sneedmca/suyu
1
0
Fork 0
mirror of https://git.suyu.dev/suyu/suyu synced 2024-12-23 18:02:46 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

shader_decode: Use BitfieldExtract instead of shift + and

Browse source
This commit is contained in:
ReinUsesLisp 2018-12-26 02:58:47 -03:00
parent 52223313b1
commit 2faad9bf23
8 changed files with 37 additions and 48 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
src/video_core/renderer_opengl/gl_shader_decompiler.cpp
View file

@ -985,6 +985,11 @@ private:
Type::Int);
}
template <Type type>
std::string BitfieldExtract(Operation operation) {
return GenerateTernary(operation, "bitfieldExtract", type, type, Type::Int, Type::Int);
}
template <Type type>
std::string BitCount(Operation operation) {
return GenerateUnary(operation, "bitCount", type, type, false);
@ -1369,6 +1374,7 @@ private:
&GLSLDecompiler::BitwiseXor<Type::Int>,
&GLSLDecompiler::BitwiseNot<Type::Int>,
&GLSLDecompiler::BitfieldInsert<Type::Int>,
&GLSLDecompiler::BitfieldExtract<Type::Int>,
&GLSLDecompiler::BitCount<Type::Int>,
&GLSLDecompiler::Add<Type::Uint>,
@ -1386,6 +1392,7 @@ private:
&GLSLDecompiler::BitwiseXor<Type::Uint>,
&GLSLDecompiler::BitwiseNot<Type::Uint>,
&GLSLDecompiler::BitfieldInsert<Type::Uint>,
&GLSLDecompiler::BitfieldExtract<Type::Uint>,
&GLSLDecompiler::BitCount<Type::Uint>,
&GLSLDecompiler::Add<Type::HalfFloat>,

5
src/video_core/shader/decode/arithmetic_integer.cpp
View file

@ -57,10 +57,9 @@ u32 ShaderIR::DecodeArithmeticInteger(BasicBlock& bb, u32 pc) {
case IAdd3Height::None:
return value;
case IAdd3Height::LowerHalfWord:
return Operation(OperationCode::IBitwiseAnd, NO_PRECISE, value, Immediate(0xffff));
return BitfieldExtract(value, 0, 16);
case IAdd3Height::UpperHalfWord:
return Operation(OperationCode::ILogicalShiftRight, NO_PRECISE, value,
Immediate(16));
return BitfieldExtract(value, 16, 16);
default:
UNIMPLEMENTED_MSG("Unhandled IADD3 height: {}", static_cast<u32>(height));
return Immediate(0);

9
src/video_core/shader/decode/bfi.cpp
View file

@ -28,13 +28,8 @@ u32 ShaderIR::DecodeBfi(BasicBlock& bb, u32 pc) {
}
}();
const Node insert = GetRegister(instr.gpr8);
const Node offset =
Operation(OperationCode::UBitwiseAnd, NO_PRECISE, packed_shift, Immediate(0xff));
Node bits =
Operation(OperationCode::ULogicalShiftRight, NO_PRECISE, packed_shift, Immediate(8));
bits = Operation(OperationCode::UBitwiseAnd, NO_PRECISE, bits, Immediate(0xff));
const Node offset = BitfieldExtract(packed_shift, 0, 8);
const Node bits = BitfieldExtract(packed_shift, 8, 8);
const Node value =
Operation(OperationCode::UBitfieldInsert, PRECISE, base, insert, offset, bits);

12
src/video_core/shader/decode/register_set_predicate.cpp
View file

@ -27,20 +27,18 @@ u32 ShaderIR::DecodeRegisterSetPredicate(BasicBlock& bb, u32 pc) {
return Immediate(static_cast<u32>(instr.r2p.immediate_mask));
}
}();
const Node mask =
Operation(OperationCode::ULogicalShiftRight, NO_PRECISE, GetRegister(instr.gpr8),
Immediate(static_cast<u32>(instr.r2p.byte)));
const Node mask = GetRegister(instr.gpr8);
const auto offset = static_cast<u32>(instr.r2p.byte) * 8;
constexpr u32 programmable_preds = 7;
for (u64 pred = 0; pred < programmable_preds; ++pred) {
const Node shift = Immediate(1u << static_cast<u32>(pred));
const auto shift = static_cast<u32>(pred);
const Node apply_compare =
Operation(OperationCode::UBitwiseAnd, NO_PRECISE, apply_mask, shift);
const Node apply_compare = BitfieldExtract(apply_mask, shift, 1);
const Node condition =
Operation(OperationCode::LogicalUNotEqual, apply_compare, Immediate(0));
const Node value_compare = Operation(OperationCode::UBitwiseAnd, NO_PRECISE, mask, shift);
const Node value_compare = BitfieldExtract(mask, offset + shift, 1);
const Node value = Operation(OperationCode::LogicalUNotEqual, value_compare, Immediate(0));
const Node code = Operation(OperationCode::LogicalAssign, GetPredicate(pred), value);

12
src/video_core/shader/decode/video.cpp
View file

@ -88,21 +88,15 @@ u32 ShaderIR::DecodeVideo(BasicBlock& bb, u32 pc) {
Node ShaderIR::GetVideoOperand(Node op, bool is_chunk, bool is_signed,
Tegra::Shader::VideoType type, u64 byte_height) {
if (!is_chunk) {
const auto offset = static_cast<u32>(byte_height * 8);
const Node shift = SignedOperation(OperationCode::ILogicalShiftRight, is_signed, NO_PRECISE,
op, Immediate(offset));
return SignedOperation(OperationCode::IBitwiseAnd, is_signed, NO_PRECISE, shift,
Immediate(0xff));
return BitfieldExtract(op, static_cast<u32>(byte_height * 8), 8);
}
const Node zero = Immediate(0);
switch (type) {
case Tegra::Shader::VideoType::Size16_Low:
return SignedOperation(OperationCode::IBitwiseAnd, is_signed, NO_PRECISE, op,
Immediate(0xffff));
return BitfieldExtract(op, 0, 16);
case Tegra::Shader::VideoType::Size16_High:
return SignedOperation(OperationCode::ILogicalShiftRight, is_signed, NO_PRECISE, op,
Immediate(16));
return BitfieldExtract(op, 16, 16);
case Tegra::Shader::VideoType::Size32:
// TODO(Rodrigo): From my hardware tests it becomes a bit "mad" when this type is used
// (1 * 1 + 0 == 0x5b800000). Until a better explanation is found: abort.

26
src/video_core/shader/decode/xmad.cpp
View file

@ -47,22 +47,10 @@ u32 ShaderIR::DecodeXmad(BasicBlock& bb, u32 pc) {
return {false, Immediate(0), Immediate(0)};
}();
if (instr.xmad.high_a) {
op_a = SignedOperation(OperationCode::ILogicalShiftRight, is_signed_a, NO_PRECISE, op_a,
Immediate(16));
} else {
op_a = SignedOperation(OperationCode::IBitwiseAnd, is_signed_a, NO_PRECISE, op_a,
Immediate(0xffff));
}
op_a = BitfieldExtract(op_a, instr.xmad.high_a ? 16 : 0, 16);
const Node original_b = op_b;
if (instr.xmad.high_b) {
op_b = SignedOperation(OperationCode::ILogicalShiftRight, is_signed_b, NO_PRECISE, op_a,
Immediate(16));
} else {
op_b = SignedOperation(OperationCode::IBitwiseAnd, is_signed_b, NO_PRECISE, op_b,
Immediate(0xffff));
}
op_b = BitfieldExtract(op_b, instr.xmad.high_b ? 16 : 0, 16);
// TODO(Rodrigo): Use an appropiate sign for this operation
Node product = Operation(OperationCode::IMul, NO_PRECISE, op_a, op_b);
@ -75,11 +63,9 @@ u32 ShaderIR::DecodeXmad(BasicBlock& bb, u32 pc) {
case Tegra::Shader::XmadMode::None:
return op_c;
case Tegra::Shader::XmadMode::CLo:
return SignedOperation(OperationCode::IBitwiseAnd, is_signed_c, NO_PRECISE, op_c,
Immediate(0xffff));
return BitfieldExtract(op_c, 0, 16);
case Tegra::Shader::XmadMode::CHi:
return SignedOperation(OperationCode::ILogicalShiftRight, is_signed_c, NO_PRECISE, op_c,
Immediate(16));
return BitfieldExtract(op_c, 16, 16);
case Tegra::Shader::XmadMode::CBcc: {
const Node shifted_b = SignedOperation(OperationCode::ILogicalShiftLeft, is_signed_b,
NO_PRECISE, original_b, Immediate(16));
@ -94,9 +80,9 @@ u32 ShaderIR::DecodeXmad(BasicBlock& bb, u32 pc) {
// TODO(Rodrigo): Use an appropiate sign for this operation
Node sum = Operation(OperationCode::IAdd, product, op_c);
if (is_merge) {
const Node a = Operation(OperationCode::IBitwiseAnd, NO_PRECISE, sum, Immediate(0xffff));
const Node a = BitfieldExtract(sum, 0, 16);
const Node b =
Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, original_b, Immediate(0xffff));
Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, original_b, Immediate(16));
sum = Operation(OperationCode::IBitwiseOr, NO_PRECISE, a, b);
}

5
src/video_core/shader/shader_ir.cpp
View file

@ -348,6 +348,11 @@ void ShaderIR::SetLocalMemory(BasicBlock& bb, Node address, Node value) {
bb.push_back(Operation(OperationCode::Assign, GetLocalMemory(address), value));
}
Node ShaderIR::BitfieldExtract(Node value, u32 offset, u32 bits) {
return Operation(OperationCode::UBitfieldExtract, NO_PRECISE, value, Immediate(offset),
Immediate(bits));
}
/*static*/ OperationCode ShaderIR::SignedToUnsignedCode(OperationCode operation_code,
bool is_signed) {
if (is_signed) {

9
src/video_core/shader/shader_ir.h
View file

@ -88,6 +88,7 @@ enum class OperationCode {
IBitwiseXor, /// (MetaArithmetic, int a, int b) -> int
IBitwiseNot, /// (MetaArithmetic, int a) -> int
IBitfieldInsert, /// (MetaArithmetic, int base, int insert, int offset, int bits) -> int
IBitfieldExtract, /// (MetaArithmetic, int value, int offset, int offset) -> int
IBitCount, /// (MetaArithmetic, int) -> int
UAdd, /// (MetaArithmetic, uint a, uint b) -> uint
@ -104,8 +105,9 @@ enum class OperationCode {
UBitwiseOr, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseXor, /// (MetaArithmetic, uint a, uint b) -> uint
UBitwiseNot, /// (MetaArithmetic, uint a) -> uint
UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint
UBitCount, /// (MetaArithmetic, uint) -> uint
UBitfieldInsert, /// (MetaArithmetic, uint base, uint insert, int offset, int bits) -> uint
UBitfieldExtract, /// (MetaArithmetic, uint value, int offset, int offset) -> uint
UBitCount, /// (MetaArithmetic, uint) -> uint
HAdd, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
HMul, /// (MetaHalfArithmetic, f16vec2 a, f16vec2 b) -> f16vec2
@ -689,6 +691,9 @@ private:
const Sampler& GetSampler(const Tegra::Shader::Sampler& sampler,
Tegra::Shader::TextureType type, bool is_array, bool is_shadow);
/// Extracts a sequence of bits from a node
Node BitfieldExtract(Node value, u32 offset, u32 bits);
void WriteTexsInstructionFloat(BasicBlock& bb, Tegra::Shader::Instruction instr, Node texture);
void WriteTexsInstructionHalfFloat(BasicBlock& bb, Tegra::Shader::Instruction instr,
Node texture);