Hi,
On Tue, Apr 26, 2011 at 2:07 PM, Diego Biurrun <[email protected]> wrote:
> ---
> libswscale/ppc/yuv2rgb_altivec.c | 152
> -------------------------------------
Leaving for Luca.
> diff --git a/libswscale/rgb2rgb_template.c b/libswscale/rgb2rgb_template.c
[..]
> @@ -593,7 +593,6 @@ static inline void planar2x_c(const uint8_t *src, uint8_t
> *dst, long srcWidth,
> }
>
> // last line
> -#if 1
> dst[0]= src[0];
>
> for (x=0; x<srcWidth-1; x++) {
> @@ -601,12 +600,6 @@ static inline void planar2x_c(const uint8_t *src,
> uint8_t *dst, long srcWidth,
> dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
> }
> dst[2*srcWidth-1]= src[srcWidth-1];
> -#else
> - for (x=0; x<srcWidth; x++) {
> - dst[2*x+0]=
> - dst[2*x+1]= src[x];
> - }
> -#endif
OK.
> @@ -160,7 +160,6 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_73)[8][8]={
> { 48, 30, 43, 25, 47, 29, 42, 24, },
> };
>
> -#if 1
> DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
> {117, 62, 158, 103, 113, 58, 155, 100, },
> { 34, 199, 21, 186, 31, 196, 17, 182, },
> @@ -171,43 +170,6 @@ DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
> {138, 83, 124, 69, 148, 93, 134, 79, },
> { 7, 172, 48, 213, 3, 168, 45, 210, },
> };
> -#elif 1
> -// tries to correct a gamma of 1.5
> -DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
> -{ 0, 143, 18, 200, 2, 156, 25, 215, },
> -{ 78, 28, 125, 64, 89, 36, 138, 74, },
> -{ 10, 180, 3, 161, 16, 195, 8, 175, },
> -{109, 51, 93, 38, 121, 60, 105, 47, },
> -{ 1, 152, 23, 210, 0, 147, 20, 205, },
> -{ 85, 33, 134, 71, 81, 30, 130, 67, },
> -{ 14, 190, 6, 171, 12, 185, 5, 166, },
> -{117, 57, 101, 44, 113, 54, 97, 41, },
> -};
> -#elif 1
> -// tries to correct a gamma of 2.0
> -DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
> -{ 0, 124, 8, 193, 0, 140, 12, 213, },
> -{ 55, 14, 104, 42, 66, 19, 119, 52, },
> -{ 3, 168, 1, 145, 6, 187, 3, 162, },
> -{ 86, 31, 70, 21, 99, 39, 82, 28, },
> -{ 0, 134, 11, 206, 0, 129, 9, 200, },
> -{ 62, 17, 114, 48, 58, 16, 109, 45, },
> -{ 5, 181, 2, 157, 4, 175, 1, 151, },
> -{ 95, 36, 78, 26, 90, 34, 74, 24, },
> -};
> -#else
> -// tries to correct a gamma of 2.5
> -DECLARE_ALIGNED(8, const uint8_t, dither_8x8_220)[8][8]={
> -{ 0, 107, 3, 187, 0, 125, 6, 212, },
> -{ 39, 7, 86, 28, 49, 11, 102, 36, },
> -{ 1, 158, 0, 131, 3, 180, 1, 151, },
> -{ 68, 19, 52, 12, 81, 25, 64, 17, },
> -{ 0, 119, 5, 203, 0, 113, 4, 195, },
> -{ 45, 9, 96, 33, 42, 8, 91, 30, },
> -{ 2, 172, 1, 144, 2, 165, 0, 137, },
> -{ 77, 23, 60, 15, 72, 21, 56, 14, },
> -};
> -#endif
Looks potentially useful. I'd leave it in for now.
> diff --git a/libswscale/x86/rgb2rgb_template.c
> b/libswscale/x86/rgb2rgb_template.c
[..]
> @@ -290,7 +290,6 @@ static inline void RENAME(rgb32to16)(const uint8_t *src,
> uint8_t *dst, long src_
> uint16_t *d = (uint16_t *)dst;
> end = s + src_size;
> mm_end = end - 15;
> -#if 1 //is faster only if multiplies are reasonably fast (FIXME figure out
> on which CPUs this is faster, on Athlon it is slightly faster)
> __asm__ volatile(
> "movq %3, %%mm5 \n\t"
> "movq %4, %%mm6 \n\t"
This looks like something where we might actually be able to gain. See
it as a FIXME (it even says so), so leave it.
> @@ -1564,17 +1480,10 @@ static inline void RENAME(planar2x)(const uint8_t
> *src, uint8_t *dst, long srcWi
> "punpckhbw %%mm3, %%mm7 \n\t"
> "punpcklbw %%mm2, %%mm4 \n\t"
> "punpckhbw %%mm2, %%mm6 \n\t"
> -#if 1
> MOVNTQ" %%mm5, (%2, %%"REG_a", 2) \n\t"
> MOVNTQ" %%mm7, 8(%2, %%"REG_a", 2) \n\t"
> MOVNTQ" %%mm4, (%3, %%"REG_a", 2) \n\t"
> MOVNTQ" %%mm6, 8(%3, %%"REG_a", 2) \n\t"
> -#else
> - "movq %%mm5, (%2, %%"REG_a", 2) \n\t"
> - "movq %%mm7, 8(%2, %%"REG_a", 2) \n\t"
> - "movq %%mm4, (%3, %%"REG_a", 2) \n\t"
> - "movq %%mm6, 8(%3, %%"REG_a", 2) \n\t"
> -#endif
> "add $8, %%"REG_a" \n\t"
> "movq -1(%0, %%"REG_a"), %%mm4 \n\t"
> "movq -1(%1, %%"REG_a"), %%mm5 \n\t"
> @@ -1605,7 +1514,6 @@ static inline void RENAME(planar2x)(const uint8_t *src,
> uint8_t *dst, long srcWi
> }
>
> // last line
> -#if 1
> dst[0]= src[0];
>
> for (x=0; x<srcWidth-1; x++) {
> @@ -1613,12 +1521,6 @@ static inline void RENAME(planar2x)(const uint8_t
> *src, uint8_t *dst, long srcWi
> dst[2*x+2]= ( src[x] + 3*src[x+1])>>2;
> }
> dst[2*srcWidth-1]= src[srcWidth-1];
> -#else
> - for (x=0; x<srcWidth; x++) {
> - dst[2*x+0]=
> - dst[2*x+1]= src[x];
> - }
> -#endif
OK.
> diff --git a/libswscale/yuv2rgb.c b/libswscale/yuv2rgb.c
> index 8ea41af..2b23930 100644
> --- a/libswscale/yuv2rgb.c
> +++ b/libswscale/yuv2rgb.c
> @@ -330,28 +330,6 @@ YUV2RGBFUNC(yuv2rgb_c_16, uint16_t, 0)
> PUTRGB(dst_1,py_1,3);
> CLOSEYUV2RGBFUNC(8)
>
> -#if 0 // Currently unused
> -// This is exactly the same code as yuv2rgb_c_32 except for the types of
> -// r, g, b, dst_1, dst_2
> -YUV2RGBFUNC(yuv2rgb_c_8, uint8_t, 0)
> - LOADCHROMA(0);
> - PUTRGB(dst_1,py_1,0);
> - PUTRGB(dst_2,py_2,0);
> -
> - LOADCHROMA(1);
> - PUTRGB(dst_2,py_2,1);
> - PUTRGB(dst_1,py_1,1);
> -
> - LOADCHROMA(2);
> - PUTRGB(dst_1,py_1,2);
> - PUTRGB(dst_2,py_2,2);
> -
> - LOADCHROMA(3);
> - PUTRGB(dst_2,py_2,3);
> - PUTRGB(dst_1,py_1,3);
> -CLOSEYUV2RGBFUNC(8)
> -#endif
> -
> // r, g, b, dst_1, dst_2
> YUV2RGBFUNC(yuv2rgb_c_12_ordered_dither, uint16_t, 0)
> const uint8_t *d16 = dither_4x4_16[y&3];
> @@ -405,36 +383,6 @@ YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0)
> PUTRGB8(dst_1,py_1,3,6);
> CLOSEYUV2RGBFUNC(8)
>
> -#if 0 // Currently unused
> -// This is exactly the same code as yuv2rgb_c_32 except for the types of
> -// r, g, b, dst_1, dst_2
> -YUV2RGBFUNC(yuv2rgb_c_4, uint8_t, 0)
> - int acc;
> -#define PUTRGB4(dst,src,i) \
> - Y = src[2*i]; \
> - acc = r[Y] + g[Y] + b[Y]; \
> - Y = src[2*i+1]; \
> - acc |= (r[Y] + g[Y] + b[Y])<<4; \
> - dst[i] = acc;
> -
> - LOADCHROMA(0);
> - PUTRGB4(dst_1,py_1,0);
> - PUTRGB4(dst_2,py_2,0);
> -
> - LOADCHROMA(1);
> - PUTRGB4(dst_2,py_2,1);
> - PUTRGB4(dst_1,py_1,1);
> -
> - LOADCHROMA(2);
> - PUTRGB4(dst_1,py_1,2);
> - PUTRGB4(dst_2,py_2,2);
> -
> - LOADCHROMA(3);
> - PUTRGB4(dst_2,py_2,3);
> - PUTRGB4(dst_1,py_1,3);
> -CLOSEYUV2RGBFUNC(4)
> -#endif
> -
> YUV2RGBFUNC(yuv2rgb_c_4_ordered_dither, uint8_t, 0)
> const uint8_t *d64 = dither_8x8_73[y&7];
> const uint8_t *d128 = dither_8x8_220[y&7];
> @@ -464,28 +412,6 @@ YUV2RGBFUNC(yuv2rgb_c_4_ordered_dither, uint8_t, 0)
> PUTRGB4D(dst_1,py_1,3,6);
> CLOSEYUV2RGBFUNC(4)
>
> -#if 0 // Currently unused
> -// This is exactly the same code as yuv2rgb_c_32 except for the types of
> -// r, g, b, dst_1, dst_2
> -YUV2RGBFUNC(yuv2rgb_c_4b, uint8_t, 0)
> - LOADCHROMA(0);
> - PUTRGB(dst_1,py_1,0);
> - PUTRGB(dst_2,py_2,0);
> -
> - LOADCHROMA(1);
> - PUTRGB(dst_2,py_2,1);
> - PUTRGB(dst_1,py_1,1);
> -
> - LOADCHROMA(2);
> - PUTRGB(dst_1,py_1,2);
> - PUTRGB(dst_2,py_2,2);
> -
> - LOADCHROMA(3);
> - PUTRGB(dst_2,py_2,3);
> - PUTRGB(dst_1,py_1,3);
> -CLOSEYUV2RGBFUNC(8)
> -#endif
> -
> YUV2RGBFUNC(yuv2rgb_c_4b_ordered_dither, uint8_t, 0)
> const uint8_t *d64 = dither_8x8_73[y&7];
> const uint8_t *d128 = dither_8x8_220[y&7];
OK.
Ronald
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
