================ @@ -0,0 +1,28 @@ +//===- MarkColdFunctions.h - ------------------------------------*- C++ -*-===// +// +// 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 +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_TRANSFORMS_INSTRUMENTATION_MARKCOLDFUNCTIONS_H +#define LLVM_TRANSFORMS_INSTRUMENTATION_MARKCOLDFUNCTIONS_H + +#include "llvm/IR/PassManager.h" +#include "llvm/Support/PGOOptions.h" + +namespace llvm { + +struct MarkColdFunctionsPass : public PassInfoMixin<MarkColdFunctionsPass> { ---------------- ellishg wrote:
I think this pass could be generalized a bit. It seems that the goal for this pr is to reduce the size overhead of `-O2` by forcing cold functions to be `optsize`/`minsize`. Another goal could be to improve the performance of `-Oz` by forcing very hot functions to be `optsize` (as opposed to `minsize`) or remove `optsize`/`minsize` from them entirely. This is actually something we explored in [this paper](https://dl.acm.org/doi/10.1145/3497776.3517764) a while back and I would eventually like to upstream something similar. I believe this could be incorporated into this pass, but the name seems to be too vague. I think "PGO" should to be in the name so users know that it is involved. Maybe `PGOForceFuncAttrsPass` or `ProfileGuidedFuncAttrsPass`? CC @kyulee-com https://github.com/llvm/llvm-project/pull/69030 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits