From: Sergey Kaplun via Tarantool-patches <tarantool-patches@dev.tarantool.org>
To: Sergey Bronnikov <estetus@gmail.com>
Cc: tarantool-patches@dev.tarantool.org
Subject: Re: [Tarantool-patches] [PATCH luajit 1/4][v2] ci: add a workflow for testing with AVX512
Date: Fri, 12 Apr 2024 13:27:40 +0300 [thread overview]
Message-ID: <ZhkMnDv9fNhiyHCx@root> (raw)
In-Reply-To: <6777a43e3012332d04493f93d6afe7fb4156af1b.1712841312.git.sergeyb@tarantool.org>
Hi, Sergey!
Thanks for the patch!
Please consider my comments below.
On 11.04.24, Sergey Bronnikov wrote:
> From: Sergey Bronnikov <sergeyb@tarantool.org>
>
> This commit adds a workflow for building and testing with enabled
> AVX512.
Typo: s/with enabled AVX512/with AVX512 enabled/
>
> Needed for tarantool/tarantool#9595
> Related to tarantool/tarantool#6787
Typo: s/Related/Relates/
| git log | grep Relates | wc -l
| 29
| git log | grep Related | wc -l
| 7
> ---
> .github/actions/setup-linux/action.yml | 12 +++++
> .github/workflows/avx512-build-testing.yml | 54 ++++++++++++++++++++++
> test/LuaJIT-tests/lib/ffi/bit64.lua | 2 +-
> 3 files changed, 67 insertions(+), 1 deletion(-)
> create mode 100644 .github/workflows/avx512-build-testing.yml
>
> diff --git a/.github/actions/setup-linux/action.yml b/.github/actions/setup-linux/action.yml
> index f0171b83..19bdcfa2 100644
> --- a/.github/actions/setup-linux/action.yml
> +++ b/.github/actions/setup-linux/action.yml
> @@ -17,3 +17,15 @@ runs:
> apt -y update
> apt -y install cmake gcc make ninja-build perl
> shell: bash
> + - name: Detect a presence of AVX512
I suppose it shouldn't be a part of setup-linux action, but just a one
step in testing-avx512.yml, see [1] for details of steps content usage.
Is it possible or did I miss some GH-action API trick here?
But anyway it should be done not in the same action.
> + id: avx512_script
> + run: |
> + # Set avx512_support environment variable to 'true' when AVX512
Typo: s/avx512_support/the avx512_support/
> + # is supported and 'false' otherwise.
The comment is a little bit misleading since the resulting values are
"0" or "1". Changing it to 'true' or 'false' looks like a good idea and
simplifies reading later.
> + #
> + # Normally `grep` has the exit status is 0 if a line is
Typo: s/Normally/Normally,/
Also, "has the exit status is" has two verbs, so I suggest to
reformulate this like the following:
| Normally, the exit status of `grep` is 0 if a line is selected, 1 if
| no lines were selected, and 2 if an error occurred.
> + # selected, 1 if no lines were selected, and 2 if an error
> + # occurred.
> + avx512_support=$(grep avx /proc/cpuinfo 2>&1 > /dev/null; [[ $? == 0 ]] && echo 0 || echo 1)
Why not just echo the "$?"?
On my laptop without avx512 support this command returns true.
| $ grep avx /proc/cpuinfo 2>&1 > /dev/null; [[ $? == 0 ]] && echo 0 || echo 1
| 0
I have no avx512 support, but avx|avx2 support. Hence, there is matching
content:
| $ grep -o -P '.{4}avx.{4}' /proc/cpuinfo
| ave avx f16
| mi1 avx2 sm
| ave avx f16
| mi1 avx2 sm
| ave avx f16
| mi1 avx2 sm
| ave avx f16
| mi1 avx2 sm
I suppose we should check for the "AVX512F" CPUID feature flag: its
supports allows to emit vcvttsd2usi [2] instruction that causes troubles.
> + echo "avx512_support=$avx512_support" >> $GITHUB_OUTPUT
Minor: It will be nice to echo some message about the value of this
variable. Hence, we can check this from the GitHub logs.
Side note: Do I understand correctly that GH-actions output is always a
string that evaluates to `true` regardless of its content?
> + shell: bash
> diff --git a/.github/workflows/avx512-build-testing.yml b/.github/workflows/avx512-build-testing.yml
> new file mode 100644
> index 00000000..d70fa661
> --- /dev/null
> +++ b/.github/workflows/avx512-build-testing.yml
<snipped>
> +
> +jobs:
> + test-avx512:
Minor: I suggest adding a name for the job to be consistent with other
jobs. Something like "LuaJIT avx512""
> + strategy:
> + fail-fast: false
> + runs-on: [self-hosted, regular, x86_64, Linux]
> + steps:
> + - uses: actions/checkout@v4
> + with:
> + fetch-depth: 0
> + submodules: recursive
> + - name: setup Linux
> + uses: ./.github/actions/setup-linux
> + - name: configure
> + if: needs.avx512_script.outputs.avx512_support == 0
> + run: >
> + cmake -S . -B ${{ env.BUILDDIR }}
> + -G Ninja
> + -DCMAKE_BUILD_TYPE=RelWithDebInfo
Please add a comment. Why do we check the non-debug build here?
If there is no reason for it, maybe it is better to check both: the
debug and release builds within a simple matrix.
> + -DCMAKE_C_FLAGS=-march=skylake-avx512
> + -DCMAKE_C_COMPILER=gcc
Do we need to specify compiler manually?
> + - name: build
> + if: needs.avx512_script.outputs.avx512_support == 0
> + run: cmake --build ${{ env.BUILDDIR }} --parallel
> + - name: run regression tests
> + if: needs.avx512_script.outputs.avx512_support == 0
> + run: cmake --build ${{ env.BUILDDIR }} --parallel --target LuaJIT-test
> diff --git a/test/LuaJIT-tests/lib/ffi/bit64.lua b/test/LuaJIT-tests/lib/ffi/bit64.lua
> index d1b47bef..cf3a96eb 100644
> --- a/test/LuaJIT-tests/lib/ffi/bit64.lua
> +++ b/test/LuaJIT-tests/lib/ffi/bit64.lua
> @@ -41,7 +41,7 @@ do --- tobit/band assorted C types
> end
> end
>
> -do --- tobit/band negative unsigned enum
> +do --- tobit/band negative unsigned enum -avx512
Should we also to set such variable in the CMakeLists.txt when we met
the corresponding condition?
> local x = ffi.new("uenum_t", -10)
> local y = tobit(x)
> local z = band(x)
> --
> 2.34.1
[1]: https://docs.github.com/en/actions/learn-github-actions/contexts#example-usage-of-the-steps-context
[2]: https://www.laruence.com/x86/VCVTTSD2USI.html
--
Best regards,
Sergey Kaplun
next prev parent reply other threads:[~2024-04-12 10:31 UTC|newest]
Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-04-11 13:22 [Tarantool-patches] [PATCH luajit 0/4][v2] Mach-O generation fixes Sergey Bronnikov via Tarantool-patches
2024-04-11 13:22 ` [Tarantool-patches] [PATCH luajit 1/4][v2] ci: add a workflow for testing with AVX512 Sergey Bronnikov via Tarantool-patches
2024-04-12 10:27 ` Sergey Kaplun via Tarantool-patches [this message]
2024-04-16 11:53 ` Sergey Bronnikov via Tarantool-patches
2024-04-18 8:24 ` Sergey Kaplun via Tarantool-patches
2024-05-05 12:29 ` Maxim Kokryashkin via Tarantool-patches
2024-06-14 13:55 ` Sergey Bronnikov via Tarantool-patches
2024-06-14 15:24 ` Sergey Bronnikov via Tarantool-patches
2024-04-11 13:22 ` [Tarantool-patches] [PATCH luajit 2/4][v2] test: introduce a helper read_file Sergey Bronnikov via Tarantool-patches
2024-04-12 10:47 ` Sergey Kaplun via Tarantool-patches
2024-04-16 12:02 ` Sergey Bronnikov via Tarantool-patches
2024-06-20 12:14 ` Sergey Bronnikov via Tarantool-patches
2024-04-11 13:22 ` [Tarantool-patches] [PATCH luajit 3/4][v2] OSX/iOS/ARM64: Fix generation of Mach-O object files Sergey Bronnikov via Tarantool-patches
2024-04-12 11:19 ` Sergey Kaplun via Tarantool-patches
2024-04-16 15:29 ` Sergey Bronnikov via Tarantool-patches
2024-06-13 15:50 ` Sergey Bronnikov via Tarantool-patches
2024-04-11 13:22 ` [Tarantool-patches] [PATCH luajit 4/4][v2] OSX/iOS/ARM64: Fix bytecode embedding in Mach-O object file Sergey Bronnikov via Tarantool-patches
2024-04-12 11:27 ` Sergey Kaplun via Tarantool-patches
2024-06-13 15:40 ` Sergey Bronnikov via Tarantool-patches
2024-06-13 15:47 ` Sergey Bronnikov via Tarantool-patches
2024-06-20 10:15 ` [Tarantool-patches] [PATCH luajit 0/4][v2] Mach-O generation fixes Sergey Kaplun via Tarantool-patches
2024-07-09 8:07 ` Sergey Kaplun via Tarantool-patches
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=ZhkMnDv9fNhiyHCx@root \
--to=tarantool-patches@dev.tarantool.org \
--cc=estetus@gmail.com \
--cc=skaplun@tarantool.org \
--subject='Re: [Tarantool-patches] [PATCH luajit 1/4][v2] ci: add a workflow for testing with AVX512' \
/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
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox