diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index 98b8dfd..a47c347 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -252,6 +261,13 @@ static Bool FUNC_NAME(R100TextureSetup)(
 	txformat |= RADEON_TXFORMAT_NON_POWER2;
     txformat |= unit << 24; /* RADEON_TXFORMAT_ST_ROUTE_STQX */
  
+    if (((w * pPix->drawable.bitsPerPixel / 8 + 31) & ~31) != txpitch)
+	RADEON_FALLBACK(("Width %d and pitch %u not compatible for repeat\n",
+			 w, txpitch));
+
+    info->texW[unit] = w;
+    info->texH[unit] = h;
+
     switch (pPict->filter) {
     case PictFilterNearest:
 	txfilter = (RADEON_MAG_FILTER_NEAREST | RADEON_MIN_FILTER_NEAREST);
@@ -505,6 +521,13 @@ static Bool FUNC_NAME(R200TextureSetup)(
 	txformat |= R200_TXFORMAT_NON_POWER2;
     txformat |= unit << R200_TXFORMAT_ST_ROUTE_SHIFT;
 
+    if (((w * pPix->drawable.bitsPerPixel / 8 + 31) & ~31) != txpitch)
+	RADEON_FALLBACK(("Width %d and pitch %u not compatible for repeat\n",
+			 w, txpitch));
+
+    info->texW[unit] = w;
+    info->texH[unit] = h;
+
     switch (pPict->filter) {
     case PictFilterNearest:
 	txfilter = (R200_MAG_FILTER_NEAREST |
@@ -791,11 +814,6 @@ #ifdef ACCEL_CP
 		 (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 +826,24 @@ #else /* ACCEL_CP */
 					  RADEON_VF_PRIM_WALK_DATA |
 					  4 << RADEON_VF_NUM_VERTICES_SHIFT));
     }
+#endif
+
+    VTX_OUT(dstX,     dstY,
+	    (float)srcX / info->texW[0],     (float)srcY / info->texH[0],
+	    (float)maskX / info->texW[1],    (float)maskY / info->texH[1]);
+    VTX_OUT(dstX,     dstY + h,
+	    (float)srcX / info->texW[0],     (float)srcYend / info->texH[0],
+	    (float)maskX / info->texW[1],    (float)maskYend / info->texH[1]);
+    VTX_OUT(dstX + w, dstY + h,
+	    (float)srcXend / info->texW[0],  (float)srcYend / info->texH[0],
+	    (float)maskXend / info->texW[1], (float)maskYend / info->texH[1]);
+    VTX_OUT(dstX + w, 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 9e3529e..85cb952 100644
--- a/src/radeon_render.c
+++ b/src/radeon_render.c
@@ -398,6 +398,9 @@ #endif
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+    if (((width * tex_bytepp + 31) & ~31) != src_pitch)
+	return FALSE;
+
 #ifndef ACCEL_CP
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
@@ -424,6 +427,9 @@ #endif
 	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);
 
@@ -621,10 +627,10 @@ FUNC_NAME(R100SubsequentCPUToScreenTextu
 
     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(25);
@@ -731,6 +737,9 @@ #endif
     txformat = RadeonGetTextureFormat(format);
     tex_bytepp = PICT_FORMAT_BPP(format) >> 3;
 
+    if (((width * tex_bytepp + 31) & ~31) != src_pitch)
+	return FALSE;
+
 #ifndef ACCEL_CP
 
 #if X_BYTE_ORDER == X_BIG_ENDIAN
@@ -757,6 +766,9 @@ #endif
 	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 +968,10 @@ FUNC_NAME(R200SubsequentCPUToScreenTextu
     
     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);
diff --git a/src/radeon.h b/src/radeon.h
index 6402e48..0b1f4bd 100644
--- a/src/radeon.h
+++ b/src/radeon.h
@@ -726,6 +726,8 @@ #endif /* XF86DRI */
 
     /* 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 14260c9..0e353b0 100644
--- a/src/radeon_commonfuncs.c
+++ b/src/radeon_commonfuncs.c
@@ -73,8 +73,7 @@ static void FUNC_NAME(RADEONInit3DEngine
 	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();
@@ -87,8 +86,6 @@ static void FUNC_NAME(RADEONInit3DEngine
             OUT_ACCEL_REG(RADEON_SE_CNTL_STATUS, RADEON_TCL_BYPASS);
 	OUT_ACCEL_REG(RADEON_SE_COORD_FMT,
 	    RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
-	    RADEON_VTX_ST0_NONPARAMETRIC |
-	    RADEON_VTX_ST1_NONPARAMETRIC |
 	    RADEON_TEX1_W_ROUTING_USE_W0);
 	FINISH_ACCEL();
     }