void
_mesa_emit_nv_temp_initialization(struct gl_context *ctx,
struct gl_program *program)
{
struct prog_instruction *inst;
GLuint i;
struct gl_shader_compiler_options* options =
&ctx->ShaderCompilerOptions[_mesa_program_target_to_index(program->Target)];
if (!options->EmitNVTempInitialization)
return;
/* We'll swizzle up a zero temporary so we can use it for the
* ARL.
*/
if (program->NumTemporaries == 0)
program->NumTemporaries = 1;
_mesa_insert_instructions(program, 0, program->NumTemporaries + 1);
for (i = 0; i < program->NumTemporaries; i++) {
struct prog_instruction *inst = &program->Instructions[i];
inst->Opcode = OPCODE_SWZ;
inst->DstReg.File = PROGRAM_TEMPORARY;
inst->DstReg.Index = i;
inst->DstReg.WriteMask = WRITEMASK_XYZW;
inst->SrcReg[0].File = PROGRAM_TEMPORARY;
inst->SrcReg[0].Index = 0;
inst->SrcReg[0].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ZERO,
SWIZZLE_ZERO);
}
inst = &program->Instructions[i];
inst->Opcode = OPCODE_ARL;
inst->DstReg.File = PROGRAM_ADDRESS;
inst->DstReg.Index = 0;
inst->DstReg.WriteMask = WRITEMASK_XYZW;
inst->SrcReg[0].File = PROGRAM_TEMPORARY;
inst->SrcReg[0].Index = 0;
inst->SrcReg[0].Swizzle = SWIZZLE_XXXX;
if (program->NumAddressRegs == 0)
program->NumAddressRegs = 1;
}
void insert_wpos_code(GLcontext *ctx, struct gl_fragment_program *fprog)
{
static const gl_state_index winstate[STATE_LENGTH]
= { STATE_INTERNAL, STATE_FB_SIZE, 0, 0, 0};
struct prog_instruction *newInst, *inst;
GLint win_size; /* state reference */
GLuint wpos_temp; /* temp register */
int i, j;
/* PARAM win_size = STATE_FB_SIZE */
win_size = _mesa_add_state_reference(fprog->Base.Parameters, winstate);
wpos_temp = fprog->Base.NumTemporaries++;
/* scan program where WPOS is used and replace with wpos_temp */
inst = fprog->Base.Instructions;
for (i = 0; i < fprog->Base.NumInstructions; i++) {
for (j=0; j < 3; j++) {
if(inst->SrcReg[j].File == PROGRAM_INPUT &&
inst->SrcReg[j].Index == FRAG_ATTRIB_WPOS) {
inst->SrcReg[j].File = PROGRAM_TEMPORARY;
inst->SrcReg[j].Index = wpos_temp;
}
}
inst++;
}
_mesa_insert_instructions(&(fprog->Base), 0, 1);
newInst = fprog->Base.Instructions;
/* invert wpos.y
* wpos_temp.xyzw = wpos.x-yzw + winsize.0y00 */
newInst[0].Opcode = OPCODE_ADD;
newInst[0].DstReg.File = PROGRAM_TEMPORARY;
newInst[0].DstReg.Index = wpos_temp;
newInst[0].DstReg.WriteMask = WRITEMASK_XYZW;
newInst[0].SrcReg[0].File = PROGRAM_INPUT;
newInst[0].SrcReg[0].Index = FRAG_ATTRIB_WPOS;
newInst[0].SrcReg[0].Swizzle = SWIZZLE_XYZW;
newInst[0].SrcReg[0].Negate = NEGATE_Y;
newInst[0].SrcReg[1].File = PROGRAM_STATE_VAR;
newInst[0].SrcReg[1].Index = win_size;
newInst[0].SrcReg[1].Swizzle = MAKE_SWIZZLE4(SWIZZLE_ZERO, SWIZZLE_Y, SWIZZLE_ZERO, SWIZZLE_ZERO);
}
/**
* 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;
}
}
}
}
/**
* 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++;
}
}
}
}