xref: /llvm-project/llvm/test/CodeGen/Hexagon/vdotprod.ll (revision 2208c97c1bec2512d4e47b6223db6d95a7037956)
1; RUN: llc -mtriple=hexagon < %s | FileCheck %s
2
3; Test that we generate a single packet for the vectorized dot product loop.
4
5; CHECK: loop0(.LBB0_[[LOOP:.]],
6; CHECK: .LBB0_[[LOOP]]:
7; CHECK: {
8; CHECK: mpyi
9; CHECK: mpyi
10; CHECK: memd
11; CHECK: memd
12; CHECK-NOT: {
13; CHECK: }{{[ \t]*}}:endloop
14
15; Function Attrs: nounwind readonly
16define i32 @f0(ptr nocapture readonly %a0, ptr nocapture readonly %a1) #0 {
17b0:
18  br label %b1
19
20b1:                                               ; preds = %b1, %b0
21  %v2 = phi ptr [ %a0, %b0 ], [ %v21, %b1 ]
22  %v3 = phi ptr [ %a1, %b0 ], [ %v22, %b1 ]
23  %v4 = phi i32 [ 0, %b0 ], [ %v19, %b1 ]
24  %v5 = phi i32 [ 0, %b0 ], [ %v14, %b1 ]
25  %v6 = phi i32 [ 0, %b0 ], [ %v18, %b1 ]
26  %v7 = load i64, ptr %v2, align 8
27  %v8 = trunc i64 %v7 to i32
28  %v9 = lshr i64 %v7, 32
29  %v10 = load i64, ptr %v3, align 8
30  %v11 = trunc i64 %v10 to i32
31  %v12 = lshr i64 %v10, 32
32  %v13 = mul nsw i32 %v11, %v8
33  %v14 = add nsw i32 %v13, %v5
34  %v15 = trunc i64 %v9 to i32
35  %v16 = trunc i64 %v12 to i32
36  %v17 = mul nsw i32 %v16, %v15
37  %v18 = add nsw i32 %v17, %v6
38  %v19 = add nsw i32 %v4, 1
39  %v20 = icmp eq i32 %v19, 199
40  %v21 = getelementptr i64, ptr %v2, i32 1
41  %v22 = getelementptr i64, ptr %v3, i32 1
42  br i1 %v20, label %b2, label %b1
43
44b2:                                               ; preds = %b1
45  %v23 = add nsw i32 %v14, %v18
46  ret i32 %v23
47}
48
49attributes #0 = { nounwind readonly }
50