summaryrefslogtreecommitdiff
path: root/src/mesa
diff options
context:
space:
mode:
authorNicolai Hähnle <nhaehnle@gmail.com>2009-07-23 21:10:37 +0200
committerNicolai Hähnle <nhaehnle@gmail.com>2009-07-27 20:32:06 +0200
commitc5cb9a337881229f1db462632fa1d64e2677f316 (patch)
tree2d9c086d5695d1d7ee35d501960fa3560fb30d1d /src/mesa
parentd65404225d8ba2c16eaffac833cb7dcfd2722a38 (diff)
r300: Remove dependency on GLcontext from compiler
Unfortunately, this does cause some code duplication (which we can hopefully eliminate eventually). Signed-off-by: Nicolai Hähnle <nhaehnle@gmail.com>
Diffstat (limited to 'src/mesa')
-rw-r--r--src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c131
-rw-r--r--src/mesa/drivers/dri/r300/compiler/radeon_compiler.h2
-rw-r--r--src/mesa/drivers/dri/r300/r300_vertprog.c2
3 files changed, 98 insertions, 37 deletions
diff --git a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
index 400408620e..a0e081fe6f 100644
--- a/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
+++ b/src/mesa/drivers/dri/r300/compiler/r3xx_vertprog.c
@@ -28,7 +28,6 @@
#include "radeon_program.h"
#include "radeon_program_alu.h"
-#include "shader/prog_optimize.h"
#include "shader/prog_print.h"
@@ -438,6 +437,100 @@ static GLboolean translate_vertex_program(struct r300_vertex_program_compiler *
return GL_TRUE;
}
+struct temporary_allocation {
+ GLuint Allocated:1;
+ GLuint HwTemp:15;
+ struct prog_instruction * LastRead;
+};
+
+static void allocate_temporary_registers(struct r300_vertex_program_compiler * compiler)
+{
+ struct prog_instruction *inst;
+ GLuint num_orig_temps = 0;
+ GLboolean hwtemps[VSF_MAX_FRAGMENT_TEMPS];
+ struct temporary_allocation * ta;
+ GLuint i, j;
+
+ compiler->code->num_temporaries = 0;
+ memset(hwtemps, 0, sizeof(hwtemps));
+
+ /* Pass 1: Count original temporaries and allocate structures */
+ for(inst = compiler->program->Instructions; inst->Opcode != OPCODE_END; inst++) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->Opcode);
+ GLuint numdsts = _mesa_num_inst_dst_regs(inst->Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->SrcReg[i].File == PROGRAM_TEMPORARY) {
+ if (inst->SrcReg[i].Index >= num_orig_temps)
+ num_orig_temps = inst->SrcReg[i].Index + 1;
+ }
+ }
+
+ if (numdsts) {
+ if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+ if (inst->DstReg.Index >= num_orig_temps)
+ num_orig_temps = inst->DstReg.Index + 1;
+ }
+ }
+ }
+
+ ta = (struct temporary_allocation*)memory_pool_malloc(&compiler->Base.Pool,
+ sizeof(struct temporary_allocation) * num_orig_temps);
+ memset(ta, 0, sizeof(struct temporary_allocation) * num_orig_temps);
+
+ /* Pass 2: Determine original temporary lifetimes */
+ for(inst = compiler->program->Instructions; inst->Opcode != OPCODE_END; inst++) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->SrcReg[i].File == PROGRAM_TEMPORARY)
+ ta[inst->SrcReg[i].Index].LastRead = inst;
+ }
+ }
+
+ /* Pass 3: Register allocation */
+ for(inst = compiler->program->Instructions; inst->Opcode != OPCODE_END; inst++) {
+ GLuint numsrcs = _mesa_num_inst_src_regs(inst->Opcode);
+ GLuint numdsts = _mesa_num_inst_dst_regs(inst->Opcode);
+
+ for (i = 0; i < numsrcs; ++i) {
+ if (inst->SrcReg[i].File == PROGRAM_TEMPORARY) {
+ GLuint orig = inst->SrcReg[i].Index;
+ inst->SrcReg[i].Index = ta[orig].HwTemp;
+
+ if (ta[orig].Allocated && inst == ta[orig].LastRead)
+ hwtemps[ta[orig].HwTemp] = GL_FALSE;
+ }
+ }
+
+ if (numdsts) {
+ if (inst->DstReg.File == PROGRAM_TEMPORARY) {
+ GLuint orig = inst->DstReg.Index;
+
+ if (!ta[orig].Allocated) {
+ for(j = 0; j < VSF_MAX_FRAGMENT_TEMPS; ++j) {
+ if (!hwtemps[j])
+ break;
+ }
+ if (j >= VSF_MAX_FRAGMENT_TEMPS) {
+ fprintf(stderr, "Out of hw temporaries\n");
+ } else {
+ ta[orig].Allocated = GL_TRUE;
+ ta[orig].HwTemp = j;
+ hwtemps[j] = GL_TRUE;
+
+ if (j >= compiler->code->num_temporaries)
+ compiler->code->num_temporaries = j + 1;
+ }
+ }
+
+ inst->DstReg.Index = ta[orig].HwTemp;
+ }
+ }
+ }
+}
+
+
/**
* Vertex engine cannot read two inputs or two constants at the same time.
* Introduce intermediate MOVs to temporary registers to account for this.
@@ -675,7 +768,7 @@ static GLboolean swizzleIsNative(GLuint opcode, struct prog_src_register reg)
-GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler, GLcontext * ctx)
+GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compiler)
{
GLboolean success;
@@ -728,7 +821,7 @@ GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compi
radeonNqssaDce(compiler->program, &nqssadce, compiler);
/* We need this step for reusing temporary registers */
- _mesa_optimize_program(ctx, compiler->program);
+ allocate_temporary_registers(compiler);
if (compiler->Base.Debug) {
fprintf(stderr, "Vertex program after NQSSADCE:\n");
@@ -738,38 +831,6 @@ GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* compi
}
assert(compiler->program->NumInstructions);
- {
- struct prog_instruction *inst;
- int max, i, tmp;
-
- inst = compiler->program->Instructions;
- max = -1;
- while (inst->Opcode != OPCODE_END) {
- tmp = _mesa_num_inst_src_regs(inst->Opcode);
- for (i = 0; i < tmp; ++i) {
- if (inst->SrcReg[i].File == PROGRAM_TEMPORARY) {
- if ((int) inst->SrcReg[i].Index > max) {
- max = inst->SrcReg[i].Index;
- }
- }
- }
-
- if (_mesa_num_inst_dst_regs(inst->Opcode)) {
- if (inst->DstReg.File == PROGRAM_TEMPORARY) {
- if ((int) inst->DstReg.Index > max) {
- max = inst->DstReg.Index;
- }
- }
- }
- ++inst;
- }
-
- /* We actually want highest index of used temporary register,
- * not the number of temporaries used.
- * These values aren't always the same.
- */
- compiler->code->num_temporaries = max + 1;
- }
success = translate_vertex_program(compiler);
diff --git a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
index f8e4b3c681..b98b1c9e6b 100644
--- a/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
+++ b/src/mesa/drivers/dri/r300/compiler/radeon_compiler.h
@@ -74,6 +74,6 @@ struct r300_vertex_program_compiler {
struct gl_program *program;
};
-GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* c, GLcontext * ctx);
+GLboolean r3xx_compile_vertex_program(struct r300_vertex_program_compiler* c);
#endif /* RADEON_COMPILER_H */
diff --git a/src/mesa/drivers/dri/r300/r300_vertprog.c b/src/mesa/drivers/dri/r300/r300_vertprog.c
index ec4ba9ca7d..dfb2a9e3d8 100644
--- a/src/mesa/drivers/dri/r300/r300_vertprog.c
+++ b/src/mesa/drivers/dri/r300/r300_vertprog.c
@@ -125,7 +125,7 @@ static struct r300_vertex_program *build_program(GLcontext *ctx,
_mesa_insert_mvp_code(ctx, (struct gl_vertex_program *)compiler.program);
}
- if (!r3xx_compile_vertex_program(&compiler, ctx))
+ if (!r3xx_compile_vertex_program(&compiler))
vp->error = GL_TRUE;
rc_destroy(&compiler.Base);