1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt -passes=slp-vectorizer -S %s | FileCheck %s
4 declare float @memread(float) readonly nounwind willreturn #0
5 declare <4 x float> @vmemread(<4 x float>)
7 define <4 x float> @memread_4x(ptr %a) {
8 ; CHECK-LABEL: @memread_4x(
10 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A:%.*]], align 16
11 ; CHECK-NEXT: [[TMP1:%.*]] = call fast <4 x float> @vmemread(<4 x float> [[TMP0]])
12 ; CHECK-NEXT: ret <4 x float> [[TMP1]]
15 %0 = load <4 x float>, ptr %a, align 16
16 %vecext = extractelement <4 x float> %0, i32 0
17 %1 = tail call fast float @memread(float %vecext) #0
18 %vecins = insertelement <4 x float> undef, float %1, i32 0
19 %vecext.1 = extractelement <4 x float> %0, i32 1
20 %2 = tail call fast float @memread(float %vecext.1) #0
21 %vecins.1 = insertelement <4 x float> %vecins, float %2, i32 1
22 %vecext.2 = extractelement <4 x float> %0, i32 2
23 %3 = tail call fast float @memread(float %vecext.2) #0
24 %vecins.2 = insertelement <4 x float> %vecins.1, float %3, i32 2
25 %vecext.3 = extractelement <4 x float> %0, i32 3
26 %4 = tail call fast float @memread(float %vecext.3) #0
27 %vecins.3 = insertelement <4 x float> %vecins.2, float %4, i32 3
28 ret <4 x float> %vecins.3
31 declare float @memwrite(float) nounwind willreturn #1
32 declare <4 x float> @vmemwrite(<4 x float>)
34 define <4 x float> @memwrite_4x(ptr %a) {
35 ; CHECK-LABEL: @memwrite_4x(
37 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A:%.*]], align 16
38 ; CHECK-NEXT: [[VECEXT:%.*]] = extractelement <4 x float> [[TMP0]], i32 0
39 ; CHECK-NEXT: [[TMP1:%.*]] = tail call fast float @memwrite(float [[VECEXT]]) #[[ATTR2:[0-9]+]]
40 ; CHECK-NEXT: [[VECINS:%.*]] = insertelement <4 x float> undef, float [[TMP1]], i32 0
41 ; CHECK-NEXT: [[VECEXT_1:%.*]] = extractelement <4 x float> [[TMP0]], i32 1
42 ; CHECK-NEXT: [[TMP2:%.*]] = tail call fast float @memwrite(float [[VECEXT_1]]) #[[ATTR2]]
43 ; CHECK-NEXT: [[VECINS_1:%.*]] = insertelement <4 x float> [[VECINS]], float [[TMP2]], i32 1
44 ; CHECK-NEXT: [[VECEXT_2:%.*]] = extractelement <4 x float> [[TMP0]], i32 2
45 ; CHECK-NEXT: [[TMP3:%.*]] = tail call fast float @memwrite(float [[VECEXT_2]]) #[[ATTR2]]
46 ; CHECK-NEXT: [[VECINS_2:%.*]] = insertelement <4 x float> [[VECINS_1]], float [[TMP3]], i32 2
47 ; CHECK-NEXT: [[VECEXT_3:%.*]] = extractelement <4 x float> [[TMP0]], i32 3
48 ; CHECK-NEXT: [[TMP4:%.*]] = tail call fast float @memwrite(float [[VECEXT_3]]) #[[ATTR2]]
49 ; CHECK-NEXT: [[VECINS_3:%.*]] = insertelement <4 x float> [[VECINS_2]], float [[TMP4]], i32 3
50 ; CHECK-NEXT: ret <4 x float> [[VECINS_3]]
53 %0 = load <4 x float>, ptr %a, align 16
54 %vecext = extractelement <4 x float> %0, i32 0
55 %1 = tail call fast float @memwrite(float %vecext) #1
56 %vecins = insertelement <4 x float> undef, float %1, i32 0
57 %vecext.1 = extractelement <4 x float> %0, i32 1
58 %2 = tail call fast float @memwrite(float %vecext.1) #1
59 %vecins.1 = insertelement <4 x float> %vecins, float %2, i32 1
60 %vecext.2 = extractelement <4 x float> %0, i32 2
61 %3 = tail call fast float @memwrite(float %vecext.2) #1
62 %vecins.2 = insertelement <4 x float> %vecins.1, float %3, i32 2
63 %vecext.3 = extractelement <4 x float> %0, i32 3
64 %4 = tail call fast float @memwrite(float %vecext.3) #1
65 %vecins.3 = insertelement <4 x float> %vecins.2, float %4, i32 3
66 ret <4 x float> %vecins.3
69 attributes #0 = { "vector-function-abi-variant"="_ZGV_LLVM_N4v_memread(vmemread)" }
70 attributes #1 = { "vector-function-abi-variant"="_ZGV_LLVM_N4v_memwrite(vmemwrite)" }