[PATCH] D46179: [X86] Lowering adds/addus/subs/subus intrinsics to native IR (LLVM part)

[Roland Scheidegger via Phabricator via llvm-commits]

sroland added inline comments.


================
Comment at: include/llvm/IR/IntrinsicsX86.td:367
 let TargetPrefix = "x86" in {  // All intrinsics start with "llvm.x86.".
-  def int_x86_sse2_padds_b : GCCBuiltin<"__builtin_ia32_paddsb128">,
+  def int_x86_sse2_padds_b : // FIXME: remove this intrinsic
               Intrinsic<[llvm_v16i8_ty], [llvm_v16i8_ty,
----------------
FWIW I don't quite agree this is really a FIXME (not without having some appropriate replacement, like a generic llvm intrinsic for saturate arithmetic).
I'd take an intrinsic over tons of simple ir ops (on the vague hope it will get recognized again unless some other optimization messed it up) any day of the week.

But otherwise looks alright to me.


Repository:
  rL LLVM

https://reviews.llvm.org/D46179