Merge branch 'shaders_cleanup'

17 files changed, 883 insertions, 570 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index b7d75426c57..6f3aab986d2 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -418,8 +418,6 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 
 	_mesa_set_mvp_with_dp4( ctx, GL_TRUE );
 
-	_mesa_set_mvp_with_dp4( ctx, GL_TRUE );
-
 	/* Initialize the software rasterizer and helper modules.
 	 */
 	_swrast_CreateContext(ctx);
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index 026c33c67cb..f7af7d4e574 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -405,12 +405,13 @@ struct r300_hw_state {
 #undef TAG
 
 struct r300_vertex_program {
+	struct gl_vertex_program *Base;
 	struct r300_vertex_program *next;
 
 	struct r300_vertex_program_key {
-		GLuint InputsRead;
-		GLuint OutputsWritten;
-		GLuint OutputsAdded;
+		GLuint FpReads;
+		GLuint FogAttr;
+		GLuint WPosAttr;
 	} key;
 	
 	struct r300_vertex_shader_hw_code {
@@ -426,13 +427,17 @@ struct r300_vertex_program {
 
 	int pos_end;
 	int num_temporaries;	/* Number of temp vars used by program */
-	int wpos_idx;
 	int inputs[VERT_ATTRIB_MAX];
 	int outputs[VERT_RESULT_MAX];
 };
 
 struct r300_vertex_program_cont {
-	struct gl_vertex_program mesa_program;	/* Must be first */
+	/* This is the unmodified vertex program mesa provided us with.
+	 * We need to keep it unchanged because we may need to create another
+	 * hw specific vertex program based on this.
+	 */
+	struct gl_vertex_program mesa_program;
+	/* This is the list of hw specific vertex programs derived from mesa_program */
 	struct r300_vertex_program *progs;
 };
 
@@ -546,7 +551,7 @@ struct r500_fragment_program_code {
 * to render with that program.
 */
 struct r300_fragment_program {
-	struct gl_fragment_program Base;
+	struct gl_program *Base;
 
 	GLboolean translated;
 	GLboolean error;
@@ -559,6 +564,23 @@ struct r300_fragment_program {
 
 	GLboolean writes_depth;
 	GLuint optimization;
+
+	struct r300_fragment_program *next;
+
+	/* attribute that we are sending the WPOS in */
+	gl_frag_attrib wpos_attr;
+	/* attribute that we are sending the fog coordinate in */
+	gl_frag_attrib fog_attr;
+};
+
+struct r300_fragment_program_cont {
+	/* This is the unmodified fragment program mesa provided us with.
+	 * We need to keep it unchanged because we may need to create another
+	 * hw specific fragment program based on this.
+	 */
+	struct gl_fragment_program Base;
+	/* This is the list of hw specific fragment programs derived from Base */
+	struct r300_fragment_program *progs;
 };
 
 struct r300_fragment_program_compiler {
@@ -633,6 +655,7 @@ struct r300_context {
 	struct r300_hw_state hw;
 
 	struct r300_vertex_program *selected_vp;
+	struct r300_fragment_program *selected_fp;
 
 	/* Vertex buffers
 	 */
@@ -664,11 +687,7 @@ extern GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 				   __DRIcontextPrivate * driContextPriv,
 				   void *sharedContextPrivate);
 
-extern void r300SelectVertexShader(r300ContextPtr r300);
 extern void r300InitShaderFuncs(struct dd_function_table *functions);
-extern int r300VertexProgUpdateParams(GLcontext * ctx,
-				      struct r300_vertex_program_cont *vp,
-				      float *dst);
 
 extern void r300InitShaderFunctions(r300ContextPtr r300);
 
diff --git a/src/mesa/drivers/dri/r300/r300_draw.c b/src/mesa/drivers/dri/r300/r300_draw.c
index 4e8b62f1865..9769ff53994 100644
--- a/src/mesa/drivers/dri/r300/r300_draw.c
+++ b/src/mesa/drivers/dri/r300/r300_draw.c
@@ -195,6 +195,11 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st
 		}
 
 		GLfloat *dst_ptr, *tmp;
+
+		/* Convert value for first element only */
+		if (input->StrideB == 0)
+			count = 1;
+
 		tmp = dst_ptr = _mesa_malloc(sizeof(GLfloat) * input->Size * count);
 
 		switch (input->Type) {
@@ -228,7 +233,11 @@ static void r300TranslateAttrib(GLcontext *ctx, GLuint attr, int count, const st
 		type = GL_FLOAT;
 		r300_attr.free_needed = GL_TRUE;
 		r300_attr.data = tmp;
-		r300_attr.stride = sizeof(GLfloat) * input->Size;
+		if (input->StrideB == 0) {
+			r300_attr.stride = 0;
+		} else {
+			r300_attr.stride = sizeof(GLfloat) * input->Size;
+		}
 		r300_attr.dwords = input->Size;
 	} else {
 		type = input->Type;
@@ -332,7 +341,7 @@ static void r300SetVertexFormat(GLcontext *ctx, const struct gl_client_array *ar
 	{
 		int i, tmp;
 
-		tmp = r300->selected_vp->key.InputsRead;
+		tmp = r300->selected_vp->Base->Base.InputsRead;
 		i = 0;
 		vbuf->num_attribs = 0;
 		while (tmp) {
@@ -428,7 +437,7 @@ static GLboolean r300TryDrawPrims(GLcontext *ctx,
 	if (r300->fallback)
 		return GL_FALSE;
 
-	r300SetupVAP(ctx, r300->selected_vp->key.InputsRead, r300->selected_vp->key.OutputsWritten);
+	r300SetupVAP(ctx, r300->selected_vp->Base->Base.InputsRead, r300->selected_vp->Base->Base.OutputsWritten);
 
 	r300UpdateShaderStates(r300);
 
diff --git a/src/mesa/drivers/dri/r300/r300_emit.c b/src/mesa/drivers/dri/r300/r300_emit.c
index c3817721dc4..feb3370f372 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_emit.c
@@ -81,17 +81,17 @@ GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead)
 	return vic_1;
 }
 
-GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
+GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes)
 {
 	GLuint ret = 0;
 
 	if (vp_writes & (1 << VERT_RESULT_HPOS))
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
 
-	if (vp_writes & (1 << VERT_RESULT_COL0) && fp_reads & FRAG_BIT_COL0)
+	if (vp_writes & (1 << VERT_RESULT_COL0))
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT;
 
-	if (vp_writes & (1 << VERT_RESULT_COL1) && fp_reads & FRAG_BIT_COL1)
+	if (vp_writes & (1 << VERT_RESULT_COL1))
 		ret |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_1_PRESENT;
 
 	/* Two sided lighting works only if all 4 colors are written */
@@ -105,26 +105,17 @@ GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
 	return ret;
 }
 
-GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads)
+GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes)
 {
 	GLuint i, ret = 0, first_free_texcoord = 0;
 
 	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-		if (vp_writes & (1 << (VERT_RESULT_TEX0 + i)) && fp_reads & FRAG_BIT_TEX(i)) {
+		if (vp_writes & (1 << (VERT_RESULT_TEX0 + i))) {
 			ret |= (4 << (3 * first_free_texcoord));
 			++first_free_texcoord;
 		}
 	}
 
-	if (fp_reads & FRAG_BIT_WPOS) {
-		ret |= (4 << (3 * first_free_texcoord));
-		++first_free_texcoord;
-	}
-
-	if (vp_writes & (1 << VERT_RESULT_FOGC) && fp_reads & FRAG_BIT_FOGC) {
-		ret |= 4 << (3 * first_free_texcoord);
-	}
-
 	if (first_free_texcoord > 8) {
 		fprintf(stderr, "\tout of free texcoords\n");
 		_mesa_exit(-1);
diff --git a/src/mesa/drivers/dri/r300/r300_emit.h b/src/mesa/drivers/dri/r300/r300_emit.h
index 2fb8b82d3a1..3f8c60ffae1 100644
--- a/src/mesa/drivers/dri/r300/r300_emit.h
+++ b/src/mesa/drivers/dri/r300/r300_emit.h
@@ -225,7 +225,7 @@ extern void r300EmitCacheFlush(r300ContextPtr rmesa);
 
 extern GLuint r300VAPInputCntl0(GLcontext * ctx, GLuint InputsRead);
 extern GLuint r300VAPInputCntl1(GLcontext * ctx, GLuint InputsRead);
-extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
-extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes, GLuint fp_reads);
+extern GLuint r300VAPOutputCntl0(GLcontext * ctx, GLuint vp_writes);
+extern GLuint r300VAPOutputCntl1(GLcontext * ctx, GLuint vp_writes);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
index abc8757ba1f..f5c4c0f4a0e 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_common.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c
@@ -50,13 +50,6 @@
 #include "radeon_program.h"
 #include "radeon_program_alu.h"
 
-static void update_params(GLcontext *ctx, struct gl_fragment_program *fp)
-{
-	/* Ask Mesa nicely to fill in ParameterValues for us */
-	if (fp->Base.Parameters)
-		_mesa_load_state_parameters(ctx, fp->Base.Parameters);
-}
-
 static void nqssadce_init(struct nqssadce_state* s)
 {
 	s->Outputs[FRAG_RESULT_COLOR].Sourced = WRITEMASK_XYZW;
@@ -74,10 +67,12 @@ static void nqssadce_init(struct nqssadce_state* s)
  */
 static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 {
-	GLuint InputsRead = compiler->fp->Base.Base.InputsRead;
+	GLuint InputsRead = compiler->fp->Base->InputsRead;
 
-	if (!(InputsRead & FRAG_BIT_WPOS))
+	if (!(InputsRead & FRAG_BIT_WPOS)) {
+		compiler->fp->wpos_attr = FRAG_ATTRIB_MAX;
 		return;
+	}
 
 	static gl_state_index tokens[STATE_LENGTH] = {
 		STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
@@ -85,10 +80,23 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 	struct prog_instruction *fpi;
 	GLuint window_index;
 	int i = 0;
+
+	for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i)
+	{
+		if (!(InputsRead & (1 << i))) {
+			InputsRead &= ~(1 << FRAG_ATTRIB_WPOS);
+			InputsRead |= 1 << i;
+			compiler->fp->Base->InputsRead = InputsRead;
+			compiler->fp->wpos_attr = i;
+			break;
+		}
+	}
+
 	GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
 
 	_mesa_insert_instructions(compiler->program, 0, 3);
 	fpi = compiler->program->Instructions;
+	i = 0;
 
 	/* perspective divide */
 	fpi[i].Opcode = OPCODE_RCP;
@@ -99,7 +107,7 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 	fpi[i].DstReg.CondMask = COND_TR;
 
 	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+	fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr;
 	fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
 	i++;
 
@@ -111,7 +119,7 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 	fpi[i].DstReg.CondMask = COND_TR;
 
 	fpi[i].SrcReg[0].File = PROGRAM_INPUT;
-	fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
+	fpi[i].SrcReg[0].Index = compiler->fp->wpos_attr;
 	fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
 
 	fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
@@ -154,6 +162,57 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 	}
 }
 
+
+/**
+ * Rewrite fragment.fogcoord to use a texture coordinate slot.
+ * Note that fogcoord is forced into an X001 pattern, and this enforcement
+ * is done here.
+ *
+ * See also the counterpart rewriting for vertex programs.
+ */
+static void rewriteFog(struct r300_fragment_program_compiler *compiler)
+{
+	struct r300_fragment_program *fp = compiler->fp;
+	GLuint InputsRead;
+	int i;
+
+	InputsRead = fp->Base->InputsRead;
+
+	if (!(InputsRead & FRAG_BIT_FOGC)) {
+		fp->fog_attr = FRAG_ATTRIB_MAX;
+		return;
+	}
+
+	for (i = FRAG_ATTRIB_TEX0; i <= FRAG_ATTRIB_TEX7; ++i)
+	{
+		if (!(InputsRead & (1 << i))) {
+			InputsRead &= ~(1 << FRAG_ATTRIB_FOGC);
+			InputsRead |= 1 << i;
+			fp->Base->InputsRead = InputsRead;
+			fp->fog_attr = i;
+			break;
+		}
+	}
+
+	{
+		struct prog_instruction *inst;
+
+		inst = compiler->program->Instructions;
+		while (inst->Opcode != OPCODE_END) {
+			const int src_regs = _mesa_num_inst_src_regs(inst->Opcode);
+			for (i = 0; i < src_regs; ++i) {
+				if (inst->SrcReg[i].File == PROGRAM_INPUT && inst->SrcReg[i].Index == FRAG_ATTRIB_FOGC) {
+					inst->SrcReg[i].Index = fp->fog_attr;
+					inst->SrcReg[i].Swizzle = combine_swizzles(
+						MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE),
+						inst->SrcReg[i].Swizzle);
+				}
+			}
+			++inst;
+		}
+	}
+}
+
 static GLuint build_dtm(GLuint depthmode)
 {
 	switch(depthmode) {
@@ -175,7 +234,7 @@ static GLuint build_func(GLuint comparefunc)
  */
 static void build_state(
 	r300ContextPtr r300,
-	struct r300_fragment_program *fp,
+	struct gl_fragment_program *fp,
 	struct r300_fragment_program_external_state *state)
 {
 	int unit;
@@ -183,7 +242,7 @@ static void build_state(
 	_mesa_bzero(state, sizeof(*state));
 
 	for(unit = 0; unit < 16; ++unit) {
-		if (fp->Base.Base.ShadowSamplers & (1 << unit)) {
+		if (fp->Base.ShadowSamplers & (1 << unit)) {
 			struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
 
 			state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
@@ -192,100 +251,149 @@ static void build_state(
 	}
 }
 
-void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp)
+static void rewrite_depth_out(struct gl_program *prog)
 {
-	r300ContextPtr r300 = R300_CONTEXT(ctx);
-	struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)fp;
-	struct r300_fragment_program_external_state state;
+	struct prog_instruction *inst;
 
-	build_state(r300, r300_fp, &state);
-	if (_mesa_memcmp(&r300_fp->state, &state, sizeof(state))) {
-		/* TODO: cache compiled programs */
-		r300_fp->translated = GL_FALSE;
-		_mesa_memcpy(&r300_fp->state, &state, sizeof(state));
-	}
-
-	if (!r300_fp->translated) {
-		struct r300_fragment_program_compiler compiler;
+	for (inst = prog->Instructions; inst->Opcode != OPCODE_END; ++inst) {
+		if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != FRAG_RESULT_DEPTH)
+			continue;
 
-		compiler.r300 = r300;
-		compiler.fp = r300_fp;
-		compiler.code = &r300_fp->code;
-		compiler.program = _mesa_clone_program(ctx, &fp->Base);
+		if (inst->DstReg.WriteMask & WRITEMASK_Z) {
+			inst->DstReg.WriteMask = WRITEMASK_W;
+		} else {
+			inst->DstReg.WriteMask = 0;
+			continue;
+		}
 
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			fflush(stdout);
-			_mesa_printf("Fragment Program: Initial program:\n");
-			_mesa_print_program(compiler.program);
-			fflush(stdout);
+		switch (inst->Opcode) {
+			case OPCODE_FRC:
+			case OPCODE_MOV:
+				inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
+				break;
+			case OPCODE_ADD:
+			case OPCODE_MAX:
+			case OPCODE_MIN:
+			case OPCODE_MUL:
+				inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
+				inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]);
+				break;
+			case OPCODE_CMP:
+			case OPCODE_MAD:
+				inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
+				inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]);
+				inst->SrcReg[2] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[2]);
+				break;
+			default:
+				// Scalar instructions needn't be reswizzled
+				break;
 		}
+	}
+}
 
-		insert_WPOS_trailer(&compiler);
+void r300TranslateFragmentShader(GLcontext *ctx, struct r300_fragment_program *fp)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_fragment_program_compiler compiler;
+
+	compiler.r300 = r300;
+	compiler.fp = fp;
+	compiler.code = &fp->code;
+	compiler.program = fp->Base;
+
+	if (RADEON_DEBUG & DEBUG_PIXEL) {
+		fflush(stdout);
+		_mesa_printf("Fragment Program: Initial program:\n");
+		_mesa_print_program(compiler.program);
+		fflush(stdout);
+	}
 
-		if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-			struct radeon_program_transformation transformations[] = {
-				{ &r500_transform_TEX, &compiler },
-				{ &radeonTransformALU, 0 },
-				{ &radeonTransformDeriv, 0 },
-				{ &radeonTransformTrigScale, 0 }
-			};
-			radeonLocalTransform(ctx, compiler.program, 4, transformations);
-		} else {
-			struct radeon_program_transformation transformations[] = {
-				{ &r300_transform_TEX, &compiler },
-				{ &radeonTransformALU, 0 },
-				{ &radeonTransformTrigSimple, 0 }
-			};
-			radeonLocalTransform(ctx, compiler.program, 3, transformations);
-		}
+	insert_WPOS_trailer(&compiler);
+
+	rewriteFog(&compiler);
+
+	rewrite_depth_out(compiler.program);
+
+	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+		struct radeon_program_transformation transformations[] = {
+			{ &r500_transform_TEX, &compiler },
+			{ &radeonTransformALU, 0 },
+			{ &radeonTransformDeriv, 0 },
+			{ &radeonTransformTrigScale, 0 }
+		};
+		radeonLocalTransform(ctx, compiler.program, 4, transformations);
+	} else {
+		struct radeon_program_transformation transformations[] = {
+			{ &r300_transform_TEX, &compiler },
+			{ &radeonTransformALU, 0 },
+			{ &radeonTransformTrigSimple, 0 }
+		};
+		radeonLocalTransform(ctx, compiler.program, 3, transformations);
+	}
 
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Fragment Program: After native rewrite:\n");
-			_mesa_print_program(compiler.program);
-			fflush(stdout);
-		}
+	if (RADEON_DEBUG & DEBUG_PIXEL) {
+		_mesa_printf("Fragment Program: After native rewrite:\n");
+		_mesa_print_program(compiler.program);
+		fflush(stdout);
+	}
 
-		if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
-			struct radeon_nqssadce_descr nqssadce = {
-				.Init = &nqssadce_init,
-				.IsNativeSwizzle = &r500FPIsNativeSwizzle,
-				.BuildSwizzle = &r500FPBuildSwizzle,
-				.RewriteDepthOut = GL_TRUE
-			};
-			radeonNqssaDce(ctx, compiler.program, &nqssadce);
-		} else {
-			struct radeon_nqssadce_descr nqssadce = {
-				.Init = &nqssadce_init,
-				.IsNativeSwizzle = &r300FPIsNativeSwizzle,
-				.BuildSwizzle = &r300FPBuildSwizzle,
-				.RewriteDepthOut = GL_TRUE
-			};
-			radeonNqssaDce(ctx, compiler.program, &nqssadce);
-		}
+	if (r300->radeon.radeonScreen->chip_family >= CHIP_FAMILY_RV515) {
+		struct radeon_nqssadce_descr nqssadce = {
+			.Init = &nqssadce_init,
+			.IsNativeSwizzle = &r500FPIsNativeSwizzle,
+			.BuildSwizzle = &r500FPBuildSwizzle
+		};
+		radeonNqssaDce(ctx, compiler.program, &nqssadce);
+	} else {
+		struct radeon_nqssadce_descr nqssadce = {
+			.Init = &nqssadce_init,
+			.IsNativeSwizzle = &r300FPIsNativeSwizzle,
+			.BuildSwizzle = &r300FPBuildSwizzle
+		};
+		radeonNqssaDce(ctx, compiler.program, &nqssadce);
+	}
 
-		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Compiler: after NqSSA-DCE:\n");
-			_mesa_print_program(compiler.program);
-			fflush(stdout);
-		}
+	if (RADEON_DEBUG & DEBUG_PIXEL) {
+		_mesa_printf("Compiler: after NqSSA-DCE:\n");
+		_mesa_print_program(compiler.program);
+		fflush(stdout);
+	}
 
-		if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler))
-			r300_fp->error = GL_TRUE;
+	if (!r300->vtbl.BuildFragmentProgramHwCode(&compiler))
+		fp->error = GL_TRUE;
 
-		/* Subtle: Rescue any parameters that have been added during transformations */
-		_mesa_free_parameter_list(fp->Base.Parameters);
-		fp->Base.Parameters = compiler.program->Parameters;
-		compiler.program->Parameters = 0;
+	fp->translated = GL_TRUE;
 
-		_mesa_reference_program(ctx, &compiler.program, NULL);
+	if (fp->error || (RADEON_DEBUG & DEBUG_PIXEL))
+		r300->vtbl.FragmentProgramDump(&fp->code);
+}
 
-		r300_fp->translated = GL_TRUE;
+struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_fragment_program_cont *fp_list;
+	struct r300_fragment_program *fp;
+	struct r300_fragment_program_external_state state;
 
-		r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
+	fp_list = (struct r300_fragment_program_cont *)ctx->FragmentProgram._Current;
+	build_state(r300, ctx->FragmentProgram._Current, &state);
 
-		if (r300_fp->error || (RADEON_DEBUG & DEBUG_PIXEL))
-			r300->vtbl.FragmentProgramDump(&r300_fp->code);
+	fp = fp_list->progs;
+	while (fp) {
+		if (_mesa_memcmp(&fp->state, &state, sizeof(state)) == 0) {
+			return r300->selected_fp = fp;
+		}
+		fp = fp->next;
 	}
 
-	update_params(ctx, fp);
+	fp = _mesa_calloc(sizeof(struct r300_fragment_program));
+
+	fp->state = state;
+	fp->translated = GL_FALSE;
+	fp->Base = _mesa_clone_program(ctx, &ctx->FragmentProgram._Current->Base);
+
+	fp->next = fp_list->progs;
+	fp_list->progs = fp;
+
+	return r300->selected_fp = fp;
 }
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.h b/src/mesa/drivers/dri/r300/r300_fragprog_common.h
index 85ea86fecb1..5e103ee4086 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_common.h
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.h
@@ -30,6 +30,10 @@
 
 #include "main/mtypes.h"
 
-extern void r300TranslateFragmentShader(GLcontext *ctx, struct gl_fragment_program *fp);
+#include "r300_context.h"
+
+extern void r300TranslateFragmentShader(GLcontext *ctx, struct r300_fragment_program *fp);
+
+struct r300_fragment_program *r300SelectFragmentShader(GLcontext *ctx);
 
 #endif
diff --git a/src/mesa/drivers/dri/r300/r300_shader.c b/src/mesa/drivers/dri/r300/r300_shader.c
index 0133b837966..62228a3786e 100644
--- a/src/mesa/drivers/dri/r300/r300_shader.c
+++ b/src/mesa/drivers/dri/r300/r300_shader.c
@@ -32,22 +32,45 @@
 #include "r300_context.h"
 #include "r300_fragprog_common.h"
 
+static void freeFragProgCache(GLcontext *ctx, struct r300_fragment_program_cont *cache)
+{
+	struct r300_fragment_program *tmp, *fp = cache->progs;
+
+	while (fp) {
+		tmp = fp->next;
+		_mesa_reference_program(ctx, &fp->Base, NULL);
+		_mesa_free(fp);
+		fp = tmp;
+	}
+}
+
+static void freeVertProgCache(GLcontext *ctx, struct r300_vertex_program_cont *cache)
+{
+	struct r300_vertex_program *tmp, *vp = cache->progs;
+
+	while (vp) {
+		tmp = vp->next;
+		_mesa_reference_vertprog(ctx, &vp->Base, NULL);
+		_mesa_free(vp);
+		vp = tmp;
+	}
+}
+
 static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target,
 					 GLuint id)
 {
 	struct r300_vertex_program_cont *vp;
-	struct r300_fragment_program *fp;
+	struct r300_fragment_program_cont *fp;
 
 	switch (target) {
 	case GL_VERTEX_STATE_PROGRAM_NV:
 	case GL_VERTEX_PROGRAM_ARB:
 		vp = CALLOC_STRUCT(r300_vertex_program_cont);
-		return _mesa_init_vertex_program(ctx, &vp->mesa_program,
-						 target, id);
+		return _mesa_init_vertex_program(ctx, &vp->mesa_program, target, id);
 
 	case GL_FRAGMENT_PROGRAM_NV:
 	case GL_FRAGMENT_PROGRAM_ARB:
-		fp = CALLOC_STRUCT(r300_fragment_program);
+		fp = CALLOC_STRUCT(r300_fragment_program_cont);
 		return _mesa_init_fragment_program(ctx, &fp->Base, target, id);
 
 	default:
@@ -59,21 +82,35 @@ static struct gl_program *r300NewProgram(GLcontext * ctx, GLenum target,
 
 static void r300DeleteProgram(GLcontext * ctx, struct gl_program *prog)
 {
+	struct r300_vertex_program_cont *vp = (struct r300_vertex_program_cont *)prog;
+	struct r300_fragment_program_cont *fp = (struct r300_fragment_program_cont *)prog;
+
+	switch (prog->Target) {
+		case GL_VERTEX_PROGRAM_ARB:
+			freeVertProgCache(ctx, vp);
+			break;
+		case GL_FRAGMENT_PROGRAM_ARB:
+			freeFragProgCache(ctx, fp);
+			break;
+	}
+
 	_mesa_delete_program(ctx, prog);
 }
 
 static void
 r300ProgramStringNotify(GLcontext * ctx, GLenum target, struct gl_program *prog)
 {
-	struct r300_vertex_program_cont *vp = (void *)prog;
-	struct r300_fragment_program *r300_fp = (struct r300_fragment_program *)prog;
+	struct r300_vertex_program_cont *vp = (struct r300_vertex_program_cont *)prog;
+	struct r300_fragment_program_cont *fp = (struct r300_fragment_program_cont *)prog;
 
 	switch (target) {
 	case GL_VERTEX_PROGRAM_ARB:
+		freeVertProgCache(ctx, vp);
 		vp->progs = NULL;
 		break;
 	case GL_FRAGMENT_PROGRAM_ARB:
-		r300_fp->translated = GL_FALSE;
+		freeFragProgCache(ctx, fp);
+		fp->progs = NULL;
 		break;
 	}
 
@@ -85,13 +122,18 @@ static GLboolean
 r300IsProgramNative(GLcontext * ctx, GLenum target, struct gl_program *prog)
 {
 	if (target == GL_FRAGMENT_PROGRAM_ARB) {
-		struct r300_fragment_program *fp = (struct r300_fragment_program *)prog;
+		struct r300_fragment_program *fp = r300SelectFragmentShader(ctx);
 		if (!fp->translated)
-			r300TranslateFragmentShader(ctx, &fp->Base);
+			r300TranslateFragmentShader(ctx, fp);
 
 		return !fp->error;
-	} else
-		return GL_TRUE;
+	} else {
+		struct r300_vertex_program *vp = r300SelectVertexShader(ctx);
+		if (!vp->translated)
+			r300TranslateVertexShader(vp);
+
+		return !vp->error;
+	}
 }
 
 void r300InitShaderFuncs(struct dd_function_table *functions)
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index c0eda977db8..4d504d14e36 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -452,9 +452,9 @@ static void r300SetPolygonOffsetState(GLcontext * ctx, GLboolean state)
 
 static GLboolean current_fragment_program_writes_depth(GLcontext* ctx)
 {
-	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
 
-	return (fp && fp->writes_depth);
+	return ctx->FragmentProgram._Current && r300->selected_fp->writes_depth;
 }
 
 static void r300SetEarlyZState(GLcontext * ctx)
@@ -1093,24 +1093,25 @@ r300FetchStateParameter(GLcontext * ctx,
  * Update R300's own internal state parameters.
  * For now just STATE_R300_WINDOW_DIMENSION
  */
-void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state)
+static void r300UpdateStateParameters(GLcontext * ctx, GLuint new_state)
 {
-	struct r300_fragment_program *fp;
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
 	struct gl_program_parameter_list *paramList;
 	GLuint i;
 
 	if (!(new_state & (_NEW_BUFFERS | _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS)))
 		return;
 
-	fp = (struct r300_fragment_program *)ctx->FragmentProgram._Current;
-	if (!fp)
+	if (!ctx->FragmentProgram._Current || !rmesa->selected_fp)
 		return;
 
-	paramList = fp->Base.Base.Parameters;
+	paramList = rmesa->selected_fp->Base->Parameters;
 
 	if (!paramList)
 		return;
 
+	_mesa_load_state_parameters(ctx, paramList);
+
 	for (i = 0; i < paramList->NumParameters; i++) {
 		if (paramList->Parameters[i].Type == PROGRAM_STATE_VAR) {
 			r300FetchStateParameter(ctx,
@@ -1225,8 +1226,7 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 {
 	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	int i;
-	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
-	struct r300_fragment_program_code *code = &fp->code.r300;
+	struct r300_fragment_program_code *code = &r300->selected_fp->code.r300;
 
 	R300_STATECHANGE(r300, fpt);
 
@@ -1266,9 +1266,9 @@ static void r300SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 
 static void r500SetupFragmentShaderTextures(GLcontext *ctx, int *tmu_mappings)
 {
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	int i;
-	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
-	struct r500_fragment_program_code *code = &fp->code.r500;
+	struct r500_fragment_program_code *code = &r300->selected_fp->code.r500;
 
 	/* find all the texture instructions and relocate the texture units */
 	for (i = 0; i < code->inst_end + 1; i++) {
@@ -1316,8 +1316,6 @@ static void r300SetupTextures(GLcontext * ctx)
 	int hw_tmu = 0;
 	int last_hw_tmu = -1;	/* -1 translates into no setup costs for fields */
 	int tmu_mappings[R300_MAX_TEXTURE_UNITS] = { -1, };
-	struct r300_fragment_program *fp = (struct r300_fragment_program *)
-	    (char *)ctx->FragmentProgram._Current;
 
 	R300_STATECHANGE(r300, txe);
 	R300_STATECHANGE(r300, tex.filter);
@@ -1420,7 +1418,7 @@ static void r300SetupTextures(GLcontext * ctx)
 	    cmdpacket0(r300->radeon.radeonScreen, R300_TX_BORDER_COLOR_0, last_hw_tmu + 1);
 
 	if (r300->radeon.radeonScreen->chip_family < CHIP_FAMILY_RV515) {
-		if (fp->Base.UsesKill && last_hw_tmu < 0) {
+		if (ctx->FragmentProgram._Current->UsesKill && last_hw_tmu < 0) {
 			// The KILL operation requires the first texture unit
 			// to be enabled.
 			r300->hw.txe.cmd[R300_TXE_ENABLE] |= 1;
@@ -1458,11 +1456,11 @@ static void r300SetupRSUnit(GLcontext * ctx)
 	hw_tcl_on = r300->options.hw_tcl_enabled;
 
 	if (hw_tcl_on)
-		OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten;
+		OutputsWritten.vp_outputs = r300->selected_vp->Base->Base.OutputsWritten;
 	else
 		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset);
 
-	InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
+	InputsRead = r300->selected_fp->Base->InputsRead;
 
 	R300_STATECHANGE(r300, ri);
 	R300_STATECHANGE(r300, rc);
@@ -1521,29 +1519,6 @@ static void r300SetupRSUnit(GLcontext * ctx)
 		++fp_reg;
 	}
 
-	if (InputsRead & FRAG_BIT_WPOS) {
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(1) | R300_RS_SEL_R(2) | R300_RS_SEL_Q(3) | R300_RS_TEX_PTR(rs_tex_count);
-		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
-		InputsRead &= ~FRAG_BIT_WPOS;
-		rs_tex_count += 4;
-		++tex_ip;
-		++fp_reg;
-	}
-
-	if (InputsRead & FRAG_BIT_FOGC) {
-		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) {
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_S(0) | R300_RS_SEL_T(R300_RS_SEL_K0) | R300_RS_SEL_R(R300_RS_SEL_K0);
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= R300_RS_SEL_Q(R300_RS_SEL_K1) | R300_RS_TEX_PTR(rs_tex_count);
-			r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R300_RS_INST_TEX_ID(tex_ip) | R300_RS_INST_TEX_CN_WRITE | R300_RS_INST_TEX_ADDR(fp_reg);
-			InputsRead &= ~FRAG_BIT_FOGC;
-			rs_tex_count += 4;
-			++tex_ip;
-			++fp_reg;
-		} else {
-			WARN_ONCE("fragprog wants fogc, vp doesn't provide it\n");
-		}
-	}
-
 	/* Setup default color if no color or tex was set */
 	if (rs_tex_count == 0 && col_ip == 0) {
 		r300->hw.rr.cmd[R300_RR_INST_0] = R300_RS_INST_COL_ID(0) | R300_RS_INST_COL_ADDR(0);
@@ -1575,11 +1550,11 @@ static void r500SetupRSUnit(GLcontext * ctx)
 	hw_tcl_on = r300->options.hw_tcl_enabled;
 
 	if (hw_tcl_on)
-		OutputsWritten.vp_outputs = r300->selected_vp->key.OutputsWritten;
+		OutputsWritten.vp_outputs = r300->selected_vp->Base->Base.OutputsWritten;
 	else
 		RENDERINPUTS_COPY(OutputsWritten.index_bitset, r300->render_inputs_bitset);
 
-	InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
+	InputsRead = r300->selected_fp->Base->InputsRead;
 
 	R300_STATECHANGE(r300, ri);
 	R300_STATECHANGE(r300, rc);
@@ -1642,36 +1617,6 @@ static void r500SetupRSUnit(GLcontext * ctx)
 		++fp_reg;
 	}
 
-	if (InputsRead & FRAG_BIT_WPOS) {
-		r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= ((rs_tex_count + 0) << R500_RS_IP_TEX_PTR_S_SHIFT) |
-			((rs_tex_count + 1) << R500_RS_IP_TEX_PTR_T_SHIFT) |
-			((rs_tex_count + 2) << R500_RS_IP_TEX_PTR_R_SHIFT) |
-			((rs_tex_count + 3) << R500_RS_IP_TEX_PTR_Q_SHIFT);
-
-		r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
-		InputsRead &= ~FRAG_BIT_WPOS;
-		rs_tex_count += 4;
-		++tex_ip;
-		++fp_reg;
-	}
-
-	if (InputsRead & FRAG_BIT_FOGC) {
-		if (R300_OUTPUTS_WRITTEN_TEST(OutputsWritten, VERT_RESULT_FOGC, _TNL_ATTRIB_FOG)) {
-			r300->hw.ri.cmd[R300_RI_INTERP_0 + tex_ip] |= (rs_tex_count << R500_RS_IP_TEX_PTR_S_SHIFT) |
-				(R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_T_SHIFT) |
-				(R500_RS_IP_PTR_K0 << R500_RS_IP_TEX_PTR_R_SHIFT) |
-				(R500_RS_IP_PTR_K1 << R500_RS_IP_TEX_PTR_Q_SHIFT);
-
-			r300->hw.rr.cmd[R300_RR_INST_0 + tex_ip] |= R500_RS_INST_TEX_ID(tex_ip) | R500_RS_INST_TEX_CN_WRITE | R500_RS_INST_TEX_ADDR(fp_reg);
-			InputsRead &= ~FRAG_BIT_FOGC;
-			rs_tex_count += 4;
-			++tex_ip;
-			++fp_reg;
-		} else {
-			WARN_ONCE("fragprog wants fogc, vp doesn't provide it\n");
-		}
-	}
-
 	/* Setup default color if no color or tex was set */
 	if (rs_tex_count == 0 && col_ip == 0) {
 		r300->hw.rr.cmd[R300_RR_INST_0] = R500_RS_INST_COL_ID(0) | R500_RS_INST_COL_ADDR(0);
@@ -2036,43 +1981,51 @@ static void r300ResetHwState(r300ContextPtr r300)
 
 void r300UpdateShaders(r300ContextPtr rmesa)
 {
-	GLcontext *ctx;
-	struct r300_fragment_program *fp;
-	int i;
-
-	ctx = rmesa->radeon.glCtx;
-	fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	GLcontext *ctx = rmesa->radeon.glCtx;
 
 	/* should only happenen once, just after context is created */
 	/* TODO: shouldn't we fallback to sw here? */
-	if (!fp) {
+	if (!ctx->FragmentProgram._Current) {
 		_mesa_fprintf(stderr, "No ctx->FragmentProgram._Current!!\n");
 		return;
 	}
 
-	if (rmesa->radeon.NewGLState && rmesa->options.hw_tcl_enabled) {
-		for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
-			rmesa->temp_attrib[i] =
-			    TNL_CONTEXT(ctx)->vb.AttribPtr[i];
-			TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
-			    &rmesa->dummy_attrib[i];
-		}
+	{
+		struct r300_fragment_program *fp;
 
-		_tnl_UpdateFixedFunctionProgram(ctx);
+		fp = r300SelectFragmentShader(ctx);
+		if (!fp->translated)
+			r300TranslateFragmentShader(ctx, fp);
 
-		for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
-			TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
-			    rmesa->temp_attrib[i];
-		}
-
-		r300SelectVertexShader(rmesa);
-		r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, rmesa->selected_vp->error);
+		r300SwitchFallback(ctx, R300_FALLBACK_FRAGMENT_PROGRAM, fp->error);
 	}
 
-	if (!fp->translated || rmesa->radeon.NewGLState)
-		r300TranslateFragmentShader(ctx, ctx->FragmentProgram._Current);
+	if (rmesa->options.hw_tcl_enabled) {
+		struct r300_vertex_program *vp;
+
+		if (rmesa->radeon.NewGLState) {
+			int i;
+			for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
+				rmesa->temp_attrib[i] =
+				    TNL_CONTEXT(ctx)->vb.AttribPtr[i];
+				TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
+				    &rmesa->dummy_attrib[i];
+			}
+
+			_tnl_UpdateFixedFunctionProgram(ctx);
+
+			for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
+				TNL_CONTEXT(ctx)->vb.AttribPtr[i] =
+				    rmesa->temp_attrib[i];
+			}
+		}
+
+		vp = r300SelectVertexShader(ctx);
+		if (!vp->translated)
+			r300TranslateVertexShader(vp);
 
-	r300SwitchFallback(ctx, R300_FALLBACK_FRAGMENT_PROGRAM, fp->error);
+		r300SwitchFallback(ctx, R300_FALLBACK_VERTEX_PROGRAM, vp->error);
+	}
 
 	r300UpdateStateParameters(ctx, _NEW_PROGRAM | _NEW_PROGRAM_CONSTANTS);
 	rmesa->radeon.NewGLState = 0;
@@ -2102,7 +2055,7 @@ static const GLfloat *get_fragmentprogram_constant(GLcontext *ctx,
 static void r300SetupPixelShader(GLcontext *ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	struct r300_fragment_program *fp = rmesa->selected_fp;
 	struct r300_fragment_program_code *code;
 	int i, k;
 
@@ -2148,8 +2101,7 @@ 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, code->const_nr * 4);
 	for (i = 0; i < code->const_nr; i++) {
-		const GLfloat *constant = get_fragmentprogram_constant(ctx,
-			&fp->Base.Base, code->constant[i]);
+		const GLfloat *constant = get_fragmentprogram_constant(ctx, fp->Base, code->constant[i]);
 		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]);
@@ -2174,7 +2126,7 @@ static void r300SetupPixelShader(GLcontext *ctx)
 static void r500SetupPixelShader(GLcontext *ctx)
 {
 	r300ContextPtr rmesa = R300_CONTEXT(ctx);
-	struct r300_fragment_program *fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
+	struct r300_fragment_program *fp = rmesa->selected_fp;
 	int i;
 	struct r500_fragment_program_code *code;
 
@@ -2210,8 +2162,7 @@ static void r500SetupPixelShader(GLcontext *ctx)
 
 	R300_STATECHANGE(rmesa, r500fp_const);
 	for (i = 0; i < code->const_nr; i++) {
-		const GLfloat *constant = get_fragmentprogram_constant(ctx,
-			&fp->Base.Base, code->constant[i]);
+		const GLfloat *constant = get_fragmentprogram_constant(ctx, fp->Base, code->constant[i]);
 		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]);
@@ -2274,20 +2225,17 @@ void r300SetupVAP(GLcontext *ctx, GLuint InputsRead, GLuint OutputsWritten)
 
 	rmesa->hw.vic.cmd[R300_VIC_CNTL_0] = r300VAPInputCntl0(ctx, InputsRead);
 	rmesa->hw.vic.cmd[R300_VIC_CNTL_1] = r300VAPInputCntl1(ctx, InputsRead);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
-	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten, ctx->FragmentProgram._Current->Base.InputsRead);
+	rmesa->hw.vof.cmd[R300_VOF_CNTL_0] = r300VAPOutputCntl0(ctx, OutputsWritten);
+	rmesa->hw.vof.cmd[R300_VOF_CNTL_1] = r300VAPOutputCntl1(ctx, OutputsWritten);
 }
 
 void r300UpdateShaderStates(r300ContextPtr rmesa)
 {
 	GLcontext *ctx;
 	ctx = rmesa->radeon.glCtx;
-	struct r300_fragment_program *r300_fp;
-
-	r300_fp = (struct r300_fragment_program *) ctx->FragmentProgram._Current;
 
 	/* should only happenen once, just after context is created */
-	if (!r300_fp)
+	if (!ctx->FragmentProgram._Current)
 		return;
 
 	r300SetEarlyZState(ctx);
@@ -2323,8 +2271,6 @@ static void r300InvalidateState(GLcontext * ctx, GLuint new_state)
 		R300_STATECHANGE(r300, cb);
 	}
 
-	r300UpdateStateParameters(ctx, new_state);
-
 	r300->radeon.NewGLState |= new_state;
 }
 
diff --git a/src/mesa/drivers/dri/r300/r300_state.h b/src/mesa/drivers/dri/r300/r300_state.h
index 23282894202..d46bf9f1796 100644
--- a/src/mesa/drivers/dri/r300/r300_state.h
+++ b/src/mesa/drivers/dri/r300/r300_state.h
@@ -52,7 +52,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 void r300UpdateViewportOffset (GLcontext * ctx);
 void r300UpdateDrawBuffer (GLcontext * ctx);
-void r300UpdateStateParameters (GLcontext * ctx, GLuint new_state);
 void r300UpdateShaders (r300ContextPtr rmesa);
 void r300UpdateShaderStates (r300ContextPtr rmesa);
 void r300InitState (r300ContextPtr r300);
diff --git a/src/mesa/drivers/dri/r300/r300_swtcl.c b/src/mesa/drivers/dri/r300/r300_swtcl.c
index ce4179208eb..56ed519cf41 100644
--- a/src/mesa/drivers/dri/r300/r300_swtcl.c
+++ b/src/mesa/drivers/dri/r300/r300_swtcl.c
@@ -76,7 +76,7 @@ void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead,  GLuint *_
 	GLuint InputsRead = 0;
 	GLuint OutputsWritten = 0;
 	int num_attrs = 0;
-	GLuint fp_reads = ctx->FragmentProgram._Current->Base.InputsRead;
+	GLuint fp_reads = rmesa->selected_fp->Base->InputsRead;
 	struct vertex_attribute *attrs = rmesa->vbuf.attribs;
 
 	rmesa->swtcl.coloroffset = rmesa->swtcl.specoffset = 0;
@@ -150,6 +150,22 @@ void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead,  GLuint *_
 		ADD_ATTR(VERT_ATTRIB_POINT_SIZE, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_POINT_SIZE, swiz, MASK_X, 0);
 	}
 
+	if (rmesa->selected_fp->wpos_attr != FRAG_ATTRIB_MAX) {
+		int tex_id = rmesa->selected_fp->wpos_attr - FRAG_ATTRIB_TEX0;
+
+		VB->AttribPtr[VERT_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_POS];
+		VB->TexCoordPtr[tex_id] = VB->AttribPtr[VERT_ATTRIB_POS];
+		RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
+	}
+
+	if (rmesa->selected_fp->fog_attr != FRAG_ATTRIB_MAX) {
+		int tex_id = rmesa->selected_fp->fog_attr - FRAG_ATTRIB_TEX0;
+
+		VB->AttribPtr[VERT_ATTRIB_TEX0 + tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+		VB->TexCoordPtr[tex_id] = VB->AttribPtr[VERT_ATTRIB_FOG];
+		RENDERINPUTS_SET(tnl->render_inputs_bitset, _TNL_ATTRIB_TEX0 + tex_id);
+	}
+
 	/**
 	 *  Sending only one texcoord component may lead to lock up,
 	 *  so for all textures always output 4 texcoord components to RS.
@@ -192,31 +208,9 @@ void r300ChooseSwtclVertexFormat(GLcontext *ctx, GLuint *_InputsRead,  GLuint *_
 		}
 	}
 
-	/* RS can't put fragment position on the pixel stack, so stuff it in texcoord if needed */
-	if (fp_reads & FRAG_BIT_WPOS) {
-		if (first_free_tex >= ctx->Const.MaxTextureUnits) {
-			fprintf(stderr, "\tout of free texcoords to write w pos\n");
-			_mesa_exit(-1);
-		}
-
-		InputsRead |= 1 << (VERT_ATTRIB_TEX0 + first_free_tex);
-		OutputsWritten |= 1 << (VERT_RESULT_TEX0 + first_free_tex);
-		EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F );
-		ADD_ATTR(VERT_ATTRIB_POS, R300_DATA_TYPE_FLOAT_4, SWTCL_OVM_TEX(first_free_tex), SWIZZLE_XYZW, MASK_XYZW, 0);
-		++first_free_tex;
-	}
-
-	if (fp_reads & FRAG_BIT_FOGC) {
-		if (first_free_tex >= ctx->Const.MaxTextureUnits) {
-			fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
-			_mesa_exit(-1);
-		}
-
-		InputsRead |= 1 << VERT_ATTRIB_FOG;
-		OutputsWritten |= 1 << VERT_RESULT_FOGC;
-		GLuint swiz = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO);
-		EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1F );
-		ADD_ATTR(VERT_ATTRIB_FOG, R300_DATA_TYPE_FLOAT_1, SWTCL_OVM_TEX(first_free_tex), swiz, MASK_XYZW, 0);
+	if (first_free_tex >= ctx->Const.MaxTextureUnits) {
+		fprintf(stderr, "\tout of free texcoords to write fog coordinate\n");
+		_mesa_exit(-1);
 	}
 
 	R300_NEWPRIM(rmesa);
@@ -497,11 +491,13 @@ void r300RenderStart(GLcontext *ctx)
 	r300ContextPtr rmesa = R300_CONTEXT( ctx );
 
 	r300ChooseRenderState(ctx);
+
+	r300UpdateShaders(rmesa);
+
 	r300PrepareVertices(ctx);
 
 	r300ValidateBuffers(ctx);
 
-	r300UpdateShaders(rmesa);
 	r300UpdateShaderStates(rmesa);
 
 	r300EmitCacheFlush(rmesa);
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index c41a8fdd621..de32013032f 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -32,12 +32,15 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "main/macros.h"
 #include "main/enums.h"
 #include "shader/program.h"
+#include "shader/programopt.h"
 #include "shader/prog_instruction.h"
+#include "shader/prog_optimize.h"
 #include "shader/prog_parameter.h"
 #include "shader/prog_print.h"
 #include "shader/prog_statevars.h"
 #include "tnl/tnl.h"
 
+#include "radeon_nqssadce.h"
 #include "r300_context.h"
 #include "r300_state.h"
 
@@ -71,15 +74,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 		u_temp_i=VSF_MAX_FRAGMENT_TEMPS-1; \
 	} while (0)
 
-int r300VertexProgUpdateParams(GLcontext * ctx,
-			       struct r300_vertex_program_cont *vp, float *dst)
+static int r300VertexProgUpdateParams(GLcontext * ctx, struct gl_vertex_program *vp, float *dst)
 {
 	int pi;
-	struct gl_vertex_program *mesa_vp = &vp->mesa_program;
 	float *dst_o = dst;
 	struct gl_program_parameter_list *paramList;
 
-	if (mesa_vp->IsNVProgram) {
+	if (vp->IsNVProgram) {
 		_mesa_load_tracked_matrices(ctx);
 
 		for (pi = 0; pi < MAX_NV_VERTEX_PROGRAM_PARAMS; pi++) {
@@ -91,16 +92,18 @@ int r300VertexProgUpdateParams(GLcontext * ctx,
 		return dst - dst_o;
 	}
 
-	assert(mesa_vp->Base.Parameters);
-	_mesa_load_state_parameters(ctx, mesa_vp->Base.Parameters);
+	if (!vp->Base.Parameters)
+		return 0;
 
-	if (mesa_vp->Base.Parameters->NumParameters * 4 >
+	_mesa_load_state_parameters(ctx, vp->Base.Parameters);
+
+	if (vp->Base.Parameters->NumParameters * 4 >
 	    VSF_MAX_FRAGMENT_LENGTH) {
 		fprintf(stderr, "%s:Params exhausted\n", __FUNCTION__);
 		_mesa_exit(-1);
 	}
 
-	paramList = mesa_vp->Base.Parameters;
+	paramList = vp->Base.Parameters;
 	for (pi = 0; pi < paramList->NumParameters; pi++) {
 		switch (paramList->Parameters[pi].Type) {
 		case PROGRAM_STATE_VAR:
@@ -933,10 +936,14 @@ static void t_inputs_outputs(struct r300_vertex_program *vp)
 {
 	int i;
 	int cur_reg;
+	GLuint OutputsWritten, InputsRead;
+
+	OutputsWritten = vp->Base->Base.OutputsWritten;
+	InputsRead = vp->Base->Base.InputsRead;
 
 	cur_reg = -1;
 	for (i = 0; i < VERT_ATTRIB_MAX; i++) {
-		if (vp->key.InputsRead & (1 << i))
+		if (InputsRead & (1 << i))
 			vp->inputs[i] = ++cur_reg;
 		else
 			vp->inputs[i] = -1;
@@ -946,13 +953,13 @@ static void t_inputs_outputs(struct r300_vertex_program *vp)
 	for (i = 0; i < VERT_RESULT_MAX; i++)
 		vp->outputs[i] = -1;
 
-	assert(vp->key.OutputsWritten & (1 << VERT_RESULT_HPOS));
+	assert(OutputsWritten & (1 << VERT_RESULT_HPOS));
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_HPOS)) {
+	if (OutputsWritten & (1 << VERT_RESULT_HPOS)) {
 		vp->outputs[VERT_RESULT_HPOS] = cur_reg++;
 	}
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
+	if (OutputsWritten & (1 << VERT_RESULT_PSIZ)) {
 		vp->outputs[VERT_RESULT_PSIZ] = cur_reg++;
 	}
 
@@ -962,46 +969,46 @@ static void t_inputs_outputs(struct r300_vertex_program *vp)
 	 * pretend it does by skipping output index reg so the colors
 	 * get written into appropriate output vectors.
 	 */
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL0)) {
+	if (OutputsWritten & (1 << VERT_RESULT_COL0)) {
 		vp->outputs[VERT_RESULT_COL0] = cur_reg++;
-	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
-		vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+	} else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+		OutputsWritten & (1 << VERT_RESULT_BFC1)) {
 		cur_reg++;
 	}
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_COL1)) {
+	if (OutputsWritten & (1 << VERT_RESULT_COL1)) {
 		vp->outputs[VERT_RESULT_COL1] = cur_reg++;
-	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0) ||
-		vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+	} else if (OutputsWritten & (1 << VERT_RESULT_BFC0) ||
+		OutputsWritten & (1 << VERT_RESULT_BFC1)) {
 		cur_reg++;
 	}
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+	if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
 		vp->outputs[VERT_RESULT_BFC0] = cur_reg++;
-	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+	} else if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
 		cur_reg++;
 	}
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC1)) {
+	if (OutputsWritten & (1 << VERT_RESULT_BFC1)) {
 		vp->outputs[VERT_RESULT_BFC1] = cur_reg++;
-	} else if (vp->key.OutputsWritten & (1 << VERT_RESULT_BFC0)) {
+	} else if (OutputsWritten & (1 << VERT_RESULT_BFC0)) {
 		cur_reg++;
 	}
 
 	for (i = VERT_RESULT_TEX0; i <= VERT_RESULT_TEX7; i++) {
-		if (vp->key.OutputsWritten & (1 << i)) {
+		if (OutputsWritten & (1 << i)) {
 			vp->outputs[i] = cur_reg++;
 		}
 	}
 
-	if (vp->key.OutputsWritten & (1 << VERT_RESULT_FOGC)) {
+	if (OutputsWritten & (1 << VERT_RESULT_FOGC)) {
 		vp->outputs[VERT_RESULT_FOGC] = cur_reg++;
 	}
 }
 
-static void r300TranslateVertexShader(struct r300_vertex_program *vp,
-				      struct prog_instruction *vpi)
+void r300TranslateVertexShader(struct r300_vertex_program *vp)
 {
+	struct prog_instruction *vpi = vp->Base->Base.Instructions;
 	int i;
 	GLuint *inst;
 	unsigned long num_operands;
@@ -1191,313 +1198,463 @@ static void r300TranslateVertexShader(struct r300_vertex_program *vp,
 	}
 }
 
-/* DP4 version seems to trigger some hw peculiarity */
-//#define PREFER_DP4
+static void insert_wpos(struct gl_program *prog, GLuint temp_index, int tex_id)
+{
+	struct prog_instruction *vpi;
+
+	_mesa_insert_instructions(prog, prog->NumInstructions - 1, 2);
+
+	vpi = &prog->Instructions[prog->NumInstructions - 3];
+
+	vpi->Opcode = OPCODE_MOV;
+
+	vpi->DstReg.File = PROGRAM_OUTPUT;
+	vpi->DstReg.Index = VERT_RESULT_HPOS;
+	vpi->DstReg.WriteMask = WRITEMASK_XYZW;
+	vpi->DstReg.CondMask = COND_TR;
+
+	vpi->SrcReg[0].File = PROGRAM_TEMPORARY;
+	vpi->SrcReg[0].Index = temp_index;
+	vpi->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+	++vpi;
+
+	vpi->Opcode = OPCODE_MOV;
 
-static void position_invariant(struct gl_program *prog)
+	vpi->DstReg.File = PROGRAM_OUTPUT;
+	vpi->DstReg.Index = VERT_RESULT_TEX0 + tex_id;
+	vpi->DstReg.WriteMask = WRITEMASK_XYZW;
+	vpi->DstReg.CondMask = COND_TR;
+
+	vpi->SrcReg[0].File = PROGRAM_TEMPORARY;
+	vpi->SrcReg[0].Index = temp_index;
+	vpi->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+	++vpi;
+
+	vpi->Opcode = OPCODE_END;
+}
+
+static void pos_as_texcoord(struct gl_program *prog, int tex_id)
 {
 	struct prog_instruction *vpi;
-	struct gl_program_parameter_list *paramList;
-	int i;
+	GLuint tempregi = prog->NumTemporaries;
 
-	gl_state_index tokens[STATE_LENGTH] = { STATE_MVP_MATRIX, 0, 0, 0, 0 };
+	prog->NumTemporaries++;
 
-	/* tokens[4] = matrix modifier */
-#ifdef PREFER_DP4
-	tokens[4] = 0;		/* not transposed or inverted */
-#else
-	tokens[4] = STATE_MATRIX_TRANSPOSE;
-#endif
-	paramList = prog->Parameters;
-
-	vpi = _mesa_alloc_instructions(prog->NumInstructions + 4);
-	_mesa_init_instructions(vpi, prog->NumInstructions + 4);
-
-	for (i = 0; i < 4; i++) {
-		GLint idx;
-		tokens[2] = tokens[3] = i;	/* matrix row[i]..row[i] */
-		idx = _mesa_add_state_reference(paramList, tokens);
-#ifdef PREFER_DP4
-		vpi[i].Opcode = OPCODE_DP4;
-		vpi[i].StringPos = 0;
-		vpi[i].Data = 0;
-
-		vpi[i].DstReg.File = PROGRAM_OUTPUT;
-		vpi[i].DstReg.Index = VERT_RESULT_HPOS;
-		vpi[i].DstReg.WriteMask = 1 << i;
-		vpi[i].DstReg.CondMask = COND_TR;
-
-		vpi[i].SrcReg[0].File = PROGRAM_STATE_VAR;
-		vpi[i].SrcReg[0].Index = idx;
-		vpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-
-		vpi[i].SrcReg[1].File = PROGRAM_INPUT;
-		vpi[i].SrcReg[1].Index = VERT_ATTRIB_POS;
-		vpi[i].SrcReg[1].Swizzle = SWIZZLE_XYZW;
-#else
-		if (i == 0)
-			vpi[i].Opcode = OPCODE_MUL;
-		else
-			vpi[i].Opcode = OPCODE_MAD;
+	for (vpi = prog->Instructions; vpi->Opcode != OPCODE_END; vpi++) {
+		if (vpi->DstReg.File == PROGRAM_OUTPUT && vpi->DstReg.Index == VERT_RESULT_HPOS) {
+			vpi->DstReg.File = PROGRAM_TEMPORARY;
+			vpi->DstReg.Index = tempregi;
+		}
+	}
 
-		vpi[i].Data = 0;
+	insert_wpos(prog, tempregi, tex_id);
 
-		if (i == 3)
-			vpi[i].DstReg.File = PROGRAM_OUTPUT;
-		else
-			vpi[i].DstReg.File = PROGRAM_TEMPORARY;
-		vpi[i].DstReg.Index = 0;
-		vpi[i].DstReg.WriteMask = 0xf;
-		vpi[i].DstReg.CondMask = COND_TR;
-
-		vpi[i].SrcReg[0].File = PROGRAM_STATE_VAR;
-		vpi[i].SrcReg[0].Index = idx;
-		vpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-
-		vpi[i].SrcReg[1].File = PROGRAM_INPUT;
-		vpi[i].SrcReg[1].Index = VERT_ATTRIB_POS;
-		vpi[i].SrcReg[1].Swizzle = MAKE_SWIZZLE4(i, i, i, i);
-
-		if (i > 0) {
-			vpi[i].SrcReg[2].File = PROGRAM_TEMPORARY;
-			vpi[i].SrcReg[2].Index = 0;
-			vpi[i].SrcReg[2].Swizzle = SWIZZLE_XYZW;
+	prog->OutputsWritten |= 1 << (VERT_RESULT_TEX0 + tex_id);
+}
+
+/**
+ * The fogcoord attribute is special in that only the first component
+ * is relevant, and the remaining components are always fixed (when read
+ * from by the fragment program) to yield an X001 pattern.
+ *
+ * We need to enforce this either in the vertex program or in the fragment
+ * program, and this code chooses not to enforce it in the vertex program.
+ * This is slightly cheaper, as long as the fragment program does not use
+ * weird swizzles.
+ *
+ * And it seems that usually, weird swizzles are not used, so...
+ *
+ * See also the counterpart rewriting for fragment programs.
+ */
+static void fog_as_texcoord(struct gl_program *prog, int tex_id)
+{
+	struct prog_instruction *vpi;
+
+	vpi = prog->Instructions;
+	while (vpi->Opcode != OPCODE_END) {
+		if (vpi->DstReg.File == PROGRAM_OUTPUT && vpi->DstReg.Index == VERT_RESULT_FOGC) {
+			vpi->DstReg.Index = VERT_RESULT_TEX0 + tex_id;
+			vpi->DstReg.WriteMask = WRITEMASK_X;
 		}
-#endif
+
+		++vpi;
 	}
 
-	_mesa_copy_instructions(&vpi[i], prog->Instructions,
-				prog->NumInstructions);
+	prog->OutputsWritten &= ~(1 << VERT_RESULT_FOGC);
+	prog->OutputsWritten |= 1 << (VERT_RESULT_TEX0 + tex_id);
+}
 
-	free(prog->Instructions);
+static int translateABS(struct gl_program *prog, int pos)
+{
+	struct prog_instruction *inst;
 
-	prog->Instructions = vpi;
+	inst = &prog->Instructions[pos];
 
-	prog->NumInstructions += 4;
-	vpi = &prog->Instructions[prog->NumInstructions - 1];
+	inst->Opcode = OPCODE_MAX;
+	inst->SrcReg[1] = inst->SrcReg[0];
+	inst->SrcReg[1].Negate ^= NEGATE_XYZW;
 
-	assert(vpi->Opcode == OPCODE_END);
+	return 0;
 }
 
-static void insert_wpos(struct r300_vertex_program *vp, struct gl_program *prog,
-			GLuint temp_index)
+static int translateDP3(struct gl_program *prog, int pos)
 {
-	struct prog_instruction *vpi;
-	struct prog_instruction *vpi_insert;
-	int i = 0;
+	struct prog_instruction *inst;
 
-	vpi = _mesa_alloc_instructions(prog->NumInstructions + 2);
-	_mesa_init_instructions(vpi, prog->NumInstructions + 2);
-	/* all but END */
-	_mesa_copy_instructions(vpi, prog->Instructions,
-				prog->NumInstructions - 1);
-	/* END */
-	_mesa_copy_instructions(&vpi[prog->NumInstructions + 1],
-				&prog->Instructions[prog->NumInstructions - 1],
-				1);
-	vpi_insert = &vpi[prog->NumInstructions - 1];
+	inst = &prog->Instructions[pos];
 
-	vpi_insert[i].Opcode = OPCODE_MOV;
+	inst->Opcode = OPCODE_DP4;
+	inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
 
-	vpi_insert[i].DstReg.File = PROGRAM_OUTPUT;
-	vpi_insert[i].DstReg.Index = VERT_RESULT_HPOS;
-	vpi_insert[i].DstReg.WriteMask = WRITEMASK_XYZW;
-	vpi_insert[i].DstReg.CondMask = COND_TR;
+	return 0;
+}
+
+static int translateDPH(struct gl_program *prog, int pos)
+{
+	struct prog_instruction *inst;
+
+	inst = &prog->Instructions[pos];
+
+	inst->Opcode = OPCODE_DP4;
+	inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE);
 
-	vpi_insert[i].SrcReg[0].File = PROGRAM_TEMPORARY;
-	vpi_insert[i].SrcReg[0].Index = temp_index;
-	vpi_insert[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-	i++;
+	return 0;
+}
 
-	vpi_insert[i].Opcode = OPCODE_MOV;
+static int translateFLR(struct gl_program *prog, int pos)
+{
+	struct prog_instruction *inst;
+	struct prog_dst_register dst;
+	int tmp_idx;
 
-	vpi_insert[i].DstReg.File = PROGRAM_OUTPUT;
-	vpi_insert[i].DstReg.Index = VERT_RESULT_TEX0 + vp->wpos_idx;
-	vpi_insert[i].DstReg.WriteMask = WRITEMASK_XYZW;
-	vpi_insert[i].DstReg.CondMask = COND_TR;
+	tmp_idx = prog->NumTemporaries++;
 
-	vpi_insert[i].SrcReg[0].File = PROGRAM_TEMPORARY;
-	vpi_insert[i].SrcReg[0].Index = temp_index;
-	vpi_insert[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-	i++;
+	_mesa_insert_instructions(prog, pos + 1, 1);
 
-	free(prog->Instructions);
+	inst = &prog->Instructions[pos];
+	dst = inst->DstReg;
 
-	prog->Instructions = vpi;
+	inst->Opcode = OPCODE_FRC;
+	inst->DstReg.File = PROGRAM_TEMPORARY;
+	inst->DstReg.Index = tmp_idx;
+	++inst;
 
-	prog->NumInstructions += i;
-	vpi = &prog->Instructions[prog->NumInstructions - 1];
+	inst->Opcode = OPCODE_ADD;
+	inst->DstReg = dst;
+	inst->SrcReg[0] = (inst-1)->SrcReg[0];
+	inst->SrcReg[1].File = PROGRAM_TEMPORARY;
+	inst->SrcReg[1].Index = tmp_idx;
+	inst->SrcReg[1].Negate = NEGATE_XYZW;
 
-	assert(vpi->Opcode == OPCODE_END);
+	return 1;
 }
 
-static void pos_as_texcoord(struct r300_vertex_program *vp,
-			    struct gl_program *prog)
+static int translateSUB(struct gl_program *prog, int pos)
 {
-	struct prog_instruction *vpi;
-	GLuint tempregi = prog->NumTemporaries;
-	/* should do something else if no temps left... */
-	prog->NumTemporaries++;
+	struct prog_instruction *inst;
 
-	for (vpi = prog->Instructions; vpi->Opcode != OPCODE_END; vpi++) {
-		if (vpi->DstReg.File == PROGRAM_OUTPUT
-		    && vpi->DstReg.Index == VERT_RESULT_HPOS) {
-			vpi->DstReg.File = PROGRAM_TEMPORARY;
-			vpi->DstReg.Index = tempregi;
-		}
-	}
-	insert_wpos(vp, prog, tempregi);
+	inst = &prog->Instructions[pos];
+
+	inst->Opcode = OPCODE_ADD;
+	inst->SrcReg[1].Negate ^= NEGATE_XYZW;
+
+	return 0;
 }
 
-static struct r300_vertex_program *build_program(struct r300_vertex_program_key
-						 *wanted_key, struct gl_vertex_program
-						 *mesa_vp, GLint wpos_idx)
+static int translateSWZ(struct gl_program *prog, int pos)
 {
-	struct r300_vertex_program *vp;
+	prog->Instructions[pos].Opcode = OPCODE_MOV;
 
-	vp = _mesa_calloc(sizeof(*vp));
-	_mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key));
-	vp->wpos_idx = wpos_idx;
+	return 0;
+}
 
-	if (mesa_vp->IsPositionInvariant) {
-		position_invariant(&mesa_vp->Base);
-	}
+static int translateXPD(struct gl_program *prog, int pos)
+{
+	struct prog_instruction *inst;
+	int tmp_idx;
+
+	tmp_idx = prog->NumTemporaries++;
+
+	_mesa_insert_instructions(prog, pos + 1, 1);
+
+	inst = &prog->Instructions[pos];
+
+	*(inst+1) = *inst;
+
+	inst->Opcode = OPCODE_MUL;
+	inst->DstReg.File = PROGRAM_TEMPORARY;
+	inst->DstReg.Index = tmp_idx;
+	inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W);
+	inst->SrcReg[1].Swizzle = combine_swizzles4(inst->SrcReg[1].Swizzle, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W);
+	++inst;
+
+	inst->Opcode = OPCODE_MAD;
+	inst->SrcReg[0].Swizzle = combine_swizzles4(inst->SrcReg[0].Swizzle, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W);
+	inst->SrcReg[1].Swizzle = combine_swizzles4(inst->SrcReg[1].Swizzle, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W);
+	inst->SrcReg[1].Negate ^= NEGATE_XYZW;
+	inst->SrcReg[2].File = PROGRAM_TEMPORARY;
+	inst->SrcReg[2].Index = tmp_idx;
+
+	return 1;
+}
+
+static void translateInsts(struct gl_program *prog)
+{
+	struct prog_instruction *inst;
+	int i;
+
+	for (i = 0; i < prog->NumInstructions; ++i) {
+		inst = &prog->Instructions[i];
 
-	if (wpos_idx > -1) {
-		pos_as_texcoord(vp, &mesa_vp->Base);
+		switch (inst->Opcode) {
+			case OPCODE_ABS:
+				i += translateABS(prog, i);
+				break;
+			case OPCODE_DP3:
+				i += translateDP3(prog, i);
+				break;
+			case OPCODE_DPH:
+				i += translateDPH(prog, i);
+				break;
+			case OPCODE_FLR:
+				i += translateFLR(prog, i);
+				break;
+			case OPCODE_SUB:
+				i += translateSUB(prog, i);
+				break;
+			case OPCODE_SWZ:
+				i += translateSWZ(prog, i);
+				break;
+			case OPCODE_XPD:
+				i += translateXPD(prog, i);
+				break;
+			default:
+				break;
+		}
 	}
+}
 
-	if (RADEON_DEBUG & DEBUG_VERTS) {
-		fprintf(stderr, "Vertex program after native rewrite:\n");
-		_mesa_print_program(&mesa_vp->Base);
-		fflush(stdout);
+#define ADD_OUTPUT(fp_attr, vp_result) \
+	do { \
+		if ((FpReads & (1 << (fp_attr))) && !(prog->OutputsWritten & (1 << (vp_result)))) { \
+			OutputsAdded |= 1 << (vp_result); \
+			count++; \
+		} \
+	} while (0)
+
+static void addArtificialOutputs(GLcontext *ctx, struct gl_program *prog)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	GLuint OutputsAdded, FpReads;
+	int i, count;
+
+	OutputsAdded = 0;
+	count = 0;
+	FpReads = r300->selected_fp->Base->InputsRead;
+
+	ADD_OUTPUT(FRAG_ATTRIB_COL0, VERT_RESULT_COL0);
+	ADD_OUTPUT(FRAG_ATTRIB_COL1, VERT_RESULT_COL1);
+
+	for (i = 0; i < 7; ++i) {
+		ADD_OUTPUT(FRAG_ATTRIB_TEX0 + i, VERT_RESULT_TEX0 + i);
 	}
 
 	/* Some outputs may be artificially added, to match the inputs of the fragment program.
 	 * Issue 16 of vertex program spec says that all vertex attributes that are unwritten by
 	 * vertex program are undefined, so just use MOV [vertex_result], CONST[0]
 	 */
-	{
-		int i, count = 0;
+	if (count > 0) {
+		struct prog_instruction *inst;
+
+		_mesa_insert_instructions(prog, prog->NumInstructions - 1, count);
+		inst = &prog->Instructions[prog->NumInstructions - 1 - count];
+
 		for (i = 0; i < VERT_RESULT_MAX; ++i) {
-			if (vp->key.OutputsAdded & (1 << i)) {
-				++count;
+			if (OutputsAdded & (1 << i)) {
+				inst->Opcode = OPCODE_MOV;
+
+				inst->DstReg.File = PROGRAM_OUTPUT;
+				inst->DstReg.Index = i;
+				inst->DstReg.WriteMask = WRITEMASK_XYZW;
+				inst->DstReg.CondMask = COND_TR;
+
+				inst->SrcReg[0].File = PROGRAM_CONSTANT;
+				inst->SrcReg[0].Index = 0;
+				inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+				++inst;
 			}
 		}
 
-		if (count > 0) {
-			struct prog_instruction *inst;
+		prog->OutputsWritten |= OutputsAdded;
+	}
+}
 
-			_mesa_insert_instructions(&mesa_vp->Base, mesa_vp->Base.NumInstructions - 1, count);
-			inst = &mesa_vp->Base.Instructions[mesa_vp->Base.NumInstructions - 1 - count];
+#undef ADD_OUTPUT
 
-			for (i = 0; i < VERT_RESULT_MAX; ++i) {
-				if (vp->key.OutputsAdded & (1 << i)) {
-					inst->Opcode = OPCODE_MOV;
+static void nqssadceInit(struct nqssadce_state* s)
+{
+	r300ContextPtr r300 = R300_CONTEXT(s->Ctx);
+	GLuint fp_reads;
 
-					inst->DstReg.File = PROGRAM_OUTPUT;
-					inst->DstReg.Index = i;
-					inst->DstReg.WriteMask = WRITEMASK_XYZW;
-					inst->DstReg.CondMask = COND_TR;
+	fp_reads = r300->selected_fp->Base->InputsRead;
+	{
+		if (fp_reads & FRAG_BIT_COL0) {
+				s->Outputs[VERT_RESULT_COL0].Sourced = WRITEMASK_XYZW;
+				s->Outputs[VERT_RESULT_BFC0].Sourced = WRITEMASK_XYZW;
+		}
 
-					inst->SrcReg[0].File = PROGRAM_CONSTANT;
-					inst->SrcReg[0].Index = 0;
-					inst->SrcReg[0].Swizzle = SWIZZLE_XYZW;
+		if (fp_reads & FRAG_BIT_COL1) {
+				s->Outputs[VERT_RESULT_COL1].Sourced = WRITEMASK_XYZW;
+				s->Outputs[VERT_RESULT_BFC1].Sourced = WRITEMASK_XYZW;
+		}
+	}
 
-					++inst;
-				}
+	{
+		int i;
+		for (i = 0; i < 8; ++i) {
+			if (fp_reads & FRAG_BIT_TEX(i)) {
+				s->Outputs[VERT_RESULT_TEX0 + i].Sourced = WRITEMASK_XYZW;
 			}
 		}
 	}
 
-	assert(mesa_vp->Base.NumInstructions);
-	vp->num_temporaries = mesa_vp->Base.NumTemporaries;
-	r300TranslateVertexShader(vp, mesa_vp->Base.Instructions);
-
-	return vp;
+	s->Outputs[VERT_RESULT_HPOS].Sourced = WRITEMASK_XYZW;
+	if (s->Program->OutputsWritten & (1 << VERT_RESULT_PSIZ))
+		s->Outputs[VERT_RESULT_PSIZ].Sourced = WRITEMASK_X;
 }
 
-static void add_outputs(struct r300_vertex_program_key *key, GLint vert)
+static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg)
 {
-	if (key->OutputsWritten & (1 << vert))
-		return;
+	(void) opcode;
+	(void) reg;
 
-	key->OutputsWritten |= 1 << vert;
-	key->OutputsAdded |= 1 << vert;
+	return GL_TRUE;
 }
 
-void r300SelectVertexShader(r300ContextPtr r300)
+static struct r300_vertex_program *build_program(GLcontext *ctx,
+						 struct r300_vertex_program_key *wanted_key,
+						 const struct gl_vertex_program *mesa_vp)
 {
-	GLcontext *ctx = ctx = r300->radeon.glCtx;
-	GLuint InputsRead;
-	struct r300_vertex_program_key wanted_key = { 0 };
-	GLint i;
-	struct r300_vertex_program_cont *vpc;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
 	struct r300_vertex_program *vp;
-	GLint wpos_idx;
+	struct gl_program *prog;
 
-	vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current;
-	wanted_key.InputsRead = vpc->mesa_program.Base.InputsRead;
-	wanted_key.OutputsWritten = vpc->mesa_program.Base.OutputsWritten;
-	InputsRead = ctx->FragmentProgram._Current->Base.InputsRead;
-
-	wpos_idx = -1;
-	if (InputsRead & FRAG_BIT_WPOS) {
-		for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
-			if (!(InputsRead & (FRAG_BIT_TEX0 << i)))
-				break;
+	vp = _mesa_calloc(sizeof(*vp));
+	vp->Base = (struct gl_vertex_program *) _mesa_clone_program(ctx, &mesa_vp->Base);
+	_mesa_memcpy(&vp->key, wanted_key, sizeof(vp->key));
 
-		if (i == ctx->Const.MaxTextureUnits) {
-			fprintf(stderr, "\tno free texcoord found\n");
-			_mesa_exit(-1);
-		}
+	prog = &vp->Base->Base;
+
+	if (RADEON_DEBUG & DEBUG_VERTS) {
+		fprintf(stderr, "Initial vertex program:\n");
+		_mesa_print_program(prog);
+		fflush(stdout);
+	}
 
-		wanted_key.OutputsWritten |= 1 << (VERT_RESULT_TEX0 + i);
-		wpos_idx = i;
+	if (vp->Base->IsPositionInvariant) {
+		_mesa_insert_mvp_code(ctx, vp->Base);
 	}
 
-	if (vpc->mesa_program.IsPositionInvariant) {
-		wanted_key.InputsRead |= (1 << VERT_ATTRIB_POS);
-		wanted_key.OutputsWritten |= (1 << VERT_RESULT_HPOS);
-	} else {
-		add_outputs(&wanted_key, VERT_RESULT_HPOS);
+	if (r300->selected_fp->wpos_attr != FRAG_ATTRIB_MAX) {
+		pos_as_texcoord(&vp->Base->Base, r300->selected_fp->wpos_attr - FRAG_ATTRIB_TEX0);
 	}
 
-	if (InputsRead & FRAG_BIT_COL0) {
-		add_outputs(&wanted_key, VERT_RESULT_COL0);
+	if (r300->selected_fp->fog_attr != FRAG_ATTRIB_MAX) {
+		fog_as_texcoord(&vp->Base->Base, r300->selected_fp->fog_attr - FRAG_ATTRIB_TEX0);
 	}
 
-	if (InputsRead & FRAG_BIT_COL1) {
-		add_outputs(&wanted_key, VERT_RESULT_COL1);
+	addArtificialOutputs(ctx, prog);
+
+	translateInsts(prog);
+
+	if (RADEON_DEBUG & DEBUG_VERTS) {
+		fprintf(stderr, "Vertex program after native rewrite:\n");
+		_mesa_print_program(prog);
+		fflush(stdout);
 	}
 
-	if (InputsRead & FRAG_BIT_FOGC) {
-		add_outputs(&wanted_key, VERT_RESULT_FOGC);
+	{
+		struct radeon_nqssadce_descr nqssadce = {
+			.Init = &nqssadceInit,
+			.IsNativeSwizzle = &swizzleIsNative,
+			.BuildSwizzle = NULL
+		};
+		radeonNqssaDce(ctx, prog, &nqssadce);
+
+		/* We need this step for reusing temporary registers */
+		_mesa_optimize_program(ctx, prog);
+
+		if (RADEON_DEBUG & DEBUG_VERTS) {
+			fprintf(stderr, "Vertex program after NQSSADCE:\n");
+			_mesa_print_program(prog);
+			fflush(stdout);
+		}
 	}
 
-	for (i = 0; i < ctx->Const.MaxTextureUnits; i++) {
-		if (InputsRead & (FRAG_BIT_TEX0 << i)) {
-			add_outputs(&wanted_key, VERT_RESULT_TEX0 + i);
+	assert(prog->NumInstructions);
+	{
+		struct prog_instruction *inst;
+		int max, i, tmp;
+
+		inst = prog->Instructions;
+		max = -1;
+		while (inst->Opcode != OPCODE_END) {
+			tmp = _mesa_num_inst_src_regs(inst->Opcode);
+			for (i = 0; i < tmp; ++i) {
+				if (inst->SrcReg[i].File == PROGRAM_TEMPORARY) {
+					if ((int) inst->SrcReg[i].Index > max) {
+						max = inst->SrcReg[i].Index;
+					}
+				}
+			}
+
+			if (_mesa_num_inst_dst_regs(inst->Opcode)) {
+				if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+					if ((int) inst->DstReg.Index > max) {
+						max = inst->DstReg.Index;
+					}
+				}
+			}
+			++inst;
 		}
+
+		/* We actually want highest index of used temporary register,
+		 * not the number of temporaries used.
+		 * These values aren't always the same.
+		 */
+		vp->num_temporaries = max + 1;
 	}
 
-	for (vp = vpc->progs; vp; vp = vp->next)
+	return vp;
+}
+
+struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	struct r300_vertex_program_key wanted_key = { 0 };
+	struct r300_vertex_program_cont *vpc;
+	struct r300_vertex_program *vp;
+
+	vpc = (struct r300_vertex_program_cont *)ctx->VertexProgram._Current;
+	wanted_key.FpReads = r300->selected_fp->Base->InputsRead;
+	wanted_key.FogAttr = r300->selected_fp->fog_attr;
+	wanted_key.WPosAttr = r300->selected_fp->wpos_attr;
+
+	for (vp = vpc->progs; vp; vp = vp->next) {
 		if (_mesa_memcmp(&vp->key, &wanted_key, sizeof(wanted_key))
 		    == 0) {
-			r300->selected_vp = vp;
-			return;
+			return r300->selected_vp = vp;
 		}
-
-	if (RADEON_DEBUG & DEBUG_VERTS) {
-		fprintf(stderr, "Initial vertex program:\n");
-		_mesa_print_program(&vpc->mesa_program.Base);
-		fflush(stdout);
 	}
 
-	vp = build_program(&wanted_key, &vpc->mesa_program, wpos_idx);
+	vp = build_program(ctx, &wanted_key, &vpc->mesa_program);
 	vp->next = vpc->progs;
 	vpc->progs = vp;
-	r300->selected_vp = vp;
+
+	return r300->selected_vp = vp;
 }
 
 #define bump_vpu_count(ptr, new_count)   do { \
@@ -1544,25 +1701,22 @@ void r300SetupVertexProgram(r300ContextPtr rmesa)
 	struct r300_vertex_program *prog = rmesa->selected_vp;
 	int inst_count = 0;
 	int param_count = 0;
-	
+
 	/* Reset state, in case we don't use something */
 	((drm_r300_cmd_header_t *) rmesa->hw.vpp.cmd)->vpu.count = 0;
 	((drm_r300_cmd_header_t *) rmesa->hw.vpi.cmd)->vpu.count = 0;
 	((drm_r300_cmd_header_t *) rmesa->hw.vps.cmd)->vpu.count = 0;
-	
+
 	R300_STATECHANGE(rmesa, vpp);
-	param_count = r300VertexProgUpdateParams(ctx,
-								(struct r300_vertex_program_cont *)
-								ctx->VertexProgram._Current,
-								(float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
+	param_count = r300VertexProgUpdateParams(ctx, prog->Base, (float *)&rmesa->hw.vpp.cmd[R300_VPP_PARAM_0]);
 	bump_vpu_count(rmesa->hw.vpp.cmd, param_count);
 	param_count /= 4;
 
 	r300EmitVertexProgram(rmesa, R300_PVS_CODE_START, &(prog->hw_code));
 	inst_count = (prog->hw_code.length / 4) - 1;
 
-	r300VapCntl(rmesa, _mesa_bitcount(prog->key.InputsRead),
-				 _mesa_bitcount(prog->key.OutputsWritten), prog->num_temporaries);
+	r300VapCntl(rmesa, _mesa_bitcount(prog->Base->Base.InputsRead),
+				 _mesa_bitcount(prog->Base->Base.OutputsWritten), prog->num_temporaries);
 
 	R300_STATECHANGE(rmesa, pvs);
 	rmesa->hw.pvs.cmd[R300_PVS_CNTL_1] = (0 << R300_PVS_FIRST_INST_SHIFT) | (inst_count << R300_PVS_XYZW_VALID_INST_SHIFT) |
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.h b/src/mesa/drivers/dri/r300/r300_vertprog.h
index b552e3fb1bd..2dab11c3378 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.h
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.h
@@ -34,4 +34,8 @@
 
 void r300SetupVertexProgram(r300ContextPtr rmesa);
 
+struct r300_vertex_program * r300SelectVertexShader(GLcontext *ctx);
+
+void r300TranslateVertexShader(struct r300_vertex_program *vp);
+
 #endif
diff --git a/src/mesa/drivers/dri/r300/radeon_nqssadce.c b/src/mesa/drivers/dri/r300/radeon_nqssadce.c
index 4a2e1cba402..840c9733b16 100644
--- a/src/mesa/drivers/dri/r300/radeon_nqssadce.c
+++ b/src/mesa/drivers/dri/r300/radeon_nqssadce.c
@@ -46,6 +46,7 @@ static struct register_state *get_reg_state(struct nqssadce_state* s, GLuint fil
 	switch(file) {
 	case PROGRAM_TEMPORARY: return &s->Temps[index];
 	case PROGRAM_OUTPUT: return &s->Outputs[index];
+	case PROGRAM_ADDRESS: return &s->Address;
 	default: return 0;
 	}
 }
@@ -56,7 +57,7 @@ static struct register_state *get_reg_state(struct nqssadce_state* s, GLuint fil
  *
  * @note Works correctly only for X, Y, Z, W swizzles, not for constant swizzles.
  */
-static struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register srcreg)
+struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register srcreg)
 {
 	struct prog_src_register tmp = srcreg;
 	int i;
@@ -114,47 +115,19 @@ static struct prog_instruction* track_used_srcreg(struct nqssadce_state* s,
 		deswz_source = sourced;
 	}
 
-	struct register_state *regstate = get_reg_state(s, inst->SrcReg[src].File, inst->SrcReg[src].Index);
+	struct register_state *regstate;
+
+	if (inst->SrcReg[src].RelAddr)
+		regstate = get_reg_state(s, PROGRAM_ADDRESS, 0);
+	else
+		regstate = get_reg_state(s, inst->SrcReg[src].File, inst->SrcReg[src].Index);
+
 	if (regstate)
 		regstate->Sourced |= deswz_source & 0xf;
 
 	return inst;
 }
 
-
-static void rewrite_depth_out(struct prog_instruction *inst)
-{
-	if (inst->DstReg.WriteMask & WRITEMASK_Z) {
-		inst->DstReg.WriteMask = WRITEMASK_W;
-	} else {
-		inst->DstReg.WriteMask = 0;
-		return;
-	}
-
-	switch (inst->Opcode) {
-	case OPCODE_FRC:
-	case OPCODE_MOV:
-		inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
-		break;
-	case OPCODE_ADD:
-	case OPCODE_MAX:
-	case OPCODE_MIN:
-	case OPCODE_MUL:
-		inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
-		inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]);
-		break;
-	case OPCODE_CMP:
-	case OPCODE_MAD:
-		inst->SrcReg[0] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[0]);
-		inst->SrcReg[1] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[1]);
-		inst->SrcReg[2] = lmul_swizzle(SWIZZLE_ZZZZ, inst->SrcReg[2]);
-		break;
-	default:
-		// Scalar instructions needn't be reswizzled
-		break;
-	}
-}
-
 static void unalias_srcregs(struct prog_instruction *inst, GLuint oldindex, GLuint newindex)
 {
 	int nsrc = _mesa_num_inst_src_regs(inst->Opcode);
@@ -189,11 +162,6 @@ static void process_instruction(struct nqssadce_state* s)
 		return;
 
 	if (inst->Opcode != OPCODE_KIL) {
-		if (s->Descr->RewriteDepthOut) {
-			if (inst->DstReg.File == PROGRAM_OUTPUT && inst->DstReg.Index == FRAG_RESULT_DEPTH)
-				rewrite_depth_out(inst);
-		}
-
 		struct register_state *regstate = get_reg_state(s, inst->DstReg.File, inst->DstReg.Index);
 		if (!regstate) {
 			_mesa_problem(s->Ctx, "NqssaDce: bad destination register (%i[%i])\n",
@@ -217,6 +185,7 @@ static void process_instruction(struct nqssadce_state* s)
 	 * might change the instruction stream under us, so we have
 	 * to be careful with the inst pointer. */
 	switch (inst->Opcode) {
+	case OPCODE_ARL:
 	case OPCODE_DDX:
 	case OPCODE_DDY:
 	case OPCODE_FRC:
@@ -227,6 +196,8 @@ static void process_instruction(struct nqssadce_state* s)
 	case OPCODE_MAX:
 	case OPCODE_MIN:
 	case OPCODE_MUL:
+	case OPCODE_SGE:
+	case OPCODE_SLT:
 		inst = track_used_srcreg(s, inst, 0, inst->DstReg.WriteMask);
 		inst = track_used_srcreg(s, inst, 1, inst->DstReg.WriteMask);
 		break;
@@ -258,12 +229,51 @@ static void process_instruction(struct nqssadce_state* s)
 	case OPCODE_TXP:
 		inst = track_used_srcreg(s, inst, 0, 0xf);
 		break;
+	case OPCODE_DST:
+		inst = track_used_srcreg(s, inst, 0, 0x6);
+		inst = track_used_srcreg(s, inst, 1, 0xa);
+		break;
+	case OPCODE_EXP:
+	case OPCODE_LOG:
+	case OPCODE_POW:
+		inst = track_used_srcreg(s, inst, 0, 0x3);
+		break;
+	case OPCODE_LIT:
+		inst = track_used_srcreg(s, inst, 0, 0xb);
+		break;
 	default:
 		_mesa_problem(s->Ctx, "NqssaDce: Unknown opcode %d\n", inst->Opcode);
 		return;
 	}
 }
 
+static void calculateInputsOutputs(struct gl_program *p)
+{
+	struct prog_instruction *inst;
+	GLuint InputsRead, OutputsWritten;
+
+	inst = p->Instructions;
+	InputsRead = 0;
+	OutputsWritten = 0;
+	while (inst->Opcode != OPCODE_END)
+	{
+		int i, num_src_regs;
+
+		num_src_regs = _mesa_num_inst_src_regs(inst->Opcode);
+		for (i = 0; i < num_src_regs; ++i) {
+			if (inst->SrcReg[i].File == PROGRAM_INPUT)
+				InputsRead |= 1 << inst->SrcReg[i].Index;
+		}
+
+		if (inst->DstReg.File == PROGRAM_OUTPUT)
+			OutputsWritten |= 1 << inst->DstReg.Index;
+
+		++inst;
+	}
+
+	p->InputsRead = InputsRead;
+	p->OutputsWritten = OutputsWritten;
+}
 
 void radeonNqssaDce(GLcontext *ctx, struct gl_program *p, struct radeon_nqssadce_descr* descr)
 {
@@ -280,4 +290,6 @@ void radeonNqssaDce(GLcontext *ctx, struct gl_program *p, struct radeon_nqssadce
 		s.IP--;
 		process_instruction(&s);
 	}
+
+	calculateInputsOutputs(p);
 }
diff --git a/src/mesa/drivers/dri/r300/radeon_nqssadce.h b/src/mesa/drivers/dri/r300/radeon_nqssadce.h
index a4f94abcb62..8626f21c25e 100644
--- a/src/mesa/drivers/dri/r300/radeon_nqssadce.h
+++ b/src/mesa/drivers/dri/r300/radeon_nqssadce.h
@@ -58,6 +58,7 @@ struct nqssadce_state {
 	 */
 	struct register_state Temps[MAX_PROGRAM_TEMPS];
 	struct register_state Outputs[VERT_RESULT_MAX];
+	struct register_state Address;
 };
 
 
@@ -83,14 +84,10 @@ struct radeon_nqssadce_descr {
 	 */
 	void (*BuildSwizzle)(struct nqssadce_state*, struct prog_dst_register dst, struct prog_src_register src);
 
-	/**
-	 * Rewrite instructions that write to DEPR.z to write to DEPR.w
-	 * instead (rewriting is done *before* the WriteMask test).
-	 */
-	GLboolean RewriteDepthOut;
 	void *Data;
 };
 
 void radeonNqssaDce(GLcontext *ctx, struct gl_program *p, struct radeon_nqssadce_descr* descr);
+struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register srcreg);
 
 #endif /* __RADEON_PROGRAM_NQSSADCE_H_ */
diff --git a/src/mesa/drivers/dri/r300/radeon_program.h b/src/mesa/drivers/dri/r300/radeon_program.h
index b411f69bc88..88474d43a22 100644
--- a/src/mesa/drivers/dri/r300/radeon_program.h
+++ b/src/mesa/drivers/dri/r300/radeon_program.h
@@ -52,6 +52,38 @@ enum {
 #define SWIZZLE_0000 MAKE_SWIZZLE4(SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO)
 #define SWIZZLE_1111 MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE)
 
+static inline GLuint get_swz(GLuint swz, GLuint idx)
+{
+	if (idx & 0x4)
+		return idx;
+	return GET_SWZ(swz, idx);
+}
+
+static inline GLuint combine_swizzles4(GLuint src, GLuint swz_x, GLuint swz_y, GLuint swz_z, GLuint swz_w)
+{
+	GLuint ret = 0;
+
+	ret |= get_swz(src, swz_x);
+	ret |= get_swz(src, swz_y) << 3;
+	ret |= get_swz(src, swz_z) << 6;
+	ret |= get_swz(src, swz_w) << 9;
+
+	return ret;
+}
+
+static inline GLuint combine_swizzles(GLuint src, GLuint swz)
+{
+	GLuint ret = 0;
+
+	ret |= get_swz(src, GET_SWZ(swz, SWIZZLE_X));
+	ret |= get_swz(src, GET_SWZ(swz, SWIZZLE_Y)) << 3;
+	ret |= get_swz(src, GET_SWZ(swz, SWIZZLE_Z)) << 6;
+	ret |= get_swz(src, GET_SWZ(swz, SWIZZLE_W)) << 9;
+
+	return ret;
+}
+
+
 /**
  * Transformation context that is passed to local transformations.
  *
diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/radeon_program_pair.c
index 906d36e5226..d6fb474cf23 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_pair.c
+++ b/src/mesa/drivers/dri/r300/radeon_program_pair.c
@@ -870,7 +870,7 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
 
 	_mesa_bzero(&s, sizeof(s));
 	s.Ctx = ctx;
-	s.Program = program;
+	s.Program = _mesa_clone_program(ctx, program);
 	s.Handler = handler;
 	s.UserData = userdata;
 	s.Debug = (RADEON_DEBUG & DEBUG_PIXEL) ? GL_TRUE : GL_FALSE;
@@ -904,6 +904,8 @@ GLboolean radeonPairProgram(GLcontext *ctx, struct gl_program *program,
 	_mesa_free(s.ValuePool);
 	_mesa_free(s.ReaderPool);
 
+	_mesa_reference_program(ctx, &s.Program, NULL);
+
 	return !s.Error;
 }