working build and load of cuda compiled .dll

build.bat

@@ -1,19 +1,64 @@
 @echo off
 setlocal
 
+@rem I rely on environment variables here.
 @rem set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4
 @rem set CUDSS_PATH=C:\Program Files\NVIDIA\cuDSS
 
-nvcc -o build/cudss_test.exe src/cudss_test.cu ^
+set nvccArch=sm_120
-    -I"%CUDA_PATH%\include" ^
+set commLayerSource=commlayer
-    -I"%CUDSS_PATH%\include" ^
+set mainSource=main
-    -L"%CUDA_PATH%\lib\x64" ^
+set CommonXCompFlags=/W3 /O2 /EHsc
-    -L"%CUDSS_PATH%\lib\12" ^
+set DllXcompilerFlags="%CommonXCompFlags% /MD"
+set ExeXcompilerFlags="%CommonXCompFlags% /MT"
+set IgnoreWarnings=-Wno-deprecated-gpu-targets
+
+set CudaIncludes=-I"%CUDA_PATH%\include" -I"%CUDSS_PATH%\include"
+set OtherIncludes=-Iinclude
+set CudaLibPaths=-L"%CUDA_PATH%\lib\x64" -L"%CUDSS_PATH%\lib\12"
+
+nvcc -c -o build/%commLayerSource%.obj src/%commLayerSource%.cu ^
+    -arch=%nvccArch% ^
+     %CudaIncludes% %OtherIncludes% %CudaLibPaths% ^
     -lcudss -lcudart ^
-    -Xcompiler "/W3 /O2"
+    %IgnoreWarnings% ^
+    -Xcompiler %DllXcompilerFlags%
+
+if %ERRORLEVEL% NEQ 0 (
+    echo Compile failed: %commLayerSource%.cu
+    exit /b 1
+)
+
+nvcc -shared -o build/%commLayerSource%.dll build/%commLayerSource%.obj ^
+    -arch=%nvccArch% ^
+    %IgnoreWarnings% ^
+    -Xcompiler %DllXcompilerFlags% ^
+    -Xlinker "/NODEFAULTLIB:LIBCMT"
+
+if %ERRORLEVEL% NEQ 0 (
+    echo DLL link failed
+    exit /b 1
+)
+
+nvcc -o build/%mainSource% src/%mainSource%.cu ^
+    -arch=%nvccArch% ^
+     %CudaIncludes% %OtherIncludes% %CudaLibPaths% ^
+    -lcudss -lcudart ^
+    %IgnoreWarnings% ^
+    -Xcompiler %ExeXcompilerFlags%
+
+
+
+@rem nvcc -o build/cudss_test.exe src/cudss_test.cu ^
+@rem     -I"%CUDA_PATH%\include" ^
+@rem     -I"%CUDSS_PATH%\include" ^
+@rem     -L"%CUDA_PATH%\lib\x64" ^
+@rem     -L"%CUDSS_PATH%\lib\12" ^
+@rem     -lcudss -lcudart ^
+@rem     -Xcompiler "/W3 /O2"
 
 if %ERRORLEVEL% EQU 0 (
-    echo Build successful: cudss_test.exe
+    echo Build successful: %mainSource%.exe
 ) else (
     echo Build failed
     exit /b 1

include/util.h

@@ -0,0 +1,33 @@
+#pragma once
+
+/* ============================================================================
+ * Logging
+ * ============================================================================ */
+
+#define LOG(fmt, ...) fprintf(stdout, "[cudss_test] " fmt "\n", ##__VA_ARGS__)
+#define LOG_ERROR(fmt, ...) fprintf(stderr, "[cudss_test ERROR] " fmt "\n", ##__VA_ARGS__)
+
+/* ============================================================================
+ * Error Checking Macros
+ * ============================================================================ */
+
+#define CUDA_CHECK(call)                                                       \
+    do {                                                                       \
+        cudaError_t err = (call);                                              \
+        if (err != cudaSuccess) {                                              \
+            LOG_ERROR("CUDA error at %s:%d - %s", __FILE__, __LINE__,          \
+                      cudaGetErrorString(err));                                \
+            exit(EXIT_FAILURE);                                                \
+        }                                                                      \
+    } while (0)
+
+#define CUDSS_CHECK(call)                                                      \
+    do {                                                                       \
+        cudssStatus_t status = (call);                                         \
+        if (status != CUDSS_STATUS_SUCCESS) {                                  \
+            LOG_ERROR("cuDSS error at %s:%d - status %d", __FILE__, __LINE__,  \
+                      (int)status);                                            \
+            exit(EXIT_FAILURE);                                                \
+        }                                                                      \
+    } while (0)
+

src/commlayer.cu

@@ -0,0 +1,14 @@
+#include <cuda_runtime.h>
+#include <iostream>
+
+#define __CSCUDSS_EXPORT extern "C" __declspec(dllexport)
+
+
+__CSCUDSS_EXPORT int libraryMain() {
+
+  std::cout << "HELLO LIBRARY" << std::endl;
+
+
+  return 0;
+
+}

src/main.cu

@@ -0,0 +1,105 @@
+#include <cuda_runtime.h>
+#include <cudss.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <iostream>
+#define WIN32_LEAN_AND_MEAN
+#include <windows.h>
+
+#include "util.h"
+
+
+
+/* ============================================================================
+ * Version Information
+ * ============================================================================ */
+
+void printCudaVersion(void) {
+    int runtimeVersion = 0;
+    int driverVersion = 0;
+    
+    CUDA_CHECK(cudaRuntimeGetVersion(&runtimeVersion));
+    CUDA_CHECK(cudaDriverGetVersion(&driverVersion));
+    
+    int runtimeMajor = runtimeVersion / 1000;
+    int runtimeMinor = (runtimeVersion % 1000) / 10;
+    
+    int driverMajor = driverVersion / 1000;
+    int driverMinor = (driverVersion % 1000) / 10;
+    
+    LOG("CUDA Runtime Version: %d.%d", runtimeMajor, runtimeMinor);
+    LOG("CUDA Driver Version:  %d.%d", driverMajor, driverMinor);
+}
+
+void printCudssVersion(void) {
+    int major = 0;
+    int minor = 0;
+    int patch = 0;
+    
+    CUDSS_CHECK(cudssGetProperty(MAJOR_VERSION, &major));
+    CUDSS_CHECK(cudssGetProperty(MINOR_VERSION, &minor));
+    CUDSS_CHECK(cudssGetProperty(PATCH_LEVEL, &patch));
+    
+    LOG("cuDSS Version: %d.%d.%d", major, minor, patch);
+}
+
+void printDeviceInfo(void) {
+    int deviceCount = 0;
+    CUDA_CHECK(cudaGetDeviceCount(&deviceCount));
+    
+    if (deviceCount == 0) {
+        LOG_ERROR("No CUDA-capable devices found");
+        exit(EXIT_FAILURE);
+    }
+    
+    int device = 0;
+    CUDA_CHECK(cudaGetDevice(&device));
+    
+    struct cudaDeviceProp props;
+    CUDA_CHECK(cudaGetDeviceProperties(&props, device));
+    
+    LOG("Device: %s (compute %d.%d)", props.name, props.major, props.minor);
+    LOG("Memory: %.2f GB", (double)props.totalGlobalMem / (1024.0 * 1024.0 * 1024.0));
+}
+
+/* ============================================================================
+ * Main Entry Point
+ * ============================================================================ */
+
+using libraryMain = int(*)();
+
+int main(int argc, char **argv) {
+    LOG("cuDSS Test Program");
+    LOG("==================");
+    
+    /* Print version information */
+    printCudaVersion();
+    printCudssVersion();
+    printDeviceInfo();
+    
+    HMODULE hLib = LoadLibraryA("commlayer.dll");
+    if (!hLib) {
+        std::cerr << "Failed to load DLL. Error: " << GetLastError() << std::endl;
+        return 1;
+    }
+    
+    libraryMain libMain = (libraryMain)GetProcAddress(hLib, "libraryMain");
+    if (!libMain) {
+        std::cerr << "Failed to find function. Error: " << GetLastError() << std::endl;
+        FreeLibrary(hLib);
+        return 1;
+    }
+    libMain();
+
+    /* Initialize cuDSS */
+    cudssHandle_t handle = NULL;
+    CUDSS_CHECK(cudssCreate(&handle));
+    LOG("cuDSS handle created successfully");
+       
+    /* Cleanup */
+    CUDSS_CHECK(cudssDestroy(handle));
+    LOG("cuDSS handle destroyed");
+    
+    LOG("Done.");
+    return 0;
+}