[PATCH] D150010: AMDGPU: fold fmed3 of fpext sources to f16 fmed3

Jay Foad via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Tue May 9 07:39:40 PDT 2023


foad added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUInstCombineIntrinsic.cpp:726
+    // Repeat floating-point width reduction done for minnum/maxnum.
+    // fmed3((fpext X), (fpext Y), (fpext Z)) -> fpext (fmed3(X, Y, Z))
+    if (matchFPExtFromF16(Src0, X) && matchFPExtFromF16(Src1, Y) &&
----------------
For constants, you need to check that they are exactly representable as half, otherwise this could change the result.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D150010/new/

https://reviews.llvm.org/D150010



More information about the llvm-commits mailing list