[PATCH] D51494: Fix unwind information for floating point registers

[Oliver Stannard via Phabricator via llvm-commits]

olista01 added inline comments.


================
Comment at: lib/Target/ARM/ARMAsmPrinter.cpp:1105
                  "Pad registers must come before restored ones");
-          Pad += 4;
+          const MachineFunction &MF = *MI->getParent()->getParent();
+          const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
----------------
We already get the MachineFunction and TargetRegisterInfo at the top of this function, and the MachineRegisterInfo should be moved up there to keep these together.


================
Comment at: test/CodeGen/ARM/unwind-fp.ll:1
+; RUN: llc < %s -mtriple=armv7a-arm-none-eabi | FileCheck %s
+; CHECK: pad #8
----------------
This test can be simplified a lot, to something like this:

  target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
  target triple = "armv7-arm-none-eabi"

  define void @foo() minsize {
  entry:
    %a = alloca i32, align 4
    call void asm sideeffect "", "r,~{d8}"(i32* %a)
    ret void
  }

Also, it would be clearer to put the CHECK line inside the IR for the function being tested, and to also check the other frame directives and the actual frame setup instructions.


Repository:
  rL LLVM

https://reviews.llvm.org/D51494