[Bug target/113022] New: GCN offloading bricked by "amdgcn: Work around XNACK register allocation problem"

[Bug target/113022] New: GCN offloading bricked by "amdgcn: Work around XNACK register allocation problem"

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113022

            Bug ID: 113022
           Summary: GCN offloading bricked by "amdgcn: Work around XNACK
                    register allocation problem"
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Keywords: openacc, openmp, testsuite-fail
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tschwinge at gcc dot gnu.org
                CC: ams at gcc dot gnu.org, jules at gcc dot gnu.org
  Target Milestone: ---
            Target: GCN

I've not seen a problem in GCN target testing, but GCN offloading -- at least
in my testing -- is bricked (for non-'-march=gfx90a'?) by the recent commit
r14-6503-g4c12bcbeb0c0fd6da4c56e7622814201daadd585 "amdgcn: Work around XNACK
register allocation problem":

    /tmp/ccwsYf5g.mkoffload.2.s:1:17: error: .amdgcn_target directive's target
id amdgcn-unknown-amdhsa--gfx900:xnack- does not match the specified target id
amdgcn-unknown-amdhsa--gfx900
            .amdgcn_target "amdgcn-unknown-amdhsa--gfx900:xnack-"
                           ^
    /tmp/ccwsYf5g.mkoffload.2.s:29:4: error: .amdhsa_reserve_xnack_mask does
not match target id
              .amdhsa_reserve_xnack_mask    0
              ^~~~~~~~~~~~~~~~~~~~~~~~~~
    [...]

Reverting that commit is my workaround for the time being.

Is maybe simply something missing in GCN 'mkoffload'?

[Bug target/113022] GCN offloading bricked by "amdgcn: Work around XNACK register allocation problem"

[ams at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113022

--- Comment #1 from Andrew Stubbs <ams at gcc dot gnu.org> ---
This is what I get for trying to get this done before vacation. :(

Yes, there's probably something in mkoffload that has to match the default
change from -mxnack=any to -mxnack=off on the older ISAs.

[Bug target/113022] GCN offloading bricked by "amdgcn: Work around XNACK register allocation problem"

[tschwinge at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113022

Thomas Schwinge <tschwinge at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|UNCONFIRMED                 |RESOLVED
           Assignee|unassigned at gcc dot gnu.org      |ams at gcc dot gnu.org

--- Comment #2 from Thomas Schwinge <tschwinge at gcc dot gnu.org> ---
Resolved via commit r14-6997-g78dff4c25c1b959e4682d7da50d00fb371849a46 "amdgcn:
Match new XNACK defaults in mkoffload".