Home | History | Annotate | Download | only in X86
      1 ; RUN: opt < %s  -loop-vectorize -mcpu=core2 -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 
      5 target 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"
      6 target 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"
     13 define 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], [2048 x i32]* @b, i64 0, i64 %indvars.iv
     19   %3 = load i32, i32* %2, align 4
     20   %4 = getelementptr inbounds [2048 x i32], [2048 x i32]* @c, i64 0, i64 %indvars.iv
     21   %5 = load i32, i32* %4, align 4
     22   %6 = add nsw i32 %5, %3
     23   %7 = getelementptr inbounds [2048 x i32], [2048 x i32]* @a, i64 0, i64 %indvars.iv
     24 
     25 ; A scalar select has a cost of 1 on core2
     26 ; CHECK: cost of 1 for VF 2 {{.*}}  select i1 %cond, i32 %6, i32 0
     27 
     28   %sel = select i1 %cond, i32 %6, i32 zeroinitializer
     29   store i32 %sel, i32* %7, align 4
     30   %indvars.iv.next = add i64 %indvars.iv, 1
     31   %lftr.wideiv = trunc i64 %indvars.iv.next to i32
     32   %exitcond = icmp eq i32 %lftr.wideiv, 256
     33   br i1 %exitcond, label %8, label %1
     34 
     35 ; <label>:8
     36   ret void
     37 }
     38 
     39 ; CHECK: Checking a loop in "vectorselect"
     40 define void @vectorselect(i1 %cond) {
     41   br label %1
     42 
     43 ; <label>:1
     44   %indvars.iv = phi i64 [ 0, %0 ], [ %indvars.iv.next, %1 ]
     45   %2 = getelementptr inbounds [2048 x i32], [2048 x i32]* @b, i64 0, i64 %indvars.iv
     46   %3 = load i32, i32* %2, align 4
     47   %4 = getelementptr inbounds [2048 x i32], [2048 x i32]* @c, i64 0, i64 %indvars.iv
     48   %5 = load i32, i32* %4, align 4
     49   %6 = add nsw i32 %5, %3
     50   %7 = getelementptr inbounds [2048 x i32], [2048 x i32]* @a, i64 0, i64 %indvars.iv
     51   %8 = icmp ult i64 %indvars.iv, 8
     52 
     53 ; A vector select has a cost of 1 on core2
     54 ; CHECK: cost of 1 for VF 2 {{.*}}  select i1 %8, i32 %6, i32 0
     55 
     56   %sel = select i1 %8, i32 %6, i32 zeroinitializer
     57   store i32 %sel, i32* %7, align 4
     58   %indvars.iv.next = add i64 %indvars.iv, 1
     59   %lftr.wideiv = trunc i64 %indvars.iv.next to i32
     60   %exitcond = icmp eq i32 %lftr.wideiv, 256
     61   br i1 %exitcond, label %9, label %1
     62 
     63 ; <label>:9
     64   ret void
     65 }
     66 
     67