akirilov-arm opened PR #1854 from simd_load_splat
to master
:
One problem with this implementation is that a SIMD load and splat operation is represented as two IR instructions -
load
andsplat
. As a result, two A64 instructions are generated - a scalar load, followed byDUP
, while the operation can be expressed with justLD1R
. Is there a relatively easy way to fix this?
cfallin submitted PR Review.
cfallin merged PR #1854.
Last updated: Dec 23 2024 at 12:05 UTC