[GIT]: libpfm4 branch, master, updated. 284c0820e4717aa2b7a34a9073a4d5a132f1536a
Status: Beta
Brought to you by:
seranian
From: seranian <ser...@us...> - 2011-06-02 10:55:20
|
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 284c0820e4717aa2b7a34a9073a4d5a132f1536a (commit) via 7e8541d287c8e1fad11f249406f5fe686339dfe2 (commit) from c02482a8bdc1e14f81dc4094c8abd3aa2de0a430 (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 284c0820e4717aa2b7a34a9073a4d5a132f1536a Author: Stephane Eranian <er...@gm...> Date: Thu Jun 2 12:49:25 2011 +0200 reduce size of AMD64 event tables Use the same technic as for Intel X86, go from: struct entry_t { int numasks; umask_t umasks[MAX]; }; to: umask_t foo[]= { }; struct entry_t { int numasks; umask_t *umasks = foo; }; Size reduction is also important, from: text data bss dec hex filename 15128 192 0 15320 3bd8 pfmlib_amd64_k7.o 59109 1152 0 60261 eb65 pfmlib_amd64_k8.o 87197 576 0 87773 156dd pfmlib_amd64_fam10h.o 56737 192 0 56929 de61 pfmlib_amd64_fam14h.o to: text data bss dec hex filename 3412 192 0 3604 e14 pfmlib_amd64_k7.o (ex lib/libpfm.a) 19961 1152 0 21113 5279 pfmlib_amd64_k8.o (ex lib/libpfm.a) 33665 576 0 34241 85c1 pfmlib_amd64_fam10h.o (ex lib/libpfm.a) 21909 192 0 22101 5655 pfmlib_amd64_fam14h.o (ex lib/libpfm.a) Overall: 37% size reduction. This is less than on Intel X86, but the umasks array was only 16-entry long and most events had 4 or more umasks. Signed-off-by: Stephane Eranian <er...@gm...> commit 7e8541d287c8e1fad11f249406f5fe686339dfe2 Author: Stephane Eranian <er...@gm...> Date: Thu Jun 2 12:47:11 2011 +0200 pfm_get_event_attr_info() returns ucode >> 8 pfm_get_event_attr_info() returns the attribute (umask) code and not the raw code which is now shifted by 8, i.e., final position in register. This make more sense and provides backward compatibility. offcore_response umasks are treated differently, the raw umask code is never shifted << 8. Signed-off-by: Stephane Eranian <er...@gm...> ----------------------------------------------------------------------- Summary of changes: lib/events/amd64_events_fam10h.h | 5233 +++++++++++++++++--------------------- lib/events/amd64_events_fam14h.h | 3044 +++++++++++----------- lib/events/amd64_events_k7.h | 405 ++-- lib/events/amd64_events_k8.h | 2520 +++++++++--------- lib/pfmlib_amd64.c | 17 +- lib/pfmlib_amd64_priv.h | 3 +- lib/pfmlib_intel_x86.c | 2 +- 7 files changed, 5336 insertions(+), 5888 deletions(-) hooks/post-receive -- libpfm4 |