ShaderGen: Implemented the fsetp instruction.

Predicate variables are now added to the generated shader code in the form of 'pX' where X is the predicate id.
These predicate variables are initialized to false on shader startup and are set via the fsetp instructions.

TODO:

* Not all the comparison types are implemented.
* Only the single-predicate version is implemented.
This commit is contained in:
Subv 2018-04-20 09:09:50 -05:00
parent d03fc77475
commit 0a5e01b710
2 changed files with 112 additions and 3 deletions

View file

@ -109,6 +109,8 @@ union OpCode {
FSETP_R = 0x5BB, FSETP_R = 0x5BB,
FSETP_C = 0x4BB, FSETP_C = 0x4BB,
FSETP_IMM = 0x36B,
FSETP_NEG_IMM = 0x37B,
EXIT = 0xE30, EXIT = 0xE30,
KIL = 0xE33, KIL = 0xE33,
@ -124,6 +126,7 @@ union OpCode {
Ffma, Ffma,
Flow, Flow,
Memory, Memory,
FloatPredicate,
Unknown, Unknown,
}; };
@ -164,6 +167,9 @@ union OpCode {
case Id::FSETP_C: case Id::FSETP_C:
case Id::KIL: case Id::KIL:
return op4; return op4;
case Id::FSETP_IMM:
case Id::FSETP_NEG_IMM:
return Id::FSETP_IMM;
} }
switch (op5) { switch (op5) {
@ -241,8 +247,9 @@ union OpCode {
info_table[Id::FMUL_C] = {Type::Arithmetic, "fmul_c"}; info_table[Id::FMUL_C] = {Type::Arithmetic, "fmul_c"};
info_table[Id::FMUL_IMM] = {Type::Arithmetic, "fmul_imm"}; info_table[Id::FMUL_IMM] = {Type::Arithmetic, "fmul_imm"};
info_table[Id::FMUL32_IMM] = {Type::Arithmetic, "fmul32_imm"}; info_table[Id::FMUL32_IMM] = {Type::Arithmetic, "fmul32_imm"};
info_table[Id::FSETP_C] = {Type::Arithmetic, "fsetp_c"}; info_table[Id::FSETP_C] = {Type::FloatPredicate, "fsetp_c"};
info_table[Id::FSETP_R] = {Type::Arithmetic, "fsetp_r"}; info_table[Id::FSETP_R] = {Type::FloatPredicate, "fsetp_r"};
info_table[Id::FSETP_IMM] = {Type::FloatPredicate, "fsetp_imm"};
info_table[Id::EXIT] = {Type::Trivial, "exit"}; info_table[Id::EXIT] = {Type::Trivial, "exit"};
info_table[Id::IPA] = {Type::Trivial, "ipa"}; info_table[Id::IPA] = {Type::Trivial, "ipa"};
info_table[Id::KIL] = {Type::Flow, "kil"}; info_table[Id::KIL] = {Type::Flow, "kil"};
@ -286,7 +293,23 @@ namespace Shader {
enum class Pred : u64 { enum class Pred : u64 {
UnusedIndex = 0x7, UnusedIndex = 0x7,
NeverExecute = 0xf, NeverExecute = 0xF,
};
enum class PredCondition : u64 {
LessThan = 1,
Equal = 2,
LessEqual = 3,
GreaterThan = 4,
NotEqual = 5,
GreaterEqual = 6,
// TODO(Subv): Other condition types
};
enum class PredOperation : u64 {
And = 0,
Or = 1,
Xor = 2,
}; };
enum class SubOp : u64 { enum class SubOp : u64 {
@ -346,6 +369,20 @@ union Instruction {
BitField<49, 1, u64> negate_c; BitField<49, 1, u64> negate_c;
} ffma; } ffma;
union {
BitField<0, 3, u64> pred0;
BitField<3, 3, u64> pred3;
BitField<7, 1, u64> abs_a;
BitField<39, 3, u64> pred39;
BitField<42, 1, u64> neg_pred;
BitField<43, 1, u64> neg_a;
BitField<44, 1, u64> abs_b;
BitField<45, 2, PredOperation> op;
BitField<47, 1, u64> ftz;
BitField<48, 4, PredCondition> cond;
BitField<56, 1, u64> neg_b;
} fsetp;
BitField<61, 1, u64> is_b_imm; BitField<61, 1, u64> is_b_imm;
BitField<60, 1, u64> is_b_gpr; BitField<60, 1, u64> is_b_gpr;
BitField<59, 1, u64> is_c_gpr; BitField<59, 1, u64> is_c_gpr;

View file

@ -278,6 +278,21 @@ private:
shader.AddLine(dest + " = " + src + ";"); shader.AddLine(dest + " = " + src + ";");
} }
/*
* Writes code that assigns a predicate boolean variable.
* @param pred The id of the predicate to write to.
* @param value The expression value to assign to the predicate.
*/
void SetPredicate(u64 pred, const std::string& value) {
using Tegra::Shader::Pred;
// Can't assign to the constant predicate.
ASSERT(pred != static_cast<u64>(Pred::UnusedIndex));
std::string variable = 'p' + std::to_string(pred);
shader.AddLine(variable + " = " + value + ';');
declr_predicates.insert(std::move(variable));
}
/* /*
* Returns whether the instruction at the specified offset is a 'sched' instruction. * Returns whether the instruction at the specified offset is a 'sched' instruction.
* Sched instructions always appear before a sequence of 3 instructions. * Sched instructions always appear before a sequence of 3 instructions.
@ -468,7 +483,57 @@ private:
} }
break; break;
} }
case OpCode::Type::FloatPredicate: {
std::string op_a = instr.fsetp.neg_a ? "-" : "";
op_a += GetRegister(instr.gpr8);
if (instr.fsetp.abs_a) {
op_a = "abs(" + op_a + ')';
}
std::string op_b{};
if (instr.is_b_imm) {
if (instr.fsetp.neg_b) {
// Only the immediate version of fsetp has a neg_b bit.
op_b += '-';
}
op_b += '(' + GetImmediate19(instr) + ')';
} else {
if (instr.is_b_gpr) {
op_b += GetRegister(instr.gpr20);
} else {
op_b += GetUniform(instr.uniform);
}
}
if (instr.fsetp.abs_b) {
op_b = "abs(" + op_b + ')';
}
using Tegra::Shader::Pred;
ASSERT_MSG(instr.fsetp.pred0 == static_cast<u64>(Pred::UnusedIndex) &&
instr.fsetp.pred39 == static_cast<u64>(Pred::UnusedIndex),
"Compound predicates are not implemented");
// We can't use the constant predicate as destination.
ASSERT(instr.fsetp.pred3 != static_cast<u64>(Pred::UnusedIndex));
using Tegra::Shader::PredCondition;
switch (instr.fsetp.cond) {
case PredCondition::LessThan:
SetPredicate(instr.fsetp.pred3, '(' + op_a + ") < (" + op_b + ')');
break;
case PredCondition::Equal:
SetPredicate(instr.fsetp.pred3, '(' + op_a + ") == (" + op_b + ')');
break;
default:
NGLOG_CRITICAL(HW_GPU, "Unhandled predicate condition: {} (a: {}, b: {})",
static_cast<unsigned>(instr.fsetp.cond.Value()), op_a, op_b);
UNREACHABLE();
}
break;
}
default: { default: {
switch (instr.opcode.EffectiveOpCode()) { switch (instr.opcode.EffectiveOpCode()) {
case OpCode::Id::EXIT: { case OpCode::Id::EXIT: {
@ -623,6 +688,12 @@ private:
declarations.AddNewLine(); declarations.AddNewLine();
++const_buffer_layout; ++const_buffer_layout;
} }
declarations.AddNewLine();
for (const auto& pred : declr_predicates) {
declarations.AddLine("bool " + pred + " = false;");
}
declarations.AddNewLine();
} }
private: private:
@ -636,6 +707,7 @@ private:
// Declarations // Declarations
std::set<std::string> declr_register; std::set<std::string> declr_register;
std::set<std::string> declr_predicates;
std::set<Attribute::Index> declr_input_attribute; std::set<Attribute::Index> declr_input_attribute;
std::set<Attribute::Index> declr_output_attribute; std::set<Attribute::Index> declr_output_attribute;
std::array<ConstBufferEntry, Maxwell3D::Regs::MaxConstBuffers> declr_const_buffers; std::array<ConstBufferEntry, Maxwell3D::Regs::MaxConstBuffers> declr_const_buffers;