[PATCH] D26620: [X86][FastISel] Fix lowering of overflow result on AVX512 targets

[Zvi Rackover via llvm-commits]

zvi added inline comments.


================
Comment at: lib/Target/X86/X86FastISel.cpp:2774
+           Ty->getTypeAtIndex(1)->getScalarSizeInBits() == 1 &&
+           "Overflow value expected to be an i1");
 
----------------
qcolombet wrote:
> This could be in a separate NFC patch.
Sure


================
Comment at: lib/Target/X86/X86FastISel.cpp:2884
 
-    unsigned ResultReg2 = FuncInfo.CreateRegs(CondTy);
+    unsigned ResultReg2 = createResultReg(&X86::GR8RegClass);
     assert((ResultReg+1) == ResultReg2 && "Nonconsecutive result registers.");
----------------
qcolombet wrote:
> Add a comment on why GR8 is the right class at this point.
Ok


Repository:
  rL LLVM

https://reviews.llvm.org/D26620