https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85459
Bug ID: 85459
Summary: [8 Regression] Larger code generated from GMP template
meta-programming
Product: gcc
Version: 8.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: glisse at gcc dot gnu.org
Target Milestone: ---
Created attachment 43982
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=43982&action=edit
preprocessed testcase
The attached testcase is a preprocessed version of
https://gmplib.org/list-archives/gmp-bugs/2014-January/003331.html, which I was
about to commit to GMP.
Compiling for x86_64 with g++-7 -O3, I get a nice .s file of size 4174. With
g++-8 -O3, I get a much larger size of 7077. Looking more closely at the
difference, I see a ton of moves. Besides, g++-8 finds fewer
__builtin_constant_p opportunities than g++-7 (__gmpq_mul instead of
__gmpq_mul_2exp for instance).
The dump I glanced at makes me suspect I?SRA, but I am likely completely off.
