[llvm-bugs] [Bug 46858] New: [PM] Code-size regressions

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=46858

            Bug ID: 46858
           Summary: [PM] Code-size regressions
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: sjoerd.meijer at arm.com
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org

Switching to the NPM shows code-size regressions compared to the LPM. For the
llvm test-suite targeting X86, these are the code-size numbers in bytes for the
15 worst regressions:

LPM     NPM     Diff    App
-----------------------------------------------------------------------
399999  404935  4936   
SingleSource/Benchmarks/Adobe-C++/simple_types_loop_invariant
413241  418065  4824   
SingleSource/Benchmarks/Adobe-C++/simple_types_constant_folding
225786  229178  3392    SingleSource/UnitTests/matrix-types-spec
194107  196283  2176    SingleSource/Benchmarks/Adobe-C++/loop_unroll
110540  111836  1296    SingleSource/Benchmarks/Adobe-C++/stepanov_abstraction
51529   52569   1040    SingleSource/Benchmarks/Adobe-C++/stepanov_vector
17345   17841   496     SingleSource/Benchmarks/Adobe-C++/functionobjects
24768   25072   304     SingleSource/Benchmarks/Misc-C++/stepanov_container
5897    6177    280     SingleSource/Benchmarks/Misc/himenobmtxpa
38193   38465   272     SingleSource/Benchmarks/Misc-C++/stepanov_v1p2
5290    5530    240     SingleSource/Benchmarks/Misc/whetstone
1883    2099    216    
SingleSource/Regression/C/gcc-c-torture/execute/GCC-C-execute-pr49886
333621  333789  168     SingleSource/Benchmarks/Linpack/linpack-pc
9423    9591    168     SingleSource/Benchmarks/Misc/oourafft
1698    1842    144    
SingleSource/Regression/C/gcc-c-torture/execute/GCC-C-execute-pr38819
9819    9939    120    
SingleSource/Benchmarks/Shootout-C++/Shootout-C++-wordfreq
5182    5302    120     SingleSource/Benchmarks/McGill/chomp

where NPM is the new pass manger, LPM the legacy pass manager, Diff = NPM -
LMP.

These numbers were obtained with Oz, using the standard cmake config file
"cmake/caches/Oz.cmake" from the test-suite.

Targeting ARM, we see similar regressions, the trend is the same, but haven't
yet tried this on the llvm test-suite.

TODO: I haven't done yet any root cause analysis on this.

-- 
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/20200727/ba8ba654/attachment.html>