[llvm-bugs] [Bug 36875] New: [llvm-mca] Add the ability to mark section(s) of code for analysis

via llvm-bugs llvm-bugs at lists.llvm.org
Fri Mar 23 04:06:09 PDT 2018


https://bugs.llvm.org/show_bug.cgi?id=36875

            Bug ID: 36875
           Summary: [llvm-mca] Add the ability to mark section(s) of code
                    for analysis
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: andrea.dibiagio at gmail.com
          Reporter: llvm-dev at redking.me.uk
                CC: atrick at apple.com, clement.courbet at gmail.com,
                    craig.topper at gmail.com, greg.bedwell at sony.com,
                    hfinkel at anl.gov, llvm-bugs at lists.llvm.org,
                    spatel+llvm at rotateright.com

Currently llvm-mca will analyse all the asm data in an input file, this often
means that it can be tricky to get analysis for a hot loop section of code
without all the prologue/epilogue/branches/return code getting in the way.

What would be ideal is if we had a way to mark the section(s) of code that we
actually want an analysis for - in raw asm this will require
comments/directives of some sort.

What would be even better would be is if we could provide these markers in
C++/IR code to make it easy to pipe the asm directly into llvm-mca, this would
be similar to IACA's IACA_START/IACA_END macros but without inserting any code
into the final object file.

Not sure if we might have to treat these marker as some sort of barrier so it
might affect the ordering of instructions?

I don't know the best way to handle multiple sections - outputting one section
after another obviously, I don't think we need any 'multi section' summary -
but we might need an user friendly way in the marker to distinguish between
sections (or maybe loc is enough)?.

Yet another thing that might be of use is if there is a way that we could
include this in debug/DWARF data to help with comparing static analysis vs
profiling.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180323/cc1219b3/attachment-0001.html>


More information about the llvm-bugs mailing list