Mercurial > libavcodec.hg
changeset 9114:3bf99f635933 libavcodec
Remove old scaler.
| author | michael |
|---|---|
| date | Tue, 03 Mar 2009 22:51:30 +0000 |
| parents | 21f500327e5c |
| children | 1b68b8e9b4c9 |
| files | Makefile imgconvert.c imgconvert_template.c imgresample.c ppc/imgresample_altivec.c ppc/imgresample_altivec.h |
| diffstat | 6 files changed, 0 insertions(+), 3371 deletions(-) [+] |
line wrap: on
line diff
--- a/Makefile Tue Mar 03 15:48:52 2009 +0000 +++ b/Makefile Tue Mar 03 22:51:30 2009 +0000 @@ -29,7 +29,6 @@ OBJS-$(CONFIG_FFT) += fft.o OBJS-$(CONFIG_GOLOMB) += golomb.o OBJS-$(CONFIG_MDCT) += mdct.o -OBJS-$(CONFIG_OLDSCALER) += imgresample.o OBJS-$(CONFIG_RDFT) += rdft.o # decoders/encoders @@ -494,7 +493,6 @@ OBJS-$(ARCH_PPC) += ppc/dsputil_ppc.o \ ALTIVEC-OBJS-$(CONFIG_H264_DECODER) += ppc/h264_altivec.o -ALTIVEC-OBJS-$(CONFIG_OLDSCALER) += ppc/imgresample_altivec.o ALTIVEC-OBJS-$(CONFIG_SNOW_DECODER) += ppc/snow_altivec.o ALTIVEC-OBJS-$(CONFIG_VC1_DECODER) += ppc/vc1dsp_altivec.o ALTIVEC-OBJS-$(CONFIG_WMV3_DECODER) += ppc/vc1dsp_altivec.o @@ -525,7 +523,6 @@ TESTS = $(addsuffix -test$(EXESUF), cabac dct eval fft h264 rangecoder snow) -TESTS-$(CONFIG_OLDSCALER) += imgresample-test$(EXESUF) TESTS-$(ARCH_X86) += x86/cpuid-test$(EXESUF) motion-test$(EXESUF) CLEANFILES = apiexample$(EXESUF)
--- a/imgconvert.c Tue Mar 03 15:48:52 2009 +0000 +++ b/imgconvert.c Tue Mar 03 22:51:30 2009 +0000 @@ -1033,357 +1033,6 @@ } } -/* XXX: totally non optimized */ - -static void yuyv422_to_yuv420p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *p, *p1; - uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = src->data[0]; - lum1 = dst->data[0]; - cb1 = dst->data[1]; - cr1 = dst->data[2]; - - for(;height >= 1; height -= 2) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[0]; - cb[0] = p[1]; - lum[1] = p[2]; - cr[0] = p[3]; - p += 4; - lum += 2; - cb++; - cr++; - } - if (w) { - lum[0] = p[0]; - cb[0] = p[1]; - cr[0] = p[3]; - cb++; - cr++; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - if (height>1) { - p = p1; - lum = lum1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[0]; - lum[1] = p[2]; - p += 4; - lum += 2; - } - if (w) { - lum[0] = p[0]; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - } - cb1 += dst->linesize[1]; - cr1 += dst->linesize[2]; - } -} - -static void uyvy422_to_yuv420p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *p, *p1; - uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = src->data[0]; - - lum1 = dst->data[0]; - cb1 = dst->data[1]; - cr1 = dst->data[2]; - - for(;height >= 1; height -= 2) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[1]; - cb[0] = p[0]; - lum[1] = p[3]; - cr[0] = p[2]; - p += 4; - lum += 2; - cb++; - cr++; - } - if (w) { - lum[0] = p[1]; - cb[0] = p[0]; - cr[0] = p[2]; - cb++; - cr++; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - if (height>1) { - p = p1; - lum = lum1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[1]; - lum[1] = p[3]; - p += 4; - lum += 2; - } - if (w) { - lum[0] = p[1]; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - } - cb1 += dst->linesize[1]; - cr1 += dst->linesize[2]; - } -} - - -static void uyvy422_to_yuv422p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *p, *p1; - uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = src->data[0]; - lum1 = dst->data[0]; - cb1 = dst->data[1]; - cr1 = dst->data[2]; - for(;height > 0; height--) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[1]; - cb[0] = p[0]; - lum[1] = p[3]; - cr[0] = p[2]; - p += 4; - lum += 2; - cb++; - cr++; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - cb1 += dst->linesize[1]; - cr1 += dst->linesize[2]; - } -} - - -static void yuyv422_to_yuv422p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *p, *p1; - uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = src->data[0]; - lum1 = dst->data[0]; - cb1 = dst->data[1]; - cr1 = dst->data[2]; - for(;height > 0; height--) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - lum[0] = p[0]; - cb[0] = p[1]; - lum[1] = p[2]; - cr[0] = p[3]; - p += 4; - lum += 2; - cb++; - cr++; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - cb1 += dst->linesize[1]; - cr1 += dst->linesize[2]; - } -} - -static void yuv422p_to_yuyv422(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - uint8_t *p, *p1; - const uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = dst->data[0]; - lum1 = src->data[0]; - cb1 = src->data[1]; - cr1 = src->data[2]; - for(;height > 0; height--) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - p[0] = lum[0]; - p[1] = cb[0]; - p[2] = lum[1]; - p[3] = cr[0]; - p += 4; - lum += 2; - cb++; - cr++; - } - p1 += dst->linesize[0]; - lum1 += src->linesize[0]; - cb1 += src->linesize[1]; - cr1 += src->linesize[2]; - } -} - -static void yuv422p_to_uyvy422(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - uint8_t *p, *p1; - const uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = dst->data[0]; - lum1 = src->data[0]; - cb1 = src->data[1]; - cr1 = src->data[2]; - for(;height > 0; height--) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 2; w -= 2) { - p[1] = lum[0]; - p[0] = cb[0]; - p[3] = lum[1]; - p[2] = cr[0]; - p += 4; - lum += 2; - cb++; - cr++; - } - p1 += dst->linesize[0]; - lum1 += src->linesize[0]; - cb1 += src->linesize[1]; - cr1 += src->linesize[2]; - } -} - -static void uyyvyy411_to_yuv411p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *p, *p1; - uint8_t *lum, *cr, *cb, *lum1, *cr1, *cb1; - int w; - - p1 = src->data[0]; - lum1 = dst->data[0]; - cb1 = dst->data[1]; - cr1 = dst->data[2]; - for(;height > 0; height--) { - p = p1; - lum = lum1; - cb = cb1; - cr = cr1; - for(w = width; w >= 4; w -= 4) { - cb[0] = p[0]; - lum[0] = p[1]; - lum[1] = p[2]; - cr[0] = p[3]; - lum[2] = p[4]; - lum[3] = p[5]; - p += 6; - lum += 4; - cb++; - cr++; - } - p1 += src->linesize[0]; - lum1 += dst->linesize[0]; - cb1 += dst->linesize[1]; - cr1 += dst->linesize[2]; - } -} - - -static void yuv420p_to_yuyv422(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int w, h; - uint8_t *line1, *line2, *linesrc = dst->data[0]; - uint8_t *lum1, *lum2, *lumsrc = src->data[0]; - uint8_t *cb1, *cb2 = src->data[1]; - uint8_t *cr1, *cr2 = src->data[2]; - - for(h = height / 2; h--;) { - line1 = linesrc; - line2 = linesrc + dst->linesize[0]; - - lum1 = lumsrc; - lum2 = lumsrc + src->linesize[0]; - - cb1 = cb2; - cr1 = cr2; - - for(w = width / 2; w--;) { - *line1++ = *lum1++; *line2++ = *lum2++; - *line1++ = *line2++ = *cb1++; - *line1++ = *lum1++; *line2++ = *lum2++; - *line1++ = *line2++ = *cr1++; - } - - linesrc += dst->linesize[0] * 2; - lumsrc += src->linesize[0] * 2; - cb2 += src->linesize[1]; - cr2 += src->linesize[2]; - } -} - -static void yuv420p_to_uyvy422(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int w, h; - uint8_t *line1, *line2, *linesrc = dst->data[0]; - uint8_t *lum1, *lum2, *lumsrc = src->data[0]; - uint8_t *cb1, *cb2 = src->data[1]; - uint8_t *cr1, *cr2 = src->data[2]; - - for(h = height / 2; h--;) { - line1 = linesrc; - line2 = linesrc + dst->linesize[0]; - - lum1 = lumsrc; - lum2 = lumsrc + src->linesize[0]; - - cb1 = cb2; - cr1 = cr2; - - for(w = width / 2; w--;) { - *line1++ = *line2++ = *cb1++; - *line1++ = *lum1++; *line2++ = *lum2++; - *line1++ = *line2++ = *cr1++; - *line1++ = *lum1++; *line2++ = *lum2++; - } - - linesrc += dst->linesize[0] * 2; - lumsrc += src->linesize[0] * 2; - cb2 += src->linesize[1]; - cr2 += src->linesize[2]; - } -} - /* 2x2 -> 1x1 */ void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, @@ -1470,622 +1119,6 @@ } } -/* XXX: add jpeg quantize code */ - -#define TRANSP_INDEX (6*6*6) - -/* this is maybe slow, but allows for extensions */ -static inline unsigned char gif_clut_index(uint8_t r, uint8_t g, uint8_t b) -{ - return (((r) / 47) % 6) * 6 * 6 + (((g) / 47) % 6) * 6 + (((b) / 47) % 6); -} - -static void build_rgb_palette(uint8_t *palette, int has_alpha) -{ - uint32_t *pal; - static const uint8_t pal_value[6] = { 0x00, 0x33, 0x66, 0x99, 0xcc, 0xff }; - int i, r, g, b; - - pal = (uint32_t *)palette; - i = 0; - for(r = 0; r < 6; r++) { - for(g = 0; g < 6; g++) { - for(b = 0; b < 6; b++) { - pal[i++] = (0xff << 24) | (pal_value[r] << 16) | - (pal_value[g] << 8) | pal_value[b]; - } - } - } - if (has_alpha) - pal[i++] = 0; - while (i < 256) - pal[i++] = 0xff000000; -} - -/* copy bit n to bits 0 ... n - 1 */ -static inline unsigned int bitcopy_n(unsigned int a, int n) -{ - int mask; - mask = (1 << n) - 1; - return (a & (0xff & ~mask)) | ((-((a >> n) & 1)) & mask); -} - -/* rgb555 handling */ - -#define RGB_NAME rgb555 - -#define RGB_IN(r, g, b, s)\ -{\ - unsigned int v = ((const uint16_t *)(s))[0];\ - r = bitcopy_n(v >> (10 - 3), 3);\ - g = bitcopy_n(v >> (5 - 3), 3);\ - b = bitcopy_n(v << 3, 3);\ -} - - -#define RGB_OUT(d, r, g, b)\ -{\ - ((uint16_t *)(d))[0] = ((r >> 3) << 10) | ((g >> 3) << 5) | (b >> 3);\ -} - -#define BPP 2 - -#include "imgconvert_template.c" - -/* rgb565 handling */ - -#define RGB_NAME rgb565 - -#define RGB_IN(r, g, b, s)\ -{\ - unsigned int v = ((const uint16_t *)(s))[0];\ - r = bitcopy_n(v >> (11 - 3), 3);\ - g = bitcopy_n(v >> (5 - 2), 2);\ - b = bitcopy_n(v << 3, 3);\ -} - -#define RGB_OUT(d, r, g, b)\ -{\ - ((uint16_t *)(d))[0] = ((r >> 3) << 11) | ((g >> 2) << 5) | (b >> 3);\ -} - -#define BPP 2 - -#include "imgconvert_template.c" - -/* bgr24 handling */ - -#define RGB_NAME bgr24 - -#define RGB_IN(r, g, b, s)\ -{\ - b = (s)[0];\ - g = (s)[1];\ - r = (s)[2];\ -} - -#define RGB_OUT(d, r, g, b)\ -{\ - (d)[0] = b;\ - (d)[1] = g;\ - (d)[2] = r;\ -} - -#define BPP 3 - -#include "imgconvert_template.c" - -#undef RGB_IN -#undef RGB_OUT -#undef BPP - -/* rgb24 handling */ - -#define RGB_NAME rgb24 -#define FMT_RGB24 - -#define RGB_IN(r, g, b, s)\ -{\ - r = (s)[0];\ - g = (s)[1];\ - b = (s)[2];\ -} - -#define RGB_OUT(d, r, g, b)\ -{\ - (d)[0] = r;\ - (d)[1] = g;\ - (d)[2] = b;\ -} - -#define BPP 3 - -#include "imgconvert_template.c" - -/* rgb32 handling */ - -#define RGB_NAME rgb32 -#define FMT_RGB32 - -#define RGB_IN(r, g, b, s)\ -{\ - unsigned int v = ((const uint32_t *)(s))[0];\ - r = (v >> 16) & 0xff;\ - g = (v >> 8) & 0xff;\ - b = v & 0xff;\ -} - -#define RGBA_IN(r, g, b, a, s)\ -{\ - unsigned int v = ((const uint32_t *)(s))[0];\ - a = (v >> 24) & 0xff;\ - r = (v >> 16) & 0xff;\ - g = (v >> 8) & 0xff;\ - b = v & 0xff;\ -} - -#define RGBA_OUT(d, r, g, b, a)\ -{\ - ((uint32_t *)(d))[0] = (a << 24) | (r << 16) | (g << 8) | b;\ -} - -#define BPP 4 - -#include "imgconvert_template.c" - -static void mono_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height, int xor_mask) -{ - const unsigned char *p; - unsigned char *q; - int v, dst_wrap, src_wrap; - int y, w; - - p = src->data[0]; - src_wrap = src->linesize[0] - ((width + 7) >> 3); - - q = dst->data[0]; - dst_wrap = dst->linesize[0] - width; - for(y=0;y<height;y++) { - w = width; - while (w >= 8) { - v = *p++ ^ xor_mask; - q[0] = -(v >> 7); - q[1] = -((v >> 6) & 1); - q[2] = -((v >> 5) & 1); - q[3] = -((v >> 4) & 1); - q[4] = -((v >> 3) & 1); - q[5] = -((v >> 2) & 1); - q[6] = -((v >> 1) & 1); - q[7] = -((v >> 0) & 1); - w -= 8; - q += 8; - } - if (w > 0) { - v = *p++ ^ xor_mask; - do { - q[0] = -((v >> 7) & 1); - q++; - v <<= 1; - } while (--w); - } - p += src_wrap; - q += dst_wrap; - } -} - -static void monowhite_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - mono_to_gray(dst, src, width, height, 0xff); -} - -static void monoblack_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - mono_to_gray(dst, src, width, height, 0x00); -} - -static void gray_to_mono(AVPicture *dst, const AVPicture *src, - int width, int height, int xor_mask) -{ - int n; - const uint8_t *s; - uint8_t *d; - int j, b, v, n1, src_wrap, dst_wrap, y; - - s = src->data[0]; - src_wrap = src->linesize[0] - width; - - d = dst->data[0]; - dst_wrap = dst->linesize[0] - ((width + 7) >> 3); - - for(y=0;y<height;y++) { - n = width; - while (n >= 8) { - v = 0; - for(j=0;j<8;j++) { - b = s[0]; - s++; - v = (v << 1) | (b >> 7); - } - d[0] = v ^ xor_mask; - d++; - n -= 8; - } - if (n > 0) { - n1 = n; - v = 0; - while (n > 0) { - b = s[0]; - s++; - v = (v << 1) | (b >> 7); - n--; - } - d[0] = (v << (8 - (n1 & 7))) ^ xor_mask; - d++; - } - s += src_wrap; - d += dst_wrap; - } -} - -static void gray_to_monowhite(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - gray_to_mono(dst, src, width, height, 0xff); -} - -static void gray_to_monoblack(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - gray_to_mono(dst, src, width, height, 0x00); -} - -static void gray_to_gray16(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int x, y, src_wrap, dst_wrap; - uint8_t *s, *d; - s = src->data[0]; - src_wrap = src->linesize[0] - width; - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width * 2; - for(y=0; y<height; y++){ - for(x=0; x<width; x++){ - *d++ = *s; - *d++ = *s++; - } - s += src_wrap; - d += dst_wrap; - } -} - -static void gray16_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int x, y, src_wrap, dst_wrap; - uint8_t *s, *d; - s = src->data[0]; - src_wrap = src->linesize[0] - width * 2; - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width; - for(y=0; y<height; y++){ - for(x=0; x<width; x++){ - *d++ = *s; - s += 2; - } - s += src_wrap; - d += dst_wrap; - } -} - -static void gray16be_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - gray16_to_gray(dst, src, width, height); -} - -static void gray16le_to_gray(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - AVPicture tmpsrc = *src; - tmpsrc.data[0]++; - gray16_to_gray(dst, &tmpsrc, width, height); -} - -static void gray16_to_gray16(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int x, y, src_wrap, dst_wrap; - uint16_t *s, *d; - s = (uint16_t*)src->data[0]; - src_wrap = (src->linesize[0] - width * 2)/2; - d = (uint16_t*)dst->data[0]; - dst_wrap = (dst->linesize[0] - width * 2)/2; - for(y=0; y<height; y++){ - for(x=0; x<width; x++){ - *d++ = bswap_16(*s++); - } - s += src_wrap; - d += dst_wrap; - } -} - - -typedef struct ConvertEntry { - void (*convert)(AVPicture *dst, - const AVPicture *src, int width, int height); -} ConvertEntry; - -/* Add each new conversion function in this table. In order to be able - to convert from any format to any format, the following constraints - must be satisfied: - - - all FF_COLOR_RGB formats must convert to and from PIX_FMT_RGB24 - - - all FF_COLOR_GRAY formats must convert to and from PIX_FMT_GRAY8 - - - all FF_COLOR_RGB formats with alpha must convert to and from PIX_FMT_RGB32 - - - PIX_FMT_YUV444P and PIX_FMT_YUVJ444P must convert to and from - PIX_FMT_RGB24. - - - PIX_FMT_422 must convert to and from PIX_FMT_422P. - - The other conversion functions are just optimizations for common cases. -*/ -static const ConvertEntry convert_table[PIX_FMT_NB][PIX_FMT_NB] = { - [PIX_FMT_YUV420P] = { - [PIX_FMT_YUYV422] = { - .convert = yuv420p_to_yuyv422, - }, - [PIX_FMT_RGB555] = { - .convert = yuv420p_to_rgb555 - }, - [PIX_FMT_RGB565] = { - .convert = yuv420p_to_rgb565 - }, - [PIX_FMT_BGR24] = { - .convert = yuv420p_to_bgr24 - }, - [PIX_FMT_RGB24] = { - .convert = yuv420p_to_rgb24 - }, - [PIX_FMT_RGB32] = { - .convert = yuv420p_to_rgb32 - }, - [PIX_FMT_UYVY422] = { - .convert = yuv420p_to_uyvy422, - }, - }, - [PIX_FMT_YUV422P] = { - [PIX_FMT_YUYV422] = { - .convert = yuv422p_to_yuyv422, - }, - [PIX_FMT_UYVY422] = { - .convert = yuv422p_to_uyvy422, - }, - }, - [PIX_FMT_YUV444P] = { - [PIX_FMT_RGB24] = { - .convert = yuv444p_to_rgb24 - }, - }, - [PIX_FMT_YUVJ420P] = { - [PIX_FMT_RGB555] = { - .convert = yuvj420p_to_rgb555 - }, - [PIX_FMT_RGB565] = { - .convert = yuvj420p_to_rgb565 - }, - [PIX_FMT_BGR24] = { - .convert = yuvj420p_to_bgr24 - }, - [PIX_FMT_RGB24] = { - .convert = yuvj420p_to_rgb24 - }, - [PIX_FMT_RGB32] = { - .convert = yuvj420p_to_rgb32 - }, - }, - [PIX_FMT_YUVJ444P] = { - [PIX_FMT_RGB24] = { - .convert = yuvj444p_to_rgb24 - }, - }, - [PIX_FMT_YUYV422] = { - [PIX_FMT_YUV420P] = { - .convert = yuyv422_to_yuv420p, - }, - [PIX_FMT_YUV422P] = { - .convert = yuyv422_to_yuv422p, - }, - }, - [PIX_FMT_UYVY422] = { - [PIX_FMT_YUV420P] = { - .convert = uyvy422_to_yuv420p, - }, - [PIX_FMT_YUV422P] = { - .convert = uyvy422_to_yuv422p, - }, - }, - [PIX_FMT_RGB24] = { - [PIX_FMT_YUV420P] = { - .convert = rgb24_to_yuv420p - }, - [PIX_FMT_RGB565] = { - .convert = rgb24_to_rgb565 - }, - [PIX_FMT_RGB555] = { - .convert = rgb24_to_rgb555 - }, - [PIX_FMT_RGB32] = { - .convert = rgb24_to_rgb32 - }, - [PIX_FMT_BGR24] = { - .convert = rgb24_to_bgr24 - }, - [PIX_FMT_GRAY8] = { - .convert = rgb24_to_gray - }, - [PIX_FMT_PAL8] = { - .convert = rgb24_to_pal8 - }, - [PIX_FMT_YUV444P] = { - .convert = rgb24_to_yuv444p - }, - [PIX_FMT_YUVJ420P] = { - .convert = rgb24_to_yuvj420p - }, - [PIX_FMT_YUVJ444P] = { - .convert = rgb24_to_yuvj444p - }, - }, - [PIX_FMT_RGB32] = { - [PIX_FMT_RGB24] = { - .convert = rgb32_to_rgb24 - }, - [PIX_FMT_BGR24] = { - .convert = rgb32_to_bgr24 - }, - [PIX_FMT_RGB565] = { - .convert = rgb32_to_rgb565 - }, - [PIX_FMT_RGB555] = { - .convert = rgb32_to_rgb555 - }, - [PIX_FMT_PAL8] = { - .convert = rgb32_to_pal8 - }, - [PIX_FMT_YUV420P] = { - .convert = rgb32_to_yuv420p - }, - [PIX_FMT_GRAY8] = { - .convert = rgb32_to_gray - }, - }, - [PIX_FMT_BGR24] = { - [PIX_FMT_RGB32] = { - .convert = bgr24_to_rgb32 - }, - [PIX_FMT_RGB24] = { - .convert = bgr24_to_rgb24 - }, - [PIX_FMT_YUV420P] = { - .convert = bgr24_to_yuv420p - }, - [PIX_FMT_GRAY8] = { - .convert = bgr24_to_gray - }, - }, - [PIX_FMT_RGB555] = { - [PIX_FMT_RGB24] = { - .convert = rgb555_to_rgb24 - }, - [PIX_FMT_RGB32] = { - .convert = rgb555_to_rgb32 - }, - [PIX_FMT_YUV420P] = { - .convert = rgb555_to_yuv420p - }, - [PIX_FMT_GRAY8] = { - .convert = rgb555_to_gray - }, - }, - [PIX_FMT_RGB565] = { - [PIX_FMT_RGB32] = { - .convert = rgb565_to_rgb32 - }, - [PIX_FMT_RGB24] = { - .convert = rgb565_to_rgb24 - }, - [PIX_FMT_YUV420P] = { - .convert = rgb565_to_yuv420p - }, - [PIX_FMT_GRAY8] = { - .convert = rgb565_to_gray - }, - }, - [PIX_FMT_GRAY16BE] = { - [PIX_FMT_GRAY8] = { - .convert = gray16be_to_gray - }, - [PIX_FMT_GRAY16LE] = { - .convert = gray16_to_gray16 - }, - }, - [PIX_FMT_GRAY16LE] = { - [PIX_FMT_GRAY8] = { - .convert = gray16le_to_gray - }, - [PIX_FMT_GRAY16BE] = { - .convert = gray16_to_gray16 - }, - }, - [PIX_FMT_GRAY8] = { - [PIX_FMT_RGB555] = { - .convert = gray_to_rgb555 - }, - [PIX_FMT_RGB565] = { - .convert = gray_to_rgb565 - }, - [PIX_FMT_RGB24] = { - .convert = gray_to_rgb24 - }, - [PIX_FMT_BGR24] = { - .convert = gray_to_bgr24 - }, - [PIX_FMT_RGB32] = { - .convert = gray_to_rgb32 - }, - [PIX_FMT_MONOWHITE] = { - .convert = gray_to_monowhite - }, - [PIX_FMT_MONOBLACK] = { - .convert = gray_to_monoblack - }, - [PIX_FMT_GRAY16LE] = { - .convert = gray_to_gray16 - }, - [PIX_FMT_GRAY16BE] = { - .convert = gray_to_gray16 - }, - }, - [PIX_FMT_MONOWHITE] = { - [PIX_FMT_GRAY8] = { - .convert = monowhite_to_gray - }, - }, - [PIX_FMT_MONOBLACK] = { - [PIX_FMT_GRAY8] = { - .convert = monoblack_to_gray - }, - }, - [PIX_FMT_PAL8] = { - [PIX_FMT_RGB555] = { - .convert = pal8_to_rgb555 - }, - [PIX_FMT_RGB565] = { - .convert = pal8_to_rgb565 - }, - [PIX_FMT_BGR24] = { - .convert = pal8_to_bgr24 - }, - [PIX_FMT_RGB24] = { - .convert = pal8_to_rgb24 - }, - [PIX_FMT_RGB32] = { - .convert = pal8_to_rgb32 - }, - }, - [PIX_FMT_UYYVYY411] = { - [PIX_FMT_YUV411P] = { - .convert = uyyvyy411_to_yuv411p, - }, - }, - -}; int avpicture_alloc(AVPicture *picture, int pix_fmt, int width, int height) @@ -2204,528 +1237,6 @@ return 0; } -#if !CONFIG_SWSCALE -static uint8_t y_ccir_to_jpeg[256]; -static uint8_t y_jpeg_to_ccir[256]; -static uint8_t c_ccir_to_jpeg[256]; -static uint8_t c_jpeg_to_ccir[256]; - -/* init various conversion tables */ -static av_cold void img_convert_init(void) -{ - int i; - uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; - - for(i = 0;i < 256; i++) { - y_ccir_to_jpeg[i] = Y_CCIR_TO_JPEG(i); - y_jpeg_to_ccir[i] = Y_JPEG_TO_CCIR(i); - c_ccir_to_jpeg[i] = C_CCIR_TO_JPEG(i); - c_jpeg_to_ccir[i] = C_JPEG_TO_CCIR(i); - } -} - -/* apply to each pixel the given table */ -static void img_apply_table(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height, const uint8_t *table1) -{ - int n; - const uint8_t *s; - uint8_t *d; - const uint8_t *table; - - table = table1; - for(;height > 0; height--) { - s = src; - d = dst; - n = width; - while (n >= 4) { - d[0] = table[s[0]]; - d[1] = table[s[1]]; - d[2] = table[s[2]]; - d[3] = table[s[3]]; - d += 4; - s += 4; - n -= 4; - } - while (n > 0) { - d[0] = table[s[0]]; - d++; - s++; - n--; - } - dst += dst_wrap; - src += src_wrap; - } -} - -/* XXX: use generic filter ? */ -/* XXX: in most cases, the sampling position is incorrect */ - -/* 4x1 -> 1x1 */ -static void shrink41(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - int w; - const uint8_t *s; - uint8_t *d; - - for(;height > 0; height--) { - s = src; - d = dst; - for(w = width;w > 0; w--) { - d[0] = (s[0] + s[1] + s[2] + s[3] + 2) >> 2; - s += 4; - d++; - } - src += src_wrap; - dst += dst_wrap; - } -} - -/* 2x1 -> 1x1 */ -static void shrink21(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - int w; - const uint8_t *s; - uint8_t *d; - - for(;height > 0; height--) { - s = src; - d = dst; - for(w = width;w > 0; w--) { - d[0] = (s[0] + s[1]) >> 1; - s += 2; - d++; - } - src += src_wrap; - dst += dst_wrap; - } -} - -/* 1x2 -> 1x1 */ -static void shrink12(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - int w; - uint8_t *d; - const uint8_t *s1, *s2; - - for(;height > 0; height--) { - s1 = src; - s2 = s1 + src_wrap; - d = dst; - for(w = width;w >= 4; w-=4) { - d[0] = (s1[0] + s2[0]) >> 1; - d[1] = (s1[1] + s2[1]) >> 1; - d[2] = (s1[2] + s2[2]) >> 1; - d[3] = (s1[3] + s2[3]) >> 1; - s1 += 4; - s2 += 4; - d += 4; - } - for(;w > 0; w--) { - d[0] = (s1[0] + s2[0]) >> 1; - s1++; - s2++; - d++; - } - src += 2 * src_wrap; - dst += dst_wrap; - } -} - -static void grow21_line(uint8_t *dst, const uint8_t *src, - int width) -{ - int w; - const uint8_t *s1; - uint8_t *d; - - s1 = src; - d = dst; - for(w = width;w >= 4; w-=4) { - d[1] = d[0] = s1[0]; - d[3] = d[2] = s1[1]; - s1 += 2; - d += 4; - } - for(;w >= 2; w -= 2) { - d[1] = d[0] = s1[0]; - s1 ++; - d += 2; - } - /* only needed if width is not a multiple of two */ - /* XXX: veryfy that */ - if (w) { - d[0] = s1[0]; - } -} - -static void grow41_line(uint8_t *dst, const uint8_t *src, - int width) -{ - int w, v; - const uint8_t *s1; - uint8_t *d; - - s1 = src; - d = dst; - for(w = width;w >= 4; w-=4) { - v = s1[0]; - d[0] = v; - d[1] = v; - d[2] = v; - d[3] = v; - s1 ++; - d += 4; - } -} - -/* 1x1 -> 2x1 */ -static void grow21(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - for(;height > 0; height--) { - grow21_line(dst, src, width); - src += src_wrap; - dst += dst_wrap; - } -} - -/* 1x1 -> 1x2 */ -static void grow12(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - for(;height > 0; height-=2) { - memcpy(dst, src, width); - dst += dst_wrap; - memcpy(dst, src, width); - dst += dst_wrap; - src += src_wrap; - } -} - -/* 1x1 -> 2x2 */ -static void grow22(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - for(;height > 0; height--) { - grow21_line(dst, src, width); - if (height%2) - src += src_wrap; - dst += dst_wrap; - } -} - -/* 1x1 -> 4x1 */ -static void grow41(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - for(;height > 0; height--) { - grow41_line(dst, src, width); - src += src_wrap; - dst += dst_wrap; - } -} - -/* 1x1 -> 4x4 */ -static void grow44(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - for(;height > 0; height--) { - grow41_line(dst, src, width); - if ((height & 3) == 1) - src += src_wrap; - dst += dst_wrap; - } -} - -/* 1x2 -> 2x1 */ -static void conv411(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height) -{ - int w, c; - const uint8_t *s1, *s2; - uint8_t *d; - - width>>=1; - - for(;height > 0; height--) { - s1 = src; - s2 = src + src_wrap; - d = dst; - for(w = width;w > 0; w--) { - c = (s1[0] + s2[0]) >> 1; - d[0] = c; - d[1] = c; - s1++; - s2++; - d += 2; - } - src += src_wrap * 2; - dst += dst_wrap; - } -} - -/* XXX: always use linesize. Return -1 if not supported */ -int img_convert(AVPicture *dst, int dst_pix_fmt, - const AVPicture *src, int src_pix_fmt, - int src_width, int src_height) -{ - static int initialized; - int i, ret, dst_width, dst_height, int_pix_fmt; - const PixFmtInfo *src_pix, *dst_pix; - const ConvertEntry *ce; - AVPicture tmp1, *tmp = &tmp1; - - if (src_pix_fmt < 0 || src_pix_fmt >= PIX_FMT_NB || - dst_pix_fmt < 0 || dst_pix_fmt >= PIX_FMT_NB) - return -1; - if (src_width <= 0 || src_height <= 0) - return 0; - - if (!initialized) { - initialized = 1; - img_convert_init(); - } - - dst_width = src_width; - dst_height = src_height; - - dst_pix = &pix_fmt_info[dst_pix_fmt]; - src_pix = &pix_fmt_info[src_pix_fmt]; - if (src_pix_fmt == dst_pix_fmt) { - /* no conversion needed: just copy */ - av_picture_copy(dst, src, dst_pix_fmt, dst_width, dst_height); - return 0; - } - - ce = &convert_table[src_pix_fmt][dst_pix_fmt]; - if (ce->convert) { - /* specific conversion routine */ - ce->convert(dst, src, dst_width, dst_height); - return 0; - } - - /* gray to YUV */ - if (is_yuv_planar(dst_pix) && - src_pix_fmt == PIX_FMT_GRAY8) { - int w, h, y; - uint8_t *d; - - if (dst_pix->color_type == FF_COLOR_YUV_JPEG) { - ff_img_copy_plane(dst->data[0], dst->linesize[0], - src->data[0], src->linesize[0], - dst_width, dst_height); - } else { - img_apply_table(dst->data[0], dst->linesize[0], - src->data[0], src->linesize[0], - dst_width, dst_height, - y_jpeg_to_ccir); - } - /* fill U and V with 128 */ - w = dst_width; - h = dst_height; - w >>= dst_pix->x_chroma_shift; - h >>= dst_pix->y_chroma_shift; - for(i = 1; i <= 2; i++) { - d = dst->data[i]; - for(y = 0; y< h; y++) { - memset(d, 128, w); - d += dst->linesize[i]; - } - } - return 0; - } - - /* YUV to gray */ - if (is_yuv_planar(src_pix) && - dst_pix_fmt == PIX_FMT_GRAY8) { - if (src_pix->color_type == FF_COLOR_YUV_JPEG) { - ff_img_copy_plane(dst->data[0], dst->linesize[0], - src->data[0], src->linesize[0], - dst_width, dst_height); - } else { - img_apply_table(dst->data[0], dst->linesize[0], - src->data[0], src->linesize[0], - dst_width, dst_height, - y_ccir_to_jpeg); - } - return 0; - } - - /* YUV to YUV planar */ - if (is_yuv_planar(dst_pix) && is_yuv_planar(src_pix)) { - int x_shift, y_shift, w, h, xy_shift; - void (*resize_func)(uint8_t *dst, int dst_wrap, - const uint8_t *src, int src_wrap, - int width, int height); - - /* compute chroma size of the smallest dimensions */ - w = dst_width; - h = dst_height; - if (dst_pix->x_chroma_shift >= src_pix->x_chroma_shift) - w >>= dst_pix->x_chroma_shift; - else - w >>= src_pix->x_chroma_shift; - if (dst_pix->y_chroma_shift >= src_pix->y_chroma_shift) - h >>= dst_pix->y_chroma_shift; - else - h >>= src_pix->y_chroma_shift; - - x_shift = (dst_pix->x_chroma_shift - src_pix->x_chroma_shift); - y_shift = (dst_pix->y_chroma_shift - src_pix->y_chroma_shift); - xy_shift = ((x_shift & 0xf) << 4) | (y_shift & 0xf); - /* there must be filters for conversion at least from and to - YUV444 format */ - switch(xy_shift) { - case 0x00: - resize_func = ff_img_copy_plane; - break; - case 0x10: - resize_func = shrink21; - break; - case 0x20: - resize_func = shrink41; - break; - case 0x01: - resize_func = shrink12; - break; - case 0x11: - resize_func = ff_shrink22; - break; - case 0x22: - resize_func = ff_shrink44; - break; - case 0xf0: - resize_func = grow21; - break; - case 0x0f: - resize_func = grow12; - break; - case 0xe0: - resize_func = grow41; - break; - case 0xff: - resize_func = grow22; - break; - case 0xee: - resize_func = grow44; - break; - case 0xf1: - resize_func = conv411; - break; - default: - /* currently not handled */ - goto no_chroma_filter; - } - - ff_img_copy_plane(dst->data[0], dst->linesize[0], - src->data[0], src->linesize[0], - dst_width, dst_height); - - for(i = 1;i <= 2; i++) - resize_func(dst->data[i], dst->linesize[i], - src->data[i], src->linesize[i], - dst_width>>dst_pix->x_chroma_shift, dst_height>>dst_pix->y_chroma_shift); - /* if yuv color space conversion is needed, we do it here on - the destination image */ - if (dst_pix->color_type != src_pix->color_type) { - const uint8_t *y_table, *c_table; - if (dst_pix->color_type == FF_COLOR_YUV) { - y_table = y_jpeg_to_ccir; - c_table = c_jpeg_to_ccir; - } else { - y_table = y_ccir_to_jpeg; - c_table = c_ccir_to_jpeg; - } - img_apply_table(dst->data[0], dst->linesize[0], - dst->data[0], dst->linesize[0], - dst_width, dst_height, - y_table); - - for(i = 1;i <= 2; i++) - img_apply_table(dst->data[i], dst->linesize[i], - dst->data[i], dst->linesize[i], - dst_width>>dst_pix->x_chroma_shift, - dst_height>>dst_pix->y_chroma_shift, - c_table); - } - return 0; - } - no_chroma_filter: - - /* try to use an intermediate format */ - if (src_pix_fmt == PIX_FMT_YUYV422 || - dst_pix_fmt == PIX_FMT_YUYV422) { - /* specific case: convert to YUV422P first */ - int_pix_fmt = PIX_FMT_YUV422P; - } else if (src_pix_fmt == PIX_FMT_UYVY422 || - dst_pix_fmt == PIX_FMT_UYVY422) { - /* specific case: convert to YUV422P first */ - int_pix_fmt = PIX_FMT_YUV422P; - } else if (src_pix_fmt == PIX_FMT_UYYVYY411 || - dst_pix_fmt == PIX_FMT_UYYVYY411) { - /* specific case: convert to YUV411P first */ - int_pix_fmt = PIX_FMT_YUV411P; - } else if ((src_pix->color_type == FF_COLOR_GRAY && - src_pix_fmt != PIX_FMT_GRAY8) || - (dst_pix->color_type == FF_COLOR_GRAY && - dst_pix_fmt != PIX_FMT_GRAY8)) { - /* gray8 is the normalized format */ - int_pix_fmt = PIX_FMT_GRAY8; - } else if ((is_yuv_planar(src_pix) && - src_pix_fmt != PIX_FMT_YUV444P && - src_pix_fmt != PIX_FMT_YUVJ444P)) { - /* yuv444 is the normalized format */ - if (src_pix->color_type == FF_COLOR_YUV_JPEG) - int_pix_fmt = PIX_FMT_YUVJ444P; - else - int_pix_fmt = PIX_FMT_YUV444P; - } else if ((is_yuv_planar(dst_pix) && - dst_pix_fmt != PIX_FMT_YUV444P && - dst_pix_fmt != PIX_FMT_YUVJ444P)) { - /* yuv444 is the normalized format */ - if (dst_pix->color_type == FF_COLOR_YUV_JPEG) - int_pix_fmt = PIX_FMT_YUVJ444P; - else - int_pix_fmt = PIX_FMT_YUV444P; - } else { - /* the two formats are rgb or gray8 or yuv[j]444p */ - if (src_pix->is_alpha && dst_pix->is_alpha) - int_pix_fmt = PIX_FMT_RGB32; - else - int_pix_fmt = PIX_FMT_RGB24; - } - if (src_pix_fmt == int_pix_fmt) - return -1; - if (avpicture_alloc(tmp, int_pix_fmt, dst_width, dst_height) < 0) - return -1; - ret = -1; - if (img_convert(tmp, int_pix_fmt, - src, src_pix_fmt, src_width, src_height) < 0) - goto fail1; - if (img_convert(dst, dst_pix_fmt, - tmp, int_pix_fmt, dst_width, dst_height) < 0) - goto fail1; - ret = 0; - fail1: - avpicture_free(tmp); - return ret; -} -#endif - /* NOTE: we scan all the pixels to have an exact information */ static int get_alpha_info_pal8(const AVPicture *src, int width, int height) { @@ -2763,9 +1274,6 @@ if (!pf->is_alpha) return 0; switch(pix_fmt) { - case PIX_FMT_RGB32: - ret = get_alpha_info_rgb32(src, width, height); - break; case PIX_FMT_PAL8: ret = get_alpha_info_pal8(src, width, height); break;
--- a/imgconvert_template.c Tue Mar 03 15:48:52 2009 +0000 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,875 +0,0 @@ -/* - * templates for image conversion routines - * Copyright (c) 2001, 2002, 2003 Fabrice Bellard - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#ifndef RGB_OUT -#define RGB_OUT(d, r, g, b) RGBA_OUT(d, r, g, b, 0xff) -#endif - -static void glue(yuv420p_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *y1_ptr, *y2_ptr, *cb_ptr, *cr_ptr; - uint8_t *d, *d1, *d2; - int w, y, cb, cr, r_add, g_add, b_add, width2; - uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; - unsigned int r, g, b; - - d = dst->data[0]; - y1_ptr = src->data[0]; - cb_ptr = src->data[1]; - cr_ptr = src->data[2]; - width2 = (width + 1) >> 1; - for(;height >= 2; height -= 2) { - d1 = d; - d2 = d + dst->linesize[0]; - y2_ptr = y1_ptr + src->linesize[0]; - for(w = width; w >= 2; w -= 2) { - YUV_TO_RGB1_CCIR(cb_ptr[0], cr_ptr[0]); - /* output 4 pixels */ - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[1]); - RGB_OUT(d1 + BPP, r, g, b); - - YUV_TO_RGB2_CCIR(r, g, b, y2_ptr[0]); - RGB_OUT(d2, r, g, b); - - YUV_TO_RGB2_CCIR(r, g, b, y2_ptr[1]); - RGB_OUT(d2 + BPP, r, g, b); - - d1 += 2 * BPP; - d2 += 2 * BPP; - - y1_ptr += 2; - y2_ptr += 2; - cb_ptr++; - cr_ptr++; - } - /* handle odd width */ - if (w) { - YUV_TO_RGB1_CCIR(cb_ptr[0], cr_ptr[0]); - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2_CCIR(r, g, b, y2_ptr[0]); - RGB_OUT(d2, r, g, b); - d1 += BPP; - d2 += BPP; - y1_ptr++; - y2_ptr++; - cb_ptr++; - cr_ptr++; - } - d += 2 * dst->linesize[0]; - y1_ptr += 2 * src->linesize[0] - width; - cb_ptr += src->linesize[1] - width2; - cr_ptr += src->linesize[2] - width2; - } - /* handle odd height */ - if (height) { - d1 = d; - for(w = width; w >= 2; w -= 2) { - YUV_TO_RGB1_CCIR(cb_ptr[0], cr_ptr[0]); - /* output 2 pixels */ - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[1]); - RGB_OUT(d1 + BPP, r, g, b); - - d1 += 2 * BPP; - - y1_ptr += 2; - cb_ptr++; - cr_ptr++; - } - /* handle width */ - if (w) { - YUV_TO_RGB1_CCIR(cb_ptr[0], cr_ptr[0]); - /* output 2 pixels */ - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - d1 += BPP; - - y1_ptr++; - cb_ptr++; - cr_ptr++; - } - } -} - -static void glue(yuvj420p_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *y1_ptr, *y2_ptr, *cb_ptr, *cr_ptr; - uint8_t *d, *d1, *d2; - int w, y, cb, cr, r_add, g_add, b_add, width2; - uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; - unsigned int r, g, b; - - d = dst->data[0]; - y1_ptr = src->data[0]; - cb_ptr = src->data[1]; - cr_ptr = src->data[2]; - width2 = (width + 1) >> 1; - for(;height >= 2; height -= 2) { - d1 = d; - d2 = d + dst->linesize[0]; - y2_ptr = y1_ptr + src->linesize[0]; - for(w = width; w >= 2; w -= 2) { - YUV_TO_RGB1(cb_ptr[0], cr_ptr[0]); - /* output 4 pixels */ - YUV_TO_RGB2(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2(r, g, b, y1_ptr[1]); - RGB_OUT(d1 + BPP, r, g, b); - - YUV_TO_RGB2(r, g, b, y2_ptr[0]); - RGB_OUT(d2, r, g, b); - - YUV_TO_RGB2(r, g, b, y2_ptr[1]); - RGB_OUT(d2 + BPP, r, g, b); - - d1 += 2 * BPP; - d2 += 2 * BPP; - - y1_ptr += 2; - y2_ptr += 2; - cb_ptr++; - cr_ptr++; - } - /* handle odd width */ - if (w) { - YUV_TO_RGB1(cb_ptr[0], cr_ptr[0]); - YUV_TO_RGB2(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2(r, g, b, y2_ptr[0]); - RGB_OUT(d2, r, g, b); - d1 += BPP; - d2 += BPP; - y1_ptr++; - y2_ptr++; - cb_ptr++; - cr_ptr++; - } - d += 2 * dst->linesize[0]; - y1_ptr += 2 * src->linesize[0] - width; - cb_ptr += src->linesize[1] - width2; - cr_ptr += src->linesize[2] - width2; - } - /* handle odd height */ - if (height) { - d1 = d; - for(w = width; w >= 2; w -= 2) { - YUV_TO_RGB1(cb_ptr[0], cr_ptr[0]); - /* output 2 pixels */ - YUV_TO_RGB2(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - - YUV_TO_RGB2(r, g, b, y1_ptr[1]); - RGB_OUT(d1 + BPP, r, g, b); - - d1 += 2 * BPP; - - y1_ptr += 2; - cb_ptr++; - cr_ptr++; - } - /* handle width */ - if (w) { - YUV_TO_RGB1(cb_ptr[0], cr_ptr[0]); - /* output 2 pixels */ - YUV_TO_RGB2(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - d1 += BPP; - - y1_ptr++; - cb_ptr++; - cr_ptr++; - } - } -} - -static void glue(RGB_NAME, _to_yuv420p)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int wrap, wrap3, width2; - int r, g, b, r1, g1, b1, w; - uint8_t *lum, *cb, *cr; - const uint8_t *p; - - lum = dst->data[0]; - cb = dst->data[1]; - cr = dst->data[2]; - - width2 = (width + 1) >> 1; - wrap = dst->linesize[0]; - wrap3 = src->linesize[0]; - p = src->data[0]; - for(;height>=2;height -= 2) { - for(w = width; w >= 2; w -= 2) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y_CCIR(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y_CCIR(r, g, b); - p += wrap3; - lum += wrap; - - RGB_IN(r, g, b, p); - r1 += r; - g1 += g; - b1 += b; - lum[0] = RGB_TO_Y_CCIR(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y_CCIR(r, g, b); - - cb[0] = RGB_TO_U_CCIR(r1, g1, b1, 2); - cr[0] = RGB_TO_V_CCIR(r1, g1, b1, 2); - - cb++; - cr++; - p += -wrap3 + 2 * BPP; - lum += -wrap + 2; - } - if (w) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y_CCIR(r, g, b); - p += wrap3; - lum += wrap; - RGB_IN(r, g, b, p); - r1 += r; - g1 += g; - b1 += b; - lum[0] = RGB_TO_Y_CCIR(r, g, b); - cb[0] = RGB_TO_U_CCIR(r1, g1, b1, 1); - cr[0] = RGB_TO_V_CCIR(r1, g1, b1, 1); - cb++; - cr++; - p += -wrap3 + BPP; - lum += -wrap + 1; - } - p += wrap3 + (wrap3 - width * BPP); - lum += wrap + (wrap - width); - cb += dst->linesize[1] - width2; - cr += dst->linesize[2] - width2; - } - /* handle odd height */ - if (height) { - for(w = width; w >= 2; w -= 2) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y_CCIR(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y_CCIR(r, g, b); - cb[0] = RGB_TO_U_CCIR(r1, g1, b1, 1); - cr[0] = RGB_TO_V_CCIR(r1, g1, b1, 1); - cb++; - cr++; - p += 2 * BPP; - lum += 2; - } - if (w) { - RGB_IN(r, g, b, p); - lum[0] = RGB_TO_Y_CCIR(r, g, b); - cb[0] = RGB_TO_U_CCIR(r, g, b, 0); - cr[0] = RGB_TO_V_CCIR(r, g, b, 0); - } - } -} - -static void glue(RGB_NAME, _to_gray)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const unsigned char *p; - unsigned char *q; - int r, g, b, dst_wrap, src_wrap; - int x, y; - - p = src->data[0]; - src_wrap = src->linesize[0] - BPP * width; - - q = dst->data[0]; - dst_wrap = dst->linesize[0] - width; - - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - RGB_IN(r, g, b, p); - q[0] = RGB_TO_Y(r, g, b); - q++; - p += BPP; - } - p += src_wrap; - q += dst_wrap; - } -} - -static void glue(gray_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const unsigned char *p; - unsigned char *q; - int r, dst_wrap, src_wrap; - int x, y; - - p = src->data[0]; - src_wrap = src->linesize[0] - width; - - q = dst->data[0]; - dst_wrap = dst->linesize[0] - BPP * width; - - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - r = p[0]; - RGB_OUT(q, r, r, r); - q += BPP; - p ++; - } - p += src_wrap; - q += dst_wrap; - } -} - -static void glue(pal8_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const unsigned char *p; - unsigned char *q; - int r, g, b, dst_wrap, src_wrap; - int x, y; - uint32_t v; - const uint32_t *palette; - - p = src->data[0]; - src_wrap = src->linesize[0] - width; - palette = (uint32_t *)src->data[1]; - - q = dst->data[0]; - dst_wrap = dst->linesize[0] - BPP * width; - - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - v = palette[p[0]]; - r = (v >> 16) & 0xff; - g = (v >> 8) & 0xff; - b = (v) & 0xff; -#ifdef RGBA_OUT - { - int a; - a = (v >> 24) & 0xff; - RGBA_OUT(q, r, g, b, a); - } -#else - RGB_OUT(q, r, g, b); -#endif - q += BPP; - p ++; - } - p += src_wrap; - q += dst_wrap; - } -} - -// RGB24 has optimized routines -#if !defined(FMT_RGB32) && !defined(FMT_RGB24) -/* alpha support */ - -static void glue(rgb32_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *s; - uint8_t *d; - int src_wrap, dst_wrap, j, y; - unsigned int v, r, g, b; -#ifdef RGBA_OUT - unsigned int a; -#endif - - s = src->data[0]; - src_wrap = src->linesize[0] - width * 4; - - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width * BPP; - - for(y=0;y<height;y++) { - for(j = 0;j < width; j++) { - v = ((const uint32_t *)(s))[0]; - r = (v >> 16) & 0xff; - g = (v >> 8) & 0xff; - b = v & 0xff; -#ifdef RGBA_OUT - a = (v >> 24) & 0xff; - RGBA_OUT(d, r, g, b, a); -#else - RGB_OUT(d, r, g, b); -#endif - s += 4; - d += BPP; - } - s += src_wrap; - d += dst_wrap; - } -} - -static void glue(RGB_NAME, _to_rgb32)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *s; - uint8_t *d; - int src_wrap, dst_wrap, j, y; - unsigned int r, g, b; -#ifdef RGBA_IN - unsigned int a; -#endif - - s = src->data[0]; - src_wrap = src->linesize[0] - width * BPP; - - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width * 4; - - for(y=0;y<height;y++) { - for(j = 0;j < width; j++) { -#ifdef RGBA_IN - RGBA_IN(r, g, b, a, s); - ((uint32_t *)(d))[0] = (a << 24) | (r << 16) | (g << 8) | b; -#else - RGB_IN(r, g, b, s); - ((uint32_t *)(d))[0] = (0xff << 24) | (r << 16) | (g << 8) | b; -#endif - d += 4; - s += BPP; - } - s += src_wrap; - d += dst_wrap; - } -} - -#endif /* !defined(FMT_RGB32) */ - -#ifndef FMT_RGB24 - -static void glue(rgb24_to_, RGB_NAME)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *s; - uint8_t *d; - int src_wrap, dst_wrap, j, y; - unsigned int r, g, b; - - s = src->data[0]; - src_wrap = src->linesize[0] - width * 3; - - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width * BPP; - - for(y=0;y<height;y++) { - for(j = 0;j < width; j++) { - r = s[0]; - g = s[1]; - b = s[2]; - RGB_OUT(d, r, g, b); - s += 3; - d += BPP; - } - s += src_wrap; - d += dst_wrap; - } -} - -static void glue(RGB_NAME, _to_rgb24)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *s; - uint8_t *d; - int src_wrap, dst_wrap, j, y; - unsigned int r, g , b; - - s = src->data[0]; - src_wrap = src->linesize[0] - width * BPP; - - d = dst->data[0]; - dst_wrap = dst->linesize[0] - width * 3; - - for(y=0;y<height;y++) { - for(j = 0;j < width; j++) { - RGB_IN(r, g, b, s) - d[0] = r; - d[1] = g; - d[2] = b; - d += 3; - s += BPP; - } - s += src_wrap; - d += dst_wrap; - } -} - -#endif /* !FMT_RGB24 */ - -#ifdef FMT_RGB24 - -static void yuv444p_to_rgb24(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *y1_ptr, *cb_ptr, *cr_ptr; - uint8_t *d, *d1; - int w, y, cb, cr, r_add, g_add, b_add; - uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; - unsigned int r, g, b; - - d = dst->data[0]; - y1_ptr = src->data[0]; - cb_ptr = src->data[1]; - cr_ptr = src->data[2]; - for(;height > 0; height --) { - d1 = d; - for(w = width; w > 0; w--) { - YUV_TO_RGB1_CCIR(cb_ptr[0], cr_ptr[0]); - - YUV_TO_RGB2_CCIR(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - d1 += BPP; - - y1_ptr++; - cb_ptr++; - cr_ptr++; - } - d += dst->linesize[0]; - y1_ptr += src->linesize[0] - width; - cb_ptr += src->linesize[1] - width; - cr_ptr += src->linesize[2] - width; - } -} - -static void yuvj444p_to_rgb24(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const uint8_t *y1_ptr, *cb_ptr, *cr_ptr; - uint8_t *d, *d1; - int w, y, cb, cr, r_add, g_add, b_add; - uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; - unsigned int r, g, b; - - d = dst->data[0]; - y1_ptr = src->data[0]; - cb_ptr = src->data[1]; - cr_ptr = src->data[2]; - for(;height > 0; height --) { - d1 = d; - for(w = width; w > 0; w--) { - YUV_TO_RGB1(cb_ptr[0], cr_ptr[0]); - - YUV_TO_RGB2(r, g, b, y1_ptr[0]); - RGB_OUT(d1, r, g, b); - d1 += BPP; - - y1_ptr++; - cb_ptr++; - cr_ptr++; - } - d += dst->linesize[0]; - y1_ptr += src->linesize[0] - width; - cb_ptr += src->linesize[1] - width; - cr_ptr += src->linesize[2] - width; - } -} - -static void rgb24_to_yuv444p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int src_wrap, x, y; - int r, g, b; - uint8_t *lum, *cb, *cr; - const uint8_t *p; - - lum = dst->data[0]; - cb = dst->data[1]; - cr = dst->data[2]; - - src_wrap = src->linesize[0] - width * BPP; - p = src->data[0]; - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - RGB_IN(r, g, b, p); - lum[0] = RGB_TO_Y_CCIR(r, g, b); - cb[0] = RGB_TO_U_CCIR(r, g, b, 0); - cr[0] = RGB_TO_V_CCIR(r, g, b, 0); - p += BPP; - cb++; - cr++; - lum++; - } - p += src_wrap; - lum += dst->linesize[0] - width; - cb += dst->linesize[1] - width; - cr += dst->linesize[2] - width; - } -} - -static void rgb24_to_yuvj420p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int wrap, wrap3, width2; - int r, g, b, r1, g1, b1, w; - uint8_t *lum, *cb, *cr; - const uint8_t *p; - - lum = dst->data[0]; - cb = dst->data[1]; - cr = dst->data[2]; - - width2 = (width + 1) >> 1; - wrap = dst->linesize[0]; - wrap3 = src->linesize[0]; - p = src->data[0]; - for(;height>=2;height -= 2) { - for(w = width; w >= 2; w -= 2) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y(r, g, b); - p += wrap3; - lum += wrap; - - RGB_IN(r, g, b, p); - r1 += r; - g1 += g; - b1 += b; - lum[0] = RGB_TO_Y(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y(r, g, b); - - cb[0] = RGB_TO_U(r1, g1, b1, 2); - cr[0] = RGB_TO_V(r1, g1, b1, 2); - - cb++; - cr++; - p += -wrap3 + 2 * BPP; - lum += -wrap + 2; - } - if (w) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y(r, g, b); - p += wrap3; - lum += wrap; - RGB_IN(r, g, b, p); - r1 += r; - g1 += g; - b1 += b; - lum[0] = RGB_TO_Y(r, g, b); - cb[0] = RGB_TO_U(r1, g1, b1, 1); - cr[0] = RGB_TO_V(r1, g1, b1, 1); - cb++; - cr++; - p += -wrap3 + BPP; - lum += -wrap + 1; - } - p += wrap3 + (wrap3 - width * BPP); - lum += wrap + (wrap - width); - cb += dst->linesize[1] - width2; - cr += dst->linesize[2] - width2; - } - /* handle odd height */ - if (height) { - for(w = width; w >= 2; w -= 2) { - RGB_IN(r, g, b, p); - r1 = r; - g1 = g; - b1 = b; - lum[0] = RGB_TO_Y(r, g, b); - - RGB_IN(r, g, b, p + BPP); - r1 += r; - g1 += g; - b1 += b; - lum[1] = RGB_TO_Y(r, g, b); - cb[0] = RGB_TO_U(r1, g1, b1, 1); - cr[0] = RGB_TO_V(r1, g1, b1, 1); - cb++; - cr++; - p += 2 * BPP; - lum += 2; - } - if (w) { - RGB_IN(r, g, b, p); - lum[0] = RGB_TO_Y(r, g, b); - cb[0] = RGB_TO_U(r, g, b, 0); - cr[0] = RGB_TO_V(r, g, b, 0); - } - } -} - -static void rgb24_to_yuvj444p(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - int src_wrap, x, y; - int r, g, b; - uint8_t *lum, *cb, *cr; - const uint8_t *p; - - lum = dst->data[0]; - cb = dst->data[1]; - cr = dst->data[2]; - - src_wrap = src->linesize[0] - width * BPP; - p = src->data[0]; - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - RGB_IN(r, g, b, p); - lum[0] = RGB_TO_Y(r, g, b); - cb[0] = RGB_TO_U(r, g, b, 0); - cr[0] = RGB_TO_V(r, g, b, 0); - p += BPP; - cb++; - cr++; - lum++; - } - p += src_wrap; - lum += dst->linesize[0] - width; - cb += dst->linesize[1] - width; - cr += dst->linesize[2] - width; - } -} - -#endif /* FMT_RGB24 */ - -#if defined(FMT_RGB24) || defined(FMT_RGB32) - -static void glue(RGB_NAME, _to_pal8)(AVPicture *dst, const AVPicture *src, - int width, int height) -{ - const unsigned char *p; - unsigned char *q; - int dst_wrap, src_wrap; - int x, y, has_alpha; - unsigned int r, g, b; - - p = src->data[0]; - src_wrap = src->linesize[0] - BPP * width; - - q = dst->data[0]; - dst_wrap = dst->linesize[0] - width; - has_alpha = 0; - - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { -#ifdef RGBA_IN - { - unsigned int a; - RGBA_IN(r, g, b, a, p); - /* crude approximation for alpha ! */ - if (a < 0x80) { - has_alpha = 1; - q[0] = TRANSP_INDEX; - } else { - q[0] = gif_clut_index(r, g, b); - } - } -#else - RGB_IN(r, g, b, p); - q[0] = gif_clut_index(r, g, b); -#endif - q++; - p += BPP; - } - p += src_wrap; - q += dst_wrap; - } - - build_rgb_palette(dst->data[1], has_alpha); -} - -#endif /* defined(FMT_RGB24) || defined(FMT_RGB32) */ - -#ifdef RGBA_IN - -static int glue(get_alpha_info_, RGB_NAME)(const AVPicture *src, - int width, int height) -{ - const unsigned char *p; - int src_wrap, ret, x, y; - unsigned int r, g, b, a; - - p = src->data[0]; - src_wrap = src->linesize[0] - BPP * width; - ret = 0; - for(y=0;y<height;y++) { - for(x=0;x<width;x++) { - RGBA_IN(r, g, b, a, p); - if (a == 0x00) { - ret |= FF_ALPHA_TRANSP; - } else if (a != 0xff) { - ret |= FF_ALPHA_SEMI_TRANSP; - } - p += BPP; - } - p += src_wrap; - } - return ret; -} - -#endif /* RGBA_IN */ - -#undef RGB_IN -#undef RGBA_IN -#undef RGB_OUT -#undef RGBA_OUT -#undef BPP -#undef RGB_NAME -#undef FMT_RGB24 -#undef FMT_RGB32
--- a/imgresample.c Tue Mar 03 15:48:52 2009 +0000 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,833 +0,0 @@ -/* - * High quality image resampling with polyphase filters - * Copyright (c) 2001 Fabrice Bellard - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/** - * @file libavcodec/imgresample.c - * High quality image resampling with polyphase filters . - * - * WARNING: This file is deprecated and will be removed after FFmpeg 0.5 - * release, do not lose your time improving it! - */ - -#include "avcodec.h" -#include "dsputil.h" -#include "imgconvert.h" -#include "libswscale/swscale.h" - -#if HAVE_ALTIVEC -#include "ppc/imgresample_altivec.h" -#endif - -#define NB_COMPONENTS 3 - -#define PHASE_BITS 4 -#define NB_PHASES (1 << PHASE_BITS) -#define NB_TAPS 4 -#define FCENTER 1 /* index of the center of the filter */ -//#define TEST 1 /* Test it */ - -#define POS_FRAC_BITS 16 -#define POS_FRAC (1 << POS_FRAC_BITS) -/* 6 bits precision is needed for MMX */ -#define FILTER_BITS 8 - -#define LINE_BUF_HEIGHT (NB_TAPS * 4) - -struct SwsContext { - const AVClass *av_class; - struct ImgReSampleContext *resampling_ctx; - enum PixelFormat src_pix_fmt, dst_pix_fmt; -}; - -typedef struct ImgReSampleContext { - int iwidth, iheight, owidth, oheight; - int topBand, bottomBand, leftBand, rightBand; - int padtop, padbottom, padleft, padright; - int pad_owidth, pad_oheight; - int h_incr, v_incr; - DECLARE_ALIGNED_8(int16_t, h_filters[NB_PHASES][NB_TAPS]); /* horizontal filters */ - DECLARE_ALIGNED_8(int16_t, v_filters[NB_PHASES][NB_TAPS]); /* vertical filters */ - uint8_t *line_buf; -} ImgReSampleContext; - -void av_build_filter(int16_t *filter, double factor, int tap_count, int phase_count, int scale, int type); - -static inline int get_phase(int pos) -{ - return ((pos) >> (POS_FRAC_BITS - PHASE_BITS)) & ((1 << PHASE_BITS) - 1); -} - -/* This function must be optimized */ -static void h_resample_fast(uint8_t *dst, int dst_width, const uint8_t *src, - int src_width, int src_start, int src_incr, - int16_t *filters) -{ - int src_pos, phase, sum, i; - const uint8_t *s; - int16_t *filter; - - src_pos = src_start; - for(i=0;i<dst_width;i++) { -#ifdef TEST - /* test */ - if ((src_pos >> POS_FRAC_BITS) < 0 || - (src_pos >> POS_FRAC_BITS) > (src_width - NB_TAPS)) - av_abort(); -#endif - s = src + (src_pos >> POS_FRAC_BITS); - phase = get_phase(src_pos); - filter = filters + phase * NB_TAPS; -#if NB_TAPS == 4 - sum = s[0] * filter[0] + - s[1] * filter[1] + - s[2] * filter[2] + - s[3] * filter[3]; -#else - { - int j; - sum = 0; - for(j=0;j<NB_TAPS;j++) - sum += s[j] * filter[j]; - } -#endif - sum = sum >> FILTER_BITS; - if (sum < 0) - sum = 0; - else if (sum > 255) - sum = 255; - dst[0] = sum; - src_pos += src_incr; - dst++; - } -} - -/* This function must be optimized */ -static void v_resample(uint8_t *dst, int dst_width, const uint8_t *src, - int wrap, int16_t *filter) -{ - int sum, i; - const uint8_t *s; - - s = src; - for(i=0;i<dst_width;i++) { -#if NB_TAPS == 4 - sum = s[0 * wrap] * filter[0] + - s[1 * wrap] * filter[1] + - s[2 * wrap] * filter[2] + - s[3 * wrap] * filter[3]; -#else - { - int j; - uint8_t *s1 = s; - - sum = 0; - for(j=0;j<NB_TAPS;j++) { - sum += s1[0] * filter[j]; - s1 += wrap; - } - } -#endif - sum = sum >> FILTER_BITS; - if (sum < 0) - sum = 0; - else if (sum > 255) - sum = 255; - dst[0] = sum; - dst++; - s++; - } -} - -#if HAVE_MMX - -#include "x86/mmx.h" - -#define FILTER4(reg) \ -{\ - s = src + (src_pos >> POS_FRAC_BITS);\ - phase = get_phase(src_pos);\ - filter = filters + phase * NB_TAPS;\ - movq_m2r(*s, reg);\ - punpcklbw_r2r(mm7, reg);\ - movq_m2r(*filter, mm6);\ - pmaddwd_r2r(reg, mm6);\ - movq_r2r(mm6, reg);\ - psrlq_i2r(32, reg);\ - paddd_r2r(mm6, reg);\ - psrad_i2r(FILTER_BITS, reg);\ - src_pos += src_incr;\ -} - -#define DUMP(reg) movq_r2m(reg, tmp); printf(#reg "=%016"PRIx64"\n", tmp.uq); - -/* XXX: do four pixels at a time */ -static void h_resample_fast4_mmx(uint8_t *dst, int dst_width, - const uint8_t *src, int src_width, - int src_start, int src_incr, int16_t *filters) -{ - int src_pos, phase; - const uint8_t *s; - int16_t *filter; - uint64_t tmp; - - src_pos = src_start; - pxor_r2r(mm7, mm7); - - while (dst_width >= 4) { - - FILTER4(mm0); - FILTER4(mm1); - FILTER4(mm2); - FILTER4(mm3); - - packuswb_r2r(mm7, mm0); - packuswb_r2r(mm7, mm1); - packuswb_r2r(mm7, mm3); - packuswb_r2r(mm7, mm2); - movq_r2m(mm0, tmp); - dst[0] = tmp & 0xFF; - movq_r2m(mm1, tmp); - dst[1] = tmp & 0xFF; - movq_r2m(mm2, tmp); - dst[2] = tmp & 0xFF; - movq_r2m(mm3, tmp); - dst[3] = tmp & 0xFF; - dst += 4; - dst_width -= 4; - } - while (dst_width > 0) { - FILTER4(mm0); - packuswb_r2r(mm7, mm0); - movq_r2m(mm0, tmp); - dst[0] = tmp & 0xFF; - dst++; - dst_width--; - } - emms(); -} - -static void v_resample4_mmx(uint8_t *dst, int dst_width, const uint8_t *src, - int wrap, int16_t *filter) -{ - int sum, i; - const uint8_t *s; - uint64_t tmp; - uint64_t coefs[4]; - - for(i=0;i<4;i++) { - tmp = filter[i]; - coefs[i] = (tmp<<48) + (tmp<<32) + (tmp<<16) + tmp; - } - - pxor_r2r(mm7, mm7); - s = src; - while (dst_width >= 4) { - movq_m2r(s[0 * wrap], mm0); - punpcklbw_r2r(mm7, mm0); - movq_m2r(s[1 * wrap], mm1); - punpcklbw_r2r(mm7, mm1); - movq_m2r(s[2 * wrap], mm2); - punpcklbw_r2r(mm7, mm2); - movq_m2r(s[3 * wrap], mm3); - punpcklbw_r2r(mm7, mm3); - - pmullw_m2r(coefs[0], mm0); - pmullw_m2r(coefs[1], mm1); - pmullw_m2r(coefs[2], mm2); - pmullw_m2r(coefs[3], mm3); - - paddw_r2r(mm1, mm0); - paddw_r2r(mm3, mm2); - paddw_r2r(mm2, mm0); - psraw_i2r(FILTER_BITS, mm0); - - packuswb_r2r(mm7, mm0); - movq_r2m(mm0, tmp); - - *(uint32_t *)dst = tmp & 0xFFFFFFFF; - dst += 4; - s += 4; - dst_width -= 4; - } - while (dst_width > 0) { - sum = s[0 * wrap] * filter[0] + - s[1 * wrap] * filter[1] + - s[2 * wrap] * filter[2] + - s[3 * wrap] * filter[3]; - sum = sum >> FILTER_BITS; - if (sum < 0) - sum = 0; - else if (sum > 255) - sum = 255; - dst[0] = sum; - dst++; - s++; - dst_width--; - } - emms(); -} -#endif /* HAVE_MMX */ - -/* slow version to handle limit cases. Does not need optimization */ -static void h_resample_slow(uint8_t *dst, int dst_width, - const uint8_t *src, int src_width, - int src_start, int src_incr, int16_t *filters) -{ - int src_pos, phase, sum, j, v, i; - const uint8_t *s, *src_end; - int16_t *filter; - - src_end = src + src_width; - src_pos = src_start; - for(i=0;i<dst_width;i++) { - s = src + (src_pos >> POS_FRAC_BITS); - phase = get_phase(src_pos); - filter = filters + phase * NB_TAPS; - sum = 0; - for(j=0;j<NB_TAPS;j++) { - if (s < src) - v = src[0]; - else if (s >= src_end) - v = src_end[-1]; - else - v = s[0]; - sum += v * filter[j]; - s++; - } - sum = sum >> FILTER_BITS; - if (sum < 0) - sum = 0; - else if (sum > 255) - sum = 255; - dst[0] = sum; - src_pos += src_incr; - dst++; - } -} - -static void h_resample(uint8_t *dst, int dst_width, const uint8_t *src, - int src_width, int src_start, int src_incr, - int16_t *filters) -{ - int n, src_end; - - if (src_start < 0) { - n = (0 - src_start + src_incr - 1) / src_incr; - h_resample_slow(dst, n, src, src_width, src_start, src_incr, filters); - dst += n; - dst_width -= n; - src_start += n * src_incr; - } - src_end = src_start + dst_width * src_incr; - if (src_end > ((src_width - NB_TAPS) << POS_FRAC_BITS)) { - n = (((src_width - NB_TAPS + 1) << POS_FRAC_BITS) - 1 - src_start) / - src_incr; - } else { - n = dst_width; - } -#if HAVE_MMX - if ((mm_flags & FF_MM_MMX) && NB_TAPS == 4) - h_resample_fast4_mmx(dst, n, - src, src_width, src_start, src_incr, filters); - else -#endif - h_resample_fast(dst, n, - src, src_width, src_start, src_incr, filters); - if (n < dst_width) { - dst += n; - dst_width -= n; - src_start += n * src_incr; - h_resample_slow(dst, dst_width, - src, src_width, src_start, src_incr, filters); - } -} - -static void component_resample(ImgReSampleContext *s, - uint8_t *output, int owrap, int owidth, int oheight, - uint8_t *input, int iwrap, int iwidth, int iheight) -{ - int src_y, src_y1, last_src_y, ring_y, phase_y, y1, y; - uint8_t *new_line, *src_line; - - last_src_y = - FCENTER - 1; - /* position of the bottom of the filter in the source image */ - src_y = (last_src_y + NB_TAPS) * POS_FRAC; - ring_y = NB_TAPS; /* position in ring buffer */ - for(y=0;y<oheight;y++) { - /* apply horizontal filter on new lines from input if needed */ - src_y1 = src_y >> POS_FRAC_BITS; - while (last_src_y < src_y1) { - if (++ring_y >= LINE_BUF_HEIGHT + NB_TAPS) - ring_y = NB_TAPS; - last_src_y++; - /* handle limit conditions : replicate line (slightly - inefficient because we filter multiple times) */ - y1 = last_src_y; - if (y1 < 0) { - y1 = 0; - } else if (y1 >= iheight) { - y1 = iheight - 1; - } - src_line = input + y1 * iwrap; - new_line = s->line_buf + ring_y * owidth; - /* apply filter and handle limit cases correctly */ - h_resample(new_line, owidth, - src_line, iwidth, - FCENTER * POS_FRAC, s->h_incr, - &s->h_filters[0][0]); - /* handle ring buffer wrapping */ - if (ring_y >= LINE_BUF_HEIGHT) { - memcpy(s->line_buf + (ring_y - LINE_BUF_HEIGHT) * owidth, - new_line, owidth); - } - } - /* apply vertical filter */ - phase_y = get_phase(src_y); -#if HAVE_MMX - /* desactivated MMX because loss of precision */ - if ((mm_flags & FF_MM_MMX) && NB_TAPS == 4 && 0) - v_resample4_mmx(output, owidth, - s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth, - &s->v_filters[phase_y][0]); - else -#endif -#if HAVE_ALTIVEC - if ((mm_flags & FF_MM_ALTIVEC) && NB_TAPS == 4 && FILTER_BITS <= 6) - v_resample16_altivec(output, owidth, - s->line_buf + (ring_y - NB_TAPS + 1) * owidth, - owidth, &s->v_filters[phase_y][0]); - else -#endif - v_resample(output, owidth, - s->line_buf + (ring_y - NB_TAPS + 1) * owidth, owidth, - &s->v_filters[phase_y][0]); - - src_y += s->v_incr; - - output += owrap; - } -} - -ImgReSampleContext *img_resample_full_init(int owidth, int oheight, - int iwidth, int iheight, - int topBand, int bottomBand, - int leftBand, int rightBand, - int padtop, int padbottom, - int padleft, int padright) -{ - ImgReSampleContext *s; - - if (!owidth || !oheight || !iwidth || !iheight) - return NULL; - - s = av_mallocz(sizeof(ImgReSampleContext)); - if (!s) - return NULL; - if((unsigned)owidth >= UINT_MAX / (LINE_BUF_HEIGHT + NB_TAPS)) - goto fail; - s->line_buf = av_mallocz(owidth * (LINE_BUF_HEIGHT + NB_TAPS)); - if (!s->line_buf) - goto fail; - - s->owidth = owidth; - s->oheight = oheight; - s->iwidth = iwidth; - s->iheight = iheight; - - s->topBand = topBand; - s->bottomBand = bottomBand; - s->leftBand = leftBand; - s->rightBand = rightBand; - - s->padtop = padtop; - s->padbottom = padbottom; - s->padleft = padleft; - s->padright = padright; - - s->pad_owidth = owidth - (padleft + padright); - s->pad_oheight = oheight - (padtop + padbottom); - - s->h_incr = ((iwidth - leftBand - rightBand) * POS_FRAC) / s->pad_owidth; - s->v_incr = ((iheight - topBand - bottomBand) * POS_FRAC) / s->pad_oheight; - - av_build_filter(&s->h_filters[0][0], (float) s->pad_owidth / - (float) (iwidth - leftBand - rightBand), NB_TAPS, NB_PHASES, 1<<FILTER_BITS, 0); - av_build_filter(&s->v_filters[0][0], (float) s->pad_oheight / - (float) (iheight - topBand - bottomBand), NB_TAPS, NB_PHASES, 1<<FILTER_BITS, 0); - - return s; -fail: - av_free(s); - return NULL; -} - -ImgReSampleContext *img_resample_init(int owidth, int oheight, - int iwidth, int iheight) -{ - return img_resample_full_init(owidth, oheight, iwidth, iheight, - 0, 0, 0, 0, 0, 0, 0, 0); -} - -void img_resample(ImgReSampleContext *s, - AVPicture *output, const AVPicture *input) -{ - int i, shift; - uint8_t* optr; - - for (i=0;i<3;i++) { - shift = (i == 0) ? 0 : 1; - - optr = output->data[i] + (((output->linesize[i] * - s->padtop) + s->padleft) >> shift); - - component_resample(s, optr, output->linesize[i], - s->pad_owidth >> shift, s->pad_oheight >> shift, - input->data[i] + (input->linesize[i] * - (s->topBand >> shift)) + (s->leftBand >> shift), - input->linesize[i], ((s->iwidth - s->leftBand - - s->rightBand) >> shift), - (s->iheight - s->topBand - s->bottomBand) >> shift); - } -} - -void img_resample_close(ImgReSampleContext *s) -{ - av_free(s->line_buf); - av_free(s); -} - -static const char *context_to_name(void* ptr) -{ - return "imgconvert"; -} - -static const AVClass context_class = { "imgresample", context_to_name, NULL }; - -struct SwsContext *sws_getContext(int srcW, int srcH, int srcFormat, - int dstW, int dstH, int dstFormat, - int flags, SwsFilter *srcFilter, - SwsFilter *dstFilter, double *param) -{ - struct SwsContext *ctx; - - ctx = av_malloc(sizeof(struct SwsContext)); - if (!ctx) { - av_log(NULL, AV_LOG_ERROR, "Cannot allocate a resampling context!\n"); - - return NULL; - } - ctx->av_class = &context_class; - - if ((srcH != dstH) || (srcW != dstW)) { - if ((srcFormat != PIX_FMT_YUV420P) || (dstFormat != PIX_FMT_YUV420P)) { - av_log(ctx, AV_LOG_INFO, "PIX_FMT_YUV420P will be used as an intermediate format for rescaling\n"); - } - ctx->resampling_ctx = img_resample_init(dstW, dstH, srcW, srcH); - } else { - ctx->resampling_ctx = av_malloc(sizeof(ImgReSampleContext)); - ctx->resampling_ctx->iheight = srcH; - ctx->resampling_ctx->iwidth = srcW; - ctx->resampling_ctx->oheight = dstH; - ctx->resampling_ctx->owidth = dstW; - } - ctx->src_pix_fmt = srcFormat; - ctx->dst_pix_fmt = dstFormat; - - return ctx; -} - -void sws_freeContext(struct SwsContext *ctx) -{ - if (!ctx) - return; - if ((ctx->resampling_ctx->iwidth != ctx->resampling_ctx->owidth) || - (ctx->resampling_ctx->iheight != ctx->resampling_ctx->oheight)) { - img_resample_close(ctx->resampling_ctx); - } else { - av_free(ctx->resampling_ctx); - } - av_free(ctx); -} - - -/** - * Checks if context is valid or reallocs a new one instead. - * If context is NULL, just calls sws_getContext() to get a new one. - * Otherwise, checks if the parameters are the same already saved in context. - * If that is the case, returns the current context. - * Otherwise, frees context and gets a new one. - * - * Be warned that srcFilter, dstFilter are not checked, they are - * assumed to remain valid. - */ -struct SwsContext *sws_getCachedContext(struct SwsContext *ctx, - int srcW, int srcH, int srcFormat, - int dstW, int dstH, int dstFormat, int flags, - SwsFilter *srcFilter, SwsFilter *dstFilter, double *param) -{ - if (ctx != NULL) { - if ((ctx->resampling_ctx->iwidth != srcW) || - (ctx->resampling_ctx->iheight != srcH) || - (ctx->src_pix_fmt != srcFormat) || - (ctx->resampling_ctx->owidth != dstW) || - (ctx->resampling_ctx->oheight != dstH) || - (ctx->dst_pix_fmt != dstFormat)) - { - sws_freeContext(ctx); - ctx = NULL; - } - } - if (ctx == NULL) { - return sws_getContext(srcW, srcH, srcFormat, - dstW, dstH, dstFormat, flags, - srcFilter, dstFilter, param); - } - return ctx; -} - -int sws_scale(struct SwsContext *ctx, uint8_t* src[], int srcStride[], - int srcSliceY, int srcSliceH, uint8_t* dst[], int dstStride[]) -{ - AVPicture src_pict, dst_pict; - int i, res = 0; - AVPicture picture_format_temp; - AVPicture picture_resample_temp, *formatted_picture, *resampled_picture; - uint8_t *buf1 = NULL, *buf2 = NULL; - enum PixelFormat current_pix_fmt; - - for (i = 0; i < 4; i++) { - src_pict.data[i] = src[i]; - src_pict.linesize[i] = srcStride[i]; - dst_pict.data[i] = dst[i]; - dst_pict.linesize[i] = dstStride[i]; - } - if ((ctx->resampling_ctx->iwidth != ctx->resampling_ctx->owidth) || - (ctx->resampling_ctx->iheight != ctx->resampling_ctx->oheight)) { - /* We have to rescale the picture, but only YUV420P rescaling is supported... */ - - if (ctx->src_pix_fmt != PIX_FMT_YUV420P) { - int size; - - /* create temporary picture for rescaling input*/ - size = avpicture_get_size(PIX_FMT_YUV420P, ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight); - buf1 = av_malloc(size); - if (!buf1) { - res = -1; - goto the_end; - } - formatted_picture = &picture_format_temp; - avpicture_fill((AVPicture*)formatted_picture, buf1, - PIX_FMT_YUV420P, ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight); - - if (img_convert((AVPicture*)formatted_picture, PIX_FMT_YUV420P, - &src_pict, ctx->src_pix_fmt, - ctx->resampling_ctx->iwidth, ctx->resampling_ctx->iheight) < 0) { - - av_log(ctx, AV_LOG_ERROR, "pixel format conversion not handled\n"); - res = -1; - goto the_end; - } - } else { - formatted_picture = &src_pict; - } - - if (ctx->dst_pix_fmt != PIX_FMT_YUV420P) { - int size; - - /* create temporary picture for rescaling output*/ - size = avpicture_get_size(PIX_FMT_YUV420P, ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight); - buf2 = av_malloc(size); - if (!buf2) { - res = -1; - goto the_end; - } - resampled_picture = &picture_resample_temp; - avpicture_fill((AVPicture*)resampled_picture, buf2, - PIX_FMT_YUV420P, ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight); - - } else { - resampled_picture = &dst_pict; - } - - /* ...and finally rescale!!! */ - img_resample(ctx->resampling_ctx, resampled_picture, formatted_picture); - current_pix_fmt = PIX_FMT_YUV420P; - } else { - resampled_picture = &src_pict; - current_pix_fmt = ctx->src_pix_fmt; - } - - if (current_pix_fmt != ctx->dst_pix_fmt) { - if (img_convert(&dst_pict, ctx->dst_pix_fmt, - resampled_picture, current_pix_fmt, - ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight) < 0) { - - av_log(ctx, AV_LOG_ERROR, "pixel format conversion not handled\n"); - - res = -1; - goto the_end; - } - } else if (resampled_picture != &dst_pict) { - av_picture_copy(&dst_pict, resampled_picture, current_pix_fmt, - ctx->resampling_ctx->owidth, ctx->resampling_ctx->oheight); - } - -the_end: - av_free(buf1); - av_free(buf2); - return res; -} - - -#ifdef TEST -#include <stdio.h> -#undef exit - -/* input */ -#define XSIZE 256 -#define YSIZE 256 -uint8_t img[XSIZE * YSIZE]; - -/* output */ -#define XSIZE1 512 -#define YSIZE1 512 -uint8_t img1[XSIZE1 * YSIZE1]; -uint8_t img2[XSIZE1 * YSIZE1]; - -void save_pgm(const char *filename, uint8_t *img, int xsize, int ysize) -{ -#undef fprintf - FILE *f; - f=fopen(filename,"w"); - fprintf(f,"P5\n%d %d\n%d\n", xsize, ysize, 255); - fwrite(img,1, xsize * ysize,f); - fclose(f); -#define fprintf please_use_av_log -} - -static void dump_filter(int16_t *filter) -{ - int i, ph; - - for(ph=0;ph<NB_PHASES;ph++) { - av_log(NULL, AV_LOG_INFO, "%2d: ", ph); - for(i=0;i<NB_TAPS;i++) { - av_log(NULL, AV_LOG_INFO, " %5.2f", filter[ph * NB_TAPS + i] / 256.0); - } - av_log(NULL, AV_LOG_INFO, "\n"); - } -} - -#if HAVE_MMX -int mm_flags; -#endif - -int main(int argc, char **argv) -{ - int x, y, v, i, xsize, ysize; - ImgReSampleContext *s; - float fact, factors[] = { 1/2.0, 3.0/4.0, 1.0, 4.0/3.0, 16.0/9.0, 2.0 }; - char buf[256]; - - /* build test image */ - for(y=0;y<YSIZE;y++) { - for(x=0;x<XSIZE;x++) { - if (x < XSIZE/2 && y < YSIZE/2) { - if (x < XSIZE/4 && y < YSIZE/4) { - if ((x % 10) <= 6 && - (y % 10) <= 6) - v = 0xff; - else - v = 0x00; - } else if (x < XSIZE/4) { - if (x & 1) - v = 0xff; - else - v = 0; - } else if (y < XSIZE/4) { - if (y & 1) - v = 0xff; - else - v = 0; - } else { - if (y < YSIZE*3/8) { - if ((y+x) & 1) - v = 0xff; - else - v = 0; - } else { - if (((x+3) % 4) <= 1 && - ((y+3) % 4) <= 1) - v = 0xff; - else - v = 0x00; - } - } - } else if (x < XSIZE/2) { - v = ((x - (XSIZE/2)) * 255) / (XSIZE/2); - } else if (y < XSIZE/2) { - v = ((y - (XSIZE/2)) * 255) / (XSIZE/2); - } else { - v = ((x + y - XSIZE) * 255) / XSIZE; - } - img[(YSIZE - y) * XSIZE + (XSIZE - x)] = v; - } - } - save_pgm("/tmp/in.pgm", img, XSIZE, YSIZE); - for(i=0;i<FF_ARRAY_ELEMS(factors);i++) { - fact = factors[i]; - xsize = (int)(XSIZE * fact); - ysize = (int)((YSIZE - 100) * fact); - s = img_resample_full_init(xsize, ysize, XSIZE, YSIZE, 50 ,50, 0, 0, 0, 0, 0, 0); - av_log(NULL, AV_LOG_INFO, "Factor=%0.2f\n", fact); - dump_filter(&s->h_filters[0][0]); - component_resample(s, img1, xsize, xsize, ysize, - img + 50 * XSIZE, XSIZE, XSIZE, YSIZE - 100); - img_resample_close(s); - - snprintf(buf, sizeof(buf), "/tmp/out%d.pgm", i); - save_pgm(buf, img1, xsize, ysize); - } - - /* mmx test */ -#if HAVE_MMX - av_log(NULL, AV_LOG_INFO, "MMX test\n"); - fact = 0.72; - xsize = (int)(XSIZE * fact); - ysize = (int)(YSIZE * fact); - mm_flags = FF_MM_MMX; - s = img_resample_init(xsize, ysize, XSIZE, YSIZE); - component_resample(s, img1, xsize, xsize, ysize, - img, XSIZE, XSIZE, YSIZE); - - mm_flags = 0; - s = img_resample_init(xsize, ysize, XSIZE, YSIZE); - component_resample(s, img2, xsize, xsize, ysize, - img, XSIZE, XSIZE, YSIZE); - if (memcmp(img1, img2, xsize * ysize) != 0) { - av_log(NULL, AV_LOG_ERROR, "mmx error\n"); - exit(1); - } - av_log(NULL, AV_LOG_INFO, "MMX OK\n"); -#endif /* HAVE_MMX */ - return 0; -} - -#endif /* TEST */
--- a/ppc/imgresample_altivec.c Tue Mar 03 15:48:52 2009 +0000 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,142 +0,0 @@ -/* - * High quality image resampling with polyphase filters - * Copyright (c) 2001 Fabrice Bellard - * - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -/** - * @file libavcodec/ppc/imgresample_altivec.c - * High quality image resampling with polyphase filters - AltiVec bits - */ - -#include "util_altivec.h" -#define FILTER_BITS 8 - -typedef union { - vector signed short v; - signed short s[8]; -} vec_ss; - -void v_resample16_altivec(uint8_t *dst, int dst_width, const uint8_t *src, - int wrap, int16_t *filter) -{ - int sum, i; - const uint8_t *s; - vector unsigned char *tv, tmp, dstv, zero; - vec_ss srchv[4], srclv[4], fv[4]; - vector signed short zeros, sumhv, sumlv; - s = src; - - for(i=0;i<4;i++) { - /* - The vec_madds later on does an implicit >>15 on the result. - Since FILTER_BITS is 8, and we have 15 bits of magnitude in - a signed short, we have just enough bits to pre-shift our - filter constants <<7 to compensate for vec_madds. - */ - fv[i].s[0] = filter[i] << (15-FILTER_BITS); - fv[i].v = vec_splat(fv[i].v, 0); - } - - zero = vec_splat_u8(0); - zeros = vec_splat_s16(0); - - - /* - When we're resampling, we'd ideally like both our input buffers, - and output buffers to be 16-byte aligned, so we can do both aligned - reads and writes. Sadly we can't always have this at the moment, so - we opt for aligned writes, as unaligned writes have a huge overhead. - To do this, do enough scalar resamples to get dst 16-byte aligned. - */ - i = (-(int)dst) & 0xf; - while(i>0) { - sum = s[0 * wrap] * filter[0] + - s[1 * wrap] * filter[1] + - s[2 * wrap] * filter[2] + - s[3 * wrap] * filter[3]; - sum = sum >> FILTER_BITS; - if (sum<0) sum = 0; else if (sum>255) sum=255; - dst[0] = sum; - dst++; - s++; - dst_width--; - i--; - } - - /* Do our altivec resampling on 16 pixels at once. */ - while(dst_width>=16) { - /* Read 16 (potentially unaligned) bytes from each of - 4 lines into 4 vectors, and split them into shorts. - Interleave the multipy/accumulate for the resample - filter with the loads to hide the 3 cycle latency - the vec_madds have. */ - tv = (vector unsigned char *) &s[0 * wrap]; - tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[i * wrap])); - srchv[0].v = (vector signed short) vec_mergeh(zero, tmp); - srclv[0].v = (vector signed short) vec_mergel(zero, tmp); - sumhv = vec_madds(srchv[0].v, fv[0].v, zeros); - sumlv = vec_madds(srclv[0].v, fv[0].v, zeros); - - tv = (vector unsigned char *) &s[1 * wrap]; - tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[1 * wrap])); - srchv[1].v = (vector signed short) vec_mergeh(zero, tmp); - srclv[1].v = (vector signed short) vec_mergel(zero, tmp); - sumhv = vec_madds(srchv[1].v, fv[1].v, sumhv); - sumlv = vec_madds(srclv[1].v, fv[1].v, sumlv); - - tv = (vector unsigned char *) &s[2 * wrap]; - tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[2 * wrap])); - srchv[2].v = (vector signed short) vec_mergeh(zero, tmp); - srclv[2].v = (vector signed short) vec_mergel(zero, tmp); - sumhv = vec_madds(srchv[2].v, fv[2].v, sumhv); - sumlv = vec_madds(srclv[2].v, fv[2].v, sumlv); - - tv = (vector unsigned char *) &s[3 * wrap]; - tmp = vec_perm(tv[0], tv[1], vec_lvsl(0, &s[3 * wrap])); - srchv[3].v = (vector signed short) vec_mergeh(zero, tmp); - srclv[3].v = (vector signed short) vec_mergel(zero, tmp); - sumhv = vec_madds(srchv[3].v, fv[3].v, sumhv); - sumlv = vec_madds(srclv[3].v, fv[3].v, sumlv); - - /* Pack the results into our destination vector, - and do an aligned write of that back to memory. */ - dstv = vec_packsu(sumhv, sumlv) ; - vec_st(dstv, 0, (vector unsigned char *) dst); - - dst+=16; - s+=16; - dst_width-=16; - } - - /* If there are any leftover pixels, resample them - with the slow scalar method. */ - while(dst_width>0) { - sum = s[0 * wrap] * filter[0] + - s[1 * wrap] * filter[1] + - s[2 * wrap] * filter[2] + - s[3 * wrap] * filter[3]; - sum = sum >> FILTER_BITS; - if (sum<0) sum = 0; else if (sum>255) sum=255; - dst[0] = sum; - dst++; - s++; - dst_width--; - } -} -
--- a/ppc/imgresample_altivec.h Tue Mar 03 15:48:52 2009 +0000 +++ /dev/null Thu Jan 01 00:00:00 1970 +0000 @@ -1,26 +0,0 @@ -/* - * This file is part of FFmpeg. - * - * FFmpeg is free software; you can redistribute it and/or - * modify it under the terms of the GNU Lesser General Public - * License as published by the Free Software Foundation; either - * version 2.1 of the License, or (at your option) any later version. - * - * FFmpeg is distributed in the hope that it will be useful, - * but WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * Lesser General Public License for more details. - * - * You should have received a copy of the GNU Lesser General Public - * License along with FFmpeg; if not, write to the Free Software - * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA - */ - -#ifndef AVCODEC_PPC_IMGRESAMPLE_ALTIVEC_H -#define AVCODEC_PPC_IMGRESAMPLE_ALTIVEC_H - -#include <stdint.h> - -void v_resample16_altivec(uint8_t *dst, int dst_width, const uint8_t *src, - int wrap, int16_t *filter); -#endif /* AVCODEC_PPC_IMGRESAMPLE_ALTIVEC_H */