1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown -mattr=+sse2 | FileCheck %s --check-prefixes=X32-SSE2
3 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3 | FileCheck %s --check-prefixes=X64,X64-SSSE3
4 ; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx | FileCheck %s --check-prefixes=X64,X64-AVX
6 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
8 define i32 @t(<2 x i64>* %val) nounwind {
11 ; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
12 ; X32-SSE2-NEXT: movl 8(%eax), %eax
17 ; X64-NEXT: movl 8(%rdi), %eax
19 %tmp2 = load <2 x i64>, <2 x i64>* %val, align 16 ; <<2 x i64>> [#uses=1]
20 %tmp3 = bitcast <2 x i64> %tmp2 to <4 x i32> ; <<4 x i32>> [#uses=1]
21 %tmp4 = extractelement <4 x i32> %tmp3, i32 2 ; <i32> [#uses=1]
25 ; Case where extractelement of load ends up as undef.
26 ; (Making sure this doesn't crash.)
27 define i32 @t2(<8 x i32>* %xp) {
35 %x = load <8 x i32>, <8 x i32>* %xp
36 %Shuff68 = shufflevector <8 x i32> %x, <8 x i32> undef, <8 x i32> <i32 undef, i32 7, i32 9, i32 undef, i32 13, i32 15, i32 1, i32 3>
37 %y = extractelement <8 x i32> %Shuff68, i32 0
41 ; This case could easily end up inf-looping in the DAG combiner due to an
42 ; low alignment load of the vector which prevents us from reliably forming a
45 define void @t3(<2 x double>* %a0) {
47 ; X32-SSE2: # %bb.0: # %bb
48 ; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
49 ; X32-SSE2-NEXT: movups (%eax), %xmm0
50 ; X32-SSE2-NEXT: movhps %xmm0, (%eax)
53 ; X64-SSSE3-LABEL: t3:
54 ; X64-SSSE3: # %bb.0: # %bb
55 ; X64-SSSE3-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
56 ; X64-SSSE3-NEXT: movsd %xmm0, (%rax)
57 ; X64-SSSE3-NEXT: retq
60 ; X64-AVX: # %bb.0: # %bb
61 ; X64-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
62 ; X64-AVX-NEXT: vmovsd %xmm0, (%rax)
65 %tmp13 = load <2 x double>, <2 x double>* %a0, align 1
66 %.sroa.3.24.vec.extract = extractelement <2 x double> %tmp13, i32 1
67 store double %.sroa.3.24.vec.extract, double* undef, align 8
71 ; Case where a load is unary shuffled, then bitcast (to a type with the same
72 ; number of elements) before extractelement.
73 ; This is testing for an assertion - the extraction was assuming that the undef
74 ; second shuffle operand was a post-bitcast type instead of a pre-bitcast type.
75 define i64 @t4(<2 x double>* %a) {
78 ; X32-SSE2-NEXT: movl {{[0-9]+}}(%esp), %eax
79 ; X32-SSE2-NEXT: movdqa (%eax), %xmm0
80 ; X32-SSE2-NEXT: movd %xmm0, %eax
81 ; X32-SSE2-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,0,1]
82 ; X32-SSE2-NEXT: movd %xmm0, %edx
87 ; X64-NEXT: movq (%rdi), %rax
89 %b = load <2 x double>, <2 x double>* %a, align 16
90 %c = shufflevector <2 x double> %b, <2 x double> %b, <2 x i32> <i32 1, i32 0>
91 %d = bitcast <2 x double> %c to <2 x i64>
92 %e = extractelement <2 x i64> %d, i32 1