151 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc -mattr=+sve -force-streaming-compatible < %s | FileCheck %s3; RUN: llc -mattr=+sme -force-streaming < %s | FileCheck %s4; RUN: llc -force-streaming-compatible < %s | FileCheck %s --check-prefix=NONEON-NOSVE5 6 7target triple = "aarch64-unknown-linux-gnu"8 9define void @hang_when_merging_stores_after_legalisation(ptr %a, <2 x i32> %b) {10; CHECK-LABEL: hang_when_merging_stores_after_legalisation:11; CHECK: // %bb.0:12; CHECK-NEXT: // kill: def $d0 killed $d0 def $z013; CHECK-NEXT: ptrue p0.s, vl414; CHECK-NEXT: mov z0.s, s015; CHECK-NEXT: mov z1.d, z0.d16; CHECK-NEXT: st2w { z0.s, z1.s }, p0, [x0]17; CHECK-NEXT: ret18;19; NONEON-NOSVE-LABEL: hang_when_merging_stores_after_legalisation:20; NONEON-NOSVE: // %bb.0:21; NONEON-NOSVE-NEXT: sub sp, sp, #3222; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 3223; NONEON-NOSVE-NEXT: str d0, [sp, #8]24; NONEON-NOSVE-NEXT: ldr w8, [sp, #8]25; NONEON-NOSVE-NEXT: stp w8, w8, [sp, #24]26; NONEON-NOSVE-NEXT: stp w8, w8, [sp, #16]27; NONEON-NOSVE-NEXT: ldr q0, [sp, #16]28; NONEON-NOSVE-NEXT: stp q0, q0, [x0]29; NONEON-NOSVE-NEXT: add sp, sp, #3230; NONEON-NOSVE-NEXT: ret31 %splat = shufflevector <2 x i32> %b, <2 x i32> poison, <8 x i32> zeroinitializer32 %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>33 store <8 x i32> %interleaved.vec, ptr %a, align 434 ret void35}36 37define void @interleave_store_without_splat(ptr %a, <4 x i32> %v1, <4 x i32> %v2) {38; CHECK-LABEL: interleave_store_without_splat:39; CHECK: // %bb.0:40; CHECK-NEXT: ptrue p0.s, vl441; CHECK-NEXT: // kill: def $q1 killed $q1 killed $z0_z1 def $z0_z142; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z0_z1 def $z0_z143; CHECK-NEXT: st2w { z0.s, z1.s }, p0, [x0]44; CHECK-NEXT: ret45;46; NONEON-NOSVE-LABEL: interleave_store_without_splat:47; NONEON-NOSVE: // %bb.0:48; NONEON-NOSVE-NEXT: sub sp, sp, #6449; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 6450; NONEON-NOSVE-NEXT: stp q0, q1, [sp, #16]51; NONEON-NOSVE-NEXT: ldr w9, [sp, #36]52; NONEON-NOSVE-NEXT: ldr w8, [sp, #20]53; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #8]54; NONEON-NOSVE-NEXT: ldr w9, [sp, #32]55; NONEON-NOSVE-NEXT: ldr w8, [sp, #16]56; NONEON-NOSVE-NEXT: stp w8, w9, [sp]57; NONEON-NOSVE-NEXT: ldr w9, [sp, #44]58; NONEON-NOSVE-NEXT: ldr w8, [sp, #28]59; NONEON-NOSVE-NEXT: ldr q1, [sp]60; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #56]61; NONEON-NOSVE-NEXT: ldr w9, [sp, #40]62; NONEON-NOSVE-NEXT: ldr w8, [sp, #24]63; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #48]64; NONEON-NOSVE-NEXT: ldr q0, [sp, #48]65; NONEON-NOSVE-NEXT: stp q1, q0, [x0]66; NONEON-NOSVE-NEXT: add sp, sp, #6467; NONEON-NOSVE-NEXT: ret68 %shuffle = shufflevector <4 x i32> %v1, <4 x i32> %v2, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>69 %interleaved = shufflevector <8 x i32> %shuffle, <8 x i32> poison, <8 x i32> <i32 0, i32 4, i32 1, i32 5, i32 2, i32 6, i32 3, i32 7>70 store <8 x i32> %interleaved, ptr %a, align 171 ret void72}73 74define void @interleave_store_legalization(ptr %a, <8 x i32> %v1, <8 x i32> %v2) {75; CHECK-LABEL: interleave_store_legalization:76; CHECK: // %bb.0:77; CHECK-NEXT: mov z5.d, z2.d78; CHECK-NEXT: // kill: def $q3 killed $q3 def $z2_z379; CHECK-NEXT: ptrue p0.s, vl480; CHECK-NEXT: mov x8, #8 // =0x881; CHECK-NEXT: mov z4.d, z0.d82; CHECK-NEXT: mov z2.d, z1.d83; CHECK-NEXT: st2w { z4.s, z5.s }, p0, [x0]84; CHECK-NEXT: st2w { z2.s, z3.s }, p0, [x0, x8, lsl #2]85; CHECK-NEXT: ret86;87; NONEON-NOSVE-LABEL: interleave_store_legalization:88; NONEON-NOSVE: // %bb.0:89; NONEON-NOSVE-NEXT: sub sp, sp, #12890; NONEON-NOSVE-NEXT: .cfi_def_cfa_offset 12891; NONEON-NOSVE-NEXT: stp q1, q3, [sp, #16]92; NONEON-NOSVE-NEXT: stp q0, q2, [sp, #80]93; NONEON-NOSVE-NEXT: ldr w9, [sp, #100]94; NONEON-NOSVE-NEXT: ldr w8, [sp, #84]95; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #72]96; NONEON-NOSVE-NEXT: ldr w9, [sp, #96]97; NONEON-NOSVE-NEXT: ldr w8, [sp, #80]98; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #64]99; NONEON-NOSVE-NEXT: ldr w9, [sp, #108]100; NONEON-NOSVE-NEXT: ldr w8, [sp, #92]101; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #120]102; NONEON-NOSVE-NEXT: ldr w9, [sp, #104]103; NONEON-NOSVE-NEXT: ldr w8, [sp, #88]104; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #112]105; NONEON-NOSVE-NEXT: ldr w9, [sp, #36]106; NONEON-NOSVE-NEXT: ldr w8, [sp, #20]107; NONEON-NOSVE-NEXT: ldr q3, [sp, #112]108; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #8]109; NONEON-NOSVE-NEXT: ldr w9, [sp, #32]110; NONEON-NOSVE-NEXT: ldr w8, [sp, #16]111; NONEON-NOSVE-NEXT: stp w8, w9, [sp]112; NONEON-NOSVE-NEXT: ldr w9, [sp, #44]113; NONEON-NOSVE-NEXT: ldr w8, [sp, #28]114; NONEON-NOSVE-NEXT: ldr q1, [sp]115; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #56]116; NONEON-NOSVE-NEXT: ldr w9, [sp, #40]117; NONEON-NOSVE-NEXT: ldr w8, [sp, #24]118; NONEON-NOSVE-NEXT: stp w8, w9, [sp, #48]119; NONEON-NOSVE-NEXT: ldp q0, q2, [sp, #48]120; NONEON-NOSVE-NEXT: stp q2, q3, [x0]121; NONEON-NOSVE-NEXT: stp q1, q0, [x0, #32]122; NONEON-NOSVE-NEXT: add sp, sp, #128123; NONEON-NOSVE-NEXT: ret124 %interleaved.vec = shufflevector <8 x i32> %v1, <8 x i32> %v2, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11,125 i32 4, i32 12, i32 5, i32 13, i32 6, i32 14, i32 7, i32 15>126 store <16 x i32> %interleaved.vec, ptr %a, align 4127 ret void128}129 130; Ensure we don't crash when trying to lower a shuffle via an extract131define void @crash_when_lowering_extract_shuffle(ptr %dst, i1 %cond) {132; CHECK-LABEL: crash_when_lowering_extract_shuffle:133; CHECK: // %bb.0:134; CHECK-NEXT: ret135;136; NONEON-NOSVE-LABEL: crash_when_lowering_extract_shuffle:137; NONEON-NOSVE: // %bb.0:138; NONEON-NOSVE-NEXT: ret139 %broadcast.splat = shufflevector <32 x i1> zeroinitializer, <32 x i1> zeroinitializer, <32 x i32> zeroinitializer140 br i1 %cond, label %exit, label %vector.body141 142vector.body:143 %1 = load <32 x i32>, ptr %dst, align 16144 %predphi = select <32 x i1> %broadcast.splat, <32 x i32> zeroinitializer, <32 x i32> %1145 store <32 x i32> %predphi, ptr %dst, align 16146 br label %exit147 148exit:149 ret void150}151