130 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 52; 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-SMEABI3; 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-prefixes=CHECK-COMMON,CHECK-AFTER-SMEABI4 5declare void @private_za_callee()6declare void @shared_za_callee() "aarch64_inout_za"7declare 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.10define void @requires_za_save() nounwind "aarch64_inout_za" {11 ; CHECK-BEFORE-SMEABI-LABEL: name: requires_za_save12 ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0):13 ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp14 ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo15 ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp16 ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp17 ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR18 ;19 ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save20 ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0):21 ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg22 ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp23 ; 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, 026 ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 027 ; 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 $sp30 ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo31 ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp32 ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp33 ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv34 ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv35 ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 036 ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x037 ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr38 ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR39 call void @private_za_callee()40 ret void41}42 43; Tests ZA state markers like "RequiresZASavePseudo" are placed before any streaming mode changes.44define 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_change46 ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0):47 ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp48 ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo49 ; CHECK-BEFORE-SMEABI-NEXT: MSRpstatesvcrImm1 1, 0, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit-def $sp, implicit $vg, implicit-def $vg, implicit-def $fpmr50 ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp51 ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp52 ; CHECK-BEFORE-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr53 ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR54 ;55 ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save_streaming_mode_change56 ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0):57 ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg58 ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp59 ; CHECK-AFTER-SMEABI-NEXT: [[MSUBXrrr:%[0-9]+]]:gpr64 = MSUBXrrr [[RDSVLI_XI]], [[RDSVLI_XI]], [[COPY]]60 ; CHECK-AFTER-SMEABI-NEXT: $sp = COPY [[MSUBXrrr]]61 ; CHECK-AFTER-SMEABI-NEXT: STPXi [[MSUBXrrr]], [[RDSVLI_XI]], %stack.0, 062 ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 063 ; CHECK-AFTER-SMEABI-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[ADDXri]]64 ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, [[COPY1]]65 ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp66 ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo67 ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 1, 0, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit-def $sp, implicit $vg, implicit-def $vg, implicit-def $fpmr68 ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp69 ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp70 ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr71 ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv72 ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv73 ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 074 ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x075 ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr76 ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR77 call void @private_za_callee()78 ret void79}80 81; Tests "InOutZAUsePseudo" is placed after the "ADJCALLSTACKDOWN" for shared ZA callees.82define void @inout_za_call() nounwind "aarch64_inout_za" {83 ; CHECK-COMMON-LABEL: name: inout_za_call84 ; CHECK-COMMON: bb.0 (%ir-block.0):85 ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp86 ; CHECK-COMMON-NEXT: InOutZAUsePseudo87 ; CHECK-COMMON-NEXT: BL @shared_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp88 ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp89 ; CHECK-COMMON-NEXT: RET_ReallyLR90 call void @shared_za_callee()91 ret void92}93 94; Tests ZA state markers like "InOutZAUsePseudo" are placed before any stack argument setup.95define 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" {96 ; CHECK-COMMON-LABEL: name: many_args_inout_za_call97 ; CHECK-COMMON: bb.0 (%ir-block.10):98 ; CHECK-COMMON-NEXT: liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x799 ; CHECK-COMMON-NEXT: {{ $}}100 ; CHECK-COMMON-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x7101 ; CHECK-COMMON-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x6102 ; CHECK-COMMON-NEXT: [[COPY2:%[0-9]+]]:gpr64 = COPY $x5103 ; CHECK-COMMON-NEXT: [[COPY3:%[0-9]+]]:gpr64 = COPY $x4104 ; CHECK-COMMON-NEXT: [[COPY4:%[0-9]+]]:gpr64 = COPY $x3105 ; CHECK-COMMON-NEXT: [[COPY5:%[0-9]+]]:gpr64 = COPY $x2106 ; CHECK-COMMON-NEXT: [[COPY6:%[0-9]+]]:gpr64 = COPY $x1107 ; CHECK-COMMON-NEXT: [[COPY7:%[0-9]+]]:gpr64 = COPY $x0108 ; CHECK-COMMON-NEXT: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.1, 0 :: (load (s64) from %fixed-stack.1, align 16)109 ; CHECK-COMMON-NEXT: [[LDRXui1:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.0, 0 :: (load (s64) from %fixed-stack.0)110 ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 16, 0, implicit-def dead $sp, implicit $sp111 ; CHECK-COMMON-NEXT: InOutZAUsePseudo112 ; CHECK-COMMON-NEXT: [[COPY8:%[0-9]+]]:gpr64sp = COPY $sp113 ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui1]], [[COPY8]], 1 :: (store (s64) into stack + 8)114 ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui]], [[COPY8]], 0 :: (store (s64) into stack)115 ; CHECK-COMMON-NEXT: $x0 = COPY [[COPY7]]116 ; CHECK-COMMON-NEXT: $x1 = COPY [[COPY6]]117 ; CHECK-COMMON-NEXT: $x2 = COPY [[COPY5]]118 ; CHECK-COMMON-NEXT: $x3 = COPY [[COPY4]]119 ; CHECK-COMMON-NEXT: $x4 = COPY [[COPY3]]120 ; CHECK-COMMON-NEXT: $x5 = COPY [[COPY2]]121 ; CHECK-COMMON-NEXT: $x6 = COPY [[COPY1]]122 ; CHECK-COMMON-NEXT: $x7 = COPY [[COPY]]123 ; 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 $x0124 ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 16, 0, implicit-def dead $sp, implicit $sp125 ; CHECK-COMMON-NEXT: [[COPY9:%[0-9]+]]:gpr64all = COPY $x0126 ; CHECK-COMMON-NEXT: RET_ReallyLR127 %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)128 ret void129}130