-
Notifications
You must be signed in to change notification settings - Fork 18
/
Copy pathMakefile
182 lines (142 loc) · 7.3 KB
/
Makefile
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
.PHONY: all clean benchmarks directories
CFLAGS := -Wall -Werror -O3 -g -fPIC
# Uncomment this line and set the correct path to use SM/TPC partitioning
# library `libsmctrl`.
#LIBSMCTRL_PATH = ../libsmctrl
ifdef LIBSMCTRL_PATH
LDFLAGS += -L$(LIBSMCTRL_PATH)
LDLIBS += -lsmctrl -lcuda
CFLAGS += -I$(LIBSMCTRL_PATH) -DSMCTRL
endif
NVCC ?= /usr/local/cuda/bin/nvcc
NVCCFLAGS := -g --ptxas-options=-v --compiler-options="$(CFLAGS)"
# There was no easy way to autodetect what targets NVCC supported prior to CUDA
# 11.5, so a lookup table is hardcoded below.
# ***If a target is listed below, that does not mean we support it.***
CUDA_VER=$(shell $(NVCC) --version | sed -n 's/^.*release \([0-9]\+\.[0-9]\+\).*$$/\1/p' | tr -d '.')
# Tesla supported up through CUDA 6.5
ifeq ($(shell expr \( $(CUDA_VER) \>= 55 \) \& \( $(CUDA_VER) \<= 65 \)), 1)
NVCCFLAGS += --generate-code arch=compute_10,code=[compute_10,sm_10]
NVCCFLAGS += --generate-code arch=compute_11,code=[compute_11,sm_11]
NVCCFLAGS += --generate-code arch=compute_12,code=[compute_12,sm_12]
NVCCFLAGS += --generate-code arch=compute_13,code=[compute_13,sm_13]
endif
# Fermi supported up through CUDA 8
ifeq ($(shell expr \( $(CUDA_VER) \>= 55 \) \& \( $(CUDA_VER) \<= 80 \)), 1)
NVCCFLAGS += --generate-code arch=compute_20,code=[compute_20,sm_20]
endif
# Kepler (first gen) supported up through CUDA 10
ifeq ($(shell expr \( $(CUDA_VER) \>= 55 \) \& \( $(CUDA_VER) \<= 100 \)), 1)
NVCCFLAGS += --generate-code arch=compute_30,code=[compute_30,sm_30]
NVCCFLAGS += --generate-code arch=compute_32,code=[compute_32,sm_32]
endif
ifeq ($(shell expr \( $(CUDA_VER) \>= 55 \) \& \( $(CUDA_VER) \<= 115 \)), 1)
NVCCFLAGS += --generate-code arch=compute_35,code=[compute_35,sm_35]
endif
ifeq ($(shell expr \( $(CUDA_VER) \>= 70 \) \& \( $(CUDA_VER) \< 115 \)), 1)
NVCCFLAGS += --generate-code arch=compute_50,code=[compute_50,sm_50]
NVCCFLAGS += --generate-code arch=compute_52,code=[compute_52,sm_52]
NVCCFLAGS += --generate-code arch=compute_53,code=[compute_53,sm_53]
endif
ifeq ($(shell expr \( $(CUDA_VER) \>= 90 \) \& \( $(CUDA_VER) \< 115 \)), 1)
NVCCFLAGS += --generate-code arch=compute_60,code=[compute_60,sm_60]
NVCCFLAGS += --generate-code arch=compute_61,code=[compute_61,sm_61]
NVCCFLAGS += --generate-code arch=compute_62,code=[compute_62,sm_62]
NVCCFLAGS += --generate-code arch=compute_70,code=[compute_70,sm_70]
endif
ifeq ($(shell expr \( $(CUDA_VER) \>= 100 \) \& \( $(CUDA_VER) \< 115 \)), 1)
NVCCFLAGS += --generate-code arch=compute_72,code=[compute_72,sm_72]
NVCCFLAGS += --generate-code arch=compute_75,code=[compute_75,sm_75]
endif
ifeq ($(shell expr \( $(CUDA_VER) \>= 112 \) \& \( $(CUDA_VER) \< 115 \)), 1)
NVCCFLAGS += --generate-code arch=compute_80,code=[compute_80,sm_80]
NVCCFLAGS += --generate-code arch=compute_86,code=[compute_86,sm_86]
endif
# Starting with CUDA 11.5, nvcc supports auto-detection with -arch=all
ifeq ($(shell expr $(CUDA_VER) \>= 115), 1)
NVCCFLAGS += --gpu-architecture all
endif
BENCHMARK_DEPENDENCIES := src/library_interface.h \
src/benchmark_gpu_utilities.h obj/benchmark_gpu_utilities.o
all: directories benchmarks bin/runner
benchmarks: bin/mandelbrot.so bin/timer_spin.so bin/multikernel.so \
bin/cpu_inorder_walk.so bin/cpu_random_walk.so bin/inorder_walk.so \
bin/random_walk.so bin/sharedmem_timer_spin.so bin/counter_spin.so \
bin/timer_spin_default_stream.so bin/stream_action.so bin/task_host.so \
bin/matrix_multiply.so
directories:
mkdir -p bin/
mkdir -p obj/
bin/mandelbrot.so: src/mandelbrot.cu $(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/mandelbrot.so \
src/mandelbrot.cu obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/timer_spin.so: src/timer_spin.cu $(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/timer_spin.so \
src/timer_spin.cu obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/counter_spin.so: src/counter_spin.cu $(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/counter_spin.so \
src/counter_spin.cu obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/timer_spin_default_stream.so: src/timer_spin_default_stream.cu \
$(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o \
bin/timer_spin_default_stream.so src/timer_spin_default_stream.cu \
obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/multikernel.so: src/multikernel.cu $(BENCHMARK_DEPENDENCIES) obj/cjson.o
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/multikernel.so \
src/multikernel.cu obj/benchmark_gpu_utilities.o obj/cjson.o $(LDLIBS)
bin/inorder_walk.so: src/inorder_walk.cu $(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/inorder_walk.so \
src/inorder_walk.cu obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/random_walk.so: src/random_walk.cu $(BENCHMARK_DEPENDENCIES)
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/random_walk.so \
src/random_walk.cu obj/benchmark_gpu_utilities.o $(LDLIBS)
bin/sharedmem_timer_spin.so: src/sharedmem_timer_spin.cu \
$(BENCHMARK_DEPENDENCIES) obj/cjson.o
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/sharedmem_timer_spin.so \
src/sharedmem_timer_spin.cu obj/benchmark_gpu_utilities.o obj/cjson.o \
$(LDLIBS)
bin/stream_action.so: src/stream_action.cu $(BENCHMARK_DEPENDENCIES) \
obj/cjson.o
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/stream_action.so \
src/stream_action.cu obj/benchmark_gpu_utilities.o obj/cjson.o $(LDLIBS)
bin/matrix_multiply.so: src/matrix_multiply.cu $(BENCHMARK_DEPENDENCIES) \
obj/cjson.o
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/matrix_multiply.so \
src/matrix_multiply.cu obj/benchmark_gpu_utilities.o obj/cjson.o \
$(LDLIBS)
bin/cpu_inorder_walk.so: src/cpu_inorder_walk.c src/library_interface.h
gcc $(CFLAGS) $(LDFLAGS) -shared -o bin/cpu_inorder_walk.so \
src/cpu_inorder_walk.c $(LDLIBS)
bin/cpu_random_walk.so: src/cpu_random_walk.c src/library_interface.h
gcc $(CFLAGS) $(LDFLAGS) -shared -o bin/cpu_random_walk.so \
src/cpu_random_walk.c $(LDLIBS)
obj/cjson.o: src/third_party/cJSON.c src/third_party/cJSON.h
gcc -c $(CFLAGS) -o obj/cjson.o src/third_party/cJSON.c
obj/parse_config.o: src/parse_config.c src/parse_config.h \
src/third_party/cJSON.h
gcc -c $(CFLAGS) -o obj/parse_config.o src/parse_config.c
obj/runner.o: src/runner.c src/library_interface.h
gcc -c $(CFLAGS) -o obj/runner.o src/runner.c
obj/task_host.o: src/task_host.c src/third_party/cJSON.h \
src/library_interface.h
gcc -c $(CFLAGS) -o obj/task_host.o src/task_host.c
obj/task_host_utilities.o: src/task_host_utilities.cu \
src/task_host_utilities.h src/library_interface.h
$(NVCC) -c $(NVCCFLAGS) -o obj/task_host_utilities.o \
src/task_host_utilities.cu
obj/benchmark_gpu_utilities.o: src/benchmark_gpu_utilities.cu \
src/benchmark_gpu_utilities.h
$(NVCC) -c $(NVCCFLAGS) -o obj/benchmark_gpu_utilities.o \
src/benchmark_gpu_utilities.cu
obj/barrier_wait.o: src/barrier_wait.c src/barrier_wait.h
gcc -c $(CFLAGS) -o obj/barrier_wait.o src/barrier_wait.c
bin/runner: obj/runner.o bin/task_host.so
gcc $(LDFLAGS) -o bin/runner obj/runner.o -ldl $(LDLIBS)
bin/task_host.so: obj/task_host.o obj/cjson.o obj/parse_config.o \
obj/task_host_utilities.o obj/barrier_wait.o
$(NVCC) $(LDFLAGS) --shared $(NVCCFLAGS) -o bin/task_host.so \
obj/task_host.o obj/cjson.o obj/task_host_utilities.o \
obj/barrier_wait.o obj/parse_config.o -lpthread -ldl -lm $(LDLIBS)
clean:
rm -f bin/*
rm -f obj/*