From: Simon Marchi <simon.marchi@efficios.com>
To: Lancelot SIX <lsix@lancelotsix.com>
Cc: gdb-patches@sourceware.org
Subject: Re: [PATCH 2/2] gdb/amdgpu: add precise-memory support
Date: Thu, 14 Sep 2023 11:51:41 -0400 [thread overview]
Message-ID: <1dbadcb5-908a-41b4-b951-90e795386764@efficios.com> (raw)
In-Reply-To: <20230913213250.cwgwrowfrawelfac@octopus>
>> @@ -1326,6 +1338,36 @@ amd_dbgapi_target::stopped_by_hw_breakpoint ()
>> return false;
>> }
>>
>> +/* Set the process's memory access reporting precision.
>> +
>> + The precision can be ::AMD_DBGAPI_MEMORY_PRECISION_PRECISE (waves wait for
>> + memory instructions to complete before executing further instructions), or
>> + ::AMD_DBGAPI_MEMORY_PRECISION_NONE (memory instructions execute normally).
>> +
>> + Returns true if the precision is supported by the architecture of all agents
>> + in the process, or false if at least one agent does not support the
>> + requested precision.
>> +
>> + An error is thrown if setting the precision results in a status other than
>> + ::AMD_DBGAPI_STATUS_SUCCESS or ::AMD_DBGAPI_STATUS_ERROR_NOT_SUPPORTED. */
>> +
>
> Would it be simpler if this helper function received a bool parameter
> instead of the amd_dbgapi_memory_precision_t one? This could avoid
> repeating this
>
> amd_dbgapi_memory_precision_t memory_precision
> = (info->precise_memory.requested
> ? AMD_DBGAPI_MEMORY_PRECISION_PRECISE
> : AMD_DBGAPI_MEMORY_PRECISION_NONE);
>
> before calling it.
Indeed. In fact, we can factor out more... I came up with:
static void
try_set_process_memory_precision (amd_dbgapi_inferior_info &info)
{
auto mode = (info.precise_memory.requested
? AMD_DBGAPI_MEMORY_PRECISION_PRECISE
: AMD_DBGAPI_MEMORY_PRECISION_NONE);
amd_dbgapi_status_t status
= amd_dbgapi_set_memory_precision (info.process_id, mode);
if (status == AMD_DBGAPI_STATUS_ERROR_NOT_SUPPORTED)
warning (_("AMDGPU precise memory access reporting could not be enabled."));
else if (status != AMD_DBGAPI_STATUS_SUCCESS)
error (_("amd_dbgapi_set_memory_precision failed (%s)"),
get_status_string (status));
}
... such that callers just need to do:
try_set_process_memory_precision (*info);
I'll put that in v2, unless you think it's a bad idea.
>> @@ -1785,6 +1873,29 @@ amd_dbgapi_remove_breakpoint_callback
>> return AMD_DBGAPI_STATUS_SUCCESS;
>> }
>>
>> +/* signal_received observer. */
>> +
>> +static void
>> +amd_dbgapi_target_signal_received (gdb_signal sig)
>> +{
>> + amd_dbgapi_inferior_info *info
>> + = get_amd_dbgapi_inferior_info (current_inferior ());
>> +
>> + if (info->process_id == AMD_DBGAPI_PROCESS_NONE)
>> + return;
>> +
>> + if (!ptid_is_gpu (inferior_thread ()->ptid))
>> + return;
>> +
>> + if (sig != GDB_SIGNAL_SEGV && sig != GDB_SIGNAL_BUS)
>> + return;
>> +
>> + if (!info->precise_memory.enabled)
>> + gdb_printf ("\
>
> I think there should be a _() surrounding the string.
Done.
>> +The @code{set amdgpu precise-memory} parameter is per-inferior. When an
> ^
> Isn't the parameter name just "amdgpu precise-memory"?
Yes, done.
>> diff --git a/gdb/testsuite/gdb.rocm/precise-memory-exec.c b/gdb/testsuite/gdb.rocm/precise-memory-exec.c
>> new file mode 100644
>> index 000000000000..f0659a63fc5a
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/precise-memory-exec.c
>> @@ -0,0 +1,44 @@
>> +/* Copyright 2021-2023 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/>. */
>> +
>> +#include <unistd.h>
>> +#include <stdlib.h>
>> +#include <stdio.h>
>> +
>> +static void
>> +second (void)
>> +{
>> +}
>> +
>> +int
>> +main (int argc, char **argv)
>> +{
>> + if (argc == 1)
>> + {
>> + /* First invocation */
>
> Should the comment end with ". "?
Done.
>
>> + int ret = execl (argv[0], argv[0], "Hello", NULL);
>> + perror ("exec");
>> + abort ();
>> + }
>> + else
>> + {
>> + /* Second invocation */
>
> Here also.
Done.
>
>> + second ();
>> + }
>> +
>> + return 0;
>> +}
>> diff --git a/gdb/testsuite/gdb.rocm/precise-memory-exec.exp b/gdb/testsuite/gdb.rocm/precise-memory-exec.exp
>> new file mode 100644
>> index 000000000000..26be6cf72146
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/precise-memory-exec.exp
>> @@ -0,0 +1,62 @@
>> +# Copyright 2021-2023 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/>.
>> +
>> +# Test that the "set amdgpu precise-memory" setting is inherited by an inferior
>> +# created following an exec.
>> +
>> +load_lib rocm.exp
>> +
>> +require allow_hipcc_tests
>> +
>> +if { ![istarget "*-linux*"] } then {
>> + continue
>> +}
>
> Should this test be integrated in allow_hipcc_test? This would avoid
> having to repeat it in multiple testcases (and all testcases do not have
> such guard).
>
> Also, I'm not sure if there is any non-linux configuration which can
> satisfy allow_hipcc_tests, which would make this test redundant.
Yeah I think we can put it in allow_hipcc_tests. If/when we want to run
this test on, let's say, Windows, we will be able to change the check in
allow_hipcc_tests.
The existing tests in the upstream repo don't have the linux check, I
guess there should have been some checks. I'll add a new preparatory
patch in v2.
>> diff --git a/gdb/testsuite/gdb.rocm/precise-memory-warning-sigsegv.cpp b/gdb/testsuite/gdb.rocm/precise-memory-warning-sigsegv.cpp
>> new file mode 100644
>> index 000000000000..58339e5391a6
>> --- /dev/null
>> +++ b/gdb/testsuite/gdb.rocm/precise-memory-warning-sigsegv.cpp
>> @@ -0,0 +1,33 @@
>> +/* Copyright 2021-2023 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/>. */
>> +
>> +#include <hip/hip_runtime.h>
>> +
>> +__global__ void
>> +kernel ()
>> +{
>> + int *p = nullptr;
>> + *p = 1;
>> +}
>> +
>> +int
>> +main (int argc, char* argv[])
>> +{
>> + hipLaunchKernelGGL (kernel, dim3 (1), dim3 (1), 0, 0);
>
> I think the "modern" way to write this would be:
>
> kernel<<<1, 1>>> ();
>
> This is mostly a remark, I don't mind using hipLaunchKernelGGL too much
> either.
I think we wanted to migrate all to the new form, let's go with the new
form.
>> diff --git a/gdb/testsuite/lib/rocm.exp b/gdb/testsuite/lib/rocm.exp
>> index 98a3b308228d..22b294a5efae 100644
>> --- a/gdb/testsuite/lib/rocm.exp
>> +++ b/gdb/testsuite/lib/rocm.exp
>> @@ -99,6 +99,56 @@ gdb_caching_proc allow_hipcc_tests {} {
>> return 1
>> }
>>
>> +# ROCM_PATH is used by hipcc as well.
>> +if {[info exists env(ROCM_PATH)]} {
>> + set rocm_path $env(ROCM_PATH)
>> +} else {
>> + set rocm_path "/opt/rocm"
>> +}
>> +
>> +# Get the gpu target to be passed to e.g., -mcpu=.
>> +#
>> +# If HCC_AMDGPU_TARGET is set in the environment, use it. Otherwise,
>> +# try reading it from the system using the rocm_agent_enumerator
>> +# utility.
>> +
>> +proc hcc_amdgpu_target {} {
>
> There is a hcc_amdgpu_targets proc which enumerates the architecture of
> each agent present on the system. This is a fairly recent addition, it
> might have been introduced after you prepared this series.
Oh, that's true, no need to introduce hcc_amdgpu_target.
>> @@ -186,3 +236,12 @@ proc hip_devices_support_debug_multi_process {} {
>> }
>> return 1
>> }
>> +
>> +# Return true if the device supports precise memory.
>
> Using hcc_amdgpu_targets, you could have a function which checks that
> all agents in the system support precise memory, not just the first one
> detected. This will reflect how `set amdgpu precise-memory` works.
Oh, right, we have this downstream. I sync'ed the patch with the
downstream code.
Thanks, will send a v2.
Simon
next prev parent reply other threads:[~2023-09-14 15:51 UTC|newest]
Thread overview: 10+ messages / expand[flat|nested] mbox.gz Atom feed top
2023-09-13 15:28 [PATCH 1/2] gdb: add inferior_cloned observable Simon Marchi
2023-09-13 15:28 ` [PATCH 2/2] gdb/amdgpu: add precise-memory support Simon Marchi
2023-09-13 21:32 ` Lancelot SIX
2023-09-14 15:51 ` Simon Marchi [this message]
2023-09-14 14:10 ` Tom Tromey
2023-09-14 16:00 ` Simon Marchi
2023-09-14 16:18 ` Tom Tromey
2023-09-14 16:18 ` Simon Marchi
2023-09-14 14:08 ` [PATCH 1/2] gdb: add inferior_cloned observable Tom Tromey
2023-09-14 15:53 ` Simon Marchi
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=1dbadcb5-908a-41b4-b951-90e795386764@efficios.com \
--to=simon.marchi@efficios.com \
--cc=gdb-patches@sourceware.org \
--cc=lsix@lancelotsix.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;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).