[Mplayerxp-cvslog] SF.net SVN: mplayerxp:[130] mplayerxp
Brought to you by:
olov
Menu
▾
▴
[Mplayerxp-cvslog] SF.net SVN: mplayerxp:[130] mplayerxp
|
From: <nic...@us...> - 2010-01-28 15:43:41
|
Revision: 130
http://mplayerxp.svn.sourceforge.net/mplayerxp/?rev=130&view=rev
Author: nickols_k
Date: 2010-01-28 15:43:33 +0000 (Thu, 28 Jan 2010)
Log Message:
-----------
minor speedup and fixes
Modified Paths:
--------------
TOOLS/asmopt.c
TOOLS/asmopt_template.h
mplayerxp/libvo/aclib_template.c
mplayerxp/libvo/osd_template.c
mplayerxp/postproc/dsp_accel.h
mplayerxp/pvector/pvector.h
mplayerxp/pvector/pvector_f32_x86.h
mplayerxp/pvector/pvector_int_x86.h
Modified: TOOLS/asmopt.c
===================================================================
--- TOOLS/asmopt.c 2010-01-28 15:41:48 UTC (rev 129)
+++ TOOLS/asmopt.c 2010-01-28 15:43:33 UTC (rev 130)
@@ -32,7 +32,7 @@
#define INIT_ARRAYS(x) \
{\
for(i=0; i<x; i++) srca[i] = i; \
- for(i=0; i<x; i++) src[i] = i+64; \
+ for(i=0; i<x; i++) src[i] = i-61; \
for(i=0; i<x; i++) dsta[i] = i+128; \
}
@@ -44,7 +44,7 @@
gettimeofday(&tv,&tz);
// s=tv.tv_usec;s*=0.000001;s+=tv.tv_sec;
return (tv.tv_sec*1000000+tv.tv_usec);
-}
+}
static inline unsigned long long int read_tsc( void )
{
Modified: TOOLS/asmopt_template.h
===================================================================
--- TOOLS/asmopt_template.h 2010-01-28 15:41:48 UTC (rev 129)
+++ TOOLS/asmopt_template.h 2010-01-28 15:43:33 UTC (rev 130)
@@ -17,7 +17,7 @@
#endif
uint8_t *out_data = dstbase;
uint8_t *in_data = src;
-
+
unsigned i,len;
i = 0;
len = asize;
@@ -30,19 +30,13 @@
for(;i<len;i+=__IVEC_SIZE){
__ivec ind,itmp[2];
ind = _ivec_loadu(&((uint8_t *)in_data)[i]);
-#if 0 /* slower but portable on non-x86 CPUs version */
- itmp[0]= _ivec_sll_s16_imm(_ivec_u16_from_lou8(ind),8);
- itmp[1]= _ivec_sll_s16_imm(_ivec_u16_from_hiu8(ind),8);
-#else
- itmp[0]= _ivec_interleave_lo_u8(izero,ind);
- itmp[1]= _ivec_interleave_hi_u8(izero,ind);
-#endif
+ itmp[0]= _ivec_s16_from_s8(ind,&itmp[1]);
_ivec_storea(&((uint16_t*)out_data)[i],itmp[0]);
_ivec_storea(&((uint16_t*)out_data)[i+__IVEC_SIZE/2],itmp[1]);
}
#endif
for(;i<len;i++)
- ((uint16_t*)out_data)[i]=((uint16_t)((uint8_t*)in_data)[i])<<8;
+ ((int16_t*)out_data)[i]=((int16_t)((int8_t*)in_data)[i]);
#ifdef HAVE_INT_PVECTOR
_ivec_empty();
_ivec_sfence();
Modified: mplayerxp/libvo/aclib_template.c
===================================================================
--- mplayerxp/libvo/aclib_template.c 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/libvo/aclib_template.c 2010-01-28 15:43:33 UTC (rev 130)
@@ -158,6 +158,7 @@
}
#undef MEM_STORE
+#undef MEM_SFENCE
#define MEM_STORE _ivec_stream
#define MEM_SFENCE _ivec_sfence();
static inline void * PVECTOR_RENAME(fast_stream_copy)(void * to, const void * from, size_t len)
@@ -167,6 +168,7 @@
}
#undef MEM_STORE
+#undef MEM_SFENCE
#define MEM_STORE _ivec_storea
#define MEM_SFENCE
static inline void * PVECTOR_RENAME(fast_memcpy)(void * to, const void * from, size_t len)
Modified: mplayerxp/libvo/osd_template.c
===================================================================
--- mplayerxp/libvo/osd_template.c 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/libvo/osd_template.c 2010-01-28 15:43:33 UTC (rev 130)
@@ -64,10 +64,11 @@
_ivec_prefetch(&src[x]);
_ivec_prefetch(&srca[x]);
/* MOVNTDQ: #GP(0) - If memory operand is not aligned on a 16-byte boundary */
+ if(!IVEC_ALIGNED(dstbase))
for(;x<w;x++){
unsigned char *dst=&dstbase[x];
if(srca[x]) *dst=((dstbase[x]*srca[x])>>8)+src[x];
- if((((long)dst)&(__IVEC_SIZE-1))==0) break; /* align on sizeof(MMREG) boundary */
+ if(IVEC_ALIGNED(dst)) break; /* align on sizeof(MMREG) boundary */
}
if((w-x)>=__IVEC_SIZE)
for(;x<w;x+=__IVEC_SIZE){
@@ -76,13 +77,17 @@
_ivec_prefetch(&src[x+__IVEC_SIZE*4]);
_ivec_prefetch(&srca[x+__IVEC_SIZE*4]);
vdest = _ivec_loada(&dstbase[x]);
- vsrc = _ivec_loadu(&src[x]);
- vsrca = _ivec_loadu(&srca[x]);
+ if(IVEC_ALIGNED(&src[x]))
+ vsrc = _ivec_loada(&src[x]);
+ else
+ vsrc = _ivec_loadu(&src[x]);
+ if(IVEC_ALIGNED(&srca[x]))
+ vsrca = _ivec_loada(&srca[x]);
+ else
+ vsrca = _ivec_loadu(&srca[x]);
vmsk = _ivec_not(_ivec_cmpeq_s8(vsrca,vzero));
- vt[0] = _ivec_u16_from_lou8(vdest);
- vt[1] = _ivec_u16_from_hiu8(vdest);
- vt[2] = _ivec_u16_from_lou8(vsrca);
- vt[3] = _ivec_u16_from_hiu8(vsrca);
+ vt[0] = _ivec_u16_from_u8(vdest,&vt[1]);
+ vt[2] = _ivec_u16_from_u8(vsrca,&vt[3]);
vt[0] = _ivec_srl_s16_imm(_ivec_mullo_s16(vt[0],vt[2]),8);
vt[1] = _ivec_srl_s16_imm(_ivec_mullo_s16(vt[1],vt[3]),8);
vt[0] = _ivec_add_s8(_ivec_u8_from_u16(vt[0],vt[1]),vsrc);
Modified: mplayerxp/postproc/dsp_accel.h
===================================================================
--- mplayerxp/postproc/dsp_accel.h 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/postproc/dsp_accel.h 2010-01-28 15:43:33 UTC (rev 130)
@@ -3,27 +3,22 @@
static void __FASTCALL__ PVECTOR_RENAME(int8_to_int16)(const int8_t* in_data, int16_t* out_data, unsigned len, int final)
{
-#ifdef HAVE_INT_PVECTOR
- __ivec izero = _ivec_setzero();
-#endif
unsigned i;
i = 0;
#ifdef HAVE_INT_PVECTOR
+ if(!IVEC_ALIGNED(out_data))
for(;i<len;i++) {
((uint16_t*)out_data)[i]=((uint16_t)((const uint8_t*)in_data)[i])<<8;
- if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ if(IVEC_ALIGNED(out_data)) break;
}
if((len-i)>=__IVEC_SIZE)
for(;i<len;i+=__IVEC_SIZE){
__ivec ind,itmp[2];
- ind = _ivec_loadu(&((const uint8_t *)in_data)[i]);
-#if 0 /* slower but portable on non-x86 CPUs version */
- itmp[0]= _ivec_sll_s16_imm(_ivec_u16_from_lou8(ind),8);
- itmp[1]= _ivec_sll_s16_imm(_ivec_u16_from_hiu8(ind),8);
-#else
- itmp[0]= _ivec_interleave_lo_u8(izero,ind);
- itmp[1]= _ivec_interleave_hi_u8(izero,ind);
-#endif
+ if(IVEC_ALIGNED(in_data))
+ ind = _ivec_loada(&((const uint8_t *)in_data)[i]);
+ else
+ ind = _ivec_loadu(&((const uint8_t *)in_data)[i]);
+ itmp[0] = _ivec_scale_u16_from_u8(ind,&itmp[1]);
if(final) {
_ivec_stream(&((uint16_t*)out_data)[i],itmp[0]);
_ivec_stream(&((uint16_t*)out_data)[i+__IVEC_SIZE/2],itmp[1]);
@@ -44,16 +39,23 @@
unsigned i;
i = 0;
#ifdef HAVE_INT_PVECTOR
+ if(!IVEC_ALIGNED(out_data))
for(;i<len;i++) {
((uint8_t*)out_data)[i]=(uint8_t)((((const uint16_t*)in_data)[i])>>8);
- if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ if(IVEC_ALIGNED(out_data)) break;
}
if((len-i)>=__IVEC_SIZE)
for(;i<len;i+=__IVEC_SIZE){
__ivec outd,itmp[2];
- itmp[0] = _ivec_sra_s16_imm(_ivec_loadu(&((const uint16_t*)in_data)[i]),8);
- itmp[1] = _ivec_sra_s16_imm(_ivec_loadu(&((const uint16_t*)in_data)[i+__IVEC_SIZE/2]),8);
- outd = _ivec_s8_from_s16(itmp[0],itmp[1]);
+ if(IVEC_ALIGNED(in_data)) {
+ itmp[0] = _ivec_loada(&((const uint16_t*)in_data)[i]);
+ itmp[1] = _ivec_loada(&((const uint16_t*)in_data)[i+__IVEC_SIZE/2]);
+ }
+ else {
+ itmp[0] = _ivec_loadu(&((const uint16_t*)in_data)[i]);
+ itmp[1] = _ivec_loadu(&((const uint16_t*)in_data)[i+__IVEC_SIZE/2]);
+ }
+ outd = _ivec_scale_s8_from_s16(itmp[0],itmp[1]);
if(final)
_ivec_stream(&((uint8_t*)out_data)[i],outd);
else
@@ -69,7 +71,6 @@
static void __FASTCALL__ PVECTOR_RENAME(int16_to_int32)(const int16_t* in_data, int32_t* out_data, unsigned len, int final)
{
#ifdef HAVE_INT_PVECTOR
- __ivec izero = _ivec_setzero();
unsigned len_mm,j;
#endif
unsigned i;
@@ -77,22 +78,20 @@
#ifdef HAVE_INT_PVECTOR
j=0;
len_mm=len&(~(__IVEC_SIZE-1));
+ if(!IVEC_ALIGNED(out_data))
for(;i<len;i++,j+=2){
((uint32_t*)out_data)[i]=((uint32_t)((const uint16_t*)in_data)[i])<<16;
- if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ if(IVEC_ALIGNED(out_data)) break;
}
if((len_mm-i)>=__IVEC_SIZE)
for(;i<len_mm;i+=__IVEC_SIZE/2,j+=__IVEC_SIZE)
{
__ivec ind,tmp[2];
- ind = _ivec_loadu(&((const 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
+ if(IVEC_ALIGNED(in_data))
+ ind = _ivec_loada(&((const uint8_t *)in_data)[j]);
+ else
+ ind = _ivec_loadu(&((const uint8_t *)in_data)[j]);
+ tmp[0]= _ivec_scale_u32_from_u16(ind,&tmp[1]);
if(final) {
_ivec_stream(&((uint8_t *)out_data)[j*2],tmp[0]);
_ivec_stream(&((uint8_t *)out_data)[j*2+__IVEC_SIZE],tmp[1]);
@@ -117,17 +116,23 @@
i=0;
#ifdef HAVE_INT_PVECTOR
j=0;
+ if(!IVEC_ALIGNED(out_data))
for(;i<len;i++,j+=2){
((uint16_t*)out_data)[i]=(uint16_t)((((const uint32_t*)in_data)[i])>>16);
- if((((long)out_data)&(__IVEC_SIZE-1))==0) break;
+ if(IVEC_ALIGNED(out_data)) break;
}
if((len-i)>=__IVEC_SIZE)
for(;i<len;i+=__IVEC_SIZE/2,j+=__IVEC_SIZE)
{
__ivec ind[2],tmp;
- ind[0]= _ivec_sra_s32_imm(_ivec_loadu(&((const uint8_t *)in_data)[j*2]),16);
- ind[1]= _ivec_sra_s32_imm(_ivec_loadu(&((const uint8_t *)in_data)[j*2+__IVEC_SIZE]),16);
- tmp = _ivec_s16_from_s32(ind[0],ind[1]);
+ if(IVEC_ALIGNED(in_data)) {
+ ind[0]=_ivec_loada(&((const uint8_t *)in_data)[j*2]);
+ ind[1]=_ivec_loada(&((const uint8_t *)in_data)[j*2+__IVEC_SIZE]);
+ } else {
+ ind[0]=_ivec_loadu(&((const uint8_t *)in_data)[j*2]);
+ ind[1]=_ivec_loadu(&((const uint8_t *)in_data)[j*2+__IVEC_SIZE]);
+ }
+ tmp = _ivec_scale_s16_from_s32(ind[0],ind[1]);
if(final)
_ivec_stream(&((uint8_t *)out_data)[j],tmp);
else
@@ -314,23 +319,27 @@
i=0;
#ifdef HAVE_F32_PVECTOR
int_max = _f32vec_broadcast(INT32_MAX-1);
- /* SSE engine sometime has unpredictable behaviour. So downscale volume on 1% here. */
- plus1 = _f32vec_broadcast(+0.99);
- minus1= _f32vec_broadcast(-0.99);
+ /* SSE float2int engine doesn't have SATURATION functionality.
+ So CLAMP volume on 0.0002% here. */
+ plus1 = _f32vec_broadcast(+0.999998);
+ minus1= _f32vec_broadcast(-0.999998);
+ if(!F32VEC_ALIGNED(out))
for(;i<len;i++) {
ftmp=((const float*)in)[i];
SATURATE(ftmp,-1.0,+1.0);
((int32_t*)out)[i]=(int32_t)lrintf((INT_MAX-1)*ftmp);
- if((((long)out)&(__F32VEC_SIZE-1))==0) break;
+ if(F32VEC_ALIGNED(out)) break;
}
_ivec_empty();
len_mm=len&(~(__F32VEC_SIZE-1));
if((len_mm-i)>=__F32VEC_SIZE/sizeof(float))
for(;i<len_mm;i+=__F32VEC_SIZE/sizeof(float)) {
__f32vec tmp;
- tmp = _f32vec_loadu(&((const float*)in)[i]);
- tmp = _f32vec_min(tmp,plus1);
- tmp = _f32vec_max(tmp,minus1);
+ if(F32VEC_ALIGNED(in))
+ tmp = _f32vec_loada(&((const float*)in)[i]);
+ else
+ tmp = _f32vec_loadu(&((const float*)in)[i]);
+ tmp = _f32vec_clamp(tmp,minus1,plus1);
tmp = _f32vec_mul(int_max,tmp);
if(final)
_f32vec_to_s32_stream(&((int32_t*)out)[i],tmp);
@@ -354,15 +363,20 @@
#endif
register unsigned i=0;
#ifdef HAVE_F32_PVECTOR
+ if(!F32VEC_ALIGNED(out))
for(;i<len;i++) {
((float*)out)[i]=(1.0/INT_MAX)*((float)((const int32_t*)in)[i]);
- if((((long)out)&(__F32VEC_SIZE-1))==0) break;
+ if(F32VEC_ALIGNED(out)) break;
}
_ivec_empty();
if((len-i)>=__F32VEC_SIZE)
for(;i<len;i+=__F32VEC_SIZE/sizeof(float)) {
__f32vec tmp;
- tmp = _f32vec_mul(rev_imax,_f32vec_from_s32u(&((const int32_t*)in)[i]));
+ if(F32VEC_ALIGNED(in))
+ tmp = _f32vec_from_s32a(&((const int32_t*)in)[i]);
+ else
+ tmp = _f32vec_from_s32u(&((const int32_t*)in)[i]);
+ tmp = _f32vec_mul(rev_imax,tmp);
if(final)
_f32vec_stream(&((float*)out)[i],tmp);
else
Modified: mplayerxp/pvector/pvector.h
===================================================================
--- mplayerxp/pvector/pvector.h 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/pvector/pvector.h 2010-01-28 15:43:33 UTC (rev 130)
@@ -47,6 +47,10 @@
#undef HAVE_F32_PVECTOR
#endif
+#undef IVEC_ALIGNED
+#define IVEC_ALIGNED(p) ((((long)((void *)(p)))&(__IVEC_SIZE-1))==0)
+#undef F32VEC_ALIGNED
+#define F32VEC_ALIGNED(p) ((((long)((void *)(p)))&(__F32VEC_SIZE-1))==0)
/*
ABBREVIATION:
@@ -130,15 +134,22 @@
__ivec _ivec_interleave_lo_u32(__ivec s1, _ivec_ s2);
__ivec _ivec_interleave_hi_u32(__ivec s1, _ivec_ s2);
- __ivec _ivec_u16_from_lou8(__ivec s); // Convert lo part of mvec from U8 to U16
- __ivec _ivec_u16_from_hiu8(__ivec s); // Convert hi part of mvec from U8 to U16
- __ivec _ivec_u32_from_lou16(__ivec s);// Convert lo part of mvec from U16 to U32
- __ivec _ivec_u32_from_hiu16(__ivec s); // Convert hi part of mvec from U16 to U32
+ __ivec _ivec_u16_from_u8(__ivec s,__ivec *hipart); // Convert ivec from U8 to U16
+ __ivec _ivec_u32_from_u16(__ivec s,__ivec *hipart);// Convert ivec from U16 to U32
+ __ivec _ivec_s16_from_s8(__ivec s,__ivec *hipart); // Convert ivec from S8 to S16
+ __ivec _ivec_s32_from_s16(__ivec s,__ivec *hipart);// Convert ivec from S16 to S32
+ __ivec _ivec_scale_u16_from_u8(__ivec s,__ivec *hipart); // Convert ivec from U8 to U16 and shift left on 8-bit
+ __ivec _ivec_scale_u32_from_u16(__ivec s,__ivec *hipart);// Convert ivec from U16 to U32 and shift left on 16-bit
+
__ivec _ivec_s16_from_s32(__ivec s1,__ivec s2); // Convert from S32 to S16
__ivec _ivec_s8_from_s16(__ivec s1,__ivec s2); // Convert from S16 to S8
__ivec _ivec_u8_from_u16(__ivec s1,__ivec s2); // Convert from U16 to U8
+ __ivec _ivec_scale_s16_from_s32(__ivec s1,__ivec s2); // Convert from S32 to S16 and shift right on 16-bit
+ __ivec _ivec_scale_s8_from_s16(__ivec s1,__ivec s2); // Convert from S16 to S8 and shift right on 8-bit
+ __ivec _ivec_scale_u8_from_u16(__ivec s1,__ivec s2); // Convert from U16 to U8 and shift right on 8-bit
+
ARITHMETIC engine:
------------------
__ivec _ivec_add_s8(__ivec s1,__ivec s2); // Add S8
@@ -202,4 +213,5 @@
---------------
__f32vec _f32vec_min(__f32vec f1, __f32vec f2); // MIN(f1,f2)
__f32vec _f32vec_max(__f32vec f1, __f32vec f2); // MAX(f1,f2)
+ __f32vec _f32vec_clamp(__f32vec f1, __f32vec minval,__f32vec maxval); // CLAMP(f1,minval,maxval);
*/
Modified: mplayerxp/pvector/pvector_f32_x86.h
===================================================================
--- mplayerxp/pvector/pvector_f32_x86.h 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/pvector/pvector_f32_x86.h 2010-01-28 15:43:33 UTC (rev 130)
@@ -317,3 +317,12 @@
}
#undef _f32vec_min
#define _f32vec_min PVECTOR_RENAME(f32_min)
+
+extern __inline __f32vec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(f32_clamp)(__f32vec f1,__f32vec minval,__f32vec maxval)
+{
+ return _f32vec_max(_f32vec_min(f1,maxval),minval);
+}
+#undef _f32vec_clamp
+#define _f32vec_clamp PVECTOR_RENAME(f32_clamp)
+
Modified: mplayerxp/pvector/pvector_int_x86.h
===================================================================
--- mplayerxp/pvector/pvector_int_x86.h 2010-01-28 15:41:48 UTC (rev 129)
+++ mplayerxp/pvector/pvector_int_x86.h 2010-01-28 15:43:33 UTC (rev 130)
@@ -414,33 +414,50 @@
#define _ivec_interleave_hi_u32 PVECTOR_RENAME(interleave_hi_u32)
extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
-PVECTOR_RENAME(u16_from_lou8)(__ivec s)
+PVECTOR_RENAME(u16_from_u8)(__ivec s,__ivec *hipart)
{
- return _ivec_interleave_lo_u8(s,_ivec_setzero());
+ __ivec filler = _ivec_setzero();
+ *hipart = _ivec_interleave_hi_u8(s,filler);
+ return _ivec_interleave_lo_u8(s,filler);
}
-#undef _ivec_u16_from_lou8
-#define _ivec_u16_from_lou8 PVECTOR_RENAME(u16_from_lou8)
+#undef _ivec_u16_from_u8
+#define _ivec_u16_from_u8 PVECTOR_RENAME(u16_from_u8)
+
extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
-PVECTOR_RENAME(u16_from_hiu8)(__ivec s)
+PVECTOR_RENAME(u32_from_u16)(__ivec s,__ivec *hipart)
{
- return _ivec_interleave_hi_u8(s,_ivec_setzero());
+ __ivec filler = _ivec_setzero();
+ *hipart = _ivec_interleave_hi_u16(s,filler);
+ return _ivec_interleave_lo_u16(s,filler);
}
-#undef _ivec_u16_from_hiu8
-#define _ivec_u16_from_hiu8 PVECTOR_RENAME(u16_from_hiu8)
+#undef _ivec_u32_from_u16
+#define _ivec_u32_from_u16 PVECTOR_RENAME(u32_from_u16)
+
extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
-PVECTOR_RENAME(u32_from_lou16)(__ivec s)
+PVECTOR_RENAME(s16_from_s8)(__ivec s,__ivec* hipart)
{
- return _ivec_interleave_lo_u16(s,_ivec_setzero());
+ const __ivec izero = _ivec_setzero();
+ __ivec filler;
+ filler = _ivec_cmpgt_s8(izero,s);
+ *hipart = _ivec_interleave_hi_u8(s,filler);
+ return _ivec_interleave_lo_u8(s,filler);
}
-#undef _ivec_u32_from_lou16
-#define _ivec_u32_from_lou16 PVECTOR_RENAME(u32_from_lou16)
+#undef _ivec_s16_from_s8
+#define _ivec_s16_from_s8 PVECTOR_RENAME(s16_from_s8)
+
extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
-PVECTOR_RENAME(u32_from_hiu16)(__ivec s)
+PVECTOR_RENAME(s32_from_s16)(__ivec s,__ivec* hipart)
{
- return _ivec_interleave_hi_u16(s,_ivec_setzero());
+ const __ivec izero = _ivec_setzero();
+ __ivec filler;
+ filler = _ivec_cmpgt_s16(izero,s);
+ *hipart = _ivec_interleave_hi_u16(s,filler);
+ return _ivec_interleave_lo_u16(s,filler);
}
-#undef _ivec_u32_from_hiu16
-#define _ivec_u32_from_hiu16 PVECTOR_RENAME(u32_from_hiu16)
+#undef _ivec_s32_from_s16
+#define _ivec_s32_from_s16 PVECTOR_RENAME(s32_from_s16)
+
+
extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
PVECTOR_RENAME(s16_from_s32)(__ivec s1, __ivec s2)
{
@@ -840,3 +857,61 @@
}
#undef _ivec_srl_s64_imm
#define _ivec_srl_s64_imm PVECTOR_RENAME(srl_s64_imm)
+
+extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(scale_u16_from_u8)(__ivec s,__ivec *hipart)
+{
+#if 0 /* slower but portable on non-x86 CPUs version */
+ __ivec tmp[2];
+ tmp[0] = _ivec_u16_from_u8(s,&tmp[1]);
+ *hipart = _ivec_sll_s16_imm(tmp[1],8);
+ return _ivec_sll_s16_imm(tmp[0],8);
+#else
+ __ivec filler = _ivec_setzero();
+ *hipart = _ivec_interleave_hi_u8(filler,s);
+ return _ivec_interleave_lo_u8(filler,s);
+#endif
+}
+#undef _ivec_scale_u16_from_u8
+#define _ivec_scale_u16_from_u8 PVECTOR_RENAME(scale_u16_from_u8)
+
+extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(scale_u32_from_u16)(__ivec s,__ivec *hipart)
+{
+#if 0 /* slower but portable on non-x86 CPUs version */
+ __ivec tmp[2];
+ tmp[0] = _ivec_u32_from_u16(s,&tmp[1]);
+ *hipart = _ivec_sll_s32_imm(tmp[1],16);
+ return _ivec_sll_s32_imm(tmp[0],16);
+#else
+ __ivec filler = _ivec_setzero();
+ *hipart = _ivec_interleave_hi_u16(filler,s);
+ return _ivec_interleave_lo_u16(filler,s);
+#endif
+}
+#undef _ivec_scale_u32_from_u16
+#define _ivec_scale_u32_from_u16 PVECTOR_RENAME(scale_u32_from_u16)
+
+extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(scale_s16_from_s32)(__ivec s1, __ivec s2)
+{
+ return _ivec_s16_from_s32(_ivec_sra_s32_imm(s1,16),_ivec_sra_s32_imm(s2,16));
+}
+#undef _ivec_scale_s16_from_s32
+#define _ivec_scale_s16_from_s32 PVECTOR_RENAME(scale_s16_from_s32)
+
+extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(scale_s8_from_s16)(__ivec s1, __ivec s2)
+{
+ return _ivec_s8_from_s16(_ivec_sra_s16_imm(s1,8),_ivec_sra_s16_imm(s2,8));
+}
+#undef _ivec_scale_s8_from_s16
+#define _ivec_scale_s8_from_s16 PVECTOR_RENAME(scale_s8_from_s16)
+extern __inline __ivec __attribute__((__gnu_inline__, __always_inline__))
+PVECTOR_RENAME(scale_u8_from_u16)(__ivec s1, __ivec s2)
+{
+ return _ivec_u8_from_u16(_ivec_sra_s16_imm(s1,8),_ivec_sra_s16_imm(s2,8));
+}
+#undef _ivec_scale_u8_from_u16
+#define _ivec_scale_u8_from_u16 PVECTOR_RENAME(scale_u8_from_u16)
+
This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site.
|
