[PATCH] D100026: [X86] Support AMX fast register allocation

Xiang Zhang via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Apr 19 20:18:34 PDT 2021


xiangzhangllvm added inline comments.


================
Comment at: llvm/lib/Target/X86/X86LowerAMXType.cpp:96
+  ConstantInt *Gran = Builder.getInt16(Granularity);
+  Value *RealRow = Builder.CreateUDiv(V, Gran);
+  Col2Row[V] = RealRow;
----------------
LuoYuanke wrote:
> Do we need to calculate the value in compile time if V is constant?
CreateUDiv already handle it.


================
Comment at: llvm/lib/Target/X86/X86PreAMXConfig.cpp:69
+    if (Operand->getType()->isX86_AMXTy())
+      return false;
+  return II->getType()->isX86_AMXTy();
----------------
LuoYuanke wrote:
> I don't understand the function. Does it mean only tilezero intrinsic return true?
And currently tileload too, this function return is "intrinsic only def tile (not use tile)" or not.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D100026/new/

https://reviews.llvm.org/D100026



More information about the llvm-commits mailing list