[llvm] [AArch64] New pass for code layout optimizations. (PR #184434)

[Florian Hahn via llvm-commits]

================
@@ -0,0 +1,262 @@
+//===-- AArch64CodeLayoutOpt.cpp - Code Layout Optimizations --===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass runs after instruction scheduling and employs code layout
+// optimizations for certain patterns.
+//
+// Option -aarch64-code-layout-opt-enable selects instruction pairs to optimize:
+//   cmp-csel:   Enable CMP/CMN-CSEL code layout optimization
+//   fcmp-fcsel: Enable FCMP-FCSEL code layout optimization
+//
+// The initial implementation induces function alignment when a supported
+// pattern is detected, and possibly instruction-alignment when a pair would
+// straddle cache-lines.
+//===----------------------------------------------------------------------===//
+
+#include "AArch64.h"
+#include "AArch64InstrInfo.h"
+#include "AArch64Subtarget.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/MathExtras.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "aarch64-code-layout-opt"
+#define DBG(...) LLVM_DEBUG(dbgs() << DEBUG_TYPE ": " << __VA_ARGS__)
+#define AARCH64_CODE_LAYOUT_OPT_NAME "AArch64 Code Layout Optimization"
+
+enum CodeLayoutOpt {
+  CmpCsel,   // Align CMP/CMN-CSEL pairs
+  FcmpFcsel, // Align FCMP-FCSEL pairs
+};
+
+static cl::bits<CodeLayoutOpt> EnableCodeAlignment(
+    "aarch64-code-layout-opt-enable", cl::Hidden, cl::CommaSeparated,
+    cl::desc("Enable code alignment optimization for instruction pairs"),
+    cl::values(
+        clEnumValN(CmpCsel, "cmp-csel", "CMP/CMN-CSEL pair alignment (32-bit)"),
+        clEnumValN(FcmpFcsel, "fcmp-fcsel", "FCMP-FCSEL pair alignment")));
+
+static cl::opt<unsigned> FunctionAlignBytes(
+    "aarch64-code-layout-opt-align-functions", cl::Hidden,
+    cl::desc("Function alignment in bytes for code layout optimization "
+             "(must be a power of 2)"),
+    cl::init(64), cl::callback([](const unsigned &Val) {
----------------
fhahn wrote:

Would be good to also test the option

https://github.com/llvm/llvm-project/pull/184434