aparler/c-llm
1
0
Fork 0
mirror of https://github.com/karpathy/llm.c.git synced 2026-07-28 20:35:09 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

don't compile/link cudnn if not asked for it

Browse source
This commit is contained in:
Erik Schultheis 2024-05-05 02:44:15 +03:00
parent 19c290d7e6
commit b087b9c819
1 changed files with 8 additions and 6 deletions
Download patch file Download diff file Expand all files Collapse all files

14
Makefile
View file

@ -18,6 +18,7 @@ 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
USE_CUDNN ?= 0
@ -81,6 +82,7 @@ ifeq ($(USE_CUDNN), 1)
NVCC_INCLUDES += -I$(CUDNN_FRONTEND_PATH)
NVCC_LDFLAGS += -lcudnn
NVCC_FLAGS += -DENABLE_CUDNN
NVCC_CUDNN = cudnn_att.o
else
$(error ✗ cuDNN not found. See the Makefile for our currently hard-coded paths / install instructions)
endif
@ -197,20 +199,20 @@ test_gpt2: test_gpt2.c
cudnn_att.o: cudnn_att.cu
$(NVCC) -c $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS)
train_gpt2cu: train_gpt2.cu cudnn_att.o
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) cudnn_att.o
train_gpt2cu: train_gpt2.cu $(NVCC_CUDNN)
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) $(NVCC_CUDNN)
train_gpt2fp32cu: train_gpt2_fp32.cu
$(NVCC) $(NVCC_FLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
test_gpt2cu: test_gpt2.cu cudnn_att.o
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) cudnn_att.o
test_gpt2cu: test_gpt2.cu $(NVCC_CUDNN)
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) $(NVCC_CUDNN)
test_gpt2fp32cu: test_gpt2_fp32.cu
$(NVCC) $(NVCC_FLAGS) $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE)
profile_gpt2cu: profile_gpt2.cu cudnn_att.o
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) -lineinfo $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) cudnn_att.o
profile_gpt2cu: profile_gpt2.cu $(NVCC_CUDNN)
$(NVCC) $(NVCC_FLAGS) $(PFLAGS) -lineinfo $< $(NVCC_LDFLAGS) $(NVCC_INCLUDES) $(NVCC_LDLIBS) $(CUDA_OUTPUT_FILE) $(NVCC_CUDNN)
clean:
$(REMOVE_FILES) $(TARGETS)