fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-04-27 04:30:37 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

brw: Avoid regioning restrictions for u2u16/i2i16 narrowing conversions

Browse source 
Cuts 0.83% of instructions on Alchemist in affected fossil-db shaders
(nearly all of which are in parallel-rdp).

Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31833>
This commit is contained in:
Kenneth Graunke 2024-10-23 00:01:52 -07:00 • committed by Marge Bot
parent 86f8b8860e
commit 51c67ad7cf
1 changed files with 10 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
src/intel/compiler/brw_from_nir.cpp
View file

@ -1124,6 +1124,16 @@ brw_from_nir_emit_alu(nir_to_brw_state &ntb, nir_alu_instr *instr,
}
}
/* If narrowing (e.g. 32 -> 16), don't do a D -> W or UD -> UW mov,
* instead just read the source as W/UW with a stride (discarding
* the top bits). This avoids the need for the destination to be
* DWord aligned due to regioning restrictions.
*/
if (brw_type_size_bits(result.type) < brw_type_size_bits(op[0].type)) {
const unsigned bits = brw_type_size_bits(result.type);
op[0] = subscript(op[0], brw_type_with_size(op[0].type, bits), 0);
}
bld.MOV(result, op[0]);
break;
}