|
| 1 | +; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5 |
| 2 | +; RUN: llc -mtriple=aarch64 -aarch64-streaming-hazard-size=0 -mattr=+sme,+sve -aarch64-new-sme-abi -stop-before=aarch64-machine-sme-abi -verify-machineinstrs < %s | FileCheck %s --check-prefixes=CHECK-COMMON,CHECK-BEFORE-SMEABI |
| 3 | +; RUN: llc -mtriple=aarch64 -aarch64-streaming-hazard-size=0 -mattr=+sme,+sve -aarch64-new-sme-abi -stop-after=aarch64-machine-sme-abi -verify-machineinstrs < %s | FileCheck %s --check-prefix=CHECK-COMMON,CHECK-AFTER-SMEABI |
| 4 | + |
| 5 | +declare void @private_za_callee() |
| 6 | +declare void @shared_za_callee() "aarch64_inout_za" |
| 7 | +declare i64 @shared_za_callee_many_args(i64, i64, i64, i64, i64, i64, i64, i64, i64, i64) "aarch64_inout_za" |
| 8 | + |
| 9 | +; Tests a "RequiresZASavePseudo" is placed after the "ADJCALLSTACKDOWN" for private ZA callees. |
| 10 | +define void @requires_za_save() nounwind "aarch64_inout_za" { |
| 11 | + ; CHECK-BEFORE-SMEABI-LABEL: name: requires_za_save |
| 12 | + ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0): |
| 13 | + ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| 14 | + ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo |
| 15 | + ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| 16 | + ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| 17 | + ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR |
| 18 | + ; |
| 19 | + ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save |
| 20 | + ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0): |
| 21 | + ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg |
| 22 | + ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp |
| 23 | + ; CHECK-AFTER-SMEABI-NEXT: [[MSUBXrrr:%[0-9]+]]:gpr64 = MSUBXrrr [[RDSVLI_XI]], [[RDSVLI_XI]], [[COPY]] |
| 24 | + ; CHECK-AFTER-SMEABI-NEXT: $sp = COPY [[MSUBXrrr]] |
| 25 | + ; CHECK-AFTER-SMEABI-NEXT: STPXi [[MSUBXrrr]], [[RDSVLI_XI]], %stack.0, 0 |
| 26 | + ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 0 |
| 27 | + ; CHECK-AFTER-SMEABI-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[ADDXri]] |
| 28 | + ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, [[COPY1]] |
| 29 | + ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| 30 | + ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo |
| 31 | + ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| 32 | + ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| 33 | + ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv |
| 34 | + ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv |
| 35 | + ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 0 |
| 36 | + ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x0 |
| 37 | + ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr |
| 38 | + ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR |
| 39 | + call void @private_za_callee() |
| 40 | + ret void |
| 41 | +} |
| 42 | + |
| 43 | +; Tests ZA state markers like "RequiresZASavePseudo" are placed before any streaming mode changes. |
| 44 | +define void @requires_za_save_streaming_mode_change() nounwind "aarch64_inout_za" "aarch64_pstate_sm_enabled" { |
| 45 | + ; CHECK-BEFORE-SMEABI-LABEL: name: requires_za_save_streaming_mode_change |
| 46 | + ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0): |
| 47 | + ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| 48 | + ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo |
| 49 | + ; CHECK-BEFORE-SMEABI-NEXT: VGSavePseudo |
| 50 | + ; CHECK-BEFORE-SMEABI-NEXT: MSRpstatesvcrImm1 1, 0, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit-def $sp, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| 51 | + ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| 52 | + ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| 53 | + ; CHECK-BEFORE-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| 54 | + ; CHECK-BEFORE-SMEABI-NEXT: VGRestorePseudo |
| 55 | + ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR |
| 56 | + ; |
| 57 | + ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save_streaming_mode_change |
| 58 | + ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0): |
| 59 | + ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg |
| 60 | + ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp |
| 61 | + ; CHECK-AFTER-SMEABI-NEXT: [[MSUBXrrr:%[0-9]+]]:gpr64 = MSUBXrrr [[RDSVLI_XI]], [[RDSVLI_XI]], [[COPY]] |
| 62 | + ; CHECK-AFTER-SMEABI-NEXT: $sp = COPY [[MSUBXrrr]] |
| 63 | + ; CHECK-AFTER-SMEABI-NEXT: STPXi [[MSUBXrrr]], [[RDSVLI_XI]], %stack.0, 0 |
| 64 | + ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 0 |
| 65 | + ; CHECK-AFTER-SMEABI-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[ADDXri]] |
| 66 | + ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, [[COPY1]] |
| 67 | + ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| 68 | + ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo |
| 69 | + ; CHECK-AFTER-SMEABI-NEXT: VGSavePseudo |
| 70 | + ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 1, 0, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit-def $sp, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| 71 | + ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| 72 | + ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| 73 | + ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| 74 | + ; CHECK-AFTER-SMEABI-NEXT: VGRestorePseudo |
| 75 | + ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv |
| 76 | + ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv |
| 77 | + ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 0 |
| 78 | + ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x0 |
| 79 | + ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr |
| 80 | + ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR |
| 81 | + call void @private_za_callee() |
| 82 | + ret void |
| 83 | +} |
| 84 | + |
| 85 | +; Tests "InOutZAUsePseudo" is placed after the "ADJCALLSTACKDOWN" for shared ZA callees. |
| 86 | +define void @inout_za_call() nounwind "aarch64_inout_za" { |
| 87 | + ; CHECK-COMMON-LABEL: name: inout_za_call |
| 88 | + ; CHECK-COMMON: bb.0 (%ir-block.0): |
| 89 | + ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| 90 | + ; CHECK-COMMON-NEXT: InOutZAUsePseudo |
| 91 | + ; CHECK-COMMON-NEXT: BL @shared_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| 92 | + ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| 93 | + ; CHECK-COMMON-NEXT: RET_ReallyLR |
| 94 | + call void @shared_za_callee() |
| 95 | + ret void |
| 96 | +} |
| 97 | + |
| 98 | +; Tests ZA state markers like "InOutZAUsePseudo" are placed before any stack argument setup. |
| 99 | +define void @many_args_inout_za_call(i64 %0, i64 %1, i64 %2, i64 %3, i64 %4, i64 %5, i64 %6, i64 %7, i64 %8, i64 %9) "aarch64_inout_za" { |
| 100 | + ; CHECK-COMMON-LABEL: name: many_args_inout_za_call |
| 101 | + ; CHECK-COMMON: bb.0 (%ir-block.10): |
| 102 | + ; CHECK-COMMON-NEXT: liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x7 |
| 103 | + ; CHECK-COMMON-NEXT: {{ $}} |
| 104 | + ; CHECK-COMMON-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x7 |
| 105 | + ; CHECK-COMMON-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x6 |
| 106 | + ; CHECK-COMMON-NEXT: [[COPY2:%[0-9]+]]:gpr64 = COPY $x5 |
| 107 | + ; CHECK-COMMON-NEXT: [[COPY3:%[0-9]+]]:gpr64 = COPY $x4 |
| 108 | + ; CHECK-COMMON-NEXT: [[COPY4:%[0-9]+]]:gpr64 = COPY $x3 |
| 109 | + ; CHECK-COMMON-NEXT: [[COPY5:%[0-9]+]]:gpr64 = COPY $x2 |
| 110 | + ; CHECK-COMMON-NEXT: [[COPY6:%[0-9]+]]:gpr64 = COPY $x1 |
| 111 | + ; CHECK-COMMON-NEXT: [[COPY7:%[0-9]+]]:gpr64 = COPY $x0 |
| 112 | + ; CHECK-COMMON-NEXT: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.1, 0 :: (load (s64) from %fixed-stack.1, align 16) |
| 113 | + ; CHECK-COMMON-NEXT: [[LDRXui1:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.0, 0 :: (load (s64) from %fixed-stack.0) |
| 114 | + ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 16, 0, implicit-def dead $sp, implicit $sp |
| 115 | + ; CHECK-COMMON-NEXT: InOutZAUsePseudo |
| 116 | + ; CHECK-COMMON-NEXT: [[COPY8:%[0-9]+]]:gpr64sp = COPY $sp |
| 117 | + ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui1]], [[COPY8]], 1 :: (store (s64) into stack + 8) |
| 118 | + ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui]], [[COPY8]], 0 :: (store (s64) into stack) |
| 119 | + ; CHECK-COMMON-NEXT: $x0 = COPY [[COPY7]] |
| 120 | + ; CHECK-COMMON-NEXT: $x1 = COPY [[COPY6]] |
| 121 | + ; CHECK-COMMON-NEXT: $x2 = COPY [[COPY5]] |
| 122 | + ; CHECK-COMMON-NEXT: $x3 = COPY [[COPY4]] |
| 123 | + ; CHECK-COMMON-NEXT: $x4 = COPY [[COPY3]] |
| 124 | + ; CHECK-COMMON-NEXT: $x5 = COPY [[COPY2]] |
| 125 | + ; CHECK-COMMON-NEXT: $x6 = COPY [[COPY1]] |
| 126 | + ; CHECK-COMMON-NEXT: $x7 = COPY [[COPY]] |
| 127 | + ; CHECK-COMMON-NEXT: BL @shared_za_callee_many_args, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit $x0, implicit $x1, implicit $x2, implicit $x3, implicit $x4, implicit $x5, implicit $x6, implicit $x7, implicit-def $sp, implicit-def $x0 |
| 128 | + ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 16, 0, implicit-def dead $sp, implicit $sp |
| 129 | + ; CHECK-COMMON-NEXT: [[COPY9:%[0-9]+]]:gpr64all = COPY $x0 |
| 130 | + ; CHECK-COMMON-NEXT: RET_ReallyLR |
| 131 | + %ret = call i64 @shared_za_callee_many_args(i64 %0, i64 %1, i64 %2, i64 %3, i64 %4, i64 %5, i64 %6, i64 %7, i64 %8, i64 %9) |
| 132 | + ret void |
| 133 | +} |
0 commit comments