1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2 ; RUN: llc -mtriple=aarch64-unknown-linux-gnu < %s | FileCheck %s
4 ; FIXME: Currently, we avoid narrowing this v4i32 load, in the
5 ; hopes of being able to fold the shift, despite it requiring stack
6 ; storage + loads. Ideally, we should narrow here and load the i32
7 ; directly from the variable offset e.g:
9 ; add x8, x0, x1, lsl #4
11 ; ldr w0, [x8, x9, lsl #2]
13 ; The AArch64TargetLowering::shouldReduceLoadWidth heuristic should
14 ; probably be updated to choose load-narrowing instead of folding the
15 ; lsl in larger vector cases.
17 define i32 @narrow_load_v4_i32_single_ele_variable_idx(ptr %ptr, i64 %off, i32 %ele) {
18 ; CHECK-LABEL: narrow_load_v4_i32_single_ele_variable_idx:
19 ; CHECK: // %bb.0: // %entry
20 ; CHECK-NEXT: sub sp, sp, #16
21 ; CHECK-NEXT: .cfi_def_cfa_offset 16
22 ; CHECK-NEXT: ldr q0, [x0, x1, lsl #4]
23 ; CHECK-NEXT: mov x8, sp
24 ; CHECK-NEXT: // kill: def $w2 killed $w2 def $x2
25 ; CHECK-NEXT: bfi x8, x2, #2, #2
26 ; CHECK-NEXT: str q0, [sp]
27 ; CHECK-NEXT: ldr w0, [x8]
28 ; CHECK-NEXT: add sp, sp, #16
31 %idx = getelementptr inbounds <4 x i32>, ptr %ptr, i64 %off
32 %x = load <4 x i32>, ptr %idx, align 8
33 %res = extractelement <4 x i32> %x, i32 %ele