Closed mgates3 closed 1 year ago
@ayarkhan Since there are so few instances of "oneMKL" currently in SLATE, it's probably easiest for you to fix those up in your upcoming PR.
slate> git grep -i onemkl
.github/ISSUE_TEMPLATE/bug_report.md: - CUDA / ROCm / oneMKL version (e.g., `nvcc --version`):
GNUmakefile:ifneq ($(filter-out auto cuda hip onemkl none, $(gpu_backend)),)
GNUmakefile: ifeq (${gpu_backend},onemkl)
GNUmakefile: $(info Note: enabling onemkl)
GNUmakefile: @echo "---------- oneMKL options"
include/slate/internal/device.hh:// For the present, use omptarget-kernels when OneMKL is used
include/slate/internal/device.hh:#if defined( BLAS_HAVE_ONEMKL )
include/slate/internal/device.hh:// Simplify checking for GPU device support (CUDA / ROCm / OneMKL).
oneMKL is a bit confusing, because the Intel oneMKL library includes both traditional CPU BLAS/LAPACK and new GPU oneapi functions. For instance, BLAS_HAVE_MKL and BLAS_HAVE_ONEMKL mean distinctly different things, CPU and GPU support respectively. Rename oneMKL to "SYCL" for flags to clarify that we mean the GPU version, not the CPU version.