[llvm] [AMDGPU] Added hot-block-rematerialize pass (PR #136631)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,1511 @@
+//===- AMDGPUHotBlockRematerialize.cpp - AMDGPU Hot BlockRematerialize ----===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+/// \file
+/// \brief AMDGPU hot block Rematerialize
+//
+//===----------------------------------------------------------------------===//
+
+#include "AMDGPU.h"
+#include "AMDGPUMIRUtils.h"
+#include "AMDGPUOccupancyAndLatencyHelper.h"
+#include "GCNRegPressure.h"
+#include "SIInstrInfo.h"
+#include "SIMachineFunctionInfo.h"
+#include "SIRegisterInfo.h"
+#include "llvm/ADT/MapVector.h"
+#include "llvm/CodeGen/LiveInterval.h"
+#include "llvm/CodeGen/LiveIntervals.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineLoopInfo.h"
+#include "llvm/CodeGen/MachinePostDominators.h"
+#include "llvm/CodeGen/SlotIndexes.h"
+
+#define DEBUG_TYPE "amdgpu-hot-block-remat"
+
+using namespace llvm;
+
+static cl::opt<bool>
+    EnableAggressive("amdgpu-remat-enable-hot-block-remat-aggressive");
+static cl::opt<unsigned> TargetOccupancy("amdgpu-remat-target-occupancy");
----------------
arsenm wrote:

Avoid more cl::opts. There is already a target occupancy from the IR 

https://github.com/llvm/llvm-project/pull/136631