mirror of https://git.suyu.dev/suyu/suyu
Merge pull request #3219 from FernandoS27/fix-bindless
Corrections and fixes to TLD4S & bindless samplers failing
This commit is contained in:
commit
eac075692b
|
@ -1292,6 +1292,7 @@ union Instruction {
|
|||
BitField<50, 1, u64> dc_flag;
|
||||
BitField<51, 1, u64> aoffi_flag;
|
||||
BitField<52, 2, u64> component;
|
||||
BitField<55, 1, u64> fp16_flag;
|
||||
|
||||
bool UsesMiscMode(TextureMiscMode mode) const {
|
||||
switch (mode) {
|
||||
|
@ -1972,7 +1973,7 @@ private:
|
|||
INST("1101-01---------", Id::TLDS, Type::Texture, "TLDS"),
|
||||
INST("110010----111---", Id::TLD4, Type::Texture, "TLD4"),
|
||||
INST("1101111011111---", Id::TLD4_B, Type::Texture, "TLD4_B"),
|
||||
INST("1101111100------", Id::TLD4S, Type::Texture, "TLD4S"),
|
||||
INST("11011111--00----", Id::TLD4S, Type::Texture, "TLD4S"),
|
||||
INST("110111110110----", Id::TMML_B, Type::Texture, "TMML_B"),
|
||||
INST("1101111101011---", Id::TMML, Type::Texture, "TMML"),
|
||||
INST("11011110011110--", Id::TXD_B, Type::Texture, "TXD_B"),
|
||||
|
|
|
@ -1076,7 +1076,7 @@ private:
|
|||
}
|
||||
|
||||
std::string GenerateTexture(Operation operation, const std::string& function_suffix,
|
||||
const std::vector<TextureIR>& extras) {
|
||||
const std::vector<TextureIR>& extras, bool sepparate_dc = false) {
|
||||
constexpr std::array coord_constructors = {"float", "vec2", "vec3", "vec4"};
|
||||
|
||||
const auto meta = std::get_if<MetaTexture>(&operation.GetMeta());
|
||||
|
@ -1091,7 +1091,8 @@ private:
|
|||
expr += "Offset";
|
||||
}
|
||||
expr += '(' + GetSampler(meta->sampler) + ", ";
|
||||
expr += coord_constructors.at(count + (has_array ? 1 : 0) + (has_shadow ? 1 : 0) - 1);
|
||||
expr += coord_constructors.at(count + (has_array ? 1 : 0) +
|
||||
(has_shadow && !sepparate_dc ? 1 : 0) - 1);
|
||||
expr += '(';
|
||||
for (std::size_t i = 0; i < count; ++i) {
|
||||
expr += Visit(operation[i]).AsFloat();
|
||||
|
@ -1104,9 +1105,14 @@ private:
|
|||
expr += ", float(" + Visit(meta->array).AsInt() + ')';
|
||||
}
|
||||
if (has_shadow) {
|
||||
expr += ", " + Visit(meta->depth_compare).AsFloat();
|
||||
if (sepparate_dc) {
|
||||
expr += "), " + Visit(meta->depth_compare).AsFloat();
|
||||
} else {
|
||||
expr += ", " + Visit(meta->depth_compare).AsFloat() + ')';
|
||||
}
|
||||
} else {
|
||||
expr += ')';
|
||||
}
|
||||
|
||||
for (const auto& variant : extras) {
|
||||
if (const auto argument = std::get_if<TextureArgument>(&variant)) {
|
||||
|
@ -1706,10 +1712,17 @@ private:
|
|||
ASSERT(meta);
|
||||
|
||||
const auto type = meta->sampler.IsShadow() ? Type::Float : Type::Int;
|
||||
return {GenerateTexture(operation, "Gather",
|
||||
{TextureAoffi{}, TextureArgument{type, meta->component}}) +
|
||||
if (meta->sampler.IsShadow()) {
|
||||
return {GenerateTexture(operation, "Gather", {TextureAoffi{}}, true) +
|
||||
GetSwizzle(meta->element),
|
||||
Type::Float};
|
||||
} else {
|
||||
return {GenerateTexture(operation, "Gather",
|
||||
{TextureAoffi{}, TextureArgument{type, meta->component}},
|
||||
false) +
|
||||
GetSwizzle(meta->element),
|
||||
Type::Float};
|
||||
}
|
||||
}
|
||||
|
||||
Expression TextureQueryDimensions(Operation operation) {
|
||||
|
|
|
@ -107,8 +107,8 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
break;
|
||||
}
|
||||
case OpCode::Id::TLD4S: {
|
||||
UNIMPLEMENTED_IF_MSG(instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI),
|
||||
"AOFFI is not implemented");
|
||||
const bool uses_aoffi = instr.tld4s.UsesMiscMode(TextureMiscMode::AOFFI);
|
||||
UNIMPLEMENTED_IF_MSG(uses_aoffi, "AOFFI is not implemented");
|
||||
|
||||
const bool depth_compare = instr.tld4s.UsesMiscMode(TextureMiscMode::DC);
|
||||
const Node op_a = GetRegister(instr.gpr8);
|
||||
|
@ -116,29 +116,40 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
|
||||
// TODO(Subv): Figure out how the sampler type is encoded in the TLD4S instruction.
|
||||
std::vector<Node> coords;
|
||||
Node dc_reg;
|
||||
if (depth_compare) {
|
||||
// Note: TLD4S coordinate encoding works just like TEXS's
|
||||
const Node op_y = GetRegister(instr.gpr8.Value() + 1);
|
||||
coords.push_back(op_a);
|
||||
coords.push_back(op_y);
|
||||
coords.push_back(op_b);
|
||||
dc_reg = uses_aoffi ? GetRegister(instr.gpr20.Value() + 1) : op_b;
|
||||
} else {
|
||||
coords.push_back(op_a);
|
||||
if (uses_aoffi) {
|
||||
const Node op_y = GetRegister(instr.gpr8.Value() + 1);
|
||||
coords.push_back(op_y);
|
||||
} else {
|
||||
coords.push_back(op_b);
|
||||
}
|
||||
dc_reg = {};
|
||||
}
|
||||
const Node component = Immediate(static_cast<u32>(instr.tld4s.component));
|
||||
|
||||
const SamplerInfo info{TextureType::Texture2D, false, depth_compare};
|
||||
const auto& sampler = GetSampler(instr.sampler, info);
|
||||
const Sampler& sampler = *GetSampler(instr.sampler, info);
|
||||
|
||||
Node4 values;
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
auto coords_copy = coords;
|
||||
MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, component, element};
|
||||
MetaTexture meta{sampler, {}, dc_reg, {}, {}, {}, {}, component, element};
|
||||
values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
|
||||
}
|
||||
|
||||
if (instr.tld4s.fp16_flag) {
|
||||
WriteTexsInstructionHalfFloat(bb, instr, values, true);
|
||||
} else {
|
||||
WriteTexsInstructionFloat(bb, instr, values, true);
|
||||
}
|
||||
break;
|
||||
}
|
||||
case OpCode::Id::TXD_B:
|
||||
|
@ -154,9 +165,17 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
const auto texture_type = instr.txd.texture_type.Value();
|
||||
const auto coord_count = GetCoordCount(texture_type);
|
||||
|
||||
const auto& sampler = is_bindless
|
||||
const Sampler* sampler = is_bindless
|
||||
? GetBindlessSampler(base_reg, {{texture_type, false, false}})
|
||||
: GetSampler(instr.sampler, {{texture_type, false, false}});
|
||||
Node4 values;
|
||||
if (sampler == nullptr) {
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
values[element] = Immediate(0);
|
||||
}
|
||||
WriteTexInstructionFloat(bb, instr, values);
|
||||
break;
|
||||
}
|
||||
if (is_bindless) {
|
||||
base_reg++;
|
||||
}
|
||||
|
@ -170,9 +189,8 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
derivates.push_back(GetRegister(derivate_reg + derivate + 1));
|
||||
}
|
||||
|
||||
Node4 values;
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
MetaTexture meta{sampler, {}, {}, {}, derivates, {}, {}, {}, element};
|
||||
MetaTexture meta{*sampler, {}, {}, {}, derivates, {}, {}, {}, element};
|
||||
values[element] = Operation(OperationCode::TextureGradient, std::move(meta), coords);
|
||||
}
|
||||
|
||||
|
@ -187,9 +205,24 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
// TODO: The new commits on the texture refactor, change the way samplers work.
|
||||
// Sadly, not all texture instructions specify the type of texture their sampler
|
||||
// uses. This must be fixed at a later instance.
|
||||
const auto& sampler =
|
||||
const Sampler* sampler =
|
||||
is_bindless ? GetBindlessSampler(instr.gpr8) : GetSampler(instr.sampler);
|
||||
|
||||
if (sampler == nullptr) {
|
||||
u32 indexer = 0;
|
||||
for (u32 element = 0; element < 4; ++element) {
|
||||
if (!instr.txq.IsComponentEnabled(element)) {
|
||||
continue;
|
||||
}
|
||||
const Node value = Immediate(0);
|
||||
SetTemporary(bb, indexer++, value);
|
||||
}
|
||||
for (u32 i = 0; i < indexer; ++i) {
|
||||
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
u32 indexer = 0;
|
||||
switch (instr.txq.query_type) {
|
||||
case Tegra::Shader::TextureQueryType::Dimension: {
|
||||
|
@ -197,7 +230,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
if (!instr.txq.IsComponentEnabled(element)) {
|
||||
continue;
|
||||
}
|
||||
MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, {}, element};
|
||||
MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, element};
|
||||
const Node value =
|
||||
Operation(OperationCode::TextureQueryDimensions, meta,
|
||||
GetRegister(instr.gpr8.Value() + (is_bindless ? 1 : 0)));
|
||||
|
@ -223,9 +256,24 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
|
||||
auto texture_type = instr.tmml.texture_type.Value();
|
||||
const bool is_array = instr.tmml.array != 0;
|
||||
const auto& sampler =
|
||||
const Sampler* sampler =
|
||||
is_bindless ? GetBindlessSampler(instr.gpr20) : GetSampler(instr.sampler);
|
||||
|
||||
if (sampler == nullptr) {
|
||||
u32 indexer = 0;
|
||||
for (u32 element = 0; element < 2; ++element) {
|
||||
if (!instr.tmml.IsComponentEnabled(element)) {
|
||||
continue;
|
||||
}
|
||||
const Node value = Immediate(0);
|
||||
SetTemporary(bb, indexer++, value);
|
||||
}
|
||||
for (u32 i = 0; i < indexer; ++i) {
|
||||
SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
|
||||
}
|
||||
break;
|
||||
}
|
||||
|
||||
std::vector<Node> coords;
|
||||
|
||||
// TODO: Add coordinates for different samplers once other texture types are implemented.
|
||||
|
@ -251,7 +299,7 @@ u32 ShaderIR::DecodeTexture(NodeBlock& bb, u32 pc) {
|
|||
continue;
|
||||
}
|
||||
auto params = coords;
|
||||
MetaTexture meta{sampler, {}, {}, {}, {}, {}, {}, {}, element};
|
||||
MetaTexture meta{*sampler, {}, {}, {}, {}, {}, {}, {}, element};
|
||||
const Node value = Operation(OperationCode::TextureQueryLod, meta, std::move(params));
|
||||
SetTemporary(bb, indexer++, value);
|
||||
}
|
||||
|
@ -307,7 +355,7 @@ ShaderIR::SamplerInfo ShaderIR::GetSamplerInfo(std::optional<SamplerInfo> sample
|
|||
sampler->is_buffer != 0};
|
||||
}
|
||||
|
||||
const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
|
||||
const Sampler* ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
|
||||
std::optional<SamplerInfo> sampler_info) {
|
||||
const auto offset = static_cast<u32>(sampler.index.Value());
|
||||
const auto info = GetSamplerInfo(sampler_info, offset);
|
||||
|
@ -319,21 +367,24 @@ const Sampler& ShaderIR::GetSampler(const Tegra::Shader::Sampler& sampler,
|
|||
if (it != used_samplers.end()) {
|
||||
ASSERT(!it->IsBindless() && it->GetType() == info.type && it->IsArray() == info.is_array &&
|
||||
it->IsShadow() == info.is_shadow && it->IsBuffer() == info.is_buffer);
|
||||
return *it;
|
||||
return &(*it);
|
||||
}
|
||||
|
||||
// Otherwise create a new mapping for this sampler
|
||||
const auto next_index = static_cast<u32>(used_samplers.size());
|
||||
return used_samplers.emplace_back(next_index, offset, info.type, info.is_array, info.is_shadow,
|
||||
return &used_samplers.emplace_back(next_index, offset, info.type, info.is_array, info.is_shadow,
|
||||
info.is_buffer);
|
||||
}
|
||||
|
||||
const Sampler& ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
|
||||
const Sampler* ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
|
||||
std::optional<SamplerInfo> sampler_info) {
|
||||
const Node sampler_register = GetRegister(reg);
|
||||
const auto [base_sampler, buffer, offset] =
|
||||
TrackCbuf(sampler_register, global_code, static_cast<s64>(global_code.size()));
|
||||
ASSERT(base_sampler != nullptr);
|
||||
if (base_sampler == nullptr) {
|
||||
return nullptr;
|
||||
}
|
||||
|
||||
const auto info = GetSamplerInfo(sampler_info, offset, buffer);
|
||||
|
||||
|
@ -346,12 +397,12 @@ const Sampler& ShaderIR::GetBindlessSampler(Tegra::Shader::Register reg,
|
|||
if (it != used_samplers.end()) {
|
||||
ASSERT(it->IsBindless() && it->GetType() == info.type && it->IsArray() == info.is_array &&
|
||||
it->IsShadow() == info.is_shadow);
|
||||
return *it;
|
||||
return &(*it);
|
||||
}
|
||||
|
||||
// Otherwise create a new mapping for this sampler
|
||||
const auto next_index = static_cast<u32>(used_samplers.size());
|
||||
return used_samplers.emplace_back(next_index, offset, buffer, info.type, info.is_array,
|
||||
return &used_samplers.emplace_back(next_index, offset, buffer, info.type, info.is_array,
|
||||
info.is_shadow, info.is_buffer);
|
||||
}
|
||||
|
||||
|
@ -395,14 +446,14 @@ void ShaderIR::WriteTexsInstructionFloat(NodeBlock& bb, Instruction instr, const
|
|||
}
|
||||
|
||||
void ShaderIR::WriteTexsInstructionHalfFloat(NodeBlock& bb, Instruction instr,
|
||||
const Node4& components) {
|
||||
const Node4& components, bool ignore_mask) {
|
||||
// TEXS.F16 destionation registers are packed in two registers in pairs (just like any half
|
||||
// float instruction).
|
||||
|
||||
Node4 values;
|
||||
u32 dest_elem = 0;
|
||||
for (u32 component = 0; component < 4; ++component) {
|
||||
if (!instr.texs.IsComponentEnabled(component))
|
||||
if (!instr.texs.IsComponentEnabled(component) && !ignore_mask)
|
||||
continue;
|
||||
values[dest_elem++] = components[component];
|
||||
}
|
||||
|
@ -438,8 +489,15 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
|
|||
"This method is not supported.");
|
||||
|
||||
const SamplerInfo info{texture_type, is_array, is_shadow, false};
|
||||
const auto& sampler =
|
||||
const Sampler* sampler =
|
||||
is_bindless ? GetBindlessSampler(*bindless_reg, info) : GetSampler(instr.sampler, info);
|
||||
Node4 values;
|
||||
if (sampler == nullptr) {
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
values[element] = Immediate(0);
|
||||
}
|
||||
return values;
|
||||
}
|
||||
|
||||
const bool lod_needed = process_mode == TextureProcessMode::LZ ||
|
||||
process_mode == TextureProcessMode::LL ||
|
||||
|
@ -478,10 +536,9 @@ Node4 ShaderIR::GetTextureCode(Instruction instr, TextureType texture_type,
|
|||
}
|
||||
}
|
||||
|
||||
Node4 values;
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
auto copy_coords = coords;
|
||||
MetaTexture meta{sampler, array, depth_compare, aoffi, {}, bias, lod, {}, element};
|
||||
MetaTexture meta{*sampler, array, depth_compare, aoffi, {}, bias, lod, {}, element};
|
||||
values[element] = Operation(read_method, meta, std::move(copy_coords));
|
||||
}
|
||||
|
||||
|
@ -594,8 +651,15 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
|
|||
u64 parameter_register = instr.gpr20.Value();
|
||||
|
||||
const SamplerInfo info{texture_type, is_array, depth_compare, false};
|
||||
const auto& sampler = is_bindless ? GetBindlessSampler(parameter_register++, info)
|
||||
const Sampler* sampler = is_bindless ? GetBindlessSampler(parameter_register++, info)
|
||||
: GetSampler(instr.sampler, info);
|
||||
Node4 values;
|
||||
if (sampler == nullptr) {
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
values[element] = Immediate(0);
|
||||
}
|
||||
return values;
|
||||
}
|
||||
|
||||
std::vector<Node> aoffi;
|
||||
if (is_aoffi) {
|
||||
|
@ -610,10 +674,9 @@ Node4 ShaderIR::GetTld4Code(Instruction instr, TextureType texture_type, bool de
|
|||
const Node component = is_bindless ? Immediate(static_cast<u32>(instr.tld4_b.component))
|
||||
: Immediate(static_cast<u32>(instr.tld4.component));
|
||||
|
||||
Node4 values;
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
auto coords_copy = coords;
|
||||
MetaTexture meta{sampler, GetRegister(array_register), dc, aoffi, {}, {}, {}, component,
|
||||
MetaTexture meta{*sampler, GetRegister(array_register), dc, aoffi, {}, {}, {}, component,
|
||||
element};
|
||||
values[element] = Operation(OperationCode::TextureGather, meta, std::move(coords_copy));
|
||||
}
|
||||
|
@ -642,7 +705,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
|
|||
// const Node aoffi_register{is_aoffi ? GetRegister(gpr20_cursor++) : nullptr};
|
||||
// const Node multisample{is_multisample ? GetRegister(gpr20_cursor++) : nullptr};
|
||||
|
||||
const auto& sampler = GetSampler(instr.sampler);
|
||||
const auto& sampler = *GetSampler(instr.sampler);
|
||||
|
||||
Node4 values;
|
||||
for (u32 element = 0; element < values.size(); ++element) {
|
||||
|
@ -655,7 +718,7 @@ Node4 ShaderIR::GetTldCode(Tegra::Shader::Instruction instr) {
|
|||
}
|
||||
|
||||
Node4 ShaderIR::GetTldsCode(Instruction instr, TextureType texture_type, bool is_array) {
|
||||
const auto& sampler = GetSampler(instr.sampler);
|
||||
const Sampler& sampler = *GetSampler(instr.sampler);
|
||||
|
||||
const std::size_t type_coord_count = GetCoordCount(texture_type);
|
||||
const bool lod_enabled = instr.tlds.GetTextureProcessMode() == TextureProcessMode::LL;
|
||||
|
|
|
@ -313,11 +313,11 @@ private:
|
|||
std::optional<u32> buffer = std::nullopt);
|
||||
|
||||
/// Accesses a texture sampler
|
||||
const Sampler& GetSampler(const Tegra::Shader::Sampler& sampler,
|
||||
const Sampler* GetSampler(const Tegra::Shader::Sampler& sampler,
|
||||
std::optional<SamplerInfo> sampler_info = std::nullopt);
|
||||
|
||||
/// Accesses a texture sampler for a bindless texture.
|
||||
const Sampler& GetBindlessSampler(Tegra::Shader::Register reg,
|
||||
const Sampler* GetBindlessSampler(Tegra::Shader::Register reg,
|
||||
std::optional<SamplerInfo> sampler_info = std::nullopt);
|
||||
|
||||
/// Accesses an image.
|
||||
|
@ -338,7 +338,7 @@ private:
|
|||
void WriteTexsInstructionFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
|
||||
const Node4& components, bool ignore_mask = false);
|
||||
void WriteTexsInstructionHalfFloat(NodeBlock& bb, Tegra::Shader::Instruction instr,
|
||||
const Node4& components);
|
||||
const Node4& components, bool ignore_mask = false);
|
||||
|
||||
Node4 GetTexCode(Tegra::Shader::Instruction instr, Tegra::Shader::TextureType texture_type,
|
||||
Tegra::Shader::TextureProcessMode process_mode, bool depth_compare,
|
||||
|
|
Loading…
Reference in New Issue