[AArch64] Stop reserved registers from being saved in prolog/epilog

[yasuna-oribe]

GCC's documentation is clear on how -ffixed-reg must behave:

  Treat the register named reg as a fixed register; generated
  code should never refer to it (except perhaps as a stack pointer,
  frame pointer or in some other fixed role).

This implies prolog/epilog code also must not save/restore explicitly fixed registers, even when it is callee-saved. Some projects rely on this (GCC's) behavior.

For example,

void f() {
  register uint64_t x28 asm("x28") = 0xee;
  asm volatile("" : "+r"(x28)); // avoid mov being eliminated
}

should not touch x28 outside of mov w28,#0xee.

For riscv64, clang behaves the same as GCC, so I am inclined to believe this is indeed a bug.

Fixes #111379.

[github-actions]

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write permissions for the repository. In which case you can instead tag reviewers by name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

[llvmbot]

@llvm/pr-subscribers-backend-aarch64

Author: None (yasuna-oribe)

Changes

GCC's man page is clear on how -ffixed-reg must behave:

  Treat the register named reg as a fixed register; generated
  code should never refer to it (except perhaps as a stack pointer,
  frame pointer or in some other fixed role).

This implies prolog/epilog code also must not save/restore explicitly fixed registers, even when it is callee-saved. Some projects rely on this (GCC's) behavior.

For example,

void f() {
  register uint64_t x28 asm("x28") = 0xee;
  asm volatile("" : "+r"(x28)); // avoid mov being eliminated
}

should not touch x28 outside of mov w28,#<!-- -->0xee.

For riscv64, clang behaves the same as GCC, so I am inclined to believe this is indeed a bug.

Fixes #111379.

---

Full diff: https://github.com/llvm/llvm-project/pull/138448.diff

1 Files Affected:

• (modified) llvm/lib/Target/AArch64/AArch64FrameLowering.cpp (+7)

diff --git a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
index 78ac57e3e92a6..2d72f8757d7c0 100644
--- a/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
+++ b/llvm/lib/Target/AArch64/AArch64FrameLowering.cpp
@@ -3619,6 +3619,13 @@ void AArch64FrameLowering::determineCalleeSaves(MachineFunction &MF,
     if (Reg == BasePointerReg)
       SavedRegs.set(Reg);
 
+    // Don't save fixed registers specified with -ffixed-reg.
+    if (AArch64::GPR64RegClass.contains(Reg) &&
+        RegInfo->isReservedReg(MF, Reg)) {
+      SavedRegs.reset(Reg);
+      continue;
+    }
+
     bool RegUsed = SavedRegs.test(Reg);
     unsigned PairedReg = AArch64::NoRegister;
     const bool RegIsGPR64 = AArch64::GPR64RegClass.contains(Reg);

[yasuna-oribe]

Yep, tests succeeded now. I'm sorry for the force push, I just noticed I'm supposed to use git --fixup.
ping @davemgreen @sjoerdmeijer @nasherm @smithp35 (Is this how I'm supposed to ping people?)