Eagerly destroy CommandBuffer commands after submission

Command buffers hold references to all encoded objects. Freeing them
eagerly significantly reduces the amount memory held before the JS GC
clears the command buffers.

Bug: dawn:262, dawn:372
Change-Id: I68dfa973f980fba8d94611ed1de3c593bdb91a63
Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/26562
Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Rafael Cintron <rafael.cintron@microsoft.com>
Commit-Queue: Austin Eng <enga@chromium.org>

src/tests/unittests/validation/QueueSubmitValidationTests.cpp

@@ -23,9 +23,10 @@ namespace {
     // Test submitting with a mapped buffer is disallowed
     TEST_F(QueueSubmitValidationTest, SubmitWithMappedBuffer) {
         // Create a map-write buffer.
+        const uint64_t kBufferSize = 4;
         wgpu::BufferDescriptor descriptor;
         descriptor.usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc;
-        descriptor.size = 4;
+        descriptor.size = kBufferSize;
         wgpu::Buffer buffer = device.CreateBuffer(&descriptor);
 
         // Create a fake copy destination buffer
@@ -36,7 +37,7 @@ namespace {
         wgpu::CommandBuffer commands;
         {
             wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
-            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, 4);
+            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, kBufferSize);
             commands = encoder.Finish();
         }
 
@@ -45,17 +46,35 @@ namespace {
         // Submitting when the buffer has never been mapped should succeed
         queue.Submit(1, &commands);
 
+        {
+            wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
+            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, kBufferSize);
+            commands = encoder.Finish();
+        }
+
         // Map the buffer, submitting when the buffer is mapped should fail
-        buffer.MapWriteAsync(nullptr, nullptr);
+        buffer.MapAsync(wgpu::MapMode::Write, 0, kBufferSize, nullptr, nullptr);
 
         // Try submitting before the callback is fired.
         ASSERT_DEVICE_ERROR(queue.Submit(1, &commands));
 
         WaitForAllOperations(device);
 
+        {
+            wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
+            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, kBufferSize);
+            commands = encoder.Finish();
+        }
+
         // Try submitting after the callback is fired.
         ASSERT_DEVICE_ERROR(queue.Submit(1, &commands));
 
+        {
+            wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
+            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, kBufferSize);
+            commands = encoder.Finish();
+        }
+
         // Unmap the buffer, queue submit should succeed
         buffer.Unmap();
         queue.Submit(1, &commands);
@@ -182,4 +201,53 @@ namespace {
         }
     }
 
+    // Test it is invalid to submit a command buffer twice
+    TEST_F(QueueSubmitValidationTest, CommandBufferSubmittedTwice) {
+        wgpu::CommandBuffer commandBuffer = device.CreateCommandEncoder().Finish();
+        wgpu::Queue queue = device.GetDefaultQueue();
+
+        // Should succeed
+        queue.Submit(1, &commandBuffer);
+
+        // Should fail because command buffer was already submitted
+        ASSERT_DEVICE_ERROR(queue.Submit(1, &commandBuffer));
+    }
+
+    // Test resubmitting failed command buffers
+    TEST_F(QueueSubmitValidationTest, CommandBufferSubmittedFailed) {
+        // Create a map-write buffer
+        const uint64_t kBufferSize = 4;
+        wgpu::BufferDescriptor descriptor;
+        descriptor.usage = wgpu::BufferUsage::MapWrite | wgpu::BufferUsage::CopySrc;
+        descriptor.size = kBufferSize;
+        wgpu::Buffer buffer = device.CreateBuffer(&descriptor);
+
+        // Create a destination buffer for the b2b copy
+        descriptor.usage = wgpu::BufferUsage::CopyDst;
+        descriptor.size = kBufferSize;
+        wgpu::Buffer targetBuffer = device.CreateBuffer(&descriptor);
+
+        // Create a command buffer that reads from the mappable buffer
+        wgpu::CommandBuffer commands;
+        {
+            wgpu::CommandEncoder encoder = device.CreateCommandEncoder();
+            encoder.CopyBufferToBuffer(buffer, 0, targetBuffer, 0, kBufferSize);
+            commands = encoder.Finish();
+        }
+
+        wgpu::Queue queue = device.GetDefaultQueue();
+
+        // Map the source buffer to force a failure
+        buffer.MapAsync(wgpu::MapMode::Write, 0, kBufferSize, nullptr, nullptr);
+
+        // Submitting a command buffer with a mapped buffer should fail
+        ASSERT_DEVICE_ERROR(queue.Submit(1, &commands));
+
+        // Unmap buffer to fix the failure
+        buffer.Unmap();
+
+        // Resubmitting any command buffer, even if the problem was fixed, should fail
+        ASSERT_DEVICE_ERROR(queue.Submit(1, &commands));
+    }
+
 }  // anonymous namespace