void rv_emit(SymbolTable *table, Node *ptr)
{
int stIndex;
if (ptr->token.tokenNumber == NUMBER) {
emit1("ldc", atoi(ptr->token.tokenValue));
} else {
SymbolRow* foundRow;
stIndex = lookup(table, ptr->token.tokenValue);
if (stIndex != -1) {
foundRow = &table->rows[stIndex];
}
if (stIndex == -1) {
stIndex = lookup(rootTable, ptr->token.tokenValue);
if (stIndex != -1) {
foundRow = &(rootTable->rows[stIndex]);
} else {
return;
}
}
if (foundRow->qual == QUAL_CONST) {
emit1("ldc", foundRow->init);
} else if (foundRow->width > 1) {
emit2("lda", foundRow->base, foundRow->offset);
} else {
emit2("lod", foundRow->base, foundRow->offset);
}
}
}
static void r300_transform_SIN_COS_SCS(struct radeon_compiler *c,
struct rc_instruction *inst,
unsigned srctmp)
{
if (inst->U.I.Opcode == RC_OPCODE_COS) {
emit1(c, inst->Prev, RC_OPCODE_COS, inst->U.I.SaturateMode, inst->U.I.DstReg,
srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
} else if (inst->U.I.Opcode == RC_OPCODE_SIN) {
emit1(c, inst->Prev, RC_OPCODE_SIN, inst->U.I.SaturateMode,
inst->U.I.DstReg, srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
} else if (inst->U.I.Opcode == RC_OPCODE_SCS) {
struct rc_dst_register moddst = inst->U.I.DstReg;
if (inst->U.I.DstReg.WriteMask & RC_MASK_X) {
moddst.WriteMask = RC_MASK_X;
emit1(c, inst->Prev, RC_OPCODE_COS, inst->U.I.SaturateMode, moddst,
srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
}
if (inst->U.I.DstReg.WriteMask & RC_MASK_Y) {
moddst.WriteMask = RC_MASK_Y;
emit1(c, inst->Prev, RC_OPCODE_SIN, inst->U.I.SaturateMode, moddst,
srcregswz(RC_FILE_TEMPORARY, srctmp, RC_SWIZZLE_WWWW));
}
}
rc_remove_instruction(inst);
}
static void transform_POW(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register tempdst = try_to_reuse_dst(c, inst);
struct rc_src_register tempsrc = srcreg(RC_FILE_TEMPORARY, tempdst.Index);
tempdst.WriteMask = RC_MASK_W;
tempsrc.Swizzle = RC_SWIZZLE_WWWW;
emit1(c, inst->Prev, RC_OPCODE_LG2, 0, tempdst, swizzle_xxxx(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_MUL, 0, tempdst, tempsrc, swizzle_xxxx(inst->U.I.SrcReg[1]));
emit1(c, inst->Prev, RC_OPCODE_EX2, inst->U.I.SaturateMode, inst->U.I.DstReg, tempsrc);
rc_remove_instruction(inst);
}
/**
* Transform the trigonometric functions COS, SIN, and SCS
* so that the input to COS and SIN is always in the range [-PI, PI].
* SCS is replaced by one COS and one SIN instruction.
*/
int r300_transform_trig_scale_vertex(struct radeon_compiler *c,
struct rc_instruction *inst,
void *unused)
{
static const float cons[4] = {0.15915494309189535, 0.5, 6.28318530717959, -3.14159265358979};
unsigned int temp;
unsigned int constant;
if (inst->U.I.Opcode != RC_OPCODE_COS &&
inst->U.I.Opcode != RC_OPCODE_SIN &&
inst->U.I.Opcode != RC_OPCODE_SCS)
return 0;
/* Repeat x in the range [-PI, PI]:
*
* repeat(x) = frac(x / 2PI + 0.5) * 2PI - PI
*/
temp = rc_find_free_temporary(c);
constant = rc_constants_add_immediate_vec4(&c->Program.Constants, cons);
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(temp, RC_MASK_W),
swizzle_xxxx(inst->U.I.SrcReg[0]),
srcregswz(RC_FILE_CONSTANT, constant, RC_SWIZZLE_XXXX),
srcregswz(RC_FILE_CONSTANT, constant, RC_SWIZZLE_YYYY));
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(temp, RC_MASK_W),
srcreg(RC_FILE_TEMPORARY, temp));
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(temp, RC_MASK_W),
srcreg(RC_FILE_TEMPORARY, temp),
srcregswz(RC_FILE_CONSTANT, constant, RC_SWIZZLE_ZZZZ),
srcregswz(RC_FILE_CONSTANT, constant, RC_SWIZZLE_WWWW));
r300_transform_SIN_COS_SCS(c, inst, temp);
return 1;
}
/* dst = ROUND(src) :
* add = src + .5
* frac = FRC(add)
* dst = add - frac
*
* According to the GLSL spec, the implementor can decide which way to round
* when the fraction is .5. We round down for .5.
*
*/
static void transform_ROUND(struct radeon_compiler* c,
struct rc_instruction* inst)
{
unsigned int mask = inst->U.I.DstReg.WriteMask;
unsigned int frac_index, add_index;
struct rc_dst_register frac_dst, add_dst;
struct rc_src_register frac_src, add_src;
/* add = src + .5 */
add_index = rc_find_free_temporary(c);
add_dst = dstregtmpmask(add_index, mask);
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, add_dst, inst->U.I.SrcReg[0],
builtin_half);
add_src = srcreg(RC_FILE_TEMPORARY, add_dst.Index);
/* frac = FRC(add) */
frac_index = rc_find_free_temporary(c);
frac_dst = dstregtmpmask(frac_index, mask);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, frac_dst, add_src);
frac_src = srcreg(RC_FILE_TEMPORARY, frac_dst.Index);
/* dst = add - frac */
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, inst->U.I.DstReg,
add_src, negate(frac_src));
rc_remove_instruction(inst);
}
/**
* Transform the trigonometric functions COS, SIN, and SCS
* to include pre-scaling by 1/(2*PI) and taking the fractional
* part, so that the input to COS and SIN is always in the range [0,1).
* SCS is replaced by one COS and one SIN instruction.
*
* @warning This transformation implicitly changes the semantics of SIN and COS!
*/
int radeonTransformTrigScale(struct radeon_compiler* c,
struct rc_instruction* inst,
void* unused)
{
static const float RCP_2PI = 0.15915494309189535;
unsigned int temp;
unsigned int constant;
unsigned int constant_swizzle;
if (inst->U.I.Opcode != RC_OPCODE_COS &&
inst->U.I.Opcode != RC_OPCODE_SIN &&
inst->U.I.Opcode != RC_OPCODE_SCS)
return 0;
temp = rc_find_free_temporary(c);
constant = rc_constants_add_immediate_scalar(&c->Program.Constants, RCP_2PI, &constant_swizzle);
emit2(c, inst->Prev, RC_OPCODE_MUL, 0, dstregtmpmask(temp, RC_MASK_W),
swizzle_xxxx(inst->U.I.SrcReg[0]),
srcregswz(RC_FILE_CONSTANT, constant, constant_swizzle));
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(temp, RC_MASK_W),
srcreg(RC_FILE_TEMPORARY, temp));
r300_transform_SIN_COS_SCS(c, inst, temp);
return 1;
}
void VirtualMachine::emit4(int v)
{
emit1(v & 255);
emit1((v >> 8) & 255);
emit1((v >> 16) & 255);
emit1((v >> 24) & 255);
}
static void transform_FLR(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, inst->U.I.SrcReg[0]);
emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
inst->U.I.SrcReg[0], negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
rc_remove_instruction(inst);
}
static void transform_ABS(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_src_register src = inst->U.I.SrcReg[0];
src.Abs = 1;
src.Negate = RC_MASK_NONE;
emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode, inst->U.I.DstReg, src);
rc_remove_instruction(inst);
}
static void transform_TRUNC(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* Definition of trunc:
* trunc(x) = (abs(x) - fract(abs(x))) * sgn(x)
*
* The multiplication by sgn(x) can be simplified using CMP:
* y * sgn(x) = (x < 0 ? -y : y)
*/
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, absolute(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_ADD, 0, dst, absolute(inst->U.I.SrcReg[0]),
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)));
emit3(c, inst->Prev, RC_OPCODE_CMP, &inst->U.I, inst->U.I.DstReg, inst->U.I.SrcReg[0],
negate(srcreg(RC_FILE_TEMPORARY, dst.Index)), srcreg(RC_FILE_TEMPORARY, dst.Index));
rc_remove_instruction(inst);
}
static void transform_CEIL(struct radeon_compiler* c,
struct rc_instruction* inst)
{
/* Assuming:
* ceil(x) = -floor(-x)
*
* After inlining floor:
* ceil(x) = -(-x-frac(-x))
*
* After simplification:
* ceil(x) = x+frac(-x)
*/
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dst, negate(inst->U.I.SrcReg[0]));
emit2(c, inst->Prev, RC_OPCODE_ADD, inst->U.I.SaturateMode, inst->U.I.DstReg,
inst->U.I.SrcReg[0], srcreg(RC_FILE_TEMPORARY, dst.Index));
rc_remove_instruction(inst);
}
void VirtualMachine::emitInstruction(const char *string)
{
int c1, c2;
int v;
while (true)
{
c1 = string[0];
c2 = string[1];
v = (hex( c1 ) << 4) | hex(c2);
emit1( v );
if (!string[2])
{
break;
}
string += 3;
}
}
void codeGen(Node *root, FILE *ucoFile)
{
Node *p; // pointer for Node
int globalSize; // the size of global variables
file = ucoFile; //
rootTable = initSymbolTable();
// step 1: process the declaration part
for (p = root->son; p; p = p->next) {
if (p->token.tokenNumber == DCL) {
processDeclaration(rootTable, p->son);
} else if (p->token.tokenNumber == FUNC_DEF) {
processFuncHeader(rootTable, p->son);
} else {
icg_error(3);
}
}
globalSize = rootTable->offset - 1;
// step 2: process the function part
for (p = root->son; p; p = p->next) {
if (p->token.tokenNumber == FUNC_DEF) {
processFunction(rootTable, p);
}
}
display(rootTable, 0);
// step 3: generate codes for starting routine
// bgn globalSize
// ldp
// call main
// end
emit1("bgn", globalSize);
emit0("ldp");
emitJump("call", "main");
emit0("end");
}
static void transform_r300_vertex_fix_LIT(struct radeon_compiler* c,
struct rc_instruction* inst)
{
struct rc_dst_register dst = try_to_reuse_dst(c, inst);
unsigned constant_swizzle;
int constant = rc_constants_add_immediate_scalar(&c->Program.Constants,
0.0000000000000000001,
&constant_swizzle);
/* MOV dst, src */
dst.WriteMask = RC_MASK_XYZW;
emit1(c, inst->Prev, RC_OPCODE_MOV, 0,
dst,
inst->U.I.SrcReg[0]);
/* MAX dst.y, src, 0.00...001 */
emit2(c, inst->Prev, RC_OPCODE_MAX, 0,
dstregtmpmask(dst.Index, RC_MASK_Y),
srcreg(RC_FILE_TEMPORARY, dst.Index),
srcregswz(RC_FILE_CONSTANT, constant, constant_swizzle));
inst->U.I.SrcReg[0] = srcreg(RC_FILE_TEMPORARY, dst.Index);
}
void dch(u32 x) { emit1(x); }
void fillb(u32 size, u32 value) { int i; for(i=0; i<size; i+=2) emit1(value|(value<<8)); }
void emit5(u32 x, u32 y, u32 z, u32 a, u32 b) { emit1(x); emit1(y); emit1(z); emit1(a); emit1(b); }
void emit3(u32 x, u32 y, u32 z) { emit1(x); emit1(y); emit1(z); }
void emit2(u32 x, u32 y) { emit1(x); emit1(y); }
static int compile_operator(compiletime c)
{
int operand_count = 0;
const char *op = pop_operator(c, &operand_count);
if (!strcmp(op, "*"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, multiply_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "**"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, power_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "/"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, divide_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "%"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, modulo_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "="))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, assign_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, ";"))
{
}
else if (!strcmp(op, "if"))
{
const char *val1 = pop_operand(c);
emit2(c, if_tc, val1);
DEBUG printf("emit '%s' %s\n", val1, op);
c->level++;
}
else if (!strcmp(op, "else"))
{
c->level--;
emit1(c, else_tc);
c->level++;
DEBUG printf("emit %s\n", op);
}
else if (!strcmp(op, "fi"))
{
c->level--;
emit1(c, fi_tc);
DEBUG printf("emit %s\n", op);
}
else if (!strcmp(op, "!"))
{
const char *val1 = pop_operand(c);
emit2(c, not_tc, val1);
DEBUG printf("emit '%s' %s\n", val1, op);
}
else if (!strcmp(op, "+"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, add_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "-"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, subtract_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, ">"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, gt_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, ">="))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, geq_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "<="))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, leq_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "<"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, lt_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "&&"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, and_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "||"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, or_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "^^"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, xor_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "&"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, bit_and_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "|"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, bit_or_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "^"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, bit_xor_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "~"))
{
const char *val1 = pop_operand(c);
emit2(c, bit_negate_tc, val1);
DEBUG printf("emit %s %s\n", val1, op);
}
else if (!strcmp(op, "<<"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, shift_left_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, ">>"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, shift_right_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, ">>>"))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, logical_shift_right_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "=="))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, eq_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "!="))
{
const char *val2 = pop_operand(c);
const char *val1 = pop_operand(c);
emit3(c, neq_tc, val1, val2);
DEBUG printf("emit %s %s %s\n", val1, op, val2);
}
else if (!strcmp(op, "fold-case"))
{
const char *val1 = pop_operand(c);
emit2(c, func_fold_case_tc, val1);
DEBUG printf("emit %s %s\n", op, val1);
}
else if (!strcmp(op, "size"))
{
const char *val1 = pop_operand(c);
emit2(c, func_size_tc, val1);
DEBUG printf("emit %s %s\n", op, val1);
}
else if (!strcmp(op, "print"))
{
const char *val1 = pop_operand(c);
emit2(c, func_print_tc, val1);
DEBUG printf("emit %s %s\n", op, val1);
}
else if (!strcmp(op, "jday"))
{
const char *val1 = pop_operand(c);
emit2(c, func_jday_tc, val1);
DEBUG printf("emit %s %s\n", op, val1);
}
else if (!strcmp(op, "dow"))
{
const char *val1 = pop_operand(c);
emit2(c, func_dow_tc, val1);
DEBUG printf("emit %s %s\n", op, val1);
}
else if (!strcmp(op, "equals"))
{
emit_code(c, func_eq_tc, operand_count);
while (operand_count--)
{
const char *val1 = pop_operand(c);
emit_value(c, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
}
else if (!strcmp(op, "contains"))
{
emit_code(c, func_contains_tc, operand_count);
while (operand_count--)
{
const char *val1 = pop_operand(c);
emit_value(c, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
}
else if (!strcmp(op, "begins-with"))
{
emit_code(c, func_begins_with_tc, operand_count);
while (operand_count--)
{
const char *val1 = pop_operand(c);
emit_value(c, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
}
else if (!strcmp(op, "ends-with"))
{
emit_code(c, func_ends_with_tc, operand_count);
while (operand_count--)
{
const char *val1 = pop_operand(c);
emit_value(c, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
}
else if (!strcmp(op, "int"))
{
const char *val1 = pop_operand(c);
emit2(c, func_is_int_tc, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
else if (!strcmp(op, "real"))
{
const char *val1 = pop_operand(c);
emit2(c, func_is_real_tc, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
else if (!strcmp(op, "nan"))
{
const char *val1 = pop_operand(c);
emit2(c, func_is_nan_tc, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
else if (!strcmp(op, "string"))
{
const char *val1 = pop_operand(c);
emit2(c, func_is_string_tc, val1);
DEBUG printf("emit %s '%s'\n", op, val1);
}
else if (!strcmp(op, "("))
{
return 1;
}
else if (!strcmp(op, ")"))
{
return 1;
}
else
{
DEBUG printf("bad operator '%s'\n", op);
return 0;
}
push_operand(c, "_STACK");
return 1;
}
/**
* Translate the trigonometric functions COS, SIN, and SCS
* using only the basic instructions
* MOV, ADD, MUL, MAD, FRC
*/
int r300_transform_trig_simple(struct radeon_compiler* c,
struct rc_instruction* inst,
void* unused)
{
unsigned int constants[2];
unsigned int tempreg;
if (inst->U.I.Opcode != RC_OPCODE_COS &&
inst->U.I.Opcode != RC_OPCODE_SIN &&
inst->U.I.Opcode != RC_OPCODE_SCS)
return 0;
tempreg = rc_find_free_temporary(c);
sincos_constants(c, constants);
if (inst->U.I.Opcode == RC_OPCODE_COS) {
/* MAD tmp.x, src, 1/(2*PI), 0.75 */
/* FRC tmp.x, tmp.x */
/* MAD tmp.z, tmp.x, 2*PI, -PI */
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_xxxx(inst->U.I.SrcReg[0]),
swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[1])),
swizzle_xxxx(srcreg(RC_FILE_CONSTANT, constants[1])));
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)));
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)),
swizzle_wwww(srcreg(RC_FILE_CONSTANT, constants[1])),
negate(swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[0]))));
sin_approx(c, inst, inst->U.I.DstReg,
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)),
constants);
} else if (inst->U.I.Opcode == RC_OPCODE_SIN) {
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_xxxx(inst->U.I.SrcReg[0]),
swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[1])),
swizzle_yyyy(srcreg(RC_FILE_CONSTANT, constants[1])));
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)));
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_W),
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)),
swizzle_wwww(srcreg(RC_FILE_CONSTANT, constants[1])),
negate(swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[0]))));
sin_approx(c, inst, inst->U.I.DstReg,
swizzle_wwww(srcreg(RC_FILE_TEMPORARY, tempreg)),
constants);
} else {
struct rc_dst_register dst;
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_XY),
swizzle_xxxx(inst->U.I.SrcReg[0]),
swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[1])),
swizzle(srcreg(RC_FILE_CONSTANT, constants[1]), RC_SWIZZLE_X, RC_SWIZZLE_Y, RC_SWIZZLE_Z, RC_SWIZZLE_W));
emit1(c, inst->Prev, RC_OPCODE_FRC, 0, dstregtmpmask(tempreg, RC_MASK_XY),
srcreg(RC_FILE_TEMPORARY, tempreg));
emit3(c, inst->Prev, RC_OPCODE_MAD, 0, dstregtmpmask(tempreg, RC_MASK_XY),
srcreg(RC_FILE_TEMPORARY, tempreg),
swizzle_wwww(srcreg(RC_FILE_CONSTANT, constants[1])),
negate(swizzle_zzzz(srcreg(RC_FILE_CONSTANT, constants[0]))));
dst = inst->U.I.DstReg;
dst.WriteMask = inst->U.I.DstReg.WriteMask & RC_MASK_X;
sin_approx(c, inst, dst,
swizzle_xxxx(srcreg(RC_FILE_TEMPORARY, tempreg)),
constants);
dst.WriteMask = inst->U.I.DstReg.WriteMask & RC_MASK_Y;
sin_approx(c, inst, dst,
swizzle_yyyy(srcreg(RC_FILE_TEMPORARY, tempreg)),
constants);
}
rc_remove_instruction(inst);
return 1;
}
static void transform_SFL(struct radeon_compiler* c,
struct rc_instruction* inst)
{
emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode, inst->U.I.DstReg, builtin_zero);
rc_remove_instruction(inst);
}
/**
* Definition of LIT (from ARB_fragment_program):
*
* tmp = VectorLoad(op0);
* if (tmp.x < 0) tmp.x = 0;
* if (tmp.y < 0) tmp.y = 0;
* if (tmp.w < -(128.0-epsilon)) tmp.w = -(128.0-epsilon);
* else if (tmp.w > 128-epsilon) tmp.w = 128-epsilon;
* result.x = 1.0;
* result.y = tmp.x;
* result.z = (tmp.x > 0) ? RoughApproxPower(tmp.y, tmp.w) : 0.0;
* result.w = 1.0;
*
* The longest path of computation is the one leading to result.z,
* consisting of 5 operations. This implementation of LIT takes
* 5 slots, if the subsequent optimization passes are clever enough
* to pair instructions correctly.
*/
static void transform_LIT(struct radeon_compiler* c,
struct rc_instruction* inst)
{
unsigned int constant;
unsigned int constant_swizzle;
unsigned int temp;
struct rc_src_register srctemp;
constant = rc_constants_add_immediate_scalar(&c->Program.Constants, -127.999999, &constant_swizzle);
if (inst->U.I.DstReg.WriteMask != RC_MASK_XYZW || inst->U.I.DstReg.File != RC_FILE_TEMPORARY) {
struct rc_instruction * inst_mov;
inst_mov = emit1(c, inst,
RC_OPCODE_MOV, 0, inst->U.I.DstReg,
srcreg(RC_FILE_TEMPORARY, rc_find_free_temporary(c)));
inst->U.I.DstReg.File = RC_FILE_TEMPORARY;
inst->U.I.DstReg.Index = inst_mov->U.I.SrcReg[0].Index;
inst->U.I.DstReg.WriteMask = RC_MASK_XYZW;
}
temp = inst->U.I.DstReg.Index;
srctemp = srcreg(RC_FILE_TEMPORARY, temp);
/* tmp.x = max(0.0, Src.x); */
/* tmp.y = max(0.0, Src.y); */
/* tmp.w = clamp(Src.z, -128+eps, 128-eps); */
emit2(c, inst->Prev, RC_OPCODE_MAX, 0,
dstregtmpmask(temp, RC_MASK_XYW),
inst->U.I.SrcReg[0],
swizzle(srcreg(RC_FILE_CONSTANT, constant),
RC_SWIZZLE_ZERO, RC_SWIZZLE_ZERO, RC_SWIZZLE_ZERO, constant_swizzle&3));
emit2(c, inst->Prev, RC_OPCODE_MIN, 0,
dstregtmpmask(temp, RC_MASK_Z),
swizzle_wwww(srctemp),
negate(srcregswz(RC_FILE_CONSTANT, constant, constant_swizzle)));
/* tmp.w = Pow(tmp.y, tmp.w) */
emit1(c, inst->Prev, RC_OPCODE_LG2, 0,
dstregtmpmask(temp, RC_MASK_W),
swizzle_yyyy(srctemp));
emit2(c, inst->Prev, RC_OPCODE_MUL, 0,
dstregtmpmask(temp, RC_MASK_W),
swizzle_wwww(srctemp),
swizzle_zzzz(srctemp));
emit1(c, inst->Prev, RC_OPCODE_EX2, 0,
dstregtmpmask(temp, RC_MASK_W),
swizzle_wwww(srctemp));
/* tmp.z = (tmp.x > 0) ? tmp.w : 0.0 */
emit3(c, inst->Prev, RC_OPCODE_CMP, inst->U.I.SaturateMode,
dstregtmpmask(temp, RC_MASK_Z),
negate(swizzle_xxxx(srctemp)),
swizzle_wwww(srctemp),
builtin_zero);
/* tmp.x, tmp.y, tmp.w = 1.0, tmp.x, 1.0 */
emit1(c, inst->Prev, RC_OPCODE_MOV, inst->U.I.SaturateMode,
dstregtmpmask(temp, RC_MASK_XYW),
swizzle(srctemp, RC_SWIZZLE_ONE, RC_SWIZZLE_X, RC_SWIZZLE_ONE, RC_SWIZZLE_ONE));
rc_remove_instruction(inst);
}