Update of /cvsroot/pure-data/pd/src
In directory sc8-pr-cvs1.sourceforge.net:/tmp/cvs-serv15939
Modified Files:
Tag: devel_0_37
d_array.c d_global.c m_pd.h m_simd.c m_simd_def.h
m_simd_sse_gcc.c m_simd_sse_gcc.h m_simd_sse_vc.c
m_simd_sse_vc.h m_simd_ve_gcc.c m_simd_ve_gcc.h
Log Message:
exporting basic simd functions
Index: m_simd.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd.c,v
retrieving revision 1.1.2.1
retrieving revision 1.1.2.2
diff -C2 -d -r1.1.2.1 -r1.1.2.2
*** m_simd.c 6 Oct 2004 18:31:18 -0000 1.1.2.1
--- m_simd.c 14 Oct 2004 11:01:34 -0000 1.1.2.2
***************
*** 50,51 ****
--- 50,107 ----
}
}
+
+ int simd_check1(t_int n, t_float* ptr1)
+ {
+ return SIMD_CHECK1(n,ptr1);
+ }
+
+ int simd_check2(t_int n, t_float* ptr1, t_float* ptr2)
+ {
+ return SIMD_CHECK2(n,ptr1,ptr2);
+ }
+
+ int simd_check3(t_int n, t_float* ptr1, t_float* ptr2, t_float* ptr3)
+ {
+ return SIMD_CHECK3(n,ptr1,ptr2,ptr3);
+ }
+
+
+
+ #ifdef DONTUSESIMD
+ t_int simd_runtime_check()
+ {
+ return 0;
+ }
+
+ /* tb: wrapper for simd functions */
+ void zerovec_simd(t_float *dst,int n)
+ {
+ zerovec_8(dst,n);
+ }
+
+ void setvec_simd(t_float *dst,t_float v,int n)
+ {
+ setvec_8(dst,v,n);
+ }
+
+ void copyvec_simd(t_float *dst,const t_float *src,int n)
+ {
+ copyvec_8(dst,src,n);
+ }
+
+ void addvec_simd(t_float *dst,const t_float *src,int n)
+ {
+ addvec_8(dst,src,n);
+ }
+
+ void testcopyvec_simd(t_float *dst,const t_float *src,int n)
+ {
+ testcopyvec_8(dst,src,n);
+ }
+
+ void testaddvec_simd(t_float *dst,const t_float *src,int n)
+ {
+ testaddvec_8(dst,src,n);
+ }
+
+ #endif /* DONTUSESIMD */
Index: m_simd_sse_gcc.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_sse_gcc.c,v
retrieving revision 1.1.2.8
retrieving revision 1.1.2.9
diff -C2 -d -r1.1.2.8 -r1.1.2.9
*** m_simd_sse_gcc.c 9 Oct 2004 16:34:22 -0000 1.1.2.8
--- m_simd_sse_gcc.c 14 Oct 2004 11:01:34 -0000 1.1.2.9
***************
*** 10,145 ****
! /* TB: adapted from the xxx_sse_vc routines */
/* dst is assumed to be aligned */
! void zerovec_sse_gcc(t_float *dst,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "xorps %%xmm0, %%xmm0 \n" /* zero value */
! "shr $4, %0 \n"
!
! /* should we do more loop unrolling? */
! /* *dst = 0 */
! "1: \n"
! "movaps %%xmm0, (%1) \n"
! "movaps %%xmm0, 4*T_FLOAT(%1) \n"
! "movaps %%xmm0, 8*T_FLOAT(%1) \n"
! "movaps %%xmm0, 12*T_FLOAT(%1) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(dst)
! :"%xmm0");
}
/* dst is assumed to be aligned */
! void setvec_sse_gcc(t_float *dst,t_float v,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shufps $0,%2,%2 \n" /* load value */
! "shr $4,%0 \n"
!
! /* should we do more loop unrolling? */
! /* *dst = v */
! "1: \n"
! "movaps %2, (%1) \n"
! "movaps %2, 4*T_FLOAT(%1) \n"
! "movaps %2, 8*T_FLOAT(%1) \n"
! "movaps %2, 12*T_FLOAT(%1) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(dst),"x"((t_float)v)
! );
}
/* dst and src are assumed to be aligned */
! void copyvec_sse_gcc(t_float *dst,const t_float *src,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shr $4, %0 \n"
!
! /* loop: *dst = *src */
! "1: \n"
! "movaps (%1), %%xmm0 \n"
! "movaps 4*T_FLOAT(%1), %%xmm1 \n"
! "movaps 8*T_FLOAT(%1), %%xmm2 \n"
! "movaps 12*T_FLOAT(%1), %%xmm3 \n"
! "movaps %%xmm0, (%2) \n"
! "movaps %%xmm1, 4*T_FLOAT(%2) \n"
! "movaps %%xmm2, 8*T_FLOAT(%2) \n"
! "movaps %%xmm3, 12*T_FLOAT(%2) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "addl $16*T_FLOAT,%2 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(src),"r"(dst)
! :"%xmm0","%xmm1","%xmm2","%xmm3");
}
/* dst and src are assumed to be aligned */
! void addvec_sse_gcc(t_float *dst,const t_float *src,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shr $4, %0 \n"
!
! /* loop: *dst += *src */
! "1: \n"
! "movaps (%2,%3),%%xmm0 \n"
! "movaps (%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,(%2,%3) \n"
! "movaps 4*T_FLOAT(%2,%3),%%xmm0 \n"
! "movaps 4*T_FLOAT(%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,4*T_FLOAT(%2,%3) \n"
- "movaps 8*T_FLOAT(%2,%3),%%xmm0 \n"
- "movaps 8*T_FLOAT(%1,%3),%%xmm1 \n"
- "addps %%xmm0,%%xmm1 \n"
- "movaps %%xmm0,8*T_FLOAT(%2,%3) \n"
! "movaps 12*T_FLOAT(%2,%3),%%xmm0 \n"
! "movaps 12*T_FLOAT(%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,12*T_FLOAT(%2,%3) \n"
! "addl $16*T_FLOAT,%3 \n"
! "loop 1b \n"
! :
! : "c"(n),"r"(src),"r"(dst),"r"(0)
! : "%xmm0","%xmm1","%xmm2","%xmm3","%xmm4","%xmm5","%xmm6","%xmm7");
}
! t_int *zero_perf_sse_gcc(t_int *w)
{
! zerovec_sse_gcc((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_sse_gcc(t_int *w)
{
! copyvec_sse_gcc((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_sse_gcc(t_int *w)
{
! setvec_sse_gcc((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_sse_gcc (t_int * w)
{
asm(
--- 10,192 ----
! /* TB: adapted from thomas' vc routines */
/* dst is assumed to be aligned */
! void zerovec_simd(t_float *dst,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "xorps %%xmm0, %%xmm0 \n" /* zero value */
! "shr $4, %0 \n"
!
! /* should we do more loop unrolling? */
! /* *dst = 0 */
! "1: \n"
! "movaps %%xmm0, (%1) \n"
! "movaps %%xmm0, 4*T_FLOAT(%1) \n"
! "movaps %%xmm0, 8*T_FLOAT(%1) \n"
! "movaps %%xmm0, 12*T_FLOAT(%1) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(dst)
! :"%xmm0");
}
/* dst is assumed to be aligned */
! void setvec_simd(t_float *dst,t_float v,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shufps $0,%2,%2 \n" /* load value */
! "shr $4,%0 \n"
!
! /* should we do more loop unrolling? */
! /* *dst = v */
! "1: \n"
! "movaps %2, (%1) \n"
! "movaps %2, 4*T_FLOAT(%1) \n"
! "movaps %2, 8*T_FLOAT(%1) \n"
! "movaps %2, 12*T_FLOAT(%1) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(dst),"x"((t_float)v)
! );
}
/* dst and src are assumed to be aligned */
! void copyvec_simd(t_float *dst,const t_float *src,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shr $4, %0 \n"
!
! /* loop: *dst = *src */
! "1: \n"
! "movaps (%1), %%xmm0 \n"
! "movaps 4*T_FLOAT(%1), %%xmm1 \n"
! "movaps 8*T_FLOAT(%1), %%xmm2 \n"
! "movaps 12*T_FLOAT(%1), %%xmm3 \n"
! "movaps %%xmm0, (%2) \n"
! "movaps %%xmm1, 4*T_FLOAT(%2) \n"
! "movaps %%xmm2, 8*T_FLOAT(%2) \n"
! "movaps %%xmm3, 12*T_FLOAT(%2) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "addl $16*T_FLOAT,%2 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(src),"r"(dst)
! :"%xmm0","%xmm1","%xmm2","%xmm3");
}
/* dst and src are assumed to be aligned */
! void addvec_simd(t_float *dst,const t_float *src,int n)
{
asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shr $4, %0 \n"
!
! /* loop: *dst += *src */
! "1: \n"
! "movaps (%2,%3),%%xmm0 \n"
! "movaps (%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,(%2,%3) \n"
!
! "movaps 4*T_FLOAT(%2,%3),%%xmm0 \n"
! "movaps 4*T_FLOAT(%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,4*T_FLOAT(%2,%3) \n"
!
! "movaps 8*T_FLOAT(%2,%3),%%xmm0 \n"
! "movaps 8*T_FLOAT(%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,8*T_FLOAT(%2,%3) \n"
!
! "movaps 12*T_FLOAT(%2,%3),%%xmm0 \n"
! "movaps 12*T_FLOAT(%1,%3),%%xmm1 \n"
! "addps %%xmm0,%%xmm1 \n"
! "movaps %%xmm0,12*T_FLOAT(%2,%3) \n"
!
! "addl $16*T_FLOAT,%3 \n"
! "loop 1b \n"
! :
! : "c"(n),"r"(src),"r"(dst),"r"(0)
! : "%xmm0","%xmm1","%xmm2","%xmm3","%xmm4","%xmm5","%xmm6","%xmm7");
! }
! /* dst is assumed to be aligned */
! void testvec_simd(t_float *dst,t_float v,int n)
! {
! asm(
! ".set T_FLOAT,4 \n" /* sizeof(t_float) */
! "shufps $0,%2,%2 \n" /* load value */
! "shr $4,%0 \n"
!
! /* should we do more loop unrolling? */
! /* *dst = v */
! "1: \n"
! "movaps %2, (%1) \n"
! "movaps %2, 4*T_FLOAT(%1) \n"
! "movaps %2, 8*T_FLOAT(%1) \n"
! "movaps %2, 12*T_FLOAT(%1) \n"
!
! "addl $16*T_FLOAT,%1 \n"
! "loop 1b \n"
! :
! :"c"(n),"r"(dst),"x"((t_float)v)
! );
! }
! /*
! * if we switch on DAZ, we shouldn't have problems with denormals
! * any more ... tb
! */
! void testcopyvec_simd(t_float *dst,const t_float *src,int n)
! {
! #ifdef DAZ
! copyvec_simd(dst,src,n);
! #else
! testcopyvec_8(dst,src,n); /* SIMD not implemented */
! #endif
! }
! void testaddvec_simd(t_float *dst,const t_float *src,int n)
! {
! #ifdef DAZ
! addvec_simd(dst,src,n);
! #else
! testaddvec_8(dst,src,n); /* SIMD not implemented */
! #endif
}
!
! t_int *zero_perf_simd(t_int *w)
{
! zerovec_simd((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_simd(t_int *w)
{
! copyvec_simd((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_simd(t_int *w)
{
! setvec_simd((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_simd (t_int * w)
{
asm(
***************
*** 181,185 ****
! t_int *scalarplus_perf_sse_gcc(t_int *w)
{
asm(
--- 228,232 ----
! t_int *scalarplus_perf_simd(t_int *w)
{
asm(
***************
*** 218,222 ****
}
! t_int *minus_perf_sse_gcc(t_int *w)
{
asm(
--- 265,269 ----
}
! t_int *minus_perf_simd(t_int *w)
{
asm(
***************
*** 257,261 ****
}
! t_int* scalarminus_perf_sse_gcc(t_int *w)
{
asm(
--- 304,308 ----
}
! t_int* scalarminus_perf_simd(t_int *w)
{
asm(
***************
*** 295,299 ****
! t_int *times_perf_sse_gcc(t_int *w)
{
asm(
--- 342,346 ----
! t_int *times_perf_simd(t_int *w)
{
asm(
***************
*** 334,338 ****
}
! t_int* scalartimes_perf_sse_gcc(t_int *w)
{
asm(
--- 381,385 ----
}
! t_int* scalartimes_perf_simd(t_int *w)
{
asm(
***************
*** 371,375 ****
}
! t_int *sqr_perf_sse_gcc(t_int *w)
{
asm(
--- 418,422 ----
}
! t_int *sqr_perf_simd(t_int *w)
{
asm(
***************
*** 407,411 ****
! t_int* over_perf_sse_gcc(t_int * w)
{
asm(
--- 454,458 ----
! t_int* over_perf_simd(t_int * w)
{
asm(
***************
*** 446,450 ****
}
! t_int* scalarover_perf_sse_gcc(t_int *w)
{
asm(
--- 493,497 ----
}
! t_int* scalarover_perf_simd(t_int *w)
{
asm(
***************
*** 484,488 ****
! t_int* min_perf_sse_gcc(t_int * w)
{
asm(
--- 531,535 ----
! t_int* min_perf_simd(t_int * w)
{
asm(
***************
*** 524,528 ****
! t_int* scalarmin_perf_sse_gcc(t_int *w)
{
asm(
--- 571,575 ----
! t_int* scalarmin_perf_simd(t_int *w)
{
asm(
***************
*** 562,566 ****
! t_int* max_perf_sse_gcc(t_int * w)
{
asm(
--- 609,613 ----
! t_int* max_perf_simd(t_int * w)
{
asm(
***************
*** 602,611 ****
! t_int* scalarmax_perf_sse_gcc(t_int *w)
{
asm(
".set T_FLOAT,4 \n"
! "shufps $0, %1, %1 \n"
"shrl $4, %3 \n" /* divide by 16 */
--- 649,658 ----
! t_int* scalarmax_perf_simd(t_int *w)
{
asm(
".set T_FLOAT,4 \n"
! "shufps $0, %1, %1 \n"
"shrl $4, %3 \n" /* divide by 16 */
Index: m_simd_ve_gcc.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_ve_gcc.c,v
retrieving revision 1.1.2.7
retrieving revision 1.1.2.8
diff -C2 -d -r1.1.2.7 -r1.1.2.8
*** m_simd_ve_gcc.c 7 Oct 2004 17:56:38 -0000 1.1.2.7
--- m_simd_ve_gcc.c 14 Oct 2004 11:01:34 -0000 1.1.2.8
***************
*** 39,43 ****
#define LoadValue(where) LoadUnaligned((const float *)(where))
! void zerovec_ve_gcc(t_float *dst,int n)
{
const vector float zero = (vector float)(0);
--- 39,43 ----
#define LoadValue(where) LoadUnaligned((const float *)(where))
! void zerovec_simd(t_float *dst,int n)
{
const vector float zero = (vector float)(0);
***************
*** 50,54 ****
}
! void setvec_ve_gcc(t_float *dst,t_float v,int n)
{
const vector float arg = LoadValue(&v);
--- 50,54 ----
}
! void setvec_simd(t_float *dst,t_float v,int n)
{
const vector float arg = LoadValue(&v);
***************
*** 61,65 ****
}
! void copyvec_ve_gcc(t_float *dst,const t_float *src,int n)
{
for(n >>= 4; n--; src += 16,dst += 16) {
--- 61,65 ----
}
! void copyvec_simd(t_float *dst,const t_float *src,int n)
{
for(n >>= 4; n--; src += 16,dst += 16) {
***************
*** 75,79 ****
}
! void addvec_ve_gcc(t_float *dst,const t_float *src,int n)
{
#ifdef USEVECLIB
--- 75,79 ----
}
! void addvec_simd(t_float *dst,const t_float *src,int n)
{
#ifdef USEVECLIB
***************
*** 99,121 ****
}
! t_int *zero_perf_ve_gcc(t_int *w)
{
! zerovec_ve_gcc((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_ve_gcc(t_int *w)
{
! copyvec_ve_gcc((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_ve_gcc(t_int *w)
{
! setvec_ve_gcc((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_ve_gcc(t_int *w)
{
#ifdef USEVECLIB
--- 99,132 ----
}
! /* no bad float testing for PPC! */
! void testcopyvec_simd(t_float *dst,const t_float *src,int n)
{
! copyvec_simd(dst,src,n);
! }
!
! void testaddvec_simd(t_float *dst,const t_float *src,int n)
! {
! addvec_simd(dst,src,n);
! }
!
! t_int *zero_perf_simd(t_int *w)
! {
! zerovec_simd((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_simd(t_int *w)
{
! copyvec_simd((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_simd(t_int *w)
{
! setvec_simd((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_simd(t_int *w)
{
#ifdef USEVECLIB
***************
*** 147,151 ****
}
! t_int *scalarplus_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 158,162 ----
}
! t_int *scalarplus_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 173,177 ****
}
! t_int *minus_perf_ve_gcc(t_int *w)
{
#if 0 //def USEVECLIB
--- 184,188 ----
}
! t_int *minus_perf_simd(t_int *w)
{
#if 0 //def USEVECLIB
***************
*** 204,208 ****
}
! t_int *scalarminus_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 215,219 ----
}
! t_int *scalarminus_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 230,234 ****
}
! t_int *times_perf_ve_gcc(t_int *w)
{
#ifdef USEVECLIB
--- 241,245 ----
}
! t_int *times_perf_simd(t_int *w)
{
#ifdef USEVECLIB
***************
*** 261,265 ****
}
! t_int *scalartimes_perf_ve_gcc(t_int *w)
{
#ifdef USEVECLIB
--- 272,276 ----
}
! t_int *scalartimes_perf_simd(t_int *w)
{
#ifdef USEVECLIB
***************
*** 292,296 ****
}
! t_int *sqr_perf_ve_gcc(t_int *w)
{
#ifdef USEVECLIB
--- 303,307 ----
}
! t_int *sqr_perf_simd(t_int *w)
{
#ifdef USEVECLIB
***************
*** 322,326 ****
}
! t_int *over_perf_ve_gcc(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
--- 333,337 ----
}
! t_int *over_perf_simd(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
***************
*** 384,388 ****
}
! t_int *scalarover_perf_ve_gcc(t_int *w)
{
t_float *dst = (t_float *)w[3];
--- 395,399 ----
}
! t_int *scalarover_perf_simd(t_int *w)
{
t_float *dst = (t_float *)w[3];
***************
*** 432,436 ****
}
! t_int *min_perf_ve_gcc(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
--- 443,447 ----
}
! t_int *min_perf_simd(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
***************
*** 458,462 ****
}
! t_int *scalarmin_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 469,473 ----
}
! t_int *scalarmin_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 484,488 ****
}
! t_int *max_perf_ve_gcc(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
--- 495,499 ----
}
! t_int *max_perf_simd(t_int *w)
{
const t_float *src1 = (const t_float *)w[1];
***************
*** 510,514 ****
}
! t_int *scalarmax_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 521,525 ----
}
! t_int *scalarmax_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 536,540 ****
}
! t_int *clip_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 547,551 ----
}
! t_int *clip_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 588,592 ****
}
! t_int *sigwrap_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 599,603 ----
}
! t_int *sigwrap_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 613,617 ****
}
! t_int *sigsqrt_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 624,628 ----
}
! t_int *sigsqrt_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 669,673 ****
/* Attention: there's a difference to sigsqrt_perform which delivers non-zero for a zero input... i don't think the latter is intended... */
! t_int *sigrsqrt_perf_ve_gcc(t_int *w)
{
const t_float *src = (const t_float *)w[1];
--- 680,684 ----
/* Attention: there's a difference to sigsqrt_perform which delivers non-zero for a zero input... i don't think the latter is intended... */
! t_int *sigrsqrt_perf_simd(t_int *w)
{
const t_float *src = (const t_float *)w[1];
***************
*** 728,730 ****
--- 739,747 ----
}
+ t_int simd_runtime_check()
+ {
+ return 1;
+ }
+
+
#endif
Index: m_pd.h
===================================================================
RCS file: /cvsroot/pure-data/pd/src/m_pd.h,v
retrieving revision 1.1.1.4.2.21
retrieving revision 1.1.1.4.2.22
diff -C2 -d -r1.1.1.4.2.21 -r1.1.1.4.2.22
*** m_pd.h 10 Oct 2004 18:16:31 -0000 1.1.1.4.2.21
--- m_pd.h 14 Oct 2004 11:01:33 -0000 1.1.1.4.2.22
***************
*** 577,580 ****
--- 577,608 ----
/* } IOhannes */
+
+ /* tb: exporting basic simd coded dsp functions { */
+
+ /* vectorized, not simd functions*/
+ /* EXTERN void zerovec_8(t_float *dst,int n); */
+ /* EXTERN void setvec_8(t_float *dst,t_float v,int n); */
+ /* EXTERN void copyvec_8(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void addvec_8(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void testcopyvec_8(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void testaddvec_8(t_float *dst,const t_float *src,int n); */
+
+ /* vectorized, simd functions *
+ * dst and src are assumed to be aligned */
+ /* EXTERN t_int simd_runtime_check(void); */
+ /* EXTERN void zerovec_simd(t_float *dst,int n); */
+ /* EXTERN void setvec_simd(t_float *dst,t_float v,int n); */
+ /* EXTERN void copyvec_simd(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void addvec_simd(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void testcopyvec_simd(t_float *dst,const t_float *src,int n); */
+ /* EXTERN void testaddvec_simd(t_float *dst,const t_float *src,int n); */
+
+ EXTERN int simd_check1(t_int n, t_float* ptr1);
+ EXTERN int simd_check2(t_int n, t_float* ptr1, t_float* ptr2);
+ EXTERN int simd_check3(t_int n, t_float* ptr1, t_float* ptr2, t_float* ptr3);
+
+ /* } tb */
+
+
/* ----------------------- utility functions for signals -------------- */
EXTERN float mtof(float);
Index: m_simd_sse_vc.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_sse_vc.c,v
retrieving revision 1.1.2.8
retrieving revision 1.1.2.9
diff -C2 -d -r1.1.2.8 -r1.1.2.9
*** m_simd_sse_vc.c 7 Oct 2004 17:56:38 -0000 1.1.2.8
--- m_simd_sse_vc.c 14 Oct 2004 11:01:34 -0000 1.1.2.9
***************
*** 10,14 ****
/* dst is assumed to be aligned */
! void zerovec_sse_vc(t_float *dst,int n)
{
__asm {
--- 10,14 ----
/* dst is assumed to be aligned */
! void zerovec_simd(t_float *dst,int n)
{
__asm {
***************
*** 33,37 ****
/* dst is assumed to be aligned */
! void setvec_sse_vc(t_float *dst,t_float v,int n)
{
__asm {
--- 33,37 ----
/* dst is assumed to be aligned */
! void setvec_simd(t_float *dst,t_float v,int n)
{
__asm {
***************
*** 59,63 ****
/* dst and src are assumed to be aligned */
! void copyvec_sse_vc(t_float *dst,const t_float *src,int n)
{
__asm {
--- 59,63 ----
/* dst and src are assumed to be aligned */
! void copyvec_simd(t_float *dst,const t_float *src,int n)
{
__asm {
***************
*** 92,96 ****
}
! void addvec_sse_vc(t_float *dst,const t_float *src,int n)
{
__asm {
--- 92,96 ----
}
! void addvec_simd(t_float *dst,const t_float *src,int n)
{
__asm {
***************
*** 139,180 ****
}
! void testcopyvec(t_float *dst,const t_float *src,int n)
{
! while(n--) {
! *dst = (PD_BIGORSMALL(*src) ? 0 : *src);
! dst++;
! src++;
! }
}
! void testaddvec(t_float *dst,const t_float *src,int n)
{
! while(n--) {
! *dst += (PD_BIGORSMALL(*src) ? 0 : *src);
! dst++;
! src++;
! }
}
! t_int *zero_perf_sse_vc(t_int *w)
{
! zerovec_sse_vc((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_sse_vc(t_int *w)
{
! copyvec_sse_vc((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_sse_vc(t_int *w)
{
! setvec_sse_vc((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_sse_vc(t_int *w)
{
__asm {
--- 139,172 ----
}
! void testcopyvec_simd(t_float *dst,const t_float *src,int n)
{
! testcopyvec_8(dst,src,n)
}
! void testaddvec_simd(t_float *dst,const t_float *src,int n)
{
! testaddvec_8(dst,src,n)
}
! t_int *zero_perf_simd(t_int *w)
{
! zerovec_simd((t_float *)w[1],w[2]);
return w+3;
}
! t_int *copy_perf_simd(t_int *w)
{
! copyvec_simd((t_float *)w[2],(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *sig_tilde_perf_simd(t_int *w)
{
! setvec_simd((t_float *)w[2],*(const t_float *)w[1],w[3]);
return w+4;
}
! t_int *plus_perf_simd(t_int *w)
{
__asm {
***************
*** 228,232 ****
}
! t_int *scalarplus_perf_sse_vc(t_int *w)
{
__asm {
--- 220,224 ----
}
! t_int *scalarplus_perf_simd(t_int *w)
{
__asm {
***************
*** 276,280 ****
}
! t_int *minus_perf_sse_vc(t_int *w)
{
__asm {
--- 268,272 ----
}
! t_int *minus_perf_simd(t_int *w)
{
__asm {
***************
*** 331,335 ****
}
! t_int *scalarminus_perf_sse_vc(t_int *w)
{
__asm {
--- 323,327 ----
}
! t_int *scalarminus_perf_simd(t_int *w)
{
__asm {
***************
*** 378,382 ****
}
! t_int *times_perf_sse_vc(t_int *w)
{
__asm {
--- 370,374 ----
}
! t_int *times_perf_simd(t_int *w)
{
__asm {
***************
*** 434,438 ****
}
! t_int *scalartimes_perf_sse_vc(t_int *w)
{
__asm {
--- 426,430 ----
}
! t_int *scalartimes_perf_simd(t_int *w)
{
__asm {
***************
*** 482,486 ****
/* no checking for 0 yet!! */
! t_int *over_perf_sse_vc(t_int *w)
{
__asm {
--- 474,478 ----
/* no checking for 0 yet!! */
! t_int *over_perf_simd(t_int *w)
{
__asm {
***************
*** 538,542 ****
}
! t_int *scalarover_perf_sse_vc(t_int *w)
{
static const float one = 1.f;
--- 530,534 ----
}
! t_int *scalarover_perf_simd(t_int *w)
{
static const float one = 1.f;
***************
*** 599,603 ****
}
! t_int *max_perf_sse_vc(t_int *w)
{
__asm {
--- 591,595 ----
}
! t_int *max_perf_simd(t_int *w)
{
__asm {
***************
*** 654,658 ****
}
! t_int *scalarmax_perf_sse_vc(t_int *w)
{
__asm {
--- 646,650 ----
}
! t_int *scalarmax_perf_simd(t_int *w)
{
__asm {
***************
*** 701,705 ****
}
! t_int *min_perf_sse_vc(t_int *w)
{
__asm {
--- 693,697 ----
}
! t_int *min_perf_simd(t_int *w)
{
__asm {
***************
*** 754,758 ****
}
! t_int *scalarmin_perf_sse_vc(t_int *w)
{
__asm {
--- 746,750 ----
}
! t_int *scalarmin_perf_simd(t_int *w)
{
__asm {
Index: m_simd_sse_gcc.h
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_sse_gcc.h,v
retrieving revision 1.1.2.14
retrieving revision 1.1.2.15
diff -C2 -d -r1.1.2.14 -r1.1.2.15
*** m_simd_sse_gcc.h 10 Oct 2004 21:59:53 -0000 1.1.2.14
--- m_simd_sse_gcc.h 14 Oct 2004 11:01:34 -0000 1.1.2.15
***************
*** 9,90 ****
#include "m_pd.h"
! /* SIMD functions for SSE with GCC */
! void zerovec_sse_gcc(t_float *dst,int n);
! void setvec_sse_gcc(t_float *dst,t_float v,int n);
! void copyvec_sse_gcc(t_float *dst,const t_float *src,int n);
! void addvec_sse_gcc(t_float *dst,const t_float *src,int n);
! //void testcopyvec_sse_gcc(t_float *dst,const t_float *src,int n);
! //void testaddvec_sse_gcc(t_float *dst,const t_float *src,int n);
!
! t_int *zero_perf_sse_gcc(t_int *w);
! t_int *copy_perf_sse_gcc(t_int *w);
! t_int *sig_tilde_perf_sse_gcc(t_int *w);
! t_int *plus_perf_sse_gcc(t_int *w);
! t_int *scalarplus_perf_sse_gcc(t_int *w);
! t_int *minus_perf_sse_gcc(t_int *w);
! t_int *scalarminus_perf_sse_gcc(t_int *w);
! t_int *times_perf_sse_gcc(t_int *w);
! t_int *scalartimes_perf_sse_gcc(t_int *w);
! t_int *sqr_perf_sse_gcc(t_int *w);
! t_int *over_perf_sse_gcc(t_int *w);
! t_int *scalarover_perf_sse_gcc(t_int *w);
! t_int *max_perf_sse_gcc(t_int *w);
! t_int *scalarmax_perf_sse_gcc(t_int *w);
! t_int *min_perf_sse_gcc(t_int *w);
! t_int *scalarmin_perf_sse_gcc(t_int *w);
! t_int *clip_perf_sse_gcc(t_int *w);
! t_int *sigwrap_perf_sse_gcc(t_int *w);
! t_int *sigsqrt_perf_sse_gcc(t_int *w);
! t_int *sigrsqrt_perf_sse_gcc(t_int *w);
!
!
! #define zerovec zerovec_sse_gcc
! #define setvec setvec_sse_gcc
! #define copyvec copyvec_sse_gcc
! #define addvec addvec_sse_gcc
!
! /* if we switch on DAZ, we shouldn't have problems with denormals
! * any more ... tb
! */
! #ifdef DAZ
! #define testcopyvec testcopyvec_8 /* SIMD not implemented */
! #define testaddvec testaddvec_8 /* SIMD not implemented */
! #else /* DAZ */
! #define testcopyvec copyvec_sse_gcc
! #define testaddvec addvec_sse_gcc
! #endif /* DAZ */
/* functions in d_ugen.c */
! #define zero_perf_simd zero_perf_sse_gcc
/* functions in d_dac.c */
! #define copy_perf_simd copy_perf_sse_gcc
/* functions in d_ctl.c */
! #define sig_tilde_perf_simd sig_tilde_perf_sse_gcc
/* functions in d_arithmetic.c */
! #define plus_perf_simd plus_perf_sse_gcc
! #define scalarplus_perf_simd scalarplus_perf_sse_gcc
! #define minus_perf_simd minus_perf_sse_gcc
! #define scalarminus_perf_simd scalarminus_perf_sse_gcc
! #define times_perf_simd times_perf_sse_gcc
! #define scalartimes_perf_simd scalartimes_perf_sse_gcc
! #define sqr_perf_simd sqr_perf_sse_gcc //8 /* SIMD not implemented */
! #define over_perf_simd over_perf_sse_gcc
! #define scalarover_perf_simd scalarover_perf_sse_gcc
! #define min_perf_simd min_perf_sse_gcc
! #define scalarmin_perf_simd scalarmin_perf_sse_gcc
! #define max_perf_simd max_perf_sse_gcc
! #define scalarmax_perf_simd scalarmax_perf_sse_gcc
/* functions in d_math.c */
! #define clip_perf_simd clip_perform /* SIMD not implemented */
! #define sigwrap_perf_simd sigwrap_perform /* SIMD not implemented */
! #define sigsqrt_perf_simd sigsqrt_perform /* SIMD not implemented */
! #define sigrsqrt_perf_simd sigrsqrt_perform /* SIMD not implemented */
!
! /* TB: runtime check */
! extern t_int simd_runtime_check(void);
#endif /* __M_SIMD_SSE_GCC_H */
--- 9,47 ----
#include "m_pd.h"
! /* SIMD functions for SSE with gcc */
/* functions in d_ugen.c */
! t_int *zero_perf_simd(t_int *w);
/* functions in d_dac.c */
! t_int *copy_perf_simd(t_int *w);
/* functions in d_ctl.c */
! t_int *sig_tilde_perf_simd(t_int *w);
/* functions in d_arithmetic.c */
! t_int *plus_perf_simd(t_int *w);
! t_int *scalarplus_perf_simd(t_int *w);
! t_int *minus_perf_simd(t_int *w);
! t_int *scalarminus_perf_simd(t_int *w);
! t_int *times_perf_simd(t_int *w);
! t_int *scalartimes_perf_simd(t_int *w);
! t_int *sqr_perf_simd(t_int *w);
! t_int *over_perf_simd(t_int *w);
! t_int *scalarover_perf_simd(t_int *w);
! t_int *max_perf_simd(t_int *w);
! t_int *scalarmax_perf_simd(t_int *w);
! t_int *min_perf_simd(t_int *w);
! t_int *scalarmin_perf_simd(t_int *w);
! t_int *clip_perf_simd(t_int *w);
! t_int *sigwrap_perf_simd(t_int *w);
! t_int *sigsqrt_perf_simd(t_int *w);
! t_int *sigrsqrt_perf_simd(t_int *w);
/* functions in d_math.c */
! #define clip_perf_simd clip_perform /* SIMD not implemented */
! #define sigwrap_perf_simd sigwrap_perform /* SIMD not implemented */
! #define sigsqrt_perf_simd sigsqrt_perform /* SIMD not implemented */
! #define sigrsqrt_perf_simd sigrsqrt_perform /* SIMD not implemented */
#endif /* __M_SIMD_SSE_GCC_H */
Index: m_simd_sse_vc.h
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_sse_vc.h,v
retrieving revision 1.1.2.8
retrieving revision 1.1.2.9
diff -C2 -d -r1.1.2.8 -r1.1.2.9
*** m_simd_sse_vc.h 7 Oct 2004 17:56:38 -0000 1.1.2.8
--- m_simd_sse_vc.h 14 Oct 2004 11:01:34 -0000 1.1.2.9
***************
*** 10,72 ****
/* SIMD functions for SSE with VC++ */
- void zerovec_sse_vc(t_float *dst,int n);
- void setvec_sse_vc(t_float *dst,t_float v,int n);
- void copyvec_sse_vc(t_float *dst,const t_float *src,int n);
- void addvec_sse_vc(t_float *dst,const t_float *src,int n);
- void testcopyvec_sse_vc(t_float *dst,const t_float *src,int n);
- void testaddvec_sse_vc(t_float *dst,const t_float *src,int n);
-
- t_int *zero_perf_sse_vc(t_int *w);
- t_int *copy_perf_sse_vc(t_int *w);
- t_int *sig_tilde_perf_sse_vc(t_int *w);
- t_int *plus_perf_sse_vc(t_int *w);
- t_int *scalarplus_perf_sse_vc(t_int *w);
- t_int *minus_perf_sse_vc(t_int *w);
- t_int *scalarminus_perf_sse_vc(t_int *w);
- t_int *times_perf_sse_vc(t_int *w);
- t_int *scalartimes_perf_sse_vc(t_int *w);
- t_int *sqr_perf_sse_vc(t_int *w);
- t_int *over_perf_sse_vc(t_int *w);
- t_int *scalarover_perf_sse_vc(t_int *w);
- t_int *max_perf_sse_vc(t_int *w);
- t_int *scalarmax_perf_sse_vc(t_int *w);
- t_int *min_perf_sse_vc(t_int *w);
- t_int *scalarmin_perf_sse_vc(t_int *w);
- t_int *clip_perf_sse_vc(t_int *w);
- t_int *sigwrap_perf_sse_vc(t_int *w);
- t_int *sigsqrt_perf_sse_vc(t_int *w);
- t_int *sigrsqrt_perf_sse_vc(t_int *w);
-
-
- #define zerovec zerovec_sse_vc
- #define setvec setvec_sse_vc
- #define copyvec copyvec_sse_vc
- #define addvec addvec_sse_vc
- #define testcopyvec testcopyvec_sse_vc
- #define testaddvec testaddvec_sse_vc
/* functions in d_ugen.c */
! #define zero_perf_simd zero_perf_sse_vc
/* functions in d_dac.c */
! #define copy_perf_simd copy_perf_sse_vc
/* functions in d_ctl.c */
! #define sig_tilde_perf_simd sig_tilde_perf_sse_vc
/* functions in d_arithmetic.c */
- #define plus_perf_simd plus_perf_sse_vc
- #define scalarplus_perf_simd scalarplus_perf_sse_vc
- #define minus_perf_simd minus_perf_sse_vc
- #define scalarminus_perf_simd scalarminus_perf_sse_vc
- #define times_perf_simd times_perf_sse_vc
- #define scalartimes_perf_simd scalartimes_perf_sse_vc
#define sqr_perf_simd sqr_perf8 /* SIMD not implemented */
#define over_perf_simd over_perf8 /* SIMD not implemented */
- #define scalarover_perf_simd scalarover_perf_sse_vc
- #define min_perf_simd min_perf_sse_vc
- #define scalarmin_perf_simd scalarmin_perf_sse_vc
- #define max_perf_simd max_perf_sse_vc
- #define scalarmax_perf_simd scalarmax_perf_sse_vc
/* functions in d_math.c */
--- 10,46 ----
/* SIMD functions for SSE with VC++ */
/* functions in d_ugen.c */
! t_int *zero_perf_simd(t_int *w);
/* functions in d_dac.c */
! t_int *copy_perf_simd(t_int *w);
/* functions in d_ctl.c */
! t_int *sig_tilde_perf_simd(t_int *w);
!
! /* functions in d_arithmetic.c */
! t_int *plus_perf_simd(t_int *w);
! t_int *scalarplus_perf_simd(t_int *w);
! t_int *minus_perf_simd(t_int *w);
! t_int *scalarminus_perf_simd(t_int *w);
! t_int *times_perf_simd(t_int *w);
! t_int *scalartimes_perf_simd(t_int *w);
! t_int *sqr_perf_simd(t_int *w);
! t_int *over_perf_simd(t_int *w);
! t_int *scalarover_perf_simd(t_int *w);
! t_int *max_perf_simd(t_int *w);
! t_int *scalarmax_perf_simd(t_int *w);
! t_int *min_perf_simd(t_int *w);
! t_int *scalarmin_perf_simd(t_int *w);
! t_int *clip_perf_simd(t_int *w);
! t_int *sigwrap_perf_simd(t_int *w);
! t_int *sigsqrt_perf_simd(t_int *w);
! t_int *sigrsqrt_perf_simd(t_int *w);
!
/* functions in d_arithmetic.c */
#define sqr_perf_simd sqr_perf8 /* SIMD not implemented */
#define over_perf_simd over_perf8 /* SIMD not implemented */
/* functions in d_math.c */
***************
*** 76,82 ****
#define sigrsqrt_perf_simd sigrsqrt_perform /* SIMD not implemented */
- /* TB: runtime check */
- t_int simd_runtime_check();
-
-
#endif /* __M_SIMD_SSE_VC_H */
--- 50,52 ----
Index: m_simd_def.h
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_def.h,v
retrieving revision 1.1.2.7
retrieving revision 1.1.2.8
diff -C2 -d -r1.1.2.7 -r1.1.2.8
*** m_simd_def.h 7 Oct 2004 17:56:38 -0000 1.1.2.7
--- m_simd_def.h 14 Oct 2004 11:01:34 -0000 1.1.2.8
***************
*** 12,23 ****
/* These are the functions that can be coded for SIMD */
- /* functions in m_simd_def.c */
- #define zerovec zerovec_8
- #define setvec setvec_8
- #define copyvec copyvec_8
- #define addvec addvec_8
- #define testcopyvec testcopyvec_8
- #define testaddvec testaddvec_8
-
/* functions in d_ugen.c */
#define zero_perf_simd zero_perf8
--- 12,15 ----
***************
*** 50,56 ****
#define sigrsqrt_perf_simd sigrsqrt_perform /* SIMD not implemented */
- /* TB: runtime check */
- #define simd_runtime_check() 0
-
-
#endif /* __M_SIMD_DEF_H */
--- 42,44 ----
Index: m_simd_ve_gcc.h
===================================================================
RCS file: /cvsroot/pure-data/pd/src/Attic/m_simd_ve_gcc.h,v
retrieving revision 1.1.2.8
retrieving revision 1.1.2.9
diff -C2 -d -r1.1.2.8 -r1.1.2.9
*** m_simd_ve_gcc.h 7 Oct 2004 17:56:38 -0000 1.1.2.8
--- m_simd_ve_gcc.h 14 Oct 2004 11:01:34 -0000 1.1.2.9
***************
*** 10,80 ****
/* SIMD functions for VE with GCC */
- void zerovec_ve_gcc(t_float *dst,int n);
- void setvec_ve_gcc(t_float *dst,t_float v,int n);
- void copyvec_ve_gcc(t_float *dst,const t_float *src,int n);
- void addvec_ve_gcc(t_float *dst,const t_float *src,int n);
-
- t_int *zero_perf_ve_gcc(t_int *w);
- t_int *copy_perf_ve_gcc(t_int *w);
- t_int *sig_tilde_perf_ve_gcc(t_int *w);
- t_int *plus_perf_ve_gcc(t_int *w);
- t_int *scalarplus_perf_ve_gcc(t_int *w);
- t_int *minus_perf_ve_gcc(t_int *w);
- t_int *scalarminus_perf_ve_gcc(t_int *w);
- t_int *times_perf_ve_gcc(t_int *w);
- t_int *scalartimes_perf_ve_gcc(t_int *w);
- t_int *sqr_perf_ve_gcc(t_int *w);
- t_int *over_perf_ve_gcc(t_int *w);
- t_int *scalarover_perf_ve_gcc(t_int *w);
- t_int *max_perf_ve_gcc(t_int *w);
- t_int *scalarmax_perf_ve_gcc(t_int *w);
- t_int *min_perf_ve_gcc(t_int *w);
- t_int *scalarmin_perf_ve_gcc(t_int *w);
- t_int *clip_perf_ve_gcc(t_int *w);
- t_int *sigwrap_perf_ve_gcc(t_int *w);
- t_int *sigsqrt_perf_ve_gcc(t_int *w);
- t_int *sigrsqrt_perf_ve_gcc(t_int *w);
-
-
- #define zerovec zerovec_ve_gcc
- #define setvec setvec_ve_gcc
- #define copyvec copyvec_ve_gcc
- #define addvec addvec_ve_gcc
- /* no bad float testing for PPC! */
- #define testcopyvec copyvec_ve_gcc
- #define testaddvec addvec_ve_gcc
/* functions in d_ugen.c */
! #define zero_perf_simd zero_perf_ve_gcc
/* functions in d_dac.c */
! #define copy_perf_simd copy_perf_ve_gcc
/* functions in d_ctl.c */
! #define sig_tilde_perf_simd sig_tilde_perf_ve_gcc
/* functions in d_arithmetic.c */
! #define plus_perf_simd plus_perf_ve_gcc
! #define scalarplus_perf_simd scalarplus_perf_ve_gcc
! #define minus_perf_simd minus_perf_ve_gcc
! #define scalarminus_perf_simd scalarminus_perf_ve_gcc
! #define times_perf_simd times_perf_ve_gcc
! #define scalartimes_perf_simd scalartimes_perf_ve_gcc
! #define sqr_perf_simd sqr_perf_ve_gcc
! #define over_perf_simd over_perf_ve_gcc
! #define scalarover_perf_simd scalarover_perf_ve_gcc
! #define min_perf_simd min_perf_ve_gcc
! #define scalarmin_perf_simd scalarmin_perf_ve_gcc
! #define max_perf_simd max_perf_ve_gcc
! #define scalarmax_perf_simd scalarmax_perf_ve_gcc
/* functions in d_math.c */
! #define clip_perf_simd clip_perf_ve_gcc
! #define sigwrap_perf_simd sigwrap_perf_ve_gcc
! #define sigsqrt_perf_simd sigsqrt_perf_ve_gcc
! #define sigrsqrt_perf_simd sigrsqrt_perf_ve_gcc
!
! /* TB: runtime check */
! #define simd_runtime_check() 1
#endif /* __M_SIMD_VE_GCC_H */
--- 10,43 ----
/* SIMD functions for VE with GCC */
/* functions in d_ugen.c */
! t_int *zero_perf_simd(t_int *w);
/* functions in d_dac.c */
! t_int *copy_perf_simd(t_int *w);
/* functions in d_ctl.c */
! t_int *sig_tilde_perf_simd(t_int *w);
/* functions in d_arithmetic.c */
! t_int *plus_perf_simd(t_int *w);
! t_int *scalarplus_perf_simd(t_int *w);
! t_int *minus_perf_simd(t_int *w);
! t_int *scalarminus_perf_simd(t_int *w);
! t_int *times_perf_simd(t_int *w);
! t_int *scalartimes_perf_simd(t_int *w);
! t_int *sqr_perf_simd(t_int *w);
! t_int *over_perf_simd(t_int *w);
! t_int *scalarover_perf_simd(t_int *w);
! t_int *max_perf_simd(t_int *w);
! t_int *scalarmax_perf_simd(t_int *w);
! t_int *min_perf_simd(t_int *w);
! t_int *scalarmin_perf_simd(t_int *w);
/* functions in d_math.c */
! t_int *clip_perf_simd(t_int *w);
! t_int *sigwrap_perf_simd(t_int *w);
! t_int *sigsqrt_perf_simd(t_int *w);
! t_int *sigrsqrt_perf_simd(t_int *w);
#endif /* __M_SIMD_VE_GCC_H */
Index: d_global.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/d_global.c,v
retrieving revision 1.1.1.2.8.8
retrieving revision 1.1.1.2.8.9
diff -C2 -d -r1.1.1.2.8.8 -r1.1.1.2.8.9
*** d_global.c 7 Oct 2004 17:56:37 -0000 1.1.1.2.8.8
--- d_global.c 14 Oct 2004 11:01:33 -0000 1.1.1.2.8.9
***************
*** 42,48 ****
while (n--)
{
! *out = (PD_BIGORSMALL(*in) ? 0 : *in);
! out++;
! in++;
}
return (w+4);
--- 42,48 ----
while (n--)
{
! *out = (PD_BIGORSMALL(*in) ? 0 : *in);
! out++;
! in++;
}
return (w+4);
***************
*** 52,56 ****
static t_int *sigsend_perfsimd(t_int *w)
{
! testcopyvec((t_float *)w[2],(t_float *)w[1],w[3]);
return (w+4);
}
--- 52,56 ----
static t_int *sigsend_perfsimd(t_int *w)
{
! testcopyvec_simd((t_float *)w[2],(t_float *)w[1],w[3]);
return (w+4);
}
***************
*** 142,148 ****
t_float *in = x->x_wherefrom;
if(in)
! copyvec((t_float *)w[2],in,w[3]);
else
! zerovec((t_float *)w[2],w[3]);
return (w+4);
}
--- 142,148 ----
t_float *in = x->x_wherefrom;
if(in)
! copyvec_simd((t_float *)w[2],in,w[3]);
else
! zerovec_simd((t_float *)w[2],w[3]);
return (w+4);
}
***************
*** 245,250 ****
static t_int *sigcatch_perfsimd(t_int *w)
{
! copyvec((t_float *)w[2],(t_float *)w[1],w[3]);
! zerovec((t_float *)w[1],w[3]);
return (w+4);
}
--- 245,250 ----
static t_int *sigcatch_perfsimd(t_int *w)
{
! copyvec_simd((t_float *)w[2],(t_float *)w[1],w[3]);
! zerovec_simd((t_float *)w[1],w[3]);
return (w+4);
}
***************
*** 324,328 ****
t_sigthrow *x = (t_sigthrow *)(w[1]);
t_float *out = x->x_whereto;
! if(out) testaddvec(out,(t_float *)w[2],w[3]);
return (w+4);
}
--- 324,328 ----
t_sigthrow *x = (t_sigthrow *)(w[1]);
t_float *out = x->x_whereto;
! if(out) testaddvec_simd(out,(t_float *)w[2],w[3]);
return (w+4);
}
Index: d_array.c
===================================================================
RCS file: /cvsroot/pure-data/pd/src/d_array.c,v
retrieving revision 1.1.1.3.2.6
retrieving revision 1.1.1.3.2.7
diff -C2 -d -r1.1.1.3.2.6 -r1.1.1.3.2.7
*** d_array.c 7 Oct 2004 23:11:56 -0000 1.1.1.3.2.6
--- d_array.c 14 Oct 2004 11:01:33 -0000 1.1.1.3.2.7
***************
*** 295,299 ****
int n = (int)(w[4]);
int maxindex;
! float *buf = x->x_vec, *fp;
int i;
--- 295,299 ----
int n = (int)(w[4]);
int maxindex;
! float *buf = x->x_vec/* , *fp unused ... tb */;
int i;
***************
*** 419,423 ****
int index = findex;
float frac, a, b, c, d, cminusb;
! static int count;
if (index < 1)
index = 1, frac = 0;
--- 419,423 ----
int index = findex;
float frac, a, b, c, d, cminusb;
! /* static int count; tb: unused */
if (index < 1)
index = 1, frac = 0;
***************
*** 594,600 ****
int mask = fnpoints - 1;
float conv = fnpoints * x->x_conv;
! int maxindex;
float *tab = x->x_vec, *addr;
! int i;
double dphase = fnpoints * x->x_phase + UNITBIT32;
--- 594,600 ----
int mask = fnpoints - 1;
float conv = fnpoints * x->x_conv;
! /* int maxindex; unused ... tb */
float *tab = x->x_vec, *addr;
! /* int i; unused ... tb */
double dphase = fnpoints * x->x_phase + UNITBIT32;
***************
*** 742,746 ****
}
else if(SIMD_CHECK2(n,in,dest))
! testcopyvec(dest,in,n);
else
testcopyvec_8(dest,in,n);
--- 742,746 ----
}
else if(SIMD_CHECK2(n,in,dest))
! testcopyvec_simd(dest,in,n);
else
testcopyvec_8(dest,in,n);
***************
*** 758,762 ****
static void tabsend_dsp(t_tabsend *x, t_signal **sp)
{
! int i, vecsize;
t_garray *a;
--- 758,762 ----
static void tabsend_dsp(t_tabsend *x, t_signal **sp)
{
! int /* i, unused ... tb */vecsize;
t_garray *a;
***************
*** 839,845 ****
t_float *from = x->x_vec;
if(from)
! copyvec((t_float *)(w[2]),from,w[3]);
else
! zerovec((t_float *)(w[2]),w[3]);
return (w+4);
}
--- 839,845 ----
t_float *from = x->x_vec;
if(from)
! copyvec_simd((t_float *)(w[2]),from,w[3]);
else
! zerovec_simd((t_float *)(w[2]),w[3]);
return (w+4);
}
***************
*** 1032,1039 ****
static void tabwrite_float(t_tabwrite *x, t_float f)
{
! int i, vecsize;
t_garray *a;
t_float *vec;
!
if (!(a = (t_garray *)pd_findbyclass(x->x_arrayname, garray_class)))
pd_error(x, "%s: no such array", x->x_arrayname->s_name);
--- 1032,1039 ----
static void tabwrite_float(t_tabwrite *x, t_float f)
{
! int /* i, unused ... tb*/vecsize;
t_garray *a;
t_float *vec;
!
if (!(a = (t_garray *)pd_findbyclass(x->x_arrayname, garray_class)))
pd_error(x, "%s: no such array", x->x_arrayname->s_name);
