116 lines · plain
1; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=4 -force-vector-width=4 -debug-only=loop-vectorize -enable-early-exit-vectorization --disable-output -stats -S 2>&1 | FileCheck %s2; REQUIRES: asserts3 4; We have 3 loops, two of them are vectorizable (with one being early-exit5; vectorized) and the third one is not.6 7; CHECK: 4 loop-vectorize - Number of loops analyzed for vectorization8; CHECK: 1 loop-vectorize - Number of early exit loops vectorized9; CHECK: 3 loop-vectorize - Number of loops vectorized10 11target 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"12 13define void @vectorized(ptr nocapture %a, i64 %size) {14entry:15 %cmp1 = icmp sle i64 %size, 016 %cmp21 = icmp sgt i64 0, %size17 %or.cond = or i1 %cmp1, %cmp2118 br i1 %or.cond, label %exit, label %loop19 20loop:21 %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]22 %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv23 %0 = load float, ptr %arrayidx, align 424 %mul = fmul float %0, %025 store float %mul, ptr %arrayidx, align 426 %iv.next = add nuw nsw i64 %iv, 127 %cmp2 = icmp sgt i64 %iv.next, %size28 br i1 %cmp2, label %exit, label %loop29 30exit: ; preds = %entry, %loop31 ret void32}33 34define i32 @early_exit_vectorized(i64 %end) {35entry:36 %p1 = alloca [1024 x i32]37 %p2 = alloca [1024 x i32]38 call void @init_mem(ptr %p1, i64 1024)39 call void @init_mem(ptr %p2, i64 1024)40 %end.clamped = and i64 %end, 102341 br label %loop42 43loop:44 %ind = phi i64 [ %ind.next, %for.inc ], [ 0, %entry ]45 %arrayidx1 = getelementptr inbounds i32, ptr %p1, i64 %ind46 %0 = load i32, ptr %arrayidx1, align 447 %arrayidx2 = getelementptr inbounds i32, ptr %p2, i64 %ind48 %1 = load i32, ptr %arrayidx2, align 449 %cmp.early = icmp eq i32 %0, %150 br i1 %cmp.early, label %found, label %for.inc51 52for.inc:53 %ind.next = add i64 %ind, 154 %cmp = icmp ult i64 %ind.next, %end.clamped55 br i1 %cmp, label %loop, label %exit56 57found:58 ret i32 159 60exit:61 ret i32 062}63 64define void @not_vectorized(ptr nocapture %a, i64 %size) {65entry:66 %cmp1 = icmp sle i64 %size, 067 %cmp21 = icmp sgt i64 0, %size68 %or.cond = or i1 %cmp1, %cmp2169 br i1 %or.cond, label %exit, label %loop70 71loop:72 %iv = phi i64 [ %iv.next, %loop ], [ 0, %entry ]73 %0 = add nsw i64 %iv, -574 %arrayidx = getelementptr inbounds float, ptr %a, i64 %075 %1 = load float, ptr %arrayidx, align 476 %2 = add nsw i64 %iv, 277 %arrayidx2 = getelementptr inbounds float, ptr %a, i64 %278 %3 = load float, ptr %arrayidx2, align 479 %mul = fmul float %1, %380 %arrayidx4 = getelementptr inbounds float, ptr %a, i64 %iv81 store float %mul, ptr %arrayidx4, align 482 %iv.next = add nuw nsw i64 %iv, 183 %cmp2 = icmp sgt i64 %iv.next, %size84 br i1 %cmp2, label %exit, label %loop85 86exit:87 ret void88}89 90define i1 @multiple_countable_exits_multiple_exit_block(ptr %A, ptr %B, i32 %N) {91entry:92 br label %loop.header93 94loop.header:95 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.latch ]96 %cond.0 = icmp eq i32 %iv, %N97 br i1 %cond.0, label %exit.0, label %loop.latch98 99loop.latch:100 %A.gep = getelementptr inbounds i32, ptr %A, i32 %iv101 %lv = load i32, ptr %A.gep, align 4102 %B.gep = getelementptr inbounds i32, ptr %B, i32 %iv103 store i32 %lv, ptr %B.gep, align 4104 %iv.next = add nuw i32 %iv, 1105 %cond.1 = icmp ult i32 %iv.next, 1000106 br i1 %cond.1, label %loop.header, label %exit.1107 108exit.0:109 ret i1 false110 111exit.1:112 ret i1 true113}114 115declare void @init_mem(ptr, i64);116