mirror of
https://github.com/NVIDIA/nccl-tests.git
synced 2026-01-14 02:47:21 +08:00
51af5572bf
* Added "verifiable", a suite of kernels for generating and verifying reduction input and output arrays in a bit-precise way. * Data corruption errors now reported in number of wrong elements instead of max deviation. * Use ncclGetLastError. * Don't run hypercube on non-powers of 2 ranks. * Fix to hypercube data verification. * Use "thread local" as the defaut CUDA capture mode. * Replaced pthread_yield -> sched_yield() * Bugfix to the cpu-side barrier/allreduce implementations.
25 lines
569 B
Makefile
25 lines
569 B
Makefile
include ../../makefiles/common.mk
|
|
|
|
.PHONY: all clean
|
|
|
|
BUILDDIR := $(abspath ../../build)
|
|
NCCLDIR := $(BUILDDIR)
|
|
NVCUFLAGS += -I$(NCCLDIR)/include/ -I../include
|
|
DST_DIR := $(BUILDDIR)/test/verifiable
|
|
|
|
all: $(DST_DIR)/self_test $(DST_DIR)/verifiable.o
|
|
|
|
clean:
|
|
rm -rf $(DST_DIR)
|
|
|
|
TEST_VERIFIABLE_SRCDIR := .
|
|
TEST_VERIFIABLE_BUILDDIR := $(DST_DIR)
|
|
include verifiable.mk
|
|
|
|
self_test: $(DST_DIR)/self_test
|
|
|
|
$(DST_DIR)/self_test: verifiable.cu verifiable.h
|
|
@printf "Linking %s\n" $@
|
|
@mkdir -p $(DST_DIR)
|
|
$(NVCC) -o $@ $(NVCUFLAGS) -DSELF_TEST=1 verifiable.cu $(NVLDFLAGS)
|