1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=aarch64-unknown-unknown | FileCheck %s 3 4; This test is reduced from the TSVC evaluation of vectorizers: 5; https://github.com/llvm/llvm-test-suite/commits/main/MultiSource/Benchmarks/TSVC/LoopRerolling-flt/tsc.c 6 7; FIXME 8; This test is currently getting vectorized with VF=2. We should be able 9; to vectorize it with VF=4. Specifically, we should be able to have 1 load of 10; <4 x float> instead of 2 loads of <2 x float>, and there should be no need 11; for shufflevectors. 12; The current issue comes from the Left-Hand-Side fmul operands. 13; These operands are coming from 4 loads which are not 14; contiguous. The score estimation needs to be corrected, so that these 4 loads 15; are not selected for vectorization. Instead we should vectorize with 16; contiguous loads, from %a plus offsets 0 to 3, or offsets 1 to 4. 17 18define void @s116_modified(ptr %a) { 19; CHECK-LABEL: @s116_modified( 20; CHECK-NEXT: [[A:%.*]] = getelementptr inbounds float, ptr [[GEP1:%.*]], i64 2 21; CHECK-NEXT: [[GEP3:%.*]] = getelementptr inbounds float, ptr [[GEP1]], i64 3 22; CHECK-NEXT: [[LD0:%.*]] = load float, ptr [[A]], align 4 23; CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[GEP1]], align 4 24; CHECK-NEXT: [[TMP2:%.*]] = load <2 x float>, ptr [[GEP3]], align 4 25; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <2 x float> [[TMP1]], <2 x float> poison, <4 x i32> <i32 0, i32 1, i32 poison, i32 poison> 26; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x float> [[TMP4]], float [[LD0]], i32 1 27; CHECK-NEXT: [[TMP6:%.*]] = call <4 x float> @llvm.vector.insert.v4f32.v2f32(<4 x float> [[TMP5]], <2 x float> [[TMP2]], i64 2) 28; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[TMP6]], <4 x i32> <i32 1, i32 1, i32 5, i32 6> 29; CHECK-NEXT: [[TMP8:%.*]] = fmul fast <4 x float> [[TMP6]], [[TMP7]] 30; CHECK-NEXT: store <4 x float> [[TMP8]], ptr [[GEP1]], align 4 31; CHECK-NEXT: ret void 32; 33 %gep1 = getelementptr inbounds float, ptr %a, i64 1 34 %gep2 = getelementptr inbounds float, ptr %a, i64 2 35 %gep3 = getelementptr inbounds float, ptr %a, i64 3 36 %gep4 = getelementptr inbounds float, ptr %a, i64 4 37 %ld0 = load float, ptr %a 38 %ld1 = load float, ptr %gep1 39 %ld2 = load float, ptr %gep2 40 %ld3 = load float, ptr %gep3 41 %ld4 = load float, ptr %gep4 42 %mul0 = fmul fast float %ld0, %ld1 43 %mul1 = fmul fast float %ld2, %ld1 44 %mul2 = fmul fast float %ld3, %ld2 45 %mul3 = fmul fast float %ld4, %ld3 46 store float %mul0, ptr %a 47 store float %mul1, ptr %gep1 48 store float %mul2, ptr %gep2 49 store float %mul3, ptr %gep3 50 ret void 51} 52 53 54