163 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 42; RUN: llc -force-streaming -verify-machineinstrs < %s | FileCheck %s3 4target triple = "aarch64-linux"5 6define void @stmopa_za32_s8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {7; CHECK-LABEL: stmopa_za32_s8:8; CHECK: // %bb.0:9; CHECK-NEXT: mov z28.d, z3.d10; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z111; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z112; CHECK-NEXT: stmopa za0.s, { z0.b, z1.b }, z2.b, z28[0]13; CHECK-NEXT: ret14 call void @llvm.aarch64.sme.stmopa.za32.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)15 ret void16}17 18define void @utmopa_za32_u8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {19; CHECK-LABEL: utmopa_za32_u8:20; CHECK: // %bb.0:21; CHECK-NEXT: mov z28.d, z3.d22; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z123; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z124; CHECK-NEXT: utmopa za0.s, { z0.b, z1.b }, z2.b, z28[0]25; CHECK-NEXT: ret26 call void @llvm.aarch64.sme.utmopa.za32.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)27 ret void28}29 30define void @ustmopa_za32_u8_s8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {31; CHECK-LABEL: ustmopa_za32_u8_s8:32; CHECK: // %bb.0:33; CHECK-NEXT: mov z28.d, z3.d34; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z135; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z136; CHECK-NEXT: ustmopa za0.s, { z0.b, z1.b }, z2.b, z28[0]37; CHECK-NEXT: ret38 call void @llvm.aarch64.sme.ustmopa.za32.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)39 ret void40}41 42define void @sutmopa_za32_s8_u8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {43; CHECK-LABEL: sutmopa_za32_s8_u8:44; CHECK: // %bb.0:45; CHECK-NEXT: mov z28.d, z3.d46; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z147; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z148; CHECK-NEXT: sutmopa za0.s, { z0.b, z1.b }, z2.b, z28[0]49; CHECK-NEXT: ret50 call void @llvm.aarch64.sme.sutmopa.za32.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)51 ret void52}53 54define void @stmopa_za32_s16(<vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zm, <vscale x 16 x i8> %zk) #0 {55; CHECK-LABEL: stmopa_za32_s16:56; CHECK: // %bb.0:57; CHECK-NEXT: mov z28.d, z3.d58; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z159; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z160; CHECK-NEXT: stmopa za0.s, { z0.h, z1.h }, z2.h, z28[0]61; CHECK-NEXT: ret62 call void @llvm.aarch64.sme.stmopa.za32.nxv8i16(i32 0, <vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zm, <vscale x 16 x i8> %zk, i32 0)63 ret void64}65 66define void @utmopa_za32_u16(<vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zm, <vscale x 16 x i8> %zk) #0 {67; CHECK-LABEL: utmopa_za32_u16:68; CHECK: // %bb.0:69; CHECK-NEXT: mov z28.d, z3.d70; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z171; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z172; CHECK-NEXT: utmopa za0.s, { z0.h, z1.h }, z2.h, z28[0]73; CHECK-NEXT: ret74 call void @llvm.aarch64.sme.utmopa.za32.nxv8i16(i32 0, <vscale x 8 x i16> %zn1, <vscale x 8 x i16> %zn2, <vscale x 8 x i16> %zm, <vscale x 16 x i8> %zk, i32 0)75 ret void76}77 78define void @ftmopa_za32_f16(<vscale x 8 x half> %zn1, <vscale x 8 x half> %zn2, <vscale x 8 x half> %zm, <vscale x 16 x i8> %zk) #0 {79; CHECK-LABEL: ftmopa_za32_f16:80; CHECK: // %bb.0:81; CHECK-NEXT: mov z28.d, z3.d82; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z183; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z184; CHECK-NEXT: ftmopa za0.s, { z0.h, z1.h }, z2.h, z28[0]85; CHECK-NEXT: ret86 call void @llvm.aarch64.sme.ftmopa.za32.nxv8f16(i32 0, <vscale x 8 x half> %zn1, <vscale x 8 x half> %zn2, <vscale x 8 x half> %zm, <vscale x 16 x i8> %zk, i32 0)87 ret void88}89 90define void @bftmopa_za32_bf16(<vscale x 8 x bfloat> %zn1, <vscale x 8 x bfloat> %zn2, <vscale x 8 x bfloat> %zm, <vscale x 16 x i8> %zk) #0 {91; CHECK-LABEL: bftmopa_za32_bf16:92; CHECK: // %bb.0:93; CHECK-NEXT: mov z28.d, z3.d94; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z195; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z196; CHECK-NEXT: bftmopa za0.s, { z0.h, z1.h }, z2.h, z28[0]97; CHECK-NEXT: ret98 call void @llvm.aarch64.sme.ftmopa.za32.nxv8bf16(i32 0, <vscale x 8 x bfloat> %zn1, <vscale x 8 x bfloat> %zn2, <vscale x 8 x bfloat> %zm, <vscale x 16 x i8> %zk, i32 0)99 ret void100}101 102define void @ftmopa_za32_f32(<vscale x 4 x float> %zn1, <vscale x 4 x float> %zn2, <vscale x 4 x float> %zm, <vscale x 16 x i8> %zk) #0 {103; CHECK-LABEL: ftmopa_za32_f32:104; CHECK: // %bb.0:105; CHECK-NEXT: mov z28.d, z3.d106; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1107; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1108; CHECK-NEXT: ftmopa za0.s, { z0.s, z1.s }, z2.s, z28[0]109; CHECK-NEXT: ret110 call void @llvm.aarch64.sme.ftmopa.za32.nxv4f32(i32 0, <vscale x 4 x float> %zn1, <vscale x 4 x float> %zn2, <vscale x 4 x float> %zm, <vscale x 16 x i8> %zk, i32 0)111 ret void112}113 114define void @ftmopa_za16_f16(<vscale x 8 x half> %zn1, <vscale x 8 x half> %zn2, <vscale x 8 x half> %zm, <vscale x 16 x i8> %zk) #0 {115; CHECK-LABEL: ftmopa_za16_f16:116; CHECK: // %bb.0:117; CHECK-NEXT: mov z28.d, z3.d118; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1119; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1120; CHECK-NEXT: ftmopa za0.h, { z0.h, z1.h }, z2.h, z28[0]121; CHECK-NEXT: ret122 call void @llvm.aarch64.sme.ftmopa.za16.nxv8f16(i32 0, <vscale x 8 x half> %zn1, <vscale x 8 x half> %zn2, <vscale x 8 x half> %zm, <vscale x 16 x i8> %zk, i32 0)123 ret void124}125 126define void @bftmopa_za16_bf16(<vscale x 8 x bfloat> %zn1, <vscale x 8 x bfloat> %zn2, <vscale x 8 x bfloat> %zm, <vscale x 16 x i8> %zk) #0 {127; CHECK-LABEL: bftmopa_za16_bf16:128; CHECK: // %bb.0:129; CHECK-NEXT: mov z28.d, z3.d130; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1131; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1132; CHECK-NEXT: bftmopa za0.h, { z0.h, z1.h }, z2.h, z28[0]133; CHECK-NEXT: ret134 call void @llvm.aarch64.sme.ftmopa.za16.nxv8bf16(i32 0, <vscale x 8 x bfloat> %zn1, <vscale x 8 x bfloat> %zn2, <vscale x 8 x bfloat> %zm, <vscale x 16 x i8> %zk, i32 0)135 ret void136}137 138define void @ftmopa_za16_f8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {139; CHECK-LABEL: ftmopa_za16_f8:140; CHECK: // %bb.0:141; CHECK-NEXT: mov z28.d, z3.d142; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1143; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1144; CHECK-NEXT: ftmopa za0.h, { z0.b, z1.b }, z2.b, z28[0]145; CHECK-NEXT: ret146 call void @llvm.aarch64.sme.ftmopa.za16.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)147 ret void148}149 150define void @ftmopa_za32_f8(<vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk) #0 {151; CHECK-LABEL: ftmopa_za32_f8:152; CHECK: // %bb.0:153; CHECK-NEXT: mov z28.d, z3.d154; CHECK-NEXT: // kill: def $z1 killed $z1 killed $z0_z1 def $z0_z1155; CHECK-NEXT: // kill: def $z0 killed $z0 killed $z0_z1 def $z0_z1156; CHECK-NEXT: ftmopa za0.s, { z0.b, z1.b }, z2.b, z28[0]157; CHECK-NEXT: ret158 call void @llvm.aarch64.sme.ftmopa.za32.nxv16i8(i32 0, <vscale x 16 x i8> %zn1, <vscale x 16 x i8> %zn2, <vscale x 16 x i8> %zm, <vscale x 16 x i8> %zk, i32 0)159 ret void160}161 162attributes #0 = {nounwind "target-features" = "+sme2,+sme-tmop,+sme-f16f16,+sme-b16b16,+sme-f8f16,+sme-f8f32,+sme2p1,+bf16" }163