llvm-6502/test/CodeGen/ARM/none-macho.ll
Tim Northover 645c5b94e2 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

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@209895 91177308-0d34-0410-b5e6-96231b3b80d8
2014-05-30 13:23:06 +00:00

100 lines
2.8 KiB
LLVM

; RUN: llc -mtriple=thumbv7m-none-macho %s -o - -relocation-model=pic -disable-fp-elim | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-NON-FAST
; RUN: llc -mtriple=thumbv7m-none-macho -O0 %s -o - -relocation-model=pic -disable-fp-elim | FileCheck %s
; RUN: llc -mtriple=thumbv7m-none-macho -filetype=obj %s -o /dev/null
; Bare-metal should probably "declare" segments just like normal MachO
; CHECK: __picsymbolstub4
; CHECK: __StaticInit
; CHECK: __text
@var = external global i32
define i32 @test_litpool() minsize {
; CHECK-LABEL: test_litpool:
%val = load i32* @var
ret i32 %val
; Lit-pool entries need to produce a "$non_lazy_ptr" version of the symbol.
; CHECK: LCPI0_0:
; CHECK-NEXT: .long L_var$non_lazy_ptr-(LPC0_0+4)
}
define i32 @test_movw_movt() {
; CHECK-LABEL: test_movw_movt:
%val = load i32* @var
ret i32 %val
; movw/movt should also address their symbols MachO-style
; CHECK: movw [[RTMP:r[0-9]+]], :lower16:(L_var$non_lazy_ptr-(LPC1_0+4))
; CHECK: movt [[RTMP]], :upper16:(L_var$non_lazy_ptr-(LPC1_0+4))
; CHECK: LPC1_0:
; CHECK: add [[RTMP]], pc
}
declare void @llvm.trap()
define void @test_trap() {
; CHECK-LABEL: test_trap:
; Bare-metal MachO gets compiled on top of normal MachO toolchain which
; understands trap natively.
call void @llvm.trap()
; CHECK: trap
ret void
}
define i32 @test_frame_ptr() {
; CHECK-LABEL: test_frame_ptr:
call void @test_trap()
; Frame pointer is r11.
; CHECK: mov r11, sp
ret i32 42
}
%big_arr = type [8 x i32]
define void @test_two_areas(%big_arr* %addr) {
; CHECK-LABEL: test_two_areas:
%val = load %big_arr* %addr
call void @test_trap()
store %big_arr %val, %big_arr* %addr
; This goes with the choice of r7 as FP (largely). FP and LR have to be stored
; consecutively on the stack for the frame record to be valid, which means we
; need the 2 register-save areas employed by iOS.
; CHECK-NON-FAST: push.w {r4, r5, r6, r7, r8, r9, r10, r11, lr}
; ...
; CHECK-NON-FAST: pop.w {r4, r5, r6, r7, r8, r9, r10, r11, pc}
ret void
}
define void @test_tail_call() {
; CHECK-LABEL: test_tail_call:
tail call void @test_trap()
; Tail calls should be available and use Thumb2 branch.
; CHECK: b.w _test_trap
ret void
}
define float @test_softfloat_calls(float %in) {
; CHECK-LABEL: test_softfloat_calls:
%sum = fadd float %in, %in
; Soft-float calls should be GNU-style rather than RTABI and should not be the
; *vfp variants used for ARMv6 iOS.
; CHECK: blx ___addsf3{{$}}
ret float %sum
}
; Even bare-metal PIC needs GOT-like behaviour, in principle. Depends a bit on
; the use-case of course, but LLVM doesn't know what that is.
; CHECK: non_lazy_symbol_pointers
; CHECK: L_var$non_lazy_ptr:
; CHECK-NEXT: .indirect_symbol _var
; All MachO objects should have this to give the linker leeway in removing
; dead code.
; CHECK: .subsections_via_symbols