237 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 52; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA323; RUN: llc --mtriple=loongarch64 -mattr=+lasx < %s | FileCheck %s --check-prefixes=CHECK,LA644 5 6define <4 x i64> @should_not_be_optimized(ptr %ptr, ptr %dst) {7; LA32-LABEL: should_not_be_optimized:8; LA32: # %bb.0:9; LA32-NEXT: ld.w $a2, $a0, 010; LA32-NEXT: ld.w $a0, $a0, 411; LA32-NEXT: st.w $a2, $a1, 012; LA32-NEXT: vinsgr2vr.w $vr0, $a2, 013; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 114; LA32-NEXT: xvreplve0.d $xr0, $xr015; LA32-NEXT: st.w $a0, $a1, 416; LA32-NEXT: ret17;18; LA64-LABEL: should_not_be_optimized:19; LA64: # %bb.0:20; LA64-NEXT: ld.d $a0, $a0, 021; LA64-NEXT: xvreplgr2vr.d $xr0, $a022; LA64-NEXT: st.d $a0, $a1, 023; LA64-NEXT: ret24 %tmp = load i64, ptr %ptr25 store i64 %tmp, ptr %dst26 %tmp1 = insertelement <4 x i64> zeroinitializer, i64 %tmp, i32 027 %tmp2 = shufflevector <4 x i64> %tmp1, <4 x i64> poison, <4 x i32> zeroinitializer28 ret <4 x i64> %tmp229}30 31define <16 x i16> @should_not_be_optimized_sext_load(ptr %ptr) {32; CHECK-LABEL: should_not_be_optimized_sext_load:33; CHECK: # %bb.0:34; CHECK-NEXT: ld.b $a0, $a0, 035; CHECK-NEXT: xvreplgr2vr.h $xr0, $a036; CHECK-NEXT: ret37 %tmp = load i8, ptr %ptr38 %tmp1 = sext i8 %tmp to i1639 %tmp2 = insertelement <16 x i16> zeroinitializer, i16 %tmp1, i32 040 %tmp3 = shufflevector <16 x i16> %tmp2, <16 x i16> poison, <16 x i32> zeroinitializer41 ret <16 x i16> %tmp342}43 44define <16 x i16> @should_not_be_optimized_zext_load(ptr %ptr) {45; CHECK-LABEL: should_not_be_optimized_zext_load:46; CHECK: # %bb.0:47; CHECK-NEXT: ld.bu $a0, $a0, 048; CHECK-NEXT: xvreplgr2vr.h $xr0, $a049; CHECK-NEXT: ret50 %tmp = load i8, ptr %ptr51 %tmp1 = zext i8 %tmp to i1652 %tmp2 = insertelement <16 x i16> zeroinitializer, i16 %tmp1, i32 053 %tmp3 = shufflevector <16 x i16> %tmp2, <16 x i16> poison, <16 x i32> zeroinitializer54 ret <16 x i16> %tmp355}56 57define <4 x i64> @xvldrepl_d_unaligned_offset(ptr %ptr) {58; LA32-LABEL: xvldrepl_d_unaligned_offset:59; LA32: # %bb.0:60; LA32-NEXT: ld.w $a1, $a0, 461; LA32-NEXT: ld.w $a0, $a0, 862; LA32-NEXT: vinsgr2vr.w $vr0, $a1, 063; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 164; LA32-NEXT: xvreplve0.d $xr0, $xr065; LA32-NEXT: ret66;67; LA64-LABEL: xvldrepl_d_unaligned_offset:68; LA64: # %bb.0:69; LA64-NEXT: addi.d $a0, $a0, 470; LA64-NEXT: xvldrepl.d $xr0, $a0, 071; LA64-NEXT: ret72 %p = getelementptr i32, ptr %ptr, i32 173 %tmp = load i64, ptr %p74 %tmp1 = insertelement <4 x i64> zeroinitializer, i64 %tmp, i32 075 %tmp2 = shufflevector <4 x i64> %tmp1, <4 x i64> poison, <4 x i32> zeroinitializer76 ret <4 x i64> %tmp277}78 79define <32 x i8> @xvldrepl_b(ptr %ptr) {80; CHECK-LABEL: xvldrepl_b:81; CHECK: # %bb.0:82; CHECK-NEXT: xvldrepl.b $xr0, $a0, 083; CHECK-NEXT: ret84 %tmp = load i8, ptr %ptr85 %tmp1 = insertelement <32 x i8> zeroinitializer, i8 %tmp, i32 086 %tmp2 = shufflevector <32 x i8> %tmp1, <32 x i8> poison, <32 x i32> zeroinitializer87 ret <32 x i8> %tmp288}89 90define <32 x i8> @xvldrepl_b_offset(ptr %ptr) {91; CHECK-LABEL: xvldrepl_b_offset:92; CHECK: # %bb.0:93; CHECK-NEXT: xvldrepl.b $xr0, $a0, 3394; CHECK-NEXT: ret95 %p = getelementptr i8, ptr %ptr, i64 3396 %tmp = load i8, ptr %p97 %tmp1 = insertelement <32 x i8> zeroinitializer, i8 %tmp, i32 098 %tmp2 = shufflevector <32 x i8> %tmp1, <32 x i8> poison, <32 x i32> zeroinitializer99 ret <32 x i8> %tmp2100}101 102 103define <16 x i16> @xvldrepl_h(ptr %ptr) {104; CHECK-LABEL: xvldrepl_h:105; CHECK: # %bb.0:106; CHECK-NEXT: xvldrepl.h $xr0, $a0, 0107; CHECK-NEXT: ret108 %tmp = load i16, ptr %ptr109 %tmp1 = insertelement <16 x i16> zeroinitializer, i16 %tmp, i32 0110 %tmp2 = shufflevector <16 x i16> %tmp1, <16 x i16> poison, <16 x i32> zeroinitializer111 ret <16 x i16> %tmp2112}113 114define <16 x i16> @xvldrepl_h_offset(ptr %ptr) {115; CHECK-LABEL: xvldrepl_h_offset:116; CHECK: # %bb.0:117; CHECK-NEXT: xvldrepl.h $xr0, $a0, 66118; CHECK-NEXT: ret119 %p = getelementptr i16, ptr %ptr, i64 33120 %tmp = load i16, ptr %p121 %tmp1 = insertelement <16 x i16> zeroinitializer, i16 %tmp, i32 0122 %tmp2 = shufflevector <16 x i16> %tmp1, <16 x i16> poison, <16 x i32> zeroinitializer123 ret <16 x i16> %tmp2124}125 126define <8 x i32> @xvldrepl_w(ptr %ptr) {127; CHECK-LABEL: xvldrepl_w:128; CHECK: # %bb.0:129; CHECK-NEXT: xvldrepl.w $xr0, $a0, 0130; CHECK-NEXT: ret131 %tmp = load i32, ptr %ptr132 %tmp1 = insertelement <8 x i32> zeroinitializer, i32 %tmp, i32 0133 %tmp2 = shufflevector <8 x i32> %tmp1, <8 x i32> poison, <8 x i32> zeroinitializer134 ret <8 x i32> %tmp2135}136 137define <8 x i32> @xvldrepl_w_offset(ptr %ptr) {138; CHECK-LABEL: xvldrepl_w_offset:139; CHECK: # %bb.0:140; CHECK-NEXT: xvldrepl.w $xr0, $a0, 132141; CHECK-NEXT: ret142 %p = getelementptr i32, ptr %ptr, i64 33143 %tmp = load i32, ptr %p144 %tmp1 = insertelement <8 x i32> zeroinitializer, i32 %tmp, i32 0145 %tmp2 = shufflevector <8 x i32> %tmp1, <8 x i32> poison, <8 x i32> zeroinitializer146 ret <8 x i32> %tmp2147}148 149 150define <4 x i64> @xvldrepl_d(ptr %ptr) {151; LA32-LABEL: xvldrepl_d:152; LA32: # %bb.0:153; LA32-NEXT: ld.w $a1, $a0, 0154; LA32-NEXT: ld.w $a0, $a0, 4155; LA32-NEXT: vinsgr2vr.w $vr0, $a1, 0156; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 1157; LA32-NEXT: xvreplve0.d $xr0, $xr0158; LA32-NEXT: ret159;160; LA64-LABEL: xvldrepl_d:161; LA64: # %bb.0:162; LA64-NEXT: xvldrepl.d $xr0, $a0, 0163; LA64-NEXT: ret164 %tmp = load i64, ptr %ptr165 %tmp1 = insertelement <4 x i64> zeroinitializer, i64 %tmp, i32 0166 %tmp2 = shufflevector <4 x i64> %tmp1, <4 x i64> poison, <4 x i32> zeroinitializer167 ret <4 x i64> %tmp2168}169 170define <4 x i64> @xvldrepl_d_offset(ptr %ptr) {171; LA32-LABEL: xvldrepl_d_offset:172; LA32: # %bb.0:173; LA32-NEXT: ld.w $a1, $a0, 264174; LA32-NEXT: ld.w $a0, $a0, 268175; LA32-NEXT: vinsgr2vr.w $vr0, $a1, 0176; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 1177; LA32-NEXT: xvreplve0.d $xr0, $xr0178; LA32-NEXT: ret179;180; LA64-LABEL: xvldrepl_d_offset:181; LA64: # %bb.0:182; LA64-NEXT: xvldrepl.d $xr0, $a0, 264183; LA64-NEXT: ret184 %p = getelementptr i64, ptr %ptr, i64 33185 %tmp = load i64, ptr %p186 %tmp1 = insertelement <4 x i64> zeroinitializer, i64 %tmp, i32 0187 %tmp2 = shufflevector <4 x i64> %tmp1, <4 x i64> poison, <4 x i32> zeroinitializer188 ret <4 x i64> %tmp2189}190 191define <8 x float> @vldrepl_w_flt(ptr %ptr) {192; CHECK-LABEL: vldrepl_w_flt:193; CHECK: # %bb.0:194; CHECK-NEXT: xvldrepl.w $xr0, $a0, 0195; CHECK-NEXT: ret196 %tmp = load float, ptr %ptr197 %tmp1 = insertelement <8 x float> zeroinitializer, float %tmp, i32 0198 %tmp2 = shufflevector <8 x float> %tmp1, <8 x float> poison, <8 x i32> zeroinitializer199 ret <8 x float> %tmp2200}201 202define <8 x float> @vldrepl_w_flt_offset(ptr %ptr) {203; CHECK-LABEL: vldrepl_w_flt_offset:204; CHECK: # %bb.0:205; CHECK-NEXT: xvldrepl.w $xr0, $a0, 264206; CHECK-NEXT: ret207 %p = getelementptr i64, ptr %ptr, i64 33208 %tmp = load float, ptr %p209 %tmp1 = insertelement <8 x float> zeroinitializer, float %tmp, i32 0210 %tmp2 = shufflevector <8 x float> %tmp1, <8 x float> poison, <8 x i32> zeroinitializer211 ret <8 x float> %tmp2212}213 214define <4 x double> @vldrepl_d_dbl(ptr %ptr) {215; CHECK-LABEL: vldrepl_d_dbl:216; CHECK: # %bb.0:217; CHECK-NEXT: xvldrepl.d $xr0, $a0, 0218; CHECK-NEXT: ret219 %tmp = load double, ptr %ptr220 %tmp1 = insertelement <4 x double> zeroinitializer, double %tmp, i32 0221 %tmp2 = shufflevector <4 x double> %tmp1, <4 x double> poison, <4 x i32> zeroinitializer222 ret <4 x double> %tmp2223}224 225define <4 x double> @vldrepl_d_dbl_offset(ptr %ptr) {226; CHECK-LABEL: vldrepl_d_dbl_offset:227; CHECK: # %bb.0:228; CHECK-NEXT: xvldrepl.d $xr0, $a0, 264229; CHECK-NEXT: ret230 %p = getelementptr i64, ptr %ptr, i64 33231 %tmp = load double, ptr %p232 %tmp1 = insertelement <4 x double> zeroinitializer, double %tmp, i32 0233 %tmp2 = shufflevector <4 x double> %tmp1, <4 x double> poison, <4 x i32> zeroinitializer234 ret <4 x double> %tmp2235}236 237