NVAPI improvements (#1512)

* First pass at incorporating nvapi into test harness.

* D3d12 Atomic Float Add via NVAPI working

* Dx12 atomic float appears to work.

* Atomic float add on Dx12.

* Added atomic64 feature addition to vk.
Fix correct output for atomic-float-byte-address.slang

* Disable atomic float failing tests.

* Upgraded VK headers.

* Detect atomic float availability on VK.

* Try to get test working for in64 atomic.

* Made HLSL prelude controlled via the render-test requirements.

* Added -enable-nvapi to premake.

* Fix D3D12Renderer when NVAPI is not available.

* Small improvements to VKRenderer.

* Improve atomic documentation in target-compatibility.md.

* Fixed NVAPI working on D3D12.

* Test for specific NVAPI features.

* Remove requiredFeatures from Renderer::Desc as was ignored. Tried to document more around nvapiExtnSlot.

* Readded requiredFeatures to Renderer::Desc

* Improve comments in the tests.

10 files changed, 111 insertions, 73 deletions

diff --git a/tests/slang-extension/atomic-float-byte-address-buffer.slang b/tests/slang-extension/atomic-float-byte-address-buffer.slang
index a332ddeac..603b92d65 100644
--- a/tests/slang-extension/atomic-float-byte-address-buffer.slang
+++ b/tests/slang-extension/atomic-float-byte-address-buffer.slang
@@ -1,18 +1,14 @@
 // atomic-float-byte-address-buffer.slang
 
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute -output-using-type
-// Produces a different result
-//TEST(compute):COMPARE_COMPUTE_EX:-dx11 -slang -compute -render-features atomic-float -output-using-type -nvapi-register u0
-// Disabled because crashes currently on vulkan 
-// https://vulkan.gpuinfo.org/listdevicescoverage.php?extension=VK_EXT_shader_atomic_float
+//TEST(compute):COMPARE_COMPUTE_EX:-dx11 -slang -compute -render-features atomic-float -output-using-type -nvapi-slot u0
 //TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -render-features atomic-float -output-using-type
-// Doesn't work on D3D12 for unknown reasons
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-d3d12 -compute -render-features atomic-float -output-using-type -compile-arg -O2 -nvapi-register u0
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-d3d12 -compute -use-dxil -render-features atomic-float -output-using-type -compile-arg -O2 -nvapi-register u0
+//TEST(compute):COMPARE_COMPUTE_EX:-d3d12 -compute -render-features atomic-float -output-using-type -compile-arg -O2 -nvapi-slot u0
+//TEST(compute):COMPARE_COMPUTE_EX:-d3d12 -compute -use-dxil -render-features atomic-float -output-using-type -compile-arg -O2 -nvapi-slot u0
 //TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute -output-using-type 
 
-// We don't use this, but having this defined makes the 0 slot available if nvapi is going to be used
-// Only strictly necessary on the dx12 path
+// The test doesn't directly use this, but having this defined makes the 0 slot available if NVAPI is going to be used
+// Only strictly necessary on the D3D11/D3D12 paths
 //TEST_INPUT:ubuffer(data=[0 0 0 0 ], stride=4):name=nvapiBuffer
 RWStructuredBuffer<int> nvapiBuffer;
 
diff --git a/tests/slang-extension/atomic-int64-byte-address-buffer.slang b/tests/slang-extension/atomic-int64-byte-address-buffer.slang
index 216e55dc2..628c675a2 100644
--- a/tests/slang-extension/atomic-int64-byte-address-buffer.slang
+++ b/tests/slang-extension/atomic-int64-byte-address-buffer.slang
@@ -4,14 +4,13 @@
 //DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute
 // No support for int64_t on fxc - we need SM6.0 and dxil
 // https://docs.microsoft.com/en-us/windows/win32/direct3dhlsl/hlsl-shader-model-6-0-features-for-direct3d-12
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -nvapi-register u0
-// Doesn't work on current version of driver. The atomics have no effect perhaps meaning they aren't being decoded correctly
-//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -profile cs_6_0 -use-dxil -render-features atomic-int64 -nvapi-register u0 -compile-arg -O2
+//DISABLE_TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -nvapi-slot u0
+//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -profile cs_6_0 -use-dxil -render-features atomic-int64 -nvapi-slot u0 -compile-arg -O2
 //TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute -render-features atomic-int64
 //TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute
 
-// We don't use this, but having this defined makes the 0 slot available if nvapi is going to be used
-// Only strictly necessary on the dx12 path
+// The test doesn't directly use this, but having this defined makes the 0 slot available if NVAPI is going to be used
+// Only strictly necessary on the D3D12 path
 //TEST_INPUT:ubuffer(data=[0 0 0 0 ], stride=4):name=nvapiBuffer
 RWStructuredBuffer<int> nvapiBuffer;
 
diff --git a/tools/gfx/d3d11/render-d3d11.cpp b/tools/gfx/d3d11/render-d3d11.cpp
index 8eafd24b4..30c982ab3 100644
--- a/tools/gfx/d3d11/render-d3d11.cpp
+++ b/tools/gfx/d3d11/render-d3d11.cpp
@@ -514,6 +514,19 @@ D3D11Renderer::ScopeNVAPI::~ScopeNVAPI()
 
 // !!!!!!!!!!!!!!!!!!!!!!!!!!!! Renderer interface !!!!!!!!!!!!!!!!!!!!!!!!!!
 
+static bool _isSupportedNVAPIOp(IUnknown* dev, uint32_t op)
+{
+#ifdef GFX_NVAPI
+    {
+        bool isSupported;
+        NvAPI_Status status = NvAPI_D3D11_IsNvShaderExtnOpCodeSupported(dev, NvU32(op), &isSupported);
+        return status == NVAPI_OK && isSupported;
+    }
+#else
+    return false;
+#endif
+}
+
 SlangResult D3D11Renderer::initialize(const Desc& desc, void* inWindowHandle)
 {
     auto windowHandle = (HWND)inWindowHandle;
@@ -647,29 +660,32 @@ SlangResult D3D11Renderer::initialize(const Desc& desc, void* inWindowHandle)
     }
 
     // NVAPI
+    if (desc.nvapiExtnSlot >= 0)
     {
-        const char* features[] = { "nvapi", "atomic-float", "atomic-int64" };
-        bool needsNvapi = false;
-        for (Index i = 0; i < SLANG_COUNT_OF(features); ++i)
+        if (SLANG_FAILED(NVAPIUtil::initialize()))
         {
-            if (desc.requiredFeatures.indexOf(features[i]) >= 0)
-            {
-                needsNvapi = true;
-                break;
-            }
+            return SLANG_E_NOT_AVAILABLE;
         }
 
-        if (needsNvapi && SLANG_SUCCEEDED(NVAPIUtil::initialize()))
+#ifdef GFX_NVAPI
+        if (NvAPI_D3D11_SetNvShaderExtnSlot(m_device, NvU32(desc.nvapiExtnSlot)) != NVAPI_OK)
         {
-            // TODO(JS): We should test for specific features here.
-            for (Index i = 0; i < SLANG_COUNT_OF(features); ++i)
-            {
-                m_features.add(features[i]);
-            }
-            m_nvapi = true;
+            return SLANG_E_NOT_AVAILABLE;
         }
-    }
 
+        if (_isSupportedNVAPIOp(m_device, NV_EXTN_OP_UINT64_ATOMIC ))
+        {
+            m_features.add("atomic-int64");
+        }
+        if (_isSupportedNVAPIOp(m_device, NV_EXTN_OP_FP32_ATOMIC))
+        {
+            m_features.add("atomic-float");
+        }
+
+        m_nvapi = true;
+#endif
+    }
+    
     // TODO: Add support for debugging to help detect leaks:
     //
     //      ComPtr<ID3D11Debug> gDebug;
diff --git a/tools/gfx/d3d12/render-d3d12.cpp b/tools/gfx/d3d12/render-d3d12.cpp
index ca44aa04d..0f23d8dd4 100644
--- a/tools/gfx/d3d12/render-d3d12.cpp
+++ b/tools/gfx/d3d12/render-d3d12.cpp
@@ -1523,6 +1523,19 @@ Result D3D12Renderer::_createDevice(DeviceCheckFlags deviceCheckFlags, const Uno
     return SLANG_OK;
 }
 
+static bool _isSupportedNVAPIOp(ID3D12Device* dev, uint32_t op)
+{
+#ifdef GFX_NVAPI
+    {
+        bool isSupported;
+        NvAPI_Status status = NvAPI_D3D12_IsNvShaderExtnOpCodeSupported(dev, NvU32(op), &isSupported);
+        return status == NVAPI_OK && isSupported;
+    }
+#else
+    return false;
+#endif
+}
+
 Result D3D12Renderer::initialize(const Desc& desc, void* inWindowHandle)
 {
     m_hwnd = (HWND)inWindowHandle;
@@ -1602,32 +1615,40 @@ Result D3D12Renderer::initialize(const Desc& desc, void* inWindowHandle)
         return SLANG_FAIL;
     }
 
+    // Set the device
+    m_device = m_deviceInfo.m_device;
+
     // NVAPI
+    if (desc.nvapiExtnSlot >= 0)
     {
-        const char* features[] = { "nvapi", "atomic-float", "atomic-int64" };
-        bool needsNvapi = false;
-        for (Index i = 0; i < SLANG_COUNT_OF(features); ++i)
+        if (SLANG_FAILED(NVAPIUtil::initialize()))
         {
-            if (desc.requiredFeatures.indexOf(features[i]) >= 0)
-            {
-                needsNvapi = true;
-                break;
-            }
+            return SLANG_E_NOT_AVAILABLE;
         }
 
-        if (needsNvapi && SLANG_SUCCEEDED(NVAPIUtil::initialize()))
+#ifdef GFX_NVAPI
+        // From DOCS: Applications are expected to bind null UAV to this slot.
+        // NOTE! We don't currently do this, but doesn't seem to be a problem.
+
+        const NvAPI_Status status = NvAPI_D3D12_SetNvShaderExtnSlotSpace(m_device, NvU32(desc.nvapiExtnSlot), NvU32(0));
+        
+        if (status != NVAPI_OK)
         {
-            // TODO(JS): We should test for specific features here.
-            for (Index i = 0; i < SLANG_COUNT_OF(features); ++i)
-            {
-                m_features.add(features[i]);
-            }
-            m_nvapi = true;
+            return SLANG_E_NOT_AVAILABLE;
         }
-    }
 
-    // Set the device
-    m_device = m_deviceInfo.m_device;
+        if (_isSupportedNVAPIOp(m_device, NV_EXTN_OP_UINT64_ATOMIC))
+        {
+            m_features.add("atomic-int64");
+        }
+        if (_isSupportedNVAPIOp(m_device, NV_EXTN_OP_FP32_ATOMIC))
+        {
+            m_features.add("atomic-float");
+        }
+
+        m_nvapi = true;
+#endif
+    }
 
     // Find what features are supported
     {
diff --git a/tools/gfx/nvapi/nvapi-include.h b/tools/gfx/nvapi/nvapi-include.h
index e3674af95..c213e0bfb 100644
--- a/tools/gfx/nvapi/nvapi-include.h
+++ b/tools/gfx/nvapi/nvapi-include.h
@@ -15,5 +15,7 @@
 #   endif
 
 #   include <nvapi.h>
+#   include <nvShaderExtnEnums.h>
+
 #endif
 
diff --git a/tools/gfx/open-gl/render-gl.cpp b/tools/gfx/open-gl/render-gl.cpp
index 5f16f07bc..caf8794c0 100644
--- a/tools/gfx/open-gl/render-gl.cpp
+++ b/tools/gfx/open-gl/render-gl.cpp
@@ -717,9 +717,12 @@ SlangResult GLRenderer::initialize(const Desc& desc, void* inWindowHandle)
         }
     }
 
-    if (m_desc.requiredFeatures.indexOf("nvapi") >= 0 && SLANG_SUCCEEDED(NVAPIUtil::initialize()))
+    if (m_desc.nvapiExtnSlot >= 0)
     {
-        m_features.add("nvapi");
+        if (SLANG_FAILED(NVAPIUtil::initialize()))
+        {
+            return SLANG_E_NOT_AVAILABLE;
+        }
     }
 
     auto extensions = glGetString(GL_EXTENSIONS);
diff --git a/tools/gfx/render.h b/tools/gfx/render.h
index 12ef1a9e9..55beb5774 100644
--- a/tools/gfx/render.h
+++ b/tools/gfx/render.h
@@ -800,12 +800,14 @@ public:
 
     struct Desc
     {
-        int width;                                      ///< Width in pixels
-        int height;                                     ///< height in pixels
+        int width = 0;                                  ///< Width in pixels
+        int height = 0;                                 ///< height in pixels
         Slang::String adapter;                          ///< Name to identify the adapter to use
-        Slang::List<Slang::String> requiredFeatures;    ///< The features enabled on this renderer
+        Slang::List<Slang::String> requiredFeatures;    ///< List of required feature names. 
+        int nvapiExtnSlot = -1;                         ///< The slot (typically UAV) used to identify NVAPI intrinsics. If >=0 NVAPI is required.
     };
 
+        // Will return with SLANG_E_NOT_AVAILABLE if NVAPI can't be initialized and nvapiExtnSlot >= 0
     virtual SlangResult initialize(const Desc& desc, void* inWindowHandle) = 0;
 
     bool hasFeature(const Slang::UnownedStringSlice& feature) { return getFeatures().indexOf(Slang::String(feature)) != Slang::Index(-1); }
diff --git a/tools/render-test/options.cpp b/tools/render-test/options.cpp
index fd4b75ed1..797439cfa 100644
--- a/tools/render-test/options.cpp
+++ b/tools/render-test/options.cpp
@@ -240,7 +240,7 @@ static SlangResult _setRendererType(RendererType type, const char* arg, Slang::W
         {
             outOptions.dontAddDefaultEntryPoints = true;
         }
-        else if (strcmp(arg, "-nvapi-register") == 0)
+        else if (strcmp(arg, "-nvapi-slot") == 0)
         {
             if (argCursor == argEnd)
             {
@@ -248,7 +248,7 @@ static SlangResult _setRendererType(RendererType type, const char* arg, Slang::W
                 return SLANG_FAIL;
             }
 
-            outOptions.nvapiRegister = (*argCursor++);
+            outOptions.nvapiExtnSlot = (*argCursor++);
         }
         else
         {
diff --git a/tools/render-test/options.h b/tools/render-test/options.h
index d311568d4..ddb903a4a 100644
--- a/tools/render-test/options.h
+++ b/tools/render-test/options.h
@@ -75,7 +75,7 @@ struct Options
 
     uint32_t computeDispatchSize[3] = { 1, 1, 1 };
 
-    Slang::String nvapiRegister;                               ///< The nvapiRegister to use.
+    Slang::String nvapiExtnSlot;                               ///< The nvapiRegister to use.
 
     static SlangResult parse(int argc, const char*const* argv, Slang::WriterHelper stdError, Options& outOptions);
 };
diff --git a/tools/render-test/render-test-main.cpp b/tools/render-test/render-test-main.cpp
index e7356901d..25486c722 100644
--- a/tools/render-test/render-test-main.cpp
+++ b/tools/render-test/render-test-main.cpp
@@ -12,6 +12,7 @@
 #include "shader-renderer-util.h"
 
 #include "../source/core/slang-io.h"
+#include "../source/core/slang-string-util.h"
 
 #include "core/slang-token-reader.h"
 
@@ -406,7 +407,7 @@ Result RenderTestApp::update(Window* window)
 static SlangResult _setSessionPrelude(const Options& options, const char* exePath, SlangSession* session)
 {
     // Let's see if we need to set up special prelude for HLSL
-    if (options.nvapiRegister.getLength())
+    if (options.nvapiExtnSlot.getLength())
     {
         String rootPath;
         SLANG_RETURN_ON_FAIL(TestToolUtil::getRootPath(exePath, rootPath));
@@ -416,7 +417,7 @@ static SlangResult _setSessionPrelude(const Options& options, const char* exePat
 
         StringBuilder buf;
         // We have to choose a slot that NVAPI will use. 
-        buf << "#define NV_SHADER_EXTN_SLOT " << options.nvapiRegister << "\n";
+        buf << "#define NV_SHADER_EXTN_SLOT " << options.nvapiExtnSlot << "\n";
 
         // Include the NVAPI header
         buf << "#include \"" << includePath << "\"\n\n";
@@ -589,13 +590,18 @@ static SlangResult _innerMain(Slang::StdWriters* stdWriters, SlangSession* sessi
         }
     }
 
+    Index nvapiExtnSlot = -1;
+
     // Let's see if we need to set up special prelude for HLSL
-    if (options.nvapiRegister.getLength())
+    if (options.nvapiExtnSlot.getLength() && options.nvapiExtnSlot[0] == 'u')
     {
-        // We require nvapi to be available on the device
-        if (options.renderFeatures.indexOf("nvapi") < 0)
+        //
+        Slang::Int value;
+        UnownedStringSlice slice = options.nvapiExtnSlot.getUnownedSlice();
+        UnownedStringSlice indexText(slice.begin() + 1 , slice.end());
+        if (SLANG_SUCCEEDED(StringUtil::parseInt(indexText, value)))
         {
-            options.renderFeatures.add("nvapi");
+            nvapiExtnSlot = Index(value);
         }
     }
 
@@ -716,16 +722,6 @@ static SlangResult _innerMain(Slang::StdWriters* stdWriters, SlangSession* sessi
 #endif
     }
 
-    if (options.nvapiRegister.getLength())
-    {
-        // We require nvapi to be available on the device
-        if (options.renderFeatures.indexOf("nvapi") < 0)
-        {
-            options.renderFeatures.add("nvapi");
-        }
-    }
-
-
     Slang::RefPtr<Renderer> renderer;
     {
         RendererUtil::CreateFunc createFunc = RendererUtil::getCreateFunc(options.rendererType);
@@ -748,6 +744,7 @@ static SlangResult _innerMain(Slang::StdWriters* stdWriters, SlangSession* sessi
         desc.height = gWindowHeight;
         desc.adapter = options.adapter;
         desc.requiredFeatures = options.renderFeatures;
+        desc.nvapiExtnSlot = int(nvapiExtnSlot);
 
         window = renderer_test::Window::create();
         SLANG_RETURN_ON_FAIL(window->initialize(gWindowWidth, gWindowHeight));
@@ -755,7 +752,9 @@ static SlangResult _innerMain(Slang::StdWriters* stdWriters, SlangSession* sessi
         SlangResult res = renderer->initialize(desc, window->getHandle());
         if (SLANG_FAILED(res))
         {
-            if (!options.onlyStartup)
+            // Returns E_NOT_AVAILABLE only when specified features are not available.
+            // Will cause to be ignored.
+            if (!options.onlyStartup && res != SLANG_E_NOT_AVAILABLE)
             {
                 fprintf(stderr, "Unable to initialize renderer %s\n", rendererName.getBuffer());
             }