libmpeg2 patches for gcc3.1 from michel

CVS patchset: 2018
CVS date: 2002/06/04 23:07:56

6 files changed, 69 insertions, 4 deletions

diff --git a/configure.in b/configure.in
index f92f6d59e..cd84bbfd7 100644
--- a/configure.in
+++ b/configure.in
@@ -148,7 +148,7 @@ AC_C_BIGENDIAN
 dnl AC_C_BIGENDIAN triggers an AC_TRY_RUN warning; we can't cross compile
 dnl xine (oh, well)
 AC_C_CONST
-AC_C_INLINE
+AC_C_ALWAYS_INLINE
 AC_TYPE_OFF_T
 AC_TYPE_SIZE_T
 AM_TYPE_PTRDIFF_T
diff --git a/src/libmpeg2/idct.c b/src/libmpeg2/idct.c
index 1c577aa5f..abc87a0a1 100644
--- a/src/libmpeg2/idct.c
+++ b/src/libmpeg2/idct.c
@@ -223,6 +223,9 @@ static void mpeg2_idct_copy_c (int16_t * block, uint8_t * dest, int stride)
 	dest[6] = CLIP (block[6]);
 	dest[7] = CLIP (block[7]);
 
+	block[0] = 0;   block[1] = 0;   block[2] = 0;   block[3] = 0;
+	block[4] = 0;   block[5] = 0;   block[6] = 0;   block[7] = 0;
+
 	dest += stride;
 	block += 8;
     } while (--i);
@@ -249,6 +252,9 @@ static void mpeg2_idct_add_c (int16_t * block, uint8_t * dest, int stride)
 	dest[6] = CLIP (block[6] + dest[6]);
 	dest[7] = CLIP (block[7] + dest[7]);
 
+	block[0] = 0;   block[1] = 0;   block[2] = 0;   block[3] = 0;
+	block[4] = 0;   block[5] = 0;   block[6] = 0;   block[7] = 0;
+
 	dest += stride;
 	block += 8;
     } while (--i);
diff --git a/src/libmpeg2/idct_altivec.c b/src/libmpeg2/idct_altivec.c
index 1d20f56b4..b982238e7 100644
--- a/src/libmpeg2/idct_altivec.c
+++ b/src/libmpeg2/idct_altivec.c
@@ -243,6 +243,22 @@ void mpeg2_idct_copy_altivec (int16_t * block, uint8_t * dest, int stride)
 	"#	lwz		%r0,  132(%r1)		\n"
 	"#	mtlr		%r0			\n"
 	"#	la		%r1,  128(%r1)		\n"
+        "       vxor            %v1,  %v1,  %v1         \n"
+        "       addi            %r9,  %r3,  16          \n"
+        "       stvx            %v1,  0,    %r3         \n"
+        "       stvx            %v1,  0,    %r9         \n"
+        "       addi            %r11, %r3,  32          \n"
+        "       stvx            %v1,  0,    %r11        \n" 
+        "       addi            %r9,  %r3,  48          \n" 
+        "       stvx            %v1,  0,    %r9         \n" 
+        "       addi            %r11, %r3,  -64         \n" 
+        "       stvx            %v1,  0,    %r11        \n" 
+        "       addi            %r9,  %r3,  -48         \n" 
+        "       stvx            %v1,  0,    %r9         \n" 
+        "       addi            %r11, %r3,  -32         \n" 
+        "       stvx            %v1,  0,    %r11        \n" 
+        "       addi            %r3,  %r3,  -16         \n" 
+        "       stvx            %v1,  0,    %r3         \n"
 	 );
 }
 
@@ -463,6 +479,21 @@ void mpeg2_idct_add_altivec (int16_t * block, uint8_t * dest, int stride)
 	"#	lwz		%r0,  196(%r1)		\n"
 	"#	mtlr		%r0			\n"
 	"#	la		%r1,  192(%r1)		\n"
+        "       addi            %r9,  %r3,  16          \n"
+        "       stvx            %v1,  0,    %r3         \n"
+        "       stvx            %v1,  0,    %r9         \n"
+        "       addi            %r11, %r3,  32          \n" 
+        "       stvx            %v1,  0,    %r11        \n" 
+        "       addi            %r9,  %r3,  48          \n" 
+        "       stvx            %v1,  0,    %r9         \n" 
+        "       addi            %r11, %r3,  -64         \n" 
+        "       stvx            %v1,  0,    %r11        \n" 
+        "       addi            %r9,  %r3,  -48         \n" 
+        "       stvx            %v1,  0,    %r9         \n" 
+        "       addi            %r11, %r3,  -32         \n"  
+        "       stvx            %v1,  0,    %r11        \n"   
+        "       addi            %r3,  %r3,  -16         \n"   
+        "       stvx            %v1,  0,    %r3         \n"
 	 );
 }
 
@@ -625,6 +656,7 @@ void mpeg2_idct_copy_altivec (vector_s16_t * block, unsigned char * dest,
     COPY (dest, vx5)	dest += stride;
     COPY (dest, vx6)	dest += stride;
     COPY (dest, vx7)
+    memset (block, 0, 64 * sizeof (signed short));
 }
 
 void mpeg2_idct_add_altivec (vector_s16_t * block, unsigned char * dest,
@@ -661,6 +693,9 @@ void mpeg2_idct_add_altivec (vector_s16_t * block, unsigned char * dest,
     ADD (dest, vx5, perm1)	dest += stride;
     ADD (dest, vx6, perm0)	dest += stride;
     ADD (dest, vx7, perm1)
+    memset (block, 0, 64 * sizeof (signed short));
 }
 
 #endif	/* __ALTIVEC__ */
+
+
diff --git a/src/libmpeg2/idct_mlib.c b/src/libmpeg2/idct_mlib.c
index 18097f51c..8fc4e41a2 100644
--- a/src/libmpeg2/idct_mlib.c
+++ b/src/libmpeg2/idct_mlib.c
@@ -28,6 +28,7 @@
 #include <mlib_status.h>
 #include <mlib_sys.h>
 #include <mlib_video.h>
+#include <string.h>
 #include <inttypes.h>
 
 #include "mpeg2_internal.h"
@@ -36,18 +37,21 @@ void mpeg2_idct_add_mlib (int16_t * block, uint8_t * dest, int stride)
 {
     mlib_VideoIDCT_IEEE_S16_S16 (block, block);
     mlib_VideoAddBlock_U8_S16 (dest, block, stride);
+    memset (block, 0, 64 * sizeof (uint16_t);
 }
 
 void mpeg2_idct_copy_mlib_non_ieee (int16_t * block, uint8_t * dest,
 				    int stride)
 {
     mlib_VideoIDCT8x8_U8_S16 (dest, block, stride);
+    memset (block, 0, 64 * sizeof (uint16_t);
 }
 
 void mpeg2_idct_add_mlib_non_ieee (int16_t * block, uint8_t * dest, int stride)
 {
     mlib_VideoIDCT8x8_S16_S16 (block, block);
     mlib_VideoAddBlock_U8_S16 (dest, block, stride);
+    memset (block, 0, 64 * sizeof (uint16_t);
 }
 
 #endif
diff --git a/src/libmpeg2/idct_mmx.c b/src/libmpeg2/idct_mmx.c
index d8822f6d9..bee54e83b 100644
--- a/src/libmpeg2/idct_mmx.c
+++ b/src/libmpeg2/idct_mmx.c
@@ -654,6 +654,25 @@ static void block_add (int16_t * block, uint8_t * dest, int stride)
     movq_r2m (mm3, *(dest+stride));
 }
 
+static inline void block_zero (int16_t * block)   {       
+  pxor_r2r (mm0, mm0);
+  movq_r2m (mm0, *(block+0*4));
+  movq_r2m (mm0, *(block+1*4));       
+  movq_r2m (mm0, *(block+2*4));       
+  movq_r2m (mm0, *(block+3*4));       
+  movq_r2m (mm0, *(block+4*4));       
+  movq_r2m (mm0, *(block+5*4));       
+  movq_r2m (mm0, *(block+6*4));       
+  movq_r2m (mm0, *(block+7*4));       
+  movq_r2m (mm0, *(block+8*4));       
+  movq_r2m (mm0, *(block+9*4));       
+  movq_r2m (mm0, *(block+10*4));       
+  movq_r2m (mm0, *(block+11*4));       
+  movq_r2m (mm0, *(block+12*4));       
+  movq_r2m (mm0, *(block+13*4));       
+  movq_r2m (mm0, *(block+14*4));       
+  movq_r2m (mm0, *(block+15*4));   
+}
 
 declare_idct (mmxext_idct, mmxext_table,
 	      mmxext_row_head, mmxext_row, mmxext_row_tail, mmxext_row_mid)
@@ -662,12 +681,14 @@ void mpeg2_idct_copy_mmxext (int16_t * block, uint8_t * dest, int stride)
 {
     mmxext_idct (block);
     block_copy (block, dest, stride);
+    block_zero (block);
 }
 
 void mpeg2_idct_add_mmxext (int16_t * block, uint8_t * dest, int stride)
 {
     mmxext_idct (block);
     block_add (block, dest, stride);
+    block_zero (block);
 }
 
 
@@ -678,12 +699,14 @@ void mpeg2_idct_copy_mmx (int16_t * block, uint8_t * dest, int stride)
 {
     mmx_idct (block);
     block_copy (block, dest, stride);
+    block_zero (block);
 }
 
 void mpeg2_idct_add_mmx (int16_t * block, uint8_t * dest, int stride)
 {
     mmx_idct (block);
     block_add (block, dest, stride);
+    block_zero (block);
 }
 
 void mpeg2_idct_mmx_init (void)
diff --git a/src/libmpeg2/slice.c b/src/libmpeg2/slice.c
index 07f92ea10..8913b208b 100644
--- a/src/libmpeg2/slice.c
+++ b/src/libmpeg2/slice.c
@@ -23,7 +23,6 @@
 
 #include "config.h"
 
-#include <string.h>
 #include <inttypes.h>
 
 #include "video_out.h"
@@ -966,7 +965,6 @@ static inline void slice_intra_DCT (picture_t * picture, int cc,
 	picture->dc_dct_pred[cc] += get_chroma_dc_dct_diff (picture);
     picture->DCTblock[0] =
 	picture->dc_dct_pred[cc] << (3 - picture->intra_dc_precision);
-    memset (picture->DCTblock + 1, 0, 63 * sizeof (int16_t));
 
     if (picture->mpeg1) {
 	if (picture->picture_coding_type != D_TYPE)
@@ -984,7 +982,6 @@ static inline void slice_intra_DCT (picture_t * picture, int cc,
 static inline void slice_non_intra_DCT (picture_t * picture, uint8_t * dest,
 					int stride)
 {
-    memset (picture->DCTblock, 0, 64 * sizeof (int16_t));
     if (picture->mpeg1)
 	get_mpeg1_non_intra_block (picture);
     else