[PATCH] D11370: Improved the interface of methods commuting operands, improved X86-FMA3 mem-folding&coalescing.

[Vyacheslav Klochkov via llvm-commits]

v_klochkov added a comment.

Please see my answer regarding the new assert in AMDGPU version of commuteInstructionImpl() method.
Thanks,
Slava


================
Comment at: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp:796
@@ -786,2 +795,3 @@
                                            AMDGPU::OpName::src1);
-  if (Src1Idx == -1)
+  assert(Src1Idx != -1 && "Should always have src1 operand");
+
----------------
qcolombet wrote:
> I still don't get why we need to turn this if into an assert.
(Src1Idx == -1) is impossible here as this place is reachable only after findCommutedOpIndices() call which filters out such situations.
This assert for Src1Idx is semantically and stylistically equivalent to the assert at the line 788 (assert for Src0Idx).


http://reviews.llvm.org/D11370