[clang] [llvm] [win][x64] Multiple fixes for import call optimization (PR #160604)

[Eli Friedman via cfe-commits]

================
@@ -439,7 +439,7 @@ let isPseudo = 1, isCall = 1, isCodeGenOnly = 1,
             Requires<[In64BitMode]>;
 
   def CALL64r_ImpCall :
-    PseudoI<(outs), (ins GR64_A:$dst), [(X86call GR64_A:$dst)]>,
+    PseudoI<(outs), (ins GR64:$dst), [(X86call GR64:$dst)]>,
----------------
efriedma-quic wrote:

I don't see how the register class here is connected to the isel scheduling cycle mentioned in the comment in X86ISelLoweringCall.cpp.  I can see how the CopyToReg itself helps, but I don't think the scheduler cares about register classes.

https://github.com/llvm/llvm-project/pull/160604