Merge branch 'repackage' of github.com:sfilippone/psblas3 into repackage

config/pac.m4

@@ -2251,7 +2251,7 @@ if test "x$pac_cv_have_cuda" == "xyes"; then
 #include <stdio.h>
 #include <cuda.h>
 
-int main(int argc, char *argv[])
+int main(int argc, char **argv)
 {
   printf("%d",CUDA_VERSION);
   return(0);

configure

@@ -10786,7 +10786,7 @@ printf %s "checking for CUDA version... " >&6; }
 #include <stdio.h>
 #include <cuda.h>
 
-int main(int argc, char *argv)
+int main(int argc, char **argv)
 {
   printf("%d",CUDA_VERSION);
   return(0);
@@ -12296,7 +12296,7 @@ fi
    	CCOPT                 : ${CCOPT}
 
 	CUDA 		      : ${HAVE_CUDA}
-	CUDA_CC               : ${CUDA_CC}
+	CUDA_CC               : ${pac_cv_cudacc}
 
 	BLAS                  : ${BLAS_LIBS}
 
@@ -12329,7 +12329,7 @@ printf "%s\n" "$as_me:
    	CCOPT                 : ${CCOPT}
 
 	CUDA 		      : ${HAVE_CUDA}
-	CUDA_CC               : ${CUDA_CC}
+	CUDA_CC               : ${pac_cv_cudacc}
 
 	BLAS                  : ${BLAS_LIBS}
 

configure.ac

@@ -969,7 +969,7 @@ AC_MSG_NOTICE([
    	CCOPT                 : ${CCOPT}
 
 	CUDA 		      : ${HAVE_CUDA}
-	CUDA_CC               : ${CUDA_CC}
+	CUDA_CC               : ${pac_cv_cudacc}
 
 	BLAS                  : ${BLAS_LIBS}
 

cuda/cuda_util.c

@@ -378,22 +378,36 @@ void cudaSync()
 {
   cudaError_t err;
   err = cudaDeviceSynchronize();
+#if 0
   if (err == cudaSuccess)
     return SPGPU_SUCCESS;	
   else {
     fprintf(stderr,"CUDA Error cudaSync: %s\n", cudaGetErrorString(err));
     return SPGPU_UNSPECIFIED;
   }
+#else
+  if (err != cudaSuccess) {
+    fprintf(stderr,"CUDA Error cudaSync: %s\n", cudaGetErrorString(err));
+  }
+  return ;
+#endif
 }
 
 void cudaReset()
 {
   cudaError_t err;
   err = cudaDeviceReset();
+#if 0
   if (err != cudaSuccess) {
     fprintf(stderr,"CUDA Error Reset: %s\n", cudaGetErrorString(err));
     return SPGPU_UNSPECIFIED;
   }
+#else
+  if (err != cudaSuccess) {
+    fprintf(stderr,"CUDA Error Reset: %s\n", cudaGetErrorString(err));
+  }
+  return ;
+#endif
 }
 
 

cuda/cuda_util.h

@@ -54,6 +54,7 @@ int freeRemoteBuffer(void* buffer);
 int gpuInit(int dev);
 int getDeviceCount();
 int getDevice();
+int getDeviceHasUVA();
 int setDevice(int dev);
 int getGPUMultiProcessors();
 int getGPUMemoryBusWidth();

cuda/elldev.c

@@ -112,7 +112,7 @@ void zeroEllDevice(void *remoteMatrix)
   else if (tmp->dataType == SPGPU_TYPE_COMPLEX_DOUBLE)
     cudaMemset((void  *)tmp->cM, 0, tmp->allocsize*sizeof(cuDoubleComplex));
   else
-    return SPGPU_UNSUPPORTED; // Unsupported params
+    return ; // Unsupported params
   //fprintf(stderr,"From allocEllDevice: %d %d %d %p %p %p\n",tmp->maxRowSize,
   //	  tmp->avgRowSize,tmp->allocsize,tmp->rS,tmp->rP,tmp->cM);
 

cuda/elldev.h

@@ -95,6 +95,7 @@ typedef struct EllDeviceParams
 	unsigned int firstIndex;
 } EllDeviceParams;
 
+int computeEllAllocPitch(int rowsCount);
 int FallocEllDevice(void** deviceMat, unsigned int rows, unsigned int maxRowSize, 
 		    unsigned int nnzeros,
 		    unsigned int columns, unsigned int elementType, 

cuda/fcusparse_fct.h

@@ -320,7 +320,7 @@ int T_spsvCSRGDevice(T_Cmat *Matrix, TYPE alpha, void *deviceX,
 }
 
 #if CUDA_VERSION >=  11030
-T_CSRGCreateSpMVDescr(T_CSRGDeviceMat *cMat)
+int T_CSRGCreateSpMVDescr(T_CSRGDeviceMat *cMat)
 {
   int64_t tr,tc,tz;
   tr = cMat->m;