1
0
Fork 0
forked from suyu/suyu

shader_ir: Fixup TEX and TEXS and partially fix TLD4 decompiling

This commit is contained in:
ReinUsesLisp 2018-12-22 01:20:57 -03:00
parent 2d9136cec6
commit 03e088a4f4
3 changed files with 71 additions and 59 deletions

View file

@ -183,9 +183,6 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
const Node texture = GetTexCode(instr, texture_type, process_mode, depth_compare, is_array); const Node texture = GetTexCode(instr, texture_type, process_mode, depth_compare, is_array);
if (depth_compare) {
SetRegister(bb, instr.gpr0, texture);
} else {
MetaComponents meta; MetaComponents meta;
std::array<Node, 4> dest; std::array<Node, 4> dest;
@ -202,9 +199,8 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
} }
std::generate(dest.begin() + dest_elem, dest.end(), [&]() { return GetRegister(RZ); }); std::generate(dest.begin() + dest_elem, dest.end(), [&]() { return GetRegister(RZ); });
bb.push_back(Operation(OperationCode::AssignComposite, std::move(meta), texture, bb.push_back(Operation(OperationCode::AssignComposite, std::move(meta), texture, dest[0],
dest[0], dest[1], dest[2], dest[3])); dest[1], dest[2], dest[3]));
}
break; break;
} }
case OpCode::Id::TEXS: { case OpCode::Id::TEXS: {
@ -272,7 +268,7 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
params.push_back(Immediate(static_cast<u32>(instr.tld4.component))); params.push_back(Immediate(static_cast<u32>(instr.tld4.component)));
const auto& sampler = GetSampler(instr.sampler, texture_type, false, depth_compare); const auto& sampler = GetSampler(instr.sampler, texture_type, false, depth_compare);
const MetaTexture meta{sampler, num_coordinates}; MetaTexture meta{sampler, num_coordinates};
const Node texture = const Node texture =
Operation(OperationCode::F4TextureGather, std::move(meta), std::move(params)); Operation(OperationCode::F4TextureGather, std::move(meta), std::move(params));
@ -331,7 +327,7 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
const auto& sampler = const auto& sampler =
GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare); GetSampler(instr.sampler, TextureType::Texture2D, false, depth_compare);
const MetaTexture meta{sampler, num_coords}; MetaTexture meta{sampler, num_coords};
WriteTexsInstructionFloat( WriteTexsInstructionFloat(
bb, instr, Operation(OperationCode::F4TextureGather, meta, std::move(params))); bb, instr, Operation(OperationCode::F4TextureGather, meta, std::move(params)));
@ -350,7 +346,7 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
switch (instr.txq.query_type) { switch (instr.txq.query_type) {
case Tegra::Shader::TextureQueryType::Dimension: { case Tegra::Shader::TextureQueryType::Dimension: {
const MetaTexture meta_texture{sampler}; MetaTexture meta_texture{sampler};
const MetaComponents meta_components{{0, 1, 2, 3}}; const MetaComponents meta_components{{0, 1, 2, 3}};
const Node texture = Operation(OperationCode::F4TextureQueryDimensions, meta_texture, const Node texture = Operation(OperationCode::F4TextureQueryDimensions, meta_texture,
@ -402,7 +398,7 @@ u32 ShaderIR::DecodeMemory(BasicBlock& bb, u32 pc) {
texture_type = TextureType::Texture2D; texture_type = TextureType::Texture2D;
} }
const MetaTexture meta_texture{sampler, static_cast<u32>(coords.size())}; MetaTexture meta_texture{sampler, static_cast<u32>(coords.size())};
const Node texture = const Node texture =
Operation(OperationCode::F4TextureQueryLod, meta_texture, std::move(coords)); Operation(OperationCode::F4TextureQueryLod, meta_texture, std::move(coords));
@ -474,7 +470,8 @@ void ShaderIR::WriteTexsInstructionFloat(BasicBlock& bb, Tegra::Shader::Instruct
Node ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type, Node ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, bool depth_compare, bool is_array, TextureProcessMode process_mode, bool depth_compare, bool is_array,
std::size_t bias_offset, std::vector<Node>&& coords) { std::size_t array_offset, std::size_t bias_offset,
std::vector<Node>&& coords) {
UNIMPLEMENTED_IF_MSG( UNIMPLEMENTED_IF_MSG(
(texture_type == TextureType::Texture3D && (is_array || depth_compare)) || (texture_type == TextureType::Texture3D && (is_array || depth_compare)) ||
(texture_type == TextureType::TextureCube && is_array && depth_compare), (texture_type == TextureType::TextureCube && is_array && depth_compare),
@ -486,26 +483,26 @@ Node ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
process_mode == TextureProcessMode::LL || process_mode == TextureProcessMode::LL ||
process_mode == TextureProcessMode::LLA; process_mode == TextureProcessMode::LLA;
// LOD selection (either via bias or explicit textureLod) not supported in GL for
// sampler2DArrayShadow and samplerCubeArrayShadow.
const bool gl_lod_supported = const bool gl_lod_supported =
!((texture_type == TextureType::Texture2D && is_array && depth_compare) || !((texture_type == Tegra::Shader::TextureType::Texture2D && is_array && depth_compare) ||
(texture_type == TextureType::TextureCube && !is_array && depth_compare)); (texture_type == Tegra::Shader::TextureType::TextureCube && is_array && depth_compare));
const OperationCode read_method = const OperationCode read_method =
lod_needed && gl_lod_supported ? OperationCode::F4TextureLod : OperationCode::F4Texture; lod_needed && gl_lod_supported ? OperationCode::F4TextureLod : OperationCode::F4Texture;
const MetaTexture meta{sampler, static_cast<u32>(coords.size())}; UNIMPLEMENTED_IF(process_mode != TextureProcessMode::None && !gl_lod_supported);
std::optional<u32> array_offset_value;
if (is_array)
array_offset_value = static_cast<u32>(array_offset);
MetaTexture meta{sampler, static_cast<u32>(coords.size()), array_offset_value};
std::vector<Node> params = std::move(coords); std::vector<Node> params = std::move(coords);
if (process_mode != TextureProcessMode::None) { if (process_mode != TextureProcessMode::None && gl_lod_supported) {
if (process_mode == TextureProcessMode::LZ) { if (process_mode == TextureProcessMode::LZ) {
if (gl_lod_supported) { params.push_back(Immediate(0.0f));
params.push_back(Immediate(0));
} else {
// Lod 0 is emulated by a big negative bias in scenarios that are not supported by
// GLSL
params.push_back(Immediate(-1000));
}
} else { } else {
// If present, lod or bias are always stored in the register indexed by the gpr20 field // If present, lod or bias are always stored in the register indexed by the gpr20 field
// with an offset depending on the usage of the other registers // with an offset depending on the usage of the other registers
@ -518,8 +515,8 @@ Node ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
Node ShaderIR::GetTexCode(Instruction instr, TextureType texture_type, Node ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, bool depth_compare, bool is_array) { TextureProcessMode process_mode, bool depth_compare, bool is_array) {
const bool lod_bias_enabled = (process_mode != Tegra::Shader::TextureProcessMode::None && const bool lod_bias_enabled =
process_mode != Tegra::Shader::TextureProcessMode::LZ); (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);
const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement( const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5); texture_type, depth_compare, is_array, lod_bias_enabled, 4, 5);
@ -536,29 +533,30 @@ Node ShaderIR::GetTexCode(Instruction instr, TextureType texture_type,
if (depth_compare && !is_array && texture_type == TextureType::Texture1D) { if (depth_compare && !is_array && texture_type == TextureType::Texture1D) {
coords.push_back(Immediate(0.0f)); coords.push_back(Immediate(0.0f));
} }
std::size_t array_offset{};
if (is_array) {
array_offset = coords.size();
coords.push_back(GetRegister(array_register));
}
if (depth_compare) { if (depth_compare) {
// Depth is always stored in the register signaled by gpr20 // Depth is always stored in the register signaled by gpr20
// or in the next register if lod or bias are used // or in the next register if lod or bias are used
const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
coords.push_back(GetRegister(depth_register)); coords.push_back(GetRegister(depth_register));
} }
if (is_array) {
coords.push_back(GetRegister(array_register));
}
// Fill ignored coordinates // Fill ignored coordinates
while (coords.size() < total_coord_count) { while (coords.size() < total_coord_count) {
coords.push_back(Immediate(0)); coords.push_back(Immediate(0));
} }
return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, 0, return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, array_offset,
std::move(coords)); 0, std::move(coords));
} }
Node ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type, Node ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
TextureProcessMode process_mode, bool depth_compare, bool is_array) { TextureProcessMode process_mode, bool depth_compare, bool is_array) {
const bool lod_bias_enabled =
const bool lod_bias_enabled = (process_mode != Tegra::Shader::TextureProcessMode::None && (process_mode != TextureProcessMode::None && process_mode != TextureProcessMode::LZ);
process_mode != Tegra::Shader::TextureProcessMode::LZ);
const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement( const auto [coord_count, total_coord_count] = ValidateAndGetCoordinateElement(
texture_type, depth_compare, is_array, lod_bias_enabled, 4, 4); texture_type, depth_compare, is_array, lod_bias_enabled, 4, 4);
@ -577,22 +575,23 @@ Node ShaderIR::GetTexsCode(Instruction instr, TextureType texture_type,
coords.push_back(GetRegister(last ? last_coord_register : coord_register + i)); coords.push_back(GetRegister(last ? last_coord_register : coord_register + i));
} }
std::size_t array_offset{};
if (is_array) {
array_offset = coords.size();
coords.push_back(GetRegister(array_register));
}
if (depth_compare) { if (depth_compare) {
// Depth is always stored in the register signaled by gpr20 // Depth is always stored in the register signaled by gpr20
// or in the next register if lod or bias are used // or in the next register if lod or bias are used
const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0); const u64 depth_register = instr.gpr20.Value() + (lod_bias_enabled ? 1 : 0);
coords.push_back(GetRegister(depth_register)); coords.push_back(GetRegister(depth_register));
} }
if (is_array) {
coords.push_back(
Operation(OperationCode::ICastFloat, NO_PRECISE, GetRegister(array_register)));
}
// Fill ignored coordinates // Fill ignored coordinates
while (coords.size() < total_coord_count) { while (coords.size() < total_coord_count) {
coords.push_back(Immediate(0)); coords.push_back(Immediate(0));
} }
return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, return GetTextureCode(instr, texture_type, process_mode, depth_compare, is_array, array_offset,
(coord_count > 2 ? 1 : 0), std::move(coords)); (coord_count > 2 ? 1 : 0), std::move(coords));
} }

View file

@ -635,8 +635,10 @@ private:
result_type)); result_type));
} }
#pragma optimize("", off)
std::string GenerateTexture(Operation operation, const std::string& func, std::string GenerateTexture(Operation operation, const std::string& func,
const std::string& extra_cast = "") { std::string extra_cast(std::string) = nullptr) {
constexpr std::array<const char*, 4> coord_constructors = {"float", "vec2", "vec3", "vec4"}; constexpr std::array<const char*, 4> coord_constructors = {"float", "vec2", "vec3", "vec4"};
const auto& meta = std::get<MetaTexture>(operation.GetMeta()); const auto& meta = std::get<MetaTexture>(operation.GetMeta());
@ -651,15 +653,17 @@ private:
expr += '('; expr += '(';
for (u32 i = 0; i < count; ++i) { for (u32 i = 0; i < count; ++i) {
const bool is_extra = i >= meta.coords_count; const bool is_extra = i >= meta.coords_count;
const bool do_cast = is_extra && !extra_cast.empty(); const bool is_array = i == meta.array_index;
if (do_cast) {
expr += extra_cast; std::string operand = Visit(operation[i]);
expr += '('; if (is_extra && extra_cast != nullptr) {
operand = extra_cast(operand);
} }
expr += Visit(operation[i]); if (is_array) {
if (do_cast) { ASSERT(!is_extra);
expr += ')'; operand = "float(ftoi(" + operand + "))";
} }
expr += operand;
if (i + 1 == meta.coords_count) { if (i + 1 == meta.coords_count) {
expr += ')'; expr += ')';
} }
@ -1065,7 +1069,14 @@ private:
} }
std::string F4TextureGather(Operation operation) { std::string F4TextureGather(Operation operation) {
return GenerateTexture(operation, "textureGather", "int"); const bool is_shadow = std::get<MetaTexture>(operation.GetMeta()).sampler.IsShadow();
if (is_shadow) {
return GenerateTexture(operation, "textureGather",
[](std::string ref_z) { return ref_z; });
} else {
return GenerateTexture(operation, "textureGather",
[](std::string comp) { return "ftoi(" + comp + ')'; });
}
} }
std::string F4TextureQueryDimensions(Operation operation) { std::string F4TextureQueryDimensions(Operation operation) {

View file

@ -265,6 +265,7 @@ struct MetaHalfArithmetic {
struct MetaTexture { struct MetaTexture {
const Sampler& sampler; const Sampler& sampler;
u32 coords_count{}; u32 coords_count{};
std::optional<u32> array_index;
}; };
struct MetaComponents { struct MetaComponents {
@ -696,7 +697,8 @@ private:
Node GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type, Node GetTextureCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
Tegra::Shader::TextureProcessMode process_mode, bool depth_compare, Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
bool is_array, std::size_t bias_offset, std::vector<Node>&& coords); bool is_array, std::size_t array_offset, std::size_t bias_offset,
std::vector<Node>&& coords);
void WriteLogicOperation(BasicBlock& bb, Tegra::Shader::Register dest, void WriteLogicOperation(BasicBlock& bb, Tegra::Shader::Register dest,
Tegra::Shader::LogicOperation logic_op, Node op_a, Node op_b, Tegra::Shader::LogicOperation logic_op, Node op_a, Node op_b,