1; RUN: opt < %s -passes=loop-vectorize -mattr=+sse4.2 -debug-only=loop-vectorize 2>&1 -S | FileCheck %s 2; REQUIRES: asserts 3; 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 16 9@b = common global [2048 x i32] zeroinitializer, align 16 10@c = common global [2048 x i32] zeroinitializer, align 16 11 12; CHECK: Checking a loop in 'scalarselect' 13define void @scalarselect(i1 %cond) { 14 br label %1 15 16; <label>:1 17 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ] 18 %2 = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 %indvars.iv 19 %3 = load i32, ptr %2, align 4 20 %4 = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 %indvars.iv 21 %5 = load i32, ptr %4, align 4 22 %6 = add nsw i32 %5, %3 23 %7 = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 %indvars.iv 24 25; CHECK: cost of 1 for VF 1 {{.*}} select i1 %cond, i32 %6, i32 0 26; CHECK: Cost of 2 for VF 2: WIDEN-SELECT ir<%sel> = select ir<%cond>, ir<%6>, ir<0> (condition is loop invariant) 27; CHECK: Cost of 2 for VF 4: WIDEN-SELECT ir<%sel> = select ir<%cond>, ir<%6>, ir<0> (condition is loop invariant) 28 29 %sel = select i1 %cond, i32 %6, i32 zeroinitializer 30 store i32 %sel, ptr %7, align 4 31 %indvars.iv.next = add i64 %indvars.iv, 1 32 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 33 %exitcond = icmp eq i32 %lftr.wideiv, 256 34 br i1 %exitcond, label %8, label %1 35 36; <label>:8 37 ret void 38} 39 40; CHECK: Checking a loop in 'vectorselect' 41define void @vectorselect(i1 %cond) { 42 br label %1 43 44; <label>:1 45 %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ] 46 %2 = getelementptr inbounds [2048 x i32], ptr @b, i64 0, i64 %indvars.iv 47 %3 = load i32, ptr %2, align 4 48 %4 = getelementptr inbounds [2048 x i32], ptr @c, i64 0, i64 %indvars.iv 49 %5 = load i32, ptr %4, align 4 50 %6 = add nsw i32 %5, %3 51 %7 = getelementptr inbounds [2048 x i32], ptr @a, i64 0, i64 %indvars.iv 52 %8 = icmp ult i64 %indvars.iv, 8 53 54; CHECK: cost of 1 for VF 1 {{.*}} select i1 %8, i32 %6, i32 0 55; 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 zeroinitializer 59 store i32 %sel, ptr %7, align 4 60 %indvars.iv.next = add i64 %indvars.iv, 1 61 %lftr.wideiv = trunc i64 %indvars.iv.next to i32 62 %exitcond = icmp eq i32 %lftr.wideiv, 256 63 br i1 %exitcond, label %9, label %1 64 65; <label>:9 66 ret void 67} 68 69