projects / mesa.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: c8949e2) | patch
ac/nir: only canonicalize 32-bit float min/max outputs on pre-GFX9
authorSamuel Pitoiset <samuel.pitoiset@gmail.com>
Wed, 24 Jan 2018 11:31:40 +0000 (12:31 +0100)
committerSamuel Pitoiset <samuel.pitoiset@gmail.com>
Fri, 26 Jan 2018 11:14:27 +0000 (12:14 +0100)
commitb9e2f78d6e0cd7d562ffd9030768bd96ecc8ecd3
treef3dfebac5c36752f6516d0bfc3c249e0f8c43683tree
parentc8949e24984266cca3593291c30ea199baef5358commit | diff
ac/nir: only canonicalize 32-bit float min/max outputs on pre-GFX9

According to LLVM, only pre-GFX9 targets do not flush denorms
for fmin/fmax.

All dEQP-VK.glsl.builtin.precision.* still pass.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
src/amd/common/ac_nir_to_llvm.c diff | blob | history