[AArch64][GlobalISel] Legalize Shifts for Smaller/Larger Vectors (#78750)

Legalize shl/lshr/ashr for smaller/larger vector widths with legal
element sizes

Smaller than legal vector types does not work at the moment as it relies
on G_ANYEXT to work with smaller than legal vector types

6 files changed, 412 insertions, 237 deletions

diff --git a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
index a868860..3b2cf31 100644
--- a/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
+++ b/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
@@ -5198,7 +5198,10 @@ LegalizerHelper::moreElementsVector(MachineInstr &MI, unsigned TypeIdx,
   case TargetOpcode::G_FMAXIMUM:
   case TargetOpcode::G_STRICT_FADD:
   case TargetOpcode::G_STRICT_FSUB:
-  case TargetOpcode::G_STRICT_FMUL: {
+  case TargetOpcode::G_STRICT_FMUL:
+  case TargetOpcode::G_SHL:
+  case TargetOpcode::G_ASHR:
+  case TargetOpcode::G_LSHR: {
     Observer.changingInstr(MI);
     moreElementsVectorSrc(MI, MoreTy, 1);
     moreElementsVectorSrc(MI, MoreTy, 2);
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index ae7ed08..fd69a7d 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -169,6 +169,8 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
       .widenScalarToNextPow2(0)
       .clampScalar(1, s32, s64)
       .clampScalar(0, s32, s64)
+      .clampNumElements(0, v8s8, v16s8)
+      .clampNumElements(0, v4s16, v8s16)
       .clampNumElements(0, v2s32, v4s32)
       .clampNumElements(0, v2s64, v2s64)
       .moreElementsToNextPow2(0)
diff --git a/llvm/test/CodeGen/AArch64/fcmp.ll b/llvm/test/CodeGen/AArch64/fcmp.ll
index 82e29d0..d3bc7fc 100644
--- a/llvm/test/CodeGen/AArch64/fcmp.ll
+++ b/llvm/test/CodeGen/AArch64/fcmp.ll
@@ -8,9 +8,9 @@
 ; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for v3f64_i32
 ; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for v3f32_float
 ; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for v3f32_i32
-; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for v7f16_half
+; CHECK-GI-FP16-NEXT:  warning: Instruction selection used fallback path for v7f16_half
 ; CHECK-GI-FP16-NEXT:  warning: Instruction selection used fallback path for v16f16_half
-; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for v7f16_i32
+; CHECK-GI-FP16-NEXT:  warning: Instruction selection used fallback path for v7f16_i32
 ; CHECK-GI-FP16-NEXT:  warning: Instruction selection used fallback path for v16f16_i32
 
 define double @f64_double(double %a, double %b, double %d, double %e) {
@@ -437,62 +437,87 @@ define <7 x half> @v7f16_half(<7 x half> %a, <7 x half> %b, <7 x half> %d, <7 x
 ;
 ; CHECK-GI-NOFP16-LABEL: v7f16_half:
 ; CHECK-GI-NOFP16:       // %bb.0: // %entry
-; CHECK-GI-NOFP16-NEXT:    mov h4, v1.h[1]
-; CHECK-GI-NOFP16-NEXT:    mov h5, v0.h[1]
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h1
-; CHECK-GI-NOFP16-NEXT:    fcvt s7, h0
-; CHECK-GI-NOFP16-NEXT:    mov h16, v1.h[2]
-; CHECK-GI-NOFP16-NEXT:    fcvt s4, h4
+; CHECK-GI-NOFP16-NEXT:    mov h4, v0.h[1]
+; CHECK-GI-NOFP16-NEXT:    mov h5, v1.h[1]
+; CHECK-GI-NOFP16-NEXT:    mov w9, #15 // =0xf
+; CHECK-GI-NOFP16-NEXT:    fcvt s6, h0
+; CHECK-GI-NOFP16-NEXT:    fcvt s7, h1
+; CHECK-GI-NOFP16-NEXT:    mov h16, v0.h[2]
+; CHECK-GI-NOFP16-NEXT:    mov h17, v1.h[2]
+; CHECK-GI-NOFP16-NEXT:    mov h19, v1.h[4]
+; CHECK-GI-NOFP16-NEXT:    mov h20, v1.h[5]
+; CHECK-GI-NOFP16-NEXT:    fcvt s18, h4
 ; CHECK-GI-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-GI-NOFP16-NEXT:    fcmp s5, s4
-; CHECK-GI-NOFP16-NEXT:    mov h4, v0.h[2]
-; CHECK-GI-NOFP16-NEXT:    mov h5, v1.h[3]
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s7, s6
+; CHECK-GI-NOFP16-NEXT:    mov h4, v0.h[3]
+; CHECK-GI-NOFP16-NEXT:    fcmp s6, s7
 ; CHECK-GI-NOFP16-NEXT:    fcvt s7, h16
-; CHECK-GI-NOFP16-NEXT:    fcvt s4, h4
-; CHECK-GI-NOFP16-NEXT:    mov h6, v0.h[3]
-; CHECK-GI-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-GI-NOFP16-NEXT:    mov h16, v0.h[4]
-; CHECK-GI-NOFP16-NEXT:    csetm w9, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s4, s7
-; CHECK-GI-NOFP16-NEXT:    fmov s4, w9
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-GI-NOFP16-NEXT:    mov h7, v1.h[4]
-; CHECK-GI-NOFP16-NEXT:    fcvt s16, h16
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[1], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s6, s5
-; CHECK-GI-NOFP16-NEXT:    mov h5, v1.h[5]
-; CHECK-GI-NOFP16-NEXT:    mov h6, v0.h[5]
-; CHECK-GI-NOFP16-NEXT:    fcvt s7, h7
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[2], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-GI-NOFP16-NEXT:    fcmp s16, s7
-; CHECK-GI-NOFP16-NEXT:    mov h7, v1.h[6]
-; CHECK-GI-NOFP16-NEXT:    mov h16, v0.h[6]
-; CHECK-GI-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-GI-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[3], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s6, s5
-; CHECK-GI-NOFP16-NEXT:    fcvt s5, h7
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h16
+; CHECK-GI-NOFP16-NEXT:    fmov s6, w9
+; CHECK-GI-NOFP16-NEXT:    fcvt s16, h17
+; CHECK-GI-NOFP16-NEXT:    mov h17, v1.h[3]
+; CHECK-GI-NOFP16-NEXT:    fcvt s19, h19
+; CHECK-GI-NOFP16-NEXT:    mov h1, v1.h[6]
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s18, s5
+; CHECK-GI-NOFP16-NEXT:    mov h18, v0.h[4]
+; CHECK-GI-NOFP16-NEXT:    fmov s5, w8
+; CHECK-GI-NOFP16-NEXT:    cset w9, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s7, s16
+; CHECK-GI-NOFP16-NEXT:    fcvt s7, h4
+; CHECK-GI-NOFP16-NEXT:    fcvt s16, h17
+; CHECK-GI-NOFP16-NEXT:    fmov s17, w9
+; CHECK-GI-NOFP16-NEXT:    mov v4.16b, v6.16b
+; CHECK-GI-NOFP16-NEXT:    fcvt s18, h18
 ; CHECK-GI-NOFP16-NEXT:    fcvt s1, h1
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[1], v17.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[1], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov h17, v0.h[5]
+; CHECK-GI-NOFP16-NEXT:    fcmp s7, s16
+; CHECK-GI-NOFP16-NEXT:    fmov s7, w8
+; CHECK-GI-NOFP16-NEXT:    mov w8, #65535 // =0xffff
+; CHECK-GI-NOFP16-NEXT:    mov h0, v0.h[6]
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[2], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    fmov s7, w8
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[2], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s18, s19
+; CHECK-GI-NOFP16-NEXT:    fcvt s17, h17
+; CHECK-GI-NOFP16-NEXT:    fcvt s18, h20
+; CHECK-GI-NOFP16-NEXT:    fmov s19, w8
 ; CHECK-GI-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[4], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s6, s5
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[5], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
+; CHECK-GI-NOFP16-NEXT:    mov v16.16b, v7.16b
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[3], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[3], v19.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[1], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    fcmp s17, s18
+; CHECK-GI-NOFP16-NEXT:    fmov s17, w8
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[4], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[4], v17.h[0]
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
 ; CHECK-GI-NOFP16-NEXT:    fcmp s0, s1
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[6], w8
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[7], w8
-; CHECK-GI-NOFP16-NEXT:    mov v0.16b, v4.16b
-; CHECK-GI-NOFP16-NEXT:    bsl v0.16b, v2.16b, v3.16b
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[2], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    fmov s0, w8
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[5], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[5], v0.h[0]
+; CHECK-GI-NOFP16-NEXT:    fmov s0, w8
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[3], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[6], v6.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[6], v0.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[4], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v4.h[7], v0.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v5.h[7], v0.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[5], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    neg v0.8h, v4.8h
+; CHECK-GI-NOFP16-NEXT:    ushl v1.8h, v5.8h, v4.8h
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[6], v7.h[0]
+; CHECK-GI-NOFP16-NEXT:    mov v16.h[7], v0.h[0]
+; CHECK-GI-NOFP16-NEXT:    sshl v0.8h, v1.8h, v0.8h
+; CHECK-GI-NOFP16-NEXT:    eor v1.16b, v0.16b, v16.16b
+; CHECK-GI-NOFP16-NEXT:    and v0.16b, v2.16b, v0.16b
+; CHECK-GI-NOFP16-NEXT:    and v1.16b, v3.16b, v1.16b
+; CHECK-GI-NOFP16-NEXT:    orr v0.16b, v0.16b, v1.16b
 ; CHECK-GI-NOFP16-NEXT:    ret
 ;
 ; CHECK-GI-FP16-LABEL: v7f16_half:
@@ -1112,90 +1137,110 @@ define <7 x i32> @v7f16_i32(<7 x half> %a, <7 x half> %b, <7 x i32> %d, <7 x i32
 ;
 ; CHECK-GI-NOFP16-LABEL: v7f16_i32:
 ; CHECK-GI-NOFP16:       // %bb.0: // %entry
-; CHECK-GI-NOFP16-NEXT:    mov h2, v1.h[1]
-; CHECK-GI-NOFP16-NEXT:    mov h3, v0.h[1]
-; CHECK-GI-NOFP16-NEXT:    mov h4, v1.h[2]
-; CHECK-GI-NOFP16-NEXT:    mov h5, v0.h[2]
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h1
-; CHECK-GI-NOFP16-NEXT:    fcvt s7, h0
+; CHECK-GI-NOFP16-NEXT:    mov h2, v0.h[1]
+; CHECK-GI-NOFP16-NEXT:    mov h3, v1.h[1]
+; CHECK-GI-NOFP16-NEXT:    mov w13, #31 // =0x1f
+; CHECK-GI-NOFP16-NEXT:    fcvt s4, h0
+; CHECK-GI-NOFP16-NEXT:    fcvt s5, h1
+; CHECK-GI-NOFP16-NEXT:    ldr s17, [sp, #40]
+; CHECK-GI-NOFP16-NEXT:    mov h6, v0.h[2]
+; CHECK-GI-NOFP16-NEXT:    mov h7, v1.h[2]
+; CHECK-GI-NOFP16-NEXT:    fmov s16, w0
 ; CHECK-GI-NOFP16-NEXT:    fcvt s2, h2
 ; CHECK-GI-NOFP16-NEXT:    fcvt s3, h3
+; CHECK-GI-NOFP16-NEXT:    fcmp s4, s5
+; CHECK-GI-NOFP16-NEXT:    mov h4, v0.h[3]
+; CHECK-GI-NOFP16-NEXT:    mov h5, v1.h[3]
+; CHECK-GI-NOFP16-NEXT:    fcvt s6, h6
+; CHECK-GI-NOFP16-NEXT:    fcvt s7, h7
+; CHECK-GI-NOFP16-NEXT:    mov v16.s[1], w1
+; CHECK-GI-NOFP16-NEXT:    cset w9, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s2, s3
+; CHECK-GI-NOFP16-NEXT:    mov h2, v0.h[4]
+; CHECK-GI-NOFP16-NEXT:    mov h3, v1.h[4]
 ; CHECK-GI-NOFP16-NEXT:    fcvt s4, h4
 ; CHECK-GI-NOFP16-NEXT:    fcvt s5, h5
-; CHECK-GI-NOFP16-NEXT:    fcmp s3, s2
-; CHECK-GI-NOFP16-NEXT:    mov h2, v1.h[3]
-; CHECK-GI-NOFP16-NEXT:    mov h3, v0.h[3]
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s7, s6
-; CHECK-GI-NOFP16-NEXT:    mov h6, v1.h[5]
-; CHECK-GI-NOFP16-NEXT:    mov h7, v0.h[5]
+; CHECK-GI-NOFP16-NEXT:    mov v16.s[2], w2
+; CHECK-GI-NOFP16-NEXT:    cset w10, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s6, s7
+; CHECK-GI-NOFP16-NEXT:    mov h6, v0.h[5]
+; CHECK-GI-NOFP16-NEXT:    mov h7, v1.h[5]
 ; CHECK-GI-NOFP16-NEXT:    fcvt s2, h2
+; CHECK-GI-NOFP16-NEXT:    mov h0, v0.h[6]
 ; CHECK-GI-NOFP16-NEXT:    fcvt s3, h3
-; CHECK-GI-NOFP16-NEXT:    csetm w9, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s5, s4
-; CHECK-GI-NOFP16-NEXT:    mov h4, v1.h[4]
-; CHECK-GI-NOFP16-NEXT:    mov h5, v0.h[4]
-; CHECK-GI-NOFP16-NEXT:    fcvt s6, h6
-; CHECK-GI-NOFP16-NEXT:    fcvt s7, h7
-; CHECK-GI-NOFP16-NEXT:    csetm w10, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s3, s2
-; CHECK-GI-NOFP16-NEXT:    fcvt s2, h4
-; CHECK-GI-NOFP16-NEXT:    mov h4, v1.h[6]
-; CHECK-GI-NOFP16-NEXT:    mov h1, v1.h[7]
-; CHECK-GI-NOFP16-NEXT:    fcvt s3, h5
-; CHECK-GI-NOFP16-NEXT:    mov h5, v0.h[6]
-; CHECK-GI-NOFP16-NEXT:    mov h0, v0.h[7]
-; CHECK-GI-NOFP16-NEXT:    csetm w11, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s7, s6
-; CHECK-GI-NOFP16-NEXT:    fcvt s1, h1
-; CHECK-GI-NOFP16-NEXT:    csetm w12, mi
-; CHECK-GI-NOFP16-NEXT:    fcmp s3, s2
-; CHECK-GI-NOFP16-NEXT:    fcvt s2, h4
-; CHECK-GI-NOFP16-NEXT:    fcvt s3, h5
-; CHECK-GI-NOFP16-NEXT:    fmov s4, w9
+; CHECK-GI-NOFP16-NEXT:    mov h1, v1.h[6]
+; CHECK-GI-NOFP16-NEXT:    cset w11, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s4, s5
+; CHECK-GI-NOFP16-NEXT:    mov v16.s[3], w3
+; CHECK-GI-NOFP16-NEXT:    fcvt s4, h6
+; CHECK-GI-NOFP16-NEXT:    ldr s6, [sp, #32]
+; CHECK-GI-NOFP16-NEXT:    fcvt s5, h7
 ; CHECK-GI-NOFP16-NEXT:    fcvt s0, h0
-; CHECK-GI-NOFP16-NEXT:    add x9, sp, #8
-; CHECK-GI-NOFP16-NEXT:    csetm w13, mi
-; CHECK-GI-NOFP16-NEXT:    fmov s5, w13
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[1], w8
-; CHECK-GI-NOFP16-NEXT:    mov x8, sp
-; CHECK-GI-NOFP16-NEXT:    fcmp s3, s2
-; CHECK-GI-NOFP16-NEXT:    fmov s2, w7
-; CHECK-GI-NOFP16-NEXT:    fmov s3, w0
-; CHECK-GI-NOFP16-NEXT:    mov v5.h[1], w12
-; CHECK-GI-NOFP16-NEXT:    ld1 { v2.s }[1], [x8]
-; CHECK-GI-NOFP16-NEXT:    mov v3.s[1], w1
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[2], w10
+; CHECK-GI-NOFP16-NEXT:    fmov s7, w4
+; CHECK-GI-NOFP16-NEXT:    cset w8, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s2, s3
+; CHECK-GI-NOFP16-NEXT:    fcvt s1, h1
+; CHECK-GI-NOFP16-NEXT:    fmov s2, w13
+; CHECK-GI-NOFP16-NEXT:    mov v7.s[1], w5
+; CHECK-GI-NOFP16-NEXT:    cset w12, mi
+; CHECK-GI-NOFP16-NEXT:    fcmp s4, s5
+; CHECK-GI-NOFP16-NEXT:    ldr s5, [sp]
+; CHECK-GI-NOFP16-NEXT:    fmov s3, w12
+; CHECK-GI-NOFP16-NEXT:    mov v2.s[1], w13
+; CHECK-GI-NOFP16-NEXT:    cset w14, mi
 ; CHECK-GI-NOFP16-NEXT:    fcmp s0, s1
-; CHECK-GI-NOFP16-NEXT:    fmov s1, w4
-; CHECK-GI-NOFP16-NEXT:    ldr s0, [sp, #24]
-; CHECK-GI-NOFP16-NEXT:    mov v5.h[2], w8
-; CHECK-GI-NOFP16-NEXT:    ld1 { v2.s }[2], [x9]
-; CHECK-GI-NOFP16-NEXT:    add x9, sp, #32
-; CHECK-GI-NOFP16-NEXT:    mov v3.s[2], w2
-; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], w5
-; CHECK-GI-NOFP16-NEXT:    csetm w8, mi
-; CHECK-GI-NOFP16-NEXT:    mov v4.h[3], w11
-; CHECK-GI-NOFP16-NEXT:    ld1 { v0.s }[1], [x9]
-; CHECK-GI-NOFP16-NEXT:    mov v5.h[3], w8
-; CHECK-GI-NOFP16-NEXT:    add x8, sp, #16
-; CHECK-GI-NOFP16-NEXT:    ld1 { v2.s }[3], [x8]
-; CHECK-GI-NOFP16-NEXT:    mov v3.s[3], w3
-; CHECK-GI-NOFP16-NEXT:    add x8, sp, #40
-; CHECK-GI-NOFP16-NEXT:    mov v1.s[2], w6
-; CHECK-GI-NOFP16-NEXT:    sshll v4.4s, v4.4h, #0
-; CHECK-GI-NOFP16-NEXT:    ld1 { v0.s }[2], [x8]
-; CHECK-GI-NOFP16-NEXT:    sshll v5.4s, v5.4h, #0
-; CHECK-GI-NOFP16-NEXT:    bit v2.16b, v3.16b, v4.16b
-; CHECK-GI-NOFP16-NEXT:    bit v0.16b, v1.16b, v5.16b
-; CHECK-GI-NOFP16-NEXT:    mov w1, v2.s[1]
-; CHECK-GI-NOFP16-NEXT:    mov w2, v2.s[2]
-; CHECK-GI-NOFP16-NEXT:    mov w3, v2.s[3]
-; CHECK-GI-NOFP16-NEXT:    fmov w0, s2
-; CHECK-GI-NOFP16-NEXT:    mov w5, v0.s[1]
-; CHECK-GI-NOFP16-NEXT:    mov w6, v0.s[2]
-; CHECK-GI-NOFP16-NEXT:    fmov w4, s0
+; CHECK-GI-NOFP16-NEXT:    fmov s0, w9
+; CHECK-GI-NOFP16-NEXT:    mov v3.s[1], w14
+; CHECK-GI-NOFP16-NEXT:    ldr s1, [sp, #24]
+; CHECK-GI-NOFP16-NEXT:    mov v7.s[2], w6
+; CHECK-GI-NOFP16-NEXT:    mov v2.s[2], w13
+; CHECK-GI-NOFP16-NEXT:    cset w9, mi
+; CHECK-GI-NOFP16-NEXT:    mov v0.s[1], w10
+; CHECK-GI-NOFP16-NEXT:    fmov w10, s5
+; CHECK-GI-NOFP16-NEXT:    fmov s5, w7
+; CHECK-GI-NOFP16-NEXT:    mov v1.s[1], v6.s[0]
+; CHECK-GI-NOFP16-NEXT:    ldr s6, [sp, #8]
+; CHECK-GI-NOFP16-NEXT:    mov v3.s[2], w9
+; CHECK-GI-NOFP16-NEXT:    mov w9, #-1 // =0xffffffff
+; CHECK-GI-NOFP16-NEXT:    fmov s4, w9
+; CHECK-GI-NOFP16-NEXT:    mov v2.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    mov v0.s[2], w11
+; CHECK-GI-NOFP16-NEXT:    mov v5.s[1], w10
+; CHECK-GI-NOFP16-NEXT:    mov v1.s[2], v17.s[0]
+; CHECK-GI-NOFP16-NEXT:    mov v4.s[1], w9
+; CHECK-GI-NOFP16-NEXT:    mov v3.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    neg v18.4s, v2.4s
+; CHECK-GI-NOFP16-NEXT:    mov v0.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    fmov w8, s6
+; CHECK-GI-NOFP16-NEXT:    mov v4.s[2], w9
+; CHECK-GI-NOFP16-NEXT:    ushl v2.4s, v3.4s, v2.4s
+; CHECK-GI-NOFP16-NEXT:    ldr s3, [sp, #16]
+; CHECK-GI-NOFP16-NEXT:    mov v5.s[2], w8
+; CHECK-GI-NOFP16-NEXT:    mov v7.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    shl v0.4s, v0.4s, #31
+; CHECK-GI-NOFP16-NEXT:    sshl v2.4s, v2.4s, v18.4s
+; CHECK-GI-NOFP16-NEXT:    mov v4.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    fmov w8, s3
+; CHECK-GI-NOFP16-NEXT:    mov v1.s[3], v0.s[0]
+; CHECK-GI-NOFP16-NEXT:    sshr v0.4s, v0.4s, #31
+; CHECK-GI-NOFP16-NEXT:    mov v5.s[3], w8
+; CHECK-GI-NOFP16-NEXT:    eor v3.16b, v2.16b, v4.16b
+; CHECK-GI-NOFP16-NEXT:    and v2.16b, v7.16b, v2.16b
+; CHECK-GI-NOFP16-NEXT:    and v1.16b, v1.16b, v3.16b
+; CHECK-GI-NOFP16-NEXT:    bsl v0.16b, v16.16b, v5.16b
+; CHECK-GI-NOFP16-NEXT:    orr v1.16b, v2.16b, v1.16b
+; CHECK-GI-NOFP16-NEXT:    mov s2, v0.s[1]
+; CHECK-GI-NOFP16-NEXT:    mov s3, v0.s[2]
+; CHECK-GI-NOFP16-NEXT:    mov s4, v0.s[3]
+; CHECK-GI-NOFP16-NEXT:    fmov w0, s0
+; CHECK-GI-NOFP16-NEXT:    mov s5, v1.s[1]
+; CHECK-GI-NOFP16-NEXT:    mov s6, v1.s[2]
+; CHECK-GI-NOFP16-NEXT:    fmov w4, s1
+; CHECK-GI-NOFP16-NEXT:    fmov w1, s2
+; CHECK-GI-NOFP16-NEXT:    fmov w2, s3
+; CHECK-GI-NOFP16-NEXT:    fmov w3, s4
+; CHECK-GI-NOFP16-NEXT:    fmov w5, s5
+; CHECK-GI-NOFP16-NEXT:    fmov w6, s6
 ; CHECK-GI-NOFP16-NEXT:    ret
 ;
 ; CHECK-GI-FP16-LABEL: v7f16_i32:
diff --git a/llvm/test/CodeGen/AArch64/icmp.ll b/llvm/test/CodeGen/AArch64/icmp.ll
index 26711ea..789bc99 100644
--- a/llvm/test/CodeGen/AArch64/icmp.ll
+++ b/llvm/test/CodeGen/AArch64/icmp.ll
@@ -2,8 +2,6 @@
 ; RUN: llc -mtriple=aarch64-none-eabi -verify-machineinstrs %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-SD
 ; RUN: llc -mtriple=aarch64-none-eabi -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
 
-; CHECK-GI:  warning: Instruction selection used fallback path for v3i32_i32
-
 define i64 @i64_i64(i64 %a, i64 %b, i64 %d, i64 %e) {
 ; CHECK-LABEL: i64_i64:
 ; CHECK:       // %bb.0: // %entry
@@ -165,11 +163,33 @@ entry:
 }
 
 define <3 x i32> @v3i32_i32(<3 x i32> %a, <3 x i32> %b, <3 x i32> %d, <3 x i32> %e) {
-; CHECK-LABEL: v3i32_i32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    cmgt v0.4s, v1.4s, v0.4s
-; CHECK-NEXT:    bsl v0.16b, v2.16b, v3.16b
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: v3i32_i32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    cmgt v0.4s, v1.4s, v0.4s
+; CHECK-SD-NEXT:    bsl v0.16b, v2.16b, v3.16b
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: v3i32_i32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov w8, #31 // =0x1f
+; CHECK-GI-NEXT:    cmgt v0.4s, v1.4s, v0.4s
+; CHECK-GI-NEXT:    fmov s4, w8
+; CHECK-GI-NEXT:    mov v4.s[1], w8
+; CHECK-GI-NEXT:    mov v4.s[2], w8
+; CHECK-GI-NEXT:    mov w8, #-1 // =0xffffffff
+; CHECK-GI-NEXT:    fmov s5, w8
+; CHECK-GI-NEXT:    mov v5.s[1], w8
+; CHECK-GI-NEXT:    mov v4.s[3], w8
+; CHECK-GI-NEXT:    mov v5.s[2], w8
+; CHECK-GI-NEXT:    neg v1.4s, v4.4s
+; CHECK-GI-NEXT:    ushl v0.4s, v0.4s, v4.4s
+; CHECK-GI-NEXT:    mov v5.s[3], w8
+; CHECK-GI-NEXT:    sshl v0.4s, v0.4s, v1.4s
+; CHECK-GI-NEXT:    eor v1.16b, v0.16b, v5.16b
+; CHECK-GI-NEXT:    and v0.16b, v2.16b, v0.16b
+; CHECK-GI-NEXT:    and v1.16b, v3.16b, v1.16b
+; CHECK-GI-NEXT:    orr v0.16b, v0.16b, v1.16b
+; CHECK-GI-NEXT:    ret
 entry:
   %c = icmp slt <3 x i32> %a, %b
   %s = select <3 x i1> %c, <3 x i32> %d, <3 x i32> %e
diff --git a/llvm/test/CodeGen/AArch64/sext.ll b/llvm/test/CodeGen/AArch64/sext.ll
index dd53780..f319721 100644
--- a/llvm/test/CodeGen/AArch64/sext.ll
+++ b/llvm/test/CodeGen/AArch64/sext.ll
@@ -2,9 +2,6 @@
 ; RUN: llc -mtriple=aarch64 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-SD
 ; RUN: llc -mtriple=aarch64 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
 
-; CHECK-GI:       warning: Instruction selection used fallback path for sext_v3i8_v3i32
-; CHECK-GI-NEXT:  warning: Instruction selection used fallback path for sext_v3i10_v3i32
-
 define i16 @sext_i8_to_i16(i8 %a) {
 ; CHECK-LABEL: sext_i8_to_i16:
 ; CHECK:       // %bb.0: // %entry
@@ -236,15 +233,31 @@ entry:
 }
 
 define <3 x i32> @sext_v3i8_v3i32(<3 x i8> %a) {
-; CHECK-LABEL: sext_v3i8_v3i32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fmov s0, w0
-; CHECK-NEXT:    mov v0.h[1], w1
-; CHECK-NEXT:    mov v0.h[2], w2
-; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    shl v0.4s, v0.4s, #24
-; CHECK-NEXT:    sshr v0.4s, v0.4s, #24
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: sext_v3i8_v3i32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    fmov s0, w0
+; CHECK-SD-NEXT:    mov v0.h[1], w1
+; CHECK-SD-NEXT:    mov v0.h[2], w2
+; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-SD-NEXT:    shl v0.4s, v0.4s, #24
+; CHECK-SD-NEXT:    sshr v0.4s, v0.4s, #24
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: sext_v3i8_v3i32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov w8, #24 // =0x18
+; CHECK-GI-NEXT:    fmov s1, w0
+; CHECK-GI-NEXT:    fmov s0, w8
+; CHECK-GI-NEXT:    mov v1.s[1], w1
+; CHECK-GI-NEXT:    mov v0.s[1], w8
+; CHECK-GI-NEXT:    mov v1.s[2], w2
+; CHECK-GI-NEXT:    mov v0.s[2], w8
+; CHECK-GI-NEXT:    mov v1.s[3], w8
+; CHECK-GI-NEXT:    mov v0.s[3], w8
+; CHECK-GI-NEXT:    neg v2.4s, v0.4s
+; CHECK-GI-NEXT:    ushl v0.4s, v1.4s, v0.4s
+; CHECK-GI-NEXT:    sshl v0.4s, v0.4s, v2.4s
+; CHECK-GI-NEXT:    ret
 entry:
   %c = sext <3 x i8> %a to <3 x i32>
   ret <3 x i32> %c
@@ -388,15 +401,31 @@ entry:
 }
 
 define <3 x i32> @sext_v3i10_v3i32(<3 x i10> %a) {
-; CHECK-LABEL: sext_v3i10_v3i32:
-; CHECK:       // %bb.0: // %entry
-; CHECK-NEXT:    fmov s0, w0
-; CHECK-NEXT:    mov v0.h[1], w1
-; CHECK-NEXT:    mov v0.h[2], w2
-; CHECK-NEXT:    ushll v0.4s, v0.4h, #0
-; CHECK-NEXT:    shl v0.4s, v0.4s, #22
-; CHECK-NEXT:    sshr v0.4s, v0.4s, #22
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: sext_v3i10_v3i32:
+; CHECK-SD:       // %bb.0: // %entry
+; CHECK-SD-NEXT:    fmov s0, w0
+; CHECK-SD-NEXT:    mov v0.h[1], w1
+; CHECK-SD-NEXT:    mov v0.h[2], w2
+; CHECK-SD-NEXT:    ushll v0.4s, v0.4h, #0
+; CHECK-SD-NEXT:    shl v0.4s, v0.4s, #22
+; CHECK-SD-NEXT:    sshr v0.4s, v0.4s, #22
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: sext_v3i10_v3i32:
+; CHECK-GI:       // %bb.0: // %entry
+; CHECK-GI-NEXT:    mov w8, #22 // =0x16
+; CHECK-GI-NEXT:    fmov s1, w0
+; CHECK-GI-NEXT:    fmov s0, w8
+; CHECK-GI-NEXT:    mov v1.s[1], w1
+; CHECK-GI-NEXT:    mov v0.s[1], w8
+; CHECK-GI-NEXT:    mov v1.s[2], w2
+; CHECK-GI-NEXT:    mov v0.s[2], w8
+; CHECK-GI-NEXT:    mov v1.s[3], w8
+; CHECK-GI-NEXT:    mov v0.s[3], w8
+; CHECK-GI-NEXT:    neg v2.4s, v0.4s
+; CHECK-GI-NEXT:    ushl v0.4s, v1.4s, v0.4s
+; CHECK-GI-NEXT:    sshl v0.4s, v0.4s, v2.4s
+; CHECK-GI-NEXT:    ret
 entry:
   %c = sext <3 x i10> %a to <3 x i32>
   ret <3 x i32> %c
diff --git a/llvm/test/CodeGen/AArch64/shift.ll b/llvm/test/CodeGen/AArch64/shift.ll
index 55ff975..15c8e17 100644
--- a/llvm/test/CodeGen/AArch64/shift.ll
+++ b/llvm/test/CodeGen/AArch64/shift.ll
@@ -3,33 +3,11 @@
 ; RUN: llc -mtriple=aarch64 -global-isel -global-isel-abort=2 -verify-machineinstrs %s -o - 2>&1 | FileCheck %s --check-prefixes=CHECK,CHECK-GI
 
 ; CHECK-GI:         warning: Instruction selection used fallback path for shl_v4i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v32i8
 ; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v2i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v16i16
 ; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v4i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v32i8
 ; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v2i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v16i16
 ; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v4i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v32i8
 ; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v2i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v16i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v3i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v7i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v3i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v7i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for shl_v3i32
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v3i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v7i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v3i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v7i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for ashr_v3i32
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v3i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v7i8
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v3i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v7i16
-; CHECK-GI-NEXT:    warning: Instruction selection used fallback path for lshr_v3i32
-
 
 define i1 @shl_i1(i1 %0, i1 %1){
 ; CHECK-SD-LABEL: shl_i1:
@@ -562,11 +540,17 @@ define <4 x i8> @shl_v4i8(<4 x i8> %0, <4 x i8> %1){
 }
 
 define <32 x i8> @shl_v32i8(<32 x i8> %0, <32 x i8> %1){
-; CHECK-LABEL: shl_v32i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushl v1.16b, v1.16b, v3.16b
-; CHECK-NEXT:    ushl v0.16b, v0.16b, v2.16b
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: shl_v32i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ushl v1.16b, v1.16b, v3.16b
+; CHECK-SD-NEXT:    ushl v0.16b, v0.16b, v2.16b
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: shl_v32i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    ushl v0.16b, v0.16b, v2.16b
+; CHECK-GI-NEXT:    ushl v1.16b, v1.16b, v3.16b
+; CHECK-GI-NEXT:    ret
     %3 = shl <32 x i8> %0, %1
     ret <32 x i8> %3
 }
@@ -583,11 +567,17 @@ define <2 x i16> @shl_v2i16(<2 x i16> %0, <2 x i16> %1){
 }
 
 define <16 x i16> @shl_v16i16(<16 x i16> %0, <16 x i16> %1){
-; CHECK-LABEL: shl_v16i16:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    ushl v1.8h, v1.8h, v3.8h
-; CHECK-NEXT:    ushl v0.8h, v0.8h, v2.8h
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: shl_v16i16:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    ushl v1.8h, v1.8h, v3.8h
+; CHECK-SD-NEXT:    ushl v0.8h, v0.8h, v2.8h
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: shl_v16i16:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    ushl v0.8h, v0.8h, v2.8h
+; CHECK-GI-NEXT:    ushl v1.8h, v1.8h, v3.8h
+; CHECK-GI-NEXT:    ret
     %3 = shl <16 x i16> %0, %1
     ret <16 x i16> %3
 }
@@ -834,20 +824,48 @@ define <4 x i64> @lshr_v4i64(<4 x i64> %0, <4 x i64> %1){
 ; ===== Vector with Non-Pow 2 Width =====
 
 define <3 x i8> @shl_v3i8(<3 x i8> %0, <3 x i8> %1){
-; CHECK-LABEL: shl_v3i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fmov s0, w3
-; CHECK-NEXT:    fmov s1, w0
-; CHECK-NEXT:    mov v0.h[1], w4
-; CHECK-NEXT:    mov v1.h[1], w1
-; CHECK-NEXT:    mov v0.h[2], w5
-; CHECK-NEXT:    mov v1.h[2], w2
-; CHECK-NEXT:    bic v0.4h, #255, lsl #8
-; CHECK-NEXT:    ushl v0.4h, v1.4h, v0.4h
-; CHECK-NEXT:    umov w0, v0.h[0]
-; CHECK-NEXT:    umov w1, v0.h[1]
-; CHECK-NEXT:    umov w2, v0.h[2]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: shl_v3i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    fmov s0, w3
+; CHECK-SD-NEXT:    fmov s1, w0
+; CHECK-SD-NEXT:    mov v0.h[1], w4
+; CHECK-SD-NEXT:    mov v1.h[1], w1
+; CHECK-SD-NEXT:    mov v0.h[2], w5
+; CHECK-SD-NEXT:    mov v1.h[2], w2
+; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
+; CHECK-SD-NEXT:    ushl v0.4h, v1.4h, v0.4h
+; CHECK-SD-NEXT:    umov w0, v0.h[0]
+; CHECK-SD-NEXT:    umov w1, v0.h[1]
+; CHECK-SD-NEXT:    umov w2, v0.h[2]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: shl_v3i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov s0, w0
+; CHECK-GI-NEXT:    fmov s1, w1
+; CHECK-GI-NEXT:    fmov s2, w3
+; CHECK-GI-NEXT:    fmov s3, w4
+; CHECK-GI-NEXT:    mov v0.b[1], v1.b[0]
+; CHECK-GI-NEXT:    fmov s1, w2
+; CHECK-GI-NEXT:    mov v2.b[1], v3.b[0]
+; CHECK-GI-NEXT:    fmov s3, w5
+; CHECK-GI-NEXT:    mov v0.b[2], v1.b[0]
+; CHECK-GI-NEXT:    mov v2.b[2], v3.b[0]
+; CHECK-GI-NEXT:    mov v0.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v2.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v2.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v2.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v2.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[7], v0.b[0]
+; CHECK-GI-NEXT:    mov v2.b[7], v0.b[0]
+; CHECK-GI-NEXT:    ushl v0.8b, v0.8b, v2.8b
+; CHECK-GI-NEXT:    umov w0, v0.b[0]
+; CHECK-GI-NEXT:    umov w1, v0.b[1]
+; CHECK-GI-NEXT:    umov w2, v0.b[2]
+; CHECK-GI-NEXT:    ret
     %3 = shl <3 x i8> %0, %1
     ret <3 x i8> %3
 }
@@ -889,23 +907,52 @@ define <3 x i32> @shl_v3i32(<3 x i32> %0, <3 x i32> %1){
 }
 
 define <3 x i8> @ashr_v3i8(<3 x i8> %0, <3 x i8> %1){
-; CHECK-LABEL: ashr_v3i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fmov s0, w0
-; CHECK-NEXT:    fmov s1, w3
-; CHECK-NEXT:    mov v0.h[1], w1
-; CHECK-NEXT:    mov v1.h[1], w4
-; CHECK-NEXT:    mov v0.h[2], w2
-; CHECK-NEXT:    mov v1.h[2], w5
-; CHECK-NEXT:    shl v0.4h, v0.4h, #8
-; CHECK-NEXT:    bic v1.4h, #255, lsl #8
-; CHECK-NEXT:    sshr v0.4h, v0.4h, #8
-; CHECK-NEXT:    neg v1.4h, v1.4h
-; CHECK-NEXT:    sshl v0.4h, v0.4h, v1.4h
-; CHECK-NEXT:    umov w0, v0.h[0]
-; CHECK-NEXT:    umov w1, v0.h[1]
-; CHECK-NEXT:    umov w2, v0.h[2]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: ashr_v3i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    fmov s0, w0
+; CHECK-SD-NEXT:    fmov s1, w3
+; CHECK-SD-NEXT:    mov v0.h[1], w1
+; CHECK-SD-NEXT:    mov v1.h[1], w4
+; CHECK-SD-NEXT:    mov v0.h[2], w2
+; CHECK-SD-NEXT:    mov v1.h[2], w5
+; CHECK-SD-NEXT:    shl v0.4h, v0.4h, #8
+; CHECK-SD-NEXT:    bic v1.4h, #255, lsl #8
+; CHECK-SD-NEXT:    sshr v0.4h, v0.4h, #8
+; CHECK-SD-NEXT:    neg v1.4h, v1.4h
+; CHECK-SD-NEXT:    sshl v0.4h, v0.4h, v1.4h
+; CHECK-SD-NEXT:    umov w0, v0.h[0]
+; CHECK-SD-NEXT:    umov w1, v0.h[1]
+; CHECK-SD-NEXT:    umov w2, v0.h[2]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: ashr_v3i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov s0, w3
+; CHECK-GI-NEXT:    fmov s1, w4
+; CHECK-GI-NEXT:    fmov s2, w1
+; CHECK-GI-NEXT:    mov v0.b[1], v1.b[0]
+; CHECK-GI-NEXT:    fmov s1, w0
+; CHECK-GI-NEXT:    mov v1.b[1], v2.b[0]
+; CHECK-GI-NEXT:    fmov s2, w5
+; CHECK-GI-NEXT:    mov v0.b[2], v2.b[0]
+; CHECK-GI-NEXT:    fmov s2, w2
+; CHECK-GI-NEXT:    mov v1.b[2], v2.b[0]
+; CHECK-GI-NEXT:    mov v0.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[7], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[7], v0.b[0]
+; CHECK-GI-NEXT:    neg v0.8b, v0.8b
+; CHECK-GI-NEXT:    sshl v0.8b, v1.8b, v0.8b
+; CHECK-GI-NEXT:    umov w0, v0.b[0]
+; CHECK-GI-NEXT:    umov w1, v0.b[1]
+; CHECK-GI-NEXT:    umov w2, v0.b[2]
+; CHECK-GI-NEXT:    ret
     %3 = ashr <3 x i8> %0, %1
     ret <3 x i8> %3
 }
@@ -951,22 +998,51 @@ define <3 x i32> @ashr_v3i32(<3 x i32> %0, <3 x i32> %1){
 }
 
 define <3 x i8> @lshr_v3i8(<3 x i8> %0, <3 x i8> %1){
-; CHECK-LABEL: lshr_v3i8:
-; CHECK:       // %bb.0:
-; CHECK-NEXT:    fmov s0, w3
-; CHECK-NEXT:    fmov s1, w0
-; CHECK-NEXT:    mov v0.h[1], w4
-; CHECK-NEXT:    mov v1.h[1], w1
-; CHECK-NEXT:    mov v0.h[2], w5
-; CHECK-NEXT:    mov v1.h[2], w2
-; CHECK-NEXT:    bic v0.4h, #255, lsl #8
-; CHECK-NEXT:    bic v1.4h, #255, lsl #8
-; CHECK-NEXT:    neg v0.4h, v0.4h
-; CHECK-NEXT:    ushl v0.4h, v1.4h, v0.4h
-; CHECK-NEXT:    umov w0, v0.h[0]
-; CHECK-NEXT:    umov w1, v0.h[1]
-; CHECK-NEXT:    umov w2, v0.h[2]
-; CHECK-NEXT:    ret
+; CHECK-SD-LABEL: lshr_v3i8:
+; CHECK-SD:       // %bb.0:
+; CHECK-SD-NEXT:    fmov s0, w3
+; CHECK-SD-NEXT:    fmov s1, w0
+; CHECK-SD-NEXT:    mov v0.h[1], w4
+; CHECK-SD-NEXT:    mov v1.h[1], w1
+; CHECK-SD-NEXT:    mov v0.h[2], w5
+; CHECK-SD-NEXT:    mov v1.h[2], w2
+; CHECK-SD-NEXT:    bic v0.4h, #255, lsl #8
+; CHECK-SD-NEXT:    bic v1.4h, #255, lsl #8
+; CHECK-SD-NEXT:    neg v0.4h, v0.4h
+; CHECK-SD-NEXT:    ushl v0.4h, v1.4h, v0.4h
+; CHECK-SD-NEXT:    umov w0, v0.h[0]
+; CHECK-SD-NEXT:    umov w1, v0.h[1]
+; CHECK-SD-NEXT:    umov w2, v0.h[2]
+; CHECK-SD-NEXT:    ret
+;
+; CHECK-GI-LABEL: lshr_v3i8:
+; CHECK-GI:       // %bb.0:
+; CHECK-GI-NEXT:    fmov s0, w3
+; CHECK-GI-NEXT:    fmov s1, w4
+; CHECK-GI-NEXT:    fmov s2, w1
+; CHECK-GI-NEXT:    mov v0.b[1], v1.b[0]
+; CHECK-GI-NEXT:    fmov s1, w0
+; CHECK-GI-NEXT:    mov v1.b[1], v2.b[0]
+; CHECK-GI-NEXT:    fmov s2, w5
+; CHECK-GI-NEXT:    mov v0.b[2], v2.b[0]
+; CHECK-GI-NEXT:    fmov s2, w2
+; CHECK-GI-NEXT:    mov v1.b[2], v2.b[0]
+; CHECK-GI-NEXT:    mov v0.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[3], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[4], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[5], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[6], v0.b[0]
+; CHECK-GI-NEXT:    mov v0.b[7], v0.b[0]
+; CHECK-GI-NEXT:    mov v1.b[7], v0.b[0]
+; CHECK-GI-NEXT:    neg v0.8b, v0.8b
+; CHECK-GI-NEXT:    ushl v0.8b, v1.8b, v0.8b
+; CHECK-GI-NEXT:    umov w0, v0.b[0]
+; CHECK-GI-NEXT:    umov w1, v0.b[1]
+; CHECK-GI-NEXT:    umov w2, v0.b[2]
+; CHECK-GI-NEXT:    ret
     %3 = lshr <3 x i8> %0, %1
     ret <3 x i8> %3
 }