* DSP utils
*/
-#include "dsputil.h"
+#include "bit_depth_template.c"
-#define BIT_DEPTH 8
+static inline void FUNC(copy_block2)(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
+{
+ int i;
+ for(i=0; i<h; i++)
+ {
+ AV_WN2P(dst , AV_RN2P(src ));
+ dst+=dstStride;
+ src+=srcStride;
+ }
+}
-#define pixel uint8_t
-#define pixel2 uint16_t
-#define pixel4 uint32_t
-#define dctcoef int16_t
+static inline void FUNC(copy_block4)(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
+{
+ int i;
+ for(i=0; i<h; i++)
+ {
+ AV_WN4P(dst , AV_RN4P(src ));
+ dst+=dstStride;
+ src+=srcStride;
+ }
+}
-#define FUNC(a) a
-#define FUNCC(a) a ## _c
-#define INIT_CLIP uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
-#define CLIP(a) cm[a]
-#define AV_RN2P AV_RN16
-#define AV_RN4P AV_RN32
-#define PIXEL_MAX ((1<<BIT_DEPTH)-1)
+static inline void FUNC(copy_block8)(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
+{
+ int i;
+ for(i=0; i<h; i++)
+ {
+ AV_WN4P(dst , AV_RN4P(src ));
+ AV_WN4P(dst+4*sizeof(pixel), AV_RN4P(src+4*sizeof(pixel)));
+ dst+=dstStride;
+ src+=srcStride;
+ }
+}
-#define no_rnd_avg_pixel4 no_rnd_avg32
-#define rnd_avg_pixel4 rnd_avg32
+static inline void FUNC(copy_block16)(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h)
+{
+ int i;
+ for(i=0; i<h; i++)
+ {
+ AV_WN4P(dst , AV_RN4P(src ));
+ AV_WN4P(dst+ 4*sizeof(pixel), AV_RN4P(src+ 4*sizeof(pixel)));
+ AV_WN4P(dst+ 8*sizeof(pixel), AV_RN4P(src+ 8*sizeof(pixel)));
+ AV_WN4P(dst+12*sizeof(pixel), AV_RN4P(src+12*sizeof(pixel)));
+ dst+=dstStride;
+ src+=srcStride;
+ }
+}
/* draw the edges of width 'w' of an image of size width, height */
//FIXME check that this is ok for mpeg4 interlaced
-static void FUNCC(draw_edges)(uint8_t *_buf, int _wrap, int width, int height, int w, int sides)
+static void FUNCC(draw_edges)(uint8_t *_buf, int _wrap, int width, int height, int w, int h, int sides)
{
pixel *buf = (pixel*)_buf;
int wrap = _wrap / sizeof(pixel);
buf -= w;
last_line = buf + (height - 1) * wrap;
if (sides & EDGE_TOP)
- for(i = 0; i < w; i++)
+ for(i = 0; i < h; i++)
memcpy(buf - (i + 1) * wrap, buf, (width + w + w) * sizeof(pixel)); // top
if (sides & EDGE_BOTTOM)
- for (i = 0; i < w; i++)
+ for (i = 0; i < h; i++)
memcpy(last_line + (i + 1) * wrap, last_line, (width + w + w) * sizeof(pixel)); // bottom
}
}
}
-static void FUNCC(add_pixels8)(uint8_t *restrict _pixels, DCTELEM *_block, int line_size)
-{
- int i;
- pixel *restrict pixels = (pixel *restrict)_pixels;
- dctcoef *block = (dctcoef*)_block;
- line_size /= sizeof(pixel);
-
- for(i=0;i<8;i++) {
- pixels[0] += block[0];
- pixels[1] += block[1];
- pixels[2] += block[2];
- pixels[3] += block[3];
- pixels[4] += block[4];
- pixels[5] += block[5];
- pixels[6] += block[6];
- pixels[7] += block[7];
- pixels += line_size;
- block += 8;
- }
+#define DCTELEM_FUNCS(dctcoef, suffix) \
+static void FUNCC(get_pixels ## suffix)(DCTELEM *restrict _block, \
+ const uint8_t *_pixels, \
+ int line_size) \
+{ \
+ const pixel *pixels = (const pixel *) _pixels; \
+ dctcoef *restrict block = (dctcoef *) _block; \
+ int i; \
+ \
+ /* read the pixels */ \
+ for(i=0;i<8;i++) { \
+ block[0] = pixels[0]; \
+ block[1] = pixels[1]; \
+ block[2] = pixels[2]; \
+ block[3] = pixels[3]; \
+ block[4] = pixels[4]; \
+ block[5] = pixels[5]; \
+ block[6] = pixels[6]; \
+ block[7] = pixels[7]; \
+ pixels += line_size / sizeof(pixel); \
+ block += 8; \
+ } \
+} \
+ \
+static void FUNCC(add_pixels8 ## suffix)(uint8_t *restrict _pixels, \
+ DCTELEM *_block, \
+ int line_size) \
+{ \
+ int i; \
+ pixel *restrict pixels = (pixel *restrict)_pixels; \
+ dctcoef *block = (dctcoef*)_block; \
+ line_size /= sizeof(pixel); \
+ \
+ for(i=0;i<8;i++) { \
+ pixels[0] += block[0]; \
+ pixels[1] += block[1]; \
+ pixels[2] += block[2]; \
+ pixels[3] += block[3]; \
+ pixels[4] += block[4]; \
+ pixels[5] += block[5]; \
+ pixels[6] += block[6]; \
+ pixels[7] += block[7]; \
+ pixels += line_size; \
+ block += 8; \
+ } \
+} \
+ \
+static void FUNCC(add_pixels4 ## suffix)(uint8_t *restrict _pixels, \
+ DCTELEM *_block, \
+ int line_size) \
+{ \
+ int i; \
+ pixel *restrict pixels = (pixel *restrict)_pixels; \
+ dctcoef *block = (dctcoef*)_block; \
+ line_size /= sizeof(pixel); \
+ \
+ for(i=0;i<4;i++) { \
+ pixels[0] += block[0]; \
+ pixels[1] += block[1]; \
+ pixels[2] += block[2]; \
+ pixels[3] += block[3]; \
+ pixels += line_size; \
+ block += 4; \
+ } \
+} \
+ \
+static void FUNCC(clear_block ## suffix)(DCTELEM *block) \
+{ \
+ memset(block, 0, sizeof(dctcoef)*64); \
+} \
+ \
+/** \
+ * memset(blocks, 0, sizeof(DCTELEM)*6*64) \
+ */ \
+static void FUNCC(clear_blocks ## suffix)(DCTELEM *blocks) \
+{ \
+ memset(blocks, 0, sizeof(dctcoef)*6*64); \
}
-static void FUNCC(add_pixels4)(uint8_t *restrict _pixels, DCTELEM *_block, int line_size)
-{
- int i;
- pixel *restrict pixels = (pixel *restrict)_pixels;
- dctcoef *block = (dctcoef*)_block;
- line_size /= sizeof(pixel);
-
- for(i=0;i<4;i++) {
- pixels[0] += block[0];
- pixels[1] += block[1];
- pixels[2] += block[2];
- pixels[3] += block[3];
- pixels += line_size;
- block += 4;
- }
-}
-
-#if 0
-
-#define PIXOP2(OPNAME, OP) \
-static void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- for(i=0; i<h; i++){\
- OP(*((uint64_t*)block), AV_RN64(pixels));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _no_rnd_pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- for(i=0; i<h; i++){\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+1);\
- OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _pixels_x2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- for(i=0; i<h; i++){\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+1);\
- OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _no_rnd_pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- for(i=0; i<h; i++){\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+line_size);\
- OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _pixels_y2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- for(i=0; i<h; i++){\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+line_size);\
- OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+1);\
- uint64_t l0= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL)\
- + 0x0202020202020202ULL;\
- uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- uint64_t l1,h1;\
-\
- pixels+=line_size;\
- for(i=0; i<h; i+=2){\
- uint64_t a= AV_RN64(pixels );\
- uint64_t b= AV_RN64(pixels+1);\
- l1= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL);\
- h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
- pixels+=line_size;\
- block +=line_size;\
- a= AV_RN64(pixels );\
- b= AV_RN64(pixels+1);\
- l0= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL)\
- + 0x0202020202020202ULL;\
- h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-static void OPNAME ## _no_rnd_pixels_xy2_c(uint8_t *block, const uint8_t *pixels, int line_size, int h)\
-{\
- int i;\
- const uint64_t a= AV_RN64(pixels );\
- const uint64_t b= AV_RN64(pixels+1);\
- uint64_t l0= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL)\
- + 0x0101010101010101ULL;\
- uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- uint64_t l1,h1;\
-\
- pixels+=line_size;\
- for(i=0; i<h; i+=2){\
- uint64_t a= AV_RN64(pixels );\
- uint64_t b= AV_RN64(pixels+1);\
- l1= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL);\
- h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
- pixels+=line_size;\
- block +=line_size;\
- a= AV_RN64(pixels );\
- b= AV_RN64(pixels+1);\
- l0= (a&0x0303030303030303ULL)\
- + (b&0x0303030303030303ULL)\
- + 0x0101010101010101ULL;\
- h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\
- + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\
- OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\
- pixels+=line_size;\
- block +=line_size;\
- }\
-}\
-\
-CALL_2X_PIXELS(OPNAME ## _pixels16_c , OPNAME ## _pixels_c , 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _pixels16_x2_c , OPNAME ## _pixels_x2_c , 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _pixels16_y2_c , OPNAME ## _pixels_y2_c , 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _pixels16_xy2_c, OPNAME ## _pixels_xy2_c, 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_x2_c , OPNAME ## _no_rnd_pixels_x2_c , 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_y2_c , OPNAME ## _no_rnd_pixels_y2_c , 8*sizeof(pixel))\
-CALL_2X_PIXELS(OPNAME ## _no_rnd_pixels16_xy2_c, OPNAME ## _no_rnd_pixels_xy2_c, 8*sizeof(pixel))
-
-#define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) )
-#else // 64 bit variant
+DCTELEM_FUNCS(DCTELEM, _16)
+#if BIT_DEPTH > 8
+DCTELEM_FUNCS(dctcoef, _32)
+#endif
#define PIXOP2(OPNAME, OP) \
static void FUNCC(OPNAME ## _pixels2)(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
CALL_2X_PIXELS(FUNCC(OPNAME ## _no_rnd_pixels16_xy2), FUNCC(OPNAME ## _no_rnd_pixels8_xy2), 8*sizeof(pixel))\
#define op_avg(a, b) a = rnd_avg_pixel4(a, b)
-#endif
#define op_put(a, b) a = b
PIXOP2(avg, op_avg)
#undef op2_avg
#undef op2_put
-#define put_h264_qpel8_mc00_c ff_put_pixels8x8_c
-#define avg_h264_qpel8_mc00_c ff_avg_pixels8x8_c
-#define put_h264_qpel16_mc00_c ff_put_pixels16x16_c
-#define avg_h264_qpel16_mc00_c ff_avg_pixels16x16_c
+#if BIT_DEPTH == 8
+# define put_h264_qpel8_mc00_8_c ff_put_pixels8x8_8_c
+# define avg_h264_qpel8_mc00_8_c ff_avg_pixels8x8_8_c
+# define put_h264_qpel16_mc00_8_c ff_put_pixels16x16_8_c
+# define avg_h264_qpel16_mc00_8_c ff_avg_pixels16x16_8_c
+#elif BIT_DEPTH == 9
+# define put_h264_qpel8_mc00_9_c ff_put_pixels8x8_9_c
+# define avg_h264_qpel8_mc00_9_c ff_avg_pixels8x8_9_c
+# define put_h264_qpel16_mc00_9_c ff_put_pixels16x16_9_c
+# define avg_h264_qpel16_mc00_9_c ff_avg_pixels16x16_9_c
+#elif BIT_DEPTH == 10
+# define put_h264_qpel8_mc00_10_c ff_put_pixels8x8_10_c
+# define avg_h264_qpel8_mc00_10_c ff_avg_pixels8x8_10_c
+# define put_h264_qpel16_mc00_10_c ff_put_pixels16x16_10_c
+# define avg_h264_qpel16_mc00_10_c ff_avg_pixels16x16_10_c
+#endif
void FUNCC(ff_put_pixels8x8)(uint8_t *dst, uint8_t *src, int stride) {
FUNCC(put_pixels8)(dst, src, stride, 8);
void FUNCC(ff_avg_pixels16x16)(uint8_t *dst, uint8_t *src, int stride) {
FUNCC(avg_pixels16)(dst, src, stride, 16);
}
-
-static void FUNCC(clear_block)(DCTELEM *block)
-{
- memset(block, 0, sizeof(dctcoef)*64);
-}
-
-/**
- * memset(blocks, 0, sizeof(DCTELEM)*6*64)
- */
-static void FUNCC(clear_blocks)(DCTELEM *blocks)
-{
- memset(blocks, 0, sizeof(dctcoef)*6*64);
-}
projects / ffmpeg / blobdiff