summaryrefslogtreecommitdiffstats
path: root/src/video_core/shader
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/shader')
-rw-r--r--src/video_core/shader/decode/arithmetic_integer.cpp15
-rw-r--r--src/video_core/shader/decode/conversion.cpp15
-rw-r--r--src/video_core/shader/decode/image.cpp360
-rw-r--r--src/video_core/shader/decode/memory.cpp63
-rw-r--r--src/video_core/shader/decode/other.cpp48
-rw-r--r--src/video_core/shader/node.h16
-rw-r--r--src/video_core/shader/node_helper.cpp14
-rw-r--r--src/video_core/shader/shader_ir.cpp3
-rw-r--r--src/video_core/shader/shader_ir.h4
9 files changed, 476 insertions, 62 deletions
diff --git a/src/video_core/shader/decode/arithmetic_integer.cpp b/src/video_core/shader/decode/arithmetic_integer.cpp
index 2fe787d6f..0f4c3103a 100644
--- a/src/video_core/shader/decode/arithmetic_integer.cpp
+++ b/src/video_core/shader/decode/arithmetic_integer.cpp
@@ -235,34 +235,30 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
case OpCode::Id::LEA_IMM:
case OpCode::Id::LEA_RZ:
case OpCode::Id::LEA_HI: {
- const auto [op_a, op_b, op_c] = [&]() -> std::tuple<Node, Node, Node> {
+ auto [op_a, op_b, op_c] = [&]() -> std::tuple<Node, Node, Node> {
switch (opcode->get().GetId()) {
case OpCode::Id::LEA_R2: {
return {GetRegister(instr.gpr20), GetRegister(instr.gpr39),
Immediate(static_cast<u32>(instr.lea.r2.entry_a))};
}
-
case OpCode::Id::LEA_R1: {
const bool neg = instr.lea.r1.neg != 0;
return {GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
GetRegister(instr.gpr20),
Immediate(static_cast<u32>(instr.lea.r1.entry_a))};
}
-
case OpCode::Id::LEA_IMM: {
const bool neg = instr.lea.imm.neg != 0;
return {Immediate(static_cast<u32>(instr.lea.imm.entry_a)),
GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
Immediate(static_cast<u32>(instr.lea.imm.entry_b))};
}
-
case OpCode::Id::LEA_RZ: {
const bool neg = instr.lea.rz.neg != 0;
return {GetConstBuffer(instr.lea.rz.cb_index, instr.lea.rz.cb_offset),
GetOperandAbsNegInteger(GetRegister(instr.gpr8), false, neg, true),
Immediate(static_cast<u32>(instr.lea.rz.entry_a))};
}
-
case OpCode::Id::LEA_HI:
default:
UNIMPLEMENTED_MSG("Unhandled LEA subinstruction: {}", opcode->get().GetName());
@@ -275,12 +271,9 @@ u32 ShaderIR::DecodeArithmeticInteger(NodeBlock& bb, u32 pc) {
UNIMPLEMENTED_IF_MSG(instr.lea.pred48 != static_cast<u64>(Pred::UnusedIndex),
"Unhandled LEA Predicate");
- const Node shifted_c =
- Operation(OperationCode::ILogicalShiftLeft, NO_PRECISE, Immediate(1), op_c);
- const Node mul_bc = Operation(OperationCode::IMul, NO_PRECISE, op_b, shifted_c);
- const Node value = Operation(OperationCode::IAdd, NO_PRECISE, op_a, mul_bc);
-
- SetRegister(bb, instr.gpr0, value);
+ Node value = Operation(OperationCode::ILogicalShiftLeft, std::move(op_a), std::move(op_c));
+ value = Operation(OperationCode::IAdd, std::move(op_b), std::move(value));
+ SetRegister(bb, instr.gpr0, std::move(value));
break;
}
diff --git a/src/video_core/shader/decode/conversion.cpp b/src/video_core/shader/decode/conversion.cpp
index 6ead42070..c72690b2b 100644
--- a/src/video_core/shader/decode/conversion.cpp
+++ b/src/video_core/shader/decode/conversion.cpp
@@ -138,18 +138,23 @@ u32 ShaderIR::DecodeConversion(NodeBlock& bb, u32 pc) {
value = GetOperandAbsNegFloat(value, instr.conversion.abs_a, instr.conversion.negate_a);
- value = [&]() {
+ value = [&] {
+ if (instr.conversion.src_size != instr.conversion.dst_size) {
+ // Rounding operations only matter when the source and destination conversion size
+ // is the same.
+ return value;
+ }
switch (instr.conversion.f2f.GetRoundingMode()) {
case Tegra::Shader::F2fRoundingOp::None:
return value;
case Tegra::Shader::F2fRoundingOp::Round:
- return Operation(OperationCode::FRoundEven, PRECISE, value);
+ return Operation(OperationCode::FRoundEven, value);
case Tegra::Shader::F2fRoundingOp::Floor:
- return Operation(OperationCode::FFloor, PRECISE, value);
+ return Operation(OperationCode::FFloor, value);
case Tegra::Shader::F2fRoundingOp::Ceil:
- return Operation(OperationCode::FCeil, PRECISE, value);
+ return Operation(OperationCode::FCeil, value);
case Tegra::Shader::F2fRoundingOp::Trunc:
- return Operation(OperationCode::FTrunc, PRECISE, value);
+ return Operation(OperationCode::FTrunc, value);
default:
UNIMPLEMENTED_MSG("Unimplemented F2F rounding mode {}",
static_cast<u32>(instr.conversion.f2f.rounding.Value()));
diff --git a/src/video_core/shader/decode/image.cpp b/src/video_core/shader/decode/image.cpp
index d2fe4ec5d..0dd7a1196 100644
--- a/src/video_core/shader/decode/image.cpp
+++ b/src/video_core/shader/decode/image.cpp
@@ -13,13 +13,247 @@
#include "video_core/engines/shader_bytecode.h"
#include "video_core/shader/node_helper.h"
#include "video_core/shader/shader_ir.h"
+#include "video_core/textures/texture.h"
namespace VideoCommon::Shader {
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
+using Tegra::Shader::PredCondition;
+using Tegra::Shader::StoreType;
+using Tegra::Texture::ComponentType;
+using Tegra::Texture::TextureFormat;
+using Tegra::Texture::TICEntry;
namespace {
+
+ComponentType GetComponentType(Tegra::Engines::SamplerDescriptor descriptor,
+ std::size_t component) {
+ const TextureFormat format{descriptor.format};
+ switch (format) {
+ case TextureFormat::R16_G16_B16_A16:
+ case TextureFormat::R32_G32_B32_A32:
+ case TextureFormat::R32_G32_B32:
+ case TextureFormat::R32_G32:
+ case TextureFormat::R16_G16:
+ case TextureFormat::R32:
+ case TextureFormat::R16:
+ case TextureFormat::R8:
+ case TextureFormat::R1:
+ if (component == 0) {
+ return descriptor.r_type;
+ }
+ if (component == 1) {
+ return descriptor.g_type;
+ }
+ if (component == 2) {
+ return descriptor.b_type;
+ }
+ if (component == 3) {
+ return descriptor.a_type;
+ }
+ break;
+ case TextureFormat::A8R8G8B8:
+ if (component == 0) {
+ return descriptor.a_type;
+ }
+ if (component == 1) {
+ return descriptor.r_type;
+ }
+ if (component == 2) {
+ return descriptor.g_type;
+ }
+ if (component == 3) {
+ return descriptor.b_type;
+ }
+ break;
+ case TextureFormat::A2B10G10R10:
+ case TextureFormat::A4B4G4R4:
+ case TextureFormat::A5B5G5R1:
+ case TextureFormat::A1B5G5R5:
+ if (component == 0) {
+ return descriptor.a_type;
+ }
+ if (component == 1) {
+ return descriptor.b_type;
+ }
+ if (component == 2) {
+ return descriptor.g_type;
+ }
+ if (component == 3) {
+ return descriptor.r_type;
+ }
+ break;
+ case TextureFormat::R32_B24G8:
+ if (component == 0) {
+ return descriptor.r_type;
+ }
+ if (component == 1) {
+ return descriptor.b_type;
+ }
+ if (component == 2) {
+ return descriptor.g_type;
+ }
+ break;
+ case TextureFormat::B5G6R5:
+ case TextureFormat::B6G5R5:
+ if (component == 0) {
+ return descriptor.b_type;
+ }
+ if (component == 1) {
+ return descriptor.g_type;
+ }
+ if (component == 2) {
+ return descriptor.r_type;
+ }
+ break;
+ case TextureFormat::G8R24:
+ case TextureFormat::G24R8:
+ case TextureFormat::G8R8:
+ case TextureFormat::G4R4:
+ if (component == 0) {
+ return descriptor.g_type;
+ }
+ if (component == 1) {
+ return descriptor.r_type;
+ }
+ break;
+ }
+ UNIMPLEMENTED_MSG("texture format not implement={}", format);
+ return ComponentType::FLOAT;
+}
+
+bool IsComponentEnabled(std::size_t component_mask, std::size_t component) {
+ constexpr u8 R = 0b0001;
+ constexpr u8 G = 0b0010;
+ constexpr u8 B = 0b0100;
+ constexpr u8 A = 0b1000;
+ constexpr std::array<u8, 16> mask = {
+ 0, (R), (G), (R | G), (B), (R | B), (G | B), (R | G | B),
+ (A), (R | A), (G | A), (R | G | A), (B | A), (R | B | A), (G | B | A), (R | G | B | A)};
+ return std::bitset<4>{mask.at(component_mask)}.test(component);
+}
+
+u32 GetComponentSize(TextureFormat format, std::size_t component) {
+ switch (format) {
+ case TextureFormat::R32_G32_B32_A32:
+ return 32;
+ case TextureFormat::R16_G16_B16_A16:
+ return 16;
+ case TextureFormat::R32_G32_B32:
+ return component <= 2 ? 32 : 0;
+ case TextureFormat::R32_G32:
+ return component <= 1 ? 32 : 0;
+ case TextureFormat::R16_G16:
+ return component <= 1 ? 16 : 0;
+ case TextureFormat::R32:
+ return component == 0 ? 32 : 0;
+ case TextureFormat::R16:
+ return component == 0 ? 16 : 0;
+ case TextureFormat::R8:
+ return component == 0 ? 8 : 0;
+ case TextureFormat::R1:
+ return component == 0 ? 1 : 0;
+ case TextureFormat::A8R8G8B8:
+ return 8;
+ case TextureFormat::A2B10G10R10:
+ return (component == 3 || component == 2 || component == 1) ? 10 : 2;
+ case TextureFormat::A4B4G4R4:
+ return 4;
+ case TextureFormat::A5B5G5R1:
+ return (component == 0 || component == 1 || component == 2) ? 5 : 1;
+ case TextureFormat::A1B5G5R5:
+ return (component == 1 || component == 2 || component == 3) ? 5 : 1;
+ case TextureFormat::R32_B24G8:
+ if (component == 0) {
+ return 32;
+ }
+ if (component == 1) {
+ return 24;
+ }
+ if (component == 2) {
+ return 8;
+ }
+ return 0;
+ case TextureFormat::B5G6R5:
+ if (component == 0 || component == 2) {
+ return 5;
+ }
+ if (component == 1) {
+ return 6;
+ }
+ return 0;
+ case TextureFormat::B6G5R5:
+ if (component == 1 || component == 2) {
+ return 5;
+ }
+ if (component == 0) {
+ return 6;
+ }
+ return 0;
+ case TextureFormat::G8R24:
+ if (component == 0) {
+ return 8;
+ }
+ if (component == 1) {
+ return 24;
+ }
+ return 0;
+ case TextureFormat::G24R8:
+ if (component == 0) {
+ return 8;
+ }
+ if (component == 1) {
+ return 24;
+ }
+ return 0;
+ case TextureFormat::G8R8:
+ return (component == 0 || component == 1) ? 8 : 0;
+ case TextureFormat::G4R4:
+ return (component == 0 || component == 1) ? 4 : 0;
+ default:
+ UNIMPLEMENTED_MSG("texture format not implement={}", format);
+ return 0;
+ }
+}
+
+std::size_t GetImageComponentMask(TextureFormat format) {
+ constexpr u8 R = 0b0001;
+ constexpr u8 G = 0b0010;
+ constexpr u8 B = 0b0100;
+ constexpr u8 A = 0b1000;
+ switch (format) {
+ case TextureFormat::R32_G32_B32_A32:
+ case TextureFormat::R16_G16_B16_A16:
+ case TextureFormat::A8R8G8B8:
+ case TextureFormat::A2B10G10R10:
+ case TextureFormat::A4B4G4R4:
+ case TextureFormat::A5B5G5R1:
+ case TextureFormat::A1B5G5R5:
+ return std::size_t{R | G | B | A};
+ case TextureFormat::R32_G32_B32:
+ case TextureFormat::R32_B24G8:
+ case TextureFormat::B5G6R5:
+ case TextureFormat::B6G5R5:
+ return std::size_t{R | G | B};
+ case TextureFormat::R32_G32:
+ case TextureFormat::R16_G16:
+ case TextureFormat::G8R24:
+ case TextureFormat::G24R8:
+ case TextureFormat::G8R8:
+ case TextureFormat::G4R4:
+ return std::size_t{R | G};
+ case TextureFormat::R32:
+ case TextureFormat::R16:
+ case TextureFormat::R8:
+ case TextureFormat::R1:
+ return std::size_t{R};
+ default:
+ UNIMPLEMENTED_MSG("texture format not implement={}", format);
+ return std::size_t{R | G | B | A};
+ }
+}
+
std::size_t GetImageTypeNumCoordinates(Tegra::Shader::ImageType image_type) {
switch (image_type) {
case Tegra::Shader::ImageType::Texture1D:
@@ -37,6 +271,39 @@ std::size_t GetImageTypeNumCoordinates(Tegra::Shader::ImageType image_type) {
}
} // Anonymous namespace
+std::pair<Node, bool> ShaderIR::GetComponentValue(ComponentType component_type, u32 component_size,
+ Node original_value) {
+ switch (component_type) {
+ case ComponentType::SNORM: {
+ // range [-1.0, 1.0]
+ auto cnv_value = Operation(OperationCode::FMul, original_value,
+ Immediate(static_cast<float>(1 << component_size) / 2.f - 1.f));
+ cnv_value = Operation(OperationCode::ICastFloat, std::move(cnv_value));
+ return {BitfieldExtract(std::move(cnv_value), 0, component_size), true};
+ }
+ case ComponentType::SINT:
+ case ComponentType::UNORM: {
+ bool is_signed = component_type == ComponentType::SINT;
+ // range [0.0, 1.0]
+ auto cnv_value = Operation(OperationCode::FMul, original_value,
+ Immediate(static_cast<float>(1 << component_size) - 1.f));
+ return {SignedOperation(OperationCode::ICastFloat, is_signed, std::move(cnv_value)),
+ is_signed};
+ }
+ case ComponentType::UINT: // range [0, (1 << component_size) - 1]
+ return {std::move(original_value), false};
+ case ComponentType::FLOAT:
+ if (component_size == 16) {
+ return {Operation(OperationCode::HCastFloat, original_value), true};
+ } else {
+ return {std::move(original_value), true};
+ }
+ default:
+ UNIMPLEMENTED_MSG("Unimplement component type={}", component_type);
+ return {std::move(original_value), true};
+ }
+}
+
u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
@@ -53,7 +320,6 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
switch (opcode->get().GetId()) {
case OpCode::Id::SULD: {
- UNIMPLEMENTED_IF(instr.suldst.mode != Tegra::Shader::SurfaceDataMode::P);
UNIMPLEMENTED_IF(instr.suldst.out_of_bounds_store !=
Tegra::Shader::OutOfBoundsStore::Ignore);
@@ -62,17 +328,89 @@ u32 ShaderIR::DecodeImage(NodeBlock& bb, u32 pc) {
: GetBindlessImage(instr.gpr39, type)};
image.MarkRead();
- u32 indexer = 0;
- for (u32 element = 0; element < 4; ++element) {
- if (!instr.suldst.IsComponentEnabled(element)) {
- continue;
+ if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::P) {
+ u32 indexer = 0;
+ for (u32 element = 0; element < 4; ++element) {
+ if (!instr.suldst.IsComponentEnabled(element)) {
+ continue;
+ }
+ MetaImage meta{image, {}, element};
+ Node value = Operation(OperationCode::ImageLoad, meta, GetCoordinates(type));
+ SetTemporary(bb, indexer++, std::move(value));
+ }
+ for (u32 i = 0; i < indexer; ++i) {
+ SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
+ }
+ } else if (instr.suldst.mode == Tegra::Shader::SurfaceDataMode::D_BA) {
+ UNIMPLEMENTED_IF(instr.suldst.GetStoreDataLayout() != StoreType::Bits32 &&
+ instr.suldst.GetStoreDataLayout() != StoreType::Bits64);
+
+ auto descriptor = [this, instr] {
+ std::optional<Tegra::Engines::SamplerDescriptor> descriptor;
+ if (instr.suldst.is_immediate) {
+ descriptor =
+ registry.ObtainBoundSampler(static_cast<u32>(instr.image.index.Value()));
+ } else {
+ const Node image_register = GetRegister(instr.gpr39);
+ const auto [base_image, buffer, offset] = TrackCbuf(
+ image_register, global_code, static_cast<s64>(global_code.size()));
+ descriptor = registry.ObtainBindlessSampler(buffer, offset);
+ }
+ if (!descriptor) {
+ UNREACHABLE_MSG("Failed to obtain image descriptor");
+ }
+ return *descriptor;
+ }();
+
+ const auto comp_mask = GetImageComponentMask(descriptor.format);
+
+ switch (instr.suldst.GetStoreDataLayout()) {
+ case StoreType::Bits32:
+ case StoreType::Bits64: {
+ u32 indexer = 0;
+ u32 shifted_counter = 0;
+ Node value = Immediate(0);
+ for (u32 element = 0; element < 4; ++element) {
+ if (!IsComponentEnabled(comp_mask, element)) {
+ continue;
+ }
+ const auto component_type = GetComponentType(descriptor, element);
+ const auto component_size = GetComponentSize(descriptor.format, element);
+ MetaImage meta{image, {}, element};
+
+ auto [converted_value, is_signed] = GetComponentValue(
+ component_type, component_size,
+ Operation(OperationCode::ImageLoad, meta, GetCoordinates(type)));
+
+ // shift element to correct position
+ const auto shifted = shifted_counter;
+ if (shifted > 0) {
+ converted_value =
+ SignedOperation(OperationCode::ILogicalShiftLeft, is_signed,
+ std::move(converted_value), Immediate(shifted));
+ }
+ shifted_counter += component_size;
+
+ // add value into result
+ value = Operation(OperationCode::UBitwiseOr, value, std::move(converted_value));
+
+ // if we shifted enough for 1 byte -> we save it into temp
+ if (shifted_counter >= 32) {
+ SetTemporary(bb, indexer++, std::move(value));
+ // reset counter and value to prepare pack next byte
+ value = Immediate(0);
+ shifted_counter = 0;
+ }
+ }
+ for (u32 i = 0; i < indexer; ++i) {
+ SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
+ }
+ break;
+ }
+ default:
+ UNREACHABLE();
+ break;
}
- MetaImage meta{image, {}, element};
- Node value = Operation(OperationCode::ImageLoad, meta, GetCoordinates(type));
- SetTemporary(bb, indexer++, std::move(value));
- }
- for (u32 i = 0; i < indexer; ++i) {
- SetRegister(bb, instr.gpr0.Value() + i, GetTemporary(i));
}
break;
}
diff --git a/src/video_core/shader/decode/memory.cpp b/src/video_core/shader/decode/memory.cpp
index b5fbc4d58..b8f63922f 100644
--- a/src/video_core/shader/decode/memory.cpp
+++ b/src/video_core/shader/decode/memory.cpp
@@ -19,7 +19,6 @@ namespace VideoCommon::Shader {
using Tegra::Shader::AtomicOp;
using Tegra::Shader::AtomicType;
using Tegra::Shader::Attribute;
-using Tegra::Shader::GlobalAtomicOp;
using Tegra::Shader::GlobalAtomicType;
using Tegra::Shader::Instruction;
using Tegra::Shader::OpCode;
@@ -28,6 +27,31 @@ using Tegra::Shader::StoreType;
namespace {
+Node GetAtomOperation(AtomicOp op, bool is_signed, Node memory, Node data) {
+ const OperationCode operation_code = [op] {
+ switch (op) {
+ case AtomicOp::Add:
+ return OperationCode::AtomicIAdd;
+ case AtomicOp::Min:
+ return OperationCode::AtomicIMin;
+ case AtomicOp::Max:
+ return OperationCode::AtomicIMax;
+ case AtomicOp::And:
+ return OperationCode::AtomicIAnd;
+ case AtomicOp::Or:
+ return OperationCode::AtomicIOr;
+ case AtomicOp::Xor:
+ return OperationCode::AtomicIXor;
+ case AtomicOp::Exch:
+ return OperationCode::AtomicIExchange;
+ default:
+ UNIMPLEMENTED_MSG("op={}", static_cast<int>(op));
+ return OperationCode::AtomicIAdd;
+ }
+ }();
+ return SignedOperation(operation_code, is_signed, std::move(memory), std::move(data));
+}
+
bool IsUnaligned(Tegra::Shader::UniformType uniform_type) {
return uniform_type == Tegra::Shader::UniformType::UnsignedByte ||
uniform_type == Tegra::Shader::UniformType::UnsignedShort;
@@ -363,10 +387,13 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
break;
}
case OpCode::Id::ATOM: {
- UNIMPLEMENTED_IF_MSG(instr.atom.operation != GlobalAtomicOp::Add, "operation={}",
- static_cast<int>(instr.atom.operation.Value()));
- UNIMPLEMENTED_IF_MSG(instr.atom.type != GlobalAtomicType::S32, "type={}",
- static_cast<int>(instr.atom.type.Value()));
+ UNIMPLEMENTED_IF_MSG(instr.atom.operation == AtomicOp::Inc ||
+ instr.atom.operation == AtomicOp::Dec ||
+ instr.atom.operation == AtomicOp::SafeAdd,
+ "operation={}", static_cast<int>(instr.atom.operation.Value()));
+ UNIMPLEMENTED_IF_MSG(instr.atom.type == GlobalAtomicType::S64 ||
+ instr.atom.type == GlobalAtomicType::U64,
+ "type={}", static_cast<int>(instr.atom.type.Value()));
const auto [real_address, base_address, descriptor] =
TrackGlobalMemory(bb, instr, true, true);
@@ -375,25 +402,29 @@ u32 ShaderIR::DecodeMemory(NodeBlock& bb, u32 pc) {
break;
}
+ const bool is_signed =
+ instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
Node gmem = MakeNode<GmemNode>(real_address, base_address, descriptor);
- Node value = Operation(OperationCode::AtomicAdd, std::move(gmem), GetRegister(instr.gpr20));
+ Node value = GetAtomOperation(static_cast<AtomicOp>(instr.atom.operation), is_signed, gmem,
+ GetRegister(instr.gpr20));
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
case OpCode::Id::ATOMS: {
- UNIMPLEMENTED_IF_MSG(instr.atoms.operation != AtomicOp::Add, "operation={}",
- static_cast<int>(instr.atoms.operation.Value()));
- UNIMPLEMENTED_IF_MSG(instr.atoms.type != AtomicType::U32, "type={}",
- static_cast<int>(instr.atoms.type.Value()));
-
+ UNIMPLEMENTED_IF_MSG(instr.atoms.operation == AtomicOp::Inc ||
+ instr.atoms.operation == AtomicOp::Dec,
+ "operation={}", static_cast<int>(instr.atoms.operation.Value()));
+ UNIMPLEMENTED_IF_MSG(instr.atoms.type == AtomicType::S64 ||
+ instr.atoms.type == AtomicType::U64,
+ "type={}", static_cast<int>(instr.atoms.type.Value()));
+ const bool is_signed =
+ instr.atoms.type == AtomicType::S32 || instr.atoms.type == AtomicType::S64;
const s32 offset = instr.atoms.GetImmediateOffset();
Node address = GetRegister(instr.gpr8);
address = Operation(OperationCode::IAdd, std::move(address), Immediate(offset));
-
- Node memory = GetSharedMemory(std::move(address));
- Node data = GetRegister(instr.gpr20);
-
- Node value = Operation(OperationCode::AtomicAdd, std::move(memory), std::move(data));
+ Node value =
+ GetAtomOperation(static_cast<AtomicOp>(instr.atoms.operation), is_signed,
+ GetSharedMemory(std::move(address)), GetRegister(instr.gpr20));
SetRegister(bb, instr.gpr0, std::move(value));
break;
}
diff --git a/src/video_core/shader/decode/other.cpp b/src/video_core/shader/decode/other.cpp
index 4944e9d69..d4f95b18c 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -11,12 +11,17 @@
namespace VideoCommon::Shader {
+using std::move;
using Tegra::Shader::ConditionCode;
using Tegra::Shader::Instruction;
+using Tegra::Shader::IpaInterpMode;
using Tegra::Shader::OpCode;
+using Tegra::Shader::PixelImap;
using Tegra::Shader::Register;
using Tegra::Shader::SystemVariable;
+using Index = Tegra::Shader::Attribute::Index;
+
u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
const Instruction instr = {program_code[pc]};
const auto opcode = OpCode::Decode(instr);
@@ -66,18 +71,24 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
bb.push_back(Operation(OperationCode::Discard));
break;
}
- case OpCode::Id::MOV_SYS: {
+ case OpCode::Id::S2R: {
const Node value = [this, instr] {
switch (instr.sys20) {
case SystemVariable::LaneId:
- LOG_WARNING(HW_GPU, "MOV_SYS instruction with LaneId is incomplete");
+ LOG_WARNING(HW_GPU, "S2R instruction with LaneId is incomplete");
return Immediate(0U);
case SystemVariable::InvocationId:
return Operation(OperationCode::InvocationId);
case SystemVariable::Ydirection:
return Operation(OperationCode::YNegate);
case SystemVariable::InvocationInfo:
- LOG_WARNING(HW_GPU, "MOV_SYS instruction with InvocationInfo is incomplete");
+ LOG_WARNING(HW_GPU, "S2R instruction with InvocationInfo is incomplete");
+ return Immediate(0U);
+ case SystemVariable::WscaleFactorXY:
+ UNIMPLEMENTED_MSG("S2R WscaleFactorXY is not implemented");
+ return Immediate(0U);
+ case SystemVariable::WscaleFactorZ:
+ UNIMPLEMENTED_MSG("S2R WscaleFactorZ is not implemented");
return Immediate(0U);
case SystemVariable::Tid: {
Node value = Immediate(0);
@@ -213,27 +224,28 @@ u32 ShaderIR::DecodeOther(NodeBlock& bb, u32 pc) {
}
case OpCode::Id::IPA: {
const bool is_physical = instr.ipa.idx && instr.gpr8.Value() != 0xff;
-
const auto attribute = instr.attribute.fmt28;
- const Tegra::Shader::IpaMode input_mode{instr.ipa.interp_mode.Value(),
- instr.ipa.sample_mode.Value()};
+ const Index index = attribute.index;
Node value = is_physical ? GetPhysicalInputAttribute(instr.gpr8)
- : GetInputAttribute(attribute.index, attribute.element);
- const Tegra::Shader::Attribute::Index index = attribute.index.Value();
- const bool is_generic = index >= Tegra::Shader::Attribute::Index::Attribute_0 &&
- index <= Tegra::Shader::Attribute::Index::Attribute_31;
- if (is_generic || is_physical) {
- // TODO(Blinkhawk): There are cases where a perspective attribute use PASS.
- // In theory by setting them as perspective, OpenGL does the perspective correction.
- // A way must figured to reverse the last step of it.
- if (input_mode.interpolation_mode == Tegra::Shader::IpaInterpMode::Multiply) {
- value = Operation(OperationCode::FMul, PRECISE, value, GetRegister(instr.gpr20));
+ : GetInputAttribute(index, attribute.element);
+
+ // Code taken from Ryujinx.
+ if (index >= Index::Attribute_0 && index <= Index::Attribute_31) {
+ const u32 location = static_cast<u32>(index) - static_cast<u32>(Index::Attribute_0);
+ if (header.ps.GetPixelImap(location) == PixelImap::Perspective) {
+ Node position_w = GetInputAttribute(Index::Position, 3);
+ value = Operation(OperationCode::FMul, move(value), move(position_w));
}
}
- value = GetSaturatedFloat(value, instr.ipa.saturate);
- SetRegister(bb, instr.gpr0, value);
+ if (instr.ipa.interp_mode == IpaInterpMode::Multiply) {
+ value = Operation(OperationCode::FMul, move(value), GetRegister(instr.gpr20));
+ }
+
+ value = GetSaturatedFloat(move(value), instr.ipa.saturate);
+
+ SetRegister(bb, instr.gpr0, move(value));
break;
}
case OpCode::Id::OUT_R: {
diff --git a/src/video_core/shader/node.h b/src/video_core/shader/node.h
index a1828546e..5fcc9da60 100644
--- a/src/video_core/shader/node.h
+++ b/src/video_core/shader/node.h
@@ -162,7 +162,21 @@ enum class OperationCode {
AtomicImageXor, /// (MetaImage, int[N] coords) -> void
AtomicImageExchange, /// (MetaImage, int[N] coords) -> void
- AtomicAdd, /// (memory, {u}int) -> {u}int
+ AtomicUExchange, /// (memory, uint) -> uint
+ AtomicUAdd, /// (memory, uint) -> uint
+ AtomicUMin, /// (memory, uint) -> uint
+ AtomicUMax, /// (memory, uint) -> uint
+ AtomicUAnd, /// (memory, uint) -> uint
+ AtomicUOr, /// (memory, uint) -> uint
+ AtomicUXor, /// (memory, uint) -> uint
+
+ AtomicIExchange, /// (memory, int) -> int
+ AtomicIAdd, /// (memory, int) -> int
+ AtomicIMin, /// (memory, int) -> int
+ AtomicIMax, /// (memory, int) -> int
+ AtomicIAnd, /// (memory, int) -> int
+ AtomicIOr, /// (memory, int) -> int
+ AtomicIXor, /// (memory, int) -> int
Branch, /// (uint branch_target) -> void
BranchIndirect, /// (uint branch_target) -> void
diff --git a/src/video_core/shader/node_helper.cpp b/src/video_core/shader/node_helper.cpp
index 76c56abb5..7bf4ff387 100644
--- a/src/video_core/shader/node_helper.cpp
+++ b/src/video_core/shader/node_helper.cpp
@@ -86,6 +86,20 @@ OperationCode SignedToUnsignedCode(OperationCode operation_code, bool is_signed)
return OperationCode::LogicalUNotEqual;
case OperationCode::LogicalIGreaterEqual:
return OperationCode::LogicalUGreaterEqual;
+ case OperationCode::AtomicIExchange:
+ return OperationCode::AtomicUExchange;
+ case OperationCode::AtomicIAdd:
+ return OperationCode::AtomicUAdd;
+ case OperationCode::AtomicIMin:
+ return OperationCode::AtomicUMin;
+ case OperationCode::AtomicIMax:
+ return OperationCode::AtomicUMax;
+ case OperationCode::AtomicIAnd:
+ return OperationCode::AtomicUAnd;
+ case OperationCode::AtomicIOr:
+ return OperationCode::AtomicUOr;
+ case OperationCode::AtomicIXor:
+ return OperationCode::AtomicUXor;
case OperationCode::INegate:
UNREACHABLE_MSG("Can't negate an unsigned integer");
return {};
diff --git a/src/video_core/shader/shader_ir.cpp b/src/video_core/shader/shader_ir.cpp
index baf7188d2..8852c8a1b 100644
--- a/src/video_core/shader/shader_ir.cpp
+++ b/src/video_core/shader/shader_ir.cpp
@@ -359,6 +359,9 @@ Node ShaderIR::GetConditionCode(Tegra::Shader::ConditionCode cc) const {
switch (cc) {
case Tegra::Shader::ConditionCode::NEU:
return GetInternalFlag(InternalFlag::Zero, true);
+ case Tegra::Shader::ConditionCode::FCSM_TR:
+ UNIMPLEMENTED_MSG("EXIT.FCSM_TR is not implemented");
+ return MakeNode<PredicateNode>(Pred::NeverExecute, false);
default:
UNIMPLEMENTED_MSG("Unimplemented condition code: {}", static_cast<u32>(cc));
return MakeNode<PredicateNode>(Pred::NeverExecute, false);
diff --git a/src/video_core/shader/shader_ir.h b/src/video_core/shader/shader_ir.h
index 0f1ebef1b..c6e7bdf50 100644
--- a/src/video_core/shader/shader_ir.h
+++ b/src/video_core/shader/shader_ir.h
@@ -312,6 +312,10 @@ private:
/// Conditionally saturates a half float pair
Node GetSaturatedHalfFloat(Node value, bool saturate = true);
+ /// Get image component value by type and size
+ std::pair<Node, bool> GetComponentValue(Tegra::Texture::ComponentType component_type,
+ u32 component_size, Node original_value);
+
/// Returns a predicate comparing two floats
Node GetPredicateComparisonFloat(Tegra::Shader::PredCondition condition, Node op_a, Node op_b);
/// Returns a predicate comparing two integers