r300/compiler: Refactor local transforms to use rc_program

Signed-off-by: Nicolai Hähnle <nhaehnle@gmail.com>

14 files changed, 664 insertions, 646 deletions

diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c
index f838179f79..fbffd3d7cd 100644
--- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c
@@ -31,16 +31,12 @@
 
 #include "../r300_reg.h"
 
-static struct prog_src_register shadow_ambient(struct gl_program *program, int tmu)
+static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
 {
-	gl_state_index fail_value_tokens[STATE_LENGTH] = {
-		STATE_INTERNAL, STATE_SHADOW_AMBIENT, 0, 0, 0
-	};
 	struct prog_src_register reg = { 0, };
 
-	fail_value_tokens[2] = tmu;
 	reg.File = PROGRAM_STATE_VAR;
-	reg.Index = _mesa_add_state_reference(program->Parameters, fail_value_tokens);
+	reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
 	reg.Swizzle = SWIZZLE_WWWW;
 	return reg;
 }
@@ -50,173 +46,146 @@ static struct prog_src_register shadow_ambient(struct gl_program *program, int t
  *  - premultiply texture coordinates for RECT
  *  - extract operand swizzles
  *  - introduce a temporary register when write masks are needed
- *
- * \todo If/when r5xx uses the radeon_program architecture, this can probably
- * be reused.
  */
 GLboolean r300_transform_TEX(
-	struct radeon_transform_context *t,
-	struct prog_instruction* orig_inst, void* data)
+	struct radeon_compiler * c,
+	struct rc_instruction* inst,
+	void* data)
 {
 	struct r300_fragment_program_compiler *compiler =
 		(struct r300_fragment_program_compiler*)data;
-	struct prog_instruction inst = *orig_inst;
-	struct prog_instruction* tgt;
-	GLboolean destredirect = GL_FALSE;
-
-	if (inst.Opcode != OPCODE_TEX &&
-	    inst.Opcode != OPCODE_TXB &&
-	    inst.Opcode != OPCODE_TXP &&
-	    inst.Opcode != OPCODE_KIL)
+
+	if (inst->I.Opcode != OPCODE_TEX &&
+	    inst->I.Opcode != OPCODE_TXB &&
+	    inst->I.Opcode != OPCODE_TXP &&
+	    inst->I.Opcode != OPCODE_KIL)
 		return GL_FALSE;
 
-	if (inst.Opcode != OPCODE_KIL &&
-	    t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
-		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst.TexSrcUnit].texture_compare_func;
+	/* ARB_shadow & EXT_shadow_funcs */
+	if (inst->I.Opcode != OPCODE_KIL &&
+	    c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) {
+		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
 
 		if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) {
-			tgt = radeonAppendInstructions(t->Program, 1);
+			inst->I.Opcode = OPCODE_MOV;
 
-			tgt->Opcode = OPCODE_MOV;
-			tgt->DstReg = inst.DstReg;
 			if (comparefunc == GL_ALWAYS) {
-				tgt->SrcReg[0].File = PROGRAM_BUILTIN;
-				tgt->SrcReg[0].Swizzle = SWIZZLE_1111;
+				inst->I.SrcReg[0].File = PROGRAM_BUILTIN;
+				inst->I.SrcReg[0].Swizzle = SWIZZLE_1111;
 			} else {
-				tgt->SrcReg[0] = shadow_ambient(t->Program, inst.TexSrcUnit);
+				inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit);
 			}
+
 			return GL_TRUE;
-		}
+		} else {
+			GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+			GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode;
+			struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
+			struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
+			struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
+			int pass, fail;
+
+			inst_rcp->I.Opcode = OPCODE_RCP;
+			inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_rcp->I.DstReg.Index = rc_find_free_temporary(c);
+			inst_rcp->I.DstReg.WriteMask = WRITEMASK_W;
+			inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0];
+			inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW;
+
+			inst_cmp->I.DstReg = inst->I.DstReg;
+			inst->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst->I.DstReg.Index = rc_find_free_temporary(c);
+			inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+
+			inst_mad->I.Opcode = OPCODE_MAD;
+			inst_mad->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_mad->I.DstReg.Index = rc_find_free_temporary(c);
+			inst_mad->I.SrcReg[0] = inst->I.SrcReg[0];
+			inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
+			inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+			inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index;
+			inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW;
+			inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+			inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index;
+			if (depthmode == 0) /* GL_LUMINANCE */
+				inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
+			else if (depthmode == 2) /* GL_ALPHA */
+				inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW;
+
+			/* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
+			 *   r  < tex  <=>      -tex+r < 0
+			 *   r >= tex  <=> not (-tex+r < 0 */
+			if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
+				inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW;
+			else
+				inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW;
+
+			inst_cmp->I.Opcode = OPCODE_CMP;
+			/* DstReg has been filled out above */
+			inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+			inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index;
+
+			if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
+				pass = 1;
+				fail = 2;
+			} else {
+				pass = 2;
+				fail = 1;
+			}
 
-		inst.DstReg.File = PROGRAM_TEMPORARY;
-		inst.DstReg.Index = radeonFindFreeTemporary(t);
-		inst.DstReg.WriteMask = WRITEMASK_XYZW;
+			inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN;
+			inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111;
+			inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit);
+		}
 	}
 
-
 	/* Hardware uses [0..1]x[0..1] range for rectangle textures
 	 * instead of [0..Width]x[0..Height].
 	 * Add a scaling instruction.
 	 */
-	if (inst.Opcode != OPCODE_KIL && inst.TexSrcTarget == TEXTURE_RECT_INDEX) {
-		gl_state_index tokens[STATE_LENGTH] = {
-			STATE_INTERNAL, STATE_R300_TEXRECT_FACTOR, 0, 0,
-			0
-		};
-
-		int tempreg = radeonFindFreeTemporary(t);
-		int factor_index;
-
-		tokens[2] = inst.TexSrcUnit;
-		factor_index = _mesa_add_state_reference(t->Program->Parameters, tokens);
-
-		tgt = radeonAppendInstructions(t->Program, 1);
-
-		tgt->Opcode = OPCODE_MUL;
-		tgt->DstReg.File = PROGRAM_TEMPORARY;
-		tgt->DstReg.Index = tempreg;
-		tgt->SrcReg[0] = inst.SrcReg[0];
-		tgt->SrcReg[1].File = PROGRAM_STATE_VAR;
-		tgt->SrcReg[1].Index = factor_index;
-
-		reset_srcreg(&inst.SrcReg[0]);
-		inst.SrcReg[0].File = PROGRAM_TEMPORARY;
-		inst.SrcReg[0].Index = tempreg;
+	if (inst->I.Opcode != OPCODE_KIL && inst->I.TexSrcTarget == TEXTURE_RECT_INDEX) {
+		struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst->Prev);
+
+		inst_mul->I.Opcode = OPCODE_MUL;
+		inst_mul->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst_mul->I.DstReg.Index = rc_find_free_temporary(c);
+		inst_mul->I.SrcReg[0] = inst->I.SrcReg[0];
+		inst_mul->I.SrcReg[1].File = PROGRAM_STATE_VAR;
+		inst_mul->I.SrcReg[1].Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_R300_TEXRECT_FACTOR, inst->I.TexSrcUnit);
+
+		reset_srcreg(&inst->I.SrcReg[0]);
+		inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+		inst->I.SrcReg[0].Index = inst_mul->I.DstReg.Index;
 	}
 
-	if (inst.Opcode != OPCODE_KIL) {
-		if (inst.DstReg.File != PROGRAM_TEMPORARY ||
-		    inst.DstReg.WriteMask != WRITEMASK_XYZW) {
-			int tempreg = radeonFindFreeTemporary(t);
-
-			inst.DstReg.File = PROGRAM_TEMPORARY;
-			inst.DstReg.Index = tempreg;
-			inst.DstReg.WriteMask = WRITEMASK_XYZW;
-			destredirect = GL_TRUE;
-		} else if (inst.SaturateMode) {
-			destredirect = GL_TRUE;
-		}
-	}
+	/* Cannot write texture to output registers or with masks */
+	if (inst->I.Opcode != OPCODE_KIL &&
+	    (inst->I.DstReg.File != PROGRAM_TEMPORARY || inst->I.DstReg.WriteMask != WRITEMASK_XYZW)) {
+		struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
+
+		inst_mov->I.Opcode = OPCODE_MOV;
+		inst_mov->I.DstReg = inst->I.DstReg;
+		inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+		inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c);
 
-	if (inst.SrcReg[0].File != PROGRAM_TEMPORARY && inst.SrcReg[0].File != PROGRAM_INPUT) {
-		int tmpreg = radeonFindFreeTemporary(t);
-		tgt = radeonAppendInstructions(t->Program, 1);
-		tgt->Opcode = OPCODE_MOV;
-		tgt->DstReg.File = PROGRAM_TEMPORARY;
-		tgt->DstReg.Index = tmpreg;
-		tgt->SrcReg[0] = inst.SrcReg[0];
-
-		reset_srcreg(&inst.SrcReg[0]);
-		inst.SrcReg[0].File = PROGRAM_TEMPORARY;
-		inst.SrcReg[0].Index = tmpreg;
+		inst->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+		inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
 	}
 
-	tgt = radeonAppendInstructions(t->Program, 1);
-	_mesa_copy_instructions(tgt, &inst, 1);
-
-	if (inst.Opcode != OPCODE_KIL &&
-	    t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
-		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst.TexSrcUnit].texture_compare_func;
-		GLuint depthmode = compiler->state.unit[inst.TexSrcUnit].depth_texture_mode;
-		int rcptemp = radeonFindFreeTemporary(t);
-		int pass, fail;
-
-		tgt = radeonAppendInstructions(t->Program, 3);
-
-		tgt[0].Opcode = OPCODE_RCP;
-		tgt[0].DstReg.File = PROGRAM_TEMPORARY;
-		tgt[0].DstReg.Index = rcptemp;
-		tgt[0].DstReg.WriteMask = WRITEMASK_W;
-		tgt[0].SrcReg[0] = inst.SrcReg[0];
-		tgt[0].SrcReg[0].Swizzle = SWIZZLE_WWWW;
-
-		tgt[1].Opcode = OPCODE_MAD;
-		tgt[1].DstReg = inst.DstReg;
-		tgt[1].DstReg.WriteMask = orig_inst->DstReg.WriteMask;
-		tgt[1].SrcReg[0] = inst.SrcReg[0];
-		tgt[1].SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
-		tgt[1].SrcReg[1].File = PROGRAM_TEMPORARY;
-		tgt[1].SrcReg[1].Index = rcptemp;
-		tgt[1].SrcReg[1].Swizzle = SWIZZLE_WWWW;
-		tgt[1].SrcReg[2].File = PROGRAM_TEMPORARY;
-		tgt[1].SrcReg[2].Index = inst.DstReg.Index;
-		if (depthmode == 0) /* GL_LUMINANCE */
-			tgt[1].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
-		else if (depthmode == 2) /* GL_ALPHA */
-			tgt[1].SrcReg[2].Swizzle = SWIZZLE_WWWW;
-
-		/* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
-		 *   r  < tex  <=>      -tex+r < 0
-		 *   r >= tex  <=> not (-tex+r < 0 */
-		if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
-			tgt[1].SrcReg[2].Negate = tgt[0].SrcReg[2].Negate ^ NEGATE_XYZW;
-		else
-			tgt[1].SrcReg[0].Negate = tgt[0].SrcReg[0].Negate ^ NEGATE_XYZW;
-
-		tgt[2].Opcode = OPCODE_CMP;
-		tgt[2].DstReg = orig_inst->DstReg;
-		tgt[2].SrcReg[0].File = PROGRAM_TEMPORARY;
-		tgt[2].SrcReg[0].Index = tgt[1].DstReg.Index;
-
-		if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
-			pass = 1;
-			fail = 2;
-		} else {
-			pass = 2;
-			fail = 1;
-		}
 
-		tgt[2].SrcReg[pass].File = PROGRAM_BUILTIN;
-		tgt[2].SrcReg[pass].Swizzle = SWIZZLE_1111;
-		tgt[2].SrcReg[fail] = shadow_ambient(t->Program, inst.TexSrcUnit);
-	} else if (destredirect) {
-		tgt = radeonAppendInstructions(t->Program, 1);
-
-		tgt->Opcode = OPCODE_MOV;
-		tgt->DstReg = orig_inst->DstReg;
-		tgt->SaturateMode = inst.SaturateMode;
-		tgt->SrcReg[0].File = PROGRAM_TEMPORARY;
-		tgt->SrcReg[0].Index = inst.DstReg.Index;
+	/* Cannot read texture coordinate from constants file */
+	if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) {
+		struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+
+		inst_mov->I.Opcode = OPCODE_MOV;
+		inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst_mov->I.DstReg.Index = rc_find_free_temporary(c);
+		inst_mov->I.SrcReg[0] = inst->I.SrcReg[0];
+
+		reset_srcreg(&inst->I.SrcReg[0]);
+		inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+		inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index;
 	}
 
 	return GL_TRUE;
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h
index 2d094c3618..0ac46dbd9c 100644
--- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h
@@ -44,6 +44,6 @@ extern void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler
 
 extern void r300FragmentProgramDump(struct rX00_fragment_program_code *c);
 
-extern GLboolean r300_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data);
+extern GLboolean r300_transform_TEX(struct radeon_compiler * c, struct rc_instruction* inst, void* data);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
index 7ac57d0d49..bdab61f32d 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
@@ -250,6 +250,8 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 
 	rewrite_depth_out(c->program);
 
+	rc_mesa_to_rc_program(&c->Base, c->program);
+
 	if (c->is_r500) {
 		struct radeon_program_transformation transformations[] = {
 			{ &r500_transform_TEX, c },
@@ -257,24 +259,22 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
 			{ &radeonTransformDeriv, 0 },
 			{ &radeonTransformTrigScale, 0 }
 		};
-		radeonLocalTransform(c->program, 4, transformations);
+		radeonLocalTransform(&c->Base, 4, transformations);
 	} else {
 		struct radeon_program_transformation transformations[] = {
 			{ &r300_transform_TEX, c },
 			{ &radeonTransformALU, 0 },
 			{ &radeonTransformTrigSimple, 0 }
 		};
-		radeonLocalTransform(c->program, 3, transformations);
+		radeonLocalTransform(&c->Base, 3, transformations);
 	}
 
 	if (c->Base.Debug) {
 		_mesa_printf("Fragment Program: After native rewrite:\n");
-		_mesa_print_program(c->program);
+		rc_print_program(&c->Base.Program);
 		fflush(stdout);
 	}
 
-	rc_mesa_to_rc_program(&c->Base, c->program);
-
 	if (c->is_r500) {
 		struct radeon_nqssadce_descr nqssadce = {
 			.Init = &nqssadce_init,
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
index c9d0996d44..90910c45c6 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -538,47 +538,43 @@ static void allocate_temporary_registers(struct r300_vertex_program_compiler * c
  * Introduce intermediate MOVs to temporary registers to account for this.
  */
 static GLboolean transform_source_conflicts(
-	struct radeon_transform_context *t,
-	struct prog_instruction* orig_inst,
+	struct radeon_compiler *c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	struct prog_instruction inst = *orig_inst;
-	struct prog_instruction * dst;
-	GLuint num_operands = _mesa_num_inst_src_regs(inst.Opcode);
+	GLuint num_operands = _mesa_num_inst_src_regs(inst->I.Opcode);
 
 	if (num_operands == 3) {
-		if (t_src_conflict(inst.SrcReg[1], inst.SrcReg[2])
-		    || t_src_conflict(inst.SrcReg[0], inst.SrcReg[2])) {
-			int tmpreg = radeonFindFreeTemporary(t);
-			struct prog_instruction * inst_mov = radeonAppendInstructions(t->Program, 1);
-			inst_mov->Opcode = OPCODE_MOV;
-			inst_mov->DstReg.File = PROGRAM_TEMPORARY;
-			inst_mov->DstReg.Index = tmpreg;
-			inst_mov->SrcReg[0] = inst.SrcReg[2];
-
-			reset_srcreg(&inst.SrcReg[2]);
-			inst.SrcReg[2].File = PROGRAM_TEMPORARY;
-			inst.SrcReg[2].Index = tmpreg;
+		if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[2])
+		    || t_src_conflict(inst->I.SrcReg[0], inst->I.SrcReg[2])) {
+			int tmpreg = rc_find_free_temporary(c);
+			struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+			inst_mov->I.Opcode = OPCODE_MOV;
+			inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_mov->I.DstReg.Index = tmpreg;
+			inst_mov->I.SrcReg[0] = inst->I.SrcReg[2];
+
+			reset_srcreg(&inst->I.SrcReg[2]);
+			inst->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+			inst->I.SrcReg[2].Index = tmpreg;
 		}
 	}
 
 	if (num_operands >= 2) {
-		if (t_src_conflict(inst.SrcReg[1], inst.SrcReg[0])) {
-			int tmpreg = radeonFindFreeTemporary(t);
-			struct prog_instruction * inst_mov = radeonAppendInstructions(t->Program, 1);
-			inst_mov->Opcode = OPCODE_MOV;
-			inst_mov->DstReg.File = PROGRAM_TEMPORARY;
-			inst_mov->DstReg.Index = tmpreg;
-			inst_mov->SrcReg[0] = inst.SrcReg[1];
-
-			reset_srcreg(&inst.SrcReg[1]);
-			inst.SrcReg[1].File = PROGRAM_TEMPORARY;
-			inst.SrcReg[1].Index = tmpreg;
+		if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[0])) {
+			int tmpreg = rc_find_free_temporary(c);
+			struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+			inst_mov->I.Opcode = OPCODE_MOV;
+			inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_mov->I.DstReg.Index = tmpreg;
+			inst_mov->I.SrcReg[0] = inst->I.SrcReg[1];
+
+			reset_srcreg(&inst->I.SrcReg[1]);
+			inst->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+			inst->I.SrcReg[1].Index = tmpreg;
 		}
 	}
 
-	dst = radeonAppendInstructions(t->Program, 1);
-	*dst = inst;
 	return GL_TRUE;
 }
 
@@ -782,16 +778,18 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler)
 
 	addArtificialOutputs(compiler);
 
+	rc_mesa_to_rc_program(&compiler->Base, compiler->program);
+
 	{
 		struct radeon_program_transformation transformations[] = {
 			{ &r300_transform_vertex_alu, 0 },
 		};
-		radeonLocalTransform(compiler->program, 1, transformations);
+		radeonLocalTransform(&compiler->Base, 1, transformations);
 	}
 
 	if (compiler->Base.Debug) {
 		fprintf(stderr, "Vertex program after native rewrite:\n");
-		_mesa_print_program(compiler->program);
+		rc_print_program(&compiler->Base.Program);
 		fflush(stdout);
 	}
 
@@ -803,17 +801,15 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler)
 		struct radeon_program_transformation transformations[] = {
 			{ &transform_source_conflicts, 0 },
 		};
-		radeonLocalTransform(compiler->program, 1, transformations);
+		radeonLocalTransform(&compiler->Base, 1, transformations);
 	}
 
 	if (compiler->Base.Debug) {
 		fprintf(stderr, "Vertex program after source conflict resolve:\n");
-		_mesa_print_program(compiler->program);
+		rc_print_program(&compiler->Base.Program);
 		fflush(stdout);
 	}
 
-	rc_mesa_to_rc_program(&compiler->Base, compiler->program);
-
 	{
 		struct radeon_nqssadce_descr nqssadce = {
 			.Init = &nqssadceInit,
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
index 6d2158dbfc..7e2faed690 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
@@ -29,152 +29,139 @@
 
 #include "../r300_reg.h"
 
-static struct prog_src_register shadow_ambient(struct gl_program *program, int tmu)
+static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
 {
-	gl_state_index fail_value_tokens[STATE_LENGTH] = {
-		STATE_INTERNAL, STATE_SHADOW_AMBIENT, 0, 0, 0
-	};
 	struct prog_src_register reg = { 0, };
 
-	fail_value_tokens[2] = tmu;
 	reg.File = PROGRAM_STATE_VAR;
-	reg.Index = _mesa_add_state_reference(program->Parameters, fail_value_tokens);
+	reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
 	reg.Swizzle = SWIZZLE_WWWW;
 	return reg;
 }
 
 /**
  * Transform TEX, TXP, TXB, and KIL instructions in the following way:
- *  - premultiply texture coordinates for RECT
- *  - extract operand swizzles
- *  - introduce a temporary register when write masks are needed
- *
+ *  - implement texture compare (shadow extensions)
+ *  - extract non-native source / destination operands
  */
 GLboolean r500_transform_TEX(
-	struct radeon_transform_context *t,
-	struct prog_instruction* orig_inst, void* data)
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
+	void* data)
 {
 	struct r300_fragment_program_compiler *compiler =
 		(struct r300_fragment_program_compiler*)data;
-	struct prog_instruction inst = *orig_inst;
-	struct prog_instruction* tgt;
-	GLboolean destredirect = GL_FALSE;
-
-	if (inst.Opcode != OPCODE_TEX &&
-	    inst.Opcode != OPCODE_TXB &&
-	    inst.Opcode != OPCODE_TXP &&
-	    inst.Opcode != OPCODE_KIL)
+
+	if (inst->I.Opcode != OPCODE_TEX &&
+	    inst->I.Opcode != OPCODE_TXB &&
+	    inst->I.Opcode != OPCODE_TXP &&
+	    inst->I.Opcode != OPCODE_KIL)
 		return GL_FALSE;
 
 	/* ARB_shadow & EXT_shadow_funcs */
-	if (inst.Opcode != OPCODE_KIL &&
-	    t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
-		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst.TexSrcUnit].texture_compare_func;
+	if (inst->I.Opcode != OPCODE_KIL &&
+	    c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) {
+		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
 
 		if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) {
-			tgt = radeonAppendInstructions(t->Program, 1);
+			inst->I.Opcode = OPCODE_MOV;
 
-			tgt->Opcode = OPCODE_MOV;
-			tgt->DstReg = inst.DstReg;
 			if (comparefunc == GL_ALWAYS) {
-				tgt->SrcReg[0].File = PROGRAM_BUILTIN;
-				tgt->SrcReg[0].Swizzle = SWIZZLE_1111;
+				inst->I.SrcReg[0].File = PROGRAM_BUILTIN;
+				inst->I.SrcReg[0].Swizzle = SWIZZLE_1111;
 			} else {
-				tgt->SrcReg[0] = shadow_ambient(t->Program, inst.TexSrcUnit);
+				inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit);
 			}
+
 			return GL_TRUE;
+		} else {
+			GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+			GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode;
+			struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
+			struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
+			struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
+			int pass, fail;
+
+			inst_rcp->I.Opcode = OPCODE_RCP;
+			inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_rcp->I.DstReg.Index = rc_find_free_temporary(c);
+			inst_rcp->I.DstReg.WriteMask = WRITEMASK_W;
+			inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0];
+			inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW;
+
+			inst_cmp->I.DstReg = inst->I.DstReg;
+			inst->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst->I.DstReg.Index = rc_find_free_temporary(c);
+			inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+
+			inst_mad->I.Opcode = OPCODE_MAD;
+			inst_mad->I.DstReg.File = PROGRAM_TEMPORARY;
+			inst_mad->I.DstReg.Index = rc_find_free_temporary(c);
+			inst_mad->I.SrcReg[0] = inst->I.SrcReg[0];
+			inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
+			inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+			inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index;
+			inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW;
+			inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+			inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index;
+			if (depthmode == 0) /* GL_LUMINANCE */
+				inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
+			else if (depthmode == 2) /* GL_ALPHA */
+				inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW;
+
+			/* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
+			 *   r  < tex  <=>      -tex+r < 0
+			 *   r >= tex  <=> not (-tex+r < 0 */
+			if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
+				inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW;
+			else
+				inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW;
+
+			inst_cmp->I.Opcode = OPCODE_CMP;
+			/* DstReg has been filled out above */
+			inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+			inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index;
+
+			if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
+				pass = 1;
+				fail = 2;
+			} else {
+				pass = 2;
+				fail = 1;
+			}
+
+			inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN;
+			inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111;
+			inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit);
 		}
+	}
 
-		inst.DstReg.File = PROGRAM_TEMPORARY;
-		inst.DstReg.Index = radeonFindFreeTemporary(t);
-		inst.DstReg.WriteMask = WRITEMASK_XYZW;
-	} else if (inst.Opcode != OPCODE_KIL && inst.DstReg.File != PROGRAM_TEMPORARY) {
-		int tempreg = radeonFindFreeTemporary(t);
+	/* Cannot write texture to output registers */
+	if (inst->I.Opcode != OPCODE_KIL && inst->I.DstReg.File != PROGRAM_TEMPORARY) {
+		struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
 
-		inst.DstReg.File = PROGRAM_TEMPORARY;
-		inst.DstReg.Index = tempreg;
-		inst.DstReg.WriteMask = WRITEMASK_XYZW;
-		destredirect = GL_TRUE;
-	}
+		inst_mov->I.Opcode = OPCODE_MOV;
+		inst_mov->I.DstReg = inst->I.DstReg;
+		inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+		inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c);
 
-	if (inst.SrcReg[0].File != PROGRAM_TEMPORARY && inst.SrcReg[0].File != PROGRAM_INPUT) {
-		int tmpreg = radeonFindFreeTemporary(t);
-		tgt = radeonAppendInstructions(t->Program, 1);
-		tgt->Opcode = OPCODE_MOV;
-		tgt->DstReg.File = PROGRAM_TEMPORARY;
-		tgt->DstReg.Index = tmpreg;
-		tgt->SrcReg[0] = inst.SrcReg[0];
-
-		reset_srcreg(&inst.SrcReg[0]);
-		inst.SrcReg[0].File = PROGRAM_TEMPORARY;
-		inst.SrcReg[0].Index = tmpreg;
+		inst->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+		inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
 	}
 
-	tgt = radeonAppendInstructions(t->Program, 1);
-	_mesa_copy_instructions(tgt, &inst, 1);
-
-	if (inst.Opcode != OPCODE_KIL &&
-	    t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
-		GLuint comparefunc = GL_NEVER + compiler->state.unit[inst.TexSrcUnit].texture_compare_func;
-		GLuint depthmode = compiler->state.unit[inst.TexSrcUnit].depth_texture_mode;
-		int rcptemp = radeonFindFreeTemporary(t);
-		int pass, fail;
-
-		tgt = radeonAppendInstructions(t->Program, 3);
-
-		tgt[0].Opcode = OPCODE_RCP;
-		tgt[0].DstReg.File = PROGRAM_TEMPORARY;
-		tgt[0].DstReg.Index = rcptemp;
-		tgt[0].DstReg.WriteMask = WRITEMASK_W;
-		tgt[0].SrcReg[0] = inst.SrcReg[0];
-		tgt[0].SrcReg[0].Swizzle = SWIZZLE_WWWW;
-
-		tgt[1].Opcode = OPCODE_MAD;
-		tgt[1].DstReg = inst.DstReg;
-		tgt[1].DstReg.WriteMask = orig_inst->DstReg.WriteMask;
-		tgt[1].SrcReg[0] = inst.SrcReg[0];
-		tgt[1].SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
-		tgt[1].SrcReg[1].File = PROGRAM_TEMPORARY;
-		tgt[1].SrcReg[1].Index = rcptemp;
-		tgt[1].SrcReg[1].Swizzle = SWIZZLE_WWWW;
-		tgt[1].SrcReg[2].File = PROGRAM_TEMPORARY;
-		tgt[1].SrcReg[2].Index = inst.DstReg.Index;
-		if (depthmode == 0) /* GL_LUMINANCE */
-			tgt[1].SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
-		else if (depthmode == 2) /* GL_ALPHA */
-			tgt[1].SrcReg[2].Swizzle = SWIZZLE_WWWW;
-
-		/* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
-		 *   r  < tex  <=>      -tex+r < 0
-		 *   r >= tex  <=> not (-tex+r < 0 */
-		if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
-			tgt[1].SrcReg[2].Negate = tgt[0].SrcReg[2].Negate ^ NEGATE_XYZW;
-		else
-			tgt[1].SrcReg[0].Negate = tgt[0].SrcReg[0].Negate ^ NEGATE_XYZW;
-
-		tgt[2].Opcode = OPCODE_CMP;
-		tgt[2].DstReg = orig_inst->DstReg;
-		tgt[2].SrcReg[0].File = PROGRAM_TEMPORARY;
-		tgt[2].SrcReg[0].Index = tgt[1].DstReg.Index;
-
-		if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
-			pass = 1;
-			fail = 2;
-		} else {
-			pass = 2;
-			fail = 1;
-		}
+	/* Cannot read texture coordinate from constants file */
+	if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) {
+		struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
 
-		tgt[2].SrcReg[pass].File = PROGRAM_BUILTIN;
-		tgt[2].SrcReg[pass].Swizzle = SWIZZLE_1111;
-		tgt[2].SrcReg[fail] = shadow_ambient(t->Program, inst.TexSrcUnit);
-	} else if (destredirect) {
-		tgt = radeonAppendInstructions(t->Program, 1);
+		inst_mov->I.Opcode = OPCODE_MOV;
+		inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst_mov->I.DstReg.Index = rc_find_free_temporary(c);
+		inst_mov->I.SrcReg[0] = inst->I.SrcReg[0];
 
-		tgt->Opcode = OPCODE_MOV;
-		tgt->DstReg = orig_inst->DstReg;
-		tgt->SrcReg[0].File = PROGRAM_TEMPORARY;
-		tgt->SrcReg[0].Index = inst.DstReg.Index;
+		reset_srcreg(&inst->I.SrcReg[0]);
+		inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+		inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index;
 	}
 
 	return GL_TRUE;
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
index e405267bb3..9091f65cd2 100644
--- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
@@ -47,6 +47,9 @@ extern GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register r
 
 extern void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src);
 
-extern GLboolean r500_transform_TEX(struct radeon_transform_context *t, struct prog_instruction* orig_inst, void* data);
+extern GLboolean r500_transform_TEX(
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
+	void* data);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.c b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
index b94e534d9c..c7923004df 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_code.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
@@ -76,3 +76,95 @@ unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * cons
 
 	return index;
 }
+
+
+/**
+ * Add a state vector to the constant list, while trying to avoid duplicates.
+ */
+unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state0, unsigned state1)
+{
+	unsigned index;
+	struct rc_constant constant;
+
+	for(index = 0; index < c->Count; ++index) {
+		if (c->Constants[index].Type == RC_CONSTANT_STATE) {
+			if (c->Constants[index].u.State[0] == state0 &&
+			    c->Constants[index].u.State[1] == state1)
+				return index;
+		}
+	}
+
+	memset(&constant, 0, sizeof(constant));
+	constant.Type = RC_CONSTANT_STATE;
+	constant.Size = 4;
+	constant.u.State[0] = state0;
+	constant.u.State[1] = state1;
+
+	return rc_constants_add(c, &constant);
+}
+
+
+/**
+ * Add an immediate vector to the constant list, while trying to avoid
+ * duplicates.
+ */
+unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data)
+{
+	unsigned index;
+	struct rc_constant constant;
+
+	for(index = 0; index < c->Count; ++index) {
+		if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) {
+			if (!memcmp(c->Constants[index].u.Immediate, data, sizeof(float)*4))
+				return index;
+		}
+	}
+
+	memset(&constant, 0, sizeof(constant));
+	constant.Type = RC_CONSTANT_IMMEDIATE;
+	constant.Size = 4;
+	memcpy(constant.u.Immediate, data, sizeof(float) * 4);
+
+	return rc_constants_add(c, &constant);
+}
+
+
+/**
+ * Add an immediate scalar to the constant list, while trying to avoid
+ * duplicates.
+ */
+unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle)
+{
+	unsigned index;
+	int free_index = -1;
+	struct rc_constant constant;
+
+	for(index = 0; index < c->Count; ++index) {
+		if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) {
+			for(unsigned comp = 0; comp < c->Constants[index].Size; ++comp) {
+				if (c->Constants[index].u.Immediate[comp] == data) {
+					*swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp);
+					return index;
+				}
+			}
+
+			if (c->Constants[index].Size < 4)
+				free_index = index;
+		}
+	}
+
+	if (free_index >= 0) {
+		unsigned comp = c->Constants[free_index].Size++;
+		c->Constants[free_index].u.Immediate[comp] = data;
+		*swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp);
+		return free_index;
+	}
+
+	memset(&constant, 0, sizeof(constant));
+	constant.Type = RC_CONSTANT_IMMEDIATE;
+	constant.Size = 1;
+	constant.u.Immediate[0] = data;
+	*swizzle = SWIZZLE_XXXX;
+
+	return rc_constants_add(c, &constant);
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.h b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
index 3e6eb97b17..9cf4ed57bb 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_code.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
@@ -35,8 +35,6 @@
 
 
 #define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0)
-#define STATE_R300_TEXRECT_FACTOR (STATE_INTERNAL_DRIVER+1)
-
 
 enum {
 	/**
@@ -50,17 +48,26 @@ enum {
 
 	/**
 	 * Constant referring to state that is known by this compiler,
-	 * i.e. *not* arbitrary Mesa (or other) state.
+	 * see RC_STATE_xxx, i.e. *not* arbitrary Mesa (or other) state.
 	 */
 	RC_CONSTANT_STATE
 };
 
+enum {
+	RC_STATE_SHADOW_AMBIENT = 0,
+
+	RC_STATE_R300_WINDOW_DIMENSION,
+	RC_STATE_R300_TEXRECT_FACTOR
+};
+
 struct rc_constant {
 	unsigned Type:2; /**< RC_CONSTANT_xxx */
+	unsigned Size:3;
+
 	union {
 		unsigned External;
 		float Immediate[4];
-		unsigned State[4];
+		unsigned State[2];
 	} u;
 };
 
@@ -75,6 +82,9 @@ void rc_constants_init(struct rc_constant_list * c);
 void rc_constants_copy(struct rc_constant_list * dst, struct rc_constant_list * src);
 void rc_constants_destroy(struct rc_constant_list * c);
 unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * constant);
+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);
 
 /**
  * Stores state that influences the compilation of a fragment program.
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
index 6630db6279..d9ee43017d 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
@@ -67,7 +67,7 @@ void rc_error(struct radeon_compiler * c, const char * fmt, ...);
 struct r300_fragment_program_compiler {
 	struct radeon_compiler Base;
 	struct rX00_fragment_program_code *code;
-	struct gl_program *program;
+	struct gl_program * program;
 	struct r300_fragment_program_external_state state;
 	GLboolean is_r500;
 };
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.c b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
index 50a0ce8743..208d3b90c8 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
@@ -48,85 +48,27 @@
  * one instruction at a time.
  */
 void radeonLocalTransform(
-	struct gl_program *program,
+	struct radeon_compiler * c,
 	int num_transformations,
 	struct radeon_program_transformation* transformations)
 {
-	struct radeon_transform_context ctx;
-	int ip;
+	struct rc_instruction * inst = c->Program.Instructions.Next;
 
-	ctx.Program = program;
-	ctx.OldInstructions = program->Instructions;
-	ctx.OldNumInstructions = program->NumInstructions;
-
-	program->Instructions = 0;
-	program->NumInstructions = 0;
-
-	for(ip = 0; ip < ctx.OldNumInstructions; ++ip) {
-		struct prog_instruction *instr = ctx.OldInstructions + ip;
+	while(inst != &c->Program.Instructions) {
+		struct rc_instruction * current = inst;
 		int i;
 
+		inst = inst->Next;
+
 		for(i = 0; i < num_transformations; ++i) {
 			struct radeon_program_transformation* t = transformations + i;
 
-			if (t->function(&ctx, instr, t->userData))
+			if (t->function(c, current, t->userData))
 				break;
 		}
-
-		if (i >= num_transformations) {
-			struct prog_instruction* dest = radeonAppendInstructions(program, 1);
-			_mesa_copy_instructions(dest, instr, 1);
-		}
-	}
-
-	_mesa_free_instructions(ctx.OldInstructions, ctx.OldNumInstructions);
-}
-
-
-static void scan_instructions(GLboolean* used, const struct prog_instruction* insts, GLuint count)
-{
-	GLuint i;
-	for (i = 0; i < count; i++) {
-		const struct prog_instruction *inst = insts + i;
-		const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
-		GLuint k;
-
-		for (k = 0; k < n; k++) {
-			if (inst->SrcReg[k].File == PROGRAM_TEMPORARY)
-				used[inst->SrcReg[k].Index] = GL_TRUE;
-		}
 	}
 }
 
-GLint radeonFindFreeTemporary(struct radeon_transform_context *t)
-{
-	GLboolean used[MAX_PROGRAM_TEMPS];
-	GLuint i;
-
-	_mesa_memset(used, 0, sizeof(used));
-	scan_instructions(used, t->Program->Instructions, t->Program->NumInstructions);
-	scan_instructions(used, t->OldInstructions, t->OldNumInstructions);
-
-	for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
-		if (!used[i])
-			return i;
-	}
-
-	return -1;
-}
-
-
-/**
- * Append the given number of instructions to the program and return a
- * pointer to the first new instruction.
- */
-struct prog_instruction *radeonAppendInstructions(struct gl_program *program, int count)
-{
-	int oldnum = program->NumInstructions;
-	_mesa_insert_instructions(program, oldnum, count);
-	return program->Instructions + oldnum;
-}
-
 
 GLint rc_find_free_temporary(struct radeon_compiler * c)
 {
@@ -206,6 +148,7 @@ void rc_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * progr
 		struct rc_constant constant;
 
 		constant.Type = RC_CONSTANT_EXTERNAL;
+		constant.Size = 4;
 		constant.u.External = i;
 
 		rc_constants_add(&c->Program.Constants, &constant);
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.h b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
index fae8c6babe..1a34b50ed7 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
@@ -88,18 +88,6 @@ static INLINE void reset_srcreg(struct prog_src_register* reg)
 
 
 /**
- * Transformation context that is passed to local transformations.
- *
- * Care must be taken with some operations during transformation,
- * e.g. finding new temporary registers must use @ref radeonFindFreeTemporary
- */
-struct radeon_transform_context {
-	struct gl_program *Program;
-	struct prog_instruction *OldInstructions;
-	GLuint OldNumInstructions;
-};
-
-/**
  * A transformation that can be passed to \ref radeonLocalTransform.
  *
  * The function will be called once for each instruction.
@@ -111,24 +99,17 @@ struct radeon_transform_context {
  */
 struct radeon_program_transformation {
 	GLboolean (*function)(
-		struct radeon_transform_context*,
-		struct prog_instruction*,
+		struct radeon_compiler*,
+		struct rc_instruction*,
 		void*);
 	void *userData;
 };
 
 void radeonLocalTransform(
-	struct gl_program *program,
+	struct radeon_compiler *c,
 	int num_transformations,
 	struct radeon_program_transformation* transformations);
 
-/**
- * Find a usable free temporary register during program transformation
- */
-GLint radeonFindFreeTemporary(struct radeon_transform_context *ctx);
-
-struct prog_instruction *radeonAppendInstructions(struct gl_program *program, int count);
-
 GLint rc_find_free_temporary(struct radeon_compiler * c);
 
 struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c);
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 dd20b4603e..609e510ff2 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
@@ -35,49 +35,52 @@
 
 #include "radeon_program_alu.h"
 
-#include "shader/prog_parameter.h"
+#include "radeon_compiler.h"
 
 
-static struct prog_instruction *emit1(struct gl_program* p,
+static struct rc_instruction *emit1(
+	struct radeon_compiler * c, struct rc_instruction * after,
 	gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
 	struct prog_src_register SrcReg)
 {
-	struct prog_instruction *fpi = radeonAppendInstructions(p, 1);
+	struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 
-	fpi->Opcode = Opcode;
-	fpi->SaturateMode = Saturate;
-	fpi->DstReg = DstReg;
-	fpi->SrcReg[0] = SrcReg;
+	fpi->I.Opcode = Opcode;
+	fpi->I.SaturateMode = Saturate;
+	fpi->I.DstReg = DstReg;
+	fpi->I.SrcReg[0] = SrcReg;
 	return fpi;
 }
 
-static struct prog_instruction *emit2(struct gl_program* p,
+static struct rc_instruction *emit2(
+	struct radeon_compiler * c, struct rc_instruction * after,
 	gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
 	struct prog_src_register SrcReg0, struct prog_src_register SrcReg1)
 {
-	struct prog_instruction *fpi = radeonAppendInstructions(p, 1);
+	struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 
-	fpi->Opcode = Opcode;
-	fpi->SaturateMode = Saturate;
-	fpi->DstReg = DstReg;
-	fpi->SrcReg[0] = SrcReg0;
-	fpi->SrcReg[1] = SrcReg1;
+	fpi->I.Opcode = Opcode;
+	fpi->I.SaturateMode = Saturate;
+	fpi->I.DstReg = DstReg;
+	fpi->I.SrcReg[0] = SrcReg0;
+	fpi->I.SrcReg[1] = SrcReg1;
 	return fpi;
 }
 
-static struct prog_instruction *emit3(struct gl_program* p,
+static struct rc_instruction *emit3(
+	struct radeon_compiler * c, struct rc_instruction * after,
 	gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
 	struct prog_src_register SrcReg0, struct prog_src_register SrcReg1,
 	struct prog_src_register SrcReg2)
 {
-	struct prog_instruction *fpi = radeonAppendInstructions(p, 1);
+	struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
 
-	fpi->Opcode = Opcode;
-	fpi->SaturateMode = Saturate;
-	fpi->DstReg = DstReg;
-	fpi->SrcReg[0] = SrcReg0;
-	fpi->SrcReg[1] = SrcReg1;
-	fpi->SrcReg[2] = SrcReg2;
+	fpi->I.Opcode = Opcode;
+	fpi->I.SaturateMode = Saturate;
+	fpi->I.DstReg = DstReg;
+	fpi->I.SrcReg[0] = SrcReg0;
+	fpi->I.SrcReg[1] = SrcReg1;
+	fpi->I.SrcReg[2] = SrcReg2;
 	return fpi;
 }
 
@@ -171,58 +174,63 @@ static struct prog_src_register scalar(struct prog_src_register reg)
 	return swizzle(reg, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X);
 }
 
-static void transform_ABS(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_ABS(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	struct prog_src_register src = inst->SrcReg[0];
+	struct prog_src_register src = inst->I.SrcReg[0];
 	src.Abs = 1;
 	src.Negate = NEGATE_NONE;
-	emit1(t->Program, OPCODE_MOV, inst->SaturateMode, inst->DstReg, src);
+	emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode, inst->I.DstReg, src);
+	rc_remove_instruction(inst);
 }
 
-static void transform_DP3(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_DP3(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	struct prog_src_register src0 = inst->SrcReg[0];
-	struct prog_src_register src1 = inst->SrcReg[1];
+	struct prog_src_register src0 = inst->I.SrcReg[0];
+	struct prog_src_register src1 = inst->I.SrcReg[1];
 	src0.Negate &= ~NEGATE_W;
 	src0.Swizzle &= ~(7 << (3 * 3));
 	src0.Swizzle |= SWIZZLE_ZERO << (3 * 3);
 	src1.Negate &= ~NEGATE_W;
 	src1.Swizzle &= ~(7 << (3 * 3));
 	src1.Swizzle |= SWIZZLE_ZERO << (3 * 3);
-	emit2(t->Program, OPCODE_DP4, inst->SaturateMode, inst->DstReg, src0, src1);
+	emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, src1);
+	rc_remove_instruction(inst);
 }
 
-static void transform_DPH(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_DPH(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	struct prog_src_register src0 = inst->SrcReg[0];
+	struct prog_src_register src0 = inst->I.SrcReg[0];
 	src0.Negate &= ~NEGATE_W;
 	src0.Swizzle &= ~(7 << (3 * 3));
 	src0.Swizzle |= SWIZZLE_ONE << (3 * 3);
-	emit2(t->Program, OPCODE_DP4, inst->SaturateMode, inst->DstReg, src0, inst->SrcReg[1]);
+	emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, inst->I.SrcReg[1]);
+	rc_remove_instruction(inst);
 }
 
 /**
  * [1, src0.y*src1.y, src0.z, src1.w]
  * So basically MUL with lotsa swizzling.
  */
-static void transform_DST(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_DST(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	emit2(t->Program, OPCODE_MUL, inst->SaturateMode, inst->DstReg,
-		swizzle(inst->SrcReg[0], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE),
-		swizzle(inst->SrcReg[1], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_ONE, SWIZZLE_W));
+	emit2(c, inst->Prev, OPCODE_MUL, inst->I.SaturateMode, inst->I.DstReg,
+		swizzle(inst->I.SrcReg[0], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE),
+		swizzle(inst->I.SrcReg[1], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_ONE, SWIZZLE_W));
+	rc_remove_instruction(inst);
 }
 
-static void transform_FLR(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_FLR(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
-	emit1(t->Program, OPCODE_FRC, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0]);
-	emit2(t->Program, OPCODE_ADD, inst->SaturateMode, inst->DstReg,
-		inst->SrcReg[0], negate(srcreg(PROGRAM_TEMPORARY, tempreg)));
+	int tempreg = rc_find_free_temporary(c);
+	emit1(c, inst->Prev, OPCODE_FRC, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0]);
+	emit2(c, inst->Prev, OPCODE_ADD, inst->I.SaturateMode, inst->I.DstReg,
+		inst->I.SrcReg[0], negate(srcreg(PROGRAM_TEMPORARY, tempreg)));
+	rc_remove_instruction(inst);
 }
 
 /**
@@ -243,152 +251,159 @@ static void transform_FLR(struct radeon_transform_context* t,
  * 5 slots, if the subsequent optimization passes are clever enough
  * to pair instructions correctly.
  */
-static void transform_LIT(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_LIT(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	static const GLfloat LitConst[4] = { -127.999999 };
-
 	GLuint constant;
 	GLuint constant_swizzle;
 	GLuint temp;
-	int needTemporary = 0;
 	struct prog_src_register srctemp;
 
-	constant = _mesa_add_unnamed_constant(t->Program->Parameters, LitConst, 1, &constant_swizzle);
+	constant = rc_constants_add_immediate_scalar(&c->Program.Constants, -127.999999, &constant_swizzle);
 
-	if (inst->DstReg.WriteMask != WRITEMASK_XYZW) {
-		needTemporary = 1;
-	} else if (inst->DstReg.File != PROGRAM_TEMPORARY) {
-		// LIT is typically followed by DP3/DP4, so there's no point
-		// in creating special code for this case
-		needTemporary = 1;
-	}
+	if (inst->I.DstReg.WriteMask != WRITEMASK_XYZW || inst->I.DstReg.File != PROGRAM_TEMPORARY) {
+		struct rc_instruction * inst_mov;
 
-	if (needTemporary) {
-		temp = radeonFindFreeTemporary(t);
-	} else {
-		temp = inst->DstReg.Index;
+		inst_mov = emit1(c, inst,
+			OPCODE_MOV, 0, inst->I.DstReg,
+			srcreg(PROGRAM_TEMPORARY, rc_find_free_temporary(c)));
+
+		inst->I.DstReg.File = PROGRAM_TEMPORARY;
+		inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+		inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
 	}
+
+	temp = inst->I.DstReg.Index;
 	srctemp = srcreg(PROGRAM_TEMPORARY, temp);
 
 	// tmp.x = max(0.0, Src.x);
 	// tmp.y = max(0.0, Src.y);
 	// tmp.w = clamp(Src.z, -128+eps, 128-eps);
-	emit2(t->Program, OPCODE_MAX, 0,
+	emit2(c, inst->Prev, OPCODE_MAX, 0,
 		dstregtmpmask(temp, WRITEMASK_XYW),
-		inst->SrcReg[0],
+		inst->I.SrcReg[0],
 		swizzle(srcreg(PROGRAM_CONSTANT, constant),
 			SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO, constant_swizzle&3));
-	emit2(t->Program, OPCODE_MIN, 0,
+	emit2(c, inst->Prev, OPCODE_MIN, 0,
 		dstregtmpmask(temp, WRITEMASK_Z),
 		swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 		negate(srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle)));
 
 	// tmp.w = Pow(tmp.y, tmp.w)
-	emit1(t->Program, OPCODE_LG2, 0,
+	emit1(c, inst->Prev, OPCODE_LG2, 0,
 		dstregtmpmask(temp, WRITEMASK_W),
 		swizzle(srctemp, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y));
-	emit2(t->Program, OPCODE_MUL, 0,
+	emit2(c, inst->Prev, OPCODE_MUL, 0,
 		dstregtmpmask(temp, WRITEMASK_W),
 		swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 		swizzle(srctemp, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z));
-	emit1(t->Program, OPCODE_EX2, 0,
+	emit1(c, inst->Prev, OPCODE_EX2, 0,
 		dstregtmpmask(temp, WRITEMASK_W),
 		swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
 
 	// tmp.z = (tmp.x > 0) ? tmp.w : 0.0
-	emit3(t->Program, OPCODE_CMP, inst->SaturateMode,
+	emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode,
 		dstregtmpmask(temp, WRITEMASK_Z),
 		negate(swizzle(srctemp, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
 		swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 		builtin_zero);
 
 	// tmp.x, tmp.y, tmp.w = 1.0, tmp.x, 1.0
-	emit1(t->Program, OPCODE_MOV, inst->SaturateMode,
+	emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode,
 		dstregtmpmask(temp, WRITEMASK_XYW),
 		swizzle(srctemp, SWIZZLE_ONE, SWIZZLE_X, SWIZZLE_ONE, SWIZZLE_ONE));
 
-	if (needTemporary)
-		emit1(t->Program, OPCODE_MOV, 0, inst->DstReg, srctemp);
+	rc_remove_instruction(inst);
 }
 
-static void transform_LRP(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_LRP(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
+	int tempreg = rc_find_free_temporary(c);
 
-	emit2(t->Program, OPCODE_ADD, 0,
+	emit2(c, inst->Prev, OPCODE_ADD, 0,
 		dstreg(PROGRAM_TEMPORARY, tempreg),
-		inst->SrcReg[1], negate(inst->SrcReg[2]));
-	emit3(t->Program, OPCODE_MAD, inst->SaturateMode,
-		inst->DstReg,
-		inst->SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[2]);
+		inst->I.SrcReg[1], negate(inst->I.SrcReg[2]));
+	emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode,
+		inst->I.DstReg,
+		inst->I.SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[2]);
+
+	rc_remove_instruction(inst);
 }
 
-static void transform_POW(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_POW(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
+	int tempreg = rc_find_free_temporary(c);
 	struct prog_dst_register tempdst = dstreg(PROGRAM_TEMPORARY, tempreg);
 	struct prog_src_register tempsrc = srcreg(PROGRAM_TEMPORARY, tempreg);
 	tempdst.WriteMask = WRITEMASK_W;
 	tempsrc.Swizzle = SWIZZLE_WWWW;
 
-	emit1(t->Program, OPCODE_LG2, 0, tempdst, scalar(inst->SrcReg[0]));
-	emit2(t->Program, OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->SrcReg[1]));
-	emit1(t->Program, OPCODE_EX2, inst->SaturateMode, inst->DstReg, tempsrc);
+	emit1(c, inst->Prev, OPCODE_LG2, 0, tempdst, scalar(inst->I.SrcReg[0]));
+	emit2(c, inst->Prev, OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->I.SrcReg[1]));
+	emit1(c, inst->Prev, OPCODE_EX2, inst->I.SaturateMode, inst->I.DstReg, tempsrc);
+
+	rc_remove_instruction(inst);
 }
 
-static void transform_RSQ(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_RSQ(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	emit1(t->Program, OPCODE_RSQ, inst->SaturateMode, inst->DstReg, absolute(inst->SrcReg[0]));
+	inst->I.SrcReg[0] = absolute(inst->I.SrcReg[0]);
 }
 
-static void transform_SGE(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_SGE(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
+	int tempreg = rc_find_free_temporary(c);
 
-	emit2(t->Program, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0], negate(inst->SrcReg[1]));
-	emit3(t->Program, OPCODE_CMP, inst->SaturateMode, inst->DstReg,
+	emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1]));
+	emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg,
 		srcreg(PROGRAM_TEMPORARY, tempreg), builtin_zero, builtin_one);
+
+	rc_remove_instruction(inst);
 }
 
-static void transform_SLT(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_SLT(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
+	int tempreg = rc_find_free_temporary(c);
 
-	emit2(t->Program, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->SrcReg[0], negate(inst->SrcReg[1]));
-	emit3(t->Program, OPCODE_CMP, inst->SaturateMode, inst->DstReg,
+	emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1]));
+	emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg,
 		srcreg(PROGRAM_TEMPORARY, tempreg), builtin_one, builtin_zero);
+
+	rc_remove_instruction(inst);
 }
 
-static void transform_SUB(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_SUB(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	emit2(t->Program, OPCODE_ADD, inst->SaturateMode, inst->DstReg, inst->SrcReg[0], negate(inst->SrcReg[1]));
+	inst->I.Opcode = OPCODE_ADD;
+	inst->I.SrcReg[1] = negate(inst->I.SrcReg[1]);
 }
 
-static void transform_SWZ(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_SWZ(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	emit1(t->Program, OPCODE_MOV, inst->SaturateMode, inst->DstReg, inst->SrcReg[0]);
+	inst->I.Opcode = OPCODE_MOV;
 }
 
-static void transform_XPD(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_XPD(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
-	int tempreg = radeonFindFreeTemporary(t);
+	int tempreg = rc_find_free_temporary(c);
 
-	emit2(t->Program, OPCODE_MUL, 0, dstreg(PROGRAM_TEMPORARY, tempreg),
-		swizzle(inst->SrcReg[0], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
-		swizzle(inst->SrcReg[1], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W));
-	emit3(t->Program, OPCODE_MAD, inst->SaturateMode, inst->DstReg,
-		swizzle(inst->SrcReg[0], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W),
-		swizzle(inst->SrcReg[1], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
+	emit2(c, inst->Prev, OPCODE_MUL, 0, dstreg(PROGRAM_TEMPORARY, tempreg),
+		swizzle(inst->I.SrcReg[0], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
+		swizzle(inst->I.SrcReg[1], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W));
+	emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode, inst->I.DstReg,
+		swizzle(inst->I.SrcReg[0], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W),
+		swizzle(inst->I.SrcReg[1], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
 		negate(srcreg(PROGRAM_TEMPORARY, tempreg)));
+
+	rc_remove_instruction(inst);
 }
 
 
@@ -406,63 +421,64 @@ static void transform_XPD(struct radeon_transform_context* t,
  *
  * @note should be applicable to R300 and R500 fragment programs.
  */
-GLboolean radeonTransformALU(struct radeon_transform_context* t,
-	struct prog_instruction* inst,
+GLboolean radeonTransformALU(
+	struct radeon_compiler * c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	switch(inst->Opcode) {
-	case OPCODE_ABS: transform_ABS(t, inst); return GL_TRUE;
-	case OPCODE_DPH: transform_DPH(t, inst); return GL_TRUE;
-	case OPCODE_DST: transform_DST(t, inst); return GL_TRUE;
-	case OPCODE_FLR: transform_FLR(t, inst); return GL_TRUE;
-	case OPCODE_LIT: transform_LIT(t, inst); return GL_TRUE;
-	case OPCODE_LRP: transform_LRP(t, inst); return GL_TRUE;
-	case OPCODE_POW: transform_POW(t, inst); return GL_TRUE;
-	case OPCODE_RSQ: transform_RSQ(t, inst); return GL_TRUE;
-	case OPCODE_SGE: transform_SGE(t, inst); return GL_TRUE;
-	case OPCODE_SLT: transform_SLT(t, inst); return GL_TRUE;
-	case OPCODE_SUB: transform_SUB(t, inst); return GL_TRUE;
-	case OPCODE_SWZ: transform_SWZ(t, inst); return GL_TRUE;
-	case OPCODE_XPD: transform_XPD(t, inst); return GL_TRUE;
+	switch(inst->I.Opcode) {
+	case OPCODE_ABS: transform_ABS(c, inst); return GL_TRUE;
+	case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE;
+	case OPCODE_DST: transform_DST(c, inst); return GL_TRUE;
+	case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE;
+	case OPCODE_LIT: transform_LIT(c, inst); return GL_TRUE;
+	case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE;
+	case OPCODE_POW: transform_POW(c, inst); return GL_TRUE;
+	case OPCODE_RSQ: transform_RSQ(c, inst); return GL_TRUE;
+	case OPCODE_SGE: transform_SGE(c, inst); return GL_TRUE;
+	case OPCODE_SLT: transform_SLT(c, inst); return GL_TRUE;
+	case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE;
+	case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE;
+	case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE;
 	default:
 		return GL_FALSE;
 	}
 }
 
 
-static void transform_r300_vertex_ABS(struct radeon_transform_context* t,
-	struct prog_instruction* inst)
+static void transform_r300_vertex_ABS(struct radeon_compiler* c,
+	struct rc_instruction* inst)
 {
 	/* Note: r500 can take absolute values, but r300 cannot. */
-	struct prog_src_register src1 = inst->SrcReg[0];
-	src1.Negate ^= NEGATE_XYZW;
-
-	emit2(t->Program, OPCODE_MAX, inst->SaturateMode, inst->DstReg, inst->SrcReg[0], src1);
+	inst->I.Opcode = OPCODE_MAX;
+	inst->I.SrcReg[1] = inst->I.SrcReg[0];
+	inst->I.SrcReg[1].Negate ^= NEGATE_XYZW;
 }
 
 /**
  * For use with radeonLocalTransform, this transforms non-native ALU
  * instructions of the r300 up to r500 vertex engine.
  */
-GLboolean r300_transform_vertex_alu(struct radeon_transform_context* t,
-	struct prog_instruction* inst,
+GLboolean r300_transform_vertex_alu(
+	struct radeon_compiler * c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	switch(inst->Opcode) {
-	case OPCODE_ABS: transform_r300_vertex_ABS(t, inst); return GL_TRUE;
-	case OPCODE_DP3: transform_DP3(t, inst); return GL_TRUE;
-	case OPCODE_DPH: transform_DPH(t, inst); return GL_TRUE;
-	case OPCODE_FLR: transform_FLR(t, inst); return GL_TRUE;
-	case OPCODE_LRP: transform_LRP(t, inst); return GL_TRUE;
-	case OPCODE_SUB: transform_SUB(t, inst); return GL_TRUE;
-	case OPCODE_SWZ: transform_SWZ(t, inst); return GL_TRUE;
-	case OPCODE_XPD: transform_XPD(t, inst); return GL_TRUE;
+	switch(inst->I.Opcode) {
+	case OPCODE_ABS: transform_r300_vertex_ABS(c, inst); return GL_TRUE;
+	case OPCODE_DP3: transform_DP3(c, inst); return GL_TRUE;
+	case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE;
+	case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE;
+	case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE;
+	case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE;
+	case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE;
+	case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE;
 	default:
 		return GL_FALSE;
 	}
 }
 
-static void sincos_constants(struct radeon_transform_context* t, GLuint *constants)
+static void sincos_constants(struct radeon_compiler* c, GLuint *constants)
 {
 	static const GLfloat SinCosConsts[2][4] = {
 		{
@@ -480,11 +496,8 @@ static void sincos_constants(struct radeon_transform_context* t, GLuint *constan
 	};
 	int i;
 
-	for(i = 0; i < 2; ++i) {
-		GLuint swz;
-		constants[i] = _mesa_add_unnamed_constant(t->Program->Parameters, SinCosConsts[i], 4, &swz);
-		ASSERT(swz == SWIZZLE_NOOP);
-	}
+	for(i = 0; i < 2; ++i)
+		constants[i] = rc_constants_add_immediate_vec4(&c->Program.Constants, SinCosConsts[i]);
 }
 
 /**
@@ -495,23 +508,24 @@ static void sincos_constants(struct radeon_transform_context* t, GLuint *constan
  * MAD tmp.y, tmp.x, |tmp.x|, -tmp.x
  * MAD dest, tmp.y, weight, tmp.x
  */
-static void sin_approx(struct radeon_transform_context* t,
+static void sin_approx(
+	struct radeon_compiler* c, struct rc_instruction * after,
 	struct prog_dst_register dst, struct prog_src_register src, const GLuint* constants)
 {
-	GLuint tempreg = radeonFindFreeTemporary(t);
+	GLuint tempreg = rc_find_free_temporary(c);
 
-	emit2(t->Program, OPCODE_MUL, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+	emit2(c, after->Prev, OPCODE_MUL, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
 		swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 		srcreg(PROGRAM_CONSTANT, constants[0]));
-	emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_X),
+	emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_X),
 		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
 		absolute(swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
 		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
-	emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_Y),
+	emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_Y),
 		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 		absolute(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
 		negate(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)));
-	emit3(t->Program, OPCODE_MAD, 0, dst,
+	emit3(c, after->Prev, OPCODE_MAD, 0, dst,
 		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
 		swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 		swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
@@ -522,78 +536,80 @@ static void sin_approx(struct radeon_transform_context* t,
  * using only the basic instructions
  *  MOV, ADD, MUL, MAD, FRC
  */
-GLboolean radeonTransformTrigSimple(struct radeon_transform_context* t,
-	struct prog_instruction* inst,
+GLboolean radeonTransformTrigSimple(struct radeon_compiler* c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	if (inst->Opcode != OPCODE_COS &&
-	    inst->Opcode != OPCODE_SIN &&
-	    inst->Opcode != OPCODE_SCS)
+	if (inst->I.Opcode != OPCODE_COS &&
+	    inst->I.Opcode != OPCODE_SIN &&
+	    inst->I.Opcode != OPCODE_SCS)
 		return GL_FALSE;
 
 	GLuint constants[2];
-	GLuint tempreg = radeonFindFreeTemporary(t);
+	GLuint tempreg = rc_find_free_temporary(c);
 
-	sincos_constants(t, constants);
+	sincos_constants(c, constants);
 
-	if (inst->Opcode == OPCODE_COS) {
+	if (inst->I.Opcode == OPCODE_COS) {
 		// MAD tmp.x, src, 1/(2*PI), 0.75
 		// FRC tmp.x, tmp.x
 		// MAD tmp.z, tmp.x, 2*PI, -PI
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
-			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
-		emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+		emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
 
-		sin_approx(t, inst->DstReg,
+		sin_approx(c, inst->Prev, inst->I.DstReg,
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			constants);
-	} else if (inst->Opcode == OPCODE_SIN) {
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
-			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+	} else if (inst->I.Opcode == OPCODE_SIN) {
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+			swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y));
-		emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+		emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
 
-		sin_approx(t, inst->DstReg,
+		sin_approx(c, inst->Prev, inst->I.DstReg,
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			constants);
 	} else {
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
-			swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+			swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W));
-		emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+		emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
 			srcreg(PROGRAM_TEMPORARY, tempreg));
-		emit3(t->Program, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+		emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
 			srcreg(PROGRAM_TEMPORARY, tempreg),
 			swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
 			negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
 
-		struct prog_dst_register dst = inst->DstReg;
+		struct prog_dst_register dst = inst->I.DstReg;
 
-		dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_X;
-		sin_approx(t, dst,
+		dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_X;
+		sin_approx(c, inst->Prev, dst,
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 			constants);
 
-		dst.WriteMask = inst->DstReg.WriteMask & WRITEMASK_Y;
-		sin_approx(t, dst,
+		dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_Y;
+		sin_approx(c, inst->Prev, dst,
 			swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
 			constants);
 	}
 
+	rc_remove_instruction(inst);
+
 	return GL_TRUE;
 }
 
@@ -606,50 +622,52 @@ GLboolean radeonTransformTrigSimple(struct radeon_transform_context* t,
  *
  * @warning This transformation implicitly changes the semantics of SIN and COS!
  */
-GLboolean radeonTransformTrigScale(struct radeon_transform_context* t,
-	struct prog_instruction* inst,
+GLboolean radeonTransformTrigScale(struct radeon_compiler* c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	if (inst->Opcode != OPCODE_COS &&
-	    inst->Opcode != OPCODE_SIN &&
-	    inst->Opcode != OPCODE_SCS)
+	if (inst->I.Opcode != OPCODE_COS &&
+	    inst->I.Opcode != OPCODE_SIN &&
+	    inst->I.Opcode != OPCODE_SCS)
 		return GL_FALSE;
 
-	static const GLfloat RCP_2PI[] = { 0.15915494309189535 };
+	static const GLfloat RCP_2PI = 0.15915494309189535;
 	GLuint temp;
 	GLuint constant;
 	GLuint constant_swizzle;
 
-	temp = radeonFindFreeTemporary(t);
-	constant = _mesa_add_unnamed_constant(t->Program->Parameters, RCP_2PI, 1, &constant_swizzle);
+	temp = rc_find_free_temporary(c);
+	constant = rc_constants_add_immediate_scalar(&c->Program.Constants, RCP_2PI, &constant_swizzle);
 
-	emit2(t->Program, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W),
-		swizzle(inst->SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+	emit2(c, inst->Prev, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W),
+		swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
 		srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle));
-	emit1(t->Program, OPCODE_FRC, 0, dstregtmpmask(temp, WRITEMASK_W),
+	emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(temp, WRITEMASK_W),
 		srcreg(PROGRAM_TEMPORARY, temp));
 
-	if (inst->Opcode == OPCODE_COS) {
-		emit1(t->Program, OPCODE_COS, inst->SaturateMode, inst->DstReg,
+	if (inst->I.Opcode == OPCODE_COS) {
+		emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, inst->I.DstReg,
 			srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
-	} else if (inst->Opcode == OPCODE_SIN) {
-		emit1(t->Program, OPCODE_SIN, inst->SaturateMode,
-			inst->DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
-	} else if (inst->Opcode == OPCODE_SCS) {
-		struct prog_dst_register moddst = inst->DstReg;
+	} else if (inst->I.Opcode == OPCODE_SIN) {
+		emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode,
+			inst->I.DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
+	} else if (inst->I.Opcode == OPCODE_SCS) {
+		struct prog_dst_register moddst = inst->I.DstReg;
 
-		if (inst->DstReg.WriteMask & WRITEMASK_X) {
+		if (inst->I.DstReg.WriteMask & WRITEMASK_X) {
 			moddst.WriteMask = WRITEMASK_X;
-			emit1(t->Program, OPCODE_COS, inst->SaturateMode, moddst,
+			emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, moddst,
 				srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
 		}
-		if (inst->DstReg.WriteMask & WRITEMASK_Y) {
+		if (inst->I.DstReg.WriteMask & WRITEMASK_Y) {
 			moddst.WriteMask = WRITEMASK_Y;
-			emit1(t->Program, OPCODE_SIN, inst->SaturateMode, moddst,
+			emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode, moddst,
 				srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
 		}
 	}
 
+	rc_remove_instruction(inst);
+
 	return GL_TRUE;
 }
 
@@ -661,21 +679,15 @@ GLboolean radeonTransformTrigScale(struct radeon_transform_context* t,
  * @warning This explicitly changes the form of DDX and DDY!
  */
 
-GLboolean radeonTransformDeriv(struct radeon_transform_context* t,
-	struct prog_instruction* inst,
+GLboolean radeonTransformDeriv(struct radeon_compiler* c,
+	struct rc_instruction* inst,
 	void* unused)
 {
-	if (inst->Opcode != OPCODE_DDX && inst->Opcode != OPCODE_DDY)
+	if (inst->I.Opcode != OPCODE_DDX && inst->I.Opcode != OPCODE_DDY)
 		return GL_FALSE;
 
-	struct prog_src_register B = inst->SrcReg[1];
-
-	B.Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE,
-						SWIZZLE_ONE, SWIZZLE_ONE);
-	B.Negate = NEGATE_XYZW;
-
-	emit2(t->Program, inst->Opcode, inst->SaturateMode, inst->DstReg,
-		inst->SrcReg[0], B);
+	inst->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE);
+	inst->I.SrcReg[1].Negate = NEGATE_XYZW;
 
 	return GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h
index 7d94a089eb..147efec6fc 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h
@@ -31,28 +31,28 @@
 #include "radeon_program.h"
 
 GLboolean radeonTransformALU(
-	struct radeon_transform_context *t,
-	struct prog_instruction*,
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
 	void*);
 
 GLboolean r300_transform_vertex_alu(
-	struct radeon_transform_context *t,
-	struct prog_instruction*,
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
 	void*);
 
 GLboolean radeonTransformTrigSimple(
-	struct radeon_transform_context *t,
-	struct prog_instruction*,
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
 	void*);
 
 GLboolean radeonTransformTrigScale(
-	struct radeon_transform_context *t,
-	struct prog_instruction*,
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
 	void*);
 
 GLboolean radeonTransformDeriv(
-	struct radeon_transform_context *t,
-	struct prog_instruction*,
+	struct radeon_compiler * c,
+	struct rc_instruction * inst,
 	void*);
 
 #endif /* __RADEON_PROGRAM_ALU_H_ */
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index b8fad4a6e7..c79601bcb1 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -1063,24 +1063,6 @@ r300FetchStateParameter(GLcontext * ctx,
 				break;
 			}
 
-		case STATE_R300_TEXRECT_FACTOR:{
-				struct gl_texture_object *t =
-				    ctx->Texture.Unit[state[2]].CurrentTex[TEXTURE_RECT_INDEX];
-
-				if (t && t->Image[0][t->BaseLevel]) {
-					struct gl_texture_image *image =
-					    t->Image[0][t->BaseLevel];
-					value[0] = 1.0 / image->Width2;
-					value[1] = 1.0 / image->Height2;
-				} else {
-					value[0] = 1.0;
-					value[1] = 1.0;
-				}
-				value[2] = 1.0;
-				value[3] = 1.0;
-				break;
-			}
-
 		default:
 			break;
 		}
@@ -2029,7 +2011,7 @@ void r300UpdateShaders(r300ContextPtr rmesa)
 	rmesa->radeon.NewGLState = 0;
 }
 
-static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, GLuint index)
+static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, GLuint index, GLfloat * buffer)
 {
 	static const GLfloat dummy[4] = { 0, 0, 0, 0 };
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
@@ -2041,6 +2023,47 @@ static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx, GLuint index)
 		return fp->Base->Parameters->ParameterValues[rcc->u.External];
 	case RC_CONSTANT_IMMEDIATE:
 		return rcc->u.Immediate;
+	case RC_CONSTANT_STATE:
+		switch(rcc->u.State[0]) {
+		case RC_STATE_SHADOW_AMBIENT: {
+			const int unit = (int) rcc->u.State[1];
+			const struct gl_texture_object *texObj = ctx->Texture.Unit[unit]._Current;
+			if (texObj) {
+				buffer[0] =
+				buffer[1] =
+				buffer[2] =
+				buffer[3] = texObj->CompareFailValue;
+			}
+			return buffer;
+		}
+
+		case RC_STATE_R300_WINDOW_DIMENSION: {
+			__DRIdrawablePrivate * drawable = radeon_get_drawable(&rmesa->radeon);
+			buffer[0] = drawable->w * 0.5f;	/* width*0.5 */
+			buffer[1] = drawable->h * 0.5f;	/* height*0.5 */
+			buffer[2] = 0.5F;	/* for moving range [-1 1] -> [0 1] */
+			buffer[3] = 1.0F;	/* not used */
+			return buffer;
+		}
+
+		case RC_STATE_R300_TEXRECT_FACTOR: {
+			struct gl_texture_object *t =
+				ctx->Texture.Unit[rcc->u.State[1]].CurrentTex[TEXTURE_RECT_INDEX];
+
+			if (t && t->Image[0][t->BaseLevel]) {
+				struct gl_texture_image *image =
+					t->Image[0][t->BaseLevel];
+				buffer[0] = 1.0 / image->Width2;
+				buffer[1] = 1.0 / image->Height2;
+			} else {
+				buffer[0] = 1.0;
+				buffer[1] = 1.0;
+			}
+			buffer[2] = 1.0;
+			buffer[3] = 1.0;
+			return buffer;
+		}
+		}
 	}
 
 	return dummy;
@@ -2096,7 +2119,8 @@ static void r300SetupPixelShader(GLcontext *ctx)
 	R300_STATECHANGE(rmesa, fpp);
 	rmesa->hw.fpp.cmd[R300_FPP_CMD_0] = cmdpacket0(rmesa->radeon.radeonScreen, R300_PFS_PARAM_0_X, fp->code.constants.Count * 4);
 	for (i = 0; i < fp->code.constants.Count; i++) {
-		const GLfloat *constant = get_fragmentprogram_constant(ctx, i);
+		GLfloat buffer[4];
+		const GLfloat *constant = get_fragmentprogram_constant(ctx, i, buffer);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat24(constant[0]);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat24(constant[1]);
 		rmesa->hw.fpp.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat24(constant[2]);
@@ -2157,7 +2181,8 @@ static void r500SetupPixelShader(GLcontext *ctx)
 
 	R300_STATECHANGE(rmesa, r500fp_const);
 	for (i = 0; i < fp->code.constants.Count; i++) {
-		const GLfloat *constant = get_fragmentprogram_constant(ctx, i);
+		GLfloat buffer[4];
+		const GLfloat *constant = get_fragmentprogram_constant(ctx, i, buffer);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 0] = r300PackFloat32(constant[0]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 1] = r300PackFloat32(constant[1]);
 		rmesa->hw.r500fp_const.cmd[R300_FPP_PARAM_0 + 4 * i + 2] = r300PackFloat32(constant[2]);