[mlir][MemRefToLLVM] Add fmin, fmax to AtomicRMW lowering

Add cases to the memref.atomicrmw lowering for floating-point min and
max, since LLVM supports these.

Reviewed By: bondhugula

Differential Revision: https://reviews.llvm.org/D158283

Author: krzysz00

mlir/lib/Conversion/MemRefToLLVM/MemRefToLLVM.cpp

@@ -1753,10 +1753,14 @@ matchSimpleAtomicOp(memref::AtomicRMWOp atomicOp) {
     return LLVM::AtomicBinOp::add;
   case arith::AtomicRMWKind::assign:
     return LLVM::AtomicBinOp::xchg;
+  case arith::AtomicRMWKind::maxf:
+    return LLVM::AtomicBinOp::fmax;
   case arith::AtomicRMWKind::maxs:
     return LLVM::AtomicBinOp::max;
   case arith::AtomicRMWKind::maxu:
     return LLVM::AtomicBinOp::umax;
+  case arith::AtomicRMWKind::minf:
+    return LLVM::AtomicBinOp::fmin;
   case arith::AtomicRMWKind::mins:
     return LLVM::AtomicBinOp::min;
   case arith::AtomicRMWKind::minu: