[AArch64] Added tests for inserting scalar result of uaddlv neon instrinsic function into a vector

[nilanjana_basu]

Summary

Inserting scalar result of 'uaddlv' neon intrinsic function to a destination vector currently makes use of the integer unit. Subsequent patches will eliminate the redundant use of the integer registers. This is an initial set of tests for this functionality.

[nilanjana_basu]

Added a test case for inserting 'uaddlv' result in a non-zero index in the destination vector

[nilanjana_basu]

Added an test file in MIR to show what happens in these cases post ISel.

[nilanjana_basu]

Updated tests based on latest llvm open source code

llvm/test/CodeGen/AArch64/aarch64-neon-vector-insert-uaddlv.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=arm64-apple-ios -o - %s | FileCheck %s
+
+declare i32 @llvm.aarch64.neon.uaddlv.i32.v8i8(<8 x i8>) #0
+declare i32 @llvm.aarch64.neon.uaddlv.i32.v16i8(<16 x i8>) #0
+declare i32 @llvm.aarch64.neon.uaddlv.i32.v4i16(<4 x i16>) #0
+declare i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16>) #0
+declare i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32>) #0
+
+define void @insert_vec_v2i32_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i32_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s0, v0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov.s v1[0], w8
+; CHECK-NEXT:    ucvtf.2s v0, v1
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = insertelement <2 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <2 x i32> %1 to <2 x float>
+  store <2 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v4i32_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v4i32_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v0[0], w8
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = insertelement <4 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <4 x i32> %1 to <4 x float>
+  store <4 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i32_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i32_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s1, v0
+; CHECK-NEXT:    stp q0, q0, [x0, #32]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v2[0], w8
+; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    stp q1, q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = insertelement <16 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <16 x i32> %1 to <16 x float>
+  store <16 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v23i32_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v23i32_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s1, v0
+; CHECK-NEXT:    stp q0, q0, [x0, #16]
+; CHECK-NEXT:    stp q0, q0, [x0, #48]
+; CHECK-NEXT:    str d0, [x0, #80]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v2[0], w8
+; CHECK-NEXT:    add x8, x0, #88
+; CHECK-NEXT:    st1.s { v0 }[2], [x8]
+; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    str q1, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = insertelement <23 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <23 x i32> %1 to <23 x float>
+  store <23 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v2i32_uaddlv_from_v16i8(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i32_uaddlv_from_v16i8:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.16b h0, v0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov.s v1[0], w8
+; CHECK-NEXT:    ucvtf.2s v0, v1
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v16i8(<16 x i8> zeroinitializer)
+  %1 = insertelement <2 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <2 x i32> %1 to <2 x float>
+  store <2 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v2i32_uaddlv_from_v8i8(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i32_uaddlv_from_v8i8:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    uaddlv.8b h1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v0[0], w8
+; CHECK-NEXT:    ucvtf.2s v0, v0
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i8(<8 x i8> zeroinitializer)
+  %1 = insertelement <2 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <2 x i32> %1 to <2 x float>
+  store <2 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v2i32_uaddlv_from_v4i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i32_uaddlv_from_v4i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    uaddlv.4h s1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v0[0], w8
+; CHECK-NEXT:    ucvtf.2s v0, v0
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v4i16(<4 x i16> zeroinitializer)
+  %1 = insertelement <2 x i32> zeroinitializer, i32 %vaddlv, i64 0
+  %2 = uitofp <2 x i32> %1 to <2 x float>
+  store <2 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v6i64_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v6i64_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    movi d0, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d2, v1
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    mov.d v1[0], x8
+; CHECK-NEXT:    str d2, [x0, #16]
+; CHECK-NEXT:    ucvtf.2d v1, v1
+; CHECK-NEXT:    fcvtn v1.2s, v1.2d
+; CHECK-NEXT:    mov.d v1[1], v0[0]
+; CHECK-NEXT:    str q1, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = insertelement <6 x i64> zeroinitializer, i64 %vaddlv, i64 0
+  %2 = uitofp <6 x i64> %1 to <6 x float>
+  store <6 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v2i64_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i64_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d1, v0
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov.d v0[0], x8
+; CHECK-NEXT:    ucvtf.2d v0, v0
+; CHECK-NEXT:    fcvtn v0.2s, v0.2d
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = insertelement <2 x i64> zeroinitializer, i64 %vaddlv, i64 0
+  %2 = uitofp <2 x i64> %1 to <2 x float>
+  store <2 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v5i64_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v5i64_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    str wzr, [x0, #16]
+; CHECK-NEXT:    movi d0, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d2, v1
+; CHECK-NEXT:    fmov x8, d2
+; CHECK-NEXT:    mov.d v1[0], x8
+; CHECK-NEXT:    ucvtf.2d v1, v1
+; CHECK-NEXT:    fcvtn v1.2s, v1.2d
+; CHECK-NEXT:    mov.d v1[1], v0[0]
+; CHECK-NEXT:    str q1, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = insertelement <5 x i64> zeroinitializer, i64 %vaddlv, i64 0
+  %2 = uitofp <5 x i64> %1 to <5 x float>
+  store <5 x float> %2, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v8i16_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v8i16_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s0, v0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov.h v1[0], w8
+; CHECK-NEXT:    ushll.4s v0, v1, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = trunc i32 %vaddlv to i16
+  %2 = insertelement <8 x i16> zeroinitializer, i16 %1, i64 0
+  %3 = uitofp <8 x i16> %2 to <8 x float>
+  store <8 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v3i16_uaddlv_from_v8i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v3i16_uaddlv_from_v8i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s0, v0
+; CHECK-NEXT:    fmov w8, s0
+; CHECK-NEXT:    mov.h v1[0], w8
+; CHECK-NEXT:    add x8, x0, #8
+; CHECK-NEXT:    ushll.4s v0, v1, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    st1.s { v0 }[2], [x8]
+; CHECK-NEXT:    str d0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = trunc i32 %vaddlv to i16
+  %2 = insertelement <3 x i16> zeroinitializer, i16 %1, i64 0
+  %3 = uitofp <3 x i16> %2 to <3 x float>
+  store <3 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i64_uaddlv_from_v4i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i64_uaddlv_from_v4i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    movi d0, #0000000000000000
+; CHECK-NEXT:    movi.2d v3, #0000000000000000
+; CHECK-NEXT:    uaddlv.4h s2, v1
+; CHECK-NEXT:    stp q1, q1, [x0, #32]
+; CHECK-NEXT:    fmov w8, s2
+; CHECK-NEXT:    mov.s v3[0], w8
+; CHECK-NEXT:    ucvtf.2d v2, v3
+; CHECK-NEXT:    fcvtn v2.2s, v2.2d
+; CHECK-NEXT:    mov.d v2[1], v0[0]
+; CHECK-NEXT:    stp q2, q1, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v4i16(<4 x i16> zeroinitializer)
+  %1 = zext i32 %vaddlv to i64
+  %2 = insertelement <16 x i64> zeroinitializer, i64 %1, i64 0
+  %3 = uitofp <16 x i64> %2 to <16 x float>
+  store <16 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i8_uaddlv_from_v8i8(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i8_uaddlv_from_v8i8:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    uaddlv.8b h1, v0
+; CHECK-NEXT:    stp q0, q0, [x0, #32]
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.b v2[0], w8
+; CHECK-NEXT:    zip1.8b v1, v2, v0
+; CHECK-NEXT:    bic.4h v1, #255, lsl #8
+; CHECK-NEXT:    ushll.4s v1, v1, #0
+; CHECK-NEXT:    ucvtf.4s v1, v1
+; CHECK-NEXT:    stp q1, q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i8(<8 x i8> zeroinitializer)
+  %1 = trunc i32 %vaddlv to i8
+  %2 = insertelement <16 x i8> zeroinitializer, i8 %1, i64 0
+  %3 = uitofp <16 x i8> %2 to <16 x float>
+  store <16 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v8i8_uaddlv_from_v8i8(ptr %0) {
+; CHECK-LABEL: insert_vec_v8i8_uaddlv_from_v8i8:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
+; CHECK-NEXT:    uaddlv.8b h1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.h v0[0], w8
+; CHECK-NEXT:    bic.4h v0, #255, lsl #8
+; CHECK-NEXT:    ushll.4s v0, v0, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i8(<8 x i8> zeroinitializer)
+  %1 = trunc i32 %vaddlv to i8
+  %2 = insertelement <8 x i8> zeroinitializer, i8 %1, i64 0
+  %3 = uitofp <8 x i8> %2 to <8 x float>
+  store <8 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v12i16_uaddlv_from_v4i16(ptr %0) {
+; CHECK-LABEL: insert_vec_v12i16_uaddlv_from_v4i16:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
+; CHECK-NEXT:    stp xzr, xzr, [x0, #32]
+; CHECK-NEXT:    uaddlv.4h s1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.h v0[0], w8
+; CHECK-NEXT:    ushll.4s v0, v0, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v4i16(<4 x i16> zeroinitializer)
+  %1 = trunc i32 %vaddlv to i16
+  %2 = insertelement <12 x i16> zeroinitializer, i16 %1, i64 0
+  %3 = uitofp <12 x i16> %2 to <12 x float>
+  store <12 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v8i32_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v8i32_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
+; CHECK-NEXT:    uaddlv.4s d1, v0
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov.s v0[0], w8
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i32
+  %2 = insertelement <8 x i32> zeroinitializer, i32 %1, i64 0
+  %3 = uitofp <8 x i32> %2 to <8 x float>
+  store <8 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i32_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i32_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v2, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d1, v0
+; CHECK-NEXT:    stp q0, q0, [x0, #32]
+; CHECK-NEXT:    fmov x8, d1
+; CHECK-NEXT:    mov.s v2[0], w8
+; CHECK-NEXT:    ucvtf.4s v1, v2
+; CHECK-NEXT:    stp q1, q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i32
+  %2 = insertelement <16 x i32> zeroinitializer, i32 %1, i64 0
+  %3 = uitofp <16 x i32> %2 to <16 x float>
+  store <16 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v4i16_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v4i16_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d0, v0
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov.h v1[0], w8
+; CHECK-NEXT:    ushll.4s v0, v1, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i16
+  %2 = insertelement <4 x i16> zeroinitializer, i16 %1, i64 0
+  %3 = uitofp <4 x i16> %2 to <4 x float>
+  store <4 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i16_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i16_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d0, v0
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    mov.h v1[0], w8
+; CHECK-NEXT:    stp q0, q0, [x0, #32]
+; CHECK-NEXT:    ushll.4s v1, v1, #0
+; CHECK-NEXT:    ucvtf.4s v1, v1
+; CHECK-NEXT:    stp q1, q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i16
+  %2 = insertelement <16 x i16> zeroinitializer, i16 %1, i64 0
+  %3 = uitofp <16 x i16> %2 to <16 x float>
+  store <16 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v8i8_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v8i8_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    stp xzr, xzr, [x0, #16]
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d0, v0
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov.h v1[0], w8
+; CHECK-NEXT:    bic.4h v1, #255, lsl #8
+; CHECK-NEXT:    ushll.4s v0, v1, #0
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i8
+  %2 = insertelement <8 x i8> zeroinitializer, i8 %1, i64 0
+  %3 = uitofp <8 x i8> %2 to <8 x float>
+  store <8 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v16i8_uaddlv_from_v4i32(ptr %0) {
+; CHECK-LABEL: insert_vec_v16i8_uaddlv_from_v4i32:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    uaddlv.4s d0, v0
+; CHECK-NEXT:    fmov x8, d0
+; CHECK-NEXT:    mov.b v1[0], w8
+; CHECK-NEXT:    zip1.8b v0, v1, v0
+; CHECK-NEXT:    movi.2d v1, #0000000000000000
+; CHECK-NEXT:    bic.4h v0, #255, lsl #8
+; CHECK-NEXT:    ushll.4s v0, v0, #0
+; CHECK-NEXT:    stp q1, q1, [x0, #32]
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    stp q0, q1, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32> zeroinitializer)
+  %1 = trunc i64 %vaddlv to i8
+  %2 = insertelement <16 x i8> zeroinitializer, i8 %1, i64 0
+  %3 = uitofp <16 x i8> %2 to <16 x float>
+  store <16 x float> %3, ptr %0, align 8
+  ret void
+}
+
+define void @insert_vec_v2i32_uaddlv_from_v8i16_nz_index(ptr %0) {
+; CHECK-LABEL: insert_vec_v2i32_uaddlv_from_v8i16_nz_index:
+; CHECK:       ; %bb.0: ; %entry
+; CHECK-NEXT:    movi.2d v0, #0000000000000000
+; CHECK-NEXT:    uaddlv.8h s1, v0
+; CHECK-NEXT:    fmov w8, s1
+; CHECK-NEXT:    mov.s v0[2], w8
+; CHECK-NEXT:    ucvtf.4s v0, v0
+; CHECK-NEXT:    str q0, [x0]
+; CHECK-NEXT:    ret
+
+entry:
+  %vaddlv = tail call i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16> zeroinitializer)
+  %1 = insertelement <4 x i32> zeroinitializer, i32 %vaddlv, i64 2
+  %2 = uitofp <4 x i32> %1 to <4 x float>
+  store <4 x float> %2, ptr %0, align 8
+  ret void
+}

llvm/test/CodeGen/AArch64/peephole-insvigpr.mir

+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
+# RUN: llc -run-pass=aarch64-mi-peephole-opt -mtriple=aarch64-unknown-linux -verify-machineinstrs -o - %s | FileCheck %s
+--- |
+  source_filename = "/Users/nilanjana/Documents/code/llvm-project/llvm/test/CodeGen/AArch64/tmp.ll"
+
+  ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)
+  declare i32 @llvm.aarch64.neon.uaddlv.i32.v8i16(<8 x i16>) #0
+
+  ; Function Attrs: nocallback nofree nosync nounwind willreturn memory(none)
+  declare i64 @llvm.aarch64.neon.uaddlv.i64.v4i32(<4 x i32>) #0
+
+  define void @insert_vec_v6i64_uaddlv_from_v4i32(ptr %0) {
+  entry:
+    ret void
+  }
+
+  define void @insert_vec_v2i32_uaddlv_from_v8i16(ptr %0) {
+  entry:
+    ret void
+  }
+
+  define void @insert_vec_v8i16_uaddlv_from_v8i16(ptr %0) {
+  entry:
+    ret void
+  }
+
+  define void @insert_vec_v16i8_uaddlv_from_v4i32(ptr %0) {
+  entry:
+    ret void
+  }
+
+  define void @insert_vec_v2i32_uaddlv_from_v8i16_nz_index(ptr %0) {
+  entry:
+    ret void
+  }
+
+  ; The optimization is not applicable when the source is not a virtual register
+  define void @insert_vec_from_gpr(i32 %v, ptr %p) {
+  entry:
+    ret void
+  }
+
+  attributes #0 = { nocallback nofree nosync nounwind willreturn memory(none) }
+
+...
+---
+name:            insert_vec_v6i64_uaddlv_from_v4i32
+registers:
+  - { id: 0, class: gpr64common, preferred-register: '' }
+  - { id: 1, class: fpr128, preferred-register: '' }
+  - { id: 2, class: fpr64, preferred-register: '' }
+  - { id: 3, class: fpr128, preferred-register: '' }
+  - { id: 4, class: fpr128, preferred-register: '' }
+  - { id: 5, class: gpr64, preferred-register: '' }
+  - { id: 6, class: fpr128, preferred-register: '' }
+  - { id: 7, class: fpr128, preferred-register: '' }
+  - { id: 8, class: fpr64, preferred-register: '' }
+  - { id: 9, class: fpr128, preferred-register: '' }
+  - { id: 10, class: fpr128, preferred-register: '' }
+  - { id: 11, class: fpr128, preferred-register: '' }
+  - { id: 12, class: fpr64, preferred-register: '' }
+  - { id: 13, class: fpr128, preferred-register: '' }
+  - { id: 14, class: fpr128, preferred-register: '' }
+  - { id: 15, class: fpr128, preferred-register: '' }
+  - { id: 16, class: gpr64all, preferred-register: '' }
+  - { id: 17, class: fpr64, preferred-register: '' }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+body:             |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: insert_vec_v6i64_uaddlv_from_v4i32
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[UADDLVv4i32v:%[0-9]+]]:fpr64 = UADDLVv4i32v [[MOVIv2d_ns]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], killed [[UADDLVv4i32v]], %subreg.dsub
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64 = COPY [[INSERT_SUBREG]].dsub
+    ; CHECK-NEXT: [[INSvi64gpr:%[0-9]+]]:fpr128 = INSvi64gpr [[MOVIv2d_ns]], 0, killed [[COPY1]]
+    ; CHECK-NEXT: [[MOVID:%[0-9]+]]:fpr64 = MOVID 0
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], killed [[MOVID]], %subreg.dsub
+    ; CHECK-NEXT: [[UCVTFv2f64_:%[0-9]+]]:fpr128 = nofpexcept UCVTFv2f64 killed [[INSvi64gpr]], implicit $fpcr
+    ; CHECK-NEXT: [[FCVTNv2i32_:%[0-9]+]]:fpr64 = nofpexcept FCVTNv2i32 killed [[UCVTFv2f64_]], implicit $fpcr
+    ; CHECK-NEXT: [[DEF2:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG2:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF2]], killed [[FCVTNv2i32_]], %subreg.dsub
+    ; CHECK-NEXT: [[INSvi64lane:%[0-9]+]]:fpr128 = INSvi64lane [[INSERT_SUBREG2]], 1, killed [[INSERT_SUBREG1]], 0
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr64 = COPY [[MOVIv2d_ns]].dsub
+    ; CHECK-NEXT: STRDui killed [[COPY2]], [[COPY]], 2 :: (store (s64) into %ir.0 + 16)
+    ; CHECK-NEXT: STRQui killed [[INSvi64lane]], [[COPY]], 0 :: (store (s128) into %ir.0, align 8)
+    ; CHECK-NEXT: RET_ReallyLR
+    %0:gpr64common = COPY $x0
+    %1:fpr128 = MOVIv2d_ns 0
+    %2:fpr64 = UADDLVv4i32v %1
+    %4:fpr128 = IMPLICIT_DEF
+    %3:fpr128 = INSERT_SUBREG %4, killed %2, %subreg.dsub
+    %5:gpr64 = COPY %3.dsub
+    %7:fpr128 = INSvi64gpr %1, 0, killed %5
+    %8:fpr64 = MOVID 0
+    %10:fpr128 = IMPLICIT_DEF
+    %9:fpr128 = INSERT_SUBREG %10, killed %8, %subreg.dsub
+    %11:fpr128 = nofpexcept UCVTFv2f64 killed %7, implicit $fpcr
+    %12:fpr64 = nofpexcept FCVTNv2i32 killed %11, implicit $fpcr
+    %14:fpr128 = IMPLICIT_DEF
+    %13:fpr128 = INSERT_SUBREG %14, killed %12, %subreg.dsub
+    %15:fpr128 = INSvi64lane %13, 1, killed %9, 0
+    %17:fpr64 = COPY %1.dsub
+    STRDui killed %17, %0, 2 :: (store (s64) into %ir.0 + 16)
+    STRQui killed %15, %0, 0 :: (store (s128) into %ir.0, align 8)
+    RET_ReallyLR
+
+...
+---
+name:            insert_vec_v2i32_uaddlv_from_v8i16
+registers:
+  - { id: 0, class: gpr64common, preferred-register: '' }
+  - { id: 1, class: fpr128, preferred-register: '' }
+  - { id: 2, class: fpr32, preferred-register: '' }
+  - { id: 3, class: fpr128, preferred-register: '' }
+  - { id: 4, class: fpr128, preferred-register: '' }
+  - { id: 5, class: gpr32, preferred-register: '' }
+  - { id: 6, class: fpr64, preferred-register: '' }
+  - { id: 7, class: fpr128, preferred-register: '' }
+  - { id: 8, class: fpr128, preferred-register: '' }
+  - { id: 9, class: fpr128, preferred-register: '' }
+  - { id: 10, class: fpr64, preferred-register: '' }
+  - { id: 11, class: fpr64, preferred-register: '' }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+body:             |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: insert_vec_v2i32_uaddlv_from_v8i16
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[UADDLVv8i16v:%[0-9]+]]:fpr32 = UADDLVv8i16v killed [[MOVIv2d_ns]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], killed [[UADDLVv8i16v]], %subreg.ssub
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[INSERT_SUBREG]].ssub
+    ; CHECK-NEXT: [[MOVID:%[0-9]+]]:fpr64 = MOVID 0
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], killed [[MOVID]], %subreg.dsub
+    ; CHECK-NEXT: [[INSvi32gpr:%[0-9]+]]:fpr128 = INSvi32gpr [[INSERT_SUBREG1]], 0, killed [[COPY1]]
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr64 = COPY [[INSvi32gpr]].dsub
+    ; CHECK-NEXT: [[UCVTFv2f32_:%[0-9]+]]:fpr64 = nofpexcept UCVTFv2f32 killed [[COPY2]], implicit $fpcr
+    ; CHECK-NEXT: STRDui killed [[UCVTFv2f32_]], [[COPY]], 0 :: (store (s64) into %ir.0)
+    ; CHECK-NEXT: RET_ReallyLR
+    %0:gpr64common = COPY $x0
+    %1:fpr128 = MOVIv2d_ns 0
+    %2:fpr32 = UADDLVv8i16v killed %1
+    %4:fpr128 = IMPLICIT_DEF
+    %3:fpr128 = INSERT_SUBREG %4, killed %2, %subreg.ssub
+    %5:gpr32 = COPY %3.ssub
+    %6:fpr64 = MOVID 0
+    %8:fpr128 = IMPLICIT_DEF
+    %7:fpr128 = INSERT_SUBREG %8, killed %6, %subreg.dsub
+    %9:fpr128 = INSvi32gpr %7, 0, killed %5
+    %10:fpr64 = COPY %9.dsub
+    %11:fpr64 = nofpexcept UCVTFv2f32 killed %10, implicit $fpcr
+    STRDui killed %11, %0, 0 :: (store (s64) into %ir.0)
+    RET_ReallyLR
+
+...
+---
+name:            insert_vec_v8i16_uaddlv_from_v8i16
+registers:
+  - { id: 0, class: gpr64common, preferred-register: '' }
+  - { id: 1, class: fpr128, preferred-register: '' }
+  - { id: 2, class: fpr32, preferred-register: '' }
+  - { id: 3, class: fpr128, preferred-register: '' }
+  - { id: 4, class: fpr128, preferred-register: '' }
+  - { id: 5, class: gpr32, preferred-register: '' }
+  - { id: 6, class: fpr64, preferred-register: '' }
+  - { id: 7, class: fpr128, preferred-register: '' }
+  - { id: 8, class: fpr128, preferred-register: '' }
+  - { id: 9, class: fpr128, preferred-register: '' }
+  - { id: 10, class: fpr64, preferred-register: '' }
+  - { id: 11, class: fpr128, preferred-register: '' }
+  - { id: 12, class: fpr128, preferred-register: '' }
+  - { id: 13, class: gpr32, preferred-register: '' }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+body:             |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: insert_vec_v8i16_uaddlv_from_v8i16
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[UADDLVv8i16v:%[0-9]+]]:fpr32 = UADDLVv8i16v killed [[MOVIv2d_ns]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], killed [[UADDLVv8i16v]], %subreg.ssub
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[INSERT_SUBREG]].ssub
+    ; CHECK-NEXT: [[MOVID:%[0-9]+]]:fpr64 = MOVID 0
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], killed [[MOVID]], %subreg.dsub
+    ; CHECK-NEXT: [[INSvi16gpr:%[0-9]+]]:fpr128 = INSvi16gpr [[INSERT_SUBREG1]], 0, killed [[COPY1]]
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:fpr64 = COPY [[INSvi16gpr]].dsub
+    ; CHECK-NEXT: [[USHLLv4i16_shift:%[0-9]+]]:fpr128 = USHLLv4i16_shift killed [[COPY2]], 0
+    ; CHECK-NEXT: [[UCVTFv4f32_:%[0-9]+]]:fpr128 = nofpexcept UCVTFv4f32 killed [[USHLLv4i16_shift]], implicit $fpcr
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:gpr32 = COPY $wzr
+    ; CHECK-NEXT: STRWui [[COPY3]], [[COPY]], 7 :: (store (s32) into %ir.0 + 28)
+    ; CHECK-NEXT: STRWui [[COPY3]], [[COPY]], 6 :: (store (s32) into %ir.0 + 24, align 8)
+    ; CHECK-NEXT: STRWui [[COPY3]], [[COPY]], 5 :: (store (s32) into %ir.0 + 20)
+    ; CHECK-NEXT: STRWui [[COPY3]], [[COPY]], 4 :: (store (s32) into %ir.0 + 16, align 8)
+    ; CHECK-NEXT: STRQui killed [[UCVTFv4f32_]], [[COPY]], 0 :: (store (s128) into %ir.0, align 8)
+    ; CHECK-NEXT: RET_ReallyLR
+    %0:gpr64common = COPY $x0
+    %1:fpr128 = MOVIv2d_ns 0
+    %2:fpr32 = UADDLVv8i16v killed %1
+    %4:fpr128 = IMPLICIT_DEF
+    %3:fpr128 = INSERT_SUBREG %4, killed %2, %subreg.ssub
+    %5:gpr32 = COPY %3.ssub
+    %6:fpr64 = MOVID 0
+    %8:fpr128 = IMPLICIT_DEF
+    %7:fpr128 = INSERT_SUBREG %8, killed %6, %subreg.dsub
+    %9:fpr128 = INSvi16gpr %7, 0, killed %5
+    %10:fpr64 = COPY %9.dsub
+    %11:fpr128 = USHLLv4i16_shift killed %10, 0
+    %12:fpr128 = nofpexcept UCVTFv4f32 killed %11, implicit $fpcr
+    %13:gpr32 = COPY $wzr
+    STRWui %13, %0, 7 :: (store (s32) into %ir.0 + 28)
+    STRWui %13, %0, 6 :: (store (s32) into %ir.0 + 24, align 8)
+    STRWui %13, %0, 5 :: (store (s32) into %ir.0 + 20)
+    STRWui %13, %0, 4 :: (store (s32) into %ir.0 + 16, align 8)
+    STRQui killed %12, %0, 0 :: (store (s128) into %ir.0, align 8)
+    RET_ReallyLR
+
+...
+---
+name:            insert_vec_v16i8_uaddlv_from_v4i32
+registers:
+  - { id: 0, class: gpr64common, preferred-register: '' }
+  - { id: 1, class: fpr128, preferred-register: '' }
+  - { id: 2, class: fpr64, preferred-register: '' }
+  - { id: 3, class: fpr128, preferred-register: '' }
+  - { id: 4, class: fpr128, preferred-register: '' }
+  - { id: 5, class: gpr64all, preferred-register: '' }
+  - { id: 6, class: gpr32, preferred-register: '' }
+  - { id: 7, class: fpr64, preferred-register: '' }
+  - { id: 8, class: fpr128, preferred-register: '' }
+  - { id: 9, class: fpr128, preferred-register: '' }
+  - { id: 10, class: fpr128, preferred-register: '' }
+  - { id: 11, class: fpr64, preferred-register: '' }
+  - { id: 12, class: fpr64, preferred-register: '' }
+  - { id: 13, class: fpr64, preferred-register: '' }
+  - { id: 14, class: fpr64, preferred-register: '' }
+  - { id: 15, class: fpr128, preferred-register: '' }
+  - { id: 16, class: fpr128, preferred-register: '' }
+  - { id: 17, class: fpr128, preferred-register: '' }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+body:             |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: insert_vec_v16i8_uaddlv_from_v4i32
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[UADDLVv4i32v:%[0-9]+]]:fpr64 = UADDLVv4i32v [[MOVIv2d_ns]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], killed [[UADDLVv4i32v]], %subreg.dsub
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr64all = COPY [[INSERT_SUBREG]].dsub
+    ; CHECK-NEXT: [[COPY2:%[0-9]+]]:gpr32 = COPY [[COPY1]].sub_32
+    ; CHECK-NEXT: [[MOVID:%[0-9]+]]:fpr64 = MOVID 0
+    ; CHECK-NEXT: [[DEF1:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG1:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF1]], killed [[MOVID]], %subreg.dsub
+    ; CHECK-NEXT: [[INSvi8gpr:%[0-9]+]]:fpr128 = INSvi8gpr [[INSERT_SUBREG1]], 0, killed [[COPY2]]
+    ; CHECK-NEXT: [[COPY3:%[0-9]+]]:fpr64 = COPY [[INSvi8gpr]].dsub
+    ; CHECK-NEXT: [[DEF2:%[0-9]+]]:fpr64 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[ZIP1v8i8_:%[0-9]+]]:fpr64 = ZIP1v8i8 killed [[COPY3]], killed [[DEF2]]
+    ; CHECK-NEXT: [[BICv4i16_:%[0-9]+]]:fpr64 = BICv4i16 [[ZIP1v8i8_]], 255, 8
+    ; CHECK-NEXT: [[USHLLv4i16_shift:%[0-9]+]]:fpr128 = USHLLv4i16_shift killed [[BICv4i16_]], 0
+    ; CHECK-NEXT: [[UCVTFv4f32_:%[0-9]+]]:fpr128 = nofpexcept UCVTFv4f32 killed [[USHLLv4i16_shift]], implicit $fpcr
+    ; CHECK-NEXT: STRQui [[MOVIv2d_ns]], [[COPY]], 3 :: (store (s128) into %ir.0 + 48, align 8)
+    ; CHECK-NEXT: STRQui [[MOVIv2d_ns]], [[COPY]], 2 :: (store (s128) into %ir.0 + 32, align 8)
+    ; CHECK-NEXT: STRQui [[MOVIv2d_ns]], [[COPY]], 1 :: (store (s128) into %ir.0 + 16, align 8)
+    ; CHECK-NEXT: STRQui killed [[UCVTFv4f32_]], [[COPY]], 0 :: (store (s128) into %ir.0, align 8)
+    ; CHECK-NEXT: RET_ReallyLR
+    %0:gpr64common = COPY $x0
+    %1:fpr128 = MOVIv2d_ns 0
+    %2:fpr64 = UADDLVv4i32v %1
+    %4:fpr128 = IMPLICIT_DEF
+    %3:fpr128 = INSERT_SUBREG %4, killed %2, %subreg.dsub
+    %5:gpr64all = COPY %3.dsub
+    %6:gpr32 = COPY %5.sub_32
+    %7:fpr64 = MOVID 0
+    %9:fpr128 = IMPLICIT_DEF
+    %8:fpr128 = INSERT_SUBREG %9, killed %7, %subreg.dsub
+    %10:fpr128 = INSvi8gpr %8, 0, killed %6
+    %11:fpr64 = COPY %10.dsub
+    %13:fpr64 = IMPLICIT_DEF
+    %12:fpr64 = ZIP1v8i8 killed %11, killed %13
+    %14:fpr64 = BICv4i16 %12, 255, 8
+    %15:fpr128 = USHLLv4i16_shift killed %14, 0
+    %16:fpr128 = nofpexcept UCVTFv4f32 killed %15, implicit $fpcr
+    STRQui %1, %0, 3 :: (store (s128) into %ir.0 + 48, align 8)
+    STRQui %1, %0, 2 :: (store (s128) into %ir.0 + 32, align 8)
+    STRQui %1, %0, 1 :: (store (s128) into %ir.0 + 16, align 8)
+    STRQui killed %16, %0, 0 :: (store (s128) into %ir.0, align 8)
+    RET_ReallyLR
+
+...
+---
+name:            insert_vec_v2i32_uaddlv_from_v8i16_nz_index
+registers:
+  - { id: 0, class: gpr64common, preferred-register: '' }
+  - { id: 1, class: fpr128, preferred-register: '' }
+  - { id: 2, class: fpr32, preferred-register: '' }
+  - { id: 3, class: fpr128, preferred-register: '' }
+  - { id: 4, class: fpr128, preferred-register: '' }
+  - { id: 5, class: gpr32, preferred-register: '' }
+  - { id: 6, class: fpr128, preferred-register: '' }
+  - { id: 7, class: fpr128, preferred-register: '' }
+  - { id: 8, class: fpr128, preferred-register: '' }
+liveins:
+  - { reg: '$x0', virtual-reg: '%0' }
+body:             |
+  bb.0.entry:
+    liveins: $x0
+
+    ; CHECK-LABEL: name: insert_vec_v2i32_uaddlv_from_v8i16_nz_index
+    ; CHECK: liveins: $x0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x0
+    ; CHECK-NEXT: [[MOVIv2d_ns:%[0-9]+]]:fpr128 = MOVIv2d_ns 0
+    ; CHECK-NEXT: [[UADDLVv8i16v:%[0-9]+]]:fpr32 = UADDLVv8i16v [[MOVIv2d_ns]]
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:fpr128 = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:fpr128 = INSERT_SUBREG [[DEF]], killed [[UADDLVv8i16v]], %subreg.ssub
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY [[INSERT_SUBREG]].ssub
+    ; CHECK-NEXT: [[INSvi32gpr:%[0-9]+]]:fpr128 = INSvi32gpr [[MOVIv2d_ns]], 2, killed [[COPY1]]
+    ; CHECK-NEXT: [[UCVTFv4f32_:%[0-9]+]]:fpr128 = nofpexcept UCVTFv4f32 killed [[INSvi32gpr]], implicit $fpcr
+    ; CHECK-NEXT: STRQui killed [[UCVTFv4f32_]], [[COPY]], 0 :: (store (s128) into %ir.0, align 8)
+    ; CHECK-NEXT: RET_ReallyLR
+    %0:gpr64common = COPY $x0
+    %1:fpr128 = MOVIv2d_ns 0
+    %2:fpr32 = UADDLVv8i16v %1
+    %4:fpr128 = IMPLICIT_DEF
+    %3:fpr128 = INSERT_SUBREG %4, killed %2, %subreg.ssub
+    %5:gpr32 = COPY %3.ssub
+    %7:fpr128 = INSvi32gpr %1, 2, killed %5
+    %8:fpr128 = nofpexcept UCVTFv4f32 killed %7, implicit $fpcr
+    STRQui killed %8, %0, 0 :: (store (s128) into %ir.0, align 8)
+    RET_ReallyLR
+
+...
+---
+name:            insert_vec_from_gpr
+registers:
+  - { id: 0, class: gpr32, preferred-register: '' }
+  - { id: 1, class: gpr64common, preferred-register: '' }
+  - { id: 2, class: gpr64, preferred-register: '' }
+  - { id: 3, class: gpr64all, preferred-register: '' }
+  - { id: 4, class: gpr64common, preferred-register: '' }
+  - { id: 5, class: gpr64common, preferred-register: '' }
+  - { id: 6, class: fpr128, preferred-register: '' }
+  - { id: 7, class: fpr128, preferred-register: '' }
+liveins:
+  - { reg: '$w0', virtual-reg: '%0' }
+  - { reg: '$x1', virtual-reg: '%1' }
+stack:
+  - { id: 0, name: '', type: default, offset: 0, size: 16, alignment: 16,
+      stack-id: default, callee-saved-register: '', callee-saved-restored: true,
+      local-offset: -16, debug-info-variable: '', debug-info-expression: '',
+      debug-info-location: '' }
+body:             |
+  bb.0.entry:
+    liveins: $w0, $x1
+
+    ; CHECK-LABEL: name: insert_vec_from_gpr
+    ; CHECK: liveins: $w0, $x1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: [[COPY:%[0-9]+]]:gpr64common = COPY $x1
+    ; CHECK-NEXT: [[COPY1:%[0-9]+]]:gpr32 = COPY $w0
+    ; CHECK-NEXT: [[DEF:%[0-9]+]]:gpr64all = IMPLICIT_DEF
+    ; CHECK-NEXT: [[INSERT_SUBREG:%[0-9]+]]:gpr64 = INSERT_SUBREG [[DEF]], [[COPY1]], %subreg.sub_32
+    ; CHECK-NEXT: [[ADDXri:%[0-9]+]]:gpr64common = ADDXri %stack.0, 0, 0
+    ; CHECK-NEXT: [[BFMXri:%[0-9]+]]:gpr64common = BFMXri [[ADDXri]], killed [[INSERT_SUBREG]], 62, 1
+    ; CHECK-NEXT: STRWui [[COPY1]], killed [[BFMXri]], 0 :: (store (s32))
+    ; CHECK-NEXT: [[LDRQui:%[0-9]+]]:fpr128 = LDRQui %stack.0, 0 :: (load (s128) from %stack.0)
+    ; CHECK-NEXT: [[INSvi32gpr:%[0-9]+]]:fpr128 = INSvi32gpr [[LDRQui]], 1, [[COPY1]]
+    ; CHECK-NEXT: STRQui killed [[INSvi32gpr]], [[COPY]], 0 :: (store (s128) into %ir.p, align 4)
+    ; CHECK-NEXT: RET_ReallyLR
+    %1:gpr64common = COPY $x1
+    %0:gpr32 = COPY $w0
+    %3:gpr64all = IMPLICIT_DEF
+    %2:gpr64 = INSERT_SUBREG %3, %0, %subreg.sub_32
+    %4:gpr64common = ADDXri %stack.0, 0, 0
+    %5:gpr64common = BFMXri %4, killed %2, 62, 1
+    STRWui %0, killed %5, 0 :: (store (s32))
+    %6:fpr128 = LDRQui %stack.0, 0 :: (load (s128) from %stack.0)
+    %7:fpr128 = INSvi32gpr %6, 1, %0
+    STRQui killed %7, %1, 0 :: (store (s128) into %ir.p, align 4)
+    RET_ReallyLR
+
+...
+