ARM: use AAPCS-style prologues for embedded MachO.

Darwin prologues save their GPRs in two stages: a narrow push of r0-r7 & lr,
followed by a wide push of the remaining registers if there are any. AAPCS uses
a single push.w instruction.

It turns out that, on average, enough registers get pushed that code is smaller
in the AAPCS prologue, which is a nice property for M-class programmers. They
also have other options available for back-traces, so can hopefully deal with
the fact that FP & LR aren't adjacent in memory.

rdar://problem/15909583

llvm-svn: 209895

1 files changed, 4 insertions, 4 deletions

diff --git a/llvm/test/CodeGen/ARM/interrupt-attr.ll b/llvm/test/CodeGen/ARM/interrupt-attr.ll
index c5be6674da1..cb67dd929f4 100644
--- a/llvm/test/CodeGen/ARM/interrupt-attr.ll
+++ b/llvm/test/CodeGen/ARM/interrupt-attr.ll
@@ -35,15 +35,15 @@ define arm_aapcscc void @irq_fn() alignstack(8) "interrupt"="IRQ" {
   ; Normal AAPCS function (r0-r3 pushed onto stack by hardware, lr set to
   ; appropriate sentinel so no special return needed).
 ; CHECK-M-LABEL: irq_fn:
-; CHECK-M: push {r4, r6, r7, lr}
-; CHECK-M: add r7, sp, #8
+; CHECK-M: push.w {r4, r10, r11, lr}
+; CHECK-M: add.w r11, sp, #8
 ; CHECK-M: mov r4, sp
 ; CHECK-M: bic r4, r4, #7
 ; CHECK-M: mov sp, r4
 ; CHECK-M: blx _bar
-; CHECK-M: sub.w r4, r7, #8
+; CHECK-M: sub.w r4, r11, #8
 ; CHECK-M: mov sp, r4
-; CHECK-M: pop {r4, r6, r7, pc}
+; CHECK-M: pop.w {r4, r10, r11, pc}
 
   call arm_aapcscc void @bar()
   ret void