94 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s | FileCheck %s3 4target triple = "aarch64-unknown-linux-gnu"5 6; Currently there is no custom lowering for vector shuffles operating on types7; bigger than NEON. However, having no support opens us up to a code generator8; hang when expanding BUILD_VECTOR. Here we just validate the promblematic case9; successfully exits code generation.10define void @hang_when_merging_stores_after_legalisation(ptr %a, <2 x i32> %b) vscale_range(2,2) #0 {11; CHECK-LABEL: hang_when_merging_stores_after_legalisation:12; CHECK: // %bb.0:13; CHECK-NEXT: // kill: def $d0 killed $d0 def $z014; CHECK-NEXT: mov z0.s, s015; CHECK-NEXT: movprfx z1, z016; CHECK-NEXT: ext z1.b, z1.b, z0.b, #1617; CHECK-NEXT: st2 { v0.4s, v1.4s }, [x0]18; CHECK-NEXT: ret19 %splat = shufflevector <2 x i32> %b, <2 x i32> poison, <8 x i32> zeroinitializer20 %interleaved.vec = shufflevector <8 x i32> %splat, <8 x i32> poison, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>21 store <8 x i32> %interleaved.vec, ptr %a, align 422 ret void23}24 25; Ensure we don't crash when trying to lower a shuffle via an extract26define void @crash_when_lowering_extract_shuffle(ptr %dst, i1 %cond) vscale_range(2,2) #0 {27; CHECK-LABEL: crash_when_lowering_extract_shuffle:28; CHECK: // %bb.0:29; CHECK-NEXT: tbnz w1, #0, .LBB1_230; CHECK-NEXT: // %bb.1: // %vector.body31; CHECK-NEXT: movi v2.2d, #000000000000000032; CHECK-NEXT: movi v0.2d, #000000000000000033; CHECK-NEXT: movi v3.2d, #000000000000000034; CHECK-NEXT: ptrue p0.s35; CHECK-NEXT: umov w8, v2.b[8]36; CHECK-NEXT: mov v0.b[1], v2.b[1]37; CHECK-NEXT: ext z3.b, z3.b, z3.b, #1638; CHECK-NEXT: ext v4.16b, v3.16b, v3.16b, #839; CHECK-NEXT: fmov s1, w840; CHECK-NEXT: mov v0.b[2], v2.b[2]41; CHECK-NEXT: mov v1.b[1], v2.b[9]42; CHECK-NEXT: mov v0.b[3], v2.b[3]43; CHECK-NEXT: mov v1.b[2], v2.b[10]44; CHECK-NEXT: mov v0.b[4], v2.b[4]45; CHECK-NEXT: mov v1.b[3], v2.b[11]46; CHECK-NEXT: mov v0.b[5], v2.b[5]47; CHECK-NEXT: mov v1.b[4], v2.b[12]48; CHECK-NEXT: mov v0.b[6], v2.b[6]49; CHECK-NEXT: mov v1.b[5], v2.b[13]50; CHECK-NEXT: mov v0.b[7], v2.b[7]51; CHECK-NEXT: mov v1.b[6], v2.b[14]52; CHECK-NEXT: uunpklo z0.h, z0.b53; CHECK-NEXT: mov v1.b[7], v2.b[15]54; CHECK-NEXT: uunpklo z2.h, z3.b55; CHECK-NEXT: uunpklo z3.h, z4.b56; CHECK-NEXT: uunpklo z0.s, z0.h57; CHECK-NEXT: uunpklo z1.h, z1.b58; CHECK-NEXT: uunpklo z2.s, z2.h59; CHECK-NEXT: uunpklo z3.s, z3.h60; CHECK-NEXT: lsl z0.s, z0.s, #3161; CHECK-NEXT: uunpklo z1.s, z1.h62; CHECK-NEXT: lsl z2.s, z2.s, #3163; CHECK-NEXT: lsl z3.s, z3.s, #3164; CHECK-NEXT: asr z0.s, z0.s, #3165; CHECK-NEXT: asr z2.s, z2.s, #3166; CHECK-NEXT: asr z3.s, z3.s, #3167; CHECK-NEXT: lsl z1.s, z1.s, #3168; CHECK-NEXT: cmpne p3.s, p0/z, z0.s, #069; CHECK-NEXT: cmpne p1.s, p0/z, z2.s, #070; CHECK-NEXT: movi v2.2d, #000000000000000071; CHECK-NEXT: cmpne p2.s, p0/z, z3.s, #072; CHECK-NEXT: asr z1.s, z1.s, #3173; CHECK-NEXT: cmpne p0.s, p0/z, z1.s, #074; CHECK-NEXT: st1w { z2.s }, p1, [x0, #2, mul vl]75; CHECK-NEXT: st1w { z2.s }, p2, [x0, #3, mul vl]76; CHECK-NEXT: st1w { z2.s }, p3, [x0]77; CHECK-NEXT: st1w { z2.s }, p0, [x0, #1, mul vl]78; CHECK-NEXT: .LBB1_2: // %exit79; CHECK-NEXT: ret80 %broadcast.splat = shufflevector <32 x i1> zeroinitializer, <32 x i1> zeroinitializer, <32 x i32> zeroinitializer81 br i1 %cond, label %exit, label %vector.body82 83vector.body:84 %1 = load <32 x i32>, ptr %dst, align 1685 %predphi = select <32 x i1> %broadcast.splat, <32 x i32> zeroinitializer, <32 x i32> %186 store <32 x i32> %predphi, ptr %dst, align 1687 br label %exit88 89exit:90 ret void91}92 93attributes #0 = { "target-features"="+sve" }94