[llvm-dev] flags to reproduce clang -O3 with opt -O3

[Mehdi AMINI via llvm-dev]

On Thu, Jul 2, 2020 at 7:50 PM Chad Verbowski <chad at verbowski.com> wrote:

> Awesome, thanks!
>
> I'd like to have the  last step (llc in your example) not perform
> additional optimization passes, such as O3, and simply use the O3 pass from
> opt in the previous line.
>
> Do you happen to know if I should use 'llc -O0 foo_o.bc -o foo.exe'
> instead to achieve this?
>

No you should use `llc -O3`: this is controlling only the backend part of
the pipeline.


>
> On Thu, Jul 2, 2020 at 6:35 PM Mehdi AMINI <joker.eph at gmail.com> wrote:
>
>>
>>
>> On Thu, Jul 2, 2020 at 2:28 PM Chad Verbowski via llvm-dev <
>> llvm-dev at lists.llvm.org> wrote:
>>
>>> Hello,
>>>
>>> I've been trying to figure out how to reproduce the results of a single
>>> clang -O3 compilation to a binary with a multi-step process using opt.
>>>
>>> Specifically I have:
>>>
>>> clang -O3 foo.c -o foo.exe
>>>
>>>
>>> which I want to replicate with:
>>>
>>> clang -O0 -c -emit-llvm foo.c
>>>
>>>
>> Using O0 will mark every function in the IR with "optnone" which prevents
>> `opt` from optimizing it. I'd try `clang -O3  -Xclang
>> -disable-llvm-passes -c -emit-llvm foo.c`
>>
>>
>>> opt -O3 foo.bc -o foo_o.bc
>>> clang foo_o.bc -o foo.exe
>>>
>>>
>> This last step won't enable optimizations in the backend, you likely
>> should try `llc -O3` instead.
>>
>> Best,
>>
>> --
>> Mehdi
>>
>>
>>
>>
>>
>>>
>>> Any hints / suggestions on what additional flags I need to produce the
>>> same binary are greatly appreciated!
>>>
>>> *What I've tried:*
>>> I've been reading the archives, and found this
>>> <http://lists.llvm.org/pipermail/llvm-dev/2017-September/117144.html>,
>>> which suggests dumping the pass arguments using:
>>>
>>> clang -mllvm -debug-pass=Structure -O3 foo.c -o foo.exe
>>>
>>> and comparing with:
>>>
>>> clang -mllvm -debug-pass=Structure -O0 -c -emit-llvm foo.c
>>> opt -debug-pass=Structure -O3 foo.bc -o foo_o.bc
>>> clang -mllvm -debug-pass=Structure foo_o.bc -o foo.exe
>>>
>>>
>>> The first has 30 "Pass Argument" statements though only these 5 are
>>> distinct. Across these 5 there are 190 distinct flags.  The multi-step
>>> compilation has only 140 distinct flags. Comparing the flags, 18 from the
>>> multi-step are missing in the 1pass, and 67 from 1pass are missing in the
>>> multistep.
>>>
>>> These appear to be opt flags, since they cause an error when trying to
>>> use them with clang (e.g. -x86-fixup-LEAs) and when used with opt
>>> causes a crash with stack dump and request to submit a bug report. Others
>>> like -attributor appear to work with opt.
>>>
>>> I'm currently blindly trying to add the 67 different flags to the opt
>>> step to see which work, and hopefully that subset will produce the same
>>> result as clang -O3.
>>>
>>> It seems like there must be an easier / more exact way of getting the
>>> opt -O3 multi-step to match the clang -O3 result.
>>>
>>> Any thoughts or insights are appreciated. Below is a sorted list of the
>>> flags missing from each for completeness.
>>>
>>> not contained in 1pass O3 (count=18)
>>>
>>> -aa-scalar-evolution
>>>
>>> -always-inline
>>>
>>> -callsite-splitting
>>>
>>> -inject-tli-mappings
>>>
>>> -ipsccp
>>>
>>> -jump-threading-correlated-propagation
>>>
>>> -livedebugvalues
>>>
>>> -loops-loop-simplify
>>>
>>> -memdep-lazy-branch-prob
>>>
>>> -openmpopt
>>>
>>> -opt-remark-emitter-instcombine
>>>
>>> -regallocfast
>>>
>>> -speculative-execution
>>>
>>> -stackmap-liveness
>>>
>>> -tbaa-scoped-noalias
>>>
>>> -vector-combine
>>>
>>> -verify
>>>
>>> -write-bitcode
>>>
>>> not contained in multi O3 (count=67)
>>>
>>> -attributor
>>>
>>> -block-freq-loop-simplify
>>>
>>> -branch-folder
>>>
>>> -break-false-deps
>>>
>>> -callsite-splitting-ipsccp
>>>
>>> -codegenprepare
>>>
>>> -consthoist
>>>
>>> -dead-mi-elimination
>>>
>>> -detect-dead-lanes
>>>
>>> -early-ifcvt
>>>
>>> -early-machinelicm
>>>
>>> -early-tailduplication
>>>
>>> -expandmemcmp
>>>
>>> -greedy
>>>
>>> -interleaved-access
>>>
>>> -iv-users
>>>
>>> -lazy-block-freq-opt-remark-emitter
>>>
>>> -livedebugvars
>>>
>>> -liveintervals
>>>
>>> -liveregmatrix
>>>
>>> -livestacks
>>>
>>> -livevars
>>>
>>> -loop-reduce
>>>
>>> -loop-simplify-lcssa-verification
>>>
>>> -lrshrink
>>>
>>> -machine-block-freq
>>>
>>> -machine-combiner
>>>
>>> -machine-cp
>>>
>>> -machine-cse
>>>
>>> -machinedomtree-machine-loops
>>>
>>> -machinelicm
>>>
>>> -machine-loops
>>>
>>> -machinepostdomtree
>>>
>>> -machinepostdomtree-block-placement
>>>
>>> -machine-scheduler
>>>
>>> -machine-sink
>>>
>>> -machine-trace-metrics
>>>
>>> -mergeicmps
>>>
>>> -objc-arc-contract
>>>
>>> -opt-phis
>>>
>>> -partially-inline-libcalls
>>>
>>> -peephole-opt
>>>
>>> -postra-machine-sink
>>>
>>> -post-RA-sched
>>>
>>> -processimpdefs
>>>
>>> -reaching-deps-analysis
>>>
>>> -rename-independent-subregs
>>>
>>> -shrink-wrap
>>>
>>> -simple-register-coalescing
>>>
>>> -slotindexes
>>>
>>> -spill-code-placement
>>>
>>> -stack-coloring
>>>
>>> -stackmap-liveness-livedebugvalues
>>>
>>> -stack-slot-coloring
>>>
>>> -tailduplication
>>>
>>> -unreachable-mbb-elimination
>>>
>>> -virtregmap
>>>
>>> -virtregrewriter
>>>
>>> -x86-avoid-SFB
>>>
>>> -x86-cf-opt
>>>
>>> -x86-cmov-conversion
>>>
>>> -x86-domain-reassignment
>>>
>>> -x86-evex-to-vex-compress
>>>
>>> -x86-execution-domain-fix
>>>
>>> -x86-fixup-bw-insts
>>>
>>> -x86-fixup-LEAs
>>>
>>> -x86-optimize-LEAs
>>>
>>>
>>> _______________________________________________
>>> LLVM Developers mailing list
>>> llvm-dev at lists.llvm.org
>>> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>>
>>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20200702/4648ba04/attachment.html>