79 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 22; RUN: llc < %s | FileCheck %s3 4; Verify that the following code can be compiled without +sme, because if the5; call is not entered in streaming-SVE mode at runtime, the codepath leading6; to the smstop/smstart pair will not be executed either.7 8target triple = "aarch64"9 10define void @streaming_compatible() #0 {11; CHECK-LABEL: streaming_compatible:12; CHECK: // %bb.0:13; CHECK-NEXT: stp d15, d14, [sp, #-80]! // 16-byte Folded Spill14; CHECK-NEXT: stp d13, d12, [sp, #16] // 16-byte Folded Spill15; CHECK-NEXT: stp d11, d10, [sp, #32] // 16-byte Folded Spill16; CHECK-NEXT: stp d9, d8, [sp, #48] // 16-byte Folded Spill17; CHECK-NEXT: stp x30, x19, [sp, #64] // 16-byte Folded Spill18; CHECK-NEXT: bl __arm_sme_state19; CHECK-NEXT: mov x19, x020; CHECK-NEXT: tbz w19, #0, .LBB0_221; CHECK-NEXT: // %bb.1:22; CHECK-NEXT: smstop sm23; CHECK-NEXT: .LBB0_2:24; CHECK-NEXT: bl non_streaming25; CHECK-NEXT: tbz w19, #0, .LBB0_426; CHECK-NEXT: // %bb.3:27; CHECK-NEXT: smstart sm28; CHECK-NEXT: .LBB0_4:29; CHECK-NEXT: ldp x30, x19, [sp, #64] // 16-byte Folded Reload30; CHECK-NEXT: ldp d9, d8, [sp, #48] // 16-byte Folded Reload31; CHECK-NEXT: ldp d11, d10, [sp, #32] // 16-byte Folded Reload32; CHECK-NEXT: ldp d13, d12, [sp, #16] // 16-byte Folded Reload33; CHECK-NEXT: ldp d15, d14, [sp], #80 // 16-byte Folded Reload34; CHECK-NEXT: ret35 call void @non_streaming()36 ret void37}38 39declare void @non_streaming()40 41 42; Verify that COALESCER_BARRIER is also supported without +sme.43 44define void @streaming_compatible_arg(float %f) #0 {45; CHECK-LABEL: streaming_compatible_arg:46; CHECK: // %bb.0:47; CHECK-NEXT: sub sp, sp, #9648; CHECK-NEXT: stp d15, d14, [sp, #16] // 16-byte Folded Spill49; CHECK-NEXT: stp d13, d12, [sp, #32] // 16-byte Folded Spill50; CHECK-NEXT: stp d11, d10, [sp, #48] // 16-byte Folded Spill51; CHECK-NEXT: stp d9, d8, [sp, #64] // 16-byte Folded Spill52; CHECK-NEXT: stp x30, x19, [sp, #80] // 16-byte Folded Spill53; CHECK-NEXT: str s0, [sp, #12] // 4-byte Spill54; CHECK-NEXT: bl __arm_sme_state55; CHECK-NEXT: mov x19, x056; CHECK-NEXT: tbz w19, #0, .LBB1_257; CHECK-NEXT: // %bb.1:58; CHECK-NEXT: smstop sm59; CHECK-NEXT: .LBB1_2:60; CHECK-NEXT: ldr s0, [sp, #12] // 4-byte Reload61; CHECK-NEXT: bl non_streaming62; CHECK-NEXT: tbz w19, #0, .LBB1_463; CHECK-NEXT: // %bb.3:64; CHECK-NEXT: smstart sm65; CHECK-NEXT: .LBB1_4:66; CHECK-NEXT: ldp x30, x19, [sp, #80] // 16-byte Folded Reload67; CHECK-NEXT: ldp d9, d8, [sp, #64] // 16-byte Folded Reload68; CHECK-NEXT: ldp d11, d10, [sp, #48] // 16-byte Folded Reload69; CHECK-NEXT: ldp d13, d12, [sp, #32] // 16-byte Folded Reload70; CHECK-NEXT: ldp d15, d14, [sp, #16] // 16-byte Folded Reload71; CHECK-NEXT: add sp, sp, #9672; CHECK-NEXT: ret73 call void @non_streaming(float %f)74 ret void75}76 77 78attributes #0 = { nounwind "aarch64_pstate_sm_compatible" }79