[flang-commits] [PATCH] D129296: Lower mask intrinsics

Fri Jul 8 06:53:12 PDT 2022

tarunprabhu added a comment.

In D129296#3638440 <https://reviews.llvm.org/D129296#3638440>, @jeanPerier wrote:

> Where do you get the "If I < 0 or I > BIT_SIZE(KIND), the result will always be 0." requirement ? My understanding from the standard it is non conforming if  I < 0 or I > BIT_SIZE(KIND), and so there is no specified behavior.

The same point was raised by @vdonaldson  on another patch (https://reviews.llvm.org/D129316) which makes the same assumption. If the experts can weigh in on that patch, I'll apply similar changes to this one.

================
Comment at: flang/lib/Lower/IntrinsicCall.cpp:3193
+      loc, resultType, resultType.getIntOrFloatBitWidth());
+  if (args[1]) {
+    mlir::Value eight = builder.createIntegerConstant(loc, resultType, 8);
----------------
jeanPerier wrote:
> I am surprised you have to do that here. the KIND argument, if present, must already be represented in the resultType that semantics selected for the intrinsic. So I think you should be able to remove this case (otherwise, this would indicate some bug with resultType).
Duh! I didn't check to see if the resultType was already correct.  If it is as expected, I'll remove the check. 

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D129296/new/

https://reviews.llvm.org/D129296