixed = ff_butterflies_fixed_sse2; + } +} diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm index d96249978a..021ff03c87 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, @@ -233,16 +324,18 @@ 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 cglobal scalarproduct_float, 3,3,2, v1, v2, offset + shl offsetd, 2 + add v1q, offsetq + add v2q, offsetq neg offsetq - shl offsetq, 2 - sub v1q, offsetq - sub v2q, offsetq xorps xmm0, xmm0 .loop: movaps xmm1, [v1q+offsetq] @@ -266,14 +359,9 @@ cglobal scalarproduct_float, 3,3,2, v1, v2, offset ;----------------------------------------------------------------------------- INIT_XMM sse cglobal butterflies_float, 3,3,3, src0, src1, len -%if ARCH_X86_64 - movsxd lenq, lend -%endif - test lenq, lenq - jz .end - shl lenq, 2 - lea src0q, [src0q + lenq] - lea src1q, [src1q + lenq] + shl lend, 2 + add src0q, lenq + add src1q, lenq neg lenq .loop: mova m0, [src0q + lenq] @@ -284,5 +372,4 @@ cglobal butterflies_float, 3,3,3, src0, src1, len mova [src0q + lenq], m0 add lenq, mmsize jl .loop -.end: REP_RET diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c index b70433031a..f211f2396b 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) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) { + fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_fma3; + fdsp->vector_fmul_add = ff_vector_fmul_add_fma3; + } } diff --git a/libavutil/x86/intmath.h b/libavutil/x86/intmath.h new file mode 100644 index 0000000000..f58b0d08da --- /dev/null +++ b/libavutil/x86/intmath.h @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2015 James Almer + * + * 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_INTMATH_H +#define AVUTIL_X86_INTMATH_H + +#include <stdint.h> +#include <stdlib.h> +#if HAVE_FAST_CLZ +#if defined(_MSC_VER) +#include <intrin.h> +#elif defined(__INTEL_COMPILER) +#include <immintrin.h> +#endif +#endif +#include "config.h" + +#if HAVE_FAST_CLZ +#if (defined(__INTEL_COMPILER) && (__INTEL_COMPILER>=1216)) || defined(_MSC_VER) +# if defined(__INTEL_COMPILER) +# define ff_log2(x) (_bit_scan_reverse((x)|1)) +# else +# define ff_log2 ff_log2_x86 +static av_always_inline av_const int ff_log2_x86(unsigned int v) +{ + unsigned long n; + _BitScanReverse(&n, v|1); + return n; +} +# endif +# define ff_log2_16bit av_log2 + +# define ff_ctz(v) _tzcnt_u32(v) + +# if ARCH_X86_64 +# define ff_ctzll(v) _tzcnt_u64(v) +# else +# define ff_ctzll ff_ctzll_x86 +static av_always_inline av_const int ff_ctzll_x86(long long v) +{ + return ((uint32_t)v == 0) ? _tzcnt_u32((uint32_t)(v >> 32)) + 32 : _tzcnt_u32((uint32_t)v); +} +# endif + +#endif /* __INTEL_COMPILER */ + +#endif /* HAVE_FAST_CLZ */ + +#if defined(__GNUC__) + +/* Our generic version of av_popcount is faster than GCC's built-in on + * CPUs that don't support the popcnt instruction. + */ +#if defined(__POPCNT__) + #define av_popcount __builtin_popcount +#if ARCH_X86_64 + #define av_popcount64 __builtin_popcountll +#endif + +#endif /* __POPCNT__ */ + +#if defined(__BMI2__) + +#if AV_GCC_VERSION_AT_LEAST(5,1) +#define av_mod_uintp2 __builtin_ia32_bzhi_si +#elif HAVE_INLINE_ASM +/* GCC releases before 5.1.0 have a broken bzhi builtin, so for those we + * implement it using inline assembly + */ +#define av_mod_uintp2 av_mod_uintp2_bmi2 +static av_always_inline av_const unsigned av_mod_uintp2_bmi2(unsigned a, unsigned p) +{ + if (av_builtin_constant_p(p)) + return a & ((1 << p) - 1); + else { + unsigned x; + __asm__ ("bzhi %2, %1, %0 \n\t" : "=r"(x) : "rm"(a), "r"(p)); + return x; + } +} +#endif /* AV_GCC_VERSION_AT_LEAST */ + +#endif /* __BMI2__ */ + +#if defined(__SSE2__) && !defined(__INTEL_COMPILER) + +#define av_clipd av_clipd_sse2 +static av_always_inline av_const double av_clipd_sse2(double a, double amin, double amax) +{ +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL >= 2 + if (amin > amax) abort(); +#endif + __asm__ ("minsd %2, %0 \n\t" + "maxsd %1, %0 \n\t" + : "+&x"(a) : "xm"(amin), "xm"(amax)); + return a; +} + +#endif /* __SSE2__ */ + +#if defined(__SSE__) && !defined(__INTEL_COMPILER) + +#define av_clipf av_clipf_sse +static av_always_inline av_const float av_clipf_sse(float a, float amin, float amax) +{ +#if defined(ASSERT_LEVEL) && ASSERT_LEVEL >= 2 + if (amin > amax) abort(); +#endif + __asm__ ("minss %2, %0 \n\t" + "maxss %1, %0 \n\t" + : "+&x"(a) : "xm"(amin), "xm"(amax)); + return a; +} + +#endif /* __SSE__ */ + +#endif /* __GNUC__ */ + +#endif /* AVUTIL_X86_INTMATH_H */ 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..317fba6fca 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 ;****************************************************************************** @@ -125,7 +125,7 @@ cglobal update_lls, 2,5,8, ctx, var, i, j, covar2 .ret: REP_RET -INIT_YMM avx +%macro UPDATE_LLS 0 cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 %define covarq ctxq mov countd, [ctxq + LLSModel.indep_count] @@ -139,6 +139,18 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 vbroadcastsd ymm6, [varq + iq*8 + 16] vbroadcastsd ymm7, [varq + iq*8 + 24] vextractf128 xmm3, ymm1, 1 +%if cpuflag(fma3) + mova ymm0, COVAR(iq ,0) + mova xmm2, COVAR(iq+2,2) + fmaddpd ymm0, ymm1, ymm4, ymm0 + fmaddpd xmm2, xmm3, xmm6, xmm2 + fmaddpd ymm1, ymm5, ymm1, COVAR(iq ,1) + fmaddpd xmm3, xmm7, xmm3, COVAR(iq+2,3) + mova COVAR(iq ,0), ymm0 + mova COVAR(iq ,1), ymm1 + mova COVAR(iq+2,2), xmm2 + mova COVAR(iq+2,3), xmm3 +%else vmulpd ymm0, ymm1, ymm4 vmulpd ymm1, ymm1, ymm5 vmulpd xmm2, xmm3, xmm6 @@ -147,12 +159,26 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 ADDPD_MEM COVAR(iq ,1), ymm1 ADDPD_MEM COVAR(iq+2,2), xmm2 ADDPD_MEM COVAR(iq+2,3), xmm3 +%endif ; cpuflag(fma3) lea jd, [iq + 4] cmp jd, count2d jg .skip4x4 .loop4x4: ; Compute all 16 pairwise products of a 4x4 block mova ymm3, [varq + jq*8] +%if cpuflag(fma3) + mova ymm0, COVAR(jq, 0) + mova ymm1, COVAR(jq, 1) + mova ymm2, COVAR(jq, 2) + fmaddpd ymm0, ymm3, ymm4, ymm0 + fmaddpd ymm1, ymm3, ymm5, ymm1 + fmaddpd ymm2, ymm3, ymm6, ymm2 + fmaddpd ymm3, ymm7, ymm3, COVAR(jq,3) + mova COVAR(jq, 0), ymm0 + mova COVAR(jq, 1), ymm1 + mova COVAR(jq, 2), ymm2 + mova COVAR(jq, 3), ymm3 +%else vmulpd ymm0, ymm3, ymm4 vmulpd ymm1, ymm3, ymm5 vmulpd ymm2, ymm3, ymm6 @@ -161,6 +187,7 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 ADDPD_MEM COVAR(jq,1), ymm1 ADDPD_MEM COVAR(jq,2), ymm2 ADDPD_MEM COVAR(jq,3), ymm3 +%endif ; cpuflag(fma3) add jd, 4 cmp jd, count2d jle .loop4x4 @@ -168,6 +195,19 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 cmp jd, countd jg .skip2x4 mova xmm3, [varq + jq*8] +%if cpuflag(fma3) + mova xmm0, COVAR(jq, 0) + mova xmm1, COVAR(jq, 1) + mova xmm2, COVAR(jq, 2) + fmaddpd xmm0, xmm3, xmm4, xmm0 + fmaddpd xmm1, xmm3, xmm5, xmm1 + fmaddpd xmm2, xmm3, xmm6, xmm2 + fmaddpd xmm3, xmm7, xmm3, COVAR(jq,3) + mova COVAR(jq, 0), xmm0 + mova COVAR(jq, 1), xmm1 + mova COVAR(jq, 2), xmm2 + mova COVAR(jq, 3), xmm3 +%else vmulpd xmm0, xmm3, xmm4 vmulpd xmm1, xmm3, xmm5 vmulpd xmm2, xmm3, xmm6 @@ -176,6 +216,7 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 ADDPD_MEM COVAR(jq,1), xmm1 ADDPD_MEM COVAR(jq,2), xmm2 ADDPD_MEM COVAR(jq,3), xmm3 +%endif ; cpuflag(fma3) .skip2x4: add id, 4 add covarq, 4*COVAR_STRIDE @@ -186,15 +227,30 @@ cglobal update_lls, 3,6,8, ctx, var, count, i, j, count2 mov jd, id .loop2x1: vmovddup xmm0, [varq + iq*8] +%if cpuflag(fma3) + mova xmm1, [varq + jq*8] + fmaddpd xmm0, xmm1, xmm0, COVAR(jq,0) + mova COVAR(jq,0), xmm0 +%else vmulpd xmm0, [varq + jq*8] ADDPD_MEM COVAR(jq,0), xmm0 +%endif ; cpuflag(fma3) inc id add covarq, COVAR_STRIDE cmp id, countd jle .loop2x1 .ret: REP_RET +%endmacro ; UPDATE_LLS +%if HAVE_AVX_EXTERNAL +INIT_YMM avx +UPDATE_LLS +%endif +%if HAVE_FMA3_EXTERNAL +INIT_YMM fma3 +UPDATE_LLS +%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 80cda29139..9f0d862b0e 100644 --- a/libavutil/x86/lls_init.c +++ b/libavutil/x86/lls_init.c @@ -3,29 +3,30 @@ * * 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); +void ff_update_lls_fma3(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) { @@ -38,4 +39,7 @@ av_cold void ff_init_lls_x86(LLSModel *m) if (EXTERNAL_AVX_FAST(cpu_flags)) { m->update_lls = ff_update_lls_avx; } + if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) { + m->update_lls = ff_update_lls_fma3; + } } diff --git a/libavutil/x86/pixelutils.asm b/libavutil/x86/pixelutils.asm new file mode 100644 index 0000000000..7af3007d0c --- /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..c24a533aea --- /dev/null +++ b/libavutil/x86/pixelutils_init.c @@ -0,0 +1,64 @@ +/* + * 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; + } + + // The best way to use SSE2 would be to do 2 SADs in parallel, + // but we'd have to modify the pixelutils API to return SIMD functions. + + // It's probably not faster to shuffle data around + // to get two lines of 8 pixels into a single 16byte register, + // so just use the MMX 8x8 version even when SSE2 is available. + 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 bb7c341341..4d1e88def0 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..a4a05b0419 100644 --- a/libavutil/x86/w64xmmtest.h +++ b/libavutil/x86/w64xmmtest.h @@ -2,23 +2,26 @@ * 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 */ +#ifndef AVUTIL_X86_W64XMMTEST_H +#define AVUTIL_X86_W64XMMTEST_H + #include <inttypes.h> #include <stdint.h> #include <stdlib.h> @@ -71,3 +74,5 @@ int __real_ ## func; \ int __wrap_ ## func; \ int __wrap_ ## func + +#endif /* AVUTIL_X86_W64XMMTEST_H */ diff --git a/libavutil/x86/x86inc.asm b/libavutil/x86/x86inc.asm index 6ad9785536..3a0a2612ba 100644 --- a/libavutil/x86/x86inc.asm +++ b/libavutil/x86/x86inc.asm @@ -1,7 +1,7 @@ ;***************************************************************************** ;* x86inc.asm: x264asm abstraction layer ;***************************************************************************** -;* Copyright (C) 2005-2015 x264 project +;* Copyright (C) 2005-2016 x264 project ;* ;* Authors: Loren Merritt <lorenm@u.washington.edu> ;* Anton Mitrofanov <BugMaster@narod.ru> @@ -183,9 +183,9 @@ %define e%1h %3 %define r%1b %2 %define e%1b %2 -%if ARCH_X86_64 == 0 - %define r%1 e%1 -%endif + %if ARCH_X86_64 == 0 + %define r%1 e%1 + %endif %endmacro DECLARE_REG_SIZE ax, al, ah @@ -295,7 +295,7 @@ DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14 %macro ASSERT 1 %if (%1) == 0 - %error assert failed + %error assertion ``%1'' failed %endif %endmacro @@ -386,8 +386,11 @@ DECLARE_REG_TMP_SIZE 0,1,2,3,4,5,6,7,8,9,10,11,12,13,14 %if %1 != 0 && required_stack_alignment > STACK_ALIGNMENT %if %1 > 0 %assign regs_used (regs_used + 1) - %elif ARCH_X86_64 && regs_used == num_args && num_args <= 4 + UNIX64 * 2 - %warning "Stack pointer will overwrite register argument" + %endif + %if ARCH_X86_64 && regs_used < 5 + UNIX64 * 3 + ; Ensure that we don't clobber any registers containing arguments. For UNIX64 we also preserve r6 (rax) + ; since it's used as a hidden argument in vararg functions to specify the number of vector registers used. + %assign regs_used 5 + UNIX64 * 3 %endif %endif %endif @@ -501,9 +504,9 @@ DECLARE_REG 14, R15, 120 %macro RET 0 WIN64_RESTORE_XMM_INTERNAL rsp POP_IF_USED 14, 13, 12, 11, 10, 9, 8, 7 -%if mmsize == 32 - vzeroupper -%endif + %if mmsize == 32 + vzeroupper + %endif AUTO_REP_RET %endmacro @@ -540,17 +543,17 @@ DECLARE_REG 14, R15, 72 %define has_epilogue regs_used > 9 || mmsize == 32 || stack_size > 0 %macro RET 0 -%if stack_size_padded > 0 -%if required_stack_alignment > STACK_ALIGNMENT - mov rsp, rstkm -%else - add rsp, stack_size_padded -%endif -%endif + %if stack_size_padded > 0 + %if required_stack_alignment > STACK_ALIGNMENT + mov rsp, rstkm + %else + add rsp, stack_size_padded + %endif + %endif POP_IF_USED 14, 13, 12, 11, 10, 9 -%if mmsize == 32 - vzeroupper -%endif + %if mmsize == 32 + vzeroupper + %endif AUTO_REP_RET %endmacro @@ -596,29 +599,29 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14 %define has_epilogue regs_used > 3 || mmsize == 32 || stack_size > 0 %macro RET 0 -%if stack_size_padded > 0 -%if required_stack_alignment > STACK_ALIGNMENT - mov rsp, rstkm -%else - add rsp, stack_size_padded -%endif -%endif + %if stack_size_padded > 0 + %if required_stack_alignment > STACK_ALIGNMENT + mov rsp, rstkm + %else + add rsp, stack_size_padded + %endif + %endif POP_IF_USED 6, 5, 4, 3 -%if mmsize == 32 - vzeroupper -%endif + %if mmsize == 32 + vzeroupper + %endif AUTO_REP_RET %endmacro %endif ;====================================================================== %if WIN64 == 0 -%macro WIN64_SPILL_XMM 1 -%endmacro -%macro WIN64_RESTORE_XMM 1 -%endmacro -%macro WIN64_PUSH_XMM 0 -%endmacro + %macro WIN64_SPILL_XMM 1 + %endmacro + %macro WIN64_RESTORE_XMM 1 + %endmacro + %macro WIN64_PUSH_XMM 0 + %endmacro %endif ; On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either @@ -631,24 +634,26 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14 %else rep ret %endif + annotate_function_size %endmacro %define last_branch_adr $$ %macro AUTO_REP_RET 0 - %ifndef cpuflags - times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ != last_branch_adr. - %elif notcpuflag(ssse3) - times ((last_branch_adr-$)>>31)+1 rep + %if notcpuflag(ssse3) + times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ == last_branch_adr. %endif ret + annotate_function_size %endmacro %macro BRANCH_INSTR 0-* %rep %0 %macro %1 1-2 %1 %2 %1 - %%branch_instr: - %xdefine last_branch_adr %%branch_instr + %if notcpuflag(ssse3) + %%branch_instr equ $ + %xdefine last_branch_adr %%branch_instr + %endif %endmacro %rotate 1 %endrep @@ -663,6 +668,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %elif %2 jmp %1 %endif + annotate_function_size %endmacro ;============================================================================= @@ -684,6 +690,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, cglobal_internal 0, %1 %+ SUFFIX, %2 %endmacro %macro cglobal_internal 2-3+ + annotate_function_size %if %1 %xdefine %%FUNCTION_PREFIX private_prefix %xdefine %%VISIBILITY hidden @@ -697,6 +704,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, CAT_XDEFINE cglobaled_, %2, 1 %endif %xdefine current_function %2 + %xdefine current_function_section __SECT__ %if FORMAT_ELF global %2:function %%VISIBILITY %else @@ -745,6 +753,24 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, [SECTION .note.GNU-stack noalloc noexec nowrite progbits] %endif +; Tell debuggers how large the function was. +; This may be invoked multiple times per function; we rely on later instances overriding earlier ones. +; This is invoked by RET and similar macros, and also cglobal does it for the previous function, +; but if the last function in a source file doesn't use any of the standard macros for its epilogue, +; then its size might be unspecified. +%macro annotate_function_size 0 + %ifdef __YASM_VER__ + %ifdef current_function + %if FORMAT_ELF + current_function_section + %%ecf equ $ + size current_function %%ecf - current_function + __SECT__ + %endif + %endif + %endif +%endmacro + ; cpuflags %assign cpuflags_mmx (1<<0) @@ -772,9 +798,11 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %assign cpuflags_atom (1<<21) %assign cpuflags_bmi1 (1<<22)|cpuflags_lzcnt %assign cpuflags_bmi2 (1<<23)|cpuflags_bmi1 +%assign cpuflags_aesni (1<<24)|cpuflags_sse42 -%define cpuflag(x) ((cpuflags & (cpuflags_ %+ x)) == (cpuflags_ %+ x)) -%define notcpuflag(x) ((cpuflags & (cpuflags_ %+ x)) != (cpuflags_ %+ x)) +; Returns a boolean value expressing whether or not the specified cpuflag is enabled. +%define cpuflag(x) (((((cpuflags & (cpuflags_ %+ x)) ^ (cpuflags_ %+ x)) - 1) >> 31) & 1) +%define notcpuflag(x) (cpuflag(x) ^ 1) ; Takes an arbitrary number of cpuflags from the above list. ; All subsequent functions (up to the next INIT_CPUFLAGS) is built for the specified cpu. @@ -843,14 +871,14 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %define movnta movntq %assign %%i 0 %rep 8 - CAT_XDEFINE m, %%i, mm %+ %%i - CAT_XDEFINE nnmm, %%i, %%i - %assign %%i %%i+1 + CAT_XDEFINE m, %%i, mm %+ %%i + CAT_XDEFINE nnmm, %%i, %%i + %assign %%i %%i+1 %endrep %rep 8 - CAT_UNDEF m, %%i - CAT_UNDEF nnmm, %%i - %assign %%i %%i+1 + CAT_UNDEF m, %%i + CAT_UNDEF nnmm, %%i + %assign %%i %%i+1 %endrep INIT_CPUFLAGS %1 %endmacro @@ -861,7 +889,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %define mmsize 16 %define num_mmregs 8 %if ARCH_X86_64 - %define num_mmregs 16 + %define num_mmregs 16 %endif %define mova movdqa %define movu movdqu @@ -869,9 +897,9 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %define movnta movntdq %assign %%i 0 %rep num_mmregs - CAT_XDEFINE m, %%i, xmm %+ %%i - CAT_XDEFINE nnxmm, %%i, %%i - %assign %%i %%i+1 + CAT_XDEFINE m, %%i, xmm %+ %%i + CAT_XDEFINE nnxmm, %%i, %%i + %assign %%i %%i+1 %endrep INIT_CPUFLAGS %1 %endmacro @@ -882,7 +910,7 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %define mmsize 32 %define num_mmregs 8 %if ARCH_X86_64 - %define num_mmregs 16 + %define num_mmregs 16 %endif %define mova movdqa %define movu movdqu @@ -890,9 +918,9 @@ BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, %define movnta movntdq %assign %%i 0 %rep num_mmregs - CAT_XDEFINE m, %%i, ymm %+ %%i - CAT_XDEFINE nnymm, %%i, %%i - %assign %%i %%i+1 + CAT_XDEFINE m, %%i, ymm %+ %%i + CAT_XDEFINE nnymm, %%i, %%i + %assign %%i %%i+1 %endrep INIT_CPUFLAGS %1 %endmacro @@ -916,7 +944,7 @@ INIT_XMM %assign i 0 %rep 16 DECLARE_MMCAST i -%assign i i+1 + %assign i i+1 %endrep ; I often want to use macros that permute their arguments. e.g. there's no @@ -934,23 +962,23 @@ INIT_XMM ; doesn't cost any cycles. %macro PERMUTE 2-* ; takes a list of pairs to swap -%rep %0/2 - %xdefine %%tmp%2 m%2 - %rotate 2 -%endrep -%rep %0/2 - %xdefine m%1 %%tmp%2 - CAT_XDEFINE nn, m%1, %1 - %rotate 2 -%endrep + %rep %0/2 + %xdefine %%tmp%2 m%2 + %rotate 2 + %endrep + %rep %0/2 + %xdefine m%1 %%tmp%2 + CAT_XDEFINE nn, m%1, %1 + %rotate 2 + %endrep %endmacro %macro SWAP 2+ ; swaps a single chain (sometimes more concise than pairs) -%ifnum %1 ; SWAP 0, 1, ... - SWAP_INTERNAL_NUM %1, %2 -%else ; SWAP m0, m1, ... - SWAP_INTERNAL_NAME %1, %2 -%endif + %ifnum %1 ; SWAP 0, 1, ... + SWAP_INTERNAL_NUM %1, %2 + %else ; SWAP m0, m1, ... + SWAP_INTERNAL_NAME %1, %2 + %endif %endmacro %macro SWAP_INTERNAL_NUM 2-* @@ -960,7 +988,7 @@ INIT_XMM %xdefine m%2 %%tmp CAT_XDEFINE nn, m%1, %1 CAT_XDEFINE nn, m%2, %2 - %rotate 1 + %rotate 1 %endrep %endmacro @@ -968,7 +996,7 @@ INIT_XMM %xdefine %%args nn %+ %1 %rep %0-1 %xdefine %%args %%args, nn %+ %2 - %rotate 1 + %rotate 1 %endrep SWAP_INTERNAL_NUM %%args %endmacro @@ -985,7 +1013,7 @@ INIT_XMM %assign %%i 0 %rep num_mmregs CAT_XDEFINE %%f, %%i, m %+ %%i - %assign %%i %%i+1 + %assign %%i %%i+1 %endrep %endmacro @@ -995,7 +1023,7 @@ INIT_XMM %rep num_mmregs CAT_XDEFINE m, %%i, %1_m %+ %%i CAT_XDEFINE nn, m %+ %%i, %%i - %assign %%i %%i+1 + %assign %%i %%i+1 %endrep %endif %endmacro @@ -1051,7 +1079,7 @@ INIT_XMM %endif CAT_XDEFINE sizeofxmm, i, 16 CAT_XDEFINE sizeofymm, i, 32 -%assign i i+1 + %assign i i+1 %endrep %undef i @@ -1170,12 +1198,12 @@ 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 aesdec, aesni, 0, 0, 0 +AVX_INSTR aesdeclast, aesni, 0, 0, 0 +AVX_INSTR aesenc, aesni, 0, 0, 0 +AVX_INSTR aesenclast, aesni, 0, 0, 0 +AVX_INSTR aesimc, aesni +AVX_INSTR aeskeygenassist, aesni AVX_INSTR andnpd, sse2, 1, 0, 0 AVX_INSTR andnps, sse, 1, 0, 0 AVX_INSTR andpd, sse2, 1, 0, 1 @@ -1428,7 +1456,7 @@ AVX_INSTR pfmul, 3dnow, 1, 0, 1 %else CAT_XDEFINE q, j, i %endif -%assign i i+1 + %assign i i+1 %endrep %undef i %undef j @@ -1455,47 +1483,44 @@ FMA_INSTR pmadcswd, pmaddwd, paddd ; This lets us use tzcnt without bumping the yasm version requirement yet. %define tzcnt rep bsf -; convert FMA4 to FMA3 if possible -%macro FMA4_INSTR 4 - %macro %1 4-8 %1, %2, %3, %4 - %if cpuflag(fma4) - v%5 %1, %2, %3, %4 - %elifidn %1, %2 - v%6 %1, %4, %3 ; %1 = %1 * %3 + %4 - %elifidn %1, %3 - v%7 %1, %2, %4 ; %1 = %2 * %1 + %4 - %elifidn %1, %4 - v%8 %1, %2, %3 ; %1 = %2 * %3 + %1 - %else - %error fma3 emulation of ``%5 %1, %2, %3, %4'' is not supported - %endif - %endmacro +; Macros for consolidating FMA3 and FMA4 using 4-operand (dst, src1, src2, src3) syntax. +; FMA3 is only possible if dst is the same as one of the src registers. +; Either src2 or src3 can be a memory operand. +%macro FMA4_INSTR 2-* + %push fma4_instr + %xdefine %$prefix %1 + %rep %0 - 1 + %macro %$prefix%2 4-6 %$prefix, %2 + %if notcpuflag(fma3) && notcpuflag(fma4) + %error use of ``%5%6'' fma instruction in cpuname function: current_function + %elif cpuflag(fma4) + v%5%6 %1, %2, %3, %4 + %elifidn %1, %2 + ; If %3 or %4 is a memory operand it needs to be encoded as the last operand. + %ifid %3 + v%{5}213%6 %2, %3, %4 + %else + v%{5}132%6 %2, %4, %3 + %endif + %elifidn %1, %3 + v%{5}213%6 %3, %2, %4 + %elifidn %1, %4 + v%{5}231%6 %4, %2, %3 + %else + %error fma3 emulation of ``%5%6 %1, %2, %3, %4'' is not supported + %endif + %endmacro + %rotate 1 + %endrep + %pop %endmacro -FMA4_INSTR fmaddpd, fmadd132pd, fmadd213pd, fmadd231pd -FMA4_INSTR fmaddps, fmadd132ps, fmadd213ps, fmadd231ps -FMA4_INSTR fmaddsd, fmadd132sd, fmadd213sd, fmadd231sd -FMA4_INSTR fmaddss, fmadd132ss, fmadd213ss, fmadd231ss - -FMA4_INSTR fmaddsubpd, fmaddsub132pd, fmaddsub213pd, fmaddsub231pd -FMA4_INSTR fmaddsubps, fmaddsub132ps, fmaddsub213ps, fmaddsub231ps -FMA4_INSTR fmsubaddpd, fmsubadd132pd, fmsubadd213pd, fmsubadd231pd -FMA4_INSTR fmsubaddps, fmsubadd132ps, fmsubadd213ps, fmsubadd231ps - -FMA4_INSTR fmsubpd, fmsub132pd, fmsub213pd, fmsub231pd -FMA4_INSTR fmsubps, fmsub132ps, fmsub213ps, fmsub231ps -FMA4_INSTR fmsubsd, fmsub132sd, fmsub213sd, fmsub231sd -FMA4_INSTR fmsubss, fmsub132ss, fmsub213ss, fmsub231ss - -FMA4_INSTR fnmaddpd, fnmadd132pd, fnmadd213pd, fnmadd231pd -FMA4_INSTR fnmaddps, fnmadd132ps, fnmadd213ps, fnmadd231ps -FMA4_INSTR fnmaddsd, fnmadd132sd, fnmadd213sd, fnmadd231sd -FMA4_INSTR fnmaddss, fnmadd132ss, fnmadd213ss, fnmadd231ss - -FMA4_INSTR fnmsubpd, fnmsub132pd, fnmsub213pd, fnmsub231pd -FMA4_INSTR fnmsubps, fnmsub132ps, fnmsub213ps, fnmsub231ps -FMA4_INSTR fnmsubsd, fnmsub132sd, fnmsub213sd, fnmsub231sd -FMA4_INSTR fnmsubss, fnmsub132ss, fnmsub213ss, fnmsub231ss +FMA4_INSTR fmadd, pd, ps, sd, ss +FMA4_INSTR fmaddsub, pd, ps +FMA4_INSTR fmsub, pd, ps, sd, ss +FMA4_INSTR fmsubadd, pd, ps +FMA4_INSTR fnmadd, pd, ps, sd, ss +FMA4_INSTR fnmsub, pd, ps, sd, ss ; workaround: vpbroadcastq is broken in x86_32 due to a yasm bug (fixed in 1.3.0) %ifdef __YASM_VER__ diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index 9f64dd13e1..b09fa813e2 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 @@ -164,13 +210,13 @@ %endif %endmacro -%macro PSIGNW_MMX 2 +%macro PSIGNW 2 +%if cpuflag(ssse3) + psignw %1, %2 +%else pxor %1, %2 psubw %1, %2 -%endmacro - -%macro PSIGNW_SSSE3 2 - psignw %1, %2 +%endif %endmacro %macro ABS1 2 @@ -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) @@ -683,3 +777,19 @@ 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 |