[PATCH] D84166: AMDGPU: Simplify f16 to i64 custom lowering

[Petar Avramovic via Phabricator via llvm-commits]

Petar.Avramovic added inline comments.


================
Comment at: llvm/test/CodeGen/AMDGPU/fptosi.f16.ll:108
 ; GCN: buffer_load_dword
-; GCN: v_cvt_f32_f16_e32
 ; SI: v_cvt_f32_f16_e32
----------------
arsenm wrote:
> Why are there test changes only for the vector cases?
Tests only check for a few instructions and here  there was no v_cvt_f32_f16_e32.
I think that they test if there was a f16 -> f32 followed by long f32 -> i64 expansion but don't check expansion.
Now f16->i32 is selected as f16->f32->i32 and f16->f32 part is still there.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84166/new/

https://reviews.llvm.org/D84166