[llvm] [AMDGPU] Add DAG mutation to improve scheduling before barriers (PR #142716)

[Shilei Tian via llvm-commits]

================
@@ -0,0 +1,73 @@
+//===--- AMDGPUBarrierLatency.cpp - AMDGPU Barrier Latency ----------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file This file contains a DAG scheduling mutation to add latency to
+///       barrier edges between ATOMIC_FENCE instructions and preceding
+///       memory accesses potentially affected by the fence.
+///       This encourages the scheduling of more instructions before
+///       ATOMIC_FENCE instructions.  ATOMIC_FENCE instructions may
+///       introduce wait counting or indicate an impending S_BARRIER
+///       wait.  Having more instructions in-flight across these
+///       constructs improves latency hiding.
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPUBarrierLatency.h"
+#include "MCTargetDesc/AMDGPUMCTargetDesc.h"
+#include "SIInstrInfo.h"
+#include "llvm/CodeGen/ScheduleDAGInstrs.h"
+
+using namespace llvm;
+
+namespace {
+
+class BarrierLatency : public ScheduleDAGMutation {
+public:
+  BarrierLatency() = default;
+  void apply(ScheduleDAGInstrs *DAG) override;
+};
+
+void BarrierLatency::apply(ScheduleDAGInstrs *DAG) {
+  const unsigned SyntheticLatency = 2000;
----------------
shiltian wrote:

nit:
```suggestion
  constexpr const unsigned SyntheticLatency = 2000;
```

https://github.com/llvm/llvm-project/pull/142716