[PATCH] D42615: [X86] Generate BT instrutions a bit more agressively

[Amaury SECHET via Phabricator via llvm-commits]

deadalnix added a comment.

@craig.topper I had no idea `bt` has lower throughput than `test`, I assumed it was the same. If that's the case, then this approach doesn't make much sense.

We should try to get rid of the high register trick that creates extra copies. I'm not sure what is creating this, any suggestions?



================
Comment at: test/CodeGen/X86/testb-je-fusion.ll:11
 ; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    testb $2, %ch
-; CHECK-NEXT:    je .LBB0_2
----------------
craig.topper wrote:
> I will say this is kinda stupid that we forced a copy just so we could do a high register trick.
Do you have nay idea what can be causing this? There are numerous test cases where this happens.


Repository:
  rL LLVM

https://reviews.llvm.org/D42615