https://gcc.gnu.org/bugzilla/show_bug.cgi?id=71277
Bug ID: 71277
Summary: SSE4.1 rounding generation should not depend on
flag_trapping_math
Product: gcc
Version: 6.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: jsm28 at gcc dot gnu.org
Target Milestone: ---
Target: i?86-*-* x86_64-*-*
For SSE4.1 code generation, i386.md supports inlining calls to rint / ceil /
floor / trunc and the float versions thereof with the roundsd / roundss
instructions. But this is conditioned on !flag_trapping_math. It should not
be so conditioned:
* For rint, the instruction raises exactly the desired exceptions.
* For ceil / floor / trunc, "inexact" should properly not be raised as per TS
18661-1. The instruction used right now does raise "inexact", but you just
need to OR 8 (ROUND_NO_EXC) into the immediate operand to avoid raising
"inexact".
(For avoidance of doubt, this bug report is only about those built-in
functions, not about the more complicated code generation for the "round"
function, where it's likely "inexact" would be harder to avoid.)
