Bug #6756: Attempt to fix repeat picture acceleration.

Always use normalized texture coordinates on R200 and fall back if a POT
texture pitch doesn't match the HW's implicit pitch.

4 files changed, 67 insertions, 23 deletions

diff --git a/src/radeon.h b/src/radeon.h
index 7ff86f98..a7b0dfda 100644
--- a/src/radeon.h
+++ b/src/radeon.h
@@ -742,6 +742,8 @@ typedef struct {
 
     /* Render */
     Bool              RenderAccel;
+    unsigned short    texW[2];
+    unsigned short    texH[2];
 #ifdef USE_XAA
     FBLinearPtr       RenderTex;
     void              (*RenderCallback)(ScrnInfoPtr);
diff --git a/src/radeon_commonfuncs.c b/src/radeon_commonfuncs.c
index 84ce19c0..70f7ddc1 100644
--- a/src/radeon_commonfuncs.c
+++ b/src/radeon_commonfuncs.c
@@ -55,6 +55,8 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
     RADEONInfoPtr  info       = RADEONPTR(pScrn);
     ACCEL_PREAMBLE();
 
+    info->texW[0] = info->texH[0] = info->texW[1] = info->texH[1] = 1;
+
     if (info->ChipFamily >= CHIP_FAMILY_R300) {
 	/* Unimplemented */
     } else if ((info->ChipFamily == CHIP_FAMILY_RV250) || 
@@ -72,8 +74,7 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 	OUT_ACCEL_REG(R200_PP_TXMULTI_CTL_0, 0);
 	OUT_ACCEL_REG(R200_SE_VTX_STATE_CNTL, 0);
 	OUT_ACCEL_REG(R200_RE_CNTL, 0x0);
-	/* XXX: correct?  Want it to be like RADEON_VTX_ST?_NONPARAMETRIC */
-	OUT_ACCEL_REG(R200_SE_VTE_CNTL, R200_VTX_ST_DENORMALIZED);
+	OUT_ACCEL_REG(R200_SE_VTE_CNTL, 0);
 	OUT_ACCEL_REG(R200_SE_VAP_CNTL, R200_VAP_FORCE_W_TO_ONE |
 	    R200_VAP_VF_MAX_VTX_NUM);
 	FINISH_ACCEL();
diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index 98b8dfdd..d0a72163 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -246,12 +246,19 @@ static Bool FUNC_NAME(R100TextureSetup)(PicturePtr pPict, PixmapPtr pPix,
 	txoffset |= RADEON_TXO_MACRO_TILE;
 
     if (pPict->repeat) {
+	if (((w * pPix->drawable.bitsPerPixel / 8 + 31) & ~31) != txpitch)
+	    RADEON_FALLBACK(("Width %d and pitch %u not compatible for repeat\n",
+			     w, (unsigned)txpitch));
+
 	txformat |= RADEONLog2(w) << RADEON_TXFORMAT_WIDTH_SHIFT;
 	txformat |= RADEONLog2(h) << RADEON_TXFORMAT_HEIGHT_SHIFT;
-    } else 
+    } else
 	txformat |= RADEON_TXFORMAT_NON_POWER2;
     txformat |= unit << 24; /* RADEON_TXFORMAT_ST_ROUTE_STQX */
- 
+
+    info->texW[unit] = 1;
+    info->texH[unit] = 1;
+
     switch (pPict->filter) {
     case PictFilterNearest:
 	txfilter = (RADEON_MAG_FILTER_NEAREST | RADEON_MIN_FILTER_NEAREST);
@@ -499,12 +506,19 @@ static Bool FUNC_NAME(R200TextureSetup)(PicturePtr pPict, PixmapPtr pPix,
 	txoffset |= R200_TXO_MACRO_TILE;
 
     if (pPict->repeat) {
+	if (((w * pPix->drawable.bitsPerPixel / 8 + 31) & ~31) != txpitch)
+	    RADEON_FALLBACK(("Width %d and pitch %u not compatible for repeat\n",
+			     w, (unsigned)txpitch));
+
 	txformat |= RADEONLog2(w) << R200_TXFORMAT_WIDTH_SHIFT;
 	txformat |= RADEONLog2(h) << R200_TXFORMAT_HEIGHT_SHIFT;
     } else
 	txformat |= R200_TXFORMAT_NON_POWER2;
     txformat |= unit << R200_TXFORMAT_ST_ROUTE_SHIFT;
 
+    info->texW[unit] = w;
+    info->texH[unit] = h;
+
     switch (pPict->filter) {
     case PictFilterNearest:
 	txfilter = (R200_MAG_FILTER_NEAREST |
@@ -791,11 +805,6 @@ static void FUNC_NAME(RadeonComposite)(PixmapPtr pDst,
 		 (4 << RADEON_CP_VC_CNTL_NUM_SHIFT));
     }
 
-    VTX_OUT(dstX,     dstY,     srcX,    srcY,    maskX,    maskY);
-    VTX_OUT(dstX,     dstY + h, srcX,    srcYend, maskX,    maskYend);
-    VTX_OUT(dstX + w, dstY + h, srcXend, srcYend, maskXend, maskYend);
-    VTX_OUT(dstX + w, dstY,     srcXend, srcY,    maskXend, maskY);
-    ADVANCE_RING();
 #else /* ACCEL_CP */
     BEGIN_ACCEL(1 + VTX_REG_COUNT * 4);
     if (info->ChipFamily < CHIP_FAMILY_R200) {
@@ -808,11 +817,32 @@ static void FUNC_NAME(RadeonComposite)(PixmapPtr pDst,
 					  RADEON_VF_PRIM_WALK_DATA |
 					  4 << RADEON_VF_NUM_VERTICES_SHIFT));
     }
+#endif
+
+    if (info->texW[0] == 1 && info->texH[0] == 1 &&
+	info->texW[1] == 1 && info->texH[1] == 1) {
+	VTX_OUT(dstX,     dstY,       srcX,     srcY,	  maskX,    maskY);
+	VTX_OUT(dstX,     dstY + h,   srcX,     srcYend,  maskX,    maskYend);
+	VTX_OUT(dstX + w, dstY + h,   srcXend,  srcYend,  maskXend, maskYend);
+	VTX_OUT(dstX + w, dstY,	      srcXend,  srcY,     maskXend, maskY);
+    } else {
+    VTX_OUT((float)dstX,     (float)dstY,
+	    (float)srcX / info->texW[0],     (float)srcY / info->texH[0],
+	    (float)maskX / info->texW[1],    (float)maskY / info->texH[1]);
+    VTX_OUT((float)dstX,     (float)(dstY + h),
+	    (float)srcX / info->texW[0],     (float)srcYend / info->texH[0],
+	    (float)maskX / info->texW[1],    (float)maskYend / info->texH[1]);
+    VTX_OUT((float)(dstX + w), (float)(dstY + h),
+	    (float)srcXend / info->texW[0],  (float)srcYend / info->texH[0],
+	    (float)maskXend / info->texW[1], (float)maskYend / info->texH[1]);
+    VTX_OUT((float)(dstX + w), (float)dstY,
+	    (float)srcXend / info->texW[0],  (float)srcY / info->texH[0],
+	    (float)maskXend / info->texW[1], (float)maskY / info->texH[1]);
+    }
 
-    VTX_OUT(dstX,     dstY,     srcX,    srcY,    maskX,    maskY);
-    VTX_OUT(dstX,     dstY + h, srcX,    srcYend, maskX,    maskYend);
-    VTX_OUT(dstX + w, dstY + h, srcXend, srcYend, maskXend, maskYend);
-    VTX_OUT(dstX + w, dstY,     srcXend, srcY,    maskXend, maskY);
+#ifdef ACCEL_CP
+    ADVANCE_RING();
+#else
     FINISH_ACCEL();
 #endif /* !ACCEL_CP */
 
diff --git a/src/radeon_render.c b/src/radeon_render.c
index 9e3529e3..054e60f7 100644
--- a/src/radeon_render.c
+++ b/src/radeon_render.c
@@ -398,6 +398,13 @@ static Bool FUNC_NAME(R100SetupTexture)(
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+    dst_pitch = (width * tex_bytepp + 63) & ~63;
+    size = dst_pitch * height;
+
+    if ((flags & XAA_RENDER_REPEAT) &&
+	(((width * tex_bytepp + 31) & ~31) != dst_pitch))
+	return FALSE;
+
 #ifndef ACCEL_CP
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
@@ -410,9 +417,6 @@ static Bool FUNC_NAME(R100SetupTexture)(
 
 #endif
 
-    dst_pitch = (width * tex_bytepp + 63) & ~63;
-    size = dst_pitch * height;
-
     if (!AllocateLinear(pScrn, size))
 	return FALSE;
 
@@ -731,6 +735,13 @@ static Bool FUNC_NAME(R200SetupTexture)(
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+    dst_pitch = (width * tex_bytepp + 63) & ~63;
+    size = dst_pitch * height;
+
+    if ((flags & XAA_RENDER_REPEAT) &&
+	(((width * tex_bytepp + 31) & ~31) != dst_pitch))
+	return FALSE;
+
 #ifndef ACCEL_CP
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
@@ -743,9 +754,6 @@ static Bool FUNC_NAME(R200SetupTexture)(
 
 #endif
 
-    dst_pitch = (width * tex_bytepp + 63) & ~63;
-    size = dst_pitch * height;
-
     if (!AllocateLinear(pScrn, size))
 	return FALSE;
 
@@ -757,6 +765,9 @@ static Bool FUNC_NAME(R200SetupTexture)(
 	txformat |= RADEON_TXFORMAT_NON_POWER2;
     }
 
+    info->texW[0] = width;
+    info->texH[0] = height;
+
     offset = info->RenderTex->offset * pScrn->bitsPerPixel / 8;
     dst = (CARD8*)(info->FB + offset);
 
@@ -956,10 +967,10 @@ FUNC_NAME(R200SubsequentCPUToScreenTexture) (
     
     r = width + l;
     b = height + t;
-    fl = srcx;
-    fr = srcx + width;
-    ft = srcy;
-    fb = srcy + height;
+    fl = (float)srcx / info->texW[0];
+    fr = (float)(srcx + width) / info->texW[0];
+    ft = (float)srcy / info->texH[0];
+    fb = (float)(srcy + height) / info->texH[0];
 
 #ifdef ACCEL_CP
     BEGIN_RING(24);