| Message ID | 20211021223850.415607-3-goldstein.w.n@gmail.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v8,1/3] String: Add support for __memcmpeq() ABI on all targets | expand |
On Thu, Oct 21, 2021 at 3:40 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > No bug. > > This commit adds tests for the new function __memcmpeq. The new tests > use the existing tests in 'test-memcmp.c' but relax the result > requirement to only check for zero or non-zero returns. > > All string tests include test-memcmpeq are passing. > --- > Note at the momement for test-memcmpeq.c the > tests only appear to do the generic versions. > > For example on x86_64 the output of test-memcmpeq.out is: > > > ``` > __memcmpeq simple_memcmp > ``` > > whereas test-memcmp.out is: > > ``` > simple_memcmp __memcmp_avx2_movbe > __memcmp_evex_movbe __memcmp_sse4_1 __memcmp_ssse3 __memcmp_sse2 > ``` > > So the test is not going through all available ifuncs. I'm not > sure if this is expected or indicates a bug in the previous > patches adding support for __memcmpeq. > This doesn't look right. If __memcmpeq is a true alias of memcmp, you should only have simple_memcmpeq and __memcmpeq.
On Thu, Oct 21, 2021 at 5:58 PM H.J. Lu <hjl.tools@gmail.com> wrote: > > On Thu, Oct 21, 2021 at 3:40 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > > > No bug. > > > > This commit adds tests for the new function __memcmpeq. The new tests > > use the existing tests in 'test-memcmp.c' but relax the result > > requirement to only check for zero or non-zero returns. > > > > All string tests include test-memcmpeq are passing. > > --- > > Note at the momement for test-memcmpeq.c the > > tests only appear to do the generic versions. > > > > For example on x86_64 the output of test-memcmpeq.out is: > > > > > > ``` > > __memcmpeq simple_memcmp > > ``` > > > > whereas test-memcmp.out is: > > > > ``` > > simple_memcmp __memcmp_avx2_movbe > > __memcmp_evex_movbe __memcmp_sse4_1 __memcmp_ssse3 __memcmp_sse2 > > ``` > > > > So the test is not going through all available ifuncs. I'm not > > sure if this is expected or indicates a bug in the previous > > patches adding support for __memcmpeq. > > > > This doesn't look right. If __memcmpeq is a true alias of memcmp, you > should only have simple_memcmpeq and __memcmpeq. Figured. Any idea what the issue might be? > > -- > H.J.
On Thu, Oct 21, 2021 at 4:47 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > On Thu, Oct 21, 2021 at 5:58 PM H.J. Lu <hjl.tools@gmail.com> wrote: > > > > On Thu, Oct 21, 2021 at 3:40 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > > > > > No bug. > > > > > > This commit adds tests for the new function __memcmpeq. The new tests > > > use the existing tests in 'test-memcmp.c' but relax the result > > > requirement to only check for zero or non-zero returns. > > > > > > All string tests include test-memcmpeq are passing. > > > --- > > > Note at the momement for test-memcmpeq.c the > > > tests only appear to do the generic versions. > > > > > > For example on x86_64 the output of test-memcmpeq.out is: > > > > > > > > > ``` > > > __memcmpeq simple_memcmp > > > ``` > > > > > > whereas test-memcmp.out is: > > > > > > ``` > > > simple_memcmp __memcmp_avx2_movbe > > > __memcmp_evex_movbe __memcmp_sse4_1 __memcmp_ssse3 __memcmp_sse2 > > > ``` > > > > > > So the test is not going through all available ifuncs. I'm not > > > sure if this is expected or indicates a bug in the previous > > > patches adding support for __memcmpeq. > > > > > > > This doesn't look right. If __memcmpeq is a true alias of memcmp, you > > should only have simple_memcmpeq and __memcmpeq. > > Figured. Any idea what the issue might be? Where is your branch? I can take a look.
On Thu, Oct 21, 2021 at 7:42 PM H.J. Lu <hjl.tools@gmail.com> wrote: > > On Thu, Oct 21, 2021 at 4:47 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > > > On Thu, Oct 21, 2021 at 5:58 PM H.J. Lu <hjl.tools@gmail.com> wrote: > > > > > > On Thu, Oct 21, 2021 at 3:40 PM Noah Goldstein <goldstein.w.n@gmail.com> wrote: > > > > > > > > No bug. > > > > > > > > This commit adds tests for the new function __memcmpeq. The new tests > > > > use the existing tests in 'test-memcmp.c' but relax the result > > > > requirement to only check for zero or non-zero returns. > > > > > > > > All string tests include test-memcmpeq are passing. > > > > --- > > > > Note at the momement for test-memcmpeq.c the > > > > tests only appear to do the generic versions. > > > > > > > > For example on x86_64 the output of test-memcmpeq.out is: > > > > > > > > > > > > ``` > > > > __memcmpeq simple_memcmp > > > > ``` > > > > > > > > whereas test-memcmp.out is: > > > > > > > > ``` > > > > simple_memcmp __memcmp_avx2_movbe > > > > __memcmp_evex_movbe __memcmp_sse4_1 __memcmp_ssse3 __memcmp_sse2 > > > > ``` > > > > > > > > So the test is not going through all available ifuncs. I'm not > > > > sure if this is expected or indicates a bug in the previous > > > > patches adding support for __memcmpeq. > > > > > > > > > > This doesn't look right. If __memcmpeq is a true alias of memcmp, you > > > should only have simple_memcmpeq and __memcmpeq. > > > > Figured. Any idea what the issue might be? > > Where is your branch? I can take a look. Made a new branch on your gitlab here: https://gitlab.com/x86-glibc/glibc/-/tree/users/goldsteinn/memcmpeq-abi > > -- > H.J.
diff --git a/string/Makefile b/string/Makefile index f0fce2a0b8..40d6fac133 100644 --- a/string/Makefile +++ b/string/Makefile @@ -48,8 +48,8 @@ routines := strcat strchr strcmp strcoll strcpy strcspn \ sigdescr_np sigabbrev_np strerrorname_np \ strerrordesc_np -strop-tests := memchr memcmp memcpy memmove mempcpy memset memccpy \ - stpcpy stpncpy strcat strchr strcmp strcpy strcspn \ +strop-tests := memchr memcmp memcpy memcmpeq memmove mempcpy memset \ + memccpy stpcpy stpncpy strcat strchr strcmp strcpy strcspn \ strlen strncmp strncpy strpbrk strrchr strspn memmem \ strstr strcasestr strnlen strcasecmp strncasecmp \ strncat rawmemchr strchrnul bcopy bzero memrchr \ diff --git a/string/test-memcmp.c b/string/test-memcmp.c index 6ddbc05d2f..e1f07cab58 100644 --- a/string/test-memcmp.c +++ b/string/test-memcmp.c @@ -17,11 +17,14 @@ <https://www.gnu.org/licenses/>. */ #define TEST_MAIN -#ifdef WIDE +#ifdef TEST_MEMCMPEQ +# define TEST_NAME "__memcmpeq" +#elif defined WIDE # define TEST_NAME "wmemcmp" #else # define TEST_NAME "memcmp" #endif + #include "test-string.h" #ifdef WIDE # include <inttypes.h> @@ -35,6 +38,7 @@ # define CHARBYTES 4 # define CHAR__MIN WCHAR_MIN # define CHAR__MAX WCHAR_MAX + int simple_wmemcmp (const wchar_t *s1, const wchar_t *s2, size_t n) { @@ -48,8 +52,11 @@ simple_wmemcmp (const wchar_t *s1, const wchar_t *s2, size_t n) } #else # include <limits.h> - -# define MEMCMP memcmp +# ifdef TEST_MEMCMPEQ +# define MEMCMP __memcmpeq +# else +# define MEMCMP memcmp +# endif # define MEMCPY memcpy # define SIMPLE_MEMCMP simple_memcmp # define CHAR char @@ -69,6 +76,12 @@ simple_memcmp (const char *s1, const char *s2, size_t n) } #endif +#ifndef BAD_RESULT +# define BAD_RESULT(result, expec) \ + (((result) == 0 && (expec)) || ((result) < 0 && (expec) >= 0) || \ + ((result) > 0 && (expec) <= 0)) +# endif + typedef int (*proto_t) (const CHAR *, const CHAR *, size_t); IMPL (SIMPLE_MEMCMP, 0) @@ -79,9 +92,7 @@ check_result (impl_t *impl, const CHAR *s1, const CHAR *s2, size_t len, int exp_result) { int result = CALL (impl, s1, s2, len); - if ((exp_result == 0 && result != 0) - || (exp_result < 0 && result >= 0) - || (exp_result > 0 && result <= 0)) + if (BAD_RESULT(result, exp_result)) { error (0, 0, "Wrong result in function %s %d %d", impl->name, result, exp_result); @@ -186,9 +197,7 @@ do_random_tests (void) { r = CALL (impl, (CHAR *) p1 + align1, (const CHAR *) p2 + align2, len); - if ((r == 0 && result) - || (r < 0 && result >= 0) - || (r > 0 && result <= 0)) + if (BAD_RESULT(r, result)) { error (0, 0, "Iteration %zd - wrong result in function %s (%zd, %zd, %zd, %zd) %ld != %d, p1 %p p2 %p", n, impl->name, align1 * CHARBYTES & 63, align2 * CHARBYTES & 63, len, pos, r, result, p1, p2); diff --git a/string/test-memcmpeq.c b/string/test-memcmpeq.c new file mode 100644 index 0000000000..8889117868 --- /dev/null +++ b/string/test-memcmpeq.c @@ -0,0 +1,21 @@ +/* Test and measure __memcmpeq functions. + Copyright (C) 2012-2021 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/>. */ + +#define BAD_RESULT(result, expec) ((!(result)) != (!(expec))) +#define TEST_MEMCMPEQ 1 +#include "test-memcmp.c"
No bug. This commit adds tests for the new function __memcmpeq. The new tests use the existing tests in 'test-memcmp.c' but relax the result requirement to only check for zero or non-zero returns. All string tests include test-memcmpeq are passing. --- Note at the momement for test-memcmpeq.c the tests only appear to do the generic versions. For example on x86_64 the output of test-memcmpeq.out is: ``` __memcmpeq simple_memcmp ``` whereas test-memcmp.out is: ``` simple_memcmp __memcmp_avx2_movbe __memcmp_evex_movbe __memcmp_sse4_1 __memcmp_ssse3 __memcmp_sse2 ``` So the test is not going through all available ifuncs. I'm not sure if this is expected or indicates a bug in the previous patches adding support for __memcmpeq. string/Makefile | 4 ++-- string/test-memcmp.c | 27 ++++++++++++++++++--------- string/test-memcmpeq.c | 21 +++++++++++++++++++++ 3 files changed, 41 insertions(+), 11 deletions(-) create mode 100644 string/test-memcmpeq.c