90 lines · plain
1; RUN: opt < %s -passes=loop-vectorize,dce,instcombine -pass-remarks=loop-vectorize -pass-remarks-analysis=loop-vectorize -pass-remarks-missed=loop-vectorize -S 2>&1 | FileCheck %s2 3target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"4 5target triple = "x86_64-unknown-linux"6 7; First loop produced diagnostic pass remark.8;CHECK: remark: {{.*}}:0:0: vectorized loop (vectorization width: 4, interleaved count: 2)9; Second loop produces diagnostic analysis remark.10;CHECK: remark: {{.*}}:0:0: vectorized loop (vectorization width: 4, interleaved count: 1)11 12; We are vectorizing with 6 runtime checks.13;CHECK-LABEL: func1x6(14;CHECK: <4 x i32>15;CHECK: ret16define void @func1x6(ptr nocapture %out, ptr nocapture %A, ptr nocapture %B, ptr nocapture %C, ptr nocapture %D, ptr nocapture %E, ptr nocapture %F) {17entry:18 br label %for.body19 20for.body: ; preds = %for.body, %entry21 %i.016 = phi i64 [ 0, %entry ], [ %inc, %for.body ]22 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %i.01623 %0 = load i32, ptr %arrayidx, align 424 %arrayidx1 = getelementptr inbounds i32, ptr %B, i64 %i.01625 %1 = load i32, ptr %arrayidx1, align 426 %add = add nsw i32 %1, %027 %arrayidx2 = getelementptr inbounds i32, ptr %C, i64 %i.01628 %2 = load i32, ptr %arrayidx2, align 429 %add3 = add nsw i32 %add, %230 %arrayidx4 = getelementptr inbounds i32, ptr %E, i64 %i.01631 %3 = load i32, ptr %arrayidx4, align 432 %add5 = add nsw i32 %add3, %333 %arrayidx6 = getelementptr inbounds i32, ptr %F, i64 %i.01634 %4 = load i32, ptr %arrayidx6, align 435 %add7 = add nsw i32 %add5, %436 %arrayidx8 = getelementptr inbounds i32, ptr %out, i64 %i.01637 store i32 %add7, ptr %arrayidx8, align 438 %inc = add i64 %i.016, 139 %exitcond = icmp eq i64 %inc, 25640 br i1 %exitcond, label %for.end, label %for.body41 42for.end: ; preds = %for.body43 ret void44}45 46; We are vectorizing with 12 runtime checks.47;CHECK-LABEL: func2x6(48;CHECK: <4 x i32>49;CHECK: ret50define void @func2x6(ptr nocapture %out, ptr nocapture %out2, ptr nocapture %A, ptr nocapture %B, ptr nocapture %C, ptr nocapture %D, ptr nocapture %E, ptr nocapture %F) {51entry:52 br label %for.body53 54for.body: ; preds = %for.body, %entry55 %i.037 = phi i64 [ 0, %entry ], [ %inc, %for.body ]56 %arrayidx = getelementptr inbounds i32, ptr %A, i64 %i.03757 %0 = load i32, ptr %arrayidx, align 458 %arrayidx1 = getelementptr inbounds i32, ptr %B, i64 %i.03759 %1 = load i32, ptr %arrayidx1, align 460 %add = add nsw i32 %1, %061 %arrayidx2 = getelementptr inbounds i32, ptr %C, i64 %i.03762 %2 = load i32, ptr %arrayidx2, align 463 %add3 = add nsw i32 %add, %264 %arrayidx4 = getelementptr inbounds i32, ptr %E, i64 %i.03765 %3 = load i32, ptr %arrayidx4, align 466 %add5 = add nsw i32 %add3, %367 %arrayidx6 = getelementptr inbounds i32, ptr %F, i64 %i.03768 %4 = load i32, ptr %arrayidx6, align 469 %add7 = add nsw i32 %add5, %470 %arrayidx8 = getelementptr inbounds i32, ptr %out, i64 %i.03771 store i32 %add7, ptr %arrayidx8, align 472 %5 = load i32, ptr %arrayidx, align 473 %6 = load i32, ptr %arrayidx1, align 474 %add11 = add nsw i32 %6, %575 %7 = load i32, ptr %arrayidx2, align 476 %add13 = add nsw i32 %add11, %777 %8 = load i32, ptr %arrayidx4, align 478 %add15 = add nsw i32 %add13, %879 %9 = load i32, ptr %arrayidx6, align 480 %add17 = add nsw i32 %add15, %981 %arrayidx18 = getelementptr inbounds i32, ptr %out2, i64 %i.03782 store i32 %add17, ptr %arrayidx18, align 483 %inc = add i64 %i.037, 184 %exitcond = icmp eq i64 %inc, 25685 br i1 %exitcond, label %for.end, label %for.body86 87for.end: ; preds = %for.body88 ret void89}90