[Perf] (DirectVulkan): keep the render pass alive for steady-state storage-image draws and skip storage-image collection for programs without them

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.cpp

@@ -1808,6 +1808,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 layoutBinding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER;
             } else if (kind == DescriptorBindingKind::StorageImage) {
                 layoutBinding.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE;
+                entry.hasStorageImages = true;
             } else {
                 layoutBinding.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER;
             }

MobileGL/MG_Backend/DirectVulkan/Renderer/ProgramFactory.h

@@ -67,6 +67,9 @@ namespace MobileGL::MG_Backend::DirectVulkan {
             Vector<Bool> storageImageUsesBindingFormatByBinding;
             Vector<String> storageBlockNameByBinding;
             Vector<Int> storageBlockIndexByBinding;
+            // Set once during ReflectLayout so the per-draw path can skip the whole
+            // storage-image preparation for the overwhelming majority of programs.
+            Bool hasStorageImages = false;
             Int globalUboBinding = -1;
             Uint32 activeVertexInputLocationMask = 0;
             Array<GLenum, kMaxVertexInputLocations> vertexInputTypes{};
@@ -99,6 +102,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     std::move(other.storageImageUsesBindingFormatByBinding);
                 storageBlockNameByBinding = std::move(other.storageBlockNameByBinding);
                 storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
+                hasStorageImages = other.hasStorageImages;
                 globalUboBinding = other.globalUboBinding;
                 activeVertexInputLocationMask = other.activeVertexInputLocationMask;
                 vertexInputTypes = other.vertexInputTypes;
@@ -110,6 +114,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
                 other.pipelineLayout = VK_NULL_HANDLE;
+                other.hasStorageImages = false;
                 other.globalUboBinding = -1;
                 other.activeVertexInputLocationMask = 0;
                 other.activeFragmentOutputLocationMask = 0;
@@ -139,6 +144,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                     std::move(other.storageImageUsesBindingFormatByBinding);
                 storageBlockNameByBinding = std::move(other.storageBlockNameByBinding);
                 storageBlockIndexByBinding = std::move(other.storageBlockIndexByBinding);
+                hasStorageImages = other.hasStorageImages;
                 globalUboBinding = other.globalUboBinding;
                 activeVertexInputLocationMask = other.activeVertexInputLocationMask;
                 vertexInputTypes = other.vertexInputTypes;
@@ -150,6 +156,7 @@ namespace MobileGL::MG_Backend::DirectVulkan {
                 other.hash = 0;
                 other.descriptorSetLayout = VK_NULL_HANDLE;
                 other.pipelineLayout = VK_NULL_HANDLE;
+                other.hasStorageImages = false;
                 other.globalUboBinding = -1;
                 other.activeVertexInputLocationMask = 0;
                 other.activeFragmentOutputLocationMask = 0;

MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.cpp

@@ -1140,6 +1140,24 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         return ok;
     }
 
+    Bool VkTextureManager::NeedsStorageImagePreparation(MG_State::GLState::ITextureObject& texture) const {
+        const auto it = m_textureResources.find(MakeTextureIdentity(&texture));
+        if (it == m_textureResources.end()) {
+            return true;
+        }
+        const TextureResource& resource = it->second;
+        if (resource.image == VK_NULL_HANDLE || resource.layout != VK_IMAGE_LAYOUT_GENERAL) {
+            return true;
+        }
+        // Mirror SyncTexture's cross-draw skip condition: any version drift means the sync
+        // path may upload or rebuild, both of which need the render pass ended first.
+        const auto* mipTexture = MG_State::GLState::AsMipmapTexture(&texture);
+        const Uint32 mipLevelCount = mipTexture != nullptr ? mipTexture->GetMipmapLevelCount() : 0u;
+        return resource.syncedContentVersion != texture.GetContentVersion() ||
+               resource.syncedTextureParamsVersion != texture.GetTextureParamsVersion() ||
+               resource.syncedMipLevelCount != mipLevelCount;
+    }
+
     Bool VkTextureManager::TransitionImageLayout(VkCommandBuffer commandBuffer, VkImage image,
                                                  VkImageLayout& trackedLayout, VkImageLayout newLayout,
                                                  VkPipelineStageFlags srcStageMask, VkPipelineStageFlags dstStageMask,

MobileGL/MG_Backend/DirectVulkan/Renderer/VkTextureManager.h

@@ -284,6 +284,11 @@ public:
                                                       VkImageLayout newLayout);
     Bool TransitionTextureForSampling(VkCommandBuffer commandBuffer, MG_State::GLState::ITextureObject& texture);
     Bool TransitionTextureForStorageImage(VkCommandBuffer commandBuffer, MG_State::GLState::ITextureObject& texture);
+    // Non-mutating probe for the per-draw storage-image fast path: true when preparing this
+    // texture as a storage image may need work that is illegal inside a render pass (resource
+    // creation, dirty-content upload, or a layout transition to GENERAL). Unknown state reports
+    // true - a false positive merely ends the render pass, a false negative would skip a barrier.
+    Bool NeedsStorageImagePreparation(MG_State::GLState::ITextureObject& texture) const;
 
     static VkImageAspectFlags ResolveSampledImageViewAspectMask(VkImageAspectFlags imageAspect);
     static VkFormat ResolveSampledImageViewFormat(VkFormat imageFormat, SamplerNumericDomain numericDomain);

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -3552,6 +3552,9 @@ void main() {
         VkCommandBuffer commandBuffer,
         const MG_State::GLState::ProgramObject& program,
         const ProgramFactory::VkProgramObject& programObj) {
+        if (!programObj.hasStorageImages) {
+            return true;
+        }
         auto& storageTextures = m_storageImageTexturesScratch;
         if (!m_uniformManager->CollectStorageImageTextures(program, programObj, storageTextures)) {
             MGLOG_E("%s: failed to collect storage images for program=%u",
@@ -3562,6 +3565,24 @@ void main() {
             return true;
         }
 
+        // Steady-state fast path: when every collected texture is already resident in GENERAL
+        // with no pending clear and no dirty content, the loop below has nothing to record, so
+        // keep the render pass alive instead of splitting it on every storage-image draw (on
+        // tiled GPUs each split is a full tile load/store). GL makes cross-draw image-store
+        // coherence the app's job (glMemoryBarrier), so no implicit barrier is owed here.
+        Bool anyNeedsPreparation = false;
+        for (auto* texture : storageTextures) {
+            MOBILEGL_ASSERT(texture != nullptr, "%s: collected a null storage texture", __func__);
+            if (m_textureManager->NeedsStorageImagePreparation(*texture) ||
+                m_clearManager->HasPendingClear(texture)) {
+                anyNeedsPreparation = true;
+                break;
+            }
+        }
+        if (!anyNeedsPreparation) {
+            return true;
+        }
+
         // Image uploads, deferred-clear materialization, and layout barriers are illegal inside
         // a classic render pass. Do this before sampler preparation as well: a texture used by
         // both a sampler and an image must stay in GENERAL, and both descriptors must name that
@@ -3571,7 +3592,6 @@ void main() {
         }
 
         for (auto* texture : storageTextures) {
-            MOBILEGL_ASSERT(texture != nullptr, "%s: collected a null storage texture", __func__);
             if (!MaterializePendingClearForTexture(commandBuffer, *texture)) {
                 MGLOG_E("%s: failed to materialize pending clear for storage textureId=%d",
                         __func__, texture->GetExternalIndex());