testing gcc optimization options

result.md

args	run_time
-Dfoo	1.942
-ftree-sink	SAME
-fauto-inc-dec	SAME
-fcrossjumping	1.925
-fif-conversion	SAME
-faggressive-loop-optimizations	SAME
-fconserve-stack	SAME
-ftree-vrp	SAME
-ftree-vrp -fdelete-null-pointer-checks	SAME
-fsched-pressure	SAME
-fmodulo-sched	SAME
-fno-lto	SAME
-ftree-ter	1.887
-frerun-cse-after-loop	SAME
-ftree-loop-if-convert	SAME
-fivopts	SAME
-fcombine-stack-adjustments	SAME
-fcompare-elim	SAME
-fearly-inlining	SAME
-fcaller-saves	SAME
-fsel-sched-pipelining-outer-loops -fsel-sched-pipelining	SAME
-fsched-spec-load-dangerous	SAME
-fipa-sra	SAME
-fomit-frame-pointer	1.776
-floop-block	SAME
-fshrink-wrap	SAME
-fbranch-count-reg	SAME
-floop-interchange	SAME
-fthread-jumps	SAME
-ffunction-cse	SAME
-fipa-reference	SAME
-fno-sched-spec	SAME
-fno-strict-overflow	SAME
-findirect-inlining -finline-functions	SAME
-findirect-inlining -finline-small-functions	SAME
-findirect-inlining -finline-functions	SAME
-findirect-inlining -finline-small-functions	SAME
-fkeep-inline-functions	SAME
-fsched2-use-superblocks	SAME
-fpeephole	SAME
-fsched-dep-count-heuristic	SAME
-fipa-pta	SAME
-ftree-builtin-call-dce	SAME
-fcse-skip-blocks	SAME
-fsplit-wide-types	SAME
-fkeep-static-consts	SAME
-fno-zero-initialized-in-bss	SAME
-ftree-sra	SAME
-fwhole-program	1.927
-fno-function-cse	SAME
-ffat-lto-objects	SAME
-fprofile-generate	3.857
-ftree-loop-im	SAME
-ftree-loop-im -funswitch-loops	SAME
-fno-ira-share-spill-slots	SAME
-ftree-vectorize	SAME
-fmerge-all-constants	SAME
-fno-align-jumps	SAME
-fzero-initialized-in-bss	SAME
-fno-align-functions	SAME
-fno-defer-pop	SAME
-fsched-group-heuristic	SAME
-fipa-profile	SAME
-fforward-propagate	SAME
-ftree-dse	SAME
-fno-keep-static-consts	SAME
-fstrict-overflow	1.919
-fstrict-overflow -fwrapv	SAME
-ftree-dce	SAME
-fno-peephole	SAME
-fno-align-loops	SAME
-fmodulo-sched-allow-regmoves -fmodulo-sched	SAME
-fgcse-sm	SAME
-fgcse-sm -fgcse-lm	SAME
-freorder-blocks-and-partition	1.935
-fno-delete-null-pointer-checks	SAME
-ftree-fre	SAME
-fno-gcse	SAME
-ftree-loop-distribute-patterns	SAME
-ftree-pta	SAME
-fno-sched-interblock	SAME
-fsched-spec-load	SAME
-fpartial-inlining	SAME
-ftree-slsr	SAME
-funsafe-loop-optimizations	SAME
-ftree-slp-vectorize	SAME
-flto-report	SAME
-flto-report -flto	1.890
-ftree-partial-pre	SAME
-ftree-loop-optimize	SAME
-ftree-switch-conversion	SAME
-fipa-cp-clone	SAME
-fno-peephole2	SAME
-fsched-rank-heuristic	SAME
-fselective-scheduling	SAME
-fgcse-las	SAME
-ftree-forwprop	SAME
-fprofile-correction	SAME
-fgcse-lm	SAME
-fsched-spec-insn-heuristic	SAME
-fno-toplevel-reorder	SAME
-free	SAME
-ftree-phiprop	SAME
-flto-partition=1to1	SAME
-flto-partition=balanced	SAME
-flto-partition=max	SAME
-flto-partition=none	SAME
-ftree-reassoc	SAME
-fdce	SAME
-floop-parallelize-all	SAME
-fsched-critical-path-heuristic	SAME
-foptimize-strlen	SAME
-fira-hoist-pressure	SAME
-fgcse-after-reload	SAME
-fpredictive-commoning	SAME
-ftree-ch	SAME
-fdse	SAME
-fvariable-expansion-in-unroller	SAME
-freschedule-modulo-scheduled-loops	SAME
-fdevirtualize	SAME
-ftree-loop-ivcanon	SAME
-fira-algorithm=priority	SAME
-fira-algorithm=CB	SAME
-fschedule-insns	SAME
-fdelayed-branch	SAME
-ftree-copy-prop	SAME
-fira-region=all	SAME
-fira-region=mixed	SAME
-fira-region=one	SAME
-ftree-loop-if-convert-stores	SAME
-fsel-sched-pipelining -fselective-scheduling	SAME
-fsel-sched-pipelining -fselective-scheduling2	SAME
-fsel-sched-pipelining -fselective-scheduling	SAME
-fsel-sched-pipelining -fselective-scheduling2	SAME
-fsplit-ivs-in-unroller	SAME
-fsplit-ivs-in-unroller -fweb	SAME
-fno-ira-share-save-slots	SAME
-flto-compression-level=0	SAME
-flto-compression-level=1	SAME
-flto-compression-level=2	SAME
-flto-compression-level=3	SAME
-flto-compression-level=4	SAME
-flto-compression-level=5	SAME
-flto-compression-level=6	SAME
-flto-compression-level=7	SAME
-flto-compression-level=8	SAME
-flto-compression-level=9	SAME
-flto	1.906
-flto -fuse-linker-plugin	1.904
-fno-align-labels	SAME
-fdelete-null-pointer-checks	SAME
-ftree-bit-ccp -ftree-ccp	SAME
-finline-functions	SAME
-ftree-loop-linear	SAME
-fprefetch-loop-arrays	SAME
-foptimize-sibling-calls	SAME
-freorder-functions	SAME
-ftree-dominator-opts	SAME
-fgcse	SAME
-finline-functions-called-once	SAME
-fguess-branch-probability	SAME
-fhoist-adjacent-loads	SAME
-fipa-pure-const	SAME
-fno-branch-count-reg	SAME
-fstrict-aliasing	SAME
-fmerge-constants	SAME
-floop-nest-optimize	SAME
-fschedule-insns2	SAME
-fipa-cp	SAME
-fcprop-registers	SAME
-floop-strip-mine	SAME
-fno-inline	SAME
-ftree-pre	SAME
-fcse-follow-jumps	SAME
-fno-merge-constants	SAME
-fgraphite-identity	SAME
-fno-guess-branch-probability	SAME
-ftree-tail-merge	SAME
-fexpensive-optimizations	SAME
-fselective-scheduling2	SAME
-fira-loop-pressure	SAME
-freorder-blocks	SAME
-ftree-loop-distribution	SAME
-fweb	SAME
-fuse-linker-plugin	SAME
-fpeephole2	SAME
-fsched-last-insn-heuristic	SAME
-finline-small-functions	SAME
-fif-conversion2	SAME

test_opt.py

args = {
    'fine_tuning':
    {
        "-fno-defer-pop":None,
        "-fforward-propagate":None,
        "-ffp-contract=style":{
            'p_range':['style', 'off', 'fast']
        },
        "-fomit-frame-pointer":None,
        "-foptimize-sibling-calls":None,
        "-foptimize-strlen":None,
        "-fno-inline":None,
        "-finline-small-functions":None,
        "-findirect-inlining":{
            'd_or':[
                '-finline-functions' ,'-finline-small-functions'
            ]
        },
        "-finline-functions":None,
        "-finline-functions-called-once":None,
        "-fearly-inlining":None,
        "-fprofile-generate":None,
        "-fipa-sra":None,
        "-finline-limit":{
            'p_type':(int, None)
        },
        "-fno-keep-inline-dllexport":None,
        "-fkeep-inline-functions":None,
        "-fkeep-static-consts":None,
        "-fno-keep-static-consts":None,
        "-fmerge-constants":None,
        "-fno-merge-constants":None,
        "-fmerge-all-constants":None,
        "-fmodulo-sched":None,
        "-fmodulo-sched-allow-regmoves" :{
            'd_and':["-fmodulo-sched"]
        },
        "-fno-branch-count-reg":None,
        "-fbranch-count-reg":None,
        "-fno-function-cse":None,
        "-ffunction-cse":None,
        "-fno-zero-initialized-in-bss":None,
        "-fzero-initialized-in-bss":None,
        "-fthread-jumps":None,
        "-fsplit-wide-types":None,
        "-fcse-follow-jumps":None,
        "-fcse-skip-blocks":None,
        "-frerun-cse-after-loop":None,
        "-fgcse":None,
        "-fno-gcse":None,
        "-fgcse-lm":None,
        "-fgcse":None,
        "-fgcse-sm":{
            'd_opt':["-fgcse-lm"]
        },
        "-fgcse-las":None,
        "-fgcse-after-reload":None,
        "-faggressive-loop-optimizations":None,
        "-funsafe-loop-optimizations":None,
        "-fcrossjumping":None,
        "-fauto-inc-dec":None,
        "-fdce":None,
        "-fdse":None,
        "-fif-conversion":None,
        "-fif-conversion2":None,
        "-fdeclone-ctor-dtor":None,
        "-fdelete-null-pointer-checks":None,
        "-fno-delete-null-pointer-checks":None,
        "-fdevirtualize":None,
        "-fdevirtualize-speculatively":None,
        "-fdevirtualize-at-ltrans":None,
        "-fexpensive-optimizations":None,
        "-free":None,
        "-fno-lifetime-dse":None,
        "-flive-range-shrinkage":None,
        "-fira-algorithm":{
            'p_range':['priority', 'CB']
        },
        "-fira-region":{
            'p_range':['all', 'mixed', 'one']
        },
        "-fira-hoist-pressure":None,
        "-fira-loop-pressure":None,
        "-fno-ira-share-save-slots":None,
        "-fno-ira-share-spill-slots":None,
        "-fira-verbose":{
            'p_type':[int, 5]
        },
        "-flra-remat":None,
        "-fdelayed-branch":None,
        "-fschedule-insns":None,
        "-fschedule-insns2":None,
        "-fno-sched-interblock":None,
        "-fno-sched-spec":None,
        "-fsched-pressure":None,
        "-fsched-spec-load":None,
        "-fsched-spec-load-dangerous":None,
        "-fsched-stalled-insns":{
            'p_type':[int, 1]
        },
        "-fsched-stalled-insns-dep":{
            'p_type':[int, 1],
        },
        "-fsched2-use-superblocks":None,
        "-fsched-group-heuristic":None,
        "-fsched-critical-path-heuristic":None,
        "-fsched-spec-insn-heuristic":None,
        "-fsched-rank-heuristic":None,
        "-fsched-last-insn-heuristic":None,
        "-fsched-dep-count-heuristic":None,
        "-freschedule-modulo-scheduled-loops":None,
        "-fselective-scheduling":None,
        "-fselective-scheduling2":None,
        "-fsel-sched-pipelining":{
            'd_or':["-fselective-scheduling", "-fselective-scheduling2"]
        },
        "-fsel-sched-pipelining-outer-loops":{
            'd_and':["-fsel-sched-pipelining"]
        },
        "-fsemantic-interposition":None,
        "-fno-semantic-interposition":None,
        "-fshrink-wrap":None,
        "-fcaller-saves":None,
        "-fcombine-stack-adjustments":None,
        "-fipa-ra":None,
        "-fconserve-stack":None,
        "-ftree-reassoc":None,
        "-ftree-pre":None,
        "-ftree-partial-pre":None,
        "-ftree-forwprop":None,
        "-ftree-fre":None,
        "-ftree-phiprop":None,
        "-fhoist-adjacent-loads":None,
        "-ftree-copy-prop":None,
        "-fipa-pure-const":None,
        "-fipa-reference":None,
        "-fipa-pta":None,
        "-fipa-profile":None,
        "-fipa-cp":None,
        "-fipa-cp-clone":None,
        "-fipa-cp-alignment":{
            'd_and':["-fipa-cp"]
        },
        "-fipa-icf":None,
        "-fisolate-erroneous-paths-dereference":{
            'd_and':["-fdelete-null-pointer-checks"]
        },
        "-fisolate-erroneous-paths-attribute":None,
        "-ftree-sink":None,
        "-ftree-bit-ccp":{
            'd_and':["-ftree-ccp"]
        },
        "-fssa-phiopt":None,
        "-ftree-switch-conversion":None,
        "-ftree-tail-merge":None,
        "-ftree-dce":None,
        "-ftree-builtin-call-dce":None,
        "-ftree-dominator-opts":None,
        "-ftree-dse":None,
        "-ftree-ch":None,
        "-ftree-loop-optimize":None,
        "-ftree-loop-linear":None,
        "-floop-interchange":None,
        "-floop-strip-mine":None,
        "-floop-block":None,
        "-floop-unroll-and-jam":None,
        "-floop-nest-optimize":None,
        "-fgraphite-identity":None,
        "-floop-nest-optimize":None,
        "-floop-parallelize-all":None,
        "-ftree-coalesce-vars":{
            'c_and':["-fno-var-tracking-assignment"]
        },
        "-ftree-loop-if-convert":None,
        "-ftree-loop-if-convert-stores":None,
        "-ftree-loop-distribution":None,
        "-ftree-loop-distribute-patterns":None,
        "-ftree-loop-im":{
            'd_opt':["-funswitch-loops"]
        },
        "-ftree-loop-ivcanon":None,
        "-fivopts":None,
        "-ftree-parallelize-loops":{
            'p_type':[int, 1]
        },
        "-ftree-pta":None,
        "-ftree-sra":None,
        "-ftree-ter":None,
        "-ftree-slsr":None,
        "-ftree-vectorize":None,
        "-ftree-loop-vectorize":None,
        "-ftree-vectorize":None,
        "-ftree-slp-vectorize":None,
        "-fvect-cost-model":{
            'p_range':["unlimited", "dynamic", "cheap"]
        },
        "-fsimd-cost-model":{
            'p_range':["unlimited", "dynamic", "cheap"]
        },
        "-ftree-vrp":{
            'd_opt':["-fdelete-null-pointer-checks"]
        },
        "-fsplit-ivs-in-unroller":{
            'd_opt':['-fweb']
        },
        "-fvariable-expansion-in-unroller":None,
        "-fpartial-inlining":None,
        "-fpredictive-commoning":None,
        "-fprefetch-loop-arrays":None,
        "-fno-peephole":None,
        "-fno-peephole2":None,
        "-fpeephole":None,
        "-fpeephole2":None,
        "-fno-guess-branch-probability":None,
        "-fguess-branch-probability":None,
        "-freorder-blocks":None,
        "-freorder-blocks-and-partition":None,
        "-freorder-functions":None,
        "-fstrict-aliasing":None,
        "-fields":None,
        "-fstrict-aliasing":None,
        "-fstrict-overflow":{
            'd_opt':['-fwrapv']
        },
        "-fno-strict-overflow":None,
        "-falign-functions":{
            'p_type':[int, None],
            'p_remark':'should be power of 2'
        },
        "-fno-align-functions":None,
        "-falign-labels":{
            'p_type':[int, None],
            'p_remark':'should be power of 2'
        },
        "-fno-align-labels":None,
        "-falign-loops":{
            'p_type':[int, None],
            'p_remark':'should be power of 2'
        },
        "-fno-align-loops":None,
        "-falign-jumps":{
            'p_type':[int, None],
            'p_remark':'should be power of 2'
        },
        "-fno-align-jumps":None,
        "-fno-toplevel-reorder":None,
        "-fweb":None,
        "-fwhole-program":None,
        "-flto":{
            'd_opt':["-fuse-linker-plugin"]
        },# needs deeper investigation
        "-fno-lto":None,
        "-flto-partition":{
            'p_range':["1to1", "balanced", "max", "none"]
        },
        "-flto-odr-type-merging":None,
        "-flto-compression-level":{
            'p_range':[0,1,2,3,4,5,6,7,8,9,None],
            'd_and':["-flto"]
        },
        "-flto-report":{
            'd_opt':["-flto"]
        },
        "-flto-report-wpa":None,
        "-fuse-linker-plugin":None,
        "-ffat-lto-objects":None,
        "-fcompare-elim":None,
        "-fcprop-registers":None,
        "-fprofile-correction":None,
        #"-fprofile-dir=path":None,
        #"-fprofile-generate=path":None,
        #"-fprofile-use=path":None,
        #"-fauto-profile=path":None,
    },
    "float_tuning":{
        "-ffloat-store":None,
        "-fexcess-precision=standard":{
            'c_or':["-ffast-math", "-funsafe-math-optimizations"]
        },
        "-fexcess-presision=fast":None,
        "-ffast-math":None,
        "-fno-math-errno":None,
        "-fmath-errno":None,
        "-funsafe-math-optimizations":None,
        "-fassociative-math":None,
        "-fno-associative-math":None,
        "-ffinite-math-only":None,
        "-fsigned-zeros":None,
        "-fno-signed-zeros":None,
        "-ftrapping-math":None,
        "-fno-trapping-math":None,
        "-frounding-math":None,
        "-fno-rounding-math":None,
        "-fsignaling-nans":None,
        "-fno-signaling-nans":None,
        "-fsingle-precision-constant":None,
        "-fcx-limited-range":None,
        "-fno-cx-limited-range":None,
        "-ffast-math":None,
        #"-fcx-fortran-rules":None,
        #"-fno-cx-fortran-rules":None,
    },
    "pgo":{
        "-fbranch-probabilities":None,
        "-fprofile-generate"
        # "-fprofile-arcs":None,
        # "-fprofile-values":None,
        # "-fprofile-use":None,
        # # "-fprofile-reorder-functions":None,
        # # "-ftracer":None,
        # # "-funroll-loops":None,
        # # "-funroll-all-loops":None,
        # # "-fpeel-loops":None,
        # # # "-frename-registers":None,
        # "-fvpt":{
        #    'd_opt':["-fprofile-arcs"]
        #},
        "-funswitch-loops":None,
        "-ffunction-sections":None,
        "-fdata-sections":None,
        "-fbranch-target-load-optimize":None,
        "-fbranch-target-load-optimize2":None,
        "-fbtr-bb-exclusive":None,
        "-fstack-protector":None,
        "-fstack-protector-all":None,
        "-fstack-protector-strong":None,
        "-fstack-protector-explicit":None,
        "-fstdarg-opt":None,
        "-fsection-anchors":None,
        # "-fschedule-fusion":None,
        # "-fmove-loop-invariants":None,
    }
}
import os
import subprocess
import time
import hashlib
SAMPLE_TIME = 10
CC = "gcc"
SOURCE = './drystone.c'
OUTFILE = './drystone'
ASMFILE = './drystone.s'
RESULT_FILE = './result.md'
DEFAULT_ARG = ' '
ignored_sections = ['float_tuning', 'pgo']

def popen(*args):
    return subprocess.Popen(
                args,
                stdin=subprocess.PIPE, stdout=subprocess.PIPE, stderr=subprocess.PIPE)

def _md5(b):
    return hashlib.md5(b).hexdigest()

resf = open(RESULT_FILE, 'w', False)
resf.write('''
args| run_time|
------|------
''')
baseline_hash = ''
def compile_and_timeit(*arg):
    global baseline_hash
    _args = [CC, SOURCE] + list(arg) + ["-o", OUTFILE]
    _args_s = ' '.join(_args[2:-2])
    p = popen(*_args)
    p.wait()
    if p.returncode:
        cc_err = p.communicate()[1]
        print("err %s" % cc_err)
        return
    else:
        print('compiled using arg: "%s"' % _args_s)
    run_time = 0
    # test file if is same
    fhash = _md5(open(OUTFILE, 'rb').read())
    if '-Dfoo' in _args:
        baseline_hash = fhash
        print('baseline hash = %s' % baseline_hash)
    else:
        if fhash == baseline_hash:
            print('arg: "%s" exact same as baseline' % _args_s)
            resf.write("%s|SAME\n" % _args_s)
            return
    for i in range(SAMPLE_TIME):
        _t1 = time.time()
        try:
            p = popen(OUTFILE)
        except OSError as ex:
            if errno == 2:
                continue
        else:
            p.wait()
            run_time += time.time() - _t1
    print('arg: "%s" reaches %.3fs' % (_args_s, run_time / SAMPLE_TIME))
    resf.write("%s|%.3f\n" %(_args_s, run_time / SAMPLE_TIME))

for sec in args:
    if sec in ignored_sections:
        continue
    print("Testing sec %s" % sec)
    for arg in ['-Dfoo'] + args[sec].keys():
        try:
            os.remove(OUTFILE)
        except:
            pass
        meta = args[sec][arg] if arg!="-Dfoo" else None
        if meta:
            if 'd_opt' in meta or 'd_or' in meta:
                if 'd_opt' in meta:
                    compile_and_timeit(arg)
                    k = 'd_opt'
                else:
                    k = 'd_or'
                for a in meta[k]:
                    compile_and_timeit(arg, a)
            if 'd_and' in meta:
                compile_and_timeit(arg, *meta['d_and'])
            if 'd_or' in meta:
                for a in meta['d_or']:
                    compile_and_timeit(arg, a)
            if 'p_range' in meta:
                for p in meta['p_range']:
                    compile_and_timeit('%s=%s' % (arg, p))
            if 'p_type' in meta:
                continue
            # TODO c_or for conflicts
        else:
            compile_and_timeit(arg)