Bring heterogeneous-hello-world back up to date. (#1935)

* Bring heterogeneous-hello-world back up to date.

* Reintroduced heterogeneous-hello-world into the premake

* No longer uses compiled bytecode for entry point, instead a loadModule
call is hardocoded with the slang file name.

* Entry point is, similarly, hardcoded for now.

* Added a bypass to slang-legalize-types for an unneeded GPUForeach check

* Run premake and change to relative path

* Removed experimental and added README

Co-authored-by: Yong He <yonghe@outlook.com>

4 files changed, 625 insertions, 0 deletions

diff --git a/examples/heterogeneous-hello-world/README.md b/examples/heterogeneous-hello-world/README.md
new file mode 100644
index 000000000..709652922
--- /dev/null
+++ b/examples/heterogeneous-hello-world/README.md
@@ -0,0 +1,4 @@
+Slang "CPU Hello World Heterogeneous" Example
+===============================
+
+This example is a work-in-progress to illustrate how a heterogeneous programming example might work.  It should NOT be used as a reference for working Slang code yet.
\ No newline at end of file
diff --git a/examples/heterogeneous-hello-world/main.cpp b/examples/heterogeneous-hello-world/main.cpp
new file mode 100644
index 000000000..9e0bb8b0f
--- /dev/null
+++ b/examples/heterogeneous-hello-world/main.cpp
@@ -0,0 +1,335 @@
+// main.cpp
+
+// This example uses the Slang gfx layer to target different APIs and execute
+// both CPU and GPU code from a single Slang file (?)
+//
+#include <slang.h>
+#include <slang-com-ptr.h>
+using Slang::ComPtr;
+
+#include "slang-gfx.h"
+#include "gfx-util/shader-cursor.h"
+#include "source/core/slang-basic.h"
+#include "../../prelude/slang-cpp-types.h"
+
+using namespace gfx;
+using namespace Slang;
+
+// Creating global ref pointers to avoid dereferencing values
+//
+ComPtr<gfx::IDevice> gDevice;
+ComPtr<gfx::IShaderProgram> gProgram;
+ComPtr<gfx::IBufferResource> gBufferResource;
+ComPtr<gfx::IResourceView> gResourceView;
+ComPtr<gfx::ITransientResourceHeap> gTransientHeap;
+ComPtr<gfx::IPipelineState> gPipelineState;
+ComPtr<gfx::ICommandQueue> gQueue;
+
+// Boilerplate types to help the slang-generated file
+//
+struct gfx_Device_0;
+struct gfx_BufferResource_0;
+struct gfx_ShaderProgram_0;
+struct gfx_ResourceView_0;
+struct gfx_TransientResourceHeap_0;
+struct gfx_PipelineState_0;
+bool executeComputation_0();
+
+// Many Slang API functions return detailed diagnostic information
+// (error messages, warnings, etc.) as a "blob" of data, or return
+// a null blob pointer instead if there were no issues.
+//
+// For convenience, we define a subroutine that will dump the information
+// in a diagnostic blob if one is produced, and skip it otherwise.
+//
+void diagnoseIfNeeded(slang::IBlob *diagnosticsBlob)
+{
+    if (diagnosticsBlob != nullptr)
+    {
+        printf("%s", (const char *)diagnosticsBlob->getBufferPointer());
+    }
+}
+
+gfx::IDevice* createDevice()
+{
+    ComPtr<gfx::IDevice> device;
+    IDevice::Desc deviceDesc = {};
+    // Changing device type would happen here. For example:
+    //deviceDesc.deviceType = DeviceType::CUDA;
+    SLANG_RETURN_NULL_ON_FAIL(gfxCreateDevice(&deviceDesc, gDevice.writeRef()));
+    return gDevice;
+}
+
+// Loads the shader code defined in `shader.slang` for use by the `gfx` layer.
+//
+gfx::IShaderProgram* loadShaderProgram(gfx::IDevice *device)
+{
+    // We need to obtain a compilation session (`slang::ISession`) that will provide
+    // a scope to all the compilation and loading of code we do.
+    //
+    ComPtr<slang::ISession> slangSession;
+    SLANG_RETURN_NULL_ON_FAIL(device->getSlangSession(slangSession.writeRef()));
+
+    // We can now start loading code into the slang session.
+    //
+    // The simplest way to load code is by calling `loadModule` with the name of a Slang
+    // module. A call to `loadModule("MyStuff")` will behave more or less as if you
+    // wrote:
+    //
+    //      import MyStuff;
+    //
+    // In a Slang shader file. The compiler will use its search paths to try to locate
+    // `MyModule.slang`, then compile and load that file. If a matching module had
+    // already been loaded previously, that would be used directly.
+    //
+    ComPtr<slang::IBlob> diagnosticsBlob;
+    slang::IModule *module = slangSession->loadModule("shader", diagnosticsBlob.writeRef());
+    diagnoseIfNeeded(diagnosticsBlob);
+    if (!module)
+        return NULL;
+
+    // Look up entry point (hardcoded for now)
+    //
+    char const *computeEntryPointName = "computeMain";
+    ComPtr<slang::IEntryPoint> computeEntryPoint;
+    SLANG_RETURN_NULL_ON_FAIL(
+        module->findEntryPointByName(computeEntryPointName, computeEntryPoint.writeRef()));
+
+    // At this point we have a few different Slang API objects that represent
+    // pieces of our code: `module`, `vertexEntryPoint`, and `fragmentEntryPoint`.
+    //
+    // A single Slang module could contain many different entry points (e.g.,
+    // four vertex entry points, three fragment entry points, and two compute
+    // shaders), and before we try to generate output code for our target API
+    // we need to identify which entry points we plan to use together.
+    //
+    // Modules and entry points are both examples of *component types* in the
+    // Slang API. The API also provides a way to build a *composite* out of
+    // other pieces, and that is what we are going to do with our module
+    // and entry points.
+    //
+    Slang::List<slang::IComponentType *> componentTypes;
+    componentTypes.add(module);
+    componentTypes.add(computeEntryPoint);
+
+    // Actually creating the composite component type is a single operation
+    // on the Slang session, but the operation could potentially fail if
+    // something about the composite was invalid (e.g., you are trying to
+    // combine multiple copies of the same module), so we need to deal
+    // with the possibility of diagnostic output.
+    //
+    ComPtr<slang::IComponentType> composedProgram;
+    SlangResult result = slangSession->createCompositeComponentType(
+        componentTypes.getBuffer(),
+        componentTypes.getCount(),
+        composedProgram.writeRef(),
+        diagnosticsBlob.writeRef());
+    diagnoseIfNeeded(diagnosticsBlob);
+    SLANG_RETURN_NULL_ON_FAIL(result);
+
+    // At this point, `composedProgram` represents the shader program
+    // we want to run, and the compute shader there have been checked.
+    // We can create a `gfx::IShaderProgram` object from `composedProgram`
+    // so it may be used by the graphics layer.
+    gfx::IShaderProgram::Desc programDesc = {};
+    programDesc.pipelineType = gfx::PipelineType::Compute;
+    programDesc.slangProgram = composedProgram.get();
+
+    gProgram = device->createProgram(programDesc);
+
+    return gProgram;
+}
+
+gfx::IBufferResource* createStructuredBuffer(
+    gfx::IDevice *device,
+    float *initialData)
+{
+    // Create a structured buffer for storing computation data
+    //
+    const int numberCount = 4;
+    int structuredBufferSize = numberCount * sizeof(float);
+
+    IBufferResource::Desc bufferDesc = {};
+    bufferDesc.sizeInBytes = numberCount * sizeof(float);
+    bufferDesc.format = gfx::Format::Unknown;
+    bufferDesc.elementSize = sizeof(float);
+    bufferDesc.allowedStates = ResourceStateSet(ResourceState::ShaderResource,
+                                                ResourceState::UnorderedAccess,
+                                                ResourceState::CopyDestination,
+                                                ResourceState::CopySource);
+    bufferDesc.defaultState = ResourceState::UnorderedAccess;
+    bufferDesc.cpuAccessFlags = AccessFlag::Write | AccessFlag::Read;
+
+    SlangResult result = device->createBufferResource(bufferDesc,
+                                                      (void *)initialData,
+                                                      gBufferResource.writeRef());
+    SLANG_RETURN_NULL_ON_FAIL(result);
+    return gBufferResource;
+}
+
+gfx::IResourceView* createBufferView(
+    gfx::IDevice* device,
+    gfx::IBufferResource* buffer)
+{
+    // Create a resource view for the structured buffer
+    //
+    gfx::IResourceView::Desc viewDesc = {};
+    viewDesc.type = gfx::IResourceView::Type::UnorderedAccess;
+    viewDesc.format = gfx::Format::Unknown;
+    SLANG_RETURN_NULL_ON_FAIL(device->createBufferView(buffer, viewDesc, gResourceView.writeRef()));
+    return gResourceView;
+}
+
+gfx::ITransientResourceHeap* buildTransientHeap(gfx::IDevice *device)
+{
+    ITransientResourceHeap::Desc transientHeapDesc = {};
+    transientHeapDesc.constantBufferSize = 4096;
+    SLANG_RETURN_NULL_ON_FAIL(
+        device->createTransientResourceHeap(transientHeapDesc, gTransientHeap.writeRef()));
+    return gTransientHeap;
+}
+
+gfx::IPipelineState* buildPipelineState(
+    gfx::IDevice *device,
+    gfx::IShaderProgram* shaderProgram)
+{
+    gfx::ComputePipelineStateDesc pipelineDesc = {};
+    pipelineDesc.program = shaderProgram;
+    SLANG_RETURN_NULL_ON_FAIL(
+        device->createComputePipelineState(pipelineDesc, gPipelineState.writeRef()));
+    return gPipelineState;
+}
+
+void printInitialValues(float *initialArray, int length)
+{
+    printf("Before:\n");
+    for (int i = 0; i < length; i++)
+    {
+        printf("%f, ", initialArray[i]);
+    }
+    printf("\n");
+}
+
+void dispatchComputation(
+    gfx::IDevice* device,
+    gfx::ITransientResourceHeap* transientHeap,
+    gfx::IPipelineState* pipelineState,
+    gfx::IResourceView* bufferView,
+    unsigned int gridDimsX,
+    unsigned int gridDimsY,
+    unsigned int gridDimsZ)
+{
+    ICommandQueue::Desc queueDesc = {ICommandQueue::QueueType::Graphics};
+    gQueue = device->createCommandQueue(queueDesc);
+
+    auto commandBuffer = transientHeap->createCommandBuffer();
+    auto encoder = commandBuffer->encodeComputeCommands();
+
+    // First, obtain a root shader object from command encoder to start parameter binding.
+    auto rootObject = encoder->bindPipeline(pipelineState);
+
+    gfx::ShaderCursor entryPointCursor(
+        rootObject->getEntryPoint(0)); // get a cursor the the first entry-point.
+    // Bind buffer view to the entry point.
+    entryPointCursor.getPath("ioBuffer").setResource(bufferView);
+
+    encoder->dispatchCompute(gridDimsX, gridDimsY, gridDimsZ);
+    encoder->endEncoding();
+    commandBuffer->close();
+    gQueue->executeCommandBuffer(commandBuffer);
+    gQueue->wait();
+}
+
+bool printOutputValues(
+    gfx::IDevice *device,
+    gfx::IBufferResource *buffer,
+    int length)
+{
+    ComPtr<ISlangBlob> resultBlob;
+    SLANG_RETURN_FALSE_ON_FAIL(device->readBufferResource(
+        buffer, 0, length * sizeof(float), resultBlob.writeRef()));
+    auto result = reinterpret_cast<const float *>(resultBlob->getBufferPointer());
+    printf("After: \n");
+    for (int i = 0; i < length; i++)
+    {
+        printf("%f, ", result[i]);
+    }
+    printf("\n");
+    return true;
+}
+
+// Boilerplate functions to help the slang-generated file and types
+
+gfx_Device_0* createDevice_0()
+{
+     return (gfx_Device_0*)createDevice();
+}
+
+gfx_BufferResource_0* createStructuredBuffer_0(gfx_Device_0* _0, FixedArray<float, 4> _1)
+{
+    return (gfx_BufferResource_0*)createStructuredBuffer((gfx::IDevice*)_0, (float*)&_1);
+}
+
+gfx_ShaderProgram_0* loadShaderProgram_0(gfx_Device_0* _0)
+{
+    return (gfx_ShaderProgram_0*)loadShaderProgram((gfx::IDevice*)_0);
+}
+
+gfx_ResourceView_0* createBufferView_0(gfx_Device_0* _0, gfx_BufferResource_0* _1)
+{
+    return (gfx_ResourceView_0*)createBufferView((gfx::IDevice*)_0, (gfx::IBufferResource*)_1);
+}
+
+gfx_TransientResourceHeap_0* buildTransientHeap_0(gfx_Device_0* _0)
+{
+    return (gfx_TransientResourceHeap_0*)buildTransientHeap((gfx::IDevice*)_0);
+}
+
+gfx_PipelineState_0* buildPipelineState_0(gfx_Device_0* _0, gfx_ShaderProgram_0* _1)
+{
+    return (gfx_PipelineState_0*)buildPipelineState((gfx::IDevice*)_0, (gfx::IShaderProgram*)_1);
+}
+
+void printInitialValues_0(FixedArray<float, 4> _0, int32_t _1)
+{
+    printInitialValues((float*)&_0, _1);
+}
+
+void dispatchComputation_0(gfx_Device_0* _0, gfx_TransientResourceHeap_0* _1, gfx_PipelineState_0* _2, gfx_ResourceView_0* _3, unsigned int gridDimsX, unsigned int gridDimsY, unsigned int gridDimsZ)
+{
+    dispatchComputation(
+        (gfx::IDevice*)_0,
+        (gfx::ITransientResourceHeap*)_1,
+        (gfx::IPipelineState*)_2,
+        (gfx::IResourceView*)_3,
+        gridDimsX,
+        gridDimsY,
+        gridDimsZ);
+}
+
+RWStructuredBuffer<float> convertBuffer_0(gfx_BufferResource_0* _0) {
+    RWStructuredBuffer<float> result;
+    result.data = (float*)_0;
+    return result;
+}
+
+gfx_BufferResource_0* unconvertBuffer_0(RWStructuredBuffer<float> _0) {
+    return (gfx_BufferResource_0*)(_0.data);
+}
+
+bool printOutputValues_0(gfx_Device_0* _0, gfx_BufferResource_0* _1, int32_t _2)
+{
+    return printOutputValues((gfx::IDevice*)_0, (gfx::IBufferResource*)_1, _2);
+}
+
+int main()
+{
+    // We construct an instance of our example application
+    // `struct` type, and then walk through the lifecyle
+    // of the application.
+
+    if (!(executeComputation_0()))
+    {
+        return -1;
+    }
+}
diff --git a/examples/heterogeneous-hello-world/shader.cpp b/examples/heterogeneous-hello-world/shader.cpp
new file mode 100644
index 000000000..0c0c24ebc
--- /dev/null
+++ b/examples/heterogeneous-hello-world/shader.cpp
@@ -0,0 +1,215 @@
+#include "../../prelude/slang-cpp-prelude.h"
+
+
+#ifdef SLANG_PRELUDE_NAMESPACE
+using namespace SLANG_PRELUDE_NAMESPACE;
+#endif
+
+Vector<uint32_t, 3> operator+(Vector<uint32_t, 3> a, Vector<uint32_t, 3> b)
+{
+    Vector<uint32_t, 3> r;
+    r.x = a.x + b.x;
+    r.y = a.y + b.y;
+    r.z = a.z + b.z;
+    return r;
+}
+
+Vector<uint32_t, 3> operator*(Vector<uint32_t, 3> a, Vector<uint32_t, 3> b)
+{
+    Vector<uint32_t, 3> r;
+    r.x = a.x * b.x;
+    r.y = a.y * b.y;
+    r.z = a.z * b.z;
+    return r;
+}
+
+Vector<uint32_t, 3> make_VecU3(uint32_t a, uint32_t b, uint32_t c)
+{
+    return Vector<uint32_t, 3>{ a, b, c};
+}
+
+size_t __computeMainSize = 668;
+unsigned char __computeMain[] = {68, 88, 66, 67, 87, 111, 81, 164, 2, 29, 72, 42, 151, 28, 13, 217, 55, 37, 7, 95, 1, 
+0, 0, 0, 156, 2, 0, 0, 5, 0, 0, 0, 52, 0, 0, 0, 8, 1, 0, 0, 24, 
+1, 0, 0, 40, 1, 0, 0, 32, 2, 0, 0, 82, 68, 69, 70, 204, 0, 0, 0, 1, 
+0, 0, 0, 88, 0, 0, 0, 1, 0, 0, 0, 28, 0, 0, 0, 0, 4, 83, 67, 0, 
+9, 16, 0, 164, 0, 0, 0, 60, 0, 0, 0, 6, 0, 0, 0, 6, 0, 0, 0, 1, 
+0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 101, 
+110, 116, 114, 121, 80, 111, 105, 110, 116, 80, 97, 114, 97, 109, 115, 95, 105, 111, 66, 117, 
+102, 102, 101, 114, 95, 48, 0, 60, 0, 0, 0, 1, 0, 0, 0, 112, 0, 0, 0, 4, 
+0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 136, 0, 0, 0, 0, 0, 0, 0, 4, 
+0, 0, 0, 2, 0, 0, 0, 148, 0, 0, 0, 0, 0, 0, 0, 36, 69, 108, 101, 109, 
+101, 110, 116, 0, 171, 171, 171, 0, 0, 3, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 
+0, 0, 0, 77, 105, 99, 114, 111, 115, 111, 102, 116, 32, 40, 82, 41, 32, 72, 76, 83, 
+76, 32, 83, 104, 97, 100, 101, 114, 32, 67, 111, 109, 112, 105, 108, 101, 114, 32, 49, 48, 
+46, 49, 0, 73, 83, 71, 78, 8, 0, 0, 0, 0, 0, 0, 0, 8, 0, 0, 0, 79, 
+83, 71, 78, 8, 0, 0, 0, 0, 0, 0, 0, 8, 0, 0, 0, 83, 72, 69, 88, 240, 
+0, 0, 0, 64, 0, 5, 0, 60, 0, 0, 0, 106, 8, 0, 1, 158, 0, 0, 4, 0, 
+224, 17, 0, 0, 0, 0, 0, 4, 0, 0, 0, 95, 0, 0, 2, 18, 0, 2, 0, 104, 
+0, 0, 2, 1, 0, 0, 0, 155, 0, 0, 4, 4, 0, 0, 0, 1, 0, 0, 0, 1, 
+0, 0, 0, 167, 0, 0, 8, 18, 0, 16, 0, 0, 0, 0, 0, 10, 0, 2, 0, 1, 
+64, 0, 0, 0, 0, 0, 0, 6, 224, 17, 0, 0, 0, 0, 0, 49, 0, 0, 7, 34, 
+0, 16, 0, 0, 0, 0, 0, 10, 0, 16, 0, 0, 0, 0, 0, 1, 64, 0, 0, 0, 
+0, 0, 63, 0, 0, 0, 7, 66, 0, 16, 0, 0, 0, 0, 0, 10, 0, 16, 0, 0, 
+0, 0, 0, 10, 0, 16, 0, 0, 0, 0, 0, 75, 0, 0, 5, 18, 0, 16, 0, 0, 
+0, 0, 0, 10, 0, 16, 0, 0, 0, 0, 0, 55, 0, 0, 9, 18, 0, 16, 0, 0, 
+0, 0, 0, 26, 0, 16, 0, 0, 0, 0, 0, 42, 0, 16, 0, 0, 0, 0, 0, 10, 
+0, 16, 0, 0, 0, 0, 0, 168, 0, 0, 8, 18, 224, 17, 0, 0, 0, 0, 0, 10, 
+0, 2, 0, 1, 64, 0, 0, 0, 0, 0, 0, 10, 0, 16, 0, 0, 0, 0, 0, 62, 
+0, 0, 1, 83, 84, 65, 84, 116, 0, 0, 0, 7, 0, 0, 0, 1, 0, 0, 0, 0, 
+0, 0, 0, 1, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 
+0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 
+0, 0, 0, 0, 0, 0, 0, };
+void computeMain_wrapper(gfx_Device_0* device, Vector<uint32_t, 3> gridDims, 
+	RWStructuredBuffer<float> buffer)
+{
+	gfx_ShaderProgram_0* shaderProgram = loadShaderProgram_0(device);
+	gfx_TransientResourceHeap_0* transientHeap = buildTransientHeap_0(device);
+	gfx_PipelineState_0* pipelineState = buildPipelineState_0(device, shaderProgram);
+	gfx_ResourceView_0* bufferView = createBufferView_0(device, unconvertBuffer_0(buffer));
+	dispatchComputation_0(device, transientHeap, pipelineState, bufferView, gridDims.x, gridDims.y, gridDims.z);
+}
+
+#line 8 "../../../examples/heterogeneous-hello-world/shader.slang"
+struct EntryPointParams_0
+{
+    RWStructuredBuffer<float> ioBuffer_0;
+};
+
+
+#line 21
+struct gfx_Device_0
+{
+};
+
+
+#line 22
+struct gfx_BufferResource_0
+{
+};
+
+
+#line 23
+struct gfx_ResourceView_0
+{
+};
+
+
+#line 8
+void _computeMain(void* _S1, void* entryPointParams_0, void* _S2)
+{
+
+#line 8
+    ComputeThreadVaryingInput* _S3 = (slang_bit_cast<ComputeThreadVaryingInput*>(_S1));
+
+    uint32_t tid_0 = (*(&_S3->groupID) * make_VecU3(4U, 1U, 1U) + *(&_S3->groupThreadID)).x;
+
+    float* _S4 = &(*(&(slang_bit_cast<EntryPointParams_0*>(entryPointParams_0))->ioBuffer_0))[tid_0];
+
+#line 12
+    float i_0 = *_S4;
+    bool _S5 = i_0 < 0.50000000000000000000f;
+
+#line 13
+    float _S6 = i_0 + i_0;
+
+#line 13
+    float _S7 = (F32_sqrt((i_0)));
+
+#line 13
+    float o_0 = _S5 ? _S6 : _S7;
+
+    float* _S8 = &(*(&(slang_bit_cast<EntryPointParams_0*>(entryPointParams_0))->ioBuffer_0))[tid_0];
+
+#line 15
+    *_S8 = o_0;
+    return;
+}
+
+
+#line 31
+gfx_Device_0* createDevice_0();
+
+gfx_BufferResource_0* createStructuredBuffer_0(gfx_Device_0* _0, FixedArray<float, 4> _1);
+
+
+gfx_ResourceView_0* createBufferView_0(gfx_Device_0* _0, gfx_BufferResource_0* _1);
+
+
+#line 4
+RWStructuredBuffer<float> convertBuffer_0(gfx_BufferResource_0* _0);
+
+
+#line 44
+void printInitialValues_0(FixedArray<float, 4> _0, int32_t _1);
+
+
+#line 50
+bool printOutputValues_0(gfx_Device_0* _0, gfx_BufferResource_0* _1, int32_t _2);
+
+
+
+
+bool executeComputation_0()
+{
+
+    FixedArray<float, 4> initialArray_0 = { 3.00000000000000000000f, -20.00000000000000000000f, -6.00000000000000000000f, 8.00000000000000000000f };
+
+
+    gfx_Device_0* _S9 = createDevice_0();
+    gfx_BufferResource_0* _S10 = createStructuredBuffer_0(_S9, initialArray_0);
+    gfx_ResourceView_0* _S11 = createBufferView_0(_S9, _S10);
+    Vector<uint32_t, 3> _S12 = make_VecU3(uint32_t(int(4)), uint32_t(int(1)), uint32_t(int(1)));
+    RWStructuredBuffer<float> _S13 = convertBuffer_0(_S10);
+
+#line 64
+    computeMain_wrapper(_S9, _S12, _S13);
+
+    printInitialValues_0(initialArray_0, int(4));
+    bool _S14 = printOutputValues_0(_S9, _S10, int(4));
+
+
+    return true;
+}
+
+// [numthreads(4, 1, 1)]
+SLANG_PRELUDE_EXPORT
+void computeMain_Thread(ComputeThreadVaryingInput* varyingInput, void* entryPointParams, void* globalParams)
+{
+    _computeMain(varyingInput, entryPointParams, globalParams);
+}
+// [numthreads(4, 1, 1)]
+SLANG_PRELUDE_EXPORT
+void computeMain_Group(ComputeVaryingInput* varyingInput, void* entryPointParams, void* globalParams)
+{
+    ComputeThreadVaryingInput threadInput = {};
+    threadInput.groupID = varyingInput->startGroupID;
+    for (uint32_t x = 0; x < 4; ++x)
+    {
+        threadInput.groupThreadID.x = x;
+        _computeMain(&threadInput, entryPointParams, globalParams);
+    }
+}
+// [numthreads(4, 1, 1)]
+SLANG_PRELUDE_EXPORT
+void computeMain(ComputeVaryingInput* varyingInput, void* entryPointParams, void* globalParams)
+{
+    ComputeVaryingInput vi = *varyingInput;
+    ComputeVaryingInput groupVaryingInput = {};
+    for (uint32_t z = vi.startGroupID.z; z < vi.endGroupID.z; ++z)
+    {
+        groupVaryingInput.startGroupID.z = z;
+        for (uint32_t y = vi.startGroupID.y; y < vi.endGroupID.y; ++y)
+        {
+            groupVaryingInput.startGroupID.y = y;
+            for (uint32_t x = vi.startGroupID.x; x < vi.endGroupID.x; ++x)
+            {
+                groupVaryingInput.startGroupID.x = x;
+                computeMain_Group(&groupVaryingInput, entryPointParams, globalParams);
+            }
+        }
+    }
+}
diff --git a/examples/heterogeneous-hello-world/shader.slang b/examples/heterogeneous-hello-world/shader.slang
new file mode 100644
index 000000000..b66640e3d
--- /dev/null
+++ b/examples/heterogeneous-hello-world/shader.slang
@@ -0,0 +1,71 @@
+// shader.slang
+
+//TEST_INPUT:ubuffer(random(float, 4096, -1.0, 1.0), stride=4):name=ioBuffer
+RWStructuredBuffer<float> convertBuffer(Ptr<gfx::BufferResource> x);
+
+[shader("compute")]
+[numthreads(4, 1, 1)]
+void computeMain(uniform RWStructuredBuffer<float> ioBuffer, uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    uint tid = dispatchThreadID.x;
+
+    float i = ioBuffer[tid];
+    float o = i < 0.5 ? (i + i) : sqrt(i);
+
+    ioBuffer[tid] = o;
+}
+
+// Forward declarations of gfx types
+//
+namespace gfx {
+    struct Device{};
+    struct BufferResource{};
+    struct ResourceView{};
+    struct TransientResourceHeap{};
+    struct PipelineState{};
+    struct ShaderProgram{};
+}
+
+// Forward declarations of cpp functions
+//
+Ptr<gfx::Device> createDevice();
+Ptr<gfx::ShaderProgram> loadShaderProgram(Ptr<gfx::Device> device);
+Ptr<gfx::BufferResource> createStructuredBuffer(
+    Ptr<gfx::Device> device,
+    float[4] initialData);
+Ptr<gfx::ResourceView> createBufferView(
+    Ptr<gfx::Device> device,
+    Ptr<gfx::BufferResource> buffer);
+Ptr<gfx::TransientResourceHeap> buildTransientHeap(
+    Ptr<gfx::Device> device);
+Ptr<gfx::PipelineState> buildPipelineState(
+    Ptr<gfx::Device> device,
+    Ptr<gfx::ShaderProgram> shaderProgram);
+void printInitialValues(float[4] initialArray, int length);
+void dispatchComputation(
+    Ptr<gfx::Device> device,
+    Ptr<gfx::TransientResourceHeap> transientHeap,
+    Ptr<gfx::PipelineState> pipelineState,
+    Ptr<gfx::ResourceView> bufferView);
+bool printOutputValues(
+    Ptr<gfx::Device> device,
+    Ptr<gfx::BufferResource> buffer,
+    int length);
+
+public bool executeComputation() {
+    // We will hard-code the size of our initial array.
+    //
+    float   initialArray[4] = { 3.0f, -20.0f, -6.0f, 8.0f };
+
+    // Declare functions
+    let device = createDevice();
+    let structuredBuffer = createStructuredBuffer(device, initialArray);
+    let bufferView = createBufferView(device, structuredBuffer);
+    __GPU_FOREACH(device, uint3(4, 1, 1), LAMBDA(uint3 dispatchThreadID)
+        { computeMain(convertBuffer(structuredBuffer), dispatchThreadID) ; });
+    printInitialValues(initialArray, 4);
+    printOutputValues(device, structuredBuffer, 4);
+
+
+    return true;
+}