static void unalias_temporary(struct nqssadce_state* s, GLuint oldindex)
{
GLuint newindex = _mesa_find_free_register(s->Program, PROGRAM_TEMPORARY);
int ip;
for(ip = 0; ip < s->IP; ++ip) {
struct prog_instruction* inst = s->Program->Instructions + ip;
if (inst->DstReg.File == PROGRAM_TEMPORARY && inst->DstReg.Index == oldindex)
inst->DstReg.Index = newindex;
unalias_srcregs(inst, oldindex, newindex);
}
unalias_srcregs(s->Program->Instructions + s->IP, oldindex, newindex);
}
static struct prog_instruction* track_used_srcreg(struct nqssadce_state* s,
struct prog_instruction *inst, GLint src, GLuint sourced)
{
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 = _mesa_find_free_register(s->Program, PROGRAM_TEMPORARY);
dstreg.WriteMask = sourced;
s->Descr->BuildSwizzle(s, dstreg, inst->SrcReg[src]);
inst = s->Program->Instructions + s->IP;
inst->SrcReg[src].File = PROGRAM_TEMPORARY;
inst->SrcReg[src].Index = dstreg.Index;
inst->SrcReg[src].Swizzle = 0;
inst->SrcReg[src].NegateBase = 0;
inst->SrcReg[src].Abs = 0;
inst->SrcReg[src].NegateAbs = 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 = get_reg_state(s, inst->SrcReg[src].File, inst->SrcReg[src].Index);
if (regstate)
regstate->Sourced |= deswz_source & 0xf;
return inst;
}
/**
* 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 r300_fragment_program_compiler *compiler)
{
GLuint InputsRead = compiler->fp->mesa_program.Base.InputsRead;
if (!(InputsRead & FRAG_BIT_WPOS))
return;
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->program->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;
}
}
}
}
/**
* Combine two programs into one. Fix instructions so the outputs of
* the first program go to the inputs of the second program.
*/
struct gl_program *
_mesa_combine_programs(GLcontext *ctx,
const struct gl_program *progA,
const struct gl_program *progB)
{
struct prog_instruction *newInst;
struct gl_program *newProg;
const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
const GLuint lenB = progB->NumInstructions;
const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
const GLuint newLength = lenA + lenB;
GLbitfield inputsB;
GLuint i;
ASSERT(progA->Target == progB->Target);
newInst = _mesa_alloc_instructions(newLength);
if (!newInst)
return GL_FALSE;
_mesa_copy_instructions(newInst, progA->Instructions, lenA);
_mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
/* adjust branch / instruction addresses for B's instructions */
for (i = 0; i < lenB; i++) {
newInst[lenA + i].BranchTarget += lenA;
}
newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
newProg->Instructions = newInst;
newProg->NumInstructions = newLength;
if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
struct gl_fragment_program *fprogA, *fprogB, *newFprog;
fprogA = (struct gl_fragment_program *) progA;
fprogB = (struct gl_fragment_program *) progB;
newFprog = (struct gl_fragment_program *) newProg;
newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
/* Connect color outputs of fprogA to color inputs of fprogB, via a
* new temporary register.
*/
if ((progA->OutputsWritten & (1 << FRAG_RESULT_COLR)) &&
(progB->InputsRead & (1 << FRAG_ATTRIB_COL0))) {
GLint tempReg = _mesa_find_free_register(newProg, PROGRAM_TEMPORARY);
if (tempReg < 0) {
_mesa_problem(ctx, "No free temp regs found in "
"_mesa_combine_programs(), using 31");
tempReg = 31;
}
/* replace writes to result.color[0] with tempReg */
replace_registers(newInst, lenA,
PROGRAM_OUTPUT, FRAG_RESULT_COLR,
PROGRAM_TEMPORARY, tempReg);
/* replace reads from input.color[0] with tempReg */
replace_registers(newInst + lenA, lenB,
PROGRAM_INPUT, FRAG_ATTRIB_COL0,
PROGRAM_TEMPORARY, tempReg);
}
inputsB = progB->InputsRead;
if (progA->OutputsWritten & (1 << FRAG_RESULT_COLR)) {
inputsB &= ~(1 << FRAG_ATTRIB_COL0);
}
newProg->InputsRead = progA->InputsRead | inputsB;
newProg->OutputsWritten = progB->OutputsWritten;
newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
}
else {
/* vertex program */
assert(0); /* XXX todo */
}
/*
* Merge parameters (uniforms, constants, etc)
*/
newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
progB->Parameters);
adjust_param_indexes(newInst + lenA, lenB, numParamsA);
return newProg;
}
/**
* Scan/rewrite program to remove reads of custom (output) registers.
* The passed type has to be PROGRAM_OUTPUT.
* On some hardware, trying to read an output register causes trouble.
* So, rewrite the program to use a temporary register in this case.
*/
void
_mesa_remove_output_reads(struct gl_program *prog, gl_register_file type)
{
GLuint i;
GLint outputMap[VARYING_SLOT_MAX];
GLuint numVaryingReads = 0;
GLboolean usedTemps[MAX_PROGRAM_TEMPS];
GLuint firstTemp = 0;
_mesa_find_used_registers(prog, PROGRAM_TEMPORARY,
usedTemps, MAX_PROGRAM_TEMPS);
assert(type == PROGRAM_OUTPUT);
for (i = 0; i < VARYING_SLOT_MAX; i++)
outputMap[i] = -1;
/* look for instructions which read from varying vars */
for (i = 0; i < prog->NumInstructions; i++) {
struct prog_instruction *inst = prog->Instructions + i;
const GLuint numSrc = _mesa_num_inst_src_regs(inst->Opcode);
GLuint j;
for (j = 0; j < numSrc; j++) {
if (inst->SrcReg[j].File == type) {
/* replace the read with a temp reg */
const GLuint var = inst->SrcReg[j].Index;
if (outputMap[var] == -1) {
numVaryingReads++;
outputMap[var] = _mesa_find_free_register(usedTemps,
MAX_PROGRAM_TEMPS,
firstTemp);
firstTemp = outputMap[var] + 1;
}
inst->SrcReg[j].File = PROGRAM_TEMPORARY;
inst->SrcReg[j].Index = outputMap[var];
}
}
}
if (numVaryingReads == 0)
return; /* nothing to be done */
/* look for instructions which write to the varying vars identified above */
for (i = 0; i < prog->NumInstructions; i++) {
struct prog_instruction *inst = prog->Instructions + i;
if (inst->DstReg.File == type &&
outputMap[inst->DstReg.Index] >= 0) {
/* change inst to write to the temp reg, instead of the varying */
inst->DstReg.File = PROGRAM_TEMPORARY;
inst->DstReg.Index = outputMap[inst->DstReg.Index];
}
}
/* insert new instructions to copy the temp vars to the varying vars */
{
struct prog_instruction *inst;
GLint endPos, var;
/* Look for END instruction and insert the new varying writes */
endPos = -1;
for (i = 0; i < prog->NumInstructions; i++) {
struct prog_instruction *inst = prog->Instructions + i;
if (inst->Opcode == OPCODE_END) {
endPos = i;
_mesa_insert_instructions(prog, i, numVaryingReads);
break;
}
}
assert(endPos >= 0);
/* insert new MOV instructions here */
inst = prog->Instructions + endPos;
for (var = 0; var < VARYING_SLOT_MAX; var++) {
if (outputMap[var] >= 0) {
/* MOV VAR[var], TEMP[tmp]; */
inst->Opcode = OPCODE_MOV;
inst->DstReg.File = type;
inst->DstReg.Index = var;
inst->SrcReg[0].File = PROGRAM_TEMPORARY;
inst->SrcReg[0].Index = outputMap[var];
inst++;
}
}
}
}
/**
* Combine two programs into one. Fix instructions so the outputs of
* the first program go to the inputs of the second program.
*/
struct gl_program *
_mesa_combine_programs(struct gl_context *ctx,
const struct gl_program *progA,
const struct gl_program *progB)
{
struct prog_instruction *newInst;
struct gl_program *newProg;
const GLuint lenA = progA->NumInstructions - 1; /* omit END instr */
const GLuint lenB = progB->NumInstructions;
const GLuint numParamsA = _mesa_num_parameters(progA->Parameters);
const GLuint newLength = lenA + lenB;
GLboolean usedTemps[MAX_PROGRAM_TEMPS];
GLuint firstTemp = 0;
GLbitfield inputsB;
GLuint i;
ASSERT(progA->Target == progB->Target);
newInst = _mesa_alloc_instructions(newLength);
if (!newInst)
return GL_FALSE;
_mesa_copy_instructions(newInst, progA->Instructions, lenA);
_mesa_copy_instructions(newInst + lenA, progB->Instructions, lenB);
/* adjust branch / instruction addresses for B's instructions */
for (i = 0; i < lenB; i++) {
newInst[lenA + i].BranchTarget += lenA;
}
newProg = ctx->Driver.NewProgram(ctx, progA->Target, 0);
newProg->Instructions = newInst;
newProg->NumInstructions = newLength;
/* find used temp regs (we may need new temps below) */
_mesa_find_used_registers(newProg, PROGRAM_TEMPORARY,
usedTemps, MAX_PROGRAM_TEMPS);
if (newProg->Target == GL_FRAGMENT_PROGRAM_ARB) {
struct gl_fragment_program *fprogA, *fprogB, *newFprog;
GLbitfield progB_inputsRead = progB->InputsRead;
GLint progB_colorFile, progB_colorIndex;
fprogA = (struct gl_fragment_program *) progA;
fprogB = (struct gl_fragment_program *) progB;
newFprog = (struct gl_fragment_program *) newProg;
newFprog->UsesKill = fprogA->UsesKill || fprogB->UsesKill;
/* We'll do a search and replace for instances
* of progB_colorFile/progB_colorIndex below...
*/
progB_colorFile = PROGRAM_INPUT;
progB_colorIndex = FRAG_ATTRIB_COL0;
/*
* The fragment program may get color from a state var rather than
* a fragment input (vertex output) if it's constant.
* See the texenvprogram.c code.
* So, search the program's parameter list now to see if the program
* gets color from a state var instead of a conventional fragment
* input register.
*/
for (i = 0; i < progB->Parameters->NumParameters; i++) {
struct gl_program_parameter *p = &progB->Parameters->Parameters[i];
if (p->Type == PROGRAM_STATE_VAR &&
p->StateIndexes[0] == STATE_INTERNAL &&
p->StateIndexes[1] == STATE_CURRENT_ATTRIB &&
(int) p->StateIndexes[2] == (int) VERT_ATTRIB_COLOR0) {
progB_inputsRead |= FRAG_BIT_COL0;
progB_colorFile = PROGRAM_STATE_VAR;
progB_colorIndex = i;
break;
}
}
/* Connect color outputs of fprogA to color inputs of fprogB, via a
* new temporary register.
*/
if ((progA->OutputsWritten & (1 << FRAG_RESULT_COLOR)) &&
(progB_inputsRead & FRAG_BIT_COL0)) {
GLint tempReg = _mesa_find_free_register(usedTemps, MAX_PROGRAM_TEMPS,
firstTemp);
if (tempReg < 0) {
_mesa_problem(ctx, "No free temp regs found in "
"_mesa_combine_programs(), using 31");
tempReg = 31;
}
firstTemp = tempReg + 1;
/* replace writes to result.color[0] with tempReg */
replace_registers(newInst, lenA,
PROGRAM_OUTPUT, FRAG_RESULT_COLOR,
PROGRAM_TEMPORARY, tempReg);
/* replace reads from the input color with tempReg */
replace_registers(newInst + lenA, lenB,
progB_colorFile, progB_colorIndex, /* search for */
PROGRAM_TEMPORARY, tempReg /* replace with */ );
}
/* compute combined program's InputsRead */
inputsB = progB_inputsRead;
if (progA->OutputsWritten & (1 << FRAG_RESULT_COLOR)) {
inputsB &= ~(1 << FRAG_ATTRIB_COL0);
}
newProg->InputsRead = progA->InputsRead | inputsB;
newProg->OutputsWritten = progB->OutputsWritten;
newProg->SamplersUsed = progA->SamplersUsed | progB->SamplersUsed;
}
else {
/* vertex program */
assert(0); /* XXX todo */
}
/*
* Merge parameters (uniforms, constants, etc)
*/
newProg->Parameters = _mesa_combine_parameter_lists(progA->Parameters,
progB->Parameters);
adjust_param_indexes(newInst + lenA, lenB, numParamsA);
return newProg;
}
