shader_decode: Implement TEXS.F16
This commit is contained in:
parent
c68c13e1aa
commit
5af82a8ed4
|
@ -219,8 +219,7 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
|
|||
if (instr.texs.fp32_flag) {
|
||||
WriteTexsInstructionFloat(bb, instr, texture);
|
||||
} else {
|
||||
UNIMPLEMENTED();
|
||||
// WriteTexsInstructionHalfFloat(bb, instr, texture);
|
||||
WriteTexsInstructionHalfFloat(bb, instr, texture);
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
@ -416,39 +415,52 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler, Textu
|
|||
return *used_samplers.emplace(entry).first;
|
||||
}
|
||||
|
||||
void ShaderIR::WriteTexsInstructionFloat(BasicBlock& bb, Tegra::Shader::Instruction instr,
|
||||
Node texture) {
|
||||
void ShaderIR::WriteTexsInstructionFloat(BasicBlock& bb, Instruction instr, Node texture) {
|
||||
// TEXS has two destination registers and a swizzle. The first two elements in the swizzle
|
||||
// go into gpr0+0 and gpr0+1, and the rest goes into gpr28+0 and gpr28+1
|
||||
|
||||
MetaComponents meta;
|
||||
std::array<Node, 4> dest;
|
||||
|
||||
std::size_t written_components = 0;
|
||||
for (u32 component = 0; component < 4; ++component) {
|
||||
if (!instr.texs.IsComponentEnabled(component)) {
|
||||
continue;
|
||||
}
|
||||
meta.components_map[written_components] = static_cast<u32>(component);
|
||||
meta.components_map[meta.count] = component;
|
||||
|
||||
if (written_components < 2) {
|
||||
if (meta.count < 2) {
|
||||
// Write the first two swizzle components to gpr0 and gpr0+1
|
||||
dest[written_components] = GetRegister(instr.gpr0.Value() + written_components % 2);
|
||||
dest[meta.count] = GetRegister(instr.gpr0.Value() + meta.count % 2);
|
||||
} else {
|
||||
ASSERT(instr.texs.HasTwoDestinations());
|
||||
// Write the rest of the swizzle components to gpr28 and gpr28+1
|
||||
dest[written_components] = GetRegister(instr.gpr28.Value() + written_components % 2);
|
||||
dest[meta.count] = GetRegister(instr.gpr28.Value() + meta.count % 2);
|
||||
}
|
||||
++meta.count;
|
||||
}
|
||||
|
||||
++written_components;
|
||||
}
|
||||
|
||||
std::generate(dest.begin() + written_components, dest.end(), [&]() { return GetRegister(RZ); });
|
||||
std::generate(dest.begin() + meta.count, dest.end(), [&]() { return GetRegister(RZ); });
|
||||
|
||||
bb.push_back(Operation(OperationCode::AssignComposite, meta, texture, dest[0], dest[1], dest[2],
|
||||
dest[3]));
|
||||
}
|
||||
|
||||
void ShaderIR::WriteTexsInstructionHalfFloat(BasicBlock& bb, Instruction instr, Node texture) {
|
||||
// TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
|
||||
// float instruction).
|
||||
|
||||
MetaComponents meta;
|
||||
for (u32 component = 0; component < 4; ++component) {
|
||||
if (!instr.texs.IsComponentEnabled(component))
|
||||
continue;
|
||||
meta.components_map[meta.count++] = component;
|
||||
}
|
||||
if (meta.count == 0)
|
||||
return;
|
||||
|
||||
bb.push_back(Operation(OperationCode::AssignCompositeHalf, meta, texture,
|
||||
GetRegister(instr.gpr0), GetRegister(instr.gpr28)));
|
||||
}
|
||||
|
||||
Node ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
|
||||
TextureProcessMode process_mode, bool depth_compare, bool is_array,
|
||||
std::size_t array_offset, std::size_t bias_offset,
|
||||
|
|
|
@ -785,6 +785,31 @@ private:
|
|||
return {};
|
||||
}
|
||||
|
||||
std::string AssignCompositeHalf(Operation operation) {
|
||||
const auto& meta = std::get<MetaComponents>(operation.GetMeta());
|
||||
|
||||
const std::string composite = code.GenerateTemporal();
|
||||
code.AddLine("vec4 " + composite + " = " + Visit(operation[0]) + ';');
|
||||
|
||||
const auto ReadComponent = [&](u32 component) {
|
||||
if (component < meta.count) {
|
||||
return composite + '[' + std::to_string(meta.GetSourceComponent(component)) + ']';
|
||||
}
|
||||
return std::string("0");
|
||||
};
|
||||
|
||||
const auto dst1 = std::get<GprNode>(*operation[1]).GetIndex();
|
||||
const std::string src1 = "vec2(" + ReadComponent(0) + ", " + ReadComponent(1) + ')';
|
||||
code.AddLine(GetRegister(dst1) + " = utof(packHalf2x16(" + src1 + "))");
|
||||
|
||||
if (meta.count > 2) {
|
||||
const auto dst2 = std::get<GprNode>(*operation[2]).GetIndex();
|
||||
const std::string src2 = "vec2(" + ReadComponent(2) + ", " + ReadComponent(3) + ')';
|
||||
code.AddLine(GetRegister(dst2) + " = utof(packHalf2x16(" + src2 + "))");
|
||||
}
|
||||
return {};
|
||||
}
|
||||
|
||||
std::string Composite(Operation operation) {
|
||||
std::string value = "vec4(";
|
||||
for (std::size_t i = 0; i < 4; ++i) {
|
||||
|
@ -1302,6 +1327,7 @@ private:
|
|||
static constexpr OperationDecompilersArray operation_decompilers = {
|
||||
&GLSLDecompiler::Assign,
|
||||
&GLSLDecompiler::AssignComposite,
|
||||
&GLSLDecompiler::AssignCompositeHalf,
|
||||
|
||||
&GLSLDecompiler::Composite,
|
||||
&GLSLDecompiler::Select,
|
||||
|
|
|
@ -46,6 +46,7 @@ constexpr u32 RZ = 0xff;
|
|||
enum class OperationCode {
|
||||
Assign, /// (float& dest, float src) -> void
|
||||
AssignComposite, /// (MetaComponents, float4 src, float&[4] dst) -> void
|
||||
AssignCompositeHalf, /// (MetaComponents, float4 src, float&[2] dst) -> void
|
||||
|
||||
Composite, /// (float[4] values) -> float4
|
||||
Select, /// (MetaArithmetic, bool pred, float a, float b) -> float
|
||||
|
@ -279,6 +280,7 @@ struct MetaTexture {
|
|||
|
||||
struct MetaComponents {
|
||||
std::array<u32, 4> components_map{};
|
||||
u32 count{};
|
||||
|
||||
u32 GetSourceComponent(u32 dest_index) const {
|
||||
return components_map[dest_index];
|
||||
|
@ -692,6 +694,8 @@ private:
|
|||
Tegra::Shader::TextureType type, bool is_array, bool is_shadow);
|
||||
|
||||
void WriteTexsInstructionFloat(BasicBlock& bb, Tegra::Shader::Instruction instr, Node texture);
|
||||
void WriteTexsInstructionHalfFloat(BasicBlock& bb, Tegra::Shader::Instruction instr,
|
||||
Node texture);
|
||||
|
||||
Node GetTexCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
|
||||
Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
|
||||
|
|
Loading…
Reference in New Issue