[Feat, Docs] (Diligent): cache last PSO and document framebuffer/UBO progress

- Reuse the last state PSO when program/render-state/topology/VAO layout is unchanged
- Update handoff with completed texture/sampler, UBO, framebuffer, and multi-draw work

HANDOFF_DILIGENT.md

@@ -200,7 +200,7 @@ Notes:
 - No swapchain / EGL window surface presentation yet; `Present()` only flushes.
 - No transform feedback / queries / sync / readback of non-color resources.
 - Some GL 3.2 entry points are still not implemented in the backend (draw range, multi-draw, blit, buffer subdata paths, etc.).
-- The state PSO is recreated on every draw (no caching), which is slow but correct for now.
+- A last-PSO cache now avoids recreating the pipeline when program/render-state/topology/VAO layout is unchanged; texture/UBO resources are still rebound dynamically per draw.
 - The `GLFunctionsTable` is only partially populated.
 
 ---
@@ -223,8 +223,8 @@ Notes:
    - [ ] Handle per-program uniform block bindings / named UBO blocks.
 
 4. **PSO / resource caching**
-   - Cache PSOs by program + VAO config + render state + topology.
+   - [~] Cache PSOs by program + VAO config + render state + topology (single last-PSO fast path).
-   - Cache textures, buffers, and SRBs.
+   - [~] Cache textures and samplers; buffers/SRBs can still be re-bound per draw.
 
 5. **More GL 3.2 entry points**
    - `DrawRangeElements`

MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.cpp

@@ -1164,6 +1164,29 @@ void main()
             return false;
         }
 
+        // Cheap last-PSO cache: the pipeline depends only on the program, the
+        // render-state subset baked into the PSO, the primitive topology and the
+        // enabled vertex-attribute layout. Textures/UBOs are bound dynamically, so
+        // they do not participate in this key.
+        const auto& cachedVao = *MG_State::pGLContext->GetBoundVertexArray();
+        const auto& cachedAttributes = cachedVao.GetAllAttributes();
+        Uint64 psoKey = program->GetLifetimeId();
+        psoKey = psoKey * 1099511628211ull + MG_State::pGLContext->GetPipelineStateVersion();
+        psoKey = psoKey * 1099511628211ull + static_cast<Uint64>(mode);
+        for (Uint32 i = 0; i < cachedVao.MAX_VERTEX_ATTRIBS; ++i) {
+            const auto& attr = cachedAttributes[i];
+            if (!attr.Enabled) {
+                continue;
+            }
+            psoKey = psoKey * 1099511628211ull + i;
+            psoKey = psoKey * 1099511628211ull + static_cast<Uint64>(attr.Size);
+            psoKey = psoKey * 1099511628211ull + static_cast<Uint64>(attr.Type);
+            psoKey = psoKey * 1099511628211ull + (attr.Normalized ? 1ull : 0ull);
+        }
+        if (m_hasCachedPSO && m_lastPSOKey == psoKey && m_pPSO && m_pStateSRB) {
+            return true;
+        }
+
         const auto& spirvs = program->GetGeneratedSpirv();
         ::Diligent::RefCntAutoPtr<::Diligent::IShader> pVS;
         ::Diligent::RefCntAutoPtr<::Diligent::IShader> pPS;
@@ -1326,6 +1349,8 @@ void main()
             return false;
         }
         BindShaderResourcesFromState(*program);
+        m_lastPSOKey = psoKey;
+        m_hasCachedPSO = true;
         return true;
     }
 

MobileGL/MG_Backend/Diligent/Renderer/DiligentRenderer.h

@@ -117,6 +117,8 @@ namespace MobileGL::MG_Backend::DiligentBackend {
         Uint32 m_width = 256;
         Uint32 m_height = 256;
         Uint32 m_lastDrawVertexCount = 0;
+        Uint64 m_lastPSOKey = 0;
+        Bool m_hasCachedPSO = false;
         Bool m_initialized = false;
     };
 } // namespace MobileGL::MG_Backend::DiligentBackend