https://bugs.llvm.org/show_bug.cgi?id=35964
Bug ID: 35964
Summary: Profile Instruction Generate/Coverage mapping: High
overhead at runtime with many threads
Product: new-bugs
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: new bugs
Assignee: unassignedb...@nondot.org
Reporter: dennis.fels...@sap.com
CC: llvm-bugs@lists.llvm.org
Created attachment 19681
--> https://bugs.llvm.org/attachment.cgi?id=19681&action=edit
Source code
Creating coverage information with -fprofile-instr-generate -fcoverage-mapping
usually doesn't create a high overhead. When using dozens of threads the
overhead seems to grow significantly in the end running 24 times as slow with
40 threads. See the following table created on a 2x Intel E5-2660v3 system:
Threads No Coverage [s] Coverage [s] Factor
1 0.011 0.011 1.0
2 0.071 0.257 3.6
3 0.216 1.419 6.57
4 0.377 4.735 12.56
5 0.678 8.097 11.94
10 3.250 37.037 11.40
20 17.223 392.021 22.76
40 84.129 2087.047 24.81
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
