summaryrefslogtreecommitdiffstats
path: root/src/video_core
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core')
-rw-r--r--src/video_core/engines/shader_bytecode.h8
-rw-r--r--src/video_core/shader/decode/image.cpp360
-rw-r--r--src/video_core/shader/decode/other.cpp12
-rw-r--r--src/video_core/shader/shader_ir.cpp3
-rw-r--r--src/video_core/shader/shader_ir.h4
-rw-r--r--src/video_core/texture_cache/surface_params.cpp6
6 files changed, 373 insertions, 20 deletions
diff --git a/src/video_core/engines/shader_bytecode.h b/src/video_core/engines/shader_bytecode.h
index 930b605af..498936f0c 100644
--- a/src/video_core/engines/shader_bytecode.h
+++ b/src/video_core/engines/shader_bytecode.h
@@ -1712,6 +1712,7 @@ public:
BRK,
DEPBAR,
VOTE,
+ VOTE_VTG,
SHFL,
FSWZADD,
BFE_C,
@@ -1758,6 +1759,7 @@ public:
IPA,
OUT_R, // Emit vertex/primitive
ISBERD,
+ BAR,
MEMBAR,
VMAD,
VSETP,
@@ -1842,7 +1844,7 @@ public:
MOV_C,
MOV_R,
MOV_IMM,
- MOV_SYS,
+ S2R,
MOV32_IMM,
SHL_C,
SHL_R,
@@ -2026,6 +2028,7 @@ private:
INST("111000110000----", Id::EXIT, Type::Flow, "EXIT"),
INST("1111000011110---", Id::DEPBAR, Type::Synch, "DEPBAR"),
INST("0101000011011---", Id::VOTE, Type::Warp, "VOTE"),
+ INST("0101000011100---", Id::VOTE_VTG, Type::Warp, "VOTE_VTG"),
INST("1110111100010---", Id::SHFL, Type::Warp, "SHFL"),
INST("0101000011111---", Id::FSWZADD, Type::Warp, "FSWZADD"),
INST("1110111111011---", Id::LD_A, Type::Memory, "LD_A"),
@@ -2063,6 +2066,7 @@ private:
INST("11100000--------", Id::IPA, Type::Trivial, "IPA"),
INST("1111101111100---", Id::OUT_R, Type::Trivial, "OUT_R"),
INST("1110111111010---", Id::ISBERD, Type::Trivial, "ISBERD"),
+ INST("1111000010101---", Id::BAR, Type::Trivial, "BAR"),
INST("1110111110011---", Id::MEMBAR, Type::Trivial, "MEMBAR"),
INST("01011111--------", Id::VMAD, Type::Video, "VMAD"),
INST("0101000011110---", Id::VSETP, Type::Video, "VSETP"),
@@ -2134,7 +2138,7 @@ private:
INST("0100110010011---", Id::MOV_C, Type::Arithmetic, "MOV_C"),
INST("0101110010011---", Id::MOV_R, Type::Arithmetic, "MOV_R"),
INST("0011100-10011---", Id::MOV_IMM, Type::Arithmetic, "MOV_IMM"),
- INST("1111000011001---", Id::MOV_SYS, Type::Trivial, "MOV_SYS"),
+ INST("1111000011001---", Id::S2R, Type::Trivial, "S2R"),
INST("000000010000----", Id::MOV32_IMM, Type::ArithmeticImmediate, "MOV32_IMM"),
INST("0100110001100---", Id::FMNMX_C, Type::Arithmetic, "FMNMX_C"),
INST("0101110001100---", Id::FMNMX_R, Type::Arithmetic, "FMNMX_R"),
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/other.cpp b/src/video_core/shader/decode/other.cpp
index e6edec459..d4f95b18c 100644
--- a/src/video_core/shader/decode/other.cpp
+++ b/src/video_core/shader/decode/other.cpp
@@ -71,18 +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);
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 80fc9b82c..ca6c976c9 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
diff --git a/src/video_core/texture_cache/surface_params.cpp b/src/video_core/texture_cache/surface_params.cpp
index 47b2aafbd..6f3ef45be 100644
--- a/src/video_core/texture_cache/surface_params.cpp
+++ b/src/video_core/texture_cache/surface_params.cpp
@@ -113,10 +113,8 @@ SurfaceParams SurfaceParams::CreateForTexture(const FormatLookupTable& lookup_ta
params.height = tic.Height();
params.depth = tic.Depth();
params.pitch = params.is_tiled ? 0 : tic.Pitch();
- if (params.target == SurfaceTarget::Texture2D && params.depth > 1) {
- params.depth = 1;
- } else if (params.target == SurfaceTarget::TextureCubemap ||
- params.target == SurfaceTarget::TextureCubeArray) {
+ if (params.target == SurfaceTarget::TextureCubemap ||
+ params.target == SurfaceTarget::TextureCubeArray) {
params.depth *= 6;
}
params.num_levels = tic.max_mip_level + 1;
generated by cgit v1.2.3 (git 2.25.1) at 2024-07-15 09:15:05 +0200