[llvm-bugs] [Bug 33169] New: [mc][gfx9+]

via llvm-bugs llvm-bugs at lists.llvm.org
Thu May 25 08:16:16 PDT 2017


            Bug ID: 33169
           Summary: [mc][gfx9+]
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Backend: AMDGPU
          Assignee: unassignedbugs at nondot.org
          Reporter: artem.tamazov at amd.com
                CC: andrey.kasaurov at gmail.com, artem.tamazov at amd.com,
                    dpreobrazhensky at luxoft.com, i.am.perminov at gmail.com,
                    kzhuravl_dev at outlook.com, llvm-bugs at lists.llvm.org,
                    Matthew.Arsenault at amd.com, nikolay.haustov at amd.com,
                    sam.kolton at amd.com, Tony.Tye at amd.com,
                    valery.pykhtin at gmail.com, yaxun.liu at amd.com

Created attachment 18509
  --> https://bugs.llvm.org/attachment.cgi?id=18509&action=edit
regressed lit tests - r301646_regr.zip

45 regressions found during AMDGPU asm weekly testing 28 Apr 2017

Examples of failing instructions:
> v_pk_fma_f16 v5, s1, v0, 0
> v_pk_fma_f16 v5, v1, 0, 0

As of r303147, such insns fail with
> error: invalid operand (violates constant bus restrictions)

Constant bus restrictions are not relevant to the constants here, because these
are not literal constants (encoded in an additional instruction word), but
inline constants.

Instructions with one constant *or* one SGPR are working fine:
> v_pk_fma_f16 v5, v1, v0, 0
> v_pk_fma_f16 v5, v1, s0, v0

Initial regression point was
> r301029 | 7995364 | kzhuravl_dev | 2017-04-21 22:57:53 +0300 (Fri, 21 Apr 2017)
> Reviewed at https://reviews.llvm.org/D32361.

Tested with r303147.

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/20170525/dd5cc9e0/attachment-0001.html>

More information about the llvm-bugs mailing list