update README

Embedded fetch shader: Fix overlapping buffer loads (#133 )
Fix overlapping buffer loads. Fixes many games
2026-08-03 11:23:49 +00:00 · 2026-07-30 05:16:37 +02:00 · 2026-07-30 05:08:38 +02:00 · 2026-07-30 00:48:08 +02:00 · 2026-07-30 00:10:40 +02:00 · 2026-07-30 00:10:40 +02:00
26 changed files with 957 additions and 400 deletions
@@ -50,7 +50,7 @@ graphical glitches, low compatibility, and poor performance.
  </tr>
  <tr>
    <td align="center">
-      <strong>Minecraft Legends</strong><br>
+      <strong>Neptunia ReVerse</strong><br>
      <img src="docs/screenshots/ps5-04.png" width="300" alt="Minecraft Legends running in KytyPS5">
    </td>
    <td align="center">
@@ -58,8 +58,20 @@ graphical glitches, low compatibility, and poor performance.
      <img src="docs/screenshots/ps5-05.png" width="300" alt="SILENT HILL: The Short Message running in KytyPS5">
    </td>
  </tr>
+  <tr>
+    <td align="center">
+      <strong>Hellboy</strong><br>
+      <img src="docs/screenshots/ps5-02.png" width="300" alt="Disgaea 6 running in KytyPS5">
+    </td>
+    <td align="center">
+      <strong>Paleo Pines</strong><br>
+      <img src="docs/screenshots/ps5-06.png" width="300" alt="Dreaming Sarah running in KytyPS5">
+    </td>
+  </tr>
 </table>

+<p align="center"><em>And many more...</em></p>
+
 ## Contributing

 Testing games and submitting detailed bug reports are useful ways to contribute. Search existing
@@ -432,11 +432,6 @@ uint64_t SysVirtualReserveAligned(uint64_t address, uint64_t size, uint64_t alig

 	pthread_mutex_lock(&g_virtual_mutex);
 	record_alloc(ret_addr, size);
-	uintptr_t page_start  = ret_addr >> 12u;
-	uintptr_t page_end    = (ret_addr + size - 1) >> 12u;
-	for (uintptr_t page = page_start; page <= page_end; page++) {
-		(*g_protects)[page] = PROT_NONE;
-	}
 	pthread_mutex_unlock(&g_virtual_mutex);

 	return ret_addr;
@@ -470,11 +465,6 @@ bool SysVirtualReserveFixed(uint64_t address, uint64_t size) {
 	if (ptr != MAP_FAILED) {
 		pthread_mutex_lock(&g_virtual_mutex);
 		record_alloc(ret_addr, size);
-		uintptr_t page_start  = ret_addr >> 12u;
-		uintptr_t page_end    = (ret_addr + size - 1) >> 12u;
-		for (uintptr_t page = page_start; page <= page_end; page++) {
-			(*g_protects)[page] = PROT_NONE;
-		}
 		pthread_mutex_unlock(&g_virtual_mutex);

 		return true;
@@ -394,6 +394,7 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 }

 void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
+	(void)SynchronizeBacking(vaddr, size);
 	std::vector<DownloadCopy> copies;
 	{
 		FaultSafeCacheLock lock(this, m_mutex);
@@ -426,7 +427,7 @@ void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size) {
 		    });
 	}
 	if (copies.empty()) {
-		EXIT("BufferCache: GPU-owned invalidation has no dirty byte ranges\n");
+		return;
 	}
 	auto downloads = RecordDownloads(copies);
 	m_scheduler.FinishCurrent();
@@ -50,6 +50,7 @@ public:
 	[[nodiscard]] bool InvalidateMemory(PageFaultAccess access, uint64_t vaddr, uint64_t size,
 	                                    PageFaultPhase phase) noexcept;
 	void               InvalidateMemory(uint64_t vaddr, uint64_t size);
+	void               ReadMemory(uint64_t vaddr, uint64_t size);
 	void               UnmapMemory(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] BufferBinding ObtainBuffer(CommandBuffer& command, uint64_t vaddr, uint64_t size,
 	                                         bool is_written = false, bool is_read = true,
@@ -111,7 +112,6 @@ private:
 	void QueueGarbageDownload(std::span<const DownloadCopy> copies, RetiredBuffer retire);
 	void RefreshInvalidatedRanges(CommandBuffer& command, CachedBuffer& cached, uint64_t vaddr,
 	                              uint64_t size, bool upload);
-	void ReadMemory(uint64_t vaddr, uint64_t size);
 	void DiscardGpuDirtyBytesLocked(uint64_t vaddr, uint64_t size, const char* operation);
 	void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);

@@ -36,7 +36,8 @@ bool GpuResourceManager::InvalidateMemory(PageFaultAccess access, uint64_t vaddr
 }

 bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept {
-	if (!m_page_manager.IsMapped(fault_vaddr, 1)) {
+	constexpr uint64_t fault_size = 8;
+	if (!IsMapped(fault_vaddr, fault_size)) {
 		return false;
 	}
 	if (CommandScheduler::InDeferredOperation()) {
@@ -47,10 +48,15 @@ bool GpuResourceManager::HandleFault(PageFaultAccess access, uint64_t fault_vadd
 	bool       handled = false;
 	const auto resolve = [this, access, fault_vaddr, &handled](CommandProcessor& cp) {
 		cp.BeginReadbackTransaction();
-		(void)m_buffer_cache.SynchronizeBacking(fault_vaddr, 1);
 		{
 			ResourceMutex::FaultScope fault(m_resource_mutex);
-			handled = m_page_manager.HandleFault(access, fault_vaddr);
+			if (access == PageFaultAccess::Write) {
+				m_buffer_cache.InvalidateMemory(fault_vaddr, fault_size);
+				m_texture_cache.InvalidateMemory(fault_vaddr, fault_size);
+			} else {
+				m_buffer_cache.ReadMemory(fault_vaddr, fault_size);
+			}
+			handled = true;
 		}
 		cp.EndReadbackTransaction();
 	};
@@ -88,15 +88,34 @@ TextureCache::~TextureCache() {

 bool TextureCache::SameBacking(const ImageInfo& cached, const ImageInfo& requested,
                               bool exact_format) {
-	const bool unit_extent =
-	    requested.extent.width == 1 && requested.extent.height == 1 && requested.extent.depth == 1;
-	return cached.data == requested.data && cached.extent == requested.extent &&
-	       cached.samples == requested.samples &&
-	       cached.bytes_per_block == requested.bytes_per_block &&
-	       (cached.type == requested.type || unit_extent) &&
-	       (exact_format
-	            ? cached.pixel_format == requested.pixel_format
-	            : ImageViewOps::FormatsCompatible(cached.pixel_format, requested.pixel_format));
+	if (cached.data.address != requested.data.address) {
+		return false;
+	}
+	if (cached.data.size != requested.data.size) {
+		return false;
+	}
+	if (cached.extent != requested.extent) {
+		return false;
+	}
+	if (cached.samples != requested.samples) {
+		return false;
+	}
+	if (cached.bytes_per_block != requested.bytes_per_block) {
+		return false;
+	}
+	if (cached.tile_mode != requested.tile_mode) {
+		return false;
+	}
+	if (!ImageViewOps::FormatsCompatible(cached.pixel_format, requested.pixel_format)) {
+		return false;
+	}
+	if (cached.type != requested.type && requested.extent != vk::Extent3D {1, 1, 1}) {
+		return false;
+	}
+	if (exact_format && cached.pixel_format != requested.pixel_format) {
+		return false;
+	}
+	return true;
 }

 TextureCache::BindingType TextureCache::UploadBinding(const Image& image) {
@@ -735,6 +754,12 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		    (requested.IsVolume() || cached.info.IsVolume())) {
 			return {ExpandImage(requested, cached_id)};
 		}
+		if (requested.tile_mode != cached.info.tile_mode) {
+			if (safe_to_delete) {
+				DeleteImages(std::array {cached_id}, cached_id);
+			}
+			return {merged_id};
+		}
 		if (requested.pixel_format != cached.info.pixel_format ||
 		    requested.data.size <= cached.info.data.size) {
 			const auto result_id = merged_id ? merged_id : cached_id;
@@ -747,12 +772,6 @@ TextureCache::OverlapResult TextureCache::ResolveOverlap(const ImageInfo& reques
 		if (requested.type == cached.info.type && requested.resources > cached.info.resources) {
 			return {ExpandImage(requested, cached_id)};
 		}
-		if (requested.tile_mode != cached.info.tile_mode) {
-			if (safe_to_delete) {
-				DeleteImages(std::array {cached_id}, cached_id);
-			}
-			return {merged_id};
-		}
 		EXIT("TextureCache: unresolvable equal-address image overlap, address=0x%016" PRIx64
 		     " requested=%ux%u "
 		     "cached=%ux%u requested_size=0x%016" PRIx64 " cached_size=0x%016" PRIx64
@@ -1122,7 +1141,7 @@ ImageId TextureCache::FindImage(ImageDesc& desc, bool exact_format) {

 		for (const auto id: candidates) {
 			const auto owner = ResolveOwner(id);
-			if (owner == nullptr || owner->info.data != desc.info.data) {
+			if (owner == nullptr) {
 				continue;
 			}
 			if (SameBacking(owner->info, desc.info, exact_format)) {
@@ -103,10 +103,7 @@ IsSupportedSampledDepthUintResource(const ShaderRecompiler::IR::ImageResource& r

 inline void ValidateStorageColorView(vk::Format image_format, vk::Format view_format,
                                     uint32_t swizzle) noexcept {
-	const auto srgb_view = SrgbStorageViewFormat(image_format);
-	const bool srgb_storage_view =
-	    srgb_view != vk::Format::eUndefined && view_format == srgb_view;
-	if ((image_format != view_format && !srgb_storage_view) ||
+	if (!ImageViewOps::FormatsCompatible(image_format, view_format) ||
 	    !IsValidImageSwizzle(swizzle)) {
 		UnsupportedColorView("storage", image_format, view_format, swizzle);
 	}
@@ -122,7 +119,10 @@ IsSupportedStorageImageResource(const ShaderRecompiler::IR::ImageResource& resou
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray) &&
 	       resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.written &&
-	       !resource.atomic && !resource.depth_compare;
+	       (!resource.atomic ||
+	        (resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint &&
+	         resource.read)) &&
+	       !resource.depth_compare;
 }

 inline void
@@ -14,13 +14,13 @@
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/hostMemory.h"
-#include "graphics/host_gpu/renderer/image/textureCommon.h"
 #include "graphics/host_gpu/renderer/debug.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
+#include "graphics/host_gpu/renderer/image/textureCommon.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
+#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/vma.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/BindingLayout.h"
@@ -318,8 +318,8 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool valid_2d_slice =
 	    (is_color_2d && descriptor.Depth() == 0 && descriptor.BaseArray5() == 0) ||
 	    (is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth());
-	const bool is_2d = resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D &&
-	                   valid_2d_slice;
+	const bool is_2d =
+	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D && valid_2d_slice;
 	const bool is_2d_array =
 	    resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray &&
 	    is_color_2d_array && descriptor.BaseArray5() <= descriptor.Depth();
@@ -340,13 +340,13 @@ static bool IsSupportedStorageTextureDescriptor(const ShaderRecompiler::IR::Imag
 	const bool supported_tile = tile == Prospero::GpuEnumValue(Prospero::TileMode::kLinear) ||
 	                            tile == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
 	                            supported_depth_tile || supported_standard_tile;
+	const auto swizzle        = descriptor.DstSelXYZW();
 	const bool supported_swizzle =
-	    IsValidImageSwizzle(descriptor.DstSelXYZW()) &&
-	    (descriptor.DstSelXYZW() == DstSel(4, 5, 6, 7) || !resource.read);
+	    IsValidImageSwizzle(swizzle) &&
+	    (swizzle == DstSel(4, 5, 6, 7) || !resource.read || resource.atomic);
 	const bool supported_mip_view = descriptor.BaseLevel() == 0 || is_1d || is_2d;
 	return (is_1d || is_1d_array || is_2d || is_2d_array || is_3d) && supported_tile &&
-	       supported_mip_view &&
-	       descriptor.BaseLevel() == descriptor.LastLevel() &&
+	       supported_mip_view && descriptor.BaseLevel() == descriptor.LastLevel() &&
 	       descriptor.LastLevel() <= descriptor.MaxMip() && descriptor.MinLod() == 0 &&
 	       supported_swizzle && descriptor.BCSwizzle() == 0 && !descriptor.MsaaDepth();
 }
@@ -377,8 +377,10 @@ void ValidateStorageTexture(const ShaderRecompiler::IR::ImageResource& resource,
 	const bool encoding_ok   = IsSupportedStorageTextureEncoding(descriptor);
 	const bool uint_resource =
 	    resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint;
-	const bool format_ok = Prospero::IsSupportedTextureFormat(format) &&
-	                       uint_resource == Prospero::IsUintTextureFormat(format);
+	const bool format_ok =
+	    Prospero::IsSupportedTextureFormat(format) &&
+	    uint_resource == Prospero::IsUintTextureFormat(format) &&
+	    (!resource.atomic || format == Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt));
 	if (resource_ok && descriptor_ok && encoding_ok && format_ok && size != 0) {
 		return;
 	}
@@ -620,8 +622,8 @@ RenderExecutor::ResolveTexture(const ShaderRecompiler::IR::ImageResource&   reso

 	const auto              pixel_format        = TextureGetFormat(format);
 	const auto              storage_view_format = SrgbStorageViewFormat(pixel_format);
-	const auto              view_format =
-	    storage && storage_view_format != vk::Format::eUndefined ? storage_view_format
+	const auto              view_format = storage && storage_view_format != vk::Format::eUndefined
+	                                          ? storage_view_format
 	                                          : pixel_format;
 	const auto              block_bytes = Prospero::BlockCompressedBytesPerBlock(format);
 	TextureCache::ImageDesc desc {};
@@ -18,15 +18,16 @@
 #include "graphics/host_gpu/renderer/depthRenderTarget.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
-#include "graphics/host_gpu/renderer/render.h"
-#include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
+#include "graphics/host_gpu/renderer/render.h"
+#include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
 #include "graphics/shader/shader.h"
 #include "kernel/eventQueue.h"
+#include "kernel/memory.h"
 #include "kernel/pthread.h"
 #include "libs/errno.h"

@@ -221,8 +222,7 @@ static void LogDrawTargetState(const char* draw_name, const RenderColorInfo& col
 	LogMrtState(draw_name, buffer, ps_input_info);
 }

-static void LogDrawInputState(const RenderCommandBuffer&   buffer,
-                               const RenderColorInfo&       color,
+static void LogDrawInputState(const RenderCommandBuffer& buffer, const RenderColorInfo& color,
                              const ShaderVertexInputInfo& vs_input_info,
                              uint32_t index_type_and_size, uint32_t index_count,
                              const void* index_addr) {
@@ -512,8 +512,7 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
 		const auto old_image = cache.ResolveOwner(target.image_id);
-		if (old_image == nullptr ||
-		    (!old_image->registered && !old_image->info.data.Empty()) ||
+		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
 		    old_image->binding.needs_rebind) {
 			if (old_image != nullptr) {
 				old_image->binding = {};
@@ -531,8 +530,7 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 			     attachment_samples, target.samples);
 		}
 		const auto& view   = target.desc.view_info;
-		const auto layout =
-		    image.binding.is_bound ? vk::ImageLayout::eGeneral
+		const auto  layout = image.binding.is_bound ? vk::ImageLayout::eGeneral
 		                                            : vk::ImageLayout::eColorAttachmentOptimal;
 		image.Transit(layout,
 		              vk::AccessFlagBits2::eColorAttachmentRead |
@@ -561,8 +559,7 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		depth.depth_meta_clear_enable =
 		    depth.htile &&
 		    cache.IsMetaCleared(depth.htile_buffer_vaddr, depth.desc.view_info.base_layer);
-		depth.depth_load_clear_enable =
-		    depth.depth_clear_enable || depth.depth_meta_clear_enable;
+		depth.depth_load_clear_enable = depth.depth_clear_enable || depth.depth_meta_clear_enable;
 		if (depth.depth_meta_clear_enable &&
 		    !cache.TouchMeta(depth.htile_buffer_vaddr, depth.desc.view_info.base_layer, false)) {
 			EXIT("failed to consume HTile clear state\n");
@@ -572,8 +569,8 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		if (attachment_samples == 0) {
 			attachment_samples = depth.samples;
 		} else if (attachment_samples != depth.samples) {
-			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n",
-			     attachment_samples, depth.samples);
+			EXIT("mixed color/depth sample counts are unsupported: %u and %u\n", attachment_samples,
+			     depth.samples);
 		}
 		const auto layout = depth_attachment_layout(depth);
 		const auto writes = depth.AttachmentWriteAspects();
@@ -595,11 +592,9 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		attachment.image_layout   = layout;
 		attachment.clear_value[0] = std::bit_cast<uint32_t>(depth.depth_clear_value);
 		attachment.clear_value[1] = depth.stencil_clear_value;
-		attachment.has_depth =
-		    static_cast<bool>(aspects & vk::ImageAspectFlagBits::eDepth);
+		attachment.has_depth      = static_cast<bool>(aspects & vk::ImageAspectFlagBits::eDepth);
 		attachment.depth_clear    = depth.depth_load_clear_enable;
-		attachment.has_stencil =
-		    static_cast<bool>(aspects & vk::ImageAspectFlagBits::eStencil);
+		attachment.has_stencil    = static_cast<bool>(aspects & vk::ImageAspectFlagBits::eStencil);
 		attachment.stencil_clear  = depth.stencil_clear_enable;
 	}
 	if (attachment_samples == 0 ||
@@ -685,6 +680,85 @@ static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer
 	                           : buffer.num_records);
 }

+struct VertexBufferRange {
+	uint64_t      base_address  = 0;
+	uint64_t      requested_end = 0;
+	uint64_t      acquired_end  = 0;
+	BufferBinding binding;
+
+	[[nodiscard]] uint64_t RequestedSize() const { return requested_end - base_address; }
+};
+
+static std::vector<BufferBinding> AcquireVertexBuffers(RenderCommandBuffer&         buffer,
+                                                       const ShaderVertexInputInfo& vs_input_info) {
+	// Collect the non-empty guest vertex ranges.
+	std::vector<VertexBufferRange> ranges;
+	ranges.reserve(vs_input_info.buffers_num);
+	for (int i = 0; i < vs_input_info.buffers_num; i++) {
+		const auto& vertex = vs_input_info.buffers[i];
+		const auto  size   = VertexBufferDescriptorSize(vertex);
+		if (size == 0) {
+			continue;
+		}
+		if (vertex.addr == 0 || size > UINT64_MAX - vertex.addr) {
+			EXIT("invalid vertex buffer range: addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
+			     vertex.addr, size);
+		}
+		ranges.push_back({vertex.addr, vertex.addr + size});
+	}
+
+	std::ranges::sort(ranges, [](const VertexBufferRange& left, const VertexBufferRange& right) {
+		return left.base_address < right.base_address;
+	});
+
+	// Merge overlapping or touching ranges before acquiring host buffers.
+	std::vector<VertexBufferRange> merged_ranges;
+	merged_ranges.reserve(ranges.size());
+	for (const auto& range: ranges) {
+		if (!merged_ranges.empty() && merged_ranges.back().requested_end >= range.base_address) {
+			merged_ranges.back().requested_end =
+			    std::max(merged_ranges.back().requested_end, range.requested_end);
+			continue;
+		}
+		merged_ranges.push_back(range);
+	}
+
+	auto& cache = buffer.GetContext().GetBufferCache();
+	for (auto& range: merged_ranges) {
+		// PPSA20298
+		const auto size =
+		    Libs::LibKernel::Memory::ClampRangeSize(range.base_address, range.RequestedSize());
+		range.acquired_end = range.base_address + size;
+		range.binding      = cache.ObtainBuffer(buffer, range.base_address, size);
+	}
+
+	// Rebuild slot bindings, offsetting non-empty slots into their acquired merged range.
+	std::vector<BufferBinding> bindings;
+	bindings.reserve(vs_input_info.buffers_num);
+	for (int i = 0; i < vs_input_info.buffers_num; i++) {
+		const auto& vertex = vs_input_info.buffers[i];
+		const auto  size   = VertexBufferDescriptorSize(vertex);
+		if (size == 0) {
+			auto owner = cache.ObtainNullBuffer();
+			bindings.push_back({owner, owner->Handle(), 0});
+			continue;
+		}
+
+		const auto range = std::ranges::find_if(merged_ranges, [&](const VertexBufferRange& value) {
+			return vertex.addr >= value.base_address && vertex.addr < value.acquired_end;
+		});
+		if (range == merged_ranges.end()) {
+			EXIT("vertex buffer address is outside the acquired range: addr=0x%016" PRIx64 "\n",
+			     vertex.addr);
+		}
+
+		auto binding = range->binding;
+		binding.offset += vertex.addr - range->base_address;
+		bindings.push_back(std::move(binding));
+	}
+	return bindings;
+}
+
 static void SetDrawDebugPhase(RenderCommandBuffer& buffer, uint64_t submit_id,
                              const DrawCallInfo& draw, uint32_t phase) {
 	EXIT_IF(draw.name == nullptr);
@@ -823,37 +897,13 @@ static std::vector<BufferBinding> PrepareVertexBuffers(uint64_t
 	(void)submit_id;

 	LogDrawPhase(draw.name, "PrepareVertexBuffers");
-	std::vector<BufferBinding> bindings;
-	bindings.reserve(vs_input_info.buffers_num);
-	for (int i = 0; i < vs_input_info.buffers_num; i++) {
-		const auto& b    = vs_input_info.buffers[i];
-		const auto  size = VertexBufferDescriptorSize(b);
-		if (size == 0) {
-			auto owner = buffer.GetContext().GetBufferCache().ObtainNullBuffer();
-			bindings.push_back({owner, owner->Handle(), 0});
-		} else {
-			bindings.push_back(
-			    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, b.addr, size));
-		}
-	}
-	return bindings;
+	return AcquireVertexBuffers(buffer, vs_input_info);
 }

 static void RebindVertexBuffers(RenderCommandBuffer&         buffer,
                                const ShaderVertexInputInfo& vs_input_info,
                                std::vector<BufferBinding>&  bindings) {
-	EXIT_IF(bindings.size() != static_cast<size_t>(vs_input_info.buffers_num));
-	for (int i = 0; i < vs_input_info.buffers_num; i++) {
-		const auto& vertex = vs_input_info.buffers[i];
-		const auto  size   = VertexBufferDescriptorSize(vertex);
-		if (size == 0) {
-			auto owner  = buffer.GetContext().GetBufferCache().ObtainNullBuffer();
-			bindings[i] = {owner, owner->Handle(), 0};
-		} else {
-			bindings[i] =
-			    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, vertex.addr, size);
-		}
-	}
+	bindings = AcquireVertexBuffers(buffer, vs_input_info);
 }

 static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffer,
@@ -1228,8 +1278,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 }

 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer,
-                              uint32_t index_count,
+void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, uint32_t index_count,
                              uint32_t flags, uint32_t render_target_slice_offset,
                              uint32_t instance_count, uint32_t first_vertex,
                              uint32_t first_instance) {
@@ -1290,7 +1339,8 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer,
 	                         instance_count,  first_instance};

 	DrawRenderState state {};
-	if (!PrepareDrawRenderState(submit_id, buffer, draw, render_target_slice_offset, false, state)) {
+	if (!PrepareDrawRenderState(submit_id, buffer, draw, render_target_slice_offset, false,
+	                            state)) {
 		ResetBindings();
 		return;
 	}
@@ -1369,8 +1419,7 @@ bool RenderExecutor::ResolveColorTargets(uint64_t submit_id, RenderCommandBuffer
 	cache.MarkGpuWritten(dst.image_id);
 	auto& source      = cache.GetImage(src.image_id);
 	auto& destination = cache.GetImage(dst.image_id);
-	destination.Resolve(source,
-	                    {src.base_mip_level, 1, src.base_array_layer, 1},
+	destination.Resolve(source, {src.base_mip_level, 1, src.base_array_layer, 1},
 	                    {dst.base_mip_level, 1, dst.base_array_layer, 1});
 	return true;
 }
@@ -442,11 +442,6 @@ bool DecodeDs(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
 	     inst.opcode == Opcode::DsReadAddtidB32)) {
 		SetUnsupported(inst, Family::DS, opcode, "DS swizzle/addtid is available only for LDS");
 	}
-	if (inst.gds && (inst.opcode == Opcode::DsAppend || inst.opcode == Opcode::DsConsume) &&
-	    inst.offset != 0u) {
-		SetUnsupported(inst, Family::DS, opcode,
-		               "GDS append/consume requires a zero instruction offset");
-	}
 	if (inst.opcode == Opcode::DsWriteAddtidB32 && data1 != 0u) {
 		SetUnsupported(inst, Family::DS, opcode,
 		               "DS write addtid data1 operand is not implemented");
@@ -1,7 +1,7 @@
 #include "graphics/shader/recompiler/emitter/SpirvEmitter.h"

-#include "graphics/shader/recompiler/ir/SrtWalker.h"
 #include "graphics/shader/recompiler/emitter/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/ir/SrtWalker.h"

 #include <algorithm>
 #include <array>
@@ -160,8 +160,7 @@ bool ValidateInstructionContract(const IR::Instruction& inst, std::string* error
 	                            inst.dst.kind != IR::OperandKind::Null)) ||
 	    ((inst.op == IR::Opcode::DsAppend || inst.op == IR::Opcode::DsConsume) &&
 	     (!ds_kind || !ds_resource || inst.src_count != 1 ||
-	      inst.dst.kind != IR::OperandKind::Register ||
-	      (kind == IR::ResourceKind::Gds && inst.memory.offset != 0))) ||
+	      inst.dst.kind != IR::OperandKind::Register)) ||
 	    ((inst.op == IR::Opcode::DsMinF32 || inst.op == IR::Opcode::DsMaxF32) &&
 	     (!ds_kind || !ds_resource || inst.src_count != 3 ||
 	      inst.dst.kind != IR::OperandKind::Null)) ||
@@ -1025,15 +1025,47 @@ void EmitDispatcherSwitch(EmitterState& state, const IR::Program& program) {
 	EmitDispatcherExit(state);
 }

+size_t BufferLoadGroupSize(const IR::BasicBlock& block, size_t first_index) {
+	const auto& first = block.instructions[first_index];
+	if (first.op != IR::Opcode::BufferLoadDword || first.memory.component_index != 0u ||
+	    first.memory.component_count <= 1u) {
+		return 1u;
+	}
+
+	size_t count = 1u;
+	while (first_index + count < block.instructions.size() &&
+	       count < first.memory.component_count) {
+		const auto& next = block.instructions[first_index + count];
+		if (next.op != IR::Opcode::BufferLoadDword || next.pc != first.pc ||
+		    next.memory.component_index != count ||
+		    next.memory.component_count != first.memory.component_count) {
+			break;
+		}
+		count++;
+	}
+	return count;
+}
+
+void EmitBlockInstructions(EmitterState& state, const IR::BasicBlock& block) {
+	for (size_t i = 0; i < block.instructions.size();) {
+		const auto count = BufferLoadGroupSize(block, i);
+		if (count > 1u) {
+			EmitBufferLoadDwordGroup(state, block.instructions.data() + i,
+			                         static_cast<uint32_t>(count));
+		} else {
+			EmitInstruction(state, block.instructions[i]);
+		}
+		i += count;
+	}
+}
+
 void EmitDispatcherBlocks(EmitterState& state, const IR::Program& program) {
 	for (const auto& block: program.blocks) {
 		if (block.id >= state.reachable_blocks.size() || !state.reachable_blocks[block.id]) {
 			continue;
 		}
 		state.builder.AddFunction({OpLabel, BlockLabel(state, block.id)});
-		for (const auto& inst: block.instructions) {
-			EmitInstruction(state, inst);
-		}
+		EmitBlockInstructions(state, block);
 		EmitDispatcherTerminator(state, block.terminator);
 	}
 }
@@ -1083,9 +1115,7 @@ void EmitFunction(EmitterState& state, const IR::Program& program) {
 			continue;
 		}
 		state.builder.AddFunction({OpLabel, BlockLabel(state, block.id)});
-		for (const auto& inst: block.instructions) {
-			EmitInstruction(state, inst);
-		}
+		EmitBlockInstructions(state, block);
 		EmitTerminator(state, block.terminator);
 	}

@@ -3,11 +3,11 @@

 #include "common/common.h"
 #include "common/stringUtils.h"
-#include "graphics/shader/recompiler/ir/BindingLayout.h"
 #include "graphics/shader/recompiler/BufferFormat.h"
+#include "graphics/shader/recompiler/emitter/SpirvBuilder.h"
+#include "graphics/shader/recompiler/ir/BindingLayout.h"
 #include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/recompiler/emitter/SpirvBuilder.h"

 #include <algorithm>
 #include <array>
@@ -990,11 +990,6 @@ uint32_t NormalizeFormatComponent(EmitterState& state, const Format::BufferForma
 uint32_t UnpackTBufferFormat(EmitterState& state, const IR::Instruction& inst,
                             const Format::BufferFormatInfo& info);

-bool EmitTypedTBufferLoad(EmitterState& state, const IR::Instruction& inst,
-                          const Format::BufferFormatInfo& info);
-
-bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst);
-
 uint32_t FormattedBufferDwordStoreComponentCount(Prospero::BufferFormat format,
                                                 uint32_t               opcode_components);

@@ -1020,6 +1015,9 @@ void EmitBufferLoadSshort(EmitterState& state, const IR::Instruction& inst);

 void EmitBufferLoadDword(EmitterState& state, const IR::Instruction& inst);

+void EmitBufferLoadDwordGroup(EmitterState& state, const IR::Instruction* instructions,
+                              uint32_t count);
+
 void EmitBufferStoreDword(EmitterState& state, const IR::Instruction& inst);

 void EmitFlatLoadUbyte(EmitterState& state, const IR::Instruction& inst);
@@ -34,16 +34,15 @@ uint32_t EmitDppWriteActiveBool(EmitterState& state, const IR::Operand& dst) {
 	    {OpShiftLeftLogical, state.uint_type, bank_bit, ConstantU32(state, 1), bank});
 	state.builder.AddFunction(
 	    {OpShiftLeftLogical, state.uint_type, row_bit, ConstantU32(state, 1), row});
-	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, bank_hit,
-	                            ConstantU32(state, dst.dpp_bank_mask), bank_bit});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, state.uint_type, bank_hit, ConstantU32(state, dst.dpp_bank_mask), bank_bit});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, row_hit, ConstantU32(state, dst.dpp_row_mask), row_bit});
 	state.builder.AddFunction(
 	    {OpINotEqual, state.bool_type, bank_active, bank_hit, ConstantU32(state, 0)});
 	state.builder.AddFunction(
 	    {OpINotEqual, state.bool_type, row_active, row_hit, ConstantU32(state, 0)});
-	state.builder.AddFunction(
-	    {OpLogicalAnd, state.bool_type, dpp_active, bank_active, row_active});
+	state.builder.AddFunction({OpLogicalAnd, state.bool_type, dpp_active, bank_active, row_active});
 	uint32_t write_active = dpp_active;
 	if (!dst.dpp_bound_ctrl) {
 		const auto target  = EmitDppTargetLane(state, dst.dpp_ctrl);
@@ -131,8 +130,7 @@ uint32_t EmitNotEqualZeroBool(EmitterState& state, uint32_t value) {
 uint32_t EmitSelectU32Value(EmitterState& state, uint32_t condition, uint32_t true_value,
                            uint32_t false_value) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpSelect, state.uint_type, ret, condition, true_value, false_value});
+	state.builder.AddFunction({OpSelect, state.uint_type, ret, condition, true_value, false_value});
 	return ret;
 }

@@ -212,12 +210,12 @@ bool IsStorageBufferMemoryKind(IR::ResourceKind kind) {

 void EmitStorageBufferOffsets(EmitterState& state) {
 	for (uint32_t i = 0; i < state.program.bindings.buffer_offset_count; i++) {
-		const auto word = EmitShaderDataDwordLoad(
-		    state, state.program.bindings.buffer_offset_dword + i / 4u);
+		const auto word =
+		    EmitShaderDataDwordLoad(state, state.program.bindings.buffer_offset_dword + i / 4u);
 		const auto shift                = ConstantU32(state, (i % 4u) * 8u + 2u);
 		state.storage_buffer_offsets[i] = EmitBinaryU32(
-		    state, OpBitwiseAnd,
-		    EmitBinaryU32(state, OpShiftRightLogical, word, shift), ConstantU32(state, 0x3fu));
+		    state, OpBitwiseAnd, EmitBinaryU32(state, OpShiftRightLogical, word, shift),
+		    ConstantU32(state, 0x3fu));
 	}
 }

@@ -329,8 +327,7 @@ uint32_t EmitRelativeAddress(EmitterState& state, const IR::Instruction& inst, u

 uint32_t EmitFlatVirtualAddress(EmitterState& state, const IR::Instruction& inst,
                                uint32_t first_src, uint32_t src_count) {
-	if (inst.memory.resource >= state.resources.addresses.size() ||
-	    src_count < 2) {
+	if (inst.memory.resource >= state.resources.addresses.size() || src_count < 2) {
 		ExitDescriptorBindingFailure(state, IR::DescriptorBindingKind::AddressMemory,
 		                             inst.memory.resource, "flat address snapshot is missing");
 	}
@@ -659,8 +656,7 @@ void EmitAtomicUpdateU32(EmitterState& state, uint32_t pointer, IR::ResourceKind
 	state.builder.AddFunction({OpBranch, preheader});
 	state.builder.AddFunction({OpLabel, preheader});
 	state.builder.AddFunction({OpAtomicLoad, state.uint_type, initial, pointer,
-	                            ConstantU32(state, scope),
-	                            ConstantU32(state, MemorySemanticsNone)});
+	                           ConstantU32(state, scope), ConstantU32(state, MemorySemanticsNone)});
 	state.builder.AddFunction({OpBranch, header});
 	state.builder.AddFunction({OpLabel, header});
 	state.builder.AddFunction(
@@ -793,8 +789,7 @@ uint32_t EmitTBufferBitcastU32ToI32(EmitterState& state, uint32_t value) {
 uint32_t EmitTBufferCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
                                       uint32_t constant) {
 	const auto ret = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {opcode, state.bool_type, ret, value, ConstantU32(state, constant)});
+	state.builder.AddFunction({opcode, state.bool_type, ret, value, ConstantU32(state, constant)});
 	return ret;
 }

@@ -961,18 +956,8 @@ uint32_t UnpackTBufferFormat(EmitterState& state, const IR::Instruction& inst,
 	return NormalizeFormatComponent(state, info, inst.memory.component_index, raw);
 }

-bool EmitTypedTBufferLoad(EmitterState& state, const IR::Instruction& inst,
-                          const Format::BufferFormatInfo& info) {
-	if (!Format::CanUseTypedBufferLoad(info.format)) {
-		return false;
-	}
-	const auto value = EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
-	                                               AddressSourceCount(inst, 0));
-	EmitStoreU32(state, inst.dst, value);
-	return true;
-}
-
-bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst) {
+bool EmitFormattedBufferLoadValueU32(EmitterState& state, const IR::Instruction& inst,
+                                     uint32_t& value) {
 	if (!IsFormattedBufferComponent(inst)) {
 		return false;
 	}
@@ -984,18 +969,29 @@ bool EmitFormattedBufferLoad(EmitterState& state, const IR::Instruction& inst) {

 	const auto info = Format::GetFormatInfo(format);
 	if (inst.memory.component_index >= info.component_count) {
-		EmitStoreU32(state, inst.dst, ConstantU32(state, 0));
+		value = ConstantU32(state, 0);
 		return true;
 	}

-	if (EmitTypedTBufferLoad(state, inst, info)) {
+	if (Format::CanUseTypedBufferLoad(info.format)) {
+		value = EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
+		                                    AddressSourceCount(inst, 0));
 		return true;
 	}

-	EmitStoreU32(state, inst.dst, UnpackTBufferFormat(state, inst, info));
+	value = UnpackTBufferFormat(state, inst, info);
 	return true;
 }

+uint32_t EmitBufferLoadDwordValueU32(EmitterState& state, const IR::Instruction& inst) {
+	uint32_t value = 0;
+	if (EmitFormattedBufferLoadValueU32(state, inst, value)) {
+		return value;
+	}
+	return EmitMemoryLoadDwordValueU32(state, inst, IR::ResourceKind::Buffer, 0,
+	                                   AddressSourceCount(inst, 0));
+}
+
 uint32_t FormattedBufferDwordStoreComponentCount(Prospero::BufferFormat format,
                                                 uint32_t               opcode_components) {
 	switch (format) {
@@ -1262,11 +1258,27 @@ void EmitBufferLoadSshort(EmitterState& state, const IR::Instruction& inst) {
 }

 void EmitBufferLoadDword(EmitterState& state, const IR::Instruction& inst) {
-	EmitGuardedByExec(state, [&]() {
-		if (EmitFormattedBufferLoad(state, inst)) {
+	EmitGuardedByExec(
+	    state, [&]() { EmitStoreU32(state, inst.dst, EmitBufferLoadDwordValueU32(state, inst)); });
+}
+
+void EmitBufferLoadDwordGroup(EmitterState& state, const IR::Instruction* instructions,
+                              uint32_t count) {
+	if (instructions == nullptr || count == 0u) {
 		return;
 	}
-		EmitMemoryLoadU32(state, inst, IR::ResourceKind::Buffer, 0, AddressSourceCount(inst, 0));
+
+	EmitGuardedByExec(state, [&]() {
+		// RDNA VMEM captures every VADDR component before making overlapping VDATA writes
+		// visible. Keep the split IR components instruction-atomic by deferring all stores.
+		std::vector<uint32_t> values;
+		values.reserve(count);
+		for (uint32_t i = 0; i < count; i++) {
+			values.push_back(EmitBufferLoadDwordValueU32(state, instructions[i]));
+		}
+		for (uint32_t i = 0; i < count; i++) {
+			EmitStoreU32(state, instructions[i].dst, values[i]);
+		}
 	});
 }

@@ -1482,8 +1494,7 @@ void EmitDsAppendConsume(EmitterState& state, const IR::Instruction& inst, uint3
 	const auto do_atomic  = state.builder.AllocateId();
 	state.builder.AddFunction({OpIEqual, state.bool_type, first_lane, subid, exec.first_lane});
 	const auto first_active = EmitLogicalAndBool(state, first_lane, exec.any_active);
-	state.builder.AddFunction(
-	    {OpLogicalAnd, state.bool_type, do_atomic, first_active, in_bounds});
+	state.builder.AddFunction({OpLogicalAnd, state.bool_type, do_atomic, first_active, in_bounds});

 	const auto atomic_value = EmitValueOrZeroIfCondition(state, do_atomic, [&]() {
 		const auto pointer = gds ? EmitGdsElementPointer(state, address.index)
@@ -1575,14 +1586,13 @@ uint32_t EmitDsSwizzleTargetLane(EmitterState& state, uint32_t subid, uint32_t c
 	const auto xored  = state.builder.AllocateId();
 	const auto base   = state.builder.AllocateId();
 	const auto target = state.builder.AllocateId();
-	state.builder.AddFunction(
-	    {OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 31)});
+	state.builder.AddFunction({OpBitwiseAnd, state.uint_type, lane, subid, ConstantU32(state, 31)});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, masked, lane, ConstantU32(state, control & 0x1fu)});
 	state.builder.AddFunction(
 	    {OpBitwiseOr, state.uint_type, ored, masked, ConstantU32(state, (control >> 5u) & 0x1fu)});
-	state.builder.AddFunction({OpBitwiseXor, state.uint_type, xored, ored,
-	                            ConstantU32(state, (control >> 10u) & 0x1fu)});
+	state.builder.AddFunction(
+	    {OpBitwiseXor, state.uint_type, xored, ored, ConstantU32(state, (control >> 10u) & 0x1fu)});
 	state.builder.AddFunction(
 	    {OpBitwiseAnd, state.uint_type, base, subid, ConstantU32(state, 0xffffffe0u)});
 	state.builder.AddFunction({OpBitwiseOr, state.uint_type, target, base, xored});
@@ -784,10 +784,10 @@ private:
 			if (incoming.empty()) {
 				return ScalarProvenance::Undefined;
 			}
+			if (*phi == ScalarProvenance::Undefined) {
 				if (incoming.size() == 1) {
 					return incoming[0];
 				}
-			if (*phi == ScalarProvenance::Undefined) {
 				*phi = AddValue({ScalarValueOp::Phi, block.start_pc});
 			}
 			m_graph.values[*phi].phi_args = std::move(incoming);
@@ -528,6 +528,42 @@ public:
 		return false;
 	}

+	uint64_t ClampRangeSize(uint64_t virtual_addr, uint64_t size) {
+		Common::LockGuard lock(m_mutex);
+
+		if (virtual_addr == 0 || size == 0 || size > UINT64_MAX - virtual_addr) {
+			return 0;
+		}
+
+		auto vma = std::upper_bound(
+		    m_ranges.begin(), m_ranges.end(), virtual_addr,
+		    [](uint64_t value, const Range& range) { return value < range.start; });
+		if (vma == m_ranges.begin()) {
+			return 0;
+		}
+		--vma;
+
+		const auto vma_end = End(vma->start, vma->size);
+		if (virtual_addr < vma->start || virtual_addr >= vma_end ||
+		    !IsCommittedRangeType(vma->type)) {
+			return 0;
+		}
+
+		uint64_t clamped_size = std::min(size, vma_end - virtual_addr);
+		uint64_t expected     = virtual_addr + clamped_size;
+		++vma;
+
+		while (vma != m_ranges.end() && vma->start == expected && IsCommittedRangeType(vma->type) &&
+		       clamped_size < size) {
+			const auto chunk = std::min(size - clamped_size, vma->size);
+			clamped_size += chunk;
+			expected += chunk;
+			++vma;
+		}
+
+		return clamped_size;
+	}
+
 	uint64_t CountPageTableEntries(bool gpu) {
 		Common::LockGuard lock(m_mutex);

@@ -884,6 +920,23 @@ bool TryReadBacking(uint64_t vaddr, void* data, uint64_t size) {
 	       g_direct_memory_backing->TryReadBacking(vaddr, data, size);
 }

+uint64_t ClampRangeSize(uint64_t vaddr, uint64_t size) {
+	EXIT_IF(g_virtual_ranges == nullptr);
+
+	const auto clamped_size = g_virtual_ranges->ClampRangeSize(vaddr, size);
+	if (clamped_size == 0) {
+		EXIT("Memory: attempted to access invalid address 0x%016" PRIx64 " with size 0x%016" PRIx64
+		     "\n",
+		     vaddr, size);
+	}
+	if (clamped_size != size) {
+		LOGF("Memory: clamped buffer range addr=0x%016" PRIx64 " size=0x%016" PRIx64
+		     " to 0x%016" PRIx64 "\n",
+		     vaddr, size, clamped_size);
+	}
+	return clamped_size;
+}
+
 void WriteBacking(uint64_t vaddr, const void* data, uint64_t size) noexcept {
 	if (!TryWriteBacking(vaddr, data, size)) {
 		EXIT("Memory: required direct-backing write failed, addr=0x%016" PRIx64
@@ -103,6 +103,7 @@ void               RegisterCallbacks(callback_func_t alloc_func, callback_func_t
 void                   SetFlexibleMemorySize(uint64_t size);
 bool                   TryWriteBacking(uint64_t vaddr, const void* data, uint64_t size);
 bool                   TryReadBacking(uint64_t vaddr, void* data, uint64_t size);
+[[nodiscard]] uint64_t ClampRangeSize(uint64_t vaddr, uint64_t size);
 void                   WriteBacking(uint64_t vaddr, const void* data, uint64_t size) noexcept;
 void                   InvalidateMemory(uint64_t vaddr, uint64_t size);
 void                   InstallGpuResources(Graphics::GpuResourceManager* resources) noexcept;
@@ -654,6 +654,8 @@ namespace LibHttp2 {
 LIB_VERSION("Http2", 1, "Http2", 1, 1);

 constexpr int HTTP2_ERROR_INVALID_ID   = -2122641152; /* 0x817B1100 */
+constexpr int HTTP2_ERROR_BEFORE_SEND  = -2122641307; /* 0x817B1065 */
+constexpr int HTTP2_ERROR_TIMEOUT      = -2122641304; /* 0x817B1068 */
 constexpr int HTTP2_ERROR_NULL_POINTER = -2122640859; /* 0x817B1225 */

 struct Http2Options {
@@ -694,12 +696,12 @@ struct Http2Request {
 	std::string                                      url;
 	uint64_t                                         content_length = 0;
 	std::vector<std::pair<std::string, std::string>> headers;
-	bool                                             sent        = false;
-	int                                              status_code = 204;
-	std::string  response_headers                                = "HTTP/2 204 No Content\r\n\r\n";
+	int                                              send_result = HTTP2_ERROR_BEFORE_SEND;
+	int                                              status_code = 0;
+	std::string                                      response_headers;
 	std::string                                      response_body;
 	size_t                                           read_offset  = 0;
-	int          async_result = 0;
+	int                                              async_result = HTTP2_ERROR_BEFORE_SEND;
 	int                                              async_event  = 0;
 	Http2Options                                     options;
 };
@@ -1114,9 +1116,9 @@ static int KYTY_SYSV_ABI Http2SendRequest(int req_id, const void* post_data, siz
 		return HTTP2_ERROR_INVALID_ID;
 	}

-	request->second.sent = true;
+	request->second.send_result = HTTP2_ERROR_TIMEOUT;

-	return 0;
+	return request->second.send_result;
 }

 static int KYTY_SYSV_ABI Http2SendRequestAsync(int req_id, const void* post_data, size_t size,
@@ -1136,8 +1138,8 @@ static int KYTY_SYSV_ABI Http2SendRequestAsync(int req_id, const void* post_data
 		return HTTP2_ERROR_INVALID_ID;
 	}

-	request->second.sent         = true;
-	request->second.async_result = 0;
+	request->second.send_result  = HTTP2_ERROR_TIMEOUT;
+	request->second.async_result = request->second.send_result;
 	request->second.async_event  = 0;

 	return 0;
@@ -1159,7 +1161,6 @@ static int KYTY_SYSV_ABI Http2WaitAsync(int req_id, Http2AsyncResult* result, ui
 		return HTTP2_ERROR_INVALID_ID;
 	}

-	request->second.sent = true;
 	*result            = {};
 	result->event_type = request->second.async_event;
 	result->req_id     = req_id;
@@ -1178,13 +1179,19 @@ static int KYTY_SYSV_ABI Http2GetStatusCode(int req_id, int* status_code) {
 		return HTTP2_ERROR_NULL_POINTER;
 	}

+	*status_code = 0;
+
 	auto request = g_http2_requests.find(req_id);
 	if (request == g_http2_requests.end()) {
 		return HTTP2_ERROR_INVALID_ID;
 	}

-	*status_code = request->second.status_code;
+	const int send_result = request->second.send_result;
+	if (send_result != 0) {
+		return send_result;
+	}

+	*status_code = request->second.status_code;
 	return 0;
 }

@@ -1200,14 +1207,21 @@ static int KYTY_SYSV_ABI Http2GetResponseContentLength(int req_id, int* result,
 		return HTTP2_ERROR_NULL_POINTER;
 	}

+	*result         = 0;
+	*content_length = 0;
+
 	auto request = g_http2_requests.find(req_id);
 	if (request == g_http2_requests.end()) {
 		return HTTP2_ERROR_INVALID_ID;
 	}

-	*result         = 0; // SCE_HTTP2_CONTENTLEN_EXIST
-	*content_length = request->second.response_body.size();
+	const int send_result = request->second.send_result;
+	if (send_result != 0) {
+		*result = -1;
+		return send_result;
+	}

+	*content_length = request->second.response_body.size();
 	return 0;
 }

@@ -1223,14 +1237,21 @@ static int KYTY_SYSV_ABI Http2GetAllResponseHeaders(int req_id, char** header,
 		return HTTP2_ERROR_NULL_POINTER;
 	}

+	*header      = nullptr;
+	*header_size = 0;
+
 	auto request = g_http2_requests.find(req_id);
 	if (request == g_http2_requests.end()) {
 		return HTTP2_ERROR_INVALID_ID;
 	}

+	const int send_result = request->second.send_result;
+	if (send_result != 0) {
+		return send_result;
+	}
+
 	*header      = const_cast<char*>(request->second.response_headers.c_str());
 	*header_size = request->second.response_headers.size();
-
 	return 0;
 }

@@ -1250,6 +1271,11 @@ static int KYTY_SYSV_ABI Http2ReadData(int req_id, void* data, size_t size) {
 		return HTTP2_ERROR_INVALID_ID;
 	}

+	const int send_result = request->second.send_result;
+	if (send_result != 0) {
+		return send_result;
+	}
+
 	const auto& body = request->second.response_body;
 	const auto  remaining =
 	    request->second.read_offset < body.size() ? body.size() - request->second.read_offset : 0;
@@ -1280,6 +1306,13 @@ static int KYTY_SYSV_ABI Http2ReadDataAsync(int req_id, void* data, size_t size,
 		return HTTP2_ERROR_INVALID_ID;
 	}

+	const int send_result = request->second.send_result;
+	if (send_result != 0) {
+		request->second.async_result = send_result;
+		request->second.async_event  = 1;
+		return 0;
+	}
+
 	const auto& body = request->second.response_body;
 	const auto  remaining =
 	    request->second.read_offset < body.size() ? body.size() - request->second.read_offset : 0;
@@ -1398,6 +1431,7 @@ LIB_DEFINE(InitNet_1_NpManager) {
 	LIB_FUNC("O80NrhUOPGY", NpManager::NpCheckPremium);
 	LIB_FUNC("eQH7nWPcAgc", NpManager::NpGetState);
 	LIB_FUNC("e-ZuhGEoeC4", NpManager::NpGetNpReachabilityState);
+	LIB_FUNC("Oad3rvY-NJQ", NpManager::NpHasSignedUp);
 }

 } // namespace LibNpManager
@@ -1738,7 +1738,8 @@ int KYTY_SYSV_ABI Accept(int s, void* addr, uint32_t* addrlen) {
 #if defined(_WIN32)
 	sockaddr_storage host_addr {};
 	int              host_addrlen = sizeof(host_addr);
-	NativeSocket accepted = ::accept(socket, reinterpret_cast<sockaddr*>(&host_addr), &host_addrlen);
+	NativeSocket     accepted =
+	    ::accept(socket, reinterpret_cast<sockaddr*>(&host_addr), &host_addrlen);
 	if (accepted == INVALID_NATIVE_SOCKET) {
 		return SetPosixSocketError();
 	}
@@ -3753,8 +3754,6 @@ int KYTY_SYSV_ABI NpGetState(int user_id, uint32_t* state) {
 int KYTY_SYSV_ABI NpGetNpReachabilityState(int user_id, uint32_t* state) {
 	PRINT_NAME();

-	constexpr int np_error_invalid_argument = -2141913085; /* 0x80550003 */
-
 	if (state == nullptr) {
 		return np_error_invalid_argument;
 	}
@@ -3767,6 +3766,20 @@ int KYTY_SYSV_ABI NpGetNpReachabilityState(int user_id, uint32_t* state) {
 	return OK;
 }

+int KYTY_SYSV_ABI NpHasSignedUp(int user_id, bool* has_signed_up) {
+	PRINT_NAME();
+
+	if (has_signed_up == nullptr) {
+		return np_error_invalid_argument;
+	}
+
+	LOGF("\t user_id = %d\n", user_id);
+
+	*has_signed_up = false;
+
+	return OK;
+}
+
 } // namespace NpManager

 } // namespace Libs::Network
@@ -184,6 +184,7 @@ int KYTY_SYSV_ABI  NpCheckPremium(int req_id, const NpCheckPremiumParameter* par
                                  NpCheckPremiumResult* result);
 int KYTY_SYSV_ABI  NpGetState(int user_id, uint32_t* state);
 int KYTY_SYSV_ABI  NpGetNpReachabilityState(int user_id, uint32_t* state);
+int KYTY_SYSV_ABI  NpHasSignedUp(int user_id, bool* has_signed_up);

 } // namespace NpManager

@@ -186,6 +186,39 @@ void TestCfgPhi() {
 	      "acyclic control-flow descriptor phi was not classified dynamic");
 }

+void TestNestedLoopPhiConvergence() {
+	Program program;
+	program.blocks.resize(4);
+	program.blocks[0].predecessors = {1};
+	program.blocks[0].successors   = {1};
+	program.blocks[1].predecessors = {0, 3};
+	program.blocks[1].successors   = {0, 2};
+	program.blocks[2].predecessors = {1};
+	program.blocks[2].successors   = {3};
+	program.blocks[3].predecessors = {2};
+	program.blocks[3].successors   = {1};
+	Instruction increment;
+	increment.op                     = Opcode::IAddU32;
+	increment.dst                    = Sgpr(0);
+	increment.src[0]                 = Sgpr(0);
+	increment.src[1]                 = Imm(1);
+	increment.src_count              = 2;
+	program.blocks[0].instructions   = {increment};
+	program.blocks[2].instructions   = {BufferUse(4, 0)};
+
+	std::string error;
+	Check(BuildScalarProvenance(program, &error), error.c_str());
+	const auto* source =
+	    GetDescriptorSource(program, program.blocks[2].instructions[0].memory.resource_source);
+	Check(source != nullptr, "nested-loop descriptor source was not attached");
+	const auto  value_id = source->dwords[0];
+	const auto& phi      = Value(program, value_id);
+	Check(phi.op == ScalarValueOp::Phi && phi.phi_args.size() == 2 &&
+	          ((phi.phi_args[0] == value_id && phi.phi_args[1] != value_id) ||
+	           (phi.phi_args[1] == value_id && phi.phi_args[0] != value_id)),
+	      "nested loop did not retain its recursive scalar provenance phi");
+}
+
 void TestDiamondReadPathsAreDynamic() {
 	std::array<uint32_t, 1> left  = {0x11111111u};
 	std::array<uint32_t, 1> right = {0x22222222u};
@@ -1045,6 +1078,7 @@ int main() {
 	try {
 		TestPerUseDescriptorDefinitions();
 		TestCfgPhi();
+		TestNestedLoopPhiConvergence();
 		TestDiamondReadPathsAreDynamic();
 		TestEquivalentConstantPhiIsStatic();
 		TestWideMoveInvalidatesAndCopiesBothDwords();
@@ -4772,6 +4772,52 @@ public:
              "successive near-capacity image transfers replaced the shared "
              "download buffer");

+      constexpr uint64_t tile_alias_offset = 0x2000000;
+      constexpr uint64_t tile_alias_size = 0x400000;
+      constexpr uint32_t tile_alias_extent = 1024;
+      std::memset(memory + tile_alias_offset, 0,
+                  static_cast<size_t>(tile_alias_size));
+      auto render_target_alias = MakeLinearDesc(
+          base + tile_alias_offset, tile_alias_size,
+          vk::Format::eR8G8B8A8Unorm,
+          Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8UNorm),
+          Prospero::ImageType::kColor2D,
+          {tile_alias_extent, tile_alias_extent, 1}, 1, 4, 1);
+      render_target_alias.type = BindingType::Storage;
+      render_target_alias.info.tile_mode =
+          Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
+      render_target_alias.view_info.usage =
+          vk::ImageUsageFlagBits::eStorage;
+      const auto render_target_alias_image =
+          texture_cache.FindImage(render_target_alias);
+
+      auto standard_4kb_alias = render_target_alias;
+      standard_4kb_alias.type = BindingType::Texture;
+      standard_4kb_alias.info.tile_mode =
+          Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB);
+      standard_4kb_alias.view_info.usage =
+          vk::ImageUsageFlagBits::eSampled;
+      const auto standard_4kb_alias_image =
+          texture_cache.FindImage(standard_4kb_alias);
+      auto repeated_standard_4kb_alias = standard_4kb_alias;
+      const auto repeated_standard_4kb_alias_image =
+          texture_cache.FindImage(repeated_standard_4kb_alias);
+      Require(
+          name, "equal-size tile-mode alias",
+          render_target_alias_image && standard_4kb_alias_image &&
+              standard_4kb_alias_image != render_target_alias_image &&
+              repeated_standard_4kb_alias_image ==
+                  standard_4kb_alias_image &&
+              texture_cache.GetImage(render_target_alias_image)
+                      .info.tile_mode ==
+                  Prospero::GpuEnumValue(
+                      Prospero::TileMode::kRenderTarget) &&
+              texture_cache.GetImage(standard_4kb_alias_image)
+                      .info.tile_mode ==
+                  Prospero::GpuEnumValue(
+                      Prospero::TileMode::kStandard4KB),
+          "equal address/size lookup reused an incompatible tiled backing");
+
      for (auto &output : ms_observer_outputs) {
        DestroyBuffer(&output);
      }
@@ -12083,6 +12129,84 @@ TestCase BufferLoadVariants() {
           O::BufferLoadDwordx4, O::VMovB32, O::BufferStoreDword, O::SEndpgm}};
 }

+TestCase BufferLoadDwordx4SnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  code.push_back(EncodeMubuf0(0x0eu, 0, true, true));
+  code.push_back(EncodeMubuf1(21, 0, 21));
+  for (u32 i = 0; i < 4; i++) {
+    AppendStoreVgpr(&code, 21 + i, 4 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "BufferLoadDwordx4SnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x11111111u, 0x22222222u, 0x33333333u, 0x44444444u,
+                  0,           0,           0,           0};
+  test.expected = {0x11111111u, 0x22222222u, 0x33333333u, 0x44444444u,
+                   0x11111111u, 0x22222222u, 0x33333333u, 0x44444444u};
+  test.opcodes = {O::VMovB32, O::BufferLoadDwordx4, O::BufferStoreDword,
+                  O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(16, 2);
+  test.has_user_data = true;
+  return test;
+}
+
+TestCase BufferLoadDwordx2SnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  code.push_back(EncodeMubuf0(0x0du, 0, true, true));
+  code.push_back(EncodeMubuf1(21, 0, 21));
+  for (u32 i = 0; i < 2; i++) {
+    AppendStoreVgpr(&code, 21 + i, 2 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "BufferLoadDwordx2SnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x11111111u, 0x22222222u, 0, 0};
+  test.expected = {0x11111111u, 0x22222222u, 0x11111111u, 0x22222222u};
+  test.opcodes = {O::VMovB32, O::BufferLoadDwordx2, O::BufferStoreDword,
+                  O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(8, 2);
+  test.has_user_data = true;
+  return test;
+}
+
+TestCase BufferLoadDwordx3SnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  code.push_back(EncodeMubuf0(0x0fu, 0, true, true));
+  code.push_back(EncodeMubuf1(21, 0, 21));
+  for (u32 i = 0; i < 3; i++) {
+    AppendStoreVgpr(&code, 21 + i, 3 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "BufferLoadDwordx3SnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x11111111u, 0x22222222u, 0x33333333u, 0, 0, 0};
+  test.expected = {0x11111111u, 0x22222222u, 0x33333333u,
+                   0x11111111u, 0x22222222u, 0x33333333u};
+  test.opcodes = {O::VMovB32, O::BufferLoadDwordx3, O::BufferStoreDword,
+                  O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(12, 2);
+  test.has_user_data = true;
+  return test;
+}
+
 TestCase BufferStoreVariants() {
  using O = ShaderOpcode;

@@ -12151,6 +12275,69 @@ TestCase BufferFormatVariants() {
  return load;
 }

+TestCase BufferLoadFormatXyzwSnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  code.push_back(EncodeMubuf0(0x03u, 0, true, true));
+  code.push_back(EncodeMubuf1(21, 0, 21));
+  for (u32 i = 0; i < 4; i++) {
+    AppendStoreVgpr(&code, 21 + i, 4 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "BufferLoadFormatXyzwSnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u,
+                  0,           0,           0,           0};
+  test.expected = {0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u,
+                   0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u};
+  test.opcodes = {O::VMovB32, O::BufferLoadFormatXyzw, O::BufferStoreDword,
+                  O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(
+      16, 2, false,
+      BufferFormat(Prospero::BufferFormat::k32_32_32_32Float));
+  test.has_user_data = true;
+  return test;
+}
+
+TestCase BufferLoadFormatXyzwInactiveExecPreservesOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovLiteral(&code, 21, 0x11111111u);
+  AppendVMovLiteral(&code, 22, 0x22222222u);
+  AppendVMovLiteral(&code, 23, 0x33333333u);
+  AppendVMovLiteral(&code, 24, 0x44444444u);
+  code.push_back(EncodeSop1(0x04, 126, InlineU32(0)));
+  code.push_back(EncodeMubuf0(0x03u, 0, true, true));
+  code.push_back(EncodeMubuf1(21, 0, 21));
+  code.push_back(EncodeSMovB32(126, InlineU32(1)));
+  code.push_back(EncodeSMovB32(127, InlineU32(0)));
+  for (u32 i = 0; i < 4; i++) {
+    AppendStoreVgpr(&code, 21 + i, 4 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "BufferLoadFormatXyzwInactiveExecPreservesOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0xaaaaaaaa, 0xbbbbbbbb, 0xcccccccc, 0xdddddddd,
+                  0,          0,          0,          0};
+  test.expected = {0xaaaaaaaau, 0xbbbbbbbbu, 0xccccccccu, 0xddddddddu,
+                   0x11111111u, 0x22222222u, 0x33333333u, 0x44444444u};
+  test.opcodes = {O::VMovB32, O::SMovB64, O::BufferLoadFormatXyzw,
+                  O::SMovB32, O::BufferStoreDword, O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(
+      16, 2, false,
+      BufferFormat(Prospero::BufferFormat::k32_32_32_32Float));
+  test.has_user_data = true;
+  return test;
+}
+
 TestCase BufferFormatStoreVariants() {
  using O = ShaderOpcode;

@@ -12459,6 +12646,63 @@ TestCase TBufferLoadVariants() {
           O::BufferStoreDword, O::SEndpgm}};
 }

+TestCase TBufferLoadFormatXyzwSnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  constexpr auto format =
+      BufferFormat(Prospero::BufferFormat::k32_32_32_32Float);
+  code.push_back(
+      EncodeMtbuf0(0x03u, format & 0xfu, (format >> 4u) & 0x7u, 0, true, true));
+  code.push_back(EncodeMtbuf1(0x03u, 21, 0, 21));
+  for (u32 i = 0; i < 4; i++) {
+    AppendStoreVgpr(&code, 21 + i, 4 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "TBufferLoadFormatXyzwSnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u,
+                  0,           0,           0,           0};
+  test.expected = {0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u,
+                   0x3f800000u, 0x40000000u, 0x40400000u, 0x40800000u};
+  test.opcodes = {O::VMovB32, O::TBufferLoadFormatXyzw,
+                  O::BufferStoreDword, O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(16, 2);
+  test.has_user_data = true;
+  return test;
+}
+
+TestCase TBufferLoadFormatXyzwPackedSnapshotsOverlappingAddress() {
+  using O = ShaderOpcode;
+
+  std::vector<u32> code;
+  AppendVMovU32(&code, 21, 0);
+  AppendVMovU32(&code, 22, 0);
+  constexpr auto format = BufferFormat(Prospero::BufferFormat::k8_8_8_8UInt);
+  code.push_back(
+      EncodeMtbuf0(0x03u, format & 0xfu, (format >> 4u) & 0x7u, 0, true, true));
+  code.push_back(EncodeMtbuf1(0x03u, 21, 0, 21));
+  for (u32 i = 0; i < 4; i++) {
+    AppendStoreVgpr(&code, 21 + i, 4 + i);
+  }
+  AppendEnd(&code);
+
+  TestCase test;
+  test.name = "TBufferLoadFormatXyzwPackedSnapshotsOverlappingAddress";
+  test.code = std::move(code);
+  test.initial = {0x44332211u, 0, 0, 0, 0, 0, 0, 0};
+  test.expected = {0x44332211u, 0, 0, 0, 0x11u, 0x22u, 0x33u, 0x44u};
+  test.opcodes = {O::VMovB32, O::TBufferLoadFormatXyzw,
+                  O::BufferStoreDword, O::SEndpgm};
+  test.user_data = MakeStructuredStorageBufferData(4, 8);
+  test.has_user_data = true;
+  return test;
+}
+
 TestCase TBufferStoreFormatX8UintWritesOneByte() {
  using O = ShaderOpcode;

@@ -13593,23 +13837,36 @@ TestCase DsAppendUsesEncodedGdsSelector() {
  using O = ShaderOpcode;

  std::vector<u32> code;
-  AppendSMovLiteral(&code, 124, 0x00000001u);
+  AppendSMovLiteral(&code, 124, 0x00000008u);
  code.push_back(EncodeDs0(0x3e, 0, true));
  code.push_back(EncodeDs1(0, 0, 0));
  code.push_back(EncodeDs0(0x3d, 0, true));
  code.push_back(EncodeDs1(1, 0, 0));
+  code.push_back(EncodeDs0(0x3e, 4, true));
+  code.push_back(EncodeDs1(2, 0, 0));
+  code.push_back(EncodeDs0(0x3d, 4, true));
+  code.push_back(EncodeDs1(3, 0, 0));
+  AppendSMovLiteral(&code, 124, 0x00080008u);
+  code.push_back(EncodeDs0(0x3e, 4, true));
+  code.push_back(EncodeDs1(4, 0, 0));
+  code.push_back(EncodeDs0(0x3d, 4, true));
+  code.push_back(EncodeDs1(5, 0, 0));
  AppendStoreVgpr(&code, 0, 0);
  AppendStoreVgpr(&code, 1, 1);
+  AppendStoreVgpr(&code, 2, 2);
+  AppendStoreVgpr(&code, 3, 3);
+  AppendStoreVgpr(&code, 4, 4);
+  AppendStoreVgpr(&code, 5, 5);
  AppendEnd(&code);

  TestCase test{
      "DsAppendGdsSelector",
      code,
      {},
-      {10, 74},
+      {10, 74, 20, 84, 40, 104},
      {O::SMovB32, O::DsAppend, O::DsConsume, O::BufferStoreDword, O::SEndpgm}};
-  test.gds_initial = {10};
-  test.expected_gds = {10};
+  test.gds_initial = {10, 20, 30, 40};
+  test.expected_gds = {10, 20, 30, 40};
  return test;
 }

@@ -14780,8 +15037,13 @@ std::vector<TestCase> MakeCases() {
  AddCase(BufferStoreDwordAppliesHostOffset);
  AddCase(BufferOffsetsUsePackedLaneAndStorageFallback);
  AddCase(BufferLoadVariants);
+  AddCase(BufferLoadDwordx2SnapshotsOverlappingAddress);
+  AddCase(BufferLoadDwordx3SnapshotsOverlappingAddress);
+  AddCase(BufferLoadDwordx4SnapshotsOverlappingAddress);
  AddCase(BufferStoreVariants);
  AddCase(BufferFormatVariants);
+  AddCase(BufferLoadFormatXyzwSnapshotsOverlappingAddress);
+  AddCase(BufferLoadFormatXyzwInactiveExecPreservesOverlappingAddress);
  AddCase(BufferFormatStoreVariants);
  AddCase(BufferStoreFormatXResource16UintWritesHalfword);
  AddCase(BufferLoadFormatXResource8UintZeroExtendsByte);
@@ -14795,6 +15057,8 @@ std::vector<TestCase> MakeCases() {
  AddCase(BufferStoreFormatXAddTidUsesLaneIndex);
  AddCase(BufferStoreFormatXDropsOutOfRangeRecord);
  AddCase(TBufferLoadVariants);
+  AddCase(TBufferLoadFormatXyzwSnapshotsOverlappingAddress);
+  AddCase(TBufferLoadFormatXyzwPackedSnapshotsOverlappingAddress);
  AddCase(TBufferLoadFormatX8UintZeroExtendsByte);
  AddCase(TBufferLoadFormatX8888UintExtractsFirstByte);
  AddCase(TBufferLoadFormatXIdxenUsesDescriptorStride);
@@ -15220,7 +15484,7 @@ void CheckRenderTargetFormatContract() {
      resource.kind = ShaderRecompiler::IR::ResourceKind::Image;
    } else if (std::strcmp(kind, "storage-no-write") == 0) {
      resource.written = false;
-    } else if (std::strcmp(kind, "storage-atomic") == 0) {
+    } else if (std::strcmp(kind, "storage-nonuint-atomic") == 0) {
      resource.atomic = true;
    } else if (std::strcmp(kind, "storage-compare") == 0) {
      resource.depth_compare = true;
@@ -15437,6 +15701,12 @@ void CheckSampledColorViews() {
  Require("SampledColorViews", "write-only uint 2D-array storage resource",
          IsSupportedStorageImageResource(storage_resource),
          "basic write-only uint 2D-array storage resource was rejected");
+  storage_resource.dimension = ShaderRecompiler::Decoder::ImageDimension::Dim2D;
+  storage_resource.read = true;
+  storage_resource.atomic = true;
+  Require("SampledColorViews", "atomic uint 2D storage resource",
+          IsSupportedStorageImageResource(storage_resource),
+          "atomic uint storage resource was rejected");

  char path[MAX_PATH]{};
  Require("SampledColorViews", "host",
@@ -15446,7 +15716,7 @@ void CheckSampledColorViews() {
       {"sampled-invalid-selector", "sampled-incompatible-format",
        "sampled-invalid-high", "sampled-depth-format", "sampled-depth-swizzle",
        "storage-incompatible-format", "storage-kind", "storage-no-write",
-        "storage-atomic", "storage-compare", "storage-mip", "storage-dimension",
+        "storage-nonuint-atomic", "storage-compare", "storage-mip", "storage-dimension",
        "volume-mip-count", "volume-slice-range"}) {
    std::string command =
        std::string("\"") + path + "\" --image-view-death " + kind;
@@ -16155,6 +16425,19 @@ ShaderTextureResource BasicUintVolumeStorageTextureDescriptor() {
           0x00700000u, 0x00000000u, 0x00000000u}};
 }

+ShaderRecompiler::IR::ImageResource AtomicStorageTextureResource() {
+  auto resource = BasicLinearStorageTextureResource();
+  resource.kind = ShaderRecompiler::IR::ResourceKind::StorageImageUint;
+  resource.read = true;
+  resource.atomic = true;
+  return resource;
+}
+
+ShaderTextureResource AtomicStorageTextureDescriptor() {
+  return {{0x304bb700u, 0xc1400000u, 0x0000001fu, 0x91b00204u, 0x00000000u,
+           0x00700000u, 0x00000000u, 0x00000000u}};
+}
+
 [[noreturn]] void RunStorageTextureDescriptorDeathCase(const char *kind) {
  auto resource = BasicStorageTextureResource();
  auto descriptor = BasicStorageTextureDescriptor();
@@ -16216,6 +16499,12 @@ ShaderTextureResource BasicUintVolumeStorageTextureDescriptor() {
  } else if (std::strcmp(kind, "uint-resource-float-format") == 0) {
    resource = BasicUintArrayStorageTextureResource();
    descriptor = BasicArrayStorageTextureDescriptor();
+  } else if (std::strcmp(kind, "atomic-format") == 0) {
+    resource = AtomicStorageTextureResource();
+    descriptor = AtomicStorageTextureDescriptor();
+    descriptor.fields[1] =
+        (descriptor.fields[1] & ~0x1ff00000u) |
+        (Prospero::GpuEnumValue(Prospero::BufferFormat::k8UInt) << 20u);
  } else if (std::strcmp(kind, "depth-tile-read") == 0) {
    resource = Ppsa14053DepthTileStorageTextureResource();
    descriptor = Ppsa14053DepthTileStorageTextureDescriptor();
@@ -16299,7 +16588,7 @@ void CheckBasicStorageTextureDescriptor() {
      "PPSA06228 R11G11B10 storage descriptor fixture is malformed");
  ValidateStorageTexture(BasicBgraStorageTextureResource(), r11g11b10,
                         0x870000);
-  ValidateStorageColorView(vk::Format::eB10G11R11UfloatPack32,
+  ValidateStorageColorView(vk::Format::eB8G8R8A8Unorm,
                           vk::Format::eB10G11R11UfloatPack32,
                           r11g11b10.DstSelXYZW());

@@ -16590,6 +16879,18 @@ void CheckBasicStorageTextureDescriptor() {
          IsValidImageSwizzle(DstSel(4, 4, 4, 4)),
          "single-channel replicated destination selection was rejected");

+  const auto atomic = AtomicStorageTextureDescriptor();
+  Require("BasicStorageTexture", "atomic R32_UINT descriptor",
+          atomic.Width5() + 1u == 128 && atomic.Height5() + 1u == 1 &&
+              atomic.Depth() + 1u == 1 &&
+              atomic.Type() ==
+                  Prospero::GpuEnumValue(Prospero::ImageType::kColor2D) &&
+              atomic.Format() ==
+                  Prospero::GpuEnumValue(Prospero::BufferFormat::k32UInt) &&
+              atomic.DstSelXYZW() == DstSel(4, 0, 0, 1),
+          "PPSA22102 image-atomic descriptor fixture is malformed");
+  ValidateStorageTexture(AtomicStorageTextureResource(), atomic, 0x10000);
+
  char path[MAX_PATH]{};
  Require("BasicStorageTexture", "host",
          GetModuleFileNameA(nullptr, path, MAX_PATH) != 0,
@@ -16598,7 +16899,7 @@ void CheckBasicStorageTextureDescriptor() {
       {"resource", "type", "tile", "mip", "swizzle", "linear-rgb1-read",
        "bgra-read", "r16-float-read", "r8-unorm-read", "yzwx-read",
        "reserved-swizzle", "array-base-out-of-range", "array-mip-view",
-        "reserved", "uint-format", "uint-resource-float-format",
+        "reserved", "uint-format", "uint-resource-float-format", "atomic-format",
        "depth-tile-read", "depth-tile-extent", "depth-tile-fmask"}) {
    std::string command = std::string("\"") + path +
                          "\" --storage-texture-descriptor-death " + kind;
@@ -405,6 +405,10 @@ void TestDirectMapQueryOffsetAndPartialMunmap() {
 	      "TryReadBacking should reject a range crossing an unmapped span");
 	Check(test, rejected_read == transaction_sentinel,
 	      "failed backing reads must not modify a destination prefix");
+	Check(test,
+	      Libs::LibKernel::Memory::ClampRangeSize(base + SceKernelPageSize - 0xf30, 0x1560) ==
+	          0xf30,
+	      "ClampRangeSize did not stop at an unmapped span");

 	info = Query(test, base + SceKernelPageSize, SceKernelVqFindNext);
 	ExpectRange(test, info, base + SceKernelPageSize * 2, base + SceKernelPageSize * 4,
@@ -473,6 +477,11 @@ void TestMunmapAcrossAdjacentFlexibleMappings() {
 	            &right, SceKernelPageSize, SceKernelProtCpuRw, SceKernelMapFixed, "adjacent_right"),
 	        "KernelMapNamedFlexibleMemory(right)");

+	Check(test,
+	      Libs::LibKernel::Memory::ClampRangeSize(base + SceKernelPageSize - 0x100, 0x200) ==
+	          0x200,
+	      "ClampRangeSize did not cross adjacent committed mappings");
+
 	CheckOk(test, Libs::LibKernel::Memory::KernelMunmap(base, SceKernelPageSize * 2),
 	        "KernelMunmap(adjacent mappings)");
 	Check(test, AvailableFlexibleMemory(test) == baseline,
Author	SHA1	Message	Date
nmzik	77aa28b27c	update README	2026-07-30 05:16:37 +02:00
nmzikandGitHub	d04938c88c	Embedded fetch shader: Fix overlapping buffer loads (#133 ) Fix overlapping buffer loads. Fixes many games	2026-07-30 05:08:38 +02:00
nmzikandGitHub	85622befb8	Fix fabricated HTTP2 success (#129 ) @StefanosCosta Thanks!	2026-07-30 00:48:08 +02:00
nmzik	3965d41d36	texture_cache: fix exact-match reuse across different tile modes	2026-07-30 00:10:40 +02:00
nmzik	c508c4a9c0	shader_recompiler: allow GDS append/consume offsets	2026-07-30 00:10:40 +02:00
ClaxtenandGitHub	e91dd39cb0	Drop redundant PROT_NONE tracking in reserve paths for Linux (#122 ) src: platform: Linux: Drop redundant PROT_NONE tracking in reserve paths * Some UE4 games, such as The Pathless, reserve a 512 GiB virtual address range during libc startup. Tracking every 4 KiB page causes a long delay and is unnecessary since the range is already PROT_NONE, and untracked pages are treated as NoAccess. Signed-off-by: Claxten <claxten10@gmail.com>	2026-07-30 00:00:21 +02:00
nmzik	cc76827e63	Fix vertex buffer ranges crossing memory mappings	2026-07-29 21:05:24 +02:00
nmzik	832bc84100	fix(shader): stabilize scalar provenance phis in cyclic CFGs	2026-07-29 21:05:24 +02:00
nmzik	65a0f0baa7	NpManager ABI	2026-07-29 21:05:24 +02:00
nmzik	b9ae2537ef	renderer: broaden compatibility	2026-07-29 18:47:26 +02:00
nmzik	0b9edaa721	graphics: broaden storage image atomic compatibility	2026-07-29 18:47:24 +02:00
nmzik	8a244677d7	fix(renderer): resolve delayed GPU page faults through buffer and texture caches	2026-07-29 18:47:19 +02:00