[RISCV] Add test case showing failure to convert gather/scatter to strided load/store...

Our optimization pass checks for loop simplify form, before doing
the transform. The loops here aren't in loop simplify form because
the exit block has two predecessors.

Reviewed By: reames

Differential Revision: https://reviews.llvm.org/D127451

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store-asm.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store-asm.ll
index 1a4f53c..7dc28d7 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store-asm.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store-asm.ll
@@ -872,3 +872,273 @@ define void @strided_load_startval_add_with_splat(i8* noalias nocapture %0, i8*
   %47 = icmp eq i32 %46, 1024
   br i1 %47, label %36, label %37
 }
+
+declare <16 x i8> @llvm.masked.gather.v16i8.v16p0i8(<16 x i8*>, i32 immarg, <16 x i1>, <16 x i8>)
+declare void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8>, <16 x i8*>, i32 immarg, <16 x i1>)
+
+define void @gather_no_scalar_remainder(i8* noalias nocapture noundef %arg, i8* noalias nocapture noundef readonly %arg1, i64 noundef %arg2) {
+; V-LABEL: gather_no_scalar_remainder:
+; V:       # %bb.0: # %bb
+; V-NEXT:    slli a2, a2, 4
+; V-NEXT:    beqz a2, .LBB13_3
+; V-NEXT:  # %bb.1: # %bb2
+; V-NEXT:    vsetivli zero, 16, e64, m4, ta, mu
+; V-NEXT:    vid.v v8
+; V-NEXT:    li a3, 5
+; V-NEXT:    li a4, 16
+; V-NEXT:  .LBB13_2: # %bb4
+; V-NEXT:    # =>This Inner Loop Header: Depth=1
+; V-NEXT:    vmul.vx v12, v8, a3
+; V-NEXT:    vsetvli zero, zero, e8, mf2, ta, mu
+; V-NEXT:    vluxei64.v v16, (a1), v12
+; V-NEXT:    vle8.v v12, (a0)
+; V-NEXT:    vadd.vv v12, v12, v16
+; V-NEXT:    vse8.v v12, (a0)
+; V-NEXT:    vsetvli zero, zero, e64, m4, ta, mu
+; V-NEXT:    vadd.vx v8, v8, a4
+; V-NEXT:    addi a2, a2, -16
+; V-NEXT:    addi a0, a0, 16
+; V-NEXT:    bnez a2, .LBB13_2
+; V-NEXT:  .LBB13_3: # %bb16
+; V-NEXT:    ret
+;
+; ZVE32F-LABEL: gather_no_scalar_remainder:
+; ZVE32F:       # %bb.0: # %bb
+; ZVE32F-NEXT:    addi sp, sp, -240
+; ZVE32F-NEXT:    .cfi_def_cfa_offset 240
+; ZVE32F-NEXT:    sd ra, 232(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s0, 224(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s1, 216(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s2, 208(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s3, 200(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s4, 192(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s5, 184(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s6, 176(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s7, 168(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s8, 160(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s9, 152(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s10, 144(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd s11, 136(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    .cfi_offset ra, -8
+; ZVE32F-NEXT:    .cfi_offset s0, -16
+; ZVE32F-NEXT:    .cfi_offset s1, -24
+; ZVE32F-NEXT:    .cfi_offset s2, -32
+; ZVE32F-NEXT:    .cfi_offset s3, -40
+; ZVE32F-NEXT:    .cfi_offset s4, -48
+; ZVE32F-NEXT:    .cfi_offset s5, -56
+; ZVE32F-NEXT:    .cfi_offset s6, -64
+; ZVE32F-NEXT:    .cfi_offset s7, -72
+; ZVE32F-NEXT:    .cfi_offset s8, -80
+; ZVE32F-NEXT:    .cfi_offset s9, -88
+; ZVE32F-NEXT:    .cfi_offset s10, -96
+; ZVE32F-NEXT:    .cfi_offset s11, -104
+; ZVE32F-NEXT:    slli a2, a2, 4
+; ZVE32F-NEXT:    beqz a2, .LBB13_3
+; ZVE32F-NEXT:  # %bb.1: # %bb2
+; ZVE32F-NEXT:    li a3, 0
+; ZVE32F-NEXT:    li a4, 15
+; ZVE32F-NEXT:    li a5, 14
+; ZVE32F-NEXT:    li a6, 13
+; ZVE32F-NEXT:    li a7, 12
+; ZVE32F-NEXT:    li t0, 11
+; ZVE32F-NEXT:    li t1, 10
+; ZVE32F-NEXT:    li t2, 9
+; ZVE32F-NEXT:    li t3, 8
+; ZVE32F-NEXT:    li t4, 7
+; ZVE32F-NEXT:    li t5, 6
+; ZVE32F-NEXT:    li t6, 5
+; ZVE32F-NEXT:    li s0, 4
+; ZVE32F-NEXT:    li s1, 3
+; ZVE32F-NEXT:    li s2, 2
+; ZVE32F-NEXT:    li s3, 1
+; ZVE32F-NEXT:    vsetivli zero, 16, e8, mf2, ta, mu
+; ZVE32F-NEXT:  .LBB13_2: # %bb4
+; ZVE32F-NEXT:    # =>This Inner Loop Header: Depth=1
+; ZVE32F-NEXT:    sd s0, 56(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd t0, 64(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd a5, 72(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd a4, 80(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd a3, 88(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd a0, 96(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    sd a2, 104(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    slli s4, a3, 2
+; ZVE32F-NEXT:    add a0, s4, a3
+; ZVE32F-NEXT:    sd a0, 48(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    slli s5, s3, 2
+; ZVE32F-NEXT:    add a0, s5, s3
+; ZVE32F-NEXT:    sd a0, 40(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    slli s6, s2, 2
+; ZVE32F-NEXT:    add a0, s6, s2
+; ZVE32F-NEXT:    sd a0, 32(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    slli s7, s1, 2
+; ZVE32F-NEXT:    add a0, s7, s1
+; ZVE32F-NEXT:    sd a0, 24(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    slli s8, s0, 2
+; ZVE32F-NEXT:    add a0, s8, s0
+; ZVE32F-NEXT:    slli s9, t6, 2
+; ZVE32F-NEXT:    add a3, s9, t6
+; ZVE32F-NEXT:    slli s10, t5, 2
+; ZVE32F-NEXT:    add s10, s10, t5
+; ZVE32F-NEXT:    slli s11, t4, 2
+; ZVE32F-NEXT:    add s11, s11, t4
+; ZVE32F-NEXT:    slli ra, t3, 2
+; ZVE32F-NEXT:    add ra, ra, t3
+; ZVE32F-NEXT:    slli s4, t2, 2
+; ZVE32F-NEXT:    add s4, s4, t2
+; ZVE32F-NEXT:    slli s5, t1, 2
+; ZVE32F-NEXT:    add s5, s5, t1
+; ZVE32F-NEXT:    slli s6, t0, 2
+; ZVE32F-NEXT:    add s6, s6, t0
+; ZVE32F-NEXT:    slli s7, a7, 2
+; ZVE32F-NEXT:    add s7, s7, a7
+; ZVE32F-NEXT:    slli s8, a6, 2
+; ZVE32F-NEXT:    add s8, s8, a6
+; ZVE32F-NEXT:    slli a2, a5, 2
+; ZVE32F-NEXT:    add a2, a2, a5
+; ZVE32F-NEXT:    slli s9, a4, 2
+; ZVE32F-NEXT:    add s9, s9, a4
+; ZVE32F-NEXT:    add a4, a1, s9
+; ZVE32F-NEXT:    sd a4, 16(sp) # 8-byte Folded Spill
+; ZVE32F-NEXT:    add a2, a1, a2
+; ZVE32F-NEXT:    add s8, a1, s8
+; ZVE32F-NEXT:    add s7, a1, s7
+; ZVE32F-NEXT:    add s6, a1, s6
+; ZVE32F-NEXT:    add s5, a1, s5
+; ZVE32F-NEXT:    add s4, a1, s4
+; ZVE32F-NEXT:    add ra, a1, ra
+; ZVE32F-NEXT:    add s11, a1, s11
+; ZVE32F-NEXT:    add s10, a1, s10
+; ZVE32F-NEXT:    add a3, a1, a3
+; ZVE32F-NEXT:    add a0, a1, a0
+; ZVE32F-NEXT:    ld a4, 24(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    add a4, a1, a4
+; ZVE32F-NEXT:    ld a5, 32(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    add a5, a1, a5
+; ZVE32F-NEXT:    mv t0, a6
+; ZVE32F-NEXT:    ld a6, 40(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    add a6, a1, a6
+; ZVE32F-NEXT:    ld s0, 48(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    add s9, a1, s0
+; ZVE32F-NEXT:    lb s9, 0(s9)
+; ZVE32F-NEXT:    lb a6, 0(a6)
+; ZVE32F-NEXT:    lb a5, 0(a5)
+; ZVE32F-NEXT:    lb a4, 0(a4)
+; ZVE32F-NEXT:    lb a0, 0(a0)
+; ZVE32F-NEXT:    lb a3, 0(a3)
+; ZVE32F-NEXT:    lb s10, 0(s10)
+; ZVE32F-NEXT:    lb s11, 0(s11)
+; ZVE32F-NEXT:    lb ra, 0(ra)
+; ZVE32F-NEXT:    lb s4, 0(s4)
+; ZVE32F-NEXT:    lb s5, 0(s5)
+; ZVE32F-NEXT:    lb s6, 0(s6)
+; ZVE32F-NEXT:    lb s7, 0(s7)
+; ZVE32F-NEXT:    lb s8, 0(s8)
+; ZVE32F-NEXT:    lb a2, 0(a2)
+; ZVE32F-NEXT:    mv s0, t6
+; ZVE32F-NEXT:    mv t6, t5
+; ZVE32F-NEXT:    mv t5, t4
+; ZVE32F-NEXT:    mv t4, t3
+; ZVE32F-NEXT:    mv t3, t2
+; ZVE32F-NEXT:    mv t2, t1
+; ZVE32F-NEXT:    mv t1, a7
+; ZVE32F-NEXT:    ld a7, 16(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    lb a7, 0(a7)
+; ZVE32F-NEXT:    sb s9, 112(sp)
+; ZVE32F-NEXT:    sb a6, 113(sp)
+; ZVE32F-NEXT:    mv a6, t0
+; ZVE32F-NEXT:    ld t0, 64(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb a5, 114(sp)
+; ZVE32F-NEXT:    ld a5, 72(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb a4, 115(sp)
+; ZVE32F-NEXT:    ld a4, 80(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb a0, 116(sp)
+; ZVE32F-NEXT:    ld a0, 96(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb a3, 117(sp)
+; ZVE32F-NEXT:    ld a3, 88(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb s10, 118(sp)
+; ZVE32F-NEXT:    sb s11, 119(sp)
+; ZVE32F-NEXT:    sb ra, 120(sp)
+; ZVE32F-NEXT:    sb s4, 121(sp)
+; ZVE32F-NEXT:    sb s5, 122(sp)
+; ZVE32F-NEXT:    sb s6, 123(sp)
+; ZVE32F-NEXT:    sb s7, 124(sp)
+; ZVE32F-NEXT:    sb s8, 125(sp)
+; ZVE32F-NEXT:    sb a2, 126(sp)
+; ZVE32F-NEXT:    ld a2, 104(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    sb a7, 127(sp)
+; ZVE32F-NEXT:    mv a7, t1
+; ZVE32F-NEXT:    mv t1, t2
+; ZVE32F-NEXT:    mv t2, t3
+; ZVE32F-NEXT:    mv t3, t4
+; ZVE32F-NEXT:    mv t4, t5
+; ZVE32F-NEXT:    mv t5, t6
+; ZVE32F-NEXT:    mv t6, s0
+; ZVE32F-NEXT:    ld s0, 56(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    addi s4, sp, 112
+; ZVE32F-NEXT:    vle8.v v8, (s4)
+; ZVE32F-NEXT:    vle8.v v9, (a0)
+; ZVE32F-NEXT:    vadd.vv v8, v9, v8
+; ZVE32F-NEXT:    vse8.v v8, (a0)
+; ZVE32F-NEXT:    addi a3, a3, 16
+; ZVE32F-NEXT:    addi s3, s3, 16
+; ZVE32F-NEXT:    addi s2, s2, 16
+; ZVE32F-NEXT:    addi s1, s1, 16
+; ZVE32F-NEXT:    addi s0, s0, 16
+; ZVE32F-NEXT:    addi t6, t6, 16
+; ZVE32F-NEXT:    addi t5, t5, 16
+; ZVE32F-NEXT:    addi t4, t4, 16
+; ZVE32F-NEXT:    addi t3, t3, 16
+; ZVE32F-NEXT:    addi t2, t2, 16
+; ZVE32F-NEXT:    addi t1, t1, 16
+; ZVE32F-NEXT:    addi t0, t0, 16
+; ZVE32F-NEXT:    addi a7, a7, 16
+; ZVE32F-NEXT:    addi a6, a6, 16
+; ZVE32F-NEXT:    addi a5, a5, 16
+; ZVE32F-NEXT:    addi a4, a4, 16
+; ZVE32F-NEXT:    addi a2, a2, -16
+; ZVE32F-NEXT:    addi a0, a0, 16
+; ZVE32F-NEXT:    bnez a2, .LBB13_2
+; ZVE32F-NEXT:  .LBB13_3: # %bb16
+; ZVE32F-NEXT:    ld ra, 232(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s0, 224(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s1, 216(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s2, 208(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s3, 200(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s4, 192(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s5, 184(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s6, 176(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s7, 168(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s8, 160(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s9, 152(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s10, 144(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    ld s11, 136(sp) # 8-byte Folded Reload
+; ZVE32F-NEXT:    addi sp, sp, 240
+; ZVE32F-NEXT:    ret
+bb:
+  %i = shl i64 %arg2, 4
+  %i3 = icmp eq i64 %i, 0
+  br i1 %i3, label %bb16, label %bb2
+
+bb2:
+  br label %bb4
+
+bb4:                                              ; preds = %bb4, %bb
+  %i5 = phi i64 [ %i13, %bb4 ], [ 0, %bb2 ]
+  %i6 = phi <16 x i64> [ %i14, %bb4 ], [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %bb2 ]
+  %i7 = mul <16 x i64> %i6, <i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5>
+  %i8 = getelementptr inbounds i8, i8* %arg1, <16 x i64> %i7
+  %i9 = call <16 x i8> @llvm.masked.gather.v16i8.v16p0i8(<16 x i8*> %i8, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i8> undef)
+  %i10 = getelementptr inbounds i8, i8* %arg, i64 %i5
+  %cast = bitcast i8* %i10 to <16 x i8>*
+  %i11 = load <16 x i8>, <16 x i8>* %cast, align 1
+  %i12 = add <16 x i8> %i11, %i9
+  %cast2 = bitcast i8* %i10 to <16 x i8>*
+  store <16 x i8> %i12, <16 x i8>* %cast2, align 1
+  %i13 = add nuw i64 %i5, 16
+  %i14 = add <16 x i64> %i6, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+  %i15 = icmp eq i64 %i13, %i
+  br i1 %i15, label %bb16, label %bb4
+
+bb16:                                             ; preds = %bb4, %bb
+  ret void
+}

diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store.ll
index 7f34be9..fd8ff26 100644 (file)
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vector-strided-load-store.ll
@@ -853,3 +853,62 @@ define void @strided_load_startval_add_with_splat(i8* noalias nocapture %0, i8*
   %47 = icmp eq i32 %46, 1024
   br i1 %47, label %36, label %37
 }
+
+declare <16 x i8> @llvm.masked.gather.v16i8.v16p0i8(<16 x i8*>, i32 immarg, <16 x i1>, <16 x i8>)
+declare void @llvm.masked.scatter.v16i8.v16p0i8(<16 x i8>, <16 x i8*>, i32 immarg, <16 x i1>)
+
+define void @gather_no_scalar_remainder(i8* noalias nocapture noundef %arg, i8* noalias nocapture noundef readonly %arg1, i64 noundef %arg2) {
+; CHECK-LABEL: @gather_no_scalar_remainder(
+; CHECK-NEXT:  bb:
+; CHECK-NEXT:    [[I:%.*]] = shl i64 [[ARG2:%.*]], 4
+; CHECK-NEXT:    [[I3:%.*]] = icmp eq i64 [[I]], 0
+; CHECK-NEXT:    br i1 [[I3]], label [[BB16:%.*]], label [[BB2:%.*]]
+; CHECK:       bb2:
+; CHECK-NEXT:    br label [[BB4:%.*]]
+; CHECK:       bb4:
+; CHECK-NEXT:    [[I5:%.*]] = phi i64 [ [[I13:%.*]], [[BB4]] ], [ 0, [[BB2]] ]
+; CHECK-NEXT:    [[I6:%.*]] = phi <16 x i64> [ [[I14:%.*]], [[BB4]] ], [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, [[BB2]] ]
+; CHECK-NEXT:    [[I7:%.*]] = mul <16 x i64> [[I6]], <i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5>
+; CHECK-NEXT:    [[I8:%.*]] = getelementptr inbounds i8, i8* [[ARG1:%.*]], <16 x i64> [[I7]]
+; CHECK-NEXT:    [[I9:%.*]] = call <16 x i8> @llvm.masked.gather.v16i8.v16p0i8(<16 x i8*> [[I8]], i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i8> undef)
+; CHECK-NEXT:    [[I10:%.*]] = getelementptr inbounds i8, i8* [[ARG:%.*]], i64 [[I5]]
+; CHECK-NEXT:    [[CAST:%.*]] = bitcast i8* [[I10]] to <16 x i8>*
+; CHECK-NEXT:    [[I11:%.*]] = load <16 x i8>, <16 x i8>* [[CAST]], align 1
+; CHECK-NEXT:    [[I12:%.*]] = add <16 x i8> [[I11]], [[I9]]
+; CHECK-NEXT:    [[CAST2:%.*]] = bitcast i8* [[I10]] to <16 x i8>*
+; CHECK-NEXT:    store <16 x i8> [[I12]], <16 x i8>* [[CAST2]], align 1
+; CHECK-NEXT:    [[I13]] = add nuw i64 [[I5]], 16
+; CHECK-NEXT:    [[I14]] = add <16 x i64> [[I6]], <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+; CHECK-NEXT:    [[I15:%.*]] = icmp eq i64 [[I13]], [[I]]
+; CHECK-NEXT:    br i1 [[I15]], label [[BB16]], label [[BB4]]
+; CHECK:       bb16:
+; CHECK-NEXT:    ret void
+;
+bb:
+  %i = shl i64 %arg2, 4
+  %i3 = icmp eq i64 %i, 0
+  br i1 %i3, label %bb16, label %bb2
+
+bb2:
+  br label %bb4
+
+bb4:                                              ; preds = %bb4, %bb
+  %i5 = phi i64 [ %i13, %bb4 ], [ 0, %bb2 ]
+  %i6 = phi <16 x i64> [ %i14, %bb4 ], [ <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, %bb2 ]
+  %i7 = mul <16 x i64> %i6, <i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5, i64 5>
+  %i8 = getelementptr inbounds i8, i8* %arg1, <16 x i64> %i7
+  %i9 = call <16 x i8> @llvm.masked.gather.v16i8.v16p0i8(<16 x i8*> %i8, i32 1, <16 x i1> <i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true, i1 true>, <16 x i8> undef)
+  %i10 = getelementptr inbounds i8, i8* %arg, i64 %i5
+  %cast = bitcast i8* %i10 to <16 x i8>*
+  %i11 = load <16 x i8>, <16 x i8>* %cast, align 1
+  %i12 = add <16 x i8> %i11, %i9
+  %cast2 = bitcast i8* %i10 to <16 x i8>*
+  store <16 x i8> %i12, <16 x i8>* %cast2, align 1
+  %i13 = add nuw i64 %i5, 16
+  %i14 = add <16 x i64> %i6, <i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16, i64 16>
+  %i15 = icmp eq i64 %i13, %i
+  br i1 %i15, label %bb16, label %bb4
+
+bb16:                                             ; preds = %bb4, %bb
+  ret void
+}