* using DSPContext - so each codec could use its local (sub)set of CPU extension

Originally committed as revision 1194 to svn://svn.ffmpeg.org/ffmpeg/trunk

1 files changed, 55 insertions, 53 deletions

diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 3f9b5893d8..aa7277ee7b 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -45,10 +45,9 @@ extern const UINT8 ff_zigzag_direct[64];
 extern UINT32 squareTbl[512];
 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
 
-void dsputil_init(void);
 
 /* minimum alignment rules ;)
-if u notice errors in the align stuff, need more alignment for some asm code for some cpu 
+if u notice errors in the align stuff, need more alignment for some asm code for some cpu
 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
 
 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
@@ -57,39 +56,20 @@ i (michael) didnt check them, these are just the alignents which i think could b
 !future video codecs might need functions with less strict alignment
 */
 
-/* pixel ops : interface with DCT */
-extern void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
-extern void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
-extern void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
-extern void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
-extern void (*ff_gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
-extern void (*ff_gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy, 
-                  int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
-extern void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
-extern int (*pix_sum)(UINT8 * pix, int line_size);
-extern int (*pix_norm1)(UINT8 * pix, int line_size);
-
-
-
+/*
 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
 void clear_blocks_c(DCTELEM *blocks);
+*/
 
 /* add and put pixel (decoding) */
 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
 
-extern op_pixels_func put_pixels_tab[2][4];
-extern op_pixels_func avg_pixels_tab[2][4];
-extern op_pixels_func put_no_rnd_pixels_tab[2][4];
-extern op_pixels_func avg_no_rnd_pixels_tab[2][4];
-extern qpel_mc_func put_qpel_pixels_tab[2][16];
-extern qpel_mc_func avg_qpel_pixels_tab[2][16];
-extern qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
-extern qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
+
 
 #define CALL_2X_PIXELS(a, b, n)\
 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
@@ -100,20 +80,46 @@ static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 /* motion estimation */
 
 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size);
-
-extern op_pixels_abs_func pix_abs16x16;
-extern op_pixels_abs_func pix_abs16x16_x2;
-extern op_pixels_abs_func pix_abs16x16_y2;
-extern op_pixels_abs_func pix_abs16x16_xy2;
-extern op_pixels_abs_func pix_abs8x8;
-extern op_pixels_abs_func pix_abs8x8_x2;
-extern op_pixels_abs_func pix_abs8x8_y2;
-extern op_pixels_abs_func pix_abs8x8_xy2;
-
+/*
 int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
 int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
 int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
 int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
+*/
+typedef struct DSPContext {
+    /* pixel ops : interface with DCT */
+    void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
+    void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
+    void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
+    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
+    void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
+    void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
+		    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
+    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
+    int (*pix_sum)(UINT8 * pix, int line_size);
+    int (*pix_norm1)(UINT8 * pix, int line_size);
+
+    /* maybe create an array for 16/8 functions */
+    op_pixels_func put_pixels_tab[2][4];
+    op_pixels_func avg_pixels_tab[2][4];
+    op_pixels_func put_no_rnd_pixels_tab[2][4];
+    op_pixels_func avg_no_rnd_pixels_tab[2][4];
+    qpel_mc_func put_qpel_pixels_tab[2][16];
+    qpel_mc_func avg_qpel_pixels_tab[2][16];
+    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
+    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
+
+    op_pixels_abs_func pix_abs16x16;
+    op_pixels_abs_func pix_abs16x16_x2;
+    op_pixels_abs_func pix_abs16x16_y2;
+    op_pixels_abs_func pix_abs16x16_xy2;
+    op_pixels_abs_func pix_abs8x8;
+    op_pixels_abs_func pix_abs8x8_x2;
+    op_pixels_abs_func pix_abs8x8_y2;
+    op_pixels_abs_func pix_abs8x8_xy2;
+} DSPContext;
+
+void dsputil_init(DSPContext* p, unsigned mask);
 
 /**
  * permute block according to permuatation.
@@ -121,8 +127,12 @@ int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
  */
 void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
 
+#define emms_c()
+
 #if defined(HAVE_MMX)
 
+#undef emms_c()
+
 #define MM_MMX    0x0001 /* standard MMX */
 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
@@ -132,6 +142,8 @@ void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable,
 extern int mm_flags;
 
 int mm_support(void);
+void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
+void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
 
 static inline void emms(void)
 {
@@ -146,54 +158,44 @@ static inline void emms(void)
 
 #define __align8 __attribute__ ((aligned (8)))
 
-void dsputil_init_mmx(void);
-void dsputil_set_bit_exact_mmx(void);
+void dsputil_init_mmx(DSPContext* c, unsigned mask);
+void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
 
 #elif defined(ARCH_ARMV4L)
 
-#define emms_c()
-
 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
    line ptimizations */
 #define __align8 __attribute__ ((aligned (4)))
 
-void dsputil_init_armv4l(void);   
+void dsputil_init_armv4l(DSPContext* c, unsigned mask);
 
 #elif defined(HAVE_MLIB)
- 
-#define emms_c()
 
 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 #define __align8 __attribute__ ((aligned (8)))
 
-void dsputil_init_mlib(void);   
+void dsputil_init_mlib(DSPContext* c, unsigned mask);
 
 #elif defined(ARCH_ALPHA)
 
-#define emms_c()
 #define __align8 __attribute__ ((aligned (8)))
 
-void dsputil_init_alpha(void);
+void dsputil_init_alpha(DSPContext* c, unsigned mask);
 
 #elif defined(ARCH_POWERPC)
 
-#define emms_c()
 #define __align8 __attribute__ ((aligned (16)))
 
-void dsputil_init_ppc(void);
+void dsputil_init_ppc(DSPContext* c, unsigned mask);
 
 #elif defined(HAVE_MMI)
 
-#define emms_c()
-
 #define __align8 __attribute__ ((aligned (16)))
 
-void dsputil_init_mmi(void);   
+void dsputil_init_mmi(DSPContext* c, unsigned mask);
 
 #else
 
-#define emms_c()
-
 #define __align8
 
 #endif
@@ -263,9 +265,9 @@ typedef struct MDCTContext {
 } MDCTContext;
 
 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
-void ff_imdct_calc(MDCTContext *s, FFTSample *output, 
+void ff_imdct_calc(MDCTContext *s, FFTSample *output,
                 const FFTSample *input, FFTSample *tmp);
-void ff_mdct_calc(MDCTContext *s, FFTSample *out, 
+void ff_mdct_calc(MDCTContext *s, FFTSample *out,
                const FFTSample *input, FFTSample *tmp);
 void ff_mdct_end(MDCTContext *s);