blob: 261eb4a5166e4e61ca373be8d8c078549f6f2aee (
plain)
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
|
GCC = gfortran
oneAPI = ifx
# These flags will be looped over to generate different datasets
FLAGS := O0 O1 O2 O3 Ofast
# Also specify a default (note the conditional assignment operation, ?=)
OPTFLAGS ?= O3
# For the parallel runs, use all available processors
NPROC := $(shell nproc)
# Make some pretty colours
MAGENTA := \033[1;35m
RESET := \033[0m
# Define the start, stop, and step for "short" runs (including triple loops)
S_START:= 100
S_END := 300
S_STEP := 100
# And the step and end for "long" runs without triple loops.
# Automatically they start at S_END defined earlier
L_END := 800
L_STEP := 500
all:
# This target simply compiles the binaries
@mkdir -p bin/
@echo "$(MAGENTA)Compiling serial and parallel binaries with $(OPTFLAGS)$(RESET)"
$(GCC) matrixproduct.f90 -o bin/$(GCC).serial.out -$(OPTFLAGS) -fexternal-blas -lopenblas -march=native
$(oneAPI) matrixproduct.f90 -o bin/$(oneAPI).serial.out -qmkl=sequential -$(OPTFLAGS) -heap-arrays -xHost
$(GCC) matrixproduct.f90 -o bin/$(GCC).parallel.out -$(OPTFLAGS) -fexternal-blas -lopenblas -march=native -fopenmp
$(oneAPI) matrixproduct.f90 -o bin/$(oneAPI).parallel.out -qmkl=parallel -$(OPTFLAGS) -heap-arrays -xHost -fopenmp
tests: clean
# Run the parametric tests and sweep over compiler flags
@mkdir -p results/
@for opt in $(FLAGS); do \
$(MAKE) all OPTFLAGS=$$opt; \
echo "$(MAGENTA)Running serial runs with $$opt$(RESET)"; \
$(MAKE) serial OPTFLAGS=$$opt; \
echo "$(MAGENTA)Running parallel runs with $$opt$(RESET)"; \
$(MAKE) parallel OPTFLAGS=$$opt; \
done
## NOTE: For the below two, it is very important that lines have the '\' postfix
## By default, these are all executed on a different shell session, so to preserve
## the enivronment variables for OMP/MKL they need to be run as one "big" command
## serially. The backslash is the 'line continuation' operator that does this.
serial: ./bin/gfortran.serial.out ./bin/ifx.serial.out
# run the serial tests. ensures that only one thread is used for blas and mkl
export OMP_NUM_THREADS=1 && \
export MKL_NUM_THREADS=1 && \
./bin/$(GCC).serial.out $(S_START) $(S_END) $(S_STEP) yes | tee results/$(GCC)_short_serial_$(OPTFLAGS).out; \
./bin/$(oneAPI).serial.out $(S_START) $(S_END) $(S_STEP) yes | tee results/$(oneAPI)_short_serial_$(OPTFLAGS).out; \
./bin/$(GCC).serial.out $(S_END) $(L_END) $(L_STEP) no | tee results/$(GCC)_long_serial_$(OPTFLAGS).out; \
./bin/$(oneAPI).serial.out $(S_END) $(L_END) $(L_STEP) no | tee results/$(oneAPI)_long_serial_$(OPTFLAGS).out;
parallel: ./bin/gfortran.parallel.out ./bin/ifx.parallel.out
# run the parallel tests. ensures that all threads are used for the loops, blas, and mkl
export OMP_NUM_THREADS=1 && \
export OMP_NUM_THREADS=${NPROC} && \
export MKL_NUM_THREADS=${NPROC} && \
./bin/$(GCC).parallel.out $(S_START) $(S_END) $(S_STEP) yes | tee results/$(GCC)_short_parallel_$(OPTFLAGS)_$(NPROC).out; \
./bin/$(oneAPI).parallel.out $(S_START) $(S_END) $(S_STEP) yes | tee results/$(oneAPI)_short_parallel_$(OPTFLAGS)_$(NPROC).out; \
./bin/$(GCC).parallel.out $(S_END) $(L_END) $(L_STEP) no | tee results/$(GCC)_long_parallel_$(OPTFLAGS)_$(NPROC).out; \
./bin/$(oneAPI).parallel.out $(S_END) $(L_END) $(L_STEP) no | tee results/$(oneAPI)_long_parallel_$(OPTFLAGS)_$(NPROC).out;
plots:
# Deprecated. Plot some things in GNUPLOT. Currently unused.
gnuplot -p plots.gnu
clean:
# Clean up after ourselves
@echo "$(MAGENTA)Cleaning bin/ and results/$(RESET)"
@rm -rf bin/
@rm -rf results/
|
