115 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; This test is a copy of mma-acc-spill.ll except that it uses mcpu=future.3; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \4; RUN: -mcpu=future -ppc-asm-full-reg-names \5; RUN: -ppc-vsr-nums-as-vr < %s | FileCheck %s6; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \7; RUN: -mcpu=future -ppc-asm-full-reg-names \8; RUN: -ppc-vsr-nums-as-vr < %s | FileCheck %s --check-prefix=CHECK-BE9 10declare <512 x i1> @llvm.ppc.mma.xvf16ger2pp(<512 x i1>, <16 x i8>, <16 x i8>)11declare <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>)12declare void @foo()13define void @intrinsics1(<16 x i8> %vc1, <16 x i8> %vc2, <16 x i8> %vc3, <16 x i8> %vc4, ptr %ptr) {14; CHECK-LABEL: intrinsics1:15; CHECK: # %bb.0:16; CHECK-NEXT: mflr r017; CHECK-NEXT: std r0, 16(r1)18; CHECK-NEXT: stdu r1, -176(r1)19; CHECK-NEXT: .cfi_def_cfa_offset 17620; CHECK-NEXT: .cfi_offset lr, 1621; CHECK-NEXT: .cfi_offset r30, -1622; CHECK-NEXT: .cfi_offset v28, -8023; CHECK-NEXT: .cfi_offset v29, -6424; CHECK-NEXT: .cfi_offset v30, -4825; CHECK-NEXT: .cfi_offset v31, -3226; CHECK-NEXT: stxv v28, 96(r1) # 16-byte Folded Spill27; CHECK-NEXT: stxv v29, 112(r1) # 16-byte Folded Spill28; CHECK-NEXT: stxv v30, 128(r1) # 16-byte Folded Spill29; CHECK-NEXT: stxv v31, 144(r1) # 16-byte Folded Spill30; CHECK-NEXT: vmr v31, v531; CHECK-NEXT: vmr v29, v332; CHECK-NEXT: vmr v30, v433; CHECK-NEXT: vmr v28, v234; CHECK-NEXT: std r30, 160(r1) # 8-byte Folded Spill35; CHECK-NEXT: ld r30, 272(r1)36; CHECK-NEXT: dmxxinstdmr512 wacc0, vsp60, vsp62, 037; CHECK-NEXT: xvf16ger2pp wacc0, v2, v438; CHECK-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc0, 039; CHECK-NEXT: stxvp vsp36, 64(r1)40; CHECK-NEXT: stxvp vsp34, 32(r1)41; CHECK-NEXT: bl foo@notoc42; CHECK-NEXT: lxvp vsp34, 64(r1)43; CHECK-NEXT: lxvp vsp36, 32(r1)44; CHECK-NEXT: dmxxinstdmr512 wacc0, vsp34, vsp36, 045; CHECK-NEXT: xvf16ger2pp wacc0, v28, v3046; CHECK-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc0, 047; CHECK-NEXT: stxv v4, 48(r30)48; CHECK-NEXT: stxv v5, 32(r30)49; CHECK-NEXT: stxv v2, 16(r30)50; CHECK-NEXT: stxv v3, 0(r30)51; CHECK-NEXT: lxv v31, 144(r1) # 16-byte Folded Reload52; CHECK-NEXT: lxv v30, 128(r1) # 16-byte Folded Reload53; CHECK-NEXT: lxv v29, 112(r1) # 16-byte Folded Reload54; CHECK-NEXT: lxv v28, 96(r1) # 16-byte Folded Reload55; CHECK-NEXT: ld r30, 160(r1) # 8-byte Folded Reload56; CHECK-NEXT: addi r1, r1, 17657; CHECK-NEXT: ld r0, 16(r1)58; CHECK-NEXT: mtlr r059; CHECK-NEXT: blr60;61; CHECK-BE-LABEL: intrinsics1:62; CHECK-BE: # %bb.0:63; CHECK-BE-NEXT: mflr r064; CHECK-BE-NEXT: std r0, 16(r1)65; CHECK-BE-NEXT: stdu r1, -256(r1)66; CHECK-BE-NEXT: .cfi_def_cfa_offset 25667; CHECK-BE-NEXT: .cfi_offset lr, 1668; CHECK-BE-NEXT: .cfi_offset r30, -1669; CHECK-BE-NEXT: .cfi_offset v28, -8070; CHECK-BE-NEXT: .cfi_offset v29, -6471; CHECK-BE-NEXT: .cfi_offset v30, -4872; CHECK-BE-NEXT: .cfi_offset v31, -3273; CHECK-BE-NEXT: stxv v28, 176(r1) # 16-byte Folded Spill74; CHECK-BE-NEXT: stxv v29, 192(r1) # 16-byte Folded Spill75; CHECK-BE-NEXT: stxv v30, 208(r1) # 16-byte Folded Spill76; CHECK-BE-NEXT: stxv v31, 224(r1) # 16-byte Folded Spill77; CHECK-BE-NEXT: vmr v31, v578; CHECK-BE-NEXT: vmr v29, v379; CHECK-BE-NEXT: vmr v30, v480; CHECK-BE-NEXT: vmr v28, v281; CHECK-BE-NEXT: std r30, 240(r1) # 8-byte Folded Spill82; CHECK-BE-NEXT: ld r30, 368(r1)83; CHECK-BE-NEXT: dmxxinstdmr512 wacc0, vsp60, vsp62, 084; CHECK-BE-NEXT: xvf16ger2pp wacc0, v2, v485; CHECK-BE-NEXT: dmxxextfdmr512 vsp36, vsp34, wacc0, 086; CHECK-BE-NEXT: stxvp vsp36, 112(r1)87; CHECK-BE-NEXT: stxvp vsp34, 144(r1)88; CHECK-BE-NEXT: bl foo89; CHECK-BE-NEXT: nop90; CHECK-BE-NEXT: lxvp vsp34, 112(r1)91; CHECK-BE-NEXT: lxvp vsp36, 144(r1)92; CHECK-BE-NEXT: dmxxinstdmr512 wacc0, vsp34, vsp36, 093; CHECK-BE-NEXT: xvf16ger2pp wacc0, v28, v3094; CHECK-BE-NEXT: dmxxextfdmr512 vsp34, vsp36, wacc0, 095; CHECK-BE-NEXT: stxv v5, 48(r30)96; CHECK-BE-NEXT: stxv v4, 32(r30)97; CHECK-BE-NEXT: stxv v3, 16(r30)98; CHECK-BE-NEXT: stxv v2, 0(r30)99; CHECK-BE-NEXT: lxv v31, 224(r1) # 16-byte Folded Reload100; CHECK-BE-NEXT: lxv v30, 208(r1) # 16-byte Folded Reload101; CHECK-BE-NEXT: lxv v29, 192(r1) # 16-byte Folded Reload102; CHECK-BE-NEXT: lxv v28, 176(r1) # 16-byte Folded Reload103; CHECK-BE-NEXT: ld r30, 240(r1) # 8-byte Folded Reload104; CHECK-BE-NEXT: addi r1, r1, 256105; CHECK-BE-NEXT: ld r0, 16(r1)106; CHECK-BE-NEXT: mtlr r0107; CHECK-BE-NEXT: blr108 %1 = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> %vc1, <16 x i8> %vc2, <16 x i8> %vc3, <16 x i8> %vc4)109 %2 = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2pp(<512 x i1> %1, <16 x i8> %vc1, <16 x i8> %vc3)110 tail call void @foo()111 %3 = tail call <512 x i1> @llvm.ppc.mma.xvf16ger2pp(<512 x i1> %2, <16 x i8> %vc1, <16 x i8> %vc3)112 store <512 x i1> %3, ptr %ptr, align 64113 ret void114}115