Skip to content

Add support for multi-arch CUDA binaries #2001

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Sbozzolo wants to merge 1 commit into
base: main
Choose a base branch
from
+20 −2
Open

Add support for multi-arch CUDA binaries #2001

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 14 additions & 2 deletions Makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -45,6 +45,7 @@ XSMM_DIR ?= ../libxsmm
# Often /opt/cuda or /usr/local/cuda, but sometimes present on machines that don't support CUDA
CUDA_DIR ?=
CUDA_ARCH ?=
CUDA_TARGETS ?=

# Often /opt/rocm, but sometimes present on machines that don't support HIP
ROCM_DIR ?=
Expand Down Expand Up @@ -188,7 +189,18 @@ CXXFLAGS ?= $(OPT) $(CXXFLAGS.$(CC_VENDOR)) $(if $(PEDANTIC),$(PEDANTICFLAGS))
FFLAGS ?= $(OPT) $(FFLAGS.$(FC_VENDOR))
LIBCXX ?= -lstdc++
NVCCFLAGS ?= -ccbin $(CXX) -Xcompiler '$(OPT)' -Xcompiler -fPIC
ifneq ($(CUDA_ARCH),)
CUDA_TARGETS_UNKNOWN := $(filter-out sm_%,$(CUDA_TARGETS))
CUDA_SMS := $(patsubst sm_%,%,$(filter sm_%,$(CUDA_TARGETS)))
CUDA_SMS := $(shell printf "%s\n" $(CUDA_SMS) | sort -n)
cuda_gencode_sm = -gencode arch=compute_$(1),code=sm_$(1)
cuda_gencode_compute = -gencode arch=compute_$(1),code=compute_$(1)
ifneq ($(strip $(CUDA_TARGETS)),)
ifneq ($(strip $(CUDA_TARGETS_UNKNOWN)),)
$(error Unknown CUDA target(s): $(CUDA_TARGETS_UNKNOWN); expected targets such as sm_80)
endif
NVCCFLAGS += $(foreach sm,$(CUDA_SMS),$(call cuda_gencode_sm,$(sm))) \
$(call cuda_gencode_compute,$(lastword $(CUDA_SMS)))
else ifneq ($(strip $(CUDA_ARCH)),)
NVCCFLAGS += -arch=$(CUDA_ARCH)
endif
HIPCCFLAGS ?= $(filter-out $(OMP_SIMD_FLAG),$(OPT)) -fPIC -munsafe-fp-atomics
Expand Down Expand Up @@ -1049,7 +1061,7 @@ print-% :
CONFIG_VARS = CC CXX FC NVCC NVCC_CXX HIPCC \
OPT CFLAGS CPPFLAGS CXXFLAGS FFLAGS NVCCFLAGS HIPCCFLAGS SYCLFLAGS \
AR ARFLAGS LDFLAGS LDLIBS LIBCXX SED \
MAGMA_DIR XSMM_DIR CUDA_DIR CUDA_ARCH MFEM_DIR PETSC_DIR NEK5K_DIR ROCM_DIR HIP_ARCH SYCL_DIR
MAGMA_DIR XSMM_DIR CUDA_DIR CUDA_ARCH CUDA_TARGETS MFEM_DIR PETSC_DIR NEK5K_DIR ROCM_DIR HIP_ARCH SYCL_DIR

# $(call needs_save,CFLAGS) returns true (a nonempty string) if CFLAGS
# was set on the command line or in config.mk (where it will appear as
Expand Down
6 changes: 6 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -62,6 +62,12 @@ $ make AVX=0
if your compiler does not support gcc-style options, if you are cross compiling, etc.

To enable CUDA support, add `CUDA_DIR=/opt/cuda` or an appropriate directory to your `make` invocation.
To target one or more CUDA architectures, pass a space-separated `CUDA_TARGETS` list:

```console
$ make CUDA_DIR=/opt/cuda CUDA_TARGETS='sm_80 sm_90'
```

To enable HIP support, add `ROCM_DIR=/opt/rocm` or an appropriate directory.
To enable SYCL support, add `SYCL_DIR=/opt/sycl` or an appropriate directory.
Note that SYCL backends require building with oneAPI compilers as well:
Expand Down
Loading