https://bugs.llvm.org/show_bug.cgi?id=43134
Bug ID: 43134
Summary: Loop unroll by N iterations takes O(N^2) time in
domtree updater
Product: libraries
Version: trunk
Hardware: PC
OS: Windows NT
Status: NEW
Severity: enhancement
Priority: P
Component: Loop Optimizer
Assignee: unassignedb...@nondot.org
Reporter: efrie...@quicinc.com
CC: alina.sbir...@gmail.com, brzy...@gmail.com,
kubakuder...@gmail.com, llvm-bugs@lists.llvm.org
Testcase:
unsigned x() {
unsigned z = 3;
unsigned LIMIT = 16000;
_Pragma("unroll(16384)") for (int i = 0; i < LIMIT; ++i)
z *= z;
return z;
}
The compile times look roughly like the following, with "clang -O2 -mllvm
-unroll-threshold=999999":
LIMIT=1000: 0.3s
LIMIT=2000: 1.0s
LIMIT=4000: 3.6s
LIMIT=8000: 14.3s
LIMIT=16000: 57.8s
This is obviously non-linear; I think it's roughly quadratic. -ftime-report
says the time is spent in unrolling. I "profiled" in a debugger, and it looks
like the time is spent updating the domtree.
Without the "-unroll-threshold", LLVM refuses to unroll the loop 16000 times,
but it will unroll 8000 times.
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
