early-access version 4029

CMakeLists.txt

@@ -305,7 +305,7 @@ find_package(ZLIB 1.2 REQUIRED)
 find_package(zstd 1.5 REQUIRED)
 
 if (NOT YUZU_USE_EXTERNAL_VULKAN_HEADERS)
-    find_package(Vulkan 1.3.256 REQUIRED)
+    find_package(Vulkan 1.3.274 REQUIRED)
 endif()
 
 if (ENABLE_LIBUSB)

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 4028.
+This is the source code for early-access 4029.
 
 ## Legal Notice
 

src/core/arm/nce/arm_nce.cpp

@@ -39,7 +39,7 @@ fpsimd_context* GetFloatingPointState(mcontext_t& host_ctx) {
 }
 
 using namespace Common::Literals;
-constexpr u32 StackSize = 32_KiB;
+constexpr u32 StackSize = 128_KiB;
 
 } // namespace
 

src/core/arm/nce/interpreter_visitor.cpp

@@ -5,8 +5,6 @@
 #include "common/bit_cast.h"
 #include "core/arm/nce/interpreter_visitor.h"
 
-#include <dynarmic/frontend/A64/decoder/a64.h>
-
 namespace Core {
 
 template <u32 BitSize>
@@ -249,6 +247,7 @@ bool InterpreterVisitor::LDR_lit_fpsimd(Imm<2> opc, Imm<19> imm19, Vec Vt) {
         return false;
     }
 
+    // Size in bytes
     const u64 size = 4 << opc.ZeroExtend();
     const u64 offset = imm19.SignExtend<u64>() << 2;
     const u64 address = this->GetPc() + offset;
@@ -530,7 +529,7 @@ bool InterpreterVisitor::SIMDImmediate(bool wback, bool postindex, size_t scale,
     }
     case MemOp::Load: {
         u128 data{};
-        m_memory.ReadBlock(address, &data, datasize);
+        m_memory.ReadBlock(address, &data, datasize / 8);
         this->SetVec(Vt, data);
         break;
     }

src/core/arm/nce/visitor_base.h

@@ -4,9 +4,15 @@
 
 #pragma once
 
+#pragma GCC diagnostic push
+#pragma GCC diagnostic ignored "-Wshadow"
+
 #include <dynarmic/frontend/A64/a64_types.h>
+#include <dynarmic/frontend/A64/decoder/a64.h>
 #include <dynarmic/frontend/imm.h>
 
+#pragma GCC diagnostic pop
+
 namespace Core {
 
 class VisitorBase {

src/shader_recompiler/backend/spirv/spirv_emit_context.cpp

@@ -74,9 +74,17 @@ spv::ImageFormat GetImageFormat(ImageFormat format) {
     throw InvalidArgument("Invalid image format {}", format);
 }
 
+Id GetImageSampledType(EmitContext& ctx, const ImageDescriptor& desc) {
+    if (desc.is_float) {
+        return ctx.F32[1];
+    } else {
+        return ctx.U32[1];
+    }
+}
+
 Id ImageType(EmitContext& ctx, const ImageDescriptor& desc) {
     const spv::ImageFormat format{GetImageFormat(desc.format)};
-    const Id type{ctx.U32[1]};
+    const Id type{GetImageSampledType(ctx, desc)};
     switch (desc.type) {
     case TextureType::Color1D:
         return ctx.TypeImage(type, spv::Dim::Dim1D, false, false, false, 2, format);

src/shader_recompiler/frontend/ir/opcodes.inc

@@ -512,7 +512,7 @@ OPCODE(ImageQueryDimensions,                                U32x4,          Opaq
 OPCODE(ImageQueryLod,                                       F32x4,          Opaque,         Opaque,                                                         )
 OPCODE(ImageGradient,                                       F32x4,          Opaque,         Opaque,         Opaque,         Opaque,         Opaque,         )
 OPCODE(ImageRead,                                           U32x4,          Opaque,         Opaque,                                                         )
-OPCODE(ImageWrite,                                          Void,           Opaque,         Opaque,         U32x4,                                          )
+OPCODE(ImageWrite,                                          Void,           Opaque,         Opaque,         Opaque,                                         )
 
 OPCODE(IsTextureScaled,                                     U1,             U32,                                                                            )
 OPCODE(IsImageScaled,                                       U1,             U32,                                                                            )

src/shader_recompiler/host_translate_info.h

@@ -19,8 +19,10 @@ struct HostTranslateInfo {
     u32 min_ssbo_alignment{};            ///< Minimum alignment supported by the device for SSBOs
     bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry
                                                 ///< passthrough shaders
-    bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
-                                        ///< control flow
+    bool support_conditional_barrier{};  ///< True when the device supports barriers in conditional
+                                         ///< control flow
+    bool support_ufloat_write_as_uint{}; ///< True when the device supports writing float images
+                                         ///< as bitcasts to uint
 };
 
 } // namespace Shader

src/shader_recompiler/ir_opt/texture_pass.cpp

@@ -372,6 +372,10 @@ TexturePixelFormat ReadTexturePixelFormat(Environment& env, const ConstBufferAdd
     return env.ReadTexturePixelFormat(GetTextureHandle(env, cbuf));
 }
 
+bool TexturePixelFormatIsFloat(Environment& env, const ConstBufferAddr& cbuf) {
+    return ReadTexturePixelFormat(env, cbuf) == TexturePixelFormat::B10G11R11_FLOAT;
+}
+
 class Descriptors {
 public:
     explicit Descriptors(TextureBufferDescriptors& texture_buffer_descriptors_,
@@ -428,8 +432,9 @@ public:
             return desc.type == existing.type && desc.format == existing.format &&
                    desc.cbuf_index == existing.cbuf_index &&
                    desc.cbuf_offset == existing.cbuf_offset && desc.count == existing.count &&
-                   desc.size_shift == existing.size_shift;
+                   desc.size_shift == existing.size_shift && desc.is_float == existing.is_float;
         })};
+        // TODO: handle is_float?
         image_descriptors[index].is_written |= desc.is_written;
         image_descriptors[index].is_read |= desc.is_read;
         return index;
@@ -500,6 +505,19 @@ void PatchTexelFetch(IR::Block& block, IR::Inst& inst, TexturePixelFormat pixel_
                               ir.FPMul(ir.ConvertSToF(32, 32, ir.BitCast<IR::U32>(w)), max_value));
     inst.ReplaceUsesWith(converted);
 }
+
+void PatchSmallFloatImageWrite(IR::Block& block, IR::Inst& inst) {
+    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
+
+    const IR::Value old_value{inst.Arg(2)};
+    const IR::F32 x(ir.BitCast<IR::F32>(IR::U32(ir.CompositeExtract(old_value, 0))));
+    const IR::F32 y(ir.BitCast<IR::F32>(IR::U32(ir.CompositeExtract(old_value, 1))));
+    const IR::F32 z(ir.BitCast<IR::F32>(IR::U32(ir.CompositeExtract(old_value, 2))));
+    const IR::F32 w(ir.BitCast<IR::F32>(IR::U32(ir.CompositeExtract(old_value, 3))));
+    const IR::Value converted = ir.CompositeConstruct(x, y, z, w);
+    inst.SetArg(2, converted);
+}
+
 } // Anonymous namespace
 
 void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo& host_info) {
@@ -531,6 +549,9 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
         inst->ReplaceOpcode(IndexedInstruction(*inst));
 
         const auto& cbuf{texture_inst.cbuf};
+        const bool is_float_write{!host_info.support_ufloat_write_as_uint &&
+                                  inst->GetOpcode() == IR::Opcode::ImageWrite &&
+                                  TexturePixelFormatIsFloat(env, cbuf)};
         auto flags{inst->Flags<IR::TextureInstInfo>()};
         bool is_multisample{false};
         switch (inst->GetOpcode()) {
@@ -603,6 +624,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
                     .format = flags.image_format,
                     .is_written = is_written,
                     .is_read = is_read,
+                    .is_float = is_float_write,
                     .cbuf_index = cbuf.index,
                     .cbuf_offset = cbuf.offset,
                     .count = cbuf.count,
@@ -662,6 +684,10 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
                 PatchTexelFetch(*texture_inst.block, *texture_inst.inst, pixel_format);
             }
         }
+
+        if (is_float_write) {
+            PatchSmallFloatImageWrite(*texture_inst.block, *inst);
+        }
     }
 }
 

src/shader_recompiler/shader_info.h

@@ -42,6 +42,7 @@ enum class TexturePixelFormat : u32 {
     R16G16B16A16_SNORM,
     R16G16_SNORM,
     R16_SNORM,
+    B10G11R11_FLOAT,
     OTHER
 };
 
@@ -129,6 +130,7 @@ struct ImageDescriptor {
     ImageFormat format;
     bool is_written;
     bool is_read;
+    bool is_float;
     u32 cbuf_index;
     u32 cbuf_offset;
     u32 count;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -417,7 +417,7 @@ void RasterizerOpenGL::DispatchCompute() {
 void RasterizerOpenGL::ResetCounter(VideoCommon::QueryType type) {
     const auto query_cache_type = MaxwellToVideoCoreQuery(type);
     if (!query_cache_type.has_value()) {
-        UNIMPLEMENTED_MSG("Reset query type: {}", type);
+        UNIMPLEMENTED_IF_MSG(type != VideoCommon::QueryType::Payload, "Reset query type: {}", type);
         return;
     }
     query_cache.ResetCounter(*query_cache_type);

src/video_core/renderer_opengl/gl_shader_cache.cpp

@@ -245,6 +245,7 @@ ShaderCache::ShaderCache(RasterizerOpenGL& rasterizer_, Core::Frontend::EmuWindo
           .min_ssbo_alignment = static_cast<u32>(device.GetShaderStorageBufferAlignment()),
           .support_geometry_shader_passthrough = device.HasGeometryShaderPassthrough(),
           .support_conditional_barrier = device.SupportsConditionalBarriers(),
+          .support_ufloat_write_as_uint = true,
       } {
     if (use_asynchronous_shaders) {
         workers = CreateWorkers();

src/video_core/renderer_vulkan/vk_buffer_cache.cpp

@@ -78,8 +78,15 @@ vk::Buffer CreateBuffer(const Device& device, const MemoryAllocator& memory_allo
 }
 } // Anonymous namespace
 
-Buffer::Buffer(BufferCacheRuntime&, VideoCommon::NullBufferParams null_params)
-    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params), tracker{4096} {}
+Buffer::Buffer(BufferCacheRuntime& runtime, VideoCommon::NullBufferParams null_params)
+    : VideoCommon::BufferBase<VideoCore::RasterizerInterface>(null_params), tracker{4096} {
+    if (runtime.device.HasNullDescriptor()) {
+        return;
+    }
+    device = &runtime.device;
+    buffer = runtime.CreateNullBuffer();
+    is_null = true;
+}
 
 Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rasterizer_,
                VAddr cpu_addr_, u64 size_bytes_)
@@ -93,8 +100,12 @@ Buffer::Buffer(BufferCacheRuntime& runtime, VideoCore::RasterizerInterface& rast
 
 VkBufferView Buffer::View(u32 offset, u32 size, VideoCore::Surface::PixelFormat format) {
     if (!device) {
-        // Null buffer, return a null descriptor
+        // Null buffer supported, return a null descriptor
         return VK_NULL_HANDLE;
+    } else if (is_null) {
+        // Null buffer not supported, adjust offset and size
+        offset = 0;
+        size = 0;
     }
     const auto it{std::ranges::find_if(views, [offset, size, format](const BufferView& view) {
         return offset == view.offset && size == view.size && format == view.format;
@@ -627,9 +638,12 @@ void BufferCacheRuntime::BindTransformFeedbackBuffers(VideoCommon::HostBindings<
 }
 
 void BufferCacheRuntime::ReserveNullBuffer() {
-    if (null_buffer) {
-        return;
+    if (!null_buffer) {
+        null_buffer = CreateNullBuffer();
     }
+}
+
+vk::Buffer BufferCacheRuntime::CreateNullBuffer() {
     VkBufferCreateInfo create_info{
         .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
         .pNext = nullptr,
@@ -644,15 +658,17 @@ void BufferCacheRuntime::ReserveNullBuffer() {
     if (device.IsExtTransformFeedbackSupported()) {
         create_info.usage |= VK_BUFFER_USAGE_TRANSFORM_FEEDBACK_BUFFER_BIT_EXT;
     }
-    null_buffer = memory_allocator.CreateBuffer(create_info, MemoryUsage::DeviceLocal);
+    vk::Buffer ret = memory_allocator.CreateBuffer(create_info, MemoryUsage::DeviceLocal);
     if (device.HasDebuggingToolAttached()) {
-        null_buffer.SetObjectNameEXT("Null buffer");
+        ret.SetObjectNameEXT("Null buffer");
     }
 
     scheduler.RequestOutsideRenderPassOperationContext();
-    scheduler.Record([buffer = *null_buffer](vk::CommandBuffer cmdbuf) {
+    scheduler.Record([buffer = *ret](vk::CommandBuffer cmdbuf) {
         cmdbuf.FillBuffer(buffer, 0, VK_WHOLE_SIZE, 0);
     });
+
+    return ret;
 }
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_buffer_cache.h

@@ -63,6 +63,7 @@ private:
     vk::Buffer buffer;
     std::vector<BufferView> views;
     VideoCommon::UsageTracker tracker;
+    bool is_null{};
 };
 
 class QuadArrayIndexBuffer;
@@ -151,6 +152,7 @@ private:
     }
 
     void ReserveNullBuffer();
+    vk::Buffer CreateNullBuffer();
 
     const Device& device;
     MemoryAllocator& memory_allocator;

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -388,6 +388,9 @@ PipelineCache::PipelineCache(RasterizerVulkan& rasterizer_, const Device& device
         .min_ssbo_alignment = static_cast<u32>(device.GetStorageBufferAlignment()),
         .support_geometry_shader_passthrough = device.IsNvGeometryShaderPassthroughSupported(),
         .support_conditional_barrier = device.SupportsConditionalBarriers(),
+        .support_ufloat_write_as_uint = driver_id != VK_DRIVER_ID_QUALCOMM_PROPRIETARY &&
+                                        driver_id != VK_DRIVER_ID_MESA_TURNIP &&
+                                        driver_id != VK_DRIVER_ID_ARM_PROPRIETARY,
     };
 
     if (device.GetMaxVertexInputAttributes() < Maxwell::NumVertexAttributes) {

src/video_core/shader_environment.cpp

@@ -76,6 +76,8 @@ static Shader::TexturePixelFormat ConvertTexturePixelFormat(const Tegra::Texture
         return Shader::TexturePixelFormat::R16G16_SNORM;
     case VideoCore::Surface::PixelFormat::R16_SNORM:
         return Shader::TexturePixelFormat::R16_SNORM;
+    case VideoCore::Surface::PixelFormat::B10G11R11_FLOAT:
+        return Shader::TexturePixelFormat::B10G11R11_FLOAT;
     default:
         return Shader::TexturePixelFormat::OTHER;
     }

src/video_core/vulkan_common/vulkan_wrapper.cpp

@@ -377,6 +377,8 @@ const char* ToString(VkResult result) noexcept {
         return "VK_OPERATION_DEFERRED_KHR";
     case VkResult::VK_OPERATION_NOT_DEFERRED_KHR:
         return "VK_OPERATION_NOT_DEFERRED_KHR";
+    case VkResult::VK_ERROR_INVALID_VIDEO_STD_PARAMETERS_KHR:
+        return "VK_ERROR_INVALID_VIDEO_STD_PARAMETERS_KHR";
     case VkResult::VK_PIPELINE_COMPILE_REQUIRED_EXT:
         return "VK_PIPELINE_COMPILE_REQUIRED_EXT";
     case VkResult::VK_RESULT_MAX_ENUM: