#include "r500_fragprog.h"
+static void reset_srcreg(struct prog_src_register* reg)
+{
+ _mesa_bzero(reg, sizeof(*reg));
+ reg->Swizzle = SWIZZLE_NOOP;
+}
+
+static struct prog_src_register shadow_ambient(struct gl_program *program, int tmu)
+{
+ gl_state_index fail_value_tokens[STATE_LENGTH] = {
+ STATE_INTERNAL, STATE_SHADOW_AMBIENT, 0, 0, 0
+ };
+ struct prog_src_register reg = { 0, };
+
+ fail_value_tokens[2] = tmu;
+ reg.File = PROGRAM_STATE_VAR;
+ reg.Index = _mesa_add_state_reference(program->Parameters, fail_value_tokens);
+ reg.Swizzle = SWIZZLE_WWWW;
+ return reg;
+}
/**
* Transform TEX, TXP, TXB, and KIL instructions in the following way:
* - introduce a temporary register when write masks are needed
*
*/
-static GLboolean transform_TEX(
- GLcontext *ctx, struct gl_program *p,
+GLboolean r500_transform_TEX(
+ struct radeon_transform_context *t,
struct prog_instruction* orig_inst, void* data)
{
- struct r500_fragment_program_compiler *compiler =
- (struct r500_fragment_program_compiler*)data;
+ struct r300_fragment_program_compiler *compiler =
+ (struct r300_fragment_program_compiler*)data;
struct prog_instruction inst = *orig_inst;
struct prog_instruction* tgt;
GLboolean destredirect = GL_FALSE;
/* ARB_shadow & EXT_shadow_funcs */
if (inst.Opcode != OPCODE_KIL &&
- p->ShadowSamplers & (1 << inst.TexSrcUnit)) {
+ t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func;
if (comparefunc == GL_NEVER || comparefunc == GL_ALWAYS) {
- tgt = radeonAppendInstructions(p, 1);
+ tgt = radeonAppendInstructions(t->Program, 1);
tgt->Opcode = OPCODE_MOV;
- tgt->DstReg.File = inst.DstReg.File;
- tgt->DstReg.Index = inst.DstReg.Index;
- tgt->DstReg.WriteMask = inst.DstReg.WriteMask;
- tgt->SrcReg[0].File = PROGRAM_BUILTIN;
- tgt->SrcReg[0].Swizzle = comparefunc == GL_ALWAYS ? SWIZZLE_1111 : SWIZZLE_0000;
+ tgt->DstReg = inst.DstReg;
+ if (comparefunc == GL_ALWAYS) {
+ tgt->SrcReg[0].File = PROGRAM_BUILTIN;
+ tgt->SrcReg[0].Swizzle = SWIZZLE_1111;
+ } else {
+ tgt->SrcReg[0] = shadow_ambient(t->Program, inst.TexSrcUnit);
+ }
return GL_TRUE;
}
inst.DstReg.File = PROGRAM_TEMPORARY;
- inst.DstReg.Index = _mesa_find_free_register(p, PROGRAM_TEMPORARY);
+ inst.DstReg.Index = radeonFindFreeTemporary(t);
inst.DstReg.WriteMask = WRITEMASK_XYZW;
+ } else if (inst.Opcode != OPCODE_KIL && inst.DstReg.File != PROGRAM_TEMPORARY) {
+ int tempreg = radeonFindFreeTemporary(t);
+
+ inst.DstReg.File = PROGRAM_TEMPORARY;
+ inst.DstReg.Index = tempreg;
+ inst.DstReg.WriteMask = WRITEMASK_XYZW;
+ destredirect = GL_TRUE;
+ }
+
+ if (inst.SrcReg[0].File != PROGRAM_TEMPORARY && inst.SrcReg[0].File != PROGRAM_INPUT) {
+ int tmpreg = radeonFindFreeTemporary(t);
+ tgt = radeonAppendInstructions(t->Program, 1);
+ tgt->Opcode = OPCODE_MOV;
+ tgt->DstReg.File = PROGRAM_TEMPORARY;
+ tgt->DstReg.Index = tmpreg;
+ tgt->SrcReg[0] = inst.SrcReg[0];
+
+ reset_srcreg(&inst.SrcReg[0]);
+ inst.SrcReg[0].File = PROGRAM_TEMPORARY;
+ inst.SrcReg[0].Index = tmpreg;
}
- tgt = radeonAppendInstructions(p, 1);
+ tgt = radeonAppendInstructions(t->Program, 1);
_mesa_copy_instructions(tgt, &inst, 1);
if (inst.Opcode != OPCODE_KIL &&
- p->ShadowSamplers & (1 << inst.TexSrcUnit)) {
+ t->Program->ShadowSamplers & (1 << inst.TexSrcUnit)) {
GLuint comparefunc = GL_NEVER + compiler->fp->state.unit[inst.TexSrcUnit].texture_compare_func;
GLuint depthmode = compiler->fp->state.unit[inst.TexSrcUnit].depth_texture_mode;
- int rcptemp = _mesa_find_free_register(p, PROGRAM_TEMPORARY);
+ int rcptemp = radeonFindFreeTemporary(t);
+ int pass, fail;
- tgt = radeonAppendInstructions(p, 3);
+ tgt = radeonAppendInstructions(t->Program, 3);
tgt[0].Opcode = OPCODE_RCP;
tgt[0].DstReg.File = PROGRAM_TEMPORARY;
* r < tex <=> -tex+r < 0
* r >= tex <=> not (-tex+r < 0 */
if (comparefunc == GL_LESS || comparefunc == GL_GEQUAL)
- tgt[1].SrcReg[2].NegateBase = tgt[0].SrcReg[2].NegateBase ^ NEGATE_XYZW;
+ tgt[1].SrcReg[2].Negate = tgt[0].SrcReg[2].Negate ^ NEGATE_XYZW;
else
- tgt[1].SrcReg[0].NegateBase = tgt[0].SrcReg[0].NegateBase ^ NEGATE_XYZW;
+ tgt[1].SrcReg[0].Negate = tgt[0].SrcReg[0].Negate ^ NEGATE_XYZW;
tgt[2].Opcode = OPCODE_CMP;
tgt[2].DstReg = orig_inst->DstReg;
tgt[2].SrcReg[0].File = PROGRAM_TEMPORARY;
tgt[2].SrcReg[0].Index = tgt[1].DstReg.Index;
- tgt[2].SrcReg[1].File = PROGRAM_BUILTIN;
- tgt[2].SrcReg[2].File = PROGRAM_BUILTIN;
if (comparefunc == GL_LESS || comparefunc == GL_GREATER) {
- tgt[2].SrcReg[1].Swizzle = SWIZZLE_1111;
- tgt[2].SrcReg[2].Swizzle = SWIZZLE_0000;
+ pass = 1;
+ fail = 2;
} else {
- tgt[2].SrcReg[1].Swizzle = SWIZZLE_0000;
- tgt[2].SrcReg[2].Swizzle = SWIZZLE_1111;
+ pass = 2;
+ fail = 1;
}
+
+ tgt[2].SrcReg[pass].File = PROGRAM_BUILTIN;
+ tgt[2].SrcReg[pass].Swizzle = SWIZZLE_1111;
+ tgt[2].SrcReg[fail] = shadow_ambient(t->Program, inst.TexSrcUnit);
} else if (destredirect) {
- tgt = radeonAppendInstructions(p, 1);
+ tgt = radeonAppendInstructions(t->Program, 1);
tgt->Opcode = OPCODE_MOV;
tgt->DstReg = orig_inst->DstReg;
return GL_TRUE;
}
-
-static void update_params(r300ContextPtr r300, struct r500_fragment_program *fp)
+GLboolean r500FPIsNativeSwizzle(GLuint opcode, struct prog_src_register reg)
{
- struct gl_fragment_program *mp = &fp->mesa_program;
-
- /* Ask Mesa nicely to fill in ParameterValues for us */
- if (mp->Base.Parameters)
- _mesa_load_state_parameters(r300->radeon.glCtx, mp->Base.Parameters);
-}
+ GLuint relevant;
+ int i;
+
+ if (opcode == OPCODE_TEX ||
+ opcode == OPCODE_TXB ||
+ opcode == OPCODE_TXP ||
+ opcode == OPCODE_KIL) {
+ if (reg.Abs)
+ return GL_FALSE;
+
+ if (reg.Negate)
+ reg.Negate ^= NEGATE_XYZW;
+
+ if (opcode == OPCODE_KIL) {
+ if (reg.Swizzle != SWIZZLE_NOOP)
+ return GL_FALSE;
+ } else {
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(reg.Swizzle, i);
+ if (swz == SWIZZLE_NIL) {
+ reg.Negate &= ~(1 << i);
+ continue;
+ }
+ if (swz >= 4)
+ return GL_FALSE;
+ }
+ }
+ if (reg.Negate)
+ return GL_FALSE;
-/**
- * Transform the program to support fragment.position.
- *
- * Introduce a small fragment at the start of the program that will be
- * the only code that directly reads the FRAG_ATTRIB_WPOS input.
- * All other code pieces that reference that input will be rewritten
- * to read from a newly allocated temporary.
- *
- * \todo if/when r5xx supports the radeon_program architecture, this is a
- * likely candidate for code sharing.
- */
-static void insert_WPOS_trailer(struct r500_fragment_program_compiler *compiler)
-{
- GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead;
+ return GL_TRUE;
+ } else if (opcode == OPCODE_DDX || opcode == OPCODE_DDY) {
+ /* DDX/MDH and DDY/MDV explicitly ignore incoming swizzles;
+ * if it doesn't fit perfectly into a .xyzw case... */
+ if (reg.Swizzle == SWIZZLE_NOOP && !reg.Abs && !reg.Negate)
+ return GL_TRUE;
- if (!(InputsRead & FRAG_BIT_WPOS))
- return;
+ return GL_FALSE;
+ } else {
+ /* ALU instructions support almost everything */
+ if (reg.Abs)
+ return GL_TRUE;
- static gl_state_index tokens[STATE_LENGTH] = {
- STATE_INTERNAL, STATE_R300_WINDOW_DIMENSION, 0, 0, 0
- };
- struct prog_instruction *fpi;
- GLuint window_index;
- int i = 0;
- GLuint tempregi = _mesa_find_free_register(compiler->program, PROGRAM_TEMPORARY);
-
- _mesa_insert_instructions(compiler->program, 0, 3);
- fpi = compiler->program->Instructions;
-
- /* perspective divide */
- fpi[i].Opcode = OPCODE_RCP;
-
- fpi[i].DstReg.File = PROGRAM_TEMPORARY;
- fpi[i].DstReg.Index = tempregi;
- fpi[i].DstReg.WriteMask = WRITEMASK_W;
- fpi[i].DstReg.CondMask = COND_TR;
-
- fpi[i].SrcReg[0].File = PROGRAM_INPUT;
- fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
- fpi[i].SrcReg[0].Swizzle = SWIZZLE_WWWW;
- i++;
-
- fpi[i].Opcode = OPCODE_MUL;
-
- fpi[i].DstReg.File = PROGRAM_TEMPORARY;
- fpi[i].DstReg.Index = tempregi;
- fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
- fpi[i].DstReg.CondMask = COND_TR;
-
- fpi[i].SrcReg[0].File = PROGRAM_INPUT;
- fpi[i].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
- fpi[i].SrcReg[0].Swizzle = SWIZZLE_XYZW;
-
- fpi[i].SrcReg[1].File = PROGRAM_TEMPORARY;
- fpi[i].SrcReg[1].Index = tempregi;
- fpi[i].SrcReg[1].Swizzle = SWIZZLE_WWWW;
- i++;
-
- /* viewport transformation */
- window_index = _mesa_add_state_reference(compiler->fp->mesa_program.Base.Parameters, tokens);
-
- fpi[i].Opcode = OPCODE_MAD;
-
- fpi[i].DstReg.File = PROGRAM_TEMPORARY;
- fpi[i].DstReg.Index = tempregi;
- fpi[i].DstReg.WriteMask = WRITEMASK_XYZ;
- fpi[i].DstReg.CondMask = COND_TR;
-
- fpi[i].SrcReg[0].File = PROGRAM_TEMPORARY;
- fpi[i].SrcReg[0].Index = tempregi;
- fpi[i].SrcReg[0].Swizzle =
- MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
- fpi[i].SrcReg[1].File = PROGRAM_STATE_VAR;
- fpi[i].SrcReg[1].Index = window_index;
- fpi[i].SrcReg[1].Swizzle =
- MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
-
- fpi[i].SrcReg[2].File = PROGRAM_STATE_VAR;
- fpi[i].SrcReg[2].Index = window_index;
- fpi[i].SrcReg[2].Swizzle =
- MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ZERO);
- i++;
-
- for (; i < compiler->program->NumInstructions; ++i) {
- int reg;
- for (reg = 0; reg < 3; reg++) {
- if (fpi[i].SrcReg[reg].File == PROGRAM_INPUT &&
- fpi[i].SrcReg[reg].Index == FRAG_ATTRIB_WPOS) {
- fpi[i].SrcReg[reg].File = PROGRAM_TEMPORARY;
- fpi[i].SrcReg[reg].Index = tempregi;
- }
+ relevant = 0;
+ for(i = 0; i < 3; ++i) {
+ GLuint swz = GET_SWZ(reg.Swizzle, i);
+ if (swz != SWIZZLE_NIL && swz != SWIZZLE_ZERO)
+ relevant |= 1 << i;
}
- }
-}
-
+ if ((reg.Negate & relevant) && ((reg.Negate & relevant) != relevant))
+ return GL_FALSE;
-static GLuint build_dtm(GLuint depthmode)
-{
- switch(depthmode) {
- default:
- case GL_LUMINANCE: return 0;
- case GL_INTENSITY: return 1;
- case GL_ALPHA: return 2;
+ return GL_TRUE;
}
}
-static GLuint build_func(GLuint comparefunc)
-{
- return comparefunc - GL_NEVER;
-}
-
-
/**
- * Collect all external state that is relevant for compiling the given
- * fragment program.
+ * Implement a MOV with a potentially non-native swizzle.
+ *
+ * The only thing we *cannot* do in an ALU instruction is per-component
+ * negation. Therefore, we split the MOV into two instructions when necessary.
*/
-static void build_state(
- r300ContextPtr r300,
- struct r500_fragment_program *fp,
- struct r500_fragment_program_external_state *state)
+void r500FPBuildSwizzle(struct nqssadce_state *s, struct prog_dst_register dst, struct prog_src_register src)
{
- int unit;
-
- _mesa_bzero(state, sizeof(*state));
-
- for(unit = 0; unit < 16; ++unit) {
- if (fp->mesa_program.Base.ShadowSamplers & (1 << unit)) {
- struct gl_texture_object* tex = r300->radeon.glCtx->Texture.Unit[unit]._Current;
-
- state->unit[unit].depth_texture_mode = build_dtm(tex->DepthMode);
- state->unit[unit].texture_compare_func = build_func(tex->CompareFunc);
- }
+ struct prog_instruction *inst;
+ GLuint negatebase[2] = { 0, 0 };
+ int i;
+
+ for(i = 0; i < 4; ++i) {
+ GLuint swz = GET_SWZ(src.Swizzle, i);
+ if (swz == SWIZZLE_NIL)
+ continue;
+ negatebase[GET_BIT(src.Negate, i)] |= 1 << i;
}
-}
-static void dump_program(struct r500_fragment_program_code *code);
+ _mesa_insert_instructions(s->Program, s->IP, (negatebase[0] ? 1 : 0) + (negatebase[1] ? 1 : 0));
+ inst = s->Program->Instructions + s->IP;
-void r500TranslateFragmentShader(r300ContextPtr r300,
- struct r500_fragment_program *fp)
-{
- struct r500_fragment_program_external_state state;
+ for(i = 0; i <= 1; ++i) {
+ if (!negatebase[i])
+ continue;
- build_state(r300, fp, &state);
- if (_mesa_memcmp(&fp->state, &state, sizeof(state))) {
- /* TODO: cache compiled programs */
- fp->translated = GL_FALSE;
- _mesa_memcpy(&fp->state, &state, sizeof(state));
+ inst->Opcode = OPCODE_MOV;
+ inst->DstReg = dst;
+ inst->DstReg.WriteMask = negatebase[i];
+ inst->SrcReg[0] = src;
+ inst++;
+ s->IP++;
}
-
- if (!fp->translated) {
- struct r500_fragment_program_compiler compiler;
-
- compiler.r300 = r300;
- compiler.fp = fp;
- compiler.code = &fp->code;
- compiler.program = _mesa_clone_program(r300->radeon.glCtx, &fp->mesa_program.Base);
-
- if (RADEON_DEBUG & DEBUG_PIXEL) {
- _mesa_printf("Compiler: Initial program:\n");
- _mesa_print_program(compiler.program);
- }
-
- insert_WPOS_trailer(&compiler);
-
- struct radeon_program_transformation transformations[1] = {
- { &transform_TEX, &compiler }
- };
- radeonLocalTransform(r300->radeon.glCtx, compiler.program,
- 1, transformations);
-
- if (RADEON_DEBUG & DEBUG_PIXEL) {
- _mesa_printf("Compiler: after all transformations:\n");
- _mesa_print_program(compiler.program);
- }
-
- fp->translated = r500FragmentProgramEmit(&compiler);
-
- _mesa_reference_program(r300->radeon.glCtx, &compiler.program, 0);
-
- r300UpdateStateParameters(r300->radeon.glCtx, _NEW_PROGRAM);
-
- if (RADEON_DEBUG & DEBUG_PIXEL) {
- fprintf(stderr, "Mesa program:\n");
- fprintf(stderr, "-------------\n");
- _mesa_print_program(&fp->mesa_program.Base);
- fflush(stdout);
- if (fp->translated)
- dump_program(&fp->code);
- }
-
- }
-
- update_params(r300, fp);
-
}
+
static char *toswiz(int swiz_val) {
switch(swiz_val) {
case 0: return "R";
return NULL;
}
-static void dump_program(struct r500_fragment_program_code *code)
+void r500FragmentProgramDump(union rX00_fragment_program_code *c)
{
-
+ struct r500_fragment_program_code *code = &c->r500;
fprintf(stderr, "R500 Fragment Program:\n--------\n");
int n;
if (code->const_nr) {
fprintf(stderr, "--------\nConstants:\n");
for (n = 0; n < code->const_nr; n++) {
- fprintf(stderr, "Constant %d: %f %f\n\t %f %f\n", n,
- code->constant[n][0], code->constant[n][1], code->constant[n][2],
- code->constant[n][3]);
+ fprintf(stderr, "Constant %d: %i[%i]\n", n,
+ code->constant[n].File, code->constant[n].Index);
}
fprintf(stderr, "--------\n");
}