CC ?= clang
CFLAGS = -Ofast -Wno-unused-result -Wno-ignored-pragmas -Wno-unknown-attributes
LDFLAGS =
LDLIBS = -lm
INCLUDES =
CFLAGS_COND = -march=native
# Find nvcc
SHELL_UNAME = $(shell uname)
REMOVE_FILES = rm -f
OUTPUT_FILE = -o $@
CUDA_OUTPUT_FILE = -o $@
# NVCC flags
# -t=0 is short for --threads, 0 = number of CPUs on the machine
NVCC_FLAGS = -O3 -t=0 --use_fast_math
NVCC_LDFLAGS = -lcublas -lcublasLt
NVCC_INCLUDES =
NVCC_LDLIBS =
NCLL_INCUDES =
NVCC_CUDNN =
# overridable flag for multi-GPU training. by default we won't build with cudnn
# because it bloats up the compile time from a few seconds to ~minute
US