From 708ab7dd69d5c98221882a8086f68f1bb02a44a3 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Sun, 9 Oct 2011 19:12:09 -0400 Subject: fmtconvert: port float_to_int16() x86 inline asm to yasm --- libavcodec/x86/fmtconvert.asm | 42 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 42 insertions(+) (limited to 'libavcodec/x86/fmtconvert.asm') diff --git a/libavcodec/x86/fmtconvert.asm b/libavcodec/x86/fmtconvert.asm index efab87d570..d314a4e14e 100644 --- a/libavcodec/x86/fmtconvert.asm +++ b/libavcodec/x86/fmtconvert.asm @@ -24,6 +24,48 @@ SECTION_TEXT +;------------------------------------------------------------------------------ +; void ff_float_to_int16(int16_t *dst, const float *src, long len); +;------------------------------------------------------------------------------ +%macro FLOAT_TO_INT16 2 +cglobal float_to_int16_%1, 3,3,%2, dst, src, len + add lenq, lenq + lea srcq, [srcq+2*lenq] + add dstq, lenq + neg lenq +.loop: +%ifidn %1, sse2 + cvtps2dq m0, [srcq+2*lenq ] + cvtps2dq m1, [srcq+2*lenq+16] + packssdw m0, m1 + mova [dstq+lenq], m0 +%else + cvtps2pi m0, [srcq+2*lenq ] + cvtps2pi m1, [srcq+2*lenq+ 8] + cvtps2pi m2, [srcq+2*lenq+16] + cvtps2pi m3, [srcq+2*lenq+24] + packssdw m0, m1 + packssdw m2, m3 + mova [dstq+lenq ], m0 + mova [dstq+lenq+8], m2 +%endif + add lenq, 16 + js .loop +%ifnidn %1, sse2 + emms +%endif + REP_RET +%endmacro + +INIT_XMM +FLOAT_TO_INT16 sse2, 2 +INIT_MMX +FLOAT_TO_INT16 sse, 0 +%define cvtps2pi pf2id +FLOAT_TO_INT16 3dnow, 0 +%undef cvtps2pi + + %macro PSWAPD_SSE 2 pshufw %1, %2, 0x4e %endmacro -- cgit v1.2.3 From 4e8e2624767f4af0eaa932c543d072fed96fd586 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Sun, 9 Oct 2011 23:52:03 -0400 Subject: fmtconvert: port int32_to_float_fmul_scalar() x86 inline asm to yasm --- libavcodec/x86/dsputil_yasm.asm | 8 ------ libavcodec/x86/fmtconvert.asm | 46 ++++++++++++++++++++++++++++++++ libavcodec/x86/fmtconvert_mmx.c | 59 +++++------------------------------------ libavutil/x86/x86util.asm | 12 +++++++++ 4 files changed, 65 insertions(+), 60 deletions(-) (limited to 'libavcodec/x86/fmtconvert.asm') diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm index 2a2108404a..fe96d8b12b 100644 --- a/libavcodec/x86/dsputil_yasm.asm +++ b/libavcodec/x86/dsputil_yasm.asm @@ -1055,14 +1055,6 @@ emu_edge mmx ; int32_t max, unsigned int len) ;----------------------------------------------------------------------------- -%macro SPLATD_MMX 1 - punpckldq %1, %1 -%endmacro - -%macro SPLATD_SSE2 1 - pshufd %1, %1, 0 -%endmacro - %macro VECTOR_CLIP_INT32 4 cglobal vector_clip_int32_%1, 5,5,%2, dst, src, min, max, len %ifidn %1, sse2 diff --git a/libavcodec/x86/fmtconvert.asm b/libavcodec/x86/fmtconvert.asm index d314a4e14e..e3eb5d2286 100644 --- a/libavcodec/x86/fmtconvert.asm +++ b/libavcodec/x86/fmtconvert.asm @@ -24,6 +24,52 @@ SECTION_TEXT +;--------------------------------------------------------------------------------- +; void int32_to_float_fmul_scalar(float *dst, const int *src, float mul, int len); +;--------------------------------------------------------------------------------- +%macro INT32_TO_FLOAT_FMUL_SCALAR 2 +%ifdef ARCH_X86_64 +cglobal int32_to_float_fmul_scalar_%1, 3,3,%2, dst, src, len +%else +cglobal int32_to_float_fmul_scalar_%1, 4,4,%2, dst, src, mul, len + movss m0, mulm +%endif + SPLATD m0 + shl lenq, 2 + add srcq, lenq + add dstq, lenq + neg lenq +.loop: +%ifidn %1, sse2 + cvtdq2ps m1, [srcq+lenq ] + cvtdq2ps m2, [srcq+lenq+16] +%else + cvtpi2ps m1, [srcq+lenq ] + cvtpi2ps m3, [srcq+lenq+ 8] + cvtpi2ps m2, [srcq+lenq+16] + cvtpi2ps m4, [srcq+lenq+24] + movlhps m1, m3 + movlhps m2, m4 +%endif + mulps m1, m0 + mulps m2, m0 + mova [dstq+lenq ], m1 + mova [dstq+lenq+16], m2 + add lenq, 32 + jl .loop + REP_RET +%endmacro + +INIT_XMM +%define SPLATD SPLATD_SSE +%define movdqa movaps +INT32_TO_FLOAT_FMUL_SCALAR sse, 5 +%undef movdqa +%define SPLATD SPLATD_SSE2 +INT32_TO_FLOAT_FMUL_SCALAR sse2, 3 +%undef SPLATD + + ;------------------------------------------------------------------------------ ; void ff_float_to_int16(int16_t *dst, const float *src, long len); ;------------------------------------------------------------------------------ diff --git a/libavcodec/x86/fmtconvert_mmx.c b/libavcodec/x86/fmtconvert_mmx.c index 6e43280d66..86957b45ff 100644 --- a/libavcodec/x86/fmtconvert_mmx.c +++ b/libavcodec/x86/fmtconvert_mmx.c @@ -26,52 +26,11 @@ #include "libavutil/x86_cpu.h" #include "libavcodec/fmtconvert.h" -static void int32_to_float_fmul_scalar_sse(float *dst, const int *src, float mul, int len) -{ - x86_reg i = -4*len; - __asm__ volatile( - "movss %3, %%xmm4 \n" - "shufps $0, %%xmm4, %%xmm4 \n" - "1: \n" - "cvtpi2ps (%2,%0), %%xmm0 \n" - "cvtpi2ps 8(%2,%0), %%xmm1 \n" - "cvtpi2ps 16(%2,%0), %%xmm2 \n" - "cvtpi2ps 24(%2,%0), %%xmm3 \n" - "movlhps %%xmm1, %%xmm0 \n" - "movlhps %%xmm3, %%xmm2 \n" - "mulps %%xmm4, %%xmm0 \n" - "mulps %%xmm4, %%xmm2 \n" - "movaps %%xmm0, (%1,%0) \n" - "movaps %%xmm2, 16(%1,%0) \n" - "add $32, %0 \n" - "jl 1b \n" - :"+r"(i) - :"r"(dst+len), "r"(src+len), "m"(mul) - ); -} - -static void int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len) -{ - x86_reg i = -4*len; - __asm__ volatile( - "movss %3, %%xmm4 \n" - "shufps $0, %%xmm4, %%xmm4 \n" - "1: \n" - "cvtdq2ps (%2,%0), %%xmm0 \n" - "cvtdq2ps 16(%2,%0), %%xmm1 \n" - "mulps %%xmm4, %%xmm0 \n" - "mulps %%xmm4, %%xmm1 \n" - "movaps %%xmm0, (%1,%0) \n" - "movaps %%xmm1, 16(%1,%0) \n" - "add $32, %0 \n" - "jl 1b \n" - :"+r"(i) - :"r"(dst+len), "r"(src+len), "m"(mul) - ); -} - #if HAVE_YASM +void ff_int32_to_float_fmul_scalar_sse (float *dst, const int *src, float mul, int len); +void ff_int32_to_float_fmul_scalar_sse2(float *dst, const int *src, float mul, int len); + void ff_float_to_int16_3dnow(int16_t *dst, const float *src, long len); void ff_float_to_int16_sse (int16_t *dst, const float *src, long len); void ff_float_to_int16_sse2 (int16_t *dst, const float *src, long len); @@ -204,8 +163,8 @@ void ff_fmt_convert_init_x86(FmtConvertContext *c, AVCodecContext *avctx) { int mm_flags = av_get_cpu_flags(); - if (mm_flags & AV_CPU_FLAG_MMX) { #if HAVE_YASM + if (mm_flags & AV_CPU_FLAG_MMX) { c->float_interleave = float_interleave_mmx; if (HAVE_AMD3DNOW && mm_flags & AV_CPU_FLAG_3DNOW) { @@ -219,21 +178,17 @@ void ff_fmt_convert_init_x86(FmtConvertContext *c, AVCodecContext *avctx) c->float_to_int16_interleave = float_to_int16_interleave_3dn2; } } -#endif if (HAVE_SSE && mm_flags & AV_CPU_FLAG_SSE) { - c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse; -#if HAVE_YASM + c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_sse; c->float_to_int16 = ff_float_to_int16_sse; c->float_to_int16_interleave = float_to_int16_interleave_sse; c->float_interleave = float_interleave_sse; -#endif } if (HAVE_SSE && mm_flags & AV_CPU_FLAG_SSE2) { - c->int32_to_float_fmul_scalar = int32_to_float_fmul_scalar_sse2; -#if HAVE_YASM + c->int32_to_float_fmul_scalar = ff_int32_to_float_fmul_scalar_sse2; c->float_to_int16 = ff_float_to_int16_sse2; c->float_to_int16_interleave = float_to_int16_interleave_sse2; -#endif } } +#endif } diff --git a/libavutil/x86/x86util.asm b/libavutil/x86/x86util.asm index 7e16c15db2..874443a2ef 100644 --- a/libavutil/x86/x86util.asm +++ b/libavutil/x86/x86util.asm @@ -536,6 +536,18 @@ %endif %endmacro +%macro SPLATD_MMX 1 + punpckldq %1, %1 +%endmacro + +%macro SPLATD_SSE 1 + shufps %1, %1, 0 +%endmacro + +%macro SPLATD_SSE2 1 + pshufd %1, %1, 0 +%endmacro + %macro CLIPW 3 ;(dst, min, max) pmaxsw %1, %2 pminsw %1, %3 -- cgit v1.2.3 From aad3429d4e34b74e4eb0b37b17f32804e217cf02 Mon Sep 17 00:00:00 2001 From: Justin Ruggles Date: Mon, 10 Oct 2011 00:43:08 -0400 Subject: fmtconvert: port float_to_int16_interleave() 2-channel x86 inline asm to yasm --- libavcodec/x86/fmtconvert.asm | 52 +++++++++++++++++++++++++++++++ libavcodec/x86/fmtconvert_mmx.c | 69 ++++++----------------------------------- 2 files changed, 61 insertions(+), 60 deletions(-) (limited to 'libavcodec/x86/fmtconvert.asm') diff --git a/libavcodec/x86/fmtconvert.asm b/libavcodec/x86/fmtconvert.asm index e3eb5d2286..854954835c 100644 --- a/libavcodec/x86/fmtconvert.asm +++ b/libavcodec/x86/fmtconvert.asm @@ -112,6 +112,58 @@ FLOAT_TO_INT16 3dnow, 0 %undef cvtps2pi +;------------------------------------------------------------------------------- +; void ff_float_to_int16_interleave2(int16_t *dst, const float **src, long len); +;------------------------------------------------------------------------------- +%macro FLOAT_TO_INT16_INTERLEAVE2 1 +cglobal float_to_int16_interleave2_%1, 3,4,2, dst, src0, src1, len + lea lenq, [4*r2q] + mov src1q, [src0q+gprsize] + mov src0q, [src0q] + add dstq, lenq + add src0q, lenq + add src1q, lenq + neg lenq +.loop: +%ifidn %1, sse2 + cvtps2dq m0, [src0q+lenq] + cvtps2dq m1, [src1q+lenq] + packssdw m0, m1 + movhlps m1, m0 + punpcklwd m0, m1 + mova [dstq+lenq], m0 +%else + cvtps2pi m0, [src0q+lenq ] + cvtps2pi m1, [src0q+lenq+8] + cvtps2pi m2, [src1q+lenq ] + cvtps2pi m3, [src1q+lenq+8] + packssdw m0, m1 + packssdw m2, m3 + mova m1, m0 + punpcklwd m0, m2 + punpckhwd m1, m2 + mova [dstq+lenq ], m0 + mova [dstq+lenq+8], m1 +%endif + add lenq, 16 + js .loop +%ifnidn %1, sse2 + emms +%endif + REP_RET +%endmacro + +INIT_MMX +%define cvtps2pi pf2id +FLOAT_TO_INT16_INTERLEAVE2 3dnow +%undef cvtps2pi +%define movdqa movaps +FLOAT_TO_INT16_INTERLEAVE2 sse +%undef movdqa +INIT_XMM +FLOAT_TO_INT16_INTERLEAVE2 sse2 + + %macro PSWAPD_SSE 2 pshufw %1, %2, 0x4e %endmacro diff --git a/libavcodec/x86/fmtconvert_mmx.c b/libavcodec/x86/fmtconvert_mmx.c index 86957b45ff..17079d3c82 100644 --- a/libavcodec/x86/fmtconvert_mmx.c +++ b/libavcodec/x86/fmtconvert_mmx.c @@ -35,13 +35,17 @@ void ff_float_to_int16_3dnow(int16_t *dst, const float *src, long len); void ff_float_to_int16_sse (int16_t *dst, const float *src, long len); void ff_float_to_int16_sse2 (int16_t *dst, const float *src, long len); +void ff_float_to_int16_interleave2_3dnow(int16_t *dst, const float **src, long len); +void ff_float_to_int16_interleave2_sse (int16_t *dst, const float **src, long len); +void ff_float_to_int16_interleave2_sse2 (int16_t *dst, const float **src, long len); + void ff_float_to_int16_interleave6_sse(int16_t *dst, const float **src, int len); void ff_float_to_int16_interleave6_3dnow(int16_t *dst, const float **src, int len); void ff_float_to_int16_interleave6_3dn2(int16_t *dst, const float **src, int len); #define ff_float_to_int16_interleave6_sse2 ff_float_to_int16_interleave6_sse -#define FLOAT_TO_INT16_INTERLEAVE(cpu, body) \ +#define FLOAT_TO_INT16_INTERLEAVE(cpu) \ /* gcc pessimizes register allocation if this is in the same function as float_to_int16_interleave_sse2*/\ static av_noinline void float_to_int16_interleave_misc_##cpu(int16_t *dst, const float **src, long len, int channels){\ DECLARE_ALIGNED(16, int16_t, tmp)[len];\ @@ -57,71 +61,16 @@ static void float_to_int16_interleave_##cpu(int16_t *dst, const float **src, lon if(channels==1)\ ff_float_to_int16_##cpu(dst, src[0], len);\ else if(channels==2){\ - x86_reg reglen = len; \ - const float *src0 = src[0];\ - const float *src1 = src[1];\ - __asm__ volatile(\ - "shl $2, %0 \n"\ - "add %0, %1 \n"\ - "add %0, %2 \n"\ - "add %0, %3 \n"\ - "neg %0 \n"\ - body\ - :"+r"(reglen), "+r"(dst), "+r"(src0), "+r"(src1)\ - );\ + ff_float_to_int16_interleave2_##cpu(dst, src, len);\ }else if(channels==6){\ ff_float_to_int16_interleave6_##cpu(dst, src, len);\ }else\ float_to_int16_interleave_misc_##cpu(dst, src, len, channels);\ } -FLOAT_TO_INT16_INTERLEAVE(3dnow, - "1: \n" - "pf2id (%2,%0), %%mm0 \n" - "pf2id 8(%2,%0), %%mm1 \n" - "pf2id (%3,%0), %%mm2 \n" - "pf2id 8(%3,%0), %%mm3 \n" - "packssdw %%mm1, %%mm0 \n" - "packssdw %%mm3, %%mm2 \n" - "movq %%mm0, %%mm1 \n" - "punpcklwd %%mm2, %%mm0 \n" - "punpckhwd %%mm2, %%mm1 \n" - "movq %%mm0, (%1,%0)\n" - "movq %%mm1, 8(%1,%0)\n" - "add $16, %0 \n" - "js 1b \n" - "femms \n" -) - -FLOAT_TO_INT16_INTERLEAVE(sse, - "1: \n" - "cvtps2pi (%2,%0), %%mm0 \n" - "cvtps2pi 8(%2,%0), %%mm1 \n" - "cvtps2pi (%3,%0), %%mm2 \n" - "cvtps2pi 8(%3,%0), %%mm3 \n" - "packssdw %%mm1, %%mm0 \n" - "packssdw %%mm3, %%mm2 \n" - "movq %%mm0, %%mm1 \n" - "punpcklwd %%mm2, %%mm0 \n" - "punpckhwd %%mm2, %%mm1 \n" - "movq %%mm0, (%1,%0)\n" - "movq %%mm1, 8(%1,%0)\n" - "add $16, %0 \n" - "js 1b \n" - "emms \n" -) - -FLOAT_TO_INT16_INTERLEAVE(sse2, - "1: \n" - "cvtps2dq (%2,%0), %%xmm0 \n" - "cvtps2dq (%3,%0), %%xmm1 \n" - "packssdw %%xmm1, %%xmm0 \n" - "movhlps %%xmm0, %%xmm1 \n" - "punpcklwd %%xmm1, %%xmm0 \n" - "movdqa %%xmm0, (%1,%0) \n" - "add $16, %0 \n" - "js 1b \n" -) +FLOAT_TO_INT16_INTERLEAVE(3dnow) +FLOAT_TO_INT16_INTERLEAVE(sse) +FLOAT_TO_INT16_INTERLEAVE(sse2) static void float_to_int16_interleave_3dn2(int16_t *dst, const float **src, long len, int channels){ if(channels==6) -- cgit v1.2.3