forked from pool/glibc
Andreas Schwab
7db7ce4082
- no-long-double.patch: Don't use long double functions if NO_LONG_DOUBLE
- Update to glibc 2.23 release.
* Unicode 8.0.0 Support
* sched_setaffinity, pthread_setaffinity_np no longer attempt to guess
the kernel-internal CPU set size
* The fts.h header can now be used with -D_FILE_OFFSET_BITS=64
* getaddrinfo now detects certain invalid responses on an internal
netlink socket
* A defect in the malloc implementation, present since glibc 2.15 (2012)
or glibc 2.10 via --enable-experimental-malloc (2009), could result in
the unnecessary serialization of memory allocation requests across
threads
* The obsolete header <regexp.h> has been removed
* The obsolete functions bdflush, create_module, get_kernel_syms,
query_module and uselib are no longer available to newly linked
binaries
* Optimized string, wcsmbs and memory functions for IBM z13.
* Newly linked programs that define a variable called signgam will no
longer have it set by the lgamma, lgammaf and lgammal functions
- Removed patches:
* dont-remove-nodelete-flag.patch
* openat64-readd-o-largefile.patch
* mntent-blank-line.patch
* opendir-o-directory-check.patch
* strcoll-remove-strdiff-opt.patch
* ld-pointer-guard.patch
* tls-dtor-list-mangling.patch
* powerpc-lock-elision-race.patch
* prelink-elf-rtype-class.patch
OBS-URL: https://build.opensuse.org/request/show/361238
OBS-URL: https://build.opensuse.org/package/show/Base:System/glibc?expand=0&rev=423
41 lines
875 B
Diff
41 lines
875 B
Diff
Fix for bnc #868622, slow memset for large block sizes.
|
|
|
|
diff --git a/sysdeps/x86_64/memset.S b/sysdeps/x86_64/memset.S
|
|
index db4fb84..9c42018 100644
|
|
--- a/sysdeps/x86_64/memset.S
|
|
+++ b/sysdeps/x86_64/memset.S
|
|
@@ -84,6 +84,9 @@ L(loop_start):
|
|
movdqu %xmm0, -48(%rdi,%rdx)
|
|
movdqu %xmm0, 48(%rdi)
|
|
movdqu %xmm0, -64(%rdi,%rdx)
|
|
+ mov __x86_shared_cache_size(%rip),%r9d # The largest cache size
|
|
+ cmp %r9,%rdx
|
|
+ ja L(nt_move)
|
|
addq %rdi, %rdx
|
|
andq $-64, %rdx
|
|
cmpq %rdx, %rcx
|
|
@@ -99,6 +102,23 @@ L(loop):
|
|
jne L(loop)
|
|
rep
|
|
ret
|
|
+L(nt_move):
|
|
+ addq %rdi, %rdx
|
|
+ andq $-64, %rdx
|
|
+ cmpq %rdx, %rcx
|
|
+ je L(return)
|
|
+ .p2align 4
|
|
+L(nt_loop):
|
|
+ movntdq %xmm0, (%rcx)
|
|
+ movntdq %xmm0, 16(%rcx)
|
|
+ movntdq %xmm0, 32(%rcx)
|
|
+ movntdq %xmm0, 48(%rcx)
|
|
+ addq $64, %rcx
|
|
+ cmpq %rcx, %rdx
|
|
+ jne L(nt_loop)
|
|
+ sfence
|
|
+ rep
|
|
+ ret
|
|
L(less_16_bytes):
|
|
movq %xmm0, %rcx
|
|
testb $24, %dl
|