Skip to content

Commit

Permalink
Adds stand alone test_MOM_EOS and time_MOM_EOS (mom-ocean#516)
Browse files Browse the repository at this point in the history
- Added simple single-thread program to invoke EOS_unit_tests.F90
- Added not-as-simple program to time the cost of calculate_density()
  and calculate_spec_vol() for both scalar and array APIs
  - Placed in new directory config_src/drivers/timing_tests/
- Renamed MOM_unit_test_driver.F90 to test_MOM_file_parser.F90
- Updated .testing/Makefile
  - Added list of programs in config_src/drivers/unit_tests
    - These are added to BUILDS if DO_UNIT_TESTS is not blank.
      (DO_UNIT_TESTS was an existing macro but it might be uneeded)
    - These programs are compiled with code coverage
  - Added list of programs in config_src/drivers/timing_tests
    - These programs are compiled with optimization and no coverage
  - Fixed rule for building UNIT_EXECS (which did not re-build properly
    because the central Makefile was trying to model the dependencies
    even though those dependencies are in the build/unit/Makefile.dep)
  - Added convenient targets build.unit, run.unit, build.timing, run.timing
- Timing tests currently time a loop over 1000 calls (so that the resolution
  of the CPU timer is not too large) and 400 samples to collect statistics
  on timings. On gaea c5 this takes about 10 seconds.
  - The results are written to stdout in json.
- Added placeholder build and run of timing_tests to GH workflow.
  - Enabled for [push,pull_request]
  - We probably will not be able to use timings from GH but I still want
    to exercise the code we know the timing programs aren't broken by a
    commit.
- Also added driver for string_functions_unit_tests
  • Loading branch information
adcroft authored Nov 10, 2023
1 parent 0f2a69d commit b15a9d4
Show file tree
Hide file tree
Showing 13 changed files with 564 additions and 38 deletions.
12 changes: 9 additions & 3 deletions .github/workflows/coverage.yml
Original file line number Diff line number Diff line change
Expand Up @@ -19,12 +19,18 @@ jobs:

- uses: ./.github/actions/testing-setup

- name: Compile unit testing
run: make -j build/unit/MOM_unit_tests
- name: Compile file parser unit tests
run: make -j build/unit/test_MOM_file_parser

- name: Run unit tests
- name: Run file parser unit tests
run: make run.cov.unit

- name: Compile unit testing
run: make -j build.unit

- name: Run (single processor) unit tests
run: make run.unit

- name: Report unit test coverage to CI (PR)
if: github.event_name == 'pull_request'
run: make report.cov.unit REQUIRE_COVERAGE_UPLOAD=true
Expand Down
17 changes: 16 additions & 1 deletion .github/workflows/perfmon.yml
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
name: Performance Monitor

on: [pull_request]
on: [push, pull_request]

jobs:
build-test-perfmon:
Expand All @@ -20,19 +20,34 @@ jobs:
- uses: ./.github/actions/testing-setup

- name: Compile optimized models
if: ${{ github.event_name == 'pull_request' }}
run: >-
make -j build.prof
MOM_TARGET_SLUG=$GITHUB_REPOSITORY
MOM_TARGET_LOCAL_BRANCH=$GITHUB_BASE_REF
DO_REGRESSION_TESTS=true
- name: Generate profile data
if: ${{ github.event_name == 'pull_request' }}
run: >-
pip install f90nml &&
make profile
DO_REGRESSION_TESTS=true
- name: Generate perf data
if: ${{ github.event_name == 'pull_request' }}
run: |
sudo sysctl -w kernel.perf_event_paranoid=2
make perf DO_REGRESSION_TESTS=true
- name: Compile timing tests
run: |
make -j build.timing
- name: Run timing tests
run: |
make -j run.timing
- name: Display timing results
run: |
make -j show.timing
71 changes: 55 additions & 16 deletions .testing/Makefile
Original file line number Diff line number Diff line change
Expand Up @@ -116,6 +116,9 @@ DO_PROFILE ?=
# Enable code coverage runs
DO_COVERAGE ?=

# Enable code coverage runs
DO_UNIT_TESTS ?=

# Report failure if coverage report is not uploaded
REQUIRE_COVERAGE_UPLOAD ?=

Expand Down Expand Up @@ -151,10 +154,16 @@ ifeq ($(DO_PROFILE), true)
BUILDS += opt/MOM6 opt_target/MOM6
endif

# Unit testing
UNIT_EXECS ?= MOM_unit_tests
# Coverage
ifeq ($(DO_COVERAGE), true)
BUILDS += cov/MOM6 $(foreach e, $(UNIT_EXECS), unit/$(e))
BUILDS += cov/MOM6
endif

# Unit testing (or coverage)
UNIT_EXECS ?= $(basename $(notdir $(wildcard ../config_src/drivers/unit_tests/*.F90) ) )
TIMING_EXECS ?= $(basename $(notdir $(wildcard ../config_src/drivers/timing_tests/*.F90) ) )
ifneq (X$(DO_COVERAGE)$(DO_UNIT_TESTS)X, XX)
BUILDS += $(foreach e, $(UNIT_EXECS), unit/$(e))
endif

ifeq ($(DO_PROFILE), false)
Expand Down Expand Up @@ -258,13 +267,15 @@ build/coupled/Makefile: MOM_ENV += $(SYMMETRIC_FCFLAGS) $(MOM_LDFLAGS)
build/nuopc/Makefile: MOM_ENV += $(SYMMETRIC_FCFLAGS) $(MOM_LDFLAGS)
build/cov/Makefile: MOM_ENV += $(COV_FCFLAGS) $(COV_LDFLAGS)
build/unit/Makefile: MOM_ENV += $(COV_FCFLAGS) $(COV_LDFLAGS)
build/timing/Makefile: MOM_ENV += $(OPT_FCFLAGS) $(MOM_LDFLAGS)

# Configure script flags
MOM_ACFLAGS := --with-framework=$(FRAMEWORK)
build/openmp/Makefile: MOM_ACFLAGS += --enable-openmp
build/coupled/Makefile: MOM_ACFLAGS += --with-driver=FMS_cap
build/nuopc/Makefile: MOM_ACFLAGS += --with-driver=nuopc_cap
build/unit/Makefile: MOM_ACFLAGS += --with-driver=unit_tests
build/timing/Makefile: MOM_ACFLAGS += --with-driver=timing_tests

# Fetch regression target source code
build/target/Makefile: | $(TARGET_CODEBASE)
Expand All @@ -276,10 +287,15 @@ build/target_codebase/configure: $(TARGET_SOURCE)


# Build executables
$(foreach e,$(UNIT_EXECS),build/unit/$(e)): build/unit/Makefile $(MOM_SOURCE)
cd $(@D) && $(TIME) $(MAKE) -j
build/%/MOM6: build/%/Makefile $(MOM_SOURCE)
cd $(@D) && $(TIME) $(MAKE) -j
build/unit/test_%: build/unit/Makefile FORCE
cd $(@D) && $(TIME) $(MAKE) $(@F) -j
build/unit/Makefile: $(foreach e,$(UNIT_EXECS),../config_src/drivers/unit_tests/$(e).F90)
build/timing/time_%: build/timing/Makefile FORCE
cd $(@D) && $(TIME) $(MAKE) $(@F) -j
build/timing/Makefile: $(foreach e,$(TIMING_EXECS),../config_src/drivers/timing_tests/$(e).F90)
build/%/MOM6: build/%/Makefile FORCE
cd $(@D) && $(TIME) $(MAKE) $(@F) -j
FORCE: ;


# Use autoconf to construct the Makefile for each target
Expand Down Expand Up @@ -655,28 +671,47 @@ test.summary:
.PHONY: run.cov.unit
run.cov.unit: build/unit/MOM_file_parser_tests.F90.gcov

$(WORKSPACE)/work/unit/std.out: build/unit/MOM_unit_tests
.PHONY: build.unit
build.unit: $(foreach f, $(UNIT_EXECS), build/unit/$(f))
.PHONY: run.unit
run.unit: $(foreach f, $(UNIT_EXECS), work/unit/$(f).out)
.PHONY: build.timing
build.timing: $(foreach f, $(TIMING_EXECS), build/timing/$(f))
.PHONY: run.timing
run.timing: $(foreach f, $(TIMING_EXECS), work/timing/$(f).out)
.PHONY: show.timing
show.timing: $(foreach f, $(TIMING_EXECS), work/timing/$(f).show)
$(WORKSPACE)/work/timing/%.show:
./tools/disp_timing.py $(@:.show=.out)

# General rule to run a unit test executable
# Pattern is to run build/unit/executable and direct output to executable.out
$(WORKSPACE)/work/unit/%.out: build/unit/%
@mkdir -p $(@D)
cd $(@D) ; $(TIME) $(MPIRUN) -n 1 $(abspath $<) 2> >(tee $*.err) > $*.out

$(WORKSPACE)/work/unit/test_MOM_file_parser.out: build/unit/test_MOM_file_parser
if [ $(REPORT_COVERAGE) ]; then \
find build/unit -name *.gcda -exec rm -f '{}' \; ; \
fi
rm -rf $(@D)
mkdir -p $(@D)
cd $(@D) \
&& $(TIME) $(MPIRUN) -n 1 $(abspath $<) 2> std.err > std.out \
&& rm -f input.nml logfile.0000*.out *_input MOM_parameter_doc.* \
&& $(TIME) $(MPIRUN) -n 1 $(abspath $<) 2> test_MOM_file_parser.err > test_MOM_file_parser.out \
|| !( \
cat std.out | tail -n 100 ; \
cat std.err | tail -n 100 ; \
cat test_MOM_file_parser.out | tail -n 100 ; \
cat test_MOM_file_parser.err | tail -n 100 ; \
)
cd $(@D) \
&& $(TIME) $(MPIRUN) -n 2 $(abspath $<) 2> p2.std.err > p2.std.out \
&& $(TIME) $(MPIRUN) -n 2 $(abspath $<) 2> p2.test_MOM_file_parser.err > p2.test_MOM_file_parser.out \
|| !( \
cat p2.std.out | tail -n 100 ; \
cat p2.std.err | tail -n 100 ; \
cat p2.test_MOM_file_parser.out | tail -n 100 ; \
cat p2.test_MOM_file_parser.err | tail -n 100 ; \
)

# NOTE: .gcov actually depends on .gcda, but .gcda is produced with std.out
# TODO: Replace $(WORKSPACE)/work/unit/std.out with *.gcda?
build/unit/MOM_file_parser_tests.F90.gcov: $(WORKSPACE)/work/unit/std.out
build/unit/MOM_file_parser_tests.F90.gcov: $(WORKSPACE)/work/unit/test_MOM_file_parser.out
cd $(@D) \
&& gcov -b *.gcda > gcov.unit.out
find $(@D) -name "*.gcov" -exec sed -i -r 's/^( *[0-9]*)\*:/ \1:/g' {} \;
Expand All @@ -693,6 +728,10 @@ report.cov.unit: build/unit/MOM_file_parser_tests.F90.gcov codecov
if [ "$(REQUIRE_COVERAGE_UPLOAD)" = true ] ; then false ; fi ; \
}

$(WORKSPACE)/work/timing/%.out: build/timing/% FORCE
@mkdir -p $(@D)
@echo Running $< in $(@D)
@cd $(@D) ; $(TIME) $(MPIRUN) -n 1 $(abspath $<) 2> $*.err > $*.out

#---
# Profiling based on FMS clocks
Expand Down
11 changes: 11 additions & 0 deletions .testing/README.rst
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,17 @@ Usage
``make clean``
Delete the MOM6 test executables and dependency builds (FMS).

``make -j build.unit``
Build the unit test programs in config_src/drivers/unit_tests

``make -j run.unit``
Run the unit test programs from config_src/drivers/unit_tests in $(WORKSPACE)/work/unit

``make -j build.timing``
Build the timing test programs in config_src/drivers/timing_tests

``make -j run.timing``
Run the timing test programs from config_src/drivers/timing_tests in $(WORKSPACE)/work/timing

Configuration
=============
Expand Down
133 changes: 133 additions & 0 deletions .testing/tools/disp_timing.py
Original file line number Diff line number Diff line change
@@ -0,0 +1,133 @@
#!/usr/bin/env python3

from __future__ import print_function

import argparse
import json
import math

scale = 1e6 # micro-seconds (should make this dynamic)


def display_timing_file(file, show_all):
"""Parse a JSON file of timing results and pretty-print the results"""

with open(file) as json_file:
timing_dict = json.load(json_file)

print("(Times measured in %5.0e seconds)" % (1./scale))
print(" Min time Module & function")
for sub in timing_dict.keys():
tmin = timing_dict[sub]['min'] * scale
print("%10.4e %s" % (tmin, sub))

if show_all:
tmean = timing_dict[sub]['mean'] * scale
tmax = timing_dict[sub]['max'] * scale
tstd = timing_dict[sub]['std'] * scale
nsamp = timing_dict[sub]['n_samples']
tsstd = tstd / math.sqrt(nsamp)
print(" (" +
"mean = %10.4e " % (tmean) +
"±%7.1e, " % (tsstd) +
"max = %10.4e, " % (tmax) +
"std = %8.2e, " % (tstd) +
"# = %d)" % (nsamp))


def compare_timing_files(file, ref, show_all, significance_threshold):
"""Read and compare two JSON files of timing results"""

with open(file) as json_file:
timing_dict = json.load(json_file)

with open(ref) as json_file:
ref_dict = json.load(json_file)

print("(Times measured in %5.0e seconds)" % (1./scale))
print(" Delta (%) Module & function")
for sub in {**ref_dict, **timing_dict}.keys():
T1 = ref_dict.get(sub)
T2 = timing_dict.get(sub)
if T1 is not None:
# stats for reference (old)
tmin1 = T1['min'] * scale
tmean1 = T1['mean'] * scale
if T2 is not None:
# stats for reference (old)
tmin2 = T2['min'] * scale
tmean2 = T2['mean'] * scale
if (T1 is not None) and (T2 is not None):
# change in actual minimum as percentage of old
dt = (tmin2 - tmin1) * 100 / tmin1
if dt < -significance_threshold:
color = '\033[92m'
elif dt > significance_threshold:
color = '\033[91m'
else:
color = ''
print("%s%+10.4f%%\033[0m %s" % (color, dt, sub))
else:
if T2 is None:
print(" removed %s" % (sub))
else:
print(" added %s" % (sub))

if show_all:
if T2 is None:
print(" --")
else:
tmax2 = T2['max'] * scale
tstd2 = T2['std'] * scale
n2 = T2['n_samples']
tsstd2 = tstd2 / math.sqrt(n2)
print(" %10.4e (" % (tmin2) +
"mean = %10.4e " % (tmean2) +
"±%7.1e, " % (tsstd2) +
"max=%10.4e, " % (tmax2) +
"std=%8.2e, " % (tstd2) +
"# = %d)" % (n2))
if T1 is None:
print(" --")
else:
tmax1 = T1['max'] * scale
tstd1 = T1['std'] * scale
n1 = T1['n_samples']
tsstd1 = tstd1 / math.sqrt(n1)
print(" %10.4e (" % (tmin1) +
"mean = %10.4e " % (tmean1) +
"±%7.1e, " % (tsstd1) +
"max=%10.4e, " % (tmax1) +
"std=%8.2e, " % (tstd1) +
"# = %d)" % (n1))


# Parse arguments
parser = argparse.ArgumentParser(
description="Beautify timing output from MOM6 timing tests."
)
parser.add_argument(
'file',
help="File to process."
)
parser.add_argument(
'-a', '--all',
action='store_true',
help="Display all metrics rather than just the minimum time."
)
parser.add_argument(
'-t', '--threshold',
default=6.0, type=float,
help="Significance threshold to flag (percentage)."
)
parser.add_argument(
'-r', '--reference',
help="Reference file to compare against."
)
args = parser.parse_args()

# Do the thing
if args.reference is None:
display_timing_file(args.file, args.all)
else:
compare_timing_files(args.file, args.reference, args.all, args.threshold)
Loading

0 comments on commit b15a9d4

Please sign in to comment.