summaryrefslogtreecommitdiff
path: root/src/mesa/drivers/dri/r300/compiler
diff options
context:
space:
mode:
Diffstat (limited to 'src/mesa/drivers/dri/r300/compiler')
-rw-r--r--src/mesa/drivers/dri/r300/compiler/Makefile75
-rw-r--r--src/mesa/drivers/dri/r300/compiler/memory_pool.c95
-rw-r--r--src/mesa/drivers/dri/r300/compiler/memory_pool.h49
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r300_fragprog.c416
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r300_fragprog.h49
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c360
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c223
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h42
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c149
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c615
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c177
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r500_fragprog.c449
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r500_fragprog.h55
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c299
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_code.c170
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_code.h206
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_compiler.c262
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_compiler.h108
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c294
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h91
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program.c181
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program.h121
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c693
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h58
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c958
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h143
26 files changed, 6338 insertions, 0 deletions
diff --git a/src/mesa/drivers/dri/r300/compiler/Makefile b/src/mesa/drivers/dri/r300/compiler/Makefile
new file mode 100644
index 0000000000..d973844192
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/Makefile
@@ -0,0 +1,75 @@
+# src/mesa/drivers/dri/r300/compiler/Makefile
+
+TOP = ../../../../../..
+include $(TOP)/configs/current
+
+LIBNAME = r300compiler
+
+C_SOURCES = \
+ radeon_code.c \
+ radeon_compiler.c \
+ radeon_nqssadce.c \
+ radeon_program.c \
+ radeon_program_alu.c \
+ radeon_program_pair.c \
+ r3xx_fragprog.c \
+ r300_fragprog.c \
+ r300_fragprog_swizzle.c \
+ r300_fragprog_emit.c \
+ r500_fragprog.c \
+ r500_fragprog_emit.c \
+ r3xx_vertprog.c \
+ r3xx_vertprog_dump.c \
+ \
+ memory_pool.c
+
+
+### Basic defines ###
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+ $(CPP_SOURCES:.cpp=.o) \
+ $(ASM_SOURCES:.S=.o)
+
+INCLUDES = \
+ -I. \
+ -I$(TOP)/include \
+ -I$(TOP)/src/mesa \
+
+
+##### TARGETS #####
+
+default: depend lib$(LIBNAME).a
+
+lib$(LIBNAME).a: $(OBJECTS) Makefile $(TOP)/configs/current
+ $(MKLIB) -o $(LIBNAME) -static $(OBJECTS)
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS)
+ rm -f depend
+ touch depend
+ $(MKDEP) $(MKDEP_OPTIONS) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) 2> /dev/null
+
+# Emacs tags
+tags:
+ etags `find . -name \*.[ch]` `find ../include`
+
+# Remove .o and backup files
+clean:
+ rm -f $(OBJECTS) lib$(LIBNAME).a depend depend.bak
+
+# Dummy target
+install:
+ @echo -n ""
+
+##### RULES #####
+
+.c.o:
+ $(CC) -c $(INCLUDES) $(CFLAGS) $(LIBRARY_DEFINES) $< -o $@
+
+.cpp.o:
+ $(CXX) -c $(INCLUDES) $(CXXFLAGS) $(LIBRARY_DEFINES) $< -o $@
+
+.S.o:
+ $(CC) -c $(INCLUDES) $(CFLAGS) $(LIBRARY_DEFINES) $< -o $@
+
+
+sinclude depend
diff --git a/src/mesa/drivers/dri/r300/compiler/memory_pool.c b/src/mesa/drivers/dri/r300/compiler/memory_pool.c
new file mode 100644
index 0000000000..37aa2b6579
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/memory_pool.c
@@ -0,0 +1,95 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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. */
+
+#include "memory_pool.h"
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+
+
+#define POOL_LARGE_ALLOC 4096
+#define POOL_ALIGN 4
+
+
+struct memory_block {
+ struct memory_block * next;
+};
+
+void memory_pool_init(struct memory_pool * pool)
+{
+ memset(pool, 0, sizeof(struct memory_pool));
+}
+
+
+void memory_pool_destroy(struct memory_pool * pool)
+{
+ while(pool->blocks) {
+ struct memory_block * block = pool->blocks;
+ pool->blocks = block->next;
+ free(block);
+ }
+}
+
+static void refill_pool(struct memory_pool * pool)
+{
+ unsigned int blocksize = pool->total_allocated;
+ struct memory_block * newblock;
+
+ if (!blocksize)
+ blocksize = 2*POOL_LARGE_ALLOC;
+
+ newblock = (struct memory_block*)malloc(blocksize);
+ newblock->next = pool->blocks;
+ pool->blocks = newblock;
+
+ pool->head = (unsigned char*)(newblock + 1);
+ pool->end = ((unsigned char*)newblock) + blocksize;
+ pool->total_allocated += blocksize;
+}
+
+
+void * memory_pool_malloc(struct memory_pool * pool, unsigned int bytes)
+{
+ if (bytes < POOL_LARGE_ALLOC) {
+ if (pool->head + bytes > pool->end)
+ refill_pool(pool);
+
+ assert(pool->head + bytes <= pool->end);
+
+ void * ptr = pool->head;
+
+ pool->head += bytes;
+ pool->head = (unsigned char*)(((unsigned long)pool->head + POOL_ALIGN - 1) & ~(POOL_ALIGN - 1));
+
+ return ptr;
+ } else {
+ struct memory_block * block = (struct memory_block*)malloc(bytes + sizeof(struct memory_block));
+
+ block->next = pool->blocks;
+ pool->blocks = block;
+
+ return (block + 1);
+ }
+}
+
+
diff --git a/src/mesa/drivers/dri/r300/compiler/memory_pool.h b/src/mesa/drivers/dri/r300/compiler/memory_pool.h
new file mode 100644
index 0000000000..ce23c319ad
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/memory_pool.h
@@ -0,0 +1,49 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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 MEMORY_POOL_H
+#define MEMORY_POOL_H
+
+struct memory_block;
+
+/**
+ * Provides a pool of memory that can quickly be allocated from, at the
+ * cost of being unable to explicitly free one of the allocated blocks.
+ * Instead, the entire pool can be freed at once.
+ *
+ * The idea is to allow one to quickly allocate a flexible amount of
+ * memory during operations like shader compilation while avoiding
+ * reference counting headaches.
+ */
+struct memory_pool {
+ unsigned char * head;
+ unsigned char * end;
+ unsigned int total_allocated;
+ struct memory_block * blocks;
+};
+
+
+void memory_pool_init(struct memory_pool * pool);
+void memory_pool_destroy(struct memory_pool * pool);
+void * memory_pool_malloc(struct memory_pool * pool, unsigned int bytes);
+
+#endif /* MEMORY_POOL_H */
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c
new file mode 100644
index 0000000000..6c9fba4914
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c
@@ -0,0 +1,416 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * 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.
+ *
+ */
+
+#include "r300_fragprog.h"
+
+#include "shader/prog_parameter.h"
+
+#include "../r300_reg.h"
+
+static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
+{
+ struct prog_src_register reg = { 0, };
+
+ reg.File = PROGRAM_STATE_VAR;
+ reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
+ reg.Swizzle = SWIZZLE_WWWW;
+ return reg;
+}
+
+/**
+ * Transform TEX, TXP, TXB, and KIL instructions in the following way:
+ * - premultiply texture coordinates for RECT
+ * - extract operand swizzles
+ * - introduce a temporary register when write masks are needed
+ */
+GLboolean r300_transform_TEX(
+ struct radeon_compiler * c,
+ struct rc_instruction* inst,
+ void* data)
+{
+ struct r300_fragment_program_compiler *compiler =
+ (struct r300_fragment_program_compiler*)data;
+
+ if (inst->I.Opcode != OPCODE_TEX &&
+ inst->I.Opcode != OPCODE_TXB &&
+ inst->I.Opcode != OPCODE_TXP &&
+ inst->I.Opcode != OPCODE_KIL)
+ return GL_FALSE;
+
+ /* ARB_shadow & EXT_shadow_funcs */
+ if (inst->I.Opcode != OPCODE_KIL &&
+ c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) {
+ GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+
+ if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) {
+ inst->I.Opcode = OPCODE_MOV;
+
+ if (comparefunc == GL_ALWAYS) {
+ inst->I.SrcReg[0].File = PROGRAM_BUILTIN;
+ inst->I.SrcReg[0].Swizzle = SWIZZLE_1111;
+ } else {
+ inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit);
+ }
+
+ return GL_TRUE;
+ } else {
+ GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+ GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode;
+ struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
+ struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
+ struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
+ int pass, fail;
+
+ inst_rcp->I.Opcode = OPCODE_RCP;
+ inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_rcp->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_rcp->I.DstReg.WriteMask = WRITEMASK_W;
+ inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0];
+ inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW;
+
+ inst_cmp->I.DstReg = inst->I.DstReg;
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = rc_find_free_temporary(c);
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+
+ inst_mad->I.Opcode = OPCODE_MAD;
+ inst_mad->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mad->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_mad->I.SrcReg[0] = inst->I.SrcReg[0];
+ inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
+ inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+ inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index;
+ inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW;
+ inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+ inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index;
+ if (depthmode == 0) /* GL_LUMINANCE */
+ inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
+ else if (depthmode == 2) /* GL_ALPHA */
+ inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW;
+
+ /* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
+ * r < tex <=> -tex+r < 0
+ * r >= tex <=> not (-tex+r < 0 */
+ if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
+ inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW;
+ else
+ inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW;
+
+ inst_cmp->I.Opcode = OPCODE_CMP;
+ /* DstReg has been filled out above */
+ inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index;
+
+ if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
+ pass = 1;
+ fail = 2;
+ } else {
+ pass = 2;
+ fail = 1;
+ }
+
+ inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN;
+ inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111;
+ inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit);
+ }
+ }
+
+ /* Hardware uses [0..1]x[0..1] range for rectangle textures
+ * instead of [0..Width]x[0..Height].
+ * Add a scaling instruction.
+ */
+ if (inst->I.Opcode != OPCODE_KIL && inst->I.TexSrcTarget == TEXTURE_RECT_INDEX) {
+ struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mul->I.Opcode = OPCODE_MUL;
+ inst_mul->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mul->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_mul->I.SrcReg[0] = inst->I.SrcReg[0];
+ inst_mul->I.SrcReg[1].File = PROGRAM_STATE_VAR;
+ inst_mul->I.SrcReg[1].Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_R300_TEXRECT_FACTOR, inst->I.TexSrcUnit);
+
+ reset_srcreg(&inst->I.SrcReg[0]);
+ inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[0].Index = inst_mul->I.DstReg.Index;
+ }
+
+ /* Cannot write texture to output registers or with masks */
+ if (inst->I.Opcode != OPCODE_KIL &&
+ (inst->I.DstReg.File != PROGRAM_TEMPORARY || inst->I.DstReg.WriteMask != WRITEMASK_XYZW)) {
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
+
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg = inst->I.DstReg;
+ inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c);
+
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+ }
+
+
+ /* Cannot read texture coordinate from constants file */
+ if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) {
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mov->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_mov->I.SrcReg[0] = inst->I.SrcReg[0];
+
+ reset_srcreg(&inst->I.SrcReg[0]);
+ inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index;
+ }
+
+ return GL_TRUE;
+}
+
+/* just some random things... */
+void r300FragmentProgramDump(struct rX00_fragment_program_code *c)
+{
+ struct r300_fragment_program_code *code = &c->code.r300;
+ int n, i, j;
+ static int pc = 0;
+
+ fprintf(stderr, "pc=%d*************************************\n", pc++);
+
+ fprintf(stderr, "Hardware program\n");
+ fprintf(stderr, "----------------\n");
+
+ for (n = 0; n <= (code->config & 3); n++) {
+ uint32_t code_addr = code->code_addr[3 - (code->config & 3) + n];
+ int alu_offset = (code_addr & R300_ALU_START_MASK) >> R300_ALU_START_SHIFT;
+ int alu_end = (code_addr & R300_ALU_SIZE_MASK) >> R300_ALU_SIZE_SHIFT;
+ int tex_offset = (code_addr & R300_TEX_START_MASK) >> R300_TEX_START_SHIFT;
+ int tex_end = (code_addr & R300_TEX_SIZE_MASK) >> R300_TEX_SIZE_SHIFT;
+
+ fprintf(stderr, "NODE %d: alu_offset: %d, tex_offset: %d, "
+ "alu_end: %d, tex_end: %d (code_addr: %08x)\n", n,
+ alu_offset, tex_offset, alu_end, tex_end, code_addr);
+
+ if (n > 0 || (code->config & R300_PFS_CNTL_FIRST_NODE_HAS_TEX)) {
+ fprintf(stderr, " TEX:\n");
+ for (i = tex_offset;
+ i <= tex_offset + tex_end;
+ ++i) {
+ const char *instr;
+
+ switch ((code->tex.
+ inst[i] >> R300_TEX_INST_SHIFT) &
+ 15) {
+ case R300_TEX_OP_LD:
+ instr = "TEX";
+ break;
+ case R300_TEX_OP_KIL:
+ instr = "KIL";
+ break;
+ case R300_TEX_OP_TXP:
+ instr = "TXP";
+ break;
+ case R300_TEX_OP_TXB:
+ instr = "TXB";
+ break;
+ default:
+ instr = "UNKNOWN";
+ }
+
+ fprintf(stderr,
+ " %s t%i, %c%i, texture[%i] (%08x)\n",
+ instr,
+ (code->tex.
+ inst[i] >> R300_DST_ADDR_SHIFT) & 31,
+ 't',
+ (code->tex.
+ inst[i] >> R300_SRC_ADDR_SHIFT) & 31,
+ (code->tex.
+ inst[i] & R300_TEX_ID_MASK) >>
+ R300_TEX_ID_SHIFT,
+ code->tex.inst[i]);
+ }
+ }
+
+ for (i = alu_offset;
+ i <= alu_offset + alu_end; ++i) {
+ char srcc[3][10], dstc[20];
+ char srca[3][10], dsta[20];
+ char argc[3][20];
+ char arga[3][20];
+ char flags[5], tmp[10];
+
+ for (j = 0; j < 3; ++j) {
+ int regc = code->alu.inst[i].rgb_addr >> (j * 6);
+ int rega = code->alu.inst[i].alpha_addr >> (j * 6);
+
+ sprintf(srcc[j], "%c%i",
+ (regc & 32) ? 'c' : 't', regc & 31);
+ sprintf(srca[j], "%c%i",
+ (rega & 32) ? 'c' : 't', rega & 31);
+ }
+
+ dstc[0] = 0;
+ sprintf(flags, "%s%s%s",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_REG_X) ? "x" : "",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_REG_Y) ? "y" : "",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_REG_Z) ? "z" : "");
+ if (flags[0] != 0) {
+ sprintf(dstc, "t%i.%s ",
+ (code->alu.inst[i].
+ rgb_addr >> R300_ALU_DSTC_SHIFT) & 31,
+ flags);
+ }
+ sprintf(flags, "%s%s%s",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_OUTPUT_X) ? "x" : "",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_OUTPUT_Y) ? "y" : "",
+ (code->alu.inst[i].
+ rgb_addr & R300_ALU_DSTC_OUTPUT_Z) ? "z" : "");
+ if (flags[0] != 0) {
+ sprintf(tmp, "o%i.%s",
+ (code->alu.inst[i].
+ rgb_addr >> R300_ALU_DSTC_SHIFT) & 31,
+ flags);
+ strcat(dstc, tmp);
+ }
+
+ dsta[0] = 0;
+ if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_REG) {
+ sprintf(dsta, "t%i.w ",
+ (code->alu.inst[i].
+ alpha_addr >> R300_ALU_DSTA_SHIFT) & 31);
+ }
+ if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_OUTPUT) {
+ sprintf(tmp, "o%i.w ",
+ (code->alu.inst[i].
+ alpha_addr >> R300_ALU_DSTA_SHIFT) & 31);
+ strcat(dsta, tmp);
+ }
+ if (code->alu.inst[i].alpha_addr & R300_ALU_DSTA_DEPTH) {
+ strcat(dsta, "Z");
+ }
+
+ fprintf(stderr,
+ "%3i: xyz: %3s %3s %3s -> %-20s (%08x)\n"
+ " w: %3s %3s %3s -> %-20s (%08x)\n", i,
+ srcc[0], srcc[1], srcc[2], dstc,
+ code->alu.inst[i].rgb_addr, srca[0], srca[1],
+ srca[2], dsta, code->alu.inst[i].alpha_addr);
+
+ for (j = 0; j < 3; ++j) {
+ int regc = code->alu.inst[i].rgb_inst >> (j * 7);
+ int rega = code->alu.inst[i].alpha_inst >> (j * 7);
+ int d;
+ char buf[20];
+
+ d = regc & 31;
+ if (d < 12) {
+ switch (d % 4) {
+ case R300_ALU_ARGC_SRC0C_XYZ:
+ sprintf(buf, "%s.xyz",
+ srcc[d / 4]);
+ break;
+ case R300_ALU_ARGC_SRC0C_XXX:
+ sprintf(buf, "%s.xxx",
+ srcc[d / 4]);
+ break;
+ case R300_ALU_ARGC_SRC0C_YYY:
+ sprintf(buf, "%s.yyy",
+ srcc[d / 4]);
+ break;
+ case R300_ALU_ARGC_SRC0C_ZZZ:
+ sprintf(buf, "%s.zzz",
+ srcc[d / 4]);
+ break;
+ }
+ } else if (d < 15) {
+ sprintf(buf, "%s.www", srca[d - 12]);
+ } else if (d == 20) {
+ sprintf(buf, "0.0");
+ } else if (d == 21) {
+ sprintf(buf, "1.0");
+ } else if (d == 22) {
+ sprintf(buf, "0.5");
+ } else if (d >= 23 && d < 32) {
+ d -= 23;
+ switch (d / 3) {
+ case 0:
+ sprintf(buf, "%s.yzx",
+ srcc[d % 3]);
+ break;
+ case 1:
+ sprintf(buf, "%s.zxy",
+ srcc[d % 3]);
+ break;
+ case 2:
+ sprintf(buf, "%s.Wzy",
+ srcc[d % 3]);
+ break;
+ }
+ } else {
+ sprintf(buf, "%i", d);
+ }
+
+ sprintf(argc[j], "%s%s%s%s",
+ (regc & 32) ? "-" : "",
+ (regc & 64) ? "|" : "",
+ buf, (regc & 64) ? "|" : "");
+
+ d = rega & 31;
+ if (d < 9) {
+ sprintf(buf, "%s.%c", srcc[d / 3],
+ 'x' + (char)(d % 3));
+ } else if (d < 12) {
+ sprintf(buf, "%s.w", srca[d - 9]);
+ } else if (d == 16) {
+ sprintf(buf, "0.0");
+ } else if (d == 17) {
+ sprintf(buf, "1.0");
+ } else if (d == 18) {
+ sprintf(buf, "0.5");
+ } else {
+ sprintf(buf, "%i", d);
+ }
+
+ sprintf(arga[j], "%s%s%s%s",
+ (rega & 32) ? "-" : "",
+ (rega & 64) ? "|" : "",
+ buf, (rega & 64) ? "|" : "");
+ }
+
+ fprintf(stderr, " xyz: %8s %8s %8s op: %08x\n"
+ " w: %8s %8s %8s op: %08x\n",
+ argc[0], argc[1], argc[2],
+ code->alu.inst[i].rgb_inst, arga[0], arga[1],
+ arga[2], code->alu.inst[i].alpha_inst);
+ }
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h
new file mode 100644
index 0000000000..0ac46dbd9c
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h
@@ -0,0 +1,49 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * 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.
+ *
+ */
+
+/*
+ * Authors:
+ * Ben Skeggs <darktama@iinet.net.au>
+ * Jerome Glisse <j.glisse@gmail.com>
+ */
+#ifndef __R300_FRAGPROG_H_
+#define __R300_FRAGPROG_H_
+
+#include "shader/program.h"
+#include "shader/prog_instruction.h"
+
+#include "radeon_compiler.h"
+#include "radeon_program.h"
+
+
+extern void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler);
+
+extern void r300FragmentProgramDump(struct rX00_fragment_program_code *c);
+
+extern GLboolean r300_transform_TEX(struct radeon_compiler * c, struct rc_instruction* inst, void* data);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
new file mode 100644
index 0000000000..305dc074ee
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c
@@ -0,0 +1,360 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * 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
+ *
+ * Emit the r300_fragment_program_code that can be understood by the hardware.
+ * Input is a pre-transformed radeon_program.
+ *
+ * \author Ben Skeggs <darktama@iinet.net.au>
+ *
+ * \author Jerome Glisse <j.glisse@gmail.com>
+ *
+ * \todo FogOption
+ */
+
+#include "r300_fragprog.h"
+
+#include "../r300_reg.h"
+
+#include "radeon_program_pair.h"
+#include "r300_fragprog_swizzle.h"
+
+
+struct r300_emit_state {
+ struct r300_fragment_program_compiler * compiler;
+
+ unsigned current_node : 2;
+ unsigned node_first_tex : 8;
+ unsigned node_first_alu : 8;
+ uint32_t node_flags;
+};
+
+#define PROG_CODE \
+ struct r300_emit_state * emit = (struct r300_emit_state*)data; \
+ struct r300_fragment_program_compiler *c = emit->compiler; \
+ struct r300_fragment_program_code *code = &c->code->code.r300
+
+#define error(fmt, args...) do { \
+ rc_error(&c->Base, "%s::%s(): " fmt "\n", \
+ __FILE__, __FUNCTION__, ##args); \
+ } while(0)
+
+
+/**
+ * Mark a temporary register as used.
+ */
+static void use_temporary(struct r300_fragment_program_code *code, GLuint index)
+{
+ if (index > code->pixsize)
+ code->pixsize = index;
+}
+
+
+static GLuint translate_rgb_opcode(struct r300_fragment_program_compiler * c, GLuint opcode)
+{
+ 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("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;
+ }
+}
+
+static GLuint translate_alpha_opcode(struct r300_fragment_program_compiler * c, GLuint opcode)
+{
+ 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("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;
+ }
+}
+
+/**
+ * Emit one paired ALU instruction.
+ */
+static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst)
+{
+ PROG_CODE;
+
+ if (code->alu.length >= R300_PFS_MAX_ALU_INST) {
+ error("Too many ALU instructions");
+ return GL_FALSE;
+ }
+
+ int ip = code->alu.length++;
+ int j;
+
+ code->alu.inst[ip].rgb_inst = translate_rgb_opcode(c, inst->RGB.Opcode);
+ code->alu.inst[ip].alpha_inst = translate_alpha_opcode(c, inst->Alpha.Opcode);
+
+ for(j = 0; j < 3; ++j) {
+ 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].rgb_addr |= src << (6*j);
+
+ 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].alpha_addr |= src << (6*j);
+
+ 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].rgb_inst |= arg << (7*j);
+
+ 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].alpha_inst |= arg << (7*j);
+ }
+
+ if (inst->RGB.Saturate)
+ code->alu.inst[ip].rgb_inst |= R300_ALU_OUTC_CLAMP;
+ if (inst->Alpha.Saturate)
+ code->alu.inst[ip].alpha_inst |= R300_ALU_OUTA_CLAMP;
+
+ if (inst->RGB.WriteMask) {
+ use_temporary(code, inst->RGB.DestIndex);
+ code->alu.inst[ip].rgb_addr |=
+ (inst->RGB.DestIndex << R300_ALU_DSTC_SHIFT) |
+ (inst->RGB.WriteMask << R300_ALU_DSTC_REG_MASK_SHIFT);
+ }
+ if (inst->RGB.OutputWriteMask) {
+ code->alu.inst[ip].rgb_addr |= (inst->RGB.OutputWriteMask << R300_ALU_DSTC_OUTPUT_MASK_SHIFT);
+ emit->node_flags |= R300_RGBA_OUT;
+ }
+
+ if (inst->Alpha.WriteMask) {
+ use_temporary(code, inst->Alpha.DestIndex);
+ code->alu.inst[ip].alpha_addr |=
+ (inst->Alpha.DestIndex << R300_ALU_DSTA_SHIFT) |
+ R300_ALU_DSTA_REG;
+ }
+ if (inst->Alpha.OutputWriteMask) {
+ code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_OUTPUT;
+ emit->node_flags |= R300_RGBA_OUT;
+ }
+ if (inst->Alpha.DepthWriteMask) {
+ code->alu.inst[ip].alpha_addr |= R300_ALU_DSTA_DEPTH;
+ emit->node_flags |= R300_W_OUT;
+ c->code->writes_depth = GL_TRUE;
+ }
+
+ return GL_TRUE;
+}
+
+
+/**
+ * Finish the current node without advancing to the next one.
+ */
+static GLboolean finish_node(struct r300_emit_state * emit)
+{
+ struct r300_fragment_program_compiler * c = emit->compiler;
+ struct r300_fragment_program_code *code = &emit->compiler->code->code.r300;
+
+ if (code->alu.length == emit->node_first_alu) {
+ /* Generate a single NOP for this node */
+ struct radeon_pair_instruction inst;
+ _mesa_bzero(&inst, sizeof(inst));
+ if (!emit_alu(emit, &inst))
+ return GL_FALSE;
+ }
+
+ unsigned alu_offset = emit->node_first_alu;
+ unsigned alu_end = code->alu.length - alu_offset - 1;
+ unsigned tex_offset = emit->node_first_tex;
+ unsigned tex_end = code->tex.length - tex_offset - 1;
+
+ if (code->tex.length == emit->node_first_tex) {
+ if (emit->current_node > 0) {
+ error("Node %i has no TEX instructions", emit->current_node);
+ return GL_FALSE;
+ }
+
+ tex_end = 0;
+ } else {
+ if (emit->current_node == 0)
+ code->config |= R300_PFS_CNTL_FIRST_NODE_HAS_TEX;
+ }
+
+ /* Write the config register.
+ * Note: The order in which the words for each node are written
+ * is not correct here and needs to be fixed up once we're entirely
+ * done
+ *
+ * Also note that the register specification from AMD is slightly
+ * incorrect in its description of this register. */
+ code->code_addr[emit->current_node] =
+ (alu_offset << R300_ALU_START_SHIFT) |
+ (alu_end << R300_ALU_SIZE_SHIFT) |
+ (tex_offset << R300_TEX_START_SHIFT) |
+ (tex_end << R300_TEX_SIZE_SHIFT) |
+ emit->node_flags;
+
+ return GL_TRUE;
+}
+
+
+/**
+ * Begin a block of texture instructions.
+ * Create the necessary indirection.
+ */
+static GLboolean begin_tex(void* data)
+{
+ PROG_CODE;
+
+ if (code->alu.length == emit->node_first_alu &&
+ code->tex.length == emit->node_first_tex) {
+ return GL_TRUE;
+ }
+
+ if (emit->current_node == 3) {
+ error("Too many texture indirections");
+ return GL_FALSE;
+ }
+
+ if (!finish_node(emit))
+ return GL_FALSE;
+
+ emit->current_node++;
+ emit->node_first_tex = code->tex.length;
+ emit->node_first_alu = code->alu.length;
+ emit->node_flags = 0;
+ return GL_TRUE;
+}
+
+
+static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* inst)
+{
+ PROG_CODE;
+
+ if (code->tex.length >= R300_PFS_MAX_TEX_INST) {
+ error("Too many TEX instructions");
+ return GL_FALSE;
+ }
+
+ GLuint unit = inst->TexSrcUnit;
+ GLuint dest = inst->DestIndex;
+ GLuint opcode;
+
+ switch(inst->Opcode) {
+ case RADEON_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break;
+ case RADEON_OPCODE_TEX: opcode = R300_TEX_OP_LD; break;
+ case RADEON_OPCODE_TXB: opcode = R300_TEX_OP_TXB; break;
+ case RADEON_OPCODE_TXP: opcode = R300_TEX_OP_TXP; break;
+ default:
+ error("Unknown texture opcode %i", inst->Opcode);
+ return GL_FALSE;
+ }
+
+ if (inst->Opcode == RADEON_OPCODE_KIL) {
+ unit = 0;
+ dest = 0;
+ } else {
+ use_temporary(code, dest);
+ }
+
+ use_temporary(code, inst->SrcIndex);
+
+ code->tex.inst[code->tex.length++] =
+ (inst->SrcIndex << 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 = {
+ .EmitPaired = &emit_alu,
+ .EmitTex = &emit_tex,
+ .BeginTexBlock = &begin_tex,
+ .MaxHwTemps = R300_PFS_NUM_TEMP_REGS
+};
+
+/**
+ * Final compilation step: Turn the intermediate radeon_program into
+ * machine-readable instructions.
+ */
+void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler)
+{
+ struct r300_emit_state emit;
+ struct r300_fragment_program_code *code = &compiler->code->code.r300;
+
+ memset(&emit, 0, sizeof(emit));
+ emit.compiler = compiler;
+
+ _mesa_bzero(code, sizeof(struct r300_fragment_program_code));
+
+ radeonPairProgram(compiler, &pair_handler, &emit);
+ if (compiler->Base.Error)
+ return;
+
+ /* Finish the program */
+ finish_node(&emit);
+
+ code->config |= emit.current_node; /* FIRST_NODE_HAS_TEX set by finish_node */
+ code->code_offset =
+ (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);
+
+ if (emit.current_node < 3) {
+ int shift = 3 - emit.current_node;
+ int i;
+ for(i = 0; i <= emit.current_node; ++i)
+ code->code_addr[shift + i] = code->code_addr[i];
+ for(i = 0; i < shift; ++i)
+ code->code_addr[i] = 0;
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c
new file mode 100644
index 0000000000..1b14cc3888
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c
@@ -0,0 +1,223 @@
+/*
+ * 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"
+#include "radeon_compiler.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.Negate = NEGATE_NONE;
+
+ if (opcode == OPCODE_KIL ||
+ opcode == OPCODE_TEX ||
+ opcode == OPCODE_TXB ||
+ opcode == OPCODE_TXP) {
+ int j;
+
+ if (reg.Abs || reg.Negate)
+ 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.Negate & relevant) && ((reg.Negate & 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.Negate = NEGATE_NONE;
+
+ while(dst.WriteMask) {
+ const struct swizzle_data *best_swizzle = 0;
+ GLuint best_matchcount = 0;
+ GLuint best_matchmask = 0;
+ 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)) {
+ /* check if the negate bit of current component
+ * is the same for already matched components */
+ if (matchmask && (!!(src.Negate & matchmask) != !!(src.Negate & (1 << comp))))
+ continue;
+
+ matchcount++;
+ matchmask |= 1 << comp;
+ }
+ }
+ if (matchcount > best_matchcount) {
+ best_swizzle = sd;
+ best_matchcount = matchcount;
+ best_matchmask = matchmask;
+ if (matchmask == (dst.WriteMask & WRITEMASK_XYZ))
+ break;
+ }
+ }
+
+ struct rc_instruction *inst = rc_insert_new_instruction(s->Compiler, s->IP->Prev);
+ inst->I.Opcode = OPCODE_MOV;
+ inst->I.DstReg = dst;
+ inst->I.DstReg.WriteMask &= (best_matchmask | WRITEMASK_W);
+ inst->I.SrcReg[0] = src;
+ inst->I.SrcReg[0].Negate = (best_matchmask & src.Negate) ? NEGATE_XYZW : NEGATE_NONE;
+ /* Note: We rely on NqSSA/DCE to set unused swizzle components to NIL */
+
+ dst.WriteMask &= ~inst->I.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/compiler/r300_fragprog_swizzle.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h
new file mode 100644
index 0000000000..231bf4eef5
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/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 "main/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/compiler/r3xx_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
new file mode 100644
index 0000000000..d39b82be71
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c
@@ -0,0 +1,149 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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. */
+
+#include "radeon_compiler.h"
+
+#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+#include "shader/prog_statevars.h"
+
+#include "radeon_nqssadce.h"
+#include "radeon_program_alu.h"
+#include "r300_fragprog.h"
+#include "r300_fragprog_swizzle.h"
+#include "r500_fragprog.h"
+
+
+static void nqssadce_init(struct nqssadce_state* s)
+{
+ struct r300_fragment_program_compiler * c = s->UserData;
+ s->Outputs[c->OutputColor].Sourced = WRITEMASK_XYZW;
+ s->Outputs[c->OutputDepth].Sourced = WRITEMASK_W;
+}
+
+static void rewrite_depth_out(struct r300_fragment_program_compiler * c)
+{
+ struct rc_instruction *rci;
+
+ for (rci = c->Base.Program.Instructions.Next; rci != &c->Base.Program.Instructions; rci = rci->Next) {
+ struct prog_instruction * inst = &rci->I;
+
+ if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != c->OutputDepth)
+ continue;
+
+ if (inst->DstReg.WriteMask & WRITEMASK_Z) {
+ inst->DstReg.WriteMask = WRITEMASK_W;
+ } else {
+ inst->DstReg.WriteMask = 0;
+ continue;
+ }
+
+ 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;
+ }
+ }
+}
+
+void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
+{
+ rewrite_depth_out(c);
+
+ if (c->is_r500) {
+ struct radeon_program_transformation transformations[] = {
+ { &r500_transform_TEX, c },
+ { &radeonTransformALU, 0 },
+ { &radeonTransformDeriv, 0 },
+ { &radeonTransformTrigScale, 0 }
+ };
+ radeonLocalTransform(&c->Base, 4, transformations);
+ } else {
+ struct radeon_program_transformation transformations[] = {
+ { &r300_transform_TEX, c },
+ { &radeonTransformALU, 0 },
+ { &radeonTransformTrigSimple, 0 }
+ };
+ radeonLocalTransform(&c->Base, 3, transformations);
+ }
+
+ if (c->Base.Debug) {
+ _mesa_printf("Fragment Program: After native rewrite:\n");
+ rc_print_program(&c->Base.Program);
+ fflush(stdout);
+ }
+
+ if (c->is_r500) {
+ struct radeon_nqssadce_descr nqssadce = {
+ .Init = &nqssadce_init,
+ .IsNativeSwizzle = &r500FPIsNativeSwizzle,
+ .BuildSwizzle = &r500FPBuildSwizzle
+ };
+ radeonNqssaDce(&c->Base, &nqssadce, c);
+ } else {
+ struct radeon_nqssadce_descr nqssadce = {
+ .Init = &nqssadce_init,
+ .IsNativeSwizzle = &r300FPIsNativeSwizzle,
+ .BuildSwizzle = &r300FPBuildSwizzle
+ };
+ radeonNqssaDce(&c->Base, &nqssadce, c);
+ }
+
+ if (c->Base.Debug) {
+ _mesa_printf("Compiler: after NqSSA-DCE:\n");
+ rc_print_program(&c->Base.Program);
+ fflush(stdout);
+ }
+
+ if (c->is_r500) {
+ r500BuildFragmentProgramHwCode(c);
+ } else {
+ r300BuildFragmentProgramHwCode(c);
+ }
+
+ rc_constants_copy(&c->code->constants, &c->Base.Program.Constants);
+
+ if (c->Base.Debug) {
+ if (c->is_r500) {
+ r500FragmentProgramDump(c->code);
+ } else {
+ r300FragmentProgramDump(c->code);
+ }
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
new file mode 100644
index 0000000000..fc9c8f805a
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -0,0 +1,615 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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. */
+
+#include "radeon_compiler.h"
+
+#include "../r300_reg.h"
+
+#include "radeon_nqssadce.h"
+#include "radeon_program.h"
+#include "radeon_program_alu.h"
+
+#include "shader/prog_print.h"
+
+
+/*
+ * Take an already-setup and valid source then swizzle it appropriately to
+ * obtain a constant ZERO or ONE source.
+ */
+#define __CONST(x, y) \
+ (PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[x]), \
+ t_swizzle(y), \
+ t_swizzle(y), \
+ t_swizzle(y), \
+ t_swizzle(y), \
+ t_src_class(vpi->SrcReg[x].File), \
+ NEGATE_NONE) | (vpi->SrcReg[x].RelAddr << 4))
+
+
+static unsigned long t_dst_mask(GLuint mask)
+{
+ /* WRITEMASK_* is equivalent to VSF_FLAG_* */
+ return mask & WRITEMASK_XYZW;
+}
+
+static unsigned long t_dst_class(gl_register_file file)
+{
+
+ switch (file) {
+ case PROGRAM_TEMPORARY:
+ return PVS_DST_REG_TEMPORARY;
+ case PROGRAM_OUTPUT:
+ return PVS_DST_REG_OUT;
+ case PROGRAM_ADDRESS:
+ return PVS_DST_REG_A0;
+ /*
+ case PROGRAM_INPUT:
+ case PROGRAM_LOCAL_PARAM:
+ case PROGRAM_ENV_PARAM:
+ case PROGRAM_NAMED_PARAM:
+ case PROGRAM_STATE_VAR:
+ case PROGRAM_WRITE_ONLY:
+ case PROGRAM_ADDRESS:
+ */
+ default:
+ fprintf(stderr, "problem in %s", __FUNCTION__);
+ _mesa_exit(-1);
+ return -1;
+ }
+}
+
+static unsigned long t_dst_index(struct r300_vertex_program_code *vp,
+ struct prog_dst_register *dst)
+{
+ if (dst->File == PROGRAM_OUTPUT)
+ return vp->outputs[dst->Index];
+
+ return dst->Index;
+}
+
+static unsigned long t_src_class(gl_register_file file)
+{
+ switch (file) {
+ case PROGRAM_TEMPORARY:
+ return PVS_SRC_REG_TEMPORARY;
+ case PROGRAM_INPUT:
+ return PVS_SRC_REG_INPUT;
+ case PROGRAM_LOCAL_PARAM:
+ case PROGRAM_ENV_PARAM:
+ case PROGRAM_NAMED_PARAM:
+ case PROGRAM_CONSTANT:
+ case PROGRAM_STATE_VAR:
+ return PVS_SRC_REG_CONSTANT;
+ /*
+ case PROGRAM_OUTPUT:
+ case PROGRAM_WRITE_ONLY:
+ case PROGRAM_ADDRESS:
+ */
+ default:
+ fprintf(stderr, "problem in %s", __FUNCTION__);
+ _mesa_exit(-1);
+ return -1;
+ }
+}
+
+static GLboolean t_src_conflict(struct prog_src_register a, struct prog_src_register b)
+{
+ unsigned long aclass = t_src_class(a.File);
+ unsigned long bclass = t_src_class(b.File);
+
+ if (aclass != bclass)
+ return GL_FALSE;
+ if (aclass == PVS_SRC_REG_TEMPORARY)
+ return GL_FALSE;
+
+ if (a.RelAddr || b.RelAddr)
+ return GL_TRUE;
+ if (a.Index != b.Index)
+ return GL_TRUE;
+
+ return GL_FALSE;
+}
+
+static INLINE unsigned long t_swizzle(GLubyte swizzle)
+{
+ /* this is in fact a NOP as the Mesa SWIZZLE_* are all identical to VSF_IN_COMPONENT_* */
+ return swizzle;
+}
+
+static unsigned long t_src_index(struct r300_vertex_program_code *vp,
+ struct prog_src_register *src)
+{
+ if (src->File == PROGRAM_INPUT) {
+ assert(vp->inputs[src->Index] != -1);
+ return vp->inputs[src->Index];
+ } else {
+ if (src->Index < 0) {
+ fprintf(stderr,
+ "negative offsets for indirect addressing do not work.\n");
+ return 0;
+ }
+ return src->Index;
+ }
+}
+
+/* these two functions should probably be merged... */
+
+static unsigned long t_src(struct r300_vertex_program_code *vp,
+ struct prog_src_register *src)
+{
+ /* src->Negate uses the NEGATE_ flags from program_instruction.h,
+ * which equal our VSF_FLAGS_ values, so it's safe to just pass it here.
+ */
+ return PVS_SRC_OPERAND(t_src_index(vp, src),
+ t_swizzle(GET_SWZ(src->Swizzle, 0)),
+ t_swizzle(GET_SWZ(src->Swizzle, 1)),
+ t_swizzle(GET_SWZ(src->Swizzle, 2)),
+ t_swizzle(GET_SWZ(src->Swizzle, 3)),
+ t_src_class(src->File),
+ src->Negate) | (src->RelAddr << 4);
+}
+
+static unsigned long t_src_scalar(struct r300_vertex_program_code *vp,
+ struct prog_src_register *src)
+{
+ /* src->Negate uses the NEGATE_ flags from program_instruction.h,
+ * which equal our VSF_FLAGS_ values, so it's safe to just pass it here.
+ */
+ return PVS_SRC_OPERAND(t_src_index(vp, src),
+ t_swizzle(GET_SWZ(src->Swizzle, 0)),
+ t_swizzle(GET_SWZ(src->Swizzle, 0)),
+ t_swizzle(GET_SWZ(src->Swizzle, 0)),
+ t_swizzle(GET_SWZ(src->Swizzle, 0)),
+ t_src_class(src->File),
+ src->Negate ? NEGATE_XYZW : NEGATE_NONE) |
+ (src->RelAddr << 4);
+}
+
+static GLboolean valid_dst(struct r300_vertex_program_code *vp,
+ struct prog_dst_register *dst)
+{
+ if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) {
+ return GL_FALSE;
+ } else if (dst->File == PROGRAM_ADDRESS) {
+ assert(dst->Index == 0);
+ }
+
+ return GL_TRUE;
+}
+
+static void ei_vector1(struct r300_vertex_program_code *vp,
+ GLuint hw_opcode,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ inst[0] = PVS_OP_DST_OPERAND(hw_opcode,
+ GL_FALSE,
+ GL_FALSE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ inst[1] = t_src(vp, &vpi->SrcReg[0]);
+ inst[2] = __CONST(0, SWIZZLE_ZERO);
+ inst[3] = __CONST(0, SWIZZLE_ZERO);
+}
+
+static void ei_vector2(struct r300_vertex_program_code *vp,
+ GLuint hw_opcode,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ inst[0] = PVS_OP_DST_OPERAND(hw_opcode,
+ GL_FALSE,
+ GL_FALSE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ inst[1] = t_src(vp, &vpi->SrcReg[0]);
+ inst[2] = t_src(vp, &vpi->SrcReg[1]);
+ inst[3] = __CONST(1, SWIZZLE_ZERO);
+}
+
+static void ei_math1(struct r300_vertex_program_code *vp,
+ GLuint hw_opcode,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ inst[0] = PVS_OP_DST_OPERAND(hw_opcode,
+ GL_TRUE,
+ GL_FALSE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ inst[1] = t_src_scalar(vp, &vpi->SrcReg[0]);
+ inst[2] = __CONST(0, SWIZZLE_ZERO);
+ inst[3] = __CONST(0, SWIZZLE_ZERO);
+}
+
+static void ei_lit(struct r300_vertex_program_code *vp,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ //LIT TMP 1.Y Z TMP 1{} {X W Z Y} TMP 1{} {Y W Z X} TMP 1{} {Y X Z W}
+
+ inst[0] = PVS_OP_DST_OPERAND(ME_LIGHT_COEFF_DX,
+ GL_TRUE,
+ GL_FALSE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ /* NOTE: Users swizzling might not work. */
+ inst[1] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W
+ PVS_SRC_SELECT_FORCE_0, // Z
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y
+ t_src_class(vpi->SrcReg[0].File),
+ vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) |
+ (vpi->SrcReg[0].RelAddr << 4);
+ inst[2] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W
+ PVS_SRC_SELECT_FORCE_0, // Z
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X
+ t_src_class(vpi->SrcReg[0].File),
+ vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) |
+ (vpi->SrcReg[0].RelAddr << 4);
+ inst[3] = PVS_SRC_OPERAND(t_src_index(vp, &vpi->SrcReg[0]), t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 1)), // Y
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 0)), // X
+ PVS_SRC_SELECT_FORCE_0, // Z
+ t_swizzle(GET_SWZ(vpi->SrcReg[0].Swizzle, 3)), // W
+ t_src_class(vpi->SrcReg[0].File),
+ vpi->SrcReg[0].Negate ? NEGATE_XYZW : NEGATE_NONE) |
+ (vpi->SrcReg[0].RelAddr << 4);
+}
+
+static void ei_mad(struct r300_vertex_program_code *vp,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ inst[0] = PVS_OP_DST_OPERAND(PVS_MACRO_OP_2CLK_MADD,
+ GL_FALSE,
+ GL_TRUE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ inst[1] = t_src(vp, &vpi->SrcReg[0]);
+ inst[2] = t_src(vp, &vpi->SrcReg[1]);
+ inst[3] = t_src(vp, &vpi->SrcReg[2]);
+}
+
+static void ei_pow(struct r300_vertex_program_code *vp,
+ struct prog_instruction *vpi,
+ GLuint * inst)
+{
+ inst[0] = PVS_OP_DST_OPERAND(ME_POWER_FUNC_FF,
+ GL_TRUE,
+ GL_FALSE,
+ t_dst_index(vp, &vpi->DstReg),
+ t_dst_mask(vpi->DstReg.WriteMask),
+ t_dst_class(vpi->DstReg.File));
+ inst[1] = t_src_scalar(vp, &vpi->SrcReg[0]);
+ inst[2] = __CONST(0, SWIZZLE_ZERO);
+ inst[3] = t_src_scalar(vp, &vpi->SrcReg[1]);
+}
+
+
+static void translate_vertex_program(struct r300_vertex_program_compiler * compiler)
+{
+ struct rc_instruction *rci;
+
+ compiler->code->pos_end = 0; /* Not supported yet */
+ compiler->code->length = 0;
+
+ compiler->SetHwInputOutput(compiler);
+
+ for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) {
+ struct prog_instruction *vpi = &rci->I;
+ GLuint *inst = compiler->code->body.d + compiler->code->length;
+
+ /* Skip instructions writing to non-existing destination */
+ if (!valid_dst(compiler->code, &vpi->DstReg))
+ continue;
+
+ if (compiler->code->length >= VSF_MAX_FRAGMENT_LENGTH) {
+ rc_error(&compiler->Base, "Vertex program has too many instructions\n");
+ return;
+ }
+
+ switch (vpi->Opcode) {
+ case OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break;
+ case OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break;
+ case OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break;
+ case OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break;
+ case OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break;
+ case OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break;
+ case OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break;
+ case OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break;
+ case OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break;
+ case OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break;
+ case OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break;
+ case OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break;
+ case OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break;
+ case OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break;
+ case OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break;
+ case OPCODE_POW: ei_pow(compiler->code, vpi, inst); break;
+ case OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break;
+ case OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break;
+ case OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break;
+ case OPCODE_SLT: ei_vector2(compiler->code, VE_SET_LESS_THAN, vpi, inst); break;
+ default:
+ rc_error(&compiler->Base, "Unknown opcode %i\n", vpi->Opcode);
+ return;
+ }
+
+ compiler->code->length += 4;
+
+ if (compiler->Base.Error)
+ return;
+ }
+}
+
+struct temporary_allocation {
+ GLuint Allocated:1;
+ GLuint HwTemp:15;
+ struct rc_instruction * LastRead;
+};
+
+static void allocate_temporary_registers(struct r300_vertex_program_compiler * compiler)
+{
+ struct rc_instruction *inst;
+ GLuint num_orig_temps = 0;
+ GLboolean hwtemps[VSF_MAX_FRAGMENT_TEMPS];
+ struct temporary_allocation * ta;
+ GLuint i, j;
+
+ compiler->code->num_temporaries = 0;
+ memset(hwtemps, 0, sizeof(hwtemps));
+
+ /* Pass 1: Count original temporaries and allocate structures */
+ for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode);
+ GLuint numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) {
+ if (inst->I.SrcReg[i].Index >= num_orig_temps)
+ num_orig_temps = inst->I.SrcReg[i].Index + 1;
+ }
+ }
+
+ if (numdsts) {
+ if (inst->I.DstReg.File == PROGRAM_TEMPORARY) {
+ if (inst->I.DstReg.Index >= num_orig_temps)
+ num_orig_temps = inst->I.DstReg.Index + 1;
+ }
+ }
+ }
+
+ ta = (struct temporary_allocation*)memory_pool_malloc(&compiler->Base.Pool,
+ sizeof(struct temporary_allocation) * num_orig_temps);
+ memset(ta, 0, sizeof(struct temporary_allocation) * num_orig_temps);
+
+ /* Pass 2: Determine original temporary lifetimes */
+ for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY)
+ ta[inst->I.SrcReg[i].Index].LastRead = inst;
+ }
+ }
+
+ /* Pass 3: Register allocation */
+ for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode);
+ GLuint numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) {
+ GLuint orig = inst->I.SrcReg[i].Index;
+ inst->I.SrcReg[i].Index = ta[orig].HwTemp;
+
+ if (ta[orig].Allocated && inst == ta[orig].LastRead)
+ hwtemps[ta[orig].HwTemp] = GL_FALSE;
+ }
+ }
+
+ if (numdsts) {
+ if (inst->I.DstReg.File == PROGRAM_TEMPORARY) {
+ GLuint orig = inst->I.DstReg.Index;
+
+ if (!ta[orig].Allocated) {
+ for(j = 0; j < VSF_MAX_FRAGMENT_TEMPS; ++j) {
+ if (!hwtemps[j])
+ break;
+ }
+ if (j >= VSF_MAX_FRAGMENT_TEMPS) {
+ fprintf(stderr, "Out of hw temporaries\n");
+ } else {
+ ta[orig].Allocated = GL_TRUE;
+ ta[orig].HwTemp = j;
+ hwtemps[j] = GL_TRUE;
+
+ if (j >= compiler->code->num_temporaries)
+ compiler->code->num_temporaries = j + 1;
+ }
+ }
+
+ inst->I.DstReg.Index = ta[orig].HwTemp;
+ }
+ }
+ }
+}
+
+
+/**
+ * Vertex engine cannot read two inputs or two constants at the same time.
+ * Introduce intermediate MOVs to temporary registers to account for this.
+ */
+static GLboolean transform_source_conflicts(
+ struct radeon_compiler *c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ GLuint num_operands = _mesa_num_inst_src_regs(inst->I.Opcode);
+
+ if (num_operands == 3) {
+ if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[2])
+ || t_src_conflict(inst->I.SrcReg[0], inst->I.SrcReg[2])) {
+ int tmpreg = rc_find_free_temporary(c);
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mov->I.DstReg.Index = tmpreg;
+ inst_mov->I.SrcReg[0] = inst->I.SrcReg[2];
+
+ reset_srcreg(&inst->I.SrcReg[2]);
+ inst->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[2].Index = tmpreg;
+ }
+ }
+
+ if (num_operands >= 2) {
+ if (t_src_conflict(inst->I.SrcReg[1], inst->I.SrcReg[0])) {
+ int tmpreg = rc_find_free_temporary(c);
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mov->I.DstReg.Index = tmpreg;
+ inst_mov->I.SrcReg[0] = inst->I.SrcReg[1];
+
+ reset_srcreg(&inst->I.SrcReg[1]);
+ inst->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[1].Index = tmpreg;
+ }
+ }
+
+ return GL_TRUE;
+}
+
+static void addArtificialOutputs(struct r300_vertex_program_compiler * compiler)
+{
+ int i;
+
+ for(i = 0; i < 32; ++i) {
+ if ((compiler->RequiredOutputs & (1 << i)) &&
+ !(compiler->Base.Program.OutputsWritten & (1 << i))) {
+ struct rc_instruction * inst = rc_insert_new_instruction(&compiler->Base, compiler->Base.Program.Instructions.Prev);
+ inst->I.Opcode = OPCODE_MOV;
+
+ inst->I.DstReg.File = PROGRAM_OUTPUT;
+ inst->I.DstReg.Index = i;
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+
+ inst->I.SrcReg[0].File = PROGRAM_CONSTANT;
+ inst->I.SrcReg[0].Index = 0;
+ inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+ compiler->Base.Program.OutputsWritten |= 1 << i;
+ }
+ }
+}
+
+static void nqssadceInit(struct nqssadce_state* s)
+{
+ struct r300_vertex_program_compiler * compiler = s->UserData;
+ int i;
+
+ for(i = 0; i < VERT_RESULT_MAX; ++i) {
+ if (compiler->RequiredOutputs & (1 << i))
+ s->Outputs[i].Sourced = WRITEMASK_XYZW;
+ }
+}
+
+static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg)
+{
+ (void) opcode;
+ (void) reg;
+
+ return GL_TRUE;
+}
+
+
+
+void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler)
+{
+ addArtificialOutputs(compiler);
+
+ {
+ struct radeon_program_transformation transformations[] = {
+ { &r300_transform_vertex_alu, 0 },
+ };
+ radeonLocalTransform(&compiler->Base, 1, transformations);
+ }
+
+ if (compiler->Base.Debug) {
+ fprintf(stderr, "Vertex program after native rewrite:\n");
+ rc_print_program(&compiler->Base.Program);
+ fflush(stdout);
+ }
+
+ {
+ /* Note: This pass has to be done seperately from ALU rewrite,
+ * otherwise non-native ALU instructions with source conflits
+ * will not be treated properly.
+ */
+ struct radeon_program_transformation transformations[] = {
+ { &transform_source_conflicts, 0 },
+ };
+ radeonLocalTransform(&compiler->Base, 1, transformations);
+ }
+
+ if (compiler->Base.Debug) {
+ fprintf(stderr, "Vertex program after source conflict resolve:\n");
+ rc_print_program(&compiler->Base.Program);
+ fflush(stdout);
+ }
+
+ {
+ struct radeon_nqssadce_descr nqssadce = {
+ .Init = &nqssadceInit,
+ .IsNativeSwizzle = &swizzleIsNative,
+ .BuildSwizzle = NULL
+ };
+ radeonNqssaDce(&compiler->Base, &nqssadce, compiler);
+
+ /* We need this step for reusing temporary registers */
+ allocate_temporary_registers(compiler);
+
+ if (compiler->Base.Debug) {
+ fprintf(stderr, "Vertex program after NQSSADCE:\n");
+ rc_print_program(&compiler->Base.Program);
+ fflush(stdout);
+ }
+ }
+
+ translate_vertex_program(compiler);
+
+ rc_constants_copy(&compiler->code->constants, &compiler->Base.Program.Constants);
+
+ compiler->code->InputsRead = compiler->Base.Program.InputsRead;
+ compiler->code->OutputsWritten = compiler->Base.Program.OutputsWritten;
+
+ if (compiler->Base.Debug) {
+ printf("Final vertex program code:\n");
+ r300_vertex_program_dump(compiler->code);
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c
new file mode 100644
index 0000000000..39cc6953ba
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog_dump.c
@@ -0,0 +1,177 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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. */
+
+#include "radeon_code.h"
+
+#include <stdio.h>
+
+static char* r300_vs_ve_ops[] = {
+ /* R300 vector ops */
+ " VE_NO_OP",
+ " VE_DOT_PRODUCT",
+ " VE_MULTIPLY",
+ " VE_ADD",
+ " VE_MULTIPLY_ADD",
+ " VE_DISTANCE_FACTOR",
+ " VE_FRACTION",
+ " VE_MAXIMUM",
+ " VE_MINIMUM",
+ "VE_SET_GREATER_THAN_EQUAL",
+ " VE_SET_LESS_THAN",
+ " VE_MULTIPLYX2_ADD",
+ " VE_MULTIPLY_CLAMP",
+ " VE_FLT2FIX_DX",
+ " VE_FLT2FIX_DX_RND",
+ /* R500 vector ops */
+ " VE_PRED_SET_EQ_PUSH",
+ " VE_PRED_SET_GT_PUSH",
+ " VE_PRED_SET_GTE_PUSH",
+ " VE_PRED_SET_NEQ_PUSH",
+ " VE_COND_WRITE_EQ",
+ " VE_COND_WRITE_GT",
+ " VE_COND_WRITE_GTE",
+ " VE_COND_WRITE_NEQ",
+ " VE_SET_GREATER_THAN",
+ " VE_SET_EQUAL",
+ " VE_SET_NOT_EQUAL",
+ " (reserved)",
+ " (reserved)",
+ " (reserved)",
+};
+
+static char* r300_vs_me_ops[] = {
+ /* R300 math ops */
+ " ME_NO_OP",
+ " ME_EXP_BASE2_DX",
+ " ME_LOG_BASE2_DX",
+ " ME_EXP_BASEE_FF",
+ " ME_LIGHT_COEFF_DX",
+ " ME_POWER_FUNC_FF",
+ " ME_RECIP_DX",
+ " ME_RECIP_FF",
+ " ME_RECIP_SQRT_DX",
+ " ME_RECIP_SQRT_FF",
+ " ME_MULTIPLY",
+ " ME_EXP_BASE2_FULL_DX",
+ " ME_LOG_BASE2_FULL_DX",
+ " ME_POWER_FUNC_FF_CLAMP_B",
+ "ME_POWER_FUNC_FF_CLAMP_B1",
+ "ME_POWER_FUNC_FF_CLAMP_01",
+ " ME_SIN",
+ " ME_COS",
+ /* R500 math ops */
+ " ME_LOG_BASE2_IEEE",
+ " ME_RECIP_IEEE",
+ " ME_RECIP_SQRT_IEEE",
+ " ME_PRED_SET_EQ",
+ " ME_PRED_SET_GT",
+ " ME_PRED_SET_GTE",
+ " ME_PRED_SET_NEQ",
+ " ME_PRED_SET_CLR",
+ " ME_PRED_SET_INV",
+ " ME_PRED_SET_POP",
+ " ME_PRED_SET_RESTORE",
+ " (reserved)",
+ " (reserved)",
+ " (reserved)",
+};
+
+/* XXX refactor to avoid clashing symbols */
+static char* r300_vs_src_debug[] = {
+ "t",
+ "i",
+ "c",
+ "a",
+};
+
+static char* r300_vs_dst_debug[] = {
+ "t",
+ "a0",
+ "o",
+ "ox",
+ "a",
+ "i",
+ "u",
+ "u",
+};
+
+static char* r300_vs_swiz_debug[] = {
+ "X",
+ "Y",
+ "Z",
+ "W",
+ "0",
+ "1",
+ "U",
+ "U",
+};
+
+
+static void r300_vs_op_dump(uint32_t op)
+{
+ printf(" dst: %d%s op: ",
+ (op >> 13) & 0x7f, r300_vs_dst_debug[(op >> 8) & 0x7]);
+ if (op & 0x80) {
+ if (op & 0x1) {
+ printf("PVS_MACRO_OP_2CLK_M2X_ADD\n");
+ } else {
+ printf(" PVS_MACRO_OP_2CLK_MADD\n");
+ }
+ } else if (op & 0x40) {
+ printf("%s\n", r300_vs_me_ops[op & 0x1f]);
+ } else {
+ printf("%s\n", r300_vs_ve_ops[op & 0x1f]);
+ }
+}
+
+static void r300_vs_src_dump(uint32_t src)
+{
+ printf(" reg: %d%s swiz: %s%s/%s%s/%s%s/%s%s\n",
+ (src >> 5) & 0x7f, r300_vs_src_debug[src & 0x3],
+ src & (1 << 25) ? "-" : " ",
+ r300_vs_swiz_debug[(src >> 13) & 0x7],
+ src & (1 << 26) ? "-" : " ",
+ r300_vs_swiz_debug[(src >> 16) & 0x7],
+ src & (1 << 27) ? "-" : " ",
+ r300_vs_swiz_debug[(src >> 19) & 0x7],
+ src & (1 << 28) ? "-" : " ",
+ r300_vs_swiz_debug[(src >> 22) & 0x7]);
+}
+
+void r300_vertex_program_dump(struct r300_vertex_program_code * vs)
+{
+ unsigned instrcount = vs->length / 4;
+ unsigned i;
+
+ for(i = 0; i < instrcount; i++) {
+ unsigned offset = i*4;
+ unsigned src;
+
+ printf("%d: op: 0x%08x", i, vs->body.d[offset]);
+ r300_vs_op_dump(vs->body.d[offset]);
+
+ for(src = 0; src < 3; ++src) {
+ printf(" src%i: 0x%08x", src, vs->body.d[offset+1+src]);
+ r300_vs_src_dump(vs->body.d[offset+1+src]);
+ }
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
new file mode 100644
index 0000000000..7e2faed690
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c
@@ -0,0 +1,449 @@
+/*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ *
+ * 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.
+ *
+ */
+
+#include "r500_fragprog.h"
+
+#include "../r300_reg.h"
+
+static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu)
+{
+ struct prog_src_register reg = { 0, };
+
+ reg.File = PROGRAM_STATE_VAR;
+ reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu);
+ reg.Swizzle = SWIZZLE_WWWW;
+ return reg;
+}
+
+/**
+ * Transform TEX, TXP, TXB, and KIL instructions in the following way:
+ * - implement texture compare (shadow extensions)
+ * - extract non-native source / destination operands
+ */
+GLboolean r500_transform_TEX(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void* data)
+{
+ struct r300_fragment_program_compiler *compiler =
+ (struct r300_fragment_program_compiler*)data;
+
+ if (inst->I.Opcode != OPCODE_TEX &&
+ inst->I.Opcode != OPCODE_TXB &&
+ inst->I.Opcode != OPCODE_TXP &&
+ inst->I.Opcode != OPCODE_KIL)
+ return GL_FALSE;
+
+ /* ARB_shadow & EXT_shadow_funcs */
+ if (inst->I.Opcode != OPCODE_KIL &&
+ c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) {
+ GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+
+ if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) {
+ inst->I.Opcode = OPCODE_MOV;
+
+ if (comparefunc == GL_ALWAYS) {
+ inst->I.SrcReg[0].File = PROGRAM_BUILTIN;
+ inst->I.SrcReg[0].Swizzle = SWIZZLE_1111;
+ } else {
+ inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit);
+ }
+
+ return GL_TRUE;
+ } else {
+ GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func;
+ GLuint depthmode = compiler->state.unit[inst->I.TexSrcUnit].depth_texture_mode;
+ struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, inst);
+ struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_rcp);
+ struct rc_instruction * inst_cmp = rc_insert_new_instruction(c, inst_mad);
+ int pass, fail;
+
+ inst_rcp->I.Opcode = OPCODE_RCP;
+ inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_rcp->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_rcp->I.DstReg.WriteMask = WRITEMASK_W;
+ inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0];
+ inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW;
+
+ inst_cmp->I.DstReg = inst->I.DstReg;
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = rc_find_free_temporary(c);
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+
+ inst_mad->I.Opcode = OPCODE_MAD;
+ inst_mad->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mad->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_mad->I.SrcReg[0] = inst->I.SrcReg[0];
+ inst_mad->I.SrcReg[0].Swizzle = SWIZZLE_ZZZZ;
+ inst_mad->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+ inst_mad->I.SrcReg[1].Index = inst_rcp->I.DstReg.Index;
+ inst_mad->I.SrcReg[1].Swizzle = SWIZZLE_WWWW;
+ inst_mad->I.SrcReg[2].File = PROGRAM_TEMPORARY;
+ inst_mad->I.SrcReg[2].Index = inst->I.DstReg.Index;
+ if (depthmode == 0) /* GL_LUMINANCE */
+ inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_Z);
+ else if (depthmode == 2) /* GL_ALPHA */
+ inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW;
+
+ /* Recall that SrcReg[0] is tex, SrcReg[2] is r and:
+ * r < tex <=> -tex+r < 0
+ * r >= tex <=> not (-tex+r < 0 */
+ if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
+ inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ NEGATE_XYZW;
+ else
+ inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW;
+
+ inst_cmp->I.Opcode = OPCODE_CMP;
+ /* DstReg has been filled out above */
+ inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index;
+
+ if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
+ pass = 1;
+ fail = 2;
+ } else {
+ pass = 2;
+ fail = 1;
+ }
+
+ inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN;
+ inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111;
+ inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit);
+ }
+ }
+
+ /* Cannot write texture to output registers */
+ if (inst->I.Opcode != OPCODE_KIL && inst->I.DstReg.File != PROGRAM_TEMPORARY) {
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst);
+
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg = inst->I.DstReg;
+ inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c);
+
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+ }
+
+ /* Cannot read texture coordinate from constants file */
+ if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) {
+ struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst->Prev);
+
+ inst_mov->I.Opcode = OPCODE_MOV;
+ inst_mov->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mov->I.DstReg.Index = rc_find_free_temporary(c);
+ inst_mov->I.SrcReg[0] = inst->I.SrcReg[0];
+
+ reset_srcreg(&inst->I.SrcReg[0]);
+ inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index;
+ }
+
+ return GL_TRUE;
+}
+
+GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg)
+{
+ GLuint relevant;
+ int i;
+
+ if (opcode == OPCODE_TEX ||
+ opcode == OPCODE_TXB ||
+ opcode == OPCODE_TXP ||
+ opcode == OPCODE_KIL) {
+ if (reg.Abs)
+ return GL_FALSE;
+
+ if (opcode == OPCODE_KIL && (reg.Swizzle != SWIZZLE_NOOP || reg.Negate != NEGATE_NONE))
+ return GL_FALSE;
+
+ if (reg.Negate)
+ reg.Negate ^= NEGATE_XYZW;
+
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(reg.Swizzle, i);
+ if (swz == SWIZZLE_NIL) {
+ reg.Negate &= ~(1 << i);
+ continue;
+ }
+ if (swz >= 4)
+ return GL_FALSE;
+ }
+
+ if (reg.Negate)
+ return GL_FALSE;
+
+ return GL_TRUE;
+ } else if (opcode == OPCODE_DDX || opcode == OPCODE_DDY) {
+ /* DDX/MDH and DDY/MDV explicitly ignore incoming swizzles;
+ * if it doesn't fit perfectly into a .xyzw case... */
+ if (reg.Swizzle == SWIZZLE_NOOP && !reg.Abs && !reg.Negate)
+ return GL_TRUE;
+
+ return GL_FALSE;
+ } else {
+ /* ALU instructions support almost everything */
+ if (reg.Abs)
+ return GL_TRUE;
+
+ relevant = 0;
+ for(i = 0; i < 3; ++i) {
+ GLuint swz = GET_SWZ(reg.Swizzle, i);
+ if (swz != SWIZZLE_NIL && swz != SWIZZLE_ZERO)
+ relevant |= 1 << i;
+ }
+ if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
+ return GL_FALSE;
+
+ return GL_TRUE;
+ }
+}
+
+/**
+ * Implement a MOV with a potentially non-native swizzle.
+ *
+ * The only thing we *cannot* do in an ALU instruction is per-component
+ * negation. Therefore, we split the MOV into two instructions when necessary.
+ */
+void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src)
+{
+ GLuint negatebase[2] = { 0, 0 };
+ int i;
+
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(src.Swizzle, i);
+ if (swz == SWIZZLE_NIL)
+ continue;
+ negatebase[GET_BIT(src.Negate, i)] |= 1 << i;
+ }
+
+ for(i = 0; i <= 1; ++i) {
+ if (!negatebase[i])
+ continue;
+
+ struct rc_instruction *inst = rc_insert_new_instruction(s->Compiler, s->IP->Prev);
+ inst->I.Opcode = OPCODE_MOV;
+ inst->I.DstReg = dst;
+ inst->I.DstReg.WriteMask = negatebase[i];
+ inst->I.SrcReg[0] = src;
+ inst->I.SrcReg[0].Negate = (i == 0) ? NEGATE_NONE : NEGATE_XYZW;
+ }
+}
+
+
+static char *toswiz(int swiz_val) {
+ switch(swiz_val) {
+ case 0: return "R";
+ case 1: return "G";
+ case 2: return "B";
+ case 3: return "A";
+ case 4: return "0";
+ case 5: return "1/2";
+ case 6: return "1";
+ case 7: return "U";
+ }
+ return NULL;
+}
+
+static char *toop(int op_val)
+{
+ char *str = NULL;
+ switch (op_val) {
+ case 0: str = "MAD"; break;
+ case 1: str = "DP3"; break;
+ case 2: str = "DP4"; break;
+ case 3: str = "D2A"; break;
+ case 4: str = "MIN"; break;
+ case 5: str = "MAX"; break;
+ case 6: str = "Reserved"; break;
+ case 7: str = "CND"; break;
+ case 8: str = "CMP"; break;
+ case 9: str = "FRC"; break;
+ case 10: str = "SOP"; break;
+ case 11: str = "MDH"; break;
+ case 12: str = "MDV"; break;
+ }
+ return str;
+}
+
+static char *to_alpha_op(int op_val)
+{
+ char *str = NULL;
+ switch (op_val) {
+ case 0: str = "MAD"; break;
+ case 1: str = "DP"; break;
+ case 2: str = "MIN"; break;
+ case 3: str = "MAX"; break;
+ case 4: str = "Reserved"; break;
+ case 5: str = "CND"; break;
+ case 6: str = "CMP"; break;
+ case 7: str = "FRC"; break;
+ case 8: str = "EX2"; break;
+ case 9: str = "LN2"; break;
+ case 10: str = "RCP"; break;
+ case 11: str = "RSQ"; break;
+ case 12: str = "SIN"; break;
+ case 13: str = "COS"; break;
+ case 14: str = "MDH"; break;
+ case 15: str = "MDV"; break;
+ }
+ return str;
+}
+
+static char *to_mask(int val)
+{
+ char *str = NULL;
+ switch(val) {
+ case 0: str = "NONE"; break;
+ case 1: str = "R"; break;
+ case 2: str = "G"; break;
+ case 3: str = "RG"; break;
+ case 4: str = "B"; break;
+ case 5: str = "RB"; break;
+ case 6: str = "GB"; break;
+ case 7: str = "RGB"; break;
+ case 8: str = "A"; break;
+ case 9: str = "AR"; break;
+ case 10: str = "AG"; break;
+ case 11: str = "ARG"; break;
+ case 12: str = "AB"; break;
+ case 13: str = "ARB"; break;
+ case 14: str = "AGB"; break;
+ case 15: str = "ARGB"; break;
+ }
+ return str;
+}
+
+static char *to_texop(int val)
+{
+ switch(val) {
+ case 0: return "NOP";
+ case 1: return "LD";
+ case 2: return "TEXKILL";
+ case 3: return "PROJ";
+ case 4: return "LODBIAS";
+ case 5: return "LOD";
+ case 6: return "DXDY";
+ }
+ return NULL;
+}
+
+void r500FragmentProgramDump(struct rX00_fragment_program_code *c)
+{
+ struct r500_fragment_program_code *code = &c->code.r500;
+ fprintf(stderr, "R500 Fragment Program:\n--------\n");
+
+ int n;
+ uint32_t inst;
+ uint32_t inst0;
+ char *str = NULL;
+
+ for (n = 0; n < code->inst_end+1; n++) {
+ inst0 = inst = code->inst[n].inst0;
+ fprintf(stderr,"%d\t0:CMN_INST 0x%08x:", n, inst);
+ switch(inst & 0x3) {
+ case R500_INST_TYPE_ALU: str = "ALU"; break;
+ case R500_INST_TYPE_OUT: str = "OUT"; break;
+ case R500_INST_TYPE_FC: str = "FC"; break;
+ case R500_INST_TYPE_TEX: str = "TEX"; break;
+ };
+ fprintf(stderr,"%s %s %s %s %s ", str,
+ inst & R500_INST_TEX_SEM_WAIT ? "TEX_WAIT" : "",
+ inst & R500_INST_LAST ? "LAST" : "",
+ inst & R500_INST_NOP ? "NOP" : "",
+ inst & R500_INST_ALU_WAIT ? "ALU WAIT" : "");
+ fprintf(stderr,"wmask: %s omask: %s\n", to_mask((inst >> 11) & 0xf),
+ to_mask((inst >> 15) & 0xf));
+
+ switch(inst0 & 0x3) {
+ case 0:
+ case 1:
+ fprintf(stderr,"\t1:RGB_ADDR 0x%08x:", code->inst[n].inst1);
+ inst = code->inst[n].inst1;
+
+ fprintf(stderr,"Addr0: %d%c, Addr1: %d%c, Addr2: %d%c, srcp:%d\n",
+ inst & 0xff, (inst & (1<<8)) ? 'c' : 't',
+ (inst >> 10) & 0xff, (inst & (1<<18)) ? 'c' : 't',
+ (inst >> 20) & 0xff, (inst & (1<<28)) ? 'c' : 't',
+ (inst >> 30));
+
+ fprintf(stderr,"\t2:ALPHA_ADDR 0x%08x:", code->inst[n].inst2);
+ inst = code->inst[n].inst2;
+ fprintf(stderr,"Addr0: %d%c, Addr1: %d%c, Addr2: %d%c, srcp:%d\n",
+ inst & 0xff, (inst & (1<<8)) ? 'c' : 't',
+ (inst >> 10) & 0xff, (inst & (1<<18)) ? 'c' : 't',
+ (inst >> 20) & 0xff, (inst & (1<<28)) ? 'c' : 't',
+ (inst >> 30));
+ fprintf(stderr,"\t3 RGB_INST: 0x%08x:", code->inst[n].inst3);
+ inst = code->inst[n].inst3;
+ fprintf(stderr,"rgb_A_src:%d %s/%s/%s %d rgb_B_src:%d %s/%s/%s %d\n",
+ (inst) & 0x3, toswiz((inst >> 2) & 0x7), toswiz((inst >> 5) & 0x7), toswiz((inst >> 8) & 0x7),
+ (inst >> 11) & 0x3,
+ (inst >> 13) & 0x3, toswiz((inst >> 15) & 0x7), toswiz((inst >> 18) & 0x7), toswiz((inst >> 21) & 0x7),
+ (inst >> 24) & 0x3);
+
+
+ fprintf(stderr,"\t4 ALPHA_INST:0x%08x:", code->inst[n].inst4);
+ inst = code->inst[n].inst4;
+ fprintf(stderr,"%s dest:%d%s alp_A_src:%d %s %d alp_B_src:%d %s %d w:%d\n", to_alpha_op(inst & 0xf),
+ (inst >> 4) & 0x7f, inst & (1<<11) ? "(rel)":"",
+ (inst >> 12) & 0x3, toswiz((inst >> 14) & 0x7), (inst >> 17) & 0x3,
+ (inst >> 19) & 0x3, toswiz((inst >> 21) & 0x7), (inst >> 24) & 0x3,
+ (inst >> 31) & 0x1);
+
+ fprintf(stderr,"\t5 RGBA_INST: 0x%08x:", code->inst[n].inst5);
+ inst = code->inst[n].inst5;
+ fprintf(stderr,"%s dest:%d%s rgb_C_src:%d %s/%s/%s %d alp_C_src:%d %s %d\n", toop(inst & 0xf),
+ (inst >> 4) & 0x7f, inst & (1<<11) ? "(rel)":"",
+ (inst >> 12) & 0x3, toswiz((inst >> 14) & 0x7), toswiz((inst >> 17) & 0x7), toswiz((inst >> 20) & 0x7),
+ (inst >> 23) & 0x3,
+ (inst >> 25) & 0x3, toswiz((inst >> 27) & 0x7), (inst >> 30) & 0x3);
+ break;
+ case 2:
+ break;
+ case 3:
+ inst = code->inst[n].inst1;
+ fprintf(stderr,"\t1:TEX_INST: 0x%08x: id: %d op:%s, %s, %s %s\n", inst, (inst >> 16) & 0xf,
+ to_texop((inst >> 22) & 0x7), (inst & (1<<25)) ? "ACQ" : "",
+ (inst & (1<<26)) ? "IGNUNC" : "", (inst & (1<<27)) ? "UNSCALED" : "SCALED");
+ inst = code->inst[n].inst2;
+ fprintf(stderr,"\t2:TEX_ADDR: 0x%08x: src: %d%s %s/%s/%s/%s dst: %d%s %s/%s/%s/%s\n", inst,
+ inst & 127, inst & (1<<7) ? "(rel)" : "",
+ toswiz((inst >> 8) & 0x3), toswiz((inst >> 10) & 0x3),
+ toswiz((inst >> 12) & 0x3), toswiz((inst >> 14) & 0x3),
+ (inst >> 16) & 127, inst & (1<<23) ? "(rel)" : "",
+ toswiz((inst >> 24) & 0x3), toswiz((inst >> 26) & 0x3),
+ toswiz((inst >> 28) & 0x3), toswiz((inst >> 30) & 0x3));
+
+ fprintf(stderr,"\t3:TEX_DXDY: 0x%08x\n", code->inst[n].inst3);
+ break;
+ }
+ fprintf(stderr,"\n");
+ }
+
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
new file mode 100644
index 0000000000..9091f65cd2
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h
@@ -0,0 +1,55 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * 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.
+ *
+ */
+
+/*
+ * Authors:
+ * Ben Skeggs <darktama@iinet.net.au>
+ * Jerome Glisse <j.glisse@gmail.com>
+ */
+#ifndef __R500_FRAGPROG_H_
+#define __R500_FRAGPROG_H_
+
+#include "shader/prog_parameter.h"
+#include "shader/prog_instruction.h"
+
+#include "radeon_compiler.h"
+#include "radeon_nqssadce.h"
+
+extern void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler);
+
+extern void r500FragmentProgramDump(struct rX00_fragment_program_code *c);
+
+extern GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg);
+
+extern void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src);
+
+extern GLboolean r500_transform_TEX(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void* data);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
new file mode 100644
index 0000000000..d694725c9b
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c
@@ -0,0 +1,299 @@
+/*
+ * Copyright (C) 2005 Ben Skeggs.
+ *
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Adaptation and modification for ATI/AMD Radeon R500 GPU chipsets.
+ *
+ * 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
+ *
+ * \author Ben Skeggs <darktama@iinet.net.au>
+ *
+ * \author Jerome Glisse <j.glisse@gmail.com>
+ *
+ * \author Corbin Simpson <MostAwesomeDude@gmail.com>
+ *
+ * \todo Depth write, WPOS/FOGC inputs
+ *
+ * \todo FogOption
+ *
+ */
+
+#include "r500_fragprog.h"
+
+#include "../r300_reg.h"
+
+#include "radeon_program_pair.h"
+
+
+#define PROG_CODE \
+ struct r300_fragment_program_compiler *c = (struct r300_fragment_program_compiler*)data; \
+ struct r500_fragment_program_code *code = &c->code->code.r500
+
+#define error(fmt, args...) do { \
+ rc_error(&c->Base, "%s::%s(): " fmt "\n", \
+ __FILE__, __FUNCTION__, ##args); \
+ } while(0)
+
+
+static GLuint translate_rgb_op(struct r300_fragment_program_compiler *c, GLuint opcode)
+{
+ switch(opcode) {
+ case OPCODE_CMP: return R500_ALU_RGBA_OP_CMP;
+ case OPCODE_DDX: return R500_ALU_RGBA_OP_MDH;
+ case OPCODE_DDY: return R500_ALU_RGBA_OP_MDV;
+ case OPCODE_DP3: return R500_ALU_RGBA_OP_DP3;
+ case OPCODE_DP4: return R500_ALU_RGBA_OP_DP4;
+ case OPCODE_FRC: return R500_ALU_RGBA_OP_FRC;
+ default:
+ error("translate_rgb_op(%d): unknown opcode\n", opcode);
+ /* fall through */
+ case OPCODE_NOP:
+ /* fall through */
+ case OPCODE_MAD: return R500_ALU_RGBA_OP_MAD;
+ case OPCODE_MAX: return R500_ALU_RGBA_OP_MAX;
+ case OPCODE_MIN: return R500_ALU_RGBA_OP_MIN;
+ case OPCODE_REPL_ALPHA: return R500_ALU_RGBA_OP_SOP;
+ }
+}
+
+static GLuint translate_alpha_op(struct r300_fragment_program_compiler *c, GLuint opcode)
+{
+ switch(opcode) {
+ case OPCODE_CMP: return R500_ALPHA_OP_CMP;
+ case OPCODE_COS: return R500_ALPHA_OP_COS;
+ case OPCODE_DDX: return R500_ALPHA_OP_MDH;
+ case OPCODE_DDY: return R500_ALPHA_OP_MDV;
+ case OPCODE_DP3: return R500_ALPHA_OP_DP;
+ case OPCODE_DP4: return R500_ALPHA_OP_DP;
+ case OPCODE_EX2: return R500_ALPHA_OP_EX2;
+ case OPCODE_FRC: return R500_ALPHA_OP_FRC;
+ case OPCODE_LG2: return R500_ALPHA_OP_LN2;
+ default:
+ error("translate_alpha_op(%d): unknown opcode\n", opcode);
+ /* fall through */
+ case OPCODE_NOP:
+ /* fall through */
+ case OPCODE_MAD: return R500_ALPHA_OP_MAD;
+ case OPCODE_MAX: return R500_ALPHA_OP_MAX;
+ case OPCODE_MIN: return R500_ALPHA_OP_MIN;
+ case OPCODE_RCP: return R500_ALPHA_OP_RCP;
+ case OPCODE_RSQ: return R500_ALPHA_OP_RSQ;
+ case OPCODE_SIN: return R500_ALPHA_OP_SIN;
+ }
+}
+
+static GLuint fix_hw_swizzle(GLuint swz)
+{
+ if (swz == 5) swz = 6;
+ if (swz == SWIZZLE_NIL) swz = 4;
+ return swz;
+}
+
+static GLuint translate_arg_rgb(struct radeon_pair_instruction *inst, int arg)
+{
+ GLuint t = inst->RGB.Arg[arg].Source;
+ int comp;
+ t |= inst->RGB.Arg[arg].Negate << 11;
+ t |= inst->RGB.Arg[arg].Abs << 12;
+
+ for(comp = 0; comp < 3; ++comp)
+ t |= fix_hw_swizzle(GET_SWZ(inst->RGB.Arg[arg].Swizzle, comp)) << (3*comp + 2);
+
+ return t;
+}
+
+static GLuint translate_arg_alpha(struct radeon_pair_instruction *inst, int i)
+{
+ GLuint t = inst->Alpha.Arg[i].Source;
+ t |= fix_hw_swizzle(inst->Alpha.Arg[i].Swizzle) << 2;
+ t |= inst->Alpha.Arg[i].Negate << 5;
+ t |= inst->Alpha.Arg[i].Abs << 6;
+ return t;
+}
+
+static void use_temporary(struct r500_fragment_program_code* code, GLuint index)
+{
+ if (index > code->max_temp_idx)
+ code->max_temp_idx = index;
+}
+
+static GLuint use_source(struct r500_fragment_program_code* code, struct radeon_pair_instruction_source src)
+{
+ if (!src.Constant)
+ use_temporary(code, src.Index);
+ return src.Index | src.Constant << 8;
+}
+
+
+/**
+ * Emit a paired ALU instruction.
+ */
+static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst)
+{
+ PROG_CODE;
+
+ if (code->inst_end >= 511) {
+ error("emit_alu: Too many instructions");
+ return GL_FALSE;
+ }
+
+ int ip = ++code->inst_end;
+
+ code->inst[ip].inst5 = translate_rgb_op(c, inst->RGB.Opcode);
+ code->inst[ip].inst4 = translate_alpha_op(c, inst->Alpha.Opcode);
+
+ if (inst->RGB.OutputWriteMask || inst->Alpha.OutputWriteMask || inst->Alpha.DepthWriteMask)
+ code->inst[ip].inst0 = R500_INST_TYPE_OUT;
+ else
+ code->inst[ip].inst0 = R500_INST_TYPE_ALU;
+ code->inst[ip].inst0 |= R500_INST_TEX_SEM_WAIT;
+
+ code->inst[ip].inst0 |= (inst->RGB.WriteMask << 11) | (inst->Alpha.WriteMask << 14);
+ code->inst[ip].inst0 |= (inst->RGB.OutputWriteMask << 15) | (inst->Alpha.OutputWriteMask << 18);
+ if (inst->Alpha.DepthWriteMask) {
+ code->inst[ip].inst4 |= R500_ALPHA_W_OMASK;
+ c->code->writes_depth = GL_TRUE;
+ }
+
+ code->inst[ip].inst4 |= R500_ALPHA_ADDRD(inst->Alpha.DestIndex);
+ code->inst[ip].inst5 |= R500_ALU_RGBA_ADDRD(inst->RGB.DestIndex);
+ use_temporary(code, inst->Alpha.DestIndex);
+ use_temporary(code, inst->RGB.DestIndex);
+
+ if (inst->RGB.Saturate)
+ code->inst[ip].inst0 |= R500_INST_RGB_CLAMP;
+ if (inst->Alpha.Saturate)
+ code->inst[ip].inst0 |= R500_INST_ALPHA_CLAMP;
+
+ code->inst[ip].inst1 |= R500_RGB_ADDR0(use_source(code, inst->RGB.Src[0]));
+ code->inst[ip].inst1 |= R500_RGB_ADDR1(use_source(code, inst->RGB.Src[1]));
+ code->inst[ip].inst1 |= R500_RGB_ADDR2(use_source(code, inst->RGB.Src[2]));
+
+ code->inst[ip].inst2 |= R500_ALPHA_ADDR0(use_source(code, inst->Alpha.Src[0]));
+ code->inst[ip].inst2 |= R500_ALPHA_ADDR1(use_source(code, inst->Alpha.Src[1]));
+ code->inst[ip].inst2 |= R500_ALPHA_ADDR2(use_source(code, inst->Alpha.Src[2]));
+
+ code->inst[ip].inst3 |= translate_arg_rgb(inst, 0) << R500_ALU_RGB_SEL_A_SHIFT;
+ code->inst[ip].inst3 |= translate_arg_rgb(inst, 1) << R500_ALU_RGB_SEL_B_SHIFT;
+ code->inst[ip].inst5 |= translate_arg_rgb(inst, 2) << R500_ALU_RGBA_SEL_C_SHIFT;
+
+ code->inst[ip].inst4 |= translate_arg_alpha(inst, 0) << R500_ALPHA_SEL_A_SHIFT;
+ code->inst[ip].inst4 |= translate_arg_alpha(inst, 1) << R500_ALPHA_SEL_B_SHIFT;
+ code->inst[ip].inst5 |= translate_arg_alpha(inst, 2) << R500_ALU_RGBA_ALPHA_SEL_C_SHIFT;
+
+ return GL_TRUE;
+}
+
+static GLuint translate_strq_swizzle(GLuint swizzle)
+{
+ GLuint swiz = 0;
+ int i;
+ for (i = 0; i < 4; i++)
+ swiz |= (GET_SWZ(swizzle, i) & 0x3) << i*2;
+ return swiz;
+}
+
+/**
+ * Emit a single TEX instruction
+ */
+static GLboolean emit_tex(void *data, struct radeon_pair_texture_instruction *inst)
+{
+ PROG_CODE;
+
+ if (code->inst_end >= 511) {
+ error("emit_tex: Too many instructions");
+ return GL_FALSE;
+ }
+
+ int ip = ++code->inst_end;
+
+ code->inst[ip].inst0 = R500_INST_TYPE_TEX
+ | (inst->WriteMask << 11)
+ | R500_INST_TEX_SEM_WAIT;
+ code->inst[ip].inst1 = R500_TEX_ID(inst->TexSrcUnit)
+ | R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED;
+
+ if (inst->TexSrcTarget == TEXTURE_RECT_INDEX)
+ code->inst[ip].inst1 |= R500_TEX_UNSCALED;
+
+ switch (inst->Opcode) {
+ case RADEON_OPCODE_KIL:
+ code->inst[ip].inst1 |= R500_TEX_INST_TEXKILL;
+ break;
+ case RADEON_OPCODE_TEX:
+ code->inst[ip].inst1 |= R500_TEX_INST_LD;
+ break;
+ case RADEON_OPCODE_TXB:
+ code->inst[ip].inst1 |= R500_TEX_INST_LODBIAS;
+ break;
+ case RADEON_OPCODE_TXP:
+ code->inst[ip].inst1 |= R500_TEX_INST_PROJ;
+ break;
+ default:
+ error("emit_tex can't handle opcode %x\n", inst->Opcode);
+ }
+
+ code->inst[ip].inst2 = R500_TEX_SRC_ADDR(inst->SrcIndex)
+ | (translate_strq_swizzle(inst->SrcSwizzle) << 8)
+ | R500_TEX_DST_ADDR(inst->DestIndex)
+ | R500_TEX_DST_R_SWIZ_R | R500_TEX_DST_G_SWIZ_G
+ | R500_TEX_DST_B_SWIZ_B | R500_TEX_DST_A_SWIZ_A;
+
+ return GL_TRUE;
+}
+
+static const struct radeon_pair_handler pair_handler = {
+ .EmitPaired = emit_paired,
+ .EmitTex = emit_tex,
+ .MaxHwTemps = 128
+};
+
+void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compiler)
+{
+ struct r500_fragment_program_code *code = &compiler->code->code.r500;
+
+ _mesa_bzero(code, sizeof(*code));
+ code->max_temp_idx = 1;
+ code->inst_end = -1;
+
+ radeonPairProgram(compiler, &pair_handler, compiler);
+ if (compiler->Base.Error)
+ return;
+
+ if ((code->inst[code->inst_end].inst0 & R500_INST_TYPE_MASK) != R500_INST_TYPE_OUT) {
+ /* This may happen when dead-code elimination is disabled or
+ * when most of the fragment program logic is leading to a KIL */
+ if (code->inst_end >= 511) {
+ rc_error(&compiler->Base, "Introducing fake OUT: Too many instructions");
+ return;
+ }
+
+ int ip = ++code->inst_end;
+ code->inst[ip].inst0 = R500_INST_TYPE_OUT | R500_INST_TEX_SEM_WAIT;
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.c b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
new file mode 100644
index 0000000000..c7923004df
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.c
@@ -0,0 +1,170 @@
+/*
+ * Copyright (C) 2009 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.
+ *
+ */
+
+#include "main/mtypes.h"
+#include "shader/prog_instruction.h"
+
+#include "radeon_code.h"
+
+void rc_constants_init(struct rc_constant_list * c)
+{
+ memset(c, 0, sizeof(*c));
+}
+
+/**
+ * Copy a constants structure, assuming that the destination structure
+ * is not initialized.
+ */
+void rc_constants_copy(struct rc_constant_list * dst, struct rc_constant_list * src)
+{
+ dst->Constants = malloc(sizeof(struct rc_constant) * src->Count);
+ memcpy(dst->Constants, src->Constants, sizeof(struct rc_constant) * src->Count);
+ dst->Count = src->Count;
+ dst->_Reserved = src->Count;
+}
+
+void rc_constants_destroy(struct rc_constant_list * c)
+{
+ free(c->Constants);
+ memset(c, 0, sizeof(*c));
+}
+
+unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * constant)
+{
+ unsigned index = c->Count;
+
+ if (c->Count >= c->_Reserved) {
+ struct rc_constant * newlist;
+
+ c->_Reserved = c->_Reserved * 2;
+ if (!c->_Reserved)
+ c->_Reserved = 16;
+
+ newlist = malloc(sizeof(struct rc_constant) * c->_Reserved);
+ memcpy(newlist, c->Constants, sizeof(struct rc_constant) * c->Count);
+
+ free(c->Constants);
+ c->Constants = newlist;
+ }
+
+ c->Constants[index] = *constant;
+ c->Count++;
+
+ return index;
+}
+
+
+/**
+ * Add a state vector to the constant list, while trying to avoid duplicates.
+ */
+unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state0, unsigned state1)
+{
+ unsigned index;
+ struct rc_constant constant;
+
+ for(index = 0; index < c->Count; ++index) {
+ if (c->Constants[index].Type == RC_CONSTANT_STATE) {
+ if (c->Constants[index].u.State[0] == state0 &&
+ c->Constants[index].u.State[1] == state1)
+ return index;
+ }
+ }
+
+ memset(&constant, 0, sizeof(constant));
+ constant.Type = RC_CONSTANT_STATE;
+ constant.Size = 4;
+ constant.u.State[0] = state0;
+ constant.u.State[1] = state1;
+
+ return rc_constants_add(c, &constant);
+}
+
+
+/**
+ * Add an immediate vector to the constant list, while trying to avoid
+ * duplicates.
+ */
+unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data)
+{
+ unsigned index;
+ struct rc_constant constant;
+
+ for(index = 0; index < c->Count; ++index) {
+ if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) {
+ if (!memcmp(c->Constants[index].u.Immediate, data, sizeof(float)*4))
+ return index;
+ }
+ }
+
+ memset(&constant, 0, sizeof(constant));
+ constant.Type = RC_CONSTANT_IMMEDIATE;
+ constant.Size = 4;
+ memcpy(constant.u.Immediate, data, sizeof(float) * 4);
+
+ return rc_constants_add(c, &constant);
+}
+
+
+/**
+ * Add an immediate scalar to the constant list, while trying to avoid
+ * duplicates.
+ */
+unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle)
+{
+ unsigned index;
+ int free_index = -1;
+ struct rc_constant constant;
+
+ for(index = 0; index < c->Count; ++index) {
+ if (c->Constants[index].Type == RC_CONSTANT_IMMEDIATE) {
+ for(unsigned comp = 0; comp < c->Constants[index].Size; ++comp) {
+ if (c->Constants[index].u.Immediate[comp] == data) {
+ *swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp);
+ return index;
+ }
+ }
+
+ if (c->Constants[index].Size < 4)
+ free_index = index;
+ }
+ }
+
+ if (free_index >= 0) {
+ unsigned comp = c->Constants[free_index].Size++;
+ c->Constants[free_index].u.Immediate[comp] = data;
+ *swizzle = MAKE_SWIZZLE4(comp, comp, comp, comp);
+ return free_index;
+ }
+
+ memset(&constant, 0, sizeof(constant));
+ constant.Type = RC_CONSTANT_IMMEDIATE;
+ constant.Size = 1;
+ constant.u.Immediate[0] = data;
+ *swizzle = SWIZZLE_XXXX;
+
+ return rc_constants_add(c, &constant);
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.h b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
new file mode 100644
index 0000000000..0806fb1b5c
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.h
@@ -0,0 +1,206 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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 RADEON_CODE_H
+#define RADEON_CODE_H
+
+#include <stdint.h>
+
+#define R300_PFS_MAX_ALU_INST 64
+#define R300_PFS_MAX_TEX_INST 32
+#define R300_PFS_MAX_TEX_INDIRECT 4
+#define R300_PFS_NUM_TEMP_REGS 32
+#define R300_PFS_NUM_CONST_REGS 32
+
+#define R500_PFS_MAX_INST 512
+#define R500_PFS_NUM_TEMP_REGS 128
+#define R500_PFS_NUM_CONST_REGS 256
+
+
+#define STATE_R300_WINDOW_DIMENSION (STATE_INTERNAL_DRIVER+0)
+
+enum {
+ /**
+ * External constants are constants whose meaning is unknown to this
+ * compiler. For example, a Mesa gl_program's constants are turned
+ * into external constants.
+ */
+ RC_CONSTANT_EXTERNAL = 0,
+
+ RC_CONSTANT_IMMEDIATE,
+
+ /**
+ * Constant referring to state that is known by this compiler,
+ * see RC_STATE_xxx, i.e. *not* arbitrary Mesa (or other) state.
+ */
+ RC_CONSTANT_STATE
+};
+
+enum {
+ RC_STATE_SHADOW_AMBIENT = 0,
+
+ RC_STATE_R300_WINDOW_DIMENSION,
+ RC_STATE_R300_TEXRECT_FACTOR
+};
+
+struct rc_constant {
+ unsigned Type:2; /**< RC_CONSTANT_xxx */
+ unsigned Size:3;
+
+ union {
+ unsigned External;
+ float Immediate[4];
+ unsigned State[2];
+ } u;
+};
+
+struct rc_constant_list {
+ struct rc_constant * Constants;
+ unsigned Count;
+
+ unsigned _Reserved;
+};
+
+void rc_constants_init(struct rc_constant_list * c);
+void rc_constants_copy(struct rc_constant_list * dst, struct rc_constant_list * src);
+void rc_constants_destroy(struct rc_constant_list * c);
+unsigned rc_constants_add(struct rc_constant_list * c, struct rc_constant * constant);
+unsigned rc_constants_add_state(struct rc_constant_list * c, unsigned state1, unsigned state2);
+unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const float * data);
+unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle);
+
+/**
+ * Stores state that influences the compilation of a fragment program.
+ */
+struct r300_fragment_program_external_state {
+ struct {
+ /**
+ * If the sampler is used as a shadow sampler,
+ * this field is:
+ * 0 - GL_LUMINANCE
+ * 1 - GL_INTENSITY
+ * 2 - GL_ALPHA
+ * depending on the depth texture mode.
+ */
+ unsigned depth_texture_mode : 2;
+
+ /**
+ * If the sampler is used as a shadow sampler,
+ * this field is (texture_compare_func - GL_NEVER).
+ * [e.g. if compare function is GL_LEQUAL, this field is 3]
+ *
+ * Otherwise, this field is 0.
+ */
+ unsigned texture_compare_func : 3;
+ } unit[16];
+};
+
+
+
+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; /**< total # of texture instructions used */
+ uint32_t inst[R300_PFS_MAX_TEX_INST];
+ } tex;
+
+ struct {
+ int length; /**< total # of ALU instructions used */
+ struct {
+ uint32_t rgb_inst;
+ uint32_t rgb_addr;
+ uint32_t alpha_inst;
+ uint32_t alpha_addr;
+ } inst[R300_PFS_MAX_ALU_INST];
+ } alu;
+
+ uint32_t config; /* US_CONFIG */
+ uint32_t pixsize; /* US_PIXSIZE */
+ uint32_t code_offset; /* US_CODE_OFFSET */
+ uint32_t code_addr[4]; /* US_CODE_ADDR */
+};
+
+
+struct r500_fragment_program_code {
+ struct {
+ uint32_t inst0;
+ uint32_t inst1;
+ uint32_t inst2;
+ uint32_t inst3;
+ uint32_t inst4;
+ uint32_t inst5;
+ } inst[R500_PFS_MAX_INST];
+
+ int inst_end; /* Number of instructions - 1; also, last instruction to be executed */
+
+ int max_temp_idx;
+};
+
+struct rX00_fragment_program_code {
+ union {
+ struct r300_fragment_program_code r300;
+ struct r500_fragment_program_code r500;
+ } code;
+
+ unsigned writes_depth:1;
+
+ struct rc_constant_list constants;
+};
+
+
+#define VSF_MAX_FRAGMENT_LENGTH (255*4)
+#define VSF_MAX_FRAGMENT_TEMPS (14)
+
+#define VSF_MAX_INPUTS 32
+#define VSF_MAX_OUTPUTS 32
+
+struct r300_vertex_program_code {
+ int length;
+ union {
+ uint32_t d[VSF_MAX_FRAGMENT_LENGTH];
+ float f[VSF_MAX_FRAGMENT_LENGTH];
+ } body;
+
+ int pos_end;
+ int num_temporaries; /* Number of temp vars used by program */
+ int inputs[VSF_MAX_INPUTS];
+ int outputs[VSF_MAX_OUTPUTS];
+
+ struct rc_constant_list constants;
+
+ uint32_t InputsRead;
+ uint32_t OutputsWritten;
+};
+
+void r300_vertex_program_dump(struct r300_vertex_program_code * vs);
+
+#endif /* RADEON_CODE_H */ \ No newline at end of file
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c
new file mode 100644
index 0000000000..da950d5289
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c
@@ -0,0 +1,262 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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. */
+
+#include "radeon_compiler.h"
+
+#include <stdarg.h>
+
+#include "radeon_program.h"
+
+
+void rc_init(struct radeon_compiler * c)
+{
+ memset(c, 0, sizeof(*c));
+
+ memory_pool_init(&c->Pool);
+ c->Program.Instructions.Prev = &c->Program.Instructions;
+ c->Program.Instructions.Next = &c->Program.Instructions;
+ c->Program.Instructions.I.Opcode = OPCODE_END;
+}
+
+void rc_destroy(struct radeon_compiler * c)
+{
+ rc_constants_destroy(&c->Program.Constants);
+ memory_pool_destroy(&c->Pool);
+ free(c->ErrorMsg);
+}
+
+void rc_debug(struct radeon_compiler * c, const char * fmt, ...)
+{
+ va_list ap;
+
+ if (!c->Debug)
+ return;
+
+ va_start(ap, fmt);
+ vfprintf(stderr, fmt, ap);
+ va_end(ap);
+}
+
+void rc_error(struct radeon_compiler * c, const char * fmt, ...)
+{
+ va_list ap;
+
+ c->Error = GL_TRUE;
+
+ if (!c->ErrorMsg) {
+ /* Only remember the first error */
+ char buf[1024];
+ int written;
+
+ va_start(ap, fmt);
+ written = vsnprintf(buf, sizeof(buf), fmt, ap);
+ va_end(ap);
+
+ if (written < sizeof(buf)) {
+ c->ErrorMsg = strdup(buf);
+ } else {
+ c->ErrorMsg = malloc(written + 1);
+
+ va_start(ap, fmt);
+ vsnprintf(c->ErrorMsg, written + 1, fmt, ap);
+ va_end(ap);
+ }
+ }
+
+ if (c->Debug) {
+ fprintf(stderr, "r300compiler error: ");
+
+ va_start(ap, fmt);
+ vfprintf(stderr, fmt, ap);
+ va_end(ap);
+ }
+}
+
+/**
+ * Rewrite the program such that everything that source the given input
+ * register will source new_input instead.
+ */
+void rc_move_input(struct radeon_compiler * c, unsigned input, struct prog_src_register new_input)
+{
+ struct rc_instruction * inst;
+
+ c->Program.InputsRead &= ~(1 << input);
+
+ for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) {
+ const unsigned numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode);
+ unsigned i;
+
+ for(i = 0; i < numsrcs; ++i) {
+ if (inst->I.SrcReg[i].File == PROGRAM_INPUT && inst->I.SrcReg[i].Index == input) {
+ inst->I.SrcReg[i].File = new_input.File;
+ inst->I.SrcReg[i].Index = new_input.Index;
+ inst->I.SrcReg[i].Swizzle = combine_swizzles(new_input.Swizzle, inst->I.SrcReg[i].Swizzle);
+ if (!inst->I.SrcReg[i].Abs) {
+ inst->I.SrcReg[i].Negate ^= new_input.Negate;
+ inst->I.SrcReg[i].Abs = new_input.Abs;
+ }
+
+ c->Program.InputsRead |= 1 << new_input.Index;
+ }
+ }
+ }
+}
+
+
+/**
+ * Rewrite the program such that everything that writes into the given
+ * output register will instead write to new_output. The new_output
+ * writemask is honoured.
+ */
+void rc_move_output(struct radeon_compiler * c, unsigned output, unsigned new_output, unsigned writemask)
+{
+ struct rc_instruction * inst;
+
+ c->Program.OutputsWritten &= ~(1 << output);
+
+ for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) {
+ const unsigned numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode);
+
+ if (numdsts) {
+ if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) {
+ inst->I.DstReg.Index = new_output;
+ inst->I.DstReg.WriteMask &= writemask;
+
+ c->Program.OutputsWritten |= 1 << new_output;
+ }
+ }
+ }
+}
+
+
+/**
+ * Rewrite the program such that a given output is duplicated.
+ */
+void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_output)
+{
+ unsigned tempreg = rc_find_free_temporary(c);
+ struct rc_instruction * inst;
+
+ for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) {
+ const unsigned numdsts = _mesa_num_inst_dst_regs(inst->I.Opcode);
+
+ if (numdsts) {
+ if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) {
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = tempreg;
+ }
+ }
+ }
+
+ inst = rc_insert_new_instruction(c, c->Program.Instructions.Prev);
+ inst->I.Opcode = OPCODE_MOV;
+ inst->I.DstReg.File = PROGRAM_OUTPUT;
+ inst->I.DstReg.Index = output;
+
+ inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[0].Index = tempreg;
+ inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+ inst = rc_insert_new_instruction(c, c->Program.Instructions.Prev);
+ inst->I.Opcode = OPCODE_MOV;
+ inst->I.DstReg.File = PROGRAM_OUTPUT;
+ inst->I.DstReg.Index = dup_output;
+
+ inst->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[0].Index = tempreg;
+ inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW;
+
+ c->Program.OutputsWritten |= 1 << dup_output;
+}
+
+
+/**
+ * Introduce standard code fragment to deal with fragment.position.
+ */
+void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsigned new_input)
+{
+ unsigned tempregi = rc_find_free_temporary(c);
+
+ c->Program.InputsRead &= ~(1 << wpos);
+ c->Program.InputsRead |= 1 << new_input;
+
+ /* perspective divide */
+ struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, &c->Program.Instructions);
+ inst_rcp->I.Opcode = OPCODE_RCP;
+
+ inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_rcp->I.DstReg.Index = tempregi;
+ inst_rcp->I.DstReg.WriteMask = WRITEMASK_W;
+
+ inst_rcp->I.SrcReg[0].File = PROGRAM_INPUT;
+ inst_rcp->I.SrcReg[0].Index = new_input;
+ inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW;
+
+ struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst_rcp);
+ inst_mul->I.Opcode = OPCODE_MUL;
+
+ inst_mul->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mul->I.DstReg.Index = tempregi;
+ inst_mul->I.DstReg.WriteMask = WRITEMASK_XYZ;
+
+ inst_mul->I.SrcReg[0].File = PROGRAM_INPUT;
+ inst_mul->I.SrcReg[0].Index = new_input;
+
+ inst_mul->I.SrcReg[1].File = PROGRAM_TEMPORARY;
+ inst_mul->I.SrcReg[1].Index = tempregi;
+ inst_mul->I.SrcReg[1].Swizzle = SWIZZLE_WWWW;
+
+ /* viewport transformation */
+ struct rc_instruction * inst_mad = rc_insert_new_instruction(c, inst_mul);
+ inst_mad->I.Opcode = OPCODE_MAD;
+
+ inst_mad->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst_mad->I.DstReg.Index = tempregi;
+ inst_mad->I.DstReg.WriteMask = WRITEMASK_XYZ;
+
+ inst_mad->I.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst_mad->I.SrcReg[0].Index = tempregi;
+ inst_mad->I.SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+ inst_mad->I.SrcReg[1].File = PROGRAM_STATE_VAR;
+ inst_mad->I.SrcReg[1].Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_R300_WINDOW_DIMENSION, 0);
+ inst_mad->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+ inst_mad->I.SrcReg[2].File = PROGRAM_STATE_VAR;
+ inst_mad->I.SrcReg[2].Index = inst_mad->I.SrcReg[1].Index;
+ inst_mad->I.SrcReg[2].Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
+
+ struct rc_instruction * inst;
+ for (inst = inst_mad->Next; inst != &c->Program.Instructions; inst = inst->Next) {
+ const unsigned numsrcs = _mesa_num_inst_src_regs(inst->I.Opcode);
+ unsigned i;
+
+ for(i = 0; i < numsrcs; i++) {
+ if (inst->I.SrcReg[i].File == PROGRAM_INPUT &&
+ inst->I.SrcReg[i].Index == wpos) {
+ inst->I.SrcReg[i].File = PROGRAM_TEMPORARY;
+ inst->I.SrcReg[i].Index = tempregi;
+ }
+ }
+ }
+}
+
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
new file mode 100644
index 0000000000..e63ab8840a
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
@@ -0,0 +1,108 @@
+/*
+ * Copyright 2009 Nicolai Hähnle <nhaehnle@gmail.com>
+ *
+ * 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
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHOR(S) AND/OR THEIR 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 RADEON_COMPILER_H
+#define RADEON_COMPILER_H
+
+#include "main/mtypes.h"
+#include "shader/prog_instruction.h"
+
+#include "memory_pool.h"
+#include "radeon_code.h"
+
+
+struct rc_instruction {
+ struct rc_instruction * Prev;
+ struct rc_instruction * Next;
+ struct prog_instruction I;
+};
+
+struct rc_program {
+ /**
+ * Instructions.Next points to the first instruction,
+ * Instructions.Prev points to the last instruction.
+ */
+ struct rc_instruction Instructions;
+
+ /* Long term, we should probably remove InputsRead & OutputsWritten,
+ * since updating dependent state can be fragile, and they aren't
+ * actually used very often. */
+ uint32_t InputsRead;
+ uint32_t OutputsWritten;
+ uint32_t ShadowSamplers; /**< Texture units used for shadow sampling. */
+
+ struct rc_constant_list Constants;
+};
+
+struct radeon_compiler {
+ struct memory_pool Pool;
+ struct rc_program Program;
+ unsigned Debug:1;
+ unsigned Error:1;
+ char * ErrorMsg;
+};
+
+void rc_init(struct radeon_compiler * c);
+void rc_destroy(struct radeon_compiler * c);
+
+void rc_debug(struct radeon_compiler * c, const char * fmt, ...);
+void rc_error(struct radeon_compiler * c, const char * fmt, ...);
+
+void rc_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * program);
+
+void rc_calculate_inputs_outputs(struct radeon_compiler * c);
+
+void rc_move_input(struct radeon_compiler * c, unsigned input, struct prog_src_register new_input);
+void rc_move_output(struct radeon_compiler * c, unsigned output, unsigned new_output, unsigned writemask);
+void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_output);
+void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsigned new_input);
+
+struct r300_fragment_program_compiler {
+ struct radeon_compiler Base;
+ struct rX00_fragment_program_code *code;
+ struct r300_fragment_program_external_state state;
+ unsigned is_r500;
+ unsigned OutputDepth;
+ unsigned OutputColor;
+
+ void * UserData;
+ void (*AllocateHwInputs)(
+ struct r300_fragment_program_compiler * c,
+ void (*allocate)(void * data, unsigned input, unsigned hwreg),
+ void * mydata);
+};
+
+void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c);
+
+
+struct r300_vertex_program_compiler {
+ struct radeon_compiler Base;
+ struct r300_vertex_program_code *code;
+ GLbitfield RequiredOutputs;
+
+ void * UserData;
+ void (*SetHwInputOutput)(struct r300_vertex_program_compiler * c);
+};
+
+void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* c);
+
+#endif /* RADEON_COMPILER_H */
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c
new file mode 100644
index 0000000000..aaaa50ad1f
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c
@@ -0,0 +1,294 @@
+/*
+ * 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
+ *
+ * "Not-quite SSA" and Dead-Code Elimination.
+ *
+ * @note This code uses SWIZZLE_NIL in a source register to indicate that
+ * the corresponding component is ignored by the corresponding instruction.
+ */
+
+#include "radeon_nqssadce.h"
+
+#include "radeon_compiler.h"
+
+
+/**
+ * Return the @ref register_state for the given register (or 0 for untracked
+ * registers, i.e. constants).
+ */
+static struct register_state *get_reg_state(struct nqssadce_state* s, GLuint file, GLuint index)
+{
+ 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;
+ }
+}
+
+
+/**
+ * Left multiplication of a register with a swizzle
+ *
+ * @note Works correctly only for X, Y, Z, W swizzles, not for constant swizzles.
+ */
+struct prog_src_register lmul_swizzle(GLuint swizzle, struct prog_src_register srcreg)
+{
+ struct prog_src_register tmp = srcreg;
+ int i;
+ tmp.Swizzle = 0;
+ tmp.Negate = NEGATE_NONE;
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(swizzle, i);
+ if (swz < 4) {
+ tmp.Swizzle |= GET_SWZ(srcreg.Swizzle, swz) << (i*3);
+ tmp.Negate |= GET_BIT(srcreg.Negate, swz) << i;
+ } else {
+ tmp.Swizzle |= swz << (i*3);
+ }
+ }
+ return tmp;
+}
+
+
+static void track_used_srcreg(struct nqssadce_state* s,
+ GLint src, GLuint sourced)
+{
+ struct prog_instruction * inst = &s->IP->I;
+ int i;
+ GLuint deswz_source = 0;
+
+ for(i = 0; i < 4; ++i) {
+ if (GET_BIT(sourced, i)) {
+ GLuint swz = GET_SWZ(inst->SrcReg[src].Swizzle, i);
+ deswz_source |= 1 << swz;
+ } else {
+ inst->SrcReg[src].Swizzle &= ~(7 << (3*i));
+ inst->SrcReg[src].Swizzle |= SWIZZLE_NIL << (3*i);
+ }
+ }
+
+ if (!s->Descr->IsNativeSwizzle(inst->Opcode, inst->SrcReg[src])) {
+ struct prog_dst_register dstreg = inst->DstReg;
+ dstreg.File = PROGRAM_TEMPORARY;
+ dstreg.Index = rc_find_free_temporary(s->Compiler);
+ dstreg.WriteMask = sourced;
+
+ s->Descr->BuildSwizzle(s, dstreg, inst->SrcReg[src]);
+
+ inst->SrcReg[src].File = PROGRAM_TEMPORARY;
+ inst->SrcReg[src].Index = dstreg.Index;
+ inst->SrcReg[src].Swizzle = 0;
+ inst->SrcReg[src].Negate = NEGATE_NONE;
+ inst->SrcReg[src].Abs = 0;
+ for(i = 0; i < 4; ++i) {
+ if (GET_BIT(sourced, i))
+ inst->SrcReg[src].Swizzle |= i << (3*i);
+ else
+ inst->SrcReg[src].Swizzle |= SWIZZLE_NIL << (3*i);
+ }
+ deswz_source = sourced;
+ }
+
+ struct register_state *regstate;
+
+ if (inst->SrcReg[src].RelAddr) {
+ regstate = get_reg_state(s, PROGRAM_ADDRESS, 0);
+ if (regstate)
+ regstate->Sourced |= WRITEMASK_X;
+ } else {
+ regstate = get_reg_state(s, inst->SrcReg[src].File, inst->SrcReg[src].Index);
+ if (regstate)
+ regstate->Sourced |= deswz_source & 0xf;
+ }
+}
+
+static void unalias_srcregs(struct rc_instruction *inst, GLuint oldindex, GLuint newindex)
+{
+ int nsrc = _mesa_num_inst_src_regs(inst->I.Opcode);
+ int i;
+ for(i = 0; i < nsrc; ++i)
+ if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY && inst->I.SrcReg[i].Index == oldindex)
+ inst->I.SrcReg[i].Index = newindex;
+}
+
+static void unalias_temporary(struct nqssadce_state* s, GLuint oldindex)
+{
+ GLuint newindex = rc_find_free_temporary(s->Compiler);
+ struct rc_instruction * inst;
+ for(inst = s->Compiler->Program.Instructions.Next; inst != s->IP; inst = inst->Next) {
+ if (inst->I.DstReg.File == PROGRAM_TEMPORARY && inst->I.DstReg.Index == oldindex)
+ inst->I.DstReg.Index = newindex;
+ unalias_srcregs(inst, oldindex, newindex);
+ }
+ unalias_srcregs(s->IP, oldindex, newindex);
+}
+
+
+/**
+ * Handle one instruction.
+ */
+static void process_instruction(struct nqssadce_state* s)
+{
+ struct prog_instruction *inst = &s->IP->I;
+ GLuint WriteMask;
+
+ if (inst->Opcode == OPCODE_END)
+ return;
+
+ if (inst->Opcode != OPCODE_KIL) {
+ struct register_state *regstate = get_reg_state(s, inst->DstReg.File, inst->DstReg.Index);
+ if (!regstate) {
+ rc_error(s->Compiler, "NqssaDce: bad destination register (%i[%i])\n",
+ inst->DstReg.File, inst->DstReg.Index);
+ return;
+ }
+
+ inst->DstReg.WriteMask &= regstate->Sourced;
+ regstate->Sourced &= ~inst->DstReg.WriteMask;
+
+ if (inst->DstReg.WriteMask == 0) {
+ struct rc_instruction * inst_remove = s->IP;
+ s->IP = s->IP->Prev;
+ rc_remove_instruction(inst_remove);
+ return;
+ }
+
+ if (inst->DstReg.File == PROGRAM_TEMPORARY && !regstate->Sourced)
+ unalias_temporary(s, inst->DstReg.Index);
+ }
+
+ WriteMask = inst->DstReg.WriteMask;
+
+ switch (inst->Opcode) {
+ case OPCODE_ARL:
+ case OPCODE_DDX:
+ case OPCODE_DDY:
+ case OPCODE_FRC:
+ case OPCODE_MOV:
+ track_used_srcreg(s, 0, WriteMask);
+ break;
+ case OPCODE_ADD:
+ case OPCODE_MAX:
+ case OPCODE_MIN:
+ case OPCODE_MUL:
+ case OPCODE_SGE:
+ case OPCODE_SLT:
+ track_used_srcreg(s, 0, WriteMask);
+ track_used_srcreg(s, 1, WriteMask);
+ break;
+ case OPCODE_CMP:
+ case OPCODE_MAD:
+ track_used_srcreg(s, 0, WriteMask);
+ track_used_srcreg(s, 1, WriteMask);
+ track_used_srcreg(s, 2, WriteMask);
+ break;
+ case OPCODE_COS:
+ case OPCODE_EX2:
+ case OPCODE_LG2:
+ case OPCODE_RCP:
+ case OPCODE_RSQ:
+ case OPCODE_SIN:
+ track_used_srcreg(s, 0, 0x1);
+ break;
+ case OPCODE_DP3:
+ track_used_srcreg(s, 0, 0x7);
+ track_used_srcreg(s, 1, 0x7);
+ break;
+ case OPCODE_DP4:
+ track_used_srcreg(s, 0, 0xf);
+ track_used_srcreg(s, 1, 0xf);
+ break;
+ case OPCODE_KIL:
+ case OPCODE_TEX:
+ case OPCODE_TXB:
+ case OPCODE_TXP:
+ track_used_srcreg(s, 0, 0xf);
+ break;
+ case OPCODE_DST:
+ track_used_srcreg(s, 0, 0x6);
+ track_used_srcreg(s, 1, 0xa);
+ break;
+ case OPCODE_EXP:
+ case OPCODE_LOG:
+ case OPCODE_POW:
+ track_used_srcreg(s, 0, 0x3);
+ break;
+ case OPCODE_LIT:
+ track_used_srcreg(s, 0, 0xb);
+ break;
+ default:
+ rc_error(s->Compiler, "NqssaDce: Unknown opcode %d\n", inst->Opcode);
+ return;
+ }
+
+ s->IP = s->IP->Prev;
+}
+
+void rc_calculate_inputs_outputs(struct radeon_compiler * c)
+{
+ struct rc_instruction *inst;
+
+ c->Program.InputsRead = 0;
+ c->Program.OutputsWritten = 0;
+
+ for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next)
+ {
+ int i;
+ int num_src_regs = _mesa_num_inst_src_regs(inst->I.Opcode);
+
+ for (i = 0; i < num_src_regs; ++i) {
+ if (inst->I.SrcReg[i].File == PROGRAM_INPUT)
+ c->Program.InputsRead |= 1 << inst->I.SrcReg[i].Index;
+ }
+
+ if (_mesa_num_inst_dst_regs(inst->I.Opcode)) {
+ if (inst->I.DstReg.File == PROGRAM_OUTPUT)
+ c->Program.OutputsWritten |= 1 << inst->I.DstReg.Index;
+ }
+ }
+}
+
+void radeonNqssaDce(struct radeon_compiler * c, struct radeon_nqssadce_descr* descr, void * data)
+{
+ struct nqssadce_state s;
+
+ _mesa_bzero(&s, sizeof(s));
+ s.Compiler = c;
+ s.Descr = descr;
+ s.UserData = data;
+ s.Descr->Init(&s);
+ s.IP = c->Program.Instructions.Prev;
+
+ while(s.IP != &c->Program.Instructions && !c->Error)
+ process_instruction(&s);
+
+ rc_calculate_inputs_outputs(c);
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h
new file mode 100644
index 0000000000..b3fc77a35a
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h
@@ -0,0 +1,91 @@
+/*
+ * 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 __RADEON_PROGRAM_NQSSADCE_H_
+#define __RADEON_PROGRAM_NQSSADCE_H_
+
+#include "radeon_program.h"
+
+struct register_state {
+ /**
+ * Bitmask indicating which components of the register are sourced
+ * by later instructions.
+ */
+ GLuint Sourced : 4;
+};
+
+/**
+ * Maintain state such as which registers are used, which registers are
+ * read from, etc.
+ */
+struct nqssadce_state {
+ struct radeon_compiler *Compiler;
+ struct radeon_nqssadce_descr *Descr;
+
+ /**
+ * All instructions after this instruction pointer have been dealt with.
+ */
+ struct rc_instruction * IP;
+
+ /**
+ * Which registers are read by subsequent instructions?
+ */
+ struct register_state Temps[MAX_PROGRAM_TEMPS];
+ struct register_state Outputs[VERT_RESULT_MAX];
+ struct register_state Address;
+
+ void * UserData;
+};
+
+
+/**
+ * This structure contains a description of the hardware in-so-far as
+ * it is required for the NqSSA-DCE pass.
+ */
+struct radeon_nqssadce_descr {
+ /**
+ * Fill in which outputs
+ */
+ void (*Init)(struct nqssadce_state *);
+
+ /**
+ * Check whether the given swizzle, absolute and negate combination
+ * can be implemented natively by the hardware for this opcode.
+ */
+ GLboolean (*IsNativeSwizzle)(GLuint opcode, struct prog_src_register reg);
+
+ /**
+ * Emit (at the current IP) the instruction MOV dst, src;
+ * The transformation will work recursively on the emitted instruction(s).
+ */
+ void (*BuildSwizzle)(struct nqssadce_state*, struct prog_dst_register dst, struct prog_src_register src);
+};
+
+void radeonNqssaDce(struct radeon_compiler * c, struct radeon_nqssadce_descr* descr, void * data);
+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/compiler/radeon_program.c b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
new file mode 100644
index 0000000000..208d3b90c8
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.c
@@ -0,0 +1,181 @@
+/*
+ * 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.
+ *
+ */
+
+#include "radeon_program.h"
+
+#include "radeon_compiler.h"
+#include "shader/prog_parameter.h"
+#include "shader/prog_print.h"
+
+
+/**
+ * Transform the given clause in the following way:
+ * 1. Replace it with an empty clause
+ * 2. For every instruction in the original clause, try the given
+ * transformations in order.
+ * 3. If one of the transformations returns GL_TRUE, assume that it
+ * has emitted the appropriate instruction(s) into the new clause;
+ * otherwise, copy the instruction verbatim.
+ *
+ * \note The transformation is currently not recursive; in other words,
+ * instructions emitted by transformations are not transformed.
+ *
+ * \note The transform is called 'local' because it can only look at
+ * one instruction at a time.
+ */
+void radeonLocalTransform(
+ struct radeon_compiler * c,
+ int num_transformations,
+ struct radeon_program_transformation* transformations)
+{
+ struct rc_instruction * inst = c->Program.Instructions.Next;
+
+ while(inst != &c->Program.Instructions) {
+ struct rc_instruction * current = inst;
+ int i;
+
+ inst = inst->Next;
+
+ for(i = 0; i < num_transformations; ++i) {
+ struct radeon_program_transformation* t = transformations + i;
+
+ if (t->function(c, current, t->userData))
+ break;
+ }
+ }
+}
+
+
+GLint rc_find_free_temporary(struct radeon_compiler * c)
+{
+ GLboolean used[MAX_PROGRAM_TEMPS];
+ GLuint i;
+
+ memset(used, 0, sizeof(used));
+
+ for (struct rc_instruction * rcinst = c->Program.Instructions.Next; rcinst != &c->Program.Instructions; rcinst = rcinst->Next) {
+ const struct prog_instruction *inst = &rcinst->I;
+ const GLuint n = _mesa_num_inst_src_regs(inst->Opcode);
+ GLuint k;
+
+ for (k = 0; k < n; k++) {
+ if (inst->SrcReg[k].File == PROGRAM_TEMPORARY)
+ used[inst->SrcReg[k].Index] = GL_TRUE;
+ }
+ }
+
+ for (i = 0; i < MAX_PROGRAM_TEMPS; i++) {
+ if (!used[i])
+ return i;
+ }
+
+ return -1;
+}
+
+
+struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c)
+{
+ struct rc_instruction * inst = memory_pool_malloc(&c->Pool, sizeof(struct rc_instruction));
+
+ inst->Prev = 0;
+ inst->Next = 0;
+
+ _mesa_init_instructions(&inst->I, 1);
+
+ return inst;
+}
+
+
+struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after)
+{
+ struct rc_instruction * inst = rc_alloc_instruction(c);
+
+ inst->Prev = after;
+ inst->Next = after->Next;
+
+ inst->Prev->Next = inst;
+ inst->Next->Prev = inst;
+
+ return inst;
+}
+
+void rc_remove_instruction(struct rc_instruction * inst)
+{
+ inst->Prev->Next = inst->Next;
+ inst->Next->Prev = inst->Prev;
+}
+
+
+void rc_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * program)
+{
+ struct prog_instruction *source;
+ unsigned int i;
+
+ for(source = program->Instructions; source->Opcode != OPCODE_END; ++source) {
+ struct rc_instruction * dest = rc_insert_new_instruction(c, c->Program.Instructions.Prev);
+ dest->I = *source;
+ }
+
+ c->Program.ShadowSamplers = program->ShadowSamplers;
+ c->Program.InputsRead = program->InputsRead;
+ c->Program.OutputsWritten = program->OutputsWritten;
+
+ for(i = 0; i < program->Parameters->NumParameters; ++i) {
+ struct rc_constant constant;
+
+ constant.Type = RC_CONSTANT_EXTERNAL;
+ constant.Size = 4;
+ constant.u.External = i;
+
+ rc_constants_add(&c->Program.Constants, &constant);
+ }
+}
+
+
+/**
+ * Print program to stderr, default options.
+ */
+void rc_print_program(const struct rc_program *prog)
+{
+ GLuint indent = 0;
+ GLuint linenum = 1;
+ struct rc_instruction *inst;
+
+ fprintf(stderr, "# Radeon Compiler Program\n");
+
+ for(inst = prog->Instructions.Next; inst != &prog->Instructions; inst = inst->Next) {
+ fprintf(stderr, "%3d: ", linenum);
+
+ /* Massive hack: We rely on the fact that the printers do not actually
+ * use the gl_program argument (last argument) in debug mode */
+ indent = _mesa_fprint_instruction_opt(
+ stderr, &inst->I,
+ indent, PROG_PRINT_DEBUG, 0);
+
+ linenum++;
+ }
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.h b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
new file mode 100644
index 0000000000..561958608c
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.h
@@ -0,0 +1,121 @@
+/*
+ * 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 __RADEON_PROGRAM_H_
+#define __RADEON_PROGRAM_H_
+
+#include "main/glheader.h"
+#include "main/macros.h"
+#include "main/enums.h"
+#include "shader/program.h"
+#include "shader/prog_instruction.h"
+
+struct radeon_compiler;
+struct rc_instruction;
+struct rc_program;
+
+enum {
+ PROGRAM_BUILTIN = PROGRAM_FILE_MAX /**< not a real register, but a special swizzle constant */
+};
+
+enum {
+ OPCODE_REPL_ALPHA = MAX_OPCODE /**< used in paired instructions */
+};
+
+#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;
+}
+
+static INLINE void reset_srcreg(struct prog_src_register* reg)
+{
+ _mesa_bzero(reg, sizeof(*reg));
+ reg->Swizzle = SWIZZLE_NOOP;
+}
+
+
+/**
+ * A transformation that can be passed to \ref radeonLocalTransform.
+ *
+ * The function will be called once for each instruction.
+ * It has to either emit the appropriate transformed code for the instruction
+ * and return GL_TRUE, or return GL_FALSE if it doesn't understand the
+ * instruction.
+ *
+ * The function gets passed the userData as last parameter.
+ */
+struct radeon_program_transformation {
+ GLboolean (*function)(
+ struct radeon_compiler*,
+ struct rc_instruction*,
+ void*);
+ void *userData;
+};
+
+void radeonLocalTransform(
+ struct radeon_compiler *c,
+ int num_transformations,
+ struct radeon_program_transformation* transformations);
+
+GLint rc_find_free_temporary(struct radeon_compiler * c);
+
+struct rc_instruction *rc_alloc_instruction(struct radeon_compiler * c);
+struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, struct rc_instruction * after);
+void rc_remove_instruction(struct rc_instruction * inst);
+
+void rc_print_program(const struct rc_program *prog);
+
+#endif
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
new file mode 100644
index 0000000000..609e510ff2
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c
@@ -0,0 +1,693 @@
+/*
+ * 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
+ *
+ * Shareable transformations that transform "special" ALU instructions
+ * into ALU instructions that are supported by hardware.
+ *
+ */
+
+#include "radeon_program_alu.h"
+
+#include "radeon_compiler.h"
+
+
+static struct rc_instruction *emit1(
+ struct radeon_compiler * c, struct rc_instruction * after,
+ gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
+ struct prog_src_register SrcReg)
+{
+ struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
+
+ fpi->I.Opcode = Opcode;
+ fpi->I.SaturateMode = Saturate;
+ fpi->I.DstReg = DstReg;
+ fpi->I.SrcReg[0] = SrcReg;
+ return fpi;
+}
+
+static struct rc_instruction *emit2(
+ struct radeon_compiler * c, struct rc_instruction * after,
+ gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
+ struct prog_src_register SrcReg0, struct prog_src_register SrcReg1)
+{
+ struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
+
+ fpi->I.Opcode = Opcode;
+ fpi->I.SaturateMode = Saturate;
+ fpi->I.DstReg = DstReg;
+ fpi->I.SrcReg[0] = SrcReg0;
+ fpi->I.SrcReg[1] = SrcReg1;
+ return fpi;
+}
+
+static struct rc_instruction *emit3(
+ struct radeon_compiler * c, struct rc_instruction * after,
+ gl_inst_opcode Opcode, GLuint Saturate, struct prog_dst_register DstReg,
+ struct prog_src_register SrcReg0, struct prog_src_register SrcReg1,
+ struct prog_src_register SrcReg2)
+{
+ struct rc_instruction *fpi = rc_insert_new_instruction(c, after);
+
+ fpi->I.Opcode = Opcode;
+ fpi->I.SaturateMode = Saturate;
+ fpi->I.DstReg = DstReg;
+ fpi->I.SrcReg[0] = SrcReg0;
+ fpi->I.SrcReg[1] = SrcReg1;
+ fpi->I.SrcReg[2] = SrcReg2;
+ return fpi;
+}
+
+static struct prog_dst_register dstreg(int file, int index)
+{
+ struct prog_dst_register dst;
+ dst.File = file;
+ dst.Index = index;
+ dst.WriteMask = WRITEMASK_XYZW;
+ dst.CondMask = COND_TR;
+ dst.CondSwizzle = SWIZZLE_NOOP;
+ dst.CondSrc = 0;
+ dst.pad = 0;
+ return dst;
+}
+
+static struct prog_dst_register dstregtmpmask(int index, int mask)
+{
+ struct prog_dst_register dst;
+ dst.File = PROGRAM_TEMPORARY;
+ dst.Index = index;
+ dst.WriteMask = mask;
+ dst.CondMask = COND_TR;
+ dst.CondSwizzle = SWIZZLE_NOOP;
+ dst.CondSrc = 0;
+ dst.pad = 0;
+ return dst;
+}
+
+static const struct prog_src_register builtin_zero = {
+ .File = PROGRAM_BUILTIN,
+ .Index = 0,
+ .Swizzle = SWIZZLE_0000
+};
+static const struct prog_src_register builtin_one = {
+ .File = PROGRAM_BUILTIN,
+ .Index = 0,
+ .Swizzle = SWIZZLE_1111
+};
+static const struct prog_src_register srcreg_undefined = {
+ .File = PROGRAM_UNDEFINED,
+ .Index = 0,
+ .Swizzle = SWIZZLE_NOOP
+};
+
+static struct prog_src_register srcreg(int file, int index)
+{
+ struct prog_src_register src = srcreg_undefined;
+ src.File = file;
+ src.Index = index;
+ return src;
+}
+
+static struct prog_src_register srcregswz(int file, int index, int swz)
+{
+ struct prog_src_register src = srcreg_undefined;
+ src.File = file;
+ src.Index = index;
+ src.Swizzle = swz;
+ return src;
+}
+
+static struct prog_src_register absolute(struct prog_src_register reg)
+{
+ struct prog_src_register newreg = reg;
+ newreg.Abs = 1;
+ newreg.Negate = NEGATE_NONE;
+ return newreg;
+}
+
+static struct prog_src_register negate(struct prog_src_register reg)
+{
+ struct prog_src_register newreg = reg;
+ newreg.Negate = newreg.Negate ^ NEGATE_XYZW;
+ return newreg;
+}
+
+static struct prog_src_register swizzle(struct prog_src_register reg, GLuint x, GLuint y, GLuint z, GLuint w)
+{
+ struct prog_src_register swizzled = reg;
+ swizzled.Swizzle = MAKE_SWIZZLE4(
+ x >= 4 ? x : GET_SWZ(reg.Swizzle, x),
+ y >= 4 ? y : GET_SWZ(reg.Swizzle, y),
+ z >= 4 ? z : GET_SWZ(reg.Swizzle, z),
+ w >= 4 ? w : GET_SWZ(reg.Swizzle, w));
+ return swizzled;
+}
+
+static struct prog_src_register scalar(struct prog_src_register reg)
+{
+ return swizzle(reg, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X);
+}
+
+static void transform_ABS(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ struct prog_src_register src = inst->I.SrcReg[0];
+ src.Abs = 1;
+ src.Negate = NEGATE_NONE;
+ emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode, inst->I.DstReg, src);
+ rc_remove_instruction(inst);
+}
+
+static void transform_DP3(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ struct prog_src_register src0 = inst->I.SrcReg[0];
+ struct prog_src_register src1 = inst->I.SrcReg[1];
+ src0.Negate &= ~NEGATE_W;
+ src0.Swizzle &= ~(7 << (3 * 3));
+ src0.Swizzle |= SWIZZLE_ZERO << (3 * 3);
+ src1.Negate &= ~NEGATE_W;
+ src1.Swizzle &= ~(7 << (3 * 3));
+ src1.Swizzle |= SWIZZLE_ZERO << (3 * 3);
+ emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, src1);
+ rc_remove_instruction(inst);
+}
+
+static void transform_DPH(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ struct prog_src_register src0 = inst->I.SrcReg[0];
+ src0.Negate &= ~NEGATE_W;
+ src0.Swizzle &= ~(7 << (3 * 3));
+ src0.Swizzle |= SWIZZLE_ONE << (3 * 3);
+ emit2(c, inst->Prev, OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, inst->I.SrcReg[1]);
+ rc_remove_instruction(inst);
+}
+
+/**
+ * [1, src0.y*src1.y, src0.z, src1.w]
+ * So basically MUL with lotsa swizzling.
+ */
+static void transform_DST(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ emit2(c, inst->Prev, OPCODE_MUL, inst->I.SaturateMode, inst->I.DstReg,
+ swizzle(inst->I.SrcReg[0], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE),
+ swizzle(inst->I.SrcReg[1], SWIZZLE_ONE, SWIZZLE_Y, SWIZZLE_ONE, SWIZZLE_W));
+ rc_remove_instruction(inst);
+}
+
+static void transform_FLR(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+ emit1(c, inst->Prev, OPCODE_FRC, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0]);
+ emit2(c, inst->Prev, OPCODE_ADD, inst->I.SaturateMode, inst->I.DstReg,
+ inst->I.SrcReg[0], negate(srcreg(PROGRAM_TEMPORARY, tempreg)));
+ rc_remove_instruction(inst);
+}
+
+/**
+ * Definition of LIT (from ARB_fragment_program):
+ *
+ * tmp = VectorLoad(op0);
+ * if (tmp.x < 0) tmp.x = 0;
+ * if (tmp.y < 0) tmp.y = 0;
+ * if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon);
+ * else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon;
+ * result.x = 1.0;
+ * result.y = tmp.x;
+ * result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0;
+ * result.w = 1.0;
+ *
+ * The longest path of computation is the one leading to result.z,
+ * consisting of 5 operations. This implementation of LIT takes
+ * 5 slots, if the subsequent optimization passes are clever enough
+ * to pair instructions correctly.
+ */
+static void transform_LIT(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ GLuint constant;
+ GLuint constant_swizzle;
+ GLuint temp;
+ struct prog_src_register srctemp;
+
+ constant = rc_constants_add_immediate_scalar(&c->Program.Constants, -127.999999, &constant_swizzle);
+
+ if (inst->I.DstReg.WriteMask != WRITEMASK_XYZW || inst->I.DstReg.File != PROGRAM_TEMPORARY) {
+ struct rc_instruction * inst_mov;
+
+ inst_mov = emit1(c, inst,
+ OPCODE_MOV, 0, inst->I.DstReg,
+ srcreg(PROGRAM_TEMPORARY, rc_find_free_temporary(c)));
+
+ inst->I.DstReg.File = PROGRAM_TEMPORARY;
+ inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index;
+ inst->I.DstReg.WriteMask = WRITEMASK_XYZW;
+ }
+
+ temp = inst->I.DstReg.Index;
+ srctemp = srcreg(PROGRAM_TEMPORARY, temp);
+
+ // tmp.x = max(0.0, Src.x);
+ // tmp.y = max(0.0, Src.y);
+ // tmp.w = clamp(Src.z, -128+eps, 128-eps);
+ emit2(c, inst->Prev, OPCODE_MAX, 0,
+ dstregtmpmask(temp, WRITEMASK_XYW),
+ inst->I.SrcReg[0],
+ swizzle(srcreg(PROGRAM_CONSTANT, constant),
+ SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ZERO, constant_swizzle&3));
+ emit2(c, inst->Prev, OPCODE_MIN, 0,
+ dstregtmpmask(temp, WRITEMASK_Z),
+ swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ negate(srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle)));
+
+ // tmp.w = Pow(tmp.y, tmp.w)
+ emit1(c, inst->Prev, OPCODE_LG2, 0,
+ dstregtmpmask(temp, WRITEMASK_W),
+ swizzle(srctemp, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y));
+ emit2(c, inst->Prev, OPCODE_MUL, 0,
+ dstregtmpmask(temp, WRITEMASK_W),
+ swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ swizzle(srctemp, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z));
+ emit1(c, inst->Prev, OPCODE_EX2, 0,
+ dstregtmpmask(temp, WRITEMASK_W),
+ swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
+
+ // tmp.z = (tmp.x > 0) ? tmp.w : 0.0
+ emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode,
+ dstregtmpmask(temp, WRITEMASK_Z),
+ negate(swizzle(srctemp, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
+ swizzle(srctemp, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ builtin_zero);
+
+ // tmp.x, tmp.y, tmp.w = 1.0, tmp.x, 1.0
+ emit1(c, inst->Prev, OPCODE_MOV, inst->I.SaturateMode,
+ dstregtmpmask(temp, WRITEMASK_XYW),
+ swizzle(srctemp, SWIZZLE_ONE, SWIZZLE_X, SWIZZLE_ONE, SWIZZLE_ONE));
+
+ rc_remove_instruction(inst);
+}
+
+static void transform_LRP(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+
+ emit2(c, inst->Prev, OPCODE_ADD, 0,
+ dstreg(PROGRAM_TEMPORARY, tempreg),
+ inst->I.SrcReg[1], negate(inst->I.SrcReg[2]));
+ emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode,
+ inst->I.DstReg,
+ inst->I.SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[2]);
+
+ rc_remove_instruction(inst);
+}
+
+static void transform_POW(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+ struct prog_dst_register tempdst = dstreg(PROGRAM_TEMPORARY, tempreg);
+ struct prog_src_register tempsrc = srcreg(PROGRAM_TEMPORARY, tempreg);
+ tempdst.WriteMask = WRITEMASK_W;
+ tempsrc.Swizzle = SWIZZLE_WWWW;
+
+ emit1(c, inst->Prev, OPCODE_LG2, 0, tempdst, scalar(inst->I.SrcReg[0]));
+ emit2(c, inst->Prev, OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->I.SrcReg[1]));
+ emit1(c, inst->Prev, OPCODE_EX2, inst->I.SaturateMode, inst->I.DstReg, tempsrc);
+
+ rc_remove_instruction(inst);
+}
+
+static void transform_RSQ(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ inst->I.SrcReg[0] = absolute(inst->I.SrcReg[0]);
+}
+
+static void transform_SGE(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+
+ emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1]));
+ emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg,
+ srcreg(PROGRAM_TEMPORARY, tempreg), builtin_zero, builtin_one);
+
+ rc_remove_instruction(inst);
+}
+
+static void transform_SLT(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+
+ emit2(c, inst->Prev, OPCODE_ADD, 0, dstreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1]));
+ emit3(c, inst->Prev, OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg,
+ srcreg(PROGRAM_TEMPORARY, tempreg), builtin_one, builtin_zero);
+
+ rc_remove_instruction(inst);
+}
+
+static void transform_SUB(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ inst->I.Opcode = OPCODE_ADD;
+ inst->I.SrcReg[1] = negate(inst->I.SrcReg[1]);
+}
+
+static void transform_SWZ(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ inst->I.Opcode = OPCODE_MOV;
+}
+
+static void transform_XPD(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ int tempreg = rc_find_free_temporary(c);
+
+ emit2(c, inst->Prev, OPCODE_MUL, 0, dstreg(PROGRAM_TEMPORARY, tempreg),
+ swizzle(inst->I.SrcReg[0], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
+ swizzle(inst->I.SrcReg[1], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W));
+ emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode, inst->I.DstReg,
+ swizzle(inst->I.SrcReg[0], SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_X, SWIZZLE_W),
+ swizzle(inst->I.SrcReg[1], SWIZZLE_Z, SWIZZLE_X, SWIZZLE_Y, SWIZZLE_W),
+ negate(srcreg(PROGRAM_TEMPORARY, tempreg)));
+
+ rc_remove_instruction(inst);
+}
+
+
+/**
+ * Can be used as a transformation for @ref radeonClauseLocalTransform,
+ * no userData necessary.
+ *
+ * Eliminates the following ALU instructions:
+ * ABS, DPH, DST, FLR, LIT, LRP, POW, SGE, SLT, SUB, SWZ, XPD
+ * using:
+ * MOV, ADD, MUL, MAD, FRC, DP3, LG2, EX2, CMP
+ *
+ * Transforms RSQ to Radeon's native RSQ by explicitly setting
+ * absolute value.
+ *
+ * @note should be applicable to R300 and R500 fragment programs.
+ */
+GLboolean radeonTransformALU(
+ struct radeon_compiler * c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ switch(inst->I.Opcode) {
+ case OPCODE_ABS: transform_ABS(c, inst); return GL_TRUE;
+ case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE;
+ case OPCODE_DST: transform_DST(c, inst); return GL_TRUE;
+ case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE;
+ case OPCODE_LIT: transform_LIT(c, inst); return GL_TRUE;
+ case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE;
+ case OPCODE_POW: transform_POW(c, inst); return GL_TRUE;
+ case OPCODE_RSQ: transform_RSQ(c, inst); return GL_TRUE;
+ case OPCODE_SGE: transform_SGE(c, inst); return GL_TRUE;
+ case OPCODE_SLT: transform_SLT(c, inst); return GL_TRUE;
+ case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE;
+ case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE;
+ case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE;
+ default:
+ return GL_FALSE;
+ }
+}
+
+
+static void transform_r300_vertex_ABS(struct radeon_compiler* c,
+ struct rc_instruction* inst)
+{
+ /* Note: r500 can take absolute values, but r300 cannot. */
+ inst->I.Opcode = OPCODE_MAX;
+ inst->I.SrcReg[1] = inst->I.SrcReg[0];
+ inst->I.SrcReg[1].Negate ^= NEGATE_XYZW;
+}
+
+/**
+ * For use with radeonLocalTransform, this transforms non-native ALU
+ * instructions of the r300 up to r500 vertex engine.
+ */
+GLboolean r300_transform_vertex_alu(
+ struct radeon_compiler * c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ switch(inst->I.Opcode) {
+ case OPCODE_ABS: transform_r300_vertex_ABS(c, inst); return GL_TRUE;
+ case OPCODE_DP3: transform_DP3(c, inst); return GL_TRUE;
+ case OPCODE_DPH: transform_DPH(c, inst); return GL_TRUE;
+ case OPCODE_FLR: transform_FLR(c, inst); return GL_TRUE;
+ case OPCODE_LRP: transform_LRP(c, inst); return GL_TRUE;
+ case OPCODE_SUB: transform_SUB(c, inst); return GL_TRUE;
+ case OPCODE_SWZ: transform_SWZ(c, inst); return GL_TRUE;
+ case OPCODE_XPD: transform_XPD(c, inst); return GL_TRUE;
+ default:
+ return GL_FALSE;
+ }
+}
+
+static void sincos_constants(struct radeon_compiler* c, GLuint *constants)
+{
+ static const GLfloat SinCosConsts[2][4] = {
+ {
+ 1.273239545, // 4/PI
+ -0.405284735, // -4/(PI*PI)
+ 3.141592654, // PI
+ 0.2225 // weight
+ },
+ {
+ 0.75,
+ 0.5,
+ 0.159154943, // 1/(2*PI)
+ 6.283185307 // 2*PI
+ }
+ };
+ int i;
+
+ for(i = 0; i < 2; ++i)
+ constants[i] = rc_constants_add_immediate_vec4(&c->Program.Constants, SinCosConsts[i]);
+}
+
+/**
+ * Approximate sin(x), where x is clamped to (-pi/2, pi/2).
+ *
+ * MUL tmp.xy, src, { 4/PI, -4/(PI^2) }
+ * MAD tmp.x, tmp.y, |src|, tmp.x
+ * MAD tmp.y, tmp.x, |tmp.x|, -tmp.x
+ * MAD dest, tmp.y, weight, tmp.x
+ */
+static void sin_approx(
+ struct radeon_compiler* c, struct rc_instruction * after,
+ struct prog_dst_register dst, struct prog_src_register src, const GLuint* constants)
+{
+ GLuint tempreg = rc_find_free_temporary(c);
+
+ emit2(c, after->Prev, OPCODE_MUL, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+ swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ srcreg(PROGRAM_CONSTANT, constants[0]));
+ emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_X),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+ absolute(swizzle(src, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+ emit3(c, after->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_Y),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ absolute(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)),
+ negate(swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X)));
+ emit3(c, after->Prev, OPCODE_MAD, 0, dst,
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+}
+
+/**
+ * Translate the trigonometric functions COS, SIN, and SCS
+ * using only the basic instructions
+ * MOV, ADD, MUL, MAD, FRC
+ */
+GLboolean radeonTransformTrigSimple(struct radeon_compiler* c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ if (inst->I.Opcode != OPCODE_COS &&
+ inst->I.Opcode != OPCODE_SIN &&
+ inst->I.Opcode != OPCODE_SCS)
+ return GL_FALSE;
+
+ GLuint constants[2];
+ GLuint tempreg = rc_find_free_temporary(c);
+
+ sincos_constants(c, constants);
+
+ if (inst->I.Opcode == OPCODE_COS) {
+ // MAD tmp.x, src, 1/(2*PI), 0.75
+ // FRC tmp.x, tmp.x
+ // MAD tmp.z, tmp.x, 2*PI, -PI
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X));
+ emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+ sin_approx(c, inst->Prev, inst->I.DstReg,
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ constants);
+ } else if (inst->I.Opcode == OPCODE_SIN) {
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y));
+ emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W));
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_W),
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+ sin_approx(c, inst->Prev, inst->I.DstReg,
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ constants);
+ } else {
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+ swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W));
+ emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+ srcreg(PROGRAM_TEMPORARY, tempreg));
+ emit3(c, inst->Prev, OPCODE_MAD, 0, dstregtmpmask(tempreg, WRITEMASK_XY),
+ srcreg(PROGRAM_TEMPORARY, tempreg),
+ swizzle(srcreg(PROGRAM_CONSTANT, constants[1]), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W),
+ negate(swizzle(srcreg(PROGRAM_CONSTANT, constants[0]), SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z, SWIZZLE_Z)));
+
+ struct prog_dst_register dst = inst->I.DstReg;
+
+ dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_X;
+ sin_approx(c, inst->Prev, dst,
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ constants);
+
+ dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_Y;
+ sin_approx(c, inst->Prev, dst,
+ swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y),
+ constants);
+ }
+
+ rc_remove_instruction(inst);
+
+ return GL_TRUE;
+}
+
+
+/**
+ * Transform the trigonometric functions COS, SIN, and SCS
+ * to include pre-scaling by 1/(2*PI) and taking the fractional
+ * part, so that the input to COS and SIN is always in the range [0,1).
+ * SCS is replaced by one COS and one SIN instruction.
+ *
+ * @warning This transformation implicitly changes the semantics of SIN and COS!
+ */
+GLboolean radeonTransformTrigScale(struct radeon_compiler* c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ if (inst->I.Opcode != OPCODE_COS &&
+ inst->I.Opcode != OPCODE_SIN &&
+ inst->I.Opcode != OPCODE_SCS)
+ return GL_FALSE;
+
+ static const GLfloat RCP_2PI = 0.15915494309189535;
+ GLuint temp;
+ GLuint constant;
+ GLuint constant_swizzle;
+
+ temp = rc_find_free_temporary(c);
+ constant = rc_constants_add_immediate_scalar(&c->Program.Constants, RCP_2PI, &constant_swizzle);
+
+ emit2(c, inst->Prev, OPCODE_MUL, 0, dstregtmpmask(temp, WRITEMASK_W),
+ swizzle(inst->I.SrcReg[0], SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
+ srcregswz(PROGRAM_CONSTANT, constant, constant_swizzle));
+ emit1(c, inst->Prev, OPCODE_FRC, 0, dstregtmpmask(temp, WRITEMASK_W),
+ srcreg(PROGRAM_TEMPORARY, temp));
+
+ if (inst->I.Opcode == OPCODE_COS) {
+ emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, inst->I.DstReg,
+ srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
+ } else if (inst->I.Opcode == OPCODE_SIN) {
+ emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode,
+ inst->I.DstReg, srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
+ } else if (inst->I.Opcode == OPCODE_SCS) {
+ struct prog_dst_register moddst = inst->I.DstReg;
+
+ if (inst->I.DstReg.WriteMask & WRITEMASK_X) {
+ moddst.WriteMask = WRITEMASK_X;
+ emit1(c, inst->Prev, OPCODE_COS, inst->I.SaturateMode, moddst,
+ srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
+ }
+ if (inst->I.DstReg.WriteMask & WRITEMASK_Y) {
+ moddst.WriteMask = WRITEMASK_Y;
+ emit1(c, inst->Prev, OPCODE_SIN, inst->I.SaturateMode, moddst,
+ srcregswz(PROGRAM_TEMPORARY, temp, SWIZZLE_WWWW));
+ }
+ }
+
+ rc_remove_instruction(inst);
+
+ return GL_TRUE;
+}
+
+/**
+ * Rewrite DDX/DDY instructions to properly work with r5xx shaders.
+ * The r5xx MDH/MDV instruction provides per-quad partial derivatives.
+ * It takes the form A*B+C. A and C are set by setting src0. B should be -1.
+ *
+ * @warning This explicitly changes the form of DDX and DDY!
+ */
+
+GLboolean radeonTransformDeriv(struct radeon_compiler* c,
+ struct rc_instruction* inst,
+ void* unused)
+{
+ if (inst->I.Opcode != OPCODE_DDX && inst->I.Opcode != OPCODE_DDY)
+ return GL_FALSE;
+
+ inst->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE);
+ inst->I.SrcReg[1].Negate = NEGATE_XYZW;
+
+ return GL_TRUE;
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h
new file mode 100644
index 0000000000..147efec6fc
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h
@@ -0,0 +1,58 @@
+/*
+ * 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 __RADEON_PROGRAM_ALU_H_
+#define __RADEON_PROGRAM_ALU_H_
+
+#include "radeon_program.h"
+
+GLboolean radeonTransformALU(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void*);
+
+GLboolean r300_transform_vertex_alu(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void*);
+
+GLboolean radeonTransformTrigSimple(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void*);
+
+GLboolean radeonTransformTrigScale(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void*);
+
+GLboolean radeonTransformDeriv(
+ struct radeon_compiler * c,
+ struct rc_instruction * inst,
+ void*);
+
+#endif /* __RADEON_PROGRAM_ALU_H_ */
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c
new file mode 100644
index 0000000000..48616ac461
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c
@@ -0,0 +1,958 @@
+/*
+ * 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
+ *
+ * Perform temporary register allocation and attempt to pair off instructions
+ * in RGB and Alpha pairs. Also attempts to optimize the TEX instruction
+ * vs. ALU instruction scheduling.
+ */
+
+#include "radeon_program_pair.h"
+
+#include "memory_pool.h"
+#include "radeon_compiler.h"
+#include "shader/prog_print.h"
+
+#define error(fmt, args...) do { \
+ rc_error(&s->Compiler->Base, "%s::%s(): " fmt "\n", \
+ __FILE__, __FUNCTION__, ##args); \
+} while(0)
+
+struct pair_state_instruction {
+ struct prog_instruction Instruction;
+ GLuint IP; /**< Position of this instruction in original program */
+
+ GLuint IsTex:1; /**< Is a texture instruction */
+ GLuint NeedRGB:1; /**< Needs the RGB ALU */
+ GLuint NeedAlpha:1; /**< Needs the Alpha ALU */
+ GLuint IsTranscendent:1; /**< Is a special transcendent instruction */
+
+ /**
+ * Number of (read and write) dependencies that must be resolved before
+ * this instruction can be scheduled.
+ */
+ GLuint NumDependencies:5;
+
+ /**
+ * Next instruction in the linked list of ready instructions.
+ */
+ struct pair_state_instruction *NextReady;
+
+ /**
+ * Values that this instruction writes
+ */
+ struct reg_value *Values[4];
+};
+
+
+/**
+ * Used to keep track of which instructions read a value.
+ */
+struct reg_value_reader {
+ struct pair_state_instruction *Reader;
+ struct reg_value_reader *Next;
+};
+
+/**
+ * Used to keep track which values are stored in each component of a
+ * PROGRAM_TEMPORARY.
+ */
+struct reg_value {
+ struct pair_state_instruction *Writer;
+ struct reg_value *Next; /**< Pointer to the next value to be written to the same PROGRAM_TEMPORARY component */
+
+ /**
+ * Unordered linked list of instructions that read from this value.
+ */
+ struct reg_value_reader *Readers;
+
+ /**
+ * Number of readers of this value. This is calculated during @ref scan_instructions
+ * and continually decremented during code emission.
+ * When this count reaches zero, the instruction that writes the @ref Next value
+ * can be scheduled.
+ */
+ GLuint NumReaders;
+};
+
+/**
+ * Used to translate a PROGRAM_INPUT or PROGRAM_TEMPORARY Mesa register
+ * to the proper hardware temporary.
+ */
+struct pair_register_translation {
+ GLuint Allocated:1;
+ GLuint HwIndex:8;
+ GLuint RefCount:23; /**< # of times this occurs in an unscheduled instruction SrcReg or DstReg */
+
+ /**
+ * Notes the value that is currently contained in each component
+ * (only used for PROGRAM_TEMPORARY registers).
+ */
+ struct reg_value *Value[4];
+};
+
+struct pair_state {
+ struct r300_fragment_program_compiler * Compiler;
+ const struct radeon_pair_handler *Handler;
+ GLboolean Verbose;
+ void *UserData;
+
+ /**
+ * Translate Mesa registers to hardware registers
+ */
+ struct pair_register_translation Inputs[FRAG_ATTRIB_MAX];
+ struct pair_register_translation Temps[MAX_PROGRAM_TEMPS];
+
+ struct {
+ GLuint RefCount; /**< # of times this occurs in an unscheduled SrcReg or DstReg */
+ } HwTemps[128];
+
+ /**
+ * Linked list of instructions that can be scheduled right now,
+ * based on which ALU/TEX resources they require.
+ */
+ struct pair_state_instruction *ReadyFullALU;
+ struct pair_state_instruction *ReadyRGB;
+ struct pair_state_instruction *ReadyAlpha;
+ struct pair_state_instruction *ReadyTEX;
+};
+
+
+static struct pair_register_translation *get_register(struct pair_state *s, GLuint file, GLuint index)
+{
+ switch(file) {
+ case PROGRAM_TEMPORARY: return &s->Temps[index];
+ case PROGRAM_INPUT: return &s->Inputs[index];
+ default: return 0;
+ }
+}
+
+static void alloc_hw_reg(struct pair_state *s, GLuint file, GLuint index, GLuint hwindex)
+{
+ struct pair_register_translation *t = get_register(s, file, index);
+ ASSERT(!s->HwTemps[hwindex].RefCount);
+ ASSERT(!t->Allocated);
+ s->HwTemps[hwindex].RefCount = t->RefCount;
+ t->Allocated = 1;
+ t->HwIndex = hwindex;
+}
+
+static GLuint get_hw_reg(struct pair_state *s, GLuint file, GLuint index)
+{
+ GLuint hwindex;
+
+ struct pair_register_translation *t = get_register(s, file, index);
+ if (!t) {
+ error("get_hw_reg: %i[%i]\n", file, index);
+ return 0;
+ }
+
+ if (t->Allocated)
+ return t->HwIndex;
+
+ for(hwindex = 0; hwindex < s->Handler->MaxHwTemps; ++hwindex)
+ if (!s->HwTemps[hwindex].RefCount)
+ break;
+
+ if (hwindex >= s->Handler->MaxHwTemps) {
+ error("Ran out of hardware temporaries");
+ return 0;
+ }
+
+ alloc_hw_reg(s, file, index, hwindex);
+ return hwindex;
+}
+
+
+static void deref_hw_reg(struct pair_state *s, GLuint hwindex)
+{
+ if (!s->HwTemps[hwindex].RefCount) {
+ error("Hwindex %i refcount error", hwindex);
+ return;
+ }
+
+ s->HwTemps[hwindex].RefCount--;
+}
+
+static void add_pairinst_to_list(struct pair_state_instruction **list, struct pair_state_instruction *pairinst)
+{
+ pairinst->NextReady = *list;
+ *list = pairinst;
+}
+
+/**
+ * The given instruction has become ready. Link it into the ready
+ * instructions.
+ */
+static void instruction_ready(struct pair_state *s, struct pair_state_instruction *pairinst)
+{
+ if (s->Verbose)
+ _mesa_printf("instruction_ready(%i)\n", pairinst->IP);
+
+ if (pairinst->IsTex)
+ add_pairinst_to_list(&s->ReadyTEX, pairinst);
+ else if (!pairinst->NeedAlpha)
+ add_pairinst_to_list(&s->ReadyRGB, pairinst);
+ else if (!pairinst->NeedRGB)
+ add_pairinst_to_list(&s->ReadyAlpha, pairinst);
+ else
+ add_pairinst_to_list(&s->ReadyFullALU, pairinst);
+}
+
+
+/**
+ * Finally rewrite ADD, MOV, MUL as the appropriate native instruction
+ * and reverse the order of arguments for CMP.
+ */
+static void final_rewrite(struct pair_state *s, struct prog_instruction *inst)
+{
+ struct prog_src_register tmp;
+
+ switch(inst->Opcode) {
+ case OPCODE_ADD:
+ inst->SrcReg[2] = inst->SrcReg[1];
+ inst->SrcReg[1].File = PROGRAM_BUILTIN;
+ inst->SrcReg[1].Swizzle = SWIZZLE_1111;
+ inst->SrcReg[1].Negate = NEGATE_NONE;
+ inst->Opcode = OPCODE_MAD;
+ break;
+ case OPCODE_CMP:
+ tmp = inst->SrcReg[2];
+ inst->SrcReg[2] = inst->SrcReg[0];
+ inst->SrcReg[0] = tmp;
+ break;
+ case OPCODE_MOV:
+ /* AMD say we should use CMP.
+ * However, when we transform
+ * KIL -r0;
+ * into
+ * CMP tmp, -r0, -r0, 0;
+ * KIL tmp;
+ * we get incorrect behaviour on R500 when r0 == 0.0.
+ * It appears that the R500 KIL hardware treats -0.0 as less
+ * than zero.
+ */
+ inst->SrcReg[1].File = PROGRAM_BUILTIN;
+ inst->SrcReg[1].Swizzle = SWIZZLE_1111;
+ inst->SrcReg[2].File = PROGRAM_BUILTIN;
+ inst->SrcReg[2].Swizzle = SWIZZLE_0000;
+ inst->Opcode = OPCODE_MAD;
+ break;
+ case OPCODE_MUL:
+ inst->SrcReg[2].File = PROGRAM_BUILTIN;
+ inst->SrcReg[2].Swizzle = SWIZZLE_0000;
+ inst->Opcode = OPCODE_MAD;
+ break;
+ default:
+ /* nothing to do */
+ break;
+ }
+}
+
+
+/**
+ * Classify an instruction according to which ALUs etc. it needs
+ */
+static void classify_instruction(struct pair_state *s,
+ struct pair_state_instruction *psi)
+{
+ psi->NeedRGB = (psi->Instruction.DstReg.WriteMask & WRITEMASK_XYZ) ? 1 : 0;
+ psi->NeedAlpha = (psi->Instruction.DstReg.WriteMask & WRITEMASK_W) ? 1 : 0;
+
+ switch(psi->Instruction.Opcode) {
+ case OPCODE_ADD:
+ case OPCODE_CMP:
+ case OPCODE_DDX:
+ case OPCODE_DDY:
+ case OPCODE_FRC:
+ case OPCODE_MAD:
+ case OPCODE_MAX:
+ case OPCODE_MIN:
+ case OPCODE_MOV:
+ case OPCODE_MUL:
+ break;
+ case OPCODE_COS:
+ case OPCODE_EX2:
+ case OPCODE_LG2:
+ case OPCODE_RCP:
+ case OPCODE_RSQ:
+ case OPCODE_SIN:
+ psi->IsTranscendent = 1;
+ psi->NeedAlpha = 1;
+ break;
+ case OPCODE_DP4:
+ psi->NeedAlpha = 1;
+ /* fall through */
+ case OPCODE_DP3:
+ psi->NeedRGB = 1;
+ break;
+ case OPCODE_KIL:
+ case OPCODE_TEX:
+ case OPCODE_TXB:
+ case OPCODE_TXP:
+ case OPCODE_END:
+ psi->IsTex = 1;
+ break;
+ default:
+ error("Unknown opcode %d\n", psi->Instruction.Opcode);
+ break;
+ }
+}
+
+
+/**
+ * Count which (input, temporary) register is read and written how often,
+ * and scan the instruction stream to find dependencies.
+ */
+static void scan_instructions(struct pair_state *s)
+{
+ struct rc_instruction *source;
+ GLuint ip;
+
+ for(source = s->Compiler->Base.Program.Instructions.Next, ip = 0;
+ source != &s->Compiler->Base.Program.Instructions;
+ source = source->Next, ++ip) {
+ struct pair_state_instruction *pairinst = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*pairinst));
+ memset(pairinst, 0, sizeof(struct pair_state_instruction));
+
+ pairinst->Instruction = source->I;
+ pairinst->IP = ip;
+ final_rewrite(s, &pairinst->Instruction);
+ classify_instruction(s, pairinst);
+
+ int nsrc = _mesa_num_inst_src_regs(pairinst->Instruction.Opcode);
+ int j;
+ for(j = 0; j < nsrc; j++) {
+ struct pair_register_translation *t =
+ get_register(s, pairinst->Instruction.SrcReg[j].File, pairinst->Instruction.SrcReg[j].Index);
+ if (!t)
+ continue;
+
+ t->RefCount++;
+
+ if (pairinst->Instruction.SrcReg[j].File == PROGRAM_TEMPORARY) {
+ int i;
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(pairinst->Instruction.SrcReg[j].Swizzle, i);
+ if (swz >= 4)
+ continue; /* constant or NIL swizzle */
+ if (!t->Value[swz])
+ continue; /* this is an undefined read */
+
+ /* Do not add a dependency if this instruction
+ * also rewrites the value. The code below adds
+ * a dependency for the DstReg, which is a superset
+ * of the SrcReg dependency. */
+ if (pairinst->Instruction.DstReg.File == PROGRAM_TEMPORARY &&
+ pairinst->Instruction.DstReg.Index == pairinst->Instruction.SrcReg[j].Index &&
+ GET_BIT(pairinst->Instruction.DstReg.WriteMask, swz))
+ continue;
+
+ struct reg_value_reader* r = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*r));
+ pairinst->NumDependencies++;
+ t->Value[swz]->NumReaders++;
+ r->Reader = pairinst;
+ r->Next = t->Value[swz]->Readers;
+ t->Value[swz]->Readers = r;
+ }
+ }
+ }
+
+ int ndst = _mesa_num_inst_dst_regs(pairinst->Instruction.Opcode);
+ if (ndst) {
+ struct pair_register_translation *t =
+ get_register(s, pairinst->Instruction.DstReg.File, pairinst->Instruction.DstReg.Index);
+ if (t) {
+ t->RefCount++;
+
+ if (pairinst->Instruction.DstReg.File == PROGRAM_TEMPORARY) {
+ int j;
+ for(j = 0; j < 4; ++j) {
+ if (!GET_BIT(pairinst->Instruction.DstReg.WriteMask, j))
+ continue;
+
+ struct reg_value* v = memory_pool_malloc(&s->Compiler->Base.Pool, sizeof(*v));
+ memset(v, 0, sizeof(struct reg_value));
+ v->Writer = pairinst;
+ if (t->Value[j]) {
+ pairinst->NumDependencies++;
+ t->Value[j]->Next = v;
+ }
+ t->Value[j] = v;
+ pairinst->Values[j] = v;
+ }
+ }
+ }
+ }
+
+ if (s->Verbose)
+ _mesa_printf("scan(%i): NumDeps = %i\n", ip, pairinst->NumDependencies);
+
+ if (!pairinst->NumDependencies)
+ instruction_ready(s, pairinst);
+ }
+
+ /* Clear the PROGRAM_TEMPORARY state */
+ int i, j;
+ for(i = 0; i < MAX_PROGRAM_TEMPS; ++i) {
+ for(j = 0; j < 4; ++j)
+ s->Temps[i].Value[j] = 0;
+ }
+}
+
+
+static void decrement_dependencies(struct pair_state *s, struct pair_state_instruction *pairinst)
+{
+ ASSERT(pairinst->NumDependencies > 0);
+ if (!--pairinst->NumDependencies)
+ instruction_ready(s, pairinst);
+}
+
+/**
+ * Update the dependency tracking state based on what the instruction
+ * at the given IP does.
+ */
+static void commit_instruction(struct pair_state *s, struct pair_state_instruction *pairinst)
+{
+ struct prog_instruction *inst = &pairinst->Instruction;
+
+ if (s->Verbose)
+ _mesa_printf("commit_instruction(%i)\n", pairinst->IP);
+
+ if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+ struct pair_register_translation *t = &s->Temps[inst->DstReg.Index];
+ deref_hw_reg(s, t->HwIndex);
+
+ int i;
+ for(i = 0; i < 4; ++i) {
+ if (!GET_BIT(inst->DstReg.WriteMask, i))
+ continue;
+
+ t->Value[i] = pairinst->Values[i];
+ if (t->Value[i]->NumReaders) {
+ struct reg_value_reader *r;
+ for(r = pairinst->Values[i]->Readers; r; r = r->Next)
+ decrement_dependencies(s, r->Reader);
+ } else if (t->Value[i]->Next) {
+ /* This happens when the only reader writes
+ * the register at the same time */
+ decrement_dependencies(s, t->Value[i]->Next->Writer);
+ }
+ }
+ }
+
+ int nsrc = _mesa_num_inst_src_regs(inst->Opcode);
+ int i;
+ for(i = 0; i < nsrc; i++) {
+ struct pair_register_translation *t = get_register(s, inst->SrcReg[i].File, inst->SrcReg[i].Index);
+ if (!t)
+ continue;
+
+ deref_hw_reg(s, get_hw_reg(s, inst->SrcReg[i].File, inst->SrcReg[i].Index));
+
+ if (inst->SrcReg[i].File != PROGRAM_TEMPORARY)
+ continue;
+
+ int j;
+ for(j = 0; j < 4; ++j) {
+ GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j);
+ if (swz >= 4)
+ continue;
+ if (!t->Value[swz])
+ continue;
+
+ /* Do not free a dependency if this instruction
+ * also rewrites the value. See scan_instructions. */
+ if (inst->DstReg.File == PROGRAM_TEMPORARY &&
+ inst->DstReg.Index == inst->SrcReg[i].Index &&
+ GET_BIT(inst->DstReg.WriteMask, swz))
+ continue;
+
+ if (!--t->Value[swz]->NumReaders) {
+ if (t->Value[swz]->Next)
+ decrement_dependencies(s, t->Value[swz]->Next->Writer);
+ }
+ }
+ }
+}
+
+
+/**
+ * Emit all ready texture instructions in a single block.
+ *
+ * Emit as a single block to (hopefully) sample many textures in parallel,
+ * and to avoid hardware indirections on R300.
+ *
+ * In R500, we don't really know when the result of a texture instruction
+ * arrives. So allocate all destinations first, to make sure they do not
+ * arrive early and overwrite a texture coordinate we're going to use later
+ * in the block.
+ */
+static void emit_all_tex(struct pair_state *s)
+{
+ struct pair_state_instruction *readytex;
+ struct pair_state_instruction *pairinst;
+
+ ASSERT(s->ReadyTEX);
+
+ // Don't let the ready list change under us!
+ readytex = s->ReadyTEX;
+ s->ReadyTEX = 0;
+
+ // Allocate destination hardware registers in one block to avoid conflicts.
+ for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) {
+ struct prog_instruction *inst = &pairinst->Instruction;
+ if (inst->Opcode != OPCODE_KIL)
+ get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
+ }
+
+ if (s->Compiler->Base.Debug)
+ _mesa_printf(" BEGIN_TEX\n");
+
+ if (s->Handler->BeginTexBlock)
+ s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->BeginTexBlock(s->UserData);
+
+ for(pairinst = readytex; pairinst; pairinst = pairinst->NextReady) {
+ struct prog_instruction *inst = &pairinst->Instruction;
+ commit_instruction(s, pairinst);
+
+ if (inst->Opcode != OPCODE_KIL)
+ inst->DstReg.Index = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
+ inst->SrcReg[0].Index = get_hw_reg(s, inst->SrcReg[0].File, inst->SrcReg[0].Index);
+
+ if (s->Compiler->Base.Debug) {
+ _mesa_printf(" ");
+ _mesa_print_instruction(inst);
+ fflush(stdout);
+ }
+
+ struct radeon_pair_texture_instruction rpti;
+
+ switch(inst->Opcode) {
+ case OPCODE_TEX: rpti.Opcode = RADEON_OPCODE_TEX; break;
+ case OPCODE_TXB: rpti.Opcode = RADEON_OPCODE_TXB; break;
+ case OPCODE_TXP: rpti.Opcode = RADEON_OPCODE_TXP; break;
+ default:
+ case OPCODE_KIL: rpti.Opcode = RADEON_OPCODE_KIL; break;
+ }
+
+ rpti.DestIndex = inst->DstReg.Index;
+ rpti.WriteMask = inst->DstReg.WriteMask;
+ rpti.TexSrcUnit = inst->TexSrcUnit;
+ rpti.TexSrcTarget = inst->TexSrcTarget;
+ rpti.SrcIndex = inst->SrcReg[0].Index;
+ rpti.SrcSwizzle = inst->SrcReg[0].Swizzle;
+
+ s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->EmitTex(s->UserData, &rpti);
+ }
+
+ if (s->Compiler->Base.Debug)
+ _mesa_printf(" END_TEX\n");
+}
+
+
+static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instruction *pair,
+ struct prog_src_register src, GLboolean rgb, GLboolean alpha)
+{
+ int candidate = -1;
+ int candidate_quality = -1;
+ int i;
+
+ if (!rgb && !alpha)
+ return 0;
+
+ GLuint constant;
+ GLuint index;
+
+ if (src.File == PROGRAM_TEMPORARY || src.File == PROGRAM_INPUT) {
+ constant = 0;
+ index = get_hw_reg(s, src.File, src.Index);
+ } else {
+ constant = 1;
+ index = src.Index;
+ }
+
+ for(i = 0; i < 3; ++i) {
+ int q = 0;
+ if (rgb) {
+ if (pair->RGB.Src[i].Used) {
+ if (pair->RGB.Src[i].Constant != constant ||
+ pair->RGB.Src[i].Index != index)
+ continue;
+ q++;
+ }
+ }
+ if (alpha) {
+ if (pair->Alpha.Src[i].Used) {
+ if (pair->Alpha.Src[i].Constant != constant ||
+ pair->Alpha.Src[i].Index != index)
+ continue;
+ q++;
+ }
+ }
+ if (q > candidate_quality) {
+ candidate_quality = q;
+ candidate = i;
+ }
+ }
+
+ if (candidate >= 0) {
+ if (rgb) {
+ pair->RGB.Src[candidate].Used = 1;
+ pair->RGB.Src[candidate].Constant = constant;
+ pair->RGB.Src[candidate].Index = index;
+ }
+ if (alpha) {
+ pair->Alpha.Src[candidate].Used = 1;
+ pair->Alpha.Src[candidate].Constant = constant;
+ pair->Alpha.Src[candidate].Index = index;
+ }
+ }
+
+ return candidate;
+}
+
+/**
+ * Fill the given ALU instruction's opcodes and source operands into the given pair,
+ * if possible.
+ */
+static GLboolean fill_instruction_into_pair(
+ struct pair_state *s,
+ struct radeon_pair_instruction *pair,
+ struct pair_state_instruction *pairinst)
+{
+ struct prog_instruction *inst = &pairinst->Instruction;
+
+ ASSERT(!pairinst->NeedRGB || pair->RGB.Opcode == OPCODE_NOP);
+ ASSERT(!pairinst->NeedAlpha || pair->Alpha.Opcode == OPCODE_NOP);
+
+ if (pairinst->NeedRGB) {
+ if (pairinst->IsTranscendent)
+ pair->RGB.Opcode = OPCODE_REPL_ALPHA;
+ else
+ pair->RGB.Opcode = inst->Opcode;
+ if (inst->SaturateMode == SATURATE_ZERO_ONE)
+ pair->RGB.Saturate = 1;
+ }
+ if (pairinst->NeedAlpha) {
+ pair->Alpha.Opcode = inst->Opcode;
+ if (inst->SaturateMode == SATURATE_ZERO_ONE)
+ pair->Alpha.Saturate = 1;
+ }
+
+ int nargs = _mesa_num_inst_src_regs(inst->Opcode);
+ int i;
+
+ /* Special case for DDX/DDY (MDH/MDV). */
+ if (inst->Opcode == OPCODE_DDX || inst->Opcode == OPCODE_DDY) {
+ if (pair->RGB.Src[0].Used || pair->Alpha.Src[0].Used)
+ return GL_FALSE;
+ else
+ nargs++;
+ }
+
+ for(i = 0; i < nargs; ++i) {
+ int source;
+ if (pairinst->NeedRGB && !pairinst->IsTranscendent) {
+ GLboolean srcrgb = GL_FALSE;
+ GLboolean srcalpha = GL_FALSE;
+ int j;
+ for(j = 0; j < 3; ++j) {
+ GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j);
+ if (swz < 3)
+ srcrgb = GL_TRUE;
+ else if (swz < 4)
+ srcalpha = GL_TRUE;
+ }
+ source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha);
+ if (source < 0)
+ return GL_FALSE;
+ pair->RGB.Arg[i].Source = source;
+ pair->RGB.Arg[i].Swizzle = inst->SrcReg[i].Swizzle & 0x1ff;
+ pair->RGB.Arg[i].Abs = inst->SrcReg[i].Abs;
+ pair->RGB.Arg[i].Negate = !!(inst->SrcReg[i].Negate & (NEGATE_X | NEGATE_Y | NEGATE_Z));
+ }
+ if (pairinst->NeedAlpha) {
+ GLboolean srcrgb = GL_FALSE;
+ GLboolean srcalpha = GL_FALSE;
+ GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, pairinst->IsTranscendent ? 0 : 3);
+ if (swz < 3)
+ srcrgb = GL_TRUE;
+ else if (swz < 4)
+ srcalpha = GL_TRUE;
+ source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha);
+ if (source < 0)
+ return GL_FALSE;
+ pair->Alpha.Arg[i].Source = source;
+ pair->Alpha.Arg[i].Swizzle = swz;
+ pair->Alpha.Arg[i].Abs = inst->SrcReg[i].Abs;
+ pair->Alpha.Arg[i].Negate = !!(inst->SrcReg[i].Negate & NEGATE_W);
+ }
+ }
+
+ return GL_TRUE;
+}
+
+
+/**
+ * Fill in the destination register information.
+ *
+ * This is split from filling in source registers because we want
+ * to avoid allocating hardware temporaries for destinations until
+ * we are absolutely certain that we're going to emit a certain
+ * instruction pairing.
+ */
+static void fill_dest_into_pair(
+ struct pair_state *s,
+ struct radeon_pair_instruction *pair,
+ struct pair_state_instruction *pairinst)
+{
+ struct prog_instruction *inst = &pairinst->Instruction;
+
+ if (inst->DstReg.File == PROGRAM_OUTPUT) {
+ if (inst->DstReg.Index == s->Compiler->OutputColor) {
+ pair->RGB.OutputWriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ;
+ pair->Alpha.OutputWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
+ } else if (inst->DstReg.Index == s->Compiler->OutputDepth) {
+ pair->Alpha.DepthWriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
+ }
+ } else {
+ GLuint hwindex = get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index);
+ if (pairinst->NeedRGB) {
+ pair->RGB.DestIndex = hwindex;
+ pair->RGB.WriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ;
+ }
+ if (pairinst->NeedAlpha) {
+ pair->Alpha.DestIndex = hwindex;
+ pair->Alpha.WriteMask |= GET_BIT(inst->DstReg.WriteMask, 3);
+ }
+ }
+}
+
+
+/**
+ * Find a good ALU instruction or pair of ALU instruction and emit it.
+ *
+ * Prefer emitting full ALU instructions, so that when we reach a point
+ * where no full ALU instruction can be emitted, we have more candidates
+ * for RGB/Alpha pairing.
+ */
+static void emit_alu(struct pair_state *s)
+{
+ struct radeon_pair_instruction pair;
+ struct pair_state_instruction *psi;
+
+ if (s->ReadyFullALU || !(s->ReadyRGB && s->ReadyAlpha)) {
+ if (s->ReadyFullALU) {
+ psi = s->ReadyFullALU;
+ s->ReadyFullALU = s->ReadyFullALU->NextReady;
+ } else if (s->ReadyRGB) {
+ psi = s->ReadyRGB;
+ s->ReadyRGB = s->ReadyRGB->NextReady;
+ } else {
+ psi = s->ReadyAlpha;
+ s->ReadyAlpha = s->ReadyAlpha->NextReady;
+ }
+
+ _mesa_bzero(&pair, sizeof(pair));
+ fill_instruction_into_pair(s, &pair, psi);
+ fill_dest_into_pair(s, &pair, psi);
+ commit_instruction(s, psi);
+ } else {
+ struct pair_state_instruction **prgb;
+ struct pair_state_instruction **palpha;
+
+ /* Some pairings might fail because they require too
+ * many source slots; try all possible pairings if necessary */
+ for(prgb = &s->ReadyRGB; *prgb; prgb = &(*prgb)->NextReady) {
+ for(palpha = &s->ReadyAlpha; *palpha; palpha = &(*palpha)->NextReady) {
+ struct pair_state_instruction * psirgb = *prgb;
+ struct pair_state_instruction * psialpha = *palpha;
+ _mesa_bzero(&pair, sizeof(pair));
+ fill_instruction_into_pair(s, &pair, psirgb);
+ if (!fill_instruction_into_pair(s, &pair, psialpha))
+ continue;
+ *prgb = (*prgb)->NextReady;
+ *palpha = (*palpha)->NextReady;
+ fill_dest_into_pair(s, &pair, psirgb);
+ fill_dest_into_pair(s, &pair, psialpha);
+ commit_instruction(s, psirgb);
+ commit_instruction(s, psialpha);
+ goto success;
+ }
+ }
+
+ /* No success in pairing; just take the first RGB instruction */
+ psi = s->ReadyRGB;
+ s->ReadyRGB = s->ReadyRGB->NextReady;
+
+ _mesa_bzero(&pair, sizeof(pair));
+ fill_instruction_into_pair(s, &pair, psi);
+ fill_dest_into_pair(s, &pair, psi);
+ commit_instruction(s, psi);
+ success: ;
+ }
+
+ if (s->Compiler->Base.Debug)
+ radeonPrintPairInstruction(&pair);
+
+ s->Compiler->Base.Error = s->Compiler->Base.Error || !s->Handler->EmitPaired(s->UserData, &pair);
+}
+
+/* Callback function for assigning input registers to hardware registers */
+static void alloc_helper(void * data, unsigned input, unsigned hwreg)
+{
+ struct pair_state * s = data;
+ alloc_hw_reg(s, PROGRAM_INPUT, input, hwreg);
+}
+
+void radeonPairProgram(
+ struct r300_fragment_program_compiler * compiler,
+ const struct radeon_pair_handler* handler, void *userdata)
+{
+ struct pair_state s;
+
+ _mesa_bzero(&s, sizeof(s));
+ s.Compiler = compiler;
+ s.Handler = handler;
+ s.UserData = userdata;
+ s.Verbose = GL_FALSE && s.Compiler->Base.Debug;
+
+ if (s.Compiler->Base.Debug)
+ _mesa_printf("Emit paired program\n");
+
+ scan_instructions(&s);
+ s.Compiler->AllocateHwInputs(s.Compiler, &alloc_helper, &s);
+
+ while(!s.Compiler->Base.Error &&
+ (s.ReadyTEX || s.ReadyRGB || s.ReadyAlpha || s.ReadyFullALU)) {
+ if (s.ReadyTEX)
+ emit_all_tex(&s);
+
+ while(s.ReadyFullALU || s.ReadyRGB || s.ReadyAlpha)
+ emit_alu(&s);
+ }
+
+ if (s.Compiler->Base.Debug)
+ _mesa_printf(" END\n");
+}
+
+
+static void print_pair_src(int i, struct radeon_pair_instruction_source* src)
+{
+ _mesa_printf(" Src%i = %s[%i]", i, src->Constant ? "CNST" : "TEMP", src->Index);
+}
+
+static const char* opcode_string(GLuint opcode)
+{
+ if (opcode == OPCODE_REPL_ALPHA)
+ return "SOP";
+ else
+ return _mesa_opcode_string(opcode);
+}
+
+static int num_pairinst_args(GLuint opcode)
+{
+ if (opcode == OPCODE_REPL_ALPHA)
+ return 0;
+ else
+ return _mesa_num_inst_src_regs(opcode);
+}
+
+static char swizzle_char(GLuint swz)
+{
+ switch(swz) {
+ case SWIZZLE_X: return 'x';
+ case SWIZZLE_Y: return 'y';
+ case SWIZZLE_Z: return 'z';
+ case SWIZZLE_W: return 'w';
+ case SWIZZLE_ZERO: return '0';
+ case SWIZZLE_ONE: return '1';
+ case SWIZZLE_NIL: return '_';
+ default: return '?';
+ }
+}
+
+void radeonPrintPairInstruction(struct radeon_pair_instruction *inst)
+{
+ int nargs;
+ int i;
+
+ _mesa_printf(" RGB: ");
+ for(i = 0; i < 3; ++i) {
+ if (inst->RGB.Src[i].Used)
+ print_pair_src(i, inst->RGB.Src + i);
+ }
+ _mesa_printf("\n");
+ _mesa_printf(" Alpha:");
+ for(i = 0; i < 3; ++i) {
+ if (inst->Alpha.Src[i].Used)
+ print_pair_src(i, inst->Alpha.Src + i);
+ }
+ _mesa_printf("\n");
+
+ _mesa_printf(" %s%s", opcode_string(inst->RGB.Opcode), inst->RGB.Saturate ? "_SAT" : "");
+ if (inst->RGB.WriteMask)
+ _mesa_printf(" TEMP[%i].%s%s%s", inst->RGB.DestIndex,
+ (inst->RGB.WriteMask & 1) ? "x" : "",
+ (inst->RGB.WriteMask & 2) ? "y" : "",
+ (inst->RGB.WriteMask & 4) ? "z" : "");
+ if (inst->RGB.OutputWriteMask)
+ _mesa_printf(" COLOR.%s%s%s",
+ (inst->RGB.OutputWriteMask & 1) ? "x" : "",
+ (inst->RGB.OutputWriteMask & 2) ? "y" : "",
+ (inst->RGB.OutputWriteMask & 4) ? "z" : "");
+ nargs = num_pairinst_args(inst->RGB.Opcode);
+ for(i = 0; i < nargs; ++i) {
+ const char* abs = inst->RGB.Arg[i].Abs ? "|" : "";
+ const char* neg = inst->RGB.Arg[i].Negate ? "-" : "";
+ _mesa_printf(", %s%sSrc%i.%c%c%c%s", neg, abs, inst->RGB.Arg[i].Source,
+ swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 0)),
+ swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 1)),
+ swizzle_char(GET_SWZ(inst->RGB.Arg[i].Swizzle, 2)),
+ abs);
+ }
+ _mesa_printf("\n");
+
+ _mesa_printf(" %s%s", opcode_string(inst->Alpha.Opcode), inst->Alpha.Saturate ? "_SAT" : "");
+ if (inst->Alpha.WriteMask)
+ _mesa_printf(" TEMP[%i].w", inst->Alpha.DestIndex);
+ if (inst->Alpha.OutputWriteMask)
+ _mesa_printf(" COLOR.w");
+ if (inst->Alpha.DepthWriteMask)
+ _mesa_printf(" DEPTH.w");
+ nargs = num_pairinst_args(inst->Alpha.Opcode);
+ for(i = 0; i < nargs; ++i) {
+ const char* abs = inst->Alpha.Arg[i].Abs ? "|" : "";
+ const char* neg = inst->Alpha.Arg[i].Negate ? "-" : "";
+ _mesa_printf(", %s%sSrc%i.%c%s", neg, abs, inst->Alpha.Arg[i].Source,
+ swizzle_char(inst->Alpha.Arg[i].Swizzle), abs);
+ }
+ _mesa_printf("\n");
+}
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h
new file mode 100644
index 0000000000..ff76178551
--- /dev/null
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h
@@ -0,0 +1,143 @@
+/*
+ * 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 __RADEON_PROGRAM_PAIR_H_
+#define __RADEON_PROGRAM_PAIR_H_
+
+#include "radeon_program.h"
+
+struct r300_fragment_program_compiler;
+
+
+/**
+ * Represents a paired instruction, as found in R300 and R500
+ * fragment programs.
+ */
+struct radeon_pair_instruction_source {
+ GLuint Index:8;
+ GLuint Constant:1;
+ GLuint Used:1;
+};
+
+struct radeon_pair_instruction_rgb {
+ GLuint Opcode:8;
+ GLuint DestIndex:8;
+ GLuint WriteMask:3;
+ GLuint OutputWriteMask:3;
+ GLuint Saturate:1;
+
+ struct radeon_pair_instruction_source Src[3];
+
+ struct {
+ GLuint Source:2;
+ GLuint Swizzle:9;
+ GLuint Abs:1;
+ GLuint Negate:1;
+ } Arg[3];
+};
+
+struct radeon_pair_instruction_alpha {
+ GLuint Opcode:8;
+ GLuint DestIndex:8;
+ GLuint WriteMask:1;
+ GLuint OutputWriteMask:1;
+ GLuint DepthWriteMask:1;
+ GLuint Saturate:1;
+
+ struct radeon_pair_instruction_source Src[3];
+
+ struct {
+ GLuint Source:2;
+ GLuint Swizzle:3;
+ GLuint Abs:1;
+ GLuint Negate:1;
+ } Arg[3];
+};
+
+struct radeon_pair_instruction {
+ struct radeon_pair_instruction_rgb RGB;
+ struct radeon_pair_instruction_alpha Alpha;
+};
+
+
+enum {
+ RADEON_OPCODE_TEX = 0,
+ RADEON_OPCODE_TXB,
+ RADEON_OPCODE_TXP,
+ RADEON_OPCODE_KIL
+};
+
+struct radeon_pair_texture_instruction {
+ GLuint Opcode:2; /**< one of RADEON_OPCODE_xxx */
+
+ GLuint DestIndex:8;
+ GLuint WriteMask:4;
+
+ GLuint TexSrcUnit:5;
+ GLuint TexSrcTarget:3;
+
+ GLuint SrcIndex:8;
+ GLuint SrcSwizzle:12;
+};
+
+
+/**
+ *
+ */
+struct radeon_pair_handler {
+ /**
+ * Write a paired instruction to the hardware.
+ *
+ * @return GL_FALSE on error.
+ */
+ GLboolean (*EmitPaired)(void*, struct radeon_pair_instruction*);
+
+ /**
+ * Write a texture instruction to the hardware.
+ * Register indices have already been rewritten to the allocated
+ * hardware register numbers.
+ *
+ * @return GL_FALSE on error.
+ */
+ GLboolean (*EmitTex)(void*, struct radeon_pair_texture_instruction*);
+
+ /**
+ * Called before a block of contiguous, independent texture
+ * instructions is emitted.
+ */
+ GLboolean (*BeginTexBlock)(void*);
+
+ unsigned MaxHwTemps;
+};
+
+void radeonPairProgram(
+ struct r300_fragment_program_compiler * compiler,
+ const struct radeon_pair_handler*, void *userdata);
+
+void radeonPrintPairInstruction(struct radeon_pair_instruction *inst);
+
+#endif /* __RADEON_PROGRAM_PAIR_H_ */