https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79268
Bug ID: 79268
Summary: [6/7 Regression] Wrong code generation for vec_xl and
vec_xst intrinsics
Product: gcc
Version: 6.3.1
Status: UNCONFIRMED
Keywords: wrong-code
Severity: major
Priority: P3
Component: target
Assignee: wschmidt at gcc dot gnu.org
Reporter: wschmidt at gcc dot gnu.org
CC: segher at gcc dot gnu.org
Target Milestone: ---
Target: powerpc64*-*-*
I had a thinko when adding the new element-reversing vector load and store
instructions, tying them to vec_xl and vec_xst rather than vec_xl_be and
vec_xst_be. This regresses these intrinsics both by generating wrong code for
them, but also by rejecting legitimate parameter combinations that vec_xl/xst
should support (such as vector pixel). This has been encountered building at
least one debian package.
I plan to attack this in two stages. First we'll need an emergency patch that
just reverts the #define of these to match vec_vsx_ld and vec_vsx_st as they
should, which we can submit to the various distros for immediate use. Later
I'll fix up the rest so that vec_xl_be and vec_xst_be make use of the new
instructions.
