[Bug c/100009] New: -march=native doesn't recognize tigerlake

[Bug c/100009] New: -march=native doesn't recognize tigerlake

[cdotts at planetiq dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

            Bug ID: 100009
           Summary: -march=native doesn't recognize tigerlake
           Product: gcc
           Version: 9.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: cdotts at planetiq dot com
  Target Milestone: ---

Copied from email chain: 
We're using GCC v9.3.0 on a newer laptop, as packaged by Ubuntu for
20.04 LTS.  Our build for host-native tools naturally enables
cpu-specific optimization with `-march=native`.  The gcc driver is
passing -march=tigerlake down to cc1 (or cc1plus).  The driver
program's auto-detection is correct in that this is tiger lake
hardware.  However, it is not supported in this release of GCC.

How exactly is the driver coming up with this code name?  I'd like to
keep the Makefile set up as-is for the whole team.  Can we override
the definition of `native` through an environment variable or
something?  Ice lake client optimizations and instruction selection
appear to be working for our usage.  We can modify the Makefile to be
sensitive to an environment variable, but I'm wondering if GCC itself
can also be overridden in this way.

Response from Jim Wilson: 
Looks like a bug in gcc-9.  tigerlake was added to
gcc/config/i386/driver-i386.c but not to the arch_names_table in i386.c.  I
would suggest filing a bug report.  I don't think there is a way to workaround
this.  It needs to be fixed in the gcc source tree.

[Bug target/100009] [9 Regression] -march=native doesn't recognize tigerlake

[redi at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-04-10
      Known to fail|                            |9.3.1
            Summary|-march=native doesn't       |[9 Regression]
                   |recognize tigerlake         |-march=native doesn't
                   |                            |recognize tigerlake
                 CC|                            |hjl.tools at gmail dot com
             Status|UNCONFIRMED                 |NEW
      Known to work|                            |9.3.0
     Ever confirmed|0                           |1

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> ---
The "tigerlake" string was added to the branch by r9-8652, which is after the
9.3.0 release, so Ubuntu must have backported it (but not changed the version
number to 9.3.1?)

[Bug target/100009] [9 Regression] -march=native doesn't work on tigerlake

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
   Target Milestone|---                         |9.4
            Summary|[9 Regression]              |[9 Regression]
                   |-march=native doesn't       |-march=native doesn't work
                   |recognize tigerlake         |on tigerlake

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
Looking at the backport it looks like -march=native on a tigerlake system
will expand to -march=tigerlake but that's not recognized later since the
required support is missing.

[Bug target/100009] [9 Regression] -march=native doesn't work on tigerlake

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

Hongtao.liu <crazylht at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |crazylht at gmail dot com

--- Comment #3 from Hongtao.liu <crazylht at gmail dot com> ---

> Response from Jim Wilson: 
> Looks like a bug in gcc-9.  tigerlake was added to
> gcc/config/i386/driver-i386.c but not to the arch_names_table in i386.c.  I
> would suggest filing a bug report.  I don't think there is a way to
> workaround this.  It needs to be fixed in the gcc source tree.

Oops,
will backport r10-2664-ga9fcfec30f70c30883f53d4b1bd533fbea0e9fb2 (tigerlake
part) to gcc9.

[Bug target/100009] [9 Regression] -march=native doesn't work on tigerlake

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

--- Comment #4 from Hongtao.liu <crazylht at gmail dot com> ---

> Oops,
> will backport r10-2664-ga9fcfec30f70c30883f53d4b1bd533fbea0e9fb2 (tigerlake
> part) to gcc9.

PTA_AVX512VP2INTERSECT is enabled in GCC10, don't plan to backport to gcc9, so
in GCC9 -march=native wouldn't enable -mavx512vp2intersect, also kl and widekl.

[Bug target/100009] [9 Regression] -march=native doesn't work on tigerlake

[crazylht at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

--- Comment #5 from Hongtao.liu <crazylht at gmail dot com> ---
(In reply to Hongtao.liu from comment #3)
> > Response from Jim Wilson: 
> > Looks like a bug in gcc-9.  tigerlake was added to
> > gcc/config/i386/driver-i386.c but not to the arch_names_table in i386.c.  I
> > would suggest filing a bug report.  I don't think there is a way to
> > workaround this.  It needs to be fixed in the gcc source tree.
> 
> Oops,
> will backport r10-2664-ga9fcfec30f70c30883f53d4b1bd533fbea0e9fb2 (tigerlake
> part) to gcc9.

Fixed by r9-9351

[Bug target/100009] [9 Regression] -march=native doesn't work on tigerlake

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100009

Richard Biener <rguenth at gcc dot gnu.org> changed:

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

--- Comment #6 from Richard Biener <rguenth at gcc dot gnu.org> ---
So fixed (hopefully).