1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
llvm-mirror/test/CodeGen/Thumb2/mve-basic.ll
Eli Friedman f704804dd2 [SelectionDAG] Don't promote the alignment of allocas beyond the stack alignment.
allocas in LLVM IR have a specified alignment. When that alignment is
specified, the alloca has at least that alignment at runtime.

If the specified type of the alloca has a higher preferred alignment,
SelectionDAG currently ignores that specified alignment, and increases
the alignment. It does this even if it would trigger stack realignment.
I don't think this makes sense, so this patch changes that.

I was looking into this for SVE in particular: for SVE, overaligning
vscale'ed types is extra expensive because it requires realigning the
stack multiple times, or using dynamic allocation. (This currently isn't
implemented.)

I updated the expected assembly for a couple tests; in particular, for
arg-copy-elide.ll, the optimization in question does not increase the
alignment the way SelectionDAG normally would. For the rest, I just
increased the specified alignment on the allocas to match what
SelectionDAG was inferring.

Differential Revision: https://reviews.llvm.org/D79532
2020-05-11 17:39:00 -07:00

61 lines
2.1 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=thumbv8.1m.main -mattr=+mve.fp -o - %s | FileCheck %s
; RUN: llc -mtriple=thumbv8.1m.main -mattr=+mve -o - %s | FileCheck %s
define arm_aapcs_vfpcc <4 x i32> @vector_add_by_value(<4 x i32> %lhs, <4 x i32>%rhs) {
; CHECK-LABEL: vector_add_by_value:
; CHECK: @ %bb.0:
; CHECK-NEXT: @APP
; CHECK-NEXT: vadd.i32 q0, q0, q1
; CHECK-NEXT: @NO_APP
; CHECK-NEXT: bx lr
%result = tail call <4 x i32> asm "vadd.i32 $0,$1,$2", "=t,t,t"(<4 x i32> %lhs, <4 x i32> %rhs)
ret <4 x i32> %result
}
define void @vector_add_by_reference(<4 x i32>* %resultp, <4 x i32>* %lhsp, <4 x i32>* %rhsp) {
; CHECK-LABEL: vector_add_by_reference:
; CHECK: @ %bb.0:
; CHECK-NEXT: vldrw.u32 q0, [r1]
; CHECK-NEXT: vldrw.u32 q1, [r2]
; CHECK-NEXT: @APP
; CHECK-NEXT: vadd.i32 q0, q0, q1
; CHECK-NEXT: @NO_APP
; CHECK-NEXT: vstrw.32 q0, [r0]
; CHECK-NEXT: bx lr
%lhs = load <4 x i32>, <4 x i32>* %lhsp, align 16
%rhs = load <4 x i32>, <4 x i32>* %rhsp, align 16
%result = tail call <4 x i32> asm "vadd.i32 $0,$1,$2", "=t,t,t"(<4 x i32> %lhs, <4 x i32> %rhs)
store <4 x i32> %result, <4 x i32>* %resultp, align 16
ret void
}
define void @vector_f64_copy(<2 x double>* %from, <2 x double>* %to) {
; CHECK-LABEL: vector_f64_copy:
; CHECK: @ %bb.0:
; CHECK-NEXT: vldrw.u32 q0, [r0]
; CHECK-NEXT: vstrw.32 q0, [r1]
; CHECK-NEXT: bx lr
%v = load <2 x double>, <2 x double>* %from, align 16
store <2 x double> %v, <2 x double>* %to, align 16
ret void
}
define arm_aapcs_vfpcc <16 x i8> @stack_slot_handling(<16 x i8> %a) #0 {
; CHECK-LABEL: stack_slot_handling:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: sub sp, #16
; CHECK-NEXT: mov r0, sp
; CHECK-NEXT: vstrw.32 q0, [r0]
; CHECK-NEXT: vldrw.u32 q0, [r0]
; CHECK-NEXT: add sp, #16
; CHECK-NEXT: bx lr
entry:
%a.addr = alloca <16 x i8>, align 8
store <16 x i8> %a, <16 x i8>* %a.addr, align 8
%0 = load <16 x i8>, <16 x i8>* %a.addr, align 8
ret <16 x i8> %0
}
attributes #0 = { noinline optnone }