Support running and comparing execution results of compute shaders in testing framework.

12 files changed, 983 insertions, 185 deletions

diff --git a/source/core/text-io.h b/source/core/text-io.h
index c914e340a..e4bdc6e2d 100644
--- a/source/core/text-io.h
+++ b/source/core/text-io.h
@@ -311,6 +311,28 @@ namespace Slang
             stream = 0;
         }
 	};
+
+	inline List<String> Split(String text, char c)
+	{
+		List<String> result;
+		StringBuilder sb;
+		for (int i = 0; i < text.Length(); i++)
+		{
+			if (text[i] == c)
+			{
+				auto str = sb.ToString();
+				if (str.Length() != 0)
+					result.Add(str);
+				sb.Clear();
+			}
+			else
+				sb << text[i];
+		}
+		auto lastStr = sb.ToString();
+		if (lastStr.Length())
+			result.Add(lastStr);
+		return result;
+	}
 }
 
 #endif
diff --git a/tests/compute/simple.slang b/tests/compute/simple.slang
new file mode 100644
index 000000000..3f43c8cc7
--- /dev/null
+++ b/tests/compute/simple.slang
@@ -0,0 +1,11 @@
+//TEST(smoke,compute):COMPARE_COMPUTE:
+
+// This is a basic test for Slang compute shader.
+
+RWStructuredBuffer<float> outputBuffer;
+
+[numthreads(4, 1, 1)]
+void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID)
+{
+    outputBuffer[dispatchThreadID.x] = float(dispatchThreadID.x);
+}
\ No newline at end of file
diff --git a/tests/compute/simple.slang.actual.txt b/tests/compute/simple.slang.actual.txt
new file mode 100644
index 000000000..e8a8ee20b
--- /dev/null
+++ b/tests/compute/simple.slang.actual.txt
@@ -0,0 +1,512 @@
+0
+1
+2
+3
+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
+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
+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
+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
+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
+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
+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
+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
+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
+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
+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
+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
+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
+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
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
+0
diff --git a/tests/compute/simple.slang.expected.txt b/tests/compute/simple.slang.expected.txt
new file mode 100644
index 000000000..fdaa30664
--- /dev/null
+++ b/tests/compute/simple.slang.expected.txt
@@ -0,0 +1,4 @@
+0.0
+1.0
+2.0
+3.0
\ No newline at end of file
diff --git a/tools/render-test/main.cpp b/tools/render-test/main.cpp
index f2688d275..6907ba40f 100644
--- a/tools/render-test/main.cpp
+++ b/tools/render-test/main.cpp
@@ -49,21 +49,24 @@ static const Vertex kVertexData[kVertexCount] = {
 
 // Global variables for state to be used for rendering...
 
-uintptr_t gConstantBufferSize;
+uintptr_t gConstantBufferSize, gComputeResultBufferSize;
 
 Buffer*         gConstantBuffer;
 InputLayout*    gInputLayout;
 Buffer*         gVertexBuffer;
+Buffer*         gComputeResultBuffer;
 ShaderProgram*  gShaderProgram;
 
 // Entry point name to use for vertex/fragment shader
 static char const* vertexEntryPointName    = "vertexMain";
 static char const* fragmentEntryPointName  = "fragmentMain";
+static char const* computeEntryPointName = "computeMain";
 
 // "Profile" to use when compiling for HLSL targets
 // TODO: does this belong here?
 static char const* vertexProfileName   = "vs_4_0";
 static char const* fragmentProfileName = "ps_4_0";
+static char const* computeProfileName = "cs_4_0";
 
 Error initializeShaders(
     ShaderCompiler* shaderCompiler)
@@ -95,13 +98,21 @@ Error initializeShaders(
 
     ShaderCompileRequest compileRequest;
     compileRequest.source                   = sourceInfo;
-    compileRequest.vertexShader.source      = sourceInfo;
-    compileRequest.vertexShader.name        = vertexEntryPointName;
-    compileRequest.vertexShader.profile     = vertexProfileName;
-    compileRequest.fragmentShader.source    = sourceInfo;
-    compileRequest.fragmentShader.name      = fragmentEntryPointName;
-    compileRequest.fragmentShader.profile   = fragmentProfileName;
-
+	if (gOptions.shaderType == ShaderProgramType::Graphics)
+	{
+		compileRequest.vertexShader.source = sourceInfo;
+		compileRequest.vertexShader.name = vertexEntryPointName;
+		compileRequest.vertexShader.profile = vertexProfileName;
+		compileRequest.fragmentShader.source = sourceInfo;
+		compileRequest.fragmentShader.name = fragmentEntryPointName;
+		compileRequest.fragmentShader.profile = fragmentProfileName;
+	}
+	else
+	{
+		compileRequest.computeShader.source = sourceInfo;
+		compileRequest.computeShader.name = computeEntryPointName;
+		compileRequest.computeShader.profile = computeProfileName;
+	}
     gShaderProgram = shaderCompiler->compileProgram(compileRequest);
     if( !gShaderProgram )
     {
@@ -111,6 +122,15 @@ Error initializeShaders(
     return Error::None;
 }
 
+void outputComputeResult(Renderer* renderer, const char * fileName)
+{
+	float* data = (float*)renderer->map(gComputeResultBuffer, MapFlavor::HostRead);
+	FILE* f = fopen(fileName, "wt");
+	for (auto i = 0u; i < gComputeResultBufferSize / sizeof(UInt); i++)
+		fprintf(f, "%.9g\n", data[i]);
+	fclose(f);
+}
+
 //
 // At initialization time, we are going to load and compile our Slang shader
 // code, and then create the D3D11 API objects we need for rendering.
@@ -138,6 +158,19 @@ Error initializeInner(
     if(!gConstantBuffer)
         return Error::Unexpected;
 
+	gComputeResultBufferSize = 512 * sizeof(float);
+	BufferDesc computeResultBufferDesc;
+	computeResultBufferDesc.size = gComputeResultBufferSize;
+	computeResultBufferDesc.flavor = BufferFlavor::Storage;
+	gComputeResultBuffer = renderer->createBuffer(computeResultBufferDesc);
+	if (!gComputeResultBufferSize)
+		return Error::Unexpected;
+	// initialize buffer to 0
+	char * ptr = (char*)renderer->map(gComputeResultBuffer, MapFlavor::HostWrite);
+	for (auto i = 0u; i < gComputeResultBufferSize; i++)
+		ptr[i] = 0;
+	renderer->unmap(gComputeResultBuffer);
+
     // Input Assembler (IA)
 
     InputElementDesc inputElements[] = {
@@ -195,6 +228,13 @@ void renderFrameInner(
     renderer->draw(3);
 }
 
+void runCompute(Renderer * renderer)
+{
+	renderer->setShaderProgram(gShaderProgram);
+	renderer->setStorageBuffer(0, gComputeResultBuffer);
+	renderer->dispatchCompute(1, 1, 1);
+}
+
 void finalize()
 {
 }
@@ -352,44 +392,52 @@ int main(
 
     // Once initialization is all complete, we show the window...
     ShowWindow(windowHandle, showCommand);
-
-    // ... and enter the event loop:
-    for(;;)
-    {
-        MSG message;
-
-        int result = PeekMessageW(&message, NULL, 0, 0, PM_REMOVE);
-        if (result != 0)
-        {
-            if (message.message == WM_QUIT)
-            {
-                return (int)message.wParam;
-            }
-
-            TranslateMessage(&message);
-            DispatchMessageW(&message);
-        }
-        else
-        {
-            // Whenver we don't have Windows events to process,
-            // we render a frame.
-
-            static const float kClearColor[] = { 0.25, 0.25, 0.25, 1.0 };
-            renderer->setClearColor(kClearColor);
-            renderer->clearFrame();
-
-            renderFrameInner(renderer);
-
-            // If we are in a mode where output is requested, we need to snapshot the back buffer here
-            if( gOptions.outputPath )
-            {
-                renderer->captureScreenShot(gOptions.outputPath);
-                return 0;
-            }
-
-            renderer->presentFrame();
-        }
-    }
+	
+	// ... and enter the event loop:
+	for (;;)
+	{
+		MSG message;
+
+		int result = PeekMessageW(&message, NULL, 0, 0, PM_REMOVE);
+		if (result != 0)
+		{
+			if (message.message == WM_QUIT)
+			{
+				return (int)message.wParam;
+			}
+
+			TranslateMessage(&message);
+			DispatchMessageW(&message);
+		}
+		else
+		{
+			// Whenver we don't have Windows events to process,
+			// we render a frame.
+			static const float kClearColor[] = { 0.25, 0.25, 0.25, 1.0 };
+			renderer->setClearColor(kClearColor);
+			renderer->clearFrame();
+
+			if (gOptions.shaderType == ShaderProgramType::Compute)
+			{
+				runCompute(renderer);
+			}
+			else
+			{
+				renderFrameInner(renderer);
+			}
+			// If we are in a mode where output is requested, we need to snapshot the back buffer here
+			if (gOptions.outputPath)
+			{
+				if (gOptions.shaderType == ShaderProgramType::Compute)
+					outputComputeResult(renderer, gOptions.outputPath);
+				else
+					renderer->captureScreenShot(gOptions.outputPath);
+				return 0;
+			}
+
+			renderer->presentFrame();
+		}
+	}
 
     return 0;
 }
diff --git a/tools/render-test/options.cpp b/tools/render-test/options.cpp
index 53f88b7a9..89a68b27f 100644
--- a/tools/render-test/options.cpp
+++ b/tools/render-test/options.cpp
@@ -92,6 +92,14 @@ void parseOptions(int* argc, char** argv)
             }
             gOptions.slangArgs[gOptions.slangArgCount++] = *argCursor++;
         }
+		else if (strcmp(arg, "-compute") == 0)
+		{
+			gOptions.shaderType = ShaderProgramType::Compute;
+		}
+		else if (strcmp(arg, "-graphics") == 0)
+		{
+			gOptions.shaderType = ShaderProgramType::Graphics;
+		}
         else
         {
             fprintf(stderr, "unknown option '%s'\n", arg);
diff --git a/tools/render-test/options.h b/tools/render-test/options.h
index fbc615d56..a36e0a809 100644
--- a/tools/render-test/options.h
+++ b/tools/render-test/options.h
@@ -31,11 +31,18 @@ enum
     kMaxSlangArgs = 16,
 };
 
+enum class ShaderProgramType
+{
+	Graphics,
+	Compute
+};
+
 struct Options
 {
     char const* appName = "render-test";
     char const* sourcePath = nullptr;
     char const* outputPath = nullptr;
+	ShaderProgramType shaderType = ShaderProgramType::Graphics;
     Mode mode = Mode::Slang;
 
     char const* slangArgs[kMaxSlangArgs];
diff --git a/tools/render-test/render-d3d11.cpp b/tools/render-test/render-d3d11.cpp
index bc86809e6..6e9e04a78 100644
--- a/tools/render-test/render-d3d11.cpp
+++ b/tools/render-test/render-d3d11.cpp
@@ -1,4 +1,4 @@
-// render-d3d11.cpp
+// render-d3d11.cpp
 #include "render-d3d11.h"
 
 #include "options.h"
@@ -418,6 +418,12 @@ public:
         return this;
     }
 
+	struct D3DBuffer
+	{
+		ID3D11UnorderedAccessView * view = nullptr;
+		ID3D11Buffer * buffer = nullptr;
+	};
+
     virtual Buffer* createBuffer(BufferDesc const& desc) override
     {
         D3D11_BUFFER_DESC dxBufferDesc = { 0 };
@@ -437,6 +443,14 @@ public:
             dxBufferDesc.CPUAccessFlags = 0;
             break;
 
+		case BufferFlavor::Storage:
+			dxBufferDesc.Usage = D3D11_USAGE_DEFAULT;
+			dxBufferDesc.BindFlags = D3D11_BIND_UNORDERED_ACCESS;
+			dxBufferDesc.CPUAccessFlags = D3D11_CPU_ACCESS_READ | D3D11_CPU_ACCESS_WRITE;
+			dxBufferDesc.StructureByteStride = sizeof(float);
+			dxBufferDesc.MiscFlags = D3D11_RESOURCE_MISC_BUFFER_STRUCTURED;
+			break;
+
         default:
             return nullptr;
         }
@@ -452,7 +466,20 @@ public:
             &dxBuffer);
         if(FAILED(hr)) return nullptr;
 
-        return (Buffer*) dxBuffer;
+		D3DBuffer * rs = new D3DBuffer();
+		rs->buffer = dxBuffer;
+		if (desc.flavor == BufferFlavor::Storage)
+		{
+			D3D11_UNORDERED_ACCESS_VIEW_DESC viewDesc;
+			memset(&viewDesc, 0, sizeof(viewDesc));
+			viewDesc.Buffer.FirstElement = 0;
+			viewDesc.Buffer.NumElements = 512;
+			viewDesc.Buffer.Flags = 0;
+			viewDesc.ViewDimension = D3D11_UAV_DIMENSION_BUFFER;
+			viewDesc.Format = DXGI_FORMAT_UNKNOWN;
+			dxDevice->CreateUnorderedAccessView(dxBuffer, &viewDesc, &rs->view);
+		}
+        return (Buffer*) rs;
     }
 
     static DXGI_FORMAT mapFormat(Format format)
@@ -535,7 +562,7 @@ public:
     {
         auto dxContext = dxImmediateContext;
 
-        auto dxBuffer = (ID3D11Buffer*) buffer;
+        auto dxBuffer = ((D3DBuffer*)buffer)->buffer;
 
         D3D11_MAP dxMapFlavor;
         switch( flavor )
@@ -543,7 +570,12 @@ public:
         case MapFlavor::WriteDiscard:
             dxMapFlavor = D3D11_MAP_WRITE_DISCARD;
             break;
-
+		case MapFlavor::HostWrite:
+			dxMapFlavor = D3D11_MAP_WRITE;
+			break;
+		case MapFlavor::HostRead:
+			dxMapFlavor = D3D11_MAP_READ;
+			break;
         default:
             return nullptr;
         }
@@ -563,7 +595,7 @@ public:
     {
         auto dxContext = dxImmediateContext;
 
-        auto dxBuffer = (ID3D11Buffer*) buffer;
+        auto dxBuffer = ((D3DBuffer*)buffer)->buffer;
 
         dxContext->Unmap(dxBuffer, 0);
     }
@@ -609,11 +641,11 @@ public:
             dxVertexOffsets[ii] = (UINT) offsets[ii];
         }
 
-        auto dxVertexBuffers = (ID3D11Buffer* const*) buffers;
+        auto dxVertexBuffers = (D3DBuffer* const*) buffers;
 
         dxContext->IASetVertexBuffers(
             (UINT) startSlot,
-            (UINT) slotCount, &dxVertexBuffers[0], &dxVertexStrides[0], &dxVertexOffsets[0]);
+            (UINT) slotCount, &(dxVertexBuffers[0])->buffer, &dxVertexStrides[0], &dxVertexOffsets[0]);
     }
 
     virtual void setShaderProgram(ShaderProgram* inProgram) override
@@ -621,7 +653,7 @@ public:
         auto dxContext = dxImmediateContext;
         
         auto program = (D3D11ShaderProgram*) inProgram;
-
+		dxContext->CSSetShader(program->dxComputeShader, NULL, 0);
         dxContext->VSSetShader(program->dxVertexShader, NULL, 0);
         dxContext->PSSetShader(program->dxPixelShader,  NULL, 0);
     }
@@ -632,14 +664,24 @@ public:
 
         // TODO: actually use those offsets
 
-        auto dxConstantBuffers = (ID3D11Buffer* const*) buffers;
-
+        auto dxConstantBuffers = (D3DBuffer* const*) buffers;
         dxContext->VSSetConstantBuffers(
-            (UINT) startSlot, (UINT) slotCount, &dxConstantBuffers[0]);
+            (UINT) startSlot, (UINT) slotCount, &dxConstantBuffers[0]->buffer);
         dxContext->VSSetConstantBuffers(
-            (UINT) startSlot, (UINT) slotCount, &dxConstantBuffers[0]);
+            (UINT) startSlot, (UINT) slotCount, &dxConstantBuffers[0]->buffer);
     }
 
+	virtual void setStorageBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) override
+	{
+		auto dxContext = dxImmediateContext;
+
+		// TODO: actually use those offsets
+
+		auto dxStorageBuffers = (D3DBuffer* const*)buffers;
+		dxContext->CSSetUnorderedAccessViews(
+			(UINT)startSlot, (UINT)slotCount, &dxStorageBuffers[0]->view, 0);
+	}
+
 
     virtual void draw(UInt vertexCount, UInt startVertex) override
     {
@@ -653,35 +695,62 @@ public:
 
     struct D3D11ShaderProgram
     {
-        ID3D11VertexShader* dxVertexShader;
-        ID3D11PixelShader*  dxPixelShader;
+        ID3D11VertexShader* dxVertexShader = nullptr;
+        ID3D11PixelShader*  dxPixelShader = nullptr;
+		ID3D11ComputeShader* dxComputeShader = nullptr;
     };
 
     virtual ShaderProgram* compileProgram(ShaderCompileRequest const& request) override
     {
-        auto dxVertexShaderBlob     = compileHLSLShader(request.vertexShader.source.path, request.vertexShader.source.text, request.vertexShader    .name,  request.vertexShader    .profile);
-        if(!dxVertexShaderBlob)     return nullptr;
+		if (request.computeShader.name)
+		{
+			auto dxComputeShaderBlob = compileHLSLShader(request.computeShader.source.path, request.computeShader.source.text, request.computeShader.name, request.computeShader.profile);
+			if (!dxComputeShaderBlob)     return nullptr;
+
+			ID3D11ComputeShader* dxComputeShader;
+
+			HRESULT csResult = dxDevice->CreateComputeShader(dxComputeShaderBlob->GetBufferPointer(), dxComputeShaderBlob->GetBufferSize(), nullptr, &dxComputeShader);
+
+			dxComputeShaderBlob->Release();
 
-        auto dxFragmentShaderBlob   = compileHLSLShader(request.fragmentShader.source.path, request.fragmentShader.source.text, request.fragmentShader  .name,  request.fragmentShader  .profile);
-        if(!dxFragmentShaderBlob)   return nullptr;
+			if (FAILED(csResult)) return nullptr;
 
-        ID3D11VertexShader* dxVertexShader;
-        ID3D11PixelShader*  dxPixelShader;
+			D3D11ShaderProgram* shaderProgram = new D3D11ShaderProgram();
+			shaderProgram->dxComputeShader = dxComputeShader;
+			return (ShaderProgram*)shaderProgram;
+		}
+		else
+		{
+			auto dxVertexShaderBlob = compileHLSLShader(request.vertexShader.source.path, request.vertexShader.source.text, request.vertexShader.name, request.vertexShader.profile);
+			if (!dxVertexShaderBlob)     return nullptr;
 
-        HRESULT vsResult = dxDevice->CreateVertexShader(  dxVertexShaderBlob  ->GetBufferPointer(),   dxVertexShaderBlob  ->GetBufferSize(), nullptr, &dxVertexShader);
-        HRESULT psResult = dxDevice->CreatePixelShader(   dxFragmentShaderBlob->GetBufferPointer(),   dxFragmentShaderBlob->GetBufferSize(), nullptr, &dxPixelShader);
+			auto dxFragmentShaderBlob = compileHLSLShader(request.fragmentShader.source.path, request.fragmentShader.source.text, request.fragmentShader.name, request.fragmentShader.profile);
+			if (!dxFragmentShaderBlob)   return nullptr;
 
-        dxVertexShaderBlob  ->Release();
-        dxFragmentShaderBlob->Release();
+			ID3D11VertexShader* dxVertexShader;
+			ID3D11PixelShader*  dxPixelShader;
 
-        if(FAILED(vsResult)) return nullptr;
-        if(FAILED(psResult)) return nullptr;
+			HRESULT vsResult = dxDevice->CreateVertexShader(dxVertexShaderBlob->GetBufferPointer(), dxVertexShaderBlob->GetBufferSize(), nullptr, &dxVertexShader);
+			HRESULT psResult = dxDevice->CreatePixelShader(dxFragmentShaderBlob->GetBufferPointer(), dxFragmentShaderBlob->GetBufferSize(), nullptr, &dxPixelShader);
 
-        D3D11ShaderProgram* shaderProgram = new D3D11ShaderProgram();
-        shaderProgram->dxVertexShader   = dxVertexShader;
-        shaderProgram->dxPixelShader    = dxPixelShader;
-        return (ShaderProgram*) shaderProgram;
+			dxVertexShaderBlob->Release();
+			dxFragmentShaderBlob->Release();
+
+			if (FAILED(vsResult)) return nullptr;
+			if (FAILED(psResult)) return nullptr;
+
+			D3D11ShaderProgram* shaderProgram = new D3D11ShaderProgram();
+			shaderProgram->dxVertexShader = dxVertexShader;
+			shaderProgram->dxPixelShader = dxPixelShader;
+			return (ShaderProgram*)shaderProgram;
+		}
     }
+
+	virtual void dispatchCompute(int x, int y, int z) override
+	{
+		auto dxContext = dxImmediateContext;
+		dxContext->Dispatch(x, y, z);
+	}
 };
 
 
diff --git a/tools/render-test/render-gl.cpp b/tools/render-test/render-gl.cpp
index f02ac36c4..55647fb25 100644
--- a/tools/render-test/render-gl.cpp
+++ b/tools/render-test/render-gl.cpp
@@ -58,6 +58,7 @@
     F(glEnableVertexAttribArray,    PFNGLENABLEVERTEXATTRIBARRAYPROC)   \
     F(glDisableVertexAttribArray,   PFNGLDISABLEVERTEXATTRIBARRAYPROC)  \
     F(glDebugMessageCallback,       PFNGLDEBUGMESSAGECALLBACKPROC)      \
+	F(glDispatchCompute,            PFNGLDISPATCHCOMPUTEPROC) \
     /* end */
 
 namespace renderer_test {
@@ -311,8 +312,12 @@ public:
         switch (flavor)
         {
         case MapFlavor::WriteDiscard:
+		case MapFlavor::HostWrite:
             access = GL_WRITE_ONLY;
             break;
+		case MapFlavor::HostRead:
+			access = GL_READ_ONLY;
+			break;
         }
 
         auto bufferID = (GLuint)(uintptr_t)buffer;
@@ -377,21 +382,32 @@ public:
         glUseProgram(programID);
     }
 
-    virtual void setConstantBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) override
-    {
-        for (UInt ii = 0; ii < slotCount; ++ii)
-        {
-            UInt slot = startSlot + ii;
+	void bindBufferImpl(int target, UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets)
+	{
+		for (UInt ii = 0; ii < slotCount; ++ii)
+		{
+			UInt slot = startSlot + ii;
 
-            Buffer* buffer = buffers[ii];
-            GLuint bufferID = (GLuint)(uintptr_t)buffer;
+			Buffer* buffer = buffers[ii];
+			GLuint bufferID = (GLuint)(uintptr_t)buffer;
 
-            assert(!offsets || !offsets[ii]);
+			assert(!offsets || !offsets[ii]);
 
-            glBindBufferBase(GL_UNIFORM_BUFFER, (GLuint) slot, bufferID);
-        }
+			glBindBufferBase(target, (GLuint)slot, bufferID);
+		}
+	}
+
+    virtual void setConstantBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) override
+    {
+		bindBufferImpl(GL_UNIFORM_BUFFER, startSlot, slotCount, buffers, offsets);
     }
 
+
+	virtual void setStorageBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) override
+	{
+		bindBufferImpl(GL_SHADER_STORAGE_BUFFER, startSlot, slotCount, buffers, offsets);
+	}
+
     void flushStateForDraw()
     {
         auto layout = this->boundInputLayout;
@@ -432,18 +448,31 @@ public:
     virtual ShaderProgram* compileProgram(ShaderCompileRequest const& request) override
     {
         auto programID = glCreateProgram();
+		if (request.computeShader.name)
+		{
+			auto computeShaderID = loadShader(GL_COMPUTE_SHADER, request.computeShader.source.text);
 
-        auto vertexShaderID   = loadShader(GL_VERTEX_SHADER,   request.vertexShader  .source.text);
-        auto fragmentShaderID = loadShader(GL_FRAGMENT_SHADER, request.fragmentShader.source.text);
+			glAttachShader(programID, computeShaderID);
 
-        glAttachShader(programID, vertexShaderID);
-        glAttachShader(programID, fragmentShaderID);
 
-        glLinkProgram(programID);
+			glLinkProgram(programID);
 
-        glDeleteShader(vertexShaderID);
-        glDeleteShader(fragmentShaderID);
+			glDeleteShader(computeShaderID);
+		}
+		else
+		{
+			auto vertexShaderID = loadShader(GL_VERTEX_SHADER, request.vertexShader.source.text);
+			auto fragmentShaderID = loadShader(GL_FRAGMENT_SHADER, request.fragmentShader.source.text);
 
+			glAttachShader(programID, vertexShaderID);
+			glAttachShader(programID, fragmentShaderID);
+
+
+			glLinkProgram(programID);
+
+			glDeleteShader(vertexShaderID);
+			glDeleteShader(fragmentShaderID);
+		}
         GLint success = GL_FALSE;
         glGetProgramiv(programID, GL_LINK_STATUS, &success);
         if( !success )
@@ -579,6 +608,11 @@ public:
 
         return shaderID;
     }
+
+	virtual void dispatchCompute(int x, int y, int z) override
+	{
+		glDispatchCompute(x, y, z);
+	}
 };
 
 
diff --git a/tools/render-test/render.h b/tools/render-test/render.h
index afa279c66..7f8f2fffa 100644
--- a/tools/render-test/render.h
+++ b/tools/render-test/render.h
@@ -20,8 +20,8 @@ struct ShaderCompileRequest
 
     struct EntryPoint
     {
-        char const* name;
-        char const* profile;
+        char const* name = nullptr;
+        char const* profile = nullptr;
 
         SourceInfo  source;
     };
@@ -29,6 +29,7 @@ struct ShaderCompileRequest
     SourceInfo source;
     EntryPoint vertexShader;
     EntryPoint fragmentShader;
+	EntryPoint computeShader;
 };
 
 class ShaderCompiler
@@ -47,6 +48,7 @@ enum class BufferFlavor
 {
     Constant,
     Vertex,
+	Storage,
 };
 
 struct BufferDesc
@@ -66,6 +68,8 @@ struct InputElementDesc
 
 enum class MapFlavor
 {
+	HostRead,
+	HostWrite,
     WriteDiscard,
 };
 
@@ -108,13 +112,17 @@ public:
     virtual void setShaderProgram(ShaderProgram* program) = 0;
 
     virtual void setConstantBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) = 0;
-
+	virtual void setStorageBuffers(UInt startSlot, UInt slotCount, Buffer* const* buffers, UInt const* offsets) = 0;
     inline void setConstantBuffer(UInt slot, Buffer* buffer, UInt offset = 0)
     {
         setConstantBuffers(slot, 1, &buffer, &offset);
     }
-
+	inline void setStorageBuffer(UInt slot, Buffer* buffer, UInt offset = 0)
+	{
+		setStorageBuffers(slot, 1, &buffer, &offset);
+	}
     virtual void draw(UInt vertexCount, UInt startVertex = 0) = 0;
+	virtual void dispatchCompute(int x, int y, int z) = 0;
 };
 
 } // renderer_test
diff --git a/tools/render-test/slang-support.cpp b/tools/render-test/slang-support.cpp
index bf8b7b9c7..63e24126c 100644
--- a/tools/render-test/slang-support.cpp
+++ b/tools/render-test/slang-support.cpp
@@ -13,94 +13,116 @@ struct SlangShaderCompilerWrapper : public ShaderCompiler
     SlangCompileTarget  target;
     SlangSourceLanguage sourceLanguage;
 
-    virtual ShaderProgram* compileProgram(ShaderCompileRequest const& request) override
-    {
-        SlangSession* slangSession = spCreateSession(NULL);
-        SlangCompileRequest* slangRequest = spCreateCompileRequest(slangSession);
-
-        spSetCodeGenTarget(slangRequest, target);
-
-        // Define a macro so that shader code in a test can detect what language we
-        // are nominally working with.
-        char const* langDefine = nullptr;
-        switch (sourceLanguage)
-        {
-        case SLANG_SOURCE_LANGUAGE_GLSL:    langDefine = "__GLSL__";    break;
-        case SLANG_SOURCE_LANGUAGE_HLSL:    langDefine = "__HLSL__";    break;
-        case SLANG_SOURCE_LANGUAGE_SLANG:   langDefine = "__SLANG__";   break;
-        default:
-            assert(!"unexpected");
-            break;
-        }
-        spAddPreprocessorDefine(slangRequest, langDefine, "1");
-
-        spProcessCommandLineArguments(slangRequest, &gOptions.slangArgs[0], gOptions.slangArgCount);
-
-        int vertexTranslationUnit = 0;
-        int fragmentTranslationUnit = 0;
-        char const* vertexEntryPointName = request.vertexShader.name;
-        char const* fragmentEntryPointName = request.fragmentShader.name;
-        if( sourceLanguage == SLANG_SOURCE_LANGUAGE_GLSL )
-        {
-            // GLSL presents unique challenges because, frankly, it got the whole
-            // compilation model wrong. One aspect of working around this is that
-            // we will compile the same source file multiple times: once per
-            // entry point, and we will have different preprocessor definitions
-            // active in each case.
-
-            vertexTranslationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
-            spAddTranslationUnitSourceString(slangRequest, vertexTranslationUnit, request.source.path, request.source.text);
-            spTranslationUnit_addPreprocessorDefine(slangRequest, vertexTranslationUnit, "__GLSL_VERTEX__", "1");
-            vertexEntryPointName = "main";
-
-            fragmentTranslationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
-            spAddTranslationUnitSourceString(slangRequest, fragmentTranslationUnit, request.source.path, request.source.text);
-            spTranslationUnit_addPreprocessorDefine(slangRequest, fragmentTranslationUnit, "__GLSL_FRAGMENT__", "1");
-            fragmentEntryPointName = "main";
-        }
-        else
-        {
-            int translationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
-            spAddTranslationUnitSourceString(slangRequest, translationUnit, request.source.path, request.source.text);
-
-            vertexTranslationUnit = translationUnit;
-            fragmentTranslationUnit = translationUnit;
-        }
-
-
-        // If we aren't dealing with true Slang input, then don't enable checking.
-        if (sourceLanguage != SLANG_SOURCE_LANGUAGE_SLANG)
-        {
-            spSetCompileFlags(slangRequest, SLANG_COMPILE_FLAG_NO_CHECKING);
-        }
-
-        int vertexEntryPoint = spAddEntryPoint(slangRequest, vertexTranslationUnit, vertexEntryPointName,   spFindProfile(slangSession, request.vertexShader.profile));
-        int fragmentEntryPoint = spAddEntryPoint(slangRequest, fragmentTranslationUnit, fragmentEntryPointName, spFindProfile(slangSession, request.fragmentShader.profile));
-
-        int compileErr = spCompile(slangRequest);
-        if(auto diagnostics = spGetDiagnosticOutput(slangRequest))
-        {
-            // TODO(tfoley): re-enable when I get a logging solution in place
-//            OutputDebugStringA(diagnostics);
-            fprintf(stderr, "%s", diagnostics);
-        }
-        if(compileErr)
-        {
-            return nullptr;
-        }
-
-
-        ShaderCompileRequest innerRequest = request;
-
-        char const* vertexCode = spGetEntryPointSource(slangRequest, vertexEntryPoint);
-        char const* fragmentCode = spGetEntryPointSource(slangRequest, fragmentEntryPoint);
-
-        innerRequest.vertexShader.source.text = vertexCode;
-        innerRequest.fragmentShader.source.text = fragmentCode;
-
-
-        auto result = innerCompiler->compileProgram(innerRequest);
-
+	virtual ShaderProgram* compileProgram(ShaderCompileRequest const& request) override
+	{
+		SlangSession* slangSession = spCreateSession(NULL);
+		SlangCompileRequest* slangRequest = spCreateCompileRequest(slangSession);
+
+		spSetCodeGenTarget(slangRequest, target);
+
+		// Define a macro so that shader code in a test can detect what language we
+		// are nominally working with.
+		char const* langDefine = nullptr;
+		switch (sourceLanguage)
+		{
+		case SLANG_SOURCE_LANGUAGE_GLSL:    langDefine = "__GLSL__";    break;
+		case SLANG_SOURCE_LANGUAGE_HLSL:    langDefine = "__HLSL__";    break;
+		case SLANG_SOURCE_LANGUAGE_SLANG:   langDefine = "__SLANG__";   break;
+		default:
+			assert(!"unexpected");
+			break;
+		}
+		spAddPreprocessorDefine(slangRequest, langDefine, "1");
+
+		spProcessCommandLineArguments(slangRequest, &gOptions.slangArgs[0], gOptions.slangArgCount);
+		int computeTranslationUnit = 0;
+		int vertexTranslationUnit = 0;
+		int fragmentTranslationUnit = 0;
+		char const* vertexEntryPointName = request.vertexShader.name;
+		char const* fragmentEntryPointName = request.fragmentShader.name;
+		char const* computeEntryPointName = request.computeShader.name;
+
+		if (sourceLanguage == SLANG_SOURCE_LANGUAGE_GLSL)
+		{
+			// GLSL presents unique challenges because, frankly, it got the whole
+			// compilation model wrong. One aspect of working around this is that
+			// we will compile the same source file multiple times: once per
+			// entry point, and we will have different preprocessor definitions
+			// active in each case.
+
+			vertexTranslationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
+			spAddTranslationUnitSourceString(slangRequest, vertexTranslationUnit, request.source.path, request.source.text);
+			spTranslationUnit_addPreprocessorDefine(slangRequest, vertexTranslationUnit, "__GLSL_VERTEX__", "1");
+			vertexEntryPointName = "main";
+
+			fragmentTranslationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
+			spAddTranslationUnitSourceString(slangRequest, fragmentTranslationUnit, request.source.path, request.source.text);
+			spTranslationUnit_addPreprocessorDefine(slangRequest, fragmentTranslationUnit, "__GLSL_FRAGMENT__", "1");
+			fragmentEntryPointName = "main";
+
+			computeTranslationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
+			spAddTranslationUnitSourceString(slangRequest, computeTranslationUnit, request.source.path, request.source.text);
+			spTranslationUnit_addPreprocessorDefine(slangRequest, computeTranslationUnit, "__GLSL_COMPUTE__", "1");
+			computeEntryPointName = "main";
+		}
+		else
+		{
+			int translationUnit = spAddTranslationUnit(slangRequest, sourceLanguage, nullptr);
+			spAddTranslationUnitSourceString(slangRequest, translationUnit, request.source.path, request.source.text);
+
+			vertexTranslationUnit = translationUnit;
+			fragmentTranslationUnit = translationUnit;
+			computeTranslationUnit = translationUnit;
+		}
+
+
+		// If we aren't dealing with true Slang input, then don't enable checking.
+		if (sourceLanguage != SLANG_SOURCE_LANGUAGE_SLANG)
+		{
+			spSetCompileFlags(slangRequest, SLANG_COMPILE_FLAG_NO_CHECKING);
+		}
+		ShaderProgram * result = nullptr;
+		if (request.computeShader.name)
+		{
+			int computeEntryPoint = spAddEntryPoint(slangRequest, computeTranslationUnit, computeEntryPointName, spFindProfile(slangSession, request.computeShader.profile));
+			int compileErr = spCompile(slangRequest);
+			if (auto diagnostics = spGetDiagnosticOutput(slangRequest))
+			{
+				fprintf(stderr, "%s", diagnostics);
+			}
+			if (!compileErr)
+			{
+				ShaderCompileRequest innerRequest = request;
+				char const* computeCode = spGetEntryPointSource(slangRequest, computeEntryPoint);
+				innerRequest.computeShader.source.text = computeCode;
+				result = innerCompiler->compileProgram(innerRequest);
+			}
+		}
+		else
+		{
+			int vertexEntryPoint = spAddEntryPoint(slangRequest, vertexTranslationUnit, vertexEntryPointName, spFindProfile(slangSession, request.vertexShader.profile));
+			int fragmentEntryPoint = spAddEntryPoint(slangRequest, fragmentTranslationUnit, fragmentEntryPointName, spFindProfile(slangSession, request.fragmentShader.profile));
+
+			int compileErr = spCompile(slangRequest);
+			if (auto diagnostics = spGetDiagnosticOutput(slangRequest))
+			{
+				// TODO(tfoley): re-enable when I get a logging solution in place
+	//            OutputDebugStringA(diagnostics);
+				fprintf(stderr, "%s", diagnostics);
+			}
+			if (!compileErr)
+			{
+				ShaderCompileRequest innerRequest = request;
+
+				char const* vertexCode = spGetEntryPointSource(slangRequest, vertexEntryPoint);
+				char const* fragmentCode = spGetEntryPointSource(slangRequest, fragmentEntryPoint);
+
+				innerRequest.vertexShader.source.text = vertexCode;
+				innerRequest.fragmentShader.source.text = fragmentCode;
+
+				result = innerCompiler->compileProgram(innerRequest);
+			}
+		}
         // We clean up the Slang compilation context and result *after*
         // we have run the downstream compiler, because Slang
         // owns the memory allocation for the generated text, and will
diff --git a/tools/slang-test/main.cpp b/tools/slang-test/main.cpp
index 5c0c81392..fb10d46f6 100644
--- a/tools/slang-test/main.cpp
+++ b/tools/slang-test/main.cpp
@@ -1110,7 +1110,55 @@ TestResult runGLSLComparisonTest(TestInput& input)
     return kTestResult_Pass;
 }
 
+TestResult doComputeComparisonTestRunImpl(TestInput& input, const char * langOption, String referenceOutput)
+{
+	// TODO: delete any existing files at the output path(s) to avoid stale outputs leading to a false pass
+	auto filePath999 = input.filePath;
+	auto outputStem = input.outputStem;
+
+	OSProcessSpawner spawner;
+
+	spawner.pushExecutablePath(String(options.binDir) + "render-test" + osGetExecutableSuffix());
+	spawner.pushArgument(filePath999);
+
+	for (auto arg : input.testOptions->args)
+	{
+		spawner.pushArgument(arg);
+	}
+
+	spawner.pushArgument(langOption);
+	spawner.pushArgument("-o");
+	spawner.pushArgument(outputStem + ".actual.txt");
+
+	if (spawnAndWait(outputStem, spawner) != kOSError_None)
+	{
+		return kTestResult_Fail;
+	}
+
+	// check against reference output
+	if (!File::Exists(outputStem + ".actual.txt"))
+		return kTestResult_Fail;
+	if (!File::Exists(referenceOutput))
+		return kTestResult_Fail;
+
+	auto actualProgramOutput = Split(File::ReadAllText(outputStem + ".actual.txt"), '\n');
+	auto referenceProgramOutput = Split(File::ReadAllText(referenceOutput), '\n');
+	if (actualProgramOutput.Count() < referenceProgramOutput.Count())
+		return kTestResult_Fail;
+	for (int i = 0; i < referenceProgramOutput.Count(); i++)
+	{
+		auto reference = StringToFloat(referenceProgramOutput[i]);
+		auto actual = StringToFloat(actualProgramOutput[i]);
+		if (abs(actual - reference) > 1e-7f)
+			return kTestResult_Fail;
+	}
+	return kTestResult_Pass;
+}
 
+TestResult doSlangComputeComparisonTest(TestInput& input)
+{
+	return doComputeComparisonTestRunImpl(input, "-slang -compute", input.outputStem + ".expected.txt");
+}
 
 TestResult doRenderComparisonTestRun(TestInput& input, char const* langOption, char const* outputKind, String* outOutput)
 {
@@ -1317,6 +1365,8 @@ TestResult runTest(
         { "COMPARE_HLSL_GLSL_RENDER",           &skipTest },
 #endif
         { "COMPARE_GLSL", &runGLSLComparisonTest },
+		{ "COMPARE_COMPUTE", &doSlangComputeComparisonTest},
+
         { "CROSS_COMPILE", &runCrossCompilerTest },
         { "EVAL", &runEvalTest },
         { nullptr, nullptr },
@@ -1621,6 +1671,9 @@ int main(
 
     auto renderTestCategory = addTestCategory("render", fullTestCategory);
 
+	auto computeTestCategory = addTestCategory("compute", fullTestCategory);
+
+
     // An un-categorized test will always belong to the `full` category
     defaultTestCategory = fullTestCategory;