xref: /llvm-project/llvm/test/Analysis/CostModel/X86/uniformshift.ll (revision 0ec028fe105b9bcdda4196cc8803205a1577c12a) 
1; RUN: llc -mtriple=x86_64-apple-darwin -mattr=+sse2 < %s | FileCheck --check-prefix=SSE2-CODEGEN %s
2; RUN: opt -mtriple=x86_64-apple-darwin -mattr=+sse2 -passes="print<cost-model>" 2>&1 -disable-output < %s | FileCheck --check-prefix=SSE2 %s
3
4define <4 x i32> @shl(<4 x i32> %vector, i32 %scalar) {
5entry:
6  ; SSE2: 'shl'
7  ; SSE2: cost of 2 {{.*}} shl
8  ; SSE2-CODEGEN: movd  %edi, %xmm1
9  ; SSE2-CODEGEN: pslld %xmm1, %xmm0
10  %insert = insertelement <4 x i32> undef, i32 %scalar, i32 0
11  %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
12  %ret = shl <4 x i32> %vector , %splat
13  ret <4 x i32> %ret
14}
15
16define <4 x i32> @ashr(<4 x i32> %vector, i32 %scalar) {
17entry:
18  ; SSE2: 'ashr'
19  ; SSE2: cost of 2 {{.*}} ashr
20  ; SSE2-CODEGEN: movd  %edi, %xmm1
21  ; SSE2-CODEGEN: psrad %xmm1, %xmm0
22  %insert = insertelement <4 x i32> undef, i32 %scalar, i32 0
23  %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
24  %ret = ashr <4 x i32> %vector , %splat
25  ret <4 x i32> %ret
26}
27
28define <4 x i32> @lshr(<4 x i32> %vector, i32 %scalar) {
29entry:
30  ; SSE2: 'lshr'
31  ; SSE2: cost of 2 {{.*}} lshr
32  ; SSE2-CODEGEN: movd  %edi, %xmm1
33  ; SSE2-CODEGEN: psrld %xmm1, %xmm0
34  %insert = insertelement <4 x i32> undef, i32 %scalar, i32 0
35  %splat = shufflevector <4 x i32> %insert, <4 x i32> undef, <4 x i32> zeroinitializer
36  %ret = lshr <4 x i32> %vector , %splat
37  ret <4 x i32> %ret
38}
39
40