[GIT]: libpfm4 branch, master, updated. 6a70f2060dd48ae61ed6cc5d613befa32416342b
Status: Beta
Brought to you by:
seranian
From: seranian <ser...@us...> - 2010-09-02 23:13:46
|
This is an automated email from the git hooks/post-receive script. It was generated because a ref change was pushed to the repository containing the project "libpfm4". The branch, master has been updated via 6a70f2060dd48ae61ed6cc5d613befa32416342b (commit) from e1de13bb63f2c1fda62b68a6197d477de9bf9a40 (commit) Those revisions listed above that are new to this repository have not appeared on any other notification email; so we list those revisions in full, below. - Log ----------------------------------------------------------------- commit 6a70f2060dd48ae61ed6cc5d613befa32416342b Author: Stephane Eranian <er...@gm...> Date: Fri Sep 3 01:09:02 2010 +0200 fix AMD64 invalid encoding of events with umask revision constraints The encoding was not picking up the right umask, e.g., for AMD K8 PROBE:ALL for instance. There is a ALL encoding up till version C, then another one from D-G. That was coming from a bogus amd64_umask_valid() and also because both pfm_amd64_get_event_info() and pfm_amd64_get_event_attr_info() were relying on attribute indexes being contiguous. They are not because some umask may be skipped. The common code was also at fault because it was not using the ainfo.idx to encode the index of the attribute. The pfmlib_sanitize_event() routine must look at the attribute type to find duplicates. Two attributes are duplicates if they have the same attr index AND the same type. On AMD64, due to the gaps in umasks, a umask and a modifier may have the same index. Signed-off-by: Stephane Eranian <er...@gm...> ----------------------------------------------------------------------- Summary of changes: lib/pfmlib_amd64.c | 89 +++++++++++++++++++++++++++++++++++--------------- lib/pfmlib_common.c | 7 ++-- 2 files changed, 66 insertions(+), 30 deletions(-) hooks/post-receive -- libpfm4 |