R3xx/R5xx: move more VAP, etc. state setup into common init3d() function

Also some minor code cleanups

1 files changed, 52 insertions, 263 deletions

diff --git a/src/radeon_exa_render.c b/src/radeon_exa_render.c
index d968dc89..9496bb66 100644
--- a/src/radeon_exa_render.c
+++ b/src/radeon_exa_render.c
@@ -1036,10 +1036,6 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
     CARD32 txenable, colorpitch;
     CARD32 blendcntl;
     int pixel_shift;
-    int has_tcl = 0; /*((info->ChipFamily != CHIP_FAMILY_RS690) &&
-		   (info->ChipFamily != CHIP_FAMILY_RS740) &&
-		   (info->ChipFamily != CHIP_FAMILY_RS400) &&
-		   (info->ChipFamily != CHIP_FAMILY_RV515)); */
     ACCEL_PREAMBLE();
 
     TRACE;
@@ -1080,234 +1076,49 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 
     RADEON_SWITCH_TO_3D();
 
-    /* setup the VAP */
-    if (has_tcl) {
-	BEGIN_ACCEL(9);
-	OUT_ACCEL_REG(R300_VAP_CNTL_STATUS, 0);
-	OUT_ACCEL_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0);
-	OUT_ACCEL_REG(R300_VAP_CNTL, ((6 << R300_PVS_NUM_SLOTS_SHIFT) |
-				      (5 << R300_PVS_NUM_CNTLRS_SHIFT) |
-				      (4 << R300_PVS_NUM_FPUS_SHIFT) |
-				      (12 << R300_VF_MAX_VTX_NUM_SHIFT)));
-    } else {
-	BEGIN_ACCEL(8);
-	OUT_ACCEL_REG(R300_VAP_CNTL_STATUS, R300_PVS_BYPASS);
-	OUT_ACCEL_REG(R300_VAP_CNTL, ((10 << R300_PVS_NUM_SLOTS_SHIFT) |
-				      (5 << R300_PVS_NUM_CNTLRS_SHIFT) |
-				      (4 << R300_PVS_NUM_FPUS_SHIFT) |
-				      (5 << R300_VF_MAX_VTX_NUM_SHIFT)));
-    }
-
-    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);
-
-    if (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();
-
-    /* setup the vertex shader */
-    if (has_tcl) {
-	if (pMask) {
-	    BEGIN_ACCEL(22);
-	    /* flush the PVS before updating??? */
-	    OUT_ACCEL_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0);
-
-	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_0,
-			  ((0 << R300_PVS_FIRST_INST_SHIFT) |
-			   (2 << R300_PVS_XYZW_VALID_INST_SHIFT) |
-			   (2 << R300_PVS_LAST_INST_SHIFT)));
-	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_1,
-			  (2 << R300_PVS_LAST_VTX_SRC_INST_SHIFT));
-	} else {
-	    BEGIN_ACCEL(18);
-	    /* flush the PVS before updating??? */
-	    OUT_ACCEL_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0);
-
-	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_0,
-			  ((0 << R300_PVS_FIRST_INST_SHIFT) |
-			   (1 << R300_PVS_XYZW_VALID_INST_SHIFT) |
-			   (1 << R300_PVS_LAST_INST_SHIFT)));
-	    OUT_ACCEL_REG(R300_VAP_PVS_CODE_CNTL_1,
-			  (1 << R300_PVS_LAST_VTX_SRC_INST_SHIFT));
-	}
-	OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
-	/* PVS inst 0 */
-	OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
-		      (R300_PVS_DST_OPCODE(R300_VE_ADD) |
-		       R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) |
-		       R300_PVS_DST_OFFSET(0) |
-		       R300_PVS_DST_WE_X | R300_PVS_DST_WE_Y |
-		       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(0) |
-		       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(0) |
-		       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(0) |
-		       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)));
-
-	/* PVS inst 1 */
-	OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
-		      (R300_PVS_DST_OPCODE(R300_VE_ADD) |
-		       R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) |
-		       R300_PVS_DST_OFFSET(1) |
-		       R300_PVS_DST_WE_X | R300_PVS_DST_WE_Y |
-		       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_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_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_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)));
-
-	if (pMask) {
-	    /* PVS inst 2 */
-	    OUT_ACCEL_REG(R300_VAP_PVS_VECTOR_DATA_REG,
-			  (R300_PVS_DST_OPCODE(R300_VE_ADD) |
-			   R300_PVS_DST_REG_TYPE(R300_PVS_DST_REG_OUT) |
-			   R300_PVS_DST_OFFSET(2) |
-			   R300_PVS_DST_WE_X | R300_PVS_DST_WE_Y |
-			   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_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_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_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_FLOW_CNTL_OPC, 0);
-
-	OUT_ACCEL_REG(R300_VAP_GB_VERT_CLIP_ADJ, 0x3f800000);
-	OUT_ACCEL_REG(R300_VAP_GB_VERT_DISC_ADJ, 0x3f800000);
-	OUT_ACCEL_REG(R300_VAP_GB_HORZ_CLIP_ADJ, 0x3f800000);
-	OUT_ACCEL_REG(R300_VAP_GB_HORZ_DISC_ADJ, 0x3f800000);
-	OUT_ACCEL_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
-	FINISH_ACCEL();
-    }
+    BEGIN_ACCEL(8);
+    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)));
 
-    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) |
 		   (2 << R300_TEX_1_COMP_CNT_SHIFT)));
 
-    OUT_ACCEL_REG(R300_TX_INVALTAGS, 0x0);
+    OUT_ACCEL_REG(R300_TX_INVALTAGS, 0);
     OUT_ACCEL_REG(R300_TX_ENABLE, txenable);
     FINISH_ACCEL();
 
@@ -1394,9 +1205,10 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 			  R300_OUT_FMT_C2_SEL_BLUE |
 			  R300_OUT_FMT_C3_SEL_ALPHA);
 	    break;
-	case PICT_r5g6b5:
 	case PICT_a1r5g5b5:
 	case PICT_x1r5g5b5:
+	    /* fix me */
+	case PICT_r5g6b5:
 	    output_fmt = (R300_OUT_FMT_C_5_6_5 |
 			  R300_OUT_FMT_C0_SEL_BLUE |
 			  R300_OUT_FMT_C1_SEL_GREEN |
@@ -1567,103 +1379,80 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 	CARD32 mask_color, mask_alpha;
 
 	if (PICT_FORMAT_RGB(pSrcPicture->format) == 0)
-	    //src_color = R300_ALU_RGB_0_0;
 	    src_color = (R500_ALU_RGB_R_SWIZ_A_0 |
 			 R500_ALU_RGB_G_SWIZ_A_0 |
 			 R500_ALU_RGB_B_SWIZ_A_0);
 	else
-	    //src_color = R300_ALU_RGB_SRC0_RGB;
 	    src_color = (R500_ALU_RGB_R_SWIZ_A_R |
 			 R500_ALU_RGB_G_SWIZ_A_G |
 			 R500_ALU_RGB_B_SWIZ_A_B);
 
 	if (PICT_FORMAT_A(pSrcPicture->format) == 0)
-	    //src_alpha = R300_ALU_ALPHA_1_0;
 	    src_alpha = R500_ALPHA_SWIZ_A_1;
 	else
-	    //src_alpha = R300_ALU_ALPHA_SRC0_A;
 	    src_alpha = R500_ALPHA_SWIZ_A_A;
 
 	if (pMask && pMaskPicture->componentAlpha) {
 	    if (RadeonBlendOp[op].src_alpha) {
 		if (PICT_FORMAT_A(pSrcPicture->format) == 0) {
-		    //src_color = R300_ALU_RGB_1_0;
-		    //src_alpha = R300_ALU_ALPHA_1_0;
 		    src_color = (R500_ALU_RGB_R_SWIZ_A_1 |
 				 R500_ALU_RGB_G_SWIZ_A_1 |
 				 R500_ALU_RGB_B_SWIZ_A_1);
 		    src_alpha = R500_ALPHA_SWIZ_A_1;
 		} else {
-		    //src_color = R300_ALU_RGB_SRC0_AAA;
-		    //src_alpha = R300_ALU_ALPHA_SRC0_A;
 		    src_color = (R500_ALU_RGB_R_SWIZ_A_A |
 				 R500_ALU_RGB_G_SWIZ_A_A |
 				 R500_ALU_RGB_B_SWIZ_A_A);
 		    src_alpha = R500_ALPHA_SWIZ_A_A;
 		}
 
-		//mask_color = R300_ALU_RGB_SRC1_RGB;
 		mask_color = (R500_ALU_RGB_R_SWIZ_B_R |
 			      R500_ALU_RGB_G_SWIZ_B_G |
 			      R500_ALU_RGB_B_SWIZ_B_B);
 
 		if (PICT_FORMAT_A(pMaskPicture->format) == 0)
-		    //mask_alpha = R300_ALU_ALPHA_1_0;
 		    mask_alpha = R500_ALPHA_SWIZ_B_1;
 		else
-		    //mask_alpha = R300_ALU_ALPHA_SRC1_A;
 		    mask_alpha = R500_ALPHA_SWIZ_B_A;
 
 	    } else {
-		//src_color = R300_ALU_RGB_SRC0_RGB;
 		src_color = (R500_ALU_RGB_R_SWIZ_A_R |
 			     R500_ALU_RGB_G_SWIZ_A_G |
 			     R500_ALU_RGB_B_SWIZ_A_B);
 
 		if (PICT_FORMAT_A(pSrcPicture->format) == 0)
-		    //src_alpha = R300_ALU_ALPHA_1_0;
 		    src_alpha = R500_ALPHA_SWIZ_A_1;
 		else
-		    //src_alpha = R300_ALU_ALPHA_SRC0_A;
 		    src_alpha = R500_ALPHA_SWIZ_A_A;
 
-		//mask_color = R300_ALU_RGB_SRC1_RGB;
 		mask_color = (R500_ALU_RGB_R_SWIZ_B_R |
 			      R500_ALU_RGB_G_SWIZ_B_G |
 			      R500_ALU_RGB_B_SWIZ_B_B);
 
 		if (PICT_FORMAT_A(pMaskPicture->format) == 0)
-		    //mask_alpha = R300_ALU_ALPHA_1_0;
 		    mask_alpha = R500_ALPHA_SWIZ_B_1;
 		else
-		    //mask_alpha = R300_ALU_ALPHA_SRC1_A;
 		    mask_alpha = R500_ALPHA_SWIZ_B_A;
 
 	    }
 	} else if (pMask) {
 	    if (PICT_FORMAT_A(pMaskPicture->format) == 0)
-		//mask_color = R300_ALU_RGB_1_0;
 		mask_color = (R500_ALU_RGB_R_SWIZ_B_1 |
 			      R500_ALU_RGB_G_SWIZ_B_1 |
 			      R500_ALU_RGB_B_SWIZ_B_1);
 	    else
-		//mask_color = R300_ALU_RGB_SRC1_AAA;
 		mask_color = (R500_ALU_RGB_R_SWIZ_B_A |
 			      R500_ALU_RGB_G_SWIZ_B_A |
 			      R500_ALU_RGB_B_SWIZ_B_A);
 
 	    if (PICT_FORMAT_A(pMaskPicture->format) == 0)
-		//mask_alpha = R300_ALU_ALPHA_1_0;
 		mask_alpha = R500_ALPHA_SWIZ_B_1;
 	    else
-		//mask_alpha = R300_ALU_ALPHA_SRC1_A;
 		mask_alpha = R500_ALPHA_SWIZ_B_A;
 	} else {
-	    //mask_color = R300_ALU_RGB_1_0;
 	    mask_color = (R500_ALU_RGB_R_SWIZ_B_1 |
 			  R500_ALU_RGB_G_SWIZ_B_1 |
 			  R500_ALU_RGB_B_SWIZ_B_1);
-	    //mask_alpha = R300_ALU_ALPHA_1_0;
 	    mask_alpha = R500_ALPHA_SWIZ_B_1;
 	}
 
@@ -1686,9 +1475,10 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 			  R300_OUT_FMT_C2_SEL_BLUE |
 			  R300_OUT_FMT_C3_SEL_ALPHA);
 	    break;
-	case PICT_r5g6b5:
 	case PICT_a1r5g5b5:
 	case PICT_x1r5g5b5:
+	    /* fix me */
+	case PICT_r5g6b5:
 	    output_fmt = (R300_OUT_FMT_C_5_6_5 |
 			  R300_OUT_FMT_C0_SEL_BLUE |
 			  R300_OUT_FMT_C1_SEL_GREEN |
@@ -1746,7 +1536,7 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 					 (R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
 					 (R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT)));
 
-	    OUT_ACCEL_REG(R300_RS_INST_COUNT, R300_INST_COUNT_RS(1) | R300_TX_OFFSET_RS(6));
+	    OUT_ACCEL_REG(R300_RS_INST_COUNT, R300_INST_COUNT_RS(0) | R300_TX_OFFSET_RS(6));
 
 	    /* src tex */
 	    OUT_ACCEL_REG(R500_RS_INST_0, ((0 << R500_RS_INST_TEX_ID_SHIFT) |
@@ -1798,9 +1588,9 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 						   R500_DY_S_SWIZ_R |
 						   R500_DY_T_SWIZ_R |
 						   R500_DY_R_SWIZ_R |
-						   R500_DY_Q_SWIZ_R)); // TEX_ADDR_DXDY
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
+						   R500_DY_Q_SWIZ_R));
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
 
 	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, (R500_INST_TYPE_TEX |
 						   R500_INST_TEX_SEM_WAIT |
@@ -1833,9 +1623,9 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 						   R500_DY_S_SWIZ_R |
 						   R500_DY_T_SWIZ_R |
 						   R500_DY_R_SWIZ_R |
-						   R500_DY_Q_SWIZ_R)); // TEX_ADDR_DXDY
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
+						   R500_DY_Q_SWIZ_R));
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
 	} else {
 	    BEGIN_ACCEL(13);
 	    OUT_ACCEL_REG(R500_GA_US_VECTOR_INDEX, 0);
@@ -1870,9 +1660,9 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 						   R500_DY_S_SWIZ_R |
 						   R500_DY_T_SWIZ_R |
 						   R500_DY_R_SWIZ_R |
-						   R500_DY_Q_SWIZ_R)); // TEX_ADDR_DXDY
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
-	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000); // mbz
+						   R500_DY_Q_SWIZ_R));
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
+	    OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, 0x00000000);
 	}
 
 	OUT_ACCEL_REG(R500_GA_US_VECTOR_DATA, (R500_INST_TYPE_OUT |
@@ -1916,14 +1706,13 @@ static Bool FUNC_NAME(R300PrepareComposite)(int op, PicturePtr pSrcPicture,
 	FINISH_ACCEL();
     }
 
-    BEGIN_ACCEL(4);
+    BEGIN_ACCEL(3);
 
     OUT_ACCEL_REG(R300_RB3D_COLOROFFSET0, dst_offset);
     OUT_ACCEL_REG(R300_RB3D_COLORPITCH0, colorpitch);
 
     blendcntl = RADEONGetBlendCntl(op, pMaskPicture, pDstPicture->format);
     OUT_ACCEL_REG(R300_RB3D_BLENDCNTL, blendcntl | R300_ALPHA_BLEND_ENABLE | R300_READ_ENABLE);
-    OUT_ACCEL_REG(R300_RB3D_ABLENDCNTL, 0);
 
     FINISH_ACCEL();