tree 05b0c632eee0bdbbb93e0b3cc845531f1ceb97a7
parent b662b5a9fd818c3abaa25ebd172781b869806391
author Diego Caballero <diegocaballero@google.com> 1687198691 -0700
committer GitHub <noreply@github.com> 1687198691 -0700
gpgsig -----BEGIN PGP SIGNATURE-----
 
 wsBcBAABCAAQBQJkkJvjCRBK7hj4Ov3rIwAA+nYIAJwMM99HMNXsJzOr6W2pt1w6
 oEiWOlU0WyRPYzlk7mLKOKP1VjMzSSuXNGPbHdJHou2+OZBwQ9f+d2RSyJNbAtJz
 KPmXQ87BKy1aaVNuNVSeOTFS8/H+Nvko7rzhBOnlcMV/icM+jCtlexrHhxfJ8RMi
 OoNCtOvT50q51w8HOUbZk3qWp/bIEKh+NPiOYQ0PdRDe3A+tY5NxZxZV/RMlhe4M
 5ag8jY4qNnaBzBxK63oIigQ0pr2/ZX88o2KXmzsM0MW0LWzRomzmcKsEs6N1AYES
 juvemQE46oYPxCJxLVZqu7tL1lV+qaSpvW0Os5frBqkpR2rNDk2n6OiMyws1VBo=
 =1977
 -----END PGP SIGNATURE-----
 

[CPU] Scalarize vector loads used in a scalar fashion (#14150)

This PR makes sure that we scalarize the vector load that are used in a
scalar fashion (i.e., vector.transfer_read + vector.extract) in matmuls.
The LLVM backend is sometimes able to scalarize this vector load but not
always. For cases where it's not scalarized, register pressure increases
as the liveness of this vector load spans across all the unrolled
iterations of the matmul. There are also some shuffles/permutations
generated that don't make too much sense from the performance point of
view.