| ; NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5 |
| ; 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 |
| ; 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-SMEABI |
| |
| declare void @private_za_callee() |
| declare void @shared_za_callee() "aarch64_inout_za" |
| declare i64 @shared_za_callee_many_args(i64, i64, i64, i64, i64, i64, i64, i64, i64, i64) "aarch64_inout_za" |
| |
| ; Tests a "RequiresZASavePseudo" is placed after the "ADJCALLSTACKDOWN" for private ZA callees. |
| define void @requires_za_save() nounwind "aarch64_inout_za" { |
| ; CHECK-BEFORE-SMEABI-LABEL: name: requires_za_save |
| ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0): |
| ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo |
| ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR |
| ; |
| ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save |
| ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0): |
| ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg |
| ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp |
| ; CHECK-AFTER-SMEABI-NEXT: [[MSUBXrrr:%[0-9]+]]:gpr64 = MSUBXrrr [[RDSVLI_XI]], [[RDSVLI_XI]], [[COPY]] |
| ; CHECK-AFTER-SMEABI-NEXT: $sp = COPY [[MSUBXrrr]] |
| ; CHECK-AFTER-SMEABI-NEXT: STPXi [[MSUBXrrr]], [[RDSVLI_XI]], %stack.0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[ADDXri]] |
| ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, [[COPY1]] |
| ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo |
| ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv |
| ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv |
| ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x0 |
| ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr |
| ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR |
| call void @private_za_callee() |
| ret void |
| } |
| |
| ; Tests ZA state markers like "RequiresZASavePseudo" are placed before any streaming mode changes. |
| define void @requires_za_save_streaming_mode_change() nounwind "aarch64_inout_za" "aarch64_pstate_sm_enabled" { |
| ; CHECK-BEFORE-SMEABI-LABEL: name: requires_za_save_streaming_mode_change |
| ; CHECK-BEFORE-SMEABI: bb.0 (%ir-block.0): |
| ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: RequiresZASavePseudo |
| ; 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 |
| ; CHECK-BEFORE-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-BEFORE-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| ; CHECK-BEFORE-SMEABI-NEXT: RET_ReallyLR |
| ; |
| ; CHECK-AFTER-SMEABI-LABEL: name: requires_za_save_streaming_mode_change |
| ; CHECK-AFTER-SMEABI: bb.0 (%ir-block.0): |
| ; CHECK-AFTER-SMEABI-NEXT: [[RDSVLI_XI:%[0-9]+]]:gpr64 = RDSVLI_XI 1, implicit $vg |
| ; CHECK-AFTER-SMEABI-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $sp |
| ; CHECK-AFTER-SMEABI-NEXT: [[MSUBXrrr:%[0-9]+]]:gpr64 = MSUBXrrr [[RDSVLI_XI]], [[RDSVLI_XI]], [[COPY]] |
| ; CHECK-AFTER-SMEABI-NEXT: $sp = COPY [[MSUBXrrr]] |
| ; CHECK-AFTER-SMEABI-NEXT: STPXi [[MSUBXrrr]], [[RDSVLI_XI]], %stack.0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: [[ADDXri:%[0-9]+]]:gpr64sp = ADDXri %stack.0, 0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[ADDXri]] |
| ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, [[COPY1]] |
| ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-AFTER-SMEABI-NEXT: RequiresZASavePseudo |
| ; 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 |
| ; CHECK-AFTER-SMEABI-NEXT: BL @private_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| ; CHECK-AFTER-SMEABI-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 1, 1, csr_aarch64_smstartstop, implicit-def dead $nzcv, implicit $vg, implicit-def $vg, implicit-def $fpmr |
| ; CHECK-AFTER-SMEABI-NEXT: MSRpstatesvcrImm1 2, 1, implicit-def $nzcv |
| ; CHECK-AFTER-SMEABI-NEXT: [[MRS:%[0-9]+]]:gpr64 = MRS 56965, implicit-def $nzcv |
| ; CHECK-AFTER-SMEABI-NEXT: $x0 = ADDXri %stack.0, 0, 0 |
| ; CHECK-AFTER-SMEABI-NEXT: RestoreZAPseudo [[MRS]], $x0, &__arm_tpidr2_restore, csr_aarch64_sme_abi_support_routines_preservemost_from_x0 |
| ; CHECK-AFTER-SMEABI-NEXT: MSR 56965, $xzr |
| ; CHECK-AFTER-SMEABI-NEXT: RET_ReallyLR |
| call void @private_za_callee() |
| ret void |
| } |
| |
| ; Tests "InOutZAUsePseudo" is placed after the "ADJCALLSTACKDOWN" for shared ZA callees. |
| define void @inout_za_call() nounwind "aarch64_inout_za" { |
| ; CHECK-COMMON-LABEL: name: inout_za_call |
| ; CHECK-COMMON: bb.0 (%ir-block.0): |
| ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-COMMON-NEXT: InOutZAUsePseudo |
| ; CHECK-COMMON-NEXT: BL @shared_za_callee, csr_aarch64_aapcs, implicit-def dead $lr, implicit $sp, implicit-def $sp |
| ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 0, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-COMMON-NEXT: RET_ReallyLR |
| call void @shared_za_callee() |
| ret void |
| } |
| |
| ; Tests ZA state markers like "InOutZAUsePseudo" are placed before any stack argument setup. |
| 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" { |
| ; CHECK-COMMON-LABEL: name: many_args_inout_za_call |
| ; CHECK-COMMON: bb.0 (%ir-block.10): |
| ; CHECK-COMMON-NEXT: liveins: $x0, $x1, $x2, $x3, $x4, $x5, $x6, $x7 |
| ; CHECK-COMMON-NEXT: {{ $}} |
| ; CHECK-COMMON-NEXT: [[COPY:%[0-9]+]]:gpr64 = COPY $x7 |
| ; CHECK-COMMON-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY $x6 |
| ; CHECK-COMMON-NEXT: [[COPY2:%[0-9]+]]:gpr64 = COPY $x5 |
| ; CHECK-COMMON-NEXT: [[COPY3:%[0-9]+]]:gpr64 = COPY $x4 |
| ; CHECK-COMMON-NEXT: [[COPY4:%[0-9]+]]:gpr64 = COPY $x3 |
| ; CHECK-COMMON-NEXT: [[COPY5:%[0-9]+]]:gpr64 = COPY $x2 |
| ; CHECK-COMMON-NEXT: [[COPY6:%[0-9]+]]:gpr64 = COPY $x1 |
| ; CHECK-COMMON-NEXT: [[COPY7:%[0-9]+]]:gpr64 = COPY $x0 |
| ; CHECK-COMMON-NEXT: [[LDRXui:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.1, 0 :: (load (s64) from %fixed-stack.1, align 16) |
| ; CHECK-COMMON-NEXT: [[LDRXui1:%[0-9]+]]:gpr64 = LDRXui %fixed-stack.0, 0 :: (load (s64) from %fixed-stack.0) |
| ; CHECK-COMMON-NEXT: ADJCALLSTACKDOWN 16, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-COMMON-NEXT: InOutZAUsePseudo |
| ; CHECK-COMMON-NEXT: [[COPY8:%[0-9]+]]:gpr64sp = COPY $sp |
| ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui1]], [[COPY8]], 1 :: (store (s64) into stack + 8) |
| ; CHECK-COMMON-NEXT: STRXui killed [[LDRXui]], [[COPY8]], 0 :: (store (s64) into stack) |
| ; CHECK-COMMON-NEXT: $x0 = COPY [[COPY7]] |
| ; CHECK-COMMON-NEXT: $x1 = COPY [[COPY6]] |
| ; CHECK-COMMON-NEXT: $x2 = COPY [[COPY5]] |
| ; CHECK-COMMON-NEXT: $x3 = COPY [[COPY4]] |
| ; CHECK-COMMON-NEXT: $x4 = COPY [[COPY3]] |
| ; CHECK-COMMON-NEXT: $x5 = COPY [[COPY2]] |
| ; CHECK-COMMON-NEXT: $x6 = COPY [[COPY1]] |
| ; CHECK-COMMON-NEXT: $x7 = COPY [[COPY]] |
| ; 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 |
| ; CHECK-COMMON-NEXT: ADJCALLSTACKUP 16, 0, implicit-def dead $sp, implicit $sp |
| ; CHECK-COMMON-NEXT: [[COPY9:%[0-9]+]]:gpr64all = COPY $x0 |
| ; CHECK-COMMON-NEXT: RET_ReallyLR |
| %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) |
| ret void |
| } |