69 lines · plain
1; RUN: opt < %s -passes=loop-vectorize -mattr=+sse4.2 -debug-only=loop-vectorize 2>&1 -S | FileCheck %s2; REQUIRES: asserts3; Make sure we use the right select kind when querying select costs.4 5target 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"6target triple = "x86_64-apple-macosx10.8.0"7 8@a = common global [2048 x i32] zeroinitializer, align 169@b = common global [2048 x i32] zeroinitializer, align 1610@c = common global [2048 x i32] zeroinitializer, align 1611 12; CHECK: Checking a loop in 'scalarselect'13define void @scalarselect(i1 %cond) {14 br label %115 16; <label>:117 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]18 %2 = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 %indvars.iv19 %3 = load i32, ptr %2, align 420 %4 = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 %indvars.iv21 %5 = load i32, ptr %4, align 422 %6 = add nsw i32 %5, %323 %7 = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 %indvars.iv24 25; CHECK: cost of 1 for VF 1 {{.*}} select i1 %cond, i32 %6, i32 026; CHECK: Cost of 2 for VF 2: WIDEN-SELECT ir<%sel> = select ir<%cond>, ir<%6>, ir<0> (condition is single-scalar)27; CHECK: Cost of 2 for VF 4: WIDEN-SELECT ir<%sel> = select ir<%cond>, ir<%6>, ir<0> (condition is single-scalar)28 29 %sel = select i1 %cond, i32 %6, i32 zeroinitializer30 store i32 %sel, ptr %7, align 431 %indvars.iv.next = add i64 %indvars.iv, 132 %lftr.wideiv = trunc i64 %indvars.iv.next to i3233 %exitcond = icmp eq i32 %lftr.wideiv, 25634 br i1 %exitcond, label %8, label %135 36; <label>:837 ret void38}39 40; CHECK: Checking a loop in 'vectorselect'41define void @vectorselect(i1 %cond) {42 br label %143 44; <label>:145 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]46 %2 = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 %indvars.iv47 %3 = load i32, ptr %2, align 448 %4 = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 %indvars.iv49 %5 = load i32, ptr %4, align 450 %6 = add nsw i32 %5, %351 %7 = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 %indvars.iv52 %8 = icmp ult i64 %indvars.iv, 853 54; CHECK: cost of 1 for VF 1 {{.*}} select i1 %8, i32 %6, i32 055; CHECK: Cost of 2 for VF 2: WIDEN-SELECT ir<%sel> = select ir<%8>, ir<%6>, ir<0>56; CHECK: Cost of 2 for VF 4: WIDEN-SELECT ir<%sel> = select ir<%8>, ir<%6>, ir<0>57 58 %sel = select i1 %8, i32 %6, i32 zeroinitializer59 store i32 %sel, ptr %7, align 460 %indvars.iv.next = add i64 %indvars.iv, 161 %lftr.wideiv = trunc i64 %indvars.iv.next to i3262 %exitcond = icmp eq i32 %lftr.wideiv, 25663 br i1 %exitcond, label %9, label %164 65; <label>:966 ret void67}68 69