[PATCH] D25999: LegalizeDAG: Implement expansion of f16 = FP_TO_FP16 f64

[Jan Vesely via llvm-commits]

jvesely added inline comments.


================
Comment at: llvm/trunk/test/CodeGen/AMDGPU/fptrunc.ll:4
+; RUN: llc -march=amdgcn -mcpu=tonga -enable-unsafe-fp-math -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GCN-UNSAFE %s
 
 ; FUNC-LABEL: {{^}}fptrunc_f64_to_f32:
----------------
GCN-UNSAFE is never checked. did you mean GCN-FAST (or use GCN-UNSAFE in the tests)?


================
Comment at: llvm/trunk/test/CodeGen/AMDGPU/fptrunc.ll:17
+; GCN-FAST: v_cvt_f16_f32_e32 v[0-9]+, [[F32]]
+define void @fptrunc_f64_to_f16(i16 addrspace(1)* %out, double %in) {
+  %result = fptrunc double %in to half
----------------
Is there a reason to use i16 instead of half? can the half test be moved to this file?


Repository:
  rL LLVM

https://reviews.llvm.org/D25999