From: Adhemerval Zanella <adhemerval.zanella@linaro.org>
To: libc-alpha@sourceware.org, Florian Weimer <fweimer@redhat.com>
Subject: Re: [PATCH 10/11] x86_64: Add glibc-hwcaps support
Date: Wed, 2 Dec 2020 09:49:58 -0300 [thread overview]
Message-ID: <1b9e38fc-3d40-bd14-be2f-103f3331b2a1@linaro.org> (raw)
In-Reply-To: <3ae02ef23f2981bb4a7602b4259c91b84ef8e939.1604946656.git.fweimer@redhat.com>
On 09/11/2020 15:41, Florian Weimer via Libc-alpha wrote:
> The subdirectories match those in the x86-64 psABI:
>
> https://gitlab.com/x86-psABIs/x86-64-ABI/-/commit/77566eb03bc6a326811cb7e9a6b9396884b67c7c
LGTM, thanks.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
> ---
> elf/Makefile | 2 +-
> elf/tst-glibc-hwcaps-cache.script | 10 ++++
> sysdeps/x86_64/Makefile | 39 +++++++++++++++
> sysdeps/x86_64/dl-hwcaps-subdirs.c | 66 ++++++++++++++++++++++++++
> sysdeps/x86_64/tst-glibc-hwcaps.c | 76 ++++++++++++++++++++++++++++++
> 5 files changed, 192 insertions(+), 1 deletion(-)
> create mode 100644 sysdeps/x86_64/dl-hwcaps-subdirs.c
> create mode 100644 sysdeps/x86_64/tst-glibc-hwcaps.c
>
> diff --git a/elf/Makefile b/elf/Makefile
> index 80e94b9ee2..0a44b19a0e 100644
> --- a/elf/Makefile
> +++ b/elf/Makefile
> @@ -1826,7 +1826,7 @@ $(objpfx)argv0test.out: tst-rtld-argv0.sh $(objpfx)ld.so \
>
> # Most likely search subdirectories, for each supported architecture.
> # Used to obtain test coverage wide test coverage.
> -glibc-hwcaps-first-subdirs-for-tests =
> +glibc-hwcaps-first-subdirs-for-tests = x86-64-v2
>
> # The test modules are parameterized by preprocessor macros.
> LDFLAGS-libmarkermod1-1.so += -Wl,-soname,libmarkermod1.so
Ok.
> diff --git a/elf/tst-glibc-hwcaps-cache.script b/elf/tst-glibc-hwcaps-cache.script
> index 46cb5fd553..2aef9fb177 100644
> --- a/elf/tst-glibc-hwcaps-cache.script
> +++ b/elf/tst-glibc-hwcaps-cache.script
> @@ -1,2 +1,12 @@
> # test-container does not support scripts in sysdeps directories, so
> # collect everything in one file.
> +
> +mkdirp 0770 $L/glibc-hwcaps/x86-64-v2
> +cp $B/elf/libmarkermod2-2.so $L/glibc-hwcaps/x86-64-v2/libmarkermod2.so
> +mkdirp 0770 $L/glibc-hwcaps/x86-64-v3
> +cp $B/elf/libmarkermod3-2.so $L/glibc-hwcaps/x86-64-v2/libmarkermod3.so
> +cp $B/elf/libmarkermod3-3.so $L/glibc-hwcaps/x86-64-v3/libmarkermod3.so
> +mkdirp 0770 $L/glibc-hwcaps/x86-64-v4
> +cp $B/elf/libmarkermod4-2.so $L/glibc-hwcaps/x86-64-v2/libmarkermod4.so
> +cp $B/elf/libmarkermod4-3.so $L/glibc-hwcaps/x86-64-v3/libmarkermod4.so
> +cp $B/elf/libmarkermod4-4.so $L/glibc-hwcaps/x86-64-v4/libmarkermod4.so
Ok.
> diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile
> index 42b97c5cc7..d1d7cb9d2e 100644
> --- a/sysdeps/x86_64/Makefile
> +++ b/sysdeps/x86_64/Makefile
> @@ -144,8 +144,47 @@ CFLAGS-tst-auditmod10b.c += $(AVX512-CFLAGS)
> CFLAGS-tst-avx512-aux.c += $(AVX512-CFLAGS)
> CFLAGS-tst-avx512mod.c += $(AVX512-CFLAGS)
> endif
> +
> +$(objpfx)tst-glibc-hwcaps: $(objpfx)libmarkermod2-1.so \
> + $(objpfx)libmarkermod3-1.so $(objpfx)libmarkermod4-1.so
> +$(objpfx)tst-glibc-hwcaps.out: \
> + $(objpfx)libmarkermod2.so \
> + $(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod2.so \
> + $(objpfx)libmarkermod3.so \
> + $(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod3.so \
> + $(objpfx)glibc-hwcaps/x86-64-v3/libmarkermod3.so \
> + $(objpfx)libmarkermod4.so \
> + $(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod4.so \
> + $(objpfx)glibc-hwcaps/x86-64-v3/libmarkermod4.so \
> + $(objpfx)glibc-hwcaps/x86-64-v4/libmarkermod4.so \
> +
> +$(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod2.so: $(objpfx)libmarkermod2-2.so
> + $(make-target-directory)
> + cp $< $@
> +$(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod3.so: $(objpfx)libmarkermod3-2.so
> + $(make-target-directory)
> + cp $< $@
> +$(objpfx)glibc-hwcaps/x86-64-v3/libmarkermod3.so: $(objpfx)libmarkermod3-3.so
> + $(make-target-directory)
> + cp $< $@
> +$(objpfx)glibc-hwcaps/x86-64-v2/libmarkermod4.so: $(objpfx)libmarkermod4-2.so
> + $(make-target-directory)
> + cp $< $@
> +$(objpfx)glibc-hwcaps/x86-64-v3/libmarkermod4.so: $(objpfx)libmarkermod4-3.so
> + $(make-target-directory)
> + cp $< $@
> +$(objpfx)glibc-hwcaps/x86-64-v4/libmarkermod4.so: $(objpfx)libmarkermod4-4.so
> + $(make-target-directory)
> + cp $< $@
> +
> +ifeq (no,$(build-hardcoded-path-in-tests))
> +# This is an ld.so.cache test, and RPATH/RUNPATH in the executable
> +# interferes with its test objectives.
> +tests-container += tst-glibc-hwcaps-cache
> endif
>
> +endif # $(subdir) == elf
> +
> ifeq ($(subdir),csu)
> gen-as-const-headers += tlsdesc.sym rtld-offsets.sym
> endif
Ok.
> diff --git a/sysdeps/x86_64/dl-hwcaps-subdirs.c b/sysdeps/x86_64/dl-hwcaps-subdirs.c
> new file mode 100644
> index 0000000000..8810a822ef
> --- /dev/null
> +++ b/sysdeps/x86_64/dl-hwcaps-subdirs.c
> @@ -0,0 +1,66 @@
> +/* Architecture-specific glibc-hwcaps subdirectories. x86 version.
> + Copyright (C) 2020 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <dl-hwcaps.h>
> +#include <cpu-features.h>
> +
> +const char _dl_hwcaps_subdirs[] = "x86-64-v4:x86-64-v3:x86-64-v2";
> +enum { subdirs_count = 3 }; /* Number of components in _dl_hwcaps_subdirs. */
> +
> +uint32_t
> +_dl_hwcaps_subdirs_active (void)
> +{
> + int active = 0;
> +
> + /* Test in reverse preference order. */
> +
> + /* x86-64-v2. */
> + if (!(CPU_FEATURE_USABLE (CMPXCHG16B)
> + && CPU_FEATURE_USABLE (LAHF64_SAHF64)
> + && CPU_FEATURE_USABLE (POPCNT)
> + && CPU_FEATURE_USABLE (SSE3)
> + && CPU_FEATURE_USABLE (SSE4_1)
> + && CPU_FEATURE_USABLE (SSE4_2)
> + && CPU_FEATURE_USABLE (SSSE3)))
> + return _dl_hwcaps_subdirs_build_bitmask (subdirs_count, active);
> + ++active;
> +
Ok, it seems to match the x86-64-ABI/low-level-sys-info.tex description.
> + /* x86-64-v3. */
> + if (!(CPU_FEATURE_USABLE (AVX)
> + && CPU_FEATURE_USABLE (AVX2)
> + && CPU_FEATURE_USABLE (BMI1)
> + && CPU_FEATURE_USABLE (BMI2)
> + && CPU_FEATURE_USABLE (F16C)
> + && CPU_FEATURE_USABLE (FMA)
> + && CPU_FEATURE_USABLE (LZCNT)
> + && CPU_FEATURE_USABLE (MOVBE)
> + && CPU_FEATURE_USABLE (OSXSAVE)))
> + return _dl_hwcaps_subdirs_build_bitmask (subdirs_count, active);
> + ++active;
Ok.
> +
> + /* x86-64-v4. */
> + if (!(CPU_FEATURE_USABLE (AVX512F)
> + && CPU_FEATURE_USABLE (AVX512BW)
> + && CPU_FEATURE_USABLE (AVX512CD)
> + && CPU_FEATURE_USABLE (AVX512DQ)
> + && CPU_FEATURE_USABLE (AVX512VL)))
> + return _dl_hwcaps_subdirs_build_bitmask (subdirs_count, active);
> + ++active;
> +
Ok.
> + return _dl_hwcaps_subdirs_build_bitmask (subdirs_count, active);
> +}
> diff --git a/sysdeps/x86_64/tst-glibc-hwcaps.c b/sysdeps/x86_64/tst-glibc-hwcaps.c
> new file mode 100644
> index 0000000000..3075a8286d
> --- /dev/null
> +++ b/sysdeps/x86_64/tst-glibc-hwcaps.c
> @@ -0,0 +1,76 @@
> +/* glibc-hwcaps subdirectory test. x86_64 version.
> + Copyright (C) 2020 Free Software Foundation, Inc.
> + This file is part of the GNU C Library.
> +
> + The GNU C Library is free software; you can redistribute it and/or
> + modify it under the terms of the GNU Lesser General Public
> + License as published by the Free Software Foundation; either
> + version 2.1 of the License, or (at your option) any later version.
> +
> + The GNU C Library 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
> + Lesser General Public License for more details.
> +
> + You should have received a copy of the GNU Lesser General Public
> + License along with the GNU C Library; if not, see
> + <https://www.gnu.org/licenses/>. */
> +
> +#include <stdio.h>
> +#include <support/check.h>
> +#include <sys/param.h>
> +#include <sys/platform/x86.h>
> +
> +extern int marker2 (void);
> +extern int marker3 (void);
> +extern int marker4 (void);
> +
> +/* Return the x86-64-vN level, 1 for the baseline. */
> +static int
> +compute_level (void)
> +{
> + const struct cpu_features *cpu_features
> + = __x86_get_cpu_features (COMMON_CPUID_INDEX_MAX);
> +
> + if (!(CPU_FEATURE_USABLE_P (cpu_features, CMPXCHG16B)
> + && CPU_FEATURE_USABLE_P (cpu_features, LAHF64_SAHF64)
> + && CPU_FEATURE_USABLE_P (cpu_features, POPCNT)
> + && CPU_FEATURE_USABLE_P (cpu_features, MMX)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSE)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSE2)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSE3)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSSE3)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSE4_1)
> + && CPU_FEATURE_USABLE_P (cpu_features, SSE4_2)))
> + return 1;
This slight deviates from _dl_hwcaps_subdirs_active, since it is considering
MMX and SSE2.
> + if (!(CPU_FEATURE_USABLE_P (cpu_features, AVX)
> + && CPU_FEATURE_USABLE_P (cpu_features, AVX2)
> + && CPU_FEATURE_USABLE_P (cpu_features, BMI1)
> + && CPU_FEATURE_USABLE_P (cpu_features, BMI2)
> + && CPU_FEATURE_USABLE_P (cpu_features, F16C)
> + && CPU_FEATURE_USABLE_P (cpu_features, FMA)
> + && CPU_FEATURE_USABLE_P (cpu_features, LZCNT)
> + && CPU_FEATURE_USABLE_P (cpu_features, MOVBE)
> + && CPU_FEATURE_USABLE_P (cpu_features, OSXSAVE)))
> + return 2;
> + if (!(CPU_FEATURE_USABLE_P (cpu_features, AVX512F)
> + && CPU_FEATURE_USABLE_P (cpu_features, AVX512BW)
> + && CPU_FEATURE_USABLE_P (cpu_features, AVX512CD)
> + && CPU_FEATURE_USABLE_P (cpu_features, AVX512DQ)
> + && CPU_FEATURE_USABLE_P (cpu_features, AVX512VL)))
> + return 3;
> + return 4;
> +}
> +
> +static int
> +do_test (void)
> +{
> + int level = compute_level ();
> + printf ("info: detected x86-64 micro-architecture level: %d\n", level);
> + TEST_COMPARE (marker2 (), MIN (level, 2));
> + TEST_COMPARE (marker3 (), MIN (level, 3));
> + TEST_COMPARE (marker4 (), MIN (level, 4));
> + return 0;
> +}
> +
> +#include <support/test-driver.c>
>
Ok.
I am wondering if it would be possible to add an internal tests where
it tests the hwcap selection without relying on underlying capabilities
so we could decouple the test from the runtime where it is ran.
It might be possible to simulate it through qemu. Without it, for x86_64
it would require to change the cpu_features internal state to simulate
different x86_64 abi-set. Not sure how feasible it would be.
next prev parent reply other threads:[~2020-12-02 12:50 UTC|newest]
Thread overview: 58+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-11-09 18:40 [PATCH v4 00/11] " Florian Weimer
2020-11-09 18:40 ` [PATCH 01/11] support: Add support_copy_file Florian Weimer
2020-11-26 16:43 ` Adhemerval Zanella
2020-11-26 17:22 ` Florian Weimer
2020-11-09 18:40 ` [PATCH 02/11] elf: Introduce enum opt_format in the ldconfig implementation Florian Weimer
2020-11-26 16:44 ` Adhemerval Zanella
2020-11-26 18:54 ` [PATCH 12/11] s390x: Add Add glibc-hwcaps support Florian Weimer
2020-12-07 8:16 ` Florian Weimer
2020-12-08 15:47 ` Stefan Liebler
2020-12-09 18:52 ` Florian Weimer
2020-12-10 10:22 ` Stefan Liebler
2020-12-10 14:54 ` Florian Weimer
2020-11-09 18:40 ` [PATCH 03/11] elf: Add glibc-hwcaps support for LD_LIBRARY_PATH Florian Weimer
2020-11-26 20:29 ` Adhemerval Zanella
2020-11-27 19:49 ` Florian Weimer
2020-11-30 18:59 ` Adhemerval Zanella
2020-12-04 12:25 ` Matheus Castanho
2020-12-04 12:43 ` Florian Weimer
2020-12-04 12:59 ` Matheus Castanho
2020-11-09 18:40 ` [PATCH 04/11] elf: Add endianness markup to ld.so.cache Florian Weimer
2020-11-27 13:56 ` Adhemerval Zanella
2020-11-27 19:49 ` Florian Weimer
2020-11-30 19:00 ` Adhemerval Zanella
2020-11-09 18:40 ` [PATCH 05/11] elf: Add extension mechanism " Florian Weimer
2020-11-27 18:01 ` Adhemerval Zanella
2020-11-27 18:55 ` Florian Weimer
2020-11-27 18:56 ` Adhemerval Zanella
2020-11-09 18:40 ` [PATCH 06/11] elf: Implement a string table for ldconfig, with tail merging Florian Weimer
2020-11-27 19:29 ` Adhemerval Zanella
2020-11-27 19:49 ` Florian Weimer
2020-11-30 19:01 ` Adhemerval Zanella
2020-11-09 18:41 ` [PATCH 07/11] elf: Implement tail merging of strings in ldconfig Florian Weimer
2020-11-30 18:41 ` Adhemerval Zanella
2020-12-01 10:28 ` Florian Weimer
2020-11-09 18:41 ` [PATCH 08/11] elf: Process glibc-hwcaps subdirectories " Florian Weimer
2020-11-30 19:46 ` Adhemerval Zanella
2020-11-09 18:41 ` [PATCH 09/11] elf: Add glibc-hwcaps subdirectory support to ld.so cache processing Florian Weimer
2020-11-26 17:11 ` Florian Weimer
2020-12-01 17:45 ` Adhemerval Zanella
2020-12-01 20:45 ` Florian Weimer
2020-12-02 12:08 ` Adhemerval Zanella
2020-12-02 12:16 ` Florian Weimer
2020-11-09 18:41 ` [PATCH 10/11] x86_64: Add glibc-hwcaps support Florian Weimer
2020-12-02 12:49 ` Adhemerval Zanella [this message]
2020-11-09 18:41 ` [PATCH 11/11] powerpc64le: " Florian Weimer
2020-12-02 13:46 ` Adhemerval Zanella
2020-12-02 13:51 ` Florian Weimer
2020-12-02 14:27 ` Adhemerval Zanella
2020-12-02 14:31 ` Florian Weimer
2020-12-02 15:34 ` Carlos Seo
2020-12-02 20:14 ` Tulio Magno Quites Machado Filho
2020-12-04 8:56 ` Florian Weimer
2020-12-04 12:35 ` Adhemerval Zanella
2020-11-09 21:56 ` [PATCH v4 00/11] " Dan Horák
2020-11-10 10:41 ` Florian Weimer
2020-11-25 15:46 ` Florian Weimer
2020-11-25 15:58 ` H.J. Lu
2020-11-25 17:20 ` Adhemerval Zanella
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=1b9e38fc-3d40-bd14-be2f-103f3331b2a1@linaro.org \
--to=adhemerval.zanella@linaro.org \
--cc=fweimer@redhat.com \
--cc=libc-alpha@sourceware.org \
/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).