94 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 22; RUN: llc -aarch64-streaming-hazard-size=0 < %s | FileCheck %s3 4target triple = "aarch64"5 6; This function would normally scavenge a stackslot from the callee-save7; area, which would lead to spilling 's0' to that stackslot before the8; smstop and filling it with 'addvl + <offset>' after the smstop because9; the frame-pointer is not available.10; This would not be valid, since the vector-length has changed so 'addvl'11; cannot be used. This is testing that the stackslot-scavenging is disabled12; when there are any streaming-mode-changing call-sequences in the13; function.14define void @test_no_stackslot_scavenging(float %f) #0 {15; CHECK-LABEL: test_no_stackslot_scavenging:16; CHECK: // %bb.0:17; CHECK-NEXT: stp d15, d14, [sp, #-96]! // 16-byte Folded Spill18; CHECK-NEXT: stp d13, d12, [sp, #16] // 16-byte Folded Spill19; CHECK-NEXT: stp d11, d10, [sp, #32] // 16-byte Folded Spill20; CHECK-NEXT: stp d9, d8, [sp, #48] // 16-byte Folded Spill21; CHECK-NEXT: str x29, [sp, #64] // 8-byte Spill22; CHECK-NEXT: stp x30, x24, [sp, #80] // 16-byte Folded Spill23; CHECK-NEXT: sub sp, sp, #1624; CHECK-NEXT: addvl sp, sp, #-125; CHECK-NEXT: str s0, [sp, #12] // 4-byte Spill26; CHECK-NEXT: //APP27; CHECK-NEXT: //NO_APP28; CHECK-NEXT: smstop sm29; CHECK-NEXT: ldr s0, [sp, #12] // 4-byte Reload30; CHECK-NEXT: bl use_f31; CHECK-NEXT: smstart sm32; CHECK-NEXT: addvl sp, sp, #133; CHECK-NEXT: add sp, sp, #1634; CHECK-NEXT: ldp x30, x24, [sp, #80] // 16-byte Folded Reload35; CHECK-NEXT: ldr x29, [sp, #64] // 8-byte Reload36; CHECK-NEXT: ldp d9, d8, [sp, #48] // 16-byte Folded Reload37; CHECK-NEXT: ldp d11, d10, [sp, #32] // 16-byte Folded Reload38; CHECK-NEXT: ldp d13, d12, [sp, #16] // 16-byte Folded Reload39; CHECK-NEXT: ldp d15, d14, [sp], #96 // 16-byte Folded Reload40; CHECK-NEXT: ret41 %ptr = alloca <vscale x 16 x i8>42 call void asm sideeffect "", "~{x24}"() nounwind43 call void @use_f(float %f)44 ret void45}46 47define void @test_no_stackslot_scavenging_with_fp(float %f, i64 %n) #0 "frame-pointer"="all" {48; CHECK-LABEL: test_no_stackslot_scavenging_with_fp:49; CHECK: // %bb.0:50; CHECK-NEXT: stp d15, d14, [sp, #-112]! // 16-byte Folded Spill51; CHECK-NEXT: stp d13, d12, [sp, #16] // 16-byte Folded Spill52; CHECK-NEXT: stp d11, d10, [sp, #32] // 16-byte Folded Spill53; CHECK-NEXT: stp d9, d8, [sp, #48] // 16-byte Folded Spill54; CHECK-NEXT: stp x29, x30, [sp, #64] // 16-byte Folded Spill55; CHECK-NEXT: add x29, sp, #6456; CHECK-NEXT: stp x28, x25, [sp, #80] // 16-byte Folded Spill57; CHECK-NEXT: stp x24, x19, [sp, #96] // 16-byte Folded Spill58; CHECK-NEXT: sub sp, sp, #1659; CHECK-NEXT: addvl sp, sp, #-160; CHECK-NEXT: lsl x9, x0, #361; CHECK-NEXT: mov x8, sp62; CHECK-NEXT: mov x19, sp63; CHECK-NEXT: str s0, [x19, #12] // 4-byte Spill64; CHECK-NEXT: add x9, x9, #1565; CHECK-NEXT: and x9, x9, #0xfffffffffffffff066; CHECK-NEXT: sub x8, x8, x967; CHECK-NEXT: mov sp, x868; CHECK-NEXT: //APP69; CHECK-NEXT: //NO_APP70; CHECK-NEXT: smstop sm71; CHECK-NEXT: ldr s0, [x19, #12] // 4-byte Reload72; CHECK-NEXT: bl use_f73; CHECK-NEXT: smstart sm74; CHECK-NEXT: sub sp, x29, #6475; CHECK-NEXT: ldp x24, x19, [sp, #96] // 16-byte Folded Reload76; CHECK-NEXT: ldp x28, x25, [sp, #80] // 16-byte Folded Reload77; CHECK-NEXT: ldp x29, x30, [sp, #64] // 16-byte Folded Reload78; CHECK-NEXT: ldp d9, d8, [sp, #48] // 16-byte Folded Reload79; CHECK-NEXT: ldp d11, d10, [sp, #32] // 16-byte Folded Reload80; CHECK-NEXT: ldp d13, d12, [sp, #16] // 16-byte Folded Reload81; CHECK-NEXT: ldp d15, d14, [sp], #112 // 16-byte Folded Reload82; CHECK-NEXT: ret83 %ptr2 = alloca i64, i64 %n, align 884 %ptr = alloca <vscale x 16 x i8>85 call void asm sideeffect "", "~{x24},~{x25}"() nounwind86 call void @use_f(float %f)87 ret void88}89 90declare void @use_f(float)91declare void @use_f_and_ptr(float, ptr)92 93attributes #0 = { nounwind "target-features"="+sve,+sme" "aarch64_pstate_sm_enabled" }94