* Boston, MA 02111-1307, USA.
*/
+#include "config.h"
#include <math.h>
#include <stdlib.h>
#include <gst/gst.h>
#include <gstcolorspace.h>
+#ifdef HAVE_LIBMMX
+#include "mmx.h"
+#endif
+
#include "yuv2rgb.h"
+static GstBuffer *gst_colorspace_yuv422P_to_rgb32(GstBuffer *src, GstColorSpaceParameters *params);
+static GstBuffer *gst_colorspace_yuv422P_to_bgr32(GstBuffer *src, GstColorSpaceParameters *params);
+static GstBuffer *gst_colorspace_yuv422P_to_bgr32_mmx(GstBuffer *src, GstColorSpaceParameters *params);
static GstBuffer *gst_colorspace_yuv422P_to_rgb24(GstBuffer *src, GstColorSpaceParameters *params);
+static GstBuffer *gst_colorspace_yuv422P_to_bgr24(GstBuffer *src, GstColorSpaceParameters *params);
static GstBuffer *gst_colorspace_yuv422P_to_rgb16(GstBuffer *src, GstColorSpaceParameters *params);
+static GstBuffer *gst_colorspace_yuv422P_to_bgr16_mmx(GstBuffer *src, GstColorSpaceParameters *params);
static void gst_colorspace_yuv_to_rgb16(GstColorSpaceYUVTables *tables,
unsigned char *lum,
unsigned char *cb,
unsigned char *out,
int cols, int rows);
+static void gst_colorspace_yuv_to_rgb24(GstColorSpaceYUVTables *tables,
+ unsigned char *lum,
+ unsigned char *cr,
+ unsigned char *cb,
+ unsigned char *out,
+ int cols, int rows);
+static void gst_colorspace_yuv_to_rgb32(GstColorSpaceYUVTables *tables,
+ unsigned char *lum,
+ unsigned char *cr,
+ unsigned char *cb,
+ unsigned char *out,
+ int cols, int rows);
+#ifdef HAVE_LIBMMX
+static void gst_colorspace_yuv_to_bgr32_mmx(GstColorSpaceYUVTables *tables,
+ unsigned char *lum,
+ unsigned char *cr,
+ unsigned char *cb,
+ unsigned char *out,
+ int cols, int rows);
+static void gst_colorspace_yuv_to_bgr16_mmx(GstColorSpaceYUVTables *tables,
+ unsigned char *lum,
+ unsigned char *cr,
+ unsigned char *cb,
+ unsigned char *out,
+ int cols, int rows);
+#endif
static GstColorSpaceYUVTables * gst_colorspace_init_yuv(long depth,
long red_mask, long green_mask, long blue_mask);
GstColorSpaceConverter gst_colorspace_yuv2rgb_get_converter(GstColorSpace src, GstColorSpace dest) {
- DEBUG("gst_colorspace_yuv2rgb_get_converter\n");
+ DEBUG("gst_colorspace_yuv2rgb_get_converter %d\n", dest);
switch(src) {
case GST_COLORSPACE_YUV422P:
switch(dest) {
+ case GST_COLORSPACE_BGR32:
+ //return gst_colorspace_yuv422P_to_bgr32;
+ return gst_colorspace_yuv422P_to_bgr32_mmx;
+ case GST_COLORSPACE_RGB32:
+ return gst_colorspace_yuv422P_to_rgb32;
case GST_COLORSPACE_RGB24:
return gst_colorspace_yuv422P_to_rgb24;
+ case GST_COLORSPACE_BGR24:
+ return gst_colorspace_yuv422P_to_bgr24;
case GST_COLORSPACE_RGB555:
case GST_COLORSPACE_RGB565:
+ case GST_COLORSPACE_BGR555:
return gst_colorspace_yuv422P_to_rgb16;
+ case GST_COLORSPACE_BGR565:
+ return gst_colorspace_yuv422P_to_bgr16_mmx;
default:
break;
}
default:
break;
}
+ g_print("gst_colorspace_yuv2rgb not implemented\n");
return NULL;
}
+static GstBuffer *gst_colorspace_yuv422P_to_bgr32(GstBuffer *src, GstColorSpaceParameters *params) {
+ static GstColorSpaceYUVTables *color_tables = NULL;
+ int size;
+ GstBuffer *buf = NULL;
+ guchar *out;
+ DEBUG("gst_colorspace_yuv422P_to_bgr32\n");
+
+ g_return_val_if_fail(params != NULL, NULL);
+
+ if (color_tables == NULL) {
+ color_tables = gst_colorspace_init_yuv(32, 0xFF0000, 0x00FF00, 0x0000FF);
+ }
+ size = params->width * params->height;
+ if (params->outbuf == NULL) {
+ buf = gst_buffer_new();
+ out = GST_BUFFER_DATA(buf) = g_malloc(size * 4);
+ GST_BUFFER_SIZE(buf) = size * 4;
+ }
+ else out = params->outbuf;
+
+ gst_colorspace_yuv_to_rgb32(color_tables,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ out,
+ params->height,
+ params->width);
+
+ if (buf) {
+ gst_buffer_unref(src);
+ return buf;
+ }
+ else return src;
+}
+
+static GstBuffer *gst_colorspace_yuv422P_to_rgb32(GstBuffer *src, GstColorSpaceParameters *params) {
+ static GstColorSpaceYUVTables *color_tables = NULL;
+ int size;
+ GstBuffer *buf = NULL;
+ guchar *out;
+ DEBUG("gst_colorspace_yuv422P_to_rgb32\n");
+
+ g_return_val_if_fail(params != NULL, NULL);
+
+ if (color_tables == NULL) {
+ color_tables = gst_colorspace_init_yuv(32, 0x0000FF, 0x00FF00, 0xFF0000);
+ }
+ size = params->width * params->height;
+ if (params->outbuf == NULL) {
+ buf = gst_buffer_new();
+ out = GST_BUFFER_DATA(buf) = g_malloc(size * 4);
+ GST_BUFFER_SIZE(buf) = size * 4;
+ }
+ else out = params->outbuf;
+
+ gst_colorspace_yuv_to_rgb32(color_tables,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ out,
+ params->height,
+ params->width);
+
+ if (buf) {
+ gst_buffer_unref(src);
+ return buf;
+ }
+ else return src;
+}
+
+static GstBuffer *gst_colorspace_yuv422P_to_bgr24(GstBuffer *src, GstColorSpaceParameters *params) {
+ static GstColorSpaceYUVTables *color_tables = NULL;
+ int size;
+ GstBuffer *buf = NULL;
+ guchar *out;
+ DEBUG("gst_colorspace_yuv422P_to_bgr24\n");
+
+ g_return_val_if_fail(params != NULL, NULL);
+
+ if (color_tables == NULL) {
+ color_tables = gst_colorspace_init_yuv(24, 0xFF0000, 0x00FF00, 0x0000FF);
+ }
+ size = params->width * params->height;
+ if (params->outbuf == NULL) {
+ buf = gst_buffer_new();
+ out = GST_BUFFER_DATA(buf) = g_malloc(size * 3);
+ GST_BUFFER_SIZE(buf) = size * 3;
+ }
+ else out = params->outbuf;
+
+ gst_colorspace_yuv_to_rgb24(color_tables,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ out,
+ params->height,
+ params->width);
+ if (buf) {
+ gst_buffer_unref(src);
+ return buf;
+ }
+ else return src;
+}
+
static GstBuffer *gst_colorspace_yuv422P_to_rgb24(GstBuffer *src, GstColorSpaceParameters *params) {
+ static GstColorSpaceYUVTables *color_tables = NULL;
+ int size;
+ GstBuffer *buf = NULL;
+ guchar *out;
DEBUG("gst_colorspace_yuv422P_to_rgb24\n");
- return src;
+ g_return_val_if_fail(params != NULL, NULL);
+
+ if (color_tables == NULL) {
+ color_tables = gst_colorspace_init_yuv(24, 0x0000FF, 0x00FF00, 0xFF0000);
+ }
+ size = params->width * params->height;
+ if (params->outbuf == NULL) {
+ buf = gst_buffer_new();
+ out = GST_BUFFER_DATA(buf) = g_malloc(size * 3);
+ GST_BUFFER_SIZE(buf) = size * 3;
+ }
+ else out = params->outbuf;
+
+ gst_colorspace_yuv_to_rgb24(color_tables,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ out,
+ params->height,
+ params->width);
+
+ if (buf) {
+ gst_buffer_unref(src);
+ return buf;
+ }
+ else return src;
}
static GstBuffer *gst_colorspace_yuv422P_to_rgb16(GstBuffer *src, GstColorSpaceParameters *params) {
static GstColorSpaceYUVTables *color_tables = NULL;
+ int size;
DEBUG("gst_colorspace_yuv422P_to_rgb16\n");
g_return_val_if_fail(params != NULL, NULL);
+ g_return_val_if_fail(params->visual != NULL, NULL);
if (color_tables == NULL) {
- color_tables = gst_colorspace_init_yuv(16, 0xF800, 0x07E0, 0x001F);
+ color_tables = gst_colorspace_init_yuv(16, params->visual->red_mask, params->visual->green_mask, params->visual->blue_mask);
}
+ size = params->width * params->height;
gst_colorspace_yuv_to_rgb16(color_tables,
- GST_BUFFER_DATA(src), // Y component
- GST_BUFFER_DATA(src)+params->width*params->height, // cr component
- GST_BUFFER_DATA(src)+params->width*params->height+
- (params->width*params->height)/4, // cb component
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
params->outbuf,
params->height,
params->width);
return src;
}
+#ifdef HAVE_LIBMMX
+static mmx_t MMX16_redmask = (mmx_t)(long long)0xf800f800f800f800LL; //dd 07c00 7c00h, 07c007c00h
+static mmx_t MMX16_grnmask = (mmx_t)(long long)0x07e007e007e007e0LL; //dd 003e0 03e0h, 003e003e0h
+
+static GstBuffer *gst_colorspace_yuv422P_to_bgr32_mmx(GstBuffer *src, GstColorSpaceParameters *params) {
+ int size;
+ GstBuffer *buf = NULL;
+ guchar *out;
+ DEBUG("gst_colorspace_yuv422P_to_rgb32_mmx\n");
+
+ g_return_val_if_fail(params != NULL, NULL);
+
+ size = params->width * params->height;
+ if (params->outbuf == NULL) {
+ buf = gst_buffer_new();
+ out = GST_BUFFER_DATA(buf) = g_malloc(size * 4);
+ GST_BUFFER_SIZE(buf) = size * 4;
+ }
+ else out = params->outbuf;
+
+ gst_colorspace_yuv_to_bgr32_mmx(NULL,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ out,
+ params->height,
+ params->width);
+
+ if (buf) {
+ gst_buffer_unref(src);
+ return buf;
+ }
+ else return src;
+}
+static GstBuffer *gst_colorspace_yuv422P_to_bgr16_mmx(GstBuffer *src, GstColorSpaceParameters *params) {
+ int size;
+ DEBUG("gst_colorspace_yuv422P_to_bgr16_mmx \n");
+
+ g_return_val_if_fail(params != NULL, NULL);
+
+ size = params->width * params->height;
+
+ gst_colorspace_yuv_to_bgr16_mmx(NULL,
+ GST_BUFFER_DATA(src), // Y component
+ GST_BUFFER_DATA(src)+size, // cr component
+ GST_BUFFER_DATA(src)+size+(size>>2), // cb component
+ params->outbuf,
+ params->height,
+ params->width);
+
+ return src;
+}
+#endif
/*
* How many 1 bits are there in the longword.
* A similar optimisation for Alpha for 64 bit has been
* prepared for, but is not yet implemented.
*/
- if(!(depth == 32)) {
+ if(!(depth == 32) && !(depth == 24)) {
r_2_pix_alloc[i + 256] |= (r_2_pix_alloc[i + 256]) << 16;
g_2_pix_alloc[i + 256] |= (g_2_pix_alloc[i + 256]) << 16;
unsigned char *lum2;
int x, y;
int cr_r;
- int cr_g;
- int cb_g;
+ int crb_g;
int cb_b;
- int cols_2 = cols/2;
+ int cols_2 = cols>>1;
row1 = (unsigned short *)out;
- row2 = row1 + cols_2 + cols_2;
- lum2 = lum + cols_2 + cols_2;
+ row2 = row1 + cols;
+ lum2 = lum + cols;
- for (y=0; y<rows; y+=2) {
- for (x=0; x<cols_2; x++) {
- int R, G, B;
+ for (y=rows>>1; y; y--) {
+ for (x=cols_2; x; x--) {
CR = *cr++;
CB = *cb++;
cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
+ crb_g = tables->Cr_g_tab[CR] + tables->Cb_g_tab[CB];
cb_b = tables->Cb_b_tab[CB];
L = tables->L_tab[(int) *lum++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
-
- *row1++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
-
-#ifdef INTERPOLATE
- if(x != cols_2 - 1) {
- CR = (CR + *cr) >> 1;
- CB = (CB + *cb) >> 1;
- cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
- cb_b = tables->Cb_b_tab[CB];
- }
-#endif
+ *row1++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
L = tables->L_tab[(int) *lum++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
-
- *row1++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row1++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
/*
* Now, do second row.
*/
-#ifdef INTERPOLATE
- if(y != rows - 2) {
- CR = (CR + *(cr + cols_2 - 1)) >> 1;
- CB = (CB + *(cb + cols_2 - 1)) >> 1;
- cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
- cb_b = tables->Cb_b_tab[CB];
- }
-#endif
-
L = tables->L_tab[(int) *lum2++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
- *row2++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row2++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
L = tables->L_tab[(int) *lum2++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
- *row2++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row2++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
}
/*
* These values are at the start of the next line, (due
* to the ++'s above),but they need to be at the start
* of the line after that.
*/
- lum += cols_2 + cols_2;
- lum2 += cols_2 + cols_2;
- row1 += cols_2 + cols_2;
- row2 += cols_2 + cols_2;
+ lum = lum2;
+ row1 = row2;
+ lum2 += cols;
+ row2 += cols;
+ }
+}
+
+static void
+gst_colorspace_yuv_to_rgb24(tables, lum, cr, cb, out, rows, cols)
+ GstColorSpaceYUVTables *tables;
+ unsigned char *lum;
+ unsigned char *cr;
+ unsigned char *cb;
+ unsigned char *out;
+ int cols, rows;
+
+{
+ int L, CR, CB;
+ unsigned char *row1, *row2;
+ unsigned char *lum2;
+ int x, y;
+ int cr_r;
+ int crb_g;
+ int cb_b;
+ int cols_2 = cols>>1;
+ int cols_3 = cols*3;
+ unsigned char pixels[4];
+
+ row1 = out;
+ row2 = row1 + cols_3;
+ lum2 = lum + cols;
+ for (y=rows>>1; y; y--) {
+ for (x=cols_2; x; x--) {
+
+ CR = *cr++;
+ CB = *cb++;
+ cr_r = tables->Cr_r_tab[CR];
+ crb_g = tables->Cr_g_tab[CR] + tables->Cb_g_tab[CB];
+ cb_b = tables->Cb_b_tab[CB];
+
+ L = tables->L_tab[(int) *lum++];
+
+ ((int *)pixels)[0] = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
+ *row1++ = pixels[0]; *row1++ = pixels[1]; *row1++ = pixels[2];
+
+ L = tables->L_tab[(int) *lum++];
+
+ ((int *)pixels)[0] = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
+ *row1++ = pixels[0]; *row1++ = pixels[1]; *row1++ = pixels[2];
+
+ /*
+ * Now, do second row.
+ */
+
+ L = tables->L_tab [(int) *lum2++];
+
+ ((int *)pixels)[0] = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
+ *row2++ = pixels[0]; *row2++ = pixels[1]; *row2++ = pixels[2];
+
+ L = tables->L_tab [(int) *lum2++];
+
+ ((int *)pixels)[0] = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
+ *row2++ = pixels[0]; *row2++ = pixels[1]; *row2++ = pixels[2];
+ }
+ lum = lum2;
+ row1 = row2;
+ lum2 += cols;
+ row2 += cols_3;
}
}
unsigned char *lum2;
int x, y;
int cr_r;
- int cr_g;
- int cb_g;
+ int crb_g;
int cb_b;
- int cols_2 = cols / 2;
+ int cols_2 = cols>>1;
- row1 = (unsigned int *)out;
- row2 = row1 + cols_2 + cols_2;
- lum2 = lum + cols_2 + cols_2;
- for (y=0; y<rows; y+=2) {
- for (x=0; x<cols_2; x++) {
- int R, G, B;
+ row1 = (guint32 *)out;
+ row2 = row1 + cols;
+ lum2 = lum + cols;
+ for (y=rows>>1; y; y--) {
+ for (x=cols_2; x; x--) {
CR = *cr++;
CB = *cb++;
cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
+ crb_g = tables->Cr_g_tab[CR] + tables->Cb_g_tab[CB];
cb_b = tables->Cb_b_tab[CB];
L = tables->L_tab[(int) *lum++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
-
- *row1++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
-
-#ifdef INTERPOLATE
- if(x != cols_2 - 1) {
- CR = (CR + *cr) >> 1;
- CB = (CB + *cb) >> 1;
- cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
- cb_b = tables->Cb_b_tab[CB];
- }
-#endif
+ *row1++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
L = tables->L_tab[(int) *lum++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
-
- *row1++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row1++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
/*
* Now, do second row.
*/
-#ifdef INTERPOLATE
- if(y != rows - 2) {
- CR = (CR + *(cr + cols_2 - 1)) >> 1;
- CB = (CB + *(cb + cols_2 - 1)) >> 1;
- cr_r = tables->Cr_r_tab[CR];
- cr_g = tables->Cr_g_tab[CR];
- cb_g = tables->Cb_g_tab[CB];
- cb_b = tables->Cb_b_tab[CB];
- }
-#endif
-
L = tables->L_tab [(int) *lum2++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
- *row2++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row2++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
L = tables->L_tab [(int) *lum2++];
- R = L + cr_r;
- G = L + cr_g + cb_g;
- B = L + cb_b;
- *row2++ = (tables->r_2_pix[R] | tables->g_2_pix[G] | tables->b_2_pix[B]);
+ *row2++ = (tables->r_2_pix[L+cr_r] | tables->g_2_pix[L+crb_g] | tables->b_2_pix[L+cb_b]);
}
- lum += cols_2 + cols_2;
- lum2 += cols_2 + cols_2;
- row1 += cols_2 + cols_2;
- row2 += cols_2 + cols_2;
+ lum = lum2;
+ row1 = row2;
+ lum2 += cols;
+ row2 += cols;
}
}
+#ifdef HAVE_LIBMMX
+
+static mmx_t MMX_80w = (mmx_t)(long long)0x0080008000800080LL; //dd 00080 0080h, 000800080h
+
+static mmx_t MMX_00FFw = (mmx_t)(long long)0x00ff00ff00ff00ffLL; //dd 000FF 00FFh, 000FF00FFh
+static mmx_t MMX_FF00w = (mmx_t)(long long)0xff00ff00ff00ff00LL; //dd 000FF 00FFh, 000FF00FFh
+
+static mmx_t MMX16_Vredcoeff = (mmx_t)(long long)0x0066006600660066LL; //dd 00066 0066h, 000660066h
+static mmx_t MMX16_Ublucoeff = (mmx_t)(long long)0x0081008100810081LL; //dd 00081 0081h, 000810081h
+static mmx_t MMX16_Ugrncoeff = (mmx_t)(long long)0xffe8ffe8ffe8ffe8LL; //dd 0FFE7 FFE7h, 0FFE7FFE7h
+static mmx_t MMX16_Vgrncoeff = (mmx_t)(long long)0xffcdffcdffcdffcdLL; //dd 0FFCC FFCCh, 0FFCCFFCCh
+
+static mmx_t MMX16_Ycoeff = (mmx_t)(long long)0x004a004a004a004aLL; //dd 0004A 004Ah, 0004A004Ah
+
+
+static mmx_t MMX32_Vredcoeff = (mmx_t)(long long)0x0059005900590059LL;
+static mmx_t MMX32_Ubluecoeff = (mmx_t)(long long)0x0072007200720072LL;
+static mmx_t MMX32_Ugrncoeff = (mmx_t)(long long)0xffeaffeaffeaffeaLL;
+static mmx_t MMX32_Vgrncoeff = (mmx_t)(long long)0xffd2ffd2ffd2ffd2LL;
+
+static void
+gst_colorspace_yuv_to_bgr16_mmx(tables, lum, cr, cb, out, rows, cols)
+ GstColorSpaceYUVTables *tables;
+ unsigned char *lum;
+ unsigned char *cr;
+ unsigned char *cb;
+ unsigned char *out;
+ int cols, rows;
+
+{
+ unsigned short *row1 = (unsigned short* )out; // 32 bit target
+ int cols8 = cols>>3;
+
+ int y, x;
+
+ for (y=rows>>1; y; y--) {
+ for (x=cols8; x; x--) {
+
+ movd_m2r(*(mmx_t *)cb, mm0); // 4 Cb 0 0 0 0 u3 u2 u1 u0
+ pxor_r2r(mm7, mm7);
+ movd_m2r(*(mmx_t *)cr, mm1); // 4 Cr 0 0 0 0 v3 v2 v1 v0
+ punpcklbw_r2r(mm7, mm0); // 4 W cb 0 u3 0 u2 0 u1 0 u0
+ punpcklbw_r2r(mm7, mm1); // 4 W cr 0 v3 0 v2 0 v1 0 v0
+ psubw_m2r(MMX_80w, mm0);
+ psubw_m2r(MMX_80w, mm1);
+ movq_r2r(mm0, mm2); // Cb 0 u3 0 u2 0 u1 0 u0
+ movq_r2r(mm1, mm3); // Cr
+ pmullw_m2r(MMX16_Ugrncoeff, mm2); // Cb2green 0 R3 0 R2 0 R1 0 R0
+ movq_m2r(*(mmx_t *)lum, mm6); // L1 l7 L6 L5 L4 L3 L2 L1 L0
+ pmullw_m2r(MMX16_Ublucoeff, mm0); // Cb2blue
+ pand_m2r(MMX_00FFw, mm6); // L1 00 L6 00 L4 00 L2 00 L0
+ pmullw_m2r(MMX16_Vgrncoeff, mm3); // Cr2green
+ movq_m2r(*(mmx_t *)lum, mm7); // L2
+ pmullw_m2r(MMX16_Vredcoeff, mm1); // Cr2red
+ // "psubw MMX_10w, %%mm6\n"
+ psrlw_i2r(8, mm7); // L2 00 L7 00 L5 00 L3 00 L1
+ pmullw_m2r(MMX16_Ycoeff, mm6); // lum1
+ // "psubw MMX_10w, %%mm7\n" // L2
+ paddw_r2r(mm3, mm2); // Cb2green + Cr2green == green
+ pmullw_m2r(MMX16_Ycoeff, mm7); // lum2
+
+ movq_r2r(mm6, mm4); // lum1
+ paddw_r2r(mm0, mm6); // lum1 +blue 00 B6 00 B4 00 B2 00 B0
+ movq_r2r(mm4, mm5); // lum1
+ paddw_r2r(mm1, mm4); // lum1 +red 00 R6 00 R4 00 R2 00 R0
+ paddw_r2r(mm2, mm5); // lum1 +green 00 G6 00 G4 00 G2 00 G0
+ psraw_i2r(6, mm4); // R1 0 .. 64
+ movq_r2r(mm7, mm3); // lum2 00 L7 00 L5 00 L3 00 L1
+ psraw_i2r(6, mm5); // G1 - .. +
+ paddw_r2r(mm0, mm7); // Lum2 +blue 00 B7 00 B5 00 B3 00 B1
+ psraw_i2r(6, mm6); // B1 0 .. 64
+ packuswb_r2r(mm4, mm4); // R1 R1
+ packuswb_r2r(mm5, mm5); // G1 G1
+ packuswb_r2r(mm6, mm6); // B1 B1
+ punpcklbw_r2r(mm4, mm4);
+ punpcklbw_r2r(mm5, mm5);
+
+ pand_m2r(MMX16_redmask, mm4);
+ psllw_i2r(3, mm5); // GREEN 1
+ punpcklbw_r2r(mm6, mm6);
+ pand_m2r(MMX16_grnmask, mm5);
+ pand_m2r(MMX16_redmask, mm6);
+ por_r2r(mm5, mm4); //
+ psrlw_i2r(11, mm6); // BLUE 1
+ movq_r2r(mm3, mm5); // lum2
+ paddw_r2r(mm1, mm3); // lum2 +red 00 R7 00 R5 00 R3 00 R1
+ paddw_r2r(mm2, mm5); // lum2 +green 00 G7 00 G5 00 G3 00 G1
+ psraw_i2r(6, mm3); // R2
+ por_r2r(mm6, mm4); // MM4
+ psraw_i2r(6, mm5); // G2
+ movq_m2r(*(mmx_t *)(lum+cols), mm6); // L3 load lum2
+ psraw_i2r(6, mm7);
+ packuswb_r2r(mm3, mm3);
+ packuswb_r2r(mm5, mm5);
+ packuswb_r2r(mm7, mm7);
+ pand_m2r(MMX_00FFw, mm6); // L3
+ punpcklbw_r2r(mm3, mm3);
+ // "psubw MMX_10w, %%mm6\n" // L3
+ punpcklbw_r2r(mm5, mm5);
+ pmullw_m2r(MMX16_Ycoeff, mm6); // lum3
+ punpcklbw_r2r(mm7, mm7);
+ psllw_i2r(3, mm5); // GREEN 2
+ pand_m2r(MMX16_redmask, mm7);
+ pand_m2r(MMX16_redmask, mm3);
+ psrlw_i2r(11, mm7); // BLUE 2
+ pand_m2r(MMX16_grnmask, mm5);
+ por_r2r(mm7, mm3);
+ movq_m2r(*(mmx_t *)(lum+cols), mm7); // L4 load lum2
+ por_r2r(mm5, mm3); //
+ psrlw_i2r(8, mm7); // L4
+ movq_r2r(mm4, mm5);
+ // "psubw MMX_10w, %%mm7\n" // L4
+ punpcklwd_r2r(mm3, mm4);
+ pmullw_m2r(MMX16_Ycoeff, mm7); // lum4
+ punpckhwd_r2r(mm3, mm5);
+
+ movq_r2m(mm4, *(row1)); // write row1
+ movq_r2m(mm5, *(row1+4)); // write row1
+
+ movq_r2r(mm6, mm4); // Lum3
+ paddw_r2r(mm0, mm6); // Lum3 +blue
+
+ movq_r2r(mm4, mm5); // Lum3
+ paddw_r2r(mm1, mm4); // Lum3 +red
+ paddw_r2r(mm2, mm5); // Lum3 +green
+ psraw_i2r(6, mm4);
+ movq_r2r(mm7, mm3); // Lum4
+ psraw_i2r(6, mm5);
+ paddw_r2r(mm0, mm7); // Lum4 +blue
+ psraw_i2r(6, mm6); // Lum3 +blue
+ movq_r2r(mm3, mm0); // Lum4
+ packuswb_r2r(mm4, mm4);
+ paddw_r2r(mm1, mm3); // Lum4 +red
+ packuswb_r2r(mm5, mm5);
+ paddw_r2r(mm2, mm0); // Lum4 +green
+ packuswb_r2r(mm6, mm6);
+ punpcklbw_r2r(mm4, mm4);
+ punpcklbw_r2r(mm5, mm5);
+ punpcklbw_r2r(mm6, mm6);
+ psllw_i2r(3, mm5); // GREEN 3
+ pand_m2r(MMX16_redmask, mm4);
+ psraw_i2r(6, mm3); // psr 6
+ psraw_i2r(6, mm0);
+ pand_m2r(MMX16_redmask, mm6); // BLUE
+ pand_m2r(MMX16_grnmask, mm5);
+ psrlw_i2r(11, mm6); // BLUE 3
+ por_r2r(mm5, mm4);
+ psraw_i2r(6, mm7);
+ por_r2r(mm6, mm4);
+ packuswb_r2r(mm3, mm3);
+ packuswb_r2r(mm0, mm0);
+ packuswb_r2r(mm7, mm7);
+ punpcklbw_r2r(mm3, mm3);
+ punpcklbw_r2r(mm0, mm0);
+ punpcklbw_r2r(mm7, mm7);
+ pand_m2r(MMX16_redmask, mm3);
+ pand_m2r(MMX16_redmask, mm7); // BLUE
+ psllw_i2r(3, mm0); // GREEN 4
+ psrlw_i2r(11, mm7);
+ pand_m2r(MMX16_grnmask, mm0);
+ por_r2r(mm7, mm3);
+ por_r2r(mm0, mm3);
+
+ movq_r2r(mm4, mm5);
+
+ punpcklwd_r2r(mm3, mm4);
+ punpckhwd_r2r(mm3, mm5);
+
+ movq_r2m(mm4, *(row1+cols));
+ movq_r2m(mm5, *(row1+cols+4));
+
+ lum+=8;
+ cr+=4;
+ cb+=4;
+ row1 +=8;
+ }
+ lum += cols;
+ row1 += cols;
+ }
+
+ emms();
+
+}
+static void
+gst_colorspace_yuv_to_bgr32_mmx(tables, lum, cr, cb, out, rows, cols)
+ GstColorSpaceYUVTables *tables;
+ unsigned char *lum;
+ unsigned char *cr;
+ unsigned char *cb;
+ unsigned char *out;
+ int cols, rows;
+
+{
+ guint32 *row1 = (guint32 *)out; // 32 bit target
+ int cols4 = cols>>2;
+
+ int y, x;
+
+ for (y=rows>>1; y; y--) {
+ for (x=cols4; x; x--) {
+
+ // create Cr (result in mm1)
+ movd_m2r(*(mmx_t *)cr, mm1); // 0 0 0 0 v3 v2 v1 v0
+ pxor_r2r(mm7, mm7); // 00 00 00 00 00 00 00 00
+ movd_m2r(*(mmx_t *)lum, mm2); // 0 0 0 0 l3 l2 l1 l0
+ punpcklbw_r2r(mm7, mm1); // 0 v3 0 v2 00 v1 00 v0
+ punpckldq_r2r(mm1, mm1); // 00 v1 00 v0 00 v1 00 v0
+ psubw_m2r(MMX_80w, mm1); // mm1-128:r1 r1 r0 r0 r1 r1 r0 r0
+
+ // create Cr_g (result in mm0)
+ movq_r2r(mm1, mm0); // r1 r1 r0 r0 r1 r1 r0 r0
+ pmullw_m2r(MMX32_Vgrncoeff, mm0); // red*-46dec=0.7136*64
+ pmullw_m2r(MMX32_Vredcoeff, mm1); // red*89dec=1.4013*64
+ psraw_i2r(6, mm0); // red=red/64
+ psraw_i2r(6, mm1); // red=red/64
+
+ // create L1 L2 (result in mm2,mm4)
+ // L2=lum+cols
+ movq_m2r(*(mmx_t *)(lum+cols),mm3); // 0 0 0 0 L3 L2 L1 L0
+ punpckldq_r2r(mm3, mm2); // L3 L2 L1 L0 l3 l2 l1 l0
+ movq_r2r(mm2, mm4); // L3 L2 L1 L0 l3 l2 l1 l0
+ pand_m2r(MMX_FF00w, mm2); // L3 0 L1 0 l3 0 l1 0
+ pand_m2r(MMX_00FFw, mm4); // 0 L2 0 L0 0 l2 0 l0
+ psrlw_i2r(8, mm2); // 0 L3 0 L1 0 l3 0 l1
+
+ // create R (result in mm6)
+ movq_r2r(mm2, mm5); // 0 L3 0 L1 0 l3 0 l1
+ movq_r2r(mm4, mm6); // 0 L2 0 L0 0 l2 0 l0
+ paddsw_r2r(mm1, mm5); // lum1+red:x R3 x R1 x r3 x r1
+ paddsw_r2r(mm1, mm6); // lum1+red:x R2 x R0 x r2 x r0
+ packuswb_r2r(mm5, mm5); // R3 R1 r3 r1 R3 R1 r3 r1
+ packuswb_r2r(mm6, mm6); // R2 R0 r2 r0 R2 R0 r2 r0
+ pxor_r2r(mm7, mm7); // 00 00 00 00 00 00 00 00
+ punpcklbw_r2r(mm5, mm6); // R3 R2 R1 R0 r3 r2 r1 r0
+
+ // create Cb (result in mm1)
+ movd_m2r(*(mmx_t *)cb, mm1); // 0 0 0 0 u3 u2 u1 u0
+ punpcklbw_r2r(mm7, mm1); // 0 u3 0 u2 00 u1 00 u0
+ punpckldq_r2r(mm1, mm1); // 00 u1 00 u0 00 u1 00 u0
+ psubw_m2r(MMX_80w, mm1); // mm1-128:u1 u1 u0 u0 u1 u1 u0 u0
+ // create Cb_g (result in mm5)
+ movq_r2r(mm1, mm5); // u1 u1 u0 u0 u1 u1 u0 u0
+ pmullw_m2r(MMX32_Ugrncoeff, mm5); // blue*-109dec=1.7129*64
+ pmullw_m2r(MMX32_Ubluecoeff, mm1); // blue*114dec=1.78125*64
+ psraw_i2r(6, mm5); // blue=red/64
+ psraw_i2r(6, mm1); // blue=blue/64
+
+ // create G (result in mm7)
+ movq_r2r(mm2, mm3); // 0 L3 0 L1 0 l3 0 l1
+ movq_r2r(mm4, mm7); // 0 L2 0 L0 0 l2 0 l1
+ paddsw_r2r(mm5, mm3); // lum1+Cb_g:x G3t x G1t x g3t x g1t
+ paddsw_r2r(mm5, mm7); // lum1+Cb_g:x G2t x G0t x g2t x g0t
+ paddsw_r2r(mm0, mm3); // lum1+Cr_g:x G3 x G1 x g3 x g1
+ paddsw_r2r(mm0, mm7); // lum1+blue:x G2 x G0 x g2 x g0
+ packuswb_r2r(mm3, mm3); // G3 G1 g3 g1 G3 G1 g3 g1
+ packuswb_r2r(mm7, mm7); // G2 G0 g2 g0 G2 G0 g2 g0
+ punpcklbw_r2r(mm3, mm7); // G3 G2 G1 G0 g3 g2 g1 g0
+
+ // create B (result in mm5)
+ movq_r2r(mm2, mm3); // 0 L3 0 L1 0 l3 0 l1
+ movq_r2r(mm4, mm5); // 0 L2 0 L0 0 l2 0 l1
+ paddsw_r2r(mm1, mm3); // lum1+blue:x B3 x B1 x b3 x b1
+ paddsw_r2r(mm1, mm5); // lum1+blue:x B2 x B0 x b2 x b0
+ packuswb_r2r(mm3, mm3); // B3 B1 b3 b1 B3 B1 b3 b1
+ packuswb_r2r(mm5, mm5); // B2 B0 b2 b0 B2 B0 b2 b0
+ punpcklbw_r2r(mm3, mm5); // B3 B2 B1 B0 b3 b2 b1 b0
+
+ // fill destination row1 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
+
+ pxor_r2r(mm2, mm2); // 0 0 0 0 0 0 0 0
+ pxor_r2r(mm4, mm4); // 0 0 0 0 0 0 0 0
+ movq_r2r(mm6, mm1); // R3 R2 R1 R0 r3 r2 r1 r0
+ movq_r2r(mm5, mm3); // B3 B2 B1 B0 b3 b2 b1 b0
+ // process lower lum
+ punpcklbw_r2r(mm4, mm1); // 0 r3 0 r2 0 r1 0 r0
+ punpcklbw_r2r(mm4, mm3); // 0 b3 0 b2 0 b1 0 b0
+ movq_r2r(mm1, mm2); // 0 r3 0 r2 0 r1 0 r0
+ movq_r2r(mm3, mm0); // 0 b3 0 b2 0 b1 0 b0
+ punpcklwd_r2r(mm1, mm3); // 0 r1 0 b1 0 r0 0 b0
+ punpckhwd_r2r(mm2, mm0); // 0 r3 0 b3 0 r2 0 b2
+
+ pxor_r2r(mm2, mm2); // 0 0 0 0 0 0 0 0
+ movq_r2r(mm7, mm1); // G3 G2 G1 G0 g3 g2 g1 g0
+ punpcklbw_r2r(mm1, mm2); // g3 0 g2 0 g1 0 g0 0
+ punpcklwd_r2r(mm4, mm2); // 0 0 g1 0 0 0 g0 0
+ por_r2r(mm3, mm2); // 0 r1 g1 b1 0 r0 g0 b0
+ movq_r2m(mm2, *(mmx_t *)row1); // wrote out ! row1
+
+ pxor_r2r(mm2, mm2); // 0 0 0 0 0 0 0 0
+ punpcklbw_r2r(mm1, mm4); // g3 0 g2 0 g1 0 g0 0
+ punpckhwd_r2r(mm2, mm4); // 0 0 g3 0 0 0 g2 0
+ por_r2r(mm0, mm4); // 0 r3 g3 b3 0 r2 g2 b2
+ movq_r2m(mm4, *(mmx_t *)(row1+2)); // wrote out ! row1
+
+ // fill destination row2 (needed are mm6=Rr,mm7=Gg,mm5=Bb)
+ // this can be done "destructive"
+ pxor_r2r(mm2, mm2); // 0 0 0 0 0 0 0 0
+ punpckhbw_r2r(mm2, mm6); // 0 R3 0 R2 0 R1 0 R0
+ punpckhbw_r2r(mm1, mm5); // G3 B3 G2 B2 G1 B1 G0 B0
+ movq_r2r(mm5, mm1); // G3 B3 G2 B2 G1 B1 G0 B0
+ punpcklwd_r2r(mm6, mm1); // 0 R1 G1 B1 0 R0 G0 B0
+ movq_r2m(mm1, *(mmx_t *)(row1+cols)); // wrote out ! row2
+ punpckhwd_r2r(mm6, mm5); // 0 R3 G3 B3 0 R2 G2 B2
+ movq_r2m(mm5, *(mmx_t *)(row1+cols+2)); // wrote out ! row2
+
+ lum+=4;
+ cr+=2;
+ cb+=2;
+ row1 +=4;
+ }
+ lum += cols;
+ row1 += cols;
+ }
+
+ emms();
+
+}
+#endif
+