Set RCU_HAVE_FENCE to false on Intel Xeon Phi
A message that is not in the archive (sent off-list or not yet received): [email protected]
----- Original Message -----
Quoted 10 lines
From: "Simon Marchi" <[email protected]>
To: [email protected]
Sent: Friday, December 6, 2013 10:17:55 PM
Subject: [lttng-dev] [PATCH userspace-rcu] SetRCU_HAVE_FENCEto false on Intel Xeon PhiIt currently fails to compile, because the Phi does not have fence instructions.
It will use lock prefixed instructions for all memory barriers, which is overkill for this platform, but at least it will work.
Merged. It's overkill, but at least it will compile. We'll see if we want to improve the memory barriers and make them cheaper as an optimization.
Thanks,
Mathieu
Quoted 34 lines
Signed-off-by Simon Marchi <[email protected]>--- configure.ac | 5 ++++- 1 files changed, 4 insertions(+), 1 deletions(-)configure.ac +0
-1@@ -140,7 +140,10 @@ AS_IF([test "x$ARCHTYPE" = "xx86"],[ #For now, using lock; addl compatibility mode even for i686, because the #Pentium III is seen as a i686, but lacks mfence instruction. #Only using fence for x86_64. - AS_IF([test "x$host_cpu" != "xi386" -a "x$host_cpu" != "xi486" -a"x$host_cpu" != "xi586" -a "x$host_cpu" != "xi686"],[
configure.ac +4
-0+ # + #k1om is the name for the Intel MIC family (Xeon Phi). It is an x86_64 + #variant but lacks fence instructions. + AS_IF([test "x$host_cpu" != "xi386" -a "x$host_cpu" != "xi486" -a"x$host_cpu" != "xi586" -a "x$host_cpu" != "xi686" -a "x$host_vendor" != "xk1om"],[
configure.ac +0
-1AC_MSG_RESULT([yes]) AC_DEFINE([CONFIG_RCU_HAVE_FENCE], [1]) ],[ --1.7.1
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
Signature · 3 lines
Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com