; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 ; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lsx %s -o - | FileCheck %s --check-prefix=LA32 ; RUN: llc --mtriple=loongarch64 --mattr=+lsx %s -o - | FileCheck %s --check-prefix=LA64 define void @load_trunc_2i64_to_2i32(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i64_to_2i32: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 2 ; LA32-NEXT: st.w $a0, $a1, 4 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i64_to_2i32: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: vshuf4i.w $vr0, $vr0, 8 ; LA64-NEXT: vstelm.d $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i64>, ptr %ptr %trunc = trunc <2 x i64> %a to <2 x i32> store <2 x i32> %trunc, ptr %dst ret void } define void @load_trunc_2i64_to_2i16(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i64_to_2i16: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: vextrins.h $vr0, $vr0, 20 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i64_to_2i16: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: vextrins.h $vr0, $vr0, 20 ; LA64-NEXT: vstelm.w $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i64>, ptr %ptr %trunc = trunc <2 x i64> %a to <2 x i16> store <2 x i16> %trunc, ptr %dst ret void } define void @load_trunc_2i64_to_2i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i64_to_2i8: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: vextrins.b $vr0, $vr0, 24 ; LA32-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i64_to_2i8: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: vextrins.b $vr0, $vr0, 24 ; LA64-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i64>, ptr %ptr %trunc = trunc <2 x i64> %a to <2 x i8> store <2 x i8> %trunc, ptr %dst ret void } define void @load_trunc_4i32_to_4i16(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_4i32_to_4i16: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: vpickev.h $vr0, $vr0, $vr0 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 1 ; LA32-NEXT: st.w $a0, $a1, 4 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_4i32_to_4i16: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: vpickev.h $vr0, $vr0, $vr0 ; LA64-NEXT: vstelm.d $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <4 x i32>, ptr %ptr %trunc = trunc <4 x i32> %a to <4 x i16> store <4 x i16> %trunc, ptr %dst ret void } define void @load_trunc_4i32_to_4i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_4i32_to_4i8: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: pcalau12i $a0, %pc_hi20(.LCPI4_0) ; LA32-NEXT: vld $vr1, $a0, %pc_lo12(.LCPI4_0) ; LA32-NEXT: vshuf.b $vr0, $vr0, $vr0, $vr1 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_4i32_to_4i8: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: pcalau12i $a0, %pc_hi20(.LCPI4_0) ; LA64-NEXT: vld $vr1, $a0, %pc_lo12(.LCPI4_0) ; LA64-NEXT: vshuf.b $vr0, $vr0, $vr0, $vr1 ; LA64-NEXT: vstelm.w $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <4 x i32>, ptr %ptr %trunc = trunc <4 x i32> %a to <4 x i8> store <4 x i8> %trunc, ptr %dst ret void } define void @load_trunc_8i16_to_8i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_8i16_to_8i8: ; LA32: # %bb.0: ; LA32-NEXT: vld $vr0, $a0, 0 ; LA32-NEXT: vpickev.b $vr0, $vr0, $vr0 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 1 ; LA32-NEXT: st.w $a0, $a1, 4 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_8i16_to_8i8: ; LA64: # %bb.0: ; LA64-NEXT: vld $vr0, $a0, 0 ; LA64-NEXT: vpickev.b $vr0, $vr0, $vr0 ; LA64-NEXT: vstelm.d $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <8 x i16>, ptr %ptr %trunc = trunc <8 x i16> %a to <8 x i8> store <8 x i8> %trunc, ptr %dst ret void } define void @load_trunc_2i32_to_2i16(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i32_to_2i16: ; LA32: # %bb.0: ; LA32-NEXT: ld.w $a2, $a0, 0 ; LA32-NEXT: ld.w $a0, $a0, 4 ; LA32-NEXT: vinsgr2vr.w $vr0, $a2, 0 ; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 1 ; LA32-NEXT: vshuf4i.h $vr0, $vr0, 8 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i32_to_2i16: ; LA64: # %bb.0: ; LA64-NEXT: ld.d $a0, $a0, 0 ; LA64-NEXT: vinsgr2vr.d $vr0, $a0, 0 ; LA64-NEXT: vshuf4i.h $vr0, $vr0, 8 ; LA64-NEXT: vstelm.w $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i32>, ptr %ptr %trunc = trunc <2 x i32> %a to <2 x i16> store <2 x i16> %trunc, ptr %dst ret void } define void @load_trunc_2i32_to_2i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i32_to_2i8: ; LA32: # %bb.0: ; LA32-NEXT: ld.w $a2, $a0, 0 ; LA32-NEXT: ld.w $a0, $a0, 4 ; LA32-NEXT: vinsgr2vr.w $vr0, $a2, 0 ; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 1 ; LA32-NEXT: vextrins.b $vr0, $vr0, 20 ; LA32-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i32_to_2i8: ; LA64: # %bb.0: ; LA64-NEXT: ld.d $a0, $a0, 0 ; LA64-NEXT: vinsgr2vr.d $vr0, $a0, 0 ; LA64-NEXT: vextrins.b $vr0, $vr0, 20 ; LA64-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i32>, ptr %ptr %trunc = trunc <2 x i32> %a to <2 x i8> store <2 x i8> %trunc, ptr %dst ret void } define void @load_trunc_4i16_to_4i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_4i16_to_4i8: ; LA32: # %bb.0: ; LA32-NEXT: ld.w $a2, $a0, 0 ; LA32-NEXT: ld.w $a0, $a0, 4 ; LA32-NEXT: vinsgr2vr.w $vr0, $a2, 0 ; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 1 ; LA32-NEXT: vpickev.b $vr0, $vr0, $vr0 ; LA32-NEXT: vpickve2gr.w $a0, $vr0, 0 ; LA32-NEXT: st.w $a0, $a1, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_4i16_to_4i8: ; LA64: # %bb.0: ; LA64-NEXT: ld.d $a0, $a0, 0 ; LA64-NEXT: vinsgr2vr.d $vr0, $a0, 0 ; LA64-NEXT: vpickev.b $vr0, $vr0, $vr0 ; LA64-NEXT: vstelm.w $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <4 x i16>, ptr %ptr %trunc = trunc <4 x i16> %a to <4 x i8> store <4 x i8> %trunc, ptr %dst ret void } define void @load_trunc_2i16_to_2i8(ptr %ptr, ptr %dst) nounwind { ; LA32-LABEL: load_trunc_2i16_to_2i8: ; LA32: # %bb.0: ; LA32-NEXT: ld.w $a0, $a0, 0 ; LA32-NEXT: vinsgr2vr.w $vr0, $a0, 0 ; LA32-NEXT: vshuf4i.b $vr0, $vr0, 8 ; LA32-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA32-NEXT: ret ; ; LA64-LABEL: load_trunc_2i16_to_2i8: ; LA64: # %bb.0: ; LA64-NEXT: ld.w $a0, $a0, 0 ; LA64-NEXT: vinsgr2vr.w $vr0, $a0, 0 ; LA64-NEXT: vshuf4i.b $vr0, $vr0, 8 ; LA64-NEXT: vstelm.h $vr0, $a1, 0, 0 ; LA64-NEXT: ret %a = load <2 x i16>, ptr %ptr %trunc = trunc <2 x i16> %a to <2 x i8> store <2 x i8> %trunc, ptr %dst ret void }