* [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
@ 2021-07-06 10:51 Anton Blanchard
2021-07-06 10:51 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
` (3 more replies)
0 siblings, 4 replies; 11+ messages in thread
From: Anton Blanchard @ 2021-07-06 10:51 UTC (permalink / raw)
To: tuliom; +Cc: libc-alpha
We use PPC_FEATURE_HAS_VSX to select a number of POWER7 optimised
functions. These functions don't use any VSX instructions, so
PPC_FEATURE_ARCH_2_06 seems like a better fit.
---
sysdeps/powerpc/powerpc64/multiarch/memchr.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/memcmp.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/memrchr.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/memset.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/stpncpy.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strchr.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strchrnul.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strcmp.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strlen.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strncase.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strncase_l.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strncmp.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strncpy.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strnlen.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strrchr.c | 2 +-
sysdeps/powerpc/powerpc64/multiarch/strstr.c | 2 +-
19 files changed, 19 insertions(+), 19 deletions(-)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr.c b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
index 0c718d4f15..c24186689e 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
@@ -30,7 +30,7 @@ extern __typeof (__memchr) __memchr_power8 attribute_hidden;
libc_ifunc (__memchr,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __memchr_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __memchr_power7
: __memchr_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
index 4fd089aba7..99559bce26 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
@@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect_memcmp, memcmp,
#endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __memcmp_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __memcmp_power7
: (hwcap & PPC_FEATURE_POWER4)
? __memcmp_power4
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
index e06d6468b8..16bb6f0042 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
@@ -30,7 +30,7 @@ extern __typeof (__memrchr) __memrchr_power8 attribute_hidden;
libc_ifunc (__memrchr,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __memrchr_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __memrchr_power7
: __memrchr_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c
index 5994bf02e6..c1aa143f60 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memset.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c
@@ -48,7 +48,7 @@ libc_ifunc (__libc_memset,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __memset_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __memset_power7 :
(hwcap & PPC_FEATURE_ARCH_2_05)
? __memset_power6 :
diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
index c0ffea2b93..b5d2d3a635 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
@@ -41,7 +41,7 @@ libc_ifunc_redirected (__redirect___rawmemchr, __rawmemchr,
(hwcap2 & PPC_FEATURE2_ARCH_3_00)
? __rawmemchr_power9 :
# endif
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __rawmemchr_power7
: __rawmemchr_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
index bebd377fd9..e7035761a7 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
@@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect___stpncpy, __stpncpy,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __stpncpy_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06)
? __stpncpy_power7
: __stpncpy_ppc);
weak_alias (__stpncpy, stpncpy)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
index dcd7774403..55ca6c85c4 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
@@ -29,7 +29,7 @@ extern __typeof (__strcasecmp) __strcasecmp_power8 attribute_hidden;
libc_ifunc (__libc_strcasecmp,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strcasecmp_power8:
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strcasecmp_power7
: __strcasecmp_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
index 96a70b8b11..1afee5d7fd 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
@@ -32,7 +32,7 @@ extern __typeof (__strcasecmp_l) __strcasecmp_l_power7 attribute_hidden;
extern __typeof (__strcasecmp_l) __libc_strcasecmp_l;
libc_ifunc (__libc_strcasecmp_l,
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strcasecmp_l_power7
: __strcasecmp_l_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr.c b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
index ea9ac1134f..27c794c6b7 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
@@ -35,7 +35,7 @@ extern __typeof (strchr) __strchr_power8 attribute_hidden;
libc_ifunc_redirected (__redirect_strchr, strchr,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strchr_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strchr_power7
: __strchr_ppc);
weak_alias (strchr, index)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
index 4688e7c3f0..4a07b4a242 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
@@ -30,7 +30,7 @@ extern __typeof (__strchrnul) __strchrnul_power8 attribute_hidden;
libc_ifunc (__strchrnul,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strchrnul_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strchrnul_power7
: __strchrnul_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
index 72f9a639bf..4b0b25fff6 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
@@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect_strcmp, strcmp,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strcmp_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06)
? __strcmp_power7
: __strcmp_ppc);
#endif
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen.c b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
index 109c8a90bd..0cd1c6faff 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strlen.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
@@ -42,7 +42,7 @@ libc_ifunc (__libc_strlen,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strlen_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strlen_power7
: __strlen_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase.c b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
index 2013a5d75a..644046bd74 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncase.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
@@ -29,7 +29,7 @@ extern __typeof (__strncasecmp) __strncasecmp_power8 attribute_hidden;
libc_ifunc (__libc_strncasecmp,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strncasecmp_power8:
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strncasecmp_power7
: __strncasecmp_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
index cad6da302d..d2d761af72 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
@@ -34,7 +34,7 @@ extern __typeof (__strncasecmp_l) __strncasecmp_l_power7 attribute_hidden;
ifunc symbol properly. */
extern __typeof (__strncasecmp_l) __libc_strncasecmp_l;
libc_ifunc (__libc_strncasecmp_l,
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strncasecmp_l_power7
: __strncasecmp_l_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
index eef524ddfb..1f689e5c05 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
@@ -43,7 +43,7 @@ libc_ifunc_redirected (__redirect_strncmp, strncmp,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strncmp_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06)
? __strncmp_power7
: (hwcap & PPC_FEATURE_POWER4)
? __strncmp_power4
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncpy.c b/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
index 7da9def358..d4d3463bd1 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
@@ -43,7 +43,7 @@ libc_ifunc_redirected (__redirect_strncpy, strncpy,
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strncpy_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06)
? __strncpy_power7
: __strncpy_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
index 264b7a752d..baf375a75a 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
@@ -31,7 +31,7 @@ extern __typeof (__strnlen) __strnlen_power8 attribute_hidden;
libc_ifunc_redirected (__redirect___strnlen, __strnlen,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strnlen_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strnlen_power7
: __strnlen_ppc);
weak_alias (__strnlen, strnlen)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
index bb06b93d19..1c9eea1817 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
@@ -33,7 +33,7 @@ extern __typeof (strrchr) __strrchr_power8 attribute_hidden;
libc_ifunc_redirected (__redirect_strrchr, strrchr,
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
? __strrchr_power8 :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strrchr_power7
: __strrchr_ppc);
weak_alias (strrchr, rindex)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strstr.c b/sysdeps/powerpc/powerpc64/multiarch/strstr.c
index bb0588844e..6582798dda 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strstr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strstr.c
@@ -30,7 +30,7 @@ extern __typeof (strstr) __strstr_power7 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc_redirected (__redirect_strstr, strstr,
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06)
? __strstr_power7
: __strstr_ppc);
#endif
--
2.31.1
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines
2021-07-06 10:51 [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Anton Blanchard
@ 2021-07-06 10:51 ` Anton Blanchard
2021-07-08 22:33 ` Tulio Magno Quites Machado Filho
2021-07-06 10:51 ` [PATCH 3/3] powerpc64: Add checks for Altivec and VSX in ifunc selection Anton Blanchard
` (2 subsequent siblings)
3 siblings, 1 reply; 11+ messages in thread
From: Anton Blanchard @ 2021-07-06 10:51 UTC (permalink / raw)
To: tuliom; +Cc: libc-alpha
A number of optimised memset routines assume the cacheline size is 128B,
so we better check before using them.
---
sysdeps/powerpc/powerpc64/multiarch/memset.c | 15 ++++++++++-----
1 file changed, 10 insertions(+), 5 deletions(-)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c
index c1aa143f60..056e911699 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memset.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c
@@ -43,16 +43,21 @@ libc_ifunc (__libc_memset,
# ifdef __LITTLE_ENDIAN__
(hwcap2 & PPC_FEATURE2_ARCH_3_1
&& hwcap2 & PPC_FEATURE2_HAS_ISEL
- && hwcap & PPC_FEATURE_HAS_VSX)
+ && hwcap & PPC_FEATURE_HAS_VSX
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power10 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power8 :
- (hwcap & PPC_FEATURE_ARCH_2_06)
+ (hwcap & PPC_FEATURE_ARCH_2_06
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power7 :
- (hwcap & PPC_FEATURE_ARCH_2_05)
+ (hwcap & PPC_FEATURE_ARCH_2_05
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power6 :
- (hwcap & PPC_FEATURE_POWER4)
+ (hwcap & PPC_FEATURE_POWER4
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power4
: __memset_ppc);
--
2.31.1
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCH 3/3] powerpc64: Add checks for Altivec and VSX in ifunc selection
2021-07-06 10:51 [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Anton Blanchard
2021-07-06 10:51 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
@ 2021-07-06 10:51 ` Anton Blanchard
2021-07-06 15:05 ` [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Adhemerval Zanella
2021-07-08 22:29 ` Tulio Magno Quites Machado Filho
3 siblings, 0 replies; 11+ messages in thread
From: Anton Blanchard @ 2021-07-06 10:51 UTC (permalink / raw)
To: tuliom; +Cc: libc-alpha
We'd like to support processors without Altivec or VSX, so check
the relevant hwcap bits before selecting them.
---
sysdeps/powerpc/powerpc64/multiarch/bzero.c | 6 ++++--
sysdeps/powerpc/powerpc64/multiarch/memchr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/memcmp.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/memcpy.c | 7 +++++--
sysdeps/powerpc/powerpc64/multiarch/memmove.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/mempcpy.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/memrchr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/memset.c | 1 +
sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/stpcpy.c | 9 ++++++---
sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strcasestr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strcat.c | 6 ++++--
sysdeps/powerpc/powerpc64/multiarch/strchr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strchrnul.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strcmp.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strcpy.c | 9 ++++++---
sysdeps/powerpc/powerpc64/multiarch/strcspn.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strlen.c | 9 ++++++---
sysdeps/powerpc/powerpc64/multiarch/strncase.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strncat.c | 6 ++++--
sysdeps/powerpc/powerpc64/multiarch/strncmp.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strnlen.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strrchr.c | 3 ++-
sysdeps/powerpc/powerpc64/multiarch/strspn.c | 3 ++-
25 files changed, 70 insertions(+), 34 deletions(-)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/bzero.c b/sysdeps/powerpc/powerpc64/multiarch/bzero.c
index 660d7dc686..6275305b46 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/bzero.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/bzero.c
@@ -38,11 +38,13 @@ libc_ifunc (__bzero,
&& hwcap & PPC_FEATURE_HAS_VSX)
? __bzero_power10 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __bzero_power8 :
(hwcap & PPC_FEATURE_HAS_VSX)
? __bzero_power7 :
- (hwcap & PPC_FEATURE_ARCH_2_05)
+ (hwcap & PPC_FEATURE_ARCH_2_05
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __bzero_power6 :
(hwcap & PPC_FEATURE_POWER4)
? __bzero_power4
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr.c b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
index c24186689e..f40013e061 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
@@ -28,7 +28,8 @@ extern __typeof (__memchr) __memchr_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc (__memchr,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __memchr_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __memchr_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
index 99559bce26..89b56c103b 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
@@ -38,7 +38,8 @@ libc_ifunc_redirected (__redirect_memcmp, memcmp,
&& hwcap & PPC_FEATURE_HAS_VSX)
? __memcmp_power10 :
#endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __memcmp_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __memcmp_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcpy.c b/sysdeps/powerpc/powerpc64/multiarch/memcpy.c
index 53ab32ef26..684ee064f2 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memcpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memcpy.c
@@ -45,9 +45,12 @@ libc_ifunc (__libc_memcpy,
(hwcap2 & PPC_FEATURE2_ARCH_3_1 && hwcap & PPC_FEATURE_HAS_VSX)
? __memcpy_power10 :
# endif
- ((hwcap2 & PPC_FEATURE2_ARCH_2_07) && use_cached_memopt)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC
+ && use_cached_memopt)
? __memcpy_power8_cached :
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __memcpy_power7 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __memcpy_a2 :
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memmove.c b/sysdeps/powerpc/powerpc64/multiarch/memmove.c
index 637b2cbf7f..50253b4554 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memmove.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memmove.c
@@ -41,7 +41,8 @@ libc_ifunc (__libc_memmove,
&& hwcap & PPC_FEATURE_HAS_VSX)
? __memmove_power10 :
#endif
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __memmove_power7
: __memmove_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c b/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c
index b37e0f35b5..563095a5ec 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/mempcpy.c
@@ -33,7 +33,8 @@ extern __typeof (__mempcpy) __mempcpy_power7 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc_redirected (__redirect___mempcpy, __mempcpy,
- (hwcap & PPC_FEATURE_HAS_VSX)
+ (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __mempcpy_power7
: __mempcpy_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
index 16bb6f0042..a8b985b06a 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
@@ -28,7 +28,8 @@ extern __typeof (__memrchr) __memrchr_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc (__memrchr,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __memrchr_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __memrchr_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c
index 056e911699..a2bc223bcc 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memset.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c
@@ -48,6 +48,7 @@ libc_ifunc (__libc_memset,
? __memset_power10 :
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC
&& GLRO(dl_cache_line_size) == 128)
? __memset_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06
diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
index b5d2d3a635..43eb459e02 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
@@ -38,7 +38,8 @@ libc_ifunc_redirected (__redirect___rawmemchr, __rawmemchr,
(hwcap2 & PPC_FEATURE2_ARCH_3_1)
&& (hwcap & PPC_FEATURE_HAS_VSX)
? __rawmemchr_power10 :
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __rawmemchr_power9 :
# endif
(hwcap & PPC_FEATURE_ARCH_2_06)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
index d4eb4285fc..5be413405e 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/stpcpy.c
@@ -32,12 +32,15 @@ extern __typeof (__stpcpy) __stpcpy_power9 attribute_hidden;
libc_ifunc_hidden (__stpcpy, __stpcpy,
# ifdef __LITTLE_ENDIAN__
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __stpcpy_power9 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __stpcpy_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __stpcpy_power7
: __stpcpy_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
index 55ca6c85c4..21ce2d279b 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
@@ -27,7 +27,8 @@ extern __typeof (__strcasecmp) __strcasecmp_power7 attribute_hidden;
extern __typeof (__strcasecmp) __strcasecmp_power8 attribute_hidden;
libc_ifunc (__libc_strcasecmp,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strcasecmp_power8:
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strcasecmp_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasestr.c b/sysdeps/powerpc/powerpc64/multiarch/strcasestr.c
index 7e4bd3b5ac..5bb3016022 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcasestr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcasestr.c
@@ -27,7 +27,8 @@ extern __typeof (__strcasestr) __strcasestr_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc (__strcasestr,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strcasestr_power8
: __strcasestr_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcat.c b/sysdeps/powerpc/powerpc64/multiarch/strcat.c
index 6d342324c4..d8d9870824 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcat.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcat.c
@@ -28,9 +28,11 @@ extern __typeof (strcat) __strcat_power8 attribute_hidden;
# undef strcat
libc_ifunc_redirected (__redirect_strcat, strcat,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strcat_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strcat_power7
: __strcat_ppc);
#endif
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr.c b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
index 27c794c6b7..62b202baf9 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
@@ -33,7 +33,8 @@ extern __typeof (strchr) __strchr_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc_redirected (__redirect_strchr, strchr,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strchr_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strchr_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
index 4a07b4a242..40e529b9d9 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
@@ -28,7 +28,8 @@ extern __typeof (__strchrnul) __strchrnul_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc (__strchrnul,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strchrnul_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strchrnul_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
index 4b0b25fff6..8132682a99 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
@@ -35,7 +35,8 @@ extern __typeof (strcmp) __strcmp_power9 attribute_hidden;
libc_ifunc_redirected (__redirect_strcmp, strcmp,
# ifdef __LITTLE_ENDIAN__
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strcmp_power9 :
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcpy.c b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c
index b733fa5a23..5af1d45cc1 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcpy.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcpy.c
@@ -32,12 +32,15 @@ extern __typeof (strcpy) __strcpy_power9 attribute_hidden;
libc_ifunc_redirected (__redirect_strcpy, strcpy,
# ifdef __LITTLE_ENDIAN__
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strcpy_power9 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strcpy_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strcpy_power7
: __strcpy_ppc);
#endif
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcspn.c b/sysdeps/powerpc/powerpc64/multiarch/strcspn.c
index 683aa104d7..8ba01c13b1 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strcspn.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strcspn.c
@@ -27,7 +27,8 @@ extern __typeof (strcspn) __strcspn_ppc attribute_hidden;
extern __typeof (strcspn) __strcspn_power8 attribute_hidden;
libc_ifunc (__libc_strcspn,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strcspn_power8
: __strcspn_ppc);
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen.c b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
index 0cd1c6faff..f1e28414e0 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strlen.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
@@ -35,12 +35,15 @@ extern __typeof (__redirect_strlen) __strlen_power10 attribute_hidden;
libc_ifunc (__libc_strlen,
# ifdef __LITTLE_ENDIAN__
- (hwcap2 & PPC_FEATURE2_ARCH_3_1)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_1
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strlen_power10 :
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strlen_power9 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strlen_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strlen_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase.c b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
index 644046bd74..2802cf2c3f 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncase.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
@@ -27,7 +27,8 @@ extern __typeof (__strncasecmp) __strncasecmp_power7 attribute_hidden;
extern __typeof (__strncasecmp) __strncasecmp_power8 attribute_hidden;
libc_ifunc (__libc_strncasecmp,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strncasecmp_power8:
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strncasecmp_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncat.c b/sysdeps/powerpc/powerpc64/multiarch/strncat.c
index 0036fca91a..9ea294a72d 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncat.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncat.c
@@ -26,9 +26,11 @@ extern __typeof (strncat) __strncat_power7 attribute_hidden;
extern __typeof (strncat) __strncat_power8 attribute_hidden;
libc_ifunc (strncat,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strncat_power8
- : (hwcap & PPC_FEATURE_HAS_VSX)
+ : (hwcap & PPC_FEATURE_ARCH_2_06
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strncat_power7
: __strncat_ppc);
#endif
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
index 1f689e5c05..2d21122854 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
@@ -38,7 +38,8 @@ extern __typeof (strncmp) __strncmp_power9 attribute_hidden;
ifunc symbol properly. */
libc_ifunc_redirected (__redirect_strncmp, strncmp,
# ifdef __LITTLE_ENDIAN__
- (hwcap2 & PPC_FEATURE2_ARCH_3_00)
+ (hwcap2 & PPC_FEATURE2_ARCH_3_00
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strncmp_power9 :
# endif
(hwcap2 & PPC_FEATURE2_ARCH_2_07)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
index baf375a75a..e68e9d9f88 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
@@ -29,7 +29,8 @@ extern __typeof (__strnlen) __strnlen_power8 attribute_hidden;
# undef strnlen
# undef __strnlen
libc_ifunc_redirected (__redirect___strnlen, __strnlen,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strnlen_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strnlen_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
index 1c9eea1817..7f0cf2a1b7 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
@@ -31,7 +31,8 @@ extern __typeof (strrchr) __strrchr_power8 attribute_hidden;
/* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
ifunc symbol properly. */
libc_ifunc_redirected (__redirect_strrchr, strrchr,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_ALTIVEC)
? __strrchr_power8 :
(hwcap & PPC_FEATURE_ARCH_2_06)
? __strrchr_power7
diff --git a/sysdeps/powerpc/powerpc64/multiarch/strspn.c b/sysdeps/powerpc/powerpc64/multiarch/strspn.c
index 70167a176b..7613ab3d55 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/strspn.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/strspn.c
@@ -27,7 +27,8 @@ extern __typeof (strspn) __strspn_ppc attribute_hidden;
extern __typeof (strspn) __strspn_power8 attribute_hidden;
libc_ifunc (__libc_strspn,
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && hwcap & PPC_FEATURE_HAS_VSX)
? __strspn_power8
: __strspn_ppc);
--
2.31.1
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
2021-07-06 10:51 [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Anton Blanchard
2021-07-06 10:51 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
2021-07-06 10:51 ` [PATCH 3/3] powerpc64: Add checks for Altivec and VSX in ifunc selection Anton Blanchard
@ 2021-07-06 15:05 ` Adhemerval Zanella
2021-07-08 22:29 ` Tulio Magno Quites Machado Filho
3 siblings, 0 replies; 11+ messages in thread
From: Adhemerval Zanella @ 2021-07-06 15:05 UTC (permalink / raw)
To: Anton Blanchard, tuliom; +Cc: libc-alpha
On 06/07/2021 07:51, Anton Blanchard via Libc-alpha wrote:
> We use PPC_FEATURE_HAS_VSX to select a number of POWER7 optimised
> functions. These functions don't use any VSX instructions, so
> PPC_FEATURE_ARCH_2_06 seems like a better fit.
I think we should rename the symbols to something more meanifull as
well, like 'symbol_vsx'. I take the idea is to support powerpc
variants with ISA 2.06 that do not support VSX.
The patch looks good thanks and I assume you are testing on microwatt
(so you would see a failure is VSX is actually being used).
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
> ---
> sysdeps/powerpc/powerpc64/multiarch/memchr.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/memcmp.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/memrchr.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/memset.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/stpncpy.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strchr.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strchrnul.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strcmp.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strlen.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strncase.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strncase_l.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strncmp.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strncpy.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strnlen.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strrchr.c | 2 +-
> sysdeps/powerpc/powerpc64/multiarch/strstr.c | 2 +-
> 19 files changed, 19 insertions(+), 19 deletions(-)
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memchr.c b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
> index 0c718d4f15..c24186689e 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/memchr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/memchr.c
> @@ -30,7 +30,7 @@ extern __typeof (__memchr) __memchr_power8 attribute_hidden;
> libc_ifunc (__memchr,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __memchr_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __memchr_power7
> : __memchr_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
> index 4fd089aba7..99559bce26 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/memcmp.c
> @@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect_memcmp, memcmp,
> #endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __memcmp_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __memcmp_power7
> : (hwcap & PPC_FEATURE_POWER4)
> ? __memcmp_power4
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
> index e06d6468b8..16bb6f0042 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/memrchr.c
> @@ -30,7 +30,7 @@ extern __typeof (__memrchr) __memrchr_power8 attribute_hidden;
> libc_ifunc (__memrchr,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __memrchr_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __memrchr_power7
> : __memrchr_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c
> index 5994bf02e6..c1aa143f60 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/memset.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c
> @@ -48,7 +48,7 @@ libc_ifunc (__libc_memset,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __memset_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __memset_power7 :
> (hwcap & PPC_FEATURE_ARCH_2_05)
> ? __memset_power6 :
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
> index c0ffea2b93..b5d2d3a635 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/rawmemchr.c
> @@ -41,7 +41,7 @@ libc_ifunc_redirected (__redirect___rawmemchr, __rawmemchr,
> (hwcap2 & PPC_FEATURE2_ARCH_3_00)
> ? __rawmemchr_power9 :
> # endif
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __rawmemchr_power7
> : __rawmemchr_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
> index bebd377fd9..e7035761a7 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/stpncpy.c
> @@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect___stpncpy, __stpncpy,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __stpncpy_power8
> - : (hwcap & PPC_FEATURE_HAS_VSX)
> + : (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __stpncpy_power7
> : __stpncpy_ppc);
> weak_alias (__stpncpy, stpncpy)
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
> index dcd7774403..55ca6c85c4 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp.c
> @@ -29,7 +29,7 @@ extern __typeof (__strcasecmp) __strcasecmp_power8 attribute_hidden;
> libc_ifunc (__libc_strcasecmp,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strcasecmp_power8:
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strcasecmp_power7
> : __strcasecmp_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
> index 96a70b8b11..1afee5d7fd 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strcasecmp_l.c
> @@ -32,7 +32,7 @@ extern __typeof (__strcasecmp_l) __strcasecmp_l_power7 attribute_hidden;
>
> extern __typeof (__strcasecmp_l) __libc_strcasecmp_l;
> libc_ifunc (__libc_strcasecmp_l,
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strcasecmp_l_power7
> : __strcasecmp_l_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchr.c b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
> index ea9ac1134f..27c794c6b7 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strchr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strchr.c
> @@ -35,7 +35,7 @@ extern __typeof (strchr) __strchr_power8 attribute_hidden;
> libc_ifunc_redirected (__redirect_strchr, strchr,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strchr_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strchr_power7
> : __strchr_ppc);
> weak_alias (strchr, index)
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
> index 4688e7c3f0..4a07b4a242 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strchrnul.c
> @@ -30,7 +30,7 @@ extern __typeof (__strchrnul) __strchrnul_power8 attribute_hidden;
> libc_ifunc (__strchrnul,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strchrnul_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strchrnul_power7
> : __strchrnul_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
> index 72f9a639bf..4b0b25fff6 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strcmp.c
> @@ -40,7 +40,7 @@ libc_ifunc_redirected (__redirect_strcmp, strcmp,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strcmp_power8
> - : (hwcap & PPC_FEATURE_HAS_VSX)
> + : (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strcmp_power7
> : __strcmp_ppc);
> #endif
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strlen.c b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
> index 109c8a90bd..0cd1c6faff 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strlen.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strlen.c
> @@ -42,7 +42,7 @@ libc_ifunc (__libc_strlen,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strlen_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strlen_power7
> : __strlen_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase.c b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
> index 2013a5d75a..644046bd74 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strncase.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase.c
> @@ -29,7 +29,7 @@ extern __typeof (__strncasecmp) __strncasecmp_power8 attribute_hidden;
> libc_ifunc (__libc_strncasecmp,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strncasecmp_power8:
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strncasecmp_power7
> : __strncasecmp_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
> index cad6da302d..d2d761af72 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strncase_l.c
> @@ -34,7 +34,7 @@ extern __typeof (__strncasecmp_l) __strncasecmp_l_power7 attribute_hidden;
> ifunc symbol properly. */
> extern __typeof (__strncasecmp_l) __libc_strncasecmp_l;
> libc_ifunc (__libc_strncasecmp_l,
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strncasecmp_l_power7
> : __strncasecmp_l_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
> index eef524ddfb..1f689e5c05 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strncmp.c
> @@ -43,7 +43,7 @@ libc_ifunc_redirected (__redirect_strncmp, strncmp,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strncmp_power8
> - : (hwcap & PPC_FEATURE_HAS_VSX)
> + : (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strncmp_power7
> : (hwcap & PPC_FEATURE_POWER4)
> ? __strncmp_power4
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strncpy.c b/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
> index 7da9def358..d4d3463bd1 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strncpy.c
> @@ -43,7 +43,7 @@ libc_ifunc_redirected (__redirect_strncpy, strncpy,
> # endif
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strncpy_power8
> - : (hwcap & PPC_FEATURE_HAS_VSX)
> + : (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strncpy_power7
> : __strncpy_ppc);
>
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
> index 264b7a752d..baf375a75a 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strnlen.c
> @@ -31,7 +31,7 @@ extern __typeof (__strnlen) __strnlen_power8 attribute_hidden;
> libc_ifunc_redirected (__redirect___strnlen, __strnlen,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strnlen_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strnlen_power7
> : __strnlen_ppc);
> weak_alias (__strnlen, strnlen)
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
> index bb06b93d19..1c9eea1817 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strrchr.c
> @@ -33,7 +33,7 @@ extern __typeof (strrchr) __strrchr_power8 attribute_hidden;
> libc_ifunc_redirected (__redirect_strrchr, strrchr,
> (hwcap2 & PPC_FEATURE2_ARCH_2_07)
> ? __strrchr_power8 :
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strrchr_power7
> : __strrchr_ppc);
> weak_alias (strrchr, rindex)
> diff --git a/sysdeps/powerpc/powerpc64/multiarch/strstr.c b/sysdeps/powerpc/powerpc64/multiarch/strstr.c
> index bb0588844e..6582798dda 100644
> --- a/sysdeps/powerpc/powerpc64/multiarch/strstr.c
> +++ b/sysdeps/powerpc/powerpc64/multiarch/strstr.c
> @@ -30,7 +30,7 @@ extern __typeof (strstr) __strstr_power7 attribute_hidden;
> /* Avoid DWARF definition DIE on ifunc symbol so that GDB can handle
> ifunc symbol properly. */
> libc_ifunc_redirected (__redirect_strstr, strstr,
> - (hwcap & PPC_FEATURE_HAS_VSX)
> + (hwcap & PPC_FEATURE_ARCH_2_06)
> ? __strstr_power7
> : __strstr_ppc);
> #endif
>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
2021-07-06 10:51 [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Anton Blanchard
` (2 preceding siblings ...)
2021-07-06 15:05 ` [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Adhemerval Zanella
@ 2021-07-08 22:29 ` Tulio Magno Quites Machado Filho
2021-07-13 19:03 ` Adhemerval Zanella
3 siblings, 1 reply; 11+ messages in thread
From: Tulio Magno Quites Machado Filho @ 2021-07-08 22:29 UTC (permalink / raw)
To: Anton Blanchard; +Cc: libc-alpha
Anton Blanchard via Libc-alpha <libc-alpha@sourceware.org> writes:
> We use PPC_FEATURE_HAS_VSX to select a number of POWER7 optimised
> functions. These functions don't use any VSX instructions, so
> PPC_FEATURE_ARCH_2_06 seems like a better fit.
I don't think we can replace PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
because it would cause issues similar to the ones fixed in patch 3/3, i.e.
this function would run on a processor that implements the Power ISA 2.06
but does not implement Altivec.
So, if testing for PPC_FEATURE_ARCH_2_06 is important, I think it has to test
for:
(hwcap & PPC_FEATURE_ARCH_2_06 && hwcap & PPC_FEATURE_HAS_ALTIVEC)
However, VSX was introduced by Power ISA 2.06. So, PPC_FEATURE_HAS_VSX
implies PPC_FEATURE_ARCH_2_06. PPC_FEATURE_HAS_VSX also implies
PPC_FEATURE_HAS_ALTIVEC.
The change would only make a difference if a processor implements both Power
ISA 2.06 and Altivec, but does not implement VSX.
Am I missing anything?
Another important point is to keep the tests in the IFUNC resolvers in sync
with sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
Otherwise, glibc tests and benchtests may not execute correctly.
--
Tulio Magno
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines
2021-07-06 10:51 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
@ 2021-07-08 22:33 ` Tulio Magno Quites Machado Filho
0 siblings, 0 replies; 11+ messages in thread
From: Tulio Magno Quites Machado Filho @ 2021-07-08 22:33 UTC (permalink / raw)
To: Anton Blanchard; +Cc: libc-alpha
Anton Blanchard via Libc-alpha <libc-alpha@sourceware.org> writes:
> A number of optimised memset routines assume the cacheline size is 128B,
> so we better check before using them.
The code looks good to me, but it's missing the same tests for bzero().
It's also important to keep sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
in sync with both functions.
Like what this patch does:
https://github.com/tuliom/glibc/commit/5b3f36c16d97db739ea5fd422d2230b4dddd2512
--
Tulio Magno
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
2021-07-08 22:29 ` Tulio Magno Quites Machado Filho
@ 2021-07-13 19:03 ` Adhemerval Zanella
2021-07-16 13:28 ` Tulio Magno Quites Machado Filho
0 siblings, 1 reply; 11+ messages in thread
From: Adhemerval Zanella @ 2021-07-13 19:03 UTC (permalink / raw)
To: Tulio Magno Quites Machado Filho, Anton Blanchard; +Cc: libc-alpha
On 08/07/2021 19:29, Tulio Magno Quites Machado Filho via Libc-alpha wrote:
> Anton Blanchard via Libc-alpha <libc-alpha@sourceware.org> writes:
>
>> We use PPC_FEATURE_HAS_VSX to select a number of POWER7 optimised
>> functions. These functions don't use any VSX instructions, so
>> PPC_FEATURE_ARCH_2_06 seems like a better fit.
>
> I don't think we can replace PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
> because it would cause issues similar to the ones fixed in patch 3/3, i.e.
> this function would run on a processor that implements the Power ISA 2.06
> but does not implement Altivec.
>
> So, if testing for PPC_FEATURE_ARCH_2_06 is important, I think it has to test
> for:
>
> (hwcap & PPC_FEATURE_ARCH_2_06 && hwcap & PPC_FEATURE_HAS_ALTIVEC)
>
> However, VSX was introduced by Power ISA 2.06. So, PPC_FEATURE_HAS_VSX
> implies PPC_FEATURE_ARCH_2_06. PPC_FEATURE_HAS_VSX also implies
> PPC_FEATURE_HAS_ALTIVEC.
> The change would only make a difference if a processor implements both Power
> ISA 2.06 and Altivec, but does not implement VSX.
>
> Am I missing anything?
My understanding is the selection change done for this patch is to enable
routines that do *not* use vectorized instructions (either VSX or altivec)
on a CPU that does have isa 2.06 (to use cmpb for instance). That's why
I suggested that we should in fact rename them and also maybe move to
a different folder than 'power7' to something more meaningful (for instance
__memchr_isa206 or something).
The selection change above you suggested is not strictly required, it
would be for memcpy/memmove (if I recall correctly they only use altivec).
>
> Another important point is to keep the tests in the IFUNC resolvers in sync
> with sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
> Otherwise, glibc tests and benchtests may not execute correctly.
>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
2021-07-13 19:03 ` Adhemerval Zanella
@ 2021-07-16 13:28 ` Tulio Magno Quites Machado Filho
2021-07-26 23:23 ` Anton Blanchard
0 siblings, 1 reply; 11+ messages in thread
From: Tulio Magno Quites Machado Filho @ 2021-07-16 13:28 UTC (permalink / raw)
To: Adhemerval Zanella, Anton Blanchard; +Cc: libc-alpha
Adhemerval Zanella via Libc-alpha <libc-alpha@sourceware.org> writes:
> On 08/07/2021 19:29, Tulio Magno Quites Machado Filho via Libc-alpha wrote:
>> I don't think we can replace PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
>> because it would cause issues similar to the ones fixed in patch 3/3, i.e.
>> this function would run on a processor that implements the Power ISA 2.06
>> but does not implement Altivec.
>>
>> So, if testing for PPC_FEATURE_ARCH_2_06 is important, I think it has to test
>> for:
>>
>> (hwcap & PPC_FEATURE_ARCH_2_06 && hwcap & PPC_FEATURE_HAS_ALTIVEC)
>>
>> However, VSX was introduced by Power ISA 2.06. So, PPC_FEATURE_HAS_VSX
>> implies PPC_FEATURE_ARCH_2_06. PPC_FEATURE_HAS_VSX also implies
>> PPC_FEATURE_HAS_ALTIVEC.
>> The change would only make a difference if a processor implements both Power
>> ISA 2.06 and Altivec, but does not implement VSX.
>>
>> Am I missing anything?
>
> My understanding is the selection change done for this patch is to enable
> routines that do *not* use vectorized instructions (either VSX or altivec)
> on a CPU that does have isa 2.06 (to use cmpb for instance).
Yes. My previous comment would only make sense for functions using Altivec,
but these functions are not. I'm sorry.
> That's why
> I suggested that we should in fact rename them and also maybe move to
> a different folder than 'power7' to something more meaningful (for instance
> __memchr_isa206 or something).
I don't have a strong opinion on this. LGTM either way.
>> Another important point is to keep the tests in the IFUNC resolvers in sync
>> with sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
>> Otherwise, glibc tests and benchtests may not execute correctly.
Heads up this paragraph is still valid.
--
Tulio Magno
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06
2021-07-16 13:28 ` Tulio Magno Quites Machado Filho
@ 2021-07-26 23:23 ` Anton Blanchard
0 siblings, 0 replies; 11+ messages in thread
From: Anton Blanchard @ 2021-07-26 23:23 UTC (permalink / raw)
To: Tulio Magno Quites Machado Filho, Adhemerval Zanella; +Cc: libc-alpha
Hi,
> >> Another important point is to keep the tests in the IFUNC
> >> resolvers in sync with
> >> sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c Otherwise,
> >> glibc tests and benchtests may not execute correctly.
>
> Heads up this paragraph is still valid.
Oops, thanks Adhemerval and Tulio for pointing this out. Will fix and
resubmit.
Thanks,
Anton
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines
2021-07-27 5:47 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
@ 2021-08-06 19:50 ` Tulio Magno Quites Machado Filho
0 siblings, 0 replies; 11+ messages in thread
From: Tulio Magno Quites Machado Filho @ 2021-08-06 19:50 UTC (permalink / raw)
To: Anton Blanchard; +Cc: libc-alpha, adhemerval.zanella
Anton Blanchard via Libc-alpha <libc-alpha@sourceware.org> writes:
> A number of optimised memset routines assume the cacheline size is 128B,
> so we better check before using them.
LGTM.
The patch is still missing the changes for bzero(), but I understand this is
already an improvement over what we had.
I plan to modify bzero later.
Pushed as f2a15dd668913c5a1388ba7e1131b25162b2ea75.
Thanks!
--
Tulio Magno
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines
2021-07-27 5:47 Anton Blanchard
@ 2021-07-27 5:47 ` Anton Blanchard
2021-08-06 19:50 ` Tulio Magno Quites Machado Filho
0 siblings, 1 reply; 11+ messages in thread
From: Anton Blanchard @ 2021-07-27 5:47 UTC (permalink / raw)
To: tuliom, adhemerval.zanella; +Cc: libc-alpha
A number of optimised memset routines assume the cacheline size is 128B,
so we better check before using them.
---
.../powerpc64/multiarch/ifunc-impl-list.c | 18 +++++++++++++-----
sysdeps/powerpc/powerpc64/multiarch/memset.c | 15 ++++++++++-----
2 files changed, 23 insertions(+), 10 deletions(-)
diff --git a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
index 32564c8f1f..a3fdcd43bd 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/ifunc-impl-list.c
@@ -35,6 +35,9 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
unsigned long int hwcap = GLRO(dl_hwcap);
unsigned long int hwcap2 = GLRO(dl_hwcap2);
+#ifdef SHARED
+ int cacheline_size = GLRO(dl_cache_line_size);
+#endif
/* hwcap contains only the latest supported ISA, the code checks which is
and fills the previous supported ones. */
@@ -90,16 +93,21 @@ __libc_ifunc_impl_list (const char *name, struct libc_ifunc_impl *array,
IFUNC_IMPL_ADD (array, i, memset,
hwcap2 & PPC_FEATURE2_ARCH_3_1
&& hwcap2 & PPC_FEATURE2_HAS_ISEL
- && hwcap & PPC_FEATURE_HAS_VSX,
+ && hwcap & PPC_FEATURE_HAS_VSX
+ && cacheline_size == 128,
__memset_power10)
#endif
- IFUNC_IMPL_ADD (array, i, memset, hwcap2 & PPC_FEATURE2_ARCH_2_07,
+ IFUNC_IMPL_ADD (array, i, memset, hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && cacheline_size == 128,
__memset_power8)
- IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_ARCH_2_06,
+ IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_ARCH_2_06
+ && cacheline_size == 128,
__memset_power7)
- IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_ARCH_2_05,
+ IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_ARCH_2_05
+ && cacheline_size == 128,
__memset_power6)
- IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_POWER4,
+ IFUNC_IMPL_ADD (array, i, memset, hwcap & PPC_FEATURE_POWER4
+ && cacheline_size == 128,
__memset_power4)
IFUNC_IMPL_ADD (array, i, memset, 1, __memset_ppc))
diff --git a/sysdeps/powerpc/powerpc64/multiarch/memset.c b/sysdeps/powerpc/powerpc64/multiarch/memset.c
index c1aa143f60..056e911699 100644
--- a/sysdeps/powerpc/powerpc64/multiarch/memset.c
+++ b/sysdeps/powerpc/powerpc64/multiarch/memset.c
@@ -43,16 +43,21 @@ libc_ifunc (__libc_memset,
# ifdef __LITTLE_ENDIAN__
(hwcap2 & PPC_FEATURE2_ARCH_3_1
&& hwcap2 & PPC_FEATURE2_HAS_ISEL
- && hwcap & PPC_FEATURE_HAS_VSX)
+ && hwcap & PPC_FEATURE_HAS_VSX
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power10 :
# endif
- (hwcap2 & PPC_FEATURE2_ARCH_2_07)
+ (hwcap2 & PPC_FEATURE2_ARCH_2_07
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power8 :
- (hwcap & PPC_FEATURE_ARCH_2_06)
+ (hwcap & PPC_FEATURE_ARCH_2_06
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power7 :
- (hwcap & PPC_FEATURE_ARCH_2_05)
+ (hwcap & PPC_FEATURE_ARCH_2_05
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power6 :
- (hwcap & PPC_FEATURE_POWER4)
+ (hwcap & PPC_FEATURE_POWER4
+ && GLRO(dl_cache_line_size) == 128)
? __memset_power4
: __memset_ppc);
--
2.31.1
^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2021-08-06 19:50 UTC | newest]
Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-07-06 10:51 [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Anton Blanchard
2021-07-06 10:51 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
2021-07-08 22:33 ` Tulio Magno Quites Machado Filho
2021-07-06 10:51 ` [PATCH 3/3] powerpc64: Add checks for Altivec and VSX in ifunc selection Anton Blanchard
2021-07-06 15:05 ` [PATCH 1/3] powerpc64: Replace some PPC_FEATURE_HAS_VSX with PPC_FEATURE_ARCH_2_06 Adhemerval Zanella
2021-07-08 22:29 ` Tulio Magno Quites Machado Filho
2021-07-13 19:03 ` Adhemerval Zanella
2021-07-16 13:28 ` Tulio Magno Quites Machado Filho
2021-07-26 23:23 ` Anton Blanchard
2021-07-27 5:47 Anton Blanchard
2021-07-27 5:47 ` [PATCH 2/3] powerpc64: Check cacheline size before using optimised memset routines Anton Blanchard
2021-08-06 19:50 ` Tulio Magno Quites Machado Filho
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).