ASTC adjustments

src/video_core/host_shaders/astc_decoder.comp

@@ -77,14 +77,8 @@ uvec4 local_buff;
 uvec4 color_endpoint_data;
 int color_bitsread = 0;
 
-// Global "vector" to be pushed into when decoding
+#define MAX_WEIGHT_VALUES 64
-// At most will require BLOCK_WIDTH x BLOCK_HEIGHT in single plane mode
+uint result_vector[MAX_WEIGHT_VALUES];
-// At most will require BLOCK_WIDTH x BLOCK_HEIGHT x 2 in dual plane mode
-// So the maximum would be 144 (12 x 12) elements, x 2 for two planes
-#define DIVCEIL(number, divisor) (number + divisor - 1) / divisor
-#define ARRAY_NUM_ELEMENTS 144
-#define VECTOR_ARRAY_SIZE DIVCEIL(ARRAY_NUM_ELEMENTS * 2, 4)
-uint result_vector[ARRAY_NUM_ELEMENTS * 2];
 
 int result_index = 0;
 uint result_vector_max_index;
@@ -492,7 +486,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
         A = ReplicateBitTo9((bitval & 1));
         switch (encoding) {
         case JUST_BITS:
-            color_values[++out_index] = FastReplicateTo8(bitval, bitlen);
+            color_values[out_index++] = FastReplicateTo8(bitval, bitlen);
             break;
         case TRIT: {
             D = QuintTritValue(val);
@@ -571,7 +565,7 @@ void DecodeColorValues(uvec4 modes, uint num_partitions, uint color_data_bits, o
             uint T = (D * C) + B;
             T ^= A;
             T = (A & 0x80) | (T >> 2);
-            color_values[++out_index] = T;
+            color_values[out_index++] = T;
         }
     }
 }
@@ -753,12 +747,12 @@ void ComputeEndpoints(out uvec4 ep1, out uvec4 ep2, uint color_endpoint_mode, ui
 #define READ_UINT_VALUES(N)                                                                        \
     uvec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = color_values[++colvals_index];                      \
+        V[i / 4][i % 4] = color_values[colvals_index++];                      \
     }
 #define READ_INT_VALUES(N)                                                                         \
     ivec4 V[2];                                                                                    \
     for (uint i = 0; i < N; i++) {                                                                 \
-        V[i / 4][i % 4] = int(color_values[++colvals_index]);                      \
+        V[i / 4][i % 4] = int(color_values[colvals_index++]);                      \
     }
 
     switch (color_endpoint_mode) {
@@ -1225,6 +1219,10 @@ void DecompressBlock(ivec3 coord) {
         FillError(coord);
         return;
     }
+    if (GetNumWeightValues(size_params, dual_plane) > MAX_WEIGHT_VALUES) {
+        FillError(coord);
+        return;
+    }
     uint partition_index = 1;
     uvec4 color_endpoint_mode = uvec4(0);
     uint ced_pointer = 0;

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -270,6 +270,10 @@ constexpr VkBorderColor ConvertBorderColor(const std::array<float, 4>& color) {
     }
 }
 
+[[nodiscard]] bool IsLdrAstcFormat(VkFormat format) {
+    return format >= VK_FORMAT_ASTC_4x4_UNORM_BLOCK && format <= VK_FORMAT_ASTC_12x12_SRGB_BLOCK;
+}
+
 [[nodiscard]] VkImageAspectFlags ImageViewAspectMask(const VideoCommon::ImageViewInfo& info) {
     if (info.IsRenderTarget()) {
         return ImageAspectMask(info.format);
@@ -2483,9 +2487,18 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
         .pNext = nullptr,
         .usage = clamped_view_usage,
     };
+    const VkImageViewASTCDecodeModeEXT astc_decode_mode{
+        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_ASTC_DECODE_MODE_EXT,
+        .pNext = &image_view_usage,
+        .decodeMode = VK_FORMAT_R8G8B8A8_UNORM,
+    };
+    const void* view_next = &image_view_usage;
+    if (device->IsExtAstcDecodeModeSupported() && IsLdrAstcFormat(format_info.format)) {
+        view_next = &astc_decode_mode;
+    }
     const VkImageViewCreateInfo create_info{
         .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
-        .pNext = &image_view_usage,
+        .pNext = view_next,
         .flags = 0,
         .image = image.Handle(),
         .viewType = VkImageViewType{},

src/video_core/vulkan_common/vulkan_device.cpp

@@ -953,8 +953,7 @@ bool Device::ComputeIsOptimalAstcSupported() const {
         VK_FORMAT_ASTC_12x10_UNORM_BLOCK, VK_FORMAT_ASTC_12x10_SRGB_BLOCK,
         VK_FORMAT_ASTC_12x12_UNORM_BLOCK, VK_FORMAT_ASTC_12x12_SRGB_BLOCK,
     };
-    if (!features.features.textureCompressionASTC_LDR ||
+    if (!features.features.textureCompressionASTC_LDR) {
-        !features.texture_compression_astc_hdr.textureCompressionASTC_HDR) {
         return false;
     }
     const auto format_feature_usage{VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT |
@@ -1505,6 +1504,12 @@ void Device::RemoveUnsuitableExtensions() {
                                        features.subgroup_size_control,
                                        VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME);
 
+    // VK_EXT_texture_compression_astc_hdr
+    extensions.texture_compression_astc_hdr = false;
+    RemoveExtensionFeatureIfUnsuitable(extensions.texture_compression_astc_hdr,
+                                       features.texture_compression_astc_hdr,
+                                       VK_EXT_TEXTURE_COMPRESSION_ASTC_HDR_EXTENSION_NAME);
+
     // VK_EXT_transform_feedback
     extensions.transform_feedback =
         features.transform_feedback.transformFeedback &&

src/video_core/vulkan_common/vulkan_device.h

@@ -85,6 +85,7 @@ VK_DEFINE_HANDLE(VmaAllocator)
 
 // Define miscellaneous extensions which may be used by the implementation here.
 #define FOR_EACH_VK_EXTENSION(EXTENSION)                                                           \
+    EXTENSION(EXT, ASTC_DECODE_MODE, astc_decode_mode)                                             \
     EXTENSION(EXT, CONDITIONAL_RENDERING, conditional_rendering)                                   \
     EXTENSION(EXT, CONSERVATIVE_RASTERIZATION, conservative_rasterization)                         \
     EXTENSION(EXT, DEPTH_RANGE_UNRESTRICTED, depth_range_unrestricted)                             \
@@ -911,10 +912,17 @@ FN_MAX_LIMIT_LIST
                features.shader_atomic_int64.shaderSharedInt64Atomics;
     }
 
+    /// Returns true if the device supports VK_EXT_conditional_rendering.
     bool IsExtConditionalRendering() const {
         return extensions.conditional_rendering;
     }
 
+    /// Returns true if the device supports VK_EXT_astc_decode_mode.
+    bool IsExtAstcDecodeModeSupported() const {
+        return extensions.astc_decode_mode;
+    }
+
+    /// Returns true if the device supports VK_EXT_external_memory_host.
     bool IsExtExternalMemoryHostSupported() const {
         return extensions.external_memory_host;
     }