Function cusparseSpMMOp 

pub unsafe extern "C" fn cusparseSpMMOp(
    plan: cusparseSpMMOpPlan_t,
    externalBuffer: *mut c_void,
) -> cusparseStatus_t

NOTE 1: NVRTC and nvJitLink are not currently available on Arm64 Android platforms.

NOTE 2: The routine does not support Android and Tegra platforms except Judy (sm87).

Experimental: The function performs the multiplication of a sparse matrix matA and a dense matrix matB with custom operators.

where

• op(A) is a sparse matrix of size $m \times k$
• op(B) is a dense matrix of size $k \times n$
• C is a dense matrix of size $m \times n$
• $\oplus$, $\otimes$, and $\text{epilogue}$ are custom add, mul, and epilogue operators respectively.

Also, for matrix A and B

Only opA == CUSPARSE_OPERATION_NON_TRANSPOSE is currently supported

The function cusparseSpMMOp_createPlan returns the size of the workspace and the compiled kernel needed by cusparseSpMMOp

The operators must have the following signature and return type

<computetype> is one of float, double, cuComplex, cuDoubleComplex, or int,

cusparseSpMMOp supports the following sparse matrix formats:

• cusparseFormat_t::CUSPARSE_FORMAT_CSR

cusparseSpMMOp supports the following index type for representing the sparse matrix matA:

• 32-bit indices (cusparseIndexType_t::CUSPARSE_INDEX_32I)
• 64-bit indices (cusparseIndexType_t::CUSPARSE_INDEX_64I)

cusparseSpMMOp supports the following data types:

Uniform-precision computation:

A/B/ C/computeType
cudaDataType_t::CUDA_R_32F
cudaDataType_t::CUDA_R_64F
cudaDataType_t::CUDA_C_32F
cudaDataType_t::CUDA_C_64F

Mixed-precision computation:

A/B	C	computeType
cudaDataType_t::CUDA_R_8I	cudaDataType_t::CUDA_R_32I	cudaDataType_t::CUDA_R_32I
cudaDataType_t::CUDA_R_8I	cudaDataType_t::CUDA_R_32F	cudaDataType_t::CUDA_R_32F
cudaDataType_t::CUDA_R_16F
cudaDataType_t::CUDA_R_16BF
cudaDataType_t::CUDA_R_16F	cudaDataType_t::CUDA_R_16F
cudaDataType_t::CUDA_R_16BF	cudaDataType_t::CUDA_R_16BF

cusparseSpMMOp supports the following algorithms:

Algorithm	Notes
cusparseSpMMOpAlg_t::CUSPARSE_SPMM_OP_ALG_DEFAULT	Default algorithm for any sparse matrix format

Performance notes:

• Row-major layout provides higher performance than column-major.

cusparseSpMMOp has the following properties:

• The routine requires extra storage
• The routine supports asynchronous execution
• Provides deterministic (bit-wise) results for each run
• The routine allows the indices of matA to be unsorted

cusparseSpMMOp supports the following optimizations:

• CUDA graph capture
• Hardware Memory Compression

Please visit cuSPARSE Library Samples - cusparseSpMMOp.