; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \ ; RUN: -disable-auto-paired-vec-st=false -ppc-asm-full-reg-names \ ; RUN: -ppc-vsr-nums-as-vr -mcpu=future < %s | FileCheck %s ; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-aix \ ; RUN: -disable-auto-paired-vec-st=false -ppc-asm-full-reg-names \ ; RUN: -ppc-vsr-nums-as-vr -mcpu=future < %s | FileCheck %s --check-prefix=AIX ; RUN: llc -verify-machineinstrs -mtriple=powerpc-unknown-aix \ ; RUN: -disable-auto-paired-vec-st=false -ppc-asm-full-reg-names \ ; RUN: -ppc-vsr-nums-as-vr -mcpu=future < %s | FileCheck %s --check-prefix=AIX32 declare void @dummy_func() declare <2048 x i1> @llvm.ppc.mma.dmsha3hash(<2048 x i1>, i32) define dso_local void @test_dmsha3hash(ptr %vopp, ptr %resp) nounwind { ; CHECK-LABEL: test_dmsha3hash: ; CHECK: # %bb.0: # %entry ; CHECK-NEXT: mflr r0 ; CHECK-NEXT: std r0, 16(r1) ; CHECK-NEXT: stdu r1, -304(r1) ; CHECK-NEXT: std r30, 288(r1) # 8-byte Folded Spill ; CHECK-NEXT: lxvp vsp34, 0(r3) ; CHECK-NEXT: lxvp vsp36, 32(r3) ; CHECK-NEXT: mr r30, r4 ; CHECK-NEXT: dmxxinstdmr512 wacc_hi1, vsp36, vsp34, 1 ; CHECK-NEXT: lxvp vsp34, 64(r3) ; CHECK-NEXT: lxvp vsp36, 96(r3) ; CHECK-NEXT: dmxxinstdmr512 wacc1, vsp36, vsp34, 0 ; CHECK-NEXT: lxvp vsp34, 128(r3) ; CHECK-NEXT: lxvp vsp36, 160(r3) ; CHECK-NEXT: dmxxinstdmr512 wacc_hi0, vsp36, vsp34, 1 ; CHECK-NEXT: lxvp vsp34, 192(r3) ; CHECK-NEXT: lxvp vsp36, 224(r3) ; CHECK-NEXT: dmxxinstdmr512 wacc0, vsp36, vsp34, 0 ; CHECK-NEXT: dmsha3hash dmrp0, 5 ; CHECK-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc1, 0 ; CHECK-NEXT: stxvp vsp36, 128(r1) ; CHECK-NEXT: stxvp vsp34, 96(r1) ; CHECK-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi1, 1 ; CHECK-NEXT: stxvp vsp36, 64(r1) ; CHECK-NEXT: stxvp vsp34, 32(r1) ; CHECK-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc0, 0 ; CHECK-NEXT: stxvp vsp36, 256(r1) ; CHECK-NEXT: stxvp vsp34, 224(r1) ; CHECK-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi0, 1 ; CHECK-NEXT: stxvp vsp36, 192(r1) ; CHECK-NEXT: stxvp vsp34, 160(r1) ; CHECK-NEXT: bl dummy_func@notoc ; CHECK-NEXT: lxvp vsp34, 128(r1) ; CHECK-NEXT: lxvp vsp36, 96(r1) ; CHECK-NEXT: dmxxinstdmr512 wacc1, vsp34, vsp36, 0 ; CHECK-NEXT: lxvp vsp34, 64(r1) ; CHECK-NEXT: lxvp vsp36, 32(r1) ; CHECK-NEXT: dmxxinstdmr512 wacc_hi1, vsp34, vsp36, 1 ; CHECK-NEXT: lxvp vsp34, 256(r1) ; CHECK-NEXT: lxvp vsp36, 224(r1) ; CHECK-NEXT: dmxxinstdmr512 wacc0, vsp34, vsp36, 0 ; CHECK-NEXT: lxvp vsp34, 192(r1) ; CHECK-NEXT: lxvp vsp36, 160(r1) ; CHECK-NEXT: dmxxinstdmr512 wacc_hi0, vsp34, vsp36, 1 ; CHECK-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc0, 0 ; CHECK-NEXT: stxvp vsp34, 224(r30) ; CHECK-NEXT: stxvp vsp36, 192(r30) ; CHECK-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi0, 1 ; CHECK-NEXT: stxvp vsp34, 160(r30) ; CHECK-NEXT: stxvp vsp36, 128(r30) ; CHECK-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc1, 0 ; CHECK-NEXT: stxvp vsp34, 96(r30) ; CHECK-NEXT: stxvp vsp36, 64(r30) ; CHECK-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi1, 1 ; CHECK-NEXT: stxvp vsp34, 32(r30) ; CHECK-NEXT: stxvp vsp36, 0(r30) ; CHECK-NEXT: ld r30, 288(r1) # 8-byte Folded Reload ; CHECK-NEXT: addi r1, r1, 304 ; CHECK-NEXT: ld r0, 16(r1) ; CHECK-NEXT: mtlr r0 ; CHECK-NEXT: blr ; ; AIX-LABEL: test_dmsha3hash: ; AIX: # %bb.0: # %entry ; AIX-NEXT: mflr r0 ; AIX-NEXT: std r0, 16(r1) ; AIX-NEXT: stdu r1, -384(r1) ; AIX-NEXT: std r31, 376(r1) # 8-byte Folded Spill ; AIX-NEXT: lxvp vsp34, 224(r3) ; AIX-NEXT: lxvp vsp36, 192(r3) ; AIX-NEXT: mr r31, r4 ; AIX-NEXT: dmxxinstdmr512 wacc_hi1, vsp36, vsp34, 1 ; AIX-NEXT: lxvp vsp34, 160(r3) ; AIX-NEXT: lxvp vsp36, 128(r3) ; AIX-NEXT: dmxxinstdmr512 wacc1, vsp36, vsp34, 0 ; AIX-NEXT: lxvp vsp34, 96(r3) ; AIX-NEXT: lxvp vsp36, 64(r3) ; AIX-NEXT: dmxxinstdmr512 wacc_hi0, vsp36, vsp34, 1 ; AIX-NEXT: lxvp vsp34, 32(r3) ; AIX-NEXT: lxvp vsp36, 0(r3) ; AIX-NEXT: dmxxinstdmr512 wacc0, vsp36, vsp34, 0 ; AIX-NEXT: dmsha3hash dmrp0, 5 ; AIX-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc1, 0 ; AIX-NEXT: stxvp vsp36, 112(r1) ; AIX-NEXT: stxvp vsp34, 144(r1) ; AIX-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi1, 1 ; AIX-NEXT: stxvp vsp36, 176(r1) ; AIX-NEXT: stxvp vsp34, 208(r1) ; AIX-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc0, 0 ; AIX-NEXT: stxvp vsp36, 240(r1) ; AIX-NEXT: stxvp vsp34, 272(r1) ; AIX-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi0, 1 ; AIX-NEXT: stxvp vsp36, 304(r1) ; AIX-NEXT: stxvp vsp34, 336(r1) ; AIX-NEXT: bl .dummy_func[PR] ; AIX-NEXT: nop ; AIX-NEXT: lxvp vsp34, 112(r1) ; AIX-NEXT: lxvp vsp36, 144(r1) ; AIX-NEXT: dmxxinstdmr512 wacc1, vsp34, vsp36, 0 ; AIX-NEXT: lxvp vsp34, 176(r1) ; AIX-NEXT: lxvp vsp36, 208(r1) ; AIX-NEXT: dmxxinstdmr512 wacc_hi1, vsp34, vsp36, 1 ; AIX-NEXT: lxvp vsp34, 240(r1) ; AIX-NEXT: lxvp vsp36, 272(r1) ; AIX-NEXT: dmxxinstdmr512 wacc0, vsp34, vsp36, 0 ; AIX-NEXT: lxvp vsp34, 304(r1) ; AIX-NEXT: lxvp vsp36, 336(r1) ; AIX-NEXT: dmxxinstdmr512 wacc_hi0, vsp34, vsp36, 1 ; AIX-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi1, 1 ; AIX-NEXT: stxvp vsp36, 224(r31) ; AIX-NEXT: stxvp vsp34, 192(r31) ; AIX-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc1, 0 ; AIX-NEXT: stxvp vsp36, 160(r31) ; AIX-NEXT: stxvp vsp34, 128(r31) ; AIX-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi0, 1 ; AIX-NEXT: stxvp vsp36, 96(r31) ; AIX-NEXT: stxvp vsp34, 64(r31) ; AIX-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc0, 0 ; AIX-NEXT: stxvp vsp36, 32(r31) ; AIX-NEXT: stxvp vsp34, 0(r31) ; AIX-NEXT: ld r31, 376(r1) # 8-byte Folded Reload ; AIX-NEXT: addi r1, r1, 384 ; AIX-NEXT: ld r0, 16(r1) ; AIX-NEXT: mtlr r0 ; AIX-NEXT: blr ; ; AIX32-LABEL: test_dmsha3hash: ; AIX32: # %bb.0: # %entry ; AIX32-NEXT: mflr r0 ; AIX32-NEXT: stw r0, 8(r1) ; AIX32-NEXT: stwu r1, -336(r1) ; AIX32-NEXT: stw r31, 332(r1) # 4-byte Folded Spill ; AIX32-NEXT: lxvp vsp34, 224(r3) ; AIX32-NEXT: lxvp vsp36, 192(r3) ; AIX32-NEXT: mr r31, r4 ; AIX32-NEXT: dmxxinstdmr512 wacc_hi1, vsp36, vsp34, 1 ; AIX32-NEXT: lxvp vsp34, 160(r3) ; AIX32-NEXT: lxvp vsp36, 128(r3) ; AIX32-NEXT: dmxxinstdmr512 wacc1, vsp36, vsp34, 0 ; AIX32-NEXT: lxvp vsp34, 96(r3) ; AIX32-NEXT: lxvp vsp36, 64(r3) ; AIX32-NEXT: dmxxinstdmr512 wacc_hi0, vsp36, vsp34, 1 ; AIX32-NEXT: lxvp vsp34, 32(r3) ; AIX32-NEXT: lxvp vsp36, 0(r3) ; AIX32-NEXT: dmxxinstdmr512 wacc0, vsp36, vsp34, 0 ; AIX32-NEXT: dmsha3hash dmrp0, 5 ; AIX32-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc1, 0 ; AIX32-NEXT: stxvp vsp36, 64(r1) ; AIX32-NEXT: stxvp vsp34, 96(r1) ; AIX32-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi1, 1 ; AIX32-NEXT: stxvp vsp36, 128(r1) ; AIX32-NEXT: stxvp vsp34, 160(r1) ; AIX32-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc0, 0 ; AIX32-NEXT: stxvp vsp36, 192(r1) ; AIX32-NEXT: stxvp vsp34, 224(r1) ; AIX32-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc_hi0, 1 ; AIX32-NEXT: stxvp vsp36, 256(r1) ; AIX32-NEXT: stxvp vsp34, 288(r1) ; AIX32-NEXT: bl .dummy_func[PR] ; AIX32-NEXT: nop ; AIX32-NEXT: lxvp vsp34, 64(r1) ; AIX32-NEXT: lxvp vsp36, 96(r1) ; AIX32-NEXT: dmxxinstdmr512 wacc1, vsp34, vsp36, 0 ; AIX32-NEXT: lxvp vsp34, 128(r1) ; AIX32-NEXT: lxvp vsp36, 160(r1) ; AIX32-NEXT: dmxxinstdmr512 wacc_hi1, vsp34, vsp36, 1 ; AIX32-NEXT: lxvp vsp34, 192(r1) ; AIX32-NEXT: lxvp vsp36, 224(r1) ; AIX32-NEXT: dmxxinstdmr512 wacc0, vsp34, vsp36, 0 ; AIX32-NEXT: lxvp vsp34, 256(r1) ; AIX32-NEXT: lxvp vsp36, 288(r1) ; AIX32-NEXT: dmxxinstdmr512 wacc_hi0, vsp34, vsp36, 1 ; AIX32-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi1, 1 ; AIX32-NEXT: stxvp vsp36, 224(r31) ; AIX32-NEXT: stxvp vsp34, 192(r31) ; AIX32-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc1, 0 ; AIX32-NEXT: stxvp vsp36, 160(r31) ; AIX32-NEXT: stxvp vsp34, 128(r31) ; AIX32-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc_hi0, 1 ; AIX32-NEXT: stxvp vsp36, 96(r31) ; AIX32-NEXT: stxvp vsp34, 64(r31) ; AIX32-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc0, 0 ; AIX32-NEXT: stxvp vsp36, 32(r31) ; AIX32-NEXT: stxvp vsp34, 0(r31) ; AIX32-NEXT: lwz r31, 332(r1) # 4-byte Folded Reload ; AIX32-NEXT: addi r1, r1, 336 ; AIX32-NEXT: lwz r0, 8(r1) ; AIX32-NEXT: mtlr r0 ; AIX32-NEXT: blr entry: %0 = load <2048 x i1>, ptr %vopp, align 64 %2 = tail call <2048 x i1> @llvm.ppc.mma.dmsha3hash(<2048 x i1> %0, i32 5) tail call void @dummy_func() %3 = tail call <2048 x i1> @llvm.ppc.mma.dmsha3hash(<2048 x i1> %0, i32 5) store <2048 x i1> %2, ptr %resp, align 64 ret void }