vulkan: enable VK_EXT_extended_dynamic_state when available (#1906)

DecodePackedColorClear had no case for R8_UNORM/UINT, R8G8_UNORM/UINT and
R16_UINT, so a CMASK fast clear of such a render target was dropped and the
target kept its old contents. UFC 6 (PPSA23566) fast-clears an R8_UINT
region map this way every frame and then writes only part of it; a capture
showed the map never cleared. DWORD fills of these formats are decoded only
when every pixel in the word holds the same value.

Co-authored-by: 4wgycvn84p-afk <346802885+4wgycvn84p-afk@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

src/graphics/host_gpu/renderer/image/imageInfo.h

@@ -416,6 +416,20 @@ inline constexpr std::array<VideoOutFormatPolicy, 8> VIDEO_OUT_FORMAT_POLICIES {
 		case vk::Format::eR32Sfloat: next.float32[0] = std::bit_cast<float>(packed); break;
 		case vk::Format::eR32Uint: next.uint32[0] = packed; break;
 		case vk::Format::eR32Sint: next.int32[0] = static_cast<int32_t>(packed); break;
+		// Narrow targets hold their channels in the low bits of the first word. UFC 6 (PPSA23566)
+		// fast-clears an R8_UINT map through CMASK every frame and then writes only part of it;
+		// without these the clear was dropped and old contents piled up.
+		case vk::Format::eR8Uint: next.uint32[0] = packed & 0xffu; break;
+		case vk::Format::eR8Unorm: next.float32[0] = unorm8(packed); break;
+		case vk::Format::eR8G8Uint:
+			next.uint32[0] = packed & 0xffu;
+			next.uint32[1] = (packed >> 8u) & 0xffu;
+			break;
+		case vk::Format::eR8G8Unorm:
+			next.float32[0] = unorm8(packed);
+			next.float32[1] = unorm8(packed >> 8u);
+			break;
+		case vk::Format::eR16Uint: next.uint32[0] = packed & 0xffffu; break;
 		case vk::Format::eR8G8B8A8Srgb:
 			next.float32[0] = srgb8(packed);
 			next.float32[1] = srgb8(packed >> 8u);
@@ -462,10 +476,16 @@ inline constexpr std::array<VideoOutFormatPolicy, 8> VIDEO_OUT_FORMAT_POLICIES {
 // Keep this separate: the first register word alone cannot describe a 64-bit color.
 [[nodiscard]] inline bool DecodeColorDwordFill(vk::Format format, uint32_t packed,
                                                vk::ClearColorValue& clear) {
-	if ((format == vk::Format::eR16Sfloat || format == vk::Format::eR16Unorm) &&
+	if ((format == vk::Format::eR16Sfloat || format == vk::Format::eR16Unorm ||
+	     format == vk::Format::eR16Uint || format == vk::Format::eR8G8Unorm ||
+	     format == vk::Format::eR8G8Uint) &&
 	    (packed & 0xffffu) != (packed >> 16u)) {
 		return false;
 	}
+	if ((format == vk::Format::eR8Unorm || format == vk::Format::eR8Uint) &&
+	    packed != (packed & 0xffu) * 0x01010101u) {
+		return false;
+	}
 	if (format == vk::Format::eR16G16B16A16Sfloat) {
 		if (packed != 0) {
 			return false;

src/graphics/presentation/window/vulkanWindow.cpp

@@ -733,6 +733,15 @@ static vk::Device VulkanCreateDevice(GraphicContext& graphics,
 		conditional_rendering.pNext = const_cast<void*>(create_info.pNext);
 		create_info.pNext = &conditional_rendering;
 	}
+	// The renderer only uses the Vulkan 1.3 core dynamic states, but RADV (Mesa 26.3) creates its
+	// PS epilog cache only when this EXT feature is enabled and crashes in vkCmdDraw when a
+	// depth- or stencil-writing fragment shader needs an epilog.
+	vk::PhysicalDeviceExtendedDynamicStateFeaturesEXT extended_dynamic_state {};
+	extended_dynamic_state.extendedDynamicState = VK_TRUE;
+	if (HasExtension(device_extensions, VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME)) {
+		extended_dynamic_state.pNext = const_cast<void*>(create_info.pNext);
+		create_info.pNext = &extended_dynamic_state;
+	}
 	create_info.pQueueCreateInfos       = &queue_create_info;
 	create_info.queueCreateInfoCount    = 1;
 	create_info.enabledExtensionCount   = static_cast<uint32_t>(device_extensions.size());
@@ -1111,7 +1120,8 @@ void WindowContext::CreateVulkan() {
 		                             VK_KHR_WORKGROUP_MEMORY_EXPLICIT_LAYOUT_EXTENSION_NAME,
 		                             VK_EXT_DEPTH_RANGE_UNRESTRICTED_EXTENSION_NAME,
 		                             VK_KHR_FRAGMENT_SHADER_BARYCENTRIC_EXTENSION_NAME,
-		                             VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME}) {
+		                             VK_EXT_CONDITIONAL_RENDERING_EXTENSION_NAME,
+		                             VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME}) {
 			if (HasExtension(available_extensions, extension)) {
 				device_extensions.push_back(extension);
 			}

tests/ShaderRecompilerComputeTests.cpp

@@ -12274,6 +12274,8 @@ public:
         PackedCase{vk::Format::eR16G16Unorm, 0x0000ffffu, {1.0f, 0.0f}},
         PackedCase{vk::Format::eR16G16Unorm, 0x80000001u,
                    {1.0f / 65535.0f, 32768.0f / 65535.0f}},
+        PackedCase{vk::Format::eR8Unorm, 0xffffff80u, {128.0f / 255.0f}},
+        PackedCase{vk::Format::eR8G8Unorm, 0xffff40ffu, {1.0f, 64.0f / 255.0f}},
     };
     for (const auto &test : packed_cases) {
       vk::ClearColorValue clear{};
@@ -12300,6 +12302,29 @@ public:
                   DecodeColorDwordFill(format, 0x3c003c00u, clear),
               "alternating R16 pixels were treated as a uniform clear");
     }
+    struct UintCase {
+      vk::Format format;
+      uint64_t word;
+      std::array<uint32_t, 4> channels;
+    };
+    for (const auto &test : {UintCase{vk::Format::eR8Uint, 0xffffff05u, {5u}},
+                             UintCase{vk::Format::eR8G8Uint, 0xffff0a05u, {5u, 10u}},
+                             UintCase{vk::Format::eR16Uint, 0xffff1234u, {0x1234u}}}) {
+      Require(name, "packed unsigned color clear",
+              DecodePackedColorClear(test.format, test.word, clear) &&
+                  std::bit_cast<std::array<uint32_t, 4>>(clear.uint32) == test.channels,
+              "narrow unsigned clear lost a channel or kept bits of the next one");
+    }
+    Require(name, "R8 DWORD fill",
+            !DecodeColorDwordFill(vk::Format::eR8Uint, 0x01020304u, clear) &&
+                DecodeColorDwordFill(vk::Format::eR8Uint, 0x07070707u, clear) &&
+                clear.uint32[0] == 7u,
+            "differing R8 pixels were treated as a uniform clear");
+    Require(name, "RG8 DWORD fill",
+            !DecodeColorDwordFill(vk::Format::eR8G8Unorm, 0x00ff00feu, clear) &&
+                DecodeColorDwordFill(vk::Format::eR8G8Unorm, 0x00ff00ffu, clear) &&
+                clear.float32[0] == 1.0f && clear.float32[1] == 0.0f,
+            "alternating RG8 pixels were treated as a uniform clear");
     Require(name, "RG32 DWORD fill",
             DecodeColorDwordFill(vk::Format::eR32G32Sfloat, 0x3f800000u, clear) &&
                 clear.float32[0] == 1.0f && clear.float32[1] == 1.0f,