[(IsTailOfShape $rhs, $lhs),
      (IsTailOfShape $lhs, $rhs),
      (IsTailOfShape $input1, $input2),
      (IsTailOfShape $input2, $input1),
      (SameElementType $input1, $input2)]>;

  // Move binary op batched RHS before reshape:
  // binary(reshape(lhs), rhs) => reshape(binary(lhs, flatten(rhs)))
  // Pattern targetted here is as follows-
  // [input, lhr, rhs] == [<1x1024x128>, <1x1024x8x16>, <1x1x8x16xf32>]

#include "mlir/IR/BuiltinOps.h"  // from @llvm-project
#include "mlir/IR/BuiltinTypes.h"  // from @llvm-project
#include "mlir/IR/MLIRContext.h"  // from @llvm-project
#include "mlir/IR/Matchers.h"  // from @llvm-project
#include "mlir/IR/Operation.h"  // from @llvm-project
#include "mlir/IR/OperationSupport.h"  // from @llvm-project
#include "mlir/IR/PatternMatch.h"  // from @llvm-project

     * characters in addition to the normal filename characters. The '*'
     * character matches any number of characters in part of a name. If
     * the expression begins with one or more '?'s then exactly that
     * many characters will be matched whereas if it ends with '?'s
     * it will match that many characters <i>or less</i>.
     * <p>

		// rather than starting all in a batch and then waiting once
		// afterwards. By running one goroutine at a time, we can take
		// advantage of runnext to bounce back and forth between
		// workers and this goroutine. In an overloaded application,
		// this can reduce GC start latency by prioritizing these
		// goroutines rather than waiting on the end of the run queue.
		<-ready

	// of gc_sys or something. The code below just pretends it is
	// internal fragmentation and matches the GC's accounting by
	// using the whole allocation slot.
	fullSize := span.elemsize
	if rate := MemProfileRate; rate > 0 {
		// Note cache c only valid while m acquired; see #47302
		//
		// N.B. Use the full size because that matches how the GC
		// will update the mem profile on the "free" side.