mirror of
https://github.com/RPCS3/llvm.git
synced 2024-12-24 04:56:51 +00:00
645c5b94e2
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
135 lines
4.1 KiB
LLVM
135 lines
4.1 KiB
LLVM
; RUN: llc -mtriple=arm-none-none-eabi -mcpu=cortex-a15 -o - %s | FileCheck --check-prefix=CHECK-A %s
|
|
; RUN: llc -mtriple=thumb-none-none-eabi -mcpu=cortex-a15 -o - %s | FileCheck --check-prefix=CHECK-A-THUMB %s
|
|
; RUN: llc -mtriple=thumb-apple-none-macho -mcpu=cortex-m3 -o - %s | FileCheck --check-prefix=CHECK-M %s
|
|
|
|
declare arm_aapcscc void @bar()
|
|
|
|
@bigvar = global [16 x i32] zeroinitializer
|
|
|
|
define arm_aapcscc void @irq_fn() alignstack(8) "interrupt"="IRQ" {
|
|
; Must save all registers except banked sp and lr (we save lr anyway because
|
|
; we actually need it at the end to execute the return ourselves).
|
|
|
|
; Also need special function return setting pc and CPSR simultaneously.
|
|
; CHECK-A-LABEL: irq_fn:
|
|
; CHECK-A: push {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: add r11, sp, #20
|
|
; CHECK-A-NOT: sub sp, sp, #{{[0-9]+}}
|
|
; CHECK-A: bic sp, sp, #7
|
|
; CHECK-A: bl bar
|
|
; CHECK-A: sub sp, r11, #20
|
|
; CHECK-A: pop {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: subs pc, lr, #4
|
|
|
|
; CHECK-A-THUMB-LABEL: irq_fn:
|
|
; CHECK-A-THUMB: push.w {r0, r1, r2, r3, r4, r7, r12, lr}
|
|
; CHECK-A-THUMB: add r7, sp, #20
|
|
; CHECK-A-THUMB: mov r4, sp
|
|
; CHECK-A-THUMB: bic r4, r4, #7
|
|
; CHECK-A-THUMB: bl bar
|
|
; CHECK-A-THUMB: sub.w r4, r7, #20
|
|
; CHECK-A-THUMB: mov sp, r4
|
|
; CHECK-A-THUMB: pop.w {r0, r1, r2, r3, r4, r7, r12, lr}
|
|
; CHECK-A-THUMB: subs pc, lr, #4
|
|
|
|
; 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.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, r11, #8
|
|
; CHECK-M: mov sp, r4
|
|
; CHECK-M: pop.w {r4, r10, r11, pc}
|
|
|
|
call arm_aapcscc void @bar()
|
|
ret void
|
|
}
|
|
|
|
; We don't push/pop r12, as it is banked for FIQ
|
|
define arm_aapcscc void @fiq_fn() alignstack(8) "interrupt"="FIQ" {
|
|
; CHECK-A-LABEL: fiq_fn:
|
|
; CHECK-A: push {r0, r1, r2, r3, r4, r5, r6, r7, r11, lr}
|
|
; 32 to get past r0, r1, ..., r7
|
|
; CHECK-A: add r11, sp, #32
|
|
; CHECK-A: sub sp, sp, #{{[0-9]+}}
|
|
; CHECK-A: bic sp, sp, #7
|
|
; [...]
|
|
; 32 must match above
|
|
; CHECK-A: sub sp, r11, #32
|
|
; CHECK-A: pop {r0, r1, r2, r3, r4, r5, r6, r7, r11, lr}
|
|
; CHECK-A: subs pc, lr, #4
|
|
|
|
; CHECK-A-THUMB-LABEL: fiq_fn:
|
|
; CHECK-M-LABEL: fiq_fn:
|
|
%val = load volatile [16 x i32]* @bigvar
|
|
store volatile [16 x i32] %val, [16 x i32]* @bigvar
|
|
ret void
|
|
}
|
|
|
|
define arm_aapcscc void @swi_fn() alignstack(8) "interrupt"="SWI" {
|
|
; CHECK-A-LABEL: swi_fn:
|
|
; CHECK-A: push {r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, lr}
|
|
; CHECK-A: add r11, sp, #44
|
|
; CHECK-A: sub sp, sp, #{{[0-9]+}}
|
|
; CHECK-A: bic sp, sp, #7
|
|
; [...]
|
|
; CHECK-A: sub sp, r11, #44
|
|
; CHECK-A: pop {r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, r10, r11, r12, lr}
|
|
; CHECK-A: subs pc, lr, #0
|
|
|
|
%val = load volatile [16 x i32]* @bigvar
|
|
store volatile [16 x i32] %val, [16 x i32]* @bigvar
|
|
ret void
|
|
}
|
|
|
|
define arm_aapcscc void @undef_fn() alignstack(8) "interrupt"="UNDEF" {
|
|
; CHECK-A-LABEL: undef_fn:
|
|
; CHECK-A: push {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: add r11, sp, #20
|
|
; CHECK-A-NOT: sub sp, sp, #{{[0-9]+}}
|
|
; CHECK-A: bic sp, sp, #7
|
|
; [...]
|
|
; CHECK-A: sub sp, r11, #20
|
|
; CHECK-A: pop {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: subs pc, lr, #0
|
|
|
|
call void @bar()
|
|
ret void
|
|
}
|
|
|
|
define arm_aapcscc void @abort_fn() alignstack(8) "interrupt"="ABORT" {
|
|
; CHECK-A-LABEL: abort_fn:
|
|
; CHECK-A: push {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: add r11, sp, #20
|
|
; CHECK-A-NOT: sub sp, sp, #{{[0-9]+}}
|
|
; CHECK-A: bic sp, sp, #7
|
|
; [...]
|
|
; CHECK-A: sub sp, r11, #20
|
|
; CHECK-A: pop {r0, r1, r2, r3, r10, r11, r12, lr}
|
|
; CHECK-A: subs pc, lr, #4
|
|
|
|
call void @bar()
|
|
ret void
|
|
}
|
|
|
|
@var = global double 0.0
|
|
|
|
; We don't save VFP regs, since it would be a massive overhead in the general
|
|
; case.
|
|
define arm_aapcscc void @floating_fn() alignstack(8) "interrupt"="IRQ" {
|
|
; CHECK-A-LABEL: floating_fn:
|
|
; CHECK-A-NOT: vpush
|
|
; CHECK-A-NOT: vstr
|
|
; CHECK-A-NOT: vstm
|
|
; CHECK-A: vadd.f64 {{d[0-9]+}}, {{d[0-9]+}}, {{d[0-9]+}}
|
|
%lhs = load volatile double* @var
|
|
%rhs = load volatile double* @var
|
|
%sum = fadd double %lhs, %rhs
|
|
store double %sum, double* @var
|
|
ret void
|
|
}
|