[llvm-bugs] [Bug 34101] New: deriche: performance difference with -polly-position=before-vectorizer
via llvm-bugs
llvm-bugs at lists.llvm.org
Mon Aug 7 05:21:27 PDT 2017
https://bugs.llvm.org/show_bug.cgi?id=34101
Bug ID: 34101
Summary: deriche: performance difference with
-polly-position=before-vectorizer
Product: Polly
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: enhancement
Priority: P
Component: Optimizer
Assignee: polly-dev at googlegroups.com
Reporter: simbuerg at fim.uni-passau.de
CC: llvm-bugs at lists.llvm.org
Created attachment 18921
--> https://bugs.llvm.org/attachment.cgi?id=18921&action=edit
Preprocessed sources with reduced deriche kernel
Hi, using the (somewhat reduced) version of polybench's deriche kernel I get
the following two (weird) results when I compare Polly before the vectorizer to
its default position:
clang -mllvm -polly -mllvm -polly-position=before-vectorizer -O3 -I utilities
-I deriche -DPOLYBENCH_USE_C99_PROTO -DEXTRALARGE_DATASET
-DPOLYBENCH_USE_RESTRICT utilities/polybench.c deriche.dir/deriche.c -lm -o
./deriche.vect.bin
clang -mllvm -polly -O3 -I utilities -I deriche -DPOLYBENCH_USE_C99_PROTO
-DEXTRALARGE_DATASET -DPOLYBENCH_USE_RESTRICT utilities/polybench.c
deriche.dir/deriche.c -lm -o ./deriche.default.bin
/local/simbuerg/benchbuild/results/pj-test-deriche-33ef47b4-c3d5-4b39-8dc3-478aac1cbbcb
master* ⇣ simbuerg at ligeti 6s
❯ time ./deriche.vect.bin
./deriche.vect.bin 23.34s user 0.02s system 99% cpu 23.357 total
/local/simbuerg/benchbuild/results/pj-test-deriche-33ef47b4-c3d5-4b39-8dc3-478aac1cbbcb
master* ⇣ simbuerg at ligeti 23s
❯ time ./deriche.default.bin ./deriche.default.bin 0.13s user 0.00s
system 96% cpu 0.141 total
Expected result: The produced code should not get slower that without polly
:-).
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20170807/3d896f12/attachment.html>
More information about the llvm-bugs
mailing list