[PATCH] D80032: [AMDGPU] Always expand ext/insertelement with divergent idx

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/SIISelLowering.cpp:9549
   // instructions.
-  if (VecSize <= 256 && (VecSize > 64 || EltSize >= 32) &&
+  // Always do this if var-idx is divergent, otherwise it will become a loop.
+  if (ExpandDynVectors &&
----------------
arsenm wrote:
> GlobalISel needs the compare and select path implemented
Yes, although that is a separate issue. GlobalISel also needs to work with non-power of two vectors for movrel. Yet another piece of work is to tune the limits, they seem to be suboptimal at least for doubles.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D80032/new/

https://reviews.llvm.org/D80032