Exverge/suyu
3
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Shader_Ir: Downgrade precision and rounding asserts to debug asserts.

Browse source 
This commit reduces the sevirity of asserts for FP precision and 
rounding as this are well known and have little to no consequences in 
gpu's accuracy.
This commit is contained in:
Fernando Sahmkow 2019-07-18 08:17:19 -04:00
parent d4b95bfc25
commit 0b65e9335e
5 changed files with 10 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
src/video_core/shader/decode/arithmetic.cpp
View file

@ -42,10 +42,10 @@ u32 ShaderIR::DecodeArithmetic(NodeBlock& bb, u32 pc) {
case OpCode::Id::FMUL_R: case OpCode::Id::FMUL_R:
case OpCode::Id::FMUL_IMM: { case OpCode::Id::FMUL_IMM: {
// FMUL does not have 'abs' bits and only the second operand has a 'neg' bit. // FMUL does not have 'abs' bits and only the second operand has a 'neg' bit.
UNIMPLEMENTED_IF_MSG(instr.fmul.tab5cb8_2 != 0, "FMUL tab5cb8_2({}) is not implemented", DEBUG_ASSERT_MSG(instr.fmul.tab5cb8_2 == 0, "FMUL tab5cb8_2({}) is not implemented",
instr.fmul.tab5cb8_2.Value()); instr.fmul.tab5cb8_2.Value());
UNIMPLEMENTED_IF_MSG( DEBUG_ASSERT_MSG(
instr.fmul.tab5c68_0 != 1, "FMUL tab5cb8_0({}) is not implemented", instr.fmul.tab5c68_0 == 1, "FMUL tab5cb8_0({}) is not implemented",
instr.fmul.tab5c68_0.Value()); // SMO typical sends 1 here which seems to be the default instr.fmul.tab5c68_0.Value()); // SMO typical sends 1 here which seems to be the default
op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b); op_b = GetOperandAbsNegFloat(op_b, false, instr.fmul.negate_b);

2
src/video_core/shader/decode/arithmetic_half_immediate.cpp
View file

@ -23,7 +23,7 @@ u32 ShaderIR::DecodeArithmeticHalfImmediate(NodeBlock& bb, u32 pc) {
LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName()); LOG_WARNING(HW_GPU, "{} FTZ not implemented", opcode->get().GetName());
} }
} else { } else {
UNIMPLEMENTED_IF(instr.alu_half_imm.precision != Tegra::Shader::HalfPrecision::None); DEBUG_ASSERT(instr.alu_half_imm.precision == Tegra::Shader::HalfPrecision::None);
} }
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a); Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.alu_half_imm.type_a);

4
src/video_core/shader/decode/ffma.cpp
View file

@ -18,9 +18,9 @@ u32 ShaderIR::DecodeFfma(NodeBlock& bb, u32 pc) {
const auto opcode = OpCode::Decode(instr); const auto opcode = OpCode::Decode(instr);
UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented"); UNIMPLEMENTED_IF_MSG(instr.ffma.cc != 0, "FFMA cc not implemented");
UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_0 != 1, "FFMA tab5980_0({}) not implemented", DEBUG_ASSERT_MSG(instr.ffma.tab5980_0 == 1, "FFMA tab5980_0({}) not implemented",
instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO instr.ffma.tab5980_0.Value()); // Seems to be 1 by default based on SMO
UNIMPLEMENTED_IF_MSG(instr.ffma.tab5980_1 != 0, "FFMA tab5980_1({}) not implemented", DEBUG_ASSERT_MSG(instr.ffma.tab5980_1 == 0, "FFMA tab5980_1({}) not implemented",
instr.ffma.tab5980_1.Value()); instr.ffma.tab5980_1.Value());
const Node op_a = GetRegister(instr.gpr8); const Node op_a = GetRegister(instr.gpr8);

2
src/video_core/shader/decode/half_set_predicate.cpp
View file

@ -18,7 +18,7 @@ u32 ShaderIR::DecodeHalfSetPredicate(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]}; const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr); const auto opcode = OpCode::Decode(instr);
UNIMPLEMENTED_IF(instr.hsetp2.ftz != 0); DEBUG_ASSERT(instr.hsetp2.ftz == 0);
Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a); Node op_a = UnpackHalfFloat(GetRegister(instr.gpr8), instr.hsetp2.type_a);
op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a); op_a = GetOperandAbsNegHalf(op_a, instr.hsetp2.abs_a, instr.hsetp2.negate_a);

4
src/video_core/shader/decode/hfma2.cpp
View file

@ -22,9 +22,9 @@ u32 ShaderIR::DecodeHfma2(NodeBlock& bb, u32 pc) {
const auto opcode = OpCode::Decode(instr); const auto opcode = OpCode::Decode(instr);
if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) { if (opcode->get().GetId() == OpCode::Id::HFMA2_RR) {
UNIMPLEMENTED_IF(instr.hfma2.rr.precision != HalfPrecision::None); DEBUG_ASSERT(instr.hfma2.rr.precision == HalfPrecision::None);
} else { } else {
UNIMPLEMENTED_IF(instr.hfma2.precision != HalfPrecision::None); DEBUG_ASSERT(instr.hfma2.precision == HalfPrecision::None);
} }
constexpr auto identity = HalfType::H0_H1; constexpr auto identity = HalfType::H0_H1;