[Mplayerxp-cvslog] SF.net SVN: mplayerxp:[112] mplayerxp
Brought to you by:
olov
Menu
▾
▴
[Mplayerxp-cvslog] SF.net SVN: mplayerxp:[112] mplayerxp
|
From: <nic...@us...> - 2010-01-19 17:51:24
|
Revision: 112
http://mplayerxp.svn.sourceforge.net/mplayerxp/?rev=112&view=rev
Author: nickols_k
Date: 2010-01-19 17:51:18 +0000 (Tue, 19 Jan 2010)
Log Message:
-----------
cumulative pathc: enable slices in vf_fmtcvt + pvector based MMX optimization
Modified Paths:
--------------
TOOLS/Makefile
TOOLS/asmopt.c
etc/codecs.conf
mplayerxp/libmpcodecs/dec_video.c
mplayerxp/postproc/dsp.c
mplayerxp/postproc/dsp_accel.h
mplayerxp/postproc/vf.c
mplayerxp/postproc/vf_scale.c
Added Paths:
-----------
mplayerxp/postproc/dsp_accelf.h
Modified: TOOLS/Makefile
===================================================================
--- TOOLS/Makefile 2010-01-18 18:37:48 UTC (rev 111)
+++ TOOLS/Makefile 2010-01-19 17:51:18 UTC (rev 112)
@@ -1,6 +1,6 @@
-include ../mplayerxp/config.mak
+include ../mplayerxp/mp_config.mak
-CFLAGS = -I..
+CFLAGS = -g3 -I..
OBJS = bios2dump$(EXESUF) \
mem2dump$(EXESUF) \
@@ -12,7 +12,7 @@
../mplayerxp/libmpdemux/freesdp/libfreesdp.a
endif
MP_LIBS += ../mplayerxp/libmpcodecs/libmpcodecs.a ../mplayerxp/libao2/libao2.a ../mplayerxp/postproc/libpostproc.a \
-../mplayerxp/postproc/libmenu/libmenu.a ../mplayerxp/input/libinput.a ../mplayerxp/libvo/libvo.a \
+../mplayerxp/input/libinput.a ../mplayerxp/libvo/libvo.a \
../mplayerxp/osdep/libosdep.a ../mplayerxp/mp_msg.o ../mplayerxp/nls/libnls.a ../mplayerxp/cpudetect.o
COMMON_LIBS = $(MP_LIBS) $(EXTRALIBS) -lm
Modified: TOOLS/asmopt.c
===================================================================
--- TOOLS/asmopt.c 2010-01-18 18:37:48 UTC (rev 111)
+++ TOOLS/asmopt.c 2010-01-19 17:51:18 UTC (rev 112)
@@ -20,66 +20,48 @@
#include "../mplayerxp/mp_config.h"
#include "../mplayerxp/cpudetect.h"
-#if defined( ARCH_X86 ) || defined(ARCH_X86_64)
-#define CAN_COMPILE_X86_ASM
-#endif
+#undef OPTIMIZE_AVX
+#undef OPTIMIZE_SSE4
+#undef OPTIMIZE_SSSE3
+#undef OPTIMIZE_SSE3
+#undef OPTIMIZE_SSE2
+#undef OPTIMIZE_SSE
+#undef OPTIMIZE_MMX2
+#undef OPTIMIZE_MMX
+#define RENAME(a) a ## _C
+#include "asmopt_template.h"
-#ifdef CAN_COMPILE_X86_ASM
-#undef HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_SSE2
-#undef HAVE_SSE3
-#undef HAVE_SSSE3
-#undef HAVE_SSE4
-#undef HAVE_AVX
-
-/*MMX versions*/
+#ifdef __MMX__
+#define OPTIMIZE_MMX
#undef RENAME
-#define HAVE_MMX
-#undef HAVE_MMX2
#define RENAME(a) a ## _MMX
#include "asmopt_template.h"
-
-/*MMX2 versions*/
+#endif
+#ifdef __SSE__
+#define OPTIMIZE_MMX2
#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
#define RENAME(a) a ## _MMX2
#include "asmopt_template.h"
-
-/*SSE2 versions*/
+#endif
+#ifdef __SSE2__
+#define OPTIMIZE_SSE2
#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
-#define HAVE_SSE
-#define HAVE_SSE2
#define RENAME(a) a ## _SSE2
#include "asmopt_template.h"
-
-/*SSE2 versions*/
+#endif
+#ifdef __SSE3__
+#define OPTIMIZE_SSE3
#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
-#define HAVE_SSE
-#define HAVE_SSE2
-#define HAVE_SSE3
#define RENAME(a) a ## _SSE3
#include "asmopt_template.h"
-
#endif
-
-/* generic version */
+#ifdef __SSE4_1__
+#define OPTIMIZE_SSE4
#undef RENAME
-#undef HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_SSE2
-#undef HAVE_SSE3
-#undef HAVE_SSSE3
-#undef HAVE_SSE4
-#undef HAVE_AVX
-#define RENAME(a) a ## _C
+#define RENAME(a) a ## _SSE4
#include "asmopt_template.h"
+#endif
#define ARR_SIZE (1024*64*2)*10
unsigned verbose=1;
@@ -155,11 +137,17 @@
gCpuCaps.hasSSE, gCpuCaps.hasSSE2);
test_simd("asmopt.gen" ,"GENERIC:",convert_C);
-#ifdef CAN_COMPILE_X86_ASM
// ordered per speed fasterst first
+#ifdef __SSE3__
if(gCpuCaps.hasSSE3) test_simd("asmopt.sse3","SSE3 :",convert_SSE3);
+#endif
+#ifdef __SSE2__
if(gCpuCaps.hasSSE2) test_simd("asmopt.sse2","SSE2 :",convert_SSE2);
+#endif
+#ifdef __MMX2__
if(gCpuCaps.hasMMX2) test_simd("asmopt.mmx2","MMX2 :",convert_MMX2);
+#endif
+#ifdef __MMX__
if(gCpuCaps.hasMMX) test_simd("asmopt.mmx", "MMX :",convert_MMX);
#endif
return 0;
Modified: etc/codecs.conf
===================================================================
--- etc/codecs.conf 2010-01-18 18:37:48 UTC (rev 111)
+++ etc/codecs.conf 2010-01-19 17:51:18 UTC (rev 112)
@@ -2219,7 +2219,7 @@
videocodec ffwmv3
info "Windows Media Video 9 DMO"
- status untested
+ status working
fourcc WMV3,wmv3
fourcc WMVP,wmvp
driver ffmpeg
Modified: mplayerxp/libmpcodecs/dec_video.c
===================================================================
--- mplayerxp/libmpcodecs/dec_video.c 2010-01-18 18:37:48 UTC (rev 111)
+++ mplayerxp/libmpcodecs/dec_video.c 2010-01-19 17:51:18 UTC (rev 112)
@@ -167,8 +167,9 @@
if(enable_gomp) {
smp_num_cpus=omp_get_num_procs();
use_vf_threads=0;
- MSG_DBG2("[mpdec] vf_flags=%08X num_cpus=%u\n",sh_video->vf_flags,smp_num_cpus);
if(((sh_video->vf_flags&MPDEC_THREAD_COND)==MPDEC_THREAD_COND) && (smp_num_cpus>1)) use_vf_threads=1;
+ if(use_vf_threads)
+ MSG_STATUS("[mpdec] will perform parallel video-filter on %u CPUs\n",smp_num_cpus);
}
#else
MSG_V("[mpdec] GOMP was not compiled-in! Using single threaded video filtering!\n");
Modified: mplayerxp/postproc/dsp.c
===================================================================
--- mplayerxp/postproc/dsp.c 2010-01-18 18:37:48 UTC (rev 111)
+++ mplayerxp/postproc/dsp.c 2010-01-19 17:51:18 UTC (rev 112)
@@ -20,6 +20,59 @@
#endif
#include "dsp.h"
+extern uint32_t load24bit(void* data, int pos);
+extern void store24bit(void* data, int pos, uint32_t expanded_value);
+
+/* MMX optimized stugff */
+#include <limits.h>
+#include "../mp_config.h"
+#include "../cpudetect.h"
+
+#undef OPTIMIZE_AVX
+#undef OPTIMIZE_SSE4
+#undef OPTIMIZE_SSSE3
+#undef OPTIMIZE_SSE3
+#undef OPTIMIZE_SSE2
+#undef OPTIMIZE_SSE
+#undef OPTIMIZE_MMX2
+#undef OPTIMIZE_MMX
+#define RENAME(a) a ## _c
+#include "dsp_accel.h"
+#include "dsp_accelf.h"
+
+#ifndef __x86_64__
+#ifdef __MMX__
+#define OPTIMIZE_MMX
+#undef RENAME
+#define RENAME(a) a ## _MMX
+#include "dsp_accel.h"
+#endif
+#ifdef __SSE__
+#define OPTIMIZE_MMX2
+#undef RENAME
+#define RENAME(a) a ## _MMX2
+#include "dsp_accel.h"
+#endif
+#endif //__x86_64__
+#ifdef __SSE2__
+#define OPTIMIZE_SSE2
+#undef RENAME
+#define RENAME(a) a ## _SSE2
+#include "dsp_accel.h"
+#endif
+#ifdef __SSE3__
+#define OPTIMIZE_SSE3
+#undef RENAME
+#define RENAME(a) a ## _SSE3
+#include "dsp_accel.h"
+#endif
+#ifdef __SSE4_1__
+#define OPTIMIZE_SSE4
+#undef RENAME
+#define RENAME(a) a ## _SSE4
+#include "dsp_accel.h"
+#endif
+
/******************************************************************************
* FIR filter implementations
******************************************************************************/
@@ -199,7 +252,7 @@
g += w[end-i-1] * (t3 + t2); // Total gain in filter
w[end-i-1] = w[n-end+i] = w[end-i-1] * (t2 - t3);
}
- }
+ }
else{ // Band stop
if (!o) // Band stop filters must have odd length
return -1;
@@ -219,9 +272,9 @@
// Normalize gain
g=1/g;
- for (i=0; i<n; i++)
+ for (i=0; i<n; i++)
w[i] *= g;
-
+
return 0;
}
@@ -244,7 +297,7 @@
int i; // Counters
int j;
_ftype_t t; // g * w[i]
-
+
// Sanity check
if(l<1 || k<1 || !w || !pw)
return -1;
@@ -277,7 +330,7 @@
/* Pre-warp the coefficients of a numerator or denominator.
Note that a0 is assumed to be 1, so there is no wrapping
- of it.
+ of it.
*/
void __FASTCALL__ prewarp(_ftype_t* a, _ftype_t fc, _ftype_t fs)
{
@@ -560,7 +613,7 @@
*w++ = 0.2810638602 - 0.5208971735*cos(k1*(_ftype_t)i) + 0.1980389663*cos(k2*(_ftype_t)i);
}
-/* Computes the 0th order modified Bessel function of the first kind.
+/* Computes the 0th order modified Bessel function of the first kind.
// (Needed to compute Kaiser window)
//
// y = sum( (x/(2*n))^2 )
@@ -654,81 +707,30 @@
bp->prev = bp->pprev = 0.0;
}
-extern uint32_t load24bit(void* data, int pos);
-extern void store24bit(void* data, int pos, uint32_t expanded_value);
-
-
-/* MMX optimized stugff */
-#include <limits.h>
-#include "../mp_config.h"
-#include "../cpudetect.h"
-
-#undef HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_3DNOW
-#undef HAVE_3DNOW2
-#undef HAVE_SSE
-#define RENAME(a) a ## _c
-#include "dsp_accel.h"
-
-#if defined( ARCH_X86 ) || defined(ARCH_X86_64)
-#define CAN_COMPILE_X86_ASM
+static void __FASTCALL__ init_change_bps(const void* in, void* out, unsigned len, unsigned inbps, unsigned outbps)
+{
+#ifdef __SSE4_1__
+ if(gCpuCaps.hasSSE41) change_bps = change_bps_SSE4;
+ else
#endif
-#ifdef CAN_COMPILE_X86_ASM
-//MMX versions
-#ifdef CAN_COMPILE_MMX
-#undef RENAME
-#define HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_3DNOW
-#define RENAME(a) a ## _MMX
-#include "dsp_accel.h"
+#ifdef __SSE3__
+ if(gCpuCaps.hasSSE3) change_bps = change_bps_SSE3;
+ else
#endif
-
-//3DNow! versions
-#ifdef CAN_COMPILE_3DNOW
-#undef RENAME
-#define HAVE_MMX
-#undef HAVE_MMX2
-#define HAVE_3DNOW
-#define RENAME(a) a ## _3DNow
-#include "dsp_accel.h"
+#ifdef __SSE2__
+ if(gCpuCaps.hasSSE2) change_bps = change_bps_SSE2;
+ else
#endif
-
-//3DNowEx! versions
-#ifdef CAN_COMPILE_3DNOW2
-#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
-#define HAVE_3DNOW
-#define HAVE_3DNOW2
-#define RENAME(a) a ## _3DNowEx
-#include "dsp_accel.h"
-#endif
-
-//MMX2 versions
-#ifdef CAN_COMPILE_MMX2
-#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
-#undef HAVE_3DNOW
-#define RENAME(a) a ## _MMX2
-#include "dsp_accel.h"
-#endif
-
-#endif
-
-static void __FASTCALL__ init_change_bps(const void* in, void* out, unsigned len, unsigned inbps, unsigned outbps)
-{
-#ifdef CAN_COMPILE_MMX
-/* disable these functions for speed reason !
+#ifndef __x86_64__
+#ifdef __SSE__
if(gCpuCaps.hasMMX2) change_bps = change_bps_MMX2;
else
- if(gCpuCaps.has3DNow) change_bps = change_bps_3DNow;
- else */
+#endif
+#ifdef __MMX__
if(gCpuCaps.hasMMX) change_bps = change_bps_MMX;
else
-#endif //CAN_COMPILE_X86_ASM
+#endif
+#endif /* __x86_64__ */
change_bps = change_bps_c;
(*change_bps)(in,out,len,inbps,outbps);
}
@@ -736,7 +738,7 @@
static void __FASTCALL__ init_float2int(void* in, void* out, int len, int bps)
{
-#ifdef CAN_COMPILE_X86_ASM
+#if 0
#ifdef CAN_COMPILE_3DNOW2
if(gCpuCaps.has3DNowExt) float2int = float2int_3DNowEx;
else
@@ -753,7 +755,7 @@
static void __FASTCALL__ init_int2float(void* in, void* out, int len, int bps)
{
-#ifdef CAN_COMPILE_X86_ASM
+#if 0
#ifdef CAN_COMPILE_3DNOW2
if(gCpuCaps.has3DNowExt) int2float = int2float_3DNowEx;
else
@@ -771,16 +773,20 @@
static int32_t __FASTCALL__ FIR_i16_init(int16_t *x,int16_t *w)
{
-#ifdef CAN_COMPILE_X86_ASM
-#ifdef CAN_COMPILE_MMX2
+#ifdef __SSE2__
+ if(gCpuCaps.hasSSE2) FIR_i16 = FIR_i16_SSE2;
+ else
+#endif
+#ifndef __x86_64__
+#ifdef __SSE__
if(gCpuCaps.hasMMX2) FIR_i16 = FIR_i16_MMX2;
else
#endif
-#ifdef CAN_COMPILE_MMX
+#ifdef __MMX__
if(gCpuCaps.hasMMX) FIR_i16 = FIR_i16_MMX;
else
#endif
-#endif /*CAN_COMPILE_X86_ASM*/
+#endif /*__x86_64__*/
FIR_i16 = FIR_i16_c;
return (*FIR_i16)(x,w);
}
@@ -788,7 +794,7 @@
static float __FASTCALL__ FIR_f32_init(float *x,float *w)
{
-#ifdef CAN_COMPILE_X86_ASM
+#if 0
// if(gCpuCaps.hasSSE) FIR_f32 = FIR_f32_SSE;
// else
#ifdef CAN_COMPILE_3DNOW
Modified: mplayerxp/postproc/dsp_accel.h
===================================================================
--- mplayerxp/postproc/dsp_accel.h 2010-01-18 18:37:48 UTC (rev 111)
+++ mplayerxp/postproc/dsp_accel.h 2010-01-19 17:51:18 UTC (rev 112)
@@ -1,18 +1,64 @@
/* DSP acceleration routines */
-#include "../mmx_defs.h"
+#include "pvector/pvector.h"
+#ifdef HAVE_INT_PVECTOR
+static __inline __m64 __attribute__((__gnu_inline__, __always_inline__))
+RENAME(_m_load)(const void *__P)
+{
+ return *(const __m64 *)__P;
+}
+#undef _m_load
+#define _m_load RENAME(_m_load)
+static __inline __m64 __attribute__((__gnu_inline__, __always_inline__))
+RENAME(_m_load_half)(const void *__P)
+{
+ return _mm_cvtsi32_si64 (*(const int *)__P);
+}
+#undef _m_load_half
+#define _m_load_half RENAME(_m_load_half)
+
+static __inline void __attribute__((__gnu_inline__, __always_inline__))
+RENAME(_m_store)(void *__P, __m64 src)
+{
+ *(__m64 *)__P = src;
+}
+#undef _m_store
+#define _m_store RENAME(_m_store)
+
+static __inline void __attribute__((__gnu_inline__, __always_inline__))
+RENAME(_m_store_half)(void *__P, __m64 src)
+{
+ *(int *)__P = _mm_cvtsi64_si32(src);
+}
+#undef _m_store_half
+#define _m_store_half RENAME(_m_store_half)
+
+static __inline void __attribute__((__gnu_inline__, __always_inline__))
+RENAME(_m_movntq)(void *__P, __m64 src)
+{
+#ifdef HAVE_MMX2
+ _mm_stream_pi(__P,src);
+#else
+ _m_store(__P,src);
+#endif
+}
+#undef _m_movntq
+#define _m_movntq RENAME(_m_movntq)
+#endif
+
static void __FASTCALL__ RENAME(change_bps)(const void* in_data, void* out_data, unsigned len, unsigned inbps, unsigned outbps)
{
-#ifdef HAVE_MMX
- unsigned len_mm;
+#ifdef HAVE_INT_PVECTOR
+ __ivec izero = _ivec_setzero();
+ unsigned len_mm,j;
#endif
- register unsigned i;
+ unsigned i;
// Change the number of bits
switch(inbps){
case 1:
switch(outbps){
case 2:
- i=0;
+ i=0;
for(;i<len;i++)
((uint16_t*)out_data)[i]=((uint16_t)((uint8_t*)in_data)[i])<<8;
break;
@@ -23,7 +69,7 @@
((uint8_t*)out_data)[3*i+2]=(((uint8_t*)in_data)[i]);
break;
case 4:
- i=0;
+ i=0;
for(;i<len;i++)
((uint32_t*)out_data)[i]=((uint32_t)((uint8_t*)in_data)[i])<<24;
break;
@@ -32,7 +78,7 @@
case 2:
switch(outbps){
case 1:
- i=0;
+ i=0;
for(;i<len;i++)
((uint8_t*)out_data)[i]=(uint8_t)((((uint16_t*)in_data)[i])>>8);
break;
@@ -44,48 +90,27 @@
break;
case 4:
i=0;
-#ifdef HAVE_MMX
- len_mm=len&(~0xF);
- for(;i<len_mm;i+=16)
+#ifdef HAVE_INT_PVECTOR
+ j=0;
+ len_mm=len&(~(__IVEC_SIZE-1));
+ for(;i<len;i++,j+=2){
+ ((uint32_t*)out_data)[i]=((uint32_t)((uint16_t*)in_data)[i])<<16;
+ if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ }
+ if((len_mm-i)>=__IVEC_SIZE)
+ for(;i<len_mm;i+=__IVEC_SIZE/2,j+=__IVEC_SIZE)
{
- __asm __volatile(
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "pxor %%mm4, %%mm4\n\t"
- "pxor %%mm5, %%mm5\n\t"
- "pxor %%mm6, %%mm6\n\t"
- "pxor %%mm7, %%mm7\n\t"
- "punpcklwd %%mm0, %%mm4\n\t"
- "punpckhwd %%mm0, %%mm5\n\t"
- "punpcklwd %%mm1, %%mm6\n\t"
- "punpckhwd %%mm1, %%mm7\n\t"
- MOVNTQ" %%mm4, (%0)\n\t"
- MOVNTQ" %%mm5, 8(%0)\n\t"
- MOVNTQ" %%mm6, 16(%0)\n\t"
- MOVNTQ" %%mm7, 24(%0)\n\t"
- "pxor %%mm4, %%mm4\n\t"
- "pxor %%mm5, %%mm5\n\t"
- "pxor %%mm6, %%mm6\n\t"
- "pxor %%mm7, %%mm7\n\t"
- "punpcklwd %%mm2, %%mm4\n\t"
- "punpckhwd %%mm2, %%mm5\n\t"
- "punpcklwd %%mm3, %%mm6\n\t"
- "punpckhwd %%mm3, %%mm7\n\t"
- MOVNTQ" %%mm4, 32(%0)\n\t"
- MOVNTQ" %%mm5, 40(%0)\n\t"
- MOVNTQ" %%mm6, 48(%0)\n\t"
- MOVNTQ" %%mm7, 56(%0)\n\t"
- ::"r"(&(((uint32_t*)out_data)[i])),"r"(&(((uint16_t*)in_data)[i]))
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
+ __ivec ind,tmp[2];
+ ind = _ivec_loadu(&((uint8_t *)in_data)[j]);
+#if 0 /* slower but portable on non-x86 CPUs version */
+ tmp[0]= _ivec_sll_s32_imm(_ivec_u32_from_lou16(ind),16);
+ tmp[1]= _ivec_sll_s32_imm(_ivec_u32_from_hiu16(ind),16);
+#else
+ tmp[0]= _ivec_interleave_lo_u16(izero,ind);
+ tmp[1]= _ivec_interleave_hi_u16(izero,ind);
#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
+ _ivec_storea(&((uint8_t *)out_data)[j*2],tmp[0]);
+ _ivec_storea(&((uint8_t *)out_data)[j*2+__IVEC_SIZE],tmp[1]);
}
#endif
for(;i<len;i++)
@@ -120,50 +145,27 @@
case 4:
switch(outbps){
case 1:
- i=0;
+ i=0;
for(;i<len;i++)
((uint8_t*)out_data)[i]=(uint8_t)((((uint32_t*)in_data)[i])>>24);
break;
case 2:
i=0;
-#ifdef HAVE_MMX
- len_mm=len&(~0xF);
- for(;i<len_mm;i+=16)
+#ifdef HAVE_INT_PVECTOR
+ j=0;
+ len_mm=len&(~(__IVEC_SIZE-1));
+ for(;i<len;i++,j+=2){
+ ((uint16_t*)out_data)[i]=(uint16_t)((((uint32_t*)in_data)[i])>>16);
+ if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ }
+ if((len-i)>=__IVEC_SIZE)
+ for(;i<len_mm;i+=__IVEC_SIZE/2,j+=__IVEC_SIZE)
{
- __asm __volatile(
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "movq 32(%1), %%mm4\n\t"
- "movq 40(%1), %%mm5\n\t"
- "movq 48(%1), %%mm6\n\t"
- "movq 56(%1), %%mm7\n\t"
- "psrad $16, %%mm0\n\t"
- "psrad $16, %%mm1\n\t"
- "psrad $16, %%mm2\n\t"
- "psrad $16, %%mm3\n\t"
- "psrad $16, %%mm4\n\t"
- "psrad $16, %%mm5\n\t"
- "psrad $16, %%mm6\n\t"
- "psrad $16, %%mm7\n\t"
- "packssdw %%mm1, %%mm0\n\t"
- "packssdw %%mm3, %%mm2\n\t"
- "packssdw %%mm5, %%mm4\n\t"
- "packssdw %%mm7, %%mm6\n\t"
- MOVNTQ" %%mm0, (%0)\n\t"
- MOVNTQ" %%mm2, 8(%0)\n\t"
- MOVNTQ" %%mm4, 16(%0)\n\t"
- MOVNTQ" %%mm6, 24(%0)\n\t"
- ::"r"(&(((uint16_t*)out_data)[i])),"r"(&(((uint32_t*)in_data)[i]))
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
+ __ivec ind[2],tmp;
+ ind[0]= _ivec_sra_s32_imm(_ivec_loadu(&((uint8_t *)in_data)[j*2]),16);
+ ind[1]= _ivec_sra_s32_imm(_ivec_loadu(&((uint8_t *)in_data)[j*2+__IVEC_SIZE]),16);
+ tmp = _ivec_s16_from_s32(ind[0],ind[1]);
+ _ivec_storea(&((uint8_t *)out_data)[j],tmp);
}
#endif
for(;i<len;i++)
@@ -176,364 +178,40 @@
((uint8_t*)out_data)[3*i+2]=(((uint8_t*)in_data)[4*i+3]);
break;
}
- break;
+ break;
}
-#ifdef HAVE_MMX
-#ifndef HAVE_MMX1
- asm volatile(SFENCE:::"memory");
+#ifdef HAVE_INT_PVECTOR
+ _ivec_sfence();
+ _ivec_empty();
#endif
- asm volatile(EMMS::
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
-#endif /* HAVE_MMX */
}
-static void __FASTCALL__ RENAME(float2int)(void* in, void* out, int len, int bps)
+static int32_t __FASTCALL__ RENAME(FIR_i16)(int16_t *x,int16_t *w)
{
-#ifdef HAVE_3DNOW
- unsigned len_mm;
- float tmp_f32[2];
-#endif
- float ftmp;
- register int i;
- switch(bps){
- case(1):
- for(i=0;i<len;i++) {
- ftmp=((float*)in)[i];
- SATURATE(ftmp,-1.0,+1.0);
- ((int8_t*)out)[i]=(int8_t)lrintf(SCHAR_MAX*ftmp);
- }
- break;
- case(2):
- i=0;
-#ifdef HAVE_3DNOW
- len_mm=len&(~15);
- tmp_f32[0]=
- tmp_f32[1]=SHRT_MAX;
- for(;i<len_mm;i+=16)
- {
- __asm __volatile(
- PREFETCH" 64(%1)\n\t"
- PREFETCHW" 32(%0)\n\t"
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "movq 32(%1), %%mm4\n\t"
- "movq 40(%1), %%mm5\n\t"
- "movq 48(%1), %%mm6\n\t"
- "movq 56(%1), %%mm7\n\t"
- "pfmul %2, %%mm0\n\t"
- "pfmul %2, %%mm1\n\t"
- "pfmul %2, %%mm2\n\t"
- "pfmul %2, %%mm3\n\t"
- "pfmul %2, %%mm4\n\t"
- "pfmul %2, %%mm5\n\t"
- "pfmul %2, %%mm6\n\t"
- "pfmul %2, %%mm7\n\t"
- "pf2id %%mm0, %%mm0\n\t"
- "pf2id %%mm1, %%mm1\n\t"
- "pf2id %%mm2, %%mm2\n\t"
- "pf2id %%mm3, %%mm3\n\t"
- "pf2id %%mm4, %%mm4\n\t"
- "pf2id %%mm5, %%mm5\n\t"
- "pf2id %%mm6, %%mm6\n\t"
- "pf2id %%mm7, %%mm7\n\t"
- "packssdw %%mm1, %%mm0\n\t"
- "packssdw %%mm3, %%mm2\n\t"
- "packssdw %%mm5, %%mm4\n\t"
- "packssdw %%mm7, %%mm6\n\t"
- "movq %%mm0, (%0)\n\t"
- "movq %%mm2, 8(%0)\n\t"
- "movq %%mm4, 16(%0)\n\t"
- "movq %%mm6, 24(%0)"
- ::"r"(&(((uint16_t*)out)[i])),"r"(&(((float*)in)[i])),"m"(tmp_f32[0])
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- }
-#endif
- for(;i<len;i++) {
- ftmp=((float*)in)[i];
- SATURATE(ftmp,-1.0,+1.0);
- ((int16_t*)out)[i]=(int16_t)lrintf(SHRT_MAX*ftmp);
- }
- break;
- case(3):
- for(i=0;i<len;i++) {
- ftmp=((float*)in)[i];
- SATURATE(ftmp,-1.0,+1.0);
- store24bit(out, i, (int32_t)lrintf((INT_MAX-1)*ftmp));
- }
- break;
- case(4):
- i=0;
-#ifdef HAVE_3DNOW
- len_mm=len&(~15);
- tmp_f32[0]=
- tmp_f32[1]=INT_MAX;
- for(;i<len_mm;i+=16)
- {
- __asm __volatile(
- PREFETCH" 64(%1)\n\t"
- PREFETCHW" 64(%0)\n\t"
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "movq 32(%1), %%mm4\n\t"
- "movq 40(%1), %%mm5\n\t"
- "movq 48(%1), %%mm6\n\t"
- "movq 56(%1), %%mm7\n\t"
- "pfmul %2, %%mm0\n\t"
- "pfmul %2, %%mm1\n\t"
- "pfmul %2, %%mm2\n\t"
- "pfmul %2, %%mm3\n\t"
- "pfmul %2, %%mm4\n\t"
- "pfmul %2, %%mm5\n\t"
- "pfmul %2, %%mm6\n\t"
- "pfmul %2, %%mm7\n\t"
- "pf2id %%mm0, %%mm0\n\t"
- "pf2id %%mm1, %%mm1\n\t"
- "pf2id %%mm2, %%mm2\n\t"
- "pf2id %%mm3, %%mm3\n\t"
- "pf2id %%mm4, %%mm4\n\t"
- "pf2id %%mm5, %%mm5\n\t"
- "pf2id %%mm6, %%mm6\n\t"
- "pf2id %%mm7, %%mm7\n\t"
- "movq %%mm0, (%0)\n\t"
- "movq %%mm1, 8(%0)\n\t"
- "movq %%mm2, 16(%0)\n\t"
- "movq %%mm3, 24(%0)\n\t"
- "movq %%mm4, 32(%0)\n\t"
- "movq %%mm5, 40(%0)\n\t"
- "movq %%mm6, 48(%0)\n\t"
- "movq %%mm7, 56(%0)"
- ::"r"(&(((uint32_t*)out)[i])),"r"(&(((float*)in)[i])),"m"(tmp_f32[0])
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- }
-#endif
- for(;i<len;i++) {
- ftmp=((float*)in)[i];
- SATURATE(ftmp,-1.0,+1.0);
- ((int32_t*)out)[i]=(int32_t)lrintf((INT_MAX-1)*ftmp);
- }
- break;
- }
-#ifdef HAVE_3DNOW
- asm volatile(EMMS::
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
-#endif
-}
+#ifdef OPTIMIZE_MMX
+ __m64 mm[8];
+ mm[0] = _m_load(&w[0]);
+ mm[1] = _m_load(&w[4]);
+ mm[2] = _m_load(&w[8]);
+ mm[3] = _m_load(&w[12]);
-static void __FASTCALL__ RENAME(int2float)(void* in, void* out, int len, int bps)
-{
-#ifdef HAVE_3DNOW
- unsigned len_mm;
- float tmp_f32[2];
-#endif
- register int i;
- switch(bps){
- case(1):
- for(i=0;i<len;i++)
- ((float*)out)[i]=(1.0/SCHAR_MAX)*((float)((int8_t*)in)[i]);
- break;
- case(2):
- i=0;
-#ifdef HAVE_3DNOW
- tmp_f32[0]=
- tmp_f32[1]=1.0/INT_MAX;
- len_mm=len&(~15);
- for(;i<len_mm;i+=16)
- {
- __asm __volatile(
- PREFETCH" 32(%1)\n\t"
- PREFETCHW" 64(%0)\n\t"
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "pxor %%mm4, %%mm4\n\t"
- "pxor %%mm5, %%mm5\n\t"
- "pxor %%mm6, %%mm6\n\t"
- "pxor %%mm7, %%mm7\n\t"
- "punpcklwd %%mm0, %%mm4\n\t"
- "punpckhwd %%mm0, %%mm5\n\t"
- "punpcklwd %%mm1, %%mm6\n\t"
- "punpckhwd %%mm1, %%mm7\n\t"
- "pi2fd %%mm4, %%mm4\n\t"
- "pi2fd %%mm5, %%mm5\n\t"
- "pi2fd %%mm6, %%mm6\n\t"
- "pi2fd %%mm7, %%mm7\n\t"
- "pfmul %2, %%mm4\n\t"
- "pfmul %2, %%mm5\n\t"
- "pfmul %2, %%mm6\n\t"
- "pfmul %2, %%mm7\n\t"
- "movq %%mm4, (%0)\n\t"
- "movq %%mm5, 8(%0)\n\t"
- "movq %%mm6, 16(%0)\n\t"
- "movq %%mm7, 24(%0)\n\t"
- "pxor %%mm4, %%mm4\n\t"
- "pxor %%mm5, %%mm5\n\t"
- "pxor %%mm6, %%mm6\n\t"
- "pxor %%mm7, %%mm7\n\t"
- "punpcklwd %%mm2, %%mm4\n\t"
- "punpckhwd %%mm2, %%mm5\n\t"
- "punpcklwd %%mm3, %%mm6\n\t"
- "punpckhwd %%mm3, %%mm7\n\t"
- "pi2fd %%mm4, %%mm4\n\t"
- "pi2fd %%mm5, %%mm5\n\t"
- "pi2fd %%mm6, %%mm6\n\t"
- "pi2fd %%mm7, %%mm7\n\t"
- "pfmul %2, %%mm4\n\t"
- "pfmul %2, %%mm5\n\t"
- "pfmul %2, %%mm6\n\t"
- "pfmul %2, %%mm7\n\t"
- "movq %%mm4, 32(%0)\n\t"
- "movq %%mm5, 40(%0)\n\t"
- "movq %%mm6, 48(%0)\n\t"
- "movq %%mm7, 56(%0)\n\t"
- "femms"
- ::"r"(&(((float*)out)[i])),"r"(&(((int16_t*)in)[i])),"m"(tmp_f32[0])
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- }
-#endif
- for(;i<len;i++)
- ((float*)out)[i]=(1.0/SHRT_MAX)*((float)((int16_t*)in)[i]);
- break;
- case(3):
- for(i=0;i<len;i++)
- ((float*)out)[i]=(1.0/INT_MAX)*((float)((int32_t)load24bit(in, i)));
- break;
- case(4):
- i=0;
-#ifdef HAVE_3DNOW
- tmp_f32[0]=
- tmp_f32[1]=1.0/INT_MAX;
- len_mm=len&(~15);
- for(;i<len_mm;i+=16)
- {
- __asm __volatile(
- PREFETCH" 64(%1)\n\t"
- PREFETCHW" 64(%0)\n\t"
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "movq 32(%1), %%mm4\n\t"
- "movq 40(%1), %%mm5\n\t"
- "movq 48(%1), %%mm6\n\t"
- "movq 56(%1), %%mm7\n\t"
- "pi2fd %%mm0, %%mm0\n\t"
- "pi2fd %%mm1, %%mm1\n\t"
- "pi2fd %%mm2, %%mm2\n\t"
- "pi2fd %%mm3, %%mm3\n\t"
- "pi2fd %%mm4, %%mm4\n\t"
- "pi2fd %%mm5, %%mm5\n\t"
- "pi2fd %%mm6, %%mm6\n\t"
- "pi2fd %%mm7, %%mm7\n\t"
- "pfmul %2, %%mm0\n\t"
- "pfmul %2, %%mm1\n\t"
- "pfmul %2, %%mm2\n\t"
- "pfmul %2, %%mm3\n\t"
- "pfmul %2, %%mm4\n\t"
- "pfmul %2, %%mm5\n\t"
- "pfmul %2, %%mm6\n\t"
- "pfmul %2, %%mm7\n\t"
- "movq %%mm0, (%0)\n\t"
- "movq %%mm1, 8(%0)\n\t"
- "movq %%mm2, 16(%0)\n\t"
- "movq %%mm3, 24(%0)\n\t"
- "movq %%mm4, 32(%0)\n\t"
- "movq %%mm5, 40(%0)\n\t"
- "movq %%mm6, 48(%0)\n\t"
- "movq %%mm7, 56(%0)\n\t"
- "femms"
- ::"r"(&(((float*)out)[i])),"r"(&(((int32_t*)in)[i])),"m"(tmp_f32[0])
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- }
-#endif
- for(;i<len;i++)
- ((float*)out)[i]=(1.0/INT_MAX)*((float)((int32_t*)in)[i]);
- break;
- }
-}
+ mm[4] = _m_pmaddwd(mm[0],_m_load(&x[0]));
+ mm[5] = _m_pmaddwd(mm[1],_m_load(&x[4]));
+ mm[6] = _m_pmaddwd(mm[2],_m_load(&x[8]));
+ mm[7] = _m_pmaddwd(mm[3],_m_load(&x[12]));
-static int32_t __FASTCALL__ RENAME(FIR_i16)(int16_t *x,int16_t *w)
-{
-#ifdef HAVE_MMX
- int32_t rval;
- __asm __volatile(
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "pmaddwd (%2), %%mm0\n\t"
- "pmaddwd 8(%2), %%mm1\n\t"
- "pmaddwd 16(%2), %%mm2\n\t"
- "pmaddwd 24(%2), %%mm3\n\t"
- "paddd %%mm1, %%mm0\n\t"
- "paddd %%mm3, %%mm2\n\t"
- "paddd %%mm2, %%mm0\n\t"
-#ifdef HAVE_MMX2
- "pshufw $0xFE, %%mm0, %%mm1\n\t"
+ mm[0] = _m_paddd(mm[4],mm[5]);
+ mm[1] = _m_paddd(mm[6],mm[7]);
+ mm[2] = _m_paddd(mm[0],mm[1]);
+#ifdef OPTIMIZE_MMX2
+ mm[0] = _m_pshufw(mm[2],0xFE);
#else
- "movq %%mm0, %%mm1\n\t"
- "psrlq $32, %%mm1\n\t"
+ mm[0] = mm[2];
+ mm[0] = _m_psrlqi(mm[0],32);
#endif
- "paddd %%mm1, %%mm0\n\t"
- "psrld $16, %%mm0\n\t"
- "movd %%mm0, %0\n\t"
- "emms"
- :"=&r"(rval):"r"(w),"r"(x)
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- return rval;
+ mm[0] = _m_paddd(mm[0],mm[2]);
+ mm[0] = _m_psrldi(mm[0],16);
+ return _mm_cvtsi64_si32(mm[0]);
#else
return ( w[0] *x[0] +w[1] *x[1] +w[2] *x[2] +w[3] *x[3]
+ w[4] *x[4] +w[5] *x[5] +w[6] *x[6] +w[7] *x[7]
@@ -542,51 +220,3 @@
#endif
}
-static float __FASTCALL__ RENAME(FIR_f32)(float *x,float *w)
-{
-#ifdef HAVE_3DNOW
- float rval;
- __asm __volatile(
- "movq (%1), %%mm0\n\t"
- "movq 8(%1), %%mm1\n\t"
- "movq 16(%1), %%mm2\n\t"
- "movq 24(%1), %%mm3\n\t"
- "movq 32(%1), %%mm4\n\t"
- "movq 40(%1), %%mm5\n\t"
- "movq 48(%1), %%mm6\n\t"
- "movq 56(%1), %%mm7\n\t"
- "pfmul (%2), %%mm0\n\t"
- "pfmul 8(%2), %%mm1\n\t"
- "pfmul 16(%2), %%mm2\n\t"
- "pfmul 24(%2), %%mm3\n\t"
- "pfmul 32(%2), %%mm4\n\t"
- "pfmul 40(%2), %%mm5\n\t"
- "pfmul 48(%2), %%mm6\n\t"
- "pfmul 56(%2), %%mm7\n\t"
- "pfadd %%mm1, %%mm0\n\t"
- "pfadd %%mm3, %%mm2\n\t"
- "pfadd %%mm5, %%mm4\n\t"
- "pfadd %%mm7, %%mm6\n\t"
- "pfadd %%mm2, %%mm0\n\t"
- "pfadd %%mm6, %%mm4\n\t"
- "pfadd %%mm4, %%mm0\n\t"
- "pfacc %%mm0, %%mm0\n\t"
- "movd %%mm0, %0\n\t"
- "femms"
- :"=&r"(rval):"r"(w),"r"(x)
- :"memory"
-#ifdef FPU_CLOBBERED
- ,FPU_CLOBBERED
-#endif
-#ifdef MMX_CLOBBERED
- ,MMX_CLOBBERED
-#endif
- );
- return rval;
-#else
- return ( w[0] *x[0] +w[1] *x[1] +w[2] *x[2] +w[3] *x[3]
- + w[4] *x[4] +w[5] *x[5] +w[6] *x[6] +w[7] *x[7]
- + w[8] *x[8] +w[9] *x[9] +w[10]*x[10]+w[11]*x[11]
- + w[12]*x[12]+w[13]*x[13]+w[14]*x[14]+w[15]*x[15] );
-#endif
-}
Added: mplayerxp/postproc/dsp_accelf.h
===================================================================
--- mplayerxp/postproc/dsp_accelf.h (rev 0)
+++ mplayerxp/postproc/dsp_accelf.h 2010-01-19 17:51:18 UTC (rev 112)
@@ -0,0 +1,354 @@
+/* DSP floating-point acceleration routines */
+
+static void __FASTCALL__ RENAME(float2int)(void* in, void* out, int len, int bps)
+{
+#ifdef HAVE_3DNOW
+ unsigned len_mm;
+ float tmp_f32[2];
+#endif
+ float ftmp;
+ register int i;
+ switch(bps){
+ case(1):
+ for(i=0;i<len;i++) {
+ ftmp=((float*)in)[i];
+ SATURATE(ftmp,-1.0,+1.0);
+ ((int8_t*)out)[i]=(int8_t)lrintf(SCHAR_MAX*ftmp);
+ }
+ break;
+ case(2):
+ i=0;
+#ifdef HAVE_3DNOW
+ len_mm=len&(~15);
+ tmp_f32[0]=
+ tmp_f32[1]=SHRT_MAX;
+ for(;i<len_mm;i+=16)
+ {
+ __asm __volatile(
+ PREFETCH" 64(%1)\n\t"
+ PREFETCHW" 32(%0)\n\t"
+ "movq (%1), %%mm0\n\t"
+ "movq 8(%1), %%mm1\n\t"
+ "movq 16(%1), %%mm2\n\t"
+ "movq 24(%1), %%mm3\n\t"
+ "movq 32(%1), %%mm4\n\t"
+ "movq 40(%1), %%mm5\n\t"
+ "movq 48(%1), %%mm6\n\t"
+ "movq 56(%1), %%mm7\n\t"
+ "pfmul %2, %%mm0\n\t"
+ "pfmul %2, %%mm1\n\t"
+ "pfmul %2, %%mm2\n\t"
+ "pfmul %2, %%mm3\n\t"
+ "pfmul %2, %%mm4\n\t"
+ "pfmul %2, %%mm5\n\t"
+ "pfmul %2, %%mm6\n\t"
+ "pfmul %2, %%mm7\n\t"
+ "pf2id %%mm0, %%mm0\n\t"
+ "pf2id %%mm1, %%mm1\n\t"
+ "pf2id %%mm2, %%mm2\n\t"
+ "pf2id %%mm3, %%mm3\n\t"
+ "pf2id %%mm4, %%mm4\n\t"
+ "pf2id %%mm5, %%mm5\n\t"
+ "pf2id %%mm6, %%mm6\n\t"
+ "pf2id %%mm7, %%mm7\n\t"
+ "packssdw %%mm1, %%mm0\n\t"
+ "packssdw %%mm3, %%mm2\n\t"
+ "packssdw %%mm5, %%mm4\n\t"
+ "packssdw %%mm7, %%mm6\n\t"
+ "movq %%mm0, (%0)\n\t"
+ "movq %%mm2, 8(%0)\n\t"
+ "movq %%mm4, 16(%0)\n\t"
+ "movq %%mm6, 24(%0)"
+ ::"r"(&(((uint16_t*)out)[i])),"r"(&(((float*)in)[i])),"m"(tmp_f32[0])
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+ }
+#endif
+ for(;i<len;i++) {
+ ftmp=((float*)in)[i];
+ SATURATE(ftmp,-1.0,+1.0);
+ ((int16_t*)out)[i]=(int16_t)lrintf(SHRT_MAX*ftmp);
+ }
+ break;
+ case(3):
+ for(i=0;i<len;i++) {
+ ftmp=((float*)in)[i];
+ SATURATE(ftmp,-1.0,+1.0);
+ store24bit(out, i, (int32_t)lrintf((INT_MAX-1)*ftmp));
+ }
+ break;
+ case(4):
+ i=0;
+#ifdef HAVE_3DNOW
+ len_mm=len&(~15);
+ tmp_f32[0]=
+ tmp_f32[1]=INT_MAX;
+ for(;i<len_mm;i+=16)
+ {
+ __asm __volatile(
+ PREFETCH" 64(%1)\n\t"
+ PREFETCHW" 64(%0)\n\t"
+ "movq (%1), %%mm0\n\t"
+ "movq 8(%1), %%mm1\n\t"
+ "movq 16(%1), %%mm2\n\t"
+ "movq 24(%1), %%mm3\n\t"
+ "movq 32(%1), %%mm4\n\t"
+ "movq 40(%1), %%mm5\n\t"
+ "movq 48(%1), %%mm6\n\t"
+ "movq 56(%1), %%mm7\n\t"
+ "pfmul %2, %%mm0\n\t"
+ "pfmul %2, %%mm1\n\t"
+ "pfmul %2, %%mm2\n\t"
+ "pfmul %2, %%mm3\n\t"
+ "pfmul %2, %%mm4\n\t"
+ "pfmul %2, %%mm5\n\t"
+ "pfmul %2, %%mm6\n\t"
+ "pfmul %2, %%mm7\n\t"
+ "pf2id %%mm0, %%mm0\n\t"
+ "pf2id %%mm1, %%mm1\n\t"
+ "pf2id %%mm2, %%mm2\n\t"
+ "pf2id %%mm3, %%mm3\n\t"
+ "pf2id %%mm4, %%mm4\n\t"
+ "pf2id %%mm5, %%mm5\n\t"
+ "pf2id %%mm6, %%mm6\n\t"
+ "pf2id %%mm7, %%mm7\n\t"
+ "movq %%mm0, (%0)\n\t"
+ "movq %%mm1, 8(%0)\n\t"
+ "movq %%mm2, 16(%0)\n\t"
+ "movq %%mm3, 24(%0)\n\t"
+ "movq %%mm4, 32(%0)\n\t"
+ "movq %%mm5, 40(%0)\n\t"
+ "movq %%mm6, 48(%0)\n\t"
+ "movq %%mm7, 56(%0)"
+ ::"r"(&(((uint32_t*)out)[i])),"r"(&(((float*)in)[i])),"m"(tmp_f32[0])
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+ }
+#endif
+ for(;i<len;i++) {
+ ftmp=((float*)in)[i];
+ SATURATE(ftmp,-1.0,+1.0);
+ ((int32_t*)out)[i]=(int32_t)lrintf((INT_MAX-1)*ftmp);
+ }
+ break;
+ }
+#ifdef HAVE_3DNOW
+ asm volatile(EMMS::
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+#endif
+}
+
+static void __FASTCALL__ RENAME(int2float)(void* in, void* out, int len, int bps)
+{
+#ifdef HAVE_3DNOW
+ unsigned len_mm;
+ float tmp_f32[2];
+#endif
+ register int i;
+ switch(bps){
+ case(1):
+ for(i=0;i<len;i++)
+ ((float*)out)[i]=(1.0/SCHAR_MAX)*((float)((int8_t*)in)[i]);
+ break;
+ case(2):
+ i=0;
+#ifdef HAVE_3DNOW
+ tmp_f32[0]=
+ tmp_f32[1]=1.0/INT_MAX;
+ len_mm=len&(~15);
+ for(;i<len_mm;i+=16)
+ {
+ __asm __volatile(
+ PREFETCH" 32(%1)\n\t"
+ PREFETCHW" 64(%0)\n\t"
+ "movq (%1), %%mm0\n\t"
+ "movq 8(%1), %%mm1\n\t"
+ "movq 16(%1), %%mm2\n\t"
+ "movq 24(%1), %%mm3\n\t"
+ "pxor %%mm4, %%mm4\n\t"
+ "pxor %%mm5, %%mm5\n\t"
+ "pxor %%mm6, %%mm6\n\t"
+ "pxor %%mm7, %%mm7\n\t"
+ "punpcklwd %%mm0, %%mm4\n\t"
+ "punpckhwd %%mm0, %%mm5\n\t"
+ "punpcklwd %%mm1, %%mm6\n\t"
+ "punpckhwd %%mm1, %%mm7\n\t"
+ "pi2fd %%mm4, %%mm4\n\t"
+ "pi2fd %%mm5, %%mm5\n\t"
+ "pi2fd %%mm6, %%mm6\n\t"
+ "pi2fd %%mm7, %%mm7\n\t"
+ "pfmul %2, %%mm4\n\t"
+ "pfmul %2, %%mm5\n\t"
+ "pfmul %2, %%mm6\n\t"
+ "pfmul %2, %%mm7\n\t"
+ "movq %%mm4, (%0)\n\t"
+ "movq %%mm5, 8(%0)\n\t"
+ "movq %%mm6, 16(%0)\n\t"
+ "movq %%mm7, 24(%0)\n\t"
+ "pxor %%mm4, %%mm4\n\t"
+ "pxor %%mm5, %%mm5\n\t"
+ "pxor %%mm6, %%mm6\n\t"
+ "pxor %%mm7, %%mm7\n\t"
+ "punpcklwd %%mm2, %%mm4\n\t"
+ "punpckhwd %%mm2, %%mm5\n\t"
+ "punpcklwd %%mm3, %%mm6\n\t"
+ "punpckhwd %%mm3, %%mm7\n\t"
+ "pi2fd %%mm4, %%mm4\n\t"
+ "pi2fd %%mm5, %%mm5\n\t"
+ "pi2fd %%mm6, %%mm6\n\t"
+ "pi2fd %%mm7, %%mm7\n\t"
+ "pfmul %2, %%mm4\n\t"
+ "pfmul %2, %%mm5\n\t"
+ "pfmul %2, %%mm6\n\t"
+ "pfmul %2, %%mm7\n\t"
+ "movq %%mm4, 32(%0)\n\t"
+ "movq %%mm5, 40(%0)\n\t"
+ "movq %%mm6, 48(%0)\n\t"
+ "movq %%mm7, 56(%0)\n\t"
+ "femms"
+ ::"r"(&(((float*)out)[i])),"r"(&(((int16_t*)in)[i])),"m"(tmp_f32[0])
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+ }
+#endif
+ for(;i<len;i++)
+ ((float*)out)[i]=(1.0/SHRT_MAX)*((float)((int16_t*)in)[i]);
+ break;
+ case(3):
+ for(i=0;i<len;i++)
+ ((float*)out)[i]=(1.0/INT_MAX)*((float)((int32_t)load24bit(in, i)));
+ break;
+ case(4):
+ i=0;
+#ifdef HAVE_3DNOW
+ tmp_f32[0]=
+ tmp_f32[1]=1.0/INT_MAX;
+ len_mm=len&(~15);
+ for(;i<len_mm;i+=16)
+ {
+ __asm __volatile(
+ PREFETCH" 64(%1)\n\t"
+ PREFETCHW" 64(%0)\n\t"
+ "movq (%1), %%mm0\n\t"
+ "movq 8(%1), %%mm1\n\t"
+ "movq 16(%1), %%mm2\n\t"
+ "movq 24(%1), %%mm3\n\t"
+ "movq 32(%1), %%mm4\n\t"
+ "movq 40(%1), %%mm5\n\t"
+ "movq 48(%1), %%mm6\n\t"
+ "movq 56(%1), %%mm7\n\t"
+ "pi2fd %%mm0, %%mm0\n\t"
+ "pi2fd %%mm1, %%mm1\n\t"
+ "pi2fd %%mm2, %%mm2\n\t"
+ "pi2fd %%mm3, %%mm3\n\t"
+ "pi2fd %%mm4, %%mm4\n\t"
+ "pi2fd %%mm5, %%mm5\n\t"
+ "pi2fd %%mm6, %%mm6\n\t"
+ "pi2fd %%mm7, %%mm7\n\t"
+ "pfmul %2, %%mm0\n\t"
+ "pfmul %2, %%mm1\n\t"
+ "pfmul %2, %%mm2\n\t"
+ "pfmul %2, %%mm3\n\t"
+ "pfmul %2, %%mm4\n\t"
+ "pfmul %2, %%mm5\n\t"
+ "pfmul %2, %%mm6\n\t"
+ "pfmul %2, %%mm7\n\t"
+ "movq %%mm0, (%0)\n\t"
+ "movq %%mm1, 8(%0)\n\t"
+ "movq %%mm2, 16(%0)\n\t"
+ "movq %%mm3, 24(%0)\n\t"
+ "movq %%mm4, 32(%0)\n\t"
+ "movq %%mm5, 40(%0)\n\t"
+ "movq %%mm6, 48(%0)\n\t"
+ "movq %%mm7, 56(%0)\n\t"
+ "femms"
+ ::"r"(&(((float*)out)[i])),"r"(&(((int32_t*)in)[i])),"m"(tmp_f32[0])
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+ }
+#endif
+ for(;i<len;i++)
+ ((float*)out)[i]=(1.0/INT_MAX)*((float)((int32_t*)in)[i]);
+ break;
+ }
+}
+
+static float __FASTCALL__ RENAME(FIR_f32)(float *x,float *w)
+{
+#ifdef HAVE_3DNOW
+ float rval;
+ __asm __volatile(
+ "movq (%1), %%mm0\n\t"
+ "movq 8(%1), %%mm1\n\t"
+ "movq 16(%1), %%mm2\n\t"
+ "movq 24(%1), %%mm3\n\t"
+ "movq 32(%1), %%mm4\n\t"
+ "movq 40(%1), %%mm5\n\t"
+ "movq 48(%1), %%mm6\n\t"
+ "movq 56(%1), %%mm7\n\t"
+ "pfmul (%2), %%mm0\n\t"
+ "pfmul 8(%2), %%mm1\n\t"
+ "pfmul 16(%2), %%mm2\n\t"
+ "pfmul 24(%2), %%mm3\n\t"
+ "pfmul 32(%2), %%mm4\n\t"
+ "pfmul 40(%2), %%mm5\n\t"
+ "pfmul 48(%2), %%mm6\n\t"
+ "pfmul 56(%2), %%mm7\n\t"
+ "pfadd %%mm1, %%mm0\n\t"
+ "pfadd %%mm3, %%mm2\n\t"
+ "pfadd %%mm5, %%mm4\n\t"
+ "pfadd %%mm7, %%mm6\n\t"
+ "pfadd %%mm2, %%mm0\n\t"
+ "pfadd %%mm6, %%mm4\n\t"
+ "pfadd %%mm4, %%mm0\n\t"
+ "pfacc %%mm0, %%mm0\n\t"
+ "movd %%mm0, %0\n\t"
+ "femms"
+ :"=&r"(rval):"r"(w),"r"(x)
+ :"memory"
+#ifdef FPU_CLOBBERED
+ ,FPU_CLOBBERED
+#endif
+#ifdef MMX_CLOBBERED
+ ,MMX_CLOBBERED
+#endif
+ );
+ return rval;
+#else
+ return ( w[0] *x[0] +w[1] *x[1] +w[2] *x[2] +w[3] *x[3]
+ + w[4] *x[4] +w[5] *x[5] +w[6] *x[6] +w[7] *x[7]
+ + w[8] *x[8] +w[9] *x[9] +w[10]*x[10]+w[11]*x[11]
+ + w[12]*x[12]+w[13]*x[13]+w[14]*x[14]+w[15]*x[15] );
+#endif
+}
Property changes on: mplayerxp/postproc/dsp_accelf.h
___________________________________________________________________
Added: svn:eol-style
+ native
Modified: mplayerxp/postproc/vf.c
===================================================================
--- mplayerxp/postproc/vf.c 2010-01-18 18:37:48 UTC (rev 111)
+++ mplayerxp/postproc/vf.c 2010-01-19 17:51:18 UTC (rev 112)
@@ -618,7 +618,7 @@
{
MSG_V("vf_reinit->config %i %i %s=> %i %i %s\n",sw,sh,vo_format_name(sfourcc),w,h,vo_format_name(fmt));
_saved=_this->prev;
- vf_scaler=vf_open_filter(_this,sh_video,"scale",NULL);
+ vf_scaler=vf_open_filter(_this,sh_video,(w==sw&&h==sh)?"fmtcvt":"scale",NULL);
if(vf_scaler)
{
void *sfnc;
Modified: mplayerxp/postproc/vf_scale.c
===================================================================
--- mplayerxp/postproc/vf_scale.c 2010-01-18 18:37:48 UTC (rev 111)
+++ mplayerxp/postproc/vf_scale.c 2010-01-19 17:51:18 UTC (rev 112)
@@ -118,11 +118,18 @@
static void __FASTCALL__ print_conf(struct vf_instance_s* vf)
{
- MSG_INFO("[vf_scale]: scaling [%dx%d,%s] -> [%dx%d,%s]\n",
+ MSG_INFO("[vf_scale]: in[%dx%d,%s] -> out[%dx%d,%s]\n",
vf->priv->sw,vf->priv->sh,vo_format_name(vf->priv->sfmt),
vf->priv->w,vf->priv->h,vo_format_name(vf->priv->ofmt));
}
+static void __FASTCALL__ print_conf_fmtcvt(struct vf_instance_s* vf)
+{
+ MSG_INFO("[vf_fmtcvt]: video[%dx%d] in[%s] -> out[%s]\n",
+ vf->priv->sw,vf->priv->sh,vo_format_name(vf->priv->sfmt),
+ vo_format_name(vf->priv->ofmt));
+}
+
static int __FASTCALL__ config(struct vf_instance_s* vf,
int width, int height, int d_width, int d_height,
unsigned int flags, unsigned int outfmt,void *tune){
@@ -272,7 +279,7 @@
vf->priv->palette[4*i+3]=0;
}
break; }
- case IMGFMT_RGB4:
+ case IMGFMT_RGB4:
case IMGFMT_RG4B: {
int i;
vf->priv->palette=malloc(4*16);
@@ -321,7 +328,7 @@
}
}
-static int __FASTCALL__ put_slice(struct vf_instance_s* vf, mp_image_t *mpi){
+static int __FASTCALL__ put_frame(struct vf_instance_s* vf, mp_image_t *mpi){
mp_image_t *dmpi;//=mpi->priv;
uint8_t *planes[3];
int stride[3];
@@ -342,7 +349,7 @@
stride[2]=mpi->stride[2];
}
}
- MSG_DBG2("vf_scale.put_slice was called\n");
+ MSG_DBG2("vf_scale.put_frame was called\n");
dmpi=vf_get_image(vf->next,vf->priv->fmt,
MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE | MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
vf->priv->w, vf->priv->h);
@@ -350,6 +357,51 @@
return vf_next_put_slice(vf,dmpi);
}
+static int __FASTCALL__ put_slice(struct vf_instance_s* vf, mp_image_t *mpi){
+ mp_image_t *dmpi;//=mpi->priv;
+ uint8_t *planes[3],*dplanes[3];
+ int stride[3],newy,newh;
+ planes[0]=mpi->planes[0];
+ stride[0]=mpi->stride[0];
+ if(mpi->flags&MP_IMGFLAG_PLANAR){
+ if(mpi->flags&MP_IMGFLAG_SWAPPED){
+ // I420/IYUV (Y,U,V)
+ planes[1]=mpi->planes[2];
+ planes[2]=mpi->planes[1];
+ stride[1]=mpi->stride[2];
+ stride[2]=mpi->stride[1];
+ } else {
+ // YV12,YVU9,IF09 (Y,V,U)
+ planes[1]=mpi->planes[1];
+ planes[2]=mpi->planes[2];
+ stride[1]=mpi->stride[1];
+ stride[2]=mpi->stride[2];
+ }
+ }
+ MSG_DBG2("vf_scale.put_slice was called[%i %i]\n",mpi->y, mpi->h);
+ dmpi=vf_get_image(vf->next,vf->priv->fmt,
+ MP_IMGTYPE_TEMP, MP_IMGFLAG_ACCEPT_STRIDE | MP_IMGFLAG_PREFER_ALIGNED_STRIDE,
+ vf->priv->w, vf->priv->h);
+ /* Try to fake first slice*/
+ dplanes[0] = dmpi->planes[0];
+ if(mpi->flags&MP_IMGFLAG_PLANAR) {
+ dplanes[1] = dmpi->planes[1];
+ dplanes[2] = dmpi->planes[2];
+ }
+ planes[0] += mpi->y*mpi->stride[0];
+ dplanes[0] += mpi->y*dmpi->stride[0];
+ if(mpi->flags&MP_IMGFLAG_PLANAR){
+ planes[1] += (mpi->y>>mpi->chroma_y_shift)*mpi->stride[1];
+ planes[2] += (mpi->y>>mpi->chroma_y_shift)*mpi->stride[2];
+ dplanes[1]+= (mpi->y>>dmpi->chroma_y_shift)*dmpi->stride[0];
+ dplanes[1]+= (mpi->y>>dmpi->chroma_y_shift)*dmpi->stride[0];
+ }
+ scale(vf->priv->ctx, vf->priv->ctx2, planes, stride, 0, mpi->h, dplanes, dmpi->stride, vf->priv->interlaced);
+ dmpi->y = mpi->y;
+ dmpi->h = mpi->h;
+ return vf_next_put_slice(vf,dmpi);
+}
+
static int __FASTCALL__ control(struct vf_instance_s* vf, int request, void* data){
int *table;
int *inv_table;
@@ -468,7 +520,7 @@
static int __FASTCALL__ vf_open(vf_instance_t *vf,const char* args){
vf->config=config;
- vf->put_slice=put_slice;
+ vf->put_slice=put_frame;
vf->query_format=query_format;
vf->control= control;
vf->uninit=uninit;
@@ -499,6 +551,13 @@
return 1;
}
+static int __FASTCALL__ vf_open_fmtcvt(vf_instance_t *vf,const char* args){
+ int retval = vf_open(vf,args);
+ vf->put_slice=put_slice;
+ vf->print_conf=print_conf_fmtcvt;
+ return retval;
+}
+
//global sws_flags from the command line
int sws_flags=2;
@@ -600,8 +659,8 @@
"fmtcvt",
"A'rpi",
"",
- VF_FLAGS_THREADS,
- vf_open
+ VF_FLAGS_THREADS|VF_FLAGS_SLICES,
+ vf_open_fmtcvt
};
//===========================================================================//
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
