R300+: consolidate some tcl/non-tcl paths

- Move more code to init3d()
- MMIO textured video seems more reliable now on newer chips

3 files changed, 69 insertions, 142 deletions

diff --git a/src/radeon_commonfuncs.c b/src/radeon_commonfuncs.c
index ec4b434c..aa94b237 100644
--- a/src/radeon_commonfuncs.c
+++ b/src/radeon_commonfuncs.c
@@ -149,9 +149,9 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 
 	/* setup the VAP */	
 	if (info->has_tcl)
-	    BEGIN_ACCEL(13);
+	    BEGIN_ACCEL(15);
 	else
-	    BEGIN_ACCEL(7);
+	    BEGIN_ACCEL(9);
 	OUT_ACCEL_REG(R300_VAP_VTX_STATE_CNTL, 0);
 	OUT_ACCEL_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0);
 	OUT_ACCEL_REG(R300_VAP_CNTL, ((5/*6*/ << R300_PVS_NUM_SLOTS_SHIFT) |
@@ -166,6 +166,27 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 	OUT_ACCEL_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
 	OUT_ACCEL_REG(R300_VAP_PSC_SGN_NORM_CNTL, 0);
 
+	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
+		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_0_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_0_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_0_SHIFT) |
+		       (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_0_SHIFT) |
+		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
+			<< R300_WRITE_ENA_0_SHIFT) |
+		       (R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_1_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_1_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_1_SHIFT) |
+		       (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_1_SHIFT) |
+		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
+			<< R300_WRITE_ENA_1_SHIFT)));
+	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_1,
+		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_2_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_2_SHIFT) |
+		       (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_2_SHIFT) |
+		       (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_2_SHIFT) |
+		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
+			<< R300_WRITE_ENA_2_SHIFT)));
+
 	if (info->has_tcl) {
 	    OUT_ACCEL_REG(R300_VAP_PVS_FLOW_CNTL_OPC, 0);
 	    OUT_ACCEL_REG(R300_VAP_GB_VERT_CLIP_ADJ, 0x3f800000);
@@ -219,21 +240,21 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 			   R300_PVS_DST_WE_Z | R300_PVS_DST_WE_W));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_X) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_Y) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_Z) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_W)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_FORCE_0)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
@@ -248,21 +269,21 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 			   R300_PVS_DST_WE_Z | R300_PVS_DST_WE_W));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(2) |
+			   R300_PVS_SRC_OFFSET(7) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_X) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_Y) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_Z) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_W)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(2) |
+			   R300_PVS_SRC_OFFSET(7) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_FORCE_0)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(2) |
+			   R300_PVS_SRC_OFFSET(7) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
@@ -310,21 +331,21 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 			   R300_PVS_DST_WE_Z | R300_PVS_DST_WE_W));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_X) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_Y) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_Z) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_W)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_FORCE_0)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(1) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
@@ -371,21 +392,21 @@ static void FUNC_NAME(RADEONInit3DEngine)(ScrnInfoPtr pScrn)
 			   R300_PVS_DST_WE_Z | R300_PVS_DST_WE_W));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(10) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_X) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_Y) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_Z) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_W)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(10) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_W(R300_PVS_SRC_SELECT_FORCE_0)));
 	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
 			  (R300_PVS_SRC_REG_TYPE(R300_PVS_SRC_REG_INPUT) |
-			   R300_PVS_SRC_OFFSET(10) |
+			   R300_PVS_SRC_OFFSET(6) |
 			   R300_PVS_SRC_SWIZZLE_X(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Y(R300_PVS_SRC_SELECT_FORCE_0) |
 			   R300_PVS_SRC_SWIZZLE_Z(R300_PVS_SRC_SELECT_FORCE_0) |
diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index c1ea4d2c..8505d871 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -1093,85 +1093,29 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
     RADEON_SWITCH_TO_3D();
 
     /* setup the VAP */
-    BEGIN_ACCEL(4);
-    if (info->has_tcl) {
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_0,
-		      ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_0_SHIFT) |
-		       (0 << R300_DST_VEC_LOC_0_SHIFT) |
-		       R300_SIGNED_0 |
-		       (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_1_SHIFT) |
-		       (1 << R300_DST_VEC_LOC_1_SHIFT) |
-		       R300_SIGNED_1));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_1,
-		      ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_2_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_2_SHIFT) |
-		       (2 << R300_DST_VEC_LOC_2_SHIFT) |
-		       R300_LAST_VEC_2 |
-		       R300_SIGNED_2));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
-		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_0_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_1_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_1_SHIFT)));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_1,
-		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_2_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_2_SHIFT)));
-    } else {
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_0,
-		      ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_0_SHIFT) |
-		       (0 << R300_DST_VEC_LOC_0_SHIFT) |
-		       R300_SIGNED_0 |
-		       (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_1_SHIFT) |
-		       (6 << R300_DST_VEC_LOC_1_SHIFT) |
-		       R300_SIGNED_1));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_1,
-		      ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_2_SHIFT) |
-		       (0 << R300_SKIP_DWORDS_2_SHIFT) |
-		       (7 << R300_DST_VEC_LOC_2_SHIFT) |
-		       R300_LAST_VEC_2 |
-		       R300_SIGNED_2));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
-		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_0_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_0_SHIFT) |
-		       (R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_1_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_1_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_1_SHIFT)));
-	OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_EXT_1,
-		      ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_2_SHIFT) |
-		       (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_2_SHIFT) |
-		       ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			<< R300_WRITE_ENA_2_SHIFT)));
-    }
-    FINISH_ACCEL();
+    if (info->has_tcl)
+	BEGIN_ACCEL(8);
+    else
+	BEGIN_ACCEL(6);
+
+    OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_0,
+		  ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
+		   (0 << R300_SKIP_DWORDS_0_SHIFT) |
+		   (0 << R300_DST_VEC_LOC_0_SHIFT) |
+		   R300_SIGNED_0 |
+		   (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
+		   (0 << R300_SKIP_DWORDS_1_SHIFT) |
+		   (6 << R300_DST_VEC_LOC_1_SHIFT) |
+		   R300_SIGNED_1));
+    OUT_ACCEL_REG(R300_VAP_PROG_STREAM_CNTL_1,
+		  ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_2_SHIFT) |
+		   (0 << R300_SKIP_DWORDS_2_SHIFT) |
+		   (7 << R300_DST_VEC_LOC_2_SHIFT) |
+		   R300_LAST_VEC_2 |
+		   R300_SIGNED_2));
 
     /* load the vertex shader */
     if (info->has_tcl) {
-	BEGIN_ACCEL(2);
 	if (pMask) {
 	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_0,
 			  ((0 << R300_PVS_FIRST_INST_SHIFT) |
@@ -1187,10 +1131,8 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_1,
 			  (4 << R300_PVS_LAST_VTX_SRC_INST_SHIFT));
 	}
-	FINISH_ACCEL();
     }
 
-    BEGIN_ACCEL(4);
     OUT_ACCEL_REG(R300_VAP_OUT_VTX_FMT_0, R300_VTX_POS_PRESENT);
     OUT_ACCEL_REG(R300_VAP_OUT_VTX_FMT_1,
 		  ((2 << R300_TEX_0_COMP_CNT_SHIFT) |
diff --git a/src/radeon_textured_videofuncs.c b/src/radeon_textured_videofuncs.c
index a94d1311..828ec225 100644
--- a/src/radeon_textured_videofuncs.c
+++ b/src/radeon_textured_videofuncs.c
@@ -192,57 +192,21 @@ FUNC_NAME(RADEONDisplayTexturedVideo)(ScrnInfoPtr pScrn, RADEONPortPrivPtr pPriv
 	txenable = R300_TEX_0_ENABLE;
 
 	/* setup the VAP */
-	if (info->has_tcl) {
-	    BEGIN_VIDEO(7);
-	    OUT_VIDEO_REG(R300_VAP_PROG_STREAM_CNTL_0,
-			  ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
-			   (0 << R300_SKIP_DWORDS_0_SHIFT) |
-			   (0 << R300_DST_VEC_LOC_0_SHIFT) |
-			   R300_SIGNED_0 |
-			   (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
-			   (0 << R300_SKIP_DWORDS_1_SHIFT) |
-			   (10 << R300_DST_VEC_LOC_1_SHIFT) |
-			   R300_LAST_VEC_1 |
-			   R300_SIGNED_1));
-	    OUT_VIDEO_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
-			  ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_0_SHIFT) |
-			   ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			    << R300_WRITE_ENA_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Z << R300_SWIZZLE_SELECT_Z_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_W << R300_SWIZZLE_SELECT_W_1_SHIFT) |
-			   ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			    << R300_WRITE_ENA_1_SHIFT)));
-	} else {
+	if (info->has_tcl)
 	    BEGIN_VIDEO(5);
-	    OUT_VIDEO_REG(R300_VAP_PROG_STREAM_CNTL_0,
-			  ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
-			   (0 << R300_SKIP_DWORDS_0_SHIFT) |
-			   (0 << R300_DST_VEC_LOC_0_SHIFT) |
-			   R300_SIGNED_0 |
-			   (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
-			   (0 << R300_SKIP_DWORDS_1_SHIFT) |
-			   (6 << R300_DST_VEC_LOC_1_SHIFT) |
-			   R300_LAST_VEC_1 |
-			   R300_SIGNED_1));
-	    OUT_VIDEO_REG(R300_VAP_PROG_STREAM_CNTL_EXT_0,
-			  ((R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_0_SHIFT) |
-			   ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			    << R300_WRITE_ENA_0_SHIFT) |
-			   (R300_SWIZZLE_SELECT_X << R300_SWIZZLE_SELECT_X_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_Y << R300_SWIZZLE_SELECT_Y_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_FP_ZERO << R300_SWIZZLE_SELECT_Z_1_SHIFT) |
-			   (R300_SWIZZLE_SELECT_FP_ONE << R300_SWIZZLE_SELECT_W_1_SHIFT) |
-			   ((R300_WRITE_ENA_X | R300_WRITE_ENA_Y | R300_WRITE_ENA_Z | R300_WRITE_ENA_W)
-			    << R300_WRITE_ENA_1_SHIFT)));
-	}
+	else
+	    BEGIN_VIDEO(3);
+
+	OUT_VIDEO_REG(R300_VAP_PROG_STREAM_CNTL_0,
+		      ((R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_0_SHIFT) |
+		       (0 << R300_SKIP_DWORDS_0_SHIFT) |
+		       (0 << R300_DST_VEC_LOC_0_SHIFT) |
+		       R300_SIGNED_0 |
+		       (R300_DATA_TYPE_FLOAT_2 << R300_DATA_TYPE_1_SHIFT) |
+		       (0 << R300_SKIP_DWORDS_1_SHIFT) |
+		       (6 << R300_DST_VEC_LOC_1_SHIFT) |
+		       R300_LAST_VEC_1 |
+		       R300_SIGNED_1));
 
 	/* load the vertex shader */
 	if (info->has_tcl) {