#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_dataflow.h"
+#include "radeon_emulate_branches.h"
#include "radeon_program_alu.h"
+#include "radeon_program_tex.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[0], RC_MASK_XYZW);
+ callback(data, c->OutputColor[1], RC_MASK_XYZW);
+ callback(data, c->OutputColor[2], RC_MASK_XYZW);
+ callback(data, c->OutputColor[3], RC_MASK_XYZW);
+ callback(data, c->OutputDepth, RC_MASK_W);
}
static void rewrite_depth_out(struct r300_fragment_program_compiler * c)
struct rc_instruction *rci;
for (rci = c->Base.Program.Instructions.Next; rci != &c->Base.Program.Instructions; rci = rci->Next) {
- struct prog_instruction * inst = &rci->I;
+ struct rc_sub_instruction * inst = &rci->U.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
}
}
+static void debug_program_log(struct r300_fragment_program_compiler* c, const char * where)
+{
+ if (c->Base.Debug) {
+ fprintf(stderr, "Fragment Program: %s\n", where);
+ rc_print_program(&c->Base.Program);
+ }
+}
+
void r3xx_compile_fragment_program(struct r300_fragment_program_compiler* c)
{
rewrite_depth_out(c);
+ debug_program_log(c, "before compilation");
+
+ /* XXX Ideally this should be done only for r3xx, but since
+ * we don't have branching support for r5xx, we use the emulation
+ * on all chipsets. */
+ rc_emulate_branches(&c->Base);
+
+ debug_program_log(c, "after emulate branches");
+
if (c->is_r500) {
struct radeon_program_transformation transformations[] = {
- { &r500_transform_TEX, c },
+ { &r500_transform_IF, 0 },
{ &radeonTransformALU, 0 },
{ &radeonTransformDeriv, 0 },
{ &radeonTransformTrigScale, 0 }
};
radeonLocalTransform(&c->Base, 4, transformations);
+
+ debug_program_log(c, "after native rewrite part 1");
+
+ c->Base.SwizzleCaps = &r500_swizzle_caps;
} else {
struct radeon_program_transformation transformations[] = {
- { &r300_transform_TEX, c },
{ &radeonTransformALU, 0 },
{ &radeonTransformTrigSimple, 0 }
};
- radeonLocalTransform(&c->Base, 3, transformations);
- }
+ radeonLocalTransform(&c->Base, 2, transformations);
- if (c->Base.Debug) {
- _mesa_printf("Fragment Program: After native rewrite:\n");
- rc_print_program(&c->Base.Program);
- fflush(stdout);
- }
+ debug_program_log(c, "after native rewrite part 1");
- 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);
+ c->Base.SwizzleCaps = &r300_swizzle_caps;
}
- if (c->Base.Debug) {
- _mesa_printf("Compiler: after NqSSA-DCE:\n");
- rc_print_program(&c->Base.Program);
- fflush(stdout);
- }
+ /* Run the common transformations too.
+ * Remember, lowering comes last! */
+ struct radeon_program_transformation common_transformations[] = {
+ { &radeonTransformTEX, c },
+ };
+ radeonLocalTransform(&c->Base, 1, common_transformations);
+
+ common_transformations[0].function = &radeonTransformALU;
+ radeonLocalTransform(&c->Base, 1, common_transformations);
+
+ if (c->Base.Error)
+ return;
+
+ debug_program_log(c, "after native rewrite part 2");
+
+ rc_dataflow_deadcode(&c->Base, &dataflow_outputs_mark_use, 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 passes");
+
+ rc_pair_translate(c);
+ if (c->Base.Error)
+ return;
+
+ debug_program_log(c, "after pair translate");
+
+ rc_pair_schedule(c);
+ if (c->Base.Error)
+ return;
+
+ debug_program_log(c, "after pair scheduling");
+
+ rc_pair_regalloc(c, c->max_temp_regs);
+
+ if (c->Base.Error)
+ return;
+
+ debug_program_log(c, "after register allocation");
if (c->is_r500) {
r500BuildFragmentProgramHwCode(c);
projects / mesa.git / blobdiff