This is the mail archive of the libstdc++@gcc.gnu.org mailing list for the libstdc++ project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] libstdc++ atomicity, x86 agnostic version


>This seems really sad and more specific tests should be performed: the 
>current performance testsuite definitely is not tuned to benchmark that.

I don't dispute this. However, if you look at the use of the atomic
functions, I only see the allocator and string uses as being performance
critical.

And, we do have benchmarks to measure performance of these items....

Current:
string_append.cc         	char                     	   1r    1u    0s        0mem   11pf 
string_append.cc         	string                   	   1r    1u    0s        0mem    1pf 
string_append.cc         	char                     	   9r    9u    0s        0mem    0pf 
string_append.cc         	string                   	  11r   11u    0s        0mem    0pf 
string_append.cc         	char                     	  89r   86u    3s        0mem    0pf 
string_append.cc         	string                   	 115r  110u    4s        0mem    0pf 
string_cons_input_iterator.cc	                         	   2r    2u    0s        0mem    2pf 
string_cons_input_iterator.cc	                         	  16r   15u    1s        0mem    0pf 
string_cons_input_iterator.cc	                         	 157r  136u   21s        0mem    0pf 

Current + patch:
string_append.cc         	char                     	   0r    0u    0s        0mem   11pf 
string_append.cc         	string                   	   2r    2u    0s        0mem    1pf 
string_append.cc         	char                     	   8r    7u    1s        0mem    0pf 
string_append.cc         	string                   	  11r   11u    0s        0mem    0pf 
string_append.cc         	char                     	  88r   83u    5s        0mem    0pf 
string_append.cc         	string                   	 114r  110u    4s        0mem    0pf 
string_cons_input_iterator.cc	                         	   2r    2u    0s        0mem    2pf 
string_cons_input_iterator.cc	                         	  15r   13u    2s        0mem    0pf 
string_cons_input_iterator.cc	                         	 156r  137u   19s        0mem    0pf 

Sooo...  although it may conflict with your (and mine!) initial
impression, I don't seeing this as impacting performance. 

However, maybe you have something in particular in mind, that is not
being tested at the moment? If so, can you share it with me, so we can
see what the performance impacts will be?

best,
benjamin


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]