gallivm: don't use integer min/max sse intrinsics with llvm >= 3.9
authorRoland Scheidegger <sroland@vmware.com>
Sun, 19 Jun 2016 01:56:11 +0000 (03:56 +0200)
committerRoland Scheidegger <sroland@vmware.com>
Mon, 20 Jun 2016 15:19:03 +0000 (17:19 +0200)
commitb0cf99165af445adc4c5c1f66a3a3e0d882211cd
treea3088217e36fb7119c6eafb9be37ffaa25a1473e
parent154c0a42a23187c61ea0a1307198fae667398eba
gallivm: don't use integer min/max sse intrinsics with llvm >= 3.9

Apparently, these are deprecated. There's some AutoUpgrade feature which
is supposed to promote these to cmp/select, which apparently doesn't work
with jit code. It is possible it's not actually even meant to work (see
the bug filed against llvm which couldn't provide an answer neither)
but in any case this is meant to be only temporary unless the intrinsics
are really illegal. So, just use the fallback code (which should be cmp/select,
we're actually doing cmp/sext/trunc/select, but in any case llvm 3.9 manages
to optimize this back to pmin/pmax in the end).

This addresses https://llvm.org/bugs/show_bug.cgi?id=28176

CC: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Tested-by: Vinson Lee <vlee@freedesktop.org>
Tested-by: Aaron Watry <awatry@gmail.com>
src/gallium/auxiliary/gallivm/lp_bld_arit.c