Merge remote branch 'origin/master' into lp-setup-llvm

Conflicts:
	src/gallium/drivers/llvmpipe/lp_setup_coef.c
	src/gallium/drivers/llvmpipe/lp_setup_coef.h
	src/gallium/drivers/llvmpipe/lp_setup_coef_intrin.c
	src/gallium/drivers/llvmpipe/lp_setup_point.c
	src/gallium/drivers/llvmpipe/lp_setup_tri.c
	src/gallium/drivers/llvmpipe/lp_state_derived.c
	src/gallium/drivers/llvmpipe/lp_state_fs.h

21 files changed, 604 insertions, 398 deletions

diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
index 3b2b06fc2b..8be32ea91f 100644
--- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
@@ -74,7 +74,7 @@ static void use_temporary(struct r300_fragment_program_code *code, unsigned int
 		code->pixsize = index;
 }
 
-static unsigned int use_source(struct r300_fragment_program_code* code, struct radeon_pair_instruction_source src)
+static unsigned int use_source(struct r300_fragment_program_code* code, struct rc_pair_instruction_source src)
 {
 	if (src.File == RC_FILE_CONSTANT) {
 		return src.Index | (1 << 5);
@@ -133,6 +133,8 @@ static unsigned int translate_alpha_opcode(struct r300_fragment_program_compiler
  */
 static int emit_alu(struct r300_emit_state * emit, struct rc_pair_instruction* inst)
 {
+	int ip;
+	int j;
 	PROG_CODE;
 
 	if (code->alu.length >= c->Base.max_alu_insts) {
@@ -140,20 +142,20 @@ static int emit_alu(struct r300_emit_state * emit, struct rc_pair_instruction* i
 		return 0;
 	}
 
-	int ip = code->alu.length++;
-	int j;
+	ip = code->alu.length++;
 
 	code->alu.inst[ip].rgb_inst = translate_rgb_opcode(c, inst->RGB.Opcode);
 	code->alu.inst[ip].alpha_inst = translate_alpha_opcode(c, inst->Alpha.Opcode);
 
 	for(j = 0; j < 3; ++j) {
 		unsigned int src = use_source(code, inst->RGB.Src[j]);
+		unsigned int arg;
 		code->alu.inst[ip].rgb_addr |= src << (6*j);
 
 		src = use_source(code, inst->Alpha.Src[j]);
 		code->alu.inst[ip].alpha_addr |= src << (6*j);
 
-		unsigned int arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle);
+		arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle);
 		arg |= inst->RGB.Arg[j].Abs << 6;
 		arg |= inst->RGB.Arg[j].Negate << 5;
 		code->alu.inst[ip].rgb_inst |= arg << (7*j);
@@ -259,6 +261,10 @@ static int finish_node(struct r300_emit_state * emit)
 {
 	struct r300_fragment_program_compiler * c = emit->compiler;
 	struct r300_fragment_program_code *code = &emit->compiler->code->code.r300;
+	unsigned alu_offset;
+	unsigned alu_end;
+	unsigned tex_offset;
+	unsigned tex_end;
 
 	if (code->alu.length == emit->node_first_alu) {
 		/* Generate a single NOP for this node */
@@ -268,10 +274,10 @@ static int finish_node(struct r300_emit_state * emit)
 			return 0;
 	}
 
-	unsigned alu_offset = emit->node_first_alu;
-	unsigned alu_end = code->alu.length - alu_offset - 1;
-	unsigned tex_offset = emit->node_first_tex;
-	unsigned tex_end = code->tex.length - tex_offset - 1;
+	alu_offset = emit->node_first_alu;
+	alu_end = code->alu.length - alu_offset - 1;
+	tex_offset = emit->node_first_tex;
+	tex_end = code->tex.length - tex_offset - 1;
 
 	if (code->tex.length == emit->node_first_tex) {
 		if (emit->current_node > 0) {
@@ -334,6 +340,9 @@ static int begin_tex(struct r300_emit_state * emit)
 
 static int emit_tex(struct r300_emit_state * emit, struct rc_instruction * inst)
 {
+	unsigned int unit;
+	unsigned int dest;
+	unsigned int opcode;
 	PROG_CODE;
 
 	if (code->tex.length >= R300_PFS_MAX_TEX_INST) {
@@ -341,9 +350,8 @@ static int emit_tex(struct r300_emit_state * emit, struct rc_instruction * inst)
 		return 0;
 	}
 
-	unsigned int unit = inst->U.I.TexSrcUnit;
-	unsigned int dest = inst->U.I.DstReg.Index;
-	unsigned int opcode;
+	unit = inst->U.I.TexSrcUnit;
+	dest = inst->U.I.DstReg.Index;
 
 	switch(inst->U.I.Opcode) {
 	case RC_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break;
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c
index caa48fe478..2d28b06539 100644
--- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c
@@ -94,6 +94,9 @@ static const struct swizzle_data* lookup_native_swizzle(unsigned int swizzle)
  */
 static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
 {
+	unsigned int relevant;
+	int j;
+
 	if (reg.Abs)
 		reg.Negate = RC_MASK_NONE;
 
@@ -101,8 +104,6 @@ static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
 	    opcode == RC_OPCODE_TEX ||
 	    opcode == RC_OPCODE_TXB ||
 	    opcode == RC_OPCODE_TXP) {
-		int j;
-
 		if (reg.Abs || reg.Negate)
 			return 0;
 
@@ -117,8 +118,7 @@ static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg)
 		return 1;
 	}
 
-	unsigned int relevant = 0;
-	int j;
+	relevant = 0;
 
 	for(j = 0; j < 3; ++j)
 		if (GET_SWZ(reg.Swizzle, j) != RC_SWIZZLE_UNUSED)
@@ -154,9 +154,10 @@ static void r300_swizzle_split(
 			unsigned int matchcount = 0;
 			unsigned int matchmask = 0;
 			for(comp = 0; comp < 3; ++comp) {
+				unsigned int swz;
 				if (!GET_BIT(mask, comp))
 					continue;
-				unsigned int swz = GET_SWZ(src.Swizzle, comp);
+				swz = GET_SWZ(src.Swizzle, comp);
 				if (swz == RC_SWIZZLE_UNUSED)
 					continue;
 				if (swz == GET_SWZ(sd->hash, comp)) {
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
index 08785716db..d7d49e514b 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -392,9 +392,9 @@ static void ei_if(struct r300_vertex_program_compiler * compiler,
 	 * don't already have one. */
 	if (!compiler->PredicateMask) {
 		unsigned int writemasks[RC_REGISTER_MAX_INDEX];
-		memset(writemasks, 0, sizeof(writemasks));
 		struct rc_instruction * inst;
 		unsigned int i;
+		memset(writemasks, 0, sizeof(writemasks));
 		for(inst = compiler->Base.Program.Instructions.Next;
 				inst != &compiler->Base.Program.Instructions;
 							inst = inst->Next) {
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
index 627ce374ef..289bb87ae5 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
@@ -39,10 +39,12 @@ int r500_transform_IF(
 	struct rc_instruction * inst,
 	void* data)
 {
+	struct rc_instruction * inst_mov;
+
 	if (inst->U.I.Opcode != RC_OPCODE_IF)
 		return 0;
 
-	struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+	inst_mov = rc_insert_new_instruction(c, inst->Prev);
 	inst_mov->U.I.Opcode = RC_OPCODE_MOV;
 	inst_mov->U.I.DstReg.WriteMask = 0;
 	inst_mov->U.I.WriteALUResult = RC_ALURESULT_W;
@@ -251,12 +253,11 @@ void r500FragmentProgramDump(struct radeon_compiler *c, void *user)
 {
   struct r300_fragment_program_compiler *compiler = (struct r300_fragment_program_compiler*)c;
   struct r500_fragment_program_code *code = &compiler->code->code.r500;
-  fprintf(stderr, "R500 Fragment Program:\n--------\n");
-
   int n, i;
   uint32_t inst;
   uint32_t inst0;
   char *str = NULL;
+  fprintf(stderr, "R500 Fragment Program:\n--------\n");
 
   for (n = 0; n < code->inst_end+1; n++) {
     inst0 = inst = code->inst[n].inst0;
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
index 54cff9169a..6f101c68eb 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
@@ -198,7 +198,7 @@ static void use_temporary(struct r500_fragment_program_code* code, unsigned int
 		code->max_temp_idx = index;
 }
 
-static unsigned int use_source(struct r500_fragment_program_code* code, struct radeon_pair_instruction_source src)
+static unsigned int use_source(struct r500_fragment_program_code* code, struct rc_pair_instruction_source src)
 {
 	if (src.File == RC_FILE_CONSTANT) {
 		return src.Index | 0x100;
@@ -227,6 +227,7 @@ static void alu_nop(struct r300_fragment_program_compiler *c, int ip)
  */
 static void emit_paired(struct r300_fragment_program_compiler *c, struct rc_pair_instruction *inst)
 {
+	int ip;
 	PROG_CODE;
 
 	if (code->inst_end >= c->Base.max_alu_insts-1) {
@@ -234,7 +235,7 @@ static void emit_paired(struct r300_fragment_program_compiler *c, struct rc_pair
 		return;
 	}
 
-	int ip = ++code->inst_end;
+	ip = ++code->inst_end;
 
 	/* Quirk: MDH/MDV (DDX/DDY) need a NOP on previous non-TEX instructions. */
 	if (inst->RGB.Opcode == RC_OPCODE_DDX || inst->Alpha.Opcode == RC_OPCODE_DDX ||
@@ -250,7 +251,7 @@ static void emit_paired(struct r300_fragment_program_compiler *c, struct rc_pair
 	if (inst->RGB.OutputWriteMask || inst->Alpha.OutputWriteMask || inst->Alpha.DepthWriteMask) {
 		code->inst[ip].inst0 = R500_INST_TYPE_OUT;
 		if (inst->WriteALUResult) {
-			error("%s: cannot write output and ALU result at the same time");
+			error("Cannot write output and ALU result at the same time");
 			return;
 		}
 	} else {
@@ -357,6 +358,7 @@ static unsigned int translate_strq_swizzle(unsigned int swizzle)
  */
 static int emit_tex(struct r300_fragment_program_compiler *c, struct rc_sub_instruction *inst)
 {
+	int ip;
 	PROG_CODE;
 
 	if (code->inst_end >= c->Base.max_alu_insts-1) {
@@ -364,7 +366,7 @@ static int emit_tex(struct r300_fragment_program_compiler *c, struct rc_sub_inst
 		return 0;
 	}
 
-	int ip = ++code->inst_end;
+	ip = ++code->inst_end;
 
 	code->inst[ip].inst0 = R500_INST_TYPE_TEX
 		| (inst->DstReg.WriteMask << 11)
@@ -407,12 +409,14 @@ static int emit_tex(struct r300_fragment_program_compiler *c, struct rc_sub_inst
 
 static void emit_flowcontrol(struct emit_state * s, struct rc_instruction * inst)
 {
+	unsigned int newip;
+
 	if (s->Code->inst_end >= s->C->max_alu_insts-1) {
 		rc_error(s->C, "emit_tex: Too many instructions");
 		return;
 	}
 
-	unsigned int newip = ++s->Code->inst_end;
+	newip = ++s->Code->inst_end;
 
 	/* Currently all loops use the same integer constant to intialize
 	 * the loop variables. */
@@ -623,6 +627,8 @@ void r500BuildFragmentProgramHwCode(struct radeon_compiler *c, void *user)
 
 	if (code->inst_end == -1 ||
 	    (code->inst[code->inst_end].inst0 & R500_INST_TYPE_MASK) != R500_INST_TYPE_OUT) {
+		int ip;
+
 		/* This may happen when dead-code elimination is disabled or
 		 * when most of the fragment program logic is leading to a KIL */
 		if (code->inst_end >= compiler->Base.max_alu_insts-1) {
@@ -630,7 +636,7 @@ void r500BuildFragmentProgramHwCode(struct radeon_compiler *c, void *user)
 			return;
 		}
 
-		int ip = ++code->inst_end;
+		ip = ++code->inst_end;
 		code->inst[ip].inst0 = R500_INST_TYPE_OUT | R500_INST_TEX_SEM_WAIT;
 	}
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.c b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
index 0eab18c344..6842fb873b 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_code.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
@@ -28,6 +28,7 @@
 #include "radeon_code.h"
 
 #include <stdlib.h>
+#include <stdio.h>
 #include <string.h>
 
 #include "radeon_program.h"
@@ -171,3 +172,16 @@ unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float da
 
 	return rc_constants_add(c, &constant);
 }
+
+void rc_constants_print(struct rc_constant_list * c)
+{
+	unsigned int i;
+	for(i = 0; i < c->Count; i++) {
+		if (c->Constants[i].Type == RC_CONSTANT_IMMEDIATE) {
+			float * values = c->Constants[i].u.Immediate;
+			fprintf(stderr, "CONST[%u] = "
+				"{ %10.4f %10.4f %10.4f %10.4f }\n",
+				i, values[0],values[1], values[2], values[3]);
+		}
+	}
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.h b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
index f76676fae8..cfb6df2cd7 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_code.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
@@ -60,6 +60,7 @@ enum {
 
 	RC_STATE_R300_WINDOW_DIMENSION,
 	RC_STATE_R300_TEXRECT_FACTOR,
+	RC_STATE_R300_TEXSCALE_FACTOR,
 	RC_STATE_R300_VIEWPORT_SCALE,
 	RC_STATE_R300_VIEWPORT_OFFSET
 };
@@ -89,6 +90,7 @@ unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * cons
 unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state1, unsigned state2);
 unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data);
 unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle);
+void rc_constants_print(struct rc_constant_list * c);
 
 /**
  * Compare functions.
@@ -126,11 +128,7 @@ struct r300_fragment_program_external_state {
 	struct {
 		/**
 		 * If the sampler is used as a shadow sampler,
-		 * this field is:
-		 *  0 - GL_LUMINANCE
-		 *  1 - GL_INTENSITY
-		 *  2 - GL_ALPHA
-		 * depending on the depth texture mode.
+		 * this field contains swizzle depending on the depth texture mode.
 		 */
 		unsigned depth_texture_swizzle:12;
 
@@ -150,13 +148,9 @@ struct r300_fragment_program_external_state {
 		unsigned compare_mode_enabled : 1;
 
 		/**
-		 * If the sampler needs to fake NPOT, this field is set.
-		 */
-		unsigned fake_npot : 1;
-
-		/**
-		 * If the sampler will recieve non-normalized coords,
-		 * this field is set.
+		 * If the sampler will receive non-normalized coords,
+		 * this field is set. The scaling factor is given by
+		 * RC_STATE_R300_TEXRECT_FACTOR.
 		 */
 		unsigned non_normalized_coords : 1;
 
@@ -166,7 +160,13 @@ struct r300_fragment_program_external_state {
 		 * If this field is \ref RC_WRAP_NONE (aka 0), no wrapping maths
 		 * will be performed on the coordinates.
 		 */
-		unsigned wrap_mode : 2;
+		unsigned wrap_mode : 3;
+
+		/**
+		 * The coords are scaled after applying the wrap mode emulation
+		 * and right before texture fetch. The scaling factor is given by
+		 * RC_STATE_R300_TEXSCALE_FACTOR. */
+		unsigned clamp_and_scale_before_fetch : 1;
 	} unit[16];
 };
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c
index e73700f84a..a27d395587 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c
@@ -29,49 +29,32 @@
 
 #include "radeon_program.h"
 
+struct read_write_mask_data {
+	void * UserData;
+	rc_read_write_mask_fn Cb;
+};
+
 static void reads_normal_callback(
-	rc_read_write_chan_fn cb,
+	void * userdata,
 	struct rc_instruction * fullinst,
-	struct rc_src_register src,
-	void * userdata)
+	struct rc_src_register * src)
 {
+	struct read_write_mask_data * cb_data = userdata;
 	unsigned int refmask = 0;
 	unsigned int chan;
 	for(chan = 0; chan < 4; chan++) {
-		refmask |= 1 << GET_SWZ(src.Swizzle, chan);
+		refmask |= 1 << GET_SWZ(src->Swizzle, chan);
 	}
 	refmask &= RC_MASK_XYZW;
 
-	if (refmask)
-		cb(userdata, fullinst, src.File, src.Index, refmask);
-
-	if (refmask && src.RelAddr)
-		cb(userdata, fullinst, RC_FILE_ADDRESS, 0, RC_MASK_X);
-}
-
-static void reads_normal(struct rc_instruction * fullinst, rc_read_write_chan_fn cb, void * userdata)
-{
-	struct rc_sub_instruction * inst = &fullinst->U.I;
-	const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->Opcode);
-
-	for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) {
-
-		if (inst->SrcReg[src].File == RC_FILE_NONE)
-			return;
+	if (refmask) {
+		cb_data->Cb(cb_data->UserData, fullinst, src->File,
+							src->Index, refmask);
+	}
 
-		if (inst->SrcReg[src].File == RC_FILE_PRESUB) {
-			unsigned int i;
-			unsigned int srcp_regs = rc_presubtract_src_reg_count(
-							inst->PreSub.Opcode);
-			for( i = 0; i < srcp_regs; i++) {
-				reads_normal_callback(cb, fullinst,
-						inst->PreSub.SrcReg[i],
-						userdata);
-			}
-		} else {
-			reads_normal_callback(cb, fullinst,
-						inst->SrcReg[src], userdata);
-		}
+	if (refmask && src->RelAddr) {
+		cb_data->Cb(cb_data->UserData, fullinst, RC_FILE_ADDRESS, 0,
+								RC_MASK_X);
 	}
 }
 
@@ -112,7 +95,7 @@ static void pair_get_src_refmasks(unsigned int * refmasks,
 	}
 }
 
-static void reads_pair(struct rc_instruction * fullinst,  rc_read_write_mask_fn cb, void * userdata)
+static void reads_pair(struct rc_instruction * fullinst, rc_read_write_mask_fn cb, void * userdata)
 {
 	struct rc_pair_instruction * inst = &fullinst->U.P;
 	unsigned int refmasks[3] = { 0, 0, 0 };
@@ -143,6 +126,74 @@ static void reads_pair(struct rc_instruction * fullinst,  rc_read_write_mask_fn
 	}
 }
 
+static void pair_sub_for_all_args(
+	struct rc_instruction * fullinst,
+	struct rc_pair_sub_instruction * sub,
+	rc_pair_read_arg_fn cb,
+	void * userdata)
+{
+	int i;
+	const struct rc_opcode_info * info = rc_get_opcode_info(sub->Opcode);
+
+	for(i = 0; i < info->NumSrcRegs; i++) {
+		cb(userdata, fullinst, &sub->Arg[i]);
+	}
+}
+
+/* This function calls the callback function (cb) for each source used by
+ * the instruction.
+ * */
+void rc_for_all_reads_src(
+	struct rc_instruction * inst,
+	rc_read_src_fn cb,
+	void * userdata)
+{
+	const struct rc_opcode_info * opcode =
+					rc_get_opcode_info(inst->U.I.Opcode);
+
+	/* This function only works with normal instructions. */
+	if (inst->Type != RC_INSTRUCTION_NORMAL) {
+		assert(0);
+		return;
+	}
+
+	for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) {
+
+		if (inst->U.I.SrcReg[src].File == RC_FILE_NONE)
+			continue;
+
+		if (inst->U.I.SrcReg[src].File == RC_FILE_PRESUB) {
+			unsigned int i;
+			unsigned int srcp_regs = rc_presubtract_src_reg_count(
+						inst->U.I.PreSub.Opcode);
+			for( i = 0; i < srcp_regs; i++) {
+				cb(userdata, inst, &inst->U.I.PreSub.SrcReg[i]);
+			}
+		} else {
+			cb(userdata, inst, &inst->U.I.SrcReg[src]);
+		}
+	}
+}
+
+/**
+ * This function calls the callback function (cb) for each arg of the RGB and
+ * alpha components.
+ */
+void rc_pair_for_all_reads_arg(
+	struct rc_instruction * inst,
+	rc_pair_read_arg_fn cb,
+	void * userdata)
+{
+	/* This function only works with pair instructions. */
+	if (inst->Type != RC_INSTRUCTION_PAIR) {
+		assert(0);
+		return;
+	}
+
+	pair_sub_for_all_args(inst, &inst->U.P.RGB, cb, userdata);
+	pair_sub_for_all_args(inst, &inst->U.P.Alpha, cb, userdata);
+}
+
 /**
  * Calls a callback function for all register reads.
  *
@@ -153,7 +204,11 @@ static void reads_pair(struct rc_instruction * fullinst,  rc_read_write_mask_fn
 void rc_for_all_reads_mask(struct rc_instruction * inst, rc_read_write_mask_fn cb, void * userdata)
 {
 	if (inst->Type == RC_INSTRUCTION_NORMAL) {
-		reads_normal(inst, cb, userdata);
+		struct read_write_mask_data cb_data;
+		cb_data.UserData = userdata;
+		cb_data.Cb = cb;
+
+		rc_for_all_reads_src(inst, reads_normal_callback, &cb_data);
 	} else {
 		reads_pair(inst, cb, userdata);
 	}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h
index 795d9cc2b6..d10ae3c7b7 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h
@@ -33,6 +33,8 @@
 struct radeon_compiler;
 struct rc_instruction;
 struct rc_swizzle_caps;
+struct rc_src_register;
+struct rc_pair_instruction_arg;
 
 
 /**
@@ -49,6 +51,16 @@ typedef void (*rc_read_write_mask_fn)(void * userdata, struct rc_instruction * i
 void rc_for_all_reads_mask(struct rc_instruction * inst, rc_read_write_mask_fn cb, void * userdata);
 void rc_for_all_writes_mask(struct rc_instruction * inst, rc_read_write_mask_fn cb, void * userdata);
 
+typedef void (*rc_read_src_fn)(void * userdata, struct rc_instruction * inst,
+			struct rc_src_register * src);
+void rc_for_all_reads_src(struct rc_instruction * inst, rc_read_src_fn cb,
+			void * userdata);
+
+typedef void (*rc_pair_read_arg_fn)(void * userdata,
+	struct rc_instruction * inst, struct rc_pair_instruction_arg * arg);
+void rc_pair_for_all_reads_arg(struct rc_instruction * inst,
+					rc_pair_read_arg_fn cb, void * userdata);
+
 typedef void (*rc_remap_register_fn)(void * userdata, struct rc_instruction * inst,
 			rc_register_file * pfile, unsigned int * pindex);
 void rc_remap_registers(struct rc_instruction * inst, rc_remap_register_fn cb, void * userdata);
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
index 9d17b4772a..87906f37b1 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_deadcode.c
@@ -106,10 +106,12 @@ static void push_loop(struct deadcode_state * s)
 
 static void push_branch(struct deadcode_state * s)
 {
+	struct branchinfo * branch;
+
 	memory_pool_array_reserve(&s->C->Pool, struct branchinfo, s->BranchStack,
 			s->BranchStackSize, s->BranchStackReserved, 1);
 
-	struct branchinfo * branch = &s->BranchStack[s->BranchStackSize++];
+	branch = &s->BranchStack[s->BranchStackSize++];
 	branch->HaveElse = 0;
 	memcpy(&branch->StoreEndif, &s->R, sizeof(s->R));
 }
@@ -152,6 +154,7 @@ static void update_instruction(struct deadcode_state * s, struct rc_instruction
 	const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
 	struct instruction_state * insts = &s->Instructions[inst->IP];
 	unsigned int usedmask = 0;
+	unsigned int srcmasks[3];
 
 	if (opcode->HasDstReg) {
 		unsigned char * pused = get_used_ptr(s, inst->U.I.DstReg.File, inst->U.I.DstReg.Index);
@@ -180,7 +183,6 @@ static void update_instruction(struct deadcode_state * s, struct rc_instruction
 		}
 	}
 
-	unsigned int srcmasks[3];
 	rc_compute_sources_for_writemask(inst, usedmask, srcmasks);
 
 	for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) {
@@ -219,6 +221,7 @@ void rc_dataflow_deadcode(struct radeon_compiler * c, void *user)
 	unsigned int nr_instructions;
 	unsigned has_temp_reladdr_src = 0;
 	rc_dataflow_mark_outputs_fn dce = (rc_dataflow_mark_outputs_fn)user;
+	unsigned int ip;
 
 	/* Give up if there is relative addressing of destination operands. */
 	for(struct rc_instruction * inst = c->Program.Instructions.Next;
@@ -349,12 +352,14 @@ void rc_dataflow_deadcode(struct radeon_compiler * c, void *user)
 		update_instruction(&s, inst);
 	}
 
-	unsigned int ip = 0;
+	ip = 0;
 	for(struct rc_instruction * inst = c->Program.Instructions.Next;
 	    inst != &c->Program.Instructions;
 	    inst = inst->Next, ++ip) {
 		const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
 		int dead = 1;
+		unsigned int srcmasks[3];
+		unsigned int usemask;
 
 		if (!opcode->HasDstReg) {
 			dead = 0;
@@ -376,8 +381,7 @@ void rc_dataflow_deadcode(struct radeon_compiler * c, void *user)
 			continue;
 		}
 
-		unsigned int srcmasks[3];
-		unsigned int usemask = s.Instructions[ip].WriteMask;
+		usemask = s.Instructions[ip].WriteMask;
 
 		if (inst->U.I.WriteALUResult == RC_ALURESULT_X)
 			usemask |= RC_MASK_X;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_branches.c b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_branches.c
index 222e5b7e02..7bede344f3 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_branches.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_branches.c
@@ -57,18 +57,21 @@ struct emulate_branch_state {
 
 static void handle_if(struct emulate_branch_state * s, struct rc_instruction * inst)
 {
+	struct branch_info * branch;
+	struct rc_instruction * inst_mov;
+
 	memory_pool_array_reserve(&s->C->Pool, struct branch_info,
 			s->Branches, s->BranchCount, s->BranchReserved, 1);
 
 	DBG("%s\n", __FUNCTION__);
 
-	struct branch_info * branch = &s->Branches[s->BranchCount++];
+	branch = &s->Branches[s->BranchCount++];
 	memset(branch, 0, sizeof(struct branch_info));
 	branch->If = inst;
 
 	/* Make a safety copy of the decision register, because we will need
 	 * it at ENDIF time and it might be overwritten in both branches. */
-	struct rc_instruction * inst_mov = rc_insert_new_instruction(s->C, inst->Prev);
+	inst_mov = rc_insert_new_instruction(s->C, inst->Prev);
 	inst_mov->U.I.Opcode = RC_OPCODE_MOV;
 	inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
 	inst_mov->U.I.DstReg.Index = rc_find_free_temporary(s->C);
@@ -84,6 +87,8 @@ static void handle_if(struct emulate_branch_state * s, struct rc_instruction * i
 
 static void handle_else(struct emulate_branch_state * s, struct rc_instruction * inst)
 {
+	struct branch_info * branch;
+
 	if (!s->BranchCount) {
 		rc_error(s->C, "Encountered ELSE outside of branches");
 		return;
@@ -91,7 +96,7 @@ static void handle_else(struct emulate_branch_state * s, struct rc_instruction *
 
 	DBG("%s\n", __FUNCTION__);
 
-	struct branch_info * branch = &s->Branches[s->BranchCount - 1];
+	branch = &s->Branches[s->BranchCount - 1];
 	branch->Else = inst;
 }
 
@@ -191,6 +196,10 @@ static void inject_cmp(struct emulate_branch_state * s,
 
 static void handle_endif(struct emulate_branch_state * s, struct rc_instruction * inst)
 {
+	struct branch_info * branch;
+	struct register_proxies IfProxies;
+	struct register_proxies ElseProxies;
+
 	if (!s->BranchCount) {
 		rc_error(s->C, "Encountered ENDIF outside of branches");
 		return;
@@ -198,9 +207,7 @@ static void handle_endif(struct emulate_branch_state * s, struct rc_instruction
 
 	DBG("%s\n", __FUNCTION__);
 
-	struct branch_info * branch = &s->Branches[s->BranchCount - 1];
-	struct register_proxies IfProxies;
-	struct register_proxies ElseProxies;
+	branch = &s->Branches[s->BranchCount - 1];
 
 	memset(&IfProxies, 0, sizeof(IfProxies));
 	memset(&ElseProxies, 0, sizeof(ElseProxies));
@@ -261,16 +268,19 @@ static void remap_output_function(void * userdata, struct rc_instruction * inst,
  */
 static void fix_output_writes(struct emulate_branch_state * s, struct rc_instruction * inst)
 {
+	const struct rc_opcode_info * opcode;
+
 	if (!s->BranchCount)
 		return;
 
-	const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
+	opcode = rc_get_opcode_info(inst->U.I.Opcode);
 
 	if (!opcode->HasDstReg)
 		return;
 
 	if (inst->U.I.DstReg.File == RC_FILE_OUTPUT) {
 		struct remap_output_data remap;
+		struct rc_instruction * inst_mov;
 
 		remap.Output = inst->U.I.DstReg.Index;
 		remap.Temporary = rc_find_free_temporary(s->C);
@@ -281,7 +291,7 @@ static void fix_output_writes(struct emulate_branch_state * s, struct rc_instruc
 			rc_remap_registers(inst, &remap_output_function, &remap);
 		}
 
-		struct rc_instruction * inst_mov = rc_insert_new_instruction(s->C, s->C->Program.Instructions.Prev);
+		inst_mov = rc_insert_new_instruction(s->C, s->C->Program.Instructions.Prev);
 		inst_mov->U.I.Opcode = RC_OPCODE_MOV;
 		inst_mov->U.I.DstReg.File = RC_FILE_OUTPUT;
 		inst_mov->U.I.DstReg.Index = remap.Output;
@@ -299,12 +309,13 @@ static void fix_output_writes(struct emulate_branch_state * s, struct rc_instruc
 void rc_emulate_branches(struct radeon_compiler *c, void *user)
 {
 	struct emulate_branch_state s;
+	struct rc_instruction * ptr;
 
 	memset(&s, 0, sizeof(s));
 	s.C = c;
 
 	/* Untypical loop because we may remove the current instruction */
-	struct rc_instruction * ptr = c->Program.Instructions.Next;
+	ptr = c->Program.Instructions.Next;
 	while(ptr != &c->Program.Instructions) {
 		struct rc_instruction * inst = ptr;
 		ptr = ptr->Next;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
index cd4fcbabb9..205eecd112 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_emulate_loops.c
@@ -386,8 +386,6 @@ static int build_loop_info(struct radeon_compiler * c, struct loop_info * loop,
 			case RC_OPCODE_SNE:
 				break;
 			default:
-				rc_error(c, "%s: expected conditional",
-								__FUNCTION__);
 				return 0;
 			}
 			loop->Cond = loop->If->Prev;
@@ -431,8 +429,10 @@ static int transform_loop(struct emulate_loop_state * s,
 
 	loop = &s->Loops[s->LoopCount++];
 
-	if (!build_loop_info(s->C, loop, inst))
+	if (!build_loop_info(s->C, loop, inst)) {
+		rc_error(s->C, "Failed to build loop info\n");
 		return 0;
+	}
 
 	if(try_unroll_loop(s->C, loop)){
 		return 1;
@@ -511,11 +511,12 @@ void rc_emulate_loops(struct radeon_compiler *c, void *user)
 	 * loops are unrolled first.
 	 */
 	for( i = s->LoopCount - 1; i >= 0; i-- ){
+		unsigned int iterations;
+
 		if(!s->Loops[i].EndLoop){
 			continue;
 		}
-		unsigned int iterations = loop_max_possible_iterations(
-					s->C, &s->Loops[i]);
+		iterations = loop_max_possible_iterations(s->C, &s->Loops[i]);
 		unroll_loop(s->C, &s->Loops[i], iterations);
 	}
 }
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c b/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c
index c15a9b1c45..41769e347e 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_optimize.c
@@ -163,7 +163,8 @@ static void copy_propagate(struct radeon_compiler * c, struct rc_instruction * i
 
 	if (inst_mov->U.I.DstReg.File != RC_FILE_TEMPORARY ||
 	    inst_mov->U.I.DstReg.RelAddr ||
-	    inst_mov->U.I.WriteALUResult)
+	    inst_mov->U.I.WriteALUResult ||
+	    inst_mov->U.I.SaturateMode)
 		return;
 
 	memset(&s, 0, sizeof(s));
@@ -410,27 +411,34 @@ static void constant_folding(struct radeon_compiler * c, struct rc_instruction *
 
 	/* Replace 0.0, 1.0 and 0.5 immediates by their explicit swizzles */
 	for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) {
+		struct rc_constant * constant;
+		struct rc_src_register newsrc;
+		int have_real_reference;
+
 		if (inst->U.I.SrcReg[src].File != RC_FILE_CONSTANT ||
 		    inst->U.I.SrcReg[src].RelAddr ||
 		    inst->U.I.SrcReg[src].Index >= c->Program.Constants.Count)
 			continue;
 
-		struct rc_constant * constant =
+		constant =
 			&c->Program.Constants.Constants[inst->U.I.SrcReg[src].Index];
 
 		if (constant->Type != RC_CONSTANT_IMMEDIATE)
 			continue;
 
-		struct rc_src_register newsrc = inst->U.I.SrcReg[src];
-		int have_real_reference = 0;
+		newsrc = inst->U.I.SrcReg[src];
+		have_real_reference = 0;
 		for(unsigned int chan = 0; chan < 4; ++chan) {
 			unsigned int swz = GET_SWZ(newsrc.Swizzle, chan);
+			unsigned int newswz;
+			float imm;
+			float baseimm;
+
 			if (swz >= 4)
 				continue;
 
-			unsigned int newswz;
-			float imm = constant->u.Immediate[swz];
-			float baseimm = imm;
+			imm = constant->u.Immediate[swz];
+			baseimm = imm;
 			if (imm < 0.0)
 				baseimm = -baseimm;
 
@@ -475,8 +483,8 @@ static void constant_folding(struct radeon_compiler * c, struct rc_instruction *
 }
 
 /**
- * This function returns a writemask that indicates wich components are
- * read by src and also written by dst.
+ * If src and dst use the same register, this function returns a writemask that
+ * indicates wich components are read by src.  Otherwise zero is returned.
  */
 static unsigned int src_reads_dst_mask(struct rc_src_register src,
 						struct rc_dst_register dst)
@@ -563,10 +571,18 @@ static int presub_helper(
 		 * s->Inst->U.I.DstReg, because if it does we must not
 		 * remove s->Inst. */
 		for(i = 0; i < info->NumSrcRegs; i++) {
-			if(s->Inst->U.I.DstReg.WriteMask !=
-					src_reads_dst_mask(inst->U.I.SrcReg[i],
-						s->Inst->U.I.DstReg)) {
-				continue;
+			unsigned int mask = src_reads_dst_mask(
+				inst->U.I.SrcReg[i], s->Inst->U.I.DstReg);
+			/* XXX We could be more aggressive here using
+			 * presubtract.  It is okay if SrcReg[i] only reads
+			 * from some of the mask components. */
+			if(s->Inst->U.I.DstReg.WriteMask != mask) {
+				if (s->Inst->U.I.DstReg.WriteMask & mask) {
+					can_remove = 0;
+					break;
+				} else {
+					continue;
+				}
 			}
 			if (cant_sub || !can_use_presub) {
 				can_remove = 0;
@@ -626,6 +642,21 @@ static void presub_replace_add(struct peephole_state *s,
 	inst->U.I.SrcReg[src_index].Index = presub_opcode;
 }
 
+static int is_presub_candidate(struct rc_instruction * inst)
+{
+	const struct rc_opcode_info * info = rc_get_opcode_info(inst->U.I.Opcode);
+	unsigned int i;
+
+	if (inst->U.I.PreSub.Opcode != RC_PRESUB_NONE || inst->U.I.SaturateMode)
+		return 0;
+
+	for(i = 0; i < info->NumSrcRegs; i++) {
+		if (src_reads_dst_mask(inst->U.I.SrcReg[i], inst->U.I.DstReg))
+			return 0;
+	}
+	return 1;
+}
+
 static int peephole_add_presub_add(
 	struct radeon_compiler * c,
 	struct rc_instruction * inst_add)
@@ -635,10 +666,7 @@ static int peephole_add_presub_add(
 	unsigned int i;
 	struct peephole_state s;
 
-	if (inst_add->U.I.PreSub.Opcode != RC_PRESUB_NONE)
-		return 0;
-
-	if (inst_add->U.I.SaturateMode)
+	if (!is_presub_candidate(inst_add))
 		return 0;
 
 	if (inst_add->U.I.SrcReg[0].Swizzle != inst_add->U.I.SrcReg[1].Swizzle)
@@ -705,10 +733,7 @@ static int peephole_add_presub_inv(
 	unsigned int i, swz, mask;
 	struct peephole_state s;
 
-	if (inst_add->U.I.PreSub.Opcode != RC_PRESUB_NONE)
-		return 0;
-
-	if (inst_add->U.I.SaturateMode)
+	if (!is_presub_candidate(inst_add))
 		return 0;
 
 	mask = inst_add->U.I.DstReg.WriteMask;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_pair_regalloc.c b/src/mesa/drivers/dri/r300/compiler/radeon_pair_regalloc.c
index c73845512f..91524f5ec6 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_pair_regalloc.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_pair_regalloc.c
@@ -322,12 +322,13 @@ void rc_pair_regalloc_inputs_only(struct radeon_compiler *cc, void *user)
 {
 	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)cc;
 	struct regalloc_state s;
+	int temp_reg_offset;
 
 	compute_live_intervals(cc, &s);
 
 	c->AllocateHwInputs(c, &alloc_input, &s);
 
-	int temp_reg_offset = 0;
+	temp_reg_offset = 0;
 	for (unsigned i = 0; i < RC_REGISTER_MAX_INDEX; i++) {
 		if (s.Input[i].Allocated && temp_reg_offset <= s.Input[i].Index)
 			temp_reg_offset = s.Input[i].Index + 1;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_pair_schedule.c b/src/mesa/drivers/dri/r300/compiler/radeon_pair_schedule.c
index 5269d65985..bfe3553460 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_pair_schedule.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_pair_schedule.c
@@ -237,11 +237,12 @@ static void commit_alu_instruction(struct schedule_state * s, struct schedule_in
 static void emit_all_tex(struct schedule_state * s, struct rc_instruction * before)
 {
 	struct schedule_instruction *readytex;
+	struct rc_instruction * inst_begin;
 
 	assert(s->ReadyTEX);
 
 	/* Node marker for R300 */
-	struct rc_instruction * inst_begin = rc_insert_new_instruction(s->C, before->Prev);
+	inst_begin = rc_insert_new_instruction(s->C, before->Prev);
 	inst_begin->U.I.Opcode = RC_OPCODE_BEGIN_TEX;
 
 	/* Link texture instructions back in */
@@ -280,6 +281,8 @@ static int destructive_merge_instructions(
 		struct rc_pair_instruction * alpha)
 {
 	const struct rc_opcode_info * opcode;
+	const struct rc_opcode_info * rgb_info;
+
 	assert(rgb->Alpha.Opcode == RC_OPCODE_NOP);
 	assert(alpha->RGB.Opcode == RC_OPCODE_NOP);
 
@@ -288,8 +291,7 @@ static int destructive_merge_instructions(
 	 * src1. */
 
 	/* Merge the rgb presubtract registers. */
-	const struct rc_opcode_info * rgb_info =
-					rc_get_opcode_info(rgb->RGB.Opcode);
+	rgb_info = rc_get_opcode_info(rgb->RGB.Opcode);
 	if (alpha->RGB.Src[RC_PAIR_PRESUB_SRC].Used) {
 		unsigned int srcp_src;
 		unsigned int srcp_regs;
@@ -301,9 +303,9 @@ static int destructive_merge_instructions(
 			unsigned int arg;
 			int free_source;
 			unsigned int one_way = 0;
-			struct radeon_pair_instruction_source srcp =
+			struct rc_pair_instruction_source srcp =
 						alpha->RGB.Src[srcp_src];
-			struct radeon_pair_instruction_source temp;
+			struct rc_pair_instruction_source temp;
 			/* 2nd arg of 1 means this is an rgb source.
 			 * 3rd arg of 0 means this is not an alpha source. */
 			free_source = rc_pair_alloc_source(rgb, 1, 0,
@@ -366,9 +368,9 @@ static int destructive_merge_instructions(
 			unsigned int arg;
 			int free_source;
 			unsigned int one_way = 0;
-			struct radeon_pair_instruction_source srcp =
+			struct rc_pair_instruction_source srcp =
 						alpha->Alpha.Src[srcp_src];
-			struct radeon_pair_instruction_source temp;
+			struct rc_pair_instruction_source temp;
 			/* 2nd arg of 0 means this is not an rgb source.
 			 * 3rd arg of 1 means this is an alpha source. */
 			free_source = rc_pair_alloc_source(rgb, 0, 1,
@@ -424,6 +426,7 @@ static int destructive_merge_instructions(
 		unsigned int oldsrc = alpha->Alpha.Arg[arg].Source;
 		rc_register_file file = 0;
 		unsigned int index = 0;
+		int source;
 
 		if (alpha->Alpha.Arg[arg].Swizzle < 3) {
 			srcrgb = 1;
@@ -435,7 +438,7 @@ static int destructive_merge_instructions(
 			index = alpha->Alpha.Src[oldsrc].Index;
 		}
 
-		int source = rc_pair_alloc_source(rgb, srcrgb, srcalpha, file, index);
+		source = rc_pair_alloc_source(rgb, srcrgb, srcalpha, file, index);
 		if (source < 0)
 			return 0;
 
@@ -475,6 +478,12 @@ static int merge_instructions(struct rc_pair_instruction * rgb, struct rc_pair_i
 {
 	struct rc_pair_instruction backup;
 
+	/*Instructions can't write output registers and ALU result at the
+	 * same time. */
+	if ((rgb->WriteALUResult && alpha->Alpha.OutputWriteMask)
+		|| (rgb->RGB.OutputWriteMask && alpha->WriteALUResult)) {
+		return 0;
+	}
 	memcpy(&backup, rgb, sizeof(struct rc_pair_instruction));
 
 	if (destructive_merge_instructions(rgb, alpha))
@@ -597,6 +606,7 @@ static void scan_read(void * data, struct rc_instruction * inst,
 {
 	struct schedule_state * s = data;
 	struct reg_value * v = get_reg_value(s, file, index, chan);
+	struct reg_value_reader * reader;
 
 	if (!v)
 		return;
@@ -610,7 +620,7 @@ static void scan_read(void * data, struct rc_instruction * inst,
 
 	DBG("%i: read %i[%i] chan %i\n", s->Current->Instruction->IP, file, index, chan);
 
-	struct reg_value_reader * reader = memory_pool_malloc(&s->C->Pool, sizeof(*reader));
+	reader = memory_pool_malloc(&s->C->Pool, sizeof(*reader));
 	reader->Reader = s->Current;
 	reader->Next = v->Readers;
 	v->Readers = reader;
@@ -630,13 +640,14 @@ static void scan_write(void * data, struct rc_instruction * inst,
 {
 	struct schedule_state * s = data;
 	struct reg_value ** pv = get_reg_valuep(s, file, index, chan);
+	struct reg_value * newv;
 
 	if (!pv)
 		return;
 
 	DBG("%i: write %i[%i] chan %i\n", s->Current->Instruction->IP, file, index, chan);
 
-	struct reg_value * newv = memory_pool_malloc(&s->C->Pool, sizeof(*newv));
+	newv = memory_pool_malloc(&s->C->Pool, sizeof(*newv));
 	memset(newv, 0, sizeof(*newv));
 
 	newv->Writer = s->Current;
@@ -659,12 +670,13 @@ static void schedule_block(struct r300_fragment_program_compiler * c,
 		struct rc_instruction * begin, struct rc_instruction * end)
 {
 	struct schedule_state s;
+	unsigned int ip;
 
 	memset(&s, 0, sizeof(s));
 	s.C = &c->Base;
 
 	/* Scan instructions for data dependencies */
-	unsigned int ip = 0;
+	ip = 0;
 	for(struct rc_instruction * inst = begin; inst != end; inst = inst->Next) {
 		s.Current = memory_pool_malloc(&c->Base.Pool, sizeof(*s.Current));
 		memset(s.Current, 0, sizeof(struct schedule_instruction));
@@ -716,12 +728,14 @@ void rc_pair_schedule(struct radeon_compiler *cc, void *user)
 	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)cc;
 	struct rc_instruction * inst = c->Base.Program.Instructions.Next;
 	while(inst != &c->Base.Program.Instructions) {
+		struct rc_instruction * first;
+
 		if (is_controlflow(inst)) {
 			inst = inst->Next;
 			continue;
 		}
 
-		struct rc_instruction * first = inst;
+		first = inst;
 
 		while(inst != &c->Base.Program.Instructions && !is_controlflow(inst))
 			inst = inst->Next;
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_pair_translate.c b/src/mesa/drivers/dri/r300/compiler/radeon_pair_translate.c
index ff82584466..c549be5218 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_pair_translate.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_pair_translate.c
@@ -148,9 +148,12 @@ static void set_pair_instruction(struct r300_fragment_program_compiler *c,
 	struct rc_pair_instruction * pair,
 	struct rc_sub_instruction * inst)
 {
+	int needrgb, needalpha, istranscendent;
+	const struct rc_opcode_info * opcode;
+	int i;
+
 	memset(pair, 0, sizeof(struct rc_pair_instruction));
 
-	int needrgb, needalpha, istranscendent;
 	classify_instruction(inst, &needrgb, &needalpha, &istranscendent);
 
 	if (needrgb) {
@@ -167,8 +170,7 @@ static void set_pair_instruction(struct r300_fragment_program_compiler *c,
 			pair->Alpha.Saturate = 1;
 	}
 
-	const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->Opcode);
-	int i;
+	opcode = rc_get_opcode_info(inst->Opcode);
 
 	/* Presubtract handling:
 	 * We need to make sure that the values used by the presubtract
@@ -330,15 +332,18 @@ void rc_pair_translate(struct radeon_compiler *cc, void *user)
 	for(struct rc_instruction * inst = c->Base.Program.Instructions.Next;
 	    inst != &c->Base.Program.Instructions;
 	    inst = inst->Next) {
+		const struct rc_opcode_info * opcode;
+		struct rc_sub_instruction copy;
+
 		if (inst->Type != RC_INSTRUCTION_NORMAL)
 			continue;
 
-		const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
+		opcode = rc_get_opcode_info(inst->U.I.Opcode);
 
 		if (opcode->HasTexture || opcode->IsFlowControl || opcode->Opcode == RC_OPCODE_KIL)
 			continue;
 
-		struct rc_sub_instruction copy = inst->U.I;
+		copy = inst->U.I;
 
 		check_opcode_support(c, &copy);
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
index 4d3e26f28c..39408845d5 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
@@ -874,13 +874,15 @@ int r300_transform_trig_simple(struct radeon_compiler* c,
 	struct rc_instruction* inst,
 	void* unused)
 {
+	unsigned int constants[2];
+	unsigned int tempreg;
+
 	if (inst->U.I.Opcode != RC_OPCODE_COS &&
 	    inst->U.I.Opcode != RC_OPCODE_SIN &&
 	    inst->U.I.Opcode != RC_OPCODE_SCS)
 		return 0;
 
-	unsigned int constants[2];
-	unsigned int tempreg = rc_find_free_temporary(c);
+	tempreg = rc_find_free_temporary(c);
 
 	sincos_constants(c, constants);
 
@@ -918,6 +920,8 @@ int r300_transform_trig_simple(struct radeon_compiler* c,
 			swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)),
 			constants);
 	} else {
+		struct rc_dst_register dst;
+
 		emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_XY),
 			swizzle_xxxx(inst->U.I.SrcReg[0]),
 			swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[1])),
@@ -929,7 +933,7 @@ int r300_transform_trig_simple(struct radeon_compiler* c,
 			swizzle_wwww(srcreg(RC_FILE_CONSTANT, constants[1])),
 			negate(swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[0]))));
 
-		struct rc_dst_register dst = inst->U.I.DstReg;
+		dst = inst->U.I.DstReg;
 
 		dst.WriteMask = inst->U.I.DstReg.WriteMask & RC_MASK_X;
 		sin_approx(c, inst, dst,
@@ -988,16 +992,16 @@ int radeonTransformTrigScale(struct radeon_compiler* c,
 	struct rc_instruction* inst,
 	void* unused)
 {
-	if (inst->U.I.Opcode != RC_OPCODE_COS &&
-	    inst->U.I.Opcode != RC_OPCODE_SIN &&
-	    inst->U.I.Opcode != RC_OPCODE_SCS)
-		return 0;
-
 	static const float RCP_2PI = 0.15915494309189535;
 	unsigned int temp;
 	unsigned int constant;
 	unsigned int constant_swizzle;
 
+	if (inst->U.I.Opcode != RC_OPCODE_COS &&
+	    inst->U.I.Opcode != RC_OPCODE_SIN &&
+	    inst->U.I.Opcode != RC_OPCODE_SCS)
+		return 0;
+
 	temp = rc_find_free_temporary(c);
 	constant = rc_constants_add_immediate_scalar(&c->Program.Constants, RCP_2PI, &constant_swizzle);
 
@@ -1020,6 +1024,10 @@ int r300_transform_trig_scale_vertex(struct radeon_compiler *c,
 	struct rc_instruction *inst,
 	void *unused)
 {
+	static const float cons[4] = {0.15915494309189535, 0.5, 6.28318530717959, -3.14159265358979};
+	unsigned int temp;
+	unsigned int constant;
+
 	if (inst->U.I.Opcode != RC_OPCODE_COS &&
 	    inst->U.I.Opcode != RC_OPCODE_SIN &&
 	    inst->U.I.Opcode != RC_OPCODE_SCS)
@@ -1030,10 +1038,6 @@ int r300_transform_trig_scale_vertex(struct radeon_compiler *c,
 	 *   repeat(x) = frac(x / 2PI + 0.5) * 2PI - PI
 	 */
 
-	static const float cons[4] = {0.15915494309189535, 0.5, 6.28318530717959, -3.14159265358979};
-	unsigned int temp;
-	unsigned int constant;
-
 	temp = rc_find_free_temporary(c);
 	constant = rc_constants_add_immediate_vec4(&c->Program.Constants, cons);
 
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h
index e0061e454b..01cdb15424 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h
@@ -55,52 +55,35 @@ struct radeon_compiler;
  */
 #define RC_PAIR_PRESUB_SRC 3
 
-struct radeon_pair_instruction_source {
+struct rc_pair_instruction_source {
 	unsigned int Used:1;
 	unsigned int File:3;
 	unsigned int Index:RC_REGISTER_INDEX_BITS;
 };
 
-struct radeon_pair_instruction_rgb {
-	unsigned int Opcode:8;
-	unsigned int DestIndex:RC_REGISTER_INDEX_BITS;
-	unsigned int WriteMask:3;
-    unsigned int Target:2;
-	unsigned int OutputWriteMask:3;
-	unsigned int Saturate:1;
-
-	struct radeon_pair_instruction_source Src[4];
-
-	struct {
-		unsigned int Source:2;
-		unsigned int Swizzle:9;
-		unsigned int Abs:1;
-		unsigned int Negate:1;
-	} Arg[3];
+struct rc_pair_instruction_arg {
+	unsigned int Source:2;
+	unsigned int Swizzle:9;
+	unsigned int Abs:1;
+	unsigned int Negate:1;
 };
 
-struct radeon_pair_instruction_alpha {
+struct rc_pair_sub_instruction {
 	unsigned int Opcode:8;
 	unsigned int DestIndex:RC_REGISTER_INDEX_BITS;
-	unsigned int WriteMask:1;
-    unsigned int Target:2;
-	unsigned int OutputWriteMask:1;
+	unsigned int WriteMask:3;
+	unsigned int Target:2;
+	unsigned int OutputWriteMask:3;
 	unsigned int DepthWriteMask:1;
 	unsigned int Saturate:1;
 
-	struct radeon_pair_instruction_source Src[4];
-
-	struct {
-		unsigned int Source:2;
-		unsigned int Swizzle:3;
-		unsigned int Abs:1;
-		unsigned int Negate:1;
-	} Arg[3];
+	struct rc_pair_instruction_source Src[4];
+	struct rc_pair_instruction_arg Arg[3];
 };
 
 struct rc_pair_instruction {
-	struct radeon_pair_instruction_rgb RGB;
-	struct radeon_pair_instruction_alpha Alpha;
+	struct rc_pair_sub_instruction RGB;
+	struct rc_pair_sub_instruction Alpha;
 
 	unsigned int WriteALUResult:2;
 	unsigned int ALUResultCompare:3;
@@ -108,7 +91,7 @@ struct rc_pair_instruction {
 };
 
 typedef void (*rc_pair_foreach_src_fn)
-			(void *, struct radeon_pair_instruction_source *);
+			(void *, struct rc_pair_instruction_source *);
 
 typedef enum {
 	RC_PAIR_SOURCE_NONE = 0,
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
index ddce590ee6..530afa5e08 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_tex.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (C) 2010 Corbin Simpson
+ * Copyright (C) 2010 Marek Olšák <maraeo@gmail.com>
  *
  * All Rights Reserved.
  *
@@ -46,31 +47,61 @@ static struct rc_src_register shadow_ambient(struct r300_fragment_program_compil
 	return reg;
 }
 
-static void lower_texture_rect(struct r300_fragment_program_compiler *compiler,
-							   struct rc_instruction *inst)
+static void scale_texcoords(struct r300_fragment_program_compiler *compiler,
+			    struct rc_instruction *inst,
+			    unsigned state_constant)
 {
-	struct rc_instruction *inst_rect;
+	struct rc_instruction *inst_mov;
+
 	unsigned temp = rc_find_free_temporary(&compiler->Base);
 
-	if (inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
-		compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords) {
-		inst_rect = rc_insert_new_instruction(&compiler->Base, inst->Prev);
+	inst_mov = rc_insert_new_instruction(&compiler->Base, inst->Prev);
 
-		inst_rect->U.I.Opcode = RC_OPCODE_MUL;
-		inst_rect->U.I.DstReg.File = RC_FILE_TEMPORARY;
-		inst_rect->U.I.DstReg.Index = temp;
-		inst_rect->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-		inst_rect->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
-		inst_rect->U.I.SrcReg[1].Index =
-				rc_constants_add_state(&compiler->Base.Program.Constants,
-									   RC_STATE_R300_TEXRECT_FACTOR, inst->U.I.TexSrcUnit);
+	inst_mov->U.I.Opcode = RC_OPCODE_MUL;
+	inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
+	inst_mov->U.I.DstReg.Index = temp;
+	inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+	inst_mov->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
+	inst_mov->U.I.SrcReg[1].Index =
+			rc_constants_add_state(&compiler->Base.Program.Constants,
+					       state_constant, inst->U.I.TexSrcUnit);
 
-		reset_srcreg(&inst->U.I.SrcReg[0]);
-		inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
-		inst->U.I.SrcReg[0].Index = temp;
+	reset_srcreg(&inst->U.I.SrcReg[0]);
+	inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+	inst->U.I.SrcReg[0].Index = temp;
+}
 
-		inst->U.I.TexSrcTarget = RC_TEXTURE_2D;
-	}
+static void projective_divide(struct r300_fragment_program_compiler *compiler,
+			      struct rc_instruction *inst)
+{
+	struct rc_instruction *inst_mul, *inst_rcp;
+
+	unsigned temp = rc_find_free_temporary(&compiler->Base);
+
+	inst_rcp = rc_insert_new_instruction(&compiler->Base, inst->Prev);
+	inst_rcp->U.I.Opcode = RC_OPCODE_RCP;
+	inst_rcp->U.I.DstReg.File = RC_FILE_TEMPORARY;
+	inst_rcp->U.I.DstReg.Index = temp;
+	inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W;
+	inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+	/* Because the input can be arbitrarily swizzled,
+	 * read the component mapped to W. */
+	inst_rcp->U.I.SrcReg[0].Swizzle =
+		RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 3));
+
+	inst_mul = rc_insert_new_instruction(&compiler->Base, inst->Prev);
+	inst_mul->U.I.Opcode = RC_OPCODE_MUL;
+	inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY;
+	inst_mul->U.I.DstReg.Index = temp;
+	inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+	inst_mul->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
+	inst_mul->U.I.SrcReg[1].Index = temp;
+	inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_WWWW;
+
+	reset_srcreg(&inst->U.I.SrcReg[0]);
+	inst->U.I.Opcode = RC_OPCODE_TEX;
+	inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+	inst->U.I.SrcReg[0].Index = temp;
 }
 
 /**
@@ -88,6 +119,9 @@ int radeonTransformTEX(
 {
 	struct r300_fragment_program_compiler *compiler =
 		(struct r300_fragment_program_compiler*)data;
+	rc_wrap_mode wrapmode = compiler->state.unit[inst->U.I.TexSrcUnit].wrap_mode;
+	int is_rect = inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
+		      compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords;
 
 	if (inst->U.I.Opcode != RC_OPCODE_TEX &&
 		inst->U.I.Opcode != RC_OPCODE_TXB &&
@@ -141,16 +175,18 @@ int radeonTransformTEX(
 				inst_rcp->U.I.DstReg.Index = tmp_recip_w;
 				inst_rcp->U.I.DstReg.WriteMask = RC_MASK_W;
 				inst_rcp->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-				inst_rcp->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW;
+				inst_rcp->U.I.SrcReg[0].Swizzle =
+					RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 3));
 			}
 
-			/* Perspective-divide r by W (if it's TXP) and add the texture sample (see below). */
+			/* Perspective-divide Z by W (if it's TXP) and add the texture sample (see below). */
 			tmp_sum = rc_find_free_temporary(c);
 			inst_mad = rc_insert_new_instruction(c, inst_rcp ? inst_rcp : inst);
 			inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
 			inst_mad->U.I.DstReg.Index = tmp_sum;
 			inst_mad->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-			inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_ZZZZ;
+			inst_mad->U.I.SrcReg[0].Swizzle =
+				RC_MAKE_SWIZZLE_SMEAR(GET_SWZ(inst->U.I.SrcReg[0].Swizzle, 2));
 			if (inst->U.I.Opcode == RC_OPCODE_TXP) {
 				inst_mad->U.I.Opcode = RC_OPCODE_MAD;
 				inst_mad->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
@@ -206,11 +242,22 @@ int radeonTransformTEX(
 		}
 	}
 
-	/* Texture wrap modes don't work on NPOT textures or texrects.
-	 *
-	 * The game plan is simple. We have two flags, fake_npot and
-	 * non_normalized_coords, as well as a tex target. The RECT tex target
-	 * will make the emitted code use non-scaled texcoords.
+	/* R300 cannot sample from rectangles and the wrap mode fallback needs
+	 * normalized coordinates anyway. */
+	if (inst->U.I.Opcode != RC_OPCODE_KIL &&
+	    is_rect && (!c->is_r500 || wrapmode != RC_WRAP_NONE)) {
+		scale_texcoords(compiler, inst, RC_STATE_R300_TEXRECT_FACTOR);
+		inst->U.I.TexSrcTarget = RC_TEXTURE_2D;
+	}
+
+	/* Divide by W if needed. */
+	if (inst->U.I.Opcode == RC_OPCODE_TXP &&
+	    (wrapmode == RC_WRAP_REPEAT || wrapmode == RC_WRAP_MIRRORED_REPEAT ||
+	     compiler->state.unit[inst->U.I.TexSrcUnit].clamp_and_scale_before_fetch)) {
+		projective_divide(compiler, inst);
+	}
+
+	/* Texture wrap modes don't work on NPOT textures.
 	 *
 	 * Non-wrapped/clamped texcoords with NPOT are free in HW. Repeat and
 	 * mirroring are not. If we need to repeat, we do:
@@ -235,128 +282,140 @@ int radeonTransformTEX(
 	 * ~ C & M. ;)
 	 */
 	if (inst->U.I.Opcode != RC_OPCODE_KIL &&
-		(inst->U.I.TexSrcTarget == RC_TEXTURE_RECT ||
-			compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot ||
-			compiler->state.unit[inst->U.I.TexSrcUnit].non_normalized_coords)) {
-		rc_wrap_mode wrapmode = compiler->state.unit[inst->U.I.TexSrcUnit].wrap_mode;
-
-		/* R300 cannot sample from rectangles. */
-		if (!c->is_r500) {
-			lower_texture_rect(compiler, inst);
-		}
-
-		if (compiler->state.unit[inst->U.I.TexSrcUnit].fake_npot &&
-			wrapmode != RC_WRAP_NONE) {
+	    wrapmode != RC_WRAP_NONE) {
+		struct rc_instruction *inst_mov;
+		unsigned temp = rc_find_free_temporary(c);
+
+		if (wrapmode == RC_WRAP_REPEAT) {
+			/* Both instructions will be paired up. */
+			struct rc_instruction *inst_frc = rc_insert_new_instruction(c, inst->Prev);
+
+			inst_frc->U.I.Opcode = RC_OPCODE_FRC;
+			inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY;
+			inst_frc->U.I.DstReg.Index = temp;
+			inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+			inst_frc->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+		} else if (wrapmode == RC_WRAP_MIRRORED_REPEAT) {
+			/*
+			 * Function:
+			 *   f(v) = 1 - abs(frac(v * 0.5) * 2 - 1)
+			 *
+			 * Code:
+			 *   MUL temp, src0, 0.5
+			 *   FRC temp, temp
+			 *   MAD temp, temp, 2, -1
+			 *   ADD temp, 1, -abs(temp)
+			 */
+
+			struct rc_instruction *inst_mul, *inst_frc, *inst_mad, *inst_add;
+			unsigned two, two_swizzle;
+
+			inst_mul = rc_insert_new_instruction(c, inst->Prev);
+
+			inst_mul->U.I.Opcode = RC_OPCODE_MUL;
+			inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY;
+			inst_mul->U.I.DstReg.Index = temp;
+			inst_mul->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+			inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+			inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_HHHH;
+
+			inst_frc = rc_insert_new_instruction(c, inst->Prev);
+
+			inst_frc->U.I.Opcode = RC_OPCODE_FRC;
+			inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY;
+			inst_frc->U.I.DstReg.Index = temp;
+			inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+			inst_frc->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+			inst_frc->U.I.SrcReg[0].Index = temp;
+			inst_frc->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
+
+			two = rc_constants_add_immediate_scalar(&c->Program.Constants, 2, &two_swizzle);
+			inst_mad = rc_insert_new_instruction(c, inst->Prev);
+
+			inst_mad->U.I.Opcode = RC_OPCODE_MAD;
+			inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
+			inst_mad->U.I.DstReg.Index = temp;
+			inst_mad->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+			inst_mad->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+			inst_mad->U.I.SrcReg[0].Index = temp;
+			inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
+			inst_mad->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
+			inst_mad->U.I.SrcReg[1].Index = two;
+			inst_mad->U.I.SrcReg[1].Swizzle = two_swizzle;
+			inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_1111;
+			inst_mad->U.I.SrcReg[2].Negate = RC_MASK_XYZ;
+
+			inst_add = rc_insert_new_instruction(c, inst->Prev);
+
+			inst_add->U.I.Opcode = RC_OPCODE_ADD;
+			inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY;
+			inst_add->U.I.DstReg.Index = temp;
+			inst_add->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+			inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111;
+			inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
+			inst_add->U.I.SrcReg[1].Index = temp;
+			inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZ0;
+			inst_add->U.I.SrcReg[1].Abs = 1;
+			inst_add->U.I.SrcReg[1].Negate = RC_MASK_XYZ;
+		} else if (wrapmode == RC_WRAP_MIRRORED_CLAMP) {
+			/*
+			 * Mirrored clamp modes are bloody simple, we just use abs
+			 * to mirror [0, 1] into [-1, 0]. This works for
+			 * all modes i.e. CLAMP, CLAMP_TO_EDGE, and CLAMP_TO_BORDER.
+			 */
 			struct rc_instruction *inst_mov;
-			unsigned temp = rc_find_free_temporary(c);
-
-			/* For NPOT fallback, we need normalized coordinates anyway. */
-			if (c->is_r500) {
-				lower_texture_rect(compiler, inst);
-			}
-
-			if (wrapmode == RC_WRAP_REPEAT) {
-				/* Both instructions will be paired up. */
-				struct rc_instruction *inst_frc = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_frc->U.I.Opcode = RC_OPCODE_FRC;
-				inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_frc->U.I.DstReg.Index = temp;
-				inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_frc->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-			} else if (wrapmode == RC_WRAP_MIRRORED_REPEAT) {
-				/*
-				 * Function:
-				 *   f(v) = 1 - abs(frac(v * 0.5) * 2 - 1)
-				 *
-				 * Code:
-				 *   MUL temp, src0, 0.5
-				 *   FRC temp, temp
-				 *   MAD temp, temp, 2, -1
-				 *   ADD temp, 1, -abs(temp)
-				 */
-
-				struct rc_instruction *inst_mul, *inst_frc, *inst_mad, *inst_add;
-				unsigned two, two_swizzle;
-
-				inst_mul = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_mul->U.I.Opcode = RC_OPCODE_MUL;
-				inst_mul->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_mul->U.I.DstReg.Index = temp;
-				inst_mul->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_mul->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-				inst_mul->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_HHHH;
-
-				inst_frc = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_frc->U.I.Opcode = RC_OPCODE_FRC;
-				inst_frc->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_frc->U.I.DstReg.Index = temp;
-				inst_frc->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_frc->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
-				inst_frc->U.I.SrcReg[0].Index = temp;
-				inst_frc->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
-
-				two = rc_constants_add_immediate_scalar(&c->Program.Constants, 2, &two_swizzle);
-				inst_mad = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_mad->U.I.Opcode = RC_OPCODE_MAD;
-				inst_mad->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_mad->U.I.DstReg.Index = temp;
-				inst_mad->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_mad->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
-				inst_mad->U.I.SrcReg[0].Index = temp;
-				inst_mad->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZ0;
-				inst_mad->U.I.SrcReg[1].File = RC_FILE_CONSTANT;
-				inst_mad->U.I.SrcReg[1].Index = two;
-				inst_mad->U.I.SrcReg[1].Swizzle = two_swizzle;
-				inst_mad->U.I.SrcReg[2].Swizzle = RC_SWIZZLE_1111;
-				inst_mad->U.I.SrcReg[2].Negate = RC_MASK_XYZ;
-
-				inst_add = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_add->U.I.Opcode = RC_OPCODE_ADD;
-				inst_add->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_add->U.I.DstReg.Index = temp;
-				inst_add->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_add->U.I.SrcReg[0].Swizzle = RC_SWIZZLE_1111;
-				inst_add->U.I.SrcReg[1].File = RC_FILE_TEMPORARY;
-				inst_add->U.I.SrcReg[1].Index = temp;
-				inst_add->U.I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZ0;
-				inst_add->U.I.SrcReg[1].Abs = 1;
-				inst_add->U.I.SrcReg[1].Negate = RC_MASK_XYZ;
-			} else if (wrapmode == RC_WRAP_MIRRORED_CLAMP) {
-				/*
-				 * Mirrored clamp modes are bloody simple, we just use abs
-				 * to mirror [0, 1] into [-1, 0]. This works for
-				 * all modes i.e. CLAMP, CLAMP_TO_EDGE, and CLAMP_TO_BORDER.
-				 */
-				struct rc_instruction *inst_mov;
-
-				inst_mov = rc_insert_new_instruction(c, inst->Prev);
-
-				inst_mov->U.I.Opcode = RC_OPCODE_MOV;
-				inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
-				inst_mov->U.I.DstReg.Index = temp;
-				inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ;
-				inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-				inst_mov->U.I.SrcReg[0].Abs = 1;
-			}
 
-			/* Preserve W for TXP/TXB. */
 			inst_mov = rc_insert_new_instruction(c, inst->Prev);
 
 			inst_mov->U.I.Opcode = RC_OPCODE_MOV;
 			inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
 			inst_mov->U.I.DstReg.Index = temp;
-			inst_mov->U.I.DstReg.WriteMask = RC_MASK_W;
+			inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ;
 			inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
-
-			reset_srcreg(&inst->U.I.SrcReg[0]);
-			inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
-			inst->U.I.SrcReg[0].Index = temp;
+			inst_mov->U.I.SrcReg[0].Abs = 1;
 		}
+
+		/* Preserve W for TXP/TXB. */
+		inst_mov = rc_insert_new_instruction(c, inst->Prev);
+
+		inst_mov->U.I.Opcode = RC_OPCODE_MOV;
+		inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
+		inst_mov->U.I.DstReg.Index = temp;
+		inst_mov->U.I.DstReg.WriteMask = RC_MASK_W;
+		inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+
+		reset_srcreg(&inst->U.I.SrcReg[0]);
+		inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+		inst->U.I.SrcReg[0].Index = temp;
+	}
+
+	if (inst->U.I.Opcode != RC_OPCODE_KIL &&
+	    compiler->state.unit[inst->U.I.TexSrcUnit].clamp_and_scale_before_fetch) {
+		struct rc_instruction *inst_mov;
+		unsigned temp = rc_find_free_temporary(c);
+
+		/* Saturate XYZ. */
+		inst_mov = rc_insert_new_instruction(c, inst->Prev);
+		inst_mov->U.I.Opcode = RC_OPCODE_MOV;
+		inst_mov->U.I.SaturateMode = RC_SATURATE_ZERO_ONE;
+		inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
+		inst_mov->U.I.DstReg.Index = temp;
+		inst_mov->U.I.DstReg.WriteMask = RC_MASK_XYZ;
+		inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+
+		/* Copy W. */
+		inst_mov = rc_insert_new_instruction(c, inst->Prev);
+		inst_mov->U.I.Opcode = RC_OPCODE_MOV;
+		inst_mov->U.I.DstReg.File = RC_FILE_TEMPORARY;
+		inst_mov->U.I.DstReg.Index = temp;
+		inst_mov->U.I.DstReg.WriteMask = RC_MASK_W;
+		inst_mov->U.I.SrcReg[0] = inst->U.I.SrcReg[0];
+
+		reset_srcreg(&inst->U.I.SrcReg[0]);
+		inst->U.I.SrcReg[0].File = RC_FILE_TEMPORARY;
+		inst->U.I.SrcReg[0].Index = temp;
+
+		scale_texcoords(compiler, inst, RC_STATE_R300_TEXSCALE_FACTOR);
 	}
 
 	/* Cannot write texture to output registers (all chips) or with masks (non-r500) */
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_remove_constants.c b/src/mesa/drivers/dri/r300/compiler/radeon_remove_constants.c
index d111319d3d..d6c808ad81 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_remove_constants.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_remove_constants.c
@@ -28,107 +28,105 @@
 #include "radeon_remove_constants.h"
 #include "radeon_dataflow.h"
 
+struct mark_used_data {
+	unsigned char * const_used;
+	unsigned * has_rel_addr;
+};
+
 static void remap_regs(void * userdata, struct rc_instruction * inst,
 			rc_register_file * pfile, unsigned int * pindex)
 {
-        unsigned *inv_remap_table = userdata;
+	unsigned *inv_remap_table = userdata;
+
+	if (*pfile == RC_FILE_CONSTANT) {
+		*pindex = inv_remap_table[*pindex];
+	}
+}
+
+static void mark_used(void * userdata, struct rc_instruction * inst,
+						struct rc_src_register * src)
+{
+	struct mark_used_data * d = userdata;
 
-        if (*pfile == RC_FILE_CONSTANT) {
-                *pindex = inv_remap_table[*pindex];
-        }
+	if (src->File == RC_FILE_CONSTANT) {
+		if (src->RelAddr) {
+			*d->has_rel_addr = 1;
+		} else {
+			d->const_used[src->Index] = 1;
+		}
+	}
 }
 
 void rc_remove_unused_constants(struct radeon_compiler *c, void *user)
 {
 	unsigned **out_remap_table = (unsigned**)user;
-        unsigned char *const_used;
-        unsigned *remap_table;
-        unsigned *inv_remap_table;
-        unsigned has_rel_addr = 0;
-        unsigned is_identity = 1;
+	unsigned char *const_used;
+	unsigned *remap_table;
+	unsigned *inv_remap_table;
+	unsigned has_rel_addr = 0;
+	unsigned is_identity = 1;
 	unsigned are_externals_remapped = 0;
-        struct rc_constant *constants = c->Program.Constants.Constants;
+	struct rc_constant *constants = c->Program.Constants.Constants;
+	struct mark_used_data d;
+	unsigned new_count;
+
+	if (!c->Program.Constants.Count) {
+		*out_remap_table = NULL;
+		return;
+	}
 
-        if (!c->Program.Constants.Count) {
-                *out_remap_table = NULL;
-                return;
-        }
+	const_used = malloc(c->Program.Constants.Count);
+	memset(const_used, 0, c->Program.Constants.Count);
 
-        const_used = malloc(c->Program.Constants.Count);
-        memset(const_used, 0, c->Program.Constants.Count);
+	d.const_used = const_used;
+	d.has_rel_addr = &has_rel_addr;
 
 	/* Pass 1: Mark used constants. */
-        for (struct rc_instruction *inst = c->Program.Instructions.Next;
-             inst != &c->Program.Instructions; inst = inst->Next) {
-                const struct rc_opcode_info *opcode = rc_get_opcode_info(inst->U.I.Opcode);
-
-                /* XXX: This loop and the if statement after it should be
-                 * replaced by a call to one of the rc_for_all_reads_* functions.
-                 * The reason it does not use one of those functions now is
-                 * because none of them have RelAddr as an argument. */
-                for (unsigned i = 0; i < opcode->NumSrcRegs; i++) {
-                        if (inst->U.I.SrcReg[i].File == RC_FILE_CONSTANT) {
-                                if (inst->U.I.SrcReg[i].RelAddr) {
-                                        has_rel_addr = 1;
-                                } else {
-                                        const_used[inst->U.I.SrcReg[i].Index] = 1;
-                                }
-                        }
-                }
-                if (inst->U.I.PreSub.Opcode != RC_PRESUB_NONE) {
-			unsigned int i;
-			unsigned int srcp_regs = rc_presubtract_src_reg_count(
-							inst->U.I.PreSub.Opcode);
-			for( i = 0; i < srcp_regs; i++) {
-                                if (inst->U.I.PreSub.SrcReg[i].File ==
-                                                        RC_FILE_CONSTANT) {
-                                        const_used[
-                                            inst->U.I.PreSub.SrcReg[i].Index] = 1;
-                                }
-                        }
-		}
-        }
+	for (struct rc_instruction *inst = c->Program.Instructions.Next;
+	     inst != &c->Program.Instructions; inst = inst->Next) {
+		rc_for_all_reads_src(inst, mark_used, &d);
+	}
 
-        /* Pass 2: If there is relative addressing, mark all externals as used. */
-        if (has_rel_addr) {
-                for (unsigned i = 0; i < c->Program.Constants.Count; i++)
-                        if (constants[i].Type == RC_CONSTANT_EXTERNAL)
-                                const_used[i] = 1;
-        }
+	/* Pass 2: If there is relative addressing, mark all externals as used. */
+	if (has_rel_addr) {
+		for (unsigned i = 0; i < c->Program.Constants.Count; i++)
+			if (constants[i].Type == RC_CONSTANT_EXTERNAL)
+				const_used[i] = 1;
+	}
 
-        /* Pass 3: Make the remapping table and remap constants.
+	/* Pass 3: Make the remapping table and remap constants.
 	 * This pass removes unused constants simply by overwriting them by other constants. */
-        remap_table = malloc(c->Program.Constants.Count * sizeof(unsigned));
-        inv_remap_table = malloc(c->Program.Constants.Count * sizeof(unsigned));
-        unsigned new_count = 0;
+	remap_table = malloc(c->Program.Constants.Count * sizeof(unsigned));
+	inv_remap_table = malloc(c->Program.Constants.Count * sizeof(unsigned));
+	new_count = 0;
 
-        for (unsigned i = 0; i < c->Program.Constants.Count; i++) {
-                if (const_used[i]) {
-                        remap_table[new_count] = i;
-                        inv_remap_table[i] = new_count;
+	for (unsigned i = 0; i < c->Program.Constants.Count; i++) {
+		if (const_used[i]) {
+			remap_table[new_count] = i;
+			inv_remap_table[i] = new_count;
 
-                        if (i != new_count) {
+			if (i != new_count) {
 				if (constants[i].Type == RC_CONSTANT_EXTERNAL)
 					are_externals_remapped = 1;
 
 				constants[new_count] = constants[i];
-                                is_identity = 0;
-                        }
-                        new_count++;
-                }
-        }
+				is_identity = 0;
+			}
+			new_count++;
+		}
+	}
 
 	/*  is_identity ==> new_count == old_count
 	 * !is_identity ==> new_count <  old_count */
 	assert( is_identity || new_count <  c->Program.Constants.Count);
 	assert(!(has_rel_addr && are_externals_remapped));
 
-        /* Pass 4: Redirect reads of all constants to their new locations. */
-        if (!is_identity) {
-                for (struct rc_instruction *inst = c->Program.Instructions.Next;
-                     inst != &c->Program.Instructions; inst = inst->Next) {
-                        rc_remap_registers(inst, remap_regs, inv_remap_table);
-                }
+	/* Pass 4: Redirect reads of all constants to their new locations. */
+	if (!is_identity) {
+		for (struct rc_instruction *inst = c->Program.Instructions.Next;
+		     inst != &c->Program.Instructions; inst = inst->Next) {
+			rc_remap_registers(inst, remap_regs, inv_remap_table);
+		}
 
 	}
 
@@ -138,12 +136,15 @@ void rc_remove_unused_constants(struct radeon_compiler *c, void *user)
 	c->Program.Constants.Count = new_count;
 
 	if (are_externals_remapped) {
-                *out_remap_table = remap_table;
-        } else {
-                *out_remap_table = NULL;
-                free(remap_table);
-        }
-
-        free(const_used);
-        free(inv_remap_table);
+		*out_remap_table = remap_table;
+	} else {
+		*out_remap_table = NULL;
+		free(remap_table);
+	}
+
+	free(const_used);
+	free(inv_remap_table);
+
+	if (c->Debug)
+		rc_constants_print(&c->Program.Constants);
 }
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_rename_regs.c b/src/mesa/drivers/dri/r300/compiler/radeon_rename_regs.c
index 31d25f9ab8..60e228be5b 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_rename_regs.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_rename_regs.c
@@ -104,13 +104,14 @@ void rc_rename_regs(struct radeon_compiler *c, void *user)
 					inst != &c->Program.Instructions;
 					inst = inst->Next) {
 		const struct rc_opcode_info * info;
+		unsigned int old_index, temp_index;
+		struct rc_dst_register * dst;
 		if(inst->Type != RC_INSTRUCTION_NORMAL) {
 			rc_error(c, "%s only works with normal instructions.",
 								__FUNCTION__);
 			return;
 		}
-		unsigned int old_index, temp_index;
-		struct rc_dst_register * dst = &inst->U.I.DstReg;
+		dst = &inst->U.I.DstReg;
 		info = rc_get_opcode_info(inst->U.I.Opcode);
 		if(!info->HasDstReg || dst->File != RC_FILE_TEMPORARY) {
 			continue;