[PATCH] D113107: Support of expression granularity for _Float16.

[Zahira Ammarguellat via Phabricator via llvm-commits]

zahiraam added inline comments.


================
Comment at: clang/test/CodeGen/X86/Float16-arithmetic.c:207
+// CHECK-NEXT:    [[EXT:%.*]] = fpext half [[TMP0]] to float
+// CHECK-NEXT:    store float [[EXT]], ptr [[RETVAL]], align 2
+// CHECK-NEXT:    [[TMP1:%.*]] = load half, ptr [[RETVAL]], align 2
----------------
zahiraam wrote:
> pengfei wrote:
> > Not sure if we need a fptrunc and store the half value. The following tests have the same problem.
> I think that's what we want?
> // CHECK-LABEL: @RealOp(
> // CHECK-NEXT:  entry:
> // CHECK-NEXT:    [[A_ADDR:%.*]] = alloca half, align 2
> // CHECK-NEXT:    store half [[A:%.*]], ptr [[A_ADDR]], align 2
> // CHECK-NEXT:    [[TMP0:%.*]] = load half, ptr [[A_ADDR]], align 2
> // CHECK-NEXT:    [[EXT:%.*]] = fpext half [[TMP0]] to float
> // CHECK-NEXT:    [[UNPROMOTION:%.*]] = fptrunc float [[EXT]] to half
> // CHECK-NEXT:    ret half [[UNPROMOTION]]
> 
> Do you agree? If this is correct, I will make the change the other operators.
But I feel like we should be returning a float no?  In which case it will be more tricky (need to calculate the Address with the promoted elementype)? @rmjccall?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D113107/new/

https://reviews.llvm.org/D113107