Eliminate most Windows IO usages in superpmi #121367
Conversation
github-actions
bot
added
the
area-CodeGen-coreclr
dotnet-policy-service
bot
added
the
community-contribution
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
huoyaoyuan
force-pushed
the
superpmi/stdio2
branch
from
f8cb342 to
e0ea225
Compare
| #include "jitmetadatalist.h" | ||
|
|
||
| fw.Print("\n"); | ||
| fs << std::endl; |
There was a problem hiding this comment.
| fs << std::endl; | |
| fs << "\n"; |
std::endl will flush the output stream. We do not want to do that for every context we are processing.
Can you please validate the performance of these changes when running superpmi with the -details argument, especially on slower storage mediums?
There was a problem hiding this comment.
Well an error occurs when doing the test with changed executable. I'm going to debug it first.
Data for main with release exe, testing the 72MB mch generated by tests\JIT\superpmi\superpmicollect:
NVMe:925ms HDD:945ms
There was a problem hiding this comment.
PR using "\n": NVMe:1050ms HDD:1060ms
Seems that flushing won't be a problem, but formatting becomes less efficiency.
There was a problem hiding this comment.
Removing fstream usage reverts the performance back to 945ms.
There was a problem hiding this comment.
@jakobbotsch is it acceptable now? Any other concerns?
2 participants
Continuation of #116422. Contains most usages except I/O redirection for new process.
Combined with little refactors for reading file contents.