mirror of
https://github.com/c64scene-ar/llvm-6502.git
synced 2024-08-30 05:29:47 +00:00
This partially fixes PR13007 (ARM CodeGen fails with large stack alignment): for ARM and Thumb2 targets, but not for Thumb1, as it seems stack alignment for Thumb1 targets hasn't been supported at all. Producing an aligned stack pointer is done by zero-ing out the lower bits of the stack pointer. The BIC instruction was used for this. However, the immediate field of the BIC instruction only allows to encode an immediate that can zero out up to a maximum of the 8 lower bits. When a larger alignment is requested, a BIC instruction cannot be used; llvm was silently producing incorrect code in this case. This commit fixes code generation for large stack aligments by using the BFC instruction instead, when the BFC instruction is available. When not, it uses 2 instructions: a right shift, followed by a left shift to zero out the lower bits. The lowering of ARM::Int_eh_sjlj_dispatchsetup still has code that unconditionally uses BIC to realign the stack pointer, so it very likely has the same problem. However, I wasn't able to produce a test case for that. This commit adds an assert so that the compiler will fail the assert instead of silently generating wrong code if this is ever reached. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225446 91177308-0d34-0410-b5e6-96231b3b80d8
96 lines
3.1 KiB
LLVM
96 lines
3.1 KiB
LLVM
; RUN: llc < %s -mcpu=cortex-a8 -align-neon-spills=0 | FileCheck %s
|
|
; RUN: llc < %s -mcpu=cortex-a8 -align-neon-spills=1 | FileCheck %s --check-prefix=NEON
|
|
target datalayout = "e-p:32:32:32-i1:8:32-i8:8:32-i16:16:32-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:32:64-v128:32:128-a0:0:32-n32-S32"
|
|
target triple = "thumbv7-apple-ios"
|
|
|
|
; CHECK: f
|
|
; This function is forced to spill a double.
|
|
; Verify that the spill slot is properly aligned.
|
|
;
|
|
; The caller-saved r4 is used as a scratch register for stack realignment.
|
|
; CHECK: push {r4, r7, lr}
|
|
; CHECK: bfc r4, #0, #3
|
|
; CHECK: mov sp, r4
|
|
define void @f(double* nocapture %p) nounwind ssp {
|
|
entry:
|
|
%0 = load double* %p, align 4
|
|
tail call void asm sideeffect "", "~{d8},~{d9},~{d10},~{d11},~{d12},~{d13},~{d14},~{d15}"() nounwind
|
|
tail call void @g() nounwind
|
|
store double %0, double* %p, align 4
|
|
ret void
|
|
}
|
|
|
|
; NEON: f
|
|
; NEON: push {r4, r7, lr}
|
|
; NEON: sub.w r4, sp, #64
|
|
; NEON: bfc r4, #0, #4
|
|
; Stack pointer must be updated before the spills.
|
|
; NEON: mov sp, r4
|
|
; NEON: vst1.64 {d8, d9, d10, d11}, [r4:128]!
|
|
; NEON: vst1.64 {d12, d13, d14, d15}, [r4:128]
|
|
; Stack pointer adjustment for the stack frame contents.
|
|
; This could legally happen before the spills.
|
|
; Since the spill slot is only 8 bytes, technically it would be fine to only
|
|
; subtract #8 here. That would leave sp less aligned than some stack slots,
|
|
; and would probably blow MFI's mind.
|
|
; NEON: sub sp, #16
|
|
; The epilog is free to use another scratch register than r4.
|
|
; NEON: add r[[R4:[0-9]+]], sp, #16
|
|
; NEON: vld1.64 {d8, d9, d10, d11}, [r[[R4]]:128]!
|
|
; NEON: vld1.64 {d12, d13, d14, d15}, [r[[R4]]:128]
|
|
; The stack pointer restore must happen after the reloads.
|
|
; NEON: mov sp,
|
|
; NEON: pop
|
|
|
|
declare void @g()
|
|
|
|
; Spill 7 d-registers.
|
|
define void @f7(double* nocapture %p) nounwind ssp {
|
|
entry:
|
|
tail call void asm sideeffect "", "~{d8},~{d9},~{d10},~{d11},~{d12},~{d13},~{d14}"() nounwind
|
|
ret void
|
|
}
|
|
|
|
; NEON: f7
|
|
; NEON: push {r4, r7, lr}
|
|
; NEON: sub.w r4, sp, #56
|
|
; NEON: bfc r4, #0, #4
|
|
; Stack pointer must be updated before the spills.
|
|
; NEON: mov sp, r4
|
|
; NEON: vst1.64 {d8, d9, d10, d11}, [r4:128]!
|
|
; NEON: vst1.64 {d12, d13}, [r4:128]
|
|
; NEON: vstr d14, [r4, #16]
|
|
; Epilog
|
|
; NEON: vld1.64 {d8, d9, d10, d11},
|
|
; NEON: vld1.64 {d12, d13},
|
|
; NEON: vldr d14,
|
|
; The stack pointer restore must happen after the reloads.
|
|
; NEON: mov sp,
|
|
; NEON: pop
|
|
|
|
; Spill 7 d-registers, leave a hole.
|
|
define void @f3plus4(double* nocapture %p) nounwind ssp {
|
|
entry:
|
|
tail call void asm sideeffect "", "~{d8},~{d9},~{d10},~{d12},~{d13},~{d14},~{d15}"() nounwind
|
|
ret void
|
|
}
|
|
|
|
; Aligned spilling only works for contiguous ranges starting from d8.
|
|
; The rest goes to the standard vpush instructions.
|
|
; NEON: f3plus4
|
|
; NEON: push {r4, r7, lr}
|
|
; NEON: vpush {d12, d13, d14, d15}
|
|
; NEON: sub.w r4, sp, #24
|
|
; NEON: bfc r4, #0, #4
|
|
; Stack pointer must be updated before the spills.
|
|
; NEON: mov sp, r4
|
|
; NEON: vst1.64 {d8, d9}, [r4:128]
|
|
; NEON: vstr d10, [r4, #16]
|
|
; Epilog
|
|
; NEON: vld1.64 {d8, d9},
|
|
; NEON: vldr d10, [{{.*}}, #16]
|
|
; The stack pointer restore must happen after the reloads.
|
|
; NEON: mov sp,
|
|
; NEON: vpop {d12, d13, d14, d15}
|
|
; NEON: pop
|