vdotprod.ll
1.45 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
; RUN: llc -march=hexagon < %s | FileCheck %s
; Test that we generate a single packet for the vectorized dot product loop.
; CHECK: loop0(.LBB0_[[LOOP:.]],
; CHECK: .LBB0_[[LOOP]]:
; CHECK: {
; CHECK: mpyi
; CHECK: mpyi
; CHECK: memd
; CHECK: memd
; CHECK-NOT: {
; CHECK: }{{[ \t]*}}:endloop
; Function Attrs: nounwind readonly
define i32 @f0(i32* nocapture readonly %a0, i32* nocapture readonly %a1) #0 {
b0:
%v0 = bitcast i32* %a0 to i64*
%v1 = bitcast i32* %a1 to i64*
br label %b1
b1: ; preds = %b1, %b0
%v2 = phi i64* [ %v0, %b0 ], [ %v21, %b1 ]
%v3 = phi i64* [ %v1, %b0 ], [ %v22, %b1 ]
%v4 = phi i32 [ 0, %b0 ], [ %v19, %b1 ]
%v5 = phi i32 [ 0, %b0 ], [ %v14, %b1 ]
%v6 = phi i32 [ 0, %b0 ], [ %v18, %b1 ]
%v7 = load i64, i64* %v2, align 8
%v8 = trunc i64 %v7 to i32
%v9 = lshr i64 %v7, 32
%v10 = load i64, i64* %v3, align 8
%v11 = trunc i64 %v10 to i32
%v12 = lshr i64 %v10, 32
%v13 = mul nsw i32 %v11, %v8
%v14 = add nsw i32 %v13, %v5
%v15 = trunc i64 %v9 to i32
%v16 = trunc i64 %v12 to i32
%v17 = mul nsw i32 %v16, %v15
%v18 = add nsw i32 %v17, %v6
%v19 = add nsw i32 %v4, 1
%v20 = icmp eq i32 %v19, 199
%v21 = getelementptr i64, i64* %v2, i32 1
%v22 = getelementptr i64, i64* %v3, i32 1
br i1 %v20, label %b2, label %b1
b2: ; preds = %b1
%v23 = add nsw i32 %v14, %v18
ret i32 %v23
}
attributes #0 = { nounwind readonly }