shader_recompiler: Implement LowerInt16ToInt32

AMD drivers 22.3.2 and later expose a bug in yuzu, where the application
would submit 16-bit integer instructions to GPUs that don't support
16-bit integers, namely GCN 4 devices.

Replace any 16-bit instructions with 32-bit ones so newer AMD drivers
will work with VK_KHR_workgroup_memory_explicit_layout.

src/common/settings.h

@@ -606,6 +606,7 @@ struct Values {
     BasicSetting<bool> dump_exefs{false, "dump_exefs"};
     BasicSetting<bool> dump_nso{false, "dump_nso"};
     BasicSetting<bool> dump_shaders{false, "dump_shaders"};
+    BasicSetting<bool> dump_macros{false, "dump_macros"};
     BasicSetting<bool> enable_fs_access_log{false, "enable_fs_access_log"};
     BasicSetting<bool> reporting_services{false, "reporting_services"};
     BasicSetting<bool> quest_flag{false, "quest_flag"};

src/core/arm/dynarmic/arm_dynarmic_cp15.cpp

@@ -20,7 +20,7 @@ struct fmt::formatter<Dynarmic::A32::CoprocReg> {
     }
     template <typename FormatContext>
     auto format(const Dynarmic::A32::CoprocReg& reg, FormatContext& ctx) {
-        return format_to(ctx.out(), "cp{}", static_cast<size_t>(reg));
+        return fmt::format_to(ctx.out(), "cp{}", static_cast<size_t>(reg));
     }
 };
 

src/core/hle/service/hid/controllers/npad.cpp

@@ -3,7 +3,9 @@
 
 #include <algorithm>
 #include <array>
+#include <chrono>
 #include <cstring>
+
 #include "common/assert.h"
 #include "common/bit_field.h"
 #include "common/common_types.h"
@@ -529,6 +531,14 @@ void Controller_NPad::OnMotionUpdate(const Core::Timing::CoreTiming& core_timing
         auto& sixaxis_left_lifo_state = controller.sixaxis_left_lifo_state;
         auto& sixaxis_right_lifo_state = controller.sixaxis_right_lifo_state;
 
+        // Clear previous state
+        sixaxis_fullkey_state = {};
+        sixaxis_handheld_state = {};
+        sixaxis_dual_left_state = {};
+        sixaxis_dual_right_state = {};
+        sixaxis_left_lifo_state = {};
+        sixaxis_right_lifo_state = {};
+
         if (controller.sixaxis_sensor_enabled && Settings::values.motion_enabled.GetValue()) {
             controller.sixaxis_at_rest = true;
             for (std::size_t e = 0; e < motion_state.size(); ++e) {
@@ -537,69 +547,55 @@ void Controller_NPad::OnMotionUpdate(const Core::Timing::CoreTiming& core_timing
             }
         }
 
+        const auto set_motion_state = [&](SixAxisSensorState& state,
+                                          const Core::HID::ControllerMotion& hid_state) {
+            using namespace std::literals::chrono_literals;
+            static constexpr SixAxisSensorState default_motion_state = {
+                .delta_time = std::chrono::nanoseconds(5ms).count(),
+                .accel = {0, 0, -1.0f},
+                .orientation =
+                    {
+                        Common::Vec3f{1.0f, 0, 0},
+                        Common::Vec3f{0, 1.0f, 0},
+                        Common::Vec3f{0, 0, 1.0f},
+                    },
+                .attribute = {1},
+            };
+            if (!controller.sixaxis_sensor_enabled) {
+                state = default_motion_state;
+                return;
+            }
+            if (!Settings::values.motion_enabled.GetValue()) {
+                state = default_motion_state;
+                return;
+            }
+            state.attribute.is_connected.Assign(1);
+            state.delta_time = std::chrono::nanoseconds(5ms).count();
+            state.accel = hid_state.accel;
+            state.gyro = hid_state.gyro;
+            state.rotation = hid_state.rotation;
+            state.orientation = hid_state.orientation;
+        };
+
         switch (controller_type) {
         case Core::HID::NpadStyleIndex::None:
             UNREACHABLE();
             break;
         case Core::HID::NpadStyleIndex::ProController:
-            sixaxis_fullkey_state.attribute.raw = 0;
-            if (controller.sixaxis_sensor_enabled) {
-                sixaxis_fullkey_state.attribute.is_connected.Assign(1);
-                sixaxis_fullkey_state.accel = motion_state[0].accel;
-                sixaxis_fullkey_state.gyro = motion_state[0].gyro;
-                sixaxis_fullkey_state.rotation = motion_state[0].rotation;
-                sixaxis_fullkey_state.orientation = motion_state[0].orientation;
-            }
+            set_motion_state(sixaxis_fullkey_state, motion_state[0]);
             break;
         case Core::HID::NpadStyleIndex::Handheld:
-            sixaxis_handheld_state.attribute.raw = 0;
-            if (controller.sixaxis_sensor_enabled) {
-                sixaxis_handheld_state.attribute.is_connected.Assign(1);
-                sixaxis_handheld_state.accel = motion_state[0].accel;
-                sixaxis_handheld_state.gyro = motion_state[0].gyro;
-                sixaxis_handheld_state.rotation = motion_state[0].rotation;
-                sixaxis_handheld_state.orientation = motion_state[0].orientation;
-            }
+            set_motion_state(sixaxis_handheld_state, motion_state[0]);
             break;
         case Core::HID::NpadStyleIndex::JoyconDual:
-            sixaxis_dual_left_state.attribute.raw = 0;
-            sixaxis_dual_right_state.attribute.raw = 0;
-            if (controller.sixaxis_sensor_enabled) {
-                // Set motion for the left joycon
-                sixaxis_dual_left_state.attribute.is_connected.Assign(1);
-                sixaxis_dual_left_state.accel = motion_state[0].accel;
-                sixaxis_dual_left_state.gyro = motion_state[0].gyro;
-                sixaxis_dual_left_state.rotation = motion_state[0].rotation;
-                sixaxis_dual_left_state.orientation = motion_state[0].orientation;
-            }
-            if (controller.sixaxis_sensor_enabled) {
-                // Set motion for the right joycon
-                sixaxis_dual_right_state.attribute.is_connected.Assign(1);
-                sixaxis_dual_right_state.accel = motion_state[1].accel;
-                sixaxis_dual_right_state.gyro = motion_state[1].gyro;
-                sixaxis_dual_right_state.rotation = motion_state[1].rotation;
-                sixaxis_dual_right_state.orientation = motion_state[1].orientation;
-            }
+            set_motion_state(sixaxis_dual_left_state, motion_state[0]);
+            set_motion_state(sixaxis_dual_right_state, motion_state[1]);
             break;
         case Core::HID::NpadStyleIndex::JoyconLeft:
-            sixaxis_left_lifo_state.attribute.raw = 0;
-            if (controller.sixaxis_sensor_enabled) {
-                sixaxis_left_lifo_state.attribute.is_connected.Assign(1);
-                sixaxis_left_lifo_state.accel = motion_state[0].accel;
-                sixaxis_left_lifo_state.gyro = motion_state[0].gyro;
-                sixaxis_left_lifo_state.rotation = motion_state[0].rotation;
-                sixaxis_left_lifo_state.orientation = motion_state[0].orientation;
-            }
+            set_motion_state(sixaxis_left_lifo_state, motion_state[0]);
             break;
         case Core::HID::NpadStyleIndex::JoyconRight:
-            sixaxis_right_lifo_state.attribute.raw = 0;
-            if (controller.sixaxis_sensor_enabled) {
-                sixaxis_right_lifo_state.attribute.is_connected.Assign(1);
-                sixaxis_right_lifo_state.accel = motion_state[1].accel;
-                sixaxis_right_lifo_state.gyro = motion_state[1].gyro;
-                sixaxis_right_lifo_state.rotation = motion_state[1].rotation;
-                sixaxis_right_lifo_state.orientation = motion_state[1].orientation;
-            }
+            set_motion_state(sixaxis_right_lifo_state, motion_state[1]);
             break;
         default:
             break;

src/core/hle/service/hid/hid.cpp

@@ -37,8 +37,7 @@ namespace Service::HID {
 // Period time is obtained by measuring the number of samples in a second on HW using a homebrew
 constexpr auto pad_update_ns = std::chrono::nanoseconds{4 * 1000 * 1000};            // (4ms, 250Hz)
 constexpr auto mouse_keyboard_update_ns = std::chrono::nanoseconds{8 * 1000 * 1000}; // (8ms, 125Hz)
-// TODO: Correct update rate for motion is 5ms. Check why some games don't behave at that speed
-constexpr auto motion_update_ns = std::chrono::nanoseconds{10 * 1000 * 1000}; // (10ms, 100Hz)
+constexpr auto motion_update_ns = std::chrono::nanoseconds{5 * 1000 * 1000};         // (5ms, 200Hz)
 
 IAppletResource::IAppletResource(Core::System& system_,
                                  KernelHelpers::ServiceContext& service_context_)

src/shader_recompiler/CMakeLists.txt

@@ -219,6 +219,7 @@ add_library(shader_recompiler STATIC
     ir_opt/global_memory_to_storage_buffer_pass.cpp
     ir_opt/identity_removal_pass.cpp
     ir_opt/lower_fp16_to_fp32.cpp
+    ir_opt/lower_int16_to_int32.cpp
     ir_opt/lower_int64_to_int32.cpp
     ir_opt/passes.h
     ir_opt/rescaling_pass.cpp

src/shader_recompiler/frontend/ir/opcodes.h

@@ -103,6 +103,6 @@ struct fmt::formatter<Shader::IR::Opcode> {
     }
     template <typename FormatContext>
     auto format(const Shader::IR::Opcode& op, FormatContext& ctx) {
-        return format_to(ctx.out(), "{}", Shader::IR::NameOf(op));
+        return fmt::format_to(ctx.out(), "{}", Shader::IR::NameOf(op));
     }
 };

src/shader_recompiler/frontend/maxwell/opcodes.h

@@ -24,6 +24,6 @@ struct fmt::formatter<Shader::Maxwell::Opcode> {
     }
     template <typename FormatContext>
     auto format(const Shader::Maxwell::Opcode& opcode, FormatContext& ctx) {
-        return format_to(ctx.out(), "{}", NameOf(opcode));
+        return fmt::format_to(ctx.out(), "{}", NameOf(opcode));
     }
 };

src/shader_recompiler/frontend/maxwell/translate_program.cpp

@@ -209,6 +209,9 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
     if (!host_info.support_int64) {
         Optimization::LowerInt64ToInt32(program);
     }
+    if (!host_info.support_int16) {
+        Optimization::LowerInt16ToInt32(program);
+    }
     Optimization::SsaRewritePass(program);
 
     Optimization::ConstantPropagationPass(program);

src/shader_recompiler/host_translate_info.h

@@ -11,6 +11,7 @@ namespace Shader {
 /// Misc information about the host
 struct HostTranslateInfo {
     bool support_float16{};      ///< True when the device supports 16-bit floats
+    bool support_int16{};        ///< True when the device supports 16-bit integers
     bool support_int64{};        ///< True when the device supports 64-bit integers
     bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered
 };

src/shader_recompiler/ir_opt/lower_int16_to_int32.cpp

@@ -0,0 +1,72 @@
+// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "shader_recompiler/frontend/ir/value.h"
+#include "shader_recompiler/ir_opt/passes.h"
+
+namespace Shader::Optimization {
+namespace {
+IR::Opcode Replace(IR::Opcode op) {
+    switch (op) {
+    case IR::Opcode::GetCbufU16:
+    case IR::Opcode::GetCbufS16:
+        return IR::Opcode::GetCbufU32;
+    case IR::Opcode::UndefU16:
+        return IR::Opcode::UndefU32;
+    case IR::Opcode::LoadGlobalU16:
+    case IR::Opcode::LoadGlobalS16:
+        return IR::Opcode::LoadGlobal32;
+    case IR::Opcode::WriteGlobalU16:
+    case IR::Opcode::WriteGlobalS16:
+        return IR::Opcode::WriteGlobal32;
+    case IR::Opcode::LoadStorageU16:
+    case IR::Opcode::LoadStorageS16:
+        return IR::Opcode::LoadStorage32;
+    case IR::Opcode::WriteStorageU16:
+    case IR::Opcode::WriteStorageS16:
+        return IR::Opcode::WriteStorage32;
+    case IR::Opcode::LoadSharedU16:
+    case IR::Opcode::LoadSharedS16:
+        return IR::Opcode::LoadSharedU32;
+    case IR::Opcode::WriteSharedU16:
+        return IR::Opcode::WriteSharedU32;
+    case IR::Opcode::SelectU16:
+        return IR::Opcode::SelectU32;
+    case IR::Opcode::BitCastU16F16:
+        return IR::Opcode::BitCastU32F32;
+    case IR::Opcode::BitCastF16U16:
+        return IR::Opcode::BitCastF32U32;
+    case IR::Opcode::ConvertS16F16:
+    case IR::Opcode::ConvertS16F32:
+        return IR::Opcode::ConvertS32F32;
+    case IR::Opcode::ConvertS16F64:
+        return IR::Opcode::ConvertS32F64;
+    case IR::Opcode::ConvertU16F16:
+    case IR::Opcode::ConvertU16F32:
+        return IR::Opcode::ConvertU32F32;
+    case IR::Opcode::ConvertU16F64:
+        return IR::Opcode::ConvertU32F64;
+    case IR::Opcode::ConvertF16S16:
+    case IR::Opcode::ConvertF32S16:
+        return IR::Opcode::ConvertF32S32;
+    case IR::Opcode::ConvertF16U16:
+    case IR::Opcode::ConvertF32U16:
+        return IR::Opcode::ConvertF32U32;
+    case IR::Opcode::ConvertF64S16:
+    case IR::Opcode::ConvertF64U16:
+        return IR::Opcode::ConvertF64U32;
+    default:
+        return op;
+    }
+}
+} // Anonymous namespace
+
+void LowerInt16ToInt32(IR::Program& program) {
+    for (IR::Block* const block : program.blocks) {
+        for (IR::Inst& inst : block->Instructions()) {
+            inst.ReplaceOpcode(Replace(inst.GetOpcode()));
+        }
+    }
+}
+
+} // namespace Shader::Optimization

src/shader_recompiler/ir_opt/passes.h

@@ -14,6 +14,7 @@ void DeadCodeEliminationPass(IR::Program& program);
 void GlobalMemoryToStorageBufferPass(IR::Program& program);
 void IdentityRemovalPass(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
+void LowerInt16ToInt32(IR::Program& program);
 void LowerInt64ToInt32(IR::Program& program);
 void RescalingPass(IR::Program& program);
 void SsaRewritePass(IR::Program& program);

src/video_core/engines/maxwell_3d.cpp

@@ -173,6 +173,8 @@ void Maxwell3D::ProcessMethodCall(u32 method, u32 argument, u32 nonshadow_argume
     case MAXWELL3D_REG_INDEX(shadow_ram_control):
         shadow_state.shadow_ram_control = static_cast<Regs::ShadowRamControl>(nonshadow_argument);
         return;
+    case MAXWELL3D_REG_INDEX(macros.upload_address):
+        return macro_engine->ClearCode(regs.macros.upload_address);
     case MAXWELL3D_REG_INDEX(macros.data):
         return macro_engine->AddCode(regs.macros.upload_address, argument);
     case MAXWELL3D_REG_INDEX(macros.bind):

src/video_core/engines/maxwell_dma.cpp

@@ -134,7 +134,8 @@ void MaxwellDMA::CopyBlockLinearToPitch() {
 
     // Deswizzle the input and copy it over.
     UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0);
-    const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in;
+    const u32 bytes_per_pixel =
+        regs.launch_dma.remap_enable ? regs.pitch_out / regs.line_length_in : 1;
     const Parameters& src_params = regs.src_params;
     const u32 width = src_params.width;
     const u32 height = src_params.height;
@@ -166,7 +167,8 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
     UNIMPLEMENTED_IF(regs.launch_dma.remap_enable != 0);
 
     const auto& dst_params = regs.dst_params;
-    const u32 bytes_per_pixel = regs.pitch_in / regs.line_length_in;
+    const u32 bytes_per_pixel =
+        regs.launch_dma.remap_enable ? regs.pitch_in / regs.line_length_in : 1;
     const u32 width = dst_params.width;
     const u32 height = dst_params.height;
     const u32 depth = dst_params.depth;
@@ -210,7 +212,8 @@ void MaxwellDMA::CopyPitchToBlockLinear() {
 }
 
 void MaxwellDMA::FastCopyBlockLinearToPitch() {
-    const u32 bytes_per_pixel = regs.pitch_out / regs.line_length_in;
+    const u32 bytes_per_pixel =
+        regs.launch_dma.remap_enable ? regs.pitch_out / regs.line_length_in : 1;
     const size_t src_size = GOB_SIZE;
     const size_t dst_size = static_cast<size_t>(regs.pitch_out) * regs.line_count;
     u32 pos_x = regs.src_params.origin.x;

src/video_core/macro/macro.cpp

@@ -2,11 +2,15 @@
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #include <cstring>
+#include <fstream>
 #include <optional>
+#include <span>
 
 #include <boost/container_hash/hash.hpp>
 
 #include "common/assert.h"
+#include "common/fs/fs.h"
+#include "common/fs/path_util.h"
 #include "common/settings.h"
 #include "video_core/macro/macro.h"
 #include "video_core/macro/macro_hle.h"
@@ -15,6 +19,23 @@
 
 namespace Tegra {
 
+static void Dump(u64 hash, std::span<const u32> code) {
+    const auto base_dir{Common::FS::GetYuzuPath(Common::FS::YuzuPath::DumpDir)};
+    const auto macro_dir{base_dir / "macros"};
+    if (!Common::FS::CreateDir(base_dir) || !Common::FS::CreateDir(macro_dir)) {
+        LOG_ERROR(Common_Filesystem, "Failed to create macro dump directories");
+        return;
+    }
+    const auto name{macro_dir / fmt::format("{:016x}.macro", hash)};
+    std::fstream macro_file(name, std::ios::out | std::ios::binary);
+    if (!macro_file) {
+        LOG_ERROR(Common_Filesystem, "Unable to open or create file at {}",
+                  Common::FS::PathToUTF8String(name));
+        return;
+    }
+    macro_file.write(reinterpret_cast<const char*>(code.data()), code.size_bytes());
+}
+
 MacroEngine::MacroEngine(Engines::Maxwell3D& maxwell3d)
     : hle_macros{std::make_unique<Tegra::HLEMacro>(maxwell3d)} {}
 
@@ -24,6 +45,11 @@ void MacroEngine::AddCode(u32 method, u32 data) {
     uploaded_macro_code[method].push_back(data);
 }
 
+void MacroEngine::ClearCode(u32 method) {
+    macro_cache.erase(method);
+    uploaded_macro_code.erase(method);
+}
+
 void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
     auto compiled_macro = macro_cache.find(method);
     if (compiled_macro != macro_cache.end()) {
@@ -54,6 +80,9 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
         if (!mid_method.has_value()) {
             cache_info.lle_program = Compile(macro_code->second);
             cache_info.hash = boost::hash_value(macro_code->second);
+            if (Settings::values.dump_macros) {
+                Dump(cache_info.hash, macro_code->second);
+            }
         } else {
             const auto& macro_cached = uploaded_macro_code[mid_method.value()];
             const auto rebased_method = method - mid_method.value();
@@ -63,6 +92,9 @@ void MacroEngine::Execute(u32 method, const std::vector<u32>& parameters) {
                         code.size() * sizeof(u32));
             cache_info.hash = boost::hash_value(code);
             cache_info.lle_program = Compile(code);
+            if (Settings::values.dump_macros) {
+                Dump(cache_info.hash, code);
+            }
         }
 
         if (auto hle_program = hle_macros->GetHLEProgram(cache_info.hash)) {

src/video_core/macro/macro.h

@@ -117,6 +117,9 @@ public:
     // Store the uploaded macro code to compile them when they're called.
     void AddCode(u32 method, u32 data);
 
+    // Clear the code associated with a method.
+    void ClearCode(u32 method);
+
     // Compiles the macro if its not in the cache, and executes the compiled macro
     void Execute(u32 method, const std::vector<u32>& parameters);
 

src/video_core/macro/macro_jit_x64.cpp

@@ -23,7 +23,8 @@ MICROPROFILE_DEFINE(MacroJitExecute, "GPU", "Execute macro JIT", MP_RGB(255, 255
 namespace Tegra {
 namespace {
 constexpr Xbyak::Reg64 STATE = Xbyak::util::rbx;
-constexpr Xbyak::Reg32 RESULT = Xbyak::util::ebp;
+constexpr Xbyak::Reg32 RESULT = Xbyak::util::r10d;
+constexpr Xbyak::Reg64 MAX_PARAMETER = Xbyak::util::r11;
 constexpr Xbyak::Reg64 PARAMETERS = Xbyak::util::r12;
 constexpr Xbyak::Reg32 METHOD_ADDRESS = Xbyak::util::r14d;
 constexpr Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
@@ -31,6 +32,7 @@ constexpr Xbyak::Reg64 BRANCH_HOLDER = Xbyak::util::r15;
 constexpr std::bitset<32> PERSISTENT_REGISTERS = Common::X64::BuildRegSet({
     STATE,
     RESULT,
+    MAX_PARAMETER,
     PARAMETERS,
     METHOD_ADDRESS,
     BRANCH_HOLDER,
@@ -80,7 +82,7 @@ private:
         u32 carry_flag{};
     };
     static_assert(offsetof(JITState, maxwell3d) == 0, "Maxwell3D is not at 0x0");
-    using ProgramType = void (*)(JITState*, const u32*);
+    using ProgramType = void (*)(JITState*, const u32*, const u32*);
 
     struct OptimizerState {
         bool can_skip_carry{};
@@ -112,7 +114,7 @@ void MacroJITx64Impl::Execute(const std::vector<u32>& parameters, u32 method) {
     JITState state{};
     state.maxwell3d = &maxwell3d;
     state.registers = {};
-    program(&state, parameters.data());
+    program(&state, parameters.data(), parameters.data() + parameters.size());
 }
 
 void MacroJITx64Impl::Compile_ALU(Macro::Opcode opcode) {
@@ -488,6 +490,7 @@ void MacroJITx64Impl::Compile() {
     // JIT state
     mov(STATE, Common::X64::ABI_PARAM1);
     mov(PARAMETERS, Common::X64::ABI_PARAM2);
+    mov(MAX_PARAMETER, Common::X64::ABI_PARAM3);
     xor_(RESULT, RESULT);
     xor_(METHOD_ADDRESS, METHOD_ADDRESS);
     xor_(BRANCH_HOLDER, BRANCH_HOLDER);
@@ -598,7 +601,22 @@ bool MacroJITx64Impl::Compile_NextInstruction() {
     return true;
 }
 
+static void WarnInvalidParameter(uintptr_t parameter, uintptr_t max_parameter) {
+    LOG_CRITICAL(HW_GPU,
+                 "Macro JIT: invalid parameter access 0x{:x} (0x{:x} is the last parameter)",
+                 parameter, max_parameter - sizeof(u32));
+}
+
 Xbyak::Reg32 MacroJITx64Impl::Compile_FetchParameter() {
+    Xbyak::Label parameter_ok{};
+    cmp(PARAMETERS, MAX_PARAMETER);
+    jb(parameter_ok, T_NEAR);
+    Common::X64::ABI_PushRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
+    mov(Common::X64::ABI_PARAM1, PARAMETERS);
+    mov(Common::X64::ABI_PARAM2, MAX_PARAMETER);
+    Common::X64::CallFarFunction(*this, &WarnInvalidParameter);
+    Common::X64::ABI_PopRegistersAndAdjustStack(*this, PersistentCallerSavedRegs(), 0);
+    L(parameter_ok);
     mov(eax, dword[PARAMETERS]);
     add(PARAMETERS, sizeof(u32));
     return eax;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -322,6 +322,7 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, Tegra::Engines::Maxw
     };
     host_info = Shader::HostTranslateInfo{
         .support_float16 = device.IsFloat16Supported(),
+        .support_int16 = device.IsShaderInt16Supported(),
         .support_int64 = device.IsShaderInt64Supported(),
         .needs_demote_reorder = driver_id == VK_DRIVER_ID_AMD_PROPRIETARY_KHR ||
                                 driver_id == VK_DRIVER_ID_AMD_OPEN_SOURCE_KHR,

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -325,6 +325,8 @@ const char* ToString(VkResult result) noexcept {
         return "VK_PIPELINE_COMPILE_REQUIRED_EXT";
     case VkResult::VK_RESULT_MAX_ENUM:
         return "VK_RESULT_MAX_ENUM";
+    case VkResult::VK_ERROR_COMPRESSION_EXHAUSTED_EXT:
+        return "VK_ERROR_COMPRESSION_EXHAUSTED_EXT";
     }
     return "Unknown";
 }

src/yuzu/configuration/configure_debug.cpp

@@ -53,6 +53,8 @@ void ConfigureDebug::SetConfiguration() {
     ui->enable_nsight_aftermath->setChecked(Settings::values.enable_nsight_aftermath.GetValue());
     ui->dump_shaders->setEnabled(runtime_lock);
     ui->dump_shaders->setChecked(Settings::values.dump_shaders.GetValue());
+    ui->dump_macros->setEnabled(runtime_lock);
+    ui->dump_macros->setChecked(Settings::values.dump_macros.GetValue());
     ui->disable_macro_jit->setEnabled(runtime_lock);
     ui->disable_macro_jit->setChecked(Settings::values.disable_macro_jit.GetValue());
     ui->disable_loop_safety_checks->setEnabled(runtime_lock);
@@ -83,6 +85,7 @@ void ConfigureDebug::ApplyConfiguration() {
     Settings::values.cpu_debug_mode = ui->enable_cpu_debugging->isChecked();
     Settings::values.enable_nsight_aftermath = ui->enable_nsight_aftermath->isChecked();
     Settings::values.dump_shaders = ui->dump_shaders->isChecked();
+    Settings::values.dump_macros = ui->dump_macros->isChecked();
     Settings::values.disable_shader_loop_safety_checks =
         ui->disable_loop_safety_checks->isChecked();
     Settings::values.disable_macro_jit = ui->disable_macro_jit->isChecked();

src/yuzu/configuration/configure_debug.ui

@@ -118,6 +118,19 @@
         </property>
        </widget>
       </item>
+      <item row="0" column="2">
+       <widget class="QCheckBox" name="dump_macros">
+        <property name="enabled">
+         <bool>true</bool>
+        </property>
+        <property name="toolTip">
+         <string>When checked, it will dump all the macro programs of the GPU</string>
+        </property>
+        <property name="text">
+         <string>Dump Maxwell Macros</string>
+        </property>
+       </widget>
+      </item>
       <item row="0" column="1">
        <widget class="QCheckBox" name="disable_macro_jit">
         <property name="enabled">