3 ; RUN: opt -passes=loop-vectorize -mtriple=arm64-apple-ios %s -S -debug -disable-output 2>&1 | FileCheck --check-prefix=CM %s
4 ; RUN: opt -passes=loop-vectorize -force-vector-width=2 -force-vector-interleave=1 %s -S | FileCheck --check-prefix=FORCED %s
6 ; Test case from PR41294.
8 ; Check scalar cost for extractvalue. The constant and loop invariant operands are free,
9 ; leaving cost 3 for scalarizing the result + 2 for executing the op with VF 2.
11 ; CM: LV: Found uniform instruction: %a = extractvalue { i64, i64 } %sv, 0
12 ; CM: LV: Found uniform instruction: %b = extractvalue { i64, i64 } %sv, 1
14 ; CM: LV: Scalar loop costs: 5.
15 ; CM: LV: Found an estimated cost of 0 for VF 2 For instruction: %a = extractvalue { i64, i64 } %sv, 0
16 ; CM-NEXT: LV: Found an estimated cost of 0 for VF 2 For instruction: %b = extractvalue { i64, i64 } %sv, 1
18 ; Check that the extractvalue operands are actually free in vector code.
20 ; FORCED-LABEL: vector.body: ; preds = %vector.body, %vector.ph
21 ; FORCED-NEXT: %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
22 ; FORCED-NEXT: %0 = add i32 %index, 0
23 ; FORCED-NEXT: %1 = extractvalue { i64, i64 } %sv, 0
24 ; FORCED-NEXT: %broadcast.splatinsert = insertelement <2 x i64> poison, i64 %1, i64 0
25 ; FORCED-NEXT: %broadcast.splat = shufflevector <2 x i64> %broadcast.splatinsert, <2 x i64> poison, <2 x i32> zeroinitializer
26 ; FORCED-NEXT: %2 = extractvalue { i64, i64 } %sv, 1
27 ; FORCED-NEXT: %broadcast.splatinsert1 = insertelement <2 x i64> poison, i64 %2, i64 0
28 ; FORCED-NEXT: %broadcast.splat2 = shufflevector <2 x i64> %broadcast.splatinsert1, <2 x i64> poison, <2 x i32> zeroinitializer
29 ; FORCED-NEXT: %3 = getelementptr i64, ptr %dst, i32 %0
30 ; FORCED-NEXT: %4 = add <2 x i64> %broadcast.splat, %broadcast.splat2
31 ; FORCED-NEXT: %5 = getelementptr i64, ptr %3, i32 0
32 ; FORCED-NEXT: store <2 x i64> %4, ptr %5, align 4
33 ; FORCED-NEXT: %index.next = add nuw i32 %index, 2
34 ; FORCED-NEXT: %6 = icmp eq i32 %index.next, 1000
35 ; FORCED-NEXT: br i1 %6, label %middle.block, label %vector.body
37 define void @test1(ptr %dst, {i64, i64} %sv) {
42 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.body ]
43 %a = extractvalue { i64, i64 } %sv, 0
44 %b = extractvalue { i64, i64 } %sv, 1
45 %addr = getelementptr i64, ptr %dst, i32 %iv
47 store i64 %add, ptr %addr
48 %iv.next = add nsw i32 %iv, 1
49 %cond = icmp ne i32 %iv.next, 1000
50 br i1 %cond, label %loop.body, label %exit
57 ; Similar to the test case above, but checks getVectorCallCost as well.
58 declare float @powf(float, float) readnone nounwind
60 ; CM: LV: Found uniform instruction: %a = extractvalue { float, float } %sv, 0
61 ; CM: LV: Found uniform instruction: %b = extractvalue { float, float } %sv, 1
63 ; CM: LV: Scalar loop costs: 14.
64 ; CM: LV: Found an estimated cost of 0 for VF 2 For instruction: %a = extractvalue { float, float } %sv, 0
65 ; CM-NEXT: LV: Found an estimated cost of 0 for VF 2 For instruction: %b = extractvalue { float, float } %sv, 1
67 ; FORCED-LABEL: define void @test_getVectorCallCost
69 ; FORCED-LABEL: vector.body: ; preds = %vector.body, %vector.ph
70 ; FORCED-NEXT: %index = phi i32 [ 0, %vector.ph ], [ %index.next, %vector.body ]
71 ; FORCED-NEXT: %0 = add i32 %index, 0
72 ; FORCED-NEXT: %1 = extractvalue { float, float } %sv, 0
73 ; FORCED-NEXT: %broadcast.splatinsert = insertelement <2 x float> poison, float %1, i64 0
74 ; FORCED-NEXT: %broadcast.splat = shufflevector <2 x float> %broadcast.splatinsert, <2 x float> poison, <2 x i32> zeroinitializer
75 ; FORCED-NEXT: %2 = extractvalue { float, float } %sv, 1
76 ; FORCED-NEXT: %broadcast.splatinsert1 = insertelement <2 x float> poison, float %2, i64 0
77 ; FORCED-NEXT: %broadcast.splat2 = shufflevector <2 x float> %broadcast.splatinsert1, <2 x float> poison, <2 x i32> zeroinitializer
78 ; FORCED-NEXT: %3 = getelementptr float, ptr %dst, i32 %0
79 ; FORCED-NEXT: %4 = call <2 x float> @llvm.pow.v2f32(<2 x float> %broadcast.splat, <2 x float> %broadcast.splat2)
80 ; FORCED-NEXT: %5 = getelementptr float, ptr %3, i32 0
81 ; FORCED-NEXT: store <2 x float> %4, ptr %5, align 4
82 ; FORCED-NEXT: %index.next = add nuw i32 %index, 2
83 ; FORCED-NEXT: %6 = icmp eq i32 %index.next, 1000
84 ; FORCED-NEXT: br i1 %6, label %middle.block, label %vector.body
86 define void @test_getVectorCallCost(ptr %dst, {float, float} %sv) {
91 %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop.body ]
92 %a = extractvalue { float, float } %sv, 0
93 %b = extractvalue { float, float } %sv, 1
94 %addr = getelementptr float, ptr %dst, i32 %iv
95 %p = call float @powf(float %a, float %b)
96 store float %p, ptr %addr
97 %iv.next = add nsw i32 %iv, 1
98 %cond = icmp ne i32 %iv.next, 1000
99 br i1 %cond, label %loop.body, label %exit