diff options
| author | Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> | 2021-06-28 13:24:24 -0700 |
|---|---|---|
| committer | Stanislav Mekhanoshin <Stanislav.Mekhanoshin@amd.com> | 2021-06-30 11:45:38 -0700 |
| commit | 381ded345bdd59edcea9a87f399f67b18be7622a (patch) | |
| tree | 47aa9596c2093f900919796381495570007a1bba /llvm/lib/CodeGen/MachineOperand.cpp | |
| parent | 822b92aae439c4ba2946980c8a27bd2c8a62d90c (diff) | |
| download | llvm-381ded345bdd59edcea9a87f399f67b18be7622a.zip llvm-381ded345bdd59edcea9a87f399f67b18be7622a.tar.gz llvm-381ded345bdd59edcea9a87f399f67b18be7622a.tar.bz2 | |
[AMDGPU] Add S_MOV_B64_IMM_PSEUDO for wide constants
This is to allow 64 bit constant rematerialization. If a constant
is split into two separate moves initializing sub0 and sub1 like
now RA cannot rematerizalize a 64 bit register.
This gives 10-20% uplift in a set of huge apps heavily using double
precession math.
Fixes: SWDEV-292645
Differential Revision: https://reviews.llvm.org/D104874
Diffstat (limited to 'llvm/lib/CodeGen/MachineOperand.cpp')
0 files changed, 0 insertions, 0 deletions