This is the mail archive of the
libc-alpha@sources.redhat.com
mailing list for the glibc project.
[PATCH] PPC64 linuxthreads perf fix
- From: Steven Munroe <sjmunroe at us dot ibm dot com>
- To: libc-alpha <libc-alpha at sources dot redhat dot com>
- Date: Thu, 11 Sep 2003 11:52:53 -0500
- Subject: [PATCH] PPC64 linuxthreads perf fix
- Organization: IBM Linux Developement
- Reply-to: sjmunroe at vnet dot ibm dot com
Due to an oversite the linuxthreads code for PPC64 is still using sync as only
form of MEMORY_BARRIER. This is really bad on the new large (Power4)
processors. Should use lwsync or eieio as appropriate.
2003-09-11 Steven Munroe <sjmunroe@us.ibm.com>
* sysdeps/powerpc/powerpc64/pt-machine.h [MEMORY_BARRIER]: Use lwsync.
[READ_MEMORY_BARRIER]: Define.
[WRITE_MEMORY_BARRIER]: Define.
--
Steven Munroe
sjmunroe@us.ibm.com
Linux on PowerPC-64 Development
GLIBC for PowerPC-64 Development
diff -urN libc23-cvstip-20030911/linuxthreads/sysdeps/powerpc/powerpc64/pt-machine.h libc23/linuxthreads/sysdeps/powerpc/powerpc64/pt-machine.h
--- libc23-cvstip-20030911/linuxthreads/sysdeps/powerpc/powerpc64/pt-machine.h 2003-07-31 14:15:43.000000000 -0500
+++ libc23/linuxthreads/sysdeps/powerpc/powerpc64/pt-machine.h 2003-09-11 12:27:34.000000000 -0500
@@ -34,7 +34,9 @@
/* For multiprocessor systems, we want to ensure all memory accesses
are completed before we reset a lock. On other systems, we still
need to make sure that the compiler has flushed everything to memory. */
-#define MEMORY_BARRIER() __asm__ __volatile__ ("sync" : : : "memory")
+#define MEMORY_BARRIER() __asm__ __volatile__ ("lwsync" : : : "memory")
+#define READ_MEMORY_BARRIER() __asm__ __volatile__ ("lwsync" : : : "memory")
+#define WRITE_MEMORY_BARRIER() __asm__ __volatile__ ("eieio" : : : "memory")
/* We want the OS to assign stack addresses. */
#define FLOATING_STACKS 1