[Perf] (MG_Backend/DirectVulkan): skip the per-draw sampled-texture walk when the bound set is unchanged (texture-bind generation + program state version); CollectSampledTextures 5.0%->0.2%, fps 228->249

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.cpp

@@ -3389,6 +3389,9 @@ void main() {
         // Begin command recording if not yet
         if (!frame.isCommandRecording) {
             m_frameContext.BeginCommandRecording();
+            // New command buffer: a program/FBO address from a previous frame may have been
+            // recycled, so start the sampled-set skip cache fresh this frame.
+            m_lastSampledSetValid = false;
         }
 
         auto* activeRenderPass = VkRenderPassManager::GetActiveRenderPass();
@@ -3397,9 +3400,31 @@ void main() {
         // which probably indicates it's been gone through codepath like `fbo attach` -> `clear` -> `fbo detach`, and
         // without draws in between to give it a chance to materialize such clear.
         // Deal with this situation here.
-        auto& sampledTextures = m_sampledTexturesScratch; // cleared by CollectSampledTextures
-        Bool hasSampledTextures = m_uniformManager->CollectSampledTextures(program, programObj, sampledTextures);
-        MOBILEGL_ASSERT(hasSampledTextures, "%s: CollectSampledTextures failed", __func__);
+        // Reuse the previous draw's sampled-texture list when the set is provably unchanged (same
+        // program+state+transform and no bind/unbind/delete since), skipping the per-draw GL walk.
+        // The layout/feedback/transition loops below still run on the list every draw, so this only
+        // elides re-resolving *which* textures are sampled, never their layout handling.
+        auto& sampledTextures = m_sampledTexturesScratch;
+        {
+            const Uint programIndex = program.GetExternalIndex();
+            const Uint32 programVersion = program.GetBackendStateVersion();
+            const Uint64 bindGeneration = MG_State::pGLContext->GetTextureBindGeneration();
+            const Bool sampledSetUnchanged =
+                m_lastSampledSetValid && m_lastSampledSetProgramIndex == programIndex &&
+                m_lastSampledSetProgramVersion == programVersion &&
+                m_lastSampledSetTransformFlags == transformFlags &&
+                m_lastSampledSetBindGeneration == bindGeneration;
+            if (!sampledSetUnchanged) {
+                const Bool hasSampledTextures =
+                    m_uniformManager->CollectSampledTextures(program, programObj, sampledTextures);
+                MOBILEGL_ASSERT(hasSampledTextures, "%s: CollectSampledTextures failed", __func__);
+                m_lastSampledSetValid = true;
+                m_lastSampledSetProgramIndex = programIndex;
+                m_lastSampledSetProgramVersion = programVersion;
+                m_lastSampledSetTransformFlags = transformFlags;
+                m_lastSampledSetBindGeneration = bindGeneration;
+            }
+        }
         MGLOG_D("SetupDraw: program=%u drawFbo=%u sampledTextureCount=%zu activeRenderPass=%s",
                 program.GetExternalIndex(), drawFbo ? drawFbo->GetExternalIndex() : 0u, sampledTextures.size(),
                 activeRenderPass ? "true" : "false");

MobileGL/MG_Backend/DirectVulkan/Renderer/VulkanRenderer.h

@@ -403,6 +403,20 @@ namespace MobileGL::MG_Backend::DirectVulkan {
         DepthMipmapResources m_depthMipmapResources;
         Vector<DeferredDepthMipmapCleanup> m_deferredDepthMipmapCleanup;
 
+        // Skip the per-draw CollectSampledTextures walk (~5% of the render thread) when the sampled
+        // texture SET is provably unchanged from the previous draw: same program (external index +
+        // backend-state version, which covers sampler-uniform reassignment / relink) and transform
+        // flags, and no texture bind/unbind/delete since (GetTextureBindGeneration). On a hit,
+        // m_sampledTexturesScratch still holds the previous draw's list and steps 2-4 (feedback /
+        // layout probe / transition) re-run on it, so layout correctness is unaffected - only the GL
+        // walk is skipped. Reset per command-buffer recording so a reused program/FBO address can't
+        // outlive a frame.
+        Bool m_lastSampledSetValid = false;
+        Uint m_lastSampledSetProgramIndex = 0;
+        Uint32 m_lastSampledSetProgramVersion = 0;
+        ProgramFactory::CompileOptionFlags m_lastSampledSetTransformFlags = {};
+        Uint64 m_lastSampledSetBindGeneration = 0;
+
         // Per-draw scratch buffers (clear keeps capacity) — these paths run for every
         // draw call and must not allocate.
         Vector<MG_State::GLState::ITextureObject*> m_sampledTexturesScratch;

MobileGL/MG_State/GLState/Core.h

@@ -102,6 +102,10 @@ namespace MobileGL {
                 const ImageTextureBinding& GetImageTextureBinding(Int unit) const;
                 void NoteTextureUnitTouched(Int unit) { m_textureState.NoteUnitTouched(unit); }
                 Int GetMaxTouchedTextureUnit() const { return m_textureState.GetMaxTouchedUnit(); }
+                // Monotonic counter bumped whenever a texture bind/unbind/delete changes which
+                // texture is bound at a unit; lets a backend skip re-resolving an unchanged
+                // per-draw sampled-texture set.
+                Uint64 GetTextureBindGeneration() const { return m_textureState.GetTextureBindGeneration(); }
                 Bool ValidateTextureName(Uint index) const;
                 Bool ValidateTextureObject(Uint index) const;
                 Int GetActiveTextureUnit() const;

MobileGL/MG_State/GLState/TextureState/TextureState.cpp

@@ -104,6 +104,10 @@ namespace MobileGL::MG_State::GLState {
                         imageBinding.Bind(nullptr, 0, GL_FALSE, 0, GL_READ_ONLY, GL_R8);
                     }
                 }
+                // Deleting a texture unbinds it from every unit above; treat that as a binding
+                // change so a cached sampled-texture set (which may hold this raw pointer) is
+                // re-resolved instead of dangling.
+                BumpTextureBindGeneration();
                 m_textureObjects.erase(index);
             }
             m_indexGenerator.Delete(index);

MobileGL/MG_State/GLState/TextureState/TextureState.h

@@ -63,10 +63,20 @@ namespace MobileGL::MG_State::GLState {
         // can stop there instead of walking all MAX_TEXTURE_IMAGE_UNITS units.
         void NoteUnitTouched(Int unit) {
             if (unit > m_maxTouchedUnit && unit < MAX_TEXTURE_IMAGE_UNITS) m_maxTouchedUnit = unit;
+            // Every texture/sampler bind entry point (glBindTexture / glBindTextureUnit /
+            // glBindTextures / glBindSampler) routes through here, so bumping the generation here
+            // - plus in MarkTextureObjectForDeletion for delete-unbind - covers every change to
+            // which texture is bound at which unit. A backend that has cached the per-draw
+            // sampled-texture set can compare this against a snapshot to skip re-resolving it when
+            // no bind changed (the block atlas + lightmap stay bound across a whole terrain batch).
+            ++m_textureBindGeneration;
         }
         Int GetMaxTouchedUnit() const { return m_maxTouchedUnit; }
+        Uint64 GetTextureBindGeneration() const { return m_textureBindGeneration; }
+        void BumpTextureBindGeneration() { ++m_textureBindGeneration; }
 
     private:
+        Uint64 m_textureBindGeneration = 0;
         Int m_maxTouchedUnit = -1;
         Int m_activeTextureUnit = 0;
         Array<TextureUnit, MAX_TEXTURE_IMAGE_UNITS> m_textureUnits;