[llvm-bugs] [Bug 40936] New: [amdgpu][regression][bisected] function calls on Raven break sub int loads from private or local memory

via llvm-bugs llvm-bugs at lists.llvm.org
Sat Mar 2 11:19:05 PST 2019


https://bugs.llvm.org/show_bug.cgi?id=40936

            Bug ID: 40936
           Summary: [amdgpu][regression][bisected] function calls on Raven
                    break sub int loads from private or local memory
           Product: new-bugs
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: jv356 at scarletmail.rutgers.edu
                CC: htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org

Created attachment 21549
  --> https://bugs.llvm.org/attachment.cgi?id=21549&action=edit
bad.asm

Bisected to:
df3568d8a9367a51c0368d47392f1bbb1b667603 is the first bad commit
commit df3568d8a9367a51c0368d47392f1bbb1b667603
Author: Matt Arsenault <Matthew.Arsenault at amd.com>
Date:   Thu Feb 28 00:40:32 2019 +0000

    AMDGPU: Enable function calls by default

    Fixes some crashes on illegal call situations which are unfortunately
    still valid IR.

    git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@355051
91177308-0d34-0410-b5e6-96231b3b80d8

The above commit regresses the following piglit tests:
vload-half-private.cl
vload-half-local.cl
vload-short-private.cl
vload-short-local.cl
vload-ushort-private.cl[0]
vload-ushort-local.cl

Note that the regression happens only on Raven (at least carrizo/iceland is
OK),
and only vec2 tests are affected.
I've verified to switching the options off on
954735141519398580615e0907a85ed0972b50cb
Fixes the issue.

[0]
https://jvesely.github.io/piglit/raven-latest-5/raven_19_02_28/program@execute@vload@vload-ushort-private.html

GCN assembly code of both the good and the bad case is attached

-- 
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/20190302/1bbd7edb/attachment-0001.html>


More information about the llvm-bugs mailing list