Re: [PATCH 4/4] x86: fold certain AVX and AVX2 templates

["Jan Beulich" <JBeulich at suse dot com>]

>>> On 15.12.17 at 11:35, <JBeulich@suse.com> wrote:
> Similarly I didn't figure out yet the reason for an anomaly when the
> "unspecified" checks in operand_type_register_match() are missing: In
> that case I've observed errors on vaddsubp{s,d}, but not on e.g.
> vaddp{s,d} with identical operands.

This turns out to have an easy explanation: The consequences
of the missing checks were simply invisible for vaddp{s,d} and
alike, as only error messages resulted in the testsuite, but no
generated code differences could be looked at. Once I've looked
at the produced code, vaddp{s,d} simply had matches on the
AVX512VL templates, while for insns without direct AVX512VL
equivalent another match can't be found, and hence an error
is being reported.

Bottom line - nothing odd anywhere with or without the (complete)
patch.

Jan