1 ; RUN: opt < %s -basicaa -slp-vectorizer -S -mcpu=corei7-avx | FileCheck %s
3 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
4 target triple = "x86_64-unknown-linux-gnu"
6 define void @test1(x86_mmx %a, x86_mmx %b, i64* %ptr) {
7 ; Ensure we can handle x86_mmx values which are primitive and can be bitcast
8 ; with integer types but can't be put into a vector.
15 %a.cast = bitcast x86_mmx %a to i64
16 %b.cast = bitcast x86_mmx %b to i64
17 %a.and = and i64 %a.cast, 42
18 %b.and = and i64 %b.cast, 42
19 %gep = getelementptr i64, i64* %ptr, i32 1
20 store i64 %a.and, i64* %ptr
21 store i64 %b.and, i64* %gep
25 define void @test2(x86_mmx %a, x86_mmx %b) {
26 ; Same as @test1 but using phi-input vectorization instead of store
34 br i1 undef, label %if.then, label %exit
37 %a.cast = bitcast x86_mmx %a to i64
38 %b.cast = bitcast x86_mmx %b to i64
39 %a.and = and i64 %a.cast, 42
40 %b.and = and i64 %b.cast, 42
44 %a.phi = phi i64 [ 0, %entry ], [ %a.and, %if.then ]
45 %b.phi = phi i64 [ 0, %entry ], [ %b.and, %if.then ]
46 tail call void @f(i64 %a.phi, i64 %b.phi)
50 define i8 @test3(i8 *%addr) {
51 ; Check that we do not vectorize types that are padded to a bigger ones.
57 %a = bitcast i8* %addr to i2*
58 %a0 = getelementptr inbounds i2, i2* %a, i64 0
59 %a1 = getelementptr inbounds i2, i2* %a, i64 1
60 %a2 = getelementptr inbounds i2, i2* %a, i64 2
61 %a3 = getelementptr inbounds i2, i2* %a, i64 3
62 %l0 = load i2, i2* %a0, align 1
63 %l1 = load i2, i2* %a1, align 1
64 %l2 = load i2, i2* %a2, align 1
65 %l3 = load i2, i2* %a3, align 1
68 %p0 = phi i2 [ %l0, %entry ]
69 %p1 = phi i2 [ %l1, %entry ]
70 %p2 = phi i2 [ %l2, %entry ]
71 %p3 = phi i2 [ %l3, %entry ]
72 %r = zext i2 %p2 to i8
76 declare void @f(i64, i64)