changeset 23495:059d7c379fb3

Enable SIMD optmizations for mp3lib on AMD64

--- a/Changelog	Fri Jun 08 14:34:25 2007 +0000
+++ b/Changelog	Fri Jun 08 14:38:25 2007 +0000
@@ -23,6 +23,7 @@
     * handle resolution switching for Real codecs
     * ffmpeg video decoder handles aspect ratio changes
     * amr now handled via libamr wrapper (http://www.penguin.cz/~utx/amr)
+    * SIMD optimizations for mp3lib under AMD64
 
     Demuxers:
     * implemented switch_video and switch_program consistently with switch_audio

--- a/mp3lib/Makefile	Fri Jun 08 14:34:25 2007 +0000
+++ b/mp3lib/Makefile	Fri Jun 08 14:38:25 2007 +0000
@@ -3,12 +3,13 @@
 LIBNAME_COMMON = libmp3.a
 
 SRCS_COMMON = sr1.c
+SRCS_COMMON-$(TARGET_MMX)     += decode_MMX.c
+SRCS_COMMON-$(TARGET_SSE)     += dct64_sse.c
 ifeq ($(TARGET_ARCH_X86_32),yes)
 SRCS_COMMON                   += decode_i586.c
-SRCS_COMMON-$(TARGET_MMX)     += decode_MMX.c dct64_MMX.c
+SRCS_COMMON-$(TARGET_MMX)     += dct64_MMX.c
 SRCS_COMMON-$(TARGET_3DNOW)   += dct36_3dnow.c dct64_3dnow.c
 SRCS_COMMON-$(TARGET_3DNOWEX) += dct36_k7.c dct64_k7.c
-SRCS_COMMON-$(TARGET_SSE)     += dct64_sse.c
 endif
 SRCS_COMMON-$(TARGET_ALTIVEC) += dct64_altivec.c
 

--- a/mp3lib/decod386.c	Fri Jun 08 14:34:25 2007 +0000
+++ b/mp3lib/decod386.c	Fri Jun 08 14:38:25 2007 +0000
@@ -102,7 +102,7 @@
 
 static synth_func_t synth_func;
 
-#if defined(CAN_COMPILE_X86_ASM) && defined(HAVE_MMX)
+#ifdef HAVE_MMX
 extern int synth_1to1_MMX( real *bandPtr,int channel,short * samples);
 #endif
 
@@ -125,7 +125,7 @@
   *pnt += 128;
 
 /* optimized for x86 */
-#if defined(CAN_COMPILE_X86_ASM)
+#ifdef ARCH_X86
   if ( synth_func )
    {
 //    printf("Calling %p, bandPtr=%p channel=%d samples=%p\n",synth_func,bandPtr,channel,samples);

--- a/mp3lib/sr1.c	Fri Jun 08 14:34:25 2007 +0000
+++ b/mp3lib/sr1.c	Fri Jun 08 14:38:25 2007 +0000
@@ -32,8 +32,10 @@
 
 #include "libvo/fastmemcpy.h"
 
-#ifdef ARCH_X86_32
-#define CAN_COMPILE_X86_ASM
+#ifdef ARCH_X86_64
+// 3DNow! and 3DNow!Ext routines don't compile under AMD64
+#undef HAVE_3DNOW
+#undef HAVE_3DNOWEX
 #endif
 
 //static FILE* mp3_file=NULL;
@@ -137,7 +139,7 @@
 //  if(MP3_frames>=7741) printf("getbits_fast: bits=%d  bitsleft=%d  wordptr=%x\n",number_of_bits,bitsleft,wordpointer);
   if((bitsleft-=number_of_bits)<0) return 0;
   if(!number_of_bits) return 0;
-#if defined(CAN_COMPILE_X86_ASM)
+#ifdef ARCH_X86
   rval = bswap_16(*((uint16_t *)wordpointer));
 #else
   /*
@@ -180,7 +182,7 @@
 
 LOCAL int stream_head_read(unsigned char *hbuf,uint32_t *newhead){
   if(mp3_read(hbuf,4) != 4) return FALSE;
-#if defined(CAN_COMPILE_X86_ASM)
+#ifdef ARCH_X86
   *newhead = bswap_32(*((uint32_t*)hbuf));
 #else
   /*
@@ -415,8 +417,6 @@
 
     make_decode_tables(outscale);
 
-#ifdef CAN_COMPILE_X86_ASM
-
 #ifdef HAVE_MMX
     if (gCpuCaps.hasMMX)
     {
@@ -451,6 +451,7 @@
     }
     else
 #endif
+#ifdef ARCH_X86_32
 #ifdef HAVE_MMX
     if (gCpuCaps.hasMMX)
     {
@@ -465,7 +466,7 @@
 	mp_msg(MSGT_DECAUDIO,MSGL_V,"mp3lib: using Pentium optimized decore!\n");
     }
     else
-#endif
+#endif /* ARCH_X86_32 */
 #ifdef HAVE_ALTIVEC
     if (gCpuCaps.hasAltiVec)
     {