projects / platform / upstream / mesa.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 4e7b71e) | patch
intel/compiler: Don't use SIMD larger than needed for workgroup
authorCaio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Fri, 8 Oct 2021 03:18:39 +0000 (20:18 -0700)
committerMarge Bot <eric+marge@anholt.net>
Tue, 26 Oct 2021 17:49:09 +0000 (17:49 +0000)
commit3072e6e0da5d226700a8d6d0683636b5c7e2f237
tree73087c348b8cd2608e97c49574780393f2faf952tree | snapshot
parent4e7b71e00cb9ae27dff4f022fab58e14ea75fa2dcommit | diff
intel/compiler: Don't use SIMD larger than needed for workgroup

Unless we are combining multiple workgroups in the same HW thread,
there's no advantage of using SIMD16 when SIMD8 already fits the
entire workgroup.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13249>
src/intel/compiler/brw_simd_selection.c diff | blob | history
src/intel/compiler/test_simd_selection.cpp diff | blob | history
Domain: Graphics System / GL;