fdurastante/psblas3
1
0
Fork 0
mirror of https://github.com/sfilippone/psblas3.git synced 2026-10-17 03:12:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fixes for compilation with CUDA

Browse Source
This commit is contained in:
sfilippone
2024-10-09 14:27:17 +02:00
parent abdf7fc05a
commit ade79bcc7e
10 changed files with 28 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
cuda/cvectordev.c
+3 -3
View File
@@ -184,7 +184,7 @@ int iscatMultiVecDeviceFloatComplex(void* deviceVec, int vectorId, int n,
}
int nrm2MultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devMultiVecA)
int nrm2MultiVecDeviceFloatComplex(float* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
@@ -194,7 +194,7 @@ int nrm2MultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devMultiV
return(i);
}
int amaxMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devMultiVecA)
int amaxMultiVecDeviceFloatComplex(float* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
@@ -204,7 +204,7 @@ int amaxMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devMultiV
return(i);
}
int asumMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devMultiVecA)
int asumMultiVecDeviceFloatComplex(float* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
cuda/cvectordev.h
+4 -3
View File
@@ -37,6 +37,7 @@
#include "vectordev.h"
#include "cuda_runtime.h"
#include "core.h"
#include "vector.h"
int registerMappedFloatComplex(void *, void **, int, cuFloatComplex);
int writeMultiVecDeviceFloatComplex(void* deviceMultiVec, cuFloatComplex* hostMultiVec);
@@ -63,9 +64,9 @@ int iscatMultiVecDeviceFloatComplex(void* deviceVec, int vectorId, int n, int fi
int hfirst, void* host_values, int indexBase, cuFloatComplex beta);
int scalMultiVecDeviceFloatComplex(cuFloatComplex alpha, void* devMultiVecA);
int nrm2MultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devVecA);
int amaxMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devVecA);
int asumMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devVecA);
int nrm2MultiVecDeviceFloatComplex(float* y_res, int n, void* devVecA);
int amaxMultiVecDeviceFloatComplex(float* y_res, int n, void* devVecA);
int asumMultiVecDeviceFloatComplex(float* y_res, int n, void* devVecA);
int dotMultiVecDeviceFloatComplex(cuFloatComplex* y_res, int n, void* devVecA, void* devVecB);
int axpbyMultiVecDeviceFloatComplex(int n, cuFloatComplex alpha, void* devVecX, cuFloatComplex beta, void* devVecY);
cuda/dnsdev.c
+8 -8
View File
@@ -178,12 +178,12 @@ int spmvDnsDeviceFloatComplex(char transa, int m, int n, int k, float complex *a
/* Note: the M,N,K choices according to TRANS have already been handled in the caller */
if (n == 1) {
status = cublasCgemv(handle, trans, m,k,
alpha, devMat->cM,devMat->pitch, x->v_,1,
beta, y->v_,1);
(const cuComplex *) alpha, devMat->cM,devMat->pitch, x->v_,1,
(const cuComplex *) beta, y->v_,1);
} else {
status = cublasCgemm(handle, trans, CUBLAS_OP_N, m,n,k,
alpha, devMat->cM,devMat->pitch, x->v_,x->pitch_,
beta, y->v_,y->pitch_);
(const cuComplex *) alpha, devMat->cM,devMat->pitch, x->v_,x->pitch_,
(const cuComplex *) beta, y->v_,y->pitch_);
}
if (status == CUBLAS_STATUS_SUCCESS)
@@ -205,12 +205,12 @@ int spmvDnsDeviceDoubleComplex(char transa, int m, int n, int k, double complex
/* Note: the M,N,K choices according to TRANS have already been handled in the caller */
if (n == 1) {
status = cublasZgemv(handle, trans, m,k,
alpha, devMat->cM,devMat->pitch, x->v_,1,
beta, y->v_,1);
(const cuDoubleComplex *) alpha, devMat->cM,devMat->pitch, x->v_,1,
(const cuDoubleComplex *) beta, y->v_,1);
} else {
status = cublasZgemm(handle, trans, CUBLAS_OP_N, m,n,k,
alpha, devMat->cM,devMat->pitch, x->v_,x->pitch_,
beta, y->v_,y->pitch_);
(const cuDoubleComplex *) alpha, devMat->cM,devMat->pitch, x->v_,x->pitch_,
(const cuDoubleComplex *) beta, y->v_,y->pitch_);
}
if (status == CUBLAS_STATUS_SUCCESS)
cuda/dvectordev.h
+1
View File
@@ -35,6 +35,7 @@
#include "vectordev.h"
#include "cuda_runtime.h"
#include "core.h"
#include "vector.h"
int registerMappedDouble(void *, void **, int, double);
int writeMultiVecDeviceDouble(void* deviceMultiVec, double* hostMultiVec);
cuda/fcusparse_fct.h
+2 -2
View File
@@ -439,7 +439,7 @@ int T_CSRGDeviceSetMatFillMode(T_Cmat *Matrix, int type)
T_CSRGDeviceMat *cMat= Matrix->mat;
cusparseFillMode_t mode=type;
CHECK_CUSPARSE(cusparseSpMatSetAttribute(cMat->spmvDescr,
CHECK_CUSPARSE(cusparseSpMatSetAttribute((*(cMat->spmvDescr)),
CUSPARSE_SPMAT_FILL_MODE,
(const void*) &mode,
sizeof(cusparseFillMode_t)));
@@ -450,7 +450,7 @@ int T_CSRGDeviceSetMatDiagType(T_Cmat *Matrix, int type)
{
T_CSRGDeviceMat *cMat= Matrix->mat;
cusparseDiagType_t cutype=type;
CHECK_CUSPARSE(cusparseSpMatSetAttribute(cMat->spmvDescr,
CHECK_CUSPARSE(cusparseSpMatSetAttribute((*(cMat->spmvDescr)),
CUSPARSE_SPMAT_DIAG_TYPE,
(const void*) &cutype,
sizeof(cusparseDiagType_t)));
cuda/ivectordev.h
+1
View File
@@ -35,6 +35,7 @@
#include "vectordev.h"
#include "cuda_runtime.h"
#include "core.h"
#include "vector.h"
int registerMappedInt(void *, void **, int, int);
int writeMultiVecDeviceInt(void* deviceMultiVec, int* hostMultiVec);
cuda/svectordev.h
+1
View File
@@ -35,6 +35,7 @@
#include "vectordev.h"
#include "cuda_runtime.h"
#include "core.h"
#include "vector.h"
int registerMappedFloat(void *, void **, int, float);
int writeMultiVecDeviceFloat(void* deviceMultiVec, float* hostMultiVec);
cuda/vectordev.h
+1
View File
@@ -34,6 +34,7 @@
#include "cuda_runtime.h"
//#include "common.h"
//#include "cintrf.h"
#include "cuda_util.h"
#include <complex.h>
struct MultiVectDevice
cuda/zvectordev.c
+3 -3
View File
@@ -183,7 +183,7 @@ int iscatMultiVecDeviceDoubleComplex(void* deviceVec, int vectorId, int n,
}
int nrm2MultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devMultiVecA)
int nrm2MultiVecDeviceDoubleComplex(double* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
@@ -192,7 +192,7 @@ int nrm2MultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devMult
return(i);
}
int amaxMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devMultiVecA)
int amaxMultiVecDeviceDoubleComplex(double* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
@@ -202,7 +202,7 @@ int amaxMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devMult
return(i);
}
int asumMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devMultiVecA)
int asumMultiVecDeviceDoubleComplex(double* y_res, int n, void* devMultiVecA)
{ int i=0;
spgpuHandle_t handle=psb_cudaGetHandle();
struct MultiVectDevice *devVecA = (struct MultiVectDevice *) devMultiVecA;
cuda/zvectordev.h
+4 -3
View File
@@ -37,6 +37,7 @@
#include "vectordev.h"
#include "cuda_runtime.h"
#include "core.h"
#include "vector.h"
int registerMappedDoubleComplex(void *, void **, int, cuDoubleComplex);
int writeMultiVecDeviceDoubleComplex(void* deviceMultiVec, cuDoubleComplex* hostMultiVec);
@@ -69,9 +70,9 @@ int iscatMultiVecDeviceDoubleComplex(void* deviceVec, int vectorId, int n,
int indexBase, cuDoubleComplex beta);
int scalMultiVecDeviceDoubleComplex(cuDoubleComplex alpha, void* devMultiVecA);
int nrm2MultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devVecA);
int amaxMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devVecA);
int asumMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n, void* devVecA);
int nrm2MultiVecDeviceDoubleComplex(double* y_res, int n, void* devVecA);
int amaxMultiVecDeviceDoubleComplex(double* y_res, int n, void* devVecA);
int asumMultiVecDeviceDoubleComplex(double* y_res, int n, void* devVecA);
int dotMultiVecDeviceDoubleComplex(cuDoubleComplex* y_res, int n,
void* devVecA, void* devVecB);