2 files changed, 96 insertions, 5 deletions

diff --git a/tests/cpu-program/gfx-smoke.slang b/tests/cpu-program/gfx-smoke.slang
index fda6d5454..2ad33052f 100644
--- a/tests/cpu-program/gfx-smoke.slang
+++ b/tests/cpu-program/gfx-smoke.slang
@@ -3,14 +3,102 @@ __target_intrinsic(cpp, "printf(\"%s\\n\", ($0).getBuffer())")
 void writeln(String text);
 
 import gfx;
+import slang;
 
 public __extern_cpp int main()
 {
     gfx.DeviceDesc deviceDesc = {};
     deviceDesc.deviceType = gfx.DeviceType.CPU;
-    Optional<gfx.IDevice> obj;
-    if (gfx.succeeded(gfx.gfxCreateDevice(&deviceDesc, obj)) && obj.hasValue)
-        writeln("succ");
-    else
+    Optional<gfx.IDevice> device;
+    gfx.gfxCreateDevice(&deviceDesc, device);
+    if (device == none)
+    {
         writeln("fail");
+        return -1;
+    }
+    
+    gfx.CommandQueueDesc queueDesc = {};
+    queueDesc.type = gfx.QueueType.Graphics;
+    Optional<gfx.ICommandQueue> queue;
+    device.value.createCommandQueue(&queueDesc, queue);
+
+    gfx.ShaderProgramDesc2 programDesc = {};
+    NativeString s =
+        "[shader(\"compute\")]"
+        "[numthreads(4, 1, 1)]"
+        "void computeMain("
+        "    uint3 sv_dispatchThreadID: SV_DispatchThreadID,"
+        "    uniform RWStructuredBuffer<float> buffer"
+        "    )"
+        "{"
+        "    var input = buffer[sv_dispatchThreadID.x];"
+        "    buffer[sv_dispatchThreadID.x] = sv_dispatchThreadID.x;"
+        "}";
+    programDesc.sourceData = s;
+    programDesc.sourceType = gfx.ShaderModuleSourceType.SlangSource;
+    programDesc.sourceDataSize = s.length;
+    programDesc.entryPointCount = 1;
+    NativeString entryPointName = "computeMain";
+    programDesc.entryPointNames = &entryPointName;
+    Optional<gfx.IShaderProgram> program;
+    Optional<slang.ISlangBlob> diagBlob;
+    device.value.createProgram2(&programDesc, program, diagBlob);
+
+    Optional<gfx.IPipelineState> pipeline;
+    gfx.ComputePipelineStateDesc pipelineDesc = {};
+    pipelineDesc.program = NativeRef<gfx.IShaderProgram>(program.value);
+    device.value.createComputePipelineState(&pipelineDesc, pipeline);
+
+    Optional<gfx.ITransientResourceHeap> transientHeap;
+    gfx.TransientResourceHeapDesc transientHeapDesc = {};
+    transientHeapDesc.constantBufferDescriptorCount = 64;
+    transientHeapDesc.constantBufferSize = 1024;
+    transientHeapDesc.srvDescriptorCount = 1024;
+    transientHeapDesc.uavDescriptorCount = 1024;
+    transientHeapDesc.samplerDescriptorCount = 256;
+    transientHeapDesc.accelerationStructureDescriptorCount = 32;
+    device.value.createTransientResourceHeap(&transientHeapDesc, transientHeap);
+
+    Optional<gfx.IBufferResource> buffer;
+    gfx.BufferResourceDesc bufferDesc = {};
+    bufferDesc.memoryType = gfx.MemoryType.DeviceLocal;
+    bufferDesc.allowedStates.add(gfx.ResourceState.UnorderedAccess);
+    bufferDesc.defaultState = gfx.ResourceState.UnorderedAccess;
+    bufferDesc.elementSize = 4;
+    bufferDesc.sizeInBytes = 256;
+    bufferDesc.type = gfx.ResourceType.Buffer;
+    device.value.createBufferResource(&bufferDesc, nullptr, buffer);
+
+    Optional<gfx.IResourceView> bufferView;
+    gfx.ResourceViewDesc viewDesc = {};
+    viewDesc.type = gfx.ResourceViewType.UnorderedAccess;
+    device.value.createBufferView(buffer.value, none, &viewDesc, bufferView);
+
+    Optional<gfx.ICommandBuffer> commandBuffer;
+    transientHeap.value.createCommandBuffer(commandBuffer);
+    Optional<gfx.IComputeCommandEncoder> encoder;
+    commandBuffer.value.encodeComputeCommands(encoder);
+    Optional<gfx.IShaderObject> rootObject;
+    encoder.value.bindPipeline(pipeline.value, rootObject);
+    Optional<gfx.IShaderObject> entryPointObject;
+    rootObject.value.getEntryPoint(0, entryPointObject);
+    gfx.ShaderOffset offset = {};
+    entryPointObject.value.setResource(&offset, bufferView.value);
+    encoder.value.dispatchCompute(1, 1, 1);
+    encoder.value.endEncoding();
+    commandBuffer.value.close();
+    
+    NativeRef<gfx.ICommandBuffer> commandBufferRef = NativeRef<gfx.ICommandBuffer>(commandBuffer.value);
+    queue.value.executeCommandBuffers(1, &commandBufferRef, none, 0);
+    queue.value.waitOnHost();
+
+    Optional<slang.ISlangBlob> blob;
+    device.value.readBufferResource(buffer.value, 0, 16, blob);
+
+    for (int i = 0; i < 4; i++)
+    {
+        float val = ((float *)blob.value.getBufferPointer())[i];
+        writeln(String(val));
+    }
+    return 0;
 }
\ No newline at end of file
diff --git a/tests/cpu-program/gfx-smoke.slang.expected b/tests/cpu-program/gfx-smoke.slang.expected
index 981dcfc29..aa0dd7680 100644
--- a/tests/cpu-program/gfx-smoke.slang.expected
+++ b/tests/cpu-program/gfx-smoke.slang.expected
@@ -2,5 +2,8 @@ result code = 0
 standard error = {
 }
 standard output = {
-succ
+0
+1
+2
+3
 }