Aron Rosenberg
2004-Aug-06 15:01 UTC
[speex-dev] [PATCH] Make SSE Run Time option. Add Win32 SSE code
All,
Attached is a patch that does two things. First it makes the use
of the current SSE code a run time option through the use
of speex_decoder_ctl() and speex_encoder_ctl
It does this twofold. First there is a modification to the configure.in
script which introduces a check based upon platform. It will compile in the
sse assembly if you are on an i?86 based platform by making a special
define. Second, it adds a new ctl value called SPEEX_SET_ASM_FLAG which
takes in an integer. The values are defined as:
#define SPEEX_SET_ASM_FLAG 200
#define SPEEX_ASM_MMX_NONE 0
#define SPEEX_ASM_MMX_BASIC 1
#define SPEEX_ASM_MMX_SSE 2
#define SPEEX_ASM_MMX_SSE_FP 4
The current Speex SSE code requires full SSE2 support which corresponds to
SPEEX_ASM_MMX_SSE_FP. None of the other defines are actively used, but they
are included since they represent different Intel/AMD processors. For
example, an AMD Duran only supports SPEEX_ASM_MMX_BASIC while Pentium 3's
and above support full SPEEX_ASM_MMX_SSE_FP
<p>The second part of the patch adds the equivalent MS Windows assembler
for
the same sections that currently have GCC x86 assembler code.
Notes about implementation: We took the easiest route when hacking in the
flag support which was to add a global flag for the entire library at
runtime and extern it in all the various files.
Jean-Marc: We looked at adding the flag into the state structures, however
they were not passed all the way down into the filters.c files and it would
have been a massive change to make it pass all the needed data. The
approach we took should be ok since on a given machine you would have the
same settings. The decoder_ctl and encoder_ctl set the same global flag
variable.
The way we setup the asm flags var should allow you to add the ARM assembly
in the exact same manor. You would add a check in the configure.in for the
platform and define a _USE_ARM and place the code in the same functions as
we did. You would then add a SPEEX_ASM_ARM 8 or something and let the
application decide to turn it on.
<p>Other Notes: This patch obsoletes ltp_sse.h and filters_sse.h .
However
the patch does not remove them. This is thge updated version of the patch
we sent in November.
<p>Comments are welcome. BTW, we have been shipping our Video
Conferencing
product which only uses the Speex codec for 6 months now and have gotten
rave reviews (PC Magazine Editors choice) for the audio and video quality.
We use Speex in Windows, Mac OS-X, and Linux as we have clients for each
platform. Keep up the great work! Check us out at
http://www.sightspeed.com and please try our beta version (Mac and Windows
Clients available now) at http://www.sightspeed.com/page.php?page=beta
Patch in .gz format attached, plain text below.
-----------------------------------------------------------
Aron Rosenberg
CTO
SightSpeed Inc.
http://www.sightspeed.com
<p><p>diff -ru speex-1.0.3-orig/configure.in
speex-1.0.3/configure.in
--- speex-1.0.3-orig/configure.in 2003-10-30 08:57:25.000000000 -0800
+++ speex-1.0.3/configure.in 2004-01-08 20:44:25.000000000 -0800
@@ -18,11 +18,11 @@
# For automake.
VERSION=$SPEEX_VERSION
PACKAGE=speex
+AC_CANONICAL_SYSTEM
AM_INIT_AUTOMAKE($PACKAGE, $VERSION, no-define)
AM_MAINTAINER_MODE
-AC_CANONICAL_HOST
AM_PROG_LIBTOOL
AC_C_BIGENDIAN
@@ -52,8 +52,18 @@
AC_CHECK_LIB(gnugetopt, getopt_long)
AC_DEFINE_UNQUOTED(VERSION, "${VERSION}")
+AC_DEFINE(GCC_COMPILER)
+echo -n "Checking for Platform ASM Speedups..."
+case "$target" in
+ i?86*)
+ CFLAGS="$CFLAGS -D_USE_SSE"
+ echo "Found SSE and SSE2"
+ ;;
+ *)
+ echo "None Found"
+ ;;
+esac
-AC_ARG_ENABLE(sse, [ --enable-sse enable SSE support], [if
test "$enableval" = yes; then AC_DEFINE(_USE_SSE) fi])
dnl Output the makefiles and version.h.
diff -ru speex-1.0.3-orig/libspeex/filters.c speex-1.0.3/libspeex/filters.c
--- speex-1.0.3-orig/libspeex/filters.c 2003-01-05 21:56:56.000000000 -0800
+++ speex-1.0.3/libspeex/filters.c 2004-01-08 20:44:41.000000000 -0800
@@ -32,8 +32,10 @@
#include "filters.h"
#include "stack_alloc.h"
+#include "speex.h"
#include <math.h>
+extern int global_use_mmx_sse;
void bw_lpc(float gamma, float *lpc_in, float *lpc_out, int order)
{
@@ -46,41 +48,548 @@
}
}
-#ifdef _USE_SSE
-#include "filters_sse.h"
-#else
-void filter_mem2(float *x, float *num, float *den, float *y, int N, int
ord, float *mem)
+
+void filter_mem2(float *x, float *_num, float *_den, float *y, int N, int
ord, float *_mem)
{
- int i,j;
- float xi,yi;
- for (i=0;i<N;i++)
- {
- xi=x[i];
- y[i] = num[0]*xi + mem[0];
- yi=y[i];
- for (j=0;j<ord-1;j++)
- {
- mem[j] = mem[j+1] + num[j+1]*xi - den[j+1]*yi;
- }
- mem[ord-1] = num[ord]*xi - den[ord]*yi;
- }
+ if (!(global_use_mmx_sse & SPEEX_ASM_MMX_SSE_FP))
+ {
+ int i,j;
+ float xi,yi;
+ for (i=0;i<N;i++)
+ {
+ xi=x[i];
+ y[i] = _num[0]*xi + _mem[0];
+ yi=y[i];
+ for (j=0;j<ord-1;j++)
+ {
+ _mem[j] = _mem[j+1] + _num[j+1]*xi - _den[j+1]*yi;
+ }
+ _mem[ord-1] = _num[ord]*xi - _den[ord]*yi;
+ }
+ }
+ else
+ {
+ float __num[20], __den[20], __mem[20];
+ float *num, *den, *mem;
+ int i;
+
+ num = (float*)(((int)(__num+4))&0xfffffff0)-1;
+ den = (float*)(((int)(__den+4))&0xfffffff0)-1;
+ mem = (float*)(((int)(__mem+4))&0xfffffff0)-1;
+
+ for (i=0;i<=10;i++)
+ num[i]=den[i]=0;
+
+ for (i=0;i<10;i++)
+ mem[i]=0;
+
+ for (i=0;i<ord+1;i++)
+ {
+ num[i]=_num[i];
+ den[i]=_den[i];
+ }
+
+ for (i=0;i<ord;i++)
+ mem[i]=_mem[i];
+
+ for (i=0;i<N;i+=4)
+ {
+ float *in1 = x+i;
+ float *in2 = y+i;
+
+#ifdef WIN32
+ /*
+ Do we need to push???
+ push eax
+ push ebx
+ push ecx
+ push edx
+ */
+ _asm
+ {
+ mov eax, num
+ mov ebx, den
+ mov ecx, mem
+
+ mov edx, in1
+ movss xmm0, [edx]
+
+ movss xmm1, [ecx]
+ addss xmm1, xmm0
+
+ mov edx, in2
+ movss [edx], xmm1
+
+ shufps xmm0, xmm0, 0x00
+ shufps xmm1, xmm1, 0x00
+
+ movaps xmm2, [eax+4]
+ movaps xmm3, [ebx+4]
+ mulps xmm2, xmm0
+ mulps xmm3, xmm1
+ movaps xmm4, [eax+20]
+ mulps xmm4, xmm0
+ addps xmm2, [ecx+4]
+ movaps xmm5, [ebx+20]
+ mulps xmm5, xmm1
+ addps xmm4, [ecx+20]
+ subps xmm2, xmm3
+ movups [ecx], xmm2
+ subps xmm4, xmm5
+ movups [ecx+16], xmm4
+
+ movss xmm2, [eax+36]
+ mulss xmm2, xmm0
+ movss xmm3, [ebx+36]
+ mulss xmm3, xmm1
+ addss xmm2, [ecx+36]
+ movss xmm4, [eax+40]
+ mulss xmm4, xmm0
+ movss xmm5, [ebx+40]
+ mulss xmm5, xmm1
+ subss xmm2, xmm3
+ movss [ecx+32], xmm2
+ subss xmm4, xmm5
+ movss [ecx+36], xmm4
+
+ mov edx, in1
+ movss xmm0, [edx+4]
+
+ movss xmm1, [ecx]
+ addss xmm1, xmm0
+
+ mov edx, in2
+ movss [edx+4], xmm1
+
+ shufps xmm0, xmm0, 0x00
+ shufps xmm1, xmm1, 0x00
+
+ movaps xmm2, [eax+4]
+ movaps xmm3, [ebx+4]
+ mulps xmm2, xmm0
+ mulps xmm3, xmm1
+ movaps xmm4, [eax+20]
+ mulps xmm4, xmm0
+ addps xmm2, [ecx+4]
+ movaps xmm5, [ebx+20]
+ mulps xmm5, xmm1
+ addps xmm4, [ecx+20]
+ subps xmm2, xmm3
+ movups [ecx], xmm2
+ subps xmm4, xmm5
+ movups [ecx+16], xmm4
+
+ movss xmm2, [eax+36]
+ mulss xmm2, xmm0
+ movss xmm3, [ebx+36]
+ mulss xmm3, xmm1
+ addss xmm2, [ecx+36]
+ movss xmm4, [eax+40]
+ mulss xmm4, xmm0
+ movss xmm5, [ebx+40]
+ mulss xmm5, xmm1
+ subss xmm2, xmm3
+ movss [ecx+32], xmm2
+ subss xmm4, xmm5
+ movss [ecx+36], xmm4
+
+ mov edx, in1
+ movss xmm0, [edx+8]
+
+ movss xmm1, [ecx]
+ addss xmm1, xmm0
+
+ mov edx, in2
+ movss [edx+8], xmm1
+
+ shufps xmm0, xmm0, 0x00
+ shufps xmm1, xmm1, 0x00
+
+ movaps xmm2, [eax+4]
+ movaps xmm3, [ebx+4]
+ mulps xmm2, xmm0
+ mulps xmm3, xmm1
+ movaps xmm4, [eax+20]
+ mulps xmm4, xmm0
+ addps xmm2, [ecx+4]
+ movaps xmm5, [ebx+20]
+ mulps xmm5, xmm1
+ addps xmm4, [ecx+20]
+ subps xmm2, xmm3
+ movups [ecx], xmm2
+ subps xmm4, xmm5
+ movups [ecx+16], xmm4
+
+ movss xmm2, [eax+36]
+ mulss xmm2, xmm0
+ movss xmm3, [ebx+36]
+ mulss xmm3, xmm1
+ addss xmm2, [ecx+36]
+ movss xmm4, [eax+40]
+ mulss xmm4, xmm0
+ movss xmm5, [ebx+40]
+ mulss xmm5, xmm1
+ subss xmm2, xmm3
+ movss [ecx+32], xmm2
+ subss xmm4, xmm5
+ movss [ecx+36], xmm4
+
+ mov edx, in1
+ movss xmm0, [edx+12]
+
+ movss xmm1, [ecx]
+ addss xmm1, xmm0
+
+ mov edx, in2
+ movss [edx+12], xmm1
+
+ shufps xmm0, xmm0, 0x00
+ shufps xmm1, xmm1, 0x00
+
+ movaps xmm2, [eax+4]
+ movaps xmm3, [ebx+4]
+ mulps xmm2, xmm0
+ mulps xmm3, xmm1
+ movaps xmm4, [eax+20]
+ mulps xmm4, xmm0
+ addps xmm2, [ecx+4]
+ movaps xmm5, [ebx+20]
+ mulps xmm5, xmm1
+ addps xmm4, [ecx+20]
+ subps xmm2, xmm3
+ movups [ecx], xmm2
+ subps xmm4, xmm5
+ movups [ecx+16], xmm4
+
+ movss xmm2, [eax+36]
+ mulss xmm2, xmm0
+ movss xmm3, [ebx+36]
+ mulss xmm3, xmm1
+ addss xmm2, [ecx+36]
+ movss xmm4, [eax+40]
+ mulss xmm4, xmm0
+ movss xmm5, [ebx+40]
+ mulss xmm5, xmm1
+ subss xmm2, xmm3
+ movss [ecx+32], xmm2
+ subss xmm4, xmm5
+ movss [ecx+36], xmm4
+ }
+ /*
+ pop edx
+ pop ecx
+ pop ebx
+ pop eax
+ */
+#endif
+#if defined(GCC_COMPILER) && defined(_USE_SSE)
+ __asm__ __volatile__
+ (
+ "\tmovss (%1), %%xmm0\n"
+ "\tmovss (%0), %%xmm1\n"
+ "\taddss %%xmm0, %%xmm1\n"
+ "\tmovss %%xmm1, (%2)\n"
+ "\tshufps $0x00, %%xmm0, %%xmm0\n"
+ "\tshufps $0x00, %%xmm1, %%xmm1\n"
+
+ "\tmovaps 4(%3), %%xmm2\n"
+ "\tmovaps 4(%4), %%xmm3\n"
+ "\tmulps %%xmm0, %%xmm2\n"
+ "\tmulps %%xmm1, %%xmm3\n"
+ "\tmovaps 20(%3), %%xmm4\n"
+ "\tmulps %%xmm0, %%xmm4\n"
+ "\taddps 4(%0), %%xmm2\n"
+ "\tmovaps 20(%4), %%xmm5\n"
+ "\tmulps %%xmm1, %%xmm5\n"
+ "\taddps 20(%0), %%xmm4\n"
+ "\tsubps %%xmm3, %%xmm2\n"
+ "\tmovups %%xmm2, (%0)\n"
+ "\tsubps %%xmm5, %%xmm4\n"
+ "\tmovups %%xmm4, 16(%0)\n"
+
+ "\tmovss 36(%3), %%xmm2\n"
+ "\tmulss %%xmm0, %%xmm2\n"
+ "\tmovss 36(%4), %%xmm3\n"
+ "\tmulss %%xmm1, %%xmm3\n"
+ "\taddss 36(%0), %%xmm2\n"
+ "\tmovss 40(%3), %%xmm4\n"
+ "\tmulss %%xmm0, %%xmm4\n"
+ "\tmovss 40(%4), %%xmm5\n"
+ "\tmulss %%xmm1, %%xmm5\n"
+ "\tsubss %%xmm3, %%xmm2\n"
+ "\tmovss %%xmm2, 32(%0)\n"
+ "\tsubss %%xmm5, %%xmm4\n"
+ "\tmovss %%xmm4, 36(%0)\n"
+
+ "\tmovss 4(%1), %%xmm0\n"
+ "\tmovss (%0), %%xmm1\n"
+ "\taddss %%xmm0, %%xmm1\n"
+ "\tmovss %%xmm1, 4(%2)\n"
+ "\tshufps $0x00, %%xmm0, %%xmm0\n"
+ "\tshufps $0x00, %%xmm1, %%xmm1\n"
+
+ "\tmovaps 4(%3), %%xmm2\n"
+ "\tmovaps 4(%4), %%xmm3\n"
+ "\tmulps %%xmm0, %%xmm2\n"
+ "\tmulps %%xmm1, %%xmm3\n"
+ "\tmovaps 20(%3), %%xmm4\n"
+ "\tmulps %%xmm0, %%xmm4\n"
+ "\taddps 4(%0), %%xmm2\n"
+ "\tmovaps 20(%4), %%xmm5\n"
+ "\tmulps %%xmm1, %%xmm5\n"
+ "\taddps 20(%0), %%xmm4\n"
+ "\tsubps %%xmm3, %%xmm2\n"
+ "\tmovups %%xmm2, (%0)\n"
+ "\tsubps %%xmm5, %%xmm4\n"
+ "\tmovups %%xmm4, 16(%0)\n"
+
+ "\tmovss 36(%3), %%xmm2\n"
+ "\tmulss %%xmm0, %%xmm2\n"
+ "\tmovss 36(%4), %%xmm3\n"
+ "\tmulss %%xmm1, %%xmm3\n"
+ "\taddss 36(%0), %%xmm2\n"
+ "\tmovss 40(%3), %%xmm4\n"
+ "\tmulss %%xmm0, %%xmm4\n"
+ "\tmovss 40(%4), %%xmm5\n"
+ "\tmulss %%xmm1, %%xmm5\n"
+ "\tsubss %%xmm3, %%xmm2\n"
+ "\tmovss %%xmm2, 32(%0)\n"
+ "\tsubss %%xmm5, %%xmm4\n"
+ "\tmovss %%xmm4, 36(%0)\n"
+
+ "\tmovss 8(%1), %%xmm0\n"
+ "\tmovss (%0), %%xmm1\n"
+ "\taddss %%xmm0, %%xmm1\n"
+ "\tmovss %%xmm1, 8(%2)\n"
+ "\tshufps $0x00, %%xmm0, %%xmm0\n"
+ "\tshufps $0x00, %%xmm1, %%xmm1\n"
+
+ "\tmovaps 4(%3), %%xmm2\n"
+ "\tmovaps 4(%4), %%xmm3\n"
+ "\tmulps %%xmm0, %%xmm2\n"
+ "\tmulps %%xmm1, %%xmm3\n"
+ "\tmovaps 20(%3), %%xmm4\n"
+ "\tmulps %%xmm0, %%xmm4\n"
+ "\taddps 4(%0), %%xmm2\n"
+ "\tmovaps 20(%4), %%xmm5\n"
+ "\tmulps %%xmm1, %%xmm5\n"
+ "\taddps 20(%0), %%xmm4\n"
+ "\tsubps %%xmm3, %%xmm2\n"
+ "\tmovups %%xmm2, (%0)\n"
+ "\tsubps %%xmm5, %%xmm4\n"
+ "\tmovups %%xmm4, 16(%0)\n"
+
+ "\tmovss 36(%3), %%xmm2\n"
+ "\tmulss %%xmm0, %%xmm2\n"
+ "\tmovss 36(%4), %%xmm3\n"
+ "\tmulss %%xmm1, %%xmm3\n"
+ "\taddss 36(%0), %%xmm2\n"
+ "\tmovss 40(%3), %%xmm4\n"
+ "\tmulss %%xmm0, %%xmm4\n"
+ "\tmovss 40(%4), %%xmm5\n"
+ "\tmulss %%xmm1, %%xmm5\n"
+ "\tsubss %%xmm3, %%xmm2\n"
+ "\tmovss %%xmm2, 32(%0)\n"
+ "\tsubss %%xmm5, %%xmm4\n"
+ "\tmovss %%xmm4, 36(%0)\n"
+
+ "\tmovss 12(%1), %%xmm0\n"
+ "\tmovss (%0), %%xmm1\n"
+ "\taddss %%xmm0, %%xmm1\n"
+ "\tmovss %%xmm1, 12(%2)\n"
+ "\tshufps $0x00, %%xmm0, %%xmm0\n"
+ "\tshufps $0x00, %%xmm1, %%xmm1\n"
+
+ "\tmovaps 4(%3), %%xmm2\n"
+ "\tmovaps 4(%4), %%xmm3\n"
+ "\tmulps %%xmm0, %%xmm2\n"
+ "\tmulps %%xmm1, %%xmm3\n"
+ "\tmovaps 20(%3), %%xmm4\n"
+ "\tmulps %%xmm0, %%xmm4\n"
+ "\taddps 4(%0), %%xmm2\n"
+ "\tmovaps 20(%4), %%xmm5\n"
+ "\tmulps %%xmm1, %%xmm5\n"
+ "\taddps 20(%0), %%xmm4\n"
+ "\tsubps %%xmm3, %%xmm2\n"
+ "\tmovups %%xmm2, (%0)\n"
+ "\tsubps %%xmm5, %%xmm4\n"
+ "\tmovups %%xmm4, 16(%0)\n"
+
+ "\tmovss 36(%3), %%xmm2\n"
+ "\tmulss %%xmm0, %%xmm2\n"
+ "\tmovss 36(%4), %%xmm3\n"
+ "\tmulss %%xmm1, %%xmm3\n"
+ "\taddss 36(%0), %%xmm2\n"
+ "\tmovss 40(%3), %%xmm4\n"
+ "\tmulss %%xmm0, %%xmm4\n"
+ "\tmovss 40(%4), %%xmm5\n"
+ "\tmulss %%xmm1, %%xmm5\n"
+ "\tsubss %%xmm3, %%xmm2\n"
+ "\tmovss %%xmm2, 32(%0)\n"
+ "\tsubss %%xmm5, %%xmm4\n"
+ "\tmovss %%xmm4, 36(%0)\n"
+
+ : : "r" (mem), "r" (x+i), "r"
(y+i), "r" (num), "r" (den)
+ : "memory" );
+#endif
+ }
+ for (i=0;i<ord;i++)
+ _mem[i]=mem[i];
+ }
}
<p>-void iir_mem2(float *x, float *den, float *y, int N, int ord, float
*mem)
+void iir_mem2(float *x, float *_den, float *y, int N, int ord, float *_mem)
{
- int i,j;
- for (i=0;i<N;i++)
- {
- y[i] = x[i] + mem[0];
- for (j=0;j<ord-1;j++)
- {
- mem[j] = mem[j+1] - den[j+1]*y[i];
- }
- mem[ord-1] = - den[ord]*y[i];
- }
-}
+ if (!(global_use_mmx_sse & SPEEX_ASM_MMX_SSE_FP))
+ {
+ int i,j;
+ for (i=0;i<N;i++)
+ {
+ y[i] = x[i] + _mem[0];
+ for (j=0;j<ord-1;j++)
+ {
+ _mem[j] = _mem[j+1] - _den[j+1]*y[i];
+ }
+ _mem[ord-1] = - _den[ord]*y[i];
+ }
+ }
+ else
+ {
+
+ float __den[20], __mem[20];
+ float *den, *mem;
+ int i;
+
+ den = (float*)(((int)(__den+4))&0xfffffff0)-1;
+ mem = (float*)(((int)(__mem+4))&0xfffffff0)-1;
+ for (i=0;i<=10;i++)
+ {
+ den[i]=0;
+ }
+ for (i=0;i<10;i++)
+ {
+ mem[i]=0;
+ }
+ for (i=0;i<ord+1;i++)
+ {
+ den[i]=_den[i];
+ }
+ for (i=0;i<ord;i++)
+ {
+ mem[i]=_mem[i];
+ }
+
+ for (i=0;i<N;i++)
+ {
+
+/* The assembler block below is equivalent to this code
+ y[i] = x[i] + mem[0];
+ for (j=0;j<ord-1;j++)
+ {
+ mem[j] = mem[j+1] - den[j+1]*y[i];
+ }
+ mem[ord-1] = - den[ord]*y[i]
+*/
+
+ float *in1 = x+i;
+ float *in2 = y+i;
+
+#ifdef WIN32
+ /*
+ Do we need to push???
+ push eax
+ push ebx
+ push ecx
+ push edx
+ */
+ _asm
+ {
+ mov eax, den
+ mov ebx, mem
+ mov ecx, in1
+ mov edx, in2
+
+ movss xmm0, [ecx]
+ movss xmm1, [ebx]
+ addss xmm1, xmm0
+
+ movss [edx], xmm1
+
+ shufps xmm0, xmm0, 0x00
+ shufps xmm1, xmm1, 0x00
+
+ movaps xmm2, [eax+4]
+ movaps xmm3, [eax+20]
+ mulps xmm2, xmm1
+ mulps xmm3, xmm1
+ movss xmm4, [eax+36]
+ movss xmm5, [eax+40]
+ mulss xmm4, xmm1
+ mulss xmm5, xmm1
+ movaps xmm6, [ebx+4]
+ subps xmm6, xmm2
+ movups [ebx], xmm6
+ movaps xmm7, [ebx+20]
+ subps xmm7, xmm3
+ movups [ebx+16], xmm7
+
+ movss xmm7, [ebx+36]
+ subss xmm7, xmm4
+ movss [ebx+32], xmm7
+ xorps xmm2, xmm2
+ subss xmm2, xmm5
+ movss [ebx+36], xmm2
+ }
+ /*
+ pop edx
+ pop ecx
+ pop ebx
+ pop eax
+ */
+#endif
+#if defined(GCC_COMPILER) && defined(_USE_SSE)
+ __asm__ __volatile__ (
+ "\tmovss (%1), %%xmm0\n"
+ "\tmovss (%0), %%xmm1\n"
+ "\taddss %%xmm0, %%xmm1\n"
+ "\tmovss %%xmm1, (%2)\n"
+ "\tshufps $0x00, %%xmm0, %%xmm0\n"
+ "\tshufps $0x00, %%xmm1, %%xmm1\n"
+
+ "\tmovaps 4(%3), %%xmm2\n"
+ "\tmovaps 20(%3), %%xmm3\n"
+ "\tmulps %%xmm1, %%xmm2\n"
+ "\tmulps %%xmm1, %%xmm3\n"
+ "\tmovss 36(%3), %%xmm4\n"
+ "\tmovss 40(%3), %%xmm5\n"
+ "\tmulss %%xmm1, %%xmm4\n"
+ "\tmulss %%xmm1, %%xmm5\n"
+ "\tmovaps 4(%0), %%xmm6\n"
+ "\tsubps %%xmm2, %%xmm6\n"
+ "\tmovups %%xmm6, (%0)\n"
+ "\tmovaps 20(%0), %%xmm7\n"
+ "\tsubps %%xmm3, %%xmm7\n"
+ "\tmovups %%xmm7, 16(%0)\n"
+
+
+ "\tmovss 36(%0), %%xmm7\n"
+ "\tsubss %%xmm4, %%xmm7\n"
+ "\tmovss %%xmm7, 32(%0)\n"
+ "\txorps %%xmm2, %%xmm2\n"
+ "\tsubss %%xmm5, %%xmm2\n"
+ "\tmovss %%xmm2, 36(%0)\n"
+
+ : : "r" (mem), "r" (x+i),
"r" (y+i), "r" (den) :
"memory" );
#endif
+ }
+ for (i=0;i<ord;i++)
+ {
+ _mem[i]=mem[i];
+ }
+ }
+}
+
void fir_mem2(float *x, float *num, float *y, int N, int ord, float *mem)
{
diff -ru speex-1.0.3-orig/libspeex/ltp.c speex-1.0.3/libspeex/ltp.c
--- speex-1.0.3-orig/libspeex/ltp.c 2003-05-05 18:20:26.000000000 -0700
+++ speex-1.0.3/libspeex/ltp.c 2004-01-08 20:44:45.000000000 -0800
@@ -34,37 +34,168 @@
#include "ltp.h"
#include "stack_alloc.h"
#include "filters.h"
+#include "speex.h"
#include "speex_bits.h"
-#ifdef _USE_SSE
-#include "ltp_sse.h"
-#else
-static float inner_prod(float *x, float *y, int len)
+extern int global_use_mmx_sse;
+
+static float inner_prod(float *a, float *b, int len)
{
- int i;
- float sum1=0,sum2=0,sum3=0,sum4=0;
- for (i=0;i<len;)
- {
- sum1 += x[i]*y[i];
- sum2 += x[i+1]*y[i+1];
- sum3 += x[i+2]*y[i+2];
- sum4 += x[i+3]*y[i+3];
- i+=4;
- }
- return sum1+sum2+sum3+sum4;
-}
+ if (!(global_use_mmx_sse & SPEEX_ASM_MMX_SSE_FP))
+ {
+/*
+//Older, slower version
+ int i;
+ float sum=0;
+ for (i=0;i<len;i++)
+ sum += a[i]*b[i];
+ return sum;
+*/ int i;
+ float sum1=0,sum2=0,sum3=0,sum4=0;
+ for (i=0;i<len;)
+ {
+ sum1 += a[i]*b[i];
+ sum2 += a[i+1]*b[i+1];
+ sum3 += a[i+2]*b[i+2];
+ sum4 += a[i+3]*b[i+3];
+ i+=4;
+ }
+ return sum1+sum2+sum3+sum4;
+ }
+ else
+ {
+ float sum;
+ float* sum1 = ∑
+
+#ifdef WIN32
+ _asm
+ {
+ push eax
+ push ebx
+ push ecx
+ push edx
+ push edi
+
+ mov eax, a
+ mov edi, b
+ mov ecx, len
+ mov edx, sum1
+
+ xorps xmm3, xmm3
+ xorps xmm4, xmm4
+ sub ecx, 20
+
+ mul20_loop:
+
+ movups xmm0, [eax]
+ movups xmm1, [edi]
+ mulps xmm1, xmm0
+
+ movups xmm5, [eax+16]
+ movups xmm6, [edi+16]
+ mulps xmm6, xmm5
+ addps xmm3, xmm1
+
+ movups xmm0, [eax+32]
+ movups xmm1, [edi+32]
+ mulps xmm1, xmm0
+ addps xmm4, xmm6
+
+ movups xmm5, [eax+48]
+ movups xmm6, [edi+48]
+ mulps xmm6, xmm5
+ addps xmm3, xmm1
+
+ movups xmm0, [eax+64]
+ movups xmm1, [edi+64]
+ mulps xmm1, xmm0
+ addps xmm4, xmm6
+ addps xmm3, xmm1
+
+ add eax, 80
+ add edi, 80
+ sub ecx, 20
+
+ jae mul20_loop
+
+ addps xmm3, xmm4
+ movhlps xmm4, xmm3
+ addps xmm3, xmm4
+ movaps xmm4, xmm3
+ shufps xmm4, xmm4, 0x55
+ addss xmm3, xmm4
+ movss [edx], xmm3
+
+ pop edi
+ pop edx
+ pop ecx
+ pop ebx
+ pop eax
+ }
+#endif
+#if defined(GCC_COMPILER) && defined(_USE_SSE)
+ __asm__ __volatile__ (
+ "\tpush %%eax\n"
+ "\tpush %%edi\n"
+ "\tpush %%ecx\n"
+ "\txorps %%xmm3, %%xmm3\n"
+ "\txorps %%xmm4, %%xmm4\n"
+
+ "\tsub $20, %%ecx\n"
+
+ ".mul20_loop%=:\n"
+
+ "\tmovups (%%eax), %%xmm0\n"
+ "\tmovups (%%edi), %%xmm1\n"
+ "\tmulps %%xmm0, %%xmm1\n"
+
+ "\tmovups 16(%%eax), %%xmm5\n"
+ "\tmovups 16(%%edi), %%xmm6\n"
+ "\tmulps %%xmm5, %%xmm6\n"
+ "\taddps %%xmm1, %%xmm3\n"
+
+ "\tmovups 32(%%eax), %%xmm0\n"
+ "\tmovups 32(%%edi), %%xmm1\n"
+ "\tmulps %%xmm0, %%xmm1\n"
+ "\taddps %%xmm6, %%xmm4\n"
+
+ "\tmovups 48(%%eax), %%xmm5\n"
+ "\tmovups 48(%%edi), %%xmm6\n"
+ "\tmulps %%xmm5, %%xmm6\n"
+ "\taddps %%xmm1, %%xmm3\n"
+
+ "\tmovups 64(%%eax), %%xmm0\n"
+ "\tmovups 64(%%edi), %%xmm1\n"
+ "\tmulps %%xmm0, %%xmm1\n"
+ "\taddps %%xmm6, %%xmm4\n"
+ "\taddps %%xmm1, %%xmm3\n"
+
+
+ "\tadd $80, %%eax\n"
+ "\tadd $80, %%edi\n"
+
+ "\tsub $20, %%ecx\n"
+
+ "\tjae .mul20_loop%=\n"
+
+ "\taddps %%xmm4, %%xmm3\n"
+
+ "\tmovhlps %%xmm3, %%xmm4\n"
+ "\taddps %%xmm4, %%xmm3\n"
+ "\tmovaps %%xmm3, %%xmm4\n"
+ "\tshufps $0x55, %%xmm4, %%xmm4\n"
+ "\taddss %%xmm4, %%xmm3\n"
+ "\tmovss %%xmm3, (%%edx)\n"
+
+ "\tpop %%ecx\n"
+ "\tpop %%edi\n"
+ "\tpop %%eax\n"
+ : : "a" (a), "D" (b), "c" (len),
"d" (&sum) : "memory");
#endif
+ return sum;
+ }
-/*Original, non-optimized version*/
-/*static float inner_prod(float *x, float *y, int len)
-{
- int i;
- float sum=0;
- for (i=0;i<len;i++)
- sum += x[i]*y[i];
- return sum;
}
-*/
<p> void open_loop_nbest_pitch(float *sw, int start, int end, int len,
int
*pitch, float *gain, int N, char *stack)
diff -ru speex-1.0.3-orig/libspeex/modes.c speex-1.0.3/libspeex/modes.c
--- speex-1.0.3-orig/libspeex/modes.c 2003-06-02 22:29:39.000000000 -0700
+++ speex-1.0.3/libspeex/modes.c 2004-01-08 20:46:26.000000000 -0800
@@ -45,6 +45,7 @@
#define NULL 0
#endif
+int global_use_mmx_sse = 0;
SpeexMode *speex_mode_list[SPEEX_NB_MODES] = {&speex_nb_mode,
&speex_wb_mode, &speex_uwb_mode};
/* Extern declarations for all codebooks we use here */
@@ -585,16 +586,27 @@
int speex_encoder_ctl(void *state, int request, void *ptr)
{
- return (*((SpeexMode**)state))->enc_ctl(state, request, ptr);
+ if (request == SPEEX_SET_ASM_FLAG)
+ {
+ global_use_mmx_sse = *((int*)ptr);
+ return 0;
+ }
+ else
+ return (*((SpeexMode**)state))->enc_ctl(state, request, ptr);
}
int speex_decoder_ctl(void *state, int request, void *ptr)
{
- return (*((SpeexMode**)state))->dec_ctl(state, request, ptr);
+ if (request == SPEEX_SET_ASM_FLAG)
+ {
+ global_use_mmx_sse = *((int*)ptr);
+ return 0;
+ }
+ else
+ return (*((SpeexMode**)state))->dec_ctl(state, request, ptr);
}
<p>-
static int nb_mode_query(void *mode, int request, void *ptr)
{
SpeexNBMode *m = (SpeexNBMode*)mode;
diff -ru speex-1.0.3-orig/libspeex/speex.h speex-1.0.3/libspeex/speex.h
--- speex-1.0.3-orig/libspeex/speex.h 2003-01-22 23:29:39.000000000 -0800
+++ speex-1.0.3/libspeex/speex.h 2004-01-08 20:46:21.000000000 -0800
@@ -156,6 +156,13 @@
/** Number of defined modes in Speex */
#define SPEEX_NB_MODES 3
+
+#define SPEEX_SET_ASM_FLAG 200
+#define SPEEX_ASM_MMX_NONE 0
+#define SPEEX_ASM_MMX_BASIC 1
+#define SPEEX_ASM_MMX_SSE 2
+#define SPEEX_ASM_MMX_SSE_FP 4
+
struct SpeexMode;
<p><p><p>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: speex-sse-patch.gz
Type: application/octet-stream
Size: 3886 bytes
Desc: speex-sse-patch.gz
Url :
http://lists.xiph.org/pipermail/speex-dev/attachments/20040108/99cf64fa/speex-sse-patch-0001.obj
Jean-Marc Valin
2004-Aug-06 15:01 UTC
[speex-dev] [PATCH] Make SSE Run Time option. Add Win32 SSE code
Hi,
Thanks for the patch. I think it's a good idea, although I can't apply
it as is. The reason is that in its current form, the SSE version is not
tested enough and isn't very clean in some aspects. For example, the
order 10 filter is hard-coded and patched to work also for order 8 (less
efficiently). Also, I think this should really go into 1.1.x (to become
1.2). I have already found a faster implementation, which is not yet in
CVS BTW.
About your SPEEX_ASM flags, I'm not sure I see the difference between
SPEEX_ASM_MMX_SSE and SPEEX_ASM_MMX_SSE_FP. Also, you're saying that the
current code makes use of SSE2, which I don't think is the case, since I
developed it on a Pentium III, which only supports SSE1. I don't think
SSE2 is important at all, since most of the SSE2 instructions are for
double precision (which Speex doesn't use at all).
Last thing, I see in your windows version a bunch of commented pops and
pushes. Those are definitely needed. You compiler may happen to produce
code without that, but there's no guarantee you won't run into problems
later because suddenly, the compiler assumes that whatever was there
before is still there.
Jean-Marc
Le ven 09/01/2004 à 00:18, Aron Rosenberg a écrit :> All,
>
> Attached is a patch that does two things. First it makes the use
> of the current SSE code a run time option through the use
> of speex_decoder_ctl() and speex_encoder_ctl
> It does this twofold. First there is a modification to the configure.in
> script which introduces a check based upon platform. It will compile in the
> sse assembly if you are on an i?86 based platform by making a special
> define. Second, it adds a new ctl value called SPEEX_SET_ASM_FLAG which
> takes in an integer. The values are defined as:
>
> #define SPEEX_SET_ASM_FLAG 200
> #define SPEEX_ASM_MMX_NONE 0
> #define SPEEX_ASM_MMX_BASIC 1
> #define SPEEX_ASM_MMX_SSE 2
> #define SPEEX_ASM_MMX_SSE_FP 4
>
> The current Speex SSE code requires full SSE2 support which corresponds to
> SPEEX_ASM_MMX_SSE_FP. None of the other defines are actively used, but they
> are included since they represent different Intel/AMD processors. For
> example, an AMD Duran only supports SPEEX_ASM_MMX_BASIC while Pentium
3's
> and above support full SPEEX_ASM_MMX_SSE_FP
>
>
> The second part of the patch adds the equivalent MS Windows assembler for
> the same sections that currently have GCC x86 assembler code.
>
> Notes about implementation: We took the easiest route when hacking in the
> flag support which was to add a global flag for the entire library at
> runtime and extern it in all the various files.
> Jean-Marc: We looked at adding the flag into the state structures, however
> they were not passed all the way down into the filters.c files and it would
> have been a massive change to make it pass all the needed data. The
> approach we took should be ok since on a given machine you would have the
> same settings. The decoder_ctl and encoder_ctl set the same global flag
> variable.
>
> The way we setup the asm flags var should allow you to add the ARM assembly
> in the exact same manor. You would add a check in the configure.in for the
> platform and define a _USE_ARM and place the code in the same functions as
> we did. You would then add a SPEEX_ASM_ARM 8 or something and let the
> application decide to turn it on.
>
>
> Other Notes: This patch obsoletes ltp_sse.h and filters_sse.h . However
> the patch does not remove them. This is thge updated version of the patch
> we sent in November.
>
>
> Comments are welcome. BTW, we have been shipping our Video Conferencing
> product which only uses the Speex codec for 6 months now and have gotten
> rave reviews (PC Magazine Editors choice) for the audio and video quality.
> We use Speex in Windows, Mac OS-X, and Linux as we have clients for each
> platform. Keep up the great work! Check us out at
> http://www.sightspeed.com and please try our beta version (Mac and Windows
> Clients available now) at http://www.sightspeed.com/page.php?page=beta
--
Jean-Marc Valin, M.Sc.A., ing. jr.
LABORIUS (http://www.gel.usherb.ca/laborius)
Université de Sherbrooke, Québec, Canada
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 190 bytes
Desc: Ceci est une partie de message numériquement signée.
Url :
http://lists.xiph.org/pipermail/speex-dev/attachments/20040109/be8389a8/signature-0001.pgp
Aron Rosenberg
2004-Aug-06 15:01 UTC
[speex-dev] [PATCH] Make SSE Run Time option. Add Win32 SSE code
Jean-Marc,
There is a big difference between SSE and SSEFP. The SSEFP means
that the CPU supports the xmm registers. All Intel chips with SSE support
do, however no current 32 bit AMD chips support the XMM registers. They
will support the SSE instructions but not those registers. You are right
about the SSE2 not being used.
The AMD Opterons are the first AMD CPU's which support xmm registers. They
will have 16 of them while the current Pentium 3's and above have only 8.
Sorry about the patch having those push pops commented out, they should be
in there.
If you check your new code into CVS we can do all the converting needed. We
are working on an Altivec version right now based on the current code, but
if you have new code that makes it easier for us since we won't have to
port it twice.
One major thing to note - In Altivec everything needs to be 16 byte aligned
for it to work efficiently. A number of the starting points right now are
only 4 byte aligned. If you can add the following macro to the variables
that get passed in, it will make everything easier. Use it as such:
ALIGN(16) unsigned int myVar;
or
static ALIGN(16) float myArray[16];
#ifdef GCC_COMPILER
#define ALIGN(n) __attribute__ ((__aligned__ (n)))
#endif
#ifdef WIN32
#define ALIGN(n) __declspec(align(n))
#endif
<p>Aron Rosenberg
SightSpeed Software
<p>At 11:23 PM 1/8/2004, you wrote:>Hi,
>
>Thanks for the patch. I think it's a good idea, although I can't
apply
>it as is. The reason is that in its current form, the SSE version is not
>tested enough and isn't very clean in some aspects. For example, the
>order 10 filter is hard-coded and patched to work also for order 8 (less
>efficiently). Also, I think this should really go into 1.1.x (to become
>1.2). I have already found a faster implementation, which is not yet in
>CVS BTW.
>
>About your SPEEX_ASM flags, I'm not sure I see the difference between
>SPEEX_ASM_MMX_SSE and SPEEX_ASM_MMX_SSE_FP. Also, you're saying that the
>current code makes use of SSE2, which I don't think is the case, since I
>developed it on a Pentium III, which only supports SSE1. I don't think
>SSE2 is important at all, since most of the SSE2 instructions are for
>double precision (which Speex doesn't use at all).
>
>Last thing, I see in your windows version a bunch of commented pops and
>pushes. Those are definitely needed. You compiler may happen to produce
>code without that, but there's no guarantee you won't run into
problems
>later because suddenly, the compiler assumes that whatever was there
>before is still there.
>
> Jean-Marc
>
>Le ven 09/01/2004 à 00:18, Aron Rosenberg a écrit :
> > All,
> >
> > Attached is a patch that does two things. First it makes the
use
> > of the current SSE code a run time option through the use
> > of speex_decoder_ctl() and speex_encoder_ctl
> > It does this twofold. First there is a modification to the
configure.in
> > script which introduces a check based upon platform. It will compile
in
> the
> > sse assembly if you are on an i?86 based platform by making a special
> > define. Second, it adds a new ctl value called SPEEX_SET_ASM_FLAG
which
> > takes in an integer. The values are defined as:
> >
> > #define SPEEX_SET_ASM_FLAG 200
> > #define SPEEX_ASM_MMX_NONE 0
> > #define SPEEX_ASM_MMX_BASIC 1
> > #define SPEEX_ASM_MMX_SSE 2
> > #define SPEEX_ASM_MMX_SSE_FP 4
> >
> > The current Speex SSE code requires full SSE2 support which
corresponds to
> > SPEEX_ASM_MMX_SSE_FP. None of the other defines are actively used, but
> they
> > are included since they represent different Intel/AMD processors. For
> > example, an AMD Duran only supports SPEEX_ASM_MMX_BASIC while Pentium
3's
> > and above support full SPEEX_ASM_MMX_SSE_FP
> >
> >
> > The second part of the patch adds the equivalent MS Windows assembler
for
> > the same sections that currently have GCC x86 assembler code.
> >
> > Notes about implementation: We took the easiest route when hacking in
the
> > flag support which was to add a global flag for the entire library at
> > runtime and extern it in all the various files.
> > Jean-Marc: We looked at adding the flag into the state structures,
however
> > they were not passed all the way down into the filters.c files and it
> would
> > have been a massive change to make it pass all the needed data. The
> > approach we took should be ok since on a given machine you would have
the
> > same settings. The decoder_ctl and encoder_ctl set the same global
flag
> > variable.
> >
> > The way we setup the asm flags var should allow you to add the ARM
> assembly
> > in the exact same manor. You would add a check in the configure.in for
the
> > platform and define a _USE_ARM and place the code in the same
functions as
> > we did. You would then add a SPEEX_ASM_ARM 8 or something and let the
> > application decide to turn it on.
> >
> >
> > Other Notes: This patch obsoletes ltp_sse.h and filters_sse.h .
However
> > the patch does not remove them. This is thge updated version of the
patch
> > we sent in November.
> >
> >
> > Comments are welcome. BTW, we have been shipping our Video
Conferencing
> > product which only uses the Speex codec for 6 months now and have
gotten
> > rave reviews (PC Magazine Editors choice) for the audio and video
quality.
> > We use Speex in Windows, Mac OS-X, and Linux as we have clients for
each
> > platform. Keep up the great work! Check us out at
> > http://www.sightspeed.com and please try our beta version (Mac and
Windows
> > Clients available now) at http://www.sightspeed.com/page.php?page=beta
>
>--
>Jean-Marc Valin, M.Sc.A., ing. jr.
>LABORIUS (http://www.gel.usherb.ca/laborius)
>Université de Sherbrooke, Québec, Canada
<p>--- >8 ----
List archives: http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to
'speex-dev-request@xiph.org'
containing only the word 'unsubscribe' in the body. No subject is
needed.
Unsubscribe messages sent to the list will be ignored/filtered.
Apparently Analagous Threads
- [LLVMdev] llvm.x86.sse2.sqrt.pd not using sqrtpd, calling a function that modifies ECX
- [LLVMdev] Can LLVM vectorize <2 x i32> type
- [LLVMdev] Optimized code analysis problems
- [LLVMdev] RFB: Would like to flip the vector shuffle legality flag
- [LLVMdev] RFB: Would like to flip the vector shuffle legality flag