changeset 3203:8312f4bc8dab

Disable SSE code and reenable FPU dct for SSE cpus (fpu code is 0.3% faster and I don't get data aligned in dct64_sse.s, so I can't finish optimizing it)

--- a/mp3lib/Makefile	Thu Nov 29 18:00:57 2001 +0000
+++ b/mp3lib/Makefile	Thu Nov 29 18:05:42 2001 +0000
@@ -15,10 +15,10 @@
 SRCS += decode_MMX.s dct64_MMX.s tabinit_MMX.s
 OBJS += decode_MMX.o dct64_MMX.o tabinit_MMX.o
 endif
-ifeq ($(TARGET_SSE),yes)
-SRCS += dct64_sse.s
-OBJS += dct64_sse.o
-endif
+#ifeq ($(TARGET_SSE),yes)
+#SRCS += dct64_sse.s
+#OBJS += dct64_sse.o
+#endif
 ifeq ($(TARGET_3DNOW),yes)
 SRCS += dct36_3dnow.s dct64_3dnow.s
 OBJS += dct36_3dnow.o dct64_3dnow.o

--- a/mp3lib/dct64_sse.s	Thu Nov 29 18:00:57 2001 +0000
+++ b/mp3lib/dct64_sse.s	Thu Nov 29 18:05:42 2001 +0000
@@ -1,9 +1,13 @@
-# This code is a translation of dct64_k7.s from MPlayer.
-# Coded by Felix Buenemann <atmosfear at users.sourceforge.net>
-#
-# TODO: - fix phases 4 and 5 (sse)
-#       - optimize scalar FPU code? (interleave with sse code)
-#
+/ This code is a translation of dct64_k7.s from MPlayer.
+/ Coded by Felix Buenemann <atmosfear at users.sourceforge.net>
+/
+/ TODO: - fix phases 4 and 5 (sse)
+/       - optimize scalar FPU code? (interleave with sse code)
+/       - fix alignment (prohibits finishing this code)
+/       - then use faster insns for aligned data
+/
+/ Note: currently code is disabled as I couldn't get input data aligned!
+/
 
 //.data
 //	.align 8

--- a/mp3lib/sr1.c	Thu Nov 29 18:00:57 2001 +0000
+++ b/mp3lib/sr1.c	Thu Nov 29 18:05:42 2001 +0000
@@ -409,8 +409,8 @@
        Note: It's ok, Since K8 will have SSE2 support and will much faster
        of P4 ;) 
      */
-      printf( "mp3lib: Using SSE%s! optimized decore.\n",(_isse>1?"2":""));
-//      printf( "mp3lib: Using Pentium%s optimized decore.\n",(_i586>1?"-MMX":""));
+//      printf( "mp3lib: Using SSE%s! optimized decore.\n",(_isse>1?"2":""));
+      printf( "mp3lib: Using Pentium%s optimized decore.\n",(_i586>1?"-MMX":""));
     else
     if(_3dnow)
       printf( "mp3lib: Using AMD 3dnow%s! optimized decore.\n",(_3dnow>1?"-dsp(k7)":""));
@@ -443,14 +443,14 @@
     tables_done_flag=1;
 
     dct36_func=dct36;
-#ifdef HAVE_SSE
+/*#ifdef HAVE_SSE
   if(_isse)
   {
     synth_func=synth_1to1_MMX;
     dct64_MMX_func=dct64_MMX_sse;
   }    
   else
-#endif
+#endif*/
 #ifdef HAVE_3DNOWEX
   if ( _3dnow > 1 )
   {
@@ -518,7 +518,7 @@
     tables_done_flag=1;
 
     dct36_func=dct36;
-#ifdef HAVE_SSE
+/*#ifdef HAVE_SSE
   if(gCpuCaps.hasSSE)
   {
     synth_func=synth_1to1_MMX;
@@ -526,7 +526,7 @@
     printf("mp3lib: using SSE optimized decore!\n");
   }    
   else
-#endif
+#endif*/
 #ifdef HAVE_3DNOWEX
   if (gCpuCaps.has3DNowExt)
   {