*/
/**
- * @file postprocess.c
+ * @file libpostproc/postprocess.c
* postprocessing.
*/
* I do not have a 3DNow! CPU -> it is untested, but no one said it does not work so it seems to work
# more or less selfinvented filters so the exactness is not too meaningful
E = Exact implementation
-e = allmost exact implementation (slightly different rounding,...)
+e = almost exact implementation (slightly different rounding,...)
a = alternative / approximate impl
c = checked against the other implementations (-vo md5)
p = partially optimized, still some work to do
//Changelog: use the Subversion log
#include "config.h"
-#include "avutil.h"
+#include "libavutil/avutil.h"
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
-#ifdef HAVE_MALLOC_H
-#include <malloc.h>
-#endif
//#undef HAVE_MMX2
-//#define HAVE_3DNOW
+//#define HAVE_AMD3DNOW
//#undef HAVE_MMX
//#undef ARCH_X86
//#define DEBUG_BRIGHTNESS
#include "postprocess.h"
#include "postprocess_internal.h"
-#ifdef HAVE_ALTIVEC_H
+unsigned postproc_version(void)
+{
+ return LIBPOSTPROC_VERSION_INT;
+}
+
+#if HAVE_ALTIVEC_H
#include <altivec.h>
#endif
#define TEMP_STRIDE 8
//#define NUM_BLOCKS_AT_ONCE 16 //not used yet
-#if defined(ARCH_X86)
+#if ARCH_X86
DECLARE_ASM_CONST(8, uint64_t, w05)= 0x0005000500050005LL;
DECLARE_ASM_CONST(8, uint64_t, w04)= 0x0004000400040004LL;
DECLARE_ASM_CONST(8, uint64_t, w20)= 0x0020002000200020LL;
};
-#if defined(ARCH_X86)
+#if ARCH_X86
static inline void prefetchnta(void *p)
{
- asm volatile( "prefetchnta (%0)\n\t"
+ __asm__ volatile( "prefetchnta (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht0(void *p)
{
- asm volatile( "prefetcht0 (%0)\n\t"
+ __asm__ volatile( "prefetcht0 (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht1(void *p)
{
- asm volatile( "prefetcht1 (%0)\n\t"
+ __asm__ volatile( "prefetcht1 (%0)\n\t"
: : "r" (p)
);
}
static inline void prefetcht2(void *p)
{
- asm volatile( "prefetcht2 (%0)\n\t"
+ __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
+/* The horizontal functions exist only in C because the MMX
+ * code is faster with vertical filters and transposing. */
/**
* Check if the given 8x8 Block is mostly "flat"
//Note: we have C, MMX, MMX2, 3DNOW version there is no 3DNOW+MMX2 one
//Plain C versions
-#if !defined (HAVE_MMX) || defined (RUNTIME_CPUDETECT)
+#if !(HAVE_MMX || HAVE_ALTIVEC) || CONFIG_RUNTIME_CPUDETECT
#define COMPILE_C
#endif
-#ifdef HAVE_ALTIVEC
+#if HAVE_ALTIVEC
#define COMPILE_ALTIVEC
#endif //HAVE_ALTIVEC
-#if defined(ARCH_X86)
+#if ARCH_X86
-#if (defined (HAVE_MMX) && !defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#if (HAVE_MMX && !HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
#define COMPILE_MMX
#endif
-#if defined (HAVE_MMX2) || defined (RUNTIME_CPUDETECT)
+#if HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT
#define COMPILE_MMX2
#endif
-#if (defined (HAVE_3DNOW) && !defined (HAVE_MMX2)) || defined (RUNTIME_CPUDETECT)
+#if (HAVE_AMD3DNOW && !HAVE_MMX2) || CONFIG_RUNTIME_CPUDETECT
#define COMPILE_3DNOW
#endif
-#endif /* defined(ARCH_X86) */
+#endif /* ARCH_X86 */
#undef HAVE_MMX
+#define HAVE_MMX 0
#undef HAVE_MMX2
-#undef HAVE_3DNOW
+#define HAVE_MMX2 0
+#undef HAVE_AMD3DNOW
+#define HAVE_AMD3DNOW 0
#undef HAVE_ALTIVEC
+#define HAVE_ALTIVEC 0
#ifdef COMPILE_C
-#undef HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_3DNOW
#define RENAME(a) a ## _C
#include "postprocess_template.c"
#endif
#ifdef COMPILE_ALTIVEC
#undef RENAME
-#define HAVE_ALTIVEC
+#undef HAVE_ALTIVEC
+#define HAVE_ALTIVEC 1
#define RENAME(a) a ## _altivec
#include "postprocess_altivec_template.c"
#include "postprocess_template.c"
//MMX versions
#ifdef COMPILE_MMX
#undef RENAME
-#define HAVE_MMX
-#undef HAVE_MMX2
-#undef HAVE_3DNOW
+#undef HAVE_MMX
+#define HAVE_MMX 1
#define RENAME(a) a ## _MMX
#include "postprocess_template.c"
#endif
//MMX2 versions
#ifdef COMPILE_MMX2
#undef RENAME
-#define HAVE_MMX
-#define HAVE_MMX2
-#undef HAVE_3DNOW
+#undef HAVE_MMX
+#undef HAVE_MMX2
+#define HAVE_MMX 1
+#define HAVE_MMX2 1
#define RENAME(a) a ## _MMX2
#include "postprocess_template.c"
#endif
//3DNOW versions
#ifdef COMPILE_3DNOW
#undef RENAME
-#define HAVE_MMX
+#undef HAVE_MMX
#undef HAVE_MMX2
-#define HAVE_3DNOW
+#undef HAVE_AMD3DNOW
+#define HAVE_MMX 1
+#define HAVE_MMX2 0
+#define HAVE_AMD3DNOW 1
#define RENAME(a) a ## _3DNow
#include "postprocess_template.c"
#endif
// minor note: the HAVE_xyz is messed up after that line so do not use it.
static inline void postProcess(const uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
- const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode_t *vm, pp_context_t *vc)
+ const QP_STORE_T QPs[], int QPStride, int isColor, pp_mode *vm, pp_context *vc)
{
PPContext *c= (PPContext *)vc;
PPMode *ppMode= (PPMode *)vm;
// Using ifs here as they are faster than function pointers although the
// difference would not be measurable here but it is much better because
// someone might exchange the CPU whithout restarting MPlayer ;)
-#ifdef RUNTIME_CPUDETECT
-#if defined(ARCH_X86)
+#if CONFIG_RUNTIME_CPUDETECT
+#if ARCH_X86
// ordered per speed fastest first
if(c->cpuCaps & PP_CPU_CAPS_MMX2)
postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
else
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#else
-#ifdef HAVE_ALTIVEC
+#if HAVE_ALTIVEC
if(c->cpuCaps & PP_CPU_CAPS_ALTIVEC)
postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
else
#endif
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#endif
-#else //RUNTIME_CPUDETECT
-#ifdef HAVE_MMX2
+#else //CONFIG_RUNTIME_CPUDETECT
+#if HAVE_MMX2
postProcess_MMX2(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-#elif defined (HAVE_3DNOW)
+#elif HAVE_AMD3DNOW
postProcess_3DNow(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-#elif defined (HAVE_MMX)
+#elif HAVE_MMX
postProcess_MMX(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
-#elif defined (HAVE_ALTIVEC)
+#elif HAVE_ALTIVEC
postProcess_altivec(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#else
postProcess_C(src, srcStride, dst, dstStride, width, height, QPs, QPStride, isColor, c);
#endif
-#endif //!RUNTIME_CPUDETECT
+#endif //!CONFIG_RUNTIME_CPUDETECT
}
//static void postProcess(uint8_t src[], int srcStride, uint8_t dst[], int dstStride, int width, int height,
"\n"
;
-pp_mode_t *pp_get_mode_by_name_and_quality(const char *name, int quality)
+pp_mode *pp_get_mode_by_name_and_quality(const char *name, int quality)
{
char temp[GET_MODE_BUFFER_SIZE];
char *p= temp;
return ppMode;
}
-void pp_free_mode(pp_mode_t *mode){
+void pp_free_mode(pp_mode *mode){
av_free(mode);
}
for(i=0; i<3; i++){
//Note: The +17*1024 is just there so i do not have to worry about r/w over the end.
- reallocAlign((void **)&c->tempBlured[i], 8, stride*mbHeight*16 + 17*1024);
- reallocAlign((void **)&c->tempBluredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
+ reallocAlign((void **)&c->tempBlurred[i], 8, stride*mbHeight*16 + 17*1024);
+ reallocAlign((void **)&c->tempBlurredPast[i], 8, 256*((height+7)&(~7))/2 + 17*1024);//FIXME size
}
reallocAlign((void **)&c->deintTemp, 8, 2*width+32);
static const AVClass av_codec_context_class = { "Postproc", context_to_name, NULL };
-pp_context_t *pp_get_context(int width, int height, int cpuCaps){
+pp_context *pp_get_context(int width, int height, int cpuCaps){
PPContext *c= av_malloc(sizeof(PPContext));
int stride= (width+15)&(~15); //assumed / will realloc if needed
int qpStride= (width+15)/16 + 2; //assumed / will realloc if needed
PPContext *c = (PPContext*)vc;
int i;
- for(i=0; i<3; i++) av_free(c->tempBlured[i]);
- for(i=0; i<3; i++) av_free(c->tempBluredPast[i]);
+ for(i=0; i<3; i++) av_free(c->tempBlurred[i]);
+ for(i=0; i<3; i++) av_free(c->tempBlurredPast[i]);
av_free(c->tempBlocks);
av_free(c->yHistogram);
uint8_t * dst[3], const int dstStride[3],
int width, int height,
const QP_STORE_T *QP_store, int QPStride,
- pp_mode_t *vm, void *vc, int pict_type)
+ pp_mode *vm, void *vc, int pict_type)
{
int mbWidth = (width+15)>>4;
int mbHeight= (height+15)>>4;
projects / frescor / ffmpeg.git / blobdiff