From 4fff623cae914340d0968855630d21ca69826b9a Mon Sep 17 00:00:00 2001
From: Juergen Keil <jkeil@users.sourceforge.net>
Date: Sat, 1 Sep 2001 17:57:20 +0000
Subject: Enable MMX accelerated code on x86 Workaround a gcc-2.95.2 compiler
 bug in put_pixels_clamped_mmx, add_pixels_clamped_mmx Fix "debug" build for
 libffmpeg

CVS patchset: 541
CVS date: 2001/09/01 17:57:20
---
 src/libffmpeg/libavcodec/dsputil.c     |  9 +++++++--
 src/libffmpeg/libavcodec/dsputil.h     | 15 ++-------------
 src/libffmpeg/libavcodec/dsputil_mmx.c |  8 +++++---
 src/libffmpeg/libavcodec/mpegvideo.c   |  2 +-
 src/libffmpeg/libavcodec/mpegvideo.h   |  2 +-
 5 files changed, 16 insertions(+), 20 deletions(-)

(limited to 'src/libffmpeg/libavcodec')

diff --git a/src/libffmpeg/libavcodec/dsputil.c b/src/libffmpeg/libavcodec/dsputil.c
index da8b24941..c29574303 100644
--- a/src/libffmpeg/libavcodec/dsputil.c
+++ b/src/libffmpeg/libavcodec/dsputil.c
@@ -20,6 +20,7 @@
 #include <stdio.h>
 #include "avcodec.h"
 #include "dsputil.h"
+#include "cpu_accel.h"
 
 void (*ff_idct)(DCTELEM *block);
 void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size);
@@ -413,9 +414,13 @@ void block_permute(INT16 *block)
 void dsputil_init(void)
 {
     int i, j;
+#ifdef ARCH_X86
+    uint32_t mm = mm_accel();
+#endif
     int use_permuted_mmx_idct;
     int accel_dsputil;
 
+
     for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;
     for(i=0;i<MAX_NEG_CROP;i++) {
         cropTbl[i] = 0;
@@ -440,8 +445,8 @@ void dsputil_init(void)
     use_permuted_mmx_idct = 1;
     accel_dsputil = 0;
 
-#ifdef HAVE_MMX
-    if (!accel_dsputil) {
+#ifdef ARCH_X86
+    if (!accel_dsputil && (mm & MM_ACCEL_X86_MMX)) {
 	dsputil_init_mmx();
 	accel_dsputil = 1;
 	/* printf("AVCODEC: Using mmx idct\n"); */
diff --git a/src/libffmpeg/libavcodec/dsputil.h b/src/libffmpeg/libavcodec/dsputil.h
index 1189747f3..f5e22e4c4 100644
--- a/src/libffmpeg/libavcodec/dsputil.h
+++ b/src/libffmpeg/libavcodec/dsputil.h
@@ -74,23 +74,12 @@ static inline int block_permute_op(int j)
 
 void block_permute(INT16 *block);
 
-#if defined(HAVE_MMX)
+#if defined(ARCH_X86)
 
-#define MM_MMX    0x0001 /* standard MMX */
-#define MM_3DNOW  0x0004 /* AMD 3DNOW */
-#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
-#define MM_SSE    0x0008 /* SSE functions */
-#define MM_SSE2   0x0010 /* PIV SSE2 functions */
+#include "cpu_accel.h"
 
 extern int mm_flags;
 
-int mm_support(void);
-
-static inline void emms(void)
-{
-    __asm __volatile ("emms;":::"memory");
-}
-
 #define emms_c() \
 {\
     if (mm_flags & MM_MMX)\
diff --git a/src/libffmpeg/libavcodec/dsputil_mmx.c b/src/libffmpeg/libavcodec/dsputil_mmx.c
index 29a069566..44b0b23d2 100644
--- a/src/libffmpeg/libavcodec/dsputil_mmx.c
+++ b/src/libffmpeg/libavcodec/dsputil_mmx.c
@@ -105,6 +105,7 @@ static void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line
     const DCTELEM *p;
     UINT8 *pix;
     int i;
+    static int p_inc = 32;  /* hack to avoid gcc-2.95.2 loop unrolling bug */
 
     /* read the pixels */
     p = block;
@@ -132,7 +133,7 @@ static void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line
 		:"m"(*p)
 		:"memory");
         pix += line_size*4;
-        p += 32;
+        p += p_inc;
     }
 }
 
@@ -141,6 +142,7 @@ static void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line
     const DCTELEM *p;
     UINT8 *pix;
     int i;
+    static int p_inc = 16;  /* hack to avoid gcc-2.95.2 loop unrolling bug */
 
     /* read the pixels */
     p = block;
@@ -172,7 +174,7 @@ static void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line
 		:"m"(*p)
 		:"memory");
         pix += line_size*2;
-        p += 16;
+        p += p_inc;
     }
 }
 
@@ -965,7 +967,7 @@ static void   sub_pixels_xy2_mmx( DCTELEM  *block, const UINT8 *pixels, int line
 
 void dsputil_init_mmx(void)
 {
-    mm_flags = mm_support();
+    mm_flags = mm_accel();
 #if 0
     printf("CPU flags:");
     if (mm_flags & MM_MMX)
diff --git a/src/libffmpeg/libavcodec/mpegvideo.c b/src/libffmpeg/libavcodec/mpegvideo.c
index ea7979bb8..5c66b3189 100644
--- a/src/libffmpeg/libavcodec/mpegvideo.c
+++ b/src/libffmpeg/libavcodec/mpegvideo.c
@@ -106,7 +106,7 @@ int MPV_common_init(MpegEncContext *s)
     else
         s->dct_unquantize = dct_unquantize_mpeg1_c;
         
-#ifdef HAVE_MMX
+#ifdef ARCH_X86
     MPV_common_init_mmx(s);
 #endif
     s->mb_width = (s->width + 15) / 16;
diff --git a/src/libffmpeg/libavcodec/mpegvideo.h b/src/libffmpeg/libavcodec/mpegvideo.h
index e4a5617e8..31243471e 100644
--- a/src/libffmpeg/libavcodec/mpegvideo.h
+++ b/src/libffmpeg/libavcodec/mpegvideo.h
@@ -188,7 +188,7 @@ void MPV_common_end(MpegEncContext *s);
 void MPV_decode_mb(MpegEncContext *s, DCTELEM block[6][64]);
 void MPV_frame_start(MpegEncContext *s);
 void MPV_frame_end(MpegEncContext *s);
-#ifdef HAVE_MMX
+#ifdef ARCH_X86
 void MPV_common_init_mmx(MpegEncContext *s);
 #endif
 
-- 
cgit v1.2.3