ASTC: decode and re-compress to BC1/BC3 on the GPU; texture cache budget

- accelerate_astc=Gpu with astc_recompression=Bc1/Bc3 now decodes ASTC
  with the GPU decoder into a temporary RGBA8 image (dedicated memory,
  freed once the GPU is past it) and compresses it with a new compute
  shader (bc_encoder.comp: bounding-box endpoints along the colours'
  covariance, inset 1/16, nearest palette entry; BC3 alpha ramp; BC1
  punch-through), then copies the blocks into the image. No CPU decode,
  so streamed textures are ready the frame they are first used instead
  of showing placeholders for seconds; the decode pass no longer waits
  for the GPU on this path.
- The CPU path (CpuAsynchronous) fills a new image with zero blocks
  before its first decode instead of showing stale VRAM, and the
  transcode pool is a quarter of the hardware threads.
- Switch 2 titles: collector thresholds near the device budget, re-read
  every 60 frames from VK_EXT_memory_budget (another game shrank it from
  7.0 to 5.4 GB).
- EDEN_NX2_HIDE_WINDOW=0 now leaves the window visible.

4 files changed, 24 insertions, 3 deletions

diff --git a/src/video_core/vulkan_common/vulkan_device.cpp b/src/video_core/vulkan_common/vulkan_device.cpp
index be09bca8d3..454e5d4e02 100644
--- a/src/video_core/vulkan_common/vulkan_device.cpp
+++ b/src/video_core/vulkan_common/vulkan_device.cpp
@@ -1612,6 +1612,20 @@ u64 Device::GetDeviceMemoryUsage() const {
     return result;
 }
 
+u64 Device::GetDeviceMemoryBudget() const {
+    if (!extensions.memory_budget) {
+        return 0;
+    }
+    VkPhysicalDeviceMemoryBudgetPropertiesEXT budget{};
+    budget.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT;
+    physical.GetMemoryProperties(&budget);
+    u64 result{};
+    for (const size_t heap : valid_heap_memory) {
+        result += budget.heapBudget[heap];
+    }
+    return result;
+}
+
 void Device::CollectPhysicalMemoryInfo() {
     // Calculate limits using memory budget
     VkPhysicalDeviceMemoryBudgetPropertiesEXT budget{};
diff --git a/src/video_core/vulkan_common/vulkan_device.h b/src/video_core/vulkan_common/vulkan_device.h
index bbf3754e39..34ee63a275 100644
--- a/src/video_core/vulkan_common/vulkan_device.h
+++ b/src/video_core/vulkan_common/vulkan_device.h
@@ -963,6 +963,9 @@ FN_MAX_LIMIT_LIST
 
     u64 GetDeviceMemoryUsage() const;
 
+    /// The driver's current memory budget for this process (VK_EXT_memory_budget), 0 without it.
+    u64 GetDeviceMemoryBudget() const;
+
     u32 GetSetsPerPool() const {
         return sets_per_pool;
     }
diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
index 23f1b0a798..27f8fa6b4c 100644
--- a/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
+++ b/src/video_core/vulkan_common/vulkan_memory_allocator.cpp
@@ -206,10 +206,12 @@ MemoryAllocator::MemoryAllocator(const Device &device_)
 
 MemoryAllocator::~MemoryAllocator() = default;
 
-vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci) const
+vk::Image MemoryAllocator::CreateImage(const VkImageCreateInfo &ci, bool dedicated) const
 {
     const VmaAllocationCreateInfo alloc_ci = {
-            .flags = VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT,
+            .flags = static_cast<VmaAllocationCreateFlags>(VMA_ALLOCATION_CREATE_WITHIN_BUDGET_BIT) |
+                     (dedicated ? VmaAllocationCreateFlags{VMA_ALLOCATION_CREATE_DEDICATED_MEMORY_BIT}
+                               : VmaAllocationCreateFlags{0}),
             .usage = VMA_MEMORY_USAGE_AUTO_PREFER_DEVICE,
             .requiredFlags = 0,
             .preferredFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
diff --git a/src/video_core/vulkan_common/vulkan_memory_allocator.h b/src/video_core/vulkan_common/vulkan_memory_allocator.h
index 7467e929d3..fa7d4b41c3 100644
--- a/src/video_core/vulkan_common/vulkan_memory_allocator.h
+++ b/src/video_core/vulkan_common/vulkan_memory_allocator.h
@@ -103,7 +103,9 @@ namespace Vulkan {
 
         MemoryAllocator(const MemoryAllocator &) = delete;
 
-        vk::Image CreateImage(const VkImageCreateInfo &ci) const;
+        /// `dedicated`: own device memory, returned to the driver on destruction (short-lived
+        /// images would otherwise leave holes in the shared blocks).
+        vk::Image CreateImage(const VkImageCreateInfo &ci, bool dedicated = false) const;
 
         vk::Buffer CreateBuffer(const VkBufferCreateInfo &ci, MemoryUsage usage) const;