Index: llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
===================================================================
--- llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
+++ llvm/lib/Target/AArch64/GISel/AArch64InstructionSelector.cpp
@@ -3994,6 +3994,8 @@
   };
 
   switch (EltSize) {
+  case 8:
+    return BuildFn(AArch64::bsub);
   case 16:
     return BuildFn(AArch64::hsub);
   case 32:
@@ -5543,7 +5545,7 @@
   if (tryOptBuildVecToSubregToReg(I, MRI))
     return true;
 
-  if (EltSize < 16 || EltSize > 64)
+  if (EltSize != 8 && EltSize != 16 && EltSize != 32 && EltSize != 64)
     return false; // Don't support all element types yet.
   const RegisterBank &RB = *RBI.getRegBank(I.getOperand(1).getReg(), MRI, TRI);
 
Index: llvm/test/CodeGen/AArch64/load-insert-zero.ll
===================================================================
--- llvm/test/CodeGen/AArch64/load-insert-zero.ll
+++ llvm/test/CodeGen/AArch64/load-insert-zero.ll
@@ -1,74 +1,161 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc < %s -mtriple=aarch64-none-eabi -mattr=+fullfp16,+bf16 | FileCheck %s
+; RUN: llc < %s -mtriple=aarch64-none-eabi -mattr=+fullfp16,+bf16 | FileCheck %s -check-prefixes=CHECK-SD
+; RUN: llc < %s -mtriple=aarch64-none-eabi -mattr=+fullfp16,+bf16 -global-isel | FileCheck %s -check-prefixes=CHECK-GI
 
 define <8 x i8> @loadv8i8(ptr %p) {
-; CHECK-LABEL: loadv8i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr b0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv8i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr b0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv8i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    mov w8, #0
+; CHECK-GI-NEXT:    ldr b0, [x0]
+; CHECK-GI-NEXT:    fmov s1, w8
+; CHECK-GI-NEXT:    mov v0.b[1], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[2], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[3], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[4], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[5], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[6], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[7], v1.b[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load i8, ptr %p
   %v = insertelement <8 x i8> zeroinitializer, i8 %l, i32 0
   ret <8 x i8> %v
 }
 
 define <16 x i8> @loadv4i8(ptr %p) {
-; CHECK-LABEL: loadv4i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr b0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr b0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    mov w8, #0
+; CHECK-GI-NEXT:    ldr b0, [x0]
+; CHECK-GI-NEXT:    fmov s1, w8
+; CHECK-GI-NEXT:    mov v0.b[1], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[2], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[3], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[4], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[5], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[6], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[7], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[8], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[9], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[10], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[11], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[12], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[13], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[14], v1.b[0]
+; CHECK-GI-NEXT:    mov v0.b[15], v1.b[0]
+; CHECK-GI-NEXT:    ret
   %l = load i8, ptr %p
   %v = insertelement <16 x i8> zeroinitializer, i8 %l, i32 0
   ret <16 x i8> %v
 }
 
 define <4 x i16> @loadv4i16(ptr %p) {
-; CHECK-LABEL: loadv4i16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4i16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4i16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    mov w8, #0
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    fmov s1, w8
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load i16, ptr %p
   %v = insertelement <4 x i16> zeroinitializer, i16 %l, i32 0
   ret <4 x i16> %v
 }
 
 define <8 x i16> @loadv8i16(ptr %p) {
-; CHECK-LABEL: loadv8i16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv8i16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv8i16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    mov w8, #0
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    fmov s1, w8
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[4], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[5], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[6], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[7], v1.h[0]
+; CHECK-GI-NEXT:    ret
   %l = load i16, ptr %p
   %v = insertelement <8 x i16> zeroinitializer, i16 %l, i32 0
   ret <8 x i16> %v
 }
 
 define <2 x i32> @loadv2i32(ptr %p) {
-; CHECK-LABEL: loadv2i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv2i32:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr s0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv2i32:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov s1, wzr
+; CHECK-GI-NEXT:    ldr s0, [x0]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load i32, ptr %p
   %v = insertelement <2 x i32> zeroinitializer, i32 %l, i32 0
   ret <2 x i32> %v
 }
 
 define <4 x i32> @loadv4i32(ptr %p) {
-; CHECK-LABEL: loadv4i32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4i32:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr s0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4i32:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov s1, wzr
+; CHECK-GI-NEXT:    ldr s0, [x0]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v1.s[0]
+; CHECK-GI-NEXT:    ret
   %l = load i32, ptr %p
   %v = insertelement <4 x i32> zeroinitializer, i32 %l, i32 0
   ret <4 x i32> %v
 }
 
 define <2 x i64> @loadv2i64(ptr %p) {
-; CHECK-LABEL: loadv2i64:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv2i64:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr d0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv2i64:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov d1, xzr
+; CHECK-GI-NEXT:    ldr d0, [x0]
+; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT:    ret
   %l = load i64, ptr %p
   %v = insertelement <2 x i64> zeroinitializer, i64 %l, i32 0
   ret <2 x i64> %v
@@ -76,74 +163,145 @@
 
 
 define <4 x half> @loadv4f16(ptr %p) {
-; CHECK-LABEL: loadv4f16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4f16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4f16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load half, ptr %p
   %v = insertelement <4 x half> zeroinitializer, half %l, i32 0
   ret <4 x half> %v
 }
 
 define <8 x half> @loadv8f16(ptr %p) {
-; CHECK-LABEL: loadv8f16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv8f16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv8f16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[4], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[5], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[6], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[7], v1.h[0]
+; CHECK-GI-NEXT:    ret
   %l = load half, ptr %p
   %v = insertelement <8 x half> zeroinitializer, half %l, i32 0
   ret <8 x half> %v
 }
 
 define <4 x bfloat> @loadv4bf16(ptr %p) {
-; CHECK-LABEL: loadv4bf16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4bf16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4bf16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load bfloat, ptr %p
   %v = insertelement <4 x bfloat> zeroinitializer, bfloat %l, i32 0
   ret <4 x bfloat> %v
 }
 
 define <8 x bfloat> @loadv8bf16(ptr %p) {
-; CHECK-LABEL: loadv8bf16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr h0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv8bf16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr h0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv8bf16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr h0, [x0]
+; CHECK-GI-NEXT:    mov v0.h[1], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[2], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[3], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[4], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[5], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[6], v1.h[0]
+; CHECK-GI-NEXT:    mov v0.h[7], v1.h[0]
+; CHECK-GI-NEXT:    ret
   %l = load bfloat, ptr %p
   %v = insertelement <8 x bfloat> zeroinitializer, bfloat %l, i32 0
   ret <8 x bfloat> %v
 }
 
 define <2 x float> @loadv2f32(ptr %p) {
-; CHECK-LABEL: loadv2f32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    // kill: def $d0 killed $d0 killed $q0
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv2f32:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr s0, [x0]
+; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv2f32:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr s0, [x0]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 killed $q0
+; CHECK-GI-NEXT:    ret
   %l = load float, ptr %p
   %v = insertelement <2 x float> zeroinitializer, float %l, i32 0
   ret <2 x float> %v
 }
 
 define <4 x float> @loadv4f32(ptr %p) {
-; CHECK-LABEL: loadv4f32:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr s0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv4f32:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr s0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv4f32:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr s0, [x0]
+; CHECK-GI-NEXT:    mov v0.s[1], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[2], v1.s[0]
+; CHECK-GI-NEXT:    mov v0.s[3], v1.s[0]
+; CHECK-GI-NEXT:    ret
   %l = load float, ptr %p
   %v = insertelement <4 x float> zeroinitializer, float %l, i32 0
   ret <4 x float> %v
 }
 
 define <2 x double> @loadv2f64(ptr %p) {
-; CHECK-LABEL: loadv2f64:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ldr d0, [x0]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: loadv2f64:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ldr d0, [x0]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: loadv2f64:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    movi d1, #0000000000000000
+; CHECK-GI-NEXT:    ldr d0, [x0]
+; CHECK-GI-NEXT:    mov v0.d[1], v1.d[0]
+; CHECK-GI-NEXT:    ret
   %l = load double, ptr %p
   %v = insertelement <2 x double> zeroinitializer, double %l, i32 0
   ret <2 x double> %v
 }
+