projects / mesa.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: bc2ccdc) | patch
nir: Add a new ALU nir_op_imul24
authorRob Clark <robdclark@chromium.org>
Wed, 25 Sep 2019 17:10:39 +0000 (10:10 -0700)
committerRob Clark <robdclark@chromium.org>
Fri, 18 Oct 2019 22:08:54 +0000 (15:08 -0700)
commit0568761f8e7b9249ec8cffbd7826c24e38bf16d6
tree174362a432ebe93bf918d9461264a3ffbfec80cdtree
parentbc2ccdc45ab6de74d72412244984b4551ece855ccommit | diff
nir: Add a new ALU nir_op_imul24

Some hardware can do 24b multiply in a single instruction, but not 32b.
However in most cases 24b is sufficient for address/offset calculation.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Eduardo Lima Mitev <elima@igalia.com>
src/compiler/nir/nir_opcodes.py diff | blob | history