[GFX] Fix d3d12 buffer view creation logic for StructuredBuffers. (#3954)

2 files changed, 110 insertions, 0 deletions

diff --git a/tools/gfx-unit-test/uint16-buffer.slang b/tools/gfx-unit-test/uint16-buffer.slang
new file mode 100644
index 000000000..bdc09c7d8
--- /dev/null
+++ b/tools/gfx-unit-test/uint16-buffer.slang
@@ -0,0 +1,14 @@
+// uint16-buffer.slang - Simple shader that takes a buffer of uint16 type and increments all elements by 1.
+
+// This is to verify that GFX can correct set correct buffer strides for structured buffer bindings.
+
+uniform RWStructuredBuffer<uint16_t> buffer;
+
+[shader("compute")]
+[numthreads(4,1,1)]
+void computeMain(
+    uint3 sv_dispatchThreadID : SV_DispatchThreadID)
+{
+    var input = buffer[sv_dispatchThreadID.x];
+    buffer[sv_dispatchThreadID.x] = input + 1;
+}
diff --git a/tools/gfx-unit-test/uint16-structured-buffer.cpp b/tools/gfx-unit-test/uint16-structured-buffer.cpp
new file mode 100644
index 000000000..8f2f2cb97
--- /dev/null
+++ b/tools/gfx-unit-test/uint16-structured-buffer.cpp
@@ -0,0 +1,96 @@
+#include "tools/unit-test/slang-unit-test.h"
+
+#include "slang-gfx.h"
+#include "gfx-test-util.h"
+#include "tools/gfx-util/shader-cursor.h"
+#include "source/core/slang-basic.h"
+
+using namespace gfx;
+
+namespace gfx_test
+{
+    void uint16BufferTestImpl(IDevice* device, UnitTestContext* context)
+    {
+        Slang::ComPtr<ITransientResourceHeap> transientHeap;
+        ITransientResourceHeap::Desc transientHeapDesc = {};
+        transientHeapDesc.constantBufferSize = 4096;
+        GFX_CHECK_CALL_ABORT(
+            device->createTransientResourceHeap(transientHeapDesc, transientHeap.writeRef()));
+
+        ComPtr<IShaderProgram> shaderProgram;
+        slang::ProgramLayout* slangReflection;
+        GFX_CHECK_CALL_ABORT(loadComputeProgram(device, shaderProgram, "uint16-buffer", "computeMain", slangReflection));
+
+        ComputePipelineStateDesc pipelineDesc = {};
+        pipelineDesc.program = shaderProgram.get();
+        ComPtr<gfx::IPipelineState> pipelineState;
+        GFX_CHECK_CALL_ABORT(
+            device->createComputePipelineState(pipelineDesc, pipelineState.writeRef()));
+
+        const int numberCount = 4;
+        uint16_t initialData[] = { 0, 1, 2, 3 };
+        IBufferResource::Desc bufferDesc = {};
+        bufferDesc.sizeInBytes = numberCount * sizeof(uint16_t);
+        bufferDesc.format = gfx::Format::Unknown;
+        // Note: we don't specify any element size here, and gfx should be able to derive the
+        // correct element size from the reflection infomation.
+        bufferDesc.elementSize = 0;
+        bufferDesc.allowedStates = ResourceStateSet(
+            ResourceState::ShaderResource,
+            ResourceState::UnorderedAccess,
+            ResourceState::CopyDestination,
+            ResourceState::CopySource);
+        bufferDesc.defaultState = ResourceState::UnorderedAccess;
+        bufferDesc.memoryType = MemoryType::DeviceLocal;
+
+        ComPtr<IBufferResource> numbersBuffer;
+        GFX_CHECK_CALL_ABORT(device->createBufferResource(
+            bufferDesc,
+            (void*)initialData,
+            numbersBuffer.writeRef()));
+
+        ComPtr<IResourceView> bufferView;
+        IResourceView::Desc viewDesc = {};
+        viewDesc.type = IResourceView::Type::UnorderedAccess;
+        viewDesc.format = Format::Unknown;
+        GFX_CHECK_CALL_ABORT(
+            device->createBufferView(numbersBuffer, nullptr, viewDesc, bufferView.writeRef()));
+
+        // We have done all the set up work, now it is time to start recording a command buffer for
+        // GPU execution.
+        {
+            ICommandQueue::Desc queueDesc = { ICommandQueue::QueueType::Graphics };
+            auto queue = device->createCommandQueue(queueDesc);
+
+            auto commandBuffer = transientHeap->createCommandBuffer();
+            auto encoder = commandBuffer->encodeComputeCommands();
+
+            auto rootObject = encoder->bindPipeline(pipelineState);
+
+            // Bind buffer view to the entry point.
+            ShaderCursor(rootObject).getPath("buffer").setResource(bufferView);
+
+            encoder->dispatchCompute(1, 1, 1);
+            encoder->endEncoding();
+            commandBuffer->close();
+            queue->executeCommandBuffer(commandBuffer);
+            queue->waitOnHost();
+        }
+
+        compareComputeResult(
+            device,
+            numbersBuffer,
+            Slang::makeArray<uint16_t>(1, 2, 3, 4));
+    }
+
+    SLANG_UNIT_TEST(uint16BufferTestD3D12)
+    {
+        runTestImpl(uint16BufferTestImpl, unitTestContext, Slang::RenderApiFlag::D3D12);
+    }
+
+    SLANG_UNIT_TEST(uint16BufferTestVulkan)
+    {
+        runTestImpl(uint16BufferTestImpl, unitTestContext, Slang::RenderApiFlag::Vulkan);
+    }
+
+}