[Bug target/110892] New: [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[Bug target/110892] New: [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[ewlu at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110892

            Bug ID: 110892
           Summary: [14 Regression] Risc-V rvv testsuite failures caused
                    by conditional move costing
           Product: gcc
           Version: 14.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ewlu at rivosinc dot com
  Target Milestone: ---

Created attachment 55682
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=55682&action=edit
testsuite failures

Seeing 129 new failures on 36 rvv test cases.

Bisected using https://github.com/patrick-rivos/riscv-gnu-toolchain/issues/127

Confirmed to work:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=cf58b896a26390c98ae5382377fcdf44e8ae209f
Confirmed to fail - rv32 newlib:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=e15d0b6680d10d7666195e9db65581364ad5e5df
Confirmed to fail - other linux/newlib risc-v targets:
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=2d73f2eb80caf328bc4dd1324d475e7bf6b56837

Seems to be caused by
https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=e15d0b6680d10d7666195e9db65581364ad5e5df
"[PATCH 3/5] [RISC-V] RISC-V Conditional Move costing [was:Generate Zicond
instruction for select pattern with condition eq or neq to 0]"

Most(all?) failures appear to be regex match failures.

[Bug target/110892] [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110892

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |14.0
           Keywords|                            |missed-optimization,
                   |                            |testsuite-fail

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Most seem like the cost model missed the vector use case of IF_THEN_ELSE ?

[Bug target/110892] [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[patrick at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110892

--- Comment #2 from Patrick O'Neill <patrick at rivosinc dot com> ---
Seems like a fix was pushed:
https://inbox.sourceware.org/gcc-patches/89dfe3c2-ef96-3000-de87-74ff83354adf@gmail.com/T/#m09608208a2dcc2bbf88a4f3419b062b4b4b349c5

We'll test it and update the issue once it's confirmed to be fixed.

[Bug target/110892] [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[patrick at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110892

--- Comment #3 from Patrick O'Neill <patrick at rivosinc dot com> ---
These failures are fixed as of d61efa3cd3378be38738bfb5139925d1505c1325

Confirmed using rv32 newlib.
https://github.com/patrick-rivos/riscv-gnu-toolchain/issues/139

[Bug target/110892] [14 Regression] Risc-V rvv testsuite failures caused by conditional move costing

[ewlu at rivosinc dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=110892

Edwin Lu <ewlu at rivosinc dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |FIXED

--- Comment #4 from Edwin Lu <ewlu at rivosinc dot com> ---
Fixed. See Patrick's comment for details