mirror of
https://github.com/yuzu-emu/yuzu.git
synced 2024-08-21 22:53:23 +02:00
shader: Implement I2F
This commit is contained in:
parent
c97d03efb9
commit
f91859efd2
@ -84,6 +84,7 @@ add_library(shader_recompiler STATIC
|
||||
frontend/maxwell/translate/impl/integer_add_three_input.cpp
|
||||
frontend/maxwell/translate/impl/integer_compare.cpp
|
||||
frontend/maxwell/translate/impl/integer_compare_and_set.cpp
|
||||
frontend/maxwell/translate/impl/integer_floating_point_conversion.cpp
|
||||
frontend/maxwell/translate/impl/integer_funnel_shift.cpp
|
||||
frontend/maxwell/translate/impl/integer_minimum_maximum.cpp
|
||||
frontend/maxwell/translate/impl/integer_popcount.cpp
|
||||
|
@ -89,6 +89,8 @@ Id EmitContext::Def(const IR::Value& value) {
|
||||
return value.U1() ? true_value : false_value;
|
||||
case IR::Type::U32:
|
||||
return Constant(U32[1], value.U32());
|
||||
case IR::Type::U64:
|
||||
return Constant(U64, value.U64());
|
||||
case IR::Type::F32:
|
||||
return Constant(F32[1], value.F32());
|
||||
case IR::Type::F64:
|
||||
|
@ -243,6 +243,7 @@ Id EmitIMul32(EmitContext& ctx, Id a, Id b);
|
||||
Id EmitINeg32(EmitContext& ctx, Id value);
|
||||
Id EmitINeg64(EmitContext& ctx, Id value);
|
||||
Id EmitIAbs32(EmitContext& ctx, Id value);
|
||||
Id EmitIAbs64(EmitContext& ctx, Id value);
|
||||
Id EmitShiftLeftLogical32(EmitContext& ctx, Id base, Id shift);
|
||||
Id EmitShiftLeftLogical64(EmitContext& ctx, Id base, Id shift);
|
||||
Id EmitShiftRightLogical32(EmitContext& ctx, Id base, Id shift);
|
||||
@ -302,16 +303,28 @@ Id EmitConvertF16F32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32F16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32F64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64F32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16S8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16S16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16S32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16S64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16U8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16U16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16U32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF16U64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32S8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32S16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32S32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32S64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32U8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32U16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32U32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF32U64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64S8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64S16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64S32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64S64(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64U8(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64U16(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64U32(EmitContext& ctx, Id value);
|
||||
Id EmitConvertF64U64(EmitContext& ctx, Id value);
|
||||
Id EmitBindlessImageSampleImplicitLod(EmitContext&);
|
||||
|
@ -102,6 +102,14 @@ Id EmitConvertF64F32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpFConvert(ctx.F64[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16S8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16S16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16S32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F16[1], value);
|
||||
}
|
||||
@ -110,6 +118,14 @@ Id EmitConvertF16S64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16U8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16U16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF16U32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F16[1], value);
|
||||
}
|
||||
@ -118,6 +134,14 @@ Id EmitConvertF16U64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F16[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF32S8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F32[1], ctx.OpUConvert(ctx.U8, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF32S16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F32[1], ctx.OpUConvert(ctx.U16, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF32S32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F32[1], value);
|
||||
}
|
||||
@ -126,6 +150,14 @@ Id EmitConvertF32S64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F32[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF32U8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F32[1], ctx.OpUConvert(ctx.U8, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF32U16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F32[1], ctx.OpUConvert(ctx.U16, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF32U32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F32[1], value);
|
||||
}
|
||||
@ -134,6 +166,14 @@ Id EmitConvertF32U64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F32[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF64S8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F64[1], ctx.OpUConvert(ctx.U8, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF64S16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F64[1], ctx.OpUConvert(ctx.U16, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF64S32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F64[1], value);
|
||||
}
|
||||
@ -142,6 +182,14 @@ Id EmitConvertF64S64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertSToF(ctx.F64[1], value);
|
||||
}
|
||||
|
||||
Id EmitConvertF64U8(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F64[1], ctx.OpUConvert(ctx.U8, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF64U16(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F64[1], ctx.OpUConvert(ctx.U16, value));
|
||||
}
|
||||
|
||||
Id EmitConvertF64U32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpConvertUToF(ctx.F64[1], value);
|
||||
}
|
||||
|
@ -70,6 +70,10 @@ Id EmitIAbs32(EmitContext& ctx, Id value) {
|
||||
return ctx.OpSAbs(ctx.U32[1], value);
|
||||
}
|
||||
|
||||
Id EmitIAbs64(EmitContext& ctx, Id value) {
|
||||
return ctx.OpSAbs(ctx.U64, value);
|
||||
}
|
||||
|
||||
Id EmitShiftLeftLogical32(EmitContext& ctx, Id base, Id shift) {
|
||||
return ctx.OpShiftLeftLogical(ctx.U32[1], base, shift);
|
||||
}
|
||||
|
@ -53,6 +53,10 @@ U64 IREmitter::Imm64(u64 value) const {
|
||||
return U64{Value{value}};
|
||||
}
|
||||
|
||||
U64 IREmitter::Imm64(s64 value) const {
|
||||
return U64{Value{static_cast<u64>(value)}};
|
||||
}
|
||||
|
||||
F64 IREmitter::Imm64(f64 value) const {
|
||||
return F64{Value{value}};
|
||||
}
|
||||
@ -363,7 +367,7 @@ U1 IREmitter::GetSparseFromOp(const Value& op) {
|
||||
}
|
||||
|
||||
F16F32F64 IREmitter::FPAdd(const F16F32F64& a, const F16F32F64& b, FpControl control) {
|
||||
if (a.Type() != a.Type()) {
|
||||
if (a.Type() != b.Type()) {
|
||||
throw InvalidArgument("Mismatching types {} and {}", a.Type(), b.Type());
|
||||
}
|
||||
switch (a.Type()) {
|
||||
@ -974,8 +978,15 @@ U32U64 IREmitter::INeg(const U32U64& value) {
|
||||
}
|
||||
}
|
||||
|
||||
U32 IREmitter::IAbs(const U32& value) {
|
||||
return Inst<U32>(Opcode::IAbs32, value);
|
||||
U32U64 IREmitter::IAbs(const U32U64& value) {
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
return Inst<U32>(Opcode::IAbs32, value);
|
||||
case Type::U64:
|
||||
return Inst<U64>(Opcode::IAbs64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
}
|
||||
}
|
||||
|
||||
U32U64 IREmitter::ShiftLeftLogical(const U32U64& base, const U32& shift) {
|
||||
@ -1074,8 +1085,25 @@ U1 IREmitter::ILessThan(const U32& lhs, const U32& rhs, bool is_signed) {
|
||||
return Inst<U1>(is_signed ? Opcode::SLessThan : Opcode::ULessThan, lhs, rhs);
|
||||
}
|
||||
|
||||
U1 IREmitter::IEqual(const U32& lhs, const U32& rhs) {
|
||||
return Inst<U1>(Opcode::IEqual, lhs, rhs);
|
||||
U1 IREmitter::IEqual(const U32U64& lhs, const U32U64& rhs) {
|
||||
if (lhs.Type() != rhs.Type()) {
|
||||
throw InvalidArgument("Mismatching types {} and {}", lhs.Type(), rhs.Type());
|
||||
}
|
||||
switch (lhs.Type()) {
|
||||
case Type::U32:
|
||||
return Inst<U1>(Opcode::IEqual, lhs, rhs);
|
||||
case Type::U64: {
|
||||
// Manually compare the unpacked values
|
||||
const Value lhs_vector{UnpackUint2x32(lhs)};
|
||||
const Value rhs_vector{UnpackUint2x32(rhs)};
|
||||
return LogicalAnd(IEqual(IR::U32{CompositeExtract(lhs_vector, 0)},
|
||||
IR::U32{CompositeExtract(rhs_vector, 0)}),
|
||||
IEqual(IR::U32{CompositeExtract(lhs_vector, 1)},
|
||||
IR::U32{CompositeExtract(rhs_vector, 1)}));
|
||||
}
|
||||
default:
|
||||
ThrowInvalidType(lhs.Type());
|
||||
}
|
||||
}
|
||||
|
||||
U1 IREmitter::ILessThanEqual(const U32& lhs, const U32& rhs, bool is_signed) {
|
||||
@ -1198,79 +1226,96 @@ U32U64 IREmitter::ConvertFToI(size_t bitsize, bool is_signed, const F16F32F64& v
|
||||
}
|
||||
}
|
||||
|
||||
F16F32F64 IREmitter::ConvertSToF(size_t bitsize, const U32U64& value) {
|
||||
switch (bitsize) {
|
||||
F16F32F64 IREmitter::ConvertSToF(size_t dest_bitsize, size_t src_bitsize, const Value& value) {
|
||||
switch (dest_bitsize) {
|
||||
case 16:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F16>(Opcode::ConvertF16S8, value);
|
||||
case 16:
|
||||
return Inst<F16>(Opcode::ConvertF16S16, value);
|
||||
case 32:
|
||||
return Inst<F16>(Opcode::ConvertF16S32, value);
|
||||
case Type::U64:
|
||||
case 64:
|
||||
return Inst<F16>(Opcode::ConvertF16S64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
}
|
||||
break;
|
||||
case 32:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F32>(Opcode::ConvertF32S8, value);
|
||||
case 16:
|
||||
return Inst<F32>(Opcode::ConvertF32S16, value);
|
||||
case 32:
|
||||
return Inst<F32>(Opcode::ConvertF32S32, value);
|
||||
case Type::U64:
|
||||
case 64:
|
||||
return Inst<F32>(Opcode::ConvertF32S64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
}
|
||||
break;
|
||||
case 64:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
return Inst<F16>(Opcode::ConvertF64S32, value);
|
||||
case Type::U64:
|
||||
return Inst<F16>(Opcode::ConvertF64S64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F64>(Opcode::ConvertF64S8, value);
|
||||
case 16:
|
||||
return Inst<F64>(Opcode::ConvertF64S16, value);
|
||||
case 32:
|
||||
return Inst<F64>(Opcode::ConvertF64S32, value);
|
||||
case 64:
|
||||
return Inst<F64>(Opcode::ConvertF64S64, value);
|
||||
}
|
||||
default:
|
||||
throw InvalidArgument("Invalid destination bitsize {}", bitsize);
|
||||
break;
|
||||
}
|
||||
throw InvalidArgument("Invalid bit size combination dst={} src={}", dest_bitsize, src_bitsize);
|
||||
}
|
||||
|
||||
F16F32F64 IREmitter::ConvertUToF(size_t bitsize, const U32U64& value) {
|
||||
switch (bitsize) {
|
||||
F16F32F64 IREmitter::ConvertUToF(size_t dest_bitsize, size_t src_bitsize, const Value& value) {
|
||||
switch (dest_bitsize) {
|
||||
case 16:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F16>(Opcode::ConvertF16U8, value);
|
||||
case 16:
|
||||
return Inst<F16>(Opcode::ConvertF16U16, value);
|
||||
case 32:
|
||||
return Inst<F16>(Opcode::ConvertF16U32, value);
|
||||
case Type::U64:
|
||||
case 64:
|
||||
return Inst<F16>(Opcode::ConvertF16U64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
}
|
||||
break;
|
||||
case 32:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F32>(Opcode::ConvertF32U8, value);
|
||||
case 16:
|
||||
return Inst<F32>(Opcode::ConvertF32U16, value);
|
||||
case 32:
|
||||
return Inst<F32>(Opcode::ConvertF32U32, value);
|
||||
case Type::U64:
|
||||
case 64:
|
||||
return Inst<F32>(Opcode::ConvertF32U64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
}
|
||||
break;
|
||||
case 64:
|
||||
switch (value.Type()) {
|
||||
case Type::U32:
|
||||
return Inst<F16>(Opcode::ConvertF64U32, value);
|
||||
case Type::U64:
|
||||
return Inst<F16>(Opcode::ConvertF64U64, value);
|
||||
default:
|
||||
ThrowInvalidType(value.Type());
|
||||
switch (src_bitsize) {
|
||||
case 8:
|
||||
return Inst<F64>(Opcode::ConvertF64U8, value);
|
||||
case 16:
|
||||
return Inst<F64>(Opcode::ConvertF64U16, value);
|
||||
case 32:
|
||||
return Inst<F64>(Opcode::ConvertF64U32, value);
|
||||
case 64:
|
||||
return Inst<F64>(Opcode::ConvertF64U64, value);
|
||||
}
|
||||
default:
|
||||
throw InvalidArgument("Invalid destination bitsize {}", bitsize);
|
||||
break;
|
||||
}
|
||||
throw InvalidArgument("Invalid bit size combination dst={} src={}", dest_bitsize, src_bitsize);
|
||||
}
|
||||
|
||||
F16F32F64 IREmitter::ConvertIToF(size_t bitsize, bool is_signed, const U32U64& value) {
|
||||
F16F32F64 IREmitter::ConvertIToF(size_t dest_bitsize, size_t src_bitsize, bool is_signed,
|
||||
const Value& value) {
|
||||
if (is_signed) {
|
||||
return ConvertSToF(bitsize, value);
|
||||
return ConvertSToF(dest_bitsize, src_bitsize, value);
|
||||
} else {
|
||||
return ConvertUToF(bitsize, value);
|
||||
return ConvertUToF(dest_bitsize, src_bitsize, value);
|
||||
}
|
||||
}
|
||||
|
||||
|
@ -29,6 +29,7 @@ public:
|
||||
[[nodiscard]] U32 Imm32(s32 value) const;
|
||||
[[nodiscard]] F32 Imm32(f32 value) const;
|
||||
[[nodiscard]] U64 Imm64(u64 value) const;
|
||||
[[nodiscard]] U64 Imm64(s64 value) const;
|
||||
[[nodiscard]] F64 Imm64(f64 value) const;
|
||||
|
||||
void Branch(Block* label);
|
||||
@ -170,7 +171,7 @@ public:
|
||||
[[nodiscard]] U32U64 ISub(const U32U64& a, const U32U64& b);
|
||||
[[nodiscard]] U32 IMul(const U32& a, const U32& b);
|
||||
[[nodiscard]] U32U64 INeg(const U32U64& value);
|
||||
[[nodiscard]] U32 IAbs(const U32& value);
|
||||
[[nodiscard]] U32U64 IAbs(const U32U64& value);
|
||||
[[nodiscard]] U32U64 ShiftLeftLogical(const U32U64& base, const U32& shift);
|
||||
[[nodiscard]] U32U64 ShiftRightLogical(const U32U64& base, const U32& shift);
|
||||
[[nodiscard]] U32U64 ShiftRightArithmetic(const U32U64& base, const U32& shift);
|
||||
@ -193,7 +194,7 @@ public:
|
||||
[[nodiscard]] U32 UMax(const U32& a, const U32& b);
|
||||
|
||||
[[nodiscard]] U1 ILessThan(const U32& lhs, const U32& rhs, bool is_signed);
|
||||
[[nodiscard]] U1 IEqual(const U32& lhs, const U32& rhs);
|
||||
[[nodiscard]] U1 IEqual(const U32U64& lhs, const U32U64& rhs);
|
||||
[[nodiscard]] U1 ILessThanEqual(const U32& lhs, const U32& rhs, bool is_signed);
|
||||
[[nodiscard]] U1 IGreaterThan(const U32& lhs, const U32& rhs, bool is_signed);
|
||||
[[nodiscard]] U1 INotEqual(const U32& lhs, const U32& rhs);
|
||||
@ -207,9 +208,12 @@ public:
|
||||
[[nodiscard]] U32U64 ConvertFToS(size_t bitsize, const F16F32F64& value);
|
||||
[[nodiscard]] U32U64 ConvertFToU(size_t bitsize, const F16F32F64& value);
|
||||
[[nodiscard]] U32U64 ConvertFToI(size_t bitsize, bool is_signed, const F16F32F64& value);
|
||||
[[nodiscard]] F16F32F64 ConvertSToF(size_t bitsize, const U32U64& value);
|
||||
[[nodiscard]] F16F32F64 ConvertUToF(size_t bitsize, const U32U64& value);
|
||||
[[nodiscard]] F16F32F64 ConvertIToF(size_t bitsize, bool is_signed, const U32U64& value);
|
||||
[[nodiscard]] F16F32F64 ConvertSToF(size_t dest_bitsize, size_t src_bitsize,
|
||||
const Value& value);
|
||||
[[nodiscard]] F16F32F64 ConvertUToF(size_t dest_bitsize, size_t src_bitsize,
|
||||
const Value& value);
|
||||
[[nodiscard]] F16F32F64 ConvertIToF(size_t dest_bitsize, size_t src_bitsize, bool is_signed,
|
||||
const Value& value);
|
||||
|
||||
[[nodiscard]] U32U64 UConvert(size_t result_bitsize, const U32U64& value);
|
||||
[[nodiscard]] F16F32F64 FPConvert(size_t result_bitsize, const F16F32F64& value);
|
||||
|
@ -247,6 +247,7 @@ OPCODE(IMul32, U32, U32,
|
||||
OPCODE(INeg32, U32, U32, )
|
||||
OPCODE(INeg64, U64, U64, )
|
||||
OPCODE(IAbs32, U32, U32, )
|
||||
OPCODE(IAbs64, U64, U64, )
|
||||
OPCODE(ShiftLeftLogical32, U32, U32, U32, )
|
||||
OPCODE(ShiftLeftLogical64, U64, U64, U32, )
|
||||
OPCODE(ShiftRightLogical32, U32, U32, U32, )
|
||||
@ -311,16 +312,28 @@ OPCODE(ConvertF16F32, F16, F32,
|
||||
OPCODE(ConvertF32F16, F32, F16, )
|
||||
OPCODE(ConvertF32F64, F32, F64, )
|
||||
OPCODE(ConvertF64F32, F64, F32, )
|
||||
OPCODE(ConvertF16S8, F16, U32, )
|
||||
OPCODE(ConvertF16S16, F16, U32, )
|
||||
OPCODE(ConvertF16S32, F16, U32, )
|
||||
OPCODE(ConvertF16S64, F16, U64, )
|
||||
OPCODE(ConvertF16U8, F16, U32, )
|
||||
OPCODE(ConvertF16U16, F16, U32, )
|
||||
OPCODE(ConvertF16U32, F16, U32, )
|
||||
OPCODE(ConvertF16U64, F16, U64, )
|
||||
OPCODE(ConvertF32S8, F32, U32, )
|
||||
OPCODE(ConvertF32S16, F32, U32, )
|
||||
OPCODE(ConvertF32S32, F32, U32, )
|
||||
OPCODE(ConvertF32S64, F32, U64, )
|
||||
OPCODE(ConvertF32U8, F32, U32, )
|
||||
OPCODE(ConvertF32U16, F32, U32, )
|
||||
OPCODE(ConvertF32U32, F32, U32, )
|
||||
OPCODE(ConvertF32U64, F32, U64, )
|
||||
OPCODE(ConvertF64S8, F64, U32, )
|
||||
OPCODE(ConvertF64S16, F64, U32, )
|
||||
OPCODE(ConvertF64S32, F64, U32, )
|
||||
OPCODE(ConvertF64S64, F64, U64, )
|
||||
OPCODE(ConvertF64U8, F64, U32, )
|
||||
OPCODE(ConvertF64U16, F64, U32, )
|
||||
OPCODE(ConvertF64U32, F64, U32, )
|
||||
OPCODE(ConvertF64U64, F64, U64, )
|
||||
|
||||
|
@ -121,6 +121,22 @@ IR::F64 TranslatorVisitor::GetDoubleCbuf(u64 insn) {
|
||||
return ir.PackDouble2x32(ir.CompositeConstruct(lower_bits, value));
|
||||
}
|
||||
|
||||
IR::U64 TranslatorVisitor::GetPackedCbuf(u64 insn) {
|
||||
union {
|
||||
u64 raw;
|
||||
BitField<20, 1, u64> unaligned;
|
||||
} const cbuf{insn};
|
||||
|
||||
if (cbuf.unaligned != 0) {
|
||||
throw NotImplementedException("Unaligned packed constant buffer read");
|
||||
}
|
||||
const auto [binding, lower_offset]{CbufAddr(insn)};
|
||||
const IR::U32 upper_offset{ir.Imm32(lower_offset.U32() + 4)};
|
||||
const IR::U32 lower_value{ir.GetCbuf(binding, lower_offset)};
|
||||
const IR::U32 upper_value{ir.GetCbuf(binding, upper_offset)};
|
||||
return ir.PackUint2x32(ir.CompositeConstruct(lower_value, upper_value));
|
||||
}
|
||||
|
||||
IR::U32 TranslatorVisitor::GetImm20(u64 insn) {
|
||||
union {
|
||||
u64 raw;
|
||||
@ -158,6 +174,11 @@ IR::F64 TranslatorVisitor::GetDoubleImm20(u64 insn) {
|
||||
return ir.Imm64(Common::BitCast<f64>(value | sign_bit));
|
||||
}
|
||||
|
||||
IR::U64 TranslatorVisitor::GetPackedImm20(u64 insn) {
|
||||
const s64 value{GetImm20(insn).U32()};
|
||||
return ir.Imm64(static_cast<u64>(static_cast<s64>(value) << 32));
|
||||
}
|
||||
|
||||
IR::U32 TranslatorVisitor::GetImm32(u64 insn) {
|
||||
union {
|
||||
u64 raw;
|
||||
|
@ -356,10 +356,12 @@ public:
|
||||
[[nodiscard]] IR::U32 GetCbuf(u64 insn);
|
||||
[[nodiscard]] IR::F32 GetFloatCbuf(u64 insn);
|
||||
[[nodiscard]] IR::F64 GetDoubleCbuf(u64 insn);
|
||||
[[nodiscard]] IR::U64 GetPackedCbuf(u64 insn);
|
||||
|
||||
[[nodiscard]] IR::U32 GetImm20(u64 insn);
|
||||
[[nodiscard]] IR::F32 GetFloatImm20(u64 insn);
|
||||
[[nodiscard]] IR::F64 GetDoubleImm20(u64 insn);
|
||||
[[nodiscard]] IR::U64 GetPackedImm20(u64 insn);
|
||||
|
||||
[[nodiscard]] IR::U32 GetImm32(u64 insn);
|
||||
[[nodiscard]] IR::F32 GetFloatImm32(u64 insn);
|
||||
|
@ -0,0 +1,173 @@
|
||||
// Copyright 2021 yuzu Emulator Project
|
||||
// Licensed under GPLv2 or any later version
|
||||
// Refer to the license.txt file included.
|
||||
|
||||
#include "common/bit_field.h"
|
||||
#include "common/common_types.h"
|
||||
#include "shader_recompiler/frontend/maxwell/translate/impl/common_encoding.h"
|
||||
#include "shader_recompiler/frontend/maxwell/translate/impl/impl.h"
|
||||
|
||||
namespace Shader::Maxwell {
|
||||
namespace {
|
||||
enum class FloatFormat : u64 {
|
||||
F16 = 1,
|
||||
F32 = 2,
|
||||
F64 = 3,
|
||||
};
|
||||
|
||||
enum class IntFormat : u64 {
|
||||
U8 = 0,
|
||||
U16 = 1,
|
||||
U32 = 2,
|
||||
U64 = 3,
|
||||
};
|
||||
|
||||
union Encoding {
|
||||
u64 raw;
|
||||
BitField<0, 8, IR::Reg> dest_reg;
|
||||
BitField<8, 2, FloatFormat> float_format;
|
||||
BitField<10, 2, IntFormat> int_format;
|
||||
BitField<13, 1, u64> is_signed;
|
||||
BitField<39, 2, FpRounding> fp_rounding;
|
||||
BitField<41, 2, u64> selector;
|
||||
BitField<47, 1, u64> cc;
|
||||
BitField<45, 1, u64> neg;
|
||||
BitField<49, 1, u64> abs;
|
||||
};
|
||||
|
||||
bool Is64(u64 insn) {
|
||||
return Encoding{insn}.int_format == IntFormat::U64;
|
||||
}
|
||||
|
||||
int BitSize(FloatFormat format) {
|
||||
switch (format) {
|
||||
case FloatFormat::F16:
|
||||
return 16;
|
||||
case FloatFormat::F32:
|
||||
return 32;
|
||||
case FloatFormat::F64:
|
||||
return 64;
|
||||
}
|
||||
throw NotImplementedException("Invalid float format {}", format);
|
||||
}
|
||||
|
||||
IR::U32 SmallAbs(TranslatorVisitor& v, const IR::U32& value, int bitsize) {
|
||||
const IR::U32 least_value{v.ir.Imm32(-(1 << (bitsize - 1)))};
|
||||
const IR::U32 mask{v.ir.ShiftRightArithmetic(value, v.ir.Imm32(bitsize - 1))};
|
||||
const IR::U32 absolute{v.ir.BitwiseXor(v.ir.IAdd(value, mask), mask)};
|
||||
const IR::U1 is_least{v.ir.IEqual(value, least_value)};
|
||||
return IR::U32{v.ir.Select(is_least, value, absolute)};
|
||||
}
|
||||
|
||||
void I2F(TranslatorVisitor& v, u64 insn, IR::U32U64 src) {
|
||||
const Encoding i2f{insn};
|
||||
if (i2f.cc != 0) {
|
||||
throw NotImplementedException("CC");
|
||||
}
|
||||
const bool is_signed{i2f.is_signed != 0};
|
||||
int src_bitsize{};
|
||||
switch (i2f.int_format) {
|
||||
case IntFormat::U8:
|
||||
src = v.ir.BitFieldExtract(src, v.ir.Imm32(static_cast<u32>(i2f.selector) * 8),
|
||||
v.ir.Imm32(8), is_signed);
|
||||
if (i2f.abs != 0) {
|
||||
src = SmallAbs(v, src, 8);
|
||||
}
|
||||
src_bitsize = 8;
|
||||
break;
|
||||
case IntFormat::U16:
|
||||
if (i2f.selector == 1 || i2f.selector == 3) {
|
||||
throw NotImplementedException("Invalid U16 selector {}", i2f.selector.Value());
|
||||
}
|
||||
src = v.ir.BitFieldExtract(src, v.ir.Imm32(static_cast<u32>(i2f.selector) * 8),
|
||||
v.ir.Imm32(16), is_signed);
|
||||
if (i2f.abs != 0) {
|
||||
src = SmallAbs(v, src, 16);
|
||||
}
|
||||
src_bitsize = 16;
|
||||
break;
|
||||
case IntFormat::U32:
|
||||
case IntFormat::U64:
|
||||
if (i2f.selector != 0) {
|
||||
throw NotImplementedException("Unexpected selector {}", i2f.selector.Value());
|
||||
}
|
||||
if (i2f.abs != 0 && is_signed) {
|
||||
src = v.ir.IAbs(src);
|
||||
}
|
||||
src_bitsize = i2f.int_format == IntFormat::U64 ? 64 : 32;
|
||||
break;
|
||||
}
|
||||
const int conversion_src_bitsize{i2f.int_format == IntFormat::U64 ? 64 : 32};
|
||||
const int dst_bitsize{BitSize(i2f.float_format)};
|
||||
IR::F16F32F64 value{v.ir.ConvertIToF(dst_bitsize, conversion_src_bitsize, is_signed, src)};
|
||||
if (i2f.neg != 0) {
|
||||
if (i2f.abs != 0 || !is_signed) {
|
||||
// We know the value is positive
|
||||
value = v.ir.FPNeg(value);
|
||||
} else {
|
||||
// Only negate if the input isn't the lowest value
|
||||
IR::U1 is_least;
|
||||
if (src_bitsize == 64) {
|
||||
is_least = v.ir.IEqual(src, v.ir.Imm64(std::numeric_limits<s64>::min()));
|
||||
} else {
|
||||
const IR::U32 least_value{v.ir.Imm32(-(1 << (src_bitsize - 1)))};
|
||||
is_least = v.ir.IEqual(src, least_value);
|
||||
}
|
||||
value = IR::F16F32F64{v.ir.Select(is_least, value, v.ir.FPNeg(value))};
|
||||
}
|
||||
}
|
||||
switch (i2f.float_format) {
|
||||
case FloatFormat::F16: {
|
||||
const IR::F16 zero{v.ir.FPConvert(16, v.ir.Imm32(0.0f))};
|
||||
v.X(i2f.dest_reg, v.ir.PackFloat2x16(v.ir.CompositeConstruct(value, zero)));
|
||||
break;
|
||||
}
|
||||
case FloatFormat::F32:
|
||||
v.F(i2f.dest_reg, value);
|
||||
break;
|
||||
case FloatFormat::F64: {
|
||||
if (!IR::IsAligned(i2f.dest_reg, 2)) {
|
||||
throw NotImplementedException("Unaligned destination {}", i2f.dest_reg.Value());
|
||||
}
|
||||
const IR::Value vector{v.ir.UnpackDouble2x32(value)};
|
||||
for (int i = 0; i < 2; ++i) {
|
||||
v.X(i2f.dest_reg + i, IR::U32{v.ir.CompositeExtract(vector, i)});
|
||||
}
|
||||
break;
|
||||
}
|
||||
default:
|
||||
throw NotImplementedException("Invalid float format {}", i2f.float_format.Value());
|
||||
}
|
||||
}
|
||||
} // Anonymous namespace
|
||||
|
||||
void TranslatorVisitor::I2F_reg(u64 insn) {
|
||||
if (Is64(insn)) {
|
||||
union {
|
||||
u64 raw;
|
||||
BitField<20, 8, IR::Reg> reg;
|
||||
} const value{insn};
|
||||
const IR::Value regs{ir.CompositeConstruct(ir.GetReg(value.reg), ir.GetReg(value.reg + 1))};
|
||||
I2F(*this, insn, ir.PackUint2x32(regs));
|
||||
} else {
|
||||
I2F(*this, insn, GetReg20(insn));
|
||||
}
|
||||
}
|
||||
|
||||
void TranslatorVisitor::I2F_cbuf(u64 insn) {
|
||||
if (Is64(insn)) {
|
||||
I2F(*this, insn, GetPackedCbuf(insn));
|
||||
} else {
|
||||
I2F(*this, insn, GetCbuf(insn));
|
||||
}
|
||||
}
|
||||
|
||||
void TranslatorVisitor::I2F_imm(u64 insn) {
|
||||
if (Is64(insn)) {
|
||||
I2F(*this, insn, GetPackedImm20(insn));
|
||||
} else {
|
||||
I2F(*this, insn, GetImm20(insn));
|
||||
}
|
||||
}
|
||||
|
||||
} // namespace Shader::Maxwell
|
@ -241,18 +241,6 @@ void TranslatorVisitor::HSETP2_imm(u64) {
|
||||
ThrowNotImplemented(Opcode::HSETP2_imm);
|
||||
}
|
||||
|
||||
void TranslatorVisitor::I2F_reg(u64) {
|
||||
ThrowNotImplemented(Opcode::I2F_reg);
|
||||
}
|
||||
|
||||
void TranslatorVisitor::I2F_cbuf(u64) {
|
||||
ThrowNotImplemented(Opcode::I2F_cbuf);
|
||||
}
|
||||
|
||||
void TranslatorVisitor::I2F_imm(u64) {
|
||||
ThrowNotImplemented(Opcode::I2F_imm);
|
||||
}
|
||||
|
||||
void TranslatorVisitor::IDE(u64) {
|
||||
ThrowNotImplemented(Opcode::IDE);
|
||||
}
|
||||
|
@ -56,7 +56,7 @@ Shader::TextureType GetType(TextureType type, bool dc) {
|
||||
}
|
||||
|
||||
IR::Value MakeCoords(TranslatorVisitor& v, IR::Reg reg, TextureType type) {
|
||||
const auto read_array{[&]() -> IR::F32 { return v.ir.ConvertUToF(32, v.X(reg)); }};
|
||||
const auto read_array{[&]() -> IR::F32 { return v.ir.ConvertUToF(32, 16, v.X(reg)); }};
|
||||
switch (type) {
|
||||
case TextureType::_1D:
|
||||
return v.F(reg);
|
||||
|
@ -65,7 +65,7 @@ IR::Value Composite(TranslatorVisitor& v, Args... regs) {
|
||||
}
|
||||
|
||||
IR::F32 ReadArray(TranslatorVisitor& v, const IR::U32& value) {
|
||||
return v.ir.ConvertUToF(32, v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(16)));
|
||||
return v.ir.ConvertUToF(32, 16, v.ir.BitFieldExtract(value, v.ir.Imm32(0), v.ir.Imm32(16)));
|
||||
}
|
||||
|
||||
IR::Value Sample(TranslatorVisitor& v, u64 insn) {
|
||||
|
@ -79,6 +79,14 @@ void VisitUsages(Info& info, IR::Inst& inst) {
|
||||
case IR::Opcode::ConvertU16F16:
|
||||
case IR::Opcode::ConvertU32F16:
|
||||
case IR::Opcode::ConvertU64F16:
|
||||
case IR::Opcode::ConvertF16S8:
|
||||
case IR::Opcode::ConvertF16S16:
|
||||
case IR::Opcode::ConvertF16S32:
|
||||
case IR::Opcode::ConvertF16S64:
|
||||
case IR::Opcode::ConvertF16U8:
|
||||
case IR::Opcode::ConvertF16U16:
|
||||
case IR::Opcode::ConvertF16U32:
|
||||
case IR::Opcode::ConvertF16U64:
|
||||
case IR::Opcode::FPAbs16:
|
||||
case IR::Opcode::FPAdd16:
|
||||
case IR::Opcode::FPCeil16:
|
||||
@ -105,6 +113,14 @@ void VisitUsages(Info& info, IR::Inst& inst) {
|
||||
case IR::Opcode::FPRoundEven64:
|
||||
case IR::Opcode::FPSaturate64:
|
||||
case IR::Opcode::FPTrunc64:
|
||||
case IR::Opcode::ConvertF64S8:
|
||||
case IR::Opcode::ConvertF64S16:
|
||||
case IR::Opcode::ConvertF64S32:
|
||||
case IR::Opcode::ConvertF64S64:
|
||||
case IR::Opcode::ConvertF64U8:
|
||||
case IR::Opcode::ConvertF64U16:
|
||||
case IR::Opcode::ConvertF64U32:
|
||||
case IR::Opcode::ConvertF64U64:
|
||||
info.uses_fp64 = true;
|
||||
break;
|
||||
default:
|
||||
@ -123,6 +139,12 @@ void VisitUsages(Info& info, IR::Inst& inst) {
|
||||
case IR::Opcode::WriteStorageU8:
|
||||
case IR::Opcode::WriteStorageS8:
|
||||
case IR::Opcode::SelectU8:
|
||||
case IR::Opcode::ConvertF16S8:
|
||||
case IR::Opcode::ConvertF16U8:
|
||||
case IR::Opcode::ConvertF32S8:
|
||||
case IR::Opcode::ConvertF32U8:
|
||||
case IR::Opcode::ConvertF64S8:
|
||||
case IR::Opcode::ConvertF64U8:
|
||||
info.uses_int8 = true;
|
||||
break;
|
||||
default:
|
||||
@ -149,6 +171,12 @@ void VisitUsages(Info& info, IR::Inst& inst) {
|
||||
case IR::Opcode::ConvertU16F16:
|
||||
case IR::Opcode::ConvertU16F32:
|
||||
case IR::Opcode::ConvertU16F64:
|
||||
case IR::Opcode::ConvertF16S16:
|
||||
case IR::Opcode::ConvertF16U16:
|
||||
case IR::Opcode::ConvertF32S16:
|
||||
case IR::Opcode::ConvertF32U16:
|
||||
case IR::Opcode::ConvertF64S16:
|
||||
case IR::Opcode::ConvertF64U16:
|
||||
info.uses_int16 = true;
|
||||
break;
|
||||
default:
|
||||
|
@ -70,6 +70,22 @@ IR::Opcode Replace(IR::Opcode op) {
|
||||
return IR::Opcode::Identity;
|
||||
case IR::Opcode::ConvertF16F32:
|
||||
return IR::Opcode::Identity;
|
||||
case IR::Opcode::ConvertF16S8:
|
||||
return IR::Opcode::ConvertF32S8;
|
||||
case IR::Opcode::ConvertF16S16:
|
||||
return IR::Opcode::ConvertF32S16;
|
||||
case IR::Opcode::ConvertF16S32:
|
||||
return IR::Opcode::ConvertF32S32;
|
||||
case IR::Opcode::ConvertF16S64:
|
||||
return IR::Opcode::ConvertF32S64;
|
||||
case IR::Opcode::ConvertF16U8:
|
||||
return IR::Opcode::ConvertF32U8;
|
||||
case IR::Opcode::ConvertF16U16:
|
||||
return IR::Opcode::ConvertF32U16;
|
||||
case IR::Opcode::ConvertF16U32:
|
||||
return IR::Opcode::ConvertF32U32;
|
||||
case IR::Opcode::ConvertF16U64:
|
||||
return IR::Opcode::ConvertF32U64;
|
||||
default:
|
||||
return op;
|
||||
}
|
||||
|
@ -227,6 +227,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
|
||||
update_descriptor_queue{update_descriptor_queue_}, render_pass_cache{render_pass_cache_},
|
||||
buffer_cache{buffer_cache_}, texture_cache{texture_cache_} {
|
||||
const auto& float_control{device.FloatControlProperties()};
|
||||
const VkDriverIdKHR driver_id{device.GetDriverID()};
|
||||
profile = Shader::Profile{
|
||||
.unified_descriptor_binding = true,
|
||||
.support_float_controls = true,
|
||||
@ -242,7 +243,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::GPU& gpu_,
|
||||
float_control.shaderSignedZeroInfNanPreserveFloat16 != VK_FALSE,
|
||||
.support_fp32_signed_zero_nan_preserve =
|
||||
float_control.shaderSignedZeroInfNanPreserveFloat32 != VK_FALSE,
|
||||
.has_broken_spirv_clamp = true, // TODO: is_intel
|
||||
.has_broken_spirv_clamp = driver_id == VK_DRIVER_ID_INTEL_PROPRIETARY_WINDOWS_KHR,
|
||||
};
|
||||
}
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user