91 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --filter "LV: Found an estimated cost of [0-9]+ for VF [0-9]+ For instruction:\s*%valB.loaded = load i32, ptr %inB, align 4"2; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+sse2 --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=SSE3; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+sse4.2 --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=SSE4; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+avx --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=AVX15; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+avx2,-fast-gather --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=AVX2-SLOWGATHER6; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+avx2,+fast-gather --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=AVX2-FASTGATHER7; RUN: opt -passes=loop-vectorize -vectorizer-maximize-bandwidth -S -mattr=+avx512bw --debug-only=loop-vectorize --disable-output < %s 2>&1 | FileCheck %s --check-prefixes=AVX5128 9; REQUIRES: asserts10 11target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"12target triple = "x86_64-unknown-linux-gnu"13 14@A = global [1024 x i8] zeroinitializer, align 12815@B = global [1024 x i32] zeroinitializer, align 12816@C = global [1024 x i32] zeroinitializer, align 12817 18define void @test() {19; SSE-LABEL: 'test'20; SSE: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i32, ptr %inB, align 421; SSE: Cost of 3000000 for VF 2: {{.*}}ir<%valB.loaded> = load22; SSE: Cost of 3000000 for VF 4: {{.*}}ir<%valB.loaded> = load23; SSE: Cost of 3000000 for VF 8: {{.*}}ir<%valB.loaded> = load24; SSE: Cost of 3000000 for VF 16: {{.*}}ir<%valB.loaded> = load25;26; AVX1-LABEL: 'test'27; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i32, ptr %inB, align 428; AVX1: Cost of 3000000 for VF 2: {{.*}}ir<%valB.loaded> = load29; AVX1: Cost of 3000000 for VF 4: {{.*}}ir<%valB.loaded> = load30; AVX1: Cost of 3000000 for VF 8: {{.*}}ir<%valB.loaded> = load31; AVX1: Cost of 3000000 for VF 16: {{.*}}ir<%valB.loaded> = load32; AVX1: Cost of 3000000 for VF 32: {{.*}}ir<%valB.loaded> = load33;34; AVX2-SLOWGATHER-LABEL: 'test'35; AVX2-SLOWGATHER: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i32, ptr %inB, align 436; AVX2-SLOWGATHER: Cost of 3000000 for VF 2: {{.*}}ir<%valB.loaded> = load37; AVX2-SLOWGATHER: Cost of 3000000 for VF 4: {{.*}}ir<%valB.loaded> = load38; AVX2-SLOWGATHER: Cost of 3000000 for VF 8: {{.*}}ir<%valB.loaded> = load39; AVX2-SLOWGATHER: Cost of 3000000 for VF 16: {{.*}}ir<%valB.loaded> = load40; AVX2-SLOWGATHER: Cost of 3000000 for VF 32: {{.*}}ir<%valB.loaded> = load41;42; AVX2-FASTGATHER-LABEL: 'test'43; AVX2-FASTGATHER: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i32, ptr %inB, align 444; AVX2-FASTGATHER: Cost of 4 for VF 2: {{.*}}ir<%valB.loaded> = load45; AVX2-FASTGATHER: Cost of 6 for VF 4: {{.*}}ir<%valB.loaded> = load46; AVX2-FASTGATHER: Cost of 12 for VF 8: {{.*}}ir<%valB.loaded> = load47; AVX2-FASTGATHER: Cost of 24 for VF 16: {{.*}}ir<%valB.loaded> = load48; AVX2-FASTGATHER: Cost of 48 for VF 32: {{.*}}ir<%valB.loaded> = load49;50; AVX512-LABEL: 'test'51; AVX512: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i32, ptr %inB, align 452; AVX512: Cost of 8 for VF 2: {{.*}}ir<%valB.loaded> = load53; AVX512: Cost of 17 for VF 4: {{.*}}ir<%valB.loaded> = load54; AVX512: Cost of 10 for VF 8: {{.*}}ir<%valB.loaded> = load55; AVX512: Cost of 18 for VF 16: {{.*}}ir<%valB.loaded> = load56; AVX512: Cost of 36 for VF 32: {{.*}}ir<%valB.loaded> = load57; AVX512: Cost of 72 for VF 64: {{.*}}ir<%valB.loaded> = load58;59entry:60 br label %for.body61 62for.body:63 %iv = phi i64 [ 0, %entry ], [ %iv.next, %end ]64 65 %inA = getelementptr inbounds [1024 x i8], ptr @A, i64 0, i64 %iv66 %valA = load i8, ptr %inA67 %canLoad = icmp ne i8 %valA, 068 br i1 %canLoad, label %load, label %mask69 70load:71 %valA.ext = sext i8 %valA to i6472 %inB = getelementptr inbounds [1024 x i32], ptr @B, i64 0, i64 %valA.ext73 %valB.loaded = load i32, ptr %inB74 br label %end75 76mask:77 br label %end78 79end:80 %valB = phi i32 [ %valB.loaded, %load ], [ 0, %mask ]81 %out = getelementptr inbounds [1024 x i32], ptr @C, i64 0, i64 %iv82 store i32 %valB, ptr %out83 84 %iv.next = add nuw nsw i64 %iv, 185 %cmp = icmp ult i64 %iv.next, 102486 br i1 %cmp, label %for.body, label %for.cond.cleanup87 88for.cond.cleanup:89 ret void90}91