cublas.md

cuBLAS Functions

Not template but overload

Example

#include <cutf/type.hpp>
#include <cutf/memory.hpp>
#include <cutf/cublas.hpp>
constexpr std::size_t N = 1<<10;

using T = half;
int main(){
	auto cublas_handle = cutf::cublas::get_ublas_unique_ptr();
	auto A = cutf::memory::get_device_unique_ptr<T>(N * N);
	auto B = cutf::memory::get_device_unique_ptr<T>(N * N);
	auto C = cutf::memory::get_device_unique_ptr<T>(N * N);
	T alpha = cutf::type::cast<T>(1.0f);
	T beta = cutf::type::cast<T>(1.0f);

	const auto status = cutf::cublas::gemm(*cublas_handle.get(),
			CUBLAS_OP_N, CUBLAS_OP_N,
			N, N, N,
			&alpha,
			A.get(), N,
			B.get(), N,
			&beta,
			C.get(), N);

}

Implemented operations

Operation	cutf::cublas::	half	float	double	cuComplex	cuDoubleComplex
amax	iamax		S	D	C	Z
amin	iamin		S	D	C	Z
asum	asum		S	D	C	Z
axpy	axpy		S	D	C	Z
copy	copy		S	D	C	Z
dot	dot		S	D
dotc	dotc				C	Z
dotu	dotu				C	Z
nrm2	nrm2		S	D	Sc	Dz
rot	rot		S	D	C/Cs	Z/Zd
rotg	rotg		S	D	C	Z
rotm	rotm		S	D
rotmg	rotmg		S	D
scal	scal		S	D	C/Cs	Z/Zd
swap	swap		S	D	C	Z
gbmv	gbmv		S	D	C	Z
gemv	gemv		S	D	C	Z
ger	ger		S	D
gerc	gerc				C	Z
geru	geru				C	Z
sbmv	sbmv		S	D
spmv	spmv		S	D
spr	spr		S	D
spr2	spr2		S	D
symv	symv		S	D	C	Z
syr	syr		S	D	C	Z
syr2	syr2		S	D	C	Z
tbmv	tbmv		S	D	C	Z
tbsv	tbsv		S	D	C	Z
tpmv	tpmv		S	D	C	Z
tpsv	tpsv		S	D	C	Z
trmv	trmv		S	D	C	Z
trsv	trsv		S	D	C	Z
hemv	hemv				C	Z
hbmv	hbmv				C	Z
hpmv	hpmv				C	Z
her	her				C	Z
her2	her2				C	Z
hpr	hpr				C	Z
hpr2	hpr2				C	Z
gemm	gemm	H	S	D	C	Z
gemm3m	gemm3m		S	D	C	Z
gemmBatched	gemm_batched	H	S	D	C	Z
gemmStridedBatched	gemm_strided_batched	H	S	D	C	Z
gemm3mStridedBatched	gemm3m_strided_batched		S	D	C	Z
symm	symm		S	D	C	Z
syrk	syrk		S	D	C	Z
syr2k	syr2k		S	D	C	Z
syrkx	syrkx		S	D	C	Z
trmm	trmm		S	D	C	Z
trsm	trsm		S	D	C	Z
trsmBatched	trsm_batched		S	D	C	Z
hemm	hemm				C	Z
herk	herk				C	Z
her2k	her2k				C	Z
herkx	herkx				C	Z
geam	geam		S	D	C	Z
dgmm	dgmm		S	D	C	Z
getrfBatched	getrf_batched		S	D	C	Z
getrsBatched	getrs_batched		S	D	C	Z
getriBatched	getri_batched		S	D	C	Z
matinvBatched	matinv_batched		S	D	C	Z
geqrfBatched	geqrf_batched		S	D	C	Z
gelsBatched	gels_batched		S	D	C	Z
tpttr	tpttr		S	D	C	Z
trttp	trttp		S	D	C	Z

Reference

• CUDA Toolkit Documentation