1 files changed, 52 insertions, 15 deletions

diff --git a/src/libffmpeg/libavcodec/libpostproc/postprocess.c b/src/libffmpeg/libavcodec/libpostproc/postprocess.c
index 11267b666..093d94aea 100644
--- a/src/libffmpeg/libavcodec/libpostproc/postprocess.c
+++ b/src/libffmpeg/libavcodec/libpostproc/postprocess.c
@@ -70,9 +70,8 @@ try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
 #include "config.h"
 #include <inttypes.h>
 #include <stdio.h>
-
-#include "xineutils.h"
-
+#include <stdlib.h>
+#include <string.h>
 #ifdef HAVE_MALLOC_H
 #include <malloc.h>
 #endif
@@ -81,9 +80,9 @@ try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
 //#undef HAVE_MMX
 //#undef ARCH_X86
 //#define DEBUG_BRIGHTNESS
-
-#define memcpy(a,b,c) xine_fast_memcpy(a,b,c)
-
+#ifdef USE_FASTMEMCPY
+#include "../fastmemcpy.h"
+#endif
 #include "postprocess.h"
 #include "postprocess_internal.h"
 
@@ -105,13 +104,13 @@ try to unroll inner for(x=0 ... loop to avoid these damn if(x ... checks
 //#define NUM_BLOCKS_AT_ONCE 16 //not used yet
 
 #ifdef ARCH_X86
-static const uint64_t __attribute__((aligned(8))) w05=		0x0005000500050005ULL;
-static const uint64_t __attribute__((aligned(8))) w20=		0x0020002000200020ULL;
-static const uint64_t __attribute__((aligned(8))) b00= 		0x0000000000000000ULL;
-static const uint64_t __attribute__((aligned(8))) b01= 		0x0101010101010101ULL;
-static const uint64_t __attribute__((aligned(8))) b02= 		0x0202020202020202ULL;
-static const uint64_t __attribute__((aligned(8))) b08= 		0x0808080808080808ULL;
-static const uint64_t __attribute__((aligned(8))) b80= 		0x8080808080808080ULL;
+static uint64_t __attribute__((aligned(8))) w05=		0x0005000500050005LL;
+static uint64_t __attribute__((aligned(8))) w20=		0x0020002000200020LL;
+static uint64_t __attribute__((aligned(8))) b00= 		0x0000000000000000LL;
+static uint64_t __attribute__((aligned(8))) b01= 		0x0101010101010101LL;
+static uint64_t __attribute__((aligned(8))) b02= 		0x0202020202020202LL;
+static uint64_t __attribute__((aligned(8))) b08= 		0x0808080808080808LL;
+static uint64_t __attribute__((aligned(8))) b80= 		0x8080808080808080LL;
 #endif
 
 
@@ -153,6 +152,44 @@ static char *replaceTable[]=
 	NULL //End Marker
 };
 
+#ifdef ARCH_X86
+static inline void unusedVariableWarningFixer()
+{
+	if(w05 + w20 + b00 + b01 + b02 + b08 + b80 == 0) b00=0;
+}
+#endif
+
+
+#ifdef ARCH_X86
+static inline void prefetchnta(void *p)
+{
+	asm volatile(	"prefetchnta (%0)\n\t"
+		: : "r" (p)
+	);
+}
+
+static inline void prefetcht0(void *p)
+{
+	asm volatile(	"prefetcht0 (%0)\n\t"
+		: : "r" (p)
+	);
+}
+
+static inline void prefetcht1(void *p)
+{
+	asm volatile(	"prefetcht1 (%0)\n\t"
+		: : "r" (p)
+	);
+}
+
+static inline void prefetcht2(void *p)
+{
+	asm volatile(	"prefetcht2 (%0)\n\t"
+		: : "r" (p)
+	);
+}
+#endif
+
 // The horizontal Functions exist only in C cuz the MMX code is faster with vertical filters and transposing
 
 /**
@@ -906,10 +943,10 @@ for(y=0; y<mbHeight; y++){
 		int i;
 		const int count= mbHeight * QPStride;
 		for(i=0; i<(count>>2); i++){
-			((uint32_t*)c->nonBQPTable)[i] = ((uint32_t*)QP_store)[i] & 0x1F1F1F1F;
+			((uint32_t*)c->nonBQPTable)[i] = ((uint32_t*)QP_store)[i] & 0x3F3F3F3F;
 		}
 		for(i<<=2; i<count; i++){
-			c->nonBQPTable[i] = QP_store[i] & 0x1F;
+			c->nonBQPTable[i] = QP_store[i] & 0x3F;
 		}
 	}