Booleans are often more efficient with register usage.
This also allows to move comparisons further.
Totals from affected shaders: (VEGA)
SGPRS: 451608 -> 450320 (-0.29 %)
VGPRS: 351448 -> 351256 (-0.05 %)
Spilled SGPRs: 105 -> 105 (0.00 %)
Spilled VGPRs: 0 -> 0 (0.00 %)
Private memory VGPRs: 0 -> 0 (0.00 %)
Scratch size: 1008 -> 1008 (0.00 %) dwords per thread
Code Size:
26555596 ->
26551080 (-0.02 %) bytes
LDS: 10323 -> 10323 (0.00 %) blocks
Max Waves: 42850 -> 42934 (0.20 %)
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4830>
return options & nir_move_const_undef;
}
case nir_instr_type_alu: {
- if (nir_op_is_vec(nir_instr_as_alu(instr)->op))
+ if (nir_op_is_vec(nir_instr_as_alu(instr)->op) ||
+ nir_instr_as_alu(instr)->op == nir_op_b2i32)
return options & nir_move_copies;
if (nir_alu_instr_is_comparison(nir_instr_as_alu(instr)))
return options & nir_move_comparisons;