[PATCH] D44785: Lowering x86 adds/addus/subs/subus intrinsics (llvm part)

[Roland Scheidegger via Phabricator via llvm-commits]

sroland added a comment.

In https://reviews.llvm.org/D44785#1077286, @craig.topper wrote:

> Is the IR being generated on the fly and then fed to the JIT? Which means
>  it doesn't go through the autoupgrade code since that is only done by the
>  bitcode reader and the ll parser? If that's the case, you'll need to
>  generate the replacement sequence directly instead of using the intrinsic.


FWIW this is correct, it's using jit so no autoupgrade, so we'll have to fix this in mesa.
However, in the past when intrinsics disappeared, we actually got an error when compiling the IR. See https://bugs.llvm.org/show_bug.cgi?id=28176 for instance: LLVM ERROR: Program used external function '_llvm.x86.sse2.pminu.b' which could not be resolved!
That of course made it a lot more obvious what's going on - no error and just calling a 0 function in the generated code is really not helpful.


Repository:
  rL LLVM

https://reviews.llvm.org/D44785