r300_fragprog: Use nqssa+dce and program_pair for emit

Share almost all code with r500_fragprog now.

This also fixes Piglit's texrect-many test, which means that the compiz
bicubic plugin should work with hardware acceleration now.

11 files changed, 521 insertions, 1688 deletions

diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile
index 9baa1e71312..6ca934204f3 100644
--- a/src/mesa/drivers/dri/r300/Makefile
+++ b/src/mesa/drivers/dri/r300/Makefile
@@ -42,6 +42,7 @@ DRIVER_SOURCES = \
 		 radeon_nqssadce.c \
 		 r300_vertprog.c \
 		 r300_fragprog.c \
+		 r300_fragprog_swizzle.c \
 		 r300_fragprog_emit.c \
 		 r500_fragprog.c \
 		 r500_fragprog_emit.c \
diff --git a/src/mesa/drivers/dri/r300/r300_context.h b/src/mesa/drivers/dri/r300/r300_context.h
index 8e9c5cee5f3..98af6d8f10a 100644
--- a/src/mesa/drivers/dri/r300/r300_context.h
+++ b/src/mesa/drivers/dri/r300/r300_context.h
@@ -683,16 +683,25 @@ struct r300_fragment_program_external_state {
 };
 
 
+struct r300_fragment_program_node {
+	int tex_offset; /**< first tex instruction */
+	int tex_end; /**< last tex instruction, relative to tex_offset */
+	int alu_offset; /**< first ALU instruction */
+	int alu_end; /**< last ALU instruction, relative to alu_offset */
+	int flags;
+};
+
 /**
  * Stores an R300 fragment program in its compiled-to-hardware form.
  */
 struct r300_fragment_program_code {
 	struct {
-		int length;
+		int length; /**< total # of texture instructions used */
 		GLuint inst[PFS_MAX_TEX_INST];
 	} tex;
 
 	struct {
+		int length; /**< total # of ALU instructions used */
 		struct {
 			GLuint inst0;
 			GLuint inst1;
@@ -701,21 +710,10 @@ struct r300_fragment_program_code {
 		} inst[PFS_MAX_ALU_INST];
 	} alu;
 
-	struct {
-		int tex_offset;
-		int tex_end;
-		int alu_offset;
-		int alu_end;
-		int flags;
-	} node[4];
+	struct r300_fragment_program_node node[4];
 	int cur_node;
 	int first_node_has_tex;
 
-	int alu_offset;
-	int alu_end;
-	int tex_offset;
-	int tex_end;
-
 	/**
 	 * Remember which program register a given hardware constant
 	 * belongs to.
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index 8a1d690ae4e..d390de54b8a 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -29,10 +29,8 @@
  * \file
  *
  * Fragment program compiler. Perform transformations on the intermediate
- * \ref radeon_program representation (which is essentially the Mesa
- * program representation plus the notion of clauses) until the program
- * is in a form where we can translate it more or less directly into
- * machine-readable form.
+ * representation until the program is in a form where we can translate
+ * it more or less directly into machine-readable form.
  *
  * \author Ben Skeggs <[email protected]>
  * \author Jerome Glisse <[email protected]>
@@ -47,8 +45,10 @@
 
 #include "r300_context.h"
 #include "r300_fragprog.h"
+#include "r300_fragprog_swizzle.h"
 #include "r300_state.h"
 
+#include "radeon_nqssadce.h"
 #include "radeon_program_alu.h"
 
 
@@ -133,25 +133,6 @@ static GLboolean transform_TEX(
 		inst.SrcReg[0].Index = tempreg;
 	}
 
-	/* Texture operations do not support swizzles etc. in hardware,
-	 * so emit an additional arithmetic operation if necessary.
-	 */
-	if (inst.SrcReg[0].Swizzle != SWIZZLE_NOOP ||
-	    inst.SrcReg[0].Abs || inst.SrcReg[0].NegateBase || inst.SrcReg[0].NegateAbs) {
-		int tempreg = radeonFindFreeTemporary(t);
-
-		tgt = radeonAppendInstructions(t->Program, 1);
-
-		tgt->Opcode = OPCODE_MOV;
-		tgt->DstReg.File = PROGRAM_TEMPORARY;
-		tgt->DstReg.Index = tempreg;
-		tgt->SrcReg[0] = inst.SrcReg[0];
-
-		reset_srcreg(&inst.SrcReg[0]);
-		inst.SrcReg[0].File = PROGRAM_TEMPORARY;
-		inst.SrcReg[0].Index = tempreg;
-	}
-
 	if (inst.Opcode != OPCODE_KIL) {
 		if (inst.DstReg.File != PROGRAM_TEMPORARY ||
 		    inst.DstReg.WriteMask != WRITEMASK_XYZW) {
@@ -339,6 +320,13 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler)
 }
 
 
+static void nqssadce_init(struct nqssadce_state* s)
+{
+	s->Outputs[FRAG_RESULT_COLR].Sourced = WRITEMASK_XYZW;
+	s->Outputs[FRAG_RESULT_DEPR].Sourced = WRITEMASK_W;
+}
+
+
 static GLuint build_dtm(GLuint depthmode)
 {
 	switch(depthmode) {
@@ -417,7 +405,20 @@ void r300TranslateFragmentShader(r300ContextPtr r300,
 			3, transformations);
 
 		if (RADEON_DEBUG & DEBUG_PIXEL) {
-			_mesa_printf("Fragment Program: After transformations:\n");
+			_mesa_printf("Fragment Program: After native rewrite:\n");
+			_mesa_print_program(compiler.program);
+		}
+
+		struct radeon_nqssadce_descr nqssadce = {
+			.Init = &nqssadce_init,
+			.IsNativeSwizzle = &r300FPIsNativeSwizzle,
+			.BuildSwizzle = &r300FPBuildSwizzle,
+			.RewriteDepthOut = GL_TRUE
+		};
+		radeonNqssaDce(r300->radeon.glCtx, compiler.program, &nqssadce);
+
+		if (RADEON_DEBUG & DEBUG_PIXEL) {
+			_mesa_printf("Compiler: after NqSSA-DCE:\n");
 			_mesa_print_program(compiler.program);
 		}
 
@@ -451,22 +452,18 @@ void r300FragmentProgramDump(
 
 	fprintf(stderr, "pc=%d*************************************\n", pc++);
 
-	fprintf(stderr, "Mesa program:\n");
-	fprintf(stderr, "-------------\n");
-	_mesa_print_program(&fp->mesa_program.Base);
-	fflush(stdout);
-
 	fprintf(stderr, "Hardware program\n");
 	fprintf(stderr, "----------------\n");
 
 	for (n = 0; n < (code->cur_node + 1); n++) {
 		fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, "
-			"alu_end: %d, tex_end: %d\n", n,
+			"alu_end: %d, tex_end: %d, flags: %08x\n", n,
 			code->node[n].alu_offset,
 			code->node[n].tex_offset,
-			code->node[n].alu_end, code->node[n].tex_end);
+			code->node[n].alu_end, code->node[n].tex_end,
+			code->node[n].flags);
 
-		if (code->tex.length) {
+		if (n > 0 || code->first_node_has_tex) {
 			fprintf(stderr, "  TEX:\n");
 			for (i = code->node[n].tex_offset;
 			     i <= code->node[n].tex_offset + code->node[n].tex_end;
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.h b/src/mesa/drivers/dri/r300/r300_fragprog.h
index c76ae627015..b3a3cd2e04c 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.h
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.h
@@ -42,33 +42,6 @@
 #include "r300_context.h"
 #include "radeon_program.h"
 
-/* supported hw opcodes */
-#define PFS_OP_MAD 0
-#define PFS_OP_DP3 1
-#define PFS_OP_DP4 2
-#define PFS_OP_MIN 3
-#define PFS_OP_MAX 4
-#define PFS_OP_CMP 5
-#define PFS_OP_FRC 6
-#define PFS_OP_EX2 7
-#define PFS_OP_LG2 8
-#define PFS_OP_RCP 9
-#define PFS_OP_RSQ 10
-#define PFS_OP_REPL_ALPHA 11
-#define PFS_OP_CMPH 12
-#define MAX_PFS_OP 12
-
-#define PFS_FLAG_SAT	(1 << 0)
-#define PFS_FLAG_ABS	(1 << 1)
-
-#define ARG_NEG			(1 << 5)
-#define ARG_ABS			(1 << 6)
-#define ARG_MASK		(127 << 0)
-#define ARG_STRIDE		7
-#define SRC_CONST		(1 << 5)
-#define SRC_MASK		(63 << 0)
-#define SRC_STRIDE		6
-
 #define DRI_CONF_FP_OPTIMIZATION_SPEED   0
 #define DRI_CONF_FP_OPTIMIZATION_QUALITY 1
 
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
index 4786b4554df..9f0b7e35349 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_emit.c
@@ -36,1674 +36,308 @@
  * \author Jerome Glisse <[email protected]>
  *
  * \todo FogOption
- *
- * \todo Verify results of opcodes for accuracy, I've only checked them in
- * specific cases.
  */
 
-#include "glheader.h"
-#include "macros.h"
-#include "enums.h"
-#include "shader/prog_instruction.h"
-#include "shader/prog_parameter.h"
-#include "shader/prog_print.h"
-
-#include "r300_context.h"
 #include "r300_fragprog.h"
-#include "r300_reg.h"
-#include "r300_state.h"
-
-/* Mapping Mesa registers to R300 temporaries */
-struct reg_acc {
-	int reg;		/* Assigned hw temp */
-	unsigned int refcount;	/* Number of uses by mesa program */
-};
-
-/**
- * Describe the current lifetime information for an R300 temporary
- */
-struct reg_lifetime {
-	/* Index of the first slot where this register is free in the sense
-	   that it can be used as a new destination register.
-	   This is -1 if the register has been assigned to a Mesa register
-	   and the last access to the register has not yet been emitted */
-	int free;
-
-	/* Index of the first slot where this register is currently reserved.
-	   This is used to stop e.g. a scalar operation from being moved
-	   before the allocation time of a register that was first allocated
-	   for a vector operation. */
-	int reserved;
-
-	/* Index of the first slot in which the register can be used as a
-	   source without losing the value that is written by the last
-	   emitted instruction that writes to the register */
-	int vector_valid;
-	int scalar_valid;
-
-	/* Index to the slot where the register was last read.
-	   This is also the first slot in which the register may be written again */
-	int vector_lastread;
-	int scalar_lastread;
-};
-
-/**
- * Store usage information about an ALU instruction slot during the
- * compilation of a fragment program.
- */
-#define SLOT_SRC_VECTOR  (1<<0)
-#define SLOT_SRC_SCALAR  (1<<3)
-#define SLOT_SRC_BOTH    (SLOT_SRC_VECTOR | SLOT_SRC_SCALAR)
-#define SLOT_OP_VECTOR   (1<<16)
-#define SLOT_OP_SCALAR   (1<<17)
-#define SLOT_OP_BOTH     (SLOT_OP_VECTOR | SLOT_OP_SCALAR)
-
-struct r300_pfs_compile_slot {
-	/* Bitmask indicating which parts of the slot are used, using SLOT_ constants
-	   defined above */
-	unsigned int used;
-
-	/* Selected sources */
-	int vsrc[3];
-	int ssrc[3];
-};
-
-/**
- * Store information during compilation of fragment programs.
- */
-struct r300_pfs_compile_state {
-	struct r300_fragment_program_compiler *compiler;
 
-	int nrslots;		/* number of ALU slots used so far */
-
-	/* Track which (parts of) slots are already filled with instructions */
-	struct r300_pfs_compile_slot slot[PFS_MAX_ALU_INST];
-
-	/* Track the validity of R300 temporaries */
-	struct reg_lifetime hwtemps[PFS_NUM_TEMP_REGS];
-
-	/* Used to map Mesa's inputs/temps onto hardware temps */
-	int temp_in_use;
-	struct reg_acc temps[PFS_NUM_TEMP_REGS];
-	struct reg_acc inputs[32];	/* don't actually need 32... */
+#include "radeon_program_pair.h"
+#include "r300_fragprog_swizzle.h"
+#include "r300_reg.h"
 
-	/* Track usage of hardware temps, for register allocation,
-	 * indirection detection, etc. */
-	GLuint used_in_node;
-	GLuint dest_in_node;
-};
 
+#define PROG_CODE \
+	struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \
+	struct r300_fragment_program_code *code = c->code
 
-/*
- * Usefull macros and values
- */
-#define ERROR(fmt, args...) do {			\
+#define error(fmt, args...) do {			\
 		fprintf(stderr, "%s::%s(): " fmt "\n",	\
 			__FILE__, __FUNCTION__, ##args);	\
-		fp->error = GL_TRUE;			\
 	} while(0)
 
-#define PFS_INVAL 0xFFFFFFFF
-#define COMPILE_STATE \
-	struct r300_fragment_program *fp = cs->compiler->fp; \
-	struct r300_fragment_program_code *code = cs->compiler->code; \
-	(void)code; (void)fp
-
-#define SWIZZLE_XYZ		0
-#define SWIZZLE_XXX		1
-#define SWIZZLE_YYY		2
-#define SWIZZLE_ZZZ		3
-#define SWIZZLE_WWW		4
-#define SWIZZLE_YZX		5
-#define SWIZZLE_ZXY		6
-#define SWIZZLE_WZY		7
-#define SWIZZLE_111		8
-#define SWIZZLE_000		9
-#define SWIZZLE_HHH		10
-
-#define swizzle(r, x, y, z, w) do_swizzle(cs, r,		\
-					  ((SWIZZLE_##x<<0)|	\
-					   (SWIZZLE_##y<<3)|	\
-					   (SWIZZLE_##z<<6)|	\
-					   (SWIZZLE_##w<<9)),	\
-					  0)
-
-#define REG_TYPE_INPUT		0
-#define REG_TYPE_OUTPUT		1
-#define REG_TYPE_TEMP		2
-#define REG_TYPE_CONST		3
-
-#define REG_TYPE_SHIFT		0
-#define REG_INDEX_SHIFT		2
-#define REG_VSWZ_SHIFT		8
-#define REG_SSWZ_SHIFT		13
-#define REG_NEGV_SHIFT		18
-#define REG_NEGS_SHIFT		19
-#define REG_ABS_SHIFT		20
-#define REG_NO_USE_SHIFT	21	// Hack for refcounting
-#define REG_VALID_SHIFT		22	// Does the register contain a defined value?
-#define REG_BUILTIN_SHIFT   23	// Is it a builtin (like all zero/all one)?
-
-#define REG_TYPE_MASK		(0x03 << REG_TYPE_SHIFT)
-#define REG_INDEX_MASK		(0x3F << REG_INDEX_SHIFT)
-#define REG_VSWZ_MASK		(0x1F << REG_VSWZ_SHIFT)
-#define REG_SSWZ_MASK		(0x1F << REG_SSWZ_SHIFT)
-#define REG_NEGV_MASK		(0x01 << REG_NEGV_SHIFT)
-#define REG_NEGS_MASK		(0x01 << REG_NEGS_SHIFT)
-#define REG_ABS_MASK		(0x01 << REG_ABS_SHIFT)
-#define REG_NO_USE_MASK		(0x01 << REG_NO_USE_SHIFT)
-#define REG_VALID_MASK		(0x01 << REG_VALID_SHIFT)
-#define REG_BUILTIN_MASK	(0x01 << REG_BUILTIN_SHIFT)
-
-#define REG(type, index, vswz, sswz, nouse, valid, builtin)	\
-	(((type << REG_TYPE_SHIFT) & REG_TYPE_MASK) |			\
-	 ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK) |		\
-	 ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK) |		\
-	 ((valid << REG_VALID_SHIFT) & REG_VALID_MASK) |		\
-	 ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK) |	\
-	 ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK) |			\
-	 ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
-#define REG_GET_TYPE(reg)						\
-	((reg & REG_TYPE_MASK) >> REG_TYPE_SHIFT)
-#define REG_GET_INDEX(reg)						\
-	((reg & REG_INDEX_MASK) >> REG_INDEX_SHIFT)
-#define REG_GET_VSWZ(reg)						\
-	((reg & REG_VSWZ_MASK) >> REG_VSWZ_SHIFT)
-#define REG_GET_SSWZ(reg)						\
-	((reg & REG_SSWZ_MASK) >> REG_SSWZ_SHIFT)
-#define REG_GET_NO_USE(reg)						\
-	((reg & REG_NO_USE_MASK) >> REG_NO_USE_SHIFT)
-#define REG_GET_VALID(reg)						\
-	((reg & REG_VALID_MASK) >> REG_VALID_SHIFT)
-#define REG_GET_BUILTIN(reg)						\
-	((reg & REG_BUILTIN_MASK) >> REG_BUILTIN_SHIFT)
-#define REG_SET_TYPE(reg, type)						\
-	reg = ((reg & ~REG_TYPE_MASK) |					\
-	       ((type << REG_TYPE_SHIFT) & REG_TYPE_MASK))
-#define REG_SET_INDEX(reg, index)					\
-	reg = ((reg & ~REG_INDEX_MASK) |				\
-	       ((index << REG_INDEX_SHIFT) & REG_INDEX_MASK))
-#define REG_SET_VSWZ(reg, vswz)						\
-	reg = ((reg & ~REG_VSWZ_MASK) |					\
-	       ((vswz << REG_VSWZ_SHIFT) & REG_VSWZ_MASK))
-#define REG_SET_SSWZ(reg, sswz)						\
-	reg = ((reg & ~REG_SSWZ_MASK) |					\
-	       ((sswz << REG_SSWZ_SHIFT) & REG_SSWZ_MASK))
-#define REG_SET_NO_USE(reg, nouse)					\
-	reg = ((reg & ~REG_NO_USE_MASK) |				\
-	       ((nouse << REG_NO_USE_SHIFT) & REG_NO_USE_MASK))
-#define REG_SET_VALID(reg, valid)					\
-	reg = ((reg & ~REG_VALID_MASK) |				\
-	       ((valid << REG_VALID_SHIFT) & REG_VALID_MASK))
-#define REG_SET_BUILTIN(reg, builtin)					\
-	reg = ((reg & ~REG_BUILTIN_MASK) |				\
-	       ((builtin << REG_BUILTIN_SHIFT) & REG_BUILTIN_MASK))
-#define REG_ABS(reg)							\
-	reg = (reg | REG_ABS_MASK)
-#define REG_NEGV(reg)							\
-	reg = (reg | REG_NEGV_MASK)
-#define REG_NEGS(reg)							\
-	reg = (reg | REG_NEGS_MASK)
-
-#define NOP_INST0 (						 \
-		(R300_ALU_OUTC_MAD) |				 \
-		(R300_ALU_ARGC_ZERO << R300_ALU_ARG0C_SHIFT) | \
-		(R300_ALU_ARGC_ZERO << R300_ALU_ARG1C_SHIFT) | \
-		(R300_ALU_ARGC_ZERO << R300_ALU_ARG2C_SHIFT))
-#define NOP_INST1 (					     \
-		((0 | SRC_CONST) << R300_ALU_SRC0C_SHIFT) | \
-		((0 | SRC_CONST) << R300_ALU_SRC1C_SHIFT) | \
-		((0 | SRC_CONST) << R300_ALU_SRC2C_SHIFT))
-#define NOP_INST2 ( \
-		(R300_ALU_OUTA_MAD) |				 \
-		(R300_ALU_ARGA_ZERO << R300_ALU_ARG0A_SHIFT) | \
-		(R300_ALU_ARGA_ZERO << R300_ALU_ARG1A_SHIFT) | \
-		(R300_ALU_ARGA_ZERO << R300_ALU_ARG2A_SHIFT))
-#define NOP_INST3 (					     \
-		((0 | SRC_CONST) << R300_ALU_SRC0A_SHIFT) | \
-		((0 | SRC_CONST) << R300_ALU_SRC1A_SHIFT) | \
-		((0 | SRC_CONST) << R300_ALU_SRC2A_SHIFT))
-
-
-/*
- * Datas structures for fragment program generation
- */
-
-/* description of r300 native hw instructions */
-static const struct {
-	const char *name;
-	int argc;
-	int v_op;
-	int s_op;
-} r300_fpop[] = {
-	/* *INDENT-OFF* */
-	{"MAD", 3, R300_ALU_OUTC_MAD, R300_ALU_OUTA_MAD},
-	{"DP3", 2, R300_ALU_OUTC_DP3, R300_ALU_OUTA_DP4},
-	{"DP4", 2, R300_ALU_OUTC_DP4, R300_ALU_OUTA_DP4},
-	{"MIN", 2, R300_ALU_OUTC_MIN, R300_ALU_OUTA_MIN},
-	{"MAX", 2, R300_ALU_OUTC_MAX, R300_ALU_OUTA_MAX},
-	{"CMP", 3, R300_ALU_OUTC_CMP, R300_ALU_OUTA_CMP},
-	{"FRC", 1, R300_ALU_OUTC_FRC, R300_ALU_OUTA_FRC},
-	{"EX2", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_EX2},
-	{"LG2", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_LG2},
-	{"RCP", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_RCP},
-	{"RSQ", 1, R300_ALU_OUTC_REPL_ALPHA, R300_ALU_OUTA_RSQ},
-	{"REPL_ALPHA", 1, R300_ALU_OUTC_REPL_ALPHA, PFS_INVAL},
-	{"CMPH", 3, R300_ALU_OUTC_CMPH, PFS_INVAL},
-	/* *INDENT-ON* */
-};
-
-/* vector swizzles r300 can support natively, with a couple of
- * cases we handle specially
- *
- * REG_VSWZ/REG_SSWZ is an index into this table
- */
-
-/* mapping from SWIZZLE_* to r300 native values for scalar insns */
-#define SWIZZLE_HALF 6
-
-#define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, \
-					  SWIZZLE_##y, \
-					  SWIZZLE_##z, \
-					  SWIZZLE_ZERO))
-/* native swizzles */
-static const struct r300_pfs_swizzle {
-	GLuint hash;		/* swizzle value this matches */
-	GLuint base;		/* base value for hw swizzle */
-	GLuint stride;		/* difference in base between arg0/1/2 */
-	GLuint flags;
-} v_swiz[] = {
-	/* *INDENT-OFF* */
-	{MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1, SLOT_SRC_SCALAR},
-	{MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1, SLOT_SRC_VECTOR},
-	{MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1, SLOT_SRC_BOTH},
-	{MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0, 0},
-	{MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0, 0},
-	{MAKE_SWZ3(HALF, HALF, HALF), R300_ALU_ARGC_HALF, 0, 0},
-	{PFS_INVAL, 0, 0, 0},
-	/* *INDENT-ON* */
-};
-
-/* used during matching of non-native swizzles */
-#define SWZ_X_MASK (7 << 0)
-#define SWZ_Y_MASK (7 << 3)
-#define SWZ_Z_MASK (7 << 6)
-#define SWZ_W_MASK (7 << 9)
-static const struct {
-	GLuint hash;		/* used to mask matching swizzle components */
-	int mask;		/* actual outmask */
-	int count;		/* count of components matched */
-} s_mask[] = {
-	/* *INDENT-OFF* */
-	{SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK, 1 | 2 | 4, 3},
-	{SWZ_X_MASK | SWZ_Y_MASK, 1 | 2, 2},
-	{SWZ_X_MASK | SWZ_Z_MASK, 1 | 4, 2},
-	{SWZ_Y_MASK | SWZ_Z_MASK, 2 | 4, 2},
-	{SWZ_X_MASK, 1, 1},
-	{SWZ_Y_MASK, 2, 1},
-	{SWZ_Z_MASK, 4, 1},
-	{PFS_INVAL, PFS_INVAL, PFS_INVAL}
-	/* *INDENT-ON* */
-};
-
-static const struct {
-	int base;		/* hw value of swizzle */
-	int stride;		/* difference between SRC0/1/2 */
-	GLuint flags;
-} s_swiz[] = {
-	/* *INDENT-OFF* */
-	{R300_ALU_ARGA_SRC0C_X, 3, SLOT_SRC_VECTOR},
-	{R300_ALU_ARGA_SRC0C_Y, 3, SLOT_SRC_VECTOR},
-	{R300_ALU_ARGA_SRC0C_Z, 3, SLOT_SRC_VECTOR},
-	{R300_ALU_ARGA_SRC0A, 1, SLOT_SRC_SCALAR},
-	{R300_ALU_ARGA_ZERO, 0, 0},
-	{R300_ALU_ARGA_ONE, 0, 0},
-	{R300_ALU_ARGA_HALF, 0, 0}
-	/* *INDENT-ON* */
-};
-
-/* boiler-plate reg, for convenience */
-static const GLuint undef = REG(REG_TYPE_TEMP,
-				0,
-				SWIZZLE_XYZ,
-				SWIZZLE_W,
-				GL_FALSE,
-				GL_FALSE,
-				GL_FALSE);
-
-/* constant one source */
-static const GLuint pfs_one = REG(REG_TYPE_CONST,
-				  0,
-				  SWIZZLE_111,
-				  SWIZZLE_ONE,
-				  GL_FALSE,
-				  GL_TRUE,
-				  GL_TRUE);
-
-/* constant half source */
-static const GLuint pfs_half = REG(REG_TYPE_CONST,
-				   0,
-				   SWIZZLE_HHH,
-				   SWIZZLE_HALF,
-				   GL_FALSE,
-				   GL_TRUE,
-				   GL_TRUE);
-
-/* constant zero source */
-static const GLuint pfs_zero = REG(REG_TYPE_CONST,
-				   0,
-				   SWIZZLE_000,
-				   SWIZZLE_ZERO,
-				   GL_FALSE,
-				   GL_TRUE,
-				   GL_TRUE);
-
-/*
- * Common functions prototypes
- */
-static void emit_arith(struct r300_pfs_compile_state *cs, int op,
-		       GLuint dest, int mask,
-		       GLuint src0, GLuint src1, GLuint src2, int flags);
-
-/**
- * Get an R300 temporary that can be written to in the given slot.
- */
-static int get_hw_temp(struct r300_pfs_compile_state *cs, int slot)
-{
-	COMPILE_STATE;
-	int r;
-
-	for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
-		if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= slot)
-			break;
-	}
-
-	if (r >= PFS_NUM_TEMP_REGS) {
-		ERROR("Out of hardware temps\n");
-		return 0;
-	}
-	// Reserved is used to avoid the following scenario:
-	//  R300 temporary X is first assigned to Mesa temporary Y during vector ops
-	//  R300 temporary X is then assigned to Mesa temporary Z for further vector ops
-	//  Then scalar ops on Mesa temporary Z are emitted and move back in time
-	//  to overwrite the value of temporary Y.
-	// End scenario.
-	cs->hwtemps[r].reserved = cs->hwtemps[r].free;
-	cs->hwtemps[r].free = -1;
-
-	// Reset to some value that won't mess things up when the user
-	// tries to read from a temporary that hasn't been assigned a value yet.
-	// In the normal case, vector_valid and scalar_valid should be set to
-	// a sane value by the first emit that writes to this temporary.
-	cs->hwtemps[r].vector_valid = 0;
-	cs->hwtemps[r].scalar_valid = 0;
-
-	if (r > code->max_temp_idx)
-		code->max_temp_idx = r;
-
-	return r;
-}
-
-/**
- * Get an R300 temporary that will act as a TEX destination register.
- */
-static int get_hw_temp_tex(struct r300_pfs_compile_state *cs)
-{
-	COMPILE_STATE;
-	int r;
-
-	for (r = 0; r < PFS_NUM_TEMP_REGS; ++r) {
-		if (cs->used_in_node & (1 << r))
-			continue;
-
-		// Note: Be very careful here
-		if (cs->hwtemps[r].free >= 0 && cs->hwtemps[r].free <= 0)
-			break;
-	}
-
-	if (r >= PFS_NUM_TEMP_REGS)
-		return get_hw_temp(cs, 0);	/* Will cause an indirection */
-
-	cs->hwtemps[r].reserved = cs->hwtemps[r].free;
-	cs->hwtemps[r].free = -1;
-
-	// Reset to some value that won't mess things up when the user
-	// tries to read from a temporary that hasn't been assigned a value yet.
-	// In the normal case, vector_valid and scalar_valid should be set to
-	// a sane value by the first emit that writes to this temporary.
-	cs->hwtemps[r].vector_valid = cs->nrslots;
-	cs->hwtemps[r].scalar_valid = cs->nrslots;
 
-	if (r > code->max_temp_idx)
-		code->max_temp_idx = r;
-
-	return r;
-}
-
-/**
- * Mark the given hardware register as free.
- */
-static void free_hw_temp(struct r300_pfs_compile_state *cs, int idx)
+static GLboolean emit_const(void* data, GLuint file, GLuint index, GLuint *hwindex)
 {
-	// Be very careful here. Consider sequences like
-	//  MAD r0, r1,r2,r3
-	//  TEX r4, ...
-	// The TEX instruction may be moved in front of the MAD instruction
-	// due to the way nodes work. We don't want to alias r1 and r4 in
-	// this case.
-	// I'm certain the register allocation could be further sanitized,
-	// but it's tricky because of stuff that can happen inside emit_tex
-	// and emit_arith.
-	cs->hwtemps[idx].free = cs->nrslots + 1;
-}
+	PROG_CODE;
 
-/**
- * Create a new Mesa temporary register.
- */
-static GLuint get_temp_reg(struct r300_pfs_compile_state *cs)
-{
-	COMPILE_STATE;
-	GLuint r = undef;
-	GLuint index;
-
-	index = ffs(~cs->temp_in_use);
-	if (!index) {
-		ERROR("Out of program temps\n");
-		return r;
-	}
-
-	cs->temp_in_use |= (1 << --index);
-	cs->temps[index].refcount = 0xFFFFFFFF;
-	cs->temps[index].reg = -1;
-
-	REG_SET_TYPE(r, REG_TYPE_TEMP);
-	REG_SET_INDEX(r, index);
-	REG_SET_VALID(r, GL_TRUE);
-	return r;
-}
-
-/**
- * Free a Mesa temporary and the associated R300 temporary.
- */
-static void free_temp(struct r300_pfs_compile_state *cs, GLuint r)
-{
-	GLuint index = REG_GET_INDEX(r);
-
-	if (!(cs->temp_in_use & (1 << index)))
-		return;
-
-	if (REG_GET_TYPE(r) == REG_TYPE_TEMP) {
-		free_hw_temp(cs, cs->temps[index].reg);
-		cs->temps[index].reg = -1;
-		cs->temp_in_use &= ~(1 << index);
-	} else if (REG_GET_TYPE(r) == REG_TYPE_INPUT) {
-		free_hw_temp(cs, cs->inputs[index].reg);
-		cs->inputs[index].reg = -1;
-	}
-}
-
-/**
- * Emit a hardware constant/parameter.
- */
-static GLuint emit_const4fv(struct r300_pfs_compile_state *cs,
-			    struct prog_src_register srcreg)
-{
-	COMPILE_STATE;
-	GLuint reg = undef;
-	int index;
-
-	for (index = 0; index < code->const_nr; ++index) {
-		if (code->constant[index].File == srcreg.File &&
-		    code->constant[index].Index == srcreg.Index)
+	for (*hwindex = 0; *hwindex < code->const_nr; ++*hwindex) {
+		if (code->constant[*hwindex].File == file &&
+		    code->constant[*hwindex].Index == index)
 			break;
 	}
 
-	if (index >= code->const_nr) {
-		if (index >= PFS_NUM_CONST_REGS) {
-			ERROR("Out of hw constants!\n");
-			return reg;
+	if (*hwindex >= code->const_nr) {
+		if (*hwindex >= PFS_NUM_CONST_REGS) {
+			error("Out of hw constants!\n");
+			return GL_FALSE;
 		}
 
 		code->const_nr++;
-		code->constant[index] = srcreg;
+		code->constant[*hwindex].File = file;
+		code->constant[*hwindex].Index = index;
 	}
 
-	REG_SET_TYPE(reg, REG_TYPE_CONST);
-	REG_SET_INDEX(reg, index);
-	REG_SET_VALID(reg, GL_TRUE);
-	return reg;
+	return GL_TRUE;
 }
 
-static INLINE GLuint negate(GLuint r)
-{
-	REG_NEGS(r);
-	REG_NEGV(r);
-	return r;
-}
 
-/* Hack, to prevent clobbering sources used multiple times when
- * emulating non-native instructions
+/**
+ * Mark a temporary register as used.
  */
-static INLINE GLuint keep(GLuint r)
-{
-	REG_SET_NO_USE(r, GL_TRUE);
-	return r;
-}
-
-static INLINE GLuint absolute(GLuint r)
-{
-	REG_ABS(r);
-	return r;
-}
-
-static int swz_native(struct r300_pfs_compile_state *cs,
-		      GLuint src, GLuint * r, GLuint arbneg)
-{
-	COMPILE_STATE;
-
-	/* Native swizzle, handle negation */
-	src = (src & ~REG_NEGS_MASK) | (((arbneg >> 3) & 1) << REG_NEGS_SHIFT);
-
-	if ((arbneg & 0x7) == 0x0) {
-		src = src & ~REG_NEGV_MASK;
-		*r = src;
-	} else if ((arbneg & 0x7) == 0x7) {
-		src |= REG_NEGV_MASK;
-		*r = src;
-	} else {
-		if (!REG_GET_VALID(*r))
-			*r = get_temp_reg(cs);
-		src |= REG_NEGV_MASK;
-		emit_arith(cs,
-			   PFS_OP_MAD,
-			   *r, arbneg & 0x7, keep(src), pfs_one, pfs_zero, 0);
-		src = src & ~REG_NEGV_MASK;
-		emit_arith(cs,
-			   PFS_OP_MAD,
-			   *r,
-			   (arbneg ^ 0x7) | WRITEMASK_W,
-			   src, pfs_one, pfs_zero, 0);
-	}
-
-	return 3;
-}
-
-static int swz_emit_partial(struct r300_pfs_compile_state *cs,
-			    GLuint src,
-			    GLuint * r, int mask, int mc, GLuint arbneg)
-{
-	COMPILE_STATE;
-	GLuint tmp;
-	GLuint wmask = 0;
-
-	if (!REG_GET_VALID(*r))
-		*r = get_temp_reg(cs);
-
-	/* A partial match, VSWZ/mask define what parts of the
-	 * desired swizzle we match
-	 */
-	if (mc + s_mask[mask].count == 3) {
-		wmask = WRITEMASK_W;
-		src |= ((arbneg >> 3) & 1) << REG_NEGS_SHIFT;
-	}
-
-	tmp = arbneg & s_mask[mask].mask;
-	if (tmp) {
-		tmp = tmp ^ s_mask[mask].mask;
-		if (tmp) {
-			emit_arith(cs,
-				   PFS_OP_MAD,
-				   *r,
-				   arbneg & s_mask[mask].mask,
-				   keep(src) | REG_NEGV_MASK,
-				   pfs_one, pfs_zero, 0);
-			if (!wmask) {
-				REG_SET_NO_USE(src, GL_TRUE);
-			} else {
-				REG_SET_NO_USE(src, GL_FALSE);
-			}
-			emit_arith(cs,
-				   PFS_OP_MAD,
-				   *r, tmp | wmask, src, pfs_one, pfs_zero, 0);
-		} else {
-			if (!wmask) {
-				REG_SET_NO_USE(src, GL_TRUE);
-			} else {
-				REG_SET_NO_USE(src, GL_FALSE);
-			}
-			emit_arith(cs,
-				   PFS_OP_MAD,
-				   *r,
-				   (arbneg & s_mask[mask].mask) | wmask,
-				   src | REG_NEGV_MASK, pfs_one, pfs_zero, 0);
-		}
-	} else {
-		if (!wmask) {
-			REG_SET_NO_USE(src, GL_TRUE);
-		} else {
-			REG_SET_NO_USE(src, GL_FALSE);
-		}
-		emit_arith(cs, PFS_OP_MAD,
-			   *r,
-			   s_mask[mask].mask | wmask,
-			   src, pfs_one, pfs_zero, 0);
-	}
-
-	return s_mask[mask].count;
-}
-
-static GLuint do_swizzle(struct r300_pfs_compile_state *cs,
-			 GLuint src, GLuint arbswz, GLuint arbneg)
-{
-	COMPILE_STATE;
-	GLuint r = undef;
-	GLuint vswz;
-	int c_mask = 0;
-	int v_match = 0;
-
-	/* If swizzling from something without an XYZW native swizzle,
-	 * emit result to a temp, and do new swizzle from the temp.
-	 */
-#if 0
-	if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
-		GLuint temp = get_temp_reg(fp);
-		emit_arith(fp,
-			   PFS_OP_MAD,
-			   temp, WRITEMASK_XYZW, src, pfs_one, pfs_zero, 0);
-		src = temp;
-	}
-#endif
-
-	if (REG_GET_VSWZ(src) != SWIZZLE_XYZ || REG_GET_SSWZ(src) != SWIZZLE_W) {
-		GLuint vsrcswz =
-		    (v_swiz[REG_GET_VSWZ(src)].
-		     hash & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK)) |
-		    REG_GET_SSWZ(src) << 9;
-		GLint i;
-
-		GLuint newswz = 0;
-		GLuint offset;
-		for (i = 0; i < 4; ++i) {
-			offset = GET_SWZ(arbswz, i);
-
-			newswz |=
-			    (offset <= 3) ? GET_SWZ(vsrcswz,
-						    offset) << i *
-			    3 : offset << i * 3;
-		}
-
-		arbswz = newswz & (SWZ_X_MASK | SWZ_Y_MASK | SWZ_Z_MASK);
-		REG_SET_SSWZ(src, GET_SWZ(newswz, 3));
-	} else {
-		/* set scalar swizzling */
-		REG_SET_SSWZ(src, GET_SWZ(arbswz, 3));
-
-	}
-	do {
-		vswz = REG_GET_VSWZ(src);
-		do {
-			int chash;
-
-			REG_SET_VSWZ(src, vswz);
-			chash = v_swiz[REG_GET_VSWZ(src)].hash &
-			    s_mask[c_mask].hash;
-
-			if (chash == (arbswz & s_mask[c_mask].hash)) {
-				if (s_mask[c_mask].count == 3) {
-					v_match += swz_native(cs,
-							      src, &r, arbneg);
-				} else {
-					v_match += swz_emit_partial(cs,
-								    src,
-								    &r,
-								    c_mask,
-								    v_match,
-								    arbneg);
-				}
-
-				if (v_match == 3)
-					return r;
-
-				/* Fill with something invalid.. all 0's was
-				 * wrong before, matched SWIZZLE_X.  So all
-				 * 1's will be okay for now
-				 */
-				arbswz |= (PFS_INVAL & s_mask[c_mask].hash);
-			}
-		} while (v_swiz[++vswz].hash != PFS_INVAL);
-		REG_SET_VSWZ(src, SWIZZLE_XYZ);
-	} while (s_mask[++c_mask].hash != PFS_INVAL);
-
-	ERROR("should NEVER get here\n");
-	return r;
-}
-
-static GLuint t_src(struct r300_pfs_compile_state *cs,
-		    struct prog_src_register fpsrc)
-{
-	COMPILE_STATE;
-	GLuint r = undef;
-
-	switch (fpsrc.File) {
-	case PROGRAM_TEMPORARY:
-		REG_SET_INDEX(r, fpsrc.Index);
-		REG_SET_VALID(r, GL_TRUE);
-		REG_SET_TYPE(r, REG_TYPE_TEMP);
-		break;
-	case PROGRAM_INPUT:
-		REG_SET_INDEX(r, fpsrc.Index);
-		REG_SET_VALID(r, GL_TRUE);
-		REG_SET_TYPE(r, REG_TYPE_INPUT);
-		break;
-	case PROGRAM_LOCAL_PARAM:
-	case PROGRAM_ENV_PARAM:
-	case PROGRAM_STATE_VAR:
-	case PROGRAM_NAMED_PARAM:
-	case PROGRAM_CONSTANT:
-		r = emit_const4fv(cs, fpsrc);
-		break;
-	case PROGRAM_BUILTIN:
-		switch(fpsrc.Swizzle) {
-		case SWIZZLE_1111: r = pfs_one; break;
-		case SWIZZLE_0000: r = pfs_zero; break;
-		default:
-			ERROR("bad PROGRAM_BUILTIN swizzle %u\n", fpsrc.Swizzle);
-			break;
-		}
-		break;
-	default:
-		ERROR("unknown SrcReg->File %x\n", fpsrc.File);
-		return r;
-	}
-
-	/* no point swizzling ONE/ZERO/HALF constants... */
-	if (REG_GET_VSWZ(r) < SWIZZLE_111 || REG_GET_SSWZ(r) < SWIZZLE_ZERO)
-		r = do_swizzle(cs, r, fpsrc.Swizzle, fpsrc.NegateBase);
-	if (fpsrc.Abs)
-		r = absolute(r);
-	if (fpsrc.NegateAbs)
-		r = negate(r);
-	return r;
-}
-
-static GLuint t_scalar_src(struct r300_pfs_compile_state *cs,
-			   struct prog_src_register fpsrc)
+static void use_temporary(struct r300_fragment_program_code *code, GLuint index)
 {
-	struct prog_src_register src = fpsrc;
-	int sc = GET_SWZ(fpsrc.Swizzle, 0);	/* X */
-
-	src.Swizzle = ((sc << 0) | (sc << 3) | (sc << 6) | (sc << 9));
-
-	return t_src(cs, src);
+	if (index > code->max_temp_idx)
+		code->max_temp_idx = index;
 }
 
-static GLuint t_dst(struct r300_pfs_compile_state *cs,
-		    struct prog_dst_register dest)
-{
-	COMPILE_STATE;
-	GLuint r = undef;
 
-	switch (dest.File) {
-	case PROGRAM_TEMPORARY:
-		REG_SET_INDEX(r, dest.Index);
-		REG_SET_VALID(r, GL_TRUE);
-		REG_SET_TYPE(r, REG_TYPE_TEMP);
-		return r;
-	case PROGRAM_OUTPUT:
-		REG_SET_TYPE(r, REG_TYPE_OUTPUT);
-		switch (dest.Index) {
-		case FRAG_RESULT_COLR:
-		case FRAG_RESULT_DEPR:
-			REG_SET_INDEX(r, dest.Index);
-			REG_SET_VALID(r, GL_TRUE);
-			return r;
-		default:
-			ERROR("Bad DstReg->Index 0x%x\n", dest.Index);
-			return r;
-		}
-	default:
-		ERROR("Bad DstReg->File 0x%x\n", dest.File);
-		return r;
-	}
-}
-
-static int t_hw_src(struct r300_pfs_compile_state *cs, GLuint src, GLboolean tex)
+static GLuint translate_rgb_opcode(GLuint opcode)
 {
-	COMPILE_STATE;
-	int idx;
-	int index = REG_GET_INDEX(src);
-
-	switch (REG_GET_TYPE(src)) {
-	case REG_TYPE_TEMP:
-		/* NOTE: if reg==-1 here, a source is being read that
-		 *       hasn't been written to. Undefined results.
-		 */
-		if (cs->temps[index].reg == -1)
-			cs->temps[index].reg = get_hw_temp(cs, cs->nrslots);
-
-		idx = cs->temps[index].reg;
-
-		if (!REG_GET_NO_USE(src) && (--cs->temps[index].refcount == 0))
-			free_temp(cs, src);
-		break;
-	case REG_TYPE_INPUT:
-		idx = cs->inputs[index].reg;
-
-		if (!REG_GET_NO_USE(src) && (--cs->inputs[index].refcount == 0))
-			free_hw_temp(cs, cs->inputs[index].reg);
-		break;
-	case REG_TYPE_CONST:
-		return (index | SRC_CONST);
+	switch(opcode) {
+	case OPCODE_CMP: return R300_ALU_OUTC_CMP;
+	case OPCODE_DP3: return R300_ALU_OUTC_DP3;
+	case OPCODE_DP4: return R300_ALU_OUTC_DP4;
+	case OPCODE_FRC: return R300_ALU_OUTC_FRC;
 	default:
-		ERROR("Invalid type for source reg\n");
-		return (0 | SRC_CONST);
+		error("translate_rgb_opcode(%i): Unknown opcode", opcode);
+		/* fall through */
+	case OPCODE_NOP:
+		/* fall through */
+	case OPCODE_MAD: return R300_ALU_OUTC_MAD;
+	case OPCODE_MAX: return R300_ALU_OUTC_MAX;
+	case OPCODE_MIN: return R300_ALU_OUTC_MIN;
+	case OPCODE_REPL_ALPHA: return R300_ALU_OUTC_REPL_ALPHA;
 	}
-
-	if (!tex)
-		cs->used_in_node |= (1 << idx);
-
-	return idx;
 }
 
-static int t_hw_dst(struct r300_pfs_compile_state *cs,
-		    GLuint dest, GLboolean tex, int slot)
+static GLuint translate_alpha_opcode(GLuint opcode)
 {
-	COMPILE_STATE;
-	int idx;
-	GLuint index = REG_GET_INDEX(dest);
-	assert(REG_GET_VALID(dest));
-
-	switch (REG_GET_TYPE(dest)) {
-	case REG_TYPE_TEMP:
-		if (cs->temps[REG_GET_INDEX(dest)].reg == -1) {
-			if (!tex) {
-				cs->temps[index].reg = get_hw_temp(cs, slot);
-			} else {
-				cs->temps[index].reg = get_hw_temp_tex(cs);
-			}
-		}
-		idx = cs->temps[index].reg;
-
-		if (!REG_GET_NO_USE(dest) && (--cs->temps[index].refcount == 0))
-			free_temp(cs, dest);
-
-		cs->dest_in_node |= (1 << idx);
-		cs->used_in_node |= (1 << idx);
-		break;
-	case REG_TYPE_OUTPUT:
-		switch (index) {
-		case FRAG_RESULT_COLR:
-			code->node[code->cur_node].flags |= R300_RGBA_OUT;
-			break;
-		case FRAG_RESULT_DEPR:
-			fp->WritesDepth = GL_TRUE;
-			code->node[code->cur_node].flags |= R300_W_OUT;
-			break;
-		}
-		return index;
-		break;
+	switch(opcode) {
+	case OPCODE_CMP: return R300_ALU_OUTA_CMP;
+	case OPCODE_DP3: return R300_ALU_OUTA_DP4;
+	case OPCODE_DP4: return R300_ALU_OUTA_DP4;
+	case OPCODE_EX2: return R300_ALU_OUTA_EX2;
+	case OPCODE_FRC: return R300_ALU_OUTA_FRC;
+	case OPCODE_LG2: return R300_ALU_OUTA_LG2;
 	default:
-		ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
-		return 0;
-	}
-
-	return idx;
-}
-
-static void emit_nop(struct r300_pfs_compile_state *cs)
-{
-	COMPILE_STATE;
-
-	if (cs->nrslots >= PFS_MAX_ALU_INST) {
-		ERROR("Out of ALU instruction slots\n");
-		return;
-	}
-
-	code->alu.inst[cs->nrslots].inst0 = NOP_INST0;
-	code->alu.inst[cs->nrslots].inst1 = NOP_INST1;
-	code->alu.inst[cs->nrslots].inst2 = NOP_INST2;
-	code->alu.inst[cs->nrslots].inst3 = NOP_INST3;
-	cs->nrslots++;
-}
-
-static void emit_tex(struct r300_pfs_compile_state *cs,
-		     struct prog_instruction *fpi, int opcode)
-{
-	COMPILE_STATE;
-	GLuint coord = t_src(cs, fpi->SrcReg[0]);
-	GLuint dest = undef;
-	GLuint din, uin;
-	int unit = fpi->TexSrcUnit;
-	int hwsrc, hwdest;
-
-	/* Ensure correct node indirection */
-	uin = cs->used_in_node;
-	din = cs->dest_in_node;
-
-	/* Resolve source/dest to hardware registers */
-	hwsrc = t_hw_src(cs, coord, GL_TRUE);
-
-	if (opcode != R300_TEX_OP_KIL) {
-		dest = t_dst(cs, fpi->DstReg);
-
-		hwdest =
-		    t_hw_dst(cs, dest, GL_TRUE,
-			     code->node[code->cur_node].alu_offset);
-
-		/* Use a temp that hasn't been used in this node, rather
-		 * than causing an indirection
-		 */
-		if (uin & (1 << hwdest)) {
-			free_hw_temp(cs, hwdest);
-			hwdest = get_hw_temp_tex(cs);
-			cs->temps[REG_GET_INDEX(dest)].reg = hwdest;
-		}
-	} else {
-		hwdest = 0;
-		unit = 0;
-	}
-
-	/* Indirection if source has been written in this node, or if the
-	 * dest has been read/written in this node
-	 */
-	if ((REG_GET_TYPE(coord) != REG_TYPE_CONST &&
-	     (din & (1 << hwsrc))) || (uin & (1 << hwdest))) {
-
-		/* Finish off current node */
-		if (code->node[code->cur_node].alu_offset == cs->nrslots)
-			emit_nop(cs);
-
-		code->node[code->cur_node].alu_end =
-		    cs->nrslots - code->node[code->cur_node].alu_offset - 1;
-		assert(code->node[code->cur_node].alu_end >= 0);
-
-		if (++code->cur_node >= PFS_MAX_TEX_INDIRECT) {
-			ERROR("too many levels of texture indirection\n");
-			return;
-		}
-
-		/* Start new node */
-		code->node[code->cur_node].tex_offset = code->tex.length;
-		code->node[code->cur_node].alu_offset = cs->nrslots;
-		code->node[code->cur_node].tex_end = -1;
-		code->node[code->cur_node].alu_end = -1;
-		code->node[code->cur_node].flags = 0;
-		cs->used_in_node = 0;
-		cs->dest_in_node = 0;
+		error("translate_rgb_opcode(%i): Unknown opcode", opcode);
+		/* fall through */
+	case OPCODE_NOP:
+		/* fall through */
+	case OPCODE_MAD: return R300_ALU_OUTA_MAD;
+	case OPCODE_MAX: return R300_ALU_OUTA_MAX;
+	case OPCODE_MIN: return R300_ALU_OUTA_MIN;
+	case OPCODE_RCP: return R300_ALU_OUTA_RCP;
+	case OPCODE_RSQ: return R300_ALU_OUTA_RSQ;
 	}
-
-	if (code->cur_node == 0)
-		code->first_node_has_tex = 1;
-
-	code->tex.inst[code->tex.length++] = 0 | (hwsrc << R300_SRC_ADDR_SHIFT)
-	    | (hwdest << R300_DST_ADDR_SHIFT)
-	    | (unit << R300_TEX_ID_SHIFT)
-	    | (opcode << R300_TEX_INST_SHIFT);
-
-	cs->dest_in_node |= (1 << hwdest);
-	if (REG_GET_TYPE(coord) != REG_TYPE_CONST)
-		cs->used_in_node |= (1 << hwsrc);
-
-	code->node[code->cur_node].tex_end++;
 }
 
 /**
- * Returns the first slot where we could possibly allow writing to dest,
- * according to register allocation.
+ * Emit one paired ALU instruction.
  */
-static int get_earliest_allowed_write(struct r300_pfs_compile_state *cs,
-				      GLuint dest, int mask)
+static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
 {
-	COMPILE_STATE;
-	int idx;
-	int pos;
-	GLuint index = REG_GET_INDEX(dest);
-	assert(REG_GET_VALID(dest));
+	PROG_CODE;
 
-	switch (REG_GET_TYPE(dest)) {
-	case REG_TYPE_TEMP:
-		if (cs->temps[index].reg == -1)
-			return 0;
-
-		idx = cs->temps[index].reg;
-		break;
-	case REG_TYPE_OUTPUT:
-		return 0;
-	default:
-		ERROR("invalid dest reg type %d\n", REG_GET_TYPE(dest));
-		return 0;
-	}
-
-	pos = cs->hwtemps[idx].reserved;
-	if (mask & WRITEMASK_XYZ) {
-		if (pos < cs->hwtemps[idx].vector_lastread)
-			pos = cs->hwtemps[idx].vector_lastread;
-	}
-	if (mask & WRITEMASK_W) {
-		if (pos < cs->hwtemps[idx].scalar_lastread)
-			pos = cs->hwtemps[idx].scalar_lastread;
+	if (code->alu.length >= PFS_MAX_ALU_INST) {
+		error("Too many ALU instructions");
+		return GL_FALSE;
 	}
 
-	return pos;
-}
-
-/**
- * Allocates a slot for an ALU instruction that can consist of
- * a vertex part or a scalar part or both.
- *
- * Sources from src (src[0] to src[argc-1]) are added to the slot in the
- * appropriate position (vector and/or scalar), and their positions are
- * recorded in the srcpos array.
- *
- * This function emits instruction code for the source fetch and the
- * argument selection. It does not emit instruction code for the
- * opcode or the destination selection.
- *
- * @return the index of the slot
- */
-static int find_and_prepare_slot(struct r300_pfs_compile_state *cs,
-				 GLboolean emit_vop,
-				 GLboolean emit_sop,
-				 int argc, GLuint * src, GLuint dest, int mask)
-{
-	COMPILE_STATE;
-	int hwsrc[3];
-	int srcpos[3];
-	unsigned int used;
-	int tempused;
-	int tempvsrc[3];
-	int tempssrc[3];
-	int pos;
-	int regnr;
-	int i, j;
+	int ip = code->alu.length++;
+	int j;
+	code->node[code->cur_node].alu_end++;
 
-	// Determine instruction slots, whether sources are required on
-	// vector or scalar side, and the smallest slot number where
-	// all source registers are available
-	used = 0;
-	if (emit_vop)
-		used |= SLOT_OP_VECTOR;
-	if (emit_sop)
-		used |= SLOT_OP_SCALAR;
+	code->alu.inst[ip].inst0 = translate_rgb_opcode(inst->RGB.Opcode);
+	code->alu.inst[ip].inst2 = translate_alpha_opcode(inst->Alpha.Opcode);
 
-	pos = get_earliest_allowed_write(cs, dest, mask);
+	for(j = 0; j < 3; ++j) {
+		GLuint src = inst->RGB.Src[j].Index | (inst->RGB.Src[j].Constant << 5);
+		if (!inst->RGB.Src[j].Constant)
+			use_temporary(code, inst->RGB.Src[j].Index);
+		code->alu.inst[ip].inst1 |= src << (6*j);
 
-	if (code->node[code->cur_node].alu_offset > pos)
-		pos = code->node[code->cur_node].alu_offset;
-	for (i = 0; i < argc; ++i) {
-		if (!REG_GET_BUILTIN(src[i])) {
-			if (emit_vop)
-				used |= v_swiz[REG_GET_VSWZ(src[i])].flags << i;
-			if (emit_sop)
-				used |= s_swiz[REG_GET_SSWZ(src[i])].flags << i;
-		}
+		src = inst->Alpha.Src[j].Index | (inst->Alpha.Src[j].Constant << 5);
+		if (!inst->Alpha.Src[j].Constant)
+			use_temporary(code, inst->Alpha.Src[j].Index);
+		code->alu.inst[ip].inst3 |= src << (6*j);
 
-		hwsrc[i] = t_hw_src(cs, src[i], GL_FALSE);	/* Note: sideeffects wrt refcounting! */
-		regnr = hwsrc[i] & 31;
+		GLuint arg = r300FPTranslateRGBSwizzle(inst->RGB.Arg[j].Source, inst->RGB.Arg[j].Swizzle);
+		arg |= inst->RGB.Arg[j].Abs << 6;
+		arg |= inst->RGB.Arg[j].Negate << 5;
+		code->alu.inst[ip].inst0 |= arg << (7*j);
 
-		if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
-			if (used & (SLOT_SRC_VECTOR << i)) {
-				if (cs->hwtemps[regnr].vector_valid > pos)
-					pos = cs->hwtemps[regnr].vector_valid;
-			}
-			if (used & (SLOT_SRC_SCALAR << i)) {
-				if (cs->hwtemps[regnr].scalar_valid > pos)
-					pos = cs->hwtemps[regnr].scalar_valid;
-			}
-		}
+		arg = r300FPTranslateAlphaSwizzle(inst->Alpha.Arg[j].Source, inst->Alpha.Arg[j].Swizzle);
+		arg |= inst->Alpha.Arg[j].Abs << 6;
+		arg |= inst->Alpha.Arg[j].Negate << 5;
+		code->alu.inst[ip].inst2 |= arg << (7*j);
 	}
 
-	// Find a slot that fits
-	for (;; ++pos) {
-		if (cs->slot[pos].used & used & SLOT_OP_BOTH)
-			continue;
-
-		if (pos >= cs->nrslots) {
-			if (cs->nrslots >= PFS_MAX_ALU_INST) {
-				ERROR("Out of ALU instruction slots\n");
-				return -1;
-			}
-
-			code->alu.inst[pos].inst0 = NOP_INST0;
-			code->alu.inst[pos].inst1 = NOP_INST1;
-			code->alu.inst[pos].inst2 = NOP_INST2;
-			code->alu.inst[pos].inst3 = NOP_INST3;
-
-			cs->nrslots++;
-		}
-		// Note: When we need both parts (vector and scalar) of a source,
-		// we always try to put them into the same position. This makes the
-		// code easier to read, and it is optimal (i.e. one doesn't gain
-		// anything by splitting the parts).
-		// It also avoids headaches with swizzles that access both parts (i.e WXY)
-		tempused = cs->slot[pos].used;
-		for (i = 0; i < 3; ++i) {
-			tempvsrc[i] = cs->slot[pos].vsrc[i];
-			tempssrc[i] = cs->slot[pos].ssrc[i];
-		}
-
-		for (i = 0; i < argc; ++i) {
-			int flags = (used >> i) & SLOT_SRC_BOTH;
-
-			if (!flags) {
-				srcpos[i] = 0;
-				continue;
-			}
-
-			for (j = 0; j < 3; ++j) {
-				if ((tempused >> j) & flags & SLOT_SRC_VECTOR) {
-					if (tempvsrc[j] != hwsrc[i])
-						continue;
-				}
-
-				if ((tempused >> j) & flags & SLOT_SRC_SCALAR) {
-					if (tempssrc[j] != hwsrc[i])
-						continue;
-				}
+	if (inst->RGB.Saturate)
+		code->alu.inst[ip].inst0 |= R300_ALU_OUTC_CLAMP;
+	if (inst->Alpha.Saturate)
+		code->alu.inst[ip].inst2 |= R300_ALU_OUTA_CLAMP;
 
-				break;
-			}
-
-			if (j == 3)
-				break;
-
-			srcpos[i] = j;
-			tempused |= flags << j;
-			if (flags & SLOT_SRC_VECTOR)
-				tempvsrc[j] = hwsrc[i];
-			if (flags & SLOT_SRC_SCALAR)
-				tempssrc[j] = hwsrc[i];
-		}
-
-		if (i == argc)
-			break;
+	if (inst->RGB.WriteMask) {
+		use_temporary(code, inst->RGB.DestIndex);
+		code->alu.inst[ip].inst1 |=
+			(inst->RGB.DestIndex << R300_ALU_DSTC_SHIFT) |
+			(inst->RGB.WriteMask << R300_ALU_DSTC_REG_MASK_SHIFT);
 	}
-
-	// Found a slot, reserve it
-	cs->slot[pos].used = tempused | (used & SLOT_OP_BOTH);
-	for (i = 0; i < 3; ++i) {
-		cs->slot[pos].vsrc[i] = tempvsrc[i];
-		cs->slot[pos].ssrc[i] = tempssrc[i];
+	if (inst->RGB.OutputWriteMask) {
+		code->alu.inst[ip].inst1 |= (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT);
+		code->node[code->cur_node].flags |= R300_RGBA_OUT;
 	}
 
-	for (i = 0; i < argc; ++i) {
-		if (REG_GET_TYPE(src[i]) == REG_TYPE_TEMP) {
-			int regnr = hwsrc[i] & 31;
-
-			if (used & (SLOT_SRC_VECTOR << i)) {
-				if (cs->hwtemps[regnr].vector_lastread < pos)
-					cs->hwtemps[regnr].vector_lastread =
-					    pos;
-			}
-			if (used & (SLOT_SRC_SCALAR << i)) {
-				if (cs->hwtemps[regnr].scalar_lastread < pos)
-					cs->hwtemps[regnr].scalar_lastread =
-					    pos;
-			}
-		}
+	if (inst->Alpha.WriteMask) {
+		use_temporary(code, inst->Alpha.DestIndex);
+		code->alu.inst[ip].inst3 |=
+			(inst->Alpha.DestIndex << R300_ALU_DSTA_SHIFT) |
+			R300_ALU_DSTA_REG;
 	}
-
-	// Emit the source fetch code
-	code->alu.inst[pos].inst1 &= ~R300_ALU_SRC_MASK;
-	code->alu.inst[pos].inst1 |=
-	    ((cs->slot[pos].vsrc[0] << R300_ALU_SRC0C_SHIFT) |
-	     (cs->slot[pos].vsrc[1] << R300_ALU_SRC1C_SHIFT) |
-	     (cs->slot[pos].vsrc[2] << R300_ALU_SRC2C_SHIFT));
-
-	code->alu.inst[pos].inst3 &= ~R300_ALU_SRC_MASK;
-	code->alu.inst[pos].inst3 |=
-	    ((cs->slot[pos].ssrc[0] << R300_ALU_SRC0A_SHIFT) |
-	     (cs->slot[pos].ssrc[1] << R300_ALU_SRC1A_SHIFT) |
-	     (cs->slot[pos].ssrc[2] << R300_ALU_SRC2A_SHIFT));
-
-	// Emit the argument selection code
-	if (emit_vop) {
-		int swz[3];
-
-		for (i = 0; i < 3; ++i) {
-			if (i < argc) {
-				swz[i] = (v_swiz[REG_GET_VSWZ(src[i])].base +
-					  (srcpos[i] *
-					   v_swiz[REG_GET_VSWZ(src[i])].
-					   stride)) | ((src[i] & REG_NEGV_MASK)
-						       ? ARG_NEG : 0) | ((src[i]
-									  &
-									  REG_ABS_MASK)
-									 ?
-									 ARG_ABS
-									 : 0);
-			} else {
-				swz[i] = R300_ALU_ARGC_ZERO;
-			}
-		}
-
-		code->alu.inst[pos].inst0 &=
-		    ~(R300_ALU_ARG0C_MASK | R300_ALU_ARG1C_MASK |
-		      R300_ALU_ARG2C_MASK);
-		code->alu.inst[pos].inst0 |=
-		    (swz[0] << R300_ALU_ARG0C_SHIFT) | (swz[1] <<
-							 R300_ALU_ARG1C_SHIFT)
-		    | (swz[2] << R300_ALU_ARG2C_SHIFT);
+	if (inst->Alpha.OutputWriteMask) {
+		code->alu.inst[ip].inst3 |= R300_ALU_DSTA_OUTPUT;
+		code->node[code->cur_node].flags |= R300_RGBA_OUT;
 	}
-
-	if (emit_sop) {
-		int swz[3];
-
-		for (i = 0; i < 3; ++i) {
-			if (i < argc) {
-				swz[i] = (s_swiz[REG_GET_SSWZ(src[i])].base +
-					  (srcpos[i] *
-					   s_swiz[REG_GET_SSWZ(src[i])].
-					   stride)) | ((src[i] & REG_NEGS_MASK)
-						       ? ARG_NEG : 0) | ((src[i]
-									  &
-									  REG_ABS_MASK)
-									 ?
-									 ARG_ABS
-									 : 0);
-			} else {
-				swz[i] = R300_ALU_ARGA_ZERO;
-			}
-		}
-
-		code->alu.inst[pos].inst2 &=
-		    ~(R300_ALU_ARG0A_MASK | R300_ALU_ARG1A_MASK |
-		      R300_ALU_ARG2A_MASK);
-		code->alu.inst[pos].inst2 |=
-		    (swz[0] << R300_ALU_ARG0A_SHIFT) | (swz[1] <<
-							 R300_ALU_ARG1A_SHIFT)
-		    | (swz[2] << R300_ALU_ARG2A_SHIFT);
+	if (inst->Alpha.DepthWriteMask) {
+		code->alu.inst[ip].inst3 |= R300_ALU_DSTA_DEPTH;
+		code->node[code->cur_node].flags |= R300_W_OUT;
+		c->fp->WritesDepth = GL_TRUE;
 	}
 
-	return pos;
+	return GL_TRUE;
 }
 
+
 /**
- * Append an ALU instruction to the instruction list.
+ * Finish the current node without advancing to the next one.
  */
-static void emit_arith(struct r300_pfs_compile_state *cs,
-		       int op,
-		       GLuint dest,
-		       int mask,
-		       GLuint src0, GLuint src1, GLuint src2, int flags)
+static GLboolean finish_node(struct r300_fragment_program_compiler *c)
 {
-	COMPILE_STATE;
-	GLuint src[3] = { src0, src1, src2 };
-	int hwdest;
-	GLboolean emit_vop, emit_sop;
-	int vop, sop, argc;
-	int pos;
-
-	vop = r300_fpop[op].v_op;
-	sop = r300_fpop[op].s_op;
-	argc = r300_fpop[op].argc;
-
-	if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT &&
-	    REG_GET_INDEX(dest) == FRAG_RESULT_DEPR) {
-		if (mask & WRITEMASK_Z) {
-			mask = WRITEMASK_W;
-		} else {
-			return;
-		}
-	}
-
-	emit_vop = GL_FALSE;
-	emit_sop = GL_FALSE;
-	if ((mask & WRITEMASK_XYZ) || vop == R300_ALU_OUTC_DP3)
-		emit_vop = GL_TRUE;
-	if ((mask & WRITEMASK_W) || vop == R300_ALU_OUTC_REPL_ALPHA)
-		emit_sop = GL_TRUE;
-
-	pos =
-	    find_and_prepare_slot(cs, emit_vop, emit_sop, argc, src, dest,
-				  mask);
-	if (pos < 0)
-		return;
-
-	hwdest = t_hw_dst(cs, dest, GL_FALSE, pos);	/* Note: Side effects wrt register allocation */
-
-	if (flags & PFS_FLAG_SAT) {
-		vop |= R300_ALU_OUTC_CLAMP;
-		sop |= R300_ALU_OUTA_CLAMP;
+	struct r300_fragment_program_code *code = c->code;
+	struct r300_fragment_program_node *node = &code->node[code->cur_node];
+
+	if (node->alu_end < 0) {
+		/* Generate a single NOP for this node */
+		struct radeon_pair_instruction inst;
+		_mesa_bzero(&inst, sizeof(inst));
+		if (!emit_alu(c, &inst))
+			return GL_FALSE;
 	}
 
-	/* Throw the pieces together and get ALU/1 */
-	if (emit_vop) {
-		code->alu.inst[pos].inst0 |= vop;
-
-		code->alu.inst[pos].inst1 |= hwdest << R300_ALU_DSTC_SHIFT;
-
-		if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
-			if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
-				code->alu.inst[pos].inst1 |=
-				    (mask & WRITEMASK_XYZ) <<
-				    R300_ALU_DSTC_OUTPUT_MASK_SHIFT;
-			} else
-				assert(0);
+	if (node->tex_end < 0) {
+		if (code->cur_node == 0) {
+			node->tex_end = 0;
 		} else {
-			code->alu.inst[pos].inst1 |=
-			    (mask & WRITEMASK_XYZ) <<
-			    R300_ALU_DSTC_REG_MASK_SHIFT;
-
-			cs->hwtemps[hwdest].vector_valid = pos + 1;
-		}
-	}
-
-	/* And now ALU/3 */
-	if (emit_sop) {
-		code->alu.inst[pos].inst2 |= sop;
-
-		if (mask & WRITEMASK_W) {
-			if (REG_GET_TYPE(dest) == REG_TYPE_OUTPUT) {
-				if (REG_GET_INDEX(dest) == FRAG_RESULT_COLR) {
-					code->alu.inst[pos].inst3 |=
-					    (hwdest << R300_ALU_DSTA_SHIFT) |
-					    R300_ALU_DSTA_OUTPUT;
-				} else if (REG_GET_INDEX(dest) ==
-					   FRAG_RESULT_DEPR) {
-					code->alu.inst[pos].inst3 |=
-					    R300_ALU_DSTA_DEPTH;
-				} else
-					assert(0);
-			} else {
-				code->alu.inst[pos].inst3 |=
-				    (hwdest << R300_ALU_DSTA_SHIFT) |
-				    R300_ALU_DSTA_REG;
-
-				cs->hwtemps[hwdest].scalar_valid = pos + 1;
-			}
+			error("Node %i has no TEX instructions", code->cur_node);
+			return GL_FALSE;
 		}
+	} else {
+		if (code->cur_node == 0)
+			code->first_node_has_tex = 1;
 	}
 
-	return;
+	return GL_TRUE;
 }
 
-static void emit_instruction(struct r300_pfs_compile_state *cs, struct prog_instruction *fpi)
-{
-	COMPILE_STATE;
-	GLuint src[3], dest;
-	int flags, mask = 0;
 
-	if (fpi->SaturateMode == SATURATE_ZERO_ONE)
-		flags = PFS_FLAG_SAT;
-	else
-		flags = 0;
+/**
+ * Begin a block of texture instructions.
+ * Create the necessary indirection.
+ */
+static GLboolean begin_tex(void* data)
+{
+	PROG_CODE;
 
-	if (fpi->Opcode != OPCODE_KIL) {
-		dest = t_dst(cs, fpi->DstReg);
-		mask = fpi->DstReg.WriteMask;
+	if (code->cur_node == 0) {
+		if (code->node[0].alu_end < 0 &&
+		    code->node[0].tex_end < 0)
+			return GL_TRUE;
 	}
 
-	switch (fpi->Opcode) {
-	case OPCODE_ADD:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				src[0], pfs_one, src[1], flags);
-		break;
-	case OPCODE_CMP:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		src[2] = t_src(cs, fpi->SrcReg[2]);
-		/* ARB_f_p - if src0.c < 0.0 ? src1.c : src2.c
-		 *    r300 - if src2.c < 0.0 ? src1.c : src0.c
-		 */
-		emit_arith(cs, PFS_OP_CMP, dest, mask,
-				src[2], src[1], src[0], flags);
-		break;
-	case OPCODE_DP3:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_DP3, dest, mask,
-				src[0], src[1], undef, flags);
-		break;
-	case OPCODE_DP4:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_DP4, dest, mask,
-				src[0], src[1], undef, flags);
-		break;
-	case OPCODE_EX2:
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_EX2, dest, mask,
-				src[0], undef, undef, flags);
-		break;
-	case OPCODE_FRC:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_FRC, dest, mask,
-				src[0], undef, undef, flags);
-		break;
-	case OPCODE_KIL:
-		emit_tex(cs, fpi, R300_TEX_OP_KIL);
-		break;
-	case OPCODE_LG2:
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_LG2, dest, mask,
-				src[0], undef, undef, flags);
-		break;
-	case OPCODE_MAD:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		src[2] = t_src(cs, fpi->SrcReg[2]);
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				src[0], src[1], src[2], flags);
-		break;
-	case OPCODE_MAX:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_MAX, dest, mask,
-				src[0], src[1], undef, flags);
-		break;
-	case OPCODE_MIN:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_MIN, dest, mask,
-				src[0], src[1], undef, flags);
-		break;
-	case OPCODE_MOV:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				src[0], pfs_one, pfs_zero, flags);
-		break;
-	case OPCODE_MUL:
-		src[0] = t_src(cs, fpi->SrcReg[0]);
-		src[1] = t_src(cs, fpi->SrcReg[1]);
-		emit_arith(cs, PFS_OP_MAD, dest, mask,
-				src[0], src[1], pfs_zero, flags);
-		break;
-	case OPCODE_RCP:
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_RCP, dest, mask,
-				src[0], undef, undef, flags);
-		break;
-	case OPCODE_RSQ:
-		src[0] = t_scalar_src(cs, fpi->SrcReg[0]);
-		emit_arith(cs, PFS_OP_RSQ, dest, mask,
-				absolute(src[0]), pfs_zero, pfs_zero, flags);
-		break;
-	case OPCODE_TEX:
-		emit_tex(cs, fpi, R300_TEX_OP_LD);
-		break;
-	case OPCODE_TXB:
-		emit_tex(cs, fpi, R300_TEX_OP_TXB);
-		break;
-	case OPCODE_TXP:
-		emit_tex(cs, fpi, R300_TEX_OP_TXP);
-		break;
-	default:
-		ERROR("unknown fpi->Opcode %d\n", fpi->Opcode);
-		break;
+	if (code->cur_node == 3) {
+		error("Too many texture indirections");
+		return GL_FALSE;
 	}
-}
-
-static GLboolean parse_program(struct r300_pfs_compile_state *cs)
-{
-	COMPILE_STATE;
-	struct prog_instruction* fpi;
 
-	for(fpi = cs->compiler->program->Instructions; fpi->Opcode != OPCODE_END; ++fpi) {
-		emit_instruction(cs, fpi);
-
-		if (fp->error)
-			return GL_FALSE;
-	}
+	if (!finish_node(c))
+		return GL_FALSE;
 
+	struct r300_fragment_program_node *node = &code->node[++code->cur_node];
+	node->alu_offset = code->alu.length;
+	node->alu_end = -1;
+	node->tex_offset = code->tex.length;
+	node->tex_end = -1;
 	return GL_TRUE;
 }
 
 
-/* - Init structures
- * - Determine what hwregs each input corresponds to
- */
-static void init_program(struct r300_pfs_compile_state *cs)
+static GLboolean emit_tex(void* data, struct prog_instruction* inst)
 {
-	COMPILE_STATE;
-	struct gl_fragment_program *mp = &fp->mesa_program;
-	GLuint InputsRead = mp->Base.InputsRead;
-	GLuint temps_used = 0;	/* for fp->temps[] */
-	int i, j;
-
-	/* New compile, reset tracking data */
-	fp->optimization =
-	    driQueryOptioni(&cs->compiler->r300->radeon.optionCache, "fp_optimization");
-	fp->translated = GL_FALSE;
-	fp->error = GL_FALSE;
-	fp->WritesDepth = GL_FALSE;
-	code->tex.length = 0;
-	code->cur_node = 0;
-	code->first_node_has_tex = 0;
-	code->const_nr = 0;
-	code->max_temp_idx = 0;
-	code->node[0].alu_end = -1;
-	code->node[0].tex_end = -1;
-
-	for (i = 0; i < PFS_MAX_ALU_INST; i++) {
-		for (j = 0; j < 3; j++) {
-			cs->slot[i].vsrc[j] = SRC_CONST;
-			cs->slot[i].ssrc[j] = SRC_CONST;
-		}
-	}
-
-	/* Work out what temps the Mesa inputs correspond to, this must match
-	 * what setup_rs_unit does, which shouldn't be a problem as rs_unit
-	 * configures itself based on the fragprog's InputsRead
-	 *
-	 * NOTE: this depends on get_hw_temp() allocating registers in order,
-	 * starting from register 0.
-	 */
+	PROG_CODE;
 
-	/* Texcoords come first */
-	for (i = 0; i < cs->compiler->r300->radeon.glCtx->Const.MaxTextureUnits; i++) {
-		if (InputsRead & (FRAG_BIT_TEX0 << i)) {
-			cs->inputs[FRAG_ATTRIB_TEX0 + i].refcount = 0;
-			cs->inputs[FRAG_ATTRIB_TEX0 + i].reg =
-			    get_hw_temp(cs, 0);
-		}
-	}
-	InputsRead &= ~FRAG_BITS_TEX_ANY;
-
-	/* fragment position treated as a texcoord */
-	if (InputsRead & FRAG_BIT_WPOS) {
-		cs->inputs[FRAG_ATTRIB_WPOS].refcount = 0;
-		cs->inputs[FRAG_ATTRIB_WPOS].reg = get_hw_temp(cs, 0);
+	if (code->tex.length >= PFS_MAX_TEX_INST) {
+		error("Too many TEX instructions");
+		return GL_FALSE;
 	}
-	InputsRead &= ~FRAG_BIT_WPOS;
 
-	/* Then primary colour */
-	if (InputsRead & FRAG_BIT_COL0) {
-		cs->inputs[FRAG_ATTRIB_COL0].refcount = 0;
-		cs->inputs[FRAG_ATTRIB_COL0].reg = get_hw_temp(cs, 0);
-	}
-	InputsRead &= ~FRAG_BIT_COL0;
+	GLuint unit = inst->TexSrcUnit;
+	GLuint dest = inst->DstReg.Index;
+	GLuint opcode;
 
-	/* Secondary color */
-	if (InputsRead & FRAG_BIT_COL1) {
-		cs->inputs[FRAG_ATTRIB_COL1].refcount = 0;
-		cs->inputs[FRAG_ATTRIB_COL1].reg = get_hw_temp(cs, 0);
+	switch(inst->Opcode) {
+	case OPCODE_KIL: opcode = R300_TEX_OP_KIL; break;
+	case OPCODE_TEX: opcode = R300_TEX_OP_LD; break;
+	case OPCODE_TXB: opcode = R300_TEX_OP_TXB; break;
+	case OPCODE_TXP: opcode = R300_TEX_OP_TXP; break;
+	default:
+		error("Unknown texture opcode %i", inst->Opcode);
+		return GL_FALSE;
 	}
-	InputsRead &= ~FRAG_BIT_COL1;
 
-	/* Anything else */
-	if (InputsRead) {
-		WARN_ONCE("Don't know how to handle inputs 0x%x\n", InputsRead);
-		/* force read from hwreg 0 for now */
-		for (i = 0; i < 32; i++)
-			if (InputsRead & (1 << i))
-				cs->inputs[i].reg = 0;
+	if (inst->Opcode == OPCODE_KIL) {
+		unit = 0;
+		dest = 0;
+	} else {
+		use_temporary(code, dest);
 	}
 
-	/* Pre-parse the program, grabbing refcounts on input/temp regs.
-	 * That way, we can free up the reg when it's no longer needed
-	 */
-	for (i = 0; i < cs->compiler->program->NumInstructions; ++i) {
-		struct prog_instruction *fpi = cs->compiler->program->Instructions + i;
-		int idx;
-
-		for (j = 0; j < 3; j++) {
-			idx = fpi->SrcReg[j].Index;
-			switch (fpi->SrcReg[j].File) {
-			case PROGRAM_TEMPORARY:
-				if (!(temps_used & (1 << idx))) {
-					cs->temps[idx].reg = -1;
-					cs->temps[idx].refcount = 1;
-					temps_used |= (1 << idx);
-				} else
-					cs->temps[idx].refcount++;
-				break;
-			case PROGRAM_INPUT:
-				cs->inputs[idx].refcount++;
-				break;
-			default:
-				break;
-			}
-		}
+	use_temporary(code, inst->SrcReg[0].Index);
 
-		idx = fpi->DstReg.Index;
-		if (fpi->DstReg.File == PROGRAM_TEMPORARY) {
-			if (!(temps_used & (1 << idx))) {
-				cs->temps[idx].reg = -1;
-				cs->temps[idx].refcount = 1;
-				temps_used |= (1 << idx);
-			} else
-				cs->temps[idx].refcount++;
-		}
-	}
-	cs->temp_in_use = temps_used;
+	code->node[code->cur_node].tex_end++;
+	code->tex.inst[code->tex.length++] =
+		(inst->SrcReg[0].Index << R300_SRC_ADDR_SHIFT) |
+		(dest << R300_DST_ADDR_SHIFT) |
+		(unit << R300_TEX_ID_SHIFT) |
+		(opcode << R300_TEX_INST_SHIFT);
+	return GL_TRUE;
 }
 
 
+static const struct radeon_pair_handler pair_handler = {
+	.EmitConst = &emit_const,
+	.EmitPaired = &emit_alu,
+	.EmitTex = &emit_tex,
+	.BeginTexBlock = &begin_tex,
+	.MaxHwTemps = PFS_NUM_TEMP_REGS
+};
+
 /**
  * Final compilation step: Turn the intermediate radeon_program into
  * machine-readable instructions.
  */
 GLboolean r300FragmentProgramEmit(struct r300_fragment_program_compiler *compiler)
 {
-	struct r300_pfs_compile_state cs;
 	struct r300_fragment_program_code *code = compiler->code;
 
-	_mesa_memset(&cs, 0, sizeof(cs));
-	cs.compiler = compiler;
-	init_program(&cs);
+	_mesa_bzero(code, sizeof(struct r300_fragment_program_code));
+	code->node[0].alu_end = -1;
+	code->node[0].tex_end = -1;
 
-	if (!parse_program(&cs))
+	if (!radeonPairProgram(compiler->r300->radeon.glCtx, compiler->program, &pair_handler, compiler))
 		return GL_FALSE;
 
-	/* Finish off */
-	code->node[code->cur_node].alu_end =
-		cs.nrslots - code->node[code->cur_node].alu_offset - 1;
-	if (code->node[code->cur_node].tex_end < 0)
-		code->node[code->cur_node].tex_end = 0;
-	code->alu_offset = 0;
-	code->alu_end = cs.nrslots - 1;
-	code->tex_offset = 0;
-	code->tex_end = code->tex.length ? code->tex.length - 1 : 0;
-	assert(code->node[code->cur_node].alu_end >= 0);
-	assert(code->alu_end >= 0);
+	if (!finish_node(compiler))
+		return GL_FALSE;
 
 	return GL_TRUE;
 }
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.c b/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.c
new file mode 100644
index 00000000000..a86d2bd4712
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.c
@@ -0,0 +1,227 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+/**
+ * @file
+ * Utilities to deal with the somewhat odd restriction on R300 fragment
+ * program swizzles.
+ */
+
+#include "r300_fragprog_swizzle.h"
+
+#include "r300_reg.h"
+#include "radeon_nqssadce.h"
+
+#define MAKE_SWZ3(x, y, z) (MAKE_SWIZZLE4(SWIZZLE_##x, SWIZZLE_##y, SWIZZLE_##z, SWIZZLE_ZERO))
+
+struct swizzle_data {
+	GLuint hash; /**< swizzle value this matches */
+	GLuint base; /**< base value for hw swizzle */
+	GLuint stride; /**< difference in base between arg0/1/2 */
+};
+
+static const struct swizzle_data native_swizzles[] = {
+	{MAKE_SWZ3(X, Y, Z), R300_ALU_ARGC_SRC0C_XYZ, 4},
+	{MAKE_SWZ3(X, X, X), R300_ALU_ARGC_SRC0C_XXX, 4},
+	{MAKE_SWZ3(Y, Y, Y), R300_ALU_ARGC_SRC0C_YYY, 4},
+	{MAKE_SWZ3(Z, Z, Z), R300_ALU_ARGC_SRC0C_ZZZ, 4},
+	{MAKE_SWZ3(W, W, W), R300_ALU_ARGC_SRC0A, 1},
+	{MAKE_SWZ3(Y, Z, X), R300_ALU_ARGC_SRC0C_YZX, 1},
+	{MAKE_SWZ3(Z, X, Y), R300_ALU_ARGC_SRC0C_ZXY, 1},
+	{MAKE_SWZ3(W, Z, Y), R300_ALU_ARGC_SRC0CA_WZY, 1},
+	{MAKE_SWZ3(ONE, ONE, ONE), R300_ALU_ARGC_ONE, 0},
+	{MAKE_SWZ3(ZERO, ZERO, ZERO), R300_ALU_ARGC_ZERO, 0}
+};
+
+static const int num_native_swizzles = sizeof(native_swizzles)/sizeof(native_swizzles[0]);
+
+
+/**
+ * Find a native RGB swizzle that matches the given swizzle.
+ * Returns 0 if none found.
+ */
+static const struct swizzle_data* lookup_native_swizzle(GLuint swizzle)
+{
+	int i, comp;
+
+	for(i = 0; i < num_native_swizzles; ++i) {
+		const struct swizzle_data* sd = &native_swizzles[i];
+		for(comp = 0; comp < 3; ++comp) {
+			GLuint swz = GET_SWZ(swizzle, comp);
+			if (swz == SWIZZLE_NIL)
+				continue;
+			if (swz != GET_SWZ(sd->hash, comp))
+				break;
+		}
+		if (comp == 3)
+			return sd;
+	}
+
+	return 0;
+}
+
+
+/**
+ * Check whether the given instruction supports the swizzle and negate
+ * combinations in the given source register.
+ */
+GLboolean r300FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg)
+{
+	if (reg.Abs)
+		reg.NegateBase = 0;
+
+	if (opcode == OPCODE_KIL ||
+	    opcode == OPCODE_TEX ||
+	    opcode == OPCODE_TXB ||
+	    opcode == OPCODE_TXP) {
+		int j;
+
+		if (reg.Abs || reg.NegateBase != (15*reg.NegateAbs))
+			return GL_FALSE;
+
+		for(j = 0; j < 4; ++j) {
+			GLuint swz = GET_SWZ(reg.Swizzle, j);
+			if (swz == SWIZZLE_NIL)
+				continue;
+			if (swz != j)
+				return GL_FALSE;
+		}
+
+		return GL_TRUE;
+	}
+
+	GLuint relevant = 0;
+	int j;
+
+	for(j = 0; j < 3; ++j)
+		if (GET_SWZ(reg.Swizzle, j) != SWIZZLE_NIL)
+			relevant |= 1 << j;
+
+	if ((reg.NegateBase & relevant) && (reg.NegateBase & relevant) != relevant)
+		return GL_FALSE;
+
+	if (!lookup_native_swizzle(reg.Swizzle))
+		return GL_FALSE;
+
+	return GL_TRUE;
+}
+
+
+/**
+ * Generate MOV dst, src using only native swizzles.
+ */
+void r300FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src)
+{
+	if (src.Abs)
+		src.NegateBase = 0;
+
+	while(dst.WriteMask) {
+		const struct swizzle_data *best_swizzle = 0;
+		GLuint best_matchcount = 0;
+		GLuint best_matchmask = 0;
+		GLboolean rgbnegate;
+		int i, comp;
+
+		for(i = 0; i < num_native_swizzles; ++i) {
+			const struct swizzle_data *sd = &native_swizzles[i];
+			GLuint matchcount = 0;
+			GLuint matchmask = 0;
+			for(comp = 0; comp < 3; ++comp) {
+				if (!GET_BIT(dst.WriteMask, comp))
+					continue;
+				GLuint swz = GET_SWZ(src.Swizzle, comp);
+				if (swz == SWIZZLE_NIL)
+					continue;
+				if (swz == GET_SWZ(sd->hash, comp)) {
+					matchcount++;
+					matchmask |= 1 << comp;
+				}
+			}
+			if (matchcount > best_matchcount) {
+				best_swizzle = sd;
+				best_matchcount = matchcount;
+				best_matchmask = matchmask;
+				if (matchmask == (dst.WriteMask & WRITEMASK_XYZ))
+					break;
+			}
+		}
+
+		if ((src.NegateBase & best_matchmask) != 0) {
+			best_matchmask &= src.NegateBase;
+			rgbnegate = !src.NegateAbs;
+		} else {
+			rgbnegate = src.NegateAbs;
+		}
+
+		struct prog_instruction *inst;
+
+		_mesa_insert_instructions(s->Program, s->IP, 1);
+		inst = s->Program->Instructions + s->IP++;
+		inst->Opcode = OPCODE_MOV;
+		inst->DstReg = dst;
+		inst->DstReg.WriteMask &= (best_matchmask | WRITEMASK_W);
+		inst->SrcReg[0] = src;
+		/* Note: We rely on NqSSA/DCE to set unused swizzle components to NIL */
+
+		dst.WriteMask &= ~inst->DstReg.WriteMask;
+	}
+}
+
+
+/**
+ * Translate an RGB (XYZ) swizzle into the hardware code for the given
+ * instruction source.
+ */
+GLuint r300FPTranslateRGBSwizzle(GLuint src, GLuint swizzle)
+{
+	const struct swizzle_data* sd = lookup_native_swizzle(swizzle);
+
+	if (!sd) {
+		_mesa_printf("Not a native swizzle: %08x\n", swizzle);
+		return 0;
+	}
+
+	return sd->base + src*sd->stride;
+}
+
+
+/**
+ * Translate an Alpha (W) swizzle into the hardware code for the given
+ * instruction source.
+ */
+GLuint r300FPTranslateAlphaSwizzle(GLuint src, GLuint swizzle)
+{
+	if (swizzle < 3)
+		return swizzle + 3*src;
+
+	switch(swizzle) {
+	case SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src;
+	case SWIZZLE_ONE: return R300_ALU_ARGA_ONE;
+	case SWIZZLE_ZERO: return R300_ALU_ARGA_ZERO;
+	default: return R300_ALU_ARGA_ONE;
+	}
+}
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.h b/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.h
new file mode 100644
index 00000000000..3da99a9dbe2
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/r300_fragprog_swizzle.h
@@ -0,0 +1,42 @@
+/*
+ * Copyright (C) 2008 Nicolai Haehnle.
+ *
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining
+ * a copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sublicense, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial
+ * portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+ * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef __R300_FRAGPROG_SWIZZLE_H_
+#define __R300_FRAGPROG_SWIZZLE_H_
+
+#include "glheader.h"
+#include "shader/prog_instruction.h"
+
+struct nqssadce_state;
+
+GLboolean r300FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg);
+void r300FPBuildSwizzle(struct nqssadce_state*, struct prog_dst_register dst, struct prog_src_register src);
+
+GLuint r300FPTranslateRGBSwizzle(GLuint src, GLuint swizzle);
+GLuint r300FPTranslateAlphaSwizzle(GLuint src, GLuint swizzle);
+
+#endif /* __R300_FRAGPROG_SWIZZLE_H_ */
diff --git a/src/mesa/drivers/dri/r300/r300_state.c b/src/mesa/drivers/dri/r300/r300_state.c
index fbe5f664181..6931de4421d 100644
--- a/src/mesa/drivers/dri/r300/r300_state.c
+++ b/src/mesa/drivers/dri/r300/r300_state.c
@@ -2497,26 +2497,17 @@ static void r300SetupPixelShader(r300ContextPtr rmesa)
 	r300SetupTextures(ctx);
 
 	R300_STATECHANGE(rmesa, fpi[0]);
-	rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, code->alu_end + 1);
-	for (i = 0; i <= code->alu_end; i++) {
-		rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst0;
-	}
-
 	R300_STATECHANGE(rmesa, fpi[1]);
-	rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, code->alu_end + 1);
-	for (i = 0; i <= code->alu_end; i++) {
-		rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst1;
-	}
-
 	R300_STATECHANGE(rmesa, fpi[2]);
-	rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, code->alu_end + 1);
-	for (i = 0; i <= code->alu_end; i++) {
-		rmesa->hw.fpi[2].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst2;
-	}
-
 	R300_STATECHANGE(rmesa, fpi[3]);
-	rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, code->alu_end + 1);
-	for (i = 0; i <= code->alu_end; i++) {
+	rmesa->hw.fpi[0].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_INST_0, code->alu.length);
+	rmesa->hw.fpi[1].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_RGB_ADDR_0, code->alu.length);
+	rmesa->hw.fpi[2].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_INST_0, code->alu.length);
+	rmesa->hw.fpi[3].cmd[R300_FPI_CMD_0] = cmdpacket0(R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
+	for (i = 0; i < code->alu.length; i++) {
+		rmesa->hw.fpi[0].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst0;
+		rmesa->hw.fpi[1].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst1;
+		rmesa->hw.fpi[2].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst2;
 		rmesa->hw.fpi[3].cmd[R300_FPI_INSTR_0 + i] = code->alu.inst[i].inst3;
 	}
 
@@ -2524,10 +2515,10 @@ static void r300SetupPixelShader(r300ContextPtr rmesa)
 	rmesa->hw.fp.cmd[R300_FP_CNTL0] = code->cur_node | (code->first_node_has_tex << 3);
 	rmesa->hw.fp.cmd[R300_FP_CNTL1] = code->max_temp_idx;
 	rmesa->hw.fp.cmd[R300_FP_CNTL2] =
-	  (code->alu_offset << R300_PFS_CNTL_ALU_OFFSET_SHIFT) |
-	  (code->alu_end << R300_PFS_CNTL_ALU_END_SHIFT) |
-	  (code->tex_offset << R300_PFS_CNTL_TEX_OFFSET_SHIFT) |
-	  (code->tex_end << R300_PFS_CNTL_TEX_END_SHIFT);
+	  (0 << R300_PFS_CNTL_ALU_OFFSET_SHIFT) |
+	  ((code->alu.length-1) << R300_PFS_CNTL_ALU_END_SHIFT) |
+	  (0 << R300_PFS_CNTL_TEX_OFFSET_SHIFT) |
+	  ((code->tex.length ? code->tex.length-1 : 0) << R300_PFS_CNTL_TEX_END_SHIFT);
 	/* I just want to say, the way these nodes are stored.. weird.. */
 	for (i = 0, k = (4 - (code->cur_node + 1)); i < 4; i++, k++) {
 		if (i < (code->cur_node + 1)) {
diff --git a/src/mesa/drivers/dri/r300/r500_fragprog.h b/src/mesa/drivers/dri/r300/r500_fragprog.h
index ed8f7f41cac..8641ceeb8f4 100644
--- a/src/mesa/drivers/dri/r300/r500_fragprog.h
+++ b/src/mesa/drivers/dri/r300/r500_fragprog.h
@@ -45,36 +45,6 @@
 #include "r300_state.h"
 #include "radeon_program.h"
 
-/* supported hw opcodes */
-#define PFS_OP_MAD 0
-#define PFS_OP_DP3 1
-#define PFS_OP_DP4 2
-#define PFS_OP_MIN 3
-#define PFS_OP_MAX 4
-#define PFS_OP_CMP 5
-#define PFS_OP_FRC 6
-#define PFS_OP_EX2 7
-#define PFS_OP_LG2 8
-#define PFS_OP_RCP 9
-#define PFS_OP_RSQ 10
-#define PFS_OP_REPL_ALPHA 11
-#define PFS_OP_CMPH 12
-#define MAX_PFS_OP 12
-
-#define PFS_FLAG_SAT	(1 << 0)
-#define PFS_FLAG_ABS	(1 << 1)
-
-#define ARG_NEG			(1 << 5)
-#define ARG_ABS			(1 << 6)
-#define ARG_MASK		(127 << 0)
-#define ARG_STRIDE		7
-#define SRC_CONST		(1 << 5)
-#define SRC_MASK		(63 << 0)
-#define SRC_STRIDE		6
-
-#define DRI_CONF_FP_OPTIMIZATION_SPEED   0
-#define DRI_CONF_FP_OPTIMIZATION_QUALITY 1
-
 struct r500_fragment_program;
 
 extern void r500TranslateFragmentShader(r300ContextPtr r300,
diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.c b/src/mesa/drivers/dri/r300/radeon_program_pair.c
index 85ddf1dc50e..8762422801e 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_pair.c
+++ b/src/mesa/drivers/dri/r300/radeon_program_pair.c
@@ -578,6 +578,9 @@ static void emit_all_tex(struct pair_state *s)
 	if (s->Debug)
 		_mesa_printf(" BEGIN_TEX\n");
 
+	if (s->Handler->BeginTexBlock)
+		s->Error = s->Error || !s->Handler->BeginTexBlock(s->UserData);
+
 	for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) {
 		int ip = pairinst - s->Instructions;
 		struct prog_instruction *inst = s->Program->Instructions + ip;
@@ -594,9 +597,6 @@ static void emit_all_tex(struct pair_state *s)
 		s->Error = s->Error || !s->Handler->EmitTex(s->UserData, inst);
 	}
 
-	if (s->Handler->EndTexBlock)
-		s->Handler->EndTexBlock(s->UserData);
-
 	if (s->Debug)
 		_mesa_printf(" END_TEX\n");
 }
diff --git a/src/mesa/drivers/dri/r300/radeon_program_pair.h b/src/mesa/drivers/dri/r300/radeon_program_pair.h
index b2bdd08d27c..4624a246298 100644
--- a/src/mesa/drivers/dri/r300/radeon_program_pair.h
+++ b/src/mesa/drivers/dri/r300/radeon_program_pair.h
@@ -110,10 +110,10 @@ struct radeon_pair_handler {
 	GLboolean (*EmitTex)(void*, struct prog_instruction*);
 
 	/**
-	 * Called after a block of contiguous, independent texture
-	 * instructions has been emitted.
+	 * Called before a block of contiguous, independent texture
+	 * instructions is emitted.
 	 */
-	void (*EndTexBlock)(void*);
+	GLboolean (*BeginTexBlock)(void*);
 
 	GLuint MaxHwTemps;
 };