Use GetMappedRange early but still get null pointer sometimes

filament/backend/CMakeLists.txt

@@ -316,6 +316,8 @@ if (FILAMENT_SUPPORTS_WEBGPU)
             src/webgpu/WebGPURenderPrimitive.h
             src/webgpu/WebGPURenderTarget.cpp
             src/webgpu/WebGPURenderTarget.h
+            src/webgpu/WebGPUStagePool.cpp
+            src/webgpu/WebGPUStagePool.h
             src/webgpu/WebGPUStrings.h
             src/webgpu/WebGPUSwapChain.cpp
             src/webgpu/WebGPUSwapChain.h

filament/backend/src/webgpu/WebGPUBufferBase.cpp

@@ -18,6 +18,7 @@
 
 #include "WebGPUConstants.h"
 #include "WebGPUQueueManager.h"
+#include "WebGPUStagePool.h"
 
 #include "DriverBase.h"
 #include <backend/BufferDescriptor.h>
@@ -29,6 +30,7 @@
 
 #include <cstdint>
 #include <cstring>
+#include <iostream>
 
 namespace filament::backend {
 
@@ -65,7 +67,7 @@ WebGPUBufferBase::WebGPUBufferBase(wgpu::Device const& device, const wgpu::Buffe
 // of 4 by padding with zeros.
 void WebGPUBufferBase::updateGPUBuffer(BufferDescriptor const& bufferDescriptor,
         const uint32_t byteOffset, wgpu::Device const& device,
-        WebGPUQueueManager* const webGPUQueueManager) {
+        WebGPUQueueManager* const webGPUQueueManager, WebGPUStagePool* const webGPUStagePool) {
     FILAMENT_CHECK_PRECONDITION(bufferDescriptor.buffer)
             << "updateGPUBuffer called with a null buffer";
     FILAMENT_CHECK_PRECONDITION(bufferDescriptor.size + byteOffset <= mBuffer.GetSize())
@@ -79,34 +81,70 @@ void WebGPUBufferBase::updateGPUBuffer(BufferDescriptor const& bufferDescriptor,
     // This may have some performance implications. That should be investigated later.
     assert_invariant(mBuffer.GetUsage() & wgpu::BufferUsage::CopyDst);
 
-    // Calculate some alignment related sizes
+    // // Calculate some alignment related sizes
     const size_t remainder = bufferDescriptor.size % FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS;
     const size_t mainBulk = bufferDescriptor.size - remainder;
     const size_t stagingBufferSize =
             remainder == 0 ? bufferDescriptor.size : mainBulk + FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS;
+    //
+    // // create a staging buffer
+    // wgpu::BufferDescriptor descriptor{
+    //     .label = "Filament WebGPU Staging Buffer",
+    //     .usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc,
+    //     .size = stagingBufferSize,
+    //     .mappedAtCreation = true };
+    // wgpu::Buffer stagingBuffer = device.CreateBuffer(&descriptor);
+    MappedStage mappedStage = webGPUStagePool->acquireBuffer(stagingBufferSize);
 
-    // create a staging buffer
-    wgpu::BufferDescriptor descriptor{
-        .label = "Filament WebGPU Staging Buffer",
-        .usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc,
-        .size = stagingBufferSize,
-        .mappedAtCreation = true };
-    wgpu::Buffer stagingBuffer = device.CreateBuffer(&descriptor);
-
-    void* mappedRange = stagingBuffer.GetMappedRange();
-    memcpy(mappedRange, bufferDescriptor.buffer, bufferDescriptor.size);
+    std::string mappedRangeIsNull = mappedStage.mappedRange
+            ? "no"
+            : "yes";
+    std::cout << "Run Yu: got mapped range on the staging buffer with size "
+              << mappedStage.buffer.GetSize() << " and it is null? " <<  mappedRangeIsNull << std::endl;
+    memcpy(mappedStage.mappedRange, bufferDescriptor.buffer, bufferDescriptor.size);
 
     // Make sure the padded memory is set to 0 to have deterministic behaviors
-    if (remainder != 0) {
-        uint8_t* paddingStart = static_cast<uint8_t*>(mappedRange) + bufferDescriptor.size;
-        memset(paddingStart, 0, FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS - remainder);
-    }
+    // if (remainder != 0) {
+    //     uint8_t* paddingStart = static_cast<uint8_t*>(mappedRange) + bufferDescriptor.size;
+    //     memset(paddingStart, 0, FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS - remainder);
+    // }
+    // size_t stagingBufferSize = stagingBuffer.GetSize();
+    // if (stagingBufferSize != bufferDescriptor.size) {
+    //     assert(stagingBufferSize > bufferDescriptor.size);
+    //     assert(stagingBufferSize % FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS == 0);
+    //     uint8_t* paddingStart = static_cast<uint8_t*>(mappedRange) + bufferDescriptor.size;
+    //     memset(paddingStart, 0, FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS - (stagingBuffer.GetSize() - bufferDescriptor.size));
+    // }
 
-    stagingBuffer.Unmap();
+    mappedStage.buffer.Unmap();
 
+    std::cout << "Run Yu: about to issue copy command with actual staging buffer of size "
+              << mappedStage.buffer.GetSize() << ", and computed size of " << stagingBufferSize
+              << ". The mBuffer size is " << mBuffer.GetSize() << std::endl;
     // Copy the staging buffer contents to the destination buffer.
-    webGPUQueueManager->getCommandEncoder().CopyBufferToBuffer(stagingBuffer, 0, mBuffer,
-            byteOffset, stagingBufferSize);
+    webGPUQueueManager->getCommandEncoder().CopyBufferToBuffer(mappedStage.buffer, 0, mBuffer,
+            byteOffset,
+            remainder == 0 ? bufferDescriptor.size
+                           : mainBulk + FILAMENT_WEBGPU_BUFFER_SIZE_MODULUS);
+    webGPUQueueManager->flush();
+
+    struct UserData final {
+        wgpu::Buffer stagingBuffer;
+        WebGPUStagePool* webGPUStagePool;
+    };
+    auto userData = std::make_unique<UserData>(
+            UserData{ .stagingBuffer = mappedStage.buffer, .webGPUStagePool = webGPUStagePool });
+    mappedStage.buffer.MapAsync(
+            wgpu::MapMode::Write, 0, stagingBufferSize, wgpu::CallbackMode::AllowSpontaneous,
+            [](wgpu::MapAsyncStatus status, const char* message, UserData* userData) {
+                if (UTILS_LIKELY(status == wgpu::MapAsyncStatus::Success)) {
+                    std::unique_ptr<UserData> data(static_cast<UserData*>(userData));
+                    userData->webGPUStagePool->addBufferToPool(userData->stagingBuffer);
+                } else {
+                    std::cout << "Run Yu: MAPPING UNSUCCESSFUL!!\n";
+                }
+            },
+            userData.release());
 }
 
 } // namespace filament::backend

filament/backend/src/webgpu/WebGPUBufferBase.h

@@ -25,6 +25,7 @@ namespace filament::backend {
 
 class BufferDescriptor;
 class WebGPUQueueManager;
+class WebGPUStagePool;
 
 /**
   * A base class for WebGPU buffer objects, providing common functionality for creating and
@@ -40,7 +41,7 @@ public:
      * ensures the calls happen in the expected sequence.
      */
     void updateGPUBuffer(BufferDescriptor const&, uint32_t byteOffset, wgpu::Device const& device,
-            WebGPUQueueManager* const webGPUQueueManager);
+            WebGPUQueueManager* const webGPUQueueManager, WebGPUStagePool* const webGPUStagePool);
 
     [[nodiscard]] wgpu::Buffer const& getBuffer() const { return mBuffer; }
 

filament/backend/src/webgpu/WebGPUDriver.cpp

@@ -107,6 +107,7 @@ WebGPUDriver::WebGPUDriver(WebGPUPlatform& platform,
       mAdapter{ mPlatform.requestAdapter(nullptr) },
       mDevice{ mPlatform.requestDevice(mAdapter) },
       mQueueManager{ mDevice },
+      mStagePool{ mDevice },
       mPipelineLayoutCache{ mDevice },
       mPipelineCache{ mDevice },
       mRenderPassMipmapGenerator{ mDevice, &mQueueManager },
@@ -856,7 +857,7 @@ void WebGPUDriver::updateIndexBuffer(Handle<HwIndexBuffer> indexBufferHandle,
     // draw calls are made.
     flush();
     handleCast<WebGPUIndexBuffer>(indexBufferHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager, &mStagePool);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 
@@ -867,14 +868,14 @@ void WebGPUDriver::updateBufferObject(Handle<HwBufferObject> bufferObjectHandle,
     // draw calls are made.
     flush();
     handleCast<WebGPUBufferObject>(bufferObjectHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager, &mStagePool);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 
 void WebGPUDriver::updateBufferObjectUnsynchronized(Handle<HwBufferObject> bufferObjectHandle,
         BufferDescriptor&& bufferDescriptor, const uint32_t byteOffset) {
     handleCast<WebGPUBufferObject>(bufferObjectHandle)
-            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager);
+            ->updateGPUBuffer(bufferDescriptor, byteOffset, mDevice, &mQueueManager, &mStagePool);
     scheduleDestroy(std::move(bufferDescriptor));
 }
 

filament/backend/src/webgpu/WebGPUDriver.h

@@ -25,6 +25,7 @@
 #include "webgpu/WebGPUPipelineLayoutCache.h"
 #include "webgpu/WebGPURenderPassMipmapGenerator.h"
 #include "webgpu/WebGPUQueueManager.h"
+#include "webgpu/WebGPUStagePool.h"
 #include "webgpu/utils/AsyncTaskCounter.h"
 #include <backend/platforms/WebGPUPlatform.h>
 
@@ -81,6 +82,7 @@ private:
     wgpu::Device mDevice = nullptr;
     wgpu::Limits mDeviceLimits = {};
     WebGPUQueueManager mQueueManager;
+    WebGPUStagePool mStagePool;
     void* mNativeWindow = nullptr;
     WebGPUSwapChain* mSwapChain = nullptr;
     uint64_t mNextFakeHandle = 1;

filament/backend/src/webgpu/WebGPUStagePool.cpp

@@ -0,0 +1,88 @@
+/*
+* Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include "WebGPUStagePool.h"
+
+#include "WebGPUConstants.h"
+
+#include <iostream>
+
+namespace filament::backend {
+
+WebGPUStagePool::WebGPUStagePool(wgpu::Device const& device) : mDevice(device) {}
+
+WebGPUStagePool::~WebGPUStagePool() = default;
+
+MappedStage WebGPUStagePool::acquireBuffer(size_t requiredSize) {
+    std::cout << "Run Yu: required size in acquireBuffer: " << requiredSize << std::endl;
+    std::cout << "Run Yu: the pool size is " << mBuffers.size() << std::endl;
+    {
+        std::lock_guard<std::mutex> lock(mMutex);
+        auto iter = mBuffers.lower_bound(requiredSize);
+        if (iter != mBuffers.end()) {
+            std::cout << "Run Yu: found buffer in the pool with size " << iter->second.GetSize()
+                      << std::endl;
+            if (iter->second.GetMapState() != wgpu::BufferMapState::Mapped) {
+                std::cout << "Run Yu: before GetMappedRange the buffer state is not mapped!\n";
+            }
+            MappedStage mappedStage = { .buffer = iter->second,
+                .mappedRange = iter->second.GetMappedRange() };
+            if (!mappedStage.mappedRange) {
+                std::cout << "Run Yu: mapped range is null in acquireBuffer!\n";
+            }
+            if (mappedStage.buffer.GetMapState() != wgpu::BufferMapState::Mapped) {
+                std::cout << "Run Yu: after GetMappedRange the buffer state is not mapped!\n";
+            }
+            mBuffers.erase(iter);
+            return mappedStage;
+        }
+    }
+    wgpu::Buffer newBuffer = createNewBuffer(requiredSize);
+    return { .buffer = newBuffer, .mappedRange = newBuffer.GetMappedRange() };
+}
+
+void WebGPUStagePool::addBufferToPool(wgpu::Buffer buffer) {
+    std::lock_guard<std::mutex> lock(mMutex);
+    std::cout << "Run Yu: adding buffer to the pool with size " << buffer.GetSize() << std::endl;
+    mBuffers.insert({buffer.GetSize(), buffer});
+    std::cout << "Run Yu: added buffer to the pool with size " << buffer.GetSize() << std::endl;
+
+    bool allMapped = true;
+    for (const auto& pair : mBuffers) {
+        auto state = pair.second.GetMapState();
+        if (state != wgpu::BufferMapState::Mapped) {
+            allMapped = false;
+            std::cout << "Run Yu: the buffer with size " << pair.second.GetSize()
+                      << " is not mapped but somehow was added to the pool, its state is "
+                      << static_cast<int>(state) << std::endl;
+        }
+    }
+    if (!allMapped) {
+        std::cout << "Run Yu: found buffers that are not mapped\n";
+    }
+}
+
+wgpu::Buffer WebGPUStagePool::createNewBuffer(size_t bufferSize) {
+    std::cout << "Run Yu: creating new buffer with size " << bufferSize << std::endl;
+    wgpu::BufferDescriptor descriptor{
+        .label = "Filament WebGPU Staging Buffer",
+        .usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc,
+        .size = bufferSize,
+        .mappedAtCreation = true };
+    return mDevice.CreateBuffer(&descriptor);
+}
+
+} // namespace filament::backend

filament/backend/src/webgpu/WebGPUStagePool.h

@@ -0,0 +1,49 @@
+/*
+* Copyright (C) 2025 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef TNT_FILAMENT_BACKEND_WEBGPUSTAGEPOOL_H
+#define TNT_FILAMENT_BACKEND_WEBGPUSTAGEPOOL_H
+
+#include <webgpu/webgpu_cpp.h>
+
+#include <map>
+#include <mutex>
+
+namespace filament::backend {
+
+struct MappedStage {
+    wgpu::Buffer buffer;
+    void* mappedRange;
+};
+
+class WebGPUStagePool {
+public:
+    WebGPUStagePool(wgpu::Device const& device);
+    ~WebGPUStagePool();
+
+    MappedStage acquireBuffer(size_t requiredSize);
+    void addBufferToPool(wgpu::Buffer buffer);
+private:
+    wgpu::Buffer createNewBuffer(size_t bufferSize);
+    std::multimap<uint32_t, wgpu::Buffer> mBuffers;
+    mutable std::mutex mMutex;
+
+    wgpu::Device mDevice;
+};
+
+}
+
+#endif // TNT_FILAMENT_BACKEND_WEBGPUSTAGEPOOL_H