diff options
Diffstat (limited to 'src/mesa')
38 files changed, 3397 insertions, 1647 deletions
diff --git a/src/mesa/drivers/dri/r300/Makefile b/src/mesa/drivers/dri/r300/Makefile index fe775eac99a..c64f940623f 100644 --- a/src/mesa/drivers/dri/r300/Makefile +++ b/src/mesa/drivers/dri/r300/Makefile @@ -54,6 +54,7 @@ DRIVER_SOURCES = \ r300_vertprog.c \ r300_fragprog_common.c \ r300_shader.c \ + radeon_mesa_to_rc.c \ r300_emit.c \ r300_swtcl.c \ $(RADEON_COMMON_SOURCES) \ diff --git a/src/mesa/drivers/dri/r300/compiler/Makefile b/src/mesa/drivers/dri/r300/compiler/Makefile index d9738441928..53fb7caa95d 100644 --- a/src/mesa/drivers/dri/r300/compiler/Makefile +++ b/src/mesa/drivers/dri/r300/compiler/Makefile @@ -8,10 +8,15 @@ LIBNAME = r300compiler C_SOURCES = \ radeon_code.c \ radeon_compiler.c \ - radeon_nqssadce.c \ radeon_program.c \ + radeon_program_print.c \ + radeon_opcodes.c \ radeon_program_alu.c \ radeon_program_pair.c \ + radeon_dataflow.c \ + radeon_dataflow_annotate.c \ + radeon_dataflow_dealias.c \ + radeon_dataflow_swizzles.c \ r3xx_fragprog.c \ r300_fragprog.c \ r300_fragprog_swizzle.c \ diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c index 6c9fba49146..94f8fdfea3f 100644 --- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.c @@ -27,17 +27,17 @@ #include "r300_fragprog.h" -#include "shader/prog_parameter.h" +#include <stdio.h> #include "../r300_reg.h" -static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu) +static struct rc_src_register shadow_ambient(struct radeon_compiler * c, int tmu) { - struct prog_src_register reg = { 0, }; + struct rc_src_register reg = { 0, }; - reg.File = PROGRAM_STATE_VAR; + reg.File = RC_FILE_CONSTANT; reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu); - reg.Swizzle = SWIZZLE_WWWW; + reg.Swizzle = RC_SWIZZLE_WWWW; return reg; } @@ -47,7 +47,7 @@ static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int t * - extract operand swizzles * - introduce a temporary register when write masks are needed */ -GLboolean r300_transform_TEX( +int r300_transform_TEX( struct radeon_compiler * c, struct rc_instruction* inst, void* data) @@ -55,77 +55,77 @@ GLboolean r300_transform_TEX( 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; + if (inst->I.Opcode != RC_OPCODE_TEX && + inst->I.Opcode != RC_OPCODE_TXB && + inst->I.Opcode != RC_OPCODE_TXP && + inst->I.Opcode != RC_OPCODE_KIL) + return 0; /* ARB_shadow & EXT_shadow_funcs */ - if (inst->I.Opcode != OPCODE_KIL && + if (inst->I.Opcode != RC_OPCODE_KIL && c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + rc_compare_func comparefunc = compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; - if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) { - inst->I.Opcode = OPCODE_MOV; + if (comparefunc == RC_COMPARE_FUNC_NEVER || comparefunc == RC_COMPARE_FUNC_ALWAYS) { + inst->I.Opcode = RC_OPCODE_MOV; - if (comparefunc == GL_ALWAYS) { - inst->I.SrcReg[0].File = PROGRAM_BUILTIN; - inst->I.SrcReg[0].Swizzle = SWIZZLE_1111; + if (comparefunc == RC_COMPARE_FUNC_ALWAYS) { + inst->I.SrcReg[0].File = RC_FILE_NONE; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_1111; } else { inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit); } - return GL_TRUE; + return 1; } 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; + rc_compare_func comparefunc = compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + unsigned int 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.Opcode = RC_OPCODE_RCP; + inst_rcp->I.DstReg.File = RC_FILE_TEMPORARY; inst_rcp->I.DstReg.Index = rc_find_free_temporary(c); - inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + inst_rcp->I.DstReg.WriteMask = RC_MASK_W; inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0]; - inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + inst_rcp->I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW; inst_cmp->I.DstReg = inst->I.DstReg; - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = rc_find_free_temporary(c); - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; - inst_mad->I.Opcode = OPCODE_MAD; - inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.Opcode = RC_OPCODE_MAD; + inst_mad->I.DstReg.File = RC_FILE_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[0].Swizzle = RC_SWIZZLE_ZZZZ; + inst_mad->I.SrcReg[1].File = RC_FILE_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[1].Swizzle = RC_SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].File = RC_FILE_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); + inst_mad->I.SrcReg[2].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_Z); else if (depthmode == 2) /* GL_ALPHA */ - inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].Swizzle = RC_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; + if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GEQUAL) + inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ RC_MASK_XYZW; else - inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW; + inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ RC_MASK_XYZW; - inst_cmp->I.Opcode = OPCODE_CMP; + inst_cmp->I.Opcode = RC_OPCODE_CMP; /* DstReg has been filled out above */ - inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_cmp->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index; - if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { + if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GREATER) { pass = 1; fail = 2; } else { @@ -133,8 +133,8 @@ GLboolean r300_transform_TEX( fail = 1; } - inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN; - inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111; + inst_cmp->I.SrcReg[pass].File = RC_FILE_NONE; + inst_cmp->I.SrcReg[pass].Swizzle = RC_SWIZZLE_1111; inst_cmp->I.SrcReg[fail] = shadow_ambient(c, inst->I.TexSrcUnit); } } @@ -143,52 +143,52 @@ GLboolean r300_transform_TEX( * instead of [0..Width]x[0..Height]. * Add a scaling instruction. */ - if (inst->I.Opcode != OPCODE_KIL && inst->I.TexSrcTarget == TEXTURE_RECT_INDEX) { + if (inst->I.Opcode != RC_OPCODE_KIL && inst->I.TexSrcTarget == RC_TEXTURE_RECT) { 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.Opcode = RC_OPCODE_MUL; + inst_mul->I.DstReg.File = RC_FILE_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].File = RC_FILE_CONSTANT; 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].File = RC_FILE_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)) { + if (inst->I.Opcode != RC_OPCODE_KIL && + (inst->I.DstReg.File != RC_FILE_TEMPORARY || inst->I.DstReg.WriteMask != RC_MASK_XYZW)) { struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst); - inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.Opcode = RC_OPCODE_MOV; inst_mov->I.DstReg = inst->I.DstReg; - inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mov->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c); - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; } /* Cannot read texture coordinate from constants file */ - if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) { + if (inst->I.SrcReg[0].File != RC_FILE_TEMPORARY && inst->I.SrcReg[0].File != RC_FILE_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.Opcode = RC_OPCODE_MOV; + inst_mov->I.DstReg.File = RC_FILE_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].File = RC_FILE_TEMPORARY; inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index; } - return GL_TRUE; + return 1; } /* just some random things... */ diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h index 0ac46dbd9cb..418df36c936 100644 --- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog.h @@ -33,9 +33,6 @@ #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" @@ -44,6 +41,6 @@ extern void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler extern void r300FragmentProgramDump(struct rX00_fragment_program_code *c); -extern GLboolean r300_transform_TEX(struct radeon_compiler * c, struct rc_instruction* inst, void* data); +extern int 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 index c7227bbd15b..2dad3dba4ab 100644 --- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_emit.c @@ -69,64 +69,64 @@ struct r300_emit_state { /** * Mark a temporary register as used. */ -static void use_temporary(struct r300_fragment_program_code *code, GLuint index) +static void use_temporary(struct r300_fragment_program_code *code, unsigned int index) { if (index > code->pixsize) code->pixsize = index; } -static GLuint translate_rgb_opcode(struct r300_fragment_program_compiler * c, GLuint opcode) +static unsigned int translate_rgb_opcode(struct r300_fragment_program_compiler * c, rc_opcode 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; + case RC_OPCODE_CMP: return R300_ALU_OUTC_CMP; + case RC_OPCODE_DP3: return R300_ALU_OUTC_DP3; + case RC_OPCODE_DP4: return R300_ALU_OUTC_DP4; + case RC_OPCODE_FRC: return R300_ALU_OUTC_FRC; default: error("translate_rgb_opcode(%i): Unknown opcode", opcode); /* fall through */ - case OPCODE_NOP: + case RC_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; + case RC_OPCODE_MAD: return R300_ALU_OUTC_MAD; + case RC_OPCODE_MAX: return R300_ALU_OUTC_MAX; + case RC_OPCODE_MIN: return R300_ALU_OUTC_MIN; + case RC_OPCODE_REPL_ALPHA: return R300_ALU_OUTC_REPL_ALPHA; } } -static GLuint translate_alpha_opcode(struct r300_fragment_program_compiler * c, GLuint opcode) +static unsigned int translate_alpha_opcode(struct r300_fragment_program_compiler * c, rc_opcode 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; + case RC_OPCODE_CMP: return R300_ALU_OUTA_CMP; + case RC_OPCODE_DP3: return R300_ALU_OUTA_DP4; + case RC_OPCODE_DP4: return R300_ALU_OUTA_DP4; + case RC_OPCODE_EX2: return R300_ALU_OUTA_EX2; + case RC_OPCODE_FRC: return R300_ALU_OUTA_FRC; + case RC_OPCODE_LG2: return R300_ALU_OUTA_LG2; default: error("translate_rgb_opcode(%i): Unknown opcode", opcode); /* fall through */ - case OPCODE_NOP: + case RC_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; + case RC_OPCODE_MAD: return R300_ALU_OUTA_MAD; + case RC_OPCODE_MAX: return R300_ALU_OUTA_MAX; + case RC_OPCODE_MIN: return R300_ALU_OUTA_MIN; + case RC_OPCODE_RCP: return R300_ALU_OUTA_RCP; + case RC_OPCODE_RSQ: return R300_ALU_OUTA_RSQ; } } /** * Emit one paired ALU instruction. */ -static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) +static int 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; + return 0; } int ip = code->alu.length++; @@ -136,7 +136,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) 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); + unsigned int 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); @@ -146,7 +146,7 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) 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); + unsigned int 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); @@ -186,17 +186,17 @@ static GLboolean emit_alu(void* data, struct radeon_pair_instruction* inst) 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; + c->code->writes_depth = 1; } - return GL_TRUE; + return 1; } /** * Finish the current node without advancing to the next one. */ -static GLboolean finish_node(struct r300_emit_state * emit) +static int finish_node(struct r300_emit_state * emit) { struct r300_fragment_program_compiler * c = emit->compiler; struct r300_fragment_program_code *code = &emit->compiler->code->code.r300; @@ -204,9 +204,9 @@ static GLboolean finish_node(struct r300_emit_state * emit) 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)); + memset(&inst, 0, sizeof(inst)); if (!emit_alu(emit, &inst)) - return GL_FALSE; + return 0; } unsigned alu_offset = emit->node_first_alu; @@ -217,7 +217,7 @@ static GLboolean finish_node(struct r300_emit_state * emit) 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; + return 0; } tex_end = 0; @@ -240,7 +240,7 @@ static GLboolean finish_node(struct r300_emit_state * emit) (tex_end << R300_TEX_SIZE_SHIFT) | emit->node_flags; - return GL_TRUE; + return 1; } @@ -248,43 +248,43 @@ static GLboolean finish_node(struct r300_emit_state * emit) * Begin a block of texture instructions. * Create the necessary indirection. */ -static GLboolean begin_tex(void* data) +static int begin_tex(void* data) { PROG_CODE; if (code->alu.length == emit->node_first_alu && code->tex.length == emit->node_first_tex) { - return GL_TRUE; + return 1; } if (emit->current_node == 3) { error("Too many texture indirections"); - return GL_FALSE; + return 0; } if (!finish_node(emit)) - return GL_FALSE; + return 0; emit->current_node++; emit->node_first_tex = code->tex.length; emit->node_first_alu = code->alu.length; emit->node_flags = 0; - return GL_TRUE; + return 1; } -static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* inst) +static int 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; + return 0; } - GLuint unit = inst->TexSrcUnit; - GLuint dest = inst->DestIndex; - GLuint opcode; + unsigned int unit = inst->TexSrcUnit; + unsigned int dest = inst->DestIndex; + unsigned int opcode; switch(inst->Opcode) { case RADEON_OPCODE_KIL: opcode = R300_TEX_OP_KIL; break; @@ -293,7 +293,7 @@ static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* in case RADEON_OPCODE_TXP: opcode = R300_TEX_OP_TXP; break; default: error("Unknown texture opcode %i", inst->Opcode); - return GL_FALSE; + return 0; } if (inst->Opcode == RADEON_OPCODE_KIL) { @@ -310,7 +310,7 @@ static GLboolean emit_tex(void* data, struct radeon_pair_texture_instruction* in (dest << R300_DST_ADDR_SHIFT) | (unit << R300_TEX_ID_SHIFT) | (opcode << R300_TEX_INST_SHIFT); - return GL_TRUE; + return 1; } @@ -333,7 +333,7 @@ void r300BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compi memset(&emit, 0, sizeof(emit)); emit.compiler = compiler; - _mesa_bzero(code, sizeof(struct r300_fragment_program_code)); + memset(code, 0, sizeof(struct r300_fragment_program_code)); radeonPairProgram(compiler, &pair_handler, &emit); if (compiler->Base.Error) diff --git a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c index 1b14cc3888b..cfa48a59e3a 100644 --- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.c @@ -33,16 +33,17 @@ #include "r300_fragprog_swizzle.h" +#include <stdio.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)) +#define MAKE_SWZ3(x, y, z) (RC_MAKE_SWIZZLE(RC_SWIZZLE_##x, RC_SWIZZLE_##y, RC_SWIZZLE_##z, RC_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 */ + unsigned int hash; /**< swizzle value this matches */ + unsigned int base; /**< base value for hw swizzle */ + unsigned int stride; /**< difference in base between arg0/1/2 */ }; static const struct swizzle_data native_swizzles[] = { @@ -65,15 +66,15 @@ static const int num_native_swizzles = sizeof(native_swizzles)/sizeof(native_swi * 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) +static const struct swizzle_data* lookup_native_swizzle(unsigned int 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) + unsigned int swz = GET_SWZ(swizzle, comp); + if (swz == RC_SWIZZLE_UNUSED) continue; if (swz != GET_SWZ(sd->hash, comp)) break; @@ -90,71 +91,72 @@ static const struct swizzle_data* lookup_native_swizzle(GLuint swizzle) * 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) +static int r300_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg) { if (reg.Abs) - reg.Negate = NEGATE_NONE; + reg.Negate = RC_MASK_NONE; - if (opcode == OPCODE_KIL || - opcode == OPCODE_TEX || - opcode == OPCODE_TXB || - opcode == OPCODE_TXP) { + if (opcode == RC_OPCODE_KIL || + opcode == RC_OPCODE_TEX || + opcode == RC_OPCODE_TXB || + opcode == RC_OPCODE_TXP) { int j; if (reg.Abs || reg.Negate) - return GL_FALSE; + return 0; for(j = 0; j < 4; ++j) { - GLuint swz = GET_SWZ(reg.Swizzle, j); - if (swz == SWIZZLE_NIL) + unsigned int swz = GET_SWZ(reg.Swizzle, j); + if (swz == RC_SWIZZLE_UNUSED) continue; if (swz != j) - return GL_FALSE; + return 0; } - return GL_TRUE; + return 1; } - GLuint relevant = 0; + unsigned int relevant = 0; int j; for(j = 0; j < 3; ++j) - if (GET_SWZ(reg.Swizzle, j) != SWIZZLE_NIL) + if (GET_SWZ(reg.Swizzle, j) != RC_SWIZZLE_UNUSED) relevant |= 1 << j; if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant)) - return GL_FALSE; + return 0; if (!lookup_native_swizzle(reg.Swizzle)) - return GL_FALSE; + return 0; - return GL_TRUE; + return 1; } -/** - * Generate MOV dst, src using only native swizzles. - */ -void r300FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src) +static void r300_swizzle_split( + struct rc_src_register src, unsigned int mask, + struct rc_swizzle_split * split) { if (src.Abs) - src.Negate = NEGATE_NONE; + src.Negate = RC_MASK_NONE; + + split->NumPhases = 0; - while(dst.WriteMask) { + while(mask) { const struct swizzle_data *best_swizzle = 0; - GLuint best_matchcount = 0; - GLuint best_matchmask = 0; + unsigned int best_matchcount = 0; + unsigned int 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; + unsigned int matchcount = 0; + unsigned int matchmask = 0; for(comp = 0; comp < 3; ++comp) { - if (!GET_BIT(dst.WriteMask, comp)) + if (!GET_BIT(mask, comp)) continue; - GLuint swz = GET_SWZ(src.Swizzle, comp); - if (swz == SWIZZLE_NIL) + unsigned int swz = GET_SWZ(src.Swizzle, comp); + if (swz == RC_SWIZZLE_UNUSED) continue; if (swz == GET_SWZ(sd->hash, comp)) { /* check if the negate bit of current component @@ -170,34 +172,35 @@ void r300FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, best_swizzle = sd; best_matchcount = matchcount; best_matchmask = matchmask; - if (matchmask == (dst.WriteMask & WRITEMASK_XYZ)) + if (matchmask == (mask & RC_MASK_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 */ + if (mask & RC_MASK_W) + best_matchmask |= RC_MASK_W; - dst.WriteMask &= ~inst->I.DstReg.WriteMask; + split->Phase[split->NumPhases++] = best_matchmask; + mask &= ~best_matchmask; } } +struct rc_swizzle_caps r300_swizzle_caps = { + .IsNative = r300_swizzle_is_native, + .Split = r300_swizzle_split +}; + /** * Translate an RGB (XYZ) swizzle into the hardware code for the given * instruction source. */ -GLuint r300FPTranslateRGBSwizzle(GLuint src, GLuint swizzle) +unsigned int r300FPTranslateRGBSwizzle(unsigned int src, unsigned int swizzle) { const struct swizzle_data* sd = lookup_native_swizzle(swizzle); if (!sd) { - _mesa_printf("Not a native swizzle: %08x\n", swizzle); + fprintf(stderr, "Not a native swizzle: %08x\n", swizzle); return 0; } @@ -209,15 +212,15 @@ GLuint r300FPTranslateRGBSwizzle(GLuint src, GLuint swizzle) * Translate an Alpha (W) swizzle into the hardware code for the given * instruction source. */ -GLuint r300FPTranslateAlphaSwizzle(GLuint src, GLuint swizzle) +unsigned int r300FPTranslateAlphaSwizzle(unsigned int src, unsigned int 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; + case RC_SWIZZLE_W: return R300_ALU_ARGA_SRC0A + src; + case RC_SWIZZLE_ONE: return R300_ALU_ARGA_ONE; + case RC_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 index 231bf4eef5f..118476af132 100644 --- a/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h +++ b/src/mesa/drivers/dri/r300/compiler/r300_fragprog_swizzle.h @@ -28,15 +28,11 @@ #ifndef __R300_FRAGPROG_SWIZZLE_H_ #define __R300_FRAGPROG_SWIZZLE_H_ -#include "main/glheader.h" -#include "shader/prog_instruction.h" +#include "radeon_swizzle.h" -struct nqssadce_state; +extern struct rc_swizzle_caps r300_swizzle_caps; -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); +unsigned int r300FPTranslateRGBSwizzle(unsigned int src, unsigned int swizzle); +unsigned int r300FPTranslateAlphaSwizzle(unsigned int src, unsigned int 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 index 76c3a7ecfd2..bf9bea685ab 100644 --- a/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c +++ b/src/mesa/drivers/dri/r300/compiler/r3xx_fragprog.c @@ -22,22 +22,20 @@ #include "radeon_compiler.h" -#include "shader/prog_parameter.h" -#include "shader/prog_print.h" -#include "shader/prog_statevars.h" +#include <stdio.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) +static void dataflow_outputs_mark_use(void * userdata, void * data, + void (*callback)(void *, unsigned int, unsigned int)) { - struct r300_fragment_program_compiler * c = s->UserData; - s->Outputs[c->OutputColor].Sourced = WRITEMASK_XYZW; - s->Outputs[c->OutputDepth].Sourced = WRITEMASK_W; + struct r300_fragment_program_compiler * c = userdata; + callback(data, c->OutputColor, RC_MASK_XYZW); + callback(data, c->OutputDepth, RC_MASK_W); } static void rewrite_depth_out(struct r300_fragment_program_compiler * c) @@ -45,35 +43,35 @@ 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; + struct rc_sub_instruction * inst = &rci->I; - if (inst->DstReg.File != PROGRAM_OUTPUT || inst->DstReg.Index != c->OutputDepth) + if (inst->DstReg.File != RC_FILE_OUTPUT || inst->DstReg.Index != c->OutputDepth) continue; - if (inst->DstReg.WriteMask & WRITEMASK_Z) { - inst->DstReg.WriteMask = WRITEMASK_W; + if (inst->DstReg.WriteMask & RC_MASK_Z) { + inst->DstReg.WriteMask = RC_MASK_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]); + case RC_OPCODE_FRC: + case RC_OPCODE_MOV: + inst->SrcReg[0] = lmul_swizzle(RC_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]); + case RC_OPCODE_ADD: + case RC_OPCODE_MAX: + case RC_OPCODE_MIN: + case RC_OPCODE_MUL: + inst->SrcReg[0] = lmul_swizzle(RC_SWIZZLE_ZZZZ, inst->SrcReg[0]); + inst->SrcReg[1] = lmul_swizzle(RC_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]); + case RC_OPCODE_CMP: + case RC_OPCODE_MAD: + inst->SrcReg[0] = lmul_swizzle(RC_SWIZZLE_ZZZZ, inst->SrcReg[0]); + inst->SrcReg[1] = lmul_swizzle(RC_SWIZZLE_ZZZZ, inst->SrcReg[1]); + inst->SrcReg[2] = lmul_swizzle(RC_SWIZZLE_ZZZZ, inst->SrcReg[2]); break; default: // Scalar instructions needn't be reswizzled @@ -94,6 +92,8 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c) { &radeonTransformTrigScale, 0 } }; radeonLocalTransform(&c->Base, 4, transformations); + + c->Base.SwizzleCaps = &r500_swizzle_caps; } else { struct radeon_program_transformation transformations[] = { { &r300_transform_TEX, c }, @@ -101,33 +101,23 @@ void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c) { &radeonTransformTrigSimple, 0 } }; radeonLocalTransform(&c->Base, 3, transformations); + + c->Base.SwizzleCaps = &r300_swizzle_caps; } if (c->Base.Debug) { - _mesa_printf("Fragment Program: After native rewrite:\n"); - rc_print_program(&c->Base.Program); + fprintf(stderr, "Fragment Program: After native rewrite:\n"); + rc_print_program(&c->Base.Program, 0); fflush(stderr); } - 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); - } + rc_dataflow_annotate(&c->Base, &dataflow_outputs_mark_use, c); + rc_dataflow_dealias(&c->Base); + rc_dataflow_swizzles(&c->Base); if (c->Base.Debug) { - _mesa_printf("Compiler: after NqSSA-DCE:\n"); - rc_print_program(&c->Base.Program); + fprintf(stderr, "Compiler: after dataflow passes:\n"); + rc_print_program(&c->Base.Program, 0); fflush(stderr); } diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c index dad27fc98e6..ac72f8cbb61 100644 --- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c +++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c @@ -22,13 +22,13 @@ #include "radeon_compiler.h" +#include <stdio.h> + #include "../r300_reg.h" -#include "radeon_nqssadce.h" -#include "radeon_program.h" +#include "radeon_dataflow.h" #include "radeon_program_alu.h" - -#include "shader/prog_print.h" +#include "radeon_swizzle.h" /* @@ -42,104 +42,83 @@ t_swizzle(y), \ t_swizzle(y), \ t_src_class(vpi->SrcReg[x].File), \ - NEGATE_NONE) | (vpi->SrcReg[x].RelAddr << 4)) + RC_MASK_NONE) | (vpi->SrcReg[x].RelAddr << 4)) -static unsigned long t_dst_mask(GLuint mask) +static unsigned long t_dst_mask(unsigned int mask) { - /* WRITEMASK_* is equivalent to VSF_FLAG_* */ - return mask & WRITEMASK_XYZW; + /* RC_MASK_* is equivalent to VSF_FLAG_* */ + return mask & RC_MASK_XYZW; } -static unsigned long t_dst_class(gl_register_file file) +static unsigned long t_dst_class(rc_register_file file) { - switch (file) { - case PROGRAM_TEMPORARY: + default: + fprintf(stderr, "%s: Bad register file %i\n", __FUNCTION__, file); + /* fall-through */ + case RC_FILE_TEMPORARY: return PVS_DST_REG_TEMPORARY; - case PROGRAM_OUTPUT: + case RC_FILE_OUTPUT: return PVS_DST_REG_OUT; - case PROGRAM_ADDRESS: + case RC_FILE_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) + struct rc_dst_register *dst) { - if (dst->File == PROGRAM_OUTPUT) + if (dst->File == RC_FILE_OUTPUT) return vp->outputs[dst->Index]; return dst->Index; } -static unsigned long t_src_class(gl_register_file file) +static unsigned long t_src_class(rc_register_file file) { switch (file) { - case PROGRAM_BUILTIN: - case PROGRAM_TEMPORARY: + default: + fprintf(stderr, "%s: Bad register file %i\n", __FUNCTION__, file); + /* fall-through */ + case RC_FILE_NONE: + case RC_FILE_TEMPORARY: return PVS_SRC_REG_TEMPORARY; - case PROGRAM_INPUT: + case RC_FILE_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: + case RC_FILE_CONSTANT: 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) +static int t_src_conflict(struct rc_src_register a, struct rc_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; + return 0; if (aclass == PVS_SRC_REG_TEMPORARY) - return GL_FALSE; + return 0; if (a.RelAddr || b.RelAddr) - return GL_TRUE; + return 1; if (a.Index != b.Index) - return GL_TRUE; + return 1; - return GL_FALSE; + return 0; } -static INLINE unsigned long t_swizzle(GLubyte swizzle) +static inline unsigned long t_swizzle(unsigned int swizzle) { - /* this is in fact a NOP as the Mesa SWIZZLE_* are all identical to VSF_IN_COMPONENT_* */ + /* this is in fact a NOP as the Mesa RC_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) + struct rc_src_register *src) { - if (src->File == PROGRAM_INPUT) { + if (src->File == RC_FILE_INPUT) { assert(vp->inputs[src->Index] != -1); return vp->inputs[src->Index]; } else { @@ -155,9 +134,9 @@ static unsigned long t_src_index(struct r300_vertex_program_code *vp, /* these two functions should probably be merged... */ static unsigned long t_src(struct r300_vertex_program_code *vp, - struct prog_src_register *src) + struct rc_src_register *src) { - /* src->Negate uses the NEGATE_ flags from program_instruction.h, + /* src->Negate uses the RC_MASK_ 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), @@ -170,9 +149,9 @@ static unsigned long t_src(struct r300_vertex_program_code *vp, } static unsigned long t_src_scalar(struct r300_vertex_program_code *vp, - struct prog_src_register *src) + struct rc_src_register *src) { - /* src->Negate uses the NEGATE_ flags from program_instruction.h, + /* src->Negate uses the RC_MASK_ 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), @@ -181,79 +160,79 @@ static unsigned long t_src_scalar(struct r300_vertex_program_code *vp, 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->Negate ? RC_MASK_XYZW : RC_MASK_NONE) | (src->RelAddr << 4); } -static GLboolean valid_dst(struct r300_vertex_program_code *vp, - struct prog_dst_register *dst) +static int valid_dst(struct r300_vertex_program_code *vp, + struct rc_dst_register *dst) { - if (dst->File == PROGRAM_OUTPUT && vp->outputs[dst->Index] == -1) { - return GL_FALSE; - } else if (dst->File == PROGRAM_ADDRESS) { + if (dst->File == RC_FILE_OUTPUT && vp->outputs[dst->Index] == -1) { + return 0; + } else if (dst->File == RC_FILE_ADDRESS) { assert(dst->Index == 0); } - return GL_TRUE; + return 1; } static void ei_vector1(struct r300_vertex_program_code *vp, - GLuint hw_opcode, - struct prog_instruction *vpi, - GLuint * inst) + unsigned int hw_opcode, + struct rc_sub_instruction *vpi, + unsigned int * inst) { inst[0] = PVS_OP_DST_OPERAND(hw_opcode, - GL_FALSE, - GL_FALSE, + 0, + 0, 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); + inst[2] = __CONST(0, RC_SWIZZLE_ZERO); + inst[3] = __CONST(0, RC_SWIZZLE_ZERO); } static void ei_vector2(struct r300_vertex_program_code *vp, - GLuint hw_opcode, - struct prog_instruction *vpi, - GLuint * inst) + unsigned int hw_opcode, + struct rc_sub_instruction *vpi, + unsigned int * inst) { inst[0] = PVS_OP_DST_OPERAND(hw_opcode, - GL_FALSE, - GL_FALSE, + 0, + 0, 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); + inst[3] = __CONST(1, RC_SWIZZLE_ZERO); } static void ei_math1(struct r300_vertex_program_code *vp, - GLuint hw_opcode, - struct prog_instruction *vpi, - GLuint * inst) + unsigned int hw_opcode, + struct rc_sub_instruction *vpi, + unsigned int * inst) { inst[0] = PVS_OP_DST_OPERAND(hw_opcode, - GL_TRUE, - GL_FALSE, + 1, + 0, 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); + inst[2] = __CONST(0, RC_SWIZZLE_ZERO); + inst[3] = __CONST(0, RC_SWIZZLE_ZERO); } static void ei_lit(struct r300_vertex_program_code *vp, - struct prog_instruction *vpi, - GLuint * inst) + struct rc_sub_instruction *vpi, + unsigned int * 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, + 1, + 0, t_dst_index(vp, &vpi->DstReg), t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File)); @@ -263,27 +242,27 @@ static void ei_lit(struct r300_vertex_program_code *vp, 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].Negate ? RC_MASK_XYZW : RC_MASK_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].Negate ? RC_MASK_XYZW : RC_MASK_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].Negate ? RC_MASK_XYZW : RC_MASK_NONE) | (vpi->SrcReg[0].RelAddr << 4); } static void ei_mad(struct r300_vertex_program_code *vp, - struct prog_instruction *vpi, - GLuint * inst) + struct rc_sub_instruction *vpi, + unsigned int * inst) { /* Remarks about hardware limitations of MAD * (please preserve this comment, as this information is _NOT_ @@ -311,22 +290,22 @@ static void ei_mad(struct r300_vertex_program_code *vp, * according to AMD docs, this should improve performance by one clock * as a nice side bonus. */ - if (vpi->SrcReg[0].File == PROGRAM_TEMPORARY && - vpi->SrcReg[1].File == PROGRAM_TEMPORARY && - vpi->SrcReg[2].File == PROGRAM_TEMPORARY && + if (vpi->SrcReg[0].File == RC_FILE_TEMPORARY && + vpi->SrcReg[1].File == RC_FILE_TEMPORARY && + vpi->SrcReg[2].File == RC_FILE_TEMPORARY && vpi->SrcReg[0].Index != vpi->SrcReg[1].Index && vpi->SrcReg[0].Index != vpi->SrcReg[2].Index && vpi->SrcReg[1].Index != vpi->SrcReg[2].Index) { inst[0] = PVS_OP_DST_OPERAND(PVS_MACRO_OP_2CLK_MADD, - GL_FALSE, - GL_TRUE, + 0, + 1, t_dst_index(vp, &vpi->DstReg), t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File)); } else { inst[0] = PVS_OP_DST_OPERAND(VE_MULTIPLY_ADD, - GL_FALSE, - GL_FALSE, + 0, + 0, t_dst_index(vp, &vpi->DstReg), t_dst_mask(vpi->DstReg.WriteMask), t_dst_class(vpi->DstReg.File)); @@ -337,17 +316,17 @@ static void ei_mad(struct r300_vertex_program_code *vp, } static void ei_pow(struct r300_vertex_program_code *vp, - struct prog_instruction *vpi, - GLuint * inst) + struct rc_sub_instruction *vpi, + unsigned int * inst) { inst[0] = PVS_OP_DST_OPERAND(ME_POWER_FUNC_FF, - GL_TRUE, - GL_FALSE, + 1, + 0, 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[2] = __CONST(0, RC_SWIZZLE_ZERO); inst[3] = t_src_scalar(vp, &vpi->SrcReg[1]); } @@ -362,8 +341,8 @@ static void translate_vertex_program(struct r300_vertex_program_compiler * compi 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; + struct rc_sub_instruction *vpi = &rci->I; + unsigned int *inst = compiler->code->body.d + compiler->code->length; /* Skip instructions writing to non-existing destination */ if (!valid_dst(compiler->code, &vpi->DstReg)) @@ -375,26 +354,26 @@ static void translate_vertex_program(struct r300_vertex_program_compiler * compi } 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; + case RC_OPCODE_ADD: ei_vector2(compiler->code, VE_ADD, vpi, inst); break; + case RC_OPCODE_ARL: ei_vector1(compiler->code, VE_FLT2FIX_DX, vpi, inst); break; + case RC_OPCODE_DP4: ei_vector2(compiler->code, VE_DOT_PRODUCT, vpi, inst); break; + case RC_OPCODE_DST: ei_vector2(compiler->code, VE_DISTANCE_VECTOR, vpi, inst); break; + case RC_OPCODE_EX2: ei_math1(compiler->code, ME_EXP_BASE2_FULL_DX, vpi, inst); break; + case RC_OPCODE_EXP: ei_math1(compiler->code, ME_EXP_BASE2_DX, vpi, inst); break; + case RC_OPCODE_FRC: ei_vector1(compiler->code, VE_FRACTION, vpi, inst); break; + case RC_OPCODE_LG2: ei_math1(compiler->code, ME_LOG_BASE2_FULL_DX, vpi, inst); break; + case RC_OPCODE_LIT: ei_lit(compiler->code, vpi, inst); break; + case RC_OPCODE_LOG: ei_math1(compiler->code, ME_LOG_BASE2_DX, vpi, inst); break; + case RC_OPCODE_MAD: ei_mad(compiler->code, vpi, inst); break; + case RC_OPCODE_MAX: ei_vector2(compiler->code, VE_MAXIMUM, vpi, inst); break; + case RC_OPCODE_MIN: ei_vector2(compiler->code, VE_MINIMUM, vpi, inst); break; + case RC_OPCODE_MOV: ei_vector1(compiler->code, VE_ADD, vpi, inst); break; + case RC_OPCODE_MUL: ei_vector2(compiler->code, VE_MULTIPLY, vpi, inst); break; + case RC_OPCODE_POW: ei_pow(compiler->code, vpi, inst); break; + case RC_OPCODE_RCP: ei_math1(compiler->code, ME_RECIP_DX, vpi, inst); break; + case RC_OPCODE_RSQ: ei_math1(compiler->code, ME_RECIP_SQRT_DX, vpi, inst); break; + case RC_OPCODE_SGE: ei_vector2(compiler->code, VE_SET_GREATER_THAN_EQUAL, vpi, inst); break; + case RC_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; @@ -408,36 +387,35 @@ static void translate_vertex_program(struct r300_vertex_program_compiler * compi } struct temporary_allocation { - GLuint Allocated:1; - GLuint HwTemp:15; + unsigned int Allocated:1; + unsigned int 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]; + unsigned int num_orig_temps = 0; + char hwtemps[VSF_MAX_FRAGMENT_TEMPS]; struct temporary_allocation * ta; - GLuint i, j; + unsigned int 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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - for (i = 0; i < numsrcs; ++i) { - if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) { + for (i = 0; i < opcode->NumSrcRegs; ++i) { + if (inst->I.SrcReg[i].File == RC_FILE_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 (opcode->HasDstReg) { + if (inst->I.DstReg.File == RC_FILE_TEMPORARY) { if (inst->I.DstReg.Index >= num_orig_temps) num_orig_temps = inst->I.DstReg.Index + 1; } @@ -450,32 +428,31 @@ static void allocate_temporary_registers(struct r300_vertex_program_compiler * c /* 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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - for (i = 0; i < numsrcs; ++i) { - if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) + for (i = 0; i < opcode->NumSrcRegs; ++i) { + if (inst->I.SrcReg[i].File == RC_FILE_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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - for (i = 0; i < numsrcs; ++i) { - if (inst->I.SrcReg[i].File == PROGRAM_TEMPORARY) { - GLuint orig = inst->I.SrcReg[i].Index; + for (i = 0; i < opcode->NumSrcRegs; ++i) { + if (inst->I.SrcReg[i].File == RC_FILE_TEMPORARY) { + unsigned int 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; + hwtemps[ta[orig].HwTemp] = 0; } } - if (numdsts) { - if (inst->I.DstReg.File == PROGRAM_TEMPORARY) { - GLuint orig = inst->I.DstReg.Index; + if (opcode->HasDstReg) { + if (inst->I.DstReg.File == RC_FILE_TEMPORARY) { + unsigned int orig = inst->I.DstReg.Index; if (!ta[orig].Allocated) { for(j = 0; j < VSF_MAX_FRAGMENT_TEMPS; ++j) { @@ -485,9 +462,9 @@ static void allocate_temporary_registers(struct r300_vertex_program_compiler * c if (j >= VSF_MAX_FRAGMENT_TEMPS) { fprintf(stderr, "Out of hw temporaries\n"); } else { - ta[orig].Allocated = GL_TRUE; + ta[orig].Allocated = 1; ta[orig].HwTemp = j; - hwtemps[j] = GL_TRUE; + hwtemps[j] = 1; if (j >= compiler->code->num_temporaries) compiler->code->num_temporaries = j + 1; @@ -505,45 +482,45 @@ static void allocate_temporary_registers(struct r300_vertex_program_compiler * c * 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( +static int transform_source_conflicts( struct radeon_compiler *c, struct rc_instruction* inst, void* unused) { - GLuint num_operands = _mesa_num_inst_src_regs(inst->I.Opcode); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - if (num_operands == 3) { + if (opcode->NumSrcRegs == 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.Opcode = RC_OPCODE_MOV; + inst_mov->I.DstReg.File = RC_FILE_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].File = RC_FILE_TEMPORARY; inst->I.SrcReg[2].Index = tmpreg; } } - if (num_operands >= 2) { + if (opcode->NumSrcRegs >= 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.Opcode = RC_OPCODE_MOV; + inst_mov->I.DstReg.File = RC_FILE_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].File = RC_FILE_TEMPORARY; inst->I.SrcReg[1].Index = tmpreg; } } - return GL_TRUE; + return 1; } static void addArtificialOutputs(struct r300_vertex_program_compiler * compiler) @@ -554,44 +531,52 @@ static void addArtificialOutputs(struct r300_vertex_program_compiler * compiler) 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.Opcode = RC_OPCODE_MOV; - inst->I.DstReg.File = PROGRAM_OUTPUT; + inst->I.DstReg.File = RC_FILE_OUTPUT; inst->I.DstReg.Index = i; - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; - inst->I.SrcReg[0].File = PROGRAM_CONSTANT; + inst->I.SrcReg[0].File = RC_FILE_CONSTANT; inst->I.SrcReg[0].Index = 0; - inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW; compiler->Base.Program.OutputsWritten |= 1 << i; } } } -static void nqssadceInit(struct nqssadce_state* s) +static void dataflow_outputs_mark_used(void * userdata, void * data, + void (*callback)(void *, unsigned int, unsigned int)) { - struct r300_vertex_program_compiler * compiler = s->UserData; + struct r300_vertex_program_compiler * c = userdata; int i; - for(i = 0; i < VERT_RESULT_MAX; ++i) { - if (compiler->RequiredOutputs & (1 << i)) - s->Outputs[i].Sourced = WRITEMASK_XYZW; + for(i = 0; i < 32; ++i) { + if (c->RequiredOutputs & (1 << i)) + callback(data, i, RC_MASK_XYZW); } } -static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg) +static int swizzle_is_native(rc_opcode opcode, struct rc_src_register reg) { (void) opcode; (void) reg; - return GL_TRUE; + return 1; } +static struct rc_swizzle_caps r300_vertprog_swizzle_caps = { + .IsNative = &swizzle_is_native, + .Split = 0 /* should never be called */ +}; + void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler) { + compiler->Base.SwizzleCaps = &r300_vertprog_swizzle_caps; + addArtificialOutputs(compiler); { @@ -603,7 +588,7 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler) if (compiler->Base.Debug) { fprintf(stderr, "Vertex program after native rewrite:\n"); - rc_print_program(&compiler->Base.Program); + rc_print_program(&compiler->Base.Program, 0); fflush(stderr); } @@ -620,26 +605,22 @@ void r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler) if (compiler->Base.Debug) { fprintf(stderr, "Vertex program after source conflict resolve:\n"); - rc_print_program(&compiler->Base.Program); + rc_print_program(&compiler->Base.Program, 0); fflush(stderr); } - { - struct radeon_nqssadce_descr nqssadce = { - .Init = &nqssadceInit, - .IsNativeSwizzle = &swizzleIsNative, - .BuildSwizzle = NULL - }; - radeonNqssaDce(&compiler->Base, &nqssadce, compiler); + rc_dataflow_annotate(&compiler->Base, &dataflow_outputs_mark_used, compiler); + rc_dataflow_dealias(&compiler->Base); + rc_dataflow_swizzles(&compiler->Base); - /* We need this step for reusing temporary registers */ - allocate_temporary_registers(compiler); + /* This invalidates dataflow annotations and should be replaced + * by a future generic register allocation pass. */ + allocate_temporary_registers(compiler); - if (compiler->Base.Debug) { - fprintf(stderr, "Vertex program after NQSSADCE:\n"); - rc_print_program(&compiler->Base.Program); - fflush(stderr); - } + if (compiler->Base.Debug) { + fprintf(stderr, "Vertex program after dataflow:\n"); + rc_print_program(&compiler->Base.Program, 0); + fflush(stderr); } translate_vertex_program(compiler); diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c index 7e2faed6908..971465e3591 100644 --- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.c @@ -27,15 +27,17 @@ #include "r500_fragprog.h" +#include <stdio.h> + #include "../r300_reg.h" -static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int tmu) +static struct rc_src_register shadow_ambient(struct radeon_compiler * c, int tmu) { - struct prog_src_register reg = { 0, }; + struct rc_src_register reg = { 0, }; - reg.File = PROGRAM_STATE_VAR; + reg.File = RC_FILE_CONSTANT; reg.Index = rc_constants_add_state(&c->Program.Constants, RC_STATE_SHADOW_AMBIENT, tmu); - reg.Swizzle = SWIZZLE_WWWW; + reg.Swizzle = RC_SWIZZLE_WWWW; return reg; } @@ -44,7 +46,7 @@ static struct prog_src_register shadow_ambient(struct radeon_compiler * c, int t * - implement texture compare (shadow extensions) * - extract non-native source / destination operands */ -GLboolean r500_transform_TEX( +int r500_transform_TEX( struct radeon_compiler * c, struct rc_instruction * inst, void* data) @@ -52,77 +54,77 @@ GLboolean r500_transform_TEX( 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; + if (inst->I.Opcode != RC_OPCODE_TEX && + inst->I.Opcode != RC_OPCODE_TXB && + inst->I.Opcode != RC_OPCODE_TXP && + inst->I.Opcode != RC_OPCODE_KIL) + return 0; /* ARB_shadow & EXT_shadow_funcs */ - if (inst->I.Opcode != OPCODE_KIL && + if (inst->I.Opcode != RC_OPCODE_KIL && c->Program.ShadowSamplers & (1 << inst->I.TexSrcUnit)) { - GLuint comparefunc = GL_NEVER + compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + rc_compare_func comparefunc = compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; - if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) { - inst->I.Opcode = OPCODE_MOV; + if (comparefunc == RC_COMPARE_FUNC_NEVER || comparefunc == RC_COMPARE_FUNC_ALWAYS) { + inst->I.Opcode = RC_OPCODE_MOV; - if (comparefunc == GL_ALWAYS) { - inst->I.SrcReg[0].File = PROGRAM_BUILTIN; - inst->I.SrcReg[0].Swizzle = SWIZZLE_1111; + if (comparefunc == RC_COMPARE_FUNC_ALWAYS) { + inst->I.SrcReg[0].File = RC_FILE_NONE; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_1111; } else { inst->I.SrcReg[0] = shadow_ambient(c, inst->I.TexSrcUnit); } - return GL_TRUE; + return 1; } 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; + rc_compare_func comparefunc = compiler->state.unit[inst->I.TexSrcUnit].texture_compare_func; + unsigned int 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.Opcode = RC_OPCODE_RCP; + inst_rcp->I.DstReg.File = RC_FILE_TEMPORARY; inst_rcp->I.DstReg.Index = rc_find_free_temporary(c); - inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + inst_rcp->I.DstReg.WriteMask = RC_MASK_W; inst_rcp->I.SrcReg[0] = inst->I.SrcReg[0]; - inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + inst_rcp->I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW; inst_cmp->I.DstReg = inst->I.DstReg; - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = rc_find_free_temporary(c); - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; - inst_mad->I.Opcode = OPCODE_MAD; - inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.Opcode = RC_OPCODE_MAD; + inst_mad->I.DstReg.File = RC_FILE_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[0].Swizzle = RC_SWIZZLE_ZZZZ; + inst_mad->I.SrcReg[1].File = RC_FILE_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[1].Swizzle = RC_SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].File = RC_FILE_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); + inst_mad->I.SrcReg[2].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_Z); else if (depthmode == 2) /* GL_ALPHA */ - inst_mad->I.SrcReg[2].Swizzle = SWIZZLE_WWWW; + inst_mad->I.SrcReg[2].Swizzle = RC_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; + if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GEQUAL) + inst_mad->I.SrcReg[2].Negate = inst_mad->I.SrcReg[2].Negate ^ RC_MASK_XYZW; else - inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ NEGATE_XYZW; + inst_mad->I.SrcReg[0].Negate = inst_mad->I.SrcReg[0].Negate ^ RC_MASK_XYZW; - inst_cmp->I.Opcode = OPCODE_CMP; + inst_cmp->I.Opcode = RC_OPCODE_CMP; /* DstReg has been filled out above */ - inst_cmp->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_cmp->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst_cmp->I.SrcReg[0].Index = inst_mad->I.DstReg.Index; - if (comparefunc == GL_LESS || comparefunc == GL_GREATER) { + if (comparefunc == RC_COMPARE_FUNC_LESS || comparefunc == RC_COMPARE_FUNC_GREATER) { pass = 1; fail = 2; } else { @@ -130,131 +132,133 @@ GLboolean r500_transform_TEX( fail = 1; } - inst_cmp->I.SrcReg[pass].File = PROGRAM_BUILTIN; - inst_cmp->I.SrcReg[pass].Swizzle = SWIZZLE_1111; + inst_cmp->I.SrcReg[pass].File = RC_FILE_NONE; + inst_cmp->I.SrcReg[pass].Swizzle = RC_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) { + if (inst->I.Opcode != RC_OPCODE_KIL && inst->I.DstReg.File != RC_FILE_TEMPORARY) { struct rc_instruction * inst_mov = rc_insert_new_instruction(c, inst); - inst_mov->I.Opcode = OPCODE_MOV; + inst_mov->I.Opcode = RC_OPCODE_MOV; inst_mov->I.DstReg = inst->I.DstReg; - inst_mov->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mov->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst_mov->I.SrcReg[0].Index = rc_find_free_temporary(c); - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; } /* Cannot read texture coordinate from constants file */ - if (inst->I.SrcReg[0].File != PROGRAM_TEMPORARY && inst->I.SrcReg[0].File != PROGRAM_INPUT) { + if (inst->I.SrcReg[0].File != RC_FILE_TEMPORARY && inst->I.SrcReg[0].File != RC_FILE_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.Opcode = RC_OPCODE_MOV; + inst_mov->I.DstReg.File = RC_FILE_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].File = RC_FILE_TEMPORARY; inst->I.SrcReg[0].Index = inst_mov->I.DstReg.Index; } - return GL_TRUE; + return 1; } -GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg) +static int r500_swizzle_is_native(rc_opcode opcode, struct rc_src_register reg) { - GLuint relevant; + unsigned int relevant; int i; - if (opcode == OPCODE_TEX || - opcode == OPCODE_TXB || - opcode == OPCODE_TXP || - opcode == OPCODE_KIL) { + if (opcode == RC_OPCODE_TEX || + opcode == RC_OPCODE_TXB || + opcode == RC_OPCODE_TXP || + opcode == RC_OPCODE_KIL) { if (reg.Abs) - return GL_FALSE; + return 0; - if (opcode == OPCODE_KIL && (reg.Swizzle != SWIZZLE_NOOP || reg.Negate != NEGATE_NONE)) - return GL_FALSE; + if (opcode == RC_OPCODE_KIL && (reg.Swizzle != RC_SWIZZLE_XYZW || reg.Negate != RC_MASK_NONE)) + return 0; if (reg.Negate) - reg.Negate ^= NEGATE_XYZW; + reg.Negate ^= RC_MASK_XYZW; for(i = 0; i < 4; ++i) { - GLuint swz = GET_SWZ(reg.Swizzle, i); - if (swz == SWIZZLE_NIL) { + unsigned int swz = GET_SWZ(reg.Swizzle, i); + if (swz == RC_SWIZZLE_UNUSED) { reg.Negate &= ~(1 << i); continue; } if (swz >= 4) - return GL_FALSE; + return 0; } if (reg.Negate) - return GL_FALSE; + return 0; - return GL_TRUE; - } else if (opcode == OPCODE_DDX || opcode == OPCODE_DDY) { + return 1; + } else if (opcode == RC_OPCODE_DDX || opcode == RC_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; + if (reg.Swizzle == RC_SWIZZLE_XYZW && !reg.Abs && !reg.Negate) + return 1; - return GL_FALSE; + return 0; } else { /* ALU instructions support almost everything */ if (reg.Abs) - return GL_TRUE; + return 1; relevant = 0; for(i = 0; i < 3; ++i) { - GLuint swz = GET_SWZ(reg.Swizzle, i); - if (swz != SWIZZLE_NIL && swz != SWIZZLE_ZERO) + unsigned int swz = GET_SWZ(reg.Swizzle, i); + if (swz != RC_SWIZZLE_UNUSED && swz != RC_SWIZZLE_ZERO) relevant |= 1 << i; } if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant)) - return GL_FALSE; + return 0; - return GL_TRUE; + return 1; } } /** - * Implement a MOV with a potentially non-native swizzle. + * Split source register access. * * The only thing we *cannot* do in an ALU instruction is per-component - * negation. Therefore, we split the MOV into two instructions when necessary. + * negation. */ -void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src) +static void r500_swizzle_split(struct rc_src_register src, unsigned int usemask, + struct rc_swizzle_split * split) { - GLuint negatebase[2] = { 0, 0 }; + unsigned int negatebase[2] = { 0, 0 }; int i; for(i = 0; i < 4; ++i) { - GLuint swz = GET_SWZ(src.Swizzle, i); - if (swz == SWIZZLE_NIL) + unsigned int swz = GET_SWZ(src.Swizzle, i); + if (swz == RC_SWIZZLE_UNUSED || !GET_BIT(usemask, i)) continue; negatebase[GET_BIT(src.Negate, i)] |= 1 << i; } + split->NumPhases = 0; + 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; + split->Phase[split->NumPhases++] = negatebase[i]; } } +struct rc_swizzle_caps r500_swizzle_caps = { + .IsNative = r500_swizzle_is_native, + .Split = r500_swizzle_split +}; static char *toswiz(int swiz_val) { switch(swiz_val) { diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h index 9091f65cd20..92ac75d5fd4 100644 --- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog.h @@ -33,21 +33,16 @@ #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" +#include "radeon_swizzle.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 struct rc_swizzle_caps r500_swizzle_caps; -extern GLboolean r500_transform_TEX( +extern int r500_transform_TEX( struct radeon_compiler * c, struct rc_instruction * inst, void* data); diff --git a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c index d694725c9bb..2f0c0d5283c 100644 --- a/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c +++ b/src/mesa/drivers/dri/r300/compiler/r500_fragprog_emit.c @@ -60,63 +60,63 @@ } while(0) -static GLuint translate_rgb_op(struct r300_fragment_program_compiler *c, GLuint opcode) +static unsigned int translate_rgb_op(struct r300_fragment_program_compiler *c, rc_opcode 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; + case RC_OPCODE_CMP: return R500_ALU_RGBA_OP_CMP; + case RC_OPCODE_DDX: return R500_ALU_RGBA_OP_MDH; + case RC_OPCODE_DDY: return R500_ALU_RGBA_OP_MDV; + case RC_OPCODE_DP3: return R500_ALU_RGBA_OP_DP3; + case RC_OPCODE_DP4: return R500_ALU_RGBA_OP_DP4; + case RC_OPCODE_FRC: return R500_ALU_RGBA_OP_FRC; default: error("translate_rgb_op(%d): unknown opcode\n", opcode); /* fall through */ - case OPCODE_NOP: + case RC_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; + case RC_OPCODE_MAD: return R500_ALU_RGBA_OP_MAD; + case RC_OPCODE_MAX: return R500_ALU_RGBA_OP_MAX; + case RC_OPCODE_MIN: return R500_ALU_RGBA_OP_MIN; + case RC_OPCODE_REPL_ALPHA: return R500_ALU_RGBA_OP_SOP; } } -static GLuint translate_alpha_op(struct r300_fragment_program_compiler *c, GLuint opcode) +static unsigned int translate_alpha_op(struct r300_fragment_program_compiler *c, rc_opcode 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; + case RC_OPCODE_CMP: return R500_ALPHA_OP_CMP; + case RC_OPCODE_COS: return R500_ALPHA_OP_COS; + case RC_OPCODE_DDX: return R500_ALPHA_OP_MDH; + case RC_OPCODE_DDY: return R500_ALPHA_OP_MDV; + case RC_OPCODE_DP3: return R500_ALPHA_OP_DP; + case RC_OPCODE_DP4: return R500_ALPHA_OP_DP; + case RC_OPCODE_EX2: return R500_ALPHA_OP_EX2; + case RC_OPCODE_FRC: return R500_ALPHA_OP_FRC; + case RC_OPCODE_LG2: return R500_ALPHA_OP_LN2; default: error("translate_alpha_op(%d): unknown opcode\n", opcode); /* fall through */ - case OPCODE_NOP: + case RC_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; + case RC_OPCODE_MAD: return R500_ALPHA_OP_MAD; + case RC_OPCODE_MAX: return R500_ALPHA_OP_MAX; + case RC_OPCODE_MIN: return R500_ALPHA_OP_MIN; + case RC_OPCODE_RCP: return R500_ALPHA_OP_RCP; + case RC_OPCODE_RSQ: return R500_ALPHA_OP_RSQ; + case RC_OPCODE_SIN: return R500_ALPHA_OP_SIN; } } -static GLuint fix_hw_swizzle(GLuint swz) +static unsigned int fix_hw_swizzle(unsigned int swz) { if (swz == 5) swz = 6; - if (swz == SWIZZLE_NIL) swz = 4; + if (swz == RC_SWIZZLE_UNUSED) swz = 4; return swz; } -static GLuint translate_arg_rgb(struct radeon_pair_instruction *inst, int arg) +static unsigned int translate_arg_rgb(struct radeon_pair_instruction *inst, int arg) { - GLuint t = inst->RGB.Arg[arg].Source; + unsigned int t = inst->RGB.Arg[arg].Source; int comp; t |= inst->RGB.Arg[arg].Negate << 11; t |= inst->RGB.Arg[arg].Abs << 12; @@ -127,22 +127,22 @@ static GLuint translate_arg_rgb(struct radeon_pair_instruction *inst, int arg) return t; } -static GLuint translate_arg_alpha(struct radeon_pair_instruction *inst, int i) +static unsigned int translate_arg_alpha(struct radeon_pair_instruction *inst, int i) { - GLuint t = inst->Alpha.Arg[i].Source; + unsigned int 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) +static void use_temporary(struct r500_fragment_program_code* code, unsigned int 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) +static unsigned int use_source(struct r500_fragment_program_code* code, struct radeon_pair_instruction_source src) { if (!src.Constant) use_temporary(code, src.Index); @@ -153,13 +153,13 @@ static GLuint use_source(struct r500_fragment_program_code* code, struct radeon_ /** * Emit a paired ALU instruction. */ -static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) +static int 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; + return 0; } int ip = ++code->inst_end; @@ -177,7 +177,7 @@ static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) 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; + c->code->writes_depth = 1; } code->inst[ip].inst4 |= R500_ALPHA_ADDRD(inst->Alpha.DestIndex); @@ -206,12 +206,12 @@ static GLboolean emit_paired(void *data, struct radeon_pair_instruction *inst) 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; + return 1; } -static GLuint translate_strq_swizzle(GLuint swizzle) +static unsigned int translate_strq_swizzle(unsigned int swizzle) { - GLuint swiz = 0; + unsigned int swiz = 0; int i; for (i = 0; i < 4; i++) swiz |= (GET_SWZ(swizzle, i) & 0x3) << i*2; @@ -221,13 +221,13 @@ static GLuint translate_strq_swizzle(GLuint swizzle) /** * Emit a single TEX instruction */ -static GLboolean emit_tex(void *data, struct radeon_pair_texture_instruction *inst) +static int 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; + return 0; } int ip = ++code->inst_end; @@ -238,7 +238,7 @@ static GLboolean emit_tex(void *data, struct radeon_pair_texture_instruction *in code->inst[ip].inst1 = R500_TEX_ID(inst->TexSrcUnit) | R500_TEX_SEM_ACQUIRE | R500_TEX_IGNORE_UNCOVERED; - if (inst->TexSrcTarget == TEXTURE_RECT_INDEX) + if (inst->TexSrcTarget == RC_TEXTURE_RECT) code->inst[ip].inst1 |= R500_TEX_UNSCALED; switch (inst->Opcode) { @@ -264,7 +264,7 @@ static GLboolean emit_tex(void *data, struct radeon_pair_texture_instruction *in | 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; + return 1; } static const struct radeon_pair_handler pair_handler = { @@ -277,7 +277,7 @@ void r500BuildFragmentProgramHwCode(struct r300_fragment_program_compiler *compi { struct r500_fragment_program_code *code = &compiler->code->code.r500; - _mesa_bzero(code, sizeof(*code)); + memset(code, 0, sizeof(*code)); code->max_temp_idx = 1; code->inst_end = -1; diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.c b/src/mesa/drivers/dri/r300/compiler/radeon_code.c index c7923004df9..1a3d8bb6412 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_code.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.c @@ -25,11 +25,13 @@ * */ -#include "main/mtypes.h" -#include "shader/prog_instruction.h" - #include "radeon_code.h" +#include <stdlib.h> +#include <string.h> + +#include "radeon_program.h" + void rc_constants_init(struct rc_constant_list * c) { memset(c, 0, sizeof(*c)); @@ -143,7 +145,7 @@ unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float da 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); + *swizzle = RC_MAKE_SWIZZLE(comp, comp, comp, comp); return index; } } @@ -156,7 +158,7 @@ unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float da 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); + *swizzle = RC_MAKE_SWIZZLE(comp, comp, comp, comp); return free_index; } @@ -164,7 +166,7 @@ unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float da constant.Type = RC_CONSTANT_IMMEDIATE; constant.Size = 1; constant.u.Immediate[0] = data; - *swizzle = SWIZZLE_XXXX; + *swizzle = RC_SWIZZLE_XXXX; return rc_constants_add(c, &constant); } diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_code.h b/src/mesa/drivers/dri/r300/compiler/radeon_code.h index 3e88554ba19..75069e841e3 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_code.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_code.h @@ -89,6 +89,23 @@ unsigned rc_constants_add_immediate_vec4(struct rc_constant_list * c, const floa unsigned rc_constants_add_immediate_scalar(struct rc_constant_list * c, float data, unsigned * swizzle); /** + * Compare functions. + * + * \note By design, RC_COMPARE_FUNC_xxx + GL_NEVER gives you + * the correct GL compare function. + */ +typedef enum { + RC_COMPARE_FUNC_NEVER = 0, + RC_COMPARE_FUNC_LESS, + RC_COMPARE_FUNC_EQUAL, + RC_COMPARE_FUNC_LEQUAL, + RC_COMPARE_FUNC_GREATER, + RC_COMPARE_FUNC_NOTEQUAL, + RC_COMPARE_FUNC_GEQUAL, + RC_COMPARE_FUNC_ALWAYS +} rc_compare_func; + +/** * Stores state that influences the compilation of a fragment program. */ struct r300_fragment_program_external_state { @@ -105,10 +122,12 @@ struct r300_fragment_program_external_state { /** * 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] + * this field specifies the compare function. + * + * Otherwise, this field is \ref RC_COMPARE_FUNC_NEVER (aka 0). * * Otherwise, this field is 0. + * \sa rc_compare_func */ unsigned texture_compare_func : 3; } unit[16]; diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c index da950d52894..d0b78ec1c88 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.c @@ -23,6 +23,8 @@ #include "radeon_compiler.h" #include <stdarg.h> +#include <stdio.h> +#include <stdlib.h> #include "radeon_program.h" @@ -34,7 +36,7 @@ void rc_init(struct radeon_compiler * 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; + c->Program.Instructions.I.Opcode = RC_OPCODE_ILLEGAL_OPCODE; } void rc_destroy(struct radeon_compiler * c) @@ -60,7 +62,7 @@ void rc_error(struct radeon_compiler * c, const char * fmt, ...) { va_list ap; - c->Error = GL_TRUE; + c->Error = 1; if (!c->ErrorMsg) { /* Only remember the first error */ @@ -91,22 +93,57 @@ void rc_error(struct radeon_compiler * c, const char * fmt, ...) } } +int rc_if_fail_helper(struct radeon_compiler * c, const char * file, int line, const char * assertion) +{ + rc_error(c, "ICE at %s:%i: assertion failed: %s\n", file, line, assertion); + return 1; +} + +/** + * Recompute c->Program.InputsRead and c->Program.OutputsWritten + * based on which inputs and outputs are actually referenced + * in program instructions. + */ +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) + { + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + int i; + + for (i = 0; i < opcode->NumSrcRegs; ++i) { + if (inst->I.SrcReg[i].File == RC_FILE_INPUT) + c->Program.InputsRead |= 1 << inst->I.SrcReg[i].Index; + } + + if (opcode->HasDstReg) { + if (inst->I.DstReg.File == RC_FILE_OUTPUT) + c->Program.OutputsWritten |= 1 << inst->I.DstReg.Index; + } + } +} + /** * 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) +void rc_move_input(struct radeon_compiler * c, unsigned input, struct rc_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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(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) { + for(i = 0; i < opcode->NumSrcRegs; ++i) { + if (inst->I.SrcReg[i].File == RC_FILE_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); @@ -134,10 +171,10 @@ void rc_move_output(struct radeon_compiler * c, unsigned output, unsigned new_ou 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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - if (numdsts) { - if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) { + if (opcode->HasDstReg) { + if (inst->I.DstReg.File == RC_FILE_OUTPUT && inst->I.DstReg.Index == output) { inst->I.DstReg.Index = new_output; inst->I.DstReg.WriteMask &= writemask; @@ -157,33 +194,33 @@ void rc_copy_output(struct radeon_compiler * c, unsigned output, unsigned dup_ou 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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); - if (numdsts) { - if (inst->I.DstReg.File == PROGRAM_OUTPUT && inst->I.DstReg.Index == output) { - inst->I.DstReg.File = PROGRAM_TEMPORARY; + if (opcode->HasDstReg) { + if (inst->I.DstReg.File == RC_FILE_OUTPUT && inst->I.DstReg.Index == output) { + inst->I.DstReg.File = RC_FILE_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.Opcode = RC_OPCODE_MOV; + inst->I.DstReg.File = RC_FILE_OUTPUT; inst->I.DstReg.Index = output; - inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst->I.SrcReg[0].Index = tempreg; - inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + inst->I.SrcReg[0].Swizzle = RC_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.Opcode = RC_OPCODE_MOV; + inst->I.DstReg.File = RC_FILE_OUTPUT; inst->I.DstReg.Index = dup_output; - inst->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[0].File = RC_FILE_TEMPORARY; inst->I.SrcReg[0].Index = tempreg; - inst->I.SrcReg[0].Swizzle = SWIZZLE_XYZW; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW; c->Program.OutputsWritten |= 1 << dup_output; } @@ -201,59 +238,59 @@ void rc_transform_fragment_wpos(struct radeon_compiler * c, unsigned wpos, unsig /* perspective divide */ struct rc_instruction * inst_rcp = rc_insert_new_instruction(c, &c->Program.Instructions); - inst_rcp->I.Opcode = OPCODE_RCP; + inst_rcp->I.Opcode = RC_OPCODE_RCP; - inst_rcp->I.DstReg.File = PROGRAM_TEMPORARY; + inst_rcp->I.DstReg.File = RC_FILE_TEMPORARY; inst_rcp->I.DstReg.Index = tempregi; - inst_rcp->I.DstReg.WriteMask = WRITEMASK_W; + inst_rcp->I.DstReg.WriteMask = RC_MASK_W; - inst_rcp->I.SrcReg[0].File = PROGRAM_INPUT; + inst_rcp->I.SrcReg[0].File = RC_FILE_INPUT; inst_rcp->I.SrcReg[0].Index = new_input; - inst_rcp->I.SrcReg[0].Swizzle = SWIZZLE_WWWW; + inst_rcp->I.SrcReg[0].Swizzle = RC_SWIZZLE_WWWW; struct rc_instruction * inst_mul = rc_insert_new_instruction(c, inst_rcp); - inst_mul->I.Opcode = OPCODE_MUL; + inst_mul->I.Opcode = RC_OPCODE_MUL; - inst_mul->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mul->I.DstReg.File = RC_FILE_TEMPORARY; inst_mul->I.DstReg.Index = tempregi; - inst_mul->I.DstReg.WriteMask = WRITEMASK_XYZ; + inst_mul->I.DstReg.WriteMask = RC_MASK_XYZ; - inst_mul->I.SrcReg[0].File = PROGRAM_INPUT; + inst_mul->I.SrcReg[0].File = RC_FILE_INPUT; inst_mul->I.SrcReg[0].Index = new_input; - inst_mul->I.SrcReg[1].File = PROGRAM_TEMPORARY; + inst_mul->I.SrcReg[1].File = RC_FILE_TEMPORARY; inst_mul->I.SrcReg[1].Index = tempregi; - inst_mul->I.SrcReg[1].Swizzle = SWIZZLE_WWWW; + inst_mul->I.SrcReg[1].Swizzle = RC_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.Opcode = RC_OPCODE_MAD; - inst_mad->I.DstReg.File = PROGRAM_TEMPORARY; + inst_mad->I.DstReg.File = RC_FILE_TEMPORARY; inst_mad->I.DstReg.Index = tempregi; - inst_mad->I.DstReg.WriteMask = WRITEMASK_XYZ; + inst_mad->I.DstReg.WriteMask = RC_MASK_XYZ; - inst_mad->I.SrcReg[0].File = PROGRAM_TEMPORARY; + inst_mad->I.SrcReg[0].File = RC_FILE_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[0].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_ZERO); - inst_mad->I.SrcReg[1].File = PROGRAM_STATE_VAR; + inst_mad->I.SrcReg[1].File = RC_FILE_CONSTANT; 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[1].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_ZERO); - inst_mad->I.SrcReg[2].File = PROGRAM_STATE_VAR; + inst_mad->I.SrcReg[2].File = RC_FILE_CONSTANT; 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); + inst_mad->I.SrcReg[2].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_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); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); unsigned i; - for(i = 0; i < numsrcs; i++) { - if (inst->I.SrcReg[i].File == PROGRAM_INPUT && + for(i = 0; i < opcode->NumSrcRegs; i++) { + if (inst->I.SrcReg[i].File == RC_FILE_INPUT && inst->I.SrcReg[i].Index == wpos) { - inst->I.SrcReg[i].File = PROGRAM_TEMPORARY; + inst->I.SrcReg[i].File = RC_FILE_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 index e63ab8840ab..87a732cd90d 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h @@ -23,35 +23,11 @@ #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" +#include "radeon_program.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 rc_swizzle_caps; struct radeon_compiler { struct memory_pool Pool; @@ -59,6 +35,14 @@ struct radeon_compiler { unsigned Debug:1; unsigned Error:1; char * ErrorMsg; + + /** + * Variables used internally, not be touched by callers + * of the compiler + */ + /*@{*/ + struct rc_swizzle_caps * SwizzleCaps; + /*@}*/ }; void rc_init(struct radeon_compiler * c); @@ -67,11 +51,26 @@ 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); +int rc_if_fail_helper(struct radeon_compiler * c, const char * file, int line, const char * assertion); + +/** + * This macro acts like an if-statement that can be used to implement + * non-aborting assertions in the compiler. + * + * It checks whether \p cond is true. If not, an internal compiler error is + * flagged and the if-clause is run. + * + * A typical use-case would be: + * + * if (rc_assert(c, condition-that-must-be-true)) + * return; + */ +#define rc_assert(c, cond) \ + (!(cond) && rc_if_fail_helper(c, __FILE__, __LINE__, #cond)) 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_input(struct radeon_compiler * c, unsigned input, struct rc_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); @@ -97,7 +96,7 @@ 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; + uint32_t RequiredOutputs; void * UserData; void (*SetHwInputOutput)(struct r300_vertex_program_compiler * c); diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c new file mode 100644 index 00000000000..af6777a7bd3 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.c @@ -0,0 +1,106 @@ +/* + * 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 "radeon_dataflow.h" + +#include "radeon_compiler.h" + + +static void add_ref_to_vector(struct rc_dataflow_ref * ref, struct rc_dataflow_vector * vector) +{ + ref->Vector = vector; + ref->Prev = &vector->Refs; + ref->Next = vector->Refs.Next; + ref->Prev->Next = ref; + ref->Next->Prev = ref; +} + +struct rc_dataflow_ref * rc_dataflow_create_ref(struct radeon_compiler * c, + struct rc_dataflow_vector * vector, struct rc_instruction * inst) +{ + struct rc_dataflow_ref * ref = memory_pool_malloc(&c->Pool, sizeof(struct rc_dataflow_ref)); + ref->ReadInstruction = inst; + ref->UseMask = 0; + + add_ref_to_vector(ref, vector); + + return ref; +} + +struct rc_dataflow_vector * rc_dataflow_create_vector(struct radeon_compiler * c, + rc_register_file file, unsigned int index, struct rc_instruction * inst) +{ + struct rc_dataflow_vector * vec = memory_pool_malloc(&c->Pool, sizeof(struct rc_dataflow_vector)); + + memset(vec, 0, sizeof(struct rc_dataflow_vector)); + vec->File = file; + vec->Index = index; + vec->WriteInstruction = inst; + + vec->Refs.Next = vec->Refs.Prev = &vec->Refs; + + return vec; +} + +void rc_dataflow_remove_ref(struct rc_dataflow_ref * ref) +{ + ref->Prev->Next = ref->Next; + ref->Next->Prev = ref->Prev; +} + +void rc_dataflow_remove_instruction(struct rc_instruction * inst) +{ + for(unsigned int i = 0; i < 3; ++i) { + if (inst->Dataflow.SrcReg[i]) { + rc_dataflow_remove_ref(inst->Dataflow.SrcReg[i]); + inst->Dataflow.SrcReg[i] = 0; + } + if (inst->Dataflow.SrcRegAddress[i]) { + rc_dataflow_remove_ref(inst->Dataflow.SrcRegAddress[i]); + inst->Dataflow.SrcRegAddress[i] = 0; + } + } + + if (inst->Dataflow.DstReg) { + while(inst->Dataflow.DstReg->Refs.Next != &inst->Dataflow.DstReg->Refs) { + struct rc_dataflow_ref * ref = inst->Dataflow.DstReg->Refs.Next; + rc_dataflow_remove_ref(ref); + if (inst->Dataflow.DstRegPrev) + add_ref_to_vector(ref, inst->Dataflow.DstRegPrev->Vector); + } + + inst->Dataflow.DstReg->WriteInstruction = 0; + inst->Dataflow.DstReg = 0; + } + + if (inst->Dataflow.DstRegPrev) { + rc_dataflow_remove_ref(inst->Dataflow.DstRegPrev); + inst->Dataflow.DstRegPrev = 0; + } + + inst->Dataflow.DstRegAliased = 0; +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h new file mode 100644 index 00000000000..c9856affe88 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow.h @@ -0,0 +1,113 @@ +/* + * 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. + * + */ + +#ifndef RADEON_DATAFLOW_H +#define RADEON_DATAFLOW_H + +#include "radeon_program_constants.h" + +struct radeon_compiler; +struct rc_instruction; +struct rc_swizzle_caps; + +struct rc_dataflow_vector; + +struct rc_dataflow_ref { + struct rc_dataflow_vector * Vector; + + /** + * Linked list of references to the above-mentioned vector. + * The linked list is \em not sorted. + */ + /*@{*/ + struct rc_dataflow_ref * Prev; + struct rc_dataflow_ref * Next; + /*@}*/ + + unsigned int UseMask:4; + struct rc_instruction * ReadInstruction; +}; + +struct rc_dataflow_vector { + rc_register_file File:3; + unsigned int Index:RC_REGISTER_INDEX_BITS; + + /** For private use in compiler passes. MUST BE RESET TO 0 by the end of each pass. + * The annotate pass uses this bit to track whether a vector is in the + * update stack. + */ + unsigned int PassBit:1; + /** Which of the components have been written with useful values */ + unsigned int ValidMask:4; + /** Which of the components are used downstream */ + unsigned int UseMask:4; + /** The instruction that produced this vector */ + struct rc_instruction * WriteInstruction; + + /** Linked list of references to this vector */ + struct rc_dataflow_ref Refs; +}; + +struct rc_instruction_dataflow { + struct rc_dataflow_ref * SrcReg[3]; + struct rc_dataflow_ref * SrcRegAddress[3]; + + /** Reference the components of the destination register + * that are carried over without being overwritten */ + struct rc_dataflow_ref * DstRegPrev; + /** Indicates whether the destination register was in use + * before this instruction */ + unsigned int DstRegAliased:1; + struct rc_dataflow_vector * DstReg; +}; + +/** + * General functions for manipulating the dataflow structures. + */ +/*@{*/ +struct rc_dataflow_ref * rc_dataflow_create_ref(struct radeon_compiler * c, + struct rc_dataflow_vector * vector, struct rc_instruction * inst); +struct rc_dataflow_vector * rc_dataflow_create_vector(struct radeon_compiler * c, + rc_register_file file, unsigned int index, struct rc_instruction * inst); +void rc_dataflow_remove_ref(struct rc_dataflow_ref * ref); + +void rc_dataflow_remove_instruction(struct rc_instruction * inst); +/*@}*/ + + +/** + * Compiler passes based on dataflow structures. + */ +/*@{*/ +typedef void (*rc_dataflow_mark_outputs_fn)(void * userdata, void * data, + void (*mark_fn)(void * data, unsigned int index, unsigned int mask)); +void rc_dataflow_annotate(struct radeon_compiler * c, rc_dataflow_mark_outputs_fn dce, void * userdata); +void rc_dataflow_dealias(struct radeon_compiler * c); +void rc_dataflow_swizzles(struct radeon_compiler * c); +/*@}*/ + +#endif /* RADEON_DATAFLOW_H */ diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_annotate.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_annotate.c new file mode 100644 index 00000000000..41d175a22f2 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_annotate.c @@ -0,0 +1,365 @@ +/* + * 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 "radeon_dataflow.h" + +#include "radeon_compiler.h" + + +struct dataflow_state { + struct radeon_compiler * C; + unsigned int DCE:1; + unsigned int UpdateRunning:1; + + struct rc_dataflow_vector * Input[RC_REGISTER_MAX_INDEX]; + struct rc_dataflow_vector * Output[RC_REGISTER_MAX_INDEX]; + struct rc_dataflow_vector * Temporary[RC_REGISTER_MAX_INDEX]; + struct rc_dataflow_vector * Address; + + struct rc_dataflow_vector ** UpdateStack; + unsigned int UpdateStackSize; + unsigned int UpdateStackReserved; +}; + +static void mark_vector_use(struct dataflow_state * s, struct rc_dataflow_vector * vector, unsigned int mask); + +static struct rc_dataflow_vector * get_register_contents(struct dataflow_state * s, + rc_register_file file, unsigned int index) +{ + if (file == RC_FILE_INPUT || file == RC_FILE_OUTPUT || file == RC_FILE_TEMPORARY) { + if (index >= RC_REGISTER_MAX_INDEX) + return 0; /* cannot happen, but be defensive */ + + if (file == RC_FILE_TEMPORARY) + return s->Temporary[index]; + if (file == RC_FILE_INPUT) + return s->Input[index]; + if (file == RC_FILE_OUTPUT) + return s->Output[index]; + } + + if (file == RC_FILE_ADDRESS) + return s->Address; + + return 0; /* can happen, constant register file */ +} + +static void mark_ref_use(struct dataflow_state * s, struct rc_dataflow_ref * ref, unsigned int mask) +{ + if (!(mask & ~ref->UseMask)) + return; + + ref->UseMask |= mask; + mark_vector_use(s, ref->Vector, ref->UseMask); +} + +static void mark_source_use(struct dataflow_state * s, struct rc_instruction * inst, + unsigned int src, unsigned int srcmask) +{ + unsigned int refmask = 0; + + for(unsigned int i = 0; i < 4; ++i) { + if (GET_BIT(srcmask, i)) + refmask |= 1 << GET_SWZ(inst->I.SrcReg[src].Swizzle, i); + } + + /* get rid of spurious bits from ZERO, ONE, etc. swizzles */ + refmask &= RC_MASK_XYZW; + + if (!refmask) + return; /* can happen if the swizzle contains constant components */ + + if (inst->Dataflow.SrcReg[src]) + mark_ref_use(s, inst->Dataflow.SrcReg[src], refmask); + + if (inst->Dataflow.SrcRegAddress[src]) + mark_ref_use(s, inst->Dataflow.SrcRegAddress[src], RC_MASK_X); +} + +static void compute_sources_for_writemask( + struct rc_instruction * inst, + unsigned int writemask, + unsigned int *srcmasks) +{ + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + + srcmasks[0] = 0; + srcmasks[1] = 0; + srcmasks[2] = 0; + + if (inst->I.Opcode == RC_OPCODE_KIL) + srcmasks[0] |= RC_MASK_XYZW; + + if (!writemask) + return; + + if (opcode->IsComponentwise) { + for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) + srcmasks[src] |= writemask; + } else if (opcode->IsStandardScalar) { + for(unsigned int src = 0; src < opcode->NumSrcRegs; ++src) + srcmasks[src] |= RC_MASK_X; + } else { + switch(inst->I.Opcode) { + case RC_OPCODE_ARL: + srcmasks[0] |= RC_MASK_X; + break; + case RC_OPCODE_DP3: + srcmasks[0] |= RC_MASK_XYZ; + srcmasks[1] |= RC_MASK_XYZ; + break; + case RC_OPCODE_DP4: + srcmasks[0] |= RC_MASK_XYZW; + srcmasks[1] |= RC_MASK_XYZW; + break; + case RC_OPCODE_TEX: + case RC_OPCODE_TXB: + case RC_OPCODE_TXP: + srcmasks[0] |= RC_MASK_XYZW; + break; + case RC_OPCODE_DST: + srcmasks[0] |= 0x6; + srcmasks[1] |= 0xa; + break; + case RC_OPCODE_EXP: + case RC_OPCODE_LOG: + srcmasks[0] |= RC_MASK_XY; + break; + case RC_OPCODE_LIT: + srcmasks[0] |= 0xb; + break; + default: + break; + } + } +} + +static void mark_instruction_source_use(struct dataflow_state * s, + struct rc_instruction * inst, unsigned int writemask) +{ + unsigned int srcmasks[3]; + + compute_sources_for_writemask(inst, writemask, srcmasks); + + for(unsigned int src = 0; src < 3; ++src) + mark_source_use(s, inst, src, srcmasks[src]); +} + +static void run_update(struct dataflow_state * s) +{ + s->UpdateRunning = 1; + + while(s->UpdateStackSize) { + struct rc_dataflow_vector * vector = s->UpdateStack[--s->UpdateStackSize]; + vector->PassBit = 0; + + if (vector->WriteInstruction) { + struct rc_instruction * inst = vector->WriteInstruction; + + if (inst->Dataflow.DstRegPrev) { + unsigned int carryover = vector->UseMask & ~inst->I.DstReg.WriteMask; + + if (carryover) + mark_ref_use(s, inst->Dataflow.DstRegPrev, carryover); + } + + mark_instruction_source_use( + s, vector->WriteInstruction, + vector->UseMask & inst->I.DstReg.WriteMask); + } + } + + s->UpdateRunning = 0; +} + +static void mark_vector_use(struct dataflow_state * s, struct rc_dataflow_vector * vector, unsigned int mask) +{ + if (!(mask & ~vector->UseMask)) + return; /* no new used bits */ + + vector->UseMask |= mask; + if (vector->PassBit) + return; + + if (s->UpdateStackSize >= s->UpdateStackReserved) { + unsigned int new_reserve = 2 * s->UpdateStackReserved; + struct rc_dataflow_vector ** new_stack; + + if (!new_reserve) + new_reserve = 16; + + new_stack = memory_pool_malloc(&s->C->Pool, new_reserve * sizeof(struct rc_dataflow_vector *)); + memcpy(new_stack, s->UpdateStack, s->UpdateStackSize * sizeof(struct rc_dataflow_vector *)); + + s->UpdateStack = new_stack; + s->UpdateStackReserved = new_reserve; + } + + s->UpdateStack[s->UpdateStackSize++] = vector; + vector->PassBit = 1; + + if (!s->UpdateRunning) + run_update(s); +} + +static void annotate_instruction(struct dataflow_state * s, struct rc_instruction * inst) +{ + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + unsigned int src; + + for(src = 0; src < opcode->NumSrcRegs; ++src) { + struct rc_dataflow_vector * vector = get_register_contents(s, inst->I.SrcReg[src].File, inst->I.SrcReg[src].Index); + if (vector) { + inst->Dataflow.SrcReg[src] = rc_dataflow_create_ref(s->C, vector, inst); + } + if (inst->I.SrcReg[src].RelAddr) { + struct rc_dataflow_vector * addr = get_register_contents(s, RC_FILE_ADDRESS, 0); + if (addr) + inst->Dataflow.SrcRegAddress[src] = rc_dataflow_create_ref(s->C, addr, inst); + } + } + + mark_instruction_source_use(s, inst, 0); /* for KIL */ + + if (opcode->HasDstReg) { + struct rc_dataflow_vector * oldvec = get_register_contents(s, inst->I.DstReg.File, inst->I.DstReg.Index); + struct rc_dataflow_vector * newvec = rc_dataflow_create_vector(s->C, inst->I.DstReg.File, inst->I.DstReg.Index, inst); + + newvec->ValidMask = inst->I.DstReg.WriteMask; + + if (oldvec) { + unsigned int carryover = oldvec->ValidMask & ~inst->I.DstReg.WriteMask; + + if (oldvec->ValidMask) + inst->Dataflow.DstRegAliased = 1; + + if (carryover) { + inst->Dataflow.DstRegPrev = rc_dataflow_create_ref(s->C, oldvec, inst); + newvec->ValidMask |= carryover; + + if (!s->DCE) + mark_ref_use(s, inst->Dataflow.DstRegPrev, carryover); + } + } + + inst->Dataflow.DstReg = newvec; + + if (newvec->File == RC_FILE_TEMPORARY) + s->Temporary[newvec->Index] = newvec; + else if (newvec->File == RC_FILE_OUTPUT) + s->Output[newvec->Index] = newvec; + else + s->Address = newvec; + + if (!s->DCE) + mark_vector_use(s, newvec, inst->I.DstReg.WriteMask); + } +} + +static void init_inputs(struct dataflow_state * s) +{ + unsigned int index; + + for(index = 0; index < 32; ++index) { + if (s->C->Program.InputsRead & (1 << index)) { + s->Input[index] = rc_dataflow_create_vector(s->C, RC_FILE_INPUT, index, 0); + s->Input[index]->ValidMask = RC_MASK_XYZW; + } + } +} + +static void mark_output_use(void * data, unsigned int index, unsigned int mask) +{ + struct dataflow_state * s = data; + struct rc_dataflow_vector * vec = s->Output[index]; + + if (vec) + mark_vector_use(s, vec, mask); +} + +void rc_dataflow_annotate(struct radeon_compiler * c, rc_dataflow_mark_outputs_fn dce, void * userdata) +{ + struct dataflow_state s; + struct rc_instruction * inst; + + memset(&s, 0, sizeof(s)); + s.C = c; + s.DCE = dce ? 1 : 0; + + init_inputs(&s); + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + annotate_instruction(&s, inst); + } + + if (s.DCE) { + dce(userdata, &s, &mark_output_use); + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + + if (opcode->HasDstReg) { + unsigned int redundant_writes = inst->I.DstReg.WriteMask & ~inst->Dataflow.DstReg->UseMask; + + inst->I.DstReg.WriteMask &= ~redundant_writes; + + if (!inst->I.DstReg.WriteMask) { + struct rc_instruction * todelete = inst; + inst = inst->Prev; + rc_remove_instruction(todelete); + continue; + } + } + + unsigned int srcmasks[3]; + compute_sources_for_writemask(inst, inst->I.DstReg.WriteMask, srcmasks); + + for(unsigned int src = 0; src < 3; ++src) { + for(unsigned int chan = 0; chan < 4; ++chan) { + if (!GET_BIT(srcmasks[src], chan)) + SET_SWZ(inst->I.SrcReg[src].Swizzle, chan, RC_SWIZZLE_UNUSED); + } + + if (inst->Dataflow.SrcReg[src]) { + if (!inst->Dataflow.SrcReg[src]->UseMask) { + rc_dataflow_remove_ref(inst->Dataflow.SrcReg[src]); + inst->Dataflow.SrcReg[src] = 0; + } + } + + if (inst->Dataflow.SrcRegAddress[src]) { + if (!inst->Dataflow.SrcRegAddress[src]->UseMask) { + rc_dataflow_remove_ref(inst->Dataflow.SrcRegAddress[src]); + inst->Dataflow.SrcRegAddress[src] = 0; + } + } + } + } + + rc_calculate_inputs_outputs(c); + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_dealias.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_dealias.c new file mode 100644 index 00000000000..4596636970a --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_dealias.c @@ -0,0 +1,150 @@ +/* + * 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 "radeon_dataflow.h" + +#include "radeon_compiler.h" + + +#define DEALIAS_LIST_SIZE 128 + +struct dealias_state { + struct radeon_compiler * C; + + unsigned int OldIndex:RC_REGISTER_INDEX_BITS; + unsigned int NewIndex:RC_REGISTER_INDEX_BITS; + unsigned int DealiasFail:1; + + struct rc_dataflow_vector * List[DEALIAS_LIST_SIZE]; + unsigned int Length; +}; + +static void push_dealias_vector(struct dealias_state * s, struct rc_dataflow_vector * vec) +{ + if (s->Length >= DEALIAS_LIST_SIZE) { + rc_debug(s->C, "%s: list size exceeded\n", __FUNCTION__); + s->DealiasFail = 1; + return; + } + + if (rc_assert(s->C, vec->File == RC_FILE_TEMPORARY && vec->Index == s->OldIndex)) + return; + + s->List[s->Length++] = vec; +} + +static void run_dealias(struct dealias_state * s) +{ + unsigned int i; + + for(i = 0; i < s->Length && !s->DealiasFail; ++i) { + struct rc_dataflow_vector * vec = s->List[i]; + struct rc_dataflow_ref * ref; + + for(ref = vec->Refs.Next; ref != &vec->Refs; ref = ref->Next) { + if (ref->ReadInstruction->Dataflow.DstRegPrev == ref) + push_dealias_vector(s, ref->ReadInstruction->Dataflow.DstReg); + } + } + + if (s->DealiasFail) + return; + + for(i = 0; i < s->Length; ++i) { + struct rc_dataflow_vector * vec = s->List[i]; + struct rc_dataflow_ref * ref; + + vec->Index = s->NewIndex; + vec->WriteInstruction->I.DstReg.Index = s->NewIndex; + + for(ref = vec->Refs.Next; ref != &vec->Refs; ref = ref->Next) { + struct rc_instruction * inst = ref->ReadInstruction; + unsigned int i; + + for(i = 0; i < 3; ++i) { + if (inst->Dataflow.SrcReg[i] == ref) { + if (rc_assert(s->C, inst->I.SrcReg[i].File == RC_FILE_TEMPORARY && + inst->I.SrcReg[i].Index == s->OldIndex)) + return; + + inst->I.SrcReg[i].Index = s->NewIndex; + } + } + } + } +} + +/** + * Breaks register aliasing to reduce multiple assignments to a single register. + * + * This affects sequences like: + * MUL r0, ...; + * MAD r0, r1, r2, r0; + * In this example, a new register will be used for the destination of the + * second MAD. + * + * The purpose of this dealiasing is to make the resulting code more SSA-like + * and therefore make it easier to move instructions around. + * This is of crucial importance for R300 fragment programs, where de-aliasing + * can help to reduce texture indirections, but other targets can benefit from + * it as well. + * + * \note When compiling GLSL, there may be some benefit gained from breaking + * up vectors whose components are unrelated. This is not done yet and should + * be investigated at some point (of course, a matching pass to re-merge + * components would be required). + */ +void rc_dataflow_dealias(struct radeon_compiler * c) +{ + struct dealias_state s; + + memset(&s, 0, sizeof(s)); + s.C = c; + + struct rc_instruction * inst; + for(inst = c->Program.Instructions.Prev; inst != &c->Program.Instructions; inst = inst->Prev) { + if (!inst->Dataflow.DstRegAliased || inst->Dataflow.DstReg->File != RC_FILE_TEMPORARY) + continue; + + if (inst->Dataflow.DstReg->UseMask & ~inst->I.DstReg.WriteMask) + continue; + + s.OldIndex = inst->I.DstReg.Index; + s.NewIndex = rc_find_free_temporary(c); + s.DealiasFail = 0; + s.Length = 0; + + inst->Dataflow.DstRegAliased = 0; + if (inst->Dataflow.DstRegPrev) { + rc_dataflow_remove_ref(inst->Dataflow.DstRegPrev); + inst->Dataflow.DstRegPrev = 0; + } + + push_dealias_vector(&s, inst->Dataflow.DstReg); + run_dealias(&s); + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_swizzles.c b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_swizzles.c new file mode 100644 index 00000000000..1aa91eff7c1 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_dataflow_swizzles.c @@ -0,0 +1,126 @@ +/* + * 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 "radeon_dataflow.h" + +#include "radeon_compiler.h" +#include "radeon_swizzle.h" + + +static void rewrite_source(struct radeon_compiler * c, + struct rc_instruction * inst, unsigned src) +{ + struct rc_swizzle_split split; + unsigned int tempreg = rc_find_free_temporary(c); + unsigned int usemask; + struct rc_dataflow_ref * oldref = inst->Dataflow.SrcReg[src]; + struct rc_dataflow_vector * vector = 0; + + usemask = 0; + for(unsigned int chan = 0; chan < 4; ++chan) { + if (GET_SWZ(inst->I.SrcReg[src].Swizzle, chan) != RC_SWIZZLE_UNUSED) + usemask |= 1 << chan; + } + + c->SwizzleCaps->Split(inst->I.SrcReg[src], usemask, &split); + + for(unsigned int phase = 0; phase < split.NumPhases; ++phase) { + struct rc_instruction * mov = rc_insert_new_instruction(c, inst->Prev); + unsigned int phase_refmask; + unsigned int masked_negate; + + mov->I.Opcode = RC_OPCODE_MOV; + mov->I.DstReg.File = RC_FILE_TEMPORARY; + mov->I.DstReg.Index = tempreg; + mov->I.DstReg.WriteMask = split.Phase[phase]; + mov->I.SrcReg[0] = inst->I.SrcReg[src]; + + phase_refmask = 0; + for(unsigned int chan = 0; chan < 4; ++chan) { + if (!GET_BIT(split.Phase[phase], chan)) + SET_SWZ(mov->I.SrcReg[0].Swizzle, chan, RC_SWIZZLE_UNUSED); + else + phase_refmask |= 1 << GET_SWZ(mov->I.SrcReg[0].Swizzle, chan); + } + + phase_refmask &= RC_MASK_XYZW; + + masked_negate = split.Phase[phase] & mov->I.SrcReg[0].Negate; + if (masked_negate == 0) + mov->I.SrcReg[0].Negate = 0; + else if (masked_negate == split.Phase[phase]) + mov->I.SrcReg[0].Negate = RC_MASK_XYZW; + + if (oldref) { + mov->Dataflow.SrcReg[0] = rc_dataflow_create_ref(c, oldref->Vector, mov); + mov->Dataflow.SrcReg[0]->UseMask = phase_refmask; + } + + mov->Dataflow.DstReg = rc_dataflow_create_vector(c, RC_FILE_TEMPORARY, tempreg, mov); + mov->Dataflow.DstReg->ValidMask = split.Phase[phase]; + + if (vector) { + mov->Dataflow.DstRegPrev = rc_dataflow_create_ref(c, vector, mov); + mov->Dataflow.DstRegPrev->UseMask = vector->ValidMask; + mov->Dataflow.DstReg->ValidMask |= vector->ValidMask; + mov->Dataflow.DstRegAliased = 1; + } + + mov->Dataflow.DstReg->UseMask = mov->Dataflow.DstReg->ValidMask; + vector = mov->Dataflow.DstReg; + } + + if (oldref) + rc_dataflow_remove_ref(oldref); + inst->Dataflow.SrcReg[src] = rc_dataflow_create_ref(c, vector, inst); + inst->Dataflow.SrcReg[src]->UseMask = usemask; + + inst->I.SrcReg[src].File = RC_FILE_TEMPORARY; + inst->I.SrcReg[src].Index = tempreg; + inst->I.SrcReg[src].Swizzle = 0; + inst->I.SrcReg[src].Negate = RC_MASK_NONE; + inst->I.SrcReg[src].Abs = 0; + for(unsigned int chan = 0; chan < 4; ++chan) { + SET_SWZ(inst->I.SrcReg[src].Swizzle, chan, + GET_BIT(usemask, chan) ? chan : RC_SWIZZLE_UNUSED); + } +} + +void rc_dataflow_swizzles(struct radeon_compiler * c) +{ + struct rc_instruction * inst; + + for(inst = c->Program.Instructions.Next; inst != &c->Program.Instructions; inst = inst->Next) { + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + unsigned int src; + + for(src = 0; src < opcode->NumSrcRegs; ++src) { + if (!c->SwizzleCaps->IsNative(inst->I.Opcode, inst->I.SrcReg[src])) + rewrite_source(c, inst, src); + } + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c b/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c deleted file mode 100644 index aaaa50ad1f9..00000000000 --- a/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.c +++ /dev/null @@ -1,294 +0,0 @@ -/* - * 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 deleted file mode 100644 index b3fc77a35a6..00000000000 --- a/src/mesa/drivers/dri/r300/compiler/radeon_nqssadce.h +++ /dev/null @@ -1,91 +0,0 @@ -/* - * 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_opcodes.c b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c new file mode 100644 index 00000000000..b7200990c22 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.c @@ -0,0 +1,347 @@ +/* + * 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 "radeon_opcodes.h" + +struct rc_opcode_info rc_opcodes[MAX_RC_OPCODE] = { + { + .Opcode = RC_OPCODE_NOP, + .Name = "NOP" + }, + { + .Opcode = RC_OPCODE_ILLEGAL_OPCODE, + .Name = "ILLEGAL OPCODE" + }, + { + .Opcode = RC_OPCODE_ABS, + .Name = "ABS", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_ADD, + .Name = "ADD", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_ARL, + .Name = "ARL", + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_CMP, + .Name = "CMP", + .NumSrcRegs = 3, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_COS, + .Name = "COS", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_DDX, + .Name = "DDX", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_DDY, + .Name = "DDY", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_DP3, + .Name = "DP3", + .NumSrcRegs = 2, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_DP4, + .Name = "DP4", + .NumSrcRegs = 2, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_DPH, + .Name = "DPH", + .NumSrcRegs = 2, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_DST, + .Name = "DST", + .NumSrcRegs = 2, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_EX2, + .Name = "EX2", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_EXP, + .Name = "EXP", + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_FLR, + .Name = "FLR", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_FRC, + .Name = "FRC", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_KIL, + .Name = "KIL", + .NumSrcRegs = 1 + }, + { + .Opcode = RC_OPCODE_LG2, + .Name = "LG2", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_LIT, + .Name = "LIT", + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_LOG, + .Name = "LOG", + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_LRP, + .Name = "LRP", + .NumSrcRegs = 3, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_MAD, + .Name = "MAD", + .NumSrcRegs = 3, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_MAX, + .Name = "MAX", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_MIN, + .Name = "MIN", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_MOV, + .Name = "MOV", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_MUL, + .Name = "MUL", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_POW, + .Name = "POW", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_RCP, + .Name = "RCP", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_RSQ, + .Name = "RSQ", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_SCS, + .Name = "SCS", + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_SEQ, + .Name = "SEQ", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SFL, + .Name = "SFL", + .NumSrcRegs = 0, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SGE, + .Name = "SGE", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SGT, + .Name = "SGT", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SIN, + .Name = "SIN", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsStandardScalar = 1 + }, + { + .Opcode = RC_OPCODE_SLE, + .Name = "SLE", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SLT, + .Name = "SLT", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SNE, + .Name = "SNE", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SUB, + .Name = "SUB", + .NumSrcRegs = 2, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_SWZ, + .Name = "SWZ", + .NumSrcRegs = 1, + .HasDstReg = 1, + .IsComponentwise = 1 + }, + { + .Opcode = RC_OPCODE_XPD, + .Name = "XPD", + .NumSrcRegs = 2, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_TEX, + .Name = "TEX", + .HasTexture = 1, + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_TXB, + .Name = "TXB", + .HasTexture = 1, + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_TXD, + .Name = "TXD", + .HasTexture = 1, + .NumSrcRegs = 3, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_TXL, + .Name = "TXL", + .HasTexture = 1, + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_TXP, + .Name = "TXP", + .HasTexture = 1, + .NumSrcRegs = 1, + .HasDstReg = 1 + }, + { + .Opcode = RC_OPCODE_REPL_ALPHA, + .Name = "REPL_ALPHA", + .HasDstReg = 1 + } +}; diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h new file mode 100644 index 00000000000..8e30bef1e3e --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_opcodes.h @@ -0,0 +1,210 @@ +/* + * 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. + * + */ + +#ifndef RADEON_OPCODES_H +#define RADEON_OPCODES_H + +#include <assert.h> + +/** + * Opcodes understood by the Radeon compiler. + */ +typedef enum { + RC_OPCODE_NOP = 0, + RC_OPCODE_ILLEGAL_OPCODE, + + /** vec4 instruction: dst.c = abs(src0.c); */ + RC_OPCODE_ABS, + + /** vec4 instruction: dst.c = src0.c + src1.c; */ + RC_OPCODE_ADD, + + /** special instruction: load address register + * dst.x = floor(src.x), where dst must be an address register */ + RC_OPCODE_ARL, + + /** vec4 instruction: dst.c = src0.c < 0.0 ? src1.c : src2.c */ + RC_OPCODE_CMP, + + /** scalar instruction: dst = cos(src0.x) */ + RC_OPCODE_COS, + + /** special instruction: take vec4 partial derivative in X direction + * dst.c = d src0.c / dx */ + RC_OPCODE_DDX, + + /** special instruction: take vec4 partial derivative in Y direction + * dst.c = d src0.c / dy */ + RC_OPCODE_DDY, + + /** scalar instruction: dst = src0.x*src1.x + src0.y*src1.y + src0.z*src1.z */ + RC_OPCODE_DP3, + + /** scalar instruction: dst = src0.x*src1.x + src0.y*src1.y + src0.z*src1.z + src0.w*src1.w */ + RC_OPCODE_DP4, + + /** scalar instruction: dst = src0.x*src1.x + src0.y*src1.y + src0.z*src1.z + src1.w */ + RC_OPCODE_DPH, + + /** special instruction, see ARB_fragment_program */ + RC_OPCODE_DST, + + /** scalar instruction: dst = 2**src0.x */ + RC_OPCODE_EX2, + + /** special instruction, see ARB_vertex_program */ + RC_OPCODE_EXP, + + /** vec4 instruction: dst.c = floor(src0.c) */ + RC_OPCODE_FLR, + + /** vec4 instruction: dst.c = src0.c - floor(src0.c) */ + RC_OPCODE_FRC, + + /** special instruction: stop execution if any component of src0 is negative */ + RC_OPCODE_KIL, + + /** scalar instruction: dst = log_2(src0.x) */ + RC_OPCODE_LG2, + + /** special instruction, see ARB_vertex_program */ + RC_OPCODE_LIT, + + /** special instruction, see ARB_vertex_program */ + RC_OPCODE_LOG, + + /** vec4 instruction: dst.c = src0.c*src1.c + (1 - src0.c)*src2.c */ + RC_OPCODE_LRP, + + /** vec4 instruction: dst.c = src0.c*src1.c + src2.c */ + RC_OPCODE_MAD, + + /** vec4 instruction: dst.c = max(src0.c, src1.c) */ + RC_OPCODE_MAX, + + /** vec4 instruction: dst.c = min(src0.c, src1.c) */ + RC_OPCODE_MIN, + + /** vec4 instruction: dst.c = src0.c */ + RC_OPCODE_MOV, + + /** vec4 instruction: dst.c = src0.c*src1.c */ + RC_OPCODE_MUL, + + /** scalar instruction: dst = src0.x ** src1.x */ + RC_OPCODE_POW, + + /** scalar instruction: dst = 1 / src0.x */ + RC_OPCODE_RCP, + + /** scalar instruction: dst = 1 / sqrt(src0.x) */ + RC_OPCODE_RSQ, + + /** special instruction, see ARB_fragment_program */ + RC_OPCODE_SCS, + + /** vec4 instruction: dst.c = (src0.c == src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SEQ, + + /** vec4 instruction: dst.c = 0.0 */ + RC_OPCODE_SFL, + + /** vec4 instruction: dst.c = (src0.c >= src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SGE, + + /** vec4 instruction: dst.c = (src0.c > src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SGT, + + /** scalar instruction: dst = sin(src0.x) */ + RC_OPCODE_SIN, + + /** vec4 instruction: dst.c = (src0.c <= src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SLE, + + /** vec4 instruction: dst.c = (src0.c < src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SLT, + + /** vec4 instruction: dst.c = (src0.c != src1.c) ? 1.0 : 0.0 */ + RC_OPCODE_SNE, + + /** vec4 instruction: dst.c = src0.c - src1.c */ + RC_OPCODE_SUB, + + /** vec4 instruction: dst.c = src0.c */ + RC_OPCODE_SWZ, + + /** special instruction, see ARB_fragment_program */ + RC_OPCODE_XPD, + + RC_OPCODE_TEX, + RC_OPCODE_TXB, + RC_OPCODE_TXD, + RC_OPCODE_TXL, + RC_OPCODE_TXP, + + /** special instruction, used in R300-R500 fragment program pair instructions + * indicates that the result of the alpha operation shall be replicated + * across all other channels */ + RC_OPCODE_REPL_ALPHA, + + MAX_RC_OPCODE +} rc_opcode; + + +struct rc_opcode_info { + rc_opcode Opcode; + const char * Name; + + /** true if the instruction reads from a texture. + * + * \note This is false for the KIL instruction, even though KIL is + * a texture instruction from a hardware point of view. */ + unsigned int HasTexture:1; + + unsigned int NumSrcRegs:2; + unsigned int HasDstReg:1; + + /** true if this is a vector instruction that operates on components in parallel + * without any cross-component interaction */ + unsigned int IsComponentwise:1; + + /** true if this instruction sources only its operands X components + * to compute one result which is smeared across all output channels */ + unsigned int IsStandardScalar:1; +}; + +extern struct rc_opcode_info rc_opcodes[MAX_RC_OPCODE]; + +static inline const struct rc_opcode_info * rc_get_opcode_info(rc_opcode opcode) +{ + assert((unsigned int)opcode < MAX_RC_OPCODE); + assert(rc_opcodes[opcode].Opcode == opcode); + + return &rc_opcodes[opcode]; +} + +#endif /* RADEON_OPCODES_H */ diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program.c b/src/mesa/drivers/dri/r300/compiler/radeon_program.c index b636f90a968..b97c48084b4 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.c @@ -27,9 +27,9 @@ #include "radeon_program.h" +#include <stdio.h> + #include "radeon_compiler.h" -#include "shader/prog_parameter.h" -#include "shader/prog_print.h" /** @@ -69,37 +69,57 @@ void radeonLocalTransform( } } +/** + * Left multiplication of a register with a swizzle + */ +struct rc_src_register lmul_swizzle(unsigned int swizzle, struct rc_src_register srcreg) +{ + struct rc_src_register tmp = srcreg; + int i; + tmp.Swizzle = 0; + tmp.Negate = 0; + for(i = 0; i < 4; ++i) { + rc_swizzle 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; +} -GLint rc_find_free_temporary(struct radeon_compiler * c) +unsigned int rc_find_free_temporary(struct radeon_compiler * c) { - GLboolean used[MAX_PROGRAM_TEMPS]; - GLuint i; + char used[RC_REGISTER_MAX_INDEX]; + unsigned int 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 nsrc = _mesa_num_inst_src_regs(inst->Opcode); - const GLuint ndst = _mesa_num_inst_dst_regs(inst->Opcode); - GLuint k; - - for (k = 0; k < nsrc; k++) { - if (inst->SrcReg[k].File == PROGRAM_TEMPORARY) - used[inst->SrcReg[k].Index] = GL_TRUE; + const struct rc_sub_instruction *inst = &rcinst->I; + const struct rc_opcode_info *opcode = rc_get_opcode_info(inst->Opcode); + unsigned int k; + + for (k = 0; k < opcode->NumSrcRegs; k++) { + if (inst->SrcReg[k].File == RC_FILE_TEMPORARY) + used[inst->SrcReg[k].Index] = 1; } - if (ndst) { - if (inst->DstReg.File == PROGRAM_TEMPORARY) - used[inst->DstReg.Index] = GL_TRUE; + if (opcode->HasDstReg) { + if (inst->DstReg.File == RC_FILE_TEMPORARY) + used[inst->DstReg.Index] = 1; } } - for (i = 0; i < MAX_PROGRAM_TEMPS; i++) { + for (i = 0; i < RC_REGISTER_MAX_INDEX; i++) { if (!used[i]) return i; } - return -1; + rc_error(c, "Ran out of temporary registers\n"); + return 0; } @@ -107,10 +127,13 @@ 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; + memset(inst, 0, sizeof(struct rc_instruction)); - _mesa_init_instructions(&inst->I, 1); + inst->I.Opcode = RC_OPCODE_ILLEGAL_OPCODE; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_XYZW; + inst->I.SrcReg[1].Swizzle = RC_SWIZZLE_XYZW; + inst->I.SrcReg[2].Swizzle = RC_SWIZZLE_XYZW; return inst; } @@ -131,80 +154,7 @@ struct rc_instruction *rc_insert_new_instruction(struct radeon_compiler * c, str void rc_remove_instruction(struct rc_instruction * inst) { + rc_dataflow_remove_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; - - int isNVProgram = 0; - - if (program->Target == GL_VERTEX_PROGRAM_ARB) { - struct gl_vertex_program * vp = (struct gl_vertex_program *) program; - isNVProgram = vp->IsNVProgram; - } - - if (isNVProgram) { - /* NV_vertex_program has a fixed-sized constant environment. - * This could be handled more efficiently for programs that - * do not use relative addressing. - */ - for(i = 0; i < 96; ++i) { - struct rc_constant constant; - - constant.Type = RC_CONSTANT_EXTERNAL; - constant.Size = 4; - constant.u.External = i; - - rc_constants_add(&c->Program.Constants, &constant); - } - } else { - 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 index 561958608ca..d38c9a420c6 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program.h @@ -28,37 +28,129 @@ #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" +#include <stdint.h> +#include <string.h> + +#include "radeon_opcodes.h" +#include "radeon_code.h" +#include "radeon_program_constants.h" +#include "radeon_dataflow.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 */ +struct rc_src_register { + rc_register_file File:3; + + /** Negative values may be used for relative addressing. */ + signed int Index:(RC_REGISTER_INDEX_BITS+1); + unsigned int RelAddr:1; + + unsigned int Swizzle:12; + + /** Take the component-wise absolute value */ + unsigned int Abs:1; + + /** Post-Abs negation. */ + unsigned int Negate:4; +}; + +struct rc_dst_register { + rc_register_file File:3; + + /** Negative values may be used for relative addressing. */ + signed int Index:(RC_REGISTER_INDEX_BITS+1); + unsigned int RelAddr:1; + + unsigned int WriteMask:4; +}; + +/** + * Instructions are maintained by the compiler in a doubly linked list + * of these structures. + * + * This instruction format is intended to be expanded for hardware-specific + * trickery. At different stages of compilation, a different set of + * instruction types may be valid. + */ +struct rc_sub_instruction { + struct rc_src_register SrcReg[3]; + struct rc_dst_register DstReg; + + /** + * Opcode of this instruction, according to \ref rc_opcode enums. + */ + rc_opcode Opcode:8; + + /** + * Saturate each value of the result to the range [0,1] or [-1,1], + * according to \ref rc_saturate_mode enums. + */ + rc_saturate_mode SaturateMode:2; + + /** + * \name Extra fields for TEX, TXB, TXD, TXL, TXP instructions. + */ + /*@{*/ + /** Source texture unit. */ + unsigned int TexSrcUnit:5; + + /** Source texture target, one of the \ref rc_texture_target enums */ + rc_texture_target TexSrcTarget:3; + + /** True if tex instruction should do shadow comparison */ + unsigned int TexShadow:1; + /*@}*/ +}; + +struct rc_instruction { + struct rc_instruction * Prev; + struct rc_instruction * Next; + + struct rc_sub_instruction I; + + /** + * Dataflow annotations. + * + * These are not supplied by the caller of the compiler, + * but filled in during compilation stages that make use of + * dataflow analysis. + */ + struct rc_instruction_dataflow Dataflow; +}; + +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; }; enum { - OPCODE_REPL_ALPHA = MAX_OPCODE /**< used in paired instructions */ + OPCODE_REPL_ALPHA = MAX_RC_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) +static inline rc_swizzle get_swz(unsigned int swz, rc_swizzle 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) +static inline unsigned int combine_swizzles4(unsigned int src, + rc_swizzle swz_x, rc_swizzle swz_y, rc_swizzle swz_z, rc_swizzle swz_w) { - GLuint ret = 0; + unsigned int ret = 0; ret |= get_swz(src, swz_x); ret |= get_swz(src, swz_y) << 3; @@ -68,22 +160,24 @@ static inline GLuint combine_swizzles4(GLuint src, GLuint swz_x, GLuint swz_y, G return ret; } -static inline GLuint combine_swizzles(GLuint src, GLuint swz) +static inline unsigned int combine_swizzles(unsigned int src, unsigned int swz) { - GLuint ret = 0; + unsigned int 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; + ret |= get_swz(src, GET_SWZ(swz, RC_SWIZZLE_X)); + ret |= get_swz(src, GET_SWZ(swz, RC_SWIZZLE_Y)) << 3; + ret |= get_swz(src, GET_SWZ(swz, RC_SWIZZLE_Z)) << 6; + ret |= get_swz(src, GET_SWZ(swz, RC_SWIZZLE_W)) << 9; return ret; } -static INLINE void reset_srcreg(struct prog_src_register* reg) +struct rc_src_register lmul_swizzle(unsigned int swizzle, struct rc_src_register srcreg); + +static inline void reset_srcreg(struct rc_src_register* reg) { - _mesa_bzero(reg, sizeof(*reg)); - reg->Swizzle = SWIZZLE_NOOP; + memset(reg, 0, sizeof(reg)); + reg->Swizzle = RC_SWIZZLE_XYZW; } @@ -92,13 +186,13 @@ static INLINE void reset_srcreg(struct prog_src_register* reg) * * 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 + * and return true, or return false if it doesn't understand the * instruction. * * The function gets passed the userData as last parameter. */ struct radeon_program_transformation { - GLboolean (*function)( + int (*function)( struct radeon_compiler*, struct rc_instruction*, void*); @@ -110,12 +204,16 @@ void radeonLocalTransform( int num_transformations, struct radeon_program_transformation* transformations); -GLint rc_find_free_temporary(struct radeon_compiler * c); +unsigned int 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); +enum { + RC_PRINT_DATAFLOW = 0x1 +}; + +void rc_print_program(const struct rc_program *prog, unsigned int flags); #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 index f23ce301cab..e25bc4ec87b 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.c @@ -40,8 +40,8 @@ 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) + rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg, + struct rc_src_register SrcReg) { struct rc_instruction *fpi = rc_insert_new_instruction(c, after); @@ -54,8 +54,8 @@ static struct rc_instruction *emit1( 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) + rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg, + struct rc_src_register SrcReg0, struct rc_src_register SrcReg1) { struct rc_instruction *fpi = rc_insert_new_instruction(c, after); @@ -69,9 +69,9 @@ static struct rc_instruction *emit2( 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) + rc_opcode Opcode, rc_saturate_mode Saturate, struct rc_dst_register DstReg, + struct rc_src_register SrcReg0, struct rc_src_register SrcReg1, + struct rc_src_register SrcReg2) { struct rc_instruction *fpi = rc_insert_new_instruction(c, after); @@ -84,131 +84,120 @@ static struct rc_instruction *emit3( return fpi; } -static struct prog_dst_register dstreg(int file, int index) +static struct rc_dst_register dstreg(int file, int index) { - struct prog_dst_register dst; + struct rc_dst_register dst; dst.File = file; dst.Index = index; - dst.WriteMask = WRITEMASK_XYZW; - dst.CondMask = COND_TR; + dst.WriteMask = RC_MASK_XYZW; dst.RelAddr = 0; - dst.CondSwizzle = SWIZZLE_NOOP; - dst.CondSrc = 0; - dst.pad = 0; return dst; } -static struct prog_dst_register dstregtmpmask(int index, int mask) +static struct rc_dst_register dstregtmpmask(int index, int mask) { - struct prog_dst_register dst = {0}; - dst.File = PROGRAM_TEMPORARY; + struct rc_dst_register dst = {0}; + dst.File = RC_FILE_TEMPORARY; dst.Index = index; dst.WriteMask = mask; dst.RelAddr = 0; - 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, +static const struct rc_src_register builtin_zero = { + .File = RC_FILE_NONE, .Index = 0, - .Swizzle = SWIZZLE_0000 + .Swizzle = RC_SWIZZLE_0000 }; -static const struct prog_src_register builtin_one = { - .File = PROGRAM_BUILTIN, +static const struct rc_src_register builtin_one = { + .File = RC_FILE_NONE, .Index = 0, - .Swizzle = SWIZZLE_1111 + .Swizzle = RC_SWIZZLE_1111 }; -static const struct prog_src_register srcreg_undefined = { - .File = PROGRAM_UNDEFINED, +static const struct rc_src_register srcreg_undefined = { + .File = RC_FILE_NONE, .Index = 0, - .Swizzle = SWIZZLE_NOOP + .Swizzle = RC_SWIZZLE_XYZW }; -static struct prog_src_register srcreg(int file, int index) +static struct rc_src_register srcreg(int file, int index) { - struct prog_src_register src = srcreg_undefined; + struct rc_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) +static struct rc_src_register srcregswz(int file, int index, int swz) { - struct prog_src_register src = srcreg_undefined; + struct rc_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) +static struct rc_src_register absolute(struct rc_src_register reg) { - struct prog_src_register newreg = reg; + struct rc_src_register newreg = reg; newreg.Abs = 1; - newreg.Negate = NEGATE_NONE; + newreg.Negate = RC_MASK_NONE; return newreg; } -static struct prog_src_register negate(struct prog_src_register reg) +static struct rc_src_register negate(struct rc_src_register reg) { - struct prog_src_register newreg = reg; - newreg.Negate = newreg.Negate ^ NEGATE_XYZW; + struct rc_src_register newreg = reg; + newreg.Negate = newreg.Negate ^ RC_MASK_XYZW; return newreg; } -static struct prog_src_register swizzle(struct prog_src_register reg, GLuint x, GLuint y, GLuint z, GLuint w) +static struct rc_src_register swizzle(struct rc_src_register reg, + rc_swizzle x, rc_swizzle y, rc_swizzle z, rc_swizzle 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)); + struct rc_src_register swizzled = reg; + swizzled.Swizzle = combine_swizzles4(reg.Swizzle, x, y, z, w); return swizzled; } -static struct prog_src_register scalar(struct prog_src_register reg) +static struct rc_src_register scalar(struct rc_src_register reg) { - return swizzle(reg, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X); + return swizzle(reg, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X); } static void transform_ABS(struct radeon_compiler* c, struct rc_instruction* inst) { - struct prog_src_register src = inst->I.SrcReg[0]; + struct rc_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); + src.Negate = RC_MASK_NONE; + emit1(c, inst->Prev, RC_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; + struct rc_src_register src0 = inst->I.SrcReg[0]; + struct rc_src_register src1 = inst->I.SrcReg[1]; + src0.Negate &= ~RC_MASK_W; src0.Swizzle &= ~(7 << (3 * 3)); - src0.Swizzle |= SWIZZLE_ZERO << (3 * 3); - src1.Negate &= ~NEGATE_W; + src0.Swizzle |= RC_SWIZZLE_ZERO << (3 * 3); + src1.Negate &= ~RC_MASK_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); + src1.Swizzle |= RC_SWIZZLE_ZERO << (3 * 3); + emit2(c, inst->Prev, RC_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; + struct rc_src_register src0 = inst->I.SrcReg[0]; + src0.Negate &= ~RC_MASK_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]); + src0.Swizzle |= RC_SWIZZLE_ONE << (3 * 3); + emit2(c, inst->Prev, RC_OPCODE_DP4, inst->I.SaturateMode, inst->I.DstReg, src0, inst->I.SrcReg[1]); rc_remove_instruction(inst); } @@ -219,9 +208,9 @@ static void transform_DPH(struct radeon_compiler* c, 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)); + emit2(c, inst->Prev, RC_OPCODE_MUL, inst->I.SaturateMode, inst->I.DstReg, + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_ONE), + swizzle(inst->I.SrcReg[1], RC_SWIZZLE_ONE, RC_SWIZZLE_Y, RC_SWIZZLE_ONE, RC_SWIZZLE_W)); rc_remove_instruction(inst); } @@ -229,9 +218,9 @@ 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))); + emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstreg(RC_FILE_TEMPORARY, tempreg), inst->I.SrcReg[0]); + emit2(c, inst->Prev, RC_OPCODE_ADD, inst->I.SaturateMode, inst->I.DstReg, + inst->I.SrcReg[0], negate(srcreg(RC_FILE_TEMPORARY, tempreg))); rc_remove_instruction(inst); } @@ -256,64 +245,64 @@ static void transform_FLR(struct radeon_compiler* c, static void transform_LIT(struct radeon_compiler* c, struct rc_instruction* inst) { - GLuint constant; - GLuint constant_swizzle; - GLuint temp; - struct prog_src_register srctemp; + unsigned int constant; + unsigned int constant_swizzle; + unsigned int temp; + struct rc_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) { + if (inst->I.DstReg.WriteMask != RC_MASK_XYZW || inst->I.DstReg.File != RC_FILE_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))); + RC_OPCODE_MOV, 0, inst->I.DstReg, + srcreg(RC_FILE_TEMPORARY, rc_find_free_temporary(c))); - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = inst_mov->I.SrcReg[0].Index; - inst->I.DstReg.WriteMask = WRITEMASK_XYZW; + inst->I.DstReg.WriteMask = RC_MASK_XYZW; } temp = inst->I.DstReg.Index; - srctemp = srcreg(PROGRAM_TEMPORARY, temp); + srctemp = srcreg(RC_FILE_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), + emit2(c, inst->Prev, RC_OPCODE_MAX, 0, + dstregtmpmask(temp, RC_MASK_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))); + swizzle(srcreg(RC_FILE_CONSTANT, constant), + RC_SWIZZLE_ZERO, RC_SWIZZLE_ZERO, RC_SWIZZLE_ZERO, constant_swizzle&3)); + emit2(c, inst->Prev, RC_OPCODE_MIN, 0, + dstregtmpmask(temp, RC_MASK_Z), + swizzle(srctemp, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + negate(srcregswz(RC_FILE_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)); + emit1(c, inst->Prev, RC_OPCODE_LG2, 0, + dstregtmpmask(temp, RC_MASK_W), + swizzle(srctemp, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y)); + emit2(c, inst->Prev, RC_OPCODE_MUL, 0, + dstregtmpmask(temp, RC_MASK_W), + swizzle(srctemp, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + swizzle(srctemp, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z)); + emit1(c, inst->Prev, RC_OPCODE_EX2, 0, + dstregtmpmask(temp, RC_MASK_W), + swizzle(srctemp, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_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), + emit3(c, inst->Prev, RC_OPCODE_CMP, inst->I.SaturateMode, + dstregtmpmask(temp, RC_MASK_Z), + negate(swizzle(srctemp, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)), + swizzle(srctemp, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_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)); + emit1(c, inst->Prev, RC_OPCODE_MOV, inst->I.SaturateMode, + dstregtmpmask(temp, RC_MASK_XYW), + swizzle(srctemp, RC_SWIZZLE_ONE, RC_SWIZZLE_X, RC_SWIZZLE_ONE, RC_SWIZZLE_ONE)); rc_remove_instruction(inst); } @@ -323,12 +312,12 @@ static void transform_LRP(struct radeon_compiler* c, { int tempreg = rc_find_free_temporary(c); - emit2(c, inst->Prev, OPCODE_ADD, 0, - dstreg(PROGRAM_TEMPORARY, tempreg), + emit2(c, inst->Prev, RC_OPCODE_ADD, 0, + dstreg(RC_FILE_TEMPORARY, tempreg), inst->I.SrcReg[1], negate(inst->I.SrcReg[2])); - emit3(c, inst->Prev, OPCODE_MAD, inst->I.SaturateMode, + emit3(c, inst->Prev, RC_OPCODE_MAD, inst->I.SaturateMode, inst->I.DstReg, - inst->I.SrcReg[0], srcreg(PROGRAM_TEMPORARY, tempreg), inst->I.SrcReg[2]); + inst->I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, tempreg), inst->I.SrcReg[2]); rc_remove_instruction(inst); } @@ -337,14 +326,14 @@ 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; + struct rc_dst_register tempdst = dstreg(RC_FILE_TEMPORARY, tempreg); + struct rc_src_register tempsrc = srcreg(RC_FILE_TEMPORARY, tempreg); + tempdst.WriteMask = RC_MASK_W; + tempsrc.Swizzle = RC_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); + emit1(c, inst->Prev, RC_OPCODE_LG2, 0, tempdst, scalar(inst->I.SrcReg[0])); + emit2(c, inst->Prev, RC_OPCODE_MUL, 0, tempdst, tempsrc, scalar(inst->I.SrcReg[1])); + emit1(c, inst->Prev, RC_OPCODE_EX2, inst->I.SaturateMode, inst->I.DstReg, tempsrc); rc_remove_instruction(inst); } @@ -360,9 +349,9 @@ static void transform_SGE(struct radeon_compiler* c, { 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); + emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dstreg(RC_FILE_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1])); + emit3(c, inst->Prev, RC_OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg, + srcreg(RC_FILE_TEMPORARY, tempreg), builtin_zero, builtin_one); rc_remove_instruction(inst); } @@ -372,9 +361,9 @@ static void transform_SLT(struct radeon_compiler* c, { 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); + emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dstreg(RC_FILE_TEMPORARY, tempreg), inst->I.SrcReg[0], negate(inst->I.SrcReg[1])); + emit3(c, inst->Prev, RC_OPCODE_CMP, inst->I.SaturateMode, inst->I.DstReg, + srcreg(RC_FILE_TEMPORARY, tempreg), builtin_one, builtin_zero); rc_remove_instruction(inst); } @@ -382,14 +371,14 @@ static void transform_SLT(struct radeon_compiler* c, static void transform_SUB(struct radeon_compiler* c, struct rc_instruction* inst) { - inst->I.Opcode = OPCODE_ADD; + inst->I.Opcode = RC_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; + inst->I.Opcode = RC_OPCODE_MOV; } static void transform_XPD(struct radeon_compiler* c, @@ -397,13 +386,13 @@ static void transform_XPD(struct radeon_compiler* c, { 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))); + emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dstreg(RC_FILE_TEMPORARY, tempreg), + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W), + swizzle(inst->I.SrcReg[1], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W)); + emit3(c, inst->Prev, RC_OPCODE_MAD, inst->I.SaturateMode, inst->I.DstReg, + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_W), + swizzle(inst->I.SrcReg[1], RC_SWIZZLE_Z, RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_W), + negate(srcreg(RC_FILE_TEMPORARY, tempreg))); rc_remove_instruction(inst); } @@ -423,27 +412,27 @@ static void transform_XPD(struct radeon_compiler* c, * * @note should be applicable to R300 and R500 fragment programs. */ -GLboolean radeonTransformALU( +int 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; + case RC_OPCODE_ABS: transform_ABS(c, inst); return 1; + case RC_OPCODE_DPH: transform_DPH(c, inst); return 1; + case RC_OPCODE_DST: transform_DST(c, inst); return 1; + case RC_OPCODE_FLR: transform_FLR(c, inst); return 1; + case RC_OPCODE_LIT: transform_LIT(c, inst); return 1; + case RC_OPCODE_LRP: transform_LRP(c, inst); return 1; + case RC_OPCODE_POW: transform_POW(c, inst); return 1; + case RC_OPCODE_RSQ: transform_RSQ(c, inst); return 1; + case RC_OPCODE_SGE: transform_SGE(c, inst); return 1; + case RC_OPCODE_SLT: transform_SLT(c, inst); return 1; + case RC_OPCODE_SUB: transform_SUB(c, inst); return 1; + case RC_OPCODE_SWZ: transform_SWZ(c, inst); return 1; + case RC_OPCODE_XPD: transform_XPD(c, inst); return 1; default: - return GL_FALSE; + return 0; } } @@ -452,37 +441,37 @@ 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.Opcode = RC_OPCODE_MAX; inst->I.SrcReg[1] = inst->I.SrcReg[0]; - inst->I.SrcReg[1].Negate ^= NEGATE_XYZW; + inst->I.SrcReg[1].Negate ^= RC_MASK_XYZW; } /** * For use with radeonLocalTransform, this transforms non-native ALU * instructions of the r300 up to r500 vertex engine. */ -GLboolean r300_transform_vertex_alu( +int 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; + case RC_OPCODE_ABS: transform_r300_vertex_ABS(c, inst); return 1; + case RC_OPCODE_DP3: transform_DP3(c, inst); return 1; + case RC_OPCODE_DPH: transform_DPH(c, inst); return 1; + case RC_OPCODE_FLR: transform_FLR(c, inst); return 1; + case RC_OPCODE_LRP: transform_LRP(c, inst); return 1; + case RC_OPCODE_SUB: transform_SUB(c, inst); return 1; + case RC_OPCODE_SWZ: transform_SWZ(c, inst); return 1; + case RC_OPCODE_XPD: transform_XPD(c, inst); return 1; default: - return GL_FALSE; + return 0; } } -static void sincos_constants(struct radeon_compiler* c, GLuint *constants) +static void sincos_constants(struct radeon_compiler* c, unsigned int *constants) { - static const GLfloat SinCosConsts[2][4] = { + static const float SinCosConsts[2][4] = { { 1.273239545, // 4/PI -0.405284735, // -4/(PI*PI) @@ -512,25 +501,25 @@ static void sincos_constants(struct radeon_compiler* c, GLuint *constants) */ static void sin_approx( struct radeon_compiler* c, struct rc_instruction * before, - struct prog_dst_register dst, struct prog_src_register src, const GLuint* constants) -{ - GLuint tempreg = rc_find_free_temporary(c); - - emit2(c, before->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, before->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, before->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, before->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)); + struct rc_dst_register dst, struct rc_src_register src, const unsigned int* constants) +{ + unsigned int tempreg = rc_find_free_temporary(c); + + emit2(c, before, RC_OPCODE_MUL, 0, dstregtmpmask(tempreg, RC_MASK_XY), + swizzle(src, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + srcreg(RC_FILE_CONSTANT, constants[0])); + emit3(c, before, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_X), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y), + absolute(swizzle(src, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)); + emit3(c, before, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_Y), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + absolute(swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)), + negate(swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X))); + emit3(c, before, RC_OPCODE_MAD, 0, dst, + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y), + swizzle(srcreg(RC_FILE_CONSTANT, constants[0]), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)); } /** @@ -538,81 +527,81 @@ static void sin_approx( * using only the basic instructions * MOV, ADD, MUL, MAD, FRC */ -GLboolean radeonTransformTrigSimple(struct radeon_compiler* c, +int 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; + if (inst->I.Opcode != RC_OPCODE_COS && + inst->I.Opcode != RC_OPCODE_SIN && + inst->I.Opcode != RC_OPCODE_SCS) + return 0; - GLuint constants[2]; - GLuint tempreg = rc_find_free_temporary(c); + unsigned int constants[2]; + unsigned int tempreg = rc_find_free_temporary(c); sincos_constants(c, constants); - if (inst->I.Opcode == OPCODE_COS) { + if (inst->I.Opcode == RC_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))); + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X)); + emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W)); + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + negate(swizzle(srcreg(RC_FILE_CONSTANT, constants[0]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z))); sin_approx(c, inst, inst->I.DstReg, - swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_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))); + } else if (inst->I.Opcode == RC_OPCODE_SIN) { + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y)); + emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W)); + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + negate(swizzle(srcreg(RC_FILE_CONSTANT, constants[0]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z))); sin_approx(c, inst, inst->I.DstReg, - swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_W, SWIZZLE_W, SWIZZLE_W, SWIZZLE_W), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_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; + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_XY), + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_W)); + emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_XY), + srcreg(RC_FILE_TEMPORARY, tempreg)); + emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_XY), + srcreg(RC_FILE_TEMPORARY, tempreg), + swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W, RC_SWIZZLE_W), + negate(swizzle(srcreg(RC_FILE_CONSTANT, constants[0]), RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z, RC_SWIZZLE_Z))); + + struct rc_dst_register dst = inst->I.DstReg; + + dst.WriteMask = inst->I.DstReg.WriteMask & RC_MASK_X; sin_approx(c, inst, dst, - swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), constants); - dst.WriteMask = inst->I.DstReg.WriteMask & WRITEMASK_Y; + dst.WriteMask = inst->I.DstReg.WriteMask & RC_MASK_Y; sin_approx(c, inst, dst, - swizzle(srcreg(PROGRAM_TEMPORARY, tempreg), SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y, SWIZZLE_Y), + swizzle(srcreg(RC_FILE_TEMPORARY, tempreg), RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y, RC_SWIZZLE_Y), constants); } rc_remove_instruction(inst); - return GL_TRUE; + return 1; } @@ -624,53 +613,53 @@ GLboolean radeonTransformTrigSimple(struct radeon_compiler* c, * * @warning This transformation implicitly changes the semantics of SIN and COS! */ -GLboolean radeonTransformTrigScale(struct radeon_compiler* c, +int 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; + if (inst->I.Opcode != RC_OPCODE_COS && + inst->I.Opcode != RC_OPCODE_SIN && + inst->I.Opcode != RC_OPCODE_SCS) + return 0; - static const GLfloat RCP_2PI = 0.15915494309189535; - GLuint temp; - GLuint constant; - GLuint constant_swizzle; + static const float RCP_2PI = 0.15915494309189535; + unsigned int temp; + unsigned int constant; + unsigned int 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)); + emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dstregtmpmask(temp, RC_MASK_W), + swizzle(inst->I.SrcReg[0], RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X, RC_SWIZZLE_X), + srcregswz(RC_FILE_CONSTANT, constant, constant_swizzle)); + emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(temp, RC_MASK_W), + srcreg(RC_FILE_TEMPORARY, temp)); + + if (inst->I.Opcode == RC_OPCODE_COS) { + emit1(c, inst->Prev, RC_OPCODE_COS, inst->I.SaturateMode, inst->I.DstReg, + srcregswz(RC_FILE_TEMPORARY, temp, RC_SWIZZLE_WWWW)); + } else if (inst->I.Opcode == RC_OPCODE_SIN) { + emit1(c, inst->Prev, RC_OPCODE_SIN, inst->I.SaturateMode, + inst->I.DstReg, srcregswz(RC_FILE_TEMPORARY, temp, RC_SWIZZLE_WWWW)); + } else if (inst->I.Opcode == RC_OPCODE_SCS) { + struct rc_dst_register moddst = inst->I.DstReg; + + if (inst->I.DstReg.WriteMask & RC_MASK_X) { + moddst.WriteMask = RC_MASK_X; + emit1(c, inst->Prev, RC_OPCODE_COS, inst->I.SaturateMode, moddst, + srcregswz(RC_FILE_TEMPORARY, temp, RC_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)); + if (inst->I.DstReg.WriteMask & RC_MASK_Y) { + moddst.WriteMask = RC_MASK_Y; + emit1(c, inst->Prev, RC_OPCODE_SIN, inst->I.SaturateMode, moddst, + srcregswz(RC_FILE_TEMPORARY, temp, RC_SWIZZLE_WWWW)); } } rc_remove_instruction(inst); - return GL_TRUE; + return 1; } /** @@ -681,15 +670,15 @@ GLboolean radeonTransformTrigScale(struct radeon_compiler* c, * @warning This explicitly changes the form of DDX and DDY! */ -GLboolean radeonTransformDeriv(struct radeon_compiler* c, +int 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; + if (inst->I.Opcode != RC_OPCODE_DDX && inst->I.Opcode != RC_OPCODE_DDY) + return 0; - inst->I.SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE, SWIZZLE_ONE); - inst->I.SrcReg[1].Negate = NEGATE_XYZW; + inst->I.SrcReg[1].Swizzle = RC_MAKE_SWIZZLE(RC_SWIZZLE_ONE, RC_SWIZZLE_ONE, RC_SWIZZLE_ONE, RC_SWIZZLE_ONE); + inst->I.SrcReg[1].Negate = RC_MASK_XYZW; - return GL_TRUE; + return 1; } diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h index 147efec6fcb..7cb5f84b7f0 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_alu.h @@ -30,27 +30,27 @@ #include "radeon_program.h" -GLboolean radeonTransformALU( +int radeonTransformALU( struct radeon_compiler * c, struct rc_instruction * inst, void*); -GLboolean r300_transform_vertex_alu( +int r300_transform_vertex_alu( struct radeon_compiler * c, struct rc_instruction * inst, void*); -GLboolean radeonTransformTrigSimple( +int radeonTransformTrigSimple( struct radeon_compiler * c, struct rc_instruction * inst, void*); -GLboolean radeonTransformTrigScale( +int radeonTransformTrigScale( struct radeon_compiler * c, struct rc_instruction * inst, void*); -GLboolean radeonTransformDeriv( +int radeonTransformDeriv( struct radeon_compiler * c, struct rc_instruction * inst, void*); diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h b/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h new file mode 100644 index 00000000000..69994f9880f --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_constants.h @@ -0,0 +1,128 @@ +/* + * 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. + * + */ + +#ifndef RADEON_PROGRAM_CONSTANTS_H +#define RADEON_PROGRAM_CONSTANTS_H + +typedef enum { + RC_SATURATE_NONE = 0, + RC_SATURATE_ZERO_ONE, + RC_SATURATE_MINUS_PLUS_ONE +} rc_saturate_mode; + +typedef enum { + RC_TEXTURE_2D_ARRAY, + RC_TEXTURE_1D_ARRAY, + RC_TEXTURE_CUBE, + RC_TEXTURE_3D, + RC_TEXTURE_RECT, + RC_TEXTURE_2D, + RC_TEXTURE_1D +} rc_texture_target; + +typedef enum { + /** + * Used to indicate unused register descriptions and + * source register that use a constant swizzle. + */ + RC_FILE_NONE = 0, + RC_FILE_TEMPORARY, + + /** + * Input register. + * + * \note The compiler attaches no implicit semantics to input registers. + * Fragment/vertex program specific semantics must be defined explicitly + * using the appropriate compiler interfaces. + */ + RC_FILE_INPUT, + + /** + * Output register. + * + * \note The compiler attaches no implicit semantics to input registers. + * Fragment/vertex program specific semantics must be defined explicitly + * using the appropriate compiler interfaces. + */ + RC_FILE_OUTPUT, + RC_FILE_ADDRESS, + + /** + * Indicates a constant from the \ref rc_constant_list . + */ + RC_FILE_CONSTANT +} rc_register_file; + +#define RC_REGISTER_INDEX_BITS 10 +#define RC_REGISTER_MAX_INDEX (1 << RC_REGISTER_INDEX_BITS) + +typedef enum { + RC_SWIZZLE_X = 0, + RC_SWIZZLE_Y, + RC_SWIZZLE_Z, + RC_SWIZZLE_W, + RC_SWIZZLE_ZERO, + RC_SWIZZLE_ONE, + RC_SWIZZLE_HALF, + RC_SWIZZLE_UNUSED +} rc_swizzle; + +#define RC_MAKE_SWIZZLE(a,b,c,d) (((a)<<0) | ((b)<<3) | ((c)<<6) | ((d)<<9)) +#define RC_MAKE_SWIZZLE_SMEAR(a) RC_MAKE_SWIZZLE((a),(a),(a),(a)) +#define GET_SWZ(swz, idx) (((swz) >> ((idx)*3)) & 0x7) +#define GET_BIT(msk, idx) (((msk) >> (idx)) & 0x1) +#define SET_SWZ(swz, idx, newv) \ + do { \ + (swz) = ((swz) & ~(7 << ((idx)*3))) | ((newv) << ((idx)*3)); \ + } while(0) + +#define RC_SWIZZLE_XYZW RC_MAKE_SWIZZLE(RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_W) +#define RC_SWIZZLE_XXXX RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_X) +#define RC_SWIZZLE_YYYY RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_Y) +#define RC_SWIZZLE_ZZZZ RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_Z) +#define RC_SWIZZLE_WWWW RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_W) +#define RC_SWIZZLE_0000 RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_ZERO) +#define RC_SWIZZLE_1111 RC_MAKE_SWIZZLE_SMEAR(RC_SWIZZLE_ONE) + +/** + * \name Bitmasks for components of vectors. + * + * Used for write masks, negation masks, etc. + */ +/*@{*/ +#define RC_MASK_NONE 0 +#define RC_MASK_X 1 +#define RC_MASK_Y 2 +#define RC_MASK_Z 4 +#define RC_MASK_W 8 +#define RC_MASK_XY (RC_MASK_X|RC_MASK_Y) +#define RC_MASK_XYZ (RC_MASK_X|RC_MASK_Y|RC_MASK_Z) +#define RC_MASK_XYW (RC_MASK_X|RC_MASK_Y|RC_MASK_W) +#define RC_MASK_XYZW (RC_MASK_X|RC_MASK_Y|RC_MASK_Z|RC_MASK_W) +/*@}*/ + +#endif /* RADEON_PROGRAM_CONSTANTS_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 index 4c26db5d244..6cda2086008 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.c @@ -35,9 +35,10 @@ #include "radeon_program_pair.h" +#include <stdio.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", \ @@ -45,19 +46,19 @@ } while(0) struct pair_state_instruction { - struct prog_instruction Instruction; - GLuint IP; /**< Position of this instruction in original program */ + struct rc_sub_instruction Instruction; + unsigned int 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 */ + unsigned int IsTex:1; /**< Is a texture instruction */ + unsigned int NeedRGB:1; /**< Needs the RGB ALU */ + unsigned int NeedAlpha:1; /**< Needs the Alpha ALU */ + unsigned int 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; + unsigned int NumDependencies:5; /** * Next instruction in the linked list of ready instructions. @@ -81,11 +82,11 @@ struct reg_value_reader { /** * Used to keep track which values are stored in each component of a - * PROGRAM_TEMPORARY. + * RC_FILE_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 */ + struct reg_value *Next; /**< Pointer to the next value to be written to the same RC_FILE_TEMPORARY component */ /** * Unordered linked list of instructions that read from this value. @@ -98,21 +99,21 @@ struct reg_value { * When this count reaches zero, the instruction that writes the @ref Next value * can be scheduled. */ - GLuint NumReaders; + unsigned int NumReaders; }; /** - * Used to translate a PROGRAM_INPUT or PROGRAM_TEMPORARY Mesa register + * Used to translate a RC_FILE_INPUT or RC_FILE_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 */ + unsigned int Allocated:1; + unsigned int HwIndex:8; + unsigned int 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). + * (only used for RC_FILE_TEMPORARY registers). */ struct reg_value *Value[4]; }; @@ -120,17 +121,17 @@ struct pair_register_translation { struct pair_state { struct r300_fragment_program_compiler * Compiler; const struct radeon_pair_handler *Handler; - GLboolean Verbose; + unsigned int 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 pair_register_translation Inputs[RC_REGISTER_MAX_INDEX]; + struct pair_register_translation Temps[RC_REGISTER_MAX_INDEX]; struct { - GLuint RefCount; /**< # of times this occurs in an unscheduled SrcReg or DstReg */ + unsigned int RefCount; /**< # of times this occurs in an unscheduled SrcReg or DstReg */ } HwTemps[128]; /** @@ -144,28 +145,28 @@ struct pair_state { }; -static struct pair_register_translation *get_register(struct pair_state *s, GLuint file, GLuint index) +static struct pair_register_translation *get_register(struct pair_state *s, rc_register_file file, unsigned int index) { switch(file) { - case PROGRAM_TEMPORARY: return &s->Temps[index]; - case PROGRAM_INPUT: return &s->Inputs[index]; + case RC_FILE_TEMPORARY: return &s->Temps[index]; + case RC_FILE_INPUT: return &s->Inputs[index]; default: return 0; } } -static void alloc_hw_reg(struct pair_state *s, GLuint file, GLuint index, GLuint hwindex) +static void alloc_hw_reg(struct pair_state *s, rc_register_file file, unsigned int index, unsigned int hwindex) { struct pair_register_translation *t = get_register(s, file, index); - ASSERT(!s->HwTemps[hwindex].RefCount); - ASSERT(!t->Allocated); + 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) +static unsigned int get_hw_reg(struct pair_state *s, rc_register_file file, unsigned int index) { - GLuint hwindex; + unsigned int hwindex; struct pair_register_translation *t = get_register(s, file, index); if (!t) { @@ -190,7 +191,7 @@ static GLuint get_hw_reg(struct pair_state *s, GLuint file, GLuint index) } -static void deref_hw_reg(struct pair_state *s, GLuint hwindex) +static void deref_hw_reg(struct pair_state *s, unsigned int hwindex) { if (!s->HwTemps[hwindex].RefCount) { error("Hwindex %i refcount error", hwindex); @@ -213,7 +214,7 @@ static void add_pairinst_to_list(struct pair_state_instruction **list, struct pa static void instruction_ready(struct pair_state *s, struct pair_state_instruction *pairinst) { if (s->Verbose) - _mesa_printf("instruction_ready(%i)\n", pairinst->IP); + fprintf(stderr, "instruction_ready(%i)\n", pairinst->IP); if (pairinst->IsTex) add_pairinst_to_list(&s->ReadyTEX, pairinst); @@ -230,24 +231,24 @@ static void instruction_ready(struct pair_state *s, struct pair_state_instructio * 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) +static void final_rewrite(struct pair_state *s, struct rc_sub_instruction *inst) { - struct prog_src_register tmp; + struct rc_src_register tmp; switch(inst->Opcode) { - case OPCODE_ADD: + case RC_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; + inst->SrcReg[1].File = RC_FILE_NONE; + inst->SrcReg[1].Swizzle = RC_SWIZZLE_1111; + inst->SrcReg[1].Negate = RC_MASK_NONE; + inst->Opcode = RC_OPCODE_MAD; break; - case OPCODE_CMP: + case RC_OPCODE_CMP: tmp = inst->SrcReg[2]; inst->SrcReg[2] = inst->SrcReg[0]; inst->SrcReg[0] = tmp; break; - case OPCODE_MOV: + case RC_OPCODE_MOV: /* AMD say we should use CMP. * However, when we transform * KIL -r0; @@ -258,16 +259,16 @@ static void final_rewrite(struct pair_state *s, struct prog_instruction *inst) * 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; + inst->SrcReg[1].File = RC_FILE_NONE; + inst->SrcReg[1].Swizzle = RC_SWIZZLE_1111; + inst->SrcReg[2].File = RC_FILE_NONE; + inst->SrcReg[2].Swizzle = RC_SWIZZLE_0000; + inst->Opcode = RC_OPCODE_MAD; break; - case OPCODE_MUL: - inst->SrcReg[2].File = PROGRAM_BUILTIN; - inst->SrcReg[2].Swizzle = SWIZZLE_0000; - inst->Opcode = OPCODE_MAD; + case RC_OPCODE_MUL: + inst->SrcReg[2].File = RC_FILE_NONE; + inst->SrcReg[2].Swizzle = RC_SWIZZLE_0000; + inst->Opcode = RC_OPCODE_MAD; break; default: /* nothing to do */ @@ -282,41 +283,40 @@ static void final_rewrite(struct pair_state *s, struct prog_instruction *inst) 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; + psi->NeedRGB = (psi->Instruction.DstReg.WriteMask & RC_MASK_XYZ) ? 1 : 0; + psi->NeedAlpha = (psi->Instruction.DstReg.WriteMask & RC_MASK_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: + case RC_OPCODE_ADD: + case RC_OPCODE_CMP: + case RC_OPCODE_DDX: + case RC_OPCODE_DDY: + case RC_OPCODE_FRC: + case RC_OPCODE_MAD: + case RC_OPCODE_MAX: + case RC_OPCODE_MIN: + case RC_OPCODE_MOV: + case RC_OPCODE_MUL: break; - case OPCODE_COS: - case OPCODE_EX2: - case OPCODE_LG2: - case OPCODE_RCP: - case OPCODE_RSQ: - case OPCODE_SIN: + case RC_OPCODE_COS: + case RC_OPCODE_EX2: + case RC_OPCODE_LG2: + case RC_OPCODE_RCP: + case RC_OPCODE_RSQ: + case RC_OPCODE_SIN: psi->IsTranscendent = 1; psi->NeedAlpha = 1; break; - case OPCODE_DP4: + case RC_OPCODE_DP4: psi->NeedAlpha = 1; /* fall through */ - case OPCODE_DP3: + case RC_OPCODE_DP3: psi->NeedRGB = 1; break; - case OPCODE_KIL: - case OPCODE_TEX: - case OPCODE_TXB: - case OPCODE_TXP: - case OPCODE_END: + case RC_OPCODE_KIL: + case RC_OPCODE_TEX: + case RC_OPCODE_TXB: + case RC_OPCODE_TXP: psi->IsTex = 1; break; default: @@ -333,7 +333,7 @@ static void classify_instruction(struct pair_state *s, static void scan_instructions(struct pair_state *s) { struct rc_instruction *source; - GLuint ip; + unsigned int ip; for(source = s->Compiler->Base.Program.Instructions.Next, ip = 0; source != &s->Compiler->Base.Program.Instructions; @@ -346,9 +346,9 @@ static void scan_instructions(struct pair_state *s) final_rewrite(s, &pairinst->Instruction); classify_instruction(s, pairinst); - int nsrc = _mesa_num_inst_src_regs(pairinst->Instruction.Opcode); + const struct rc_opcode_info * opcode = rc_get_opcode_info(pairinst->Instruction.Opcode); int j; - for(j = 0; j < nsrc; j++) { + for(j = 0; j < opcode->NumSrcRegs; j++) { struct pair_register_translation *t = get_register(s, pairinst->Instruction.SrcReg[j].File, pairinst->Instruction.SrcReg[j].Index); if (!t) @@ -356,10 +356,10 @@ static void scan_instructions(struct pair_state *s) t->RefCount++; - if (pairinst->Instruction.SrcReg[j].File == PROGRAM_TEMPORARY) { + if (pairinst->Instruction.SrcReg[j].File == RC_FILE_TEMPORARY) { int i; for(i = 0; i < 4; ++i) { - GLuint swz = GET_SWZ(pairinst->Instruction.SrcReg[j].Swizzle, i); + unsigned int swz = GET_SWZ(pairinst->Instruction.SrcReg[j].Swizzle, i); if (swz >= 4) continue; /* constant or NIL swizzle */ if (!t->Value[swz]) @@ -369,7 +369,7 @@ static void scan_instructions(struct pair_state *s) * 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 && + if (pairinst->Instruction.DstReg.File == RC_FILE_TEMPORARY && pairinst->Instruction.DstReg.Index == pairinst->Instruction.SrcReg[j].Index && GET_BIT(pairinst->Instruction.DstReg.WriteMask, swz)) continue; @@ -384,14 +384,13 @@ static void scan_instructions(struct pair_state *s) } } - int ndst = _mesa_num_inst_dst_regs(pairinst->Instruction.Opcode); - if (ndst) { + if (opcode->HasDstReg) { 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) { + if (pairinst->Instruction.DstReg.File == RC_FILE_TEMPORARY) { int j; for(j = 0; j < 4; ++j) { if (!GET_BIT(pairinst->Instruction.DstReg.WriteMask, j)) @@ -412,15 +411,15 @@ static void scan_instructions(struct pair_state *s) } if (s->Verbose) - _mesa_printf("scan(%i): NumDeps = %i\n", ip, pairinst->NumDependencies); + fprintf(stderr, "scan(%i): NumDeps = %i\n", ip, pairinst->NumDependencies); if (!pairinst->NumDependencies) instruction_ready(s, pairinst); } - /* Clear the PROGRAM_TEMPORARY state */ + /* Clear the RC_FILE_TEMPORARY state */ int i, j; - for(i = 0; i < MAX_PROGRAM_TEMPS; ++i) { + for(i = 0; i < RC_REGISTER_MAX_INDEX; ++i) { for(j = 0; j < 4; ++j) s->Temps[i].Value[j] = 0; } @@ -429,7 +428,7 @@ static void scan_instructions(struct pair_state *s) static void decrement_dependencies(struct pair_state *s, struct pair_state_instruction *pairinst) { - ASSERT(pairinst->NumDependencies > 0); + assert(pairinst->NumDependencies > 0); if (!--pairinst->NumDependencies) instruction_ready(s, pairinst); } @@ -440,12 +439,12 @@ static void decrement_dependencies(struct pair_state *s, struct pair_state_instr */ static void commit_instruction(struct pair_state *s, struct pair_state_instruction *pairinst) { - struct prog_instruction *inst = &pairinst->Instruction; + struct rc_sub_instruction *inst = &pairinst->Instruction; if (s->Verbose) - _mesa_printf("commit_instruction(%i)\n", pairinst->IP); + fprintf(stderr, "commit_instruction(%i)\n", pairinst->IP); - if (inst->DstReg.File == PROGRAM_TEMPORARY) { + if (inst->DstReg.File == RC_FILE_TEMPORARY) { struct pair_register_translation *t = &s->Temps[inst->DstReg.Index]; deref_hw_reg(s, t->HwIndex); @@ -467,21 +466,21 @@ static void commit_instruction(struct pair_state *s, struct pair_state_instructi } } - int nsrc = _mesa_num_inst_src_regs(inst->Opcode); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->Opcode); int i; - for(i = 0; i < nsrc; i++) { + for(i = 0; i < opcode->NumSrcRegs; 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) + if (inst->SrcReg[i].File != RC_FILE_TEMPORARY) continue; int j; for(j = 0; j < 4; ++j) { - GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j); + unsigned int swz = GET_SWZ(inst->SrcReg[i].Swizzle, j); if (swz >= 4) continue; if (!t->Value[swz]) @@ -489,7 +488,7 @@ static void commit_instruction(struct pair_state *s, struct pair_state_instructi /* Do not free a dependency if this instruction * also rewrites the value. See scan_instructions. */ - if (inst->DstReg.File == PROGRAM_TEMPORARY && + if (inst->DstReg.File == RC_FILE_TEMPORARY && inst->DstReg.Index == inst->SrcReg[i].Index && GET_BIT(inst->DstReg.WriteMask, swz)) continue; @@ -519,7 +518,7 @@ static void emit_all_tex(struct pair_state *s) struct pair_state_instruction *readytex; struct pair_state_instruction *pairinst; - ASSERT(s->ReadyTEX); + assert(s->ReadyTEX); // Don't let the ready list change under us! readytex = s->ReadyTEX; @@ -527,39 +526,37 @@ static void emit_all_tex(struct pair_state *s) // 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) + struct rc_sub_instruction *inst = &pairinst->Instruction; + if (inst->Opcode != RC_OPCODE_KIL) get_hw_reg(s, inst->DstReg.File, inst->DstReg.Index); } if (s->Compiler->Base.Debug) - _mesa_printf(" BEGIN_TEX\n"); + fprintf(stderr, " 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; + struct rc_sub_instruction *inst = &pairinst->Instruction; commit_instruction(s, pairinst); - if (inst->Opcode != OPCODE_KIL) + if (inst->Opcode != RC_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(stderr); + /* Should print the TEX instruction here */ } 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; + case RC_OPCODE_TEX: rpti.Opcode = RADEON_OPCODE_TEX; break; + case RC_OPCODE_TXB: rpti.Opcode = RADEON_OPCODE_TXB; break; + case RC_OPCODE_TXP: rpti.Opcode = RADEON_OPCODE_TXP; break; default: - case OPCODE_KIL: rpti.Opcode = RADEON_OPCODE_KIL; break; + case RC_OPCODE_KIL: rpti.Opcode = RADEON_OPCODE_KIL; break; } rpti.DestIndex = inst->DstReg.Index; @@ -573,12 +570,12 @@ static void emit_all_tex(struct pair_state *s) } if (s->Compiler->Base.Debug) - _mesa_printf(" END_TEX\n"); + fprintf(stderr, " 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) + struct rc_src_register src, unsigned int rgb, unsigned int alpha) { int candidate = -1; int candidate_quality = -1; @@ -587,10 +584,10 @@ static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instructio if (!rgb && !alpha) return 0; - GLuint constant; - GLuint index; + unsigned int constant; + unsigned int index; - if (src.File == PROGRAM_TEMPORARY || src.File == PROGRAM_INPUT) { + if (src.File == RC_FILE_TEMPORARY || src.File == RC_FILE_INPUT) { constant = 0; index = get_hw_reg(s, src.File, src.Index); } else { @@ -642,37 +639,38 @@ static int alloc_pair_source(struct pair_state *s, struct radeon_pair_instructio * Fill the given ALU instruction's opcodes and source operands into the given pair, * if possible. */ -static GLboolean fill_instruction_into_pair( +static int fill_instruction_into_pair( struct pair_state *s, struct radeon_pair_instruction *pair, struct pair_state_instruction *pairinst) { - struct prog_instruction *inst = &pairinst->Instruction; + struct rc_sub_instruction *inst = &pairinst->Instruction; - ASSERT(!pairinst->NeedRGB || pair->RGB.Opcode == OPCODE_NOP); - ASSERT(!pairinst->NeedAlpha || pair->Alpha.Opcode == OPCODE_NOP); + assert(!pairinst->NeedRGB || pair->RGB.Opcode == RC_OPCODE_NOP); + assert(!pairinst->NeedAlpha || pair->Alpha.Opcode == RC_OPCODE_NOP); if (pairinst->NeedRGB) { if (pairinst->IsTranscendent) - pair->RGB.Opcode = OPCODE_REPL_ALPHA; + pair->RGB.Opcode = RC_OPCODE_REPL_ALPHA; else pair->RGB.Opcode = inst->Opcode; - if (inst->SaturateMode == SATURATE_ZERO_ONE) + if (inst->SaturateMode == RC_SATURATE_ZERO_ONE) pair->RGB.Saturate = 1; } if (pairinst->NeedAlpha) { pair->Alpha.Opcode = inst->Opcode; - if (inst->SaturateMode == SATURATE_ZERO_ONE) + if (inst->SaturateMode == RC_SATURATE_ZERO_ONE) pair->Alpha.Saturate = 1; } - int nargs = _mesa_num_inst_src_regs(inst->Opcode); + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->Opcode); + int nargs = opcode->NumSrcRegs; int i; /* Special case for DDX/DDY (MDH/MDV). */ - if (inst->Opcode == OPCODE_DDX || inst->Opcode == OPCODE_DDY) { + if (inst->Opcode == RC_OPCODE_DDX || inst->Opcode == RC_OPCODE_DDY) { if (pair->RGB.Src[0].Used || pair->Alpha.Src[0].Used) - return GL_FALSE; + return 0; else nargs++; } @@ -680,43 +678,43 @@ static GLboolean fill_instruction_into_pair( for(i = 0; i < nargs; ++i) { int source; if (pairinst->NeedRGB && !pairinst->IsTranscendent) { - GLboolean srcrgb = GL_FALSE; - GLboolean srcalpha = GL_FALSE; + unsigned int srcrgb = 0; + unsigned int srcalpha = 0; int j; for(j = 0; j < 3; ++j) { - GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, j); + unsigned int swz = GET_SWZ(inst->SrcReg[i].Swizzle, j); if (swz < 3) - srcrgb = GL_TRUE; + srcrgb = 1; else if (swz < 4) - srcalpha = GL_TRUE; + srcalpha = 1; } source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha); if (source < 0) - return GL_FALSE; + return 0; 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)); + pair->RGB.Arg[i].Negate = !!(inst->SrcReg[i].Negate & (RC_MASK_X | RC_MASK_Y | RC_MASK_Z)); } if (pairinst->NeedAlpha) { - GLboolean srcrgb = GL_FALSE; - GLboolean srcalpha = GL_FALSE; - GLuint swz = GET_SWZ(inst->SrcReg[i].Swizzle, pairinst->IsTranscendent ? 0 : 3); + unsigned int srcrgb = 0; + unsigned int srcalpha = 0; + unsigned int swz = GET_SWZ(inst->SrcReg[i].Swizzle, pairinst->IsTranscendent ? 0 : 3); if (swz < 3) - srcrgb = GL_TRUE; + srcrgb = 1; else if (swz < 4) - srcalpha = GL_TRUE; + srcalpha = 1; source = alloc_pair_source(s, pair, inst->SrcReg[i], srcrgb, srcalpha); if (source < 0) - return GL_FALSE; + return 0; 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); + pair->Alpha.Arg[i].Negate = !!(inst->SrcReg[i].Negate & RC_MASK_W); } } - return GL_TRUE; + return 1; } @@ -733,20 +731,20 @@ static void fill_dest_into_pair( struct radeon_pair_instruction *pair, struct pair_state_instruction *pairinst) { - struct prog_instruction *inst = &pairinst->Instruction; + struct rc_sub_instruction *inst = &pairinst->Instruction; - if (inst->DstReg.File == PROGRAM_OUTPUT) { + if (inst->DstReg.File == RC_FILE_OUTPUT) { if (inst->DstReg.Index == s->Compiler->OutputColor) { - pair->RGB.OutputWriteMask |= inst->DstReg.WriteMask & WRITEMASK_XYZ; + pair->RGB.OutputWriteMask |= inst->DstReg.WriteMask & RC_MASK_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); + unsigned int 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; + pair->RGB.WriteMask |= inst->DstReg.WriteMask & RC_MASK_XYZ; } if (pairinst->NeedAlpha) { pair->Alpha.DestIndex = hwindex; @@ -780,7 +778,7 @@ static void emit_alu(struct pair_state *s) s->ReadyAlpha = s->ReadyAlpha->NextReady; } - _mesa_bzero(&pair, sizeof(pair)); + memset(&pair, 0, sizeof(pair)); fill_instruction_into_pair(s, &pair, psi); fill_dest_into_pair(s, &pair, psi); commit_instruction(s, psi); @@ -794,7 +792,7 @@ static void emit_alu(struct pair_state *s) 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)); + memset(&pair, 0, sizeof(pair)); fill_instruction_into_pair(s, &pair, psirgb); if (!fill_instruction_into_pair(s, &pair, psialpha)) continue; @@ -812,7 +810,7 @@ static void emit_alu(struct pair_state *s) psi = s->ReadyRGB; s->ReadyRGB = s->ReadyRGB->NextReady; - _mesa_bzero(&pair, sizeof(pair)); + memset(&pair, 0, sizeof(pair)); fill_instruction_into_pair(s, &pair, psi); fill_dest_into_pair(s, &pair, psi); commit_instruction(s, psi); @@ -829,7 +827,7 @@ static void emit_alu(struct pair_state *s) static void alloc_helper(void * data, unsigned input, unsigned hwreg) { struct pair_state * s = data; - alloc_hw_reg(s, PROGRAM_INPUT, input, hwreg); + alloc_hw_reg(s, RC_FILE_INPUT, input, hwreg); } void radeonPairProgram( @@ -838,14 +836,14 @@ void radeonPairProgram( { struct pair_state s; - _mesa_bzero(&s, sizeof(s)); + memset(&s, 0, sizeof(s)); s.Compiler = compiler; s.Handler = handler; s.UserData = userdata; - s.Verbose = GL_FALSE && s.Compiler->Base.Debug; + s.Verbose = 0 && s.Compiler->Base.Debug; if (s.Compiler->Base.Debug) - _mesa_printf("Emit paired program\n"); + fprintf(stderr, "Emit paired program\n"); scan_instructions(&s); s.Compiler->AllocateHwInputs(s.Compiler, &alloc_helper, &s); @@ -860,41 +858,36 @@ void radeonPairProgram( } if (s.Compiler->Base.Debug) - _mesa_printf(" END\n"); + fprintf(stderr, " 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); + fprintf(stderr, " Src%i = %s[%i]", i, src->Constant ? "CNST" : "TEMP", src->Index); } -static const char* opcode_string(GLuint opcode) +static const char* opcode_string(rc_opcode opcode) { - if (opcode == OPCODE_REPL_ALPHA) - return "SOP"; - else - return _mesa_opcode_string(opcode); + return rc_get_opcode_info(opcode)->Name; } -static int num_pairinst_args(GLuint opcode) +static int num_pairinst_args(rc_opcode opcode) { - if (opcode == OPCODE_REPL_ALPHA) - return 0; - else - return _mesa_num_inst_src_regs(opcode); + return rc_get_opcode_info(opcode)->NumSrcRegs; } -static char swizzle_char(GLuint swz) +static char swizzle_char(rc_swizzle 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 '_'; + case RC_SWIZZLE_X: return 'x'; + case RC_SWIZZLE_Y: return 'y'; + case RC_SWIZZLE_Z: return 'z'; + case RC_SWIZZLE_W: return 'w'; + case RC_SWIZZLE_ZERO: return '0'; + case RC_SWIZZLE_ONE: return '1'; + case RC_SWIZZLE_HALF: return 'H'; + case RC_SWIZZLE_UNUSED: return '_'; default: return '?'; } } @@ -904,27 +897,27 @@ void radeonPrintPairInstruction(struct radeon_pair_instruction *inst) int nargs; int i; - _mesa_printf(" RGB: "); + fprintf(stderr, " 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:"); + fprintf(stderr, "\n"); + fprintf(stderr, " Alpha:"); for(i = 0; i < 3; ++i) { if (inst->Alpha.Src[i].Used) print_pair_src(i, inst->Alpha.Src + i); } - _mesa_printf("\n"); + fprintf(stderr, "\n"); - _mesa_printf(" %s%s", opcode_string(inst->RGB.Opcode), inst->RGB.Saturate ? "_SAT" : ""); + fprintf(stderr, " %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, + fprintf(stderr, " 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", + fprintf(stderr, " COLOR.%s%s%s", (inst->RGB.OutputWriteMask & 1) ? "x" : "", (inst->RGB.OutputWriteMask & 2) ? "y" : "", (inst->RGB.OutputWriteMask & 4) ? "z" : ""); @@ -932,27 +925,27 @@ void radeonPrintPairInstruction(struct radeon_pair_instruction *inst) 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, + fprintf(stderr, ", %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"); + fprintf(stderr, "\n"); - _mesa_printf(" %s%s", opcode_string(inst->Alpha.Opcode), inst->Alpha.Saturate ? "_SAT" : ""); + fprintf(stderr, " %s%s", opcode_string(inst->Alpha.Opcode), inst->Alpha.Saturate ? "_SAT" : ""); if (inst->Alpha.WriteMask) - _mesa_printf(" TEMP[%i].w", inst->Alpha.DestIndex); + fprintf(stderr, " TEMP[%i].w", inst->Alpha.DestIndex); if (inst->Alpha.OutputWriteMask) - _mesa_printf(" COLOR.w"); + fprintf(stderr, " COLOR.w"); if (inst->Alpha.DepthWriteMask) - _mesa_printf(" DEPTH.w"); + fprintf(stderr, " 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, + fprintf(stderr, ", %s%sSrc%i.%c%s", neg, abs, inst->Alpha.Arg[i].Source, swizzle_char(inst->Alpha.Arg[i].Swizzle), abs); } - _mesa_printf("\n"); + fprintf(stderr, "\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 index ff761785518..da2bcc5d899 100644 --- a/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_pair.h @@ -38,43 +38,43 @@ struct r300_fragment_program_compiler; * fragment programs. */ struct radeon_pair_instruction_source { - GLuint Index:8; - GLuint Constant:1; - GLuint Used:1; + unsigned int Index:8; + unsigned int Constant:1; + unsigned int Used:1; }; struct radeon_pair_instruction_rgb { - GLuint Opcode:8; - GLuint DestIndex:8; - GLuint WriteMask:3; - GLuint OutputWriteMask:3; - GLuint Saturate:1; + unsigned int Opcode:8; + unsigned int DestIndex:8; + unsigned int WriteMask:3; + unsigned int OutputWriteMask:3; + unsigned int Saturate:1; struct radeon_pair_instruction_source Src[3]; struct { - GLuint Source:2; - GLuint Swizzle:9; - GLuint Abs:1; - GLuint Negate:1; + unsigned int Source:2; + unsigned int Swizzle:9; + unsigned int Abs:1; + unsigned int 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; + unsigned int Opcode:8; + unsigned int DestIndex:8; + unsigned int WriteMask:1; + unsigned int OutputWriteMask:1; + unsigned int DepthWriteMask:1; + unsigned int Saturate:1; struct radeon_pair_instruction_source Src[3]; struct { - GLuint Source:2; - GLuint Swizzle:3; - GLuint Abs:1; - GLuint Negate:1; + unsigned int Source:2; + unsigned int Swizzle:3; + unsigned int Abs:1; + unsigned int Negate:1; } Arg[3]; }; @@ -92,16 +92,16 @@ enum { }; struct radeon_pair_texture_instruction { - GLuint Opcode:2; /**< one of RADEON_OPCODE_xxx */ + unsigned int Opcode:2; /**< one of RADEON_OPCODE_xxx */ - GLuint DestIndex:8; - GLuint WriteMask:4; + unsigned int DestIndex:8; + unsigned int WriteMask:4; - GLuint TexSrcUnit:5; - GLuint TexSrcTarget:3; + unsigned int TexSrcUnit:5; + unsigned int TexSrcTarget:3; - GLuint SrcIndex:8; - GLuint SrcSwizzle:12; + unsigned int SrcIndex:8; + unsigned int SrcSwizzle:12; }; @@ -112,24 +112,24 @@ struct radeon_pair_handler { /** * Write a paired instruction to the hardware. * - * @return GL_FALSE on error. + * @return 0 on error. */ - GLboolean (*EmitPaired)(void*, struct radeon_pair_instruction*); + int (*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. + * @return 0 on error. */ - GLboolean (*EmitTex)(void*, struct radeon_pair_texture_instruction*); + int (*EmitTex)(void*, struct radeon_pair_texture_instruction*); /** * Called before a block of contiguous, independent texture * instructions is emitted. */ - GLboolean (*BeginTexBlock)(void*); + int (*BeginTexBlock)(void*); unsigned MaxHwTemps; }; diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c b/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c new file mode 100644 index 00000000000..38060ea3adf --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_program_print.c @@ -0,0 +1,214 @@ +/* + * Copyright 2009 Nicolai Hähnle <[email protected]> + * + * 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_program.h" + +#include <stdio.h> + +static void print_comment(FILE * f) +{ + fprintf(f, " # "); +} + +static const char * textarget_to_string(rc_texture_target target) +{ + switch(target) { + case RC_TEXTURE_2D_ARRAY: return "2D_ARRAY"; + case RC_TEXTURE_1D_ARRAY: return "1D_ARRAY"; + case RC_TEXTURE_CUBE: return "CUBE"; + case RC_TEXTURE_3D: return "3D"; + case RC_TEXTURE_RECT: return "RECT"; + case RC_TEXTURE_2D: return "2D"; + case RC_TEXTURE_1D: return "1D"; + default: return "BAD_TEXTURE_TARGET"; + } +} + +static void rc_print_register(FILE * f, rc_register_file file, int index, unsigned int reladdr) +{ + if (file == RC_FILE_NONE) { + fprintf(f, "none"); + } else { + const char * filename; + switch(file) { + case RC_FILE_TEMPORARY: filename = "temp"; break; + case RC_FILE_INPUT: filename = "input"; break; + case RC_FILE_OUTPUT: filename = "output"; break; + case RC_FILE_ADDRESS: filename = "addr"; break; + case RC_FILE_CONSTANT: filename = "const"; break; + default: filename = "BAD FILE"; break; + } + fprintf(f, "%s[%i%s]", filename, index, reladdr ? " + addr[0]" : ""); + } +} + +static void rc_print_mask(FILE * f, unsigned int mask) +{ + if (mask & RC_MASK_X) fprintf(f, "x"); + if (mask & RC_MASK_Y) fprintf(f, "y"); + if (mask & RC_MASK_Z) fprintf(f, "z"); + if (mask & RC_MASK_W) fprintf(f, "w"); +} + +static void rc_print_dst_register(FILE * f, struct rc_dst_register dst) +{ + rc_print_register(f, dst.File, dst.Index, dst.RelAddr); + if (dst.WriteMask != RC_MASK_XYZW) { + fprintf(f, "."); + rc_print_mask(f, dst.WriteMask); + } +} + +static void rc_print_swizzle(FILE * f, unsigned int swizzle, unsigned int negate) +{ + unsigned int comp; + for(comp = 0; comp < 4; ++comp) { + rc_swizzle swz = GET_SWZ(swizzle, comp); + if (GET_BIT(negate, comp)) + fprintf(f, "-"); + switch(swz) { + case RC_SWIZZLE_X: fprintf(f, "x"); break; + case RC_SWIZZLE_Y: fprintf(f, "y"); break; + case RC_SWIZZLE_Z: fprintf(f, "z"); break; + case RC_SWIZZLE_W: fprintf(f, "w"); break; + case RC_SWIZZLE_ZERO: fprintf(f, "0"); break; + case RC_SWIZZLE_ONE: fprintf(f, "1"); break; + case RC_SWIZZLE_HALF: fprintf(f, "H"); break; + case RC_SWIZZLE_UNUSED: fprintf(f, "_"); break; + } + } +} + +static void rc_print_src_register(FILE * f, struct rc_src_register src) +{ + int trivial_negate = (src.Negate == RC_MASK_NONE || src.Negate == RC_MASK_XYZW); + + if (src.Negate == RC_MASK_XYZW) + fprintf(f, "-"); + if (src.Abs) + fprintf(f, "|"); + + rc_print_register(f, src.File, src.Index, src.RelAddr); + + if (src.Abs && !trivial_negate) + fprintf(f, "|"); + + if (src.Swizzle != RC_SWIZZLE_XYZW || !trivial_negate) { + fprintf(f, "."); + rc_print_swizzle(f, src.Swizzle, trivial_negate ? 0 : src.Negate); + } + + if (src.Abs && trivial_negate) + fprintf(f, "|"); +} + +static void rc_print_ref(FILE * f, struct rc_dataflow_ref * ref) +{ + fprintf(f, "ref(%p", ref->Vector); + + if (ref->UseMask != RC_MASK_XYZW) { + fprintf(f, "."); + rc_print_mask(f, ref->UseMask); + } + + fprintf(f, ")"); +} + +static void rc_print_instruction(FILE * f, unsigned int flags, struct rc_instruction * inst) +{ + const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->I.Opcode); + unsigned int reg; + + fprintf(f, "%s", opcode->Name); + + switch(inst->I.SaturateMode) { + case RC_SATURATE_NONE: break; + case RC_SATURATE_ZERO_ONE: fprintf(f, "_SAT"); break; + case RC_SATURATE_MINUS_PLUS_ONE: fprintf(f, "_SAT2"); break; + default: fprintf(f, "_BAD_SAT"); break; + } + + if (opcode->HasDstReg) { + fprintf(f, " "); + rc_print_dst_register(f, inst->I.DstReg); + if (opcode->NumSrcRegs) + fprintf(f, ","); + } + + for(reg = 0; reg < opcode->NumSrcRegs; ++reg) { + if (reg > 0) + fprintf(f, ","); + fprintf(f, " "); + rc_print_src_register(f, inst->I.SrcReg[reg]); + } + + if (opcode->HasTexture) { + fprintf(f, ", %s%s[%u]", + textarget_to_string(inst->I.TexSrcTarget), + inst->I.TexShadow ? "SHADOW" : "", + inst->I.TexSrcUnit); + } + + fprintf(f, ";\n"); + + if (flags & RC_PRINT_DATAFLOW) { + print_comment(f); + + fprintf(f, "Dst = %p", inst->Dataflow.DstReg); + if (inst->Dataflow.DstRegAliased) + fprintf(f, " aliased"); + if (inst->Dataflow.DstRegPrev) { + fprintf(f, " from "); + rc_print_ref(f, inst->Dataflow.DstRegPrev); + } + + for(reg = 0; reg < opcode->NumSrcRegs; ++reg) { + fprintf(f, ", "); + if (inst->Dataflow.SrcReg[reg]) + rc_print_ref(f, inst->Dataflow.SrcReg[reg]); + else + fprintf(f, "<no ref>"); + } + + fprintf(f, "\n"); + } +} + +/** + * Print program to stderr, default options. + */ +void rc_print_program(const struct rc_program *prog, unsigned int flags) +{ + unsigned int linenum = 0; + struct rc_instruction *inst; + + fprintf(stderr, "# Radeon Compiler Program%s\n", + flags & RC_PRINT_DATAFLOW ? " (with dataflow annotations)" : ""); + + for(inst = prog->Instructions.Next; inst != &prog->Instructions; inst = inst->Next) { + fprintf(stderr, "%3d: ", linenum); + + rc_print_instruction(stderr, flags, inst); + + linenum++; + } +} diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_swizzle.h b/src/mesa/drivers/dri/r300/compiler/radeon_swizzle.h new file mode 100644 index 00000000000..c81d5f7a5e9 --- /dev/null +++ b/src/mesa/drivers/dri/r300/compiler/radeon_swizzle.h @@ -0,0 +1,57 @@ +/* + * 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. + * + */ + +#ifndef RADEON_SWIZZLE_H +#define RADEON_SWIZZLE_H + +#include "radeon_program.h" + +struct rc_swizzle_split { + unsigned char NumPhases; + unsigned char Phase[4]; +}; + +/** + * Describe the swizzling capability of target hardware. + */ +struct rc_swizzle_caps { + /** + * Check whether the given swizzle, absolute and negate combination + * can be implemented natively by the hardware for this opcode. + * + * \return 1 if the swizzle is native for the given opcode + */ + int (*IsNative)(rc_opcode opcode, struct rc_src_register reg); + + /** + * Determine how to split access to the masked channels of the + * given source register to obtain ALU-native swizzles. + */ + void (*Split)(struct rc_src_register reg, unsigned int mask, struct rc_swizzle_split * split); +}; + +#endif /* RADEON_SWIZZLE_H */ diff --git a/src/mesa/drivers/dri/r300/r300_fragprog_common.c b/src/mesa/drivers/dri/r300/r300_fragprog_common.c index 0bdc90b4a84..39618b78dc6 100644 --- a/src/mesa/drivers/dri/r300/r300_fragprog_common.c +++ b/src/mesa/drivers/dri/r300/r300_fragprog_common.c @@ -44,6 +44,7 @@ #include "compiler/radeon_compiler.h" +#include "radeon_mesa_to_rc.h" #include "r300_state.h" @@ -131,7 +132,7 @@ static void insert_WPOS_trailer(struct r300_fragment_program_compiler *compiler, */ static void rewriteFog(struct r300_fragment_program_compiler *compiler, struct r300_fragment_program * fp) { - struct prog_src_register src; + struct rc_src_register src; int i; fp->fog_attr = FRAG_ATTRIB_MAX; @@ -155,7 +156,7 @@ static void rewriteFog(struct r300_fragment_program_compiler *compiler, struct r } memset(&src, 0, sizeof(src)); - src.File = PROGRAM_INPUT; + src.File = RC_FILE_INPUT; src.Index = fp->fog_attr; src.Swizzle = MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_ZERO, SWIZZLE_ZERO, SWIZZLE_ONE); rc_move_input(&compiler->Base, FRAG_ATTRIB_FOGC, src); @@ -232,7 +233,7 @@ static void translate_fragment_program(GLcontext *ctx, struct r300_fragment_prog fflush(stderr); } - rc_mesa_to_rc_program(&compiler.Base, &cont->Base.Base); + radeon_mesa_to_rc_program(&compiler.Base, &cont->Base.Base); insert_WPOS_trailer(&compiler, fp); diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c index 2f7b67c1431..fb8d6bceda8 100644 --- a/src/mesa/drivers/dri/r300/r300_vertprog.c +++ b/src/mesa/drivers/dri/r300/r300_vertprog.c @@ -41,7 +41,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE. #include "tnl/tnl.h" #include "compiler/radeon_compiler.h" -#include "compiler/radeon_nqssadce.h" +#include "radeon_mesa_to_rc.h" #include "r300_context.h" #include "r300_fragprog_common.h" #include "r300_state.h" @@ -217,20 +217,20 @@ static void initialize_NV_registers(struct radeon_compiler * compiler) for(reg = 0; reg < 12; ++reg) { inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions); - inst->I.Opcode = OPCODE_MOV; - inst->I.DstReg.File = PROGRAM_TEMPORARY; + inst->I.Opcode = RC_OPCODE_MOV; + inst->I.DstReg.File = RC_FILE_TEMPORARY; inst->I.DstReg.Index = reg; - inst->I.SrcReg[0].File = PROGRAM_BUILTIN; - inst->I.SrcReg[0].Swizzle = SWIZZLE_0000; + inst->I.SrcReg[0].File = RC_FILE_NONE; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_0000; } inst = rc_insert_new_instruction(compiler, &compiler->Program.Instructions); - inst->I.Opcode = OPCODE_ARL; - inst->I.DstReg.File = PROGRAM_ADDRESS; + inst->I.Opcode = RC_OPCODE_ARL; + inst->I.DstReg.File = RC_FILE_ADDRESS; inst->I.DstReg.Index = 0; inst->I.DstReg.WriteMask = WRITEMASK_X; - inst->I.SrcReg[0].File = PROGRAM_BUILTIN; - inst->I.SrcReg[0].Swizzle = SWIZZLE_0000; + inst->I.SrcReg[0].File = RC_FILE_NONE; + inst->I.SrcReg[0].Swizzle = RC_SWIZZLE_0000; } static struct r300_vertex_program *build_program(GLcontext *ctx, @@ -261,7 +261,7 @@ static struct r300_vertex_program *build_program(GLcontext *ctx, _mesa_insert_mvp_code(ctx, vp->Base); } - rc_mesa_to_rc_program(&compiler.Base, &vp->Base->Base); + radeon_mesa_to_rc_program(&compiler.Base, &vp->Base->Base); if (mesa_vp->IsNVProgram) initialize_NV_registers(&compiler.Base); diff --git a/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.c b/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.c new file mode 100644 index 00000000000..fb9bb9ce918 --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.c @@ -0,0 +1,223 @@ +/* + * 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 "radeon_mesa_to_rc.h" + +#include "main/mtypes.h" +#include "shader/prog_instruction.h" +#include "shader/prog_parameter.h" + +#include "compiler/radeon_compiler.h" +#include "compiler/radeon_program.h" + + +static rc_opcode translate_opcode(gl_inst_opcode opcode) +{ + switch(opcode) { + case OPCODE_NOP: return RC_OPCODE_NOP; + case OPCODE_ABS: return RC_OPCODE_ABS; + case OPCODE_ADD: return RC_OPCODE_ADD; + case OPCODE_ARL: return RC_OPCODE_ARL; + case OPCODE_CMP: return RC_OPCODE_CMP; + case OPCODE_COS: return RC_OPCODE_COS; + case OPCODE_DDX: return RC_OPCODE_DDX; + case OPCODE_DDY: return RC_OPCODE_DDY; + case OPCODE_DP3: return RC_OPCODE_DP3; + case OPCODE_DP4: return RC_OPCODE_DP4; + case OPCODE_DPH: return RC_OPCODE_DPH; + case OPCODE_DST: return RC_OPCODE_DST; + case OPCODE_EX2: return RC_OPCODE_EX2; + case OPCODE_EXP: return RC_OPCODE_EXP; + case OPCODE_FLR: return RC_OPCODE_FLR; + case OPCODE_FRC: return RC_OPCODE_FRC; + case OPCODE_KIL: return RC_OPCODE_KIL; + case OPCODE_LG2: return RC_OPCODE_LG2; + case OPCODE_LIT: return RC_OPCODE_LIT; + case OPCODE_LOG: return RC_OPCODE_LOG; + case OPCODE_LRP: return RC_OPCODE_LRP; + case OPCODE_MAD: return RC_OPCODE_MAD; + case OPCODE_MAX: return RC_OPCODE_MAX; + case OPCODE_MIN: return RC_OPCODE_MIN; + case OPCODE_MOV: return RC_OPCODE_MOV; + case OPCODE_MUL: return RC_OPCODE_MUL; + case OPCODE_POW: return RC_OPCODE_POW; + case OPCODE_RCP: return RC_OPCODE_RCP; + case OPCODE_RSQ: return RC_OPCODE_RSQ; + case OPCODE_SCS: return RC_OPCODE_SCS; + case OPCODE_SEQ: return RC_OPCODE_SEQ; + case OPCODE_SFL: return RC_OPCODE_SFL; + case OPCODE_SGE: return RC_OPCODE_SGE; + case OPCODE_SGT: return RC_OPCODE_SGT; + case OPCODE_SIN: return RC_OPCODE_SIN; + case OPCODE_SLE: return RC_OPCODE_SLE; + case OPCODE_SLT: return RC_OPCODE_SLT; + case OPCODE_SNE: return RC_OPCODE_SNE; + case OPCODE_SUB: return RC_OPCODE_SUB; + case OPCODE_SWZ: return RC_OPCODE_SWZ; + case OPCODE_TEX: return RC_OPCODE_TEX; + case OPCODE_TXB: return RC_OPCODE_TXB; + case OPCODE_TXD: return RC_OPCODE_TXD; + case OPCODE_TXL: return RC_OPCODE_TXL; + case OPCODE_TXP: return RC_OPCODE_TXP; + case OPCODE_XPD: return RC_OPCODE_XPD; + default: return RC_OPCODE_ILLEGAL_OPCODE; + } +} + +static rc_saturate_mode translate_saturate(unsigned int saturate) +{ + switch(saturate) { + default: + case SATURATE_OFF: return RC_SATURATE_NONE; + case SATURATE_ZERO_ONE: return RC_SATURATE_ZERO_ONE; + } +} + +static rc_register_file translate_register_file(unsigned int file) +{ + switch(file) { + case PROGRAM_TEMPORARY: return RC_FILE_TEMPORARY; + case PROGRAM_INPUT: return RC_FILE_INPUT; + case PROGRAM_OUTPUT: return RC_FILE_OUTPUT; + case PROGRAM_LOCAL_PARAM: + case PROGRAM_ENV_PARAM: + case PROGRAM_STATE_VAR: + case PROGRAM_NAMED_PARAM: + case PROGRAM_CONSTANT: + case PROGRAM_UNIFORM: return RC_FILE_CONSTANT; + case PROGRAM_ADDRESS: return RC_FILE_ADDRESS; + default: return RC_FILE_NONE; + } +} + +static void translate_srcreg(struct rc_src_register * dest, struct prog_src_register * src) +{ + dest->File = translate_register_file(src->File); + dest->Index = src->Index; + dest->RelAddr = src->RelAddr; + dest->Swizzle = src->Swizzle; + dest->Abs = src->Abs; + dest->Negate = src->Negate; +} + +static void translate_dstreg(struct rc_dst_register * dest, struct prog_dst_register * src) +{ + dest->File = translate_register_file(src->File); + dest->Index = src->Index; + dest->RelAddr = src->RelAddr; + dest->WriteMask = src->WriteMask; +} + +static rc_texture_target translate_tex_target(gl_texture_index target) +{ + switch(target) { + case TEXTURE_2D_ARRAY_INDEX: return RC_TEXTURE_2D_ARRAY; + case TEXTURE_1D_ARRAY_INDEX: return RC_TEXTURE_1D_ARRAY; + case TEXTURE_CUBE_INDEX: return RC_TEXTURE_CUBE; + case TEXTURE_3D_INDEX: return RC_TEXTURE_3D; + case TEXTURE_RECT_INDEX: return RC_TEXTURE_RECT; + default: + case TEXTURE_2D_INDEX: return RC_TEXTURE_2D; + case TEXTURE_1D_INDEX: return RC_TEXTURE_1D; + } +} + +static void translate_instruction(struct radeon_compiler * c, + struct rc_instruction * dest, struct prog_instruction * src) +{ + const struct rc_opcode_info * opcode; + unsigned int i; + + dest->I.Opcode = translate_opcode(src->Opcode); + if (dest->I.Opcode == RC_OPCODE_ILLEGAL_OPCODE) { + rc_error(c, "Unsupported opcode %i\n", src->Opcode); + return; + } + dest->I.SaturateMode = translate_saturate(src->SaturateMode); + + opcode = rc_get_opcode_info(dest->I.Opcode); + + for(i = 0; i < opcode->NumSrcRegs; ++i) + translate_srcreg(&dest->I.SrcReg[i], &src->SrcReg[i]); + + if (opcode->HasDstReg) + translate_dstreg(&dest->I.DstReg, &src->DstReg); + + if (opcode->HasTexture) { + dest->I.TexSrcUnit = src->TexSrcUnit; + dest->I.TexSrcTarget = translate_tex_target(src->TexSrcTarget); + dest->I.TexShadow = src->TexShadow; + } +} + +void radeon_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); + translate_instruction(c, dest, source); + } + + c->Program.ShadowSamplers = program->ShadowSamplers; + c->Program.InputsRead = program->InputsRead; + c->Program.OutputsWritten = program->OutputsWritten; + + int isNVProgram = 0; + + if (program->Target == GL_VERTEX_PROGRAM_ARB) { + struct gl_vertex_program * vp = (struct gl_vertex_program *) program; + isNVProgram = vp->IsNVProgram; + } + + if (isNVProgram) { + /* NV_vertex_program has a fixed-sized constant environment. + * This could be handled more efficiently for programs that + * do not use relative addressing. + */ + for(i = 0; i < 96; ++i) { + struct rc_constant constant; + + constant.Type = RC_CONSTANT_EXTERNAL; + constant.Size = 4; + constant.u.External = i; + + rc_constants_add(&c->Program.Constants, &constant); + } + } else { + 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); + } + } +} diff --git a/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.h b/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.h new file mode 100644 index 00000000000..9511a04f36f --- /dev/null +++ b/src/mesa/drivers/dri/r300/radeon_mesa_to_rc.h @@ -0,0 +1,36 @@ +/* + * 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. + * + */ + +#ifndef RADEON_MESA_TO_RC_H +#define RADEON_MESA_TO_RC_H + +struct gl_program; +struct radeon_compiler; + +void radeon_mesa_to_rc_program(struct radeon_compiler * c, struct gl_program * program); + +#endif /* RADEON_MESA_TO_RC_H */ |