546 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 22; RUN: opt < %s -passes=interleaved-access -S | FileCheck %s3 4target triple = "aarch64-linux-gnu"5 6define void @deinterleave_nxi8_factor2(ptr %ptr) #0 {7; CHECK-LABEL: define void @deinterleave_nxi8_factor28; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0:[0-9]+]] {9; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.aarch64.sve.ld2.sret.nxv16i8(<vscale x 16 x i1> splat (i1 true), ptr [[PTR]])10; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[LDN]], 111; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } [[LDN]], 012; CHECK-NEXT: ret void13;14 %load = load <vscale x 32 x i8>, ptr %ptr, align 115 %deinterleave = tail call { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.vector.deinterleave2.nxv32i8(<vscale x 32 x i8> %load)16 %extract1 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %deinterleave, 117 %extract2 = extractvalue { <vscale x 16 x i8>, <vscale x 16 x i8> } %deinterleave, 018 ret void19}20 21define void @deinterleave_nxi16_factor2(ptr %ptr) #0 {22; CHECK-LABEL: define void @deinterleave_nxi16_factor223; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {24; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.aarch64.sve.ld2.sret.nxv8i16(<vscale x 8 x i1> splat (i1 true), ptr [[PTR]])25; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[LDN]], 026; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } [[LDN]], 127; CHECK-NEXT: ret void28;29 %load = load <vscale x 16 x i16>, ptr %ptr, align 230 %deinterleave = tail call { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.vector.deinterleave2.nxv16i16(<vscale x 16 x i16> %load)31 %extract1 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %deinterleave, 032 %extract2 = extractvalue { <vscale x 8 x i16>, <vscale x 8 x i16> } %deinterleave, 133 ret void34}35 36define void @deinterleave_nx8xi32_factor2(ptr %ptr) #0 {37; CHECK-LABEL: define void @deinterleave_nx8xi32_factor238; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {39; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> splat (i1 true), ptr [[PTR]])40; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 041; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 142; CHECK-NEXT: ret void43;44 %load = load <vscale x 8 x i32>, ptr %ptr, align 445 %deinterleave = tail call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32> %load)46 %extract1 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave, 047 %extract2 = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } %deinterleave, 148 ret void49}50 51define void @deinterleave_nxi64_factor2(ptr %ptr) #0 {52; CHECK-LABEL: define void @deinterleave_nxi64_factor253; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {54; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld2.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[PTR]])55; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN]], 056; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN]], 157; CHECK-NEXT: ret void58;59 %load = load <vscale x 4 x i64>, ptr %ptr, align 860 %deinterleave = tail call { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64> %load)61 %extract1 = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %deinterleave, 062 %extract2 = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64> } %deinterleave, 163 ret void64}65 66define void @deinterleave_nxfloat_factor2(ptr %ptr) #0 {67; CHECK-LABEL: define void @deinterleave_nxfloat_factor268; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {69; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.aarch64.sve.ld2.sret.nxv4f32(<vscale x 4 x i1> splat (i1 true), ptr [[PTR]])70; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[LDN]], 071; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } [[LDN]], 172; CHECK-NEXT: ret void73;74 %load = load <vscale x 8 x float>, ptr %ptr, align 475 %deinterleave = tail call { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.vector.deinterleave2.nxv8f32(<vscale x 8 x float> %load)76 %extract1 = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } %deinterleave, 077 %extract2 = extractvalue { <vscale x 4 x float>, <vscale x 4 x float> } %deinterleave, 178 ret void79}80 81define void @deinterleave_nxdouble_factor2(ptr %ptr) #0 {82; CHECK-LABEL: define void @deinterleave_nxdouble_factor283; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {84; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> splat (i1 true), ptr [[PTR]])85; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN]], 086; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN]], 187; CHECK-NEXT: ret void88;89 %load = load <vscale x 4 x double>, ptr %ptr, align 890 %deinterleave = tail call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double> %load)91 %extract1 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %deinterleave, 092 %extract2 = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } %deinterleave, 193 ret void94}95 96define void @deinterleave_nxptr_factor2(ptr %ptr) #0 {97; CHECK-LABEL: define void @deinterleave_nxptr_factor298; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {99; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 2 x ptr>, <vscale x 2 x ptr> } @llvm.aarch64.sve.ld2.sret.nxv2p0(<vscale x 2 x i1> splat (i1 true), ptr [[PTR]])100; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 2 x ptr>, <vscale x 2 x ptr> } [[LDN]], 0101; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 2 x ptr>, <vscale x 2 x ptr> } [[LDN]], 1102; CHECK-NEXT: ret void103;104 %load = load <vscale x 4 x ptr>, ptr %ptr, align 8105 %deinterleave = tail call { <vscale x 2 x ptr>, <vscale x 2 x ptr> } @llvm.vector.deinterleave2.nxv4p0(<vscale x 4 x ptr> %load)106 %extract1 = extractvalue { <vscale x 2 x ptr>, <vscale x 2 x ptr> } %deinterleave, 0107 %extract2 = extractvalue { <vscale x 2 x ptr>, <vscale x 2 x ptr> } %deinterleave, 1108 ret void109}110 111define void @interleave_nxi8_factor2(ptr %ptr, <vscale x 16 x i8> %l, <vscale x 16 x i8> %r) #0 {112; CHECK-LABEL: define void @interleave_nxi8_factor2113; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 16 x i8> [[L:%.*]], <vscale x 16 x i8> [[R:%.*]]) #[[ATTR0]] {114; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv16i8(<vscale x 16 x i8> [[L]], <vscale x 16 x i8> [[R]], <vscale x 16 x i1> splat (i1 true), ptr [[PTR]])115; CHECK-NEXT: ret void116;117 %interleave = tail call <vscale x 32 x i8> @llvm.vector.interleave2.nxv32i8(<vscale x 16 x i8> %l, <vscale x 16 x i8> %r)118 store <vscale x 32 x i8> %interleave, ptr %ptr, align 1119 ret void120}121 122define void @interleave_nxi16_factor2(ptr %ptr, <vscale x 8 x i16> %l, <vscale x 8 x i16> %r) #0 {123; CHECK-LABEL: define void @interleave_nxi16_factor2124; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 8 x i16> [[L:%.*]], <vscale x 8 x i16> [[R:%.*]]) #[[ATTR0]] {125; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv8i16(<vscale x 8 x i16> [[L]], <vscale x 8 x i16> [[R]], <vscale x 8 x i1> splat (i1 true), ptr [[PTR]])126; CHECK-NEXT: ret void127;128 %interleave = tail call <vscale x 16 x i16> @llvm.vector.interleave2.nxv16i16(<vscale x 8 x i16> %l, <vscale x 8 x i16> %r)129 store <vscale x 16 x i16> %interleave, ptr %ptr, align 2130 ret void131}132 133define void @interleave_nxi32_factor2(ptr %ptr, <vscale x 4 x i32> %l, <vscale x 4 x i32> %r) #0 {134; CHECK-LABEL: define void @interleave_nxi32_factor2135; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 4 x i32> [[L:%.*]], <vscale x 4 x i32> [[R:%.*]]) #[[ATTR0]] {136; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv4i32(<vscale x 4 x i32> [[L]], <vscale x 4 x i32> [[R]], <vscale x 4 x i1> splat (i1 true), ptr [[PTR]])137; CHECK-NEXT: ret void138;139 %interleave = tail call <vscale x 8 x i32> @llvm.vector.interleave2.nxv8i32(<vscale x 4 x i32> %l, <vscale x 4 x i32> %r)140 store <vscale x 8 x i32> %interleave, ptr %ptr, align 4141 ret void142}143 144define void @interleave_nxi64_factor2(ptr %ptr, <vscale x 2 x i64> %l, <vscale x 2 x i64> %r) #0 {145; CHECK-LABEL: define void @interleave_nxi64_factor2146; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 2 x i64> [[L:%.*]], <vscale x 2 x i64> [[R:%.*]]) #[[ATTR0]] {147; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv2i64(<vscale x 2 x i64> [[L]], <vscale x 2 x i64> [[R]], <vscale x 2 x i1> splat (i1 true), ptr [[PTR]])148; CHECK-NEXT: ret void149;150 %interleave = tail call <vscale x 4 x i64> @llvm.vector.interleave2.nxv4i64(<vscale x 2 x i64> %l, <vscale x 2 x i64> %r)151 store <vscale x 4 x i64> %interleave, ptr %ptr, align 8152 ret void153}154 155define void @interleave_nxfloat_factor2(ptr %ptr, <vscale x 4 x float> %l, <vscale x 4 x float> %r) #0 {156; CHECK-LABEL: define void @interleave_nxfloat_factor2157; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 4 x float> [[L:%.*]], <vscale x 4 x float> [[R:%.*]]) #[[ATTR0]] {158; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv4f32(<vscale x 4 x float> [[L]], <vscale x 4 x float> [[R]], <vscale x 4 x i1> splat (i1 true), ptr [[PTR]])159; CHECK-NEXT: ret void160;161 %interleave = tail call <vscale x 8 x float> @llvm.vector.interleave2.nxv8f32(<vscale x 4 x float> %l, <vscale x 4 x float> %r)162 store <vscale x 8 x float> %interleave, ptr %ptr, align 4163 ret void164}165 166define void @interleave_nxdouble_factor2(ptr %ptr, <vscale x 2 x double> %l, <vscale x 2 x double> %r) #0 {167; CHECK-LABEL: define void @interleave_nxdouble_factor2168; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 2 x double> [[L:%.*]], <vscale x 2 x double> [[R:%.*]]) #[[ATTR0]] {169; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv2f64(<vscale x 2 x double> [[L]], <vscale x 2 x double> [[R]], <vscale x 2 x i1> splat (i1 true), ptr [[PTR]])170; CHECK-NEXT: ret void171;172 %interleave = tail call <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double> %l, <vscale x 2 x double> %r)173 store <vscale x 4 x double> %interleave, ptr %ptr, align 4174 ret void175}176 177define void @interleave_nxptr_factor2(ptr %ptr, <vscale x 2 x ptr> %l, <vscale x 2 x ptr> %r) #0 {178; CHECK-LABEL: define void @interleave_nxptr_factor2179; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 2 x ptr> [[L:%.*]], <vscale x 2 x ptr> [[R:%.*]]) #[[ATTR0]] {180; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv2p0(<vscale x 2 x ptr> [[L]], <vscale x 2 x ptr> [[R]], <vscale x 2 x i1> splat (i1 true), ptr [[PTR]])181; CHECK-NEXT: ret void182;183 %interleave = tail call <vscale x 4 x ptr> @llvm.vector.interleave2.nxv4p0(<vscale x 2 x ptr> %l, <vscale x 2 x ptr> %r)184 store <vscale x 4 x ptr> %interleave, ptr %ptr, align 4185 ret void186}187 188define void @interleave_nxi8_factor2_masked_store_splatmask(ptr %ptr, <vscale x 16 x i8> %l, <vscale x 16 x i8> %r, i1 %mask) #0 {189; CHECK-LABEL: define void @interleave_nxi8_factor2_masked_store_splatmask190; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 16 x i8> [[L:%.*]], <vscale x 16 x i8> [[R:%.*]], i1 [[MASK:%.*]]) #[[ATTR0]] {191; CHECK-NEXT: [[INTERLEAVE:%.*]] = tail call <vscale x 32 x i8> @llvm.vector.interleave2.nxv32i8(<vscale x 16 x i8> [[L]], <vscale x 16 x i8> [[R]])192; CHECK-NEXT: [[MASK_INS:%.*]] = insertelement <vscale x 32 x i1> poison, i1 [[MASK]], i64 0193; CHECK-NEXT: [[MASK_SPLAT:%.*]] = shufflevector <vscale x 32 x i1> [[MASK_INS]], <vscale x 32 x i1> poison, <vscale x 32 x i32> zeroinitializer194; CHECK-NEXT: call void @llvm.masked.store.nxv32i8.p0(<vscale x 32 x i8> [[INTERLEAVE]], ptr align 1 [[PTR]], <vscale x 32 x i1> [[MASK_SPLAT]])195; CHECK-NEXT: ret void196;197 %interleave = tail call <vscale x 32 x i8> @llvm.vector.interleave2.nxv32i8(<vscale x 16 x i8> %l, <vscale x 16 x i8> %r)198 %mask.ins = insertelement <vscale x 32 x i1> poison, i1 %mask, i64 0199 %mask.splat = shufflevector <vscale x 32 x i1> %mask.ins, <vscale x 32 x i1> poison, <vscale x 32 x i32> zeroinitializer200 tail call void @llvm.masked.store.nxv32i8.p0(<vscale x 32 x i8> %interleave, ptr %ptr, i32 1, <vscale x 32 x i1> %mask.splat)201 ret void202}203 204;;; Check that we 'legalize' operations that are wider than the target supports.205 206define void @deinterleave_wide_nxi32_factor2(ptr %ptr) #0 {207; CHECK-LABEL: define void @deinterleave_wide_nxi32_factor2208; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {209; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[PTR]], i64 0210; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> splat (i1 true), ptr [[TMP1]])211; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 0212; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP2]], i64 0)213; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN]], 1214; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> poison, <vscale x 4 x i32> [[TMP4]], i64 0)215; CHECK-NEXT: [[TMP6:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[PTR]], i64 2216; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> splat (i1 true), ptr [[TMP6]])217; CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 0218; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP3]], <vscale x 4 x i32> [[TMP7]], i64 4)219; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN1]], 1220; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP5]], <vscale x 4 x i32> [[TMP9]], i64 4)221; CHECK-NEXT: [[TMP11:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[PTR]], i64 4222; CHECK-NEXT: [[LDN2:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> splat (i1 true), ptr [[TMP11]])223; CHECK-NEXT: [[TMP12:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN2]], 0224; CHECK-NEXT: [[TMP13:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP8]], <vscale x 4 x i32> [[TMP12]], i64 8)225; CHECK-NEXT: [[TMP14:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN2]], 1226; CHECK-NEXT: [[TMP15:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP10]], <vscale x 4 x i32> [[TMP14]], i64 8)227; CHECK-NEXT: [[TMP16:%.*]] = getelementptr <vscale x 4 x i32>, ptr [[PTR]], i64 6228; CHECK-NEXT: [[LDN3:%.*]] = call { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.aarch64.sve.ld2.sret.nxv4i32(<vscale x 4 x i1> splat (i1 true), ptr [[TMP16]])229; CHECK-NEXT: [[TMP17:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN3]], 0230; CHECK-NEXT: [[TMP18:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP13]], <vscale x 4 x i32> [[TMP17]], i64 12)231; CHECK-NEXT: [[TMP19:%.*]] = extractvalue { <vscale x 4 x i32>, <vscale x 4 x i32> } [[LDN3]], 1232; CHECK-NEXT: [[TMP20:%.*]] = call <vscale x 16 x i32> @llvm.vector.insert.nxv16i32.nxv4i32(<vscale x 16 x i32> [[TMP15]], <vscale x 4 x i32> [[TMP19]], i64 12)233; CHECK-NEXT: [[TMP21:%.*]] = insertvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } poison, <vscale x 16 x i32> [[TMP18]], 0234; CHECK-NEXT: [[TMP22:%.*]] = insertvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } [[TMP21]], <vscale x 16 x i32> [[TMP20]], 1235; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } [[TMP22]], 0236; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } [[TMP22]], 1237; CHECK-NEXT: ret void238;239 %load = load <vscale x 32 x i32>, ptr %ptr, align 4240 %deinterleave = tail call { <vscale x 16 x i32>, <vscale x 16 x i32> } @llvm.vector.deinterleave2.nxv32i32(<vscale x 32 x i32> %load)241 %extract1 = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } %deinterleave, 0242 %extract2 = extractvalue { <vscale x 16 x i32>, <vscale x 16 x i32> } %deinterleave, 1243 ret void244}245 246define void @deinterleave_wide_nxdouble_factor2(ptr %ptr) #0 {247; CHECK-LABEL: define void @deinterleave_wide_nxdouble_factor2248; CHECK-SAME: (ptr [[PTR:%.*]]) #[[ATTR0]] {249; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[PTR]], i64 0250; CHECK-NEXT: [[LDN:%.*]] = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])251; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN]], 0252; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP2]], i64 0)253; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN]], 1254; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> poison, <vscale x 2 x double> [[TMP4]], i64 0)255; CHECK-NEXT: [[TMP6:%.*]] = getelementptr <vscale x 2 x double>, ptr [[PTR]], i64 2256; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.aarch64.sve.ld2.sret.nxv2f64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP6]])257; CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN1]], 0258; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP3]], <vscale x 2 x double> [[TMP7]], i64 2)259; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x double>, <vscale x 2 x double> } [[LDN1]], 1260; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 4 x double> @llvm.vector.insert.nxv4f64.nxv2f64(<vscale x 4 x double> [[TMP5]], <vscale x 2 x double> [[TMP9]], i64 2)261; CHECK-NEXT: [[TMP11:%.*]] = insertvalue { <vscale x 4 x double>, <vscale x 4 x double> } poison, <vscale x 4 x double> [[TMP8]], 0262; CHECK-NEXT: [[TMP12:%.*]] = insertvalue { <vscale x 4 x double>, <vscale x 4 x double> } [[TMP11]], <vscale x 4 x double> [[TMP10]], 1263; CHECK-NEXT: [[EXTRACT1:%.*]] = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } [[TMP12]], 0264; CHECK-NEXT: [[EXTRACT2:%.*]] = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } [[TMP12]], 1265; CHECK-NEXT: ret void266;267 %load = load <vscale x 8 x double>, ptr %ptr, align 8268 %deinterleave = tail call { <vscale x 4 x double>, <vscale x 4 x double> } @llvm.vector.deinterleave2.nxv8f64(<vscale x 8 x double> %load)269 %extract1 = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %deinterleave, 0270 %extract2 = extractvalue { <vscale x 4 x double>, <vscale x 4 x double> } %deinterleave, 1271 ret void272}273 274define void @interleave_wide_nxdouble_factor2(ptr %ptr, <vscale x 4 x double> %l, <vscale x 4 x double> %r) #0 {275; CHECK-LABEL: define void @interleave_wide_nxdouble_factor2276; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 4 x double> [[L:%.*]], <vscale x 4 x double> [[R:%.*]]) #[[ATTR0]] {277; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x double>, ptr [[PTR]], i64 0278; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[L]], i64 0)279; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[R]], i64 0)280; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv2f64(<vscale x 2 x double> [[TMP2]], <vscale x 2 x double> [[TMP3]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])281; CHECK-NEXT: [[TMP4:%.*]] = getelementptr <vscale x 2 x double>, ptr [[PTR]], i64 2282; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[L]], i64 2)283; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x double> @llvm.vector.extract.nxv2f64.nxv4f64(<vscale x 4 x double> [[R]], i64 2)284; CHECK-NEXT: call void @llvm.aarch64.sve.st2.nxv2f64(<vscale x 2 x double> [[TMP5]], <vscale x 2 x double> [[TMP6]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP4]])285; CHECK-NEXT: ret void286;287 %interleave = tail call <vscale x 8 x double> @llvm.vector.interleave2.nxv8f64(<vscale x 4 x double> %l, <vscale x 4 x double> %r)288 store <vscale x 8 x double> %interleave, ptr %ptr, align 4289 ret void290}291 292define void @deinterleave1_nxi64_factor3(ptr %ptr, <vscale x 4 x i64>* %s1, <vscale x 4 x i64>* %s2, <vscale x 4 x i64>* %s3) #0 {293; CHECK-LABEL: define void @deinterleave1_nxi64_factor3294; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] {295; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0296; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])297; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 0298; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)299; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 1300; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)301; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 2302; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> poison, <vscale x 2 x i64> [[TMP6]], i64 0)303; CHECK-NEXT: [[TMP8:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3304; CHECK-NEXT: [[LDN2:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP8]])305; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 0306; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP9]], i64 2)307; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 1308; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP11]], i64 2)309; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 2310; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 4 x i64> @llvm.vector.insert.nxv4i64.nxv2i64(<vscale x 4 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP13]], i64 2)311; CHECK-NEXT: [[TMP15:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } poison, <vscale x 4 x i64> [[TMP10]], 0312; CHECK-NEXT: [[TMP16:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP15]], <vscale x 4 x i64> [[TMP12]], 1313; CHECK-NEXT: [[TMP17:%.*]] = insertvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP16]], <vscale x 4 x i64> [[TMP14]], 2314; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 0315; CHECK-NEXT: [[TMP19:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 1316; CHECK-NEXT: [[TMP20:%.*]] = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } [[TMP17]], 2317; CHECK-NEXT: store <vscale x 4 x i64> [[TMP18]], ptr [[S1]], align 32318; CHECK-NEXT: store <vscale x 4 x i64> [[TMP19]], ptr [[S2]], align 32319; CHECK-NEXT: store <vscale x 4 x i64> [[TMP20]], ptr [[S3]], align 32320; CHECK-NEXT: ret void321;322 %wide.vec = load <vscale x 12 x i64>, ptr %ptr, align 8323 %ldN = tail call { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave3.nxv12i64(<vscale x 12 x i64> %wide.vec)324 325 %3 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 0326 %4 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 1327 %5 = extractvalue { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } %ldN, 2328 329 store <vscale x 4 x i64> %3, <vscale x 4 x i64>* %s1330 store <vscale x 4 x i64> %4, <vscale x 4 x i64>* %s2331 store <vscale x 4 x i64> %5, <vscale x 4 x i64>* %s3332 ret void333}334 335define void @deinterleave2_nxi64_factor3(ptr %ptr, <vscale x 8 x i64>* %s1, <vscale x 8 x i64>* %s2, <vscale x 8 x i64>* %s3) #0 {336; CHECK-LABEL: define void @deinterleave2_nxi64_factor3337; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] {338; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0339; CHECK-NEXT: [[LDN1:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])340; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 0341; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP2]], i64 0)342; CHECK-NEXT: [[TMP4:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 1343; CHECK-NEXT: [[TMP5:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP4]], i64 0)344; CHECK-NEXT: [[TMP6:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN1]], 2345; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> poison, <vscale x 2 x i64> [[TMP6]], i64 0)346; CHECK-NEXT: [[TMP8:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3347; CHECK-NEXT: [[LDN2:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP8]])348; CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 0349; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP9]], i64 2)350; CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 1351; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP5]], <vscale x 2 x i64> [[TMP11]], i64 2)352; CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN2]], 2353; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP13]], i64 2)354; CHECK-NEXT: [[TMP15:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 6355; CHECK-NEXT: [[LDN3:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP15]])356; CHECK-NEXT: [[TMP16:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 0357; CHECK-NEXT: [[TMP17:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP16]], i64 4)358; CHECK-NEXT: [[TMP18:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 1359; CHECK-NEXT: [[TMP19:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP12]], <vscale x 2 x i64> [[TMP18]], i64 4)360; CHECK-NEXT: [[TMP20:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN3]], 2361; CHECK-NEXT: [[TMP21:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP20]], i64 4)362; CHECK-NEXT: [[TMP22:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 9363; CHECK-NEXT: [[LDN4:%.*]] = call { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.aarch64.sve.ld3.sret.nxv2i64(<vscale x 2 x i1> splat (i1 true), ptr [[TMP22]])364; CHECK-NEXT: [[TMP23:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 0365; CHECK-NEXT: [[TMP24:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP17]], <vscale x 2 x i64> [[TMP23]], i64 6)366; CHECK-NEXT: [[TMP25:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 1367; CHECK-NEXT: [[TMP26:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP19]], <vscale x 2 x i64> [[TMP25]], i64 6)368; CHECK-NEXT: [[TMP27:%.*]] = extractvalue { <vscale x 2 x i64>, <vscale x 2 x i64>, <vscale x 2 x i64> } [[LDN4]], 2369; CHECK-NEXT: [[TMP28:%.*]] = call <vscale x 8 x i64> @llvm.vector.insert.nxv8i64.nxv2i64(<vscale x 8 x i64> [[TMP21]], <vscale x 2 x i64> [[TMP27]], i64 6)370; CHECK-NEXT: [[TMP29:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } poison, <vscale x 8 x i64> [[TMP24]], 0371; CHECK-NEXT: [[TMP30:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP29]], <vscale x 8 x i64> [[TMP26]], 1372; CHECK-NEXT: [[TMP31:%.*]] = insertvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP30]], <vscale x 8 x i64> [[TMP28]], 2373; CHECK-NEXT: [[TMP32:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 0374; CHECK-NEXT: [[TMP33:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 1375; CHECK-NEXT: [[TMP34:%.*]] = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } [[TMP31]], 2376; CHECK-NEXT: store <vscale x 8 x i64> [[TMP32]], ptr [[S1]], align 64377; CHECK-NEXT: store <vscale x 8 x i64> [[TMP33]], ptr [[S2]], align 64378; CHECK-NEXT: store <vscale x 8 x i64> [[TMP34]], ptr [[S3]], align 64379; CHECK-NEXT: ret void380;381 %wide.vec = load <vscale x 24 x i64>, ptr %ptr, align 8382 %ldN = tail call { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } @llvm.vector.deinterleave3.nxv24i64(<vscale x 24 x i64> %wide.vec)383 384 %3 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 0385 %4 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 1386 %5 = extractvalue { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } %ldN, 2387 388 store <vscale x 8 x i64> %3, <vscale x 8 x i64>* %s1389 store <vscale x 8 x i64> %4, <vscale x 8 x i64>* %s2390 store <vscale x 8 x i64> %5, <vscale x 8 x i64>* %s3391 ret void392}393 394define void @deinterleave_neg1_nxi64_factor3(ptr %ptr, <vscale x 1 x i64>* %s1, <vscale x 1 x i64>* %s2, <vscale x 1 x i64>* %s3) #0 {395; CHECK-LABEL: define void @deinterleave_neg1_nxi64_factor3396; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] {397; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 3 x i64>, ptr [[PTR]], align 8398; CHECK-NEXT: [[LDN:%.*]] = tail call { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64> [[WIDE_VEC]])399; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 0400; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 1401; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } [[LDN]], 2402; CHECK-NEXT: store <vscale x 1 x i64> [[TMP1]], ptr [[S1]], align 8403; CHECK-NEXT: store <vscale x 1 x i64> [[TMP2]], ptr [[S2]], align 8404; CHECK-NEXT: store <vscale x 1 x i64> [[TMP3]], ptr [[S3]], align 8405; CHECK-NEXT: ret void406;407 %wide.vec = load <vscale x 3 x i64>, ptr %ptr, align 8408 %ldN = tail call { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64> %wide.vec)409 410 %3 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 0411 %4 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 1412 %5 = extractvalue { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } %ldN, 2413 414 store <vscale x 1 x i64> %3, <vscale x 1 x i64>* %s1415 store <vscale x 1 x i64> %4, <vscale x 1 x i64>* %s2416 store <vscale x 1 x i64> %5, <vscale x 1 x i64>* %s3417 ret void418}419 420define void @deinterleave_neg2_nxi8_factor3(ptr %ptr, <vscale x 8 x i8>* %s1, <vscale x 8 x i8>* %s2, <vscale x 8 x i8>* %s3) #0 {421; CHECK-LABEL: define void @deinterleave_neg2_nxi8_factor3422; CHECK-SAME: (ptr [[PTR:%.*]], ptr [[S1:%.*]], ptr [[S2:%.*]], ptr [[S3:%.*]]) #[[ATTR0]] {423; CHECK-NEXT: [[WIDE_VEC:%.*]] = load <vscale x 24 x i8>, ptr [[PTR]], align 8424; CHECK-NEXT: [[LDN:%.*]] = tail call { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv24i8(<vscale x 24 x i8> [[WIDE_VEC]])425; CHECK-NEXT: [[TMP1:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 0426; CHECK-NEXT: [[TMP2:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 1427; CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } [[LDN]], 2428; CHECK-NEXT: store <vscale x 8 x i8> [[TMP1]], ptr [[S1]], align 8429; CHECK-NEXT: store <vscale x 8 x i8> [[TMP2]], ptr [[S2]], align 8430; CHECK-NEXT: store <vscale x 8 x i8> [[TMP3]], ptr [[S3]], align 8431; CHECK-NEXT: ret void432;433 %wide.vec = load <vscale x 24 x i8>, ptr %ptr, align 8434 %ldN = tail call { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv12i8(<vscale x 24 x i8> %wide.vec)435 436 %3 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 0437 %4 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 1438 %5 = extractvalue { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } %ldN, 2439 440 store <vscale x 8 x i8> %3, <vscale x 8 x i8>* %s1441 store <vscale x 8 x i8> %4, <vscale x 8 x i8>* %s2442 store <vscale x 8 x i8> %5, <vscale x 8 x i8>* %s3443 ret void444}445 446define void @interleave1_nxi64_factor3(ptr %ptr, <vscale x 8 x i64> %s1, <vscale x 8 x i64> %s2, <vscale x 8 x i64> %s3) #0 {447; CHECK-LABEL: define void @interleave1_nxi64_factor3448; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 8 x i64> [[S1:%.*]], <vscale x 8 x i64> [[S2:%.*]], <vscale x 8 x i64> [[S3:%.*]]) #[[ATTR0]] {449; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0450; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 0)451; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 0)452; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 0)453; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])454; CHECK-NEXT: [[TMP5:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3455; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 2)456; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 2)457; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 2)458; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP5]])459; CHECK-NEXT: [[TMP9:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 6460; CHECK-NEXT: [[TMP10:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 4)461; CHECK-NEXT: [[TMP11:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 4)462; CHECK-NEXT: [[TMP12:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 4)463; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP10]], <vscale x 2 x i64> [[TMP11]], <vscale x 2 x i64> [[TMP12]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP9]])464; CHECK-NEXT: [[TMP13:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 9465; CHECK-NEXT: [[TMP14:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S1]], i64 6)466; CHECK-NEXT: [[TMP15:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S2]], i64 6)467; CHECK-NEXT: [[TMP16:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv8i64(<vscale x 8 x i64> [[S3]], i64 6)468; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP14]], <vscale x 2 x i64> [[TMP15]], <vscale x 2 x i64> [[TMP16]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP13]])469; CHECK-NEXT: ret void470;471 %interleave = tail call <vscale x 24 x i64> @llvm.vector.interleave3.nxv24i64(<vscale x 8 x i64> %s1, <vscale x 8 x i64> %s2, <vscale x 8 x i64> %s3)472 473 store <vscale x 24 x i64> %interleave, ptr %ptr, align 4474 ret void475}476 477define void @interleave2_nxi64_factor3(ptr %ptr, <vscale x 4 x i64> %s1, <vscale x 4 x i64> %s2, <vscale x 4 x i64> %s3) #0 {478; CHECK-LABEL: define void @interleave2_nxi64_factor3479; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 4 x i64> [[S1:%.*]], <vscale x 4 x i64> [[S2:%.*]], <vscale x 4 x i64> [[S3:%.*]]) #[[ATTR0]] {480; CHECK-NEXT: [[TMP1:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 0481; CHECK-NEXT: [[TMP2:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S1]], i64 0)482; CHECK-NEXT: [[TMP3:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S2]], i64 0)483; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S3]], i64 0)484; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP2]], <vscale x 2 x i64> [[TMP3]], <vscale x 2 x i64> [[TMP4]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP1]])485; CHECK-NEXT: [[TMP5:%.*]] = getelementptr <vscale x 2 x i64>, ptr [[PTR]], i64 3486; CHECK-NEXT: [[TMP6:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S1]], i64 2)487; CHECK-NEXT: [[TMP7:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S2]], i64 2)488; CHECK-NEXT: [[TMP8:%.*]] = call <vscale x 2 x i64> @llvm.vector.extract.nxv2i64.nxv4i64(<vscale x 4 x i64> [[S3]], i64 2)489; CHECK-NEXT: call void @llvm.aarch64.sve.st3.nxv2i64(<vscale x 2 x i64> [[TMP6]], <vscale x 2 x i64> [[TMP7]], <vscale x 2 x i64> [[TMP8]], <vscale x 2 x i1> splat (i1 true), ptr [[TMP5]])490; CHECK-NEXT: ret void491;492 %interleave = tail call <vscale x 12 x i64> @llvm.vector.interleave3.nxv12i64(<vscale x 4 x i64> %s1, <vscale x 4 x i64> %s2, <vscale x 4 x i64> %s3)493 494 store <vscale x 12 x i64> %interleave, ptr %ptr, align 4495 ret void496}497 498define void @interleave_neg_nxi8_factor3(ptr %ptr, <vscale x 8 x i8> %s1, <vscale x 8 x i8> %s2, <vscale x 8 x i8> %s3) #0 {499; CHECK-LABEL: define void @interleave_neg_nxi8_factor3500; CHECK-SAME: (ptr [[PTR:%.*]], <vscale x 8 x i8> [[S1:%.*]], <vscale x 8 x i8> [[S2:%.*]], <vscale x 8 x i8> [[S3:%.*]]) #[[ATTR0]] {501; CHECK-NEXT: [[INTERLEAVE:%.*]] = tail call <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8> [[S1]], <vscale x 8 x i8> [[S2]], <vscale x 8 x i8> [[S3]])502; CHECK-NEXT: store <vscale x 24 x i8> [[INTERLEAVE]], ptr [[PTR]], align 4503; CHECK-NEXT: ret void504;505 %interleave = tail call <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8> %s1, <vscale x 8 x i8> %s2, <vscale x 8 x i8> %s3)506 507 store <vscale x 24 x i8> %interleave, ptr %ptr, align 4508 ret void509}510 511declare { <vscale x 16 x i8>, <vscale x 16 x i8> } @llvm.vector.deinterleave2.nxv32i8(<vscale x 32 x i8>)512declare { <vscale x 8 x i16>, <vscale x 8 x i16> } @llvm.vector.deinterleave2.nxv16i16(<vscale x 16 x i16>)513declare { <vscale x 4 x i32>, <vscale x 4 x i32> } @llvm.vector.deinterleave2.nxv8i32(<vscale x 8 x i32>)514declare { <vscale x 2 x i64>, <vscale x 2 x i64> } @llvm.vector.deinterleave2.nxv4i64(<vscale x 4 x i64>)515declare { <vscale x 4 x float>, <vscale x 4 x float> } @llvm.vector.deinterleave2.nxv8f32(<vscale x 8 x float>)516declare { <vscale x 2 x double>, <vscale x 2 x double> } @llvm.vector.deinterleave2.nxv4f64(<vscale x 4 x double>)517declare { <vscale x 2 x ptr>, <vscale x 2 x ptr> } @llvm.vector.deinterleave2.nxv4p0(<vscale x 4 x ptr>)518 519; Larger deinterleaves to test 'legalization'520declare { <vscale x 16 x i32>, <vscale x 16 x i32> } @llvm.vector.deinterleave2.nxv32i32(<vscale x 32 x i32>)521declare { <vscale x 4 x double>, <vscale x 4 x double> } @llvm.vector.deinterleave2.nxv8f64(<vscale x 8 x double>)522 523declare <vscale x 32 x i8> @llvm.vector.interleave2.nxv32i8(<vscale x 16 x i8>, <vscale x 16 x i8>)524declare <vscale x 16 x i16> @llvm.vector.interleave2.nxv16i16(<vscale x 8 x i16>, <vscale x 8 x i16>)525declare <vscale x 8 x i32> @llvm.vector.interleave2.nxv8i32(<vscale x 4 x i32>, <vscale x 4 x i32>)526declare <vscale x 4 x i64> @llvm.vector.interleave2.nxv4i64(<vscale x 2 x i64>, <vscale x 2 x i64>)527declare <vscale x 8 x float> @llvm.vector.interleave2.nxv8f32(<vscale x 4 x float>, <vscale x 4 x float>)528declare <vscale x 4 x double> @llvm.vector.interleave2.nxv4f64(<vscale x 2 x double>, <vscale x 2 x double>)529declare <vscale x 4 x ptr> @llvm.vector.interleave2.nxv4p0(<vscale x 2 x ptr>, <vscale x 2 x ptr>)530 531; Larger interleaves to test 'legalization'532declare <vscale x 8 x double> @llvm.vector.interleave2.nxv8f64(<vscale x 4 x double>, <vscale x 4 x double>)533 534; De-Interleaves with Factor=3535declare { <vscale x 1 x i64>, <vscale x 1 x i64>, <vscale x 1 x i64> } @llvm.vector.deinterleave3.nxv3i64(<vscale x 3 x i64>)536declare { <vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64> } @llvm.vector.deinterleave3.nxv12i64(<vscale x 12 x i64>)537declare { <vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64> } @llvm.vector.deinterleave3.nxv24i64(<vscale x 24 x i64>)538declare { <vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8> } @llvm.vector.deinterleave3.nxv24i8(<vscale x 24 x i8>)539 540; Interleaves with Factor=3541declare <vscale x 24 x i8> @llvm.vector.interleave3.nxv24i8(<vscale x 8 x i8>, <vscale x 8 x i8>, <vscale x 8 x i8>)542declare <vscale x 24 x i64> @llvm.vector.interleave3.nxv24i64(<vscale x 8 x i64>, <vscale x 8 x i64>, <vscale x 8 x i64>)543declare <vscale x 12 x i64> @llvm.vector.interleave3.nxv12i64(<vscale x 4 x i64>, <vscale x 4 x i64>, <vscale x 4 x i64>)544 545attributes #0 = { vscale_range(1,16) "target-features"="+sve" }546