#include "../r300_reg.h"
+#include "radeon_compiler_util.h"
#include "radeon_dataflow.h"
#include "radeon_program_alu.h"
#include "radeon_swizzle.h"
/* Reserve a temporary to use as our predicate stack counter, if we
* don't already have one. */
if (!compiler->PredicateMask) {
- unsigned int writemasks[R300_VS_MAX_TEMPS];
- memset(writemasks, 0, sizeof(writemasks));
+ unsigned int writemasks[RC_REGISTER_MAX_INDEX];
struct rc_instruction * inst;
unsigned int i;
+ memset(writemasks, 0, sizeof(writemasks));
for(inst = compiler->Base.Program.Instructions.Next;
inst != &compiler->Base.Program.Instructions;
inst = inst->Next) {
rc_for_all_writes_mask(inst, mark_write, writemasks);
}
- for(i = 0; i < R300_VS_MAX_TEMPS; i++) {
+ for(i = 0; i < compiler->Base.max_temp_regs; i++) {
unsigned int mask = ~writemasks[i] & RC_MASK_XYZW;
/* Only the W component can be used fo the predicate
* stack counter. */
break;
}
}
- if (i == R300_VS_MAX_TEMPS) {
+ if (i == compiler->Base.max_temp_regs) {
rc_error(&compiler->Base, "No free temporary to use for"
" predicate stack counter.\n");
return;
inst[3] = 0;
}
-static void translate_vertex_program(struct r300_vertex_program_compiler * compiler)
+static void translate_vertex_program(struct radeon_compiler *c, void *user)
{
+ struct r300_vertex_program_compiler *compiler = (struct r300_vertex_program_compiler*)c;
struct rc_instruction *rci;
struct loop * loops = NULL;
compiler->code->pos_end = 0; /* Not supported yet */
compiler->code->length = 0;
+ compiler->code->num_temporaries = 0;
compiler->SetHwInputOutput(compiler);
for(rci = compiler->Base.Program.Instructions.Next; rci != &compiler->Base.Program.Instructions; rci = rci->Next) {
struct rc_sub_instruction *vpi = &rci->U.I;
unsigned int *inst = compiler->code->body.d + compiler->code->length;
+ const struct rc_opcode_info *info = rc_get_opcode_info(vpi->Opcode);
/* Skip instructions writing to non-existing destination */
if (!valid_dst(compiler->code, &vpi->DstReg))
continue;
- if (rc_get_opcode_info(vpi->Opcode)->HasDstReg) {
+ if (info->HasDstReg) {
/* Relative addressing of destination operands is not supported yet. */
if (vpi->DstReg.RelAddr) {
rc_error(&compiler->Base, "Vertex program does not support relative "
}
}
- if (compiler->code->length >= R500_VS_MAX_ALU_DWORDS ||
- (compiler->code->length >= R300_VS_MAX_ALU_DWORDS && !compiler->Base.is_r500)) {
+ if (compiler->code->length >= c->max_alu_insts * 4) {
rc_error(&compiler->Base, "Vertex program has too many instructions\n");
return;
}
}
default:
- rc_error(&compiler->Base, "Unknown opcode %s\n", rc_get_opcode_info(vpi->Opcode)->Name);
+ rc_error(&compiler->Base, "Unknown opcode %s\n", info->Name);
return;
}
<< PVS_DST_PRED_SENSE_SHIFT);
}
+ /* Update the number of temporaries. */
+ if (info->HasDstReg && vpi->DstReg.File == RC_FILE_TEMPORARY &&
+ vpi->DstReg.Index >= compiler->code->num_temporaries)
+ compiler->code->num_temporaries = vpi->DstReg.Index + 1;
+
+ for (unsigned i = 0; i < info->NumSrcRegs; i++)
+ if (vpi->SrcReg[i].File == RC_FILE_TEMPORARY &&
+ vpi->SrcReg[i].Index >= compiler->code->num_temporaries)
+ compiler->code->num_temporaries = vpi->SrcReg[i].Index + 1;
+
+ if (compiler->PredicateMask)
+ if (compiler->PredicateIndex >= compiler->code->num_temporaries)
+ compiler->code->num_temporaries = compiler->PredicateIndex + 1;
+
+ if (compiler->code->num_temporaries > compiler->Base.max_temp_regs) {
+ rc_error(&compiler->Base, "Too many temporaries.\n");
+ return;
+ }
+
compiler->code->length += 4;
if (compiler->Base.Error)
struct rc_instruction * LastRead;
};
-static void allocate_temporary_registers(struct r300_vertex_program_compiler * compiler)
+static void allocate_temporary_registers(struct radeon_compiler *c, void *user)
{
+ struct r300_vertex_program_compiler *compiler = (struct r300_vertex_program_compiler*)c;
struct rc_instruction *inst;
struct rc_instruction *end_loop = NULL;
unsigned int num_orig_temps = 0;
- char hwtemps[R300_VS_MAX_TEMPS];
+ char hwtemps[RC_REGISTER_MAX_INDEX];
struct temporary_allocation * ta;
unsigned int i, j;
+ struct rc_instruction *last_inst_src_reladdr = NULL;
memset(hwtemps, 0, sizeof(hwtemps));
+ rc_recompute_ips(c);
+
/* Pass 1: Count original temporaries. */
for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
}
}
}
- compiler->code->num_temporaries = num_orig_temps;
- /* Pass 2: If there is relative addressing of temporaries, we cannot change register indices. Give up. */
+ /* Pass 2: If there is relative addressing of dst temporaries, we cannot change register indices. Give up.
+ * For src temporaries, save the last instruction which uses relative addressing. */
for (inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
const struct rc_opcode_info *opcode = rc_get_opcode_info(inst->U.I.Opcode);
for (i = 0; i < opcode->NumSrcRegs; ++i) {
if (inst->U.I.SrcReg[i].File == RC_FILE_TEMPORARY &&
inst->U.I.SrcReg[i].RelAddr) {
- return;
+ last_inst_src_reladdr = inst;
}
}
}
- compiler->code->num_temporaries = 0;
ta = (struct temporary_allocation*)memory_pool_malloc(&compiler->Base.Pool,
sizeof(struct temporary_allocation) * num_orig_temps);
memset(ta, 0, sizeof(struct temporary_allocation) * num_orig_temps);
}
for (i = 0; i < opcode->NumSrcRegs; ++i) {
- if (inst->U.I.SrcReg[i].File == RC_FILE_TEMPORARY)
- ta[inst->U.I.SrcReg[i].Index].LastRead =
- end_loop ? end_loop : inst;
+ if (inst->U.I.SrcReg[i].File == RC_FILE_TEMPORARY) {
+ struct rc_instruction *last_read;
+
+ /* From "last_inst_src_reladdr", "end_loop", and "inst",
+ * select the instruction with the highest instruction index (IP).
+ * Note that "end_loop", if available, has always a higher index than "inst". */
+ if (last_inst_src_reladdr) {
+ if (end_loop) {
+ last_read = last_inst_src_reladdr->IP > end_loop->IP ?
+ last_inst_src_reladdr : end_loop;
+ } else {
+ last_read = last_inst_src_reladdr->IP > inst->IP ?
+ last_inst_src_reladdr : inst;
+ }
+ } else {
+ last_read = end_loop ? end_loop : inst;
+ }
+
+ ta[inst->U.I.SrcReg[i].Index].LastRead = last_read;
+ }
}
}
for(inst = compiler->Base.Program.Instructions.Next; inst != &compiler->Base.Program.Instructions; inst = inst->Next) {
const struct rc_opcode_info * opcode = rc_get_opcode_info(inst->U.I.Opcode);
- for (i = 0; i < opcode->NumSrcRegs; ++i) {
- if (inst->U.I.SrcReg[i].File == RC_FILE_TEMPORARY) {
- unsigned int orig = inst->U.I.SrcReg[i].Index;
- inst->U.I.SrcReg[i].Index = ta[orig].HwTemp;
+ if (!last_inst_src_reladdr || last_inst_src_reladdr->IP < inst->IP) {
+ for (i = 0; i < opcode->NumSrcRegs; ++i) {
+ if (inst->U.I.SrcReg[i].File == RC_FILE_TEMPORARY) {
+ unsigned int orig = inst->U.I.SrcReg[i].Index;
+ inst->U.I.SrcReg[i].Index = ta[orig].HwTemp;
- if (ta[orig].Allocated && inst == ta[orig].LastRead)
- hwtemps[ta[orig].HwTemp] = 0;
+ if (ta[orig].Allocated && inst == ta[orig].LastRead)
+ hwtemps[ta[orig].HwTemp] = 0;
+ }
}
}
unsigned int orig = inst->U.I.DstReg.Index;
if (!ta[orig].Allocated) {
- for(j = 0; j < R300_VS_MAX_TEMPS; ++j) {
+ for(j = 0; j < c->max_temp_regs; ++j) {
if (!hwtemps[j])
break;
}
- if (j >= R300_VS_MAX_TEMPS) {
- fprintf(stderr, "Out of hw temporaries\n");
+ ta[orig].Allocated = 1;
+ if (last_inst_src_reladdr &&
+ last_inst_src_reladdr->IP > inst->IP) {
+ ta[orig].HwTemp = orig;
} else {
- ta[orig].Allocated = 1;
ta[orig].HwTemp = j;
- hwtemps[j] = 1;
-
- if (j >= compiler->code->num_temporaries)
- compiler->code->num_temporaries = j + 1;
}
+ hwtemps[ta[orig].HwTemp] = 1;
}
inst->U.I.DstReg.Index = ta[orig].HwTemp;
return 1;
}
-static void addArtificialOutputs(struct r300_vertex_program_compiler * compiler)
+static void rc_vs_add_artificial_outputs(struct radeon_compiler *c, void *user)
{
+ struct r300_vertex_program_compiler * compiler = (struct r300_vertex_program_compiler*)c;
int i;
for(i = 0; i < 32; ++i) {
}
}
-static void rc_emulate_negative_addressing(struct r300_vertex_program_compiler *c)
+static void rc_emulate_negative_addressing(struct radeon_compiler *compiler, void *user)
{
+ struct r300_vertex_program_compiler * c = (struct r300_vertex_program_compiler*)compiler;
struct rc_instruction *inst, *lastARL = NULL;
int min_offset = 0;
transform_negative_addressing(c, lastARL, inst, min_offset);
}
-static void debug_program_log(struct r300_vertex_program_compiler* c, const char * where)
-{
- if (c->Base.Debug) {
- fprintf(stderr, "Vertex Program: %s\n", where);
- rc_print_program(&c->Base.Program);
- }
-}
-
-
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 *c)
{
+ int is_r500 = c->Base.is_r500;
+ int kill_consts = c->Base.remove_unused_constants;
+ int opt = !c->Base.disable_optimizations;
+
+ /* Lists of instruction transformations. */
+ struct radeon_program_transformation alu_rewrite_r500[] = {
+ { &r300_transform_vertex_alu, 0 },
+ { &r300_transform_trig_scale_vertex, 0 },
+ { 0, 0 }
+ };
+
+ struct radeon_program_transformation alu_rewrite_r300[] = {
+ { &r300_transform_vertex_alu, 0 },
+ { &r300_transform_trig_simple, 0 },
+ { 0, 0 }
+ };
+
+ /* Note: These passes have to be done seperately from ALU rewrite,
+ * otherwise non-native ALU instructions with source conflits
+ * or non-native modifiers will not be treated properly.
+ */
+ struct radeon_program_transformation emulate_modifiers[] = {
+ { &transform_nonnative_modifiers, 0 },
+ { 0, 0 }
+ };
+
+ struct radeon_program_transformation resolve_src_conflicts[] = {
+ { &transform_source_conflicts, 0 },
+ { 0, 0 }
+ };
+
+ /* List of compiler passes. */
+ struct radeon_compiler_pass vs_list[] = {
+ /* NAME DUMP PREDICATE FUNCTION PARAM */
+ {"add artificial outputs", 0, 1, rc_vs_add_artificial_outputs, NULL},
+ {"transform loops", 1, 1, rc_transform_loops, NULL},
+ {"emulate branches", 1, !is_r500, rc_emulate_branches, NULL},
+ {"emulate negative addressing", 1, 1, rc_emulate_negative_addressing, NULL},
+ {"native rewrite", 1, is_r500, rc_local_transform, alu_rewrite_r500},
+ {"native rewrite", 1, !is_r500, rc_local_transform, alu_rewrite_r300},
+ {"emulate modifiers", 1, !is_r500, rc_local_transform, emulate_modifiers},
+ {"deadcode", 1, opt, rc_dataflow_deadcode, dataflow_outputs_mark_used},
+ {"dataflow optimize", 1, opt, rc_optimize, NULL},
+ /* This pass must be done after optimizations. */
+ {"source conflict resolve", 1, 1, rc_local_transform, resolve_src_conflicts},
+ {"register allocation", 1, opt, allocate_temporary_registers, NULL},
+ {"dead constants", 1, kill_consts, rc_remove_unused_constants, &c->code->constants_remap_table},
+ {"final code validation", 0, 1, rc_validate_final_shader, NULL},
+ {"machine code generation", 0, 1, translate_vertex_program, NULL},
+ {"dump machine code", 0, c->Base.Debug & RC_DBG_LOG, r300_vertex_program_dump, NULL},
+ {NULL, 0, 0, NULL, NULL}
+ };
+
+ c->Base.type = RC_VERTEX_PROGRAM;
c->Base.SwizzleCaps = &r300_vertprog_swizzle_caps;
- addArtificialOutputs(c);
-
- debug_program_log(c, "before compilation");
-
- if (c->Base.is_r500)
- rc_transform_loops(&c->Base, R500_VS_MAX_ALU);
- else
- rc_transform_loops(&c->Base, R300_VS_MAX_ALU);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after emulate loops");
-
- if (!c->Base.is_r500) {
- rc_emulate_branches(&c->Base);
- if (c->Base.Error)
- return;
- debug_program_log(c, "after emulate branches");
- }
-
- rc_emulate_negative_addressing(c);
-
- debug_program_log(c, "after negative addressing emulation");
-
- if (c->Base.is_r500) {
- struct radeon_program_transformation transformations[] = {
- { &r300_transform_vertex_alu, 0 },
- { &r300_transform_trig_scale_vertex, 0 }
- };
- radeonLocalTransform(&c->Base, 2, transformations);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after native rewrite");
- } else {
- struct radeon_program_transformation transformations[] = {
- { &r300_transform_vertex_alu, 0 },
- { &radeonTransformTrigSimple, 0 }
- };
- radeonLocalTransform(&c->Base, 2, transformations);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after native rewrite");
-
- /* Note: This pass has to be done seperately from ALU rewrite,
- * because it needs to check every instruction.
- */
- struct radeon_program_transformation transformations2[] = {
- { &transform_nonnative_modifiers, 0 },
- };
- radeonLocalTransform(&c->Base, 1, transformations2);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after emulate modifiers");
- }
-
- {
- /* Note: This pass has to be done seperately from ALU rewrite,
- * otherwise non-native ALU instructions with source conflits
- * will not be treated properly.
- */
- struct radeon_program_transformation transformations[] = {
- { &transform_source_conflicts, 0 },
- };
- radeonLocalTransform(&c->Base, 1, transformations);
- if (c->Base.Error)
- return;
- }
-
- debug_program_log(c, "after source conflict resolve");
-
- rc_dataflow_deadcode(&c->Base, &dataflow_outputs_mark_used, c);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after deadcode");
-
- rc_dataflow_swizzles(&c->Base);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after dataflow");
-
- allocate_temporary_registers(c);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after register allocation");
-
- if (c->Base.remove_unused_constants) {
- rc_remove_unused_constants(&c->Base,
- &c->code->constants_remap_table);
- if (c->Base.Error)
- return;
-
- debug_program_log(c, "after constants cleanup");
- }
-
- translate_vertex_program(c);
- if (c->Base.Error)
- return;
-
- rc_constants_copy(&c->code->constants, &c->Base.Program.Constants);
+ rc_run_compiler(&c->Base, vs_list);
c->code->InputsRead = c->Base.Program.InputsRead;
c->code->OutputsWritten = c->Base.Program.OutputsWritten;
-
- if (c->Base.Debug) {
- fprintf(stderr, "Final vertex program code:\n");
- r300_vertex_program_dump(c);
- }
-
- /* Check the number of constants. */
- if (!c->Base.Error &&
- c->Base.Program.Constants.Count > 256) {
- rc_error(&c->Base, "Too many constants. Max: 256, Got: %i\n",
- c->Base.Program.Constants.Count);
- }
+ rc_constants_copy(&c->code->constants, &c->Base.Program.Constants);
}
projects / mesa.git / blobdiff