0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
1 ; RUN: llc < %s -march=x86 -mattr=+sse2 -mcpu=yonah | FileCheck %s
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
2 ; RUN: llc < %s -march=x86-64 -mattr=+sse2 -mcpu=core2 | FileCheck %s
|
83
|
3 ; RUN: llc < %s -march=x86-64 -mattr=+avx -mcpu=btver2 | FileCheck %s
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
4
|
77
|
5 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
6
|
0
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
7 define i32 @t(<2 x i64>* %val) nounwind {
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
8 ; CHECK-LABEL: t:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
9 ; CHECK-NOT: movd
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
10 ; CHECK: movl 8(
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
11 ; CHECK-NEXT: ret
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
12 %tmp2 = load <2 x i64>* %val, align 16 ; <<2 x i64>> [#uses=1]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
13 %tmp3 = bitcast <2 x i64> %tmp2 to <4 x i32> ; <<4 x i32>> [#uses=1]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
14 %tmp4 = extractelement <4 x i32> %tmp3, i32 2 ; <i32> [#uses=1]
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
15 ret i32 %tmp4
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
16 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
17
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
18 ; Case where extractelement of load ends up as undef.
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
19 ; (Making sure this doesn't crash.)
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
20 define i32 @t2(<8 x i32>* %xp) {
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
21 ; CHECK-LABEL: t2:
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
22 ; CHECK: ret
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
23 %x = load <8 x i32>* %xp
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
24 %Shuff68 = shufflevector <8 x i32> %x, <8 x i32> undef, <8 x i32> <i32
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
25 undef, i32 7, i32 9, i32 undef, i32 13, i32 15, i32 1, i32 3>
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
26 %y = extractelement <8 x i32> %Shuff68, i32 0
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
27 ret i32 %y
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
28 }
|
77
|
29
|
|
30 ; This case could easily end up inf-looping in the DAG combiner due to an
|
|
31 ; low alignment load of the vector which prevents us from reliably forming a
|
|
32 ; narrow load.
|
83
|
33
|
|
34 ; The expected codegen is identical for the AVX case except
|
|
35 ; load/store instructions will have a leading 'v', so we don't
|
|
36 ; need to special-case the checks.
|
|
37
|
77
|
38 define void @t3() {
|
|
39 ; CHECK-LABEL: t3:
|
83
|
40 ; CHECK: movupd
|
|
41 ; CHECK: movhpd
|
77
|
42
|
|
43 bb:
|
|
44 %tmp13 = load <2 x double>* undef, align 1
|
|
45 %.sroa.3.24.vec.extract = extractelement <2 x double> %tmp13, i32 1
|
|
46 store double %.sroa.3.24.vec.extract, double* undef, align 8
|
|
47 unreachable
|
|
48 }
|
83
|
49
|
|
50 ; Case where a load is unary shuffled, then bitcast (to a type with the same
|
|
51 ; number of elements) before extractelement.
|
|
52 ; This is testing for an assertion - the extraction was assuming that the undef
|
|
53 ; second shuffle operand was a post-bitcast type instead of a pre-bitcast type.
|
|
54 define i64 @t4(<2 x double>* %a) {
|
|
55 ; CHECK-LABEL: t4:
|
|
56 ; CHECK: mov
|
|
57 ; CHECK: ret
|
|
58 %b = load <2 x double>* %a, align 16
|
|
59 %c = shufflevector <2 x double> %b, <2 x double> %b, <2 x i32> <i32 1, i32 0>
|
|
60 %d = bitcast <2 x double> %c to <2 x i64>
|
|
61 %e = extractelement <2 x i64> %d, i32 1
|
|
62 ret i64 %e
|
|
63 }
|
|
64
|