Tex vid: rework part 2

- use pPriv->w/h directly for tex coords
- take src x/y offset into account when calculating tex coords
- when copying data, only optimize for y clipping. In order
  to deal with the x clipping optimization, the copy routines
  or tex coords would need to be fixed.  This should fix clipping
  problems with the current code.

4 files changed, 126 insertions, 129 deletions

diff --git a/src/r600_textured_videofuncs.c b/src/r600_textured_videofuncs.c
index 27bdf4d5..23e7f405 100644
--- a/src/r600_textured_videofuncs.c
+++ b/src/r600_textured_videofuncs.c
@@ -349,7 +349,7 @@ R600DisplayTexturedVideo(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	tex_res.format              = FMT_8;
 	tex_res.w                   = pPriv->w >> 1;
 	tex_res.h                   = pPriv->h >> 1;
-	tex_res.pitch               = accel_state->src_pitch[0] >> 1;
+	tex_res.pitch               = ((accel_state->src_pitch[0] >> 1) + 255) & ~255;
 	tex_res.dst_sel_x           = SQ_SEL_X; /* V or U */
 	tex_res.dst_sel_y           = SQ_SEL_1;
 	tex_res.dst_sel_z           = SQ_SEL_1;
@@ -373,7 +373,7 @@ R600DisplayTexturedVideo(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	tex_res.format              = FMT_8;
 	tex_res.w                   = pPriv->w >> 1;
 	tex_res.h                   = pPriv->h >> 1;
-	tex_res.pitch               = accel_state->src_pitch[0] >> 1;
+	tex_res.pitch               = ((accel_state->src_pitch[0] >> 1) + 255) & ~255;
 	tex_res.dst_sel_x           = SQ_SEL_X; /* V or U */
 	tex_res.dst_sel_y           = SQ_SEL_1;
 	tex_res.dst_sel_z           = SQ_SEL_1;
@@ -559,10 +559,12 @@ R600DisplayTexturedVideo(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	dstw = pBox->x2 - pBox->x1;
 	dsth = pBox->y2 - pBox->y1;
 
-	srcX = ((pBox->x1 - pPriv->drw_x) *
-		pPriv->src_w) / pPriv->dst_w;
-	srcY = ((pBox->y1 - pPriv->drw_y) *
-		pPriv->src_h) / pPriv->dst_h;
+	srcX = pPriv->src_x;
+	srcX += ((pBox->x1 - pPriv->drw_x) *
+		 pPriv->src_w) / pPriv->dst_w;
+	srcY = pPriv->src_y;
+	srcY += ((pBox->y1 - pPriv->drw_y) *
+		 pPriv->src_h) / pPriv->dst_h;
 
 	srcw = (pPriv->src_w * dstw) / pPriv->dst_w;
 	srch = (pPriv->src_h * dsth) / pPriv->dst_h;
diff --git a/src/radeon_textured_video.c b/src/radeon_textured_video.c
index 3b7bd714..b6cedb24 100644
--- a/src/radeon_textured_video.c
+++ b/src/radeon_textured_video.c
@@ -237,7 +237,7 @@ RADEONPutImageTextured(ScrnInfoPtr pScrn,
     int srcPitch, srcPitch2, dstPitch, dstPitch2 = 0;
     int s2offset, s3offset, tmp;
     int d2line, d3line;
-    int top, left, npixels, nlines, size;
+    int top, nlines, size;
     BoxRec dstBox;
     int dst_width = width, dst_height = height;
     int hw_align;
@@ -359,13 +359,10 @@ RADEONPutImageTextured(ScrnInfoPtr pScrn,
 
     /* copy data */
     top = (y1 >> 16) & ~1;
-    left = (x1 >> 16) & ~1;
-    npixels = ((((x2 + 0xffff) >> 16) + 1) & ~1) - left;
     nlines = ((((y2 + 0xffff) >> 16) + 1) & ~1) - top;
 
     pPriv->src_offset = pPriv->video_offset + info->fbLocation + pScrn->fbOffset;
     pPriv->src_addr = (uint8_t *)(info->FB + pPriv->video_offset + (top * dstPitch));
-    pPriv->src_addr += left << 1;
     pPriv->src_pitch = dstPitch;
 
     pPriv->planeu_offset = dstPitch * dst_height;
@@ -379,46 +376,40 @@ RADEONPutImageTextured(ScrnInfoPtr pScrn,
     switch(id) {
     case FOURCC_YV12:
     case FOURCC_I420:
-	s2offset = srcPitch * ((height + 1) & ~1);
-	s3offset = s2offset + (srcPitch2 * ((height + 1) >> 1));
-	tmp = ((top >> 1) * srcPitch2) + (left >> 1);
-	s2offset += tmp;
-	s3offset += tmp;
+	s2offset = (srcPitch * ((height + 1) & ~1)) + ((top >> 1) * srcPitch2);
+	s3offset = (s2offset + (srcPitch2 * ((height + 1) >> 1))) + ((top >> 1) * srcPitch2);
 	if (pPriv->bicubic_state != BICUBIC_OFF) {
 	    if (id == FOURCC_I420) {
 		tmp = s2offset;
 		s2offset = s3offset;
 		s3offset = tmp;
 	    }
-	    RADEONCopyMungedData(pScrn, buf + (top * srcPitch) + left,
-				 buf + s2offset, buf + s3offset, pPriv->src_addr,
-				 srcPitch, srcPitch2, dstPitch, nlines, npixels);
+	    RADEONCopyMungedData(pScrn, buf + (top * srcPitch),
+				 buf + s2offset, buf + s3offset, pPriv->src_addr + (top * dstPitch),
+				 srcPitch, srcPitch2, dstPitch, nlines, width);
 	} else {
 	    if (id == FOURCC_YV12) {
 		tmp = s2offset;
 		s2offset = s3offset;
 		s3offset = tmp;
 	    }
-	    d2line = pPriv->planeu_offset;
-	    d3line = pPriv->planev_offset;
-	    tmp = ((top >> 1) * dstPitch2) - (top * dstPitch);
-	    d2line += tmp;
-	    d3line += tmp;
+	    d2line = pPriv->planeu_offset + ((top >> 1) * dstPitch2);
+	    d3line = pPriv->planev_offset + ((top >> 1) * dstPitch2);
 
 	    if (info->ChipFamily >= CHIP_FAMILY_R600) {
-		R600CopyData(pScrn, buf + (top * srcPitch) + left, pPriv->src_addr + left,
-			     srcPitch, dstPitch, nlines, npixels, 1);
-		R600CopyData(pScrn, buf + s2offset,  pPriv->src_addr + d2line + (left >> 1),
-			     srcPitch2, dstPitch >> 1, (nlines + 1) >> 1, npixels >> 1, 1);
-		R600CopyData(pScrn, buf + s3offset, pPriv->src_addr + d3line + (left >> 1),
-			     srcPitch2, dstPitch >> 1, (nlines + 1) >> 1, npixels >> 1, 1);
+		R600CopyData(pScrn, buf + (top * srcPitch), pPriv->src_addr + (top * dstPitch),
+			     srcPitch, dstPitch, nlines, width, 1);
+		R600CopyData(pScrn, buf + s2offset,  pPriv->src_addr + d2line,
+			     srcPitch2, dstPitch2, (nlines + 1) >> 1, width >> 1, 1);
+		R600CopyData(pScrn, buf + s3offset, pPriv->src_addr + d3line,
+			     srcPitch2, dstPitch2, (nlines + 1) >> 1, width >> 1, 1);
 	    } else {
-		RADEONCopyData(pScrn, buf + (top * srcPitch) + left, pPriv->src_addr + left,
-			       srcPitch, dstPitch, nlines, npixels, 1);
-		RADEONCopyData(pScrn, buf + s2offset,  pPriv->src_addr + d2line + (left >> 1),
-			       srcPitch2, dstPitch2, (nlines + 1) >> 1, npixels >> 1, 1);
-		RADEONCopyData(pScrn, buf + s3offset, pPriv->src_addr + d3line + (left >> 1),
-			       srcPitch2, dstPitch2, (nlines + 1) >> 1, npixels >> 1, 1);
+		RADEONCopyData(pScrn, buf + (top * srcPitch), pPriv->src_addr + (top * dstPitch),
+			       srcPitch, dstPitch, nlines, width, 1);
+		RADEONCopyData(pScrn, buf + s2offset,  pPriv->src_addr + d2line,
+			       srcPitch2, dstPitch2, (nlines + 1) >> 1, width >> 1, 1);
+		RADEONCopyData(pScrn, buf + s3offset, pPriv->src_addr + d3line,
+			       srcPitch2, dstPitch2, (nlines + 1) >> 1, width >> 1, 1);
 	    }
 	}
 	break;
@@ -426,11 +417,13 @@ RADEONPutImageTextured(ScrnInfoPtr pScrn,
     case FOURCC_YUY2:
     default:
 	if (info->ChipFamily >= CHIP_FAMILY_R600)
-	    R600CopyData(pScrn, buf + (top * srcPitch) + (left << 1),
-			 pPriv->src_addr, srcPitch, dstPitch, nlines, npixels, 2);
+	    R600CopyData(pScrn, buf + (top * srcPitch),
+			 pPriv->src_addr + (top * dstPitch),
+			 srcPitch, dstPitch, nlines, width, 2);
 	else
-	    RADEONCopyData(pScrn, buf + (top * srcPitch) + (left << 1),
-			   pPriv->src_addr, srcPitch, dstPitch, nlines, npixels, 2);
+	    RADEONCopyData(pScrn, buf + (top * srcPitch),
+			   pPriv->src_addr + (top * dstPitch),
+			   srcPitch, dstPitch, nlines, width, 2);
 	break;
     }
 
@@ -449,6 +442,8 @@ RADEONPutImageTextured(ScrnInfoPtr pScrn,
     pPriv->id = id;
     pPriv->src_w = src_w;
     pPriv->src_h = src_h;
+    pPriv->src_x = src_x;
+    pPriv->src_y = src_y;
     pPriv->drw_x = drw_x;
     pPriv->drw_y = drw_y;
     pPriv->dst_w = drw_w;
diff --git a/src/radeon_textured_videofuncs.c b/src/radeon_textured_videofuncs.c
index caf8dce0..dd6f862c 100644
--- a/src/radeon_textured_videofuncs.c
+++ b/src/radeon_textured_videofuncs.c
@@ -388,10 +388,12 @@ FUNC_NAME(RADEONDisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv
 	dstw = pBox->x2 - pBox->x1;
 	dsth = pBox->y2 - pBox->y1;
 
-	srcX = ((pBox->x1 - pPriv->drw_x) *
-		pPriv->src_w) / pPriv->dst_w;
-	srcY = ((pBox->y1 - pPriv->drw_y) *
-		pPriv->src_h) / pPriv->dst_h;
+	srcX = pPriv->src_x;
+	srcX += ((pBox->x1 - pPriv->drw_x) *
+		 pPriv->src_w) / pPriv->dst_w;
+	srcY = pPriv->src_y;
+	srcY += ((pBox->y1 - pPriv->drw_y) *
+		 pPriv->src_h) / pPriv->dst_h;
 
 	srcw = (pPriv->src_w * dstw) / pPriv->dst_w;
 	srch = (pPriv->src_h * dsth) / pPriv->dst_h;
@@ -546,9 +548,6 @@ FUNC_NAME(R200DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 
     FINISH_ACCEL();
 
-    info->accel_state->texW[0] = pPriv->w;
-    info->accel_state->texH[0] = pPriv->h;
-
     txfilter =  R200_MAG_FILTER_LINEAR |
 	R200_MIN_FILTER_LINEAR |
 	R200_CLAMP_S_CLAMP_LAST |
@@ -915,10 +914,12 @@ FUNC_NAME(R200DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	dstw = pBox->x2 - pBox->x1;
 	dsth = pBox->y2 - pBox->y1;
 
-	srcX = ((pBox->x1 - pPriv->drw_x) *
-		pPriv->src_w) / pPriv->dst_w;
-	srcY = ((pBox->y1 - pPriv->drw_y) *
-		pPriv->src_h) / pPriv->dst_h;
+	srcX = pPriv->src_x;
+	srcX += ((pBox->x1 - pPriv->drw_x) *
+		 pPriv->src_w) / pPriv->dst_w;
+	srcY = pPriv->src_y;
+	srcY += ((pBox->y1 - pPriv->drw_y) *
+		 pPriv->src_h) / pPriv->dst_h;
 
 	srcw = (pPriv->src_w * dstw) / pPriv->dst_w;
 	srch = (pPriv->src_h * dsth) / pPriv->dst_h;
@@ -927,25 +928,25 @@ FUNC_NAME(R200DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	    /*
 	     * Just render a rect (using three coords).
 	     */
-	    VTX_OUT_6((float)dstX,                                (float)(dstY + dsth),
-		      (float)srcX / info->accel_state->texW[0],          (float)(srcY + srch) / info->accel_state->texH[0],
-		      (float)srcX / info->accel_state->texW[0],          (float)(srcY + srch) / info->accel_state->texH[0]);
-	    VTX_OUT_6((float)(dstX + dstw),                       (float)(dstY + dsth),
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)(srcY + srch) / info->accel_state->texH[0],
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)(srcY + srch) / info->accel_state->texH[0]);
-	    VTX_OUT_6((float)(dstX + dstw),                       (float)dstY,
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0],
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0]);
+	    VTX_OUT_6((float)dstX,                     (float)(dstY + dsth),
+		      (float)srcX / pPriv->w,          (float)(srcY + srch) / pPriv->h,
+		      (float)srcX / pPriv->w,          (float)(srcY + srch) / pPriv->h);
+	    VTX_OUT_6((float)(dstX + dstw),            (float)(dstY + dsth),
+		      (float)(srcX + srcw) / pPriv->w, (float)(srcY + srch) / pPriv->h,
+		      (float)(srcX + srcw) / pPriv->w, (float)(srcY + srch) / pPriv->h);
+	    VTX_OUT_6((float)(dstX + dstw),            (float)dstY,
+		      (float)(srcX + srcw) / pPriv->w, (float)srcY / pPriv->h,
+		      (float)(srcX + srcw) / pPriv->w, (float)srcY / pPriv->h);
 	} else {
 	    /*
 	     * Just render a rect (using three coords).
 	     */
-	    VTX_OUT_4((float)dstX,                                       (float)(dstY + dsth),
-		      (float)srcX / info->accel_state->texW[0],          (float)(srcY + srch) / info->accel_state->texH[0]);
-	    VTX_OUT_4((float)(dstX + dstw),                              (float)(dstY + dsth),
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)(srcY + srch) / info->accel_state->texH[0]);
-	    VTX_OUT_4((float)(dstX + dstw),                              (float)dstY,
-		      (float)(srcX + srcw) / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0]);
+	    VTX_OUT_4((float)dstX,                     (float)(dstY + dsth),
+		      (float)srcX / pPriv->w,          (float)(srcY + srch) / pPriv->h);
+	    VTX_OUT_4((float)(dstX + dstw),            (float)(dstY + dsth),
+		      (float)(srcX + srcw) / pPriv->w, (float)(srcY + srch) / pPriv->h);
+	    VTX_OUT_4((float)(dstX + dstw),            (float)dstY,
+		      (float)(srcX + srcw) / pPriv->w, (float)srcY / pPriv->h);
 	}
 
 	pBox++;
@@ -1076,9 +1077,6 @@ FUNC_NAME(R300DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 		 (((pPriv->h - 1) & 0x7ff) << R300_TXHEIGHT_SHIFT) |
 		 R300_TXPITCH_EN);
 
-    info->accel_state->texW[0] = pPriv->w;
-    info->accel_state->texH[0] = pPriv->h;
-
     txfilter = (R300_TX_CLAMP_S(R300_TX_CLAMP_CLAMP_LAST) |
 		R300_TX_CLAMP_T(R300_TX_CLAMP_CLAMP_LAST) |
 		R300_TX_MAG_FILTER_LINEAR |
@@ -2252,10 +2250,12 @@ FUNC_NAME(R300DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	dstw = pBox->x2 - pBox->x1;
 	dsth = pBox->y2 - pBox->y1;
 
-	srcX = ((pBox->x1 - pPriv->drw_x) *
-		pPriv->src_w) / pPriv->dst_w;
-	srcY = ((pBox->y1 - pPriv->drw_y) *
-		pPriv->src_h) / pPriv->dst_h;
+	srcX = pPriv->src_x;
+	srcX += ((pBox->x1 - pPriv->drw_x) *
+		 pPriv->src_w) / pPriv->dst_w;
+	srcY = pPriv->src_y;
+	srcY += ((pBox->y1 - pPriv->drw_y) *
+		 pPriv->src_h) / pPriv->dst_h;
 
 	srcw = (pPriv->src_w * dstw) / pPriv->dst_w;
 	srch = (pPriv->src_h * dsth) / pPriv->dst_h;
@@ -2315,57 +2315,57 @@ FUNC_NAME(R300DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 		 * have to deal with the legacy handling.
 		 */
 	    if (use_quad) {
-		VTX_OUT_6((float)dstX,                                       (float)dstY,
-			  (float)srcX / info->accel_state->texW[0],          (float)srcY / info->accel_state->texH[0],
-			  (float)srcX + 0.5,                                 (float)srcY + 0.5);
-		VTX_OUT_6((float)dstX,                                       (float)(dstY + dsth),
-			  (float)srcX / info->accel_state->texW[0],          (float)(srcY + srch) / info->accel_state->texH[0],
-			  (float)srcX + 0.5,                                 (float)(srcY + srch) + 0.5);
-		VTX_OUT_6((float)(dstX + dstw),                              (float)(dstY + dsth),
-			  (float)(srcX + srcw) / info->accel_state->texW[0], (float)(srcY + srch) / info->accel_state->texH[0],
-			  (float)(srcX + srcw) + 0.5,                        (float)(srcY + srch) + 0.5);
-		VTX_OUT_6((float)(dstX + dstw),                              (float)dstY,
-			  (float)(srcX + srcw) / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0],
-			  (float)(srcX + srcw) + 0.5,                        (float)srcY + 0.5);
+		VTX_OUT_6((float)dstX,                     (float)dstY,
+			  (float)srcX / pPriv->w,          (float)srcY / pPriv->h,
+			  (float)srcX + 0.5,               (float)srcY + 0.5);
+		VTX_OUT_6((float)dstX,                     (float)(dstY + dsth),
+			  (float)srcX / pPriv->w,          (float)(srcY + srch) / pPriv->h,
+			  (float)srcX + 0.5,               (float)(srcY + srch) + 0.5);
+		VTX_OUT_6((float)(dstX + dstw),            (float)(dstY + dsth),
+			  (float)(srcX + srcw) / pPriv->w, (float)(srcY + srch) / pPriv->h,
+			  (float)(srcX + srcw) + 0.5,      (float)(srcY + srch) + 0.5);
+		VTX_OUT_6((float)(dstX + dstw),            (float)dstY,
+			  (float)(srcX + srcw) / pPriv->w, (float)srcY / pPriv->h,
+			  (float)(srcX + srcw) + 0.5,      (float)srcY + 0.5);
 	    } else {
-		VTX_OUT_6((float)dstX,                                       (float)dstY,
-			  (float)srcX / info->accel_state->texW[0],          (float)srcY / info->accel_state->texH[0],
-			  (float)srcX + 0.5,                                 (float)srcY + 0.5);
-		VTX_OUT_6((float)dstX,                                       (float)(dstY + dstw + dsth),
-			  (float)srcX / info->accel_state->texW[0],
-			  ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / info->accel_state->texH[0],
+		VTX_OUT_6((float)dstX,                     (float)dstY,
+			  (float)srcX / pPriv->w,          (float)srcY / pPriv->h,
+			  (float)srcX + 0.5,               (float)srcY + 0.5);
+		VTX_OUT_6((float)dstX,                     (float)(dstY + dstw + dsth),
+			  (float)srcX / pPriv->w,
+			  ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / pPriv->h,
 			  (float)srcX + 0.5,
 			  (float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0) + 0.5);
-		VTX_OUT_6((float)(dstX + dstw + dsth),                       (float)dstY,
-			  ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / info->accel_state->texW[0],
-			  (float)srcY / info->accel_state->texH[0],
+		VTX_OUT_6((float)(dstX + dstw + dsth),     (float)dstY,
+			  ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / pPriv->w,
+			  (float)srcY / pPriv->h,
 			  (float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0) + 0.5,
 			  (float)srcY + 0.5);
 	    }
 	} else {
 	    if (use_quad) {
-		VTX_OUT_4((float)dstX,                                       (float)dstY,
-			  (float)srcX / info->accel_state->texW[0],          (float)srcY / info->accel_state->texH[0]);
-		VTX_OUT_4((float)dstX,                                       (float)(dstY + dsth),
-			  (float)srcX / info->accel_state->texW[0],          (float)(srcY + srch) / info->accel_state->texH[0]);
-		VTX_OUT_4((float)(dstX + dstw),                              (float)(dstY + dsth),
-			  (float)(srcX + srcw) / info->accel_state->texW[0], (float)(srcY + srch) / info->accel_state->texH[0]);
-		VTX_OUT_4((float)(dstX + dstw),                              (float)dstY,
-			  (float)(srcX + srcw) / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0]);
+		VTX_OUT_4((float)dstX,                     (float)dstY,
+			  (float)srcX / pPriv->w,          (float)srcY / pPriv->h);
+		VTX_OUT_4((float)dstX,                     (float)(dstY + dsth),
+			  (float)srcX / pPriv->w,          (float)(srcY + srch) / pPriv->h);
+		VTX_OUT_4((float)(dstX + dstw),            (float)(dstY + dsth),
+			  (float)(srcX + srcw) / pPriv->w, (float)(srcY + srch) / pPriv->h);
+		VTX_OUT_4((float)(dstX + dstw),            (float)dstY,
+			  (float)(srcX + srcw) / pPriv->w, (float)srcY / pPriv->h);
 	    } else {
 		/*
 		 * Render a big, scissored triangle. This means
 		 * increasing the triangle size and adjusting
 		 * texture coordinates.
 		 */
-		VTX_OUT_4((float)dstX,                              (float)dstY,
-			  (float)srcX / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0]);
-		VTX_OUT_4((float)dstX,                              (float)(dstY + dsth + dstw),
-			  (float)srcX / info->accel_state->texW[0],
-			  ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / info->accel_state->texH[0]);
-		VTX_OUT_4((float)(dstX + dstw + dsth),              (float)dstY,
-			  ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / info->accel_state->texW[0],
-			  (float)srcY / info->accel_state->texH[0]);
+		VTX_OUT_4((float)dstX,                 (float)dstY,
+			  (float)srcX / pPriv->w,      (float)srcY / pPriv->h);
+		VTX_OUT_4((float)dstX,                 (float)(dstY + dsth + dstw),
+			  (float)srcX / pPriv->w,
+			  ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / pPriv->h);
+		VTX_OUT_4((float)(dstX + dstw + dsth), (float)dstY,
+			  ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / pPriv->w,
+			  (float)srcY / pPriv->h);
 	    }
 	}
 
@@ -2503,9 +2503,6 @@ FUNC_NAME(R500DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 		 (((pPriv->h - 1) & 0x7ff) << R300_TXHEIGHT_SHIFT) |
 		 R300_TXPITCH_EN);
 
-    info->accel_state->texW[0] = pPriv->w;
-    info->accel_state->texH[0] = pPriv->h;
-
     txfilter = (R300_TX_CLAMP_S(R300_TX_CLAMP_CLAMP_LAST) |
 		R300_TX_CLAMP_T(R300_TX_CLAMP_CLAMP_LAST) |
 		R300_TX_MAG_FILTER_LINEAR |
@@ -3805,10 +3802,12 @@ FUNC_NAME(R500DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	dstw = pBox->x2 - pBox->x1;
 	dsth = pBox->y2 - pBox->y1;
 
-	srcX = ((pBox->x1 - pPriv->drw_x) *
-		pPriv->src_w) / pPriv->dst_w;
-	srcY = ((pBox->y1 - pPriv->drw_y) *
-		pPriv->src_h) / pPriv->dst_h;
+	srcX = pPriv->src_x;
+	srcX += ((pBox->x1 - pPriv->drw_x) *
+		 pPriv->src_w) / pPriv->dst_w;
+	srcY = pPriv->src_y;
+	srcY += ((pBox->y1 - pPriv->drw_y) *
+		 pPriv->src_h) / pPriv->dst_h;
 
 	srcw = (pPriv->src_w * dstw) / pPriv->dst_w;
 	srch = (pPriv->src_h * dsth) / pPriv->dst_h;
@@ -3834,15 +3833,15 @@ FUNC_NAME(R500DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 					  (3 << RADEON_VF_NUM_VERTICES_SHIFT)));
 #endif
 	if (pPriv->bicubic_enabled) {
-	    VTX_OUT_6((float)dstX,                                       (float)dstY,
-		      (float)srcX / info->accel_state->texW[0],          (float)srcY / info->accel_state->texH[0],
-		      (float)srcX + 0.5,                                 (float)srcY + 0.5);
-	    VTX_OUT_6((float)dstX,                                       (float)(dstY + dstw + dsth),
-		      (float)srcX / info->accel_state->texW[0],          ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / info->accel_state->texH[0],
-		      (float)srcX + 0.5,                                 (float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0) + 0.5);
+	    VTX_OUT_6((float)dstX,            (float)dstY,
+		      (float)srcX / pPriv->w, (float)srcY / pPriv->h,
+		      (float)srcX + 0.5,      (float)srcY + 0.5);
+	    VTX_OUT_6((float)dstX,            (float)(dstY + dstw + dsth),
+		      (float)srcX / pPriv->w, ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / pPriv->h,
+		      (float)srcX + 0.5,      (float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0) + 0.5);
 	    VTX_OUT_6((float)(dstX + dstw + dsth),                       (float)dstY,
-		      ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / info->accel_state->texW[0],
-		      (float)srcY / info->accel_state->texH[0],
+		      ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / pPriv->w,
+		      (float)srcY / pPriv->h,
 		      (float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0) + 0.5,
 		      (float)srcY + 0.5);
 	} else {
@@ -3851,13 +3850,13 @@ FUNC_NAME(R500DisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv)
 	     * increasing the triangle size and adjusting
 	     * texture coordinates.
 	     */
-	    VTX_OUT_4((float)dstX,                              (float)dstY,
-		      (float)srcX / info->accel_state->texW[0], (float)srcY / info->accel_state->texH[0]);
+	    VTX_OUT_4((float)dstX,            (float)dstY,
+		      (float)srcX / pPriv->w, (float)srcY / pPriv->h);
 	    VTX_OUT_4((float)dstX,                              (float)(dstY + dsth + dstw),
-		      (float)srcX / info->accel_state->texW[0], ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / info->accel_state->texH[0]);
+		      (float)srcX / pPriv->w, ((float)srcY + (float)srch * (((float)dstw / (float)dsth) + 1.0)) / pPriv->h);
 	    VTX_OUT_4((float)(dstX + dstw + dsth),              (float)dstY,
-		      ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / info->accel_state->texW[0],
-		      (float)srcY / info->accel_state->texH[0]);
+		      ((float)srcX + (float)srcw * (((float)dsth / (float)dstw) + 1.0)) / pPriv->w,
+		      (float)srcY / pPriv->h);
 	}
 
 	/* flushing is pipelined, free/finish is not */
diff --git a/src/radeon_video.h b/src/radeon_video.h
index 3f8f5e0e..0cf8168b 100644
--- a/src/radeon_video.h
+++ b/src/radeon_video.h
@@ -118,6 +118,7 @@ typedef struct {
     int src_w, src_h, dst_w, dst_h;
     int w, h;
     int drw_x, drw_y;
+    int src_x, src_y;
     int vsync;
 } RADEONPortPrivRec, *RADEONPortPrivPtr;