cache directwrite

implementations/Crafter.Graphics-Device.cpp

@@ -780,6 +780,7 @@ void Device::Initialize() {
     CheckVkResult(vkCreateCommandPool(device, &commandPoolcreateInfo, NULL, &commandPool));
 
 	vkGetPhysicalDeviceMemoryProperties(physDevice, &memoryProperties);
+	CacheUploadStrategy();
 
     vkGetAccelerationStructureBuildSizesKHR = reinterpret_cast<PFN_vkGetAccelerationStructureBuildSizesKHR>(vkGetInstanceProcAddr(instance, "vkGetAccelerationStructureBuildSizesKHR"));
     vkCreateAccelerationStructureKHR = reinterpret_cast<PFN_vkCreateAccelerationStructureKHR>(vkGetInstanceProcAddr(instance, "vkCreateAccelerationStructureKHR"));
@@ -842,7 +843,7 @@ std::uint32_t Device::GetMemoryType(uint32_t typeBits, VkMemoryPropertyFlags req
 	throw std::runtime_error("Could not find a matching memory type");
 }
 
-bool Device::PreferDirectDeviceWrite(VkDeviceSize size) {
+void Device::CacheUploadStrategy() {
 	constexpr VkMemoryPropertyFlags directFlags =
 		VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT | VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT;
 
@@ -858,7 +859,8 @@ bool Device::PreferDirectDeviceWrite(VkDeviceSize size) {
 		}
 	}
 	if (!haveDirectType) {
-		return false;
+		directWriteBudget = 0;
+		return;
 	}
 
 	// 2. Compare the host-visible device-local heap against the largest plain
@@ -871,10 +873,11 @@ bool Device::PreferDirectDeviceWrite(VkDeviceSize size) {
 		}
 	}
 
-	// 3. ReBAR / UMA: the BAR heap covers (almost) all of VRAM -> direct.
+	// 3. ReBAR / UMA: the BAR heap covers (almost) all of VRAM -> map any size.
 	//    barHeap >= 0.9 * vram, written to dodge floating point / overflow.
 	if (barHeap >= vram - vram / 10) {
-		return true;
+		directWriteBudget = std::numeric_limits<VkDeviceSize>::max();
+		return;
 	}
 
 	// 4. Small BAR window: direct only for buffers small enough that several
@@ -884,5 +887,11 @@ bool Device::PreferDirectDeviceWrite(VkDeviceSize size) {
 	//    both a 256 MiB and a 4 GiB window). VK_EXT_memory_budget would refine
 	//    this to the window's *remaining* space; see the header note.
 	constexpr VkDeviceSize barWindowFraction = 8;
-	return size <= barHeap / barWindowFraction;
+	directWriteBudget = barHeap / barWindowFraction;
+}
+
+bool Device::PreferDirectDeviceWrite(VkDeviceSize size) {
+	// directWriteBudget caches the size-independent decision (CacheUploadStrategy):
+	// 0 -> never direct, max -> ReBAR/UMA so any size, else the small-window cap.
+	return directWriteBudget != 0 && size <= directWriteBudget;
 }

interfaces/Crafter.Graphics-Device.cppm

@@ -232,8 +232,22 @@ export namespace Crafter {
         // small-window budget would consult VK_EXT_memory_budget for the
         // window's *remaining* space rather than its total size; that needs the
         // extension enabled and is a follow-up.
+        //
+        // Everything except the final size comparison depends only on
+        // memoryProperties, so it is derived once by CacheUploadStrategy (called
+        // at device creation) into the directWrite* members below; this leaves
+        // the per-allocation call a single bounds check.
         static bool PreferDirectDeviceWrite(VkDeviceSize size);
 
+        // Size-independent half of PreferDirectDeviceWrite, cached from
+        // memoryProperties. directWriteBudget is the largest buffer that may be
+        // written directly: 0 when no resizable BAR exists (always stage),
+        // VkDeviceSize max for ReBAR/UMA (always map), or the small-window
+        // per-buffer cap otherwise. Recompute via CacheUploadStrategy whenever
+        // memoryProperties changes (device creation does this once).
+        inline static VkDeviceSize directWriteBudget = 0;
+        static void CacheUploadStrategy();
+
         // ─── Wayland key repeat ────────────────────────────────────────
         // TickKeyRepeats fires onRawKeyDown / onRawKeyHold / onTextInput on
         // the focused window for whichever key is currently repeating.

tests/UploadStrategy/main.cpp

@@ -80,6 +80,10 @@ void SetMemory(std::initializer_list<TypeSpec> types,
         Device::memoryProperties.memoryHeaps[h].flags = s.flags;
         ++h;
     }
+    // PreferDirectDeviceWrite reads the cache CacheUploadStrategy derives from
+    // memoryProperties (device creation does this once); refresh it here since
+    // the test mutates the layout directly.
+    Device::CacheUploadStrategy();
 }
 
 constexpr VkMemoryHeapFlags HEAP_DEVICE_LOCAL = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT;