Enable `gfx::CUDADevice` on linux. (#1756)

8 files changed, 97 insertions, 74 deletions

diff --git a/.gitignore b/.gitignore
index fad8a1030..fe322c934 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,6 +1,7 @@
 *.exe
 *.pdb
 .vs
+.vscode
 *.VC.opendb
 *.VC.db
 *.vcxproj.user
@@ -15,6 +16,7 @@
 bin/
 intermediate/
 build.*/
+Makefile
 
 # Files generated by test runner.
 #
diff --git a/premake5.lua b/premake5.lua
index 13debb57b..7052d1de9 100644
--- a/premake5.lua
+++ b/premake5.lua
@@ -351,6 +351,48 @@ function getBuildLocationName()
     end 
 end
 
+-- Adds CUDA dependency to a project
+function addCUDAIfEnabled()
+    if type(cudaPath) == "string" and isTargetWindows then
+        filter {}
+        includedirs { cudaPath .. "/include" }
+        includedirs { cudaPath .. "/include", cudaPath .. "/common/inc" }
+        links { "cuda", "cudart" }
+        if optixPath then
+            defines { "RENDER_TEST_OPTIX" }
+            includedirs { optixPath .. "include/" }
+        end
+        
+        filter { "platforms:x86" }
+            libdirs { cudaPath .. "/lib/Win32/" }
+           
+        filter { "platforms:x64" }
+            libdirs { cudaPath .. "/lib/x64/" }       
+        filter {}
+        return true
+    elseif enableCuda then
+        filter {}
+        if type(cudaPath) == "string" then
+            includedirs { cudaPath .. "/include" }
+            includedirs { cudaPath .. "/include" }
+            if optixPath then
+                defines { "GFX_OPTIX" }
+                includedirs { optixPath .. "include/" }
+            end
+            filter { "platforms:x86" }
+            libdirs { cudaPath .. "/lib32/" }
+            filter { "platforms:x64" }
+            libdirs { cudaPath .. "/lib64/" }
+            filter {}
+            links { "cuda", "cudart" }
+        else
+            print "Error: CUDA is enabled but --cuda-sdk-path is not specified."
+        end
+        return true
+    end
+    return false
+end
+
 --
 -- Next we will define a helper routine that all of our
 -- projects will bottleneck through. Here `name` is
@@ -500,7 +542,7 @@ function toolSharedLibrary(name)
     -- specifying that the project lives under the `tools/` path.
     --
     baseSlangProject(name .. "-tool", "tools/" .. name)
-    
+
     defines { "SLANG_SHARED_LIBRARY_TOOL" }
    
     kind "SharedLib"
@@ -547,6 +589,8 @@ function example(name)
             links {"X11"}
         end
     end
+
+    addCUDAIfEnabled();
 end
 
 --
@@ -716,7 +760,6 @@ toolSharedLibrary "render-test"
     
     includedirs { ".", "external", "source", "tools/gfx", "tools/platform" }
     links { "core", "slang", "gfx", "gfx-util", "platform" }
-   
     if isTargetWindows then    
         addSourceDir "tools/render-test/windows"
         
@@ -727,27 +770,14 @@ toolSharedLibrary "render-test"
         -- directory into the output directory.
         -- d3dcompiler_47.dll is copied from the external/slang-binaries submodule.
         postbuildcommands { '"$(SolutionDir)tools\\copy-hlsl-libs.bat" "$(WindowsSdkDir)Redist/D3D/%{cfg.platform:lower()}/" "%{cfg.targetdir}/" "windows-%{cfg.platform:lower()}"'}    
+        if (type(cudaPath) == "string") then
+            addSourceDir "tools/render-test/cuda"
+        end
     end
-   
-    if type(cudaPath) == "string" and isTargetWindows then
-        addSourceDir "tools/render-test/cuda"
+    if addCUDAIfEnabled() then
         defines { "RENDER_TEST_CUDA" }
-        includedirs { cudaPath .. "/include" }
-        includedirs { cudaPath .. "/include", cudaPath .. "/common/inc" }
-        links { "cuda", "cudart" }
-        if optixPath then
-            defines { "RENDER_TEST_OPTIX" }
-            includedirs { optixPath .. "include/" }
-        end
-        
-        filter { "platforms:x86" }
-            libdirs { cudaPath .. "/lib/Win32/" }
-           
-        filter { "platforms:x64" }
-            libdirs { cudaPath .. "/lib/x64/" }       
-        
     end
-  
+
 --
 -- `gfx` is a abstraction layer for different GPU platforms.
 --
@@ -769,6 +799,7 @@ tool "gfx"
     -- Will compile across targets
     addSourceDir "tools/gfx/cpu"
     addSourceDir "tools/gfx/nvapi"
+    addSourceDir "tools/gfx/cuda"
 
     -- To special case that we may be building using cygwin on windows. If 'true windows' we build for dx12/vk and run the script
     -- If not we assume it's a cygwin/mingw type situation and remove files that aren't appropriate
@@ -786,22 +817,6 @@ tool "gfx"
         addSourceDir "tools/gfx/d3d" 
         addSourceDir "tools/gfx/d3d11"
         addSourceDir "tools/gfx/d3d12"
-        addSourceDir "tools/gfx/cuda"
-
-        if type(cudaPath) == "string" then
-            defines { "GFX_ENABLE_CUDA" }
-            includedirs { cudaPath .. "/include" }
-            includedirs { cudaPath .. "/include", cudaPath .. "/common/inc" }
-            if optixPath then
-                defines { "GFX_OPTIX" }
-                includedirs { optixPath .. "include/" }
-            end
-            links { "cuda", "cudart" }   
-            filter { "platforms:x86" }
-                libdirs { cudaPath .. "/lib/Win32/" }
-            filter { "platforms:x64" }
-                libdirs { cudaPath .. "/lib/x64/" }
-        end
     elseif targetDetail == "mingw" or targetDetail == "cygwin" then
         -- Don't support any render techs...
     elseif os.target() == "macosx" then
@@ -836,6 +851,9 @@ tool "gfx"
             links { "nvapi64" }
             
     end
+    if addCUDAIfEnabled() then
+        defines { "GFX_ENABLE_CUDA" }
+    end
 
 --
 -- `gfx-util` is a static library containing utilities and helpers for using
diff --git a/tools/gfx/cuda/render-cuda.cpp b/tools/gfx/cuda/render-cuda.cpp
index 89aaa33aa..3e94f5571 100644
--- a/tools/gfx/cuda/render-cuda.cpp
+++ b/tools/gfx/cuda/render-cuda.cpp
@@ -1,20 +1,19 @@
 #include "render-cuda.h"
-#include "slang.h"
-#include "slang-com-ptr.h"
-#include "slang-com-helper.h"
+
+#ifdef GFX_ENABLE_CUDA
+#include <cuda.h>
+#include <cuda_runtime_api.h>
 #include "core/slang-basic.h"
 #include "core/slang-blob.h"
+#include "core/slang-std-writers.h"
 
+#include "slang.h"
+#include "slang-com-ptr.h"
+#include "slang-com-helper.h"
 #include "../command-writer.h"
 #include "../renderer-shared.h"
 #include "../render-graphics-common.h"
 #include "../slang-context.h"
-
-#ifdef GFX_ENABLE_CUDA
-#include <cuda.h>
-#include <cuda_runtime_api.h>
-#include "core/slang-std-writers.h"
-
 #endif
 
 namespace gfx
@@ -466,7 +465,7 @@ public:
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        setData(ShaderOffset const& offset, void const* data, size_t size)
+        setData(ShaderOffset const& offset, void const* data, size_t size) override
     {
         size = Math::Min(size, bufferResource->getDesc()->sizeInBytes - offset.uniformOffset);
         SLANG_CUDA_RETURN_ON_FAIL(cudaMemcpy(
@@ -488,7 +487,7 @@ public:
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        getObject(ShaderOffset const& offset, IShaderObject** object)
+        getObject(ShaderOffset const& offset, IShaderObject** object) override
     {
         auto subObjectIndex =
             getLayout()->m_bindingRanges[offset.bindingRangeIndex].baseIndex + offset.bindingArrayIndex;
@@ -507,7 +506,7 @@ public:
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        setObject(ShaderOffset const& offset, IShaderObject* object)
+        setObject(ShaderOffset const& offset, IShaderObject* object) override
     {
         auto layout = getLayout();
 
@@ -625,7 +624,7 @@ public:
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        setResource(ShaderOffset const& offset, IResourceView* resourceView)
+        setResource(ShaderOffset const& offset, IResourceView* resourceView) override
     {
         auto layout = getLayout();
 
@@ -669,14 +668,14 @@ public:
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        setSampler(ShaderOffset const& offset, ISamplerState* sampler)
+        setSampler(ShaderOffset const& offset, ISamplerState* sampler) override
     {
         SLANG_UNUSED(sampler);
         SLANG_UNUSED(offset);
         return SLANG_OK;
     }
     virtual SLANG_NO_THROW Result SLANG_MCALL setCombinedTextureSampler(
-        ShaderOffset const& offset, IResourceView* textureView, ISamplerState* sampler)
+        ShaderOffset const& offset, IResourceView* textureView, ISamplerState* sampler) override
     {
         SLANG_UNUSED(sampler);
         setResource(offset, textureView);
@@ -758,7 +757,7 @@ public:
     }
 
     virtual SLANG_NO_THROW Result SLANG_MCALL
-        setDeviceData(size_t offset, void* data, size_t size)
+        setDeviceData(size_t offset, void* data, size_t size) override
     {
         size = Math::Min(size, uniformBufferSize - offset);
         SLANG_CUDA_RETURN_ON_FAIL(cudaMemcpy(
@@ -1004,8 +1003,8 @@ public:
                 *outObject = nullptr;
                 return SLANG_E_NO_INTERFACE;
             }
-            virtual SLANG_NO_THROW uint32_t SLANG_MCALL addRef() { return 1; }
-            virtual SLANG_NO_THROW uint32_t SLANG_MCALL release() { return 1; }
+            virtual SLANG_NO_THROW uint32_t SLANG_MCALL addRef() override { return 1; }
+            virtual SLANG_NO_THROW uint32_t SLANG_MCALL release() override { return 1; }
 
         public:
             CommandWriter* m_writer;
@@ -1064,8 +1063,8 @@ public:
                 *outObject = nullptr;
                 return SLANG_E_NO_INTERFACE;
             }
-            virtual SLANG_NO_THROW uint32_t SLANG_MCALL addRef() { return 1; }
-            virtual SLANG_NO_THROW uint32_t SLANG_MCALL release() { return 1; }
+            virtual SLANG_NO_THROW uint32_t SLANG_MCALL addRef() override { return 1; }
+            virtual SLANG_NO_THROW uint32_t SLANG_MCALL release() override { return 1; }
 
         public:
             CommandWriter* m_writer;
@@ -1087,7 +1086,7 @@ public:
             }
 
             virtual SLANG_NO_THROW void SLANG_MCALL
-                uploadBufferData(IBufferResource* dst, size_t offset, size_t size, void* data)
+                uploadBufferData(IBufferResource* dst, size_t offset, size_t size, void* data) override
             {
                 m_writer->uploadBufferData(dst, offset, size, data);
             }
diff --git a/tools/gfx/render.cpp b/tools/gfx/render.cpp
index e7d30b728..183afaffd 100644
--- a/tools/gfx/render.cpp
+++ b/tools/gfx/render.cpp
@@ -92,11 +92,15 @@ extern "C"
             }
             break;
 #elif SLANG_LINUX_FAMILY
-            case DeviceType::Default:
-            case DeviceType::Vulkan:
-            {
-                return createVKDevice(desc, outDevice);
-            }
+        case DeviceType::Default:
+        case DeviceType::Vulkan:
+        {
+            return createVKDevice(desc, outDevice);
+        }
+        case DeviceType::CUDA:
+        {
+            return createCUDADevice(desc, outDevice);
+        }
 #endif
         case DeviceType::CPU:
             {
diff --git a/tools/render-test/bind-location.h b/tools/render-test/bind-location.h
index cfa88cce0..33670bf4b 100644
--- a/tools/render-test/bind-location.h
+++ b/tools/render-test/bind-location.h
@@ -1,10 +1,10 @@
 #ifndef BIND_LOCATION_H
 #define BIND_LOCATION_H
 
-#include "core/slang-basic.h"
-#include "core/slang-free-list.h"
-#include "core/slang-memory-arena.h"
-#include "core/slang-writer.h"
+#include "source/core/slang-basic.h"
+#include "source/core/slang-free-list.h"
+#include "source/core/slang-memory-arena.h"
+#include "source/core/slang-writer.h"
 
 #include "slang.h"
 
diff --git a/tools/render-test/cuda/cuda-compute-util.cpp b/tools/render-test/cuda/cuda-compute-util.cpp
index ea7dab163..bd77919a2 100644
--- a/tools/render-test/cuda/cuda-compute-util.cpp
+++ b/tools/render-test/cuda/cuda-compute-util.cpp
@@ -1,11 +1,11 @@
 
 #include "cuda-compute-util.h"
 
-#include "../../slang-com-helper.h"
+#include "slang-com-helper.h"
 
-#include "../../source/core/slang-std-writers.h"
-#include "../../source/core/slang-token-reader.h"
-#include "../../source/core/slang-semantic-version.h"
+#include "source/core/slang-std-writers.h"
+#include "source/core/slang-token-reader.h"
+#include "source/core/slang-semantic-version.h"
 
 #include "../bind-location.h"
 
diff --git a/tools/render-test/cuda/cuda-compute-util.h b/tools/render-test/cuda/cuda-compute-util.h
index ac0c6bba3..9c6c8b9b4 100644
--- a/tools/render-test/cuda/cuda-compute-util.h
+++ b/tools/render-test/cuda/cuda-compute-util.h
@@ -4,7 +4,7 @@
 #include "../slang-support.h"
 #include "../options.h"
 
-#include "../../source/core/slang-smart-pointer.h"
+#include "source/core/slang-smart-pointer.h"
 
 namespace renderer_test {
 
diff --git a/tools/render-test/shader-input-layout.h b/tools/render-test/shader-input-layout.h
index 2c8dbe821..253f065f6 100644
--- a/tools/render-test/shader-input-layout.h
+++ b/tools/render-test/shader-input-layout.h
@@ -1,10 +1,10 @@
 #ifndef SLANG_TEST_SHADER_INPUT_LAYOUT_H
 #define SLANG_TEST_SHADER_INPUT_LAYOUT_H
 
-#include "core/slang-basic.h"
-#include "core/slang-random-generator.h"
+#include "source/core/slang-basic.h"
+#include "source/core/slang-random-generator.h"
 
-#include "core/slang-writer.h"
+#include "source/core/slang-writer.h"
 
 
 #include "bind-location.h"