Mirror of the gdb-patches mailing list
 help / color / mirror / Atom feed
From: Luis <luis.machado.foss@gmail.com>
To: Lancelot SIX <Lancelot.Six@amd.com>,
	Luis Machado <luis.machado@amd.com>,
	 gdb-patches@sourceware.org
Subject: Re: [PATCH] gdb/testsuite: deduplicate --offload-arch targets in hcc_amdgpu_targets
Date: Fri, 3 Jul 2026 09:33:51 +0100	[thread overview]
Message-ID: <4dfd7ace-ade6-434f-8a63-f447f383dfb0@gmail.com> (raw)
In-Reply-To: <0cc2ad49-8ced-4f1b-a236-5f0efff00ddf@amd.com>

On 01/07/2026 17:00, Lancelot SIX wrote:
> 
> 
> On 29/06/2026 14:56, Luis Machado wrote:
>> On systems with multiple identical GPUs, find_amdgpu_devices returns one
>> entry per device, causing hcc_amdgpu_targets to emit redundant
>> --offload-arch flags when compiling HIP test programs (e.g. eight
>> --offload-arch=gfx942 flags on an 8-GPU MI300 system).
>>
>> Deduplicate the target list in hcc_amdgpu_targets, preserving order, so
>> each unique architecture appears exactly once regardless of how many
>> physical devices share it.  The same deduplication is applied when the
>> list comes from the HCC_AMDGPU_TARGET environment variable.  An array
>> is used for the seen-set to give O(1) membership tests.
>>
>> Add gdb.rocm/hcc-amdgpu-targets.exp to unit-test the deduplication
>> logic across both the env-var and device-enumeration code paths, without
>> requiring a GPU.
> 
> Hi Luis,
> 
> This LGTM.  Thanks for doing this.
> 
> Best,
> Lancelot.
> 
> Approved-by: Lancelot Six <lancelot.six@amd.com> (amdgpu)
> 

Thanks. Pushed now.


>> ---
>>   gdb/testsuite/gdb.rocm/hcc-amdgpu-targets.exp | 80 +++++++++++++++++++
>>   gdb/testsuite/lib/rocm.exp                    | 20 ++++-
>>   2 files changed, 97 insertions(+), 3 deletions(-)
>>   create mode 100644 gdb/testsuite/gdb.rocm/hcc-amdgpu-targets.exp
>>
>> diff --git a/gdb/testsuite/gdb.rocm/hcc-amdgpu-targets.exp b/gdb/ 
>> testsuite/gdb.rocm/hcc-amdgpu-targets.exp
>> new file mode 100644
>> index 00000000000..8f04cca7981
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/hcc-amdgpu-targets.exp
>> @@ -0,0 +1,80 @@
>> +# Copyright (C) 2026 Free Software Foundation, Inc.
>> +
>> +# This file is part of GDB.
>> +
>> +# This program is free software; you can redistribute it and/or modify
>> +# it under the terms of the GNU General Public License as published by
>> +# the Free Software Foundation; either version 3 of the License, or
>> +# (at your option) any later version.
>> +
>> +# This program is distributed in the hope that it will be useful,
>> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
>> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
>> +# GNU General Public License for more details.
>> +
>> +# You should have received a copy of the GNU General Public License
>> +# along with this program.  If not, see <http://www.gnu.org/licenses/>.
>> +
>> +# Unit tests for hcc_amdgpu_targets.  These exercise the deduplication
>> +# logic across both the env-var and device-enumeration code paths, and
>> +# do not require a GPU or a running GDB instance.
>> +
>> +load_lib rocm.exp
>> +
>> +# Run BODY with find_amdgpu_devices stubbed to return DEVICES, then
>> +# restore the original proc.
>> +proc with_stub_devices {devices body} {
>> +    rename find_amdgpu_devices __saved_find_amdgpu_devices
>> +    # tclint-disable-next-line command-args
>> +    proc find_amdgpu_devices {} [list return $devices]
>> +    set code [catch {uplevel 1 $body} result]
>> +    rename find_amdgpu_devices {}
>> +    rename __saved_find_amdgpu_devices find_amdgpu_devices
>> +    return -code $code $result
>> +}
>> +
>> +# Tests using find_amdgpu_devices (no HCC_AMDGPU_TARGET env var).
>> +
>> +save_vars { env(HCC_AMDGPU_TARGET) } {
>> +    unset -nocomplain ::env(HCC_AMDGPU_TARGET)
>> +
>> +    with_stub_devices {gfx942 gfx942 gfx942} {
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {gfx942}} "duplicates removed: 3x gfx942 
>> produces single entry"
>> +    }
>> +
>> +    with_stub_devices {gfx942 gfx942 gfx1100 gfx1100 gfx942} {
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {gfx942 gfx1100}} "duplicates removed, 
>> order preserved: mixed list"
>> +    }
>> +
>> +    with_stub_devices {gfx906 gfx90a} {
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {gfx906 gfx90a}} "no duplicates: distinct 
>> devices unchanged"
>> +    }
>> +
>> +    with_stub_devices {} {
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {}} "empty device list returned as-is"
>> +    }
>> +}
>> +
>> +# Tests using HCC_AMDGPU_TARGET env var.
>> +
>> +save_vars { env(HCC_AMDGPU_TARGET) } {
>> +    set ::env(HCC_AMDGPU_TARGET) "gfx942,gfx942,gfx942"
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {gfx942}} "env var: duplicates removed"
>> +}
>> +
>> +save_vars { env(HCC_AMDGPU_TARGET) } {
>> +    set ::env(HCC_AMDGPU_TARGET) "gfx906,gfx90a,gfx906"
>> +    set result [hcc_amdgpu_targets]
>> +    verbose -log "hcc_amdgpu_targets: got $result"
>> +    gdb_assert {$result eq {gfx906 gfx90a}} "env var: order preserved 
>> when deduplicating"
>> +}
>> diff --git a/gdb/testsuite/lib/rocm.exp b/gdb/testsuite/lib/rocm.exp
>> index 15ee22dad26..ca1ae3f1a8b 100644
>> --- a/gdb/testsuite/lib/rocm.exp
>> +++ b/gdb/testsuite/lib/rocm.exp
>> @@ -80,20 +80,34 @@ gdb_caching_proc find_amdgpu_devices {} {
>>       return $hip_gpu_devices
>>   }
>> -# Get the list of GPU targets to compile for.
>> +# Get the list of unique GPU targets to compile for.
>>   #
>>   # If HCC_AMDGPU_TARGET is set in the environment, use it.
>>   # Otherwise, consider the devices available on the system.
>> +#
>> +# Duplicates are removed so that systems with multiple identical
>> +# GPUs do not produce redundant --offload-arch flags.
>>   proc hcc_amdgpu_targets {} {
>>       # First, look for HCC_AMDGPU_TARGET (same env var hipcc uses).
>>       if {[info exists ::env(HCC_AMDGPU_TARGET)]} {
>>       # We don't verify the contents of HCC_AMDGPU_TARGET.
>>       # That's the toolchain's job.
>> -    return [split $::env(HCC_AMDGPU_TARGET) ","]
>> +    set targets [split $::env(HCC_AMDGPU_TARGET) ","]
>> +    } else {
>> +    set targets [find_amdgpu_devices]
>>       }
>> -    return [find_amdgpu_devices]
>> +    # Remove duplicates while preserving order.
>> +    array set seen {}
>> +    set unique {}
>> +    foreach t $targets {
>> +    if {![info exists seen($t)]} {
>> +        set seen($t) 1
>> +        lappend unique $t
>> +    }
>> +    }
>> +    return $unique
>>   }
>>   gdb_caching_proc allow_hipcc_tests {} {
> 


      reply	other threads:[~2026-07-03  8:34 UTC|newest]

Thread overview: 3+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-06-29 13:56 Luis Machado
2026-07-01 16:00 ` Lancelot SIX
2026-07-03  8:33   ` Luis [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=4dfd7ace-ade6-434f-8a63-f447f383dfb0@gmail.com \
    --to=luis.machado.foss@gmail.com \
    --cc=Lancelot.Six@amd.com \
    --cc=gdb-patches@sourceware.org \
    --cc=luis.machado@amd.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox