81 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 i64, ptr %inB, align 8"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=AVX26; 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=AVX27; 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@C = global [1024 x i64] zeroinitializer, align 12816 17define void @test(ptr %B) {18; SSE-LABEL: 'test'19; SSE: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i64, ptr %inB, align 820; SSE: Cost of 3000000 for VF 2: {{.*}}ir<%valB.loaded> = load21; SSE: Cost of 3000000 for VF 4: {{.*}}ir<%valB.loaded> = load22; SSE: Cost of 3000000 for VF 8: {{.*}}ir<%valB.loaded> = load23; SSE: Cost of 3000000 for VF 16: {{.*}}ir<%valB.loaded> = load24;25; AVX1-LABEL: 'test'26; AVX1: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i64, ptr %inB, align 827; AVX1: Cost of 2 for VF 2: {{.*}}ir<%valB.loaded> = load28; AVX1: Cost of 2 for VF 4: {{.*}}ir<%valB.loaded> = load29; AVX1: Cost of 4 for VF 8: {{.*}}ir<%valB.loaded> = load30; AVX1: Cost of 8 for VF 16: {{.*}}ir<%valB.loaded> = load31; AVX1: Cost of 16 for VF 32: {{.*}}ir<%valB.loaded> = load32;33; AVX2-LABEL: 'test'34; AVX2: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i64, ptr %inB, align 835; AVX2: Cost of 2 for VF 2: {{.*}}ir<%valB.loaded> = load36; AVX2: Cost of 2 for VF 4: {{.*}}ir<%valB.loaded> = load37; AVX2: Cost of 4 for VF 8: {{.*}}ir<%valB.loaded> = load38; AVX2: Cost of 8 for VF 16: {{.*}}ir<%valB.loaded> = load39; AVX2: Cost of 16 for VF 32: {{.*}}ir<%valB.loaded> = load40;41; AVX512-LABEL: 'test'42; AVX512: LV: Found an estimated cost of 1 for VF 1 For instruction: %valB.loaded = load i64, ptr %inB, align 843; AVX512: Cost of 1 for VF 2: {{.*}}ir<%valB.loaded> = load44; AVX512: Cost of 1 for VF 4: {{.*}}ir<%valB.loaded> = load45; AVX512: Cost of 1 for VF 8: {{.*}}ir<%valB.loaded> = load46; AVX512: Cost of 2 for VF 16: {{.*}}ir<%valB.loaded> = load47; AVX512: Cost of 4 for VF 32: {{.*}}ir<%valB.loaded> = load48; AVX512: Cost of 8 for VF 64: {{.*}}ir<%valB.loaded> = load49;50entry:51 br label %for.body52 53for.body:54 %iv = phi i64 [ 0, %entry ], [ %iv.next, %end ]55 56 %inA = getelementptr inbounds [1024 x i8], ptr @A, i64 0, i64 %iv57 %valA = load i8, ptr %inA58 %canLoad = icmp ne i8 %valA, 059 br i1 %canLoad, label %load, label %mask60 61load:62 %inB = getelementptr inbounds [1024 x i64], ptr %B, i64 0, i64 %iv63 %valB.loaded = load i64, ptr %inB64 br label %end65 66mask:67 br label %end68 69end:70 %valB = phi i64 [ %valB.loaded, %load ], [ 0, %mask ]71 %out = getelementptr inbounds [1024 x i64], ptr @C, i64 0, i64 %iv72 store i64 %valB, ptr %out73 74 %iv.next = add nuw nsw i64 %iv, 175 %cmp = icmp ult i64 %iv.next, 102476 br i1 %cmp, label %for.body, label %for.cond.cleanup77 78for.cond.cleanup:79 ret void80}81