[dynarmic] backport WAITPKG based spinlocks

Signed-off-by: lizzie <lizzie@eden-emu.dev>

src/dynarmic/src/dynarmic/backend/x64/block_of_code.cpp

@@ -188,6 +188,8 @@ HostFeature GetHostFeatures() {
         features |= HostFeature::LZCNT;
     if (cpu_info.has(Cpu::tGFNI))
         features |= HostFeature::GFNI;
+    if (cpu_info.has(Cpu::tWAITPKG))
+        features |= HostFeature::WAITPKG;
 
     if (cpu_info.has(Cpu::tBMI2)) {
         // BMI2 instructions such as pdep and pext have been very slow up until Zen 3.

src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.cpp.inc

@@ -430,10 +430,11 @@ void AxxEmitX64::EmitExclusiveWriteMemoryInline(AxxEmitContext& ctx, IR::Inst* i
     const Xbyak::Reg64 vaddr = ctx.reg_alloc.UseGpr(args[1]);
     const Xbyak::Reg32 status = ctx.reg_alloc.ScratchGpr().cvt32();
     const Xbyak::Reg64 tmp = ctx.reg_alloc.ScratchGpr();
+    const Xbyak::Reg64 tmp2 = ctx.reg_alloc.ScratchGpr();
 
     const auto wrapped_fn = exclusive_write_fallbacks[std::make_tuple(ordered, bitsize, vaddr.getIdx(), value.getIdx())];
 
-    EmitExclusiveLock(code, conf, tmp, eax);
+    EmitExclusiveLock(code, conf, tmp, tmp2);
 
     SharedLabel end = GenSharedLabel();
 

src/dynarmic/src/dynarmic/backend/x64/emit_x64_memory.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 /* This file is part of the dynarmic project.
  * Copyright (c) 2022 MerryMage
  * SPDX-License-Identifier: 0BSD
@@ -343,7 +346,7 @@ void EmitExclusiveLock(BlockOfCode& code, const UserConfig& conf, Xbyak::Reg64 p
     }
 
     code.mov(pointer, mcl::bit_cast<u64>(GetExclusiveMonitorLockPointer(conf.global_monitor)));
-    EmitSpinLockLock(code, pointer, tmp);
+    EmitSpinLockLock(code, pointer, tmp, code.HasHostFeature(HostFeature::WAITPKG));
 }
 
 template<typename UserConfig>

src/dynarmic/src/dynarmic/backend/x64/host_feature.h

@@ -35,9 +35,10 @@ enum class HostFeature : u64 {
     BMI2 = 1ULL << 19,
     LZCNT = 1ULL << 20,
     GFNI = 1ULL << 21,
+    WAITPKG = 1ULL << 22,
 
     // Zen-based BMI2
-    FastBMI2 = 1ULL << 22,
+    FastBMI2 = 1ULL << 23,
 
     // Orthographic AVX512 features on 128 and 256 vectors
     AVX512_Ortho = AVX512F | AVX512VL,

src/dynarmic/src/dynarmic/common/spin_lock_x64.cpp

@@ -22,17 +22,46 @@ static const auto default_cg_mode = nullptr; //Allow RWE
 
 namespace Dynarmic {
 
-void EmitSpinLockLock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp) {
+void EmitSpinLockLock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp, bool waitpkg) {
+    // TODO: this is because we lack regalloc - so better to be safe :(
+    if (waitpkg) {
+        code.push(Xbyak::util::eax);
+        code.push(Xbyak::util::ebx);
+        code.push(Xbyak::util::edx);
+    }
     Xbyak::Label start, loop;
-
     code.jmp(start, code.T_NEAR);
     code.L(loop);
-    code.pause();
+    if (waitpkg) {
+        // TODO: This clobbers EAX and EDX did we tell the regalloc?
+        // ARM ptr for address-monitoring
+        code.umonitor(ptr);
+        // tmp.bit[0] = 0: C0.1 | Slow Wakup | Better Savings
+        // tmp.bit[0] = 1: C0.2 | Fast Wakup | Lesser Savings
+        // edx:eax is implicitly used as a 64-bit deadline timestamp
+        // Use the maximum so that we use the operating system's maximum
+        // allowed wait time within the IA32_UMWAIT_CONTROL register
+        // Enter power state designated by tmp and wait for a write to lock_ptr
+        code.mov(Xbyak::util::eax, 0xFFFFFFFF);
+        code.mov(Xbyak::util::edx, Xbyak::util::eax);
+        // TODO: We can only be here because tmp is 1 already - however we repeatedly overwrite it...
+        code.mov(Xbyak::util::ebx, 1);
+        code.umwait(Xbyak::util::ebx);
+        // CF == 1 if we hit the OS-timeout in IA32_UMWAIT_CONTROL without a write
+        // CF == 0 if we exited the wait for any other reason
+    } else {
+        code.pause();
+    }
     code.L(start);
     code.mov(tmp, 1);
     /*code.lock();*/ code.xchg(code.dword[ptr], tmp);
     code.test(tmp, tmp);
     code.jnz(loop, code.T_NEAR);
+    if (waitpkg) {
+        code.pop(Xbyak::util::edx);
+        code.pop(Xbyak::util::ebx);
+        code.pop(Xbyak::util::eax);
+    }
 }
 
 void EmitSpinLockUnlock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp) {
@@ -60,7 +89,7 @@ void SpinLockImpl::Initialize() {
 
     code.align();
     lock = code.getCurr<void (*)(volatile int*)>();
-    EmitSpinLockLock(code, ABI_PARAM1, code.eax);
+    EmitSpinLockLock(code, ABI_PARAM1, code.eax, false);
     code.ret();
 
     code.align();

src/dynarmic/src/dynarmic/common/spin_lock_x64.h

@@ -1,3 +1,6 @@
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
 /* This file is part of the dynarmic project.
  * Copyright (c) 2022 MerryMage
  * SPDX-License-Identifier: 0BSD
@@ -9,7 +12,7 @@
 
 namespace Dynarmic {
 
-void EmitSpinLockLock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp);
+void EmitSpinLockLock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp, bool waitpkg);
 void EmitSpinLockUnlock(Xbyak::CodeGenerator& code, Xbyak::Reg64 ptr, Xbyak::Reg32 tmp);
 
 }  // namespace Dynarmic

src/tests/CMakeLists.txt

@@ -1,3 +1,6 @@
+# SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+# SPDX-License-Identifier: GPL-3.0-or-later
+
 # SPDX-FileCopyrightText: 2018 yuzu Emulator Project
 # SPDX-License-Identifier: GPL-2.0-or-later
 
@@ -21,7 +24,7 @@ add_executable(tests
 
 create_target_directory_groups(tests)
 
-target_link_libraries(tests PRIVATE common core input_common)
+target_link_libraries(tests PRIVATE common core input_common video_core)
 target_link_libraries(tests PRIVATE ${PLATFORM_LIBRARIES} Catch2::Catch2WithMain Threads::Threads)
 
 add_test(NAME tests COMMAND tests)

src/video_core/engines/maxwell_dma.cpp

@@ -156,6 +156,8 @@ void MaxwellDMA::Launch() {
 }
 
 void MaxwellDMA::CopyBlockLinearToPitch() {
+   
+
     u32 bytes_per_pixel = 1;
     DMA::ImageOperand src_operand;
     src_operand.bytes_per_pixel = bytes_per_pixel;

src/video_core/renderer_vulkan/blit_image.cpp

@@ -46,6 +46,38 @@ namespace Vulkan {
 using VideoCommon::ImageViewType;
 
 namespace {
+    
+[[nodiscard]] VkImageAspectFlags AspectMaskFromFormat(VideoCore::Surface::PixelFormat format) {
+    using VideoCore::Surface::SurfaceType;
+    switch (VideoCore::Surface::GetFormatType(format)) {
+    case SurfaceType::ColorTexture:
+        return VK_IMAGE_ASPECT_COLOR_BIT;
+    case SurfaceType::Depth:
+        return VK_IMAGE_ASPECT_DEPTH_BIT;
+    case SurfaceType::Stencil:
+        return VK_IMAGE_ASPECT_STENCIL_BIT;
+    case SurfaceType::DepthStencil:
+        return VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT;
+    default:
+        return VK_IMAGE_ASPECT_COLOR_BIT;
+    }
+}
+
+[[nodiscard]] VkImageSubresourceRange SubresourceRangeFromView(const ImageView& image_view) {
+    auto range = image_view.range;
+    if ((image_view.flags & VideoCommon::ImageViewFlagBits::Slice) != VideoCommon::ImageViewFlagBits{}) {
+        range.base.layer = 0;
+        range.extent.layers = 1;
+    }
+    return VkImageSubresourceRange{
+        .aspectMask = AspectMaskFromFormat(image_view.format),
+        .baseMipLevel = static_cast<u32>(range.base.level),
+        .levelCount = static_cast<u32>(range.extent.levels),
+        .baseArrayLayer = static_cast<u32>(range.base.layer),
+        .layerCount = static_cast<u32>(range.extent.layers),
+    };
+}
+
 struct PushConstants {
     std::array<float, 2> tex_scale;
     std::array<float, 2> tex_offset;
@@ -417,6 +449,40 @@ void TransitionImageLayout(vk::CommandBuffer& cmdbuf, VkImage image, VkImageLayo
                            0, barrier);
 }
 
+void RecordShaderReadBarrier(Scheduler& scheduler, const ImageView& image_view) {
+    const VkImage image = image_view.ImageHandle();
+    const VkImageSubresourceRange subresource_range = SubresourceRangeFromView(image_view);
+    scheduler.RequestOutsideRenderPassOperationContext();
+    scheduler.Record([image, subresource_range](vk::CommandBuffer cmdbuf) {
+        const VkImageMemoryBarrier barrier{
+            .sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
+            .pNext = nullptr,
+            .srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
+                             VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT |
+                             VK_ACCESS_SHADER_WRITE_BIT |
+                             VK_ACCESS_TRANSFER_WRITE_BIT,
+            .dstAccessMask = VK_ACCESS_SHADER_READ_BIT,
+            .oldLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .newLayout = VK_IMAGE_LAYOUT_GENERAL,
+            .srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
+            .image = image,
+            .subresourceRange = subresource_range,
+        };
+        cmdbuf.PipelineBarrier(
+            VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT |
+                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT |
+                VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                VK_PIPELINE_STAGE_TRANSFER_BIT |
+                VK_PIPELINE_STAGE_EARLY_FRAGMENT_TESTS_BIT |
+                VK_PIPELINE_STAGE_LATE_FRAGMENT_TESTS_BIT,
+            VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
+                VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
+            0,
+            barrier);
+    });
+}
+
 void BeginRenderPass(vk::CommandBuffer& cmdbuf, const Framebuffer* framebuffer) {
     const VkRenderPass render_pass = framebuffer->RenderPass();
     const VkFramebuffer framebuffer_handle = framebuffer->Handle();
@@ -484,7 +550,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, Scheduler& scheduler_,
 
 BlitImageHelper::~BlitImageHelper() = default;
 
-void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_view,
+void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
                                 const Region2D& dst_region, const Region2D& src_region,
                                 Tegra::Engines::Fermi2D::Filter filter,
                                 Tegra::Engines::Fermi2D::Operation operation) {
@@ -496,10 +562,12 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
     const VkPipelineLayout layout = *one_texture_pipeline_layout;
     const VkSampler sampler = is_linear ? *linear_sampler : *nearest_sampler;
     const VkPipeline pipeline = FindOrEmplaceColorPipeline(key);
+    const VkImageView src_view = src_image_view.Handle(Shader::TextureType::Color2D);
+
+    RecordShaderReadBarrier(scheduler, src_image_view);
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record([this, dst_region, src_region, pipeline, layout, sampler,
                       src_view](vk::CommandBuffer cmdbuf) {
-        // TODO: Barriers
         const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
         UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
@@ -538,7 +606,7 @@ void BlitImageHelper::BlitColor(const Framebuffer* dst_framebuffer, VkImageView
 }
 
 void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
-                                       VkImageView src_depth_view, VkImageView src_stencil_view,
+                                       ImageView& src_image_view,
                                        const Region2D& dst_region, const Region2D& src_region,
                                        Tegra::Engines::Fermi2D::Filter filter,
                                        Tegra::Engines::Fermi2D::Operation operation) {
@@ -554,10 +622,13 @@ void BlitImageHelper::BlitDepthStencil(const Framebuffer* dst_framebuffer,
     const VkPipelineLayout layout = *two_textures_pipeline_layout;
     const VkSampler sampler = *nearest_sampler;
     const VkPipeline pipeline = FindOrEmplaceDepthStencilPipeline(key);
+    const VkImageView src_depth_view = src_image_view.DepthView();
+    const VkImageView src_stencil_view = src_image_view.StencilView();
+
+    RecordShaderReadBarrier(scheduler, src_image_view);
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record([dst_region, src_region, pipeline, layout, sampler, src_depth_view,
                       src_stencil_view, this](vk::CommandBuffer cmdbuf) {
-        // TODO: Barriers
         const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
         UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
                                        src_stencil_view);
@@ -692,6 +763,7 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb
     const VkSampler sampler = *nearest_sampler;
     const VkExtent2D extent = GetConversionExtent(src_image_view);
 
+    RecordShaderReadBarrier(scheduler, src_image_view);
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record([pipeline, layout, sampler, src_view, extent, this](vk::CommandBuffer cmdbuf) {
         const VkOffset2D offset{
@@ -717,7 +789,6 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb
         const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
         UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
 
-        // TODO: Barriers
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
         cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
                                   nullptr);
@@ -737,6 +808,7 @@ void BlitImageHelper::ConvertDepthStencil(VkPipeline pipeline, const Framebuffer
     const VkSampler sampler = *nearest_sampler;
     const VkExtent2D extent = GetConversionExtent(src_image_view);
 
+    RecordShaderReadBarrier(scheduler, src_image_view);
     scheduler.RequestRenderpass(dst_framebuffer);
     scheduler.Record([pipeline, layout, sampler, src_depth_view, src_stencil_view, extent,
                       this](vk::CommandBuffer cmdbuf) {
@@ -763,7 +835,6 @@ void BlitImageHelper::ConvertDepthStencil(VkPipeline pipeline, const Framebuffer
         const VkDescriptorSet descriptor_set = two_textures_descriptor_allocator.Commit();
         UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
                                        src_stencil_view);
-        // TODO: Barriers
         cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
         cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
                                   nullptr);

src/video_core/renderer_vulkan/blit_image.h

@@ -1,4 +1,7 @@
-// SPDX-FileCopyrightText: Copyright 2020 yuzu Emulator Project
+// SPDX-FileCopyrightText: Copyright 2025 Eden Emulator Project
+// SPDX-License-Identifier: GPL-3.0-or-later
+
+// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
 #pragma once
@@ -43,7 +46,7 @@ public:
                              StateTracker& state_tracker, DescriptorPool& descriptor_pool);
     ~BlitImageHelper();
 
-    void BlitColor(const Framebuffer* dst_framebuffer, VkImageView src_image_view,
+    void BlitColor(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
                    const Region2D& dst_region, const Region2D& src_region,
                    Tegra::Engines::Fermi2D::Filter filter,
                    Tegra::Engines::Fermi2D::Operation operation);
@@ -52,9 +55,9 @@ public:
                    VkImage src_image, VkSampler src_sampler, const Region2D& dst_region,
                    const Region2D& src_region, const Extent3D& src_size);
 
-    void BlitDepthStencil(const Framebuffer* dst_framebuffer, VkImageView src_depth_view,
-                          VkImageView src_stencil_view, const Region2D& dst_region,
-                          const Region2D& src_region, Tegra::Engines::Fermi2D::Filter filter,
+    void BlitDepthStencil(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
+                          const Region2D& dst_region, const Region2D& src_region,
+                          Tegra::Engines::Fermi2D::Filter filter,
                           Tegra::Engines::Fermi2D::Operation operation);
 
     void ConvertD32ToR32(const Framebuffer* dst_framebuffer, const ImageView& src_image_view);

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -1086,8 +1086,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
         return;
     }
     if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT && !is_src_msaa && !is_dst_msaa) {
-        blit_image_helper.BlitColor(dst_framebuffer, src.Handle(Shader::TextureType::Color2D),
-                                    dst_region, src_region, filter, operation);
+        blit_image_helper.BlitColor(dst_framebuffer, src, dst_region, src_region, filter,
+                                    operation);
         return;
     }
     ASSERT(src.format == dst.format);
@@ -1106,8 +1106,8 @@ void TextureCacheRuntime::BlitImage(Framebuffer* dst_framebuffer, ImageView& dst
         }();
         if (!can_blit_depth_stencil) {
             UNIMPLEMENTED_IF(is_src_msaa || is_dst_msaa);
-            blit_image_helper.BlitDepthStencil(dst_framebuffer, src.DepthView(), src.StencilView(),
-                                               dst_region, src_region, filter, operation);
+            blit_image_helper.BlitDepthStencil(dst_framebuffer, src, dst_region, src_region,
+                                               filter, operation);
             return;
         }
     }
@@ -1968,18 +1968,17 @@ bool Image::BlitScaleHelper(bool scale_up) {
             blit_framebuffer =
                 std::make_unique<Framebuffer>(*runtime, view_ptr, nullptr, extent, scale_up);
         }
-        const auto color_view = blit_view->Handle(Shader::TextureType::Color2D);
 
-        runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), color_view, dst_region,
+        runtime->blit_image_helper.BlitColor(blit_framebuffer.get(), *blit_view, dst_region,
                                              src_region, operation, BLIT_OPERATION);
     } else if (aspect_mask == (VK_IMAGE_ASPECT_DEPTH_BIT | VK_IMAGE_ASPECT_STENCIL_BIT)) {
         if (!blit_framebuffer) {
             blit_framebuffer =
                 std::make_unique<Framebuffer>(*runtime, nullptr, view_ptr, extent, scale_up);
         }
-        runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), blit_view->DepthView(),
-                                                    blit_view->StencilView(), dst_region,
-                                                    src_region, operation, BLIT_OPERATION);
+        runtime->blit_image_helper.BlitDepthStencil(blit_framebuffer.get(), *blit_view,
+                                                    dst_region, src_region, operation,
+                                                    BLIT_OPERATION);
     } else {
         // TODO: Use helper blits where applicable
         flags &= ~ImageFlagBits::Rescaled;

src/video_core/vulkan_common/vma.h

@@ -10,4 +10,12 @@
 #define VMA_STATIC_VULKAN_FUNCTIONS 0
 #define VMA_DYNAMIC_VULKAN_FUNCTIONS 1
 
+#ifdef _MSC_VER
+#pragma warning( push )
+#pragma warning( disable : 4189 )
+#endif
 #include "vk_mem_alloc.h"
+
+#ifdef _MSC_VER
+#pragma warning( pop )
+#endif