1 ; RUN: llc < %s | FileCheck -check-prefix=ENABLED %s
2 ; RUN: llc -disable-nvptx-load-store-vectorizer < %s | FileCheck -check-prefix=DISABLED %s
3 target triple = "nvptx64-nvidia-cuda"
5 ; Check that the load-store vectorizer is enabled by default for nvptx, and
6 ; that it's disabled by the appropriate flag.
8 ; ENABLED: ld.v2.{{.}}32
10 ; DISABLED: ld.{{.}}32
11 define i32 @f(i32* %p) {
12 %p.1 = getelementptr i32, i32* %p, i32 1
13 %v0 = load i32, i32* %p, align 8
14 %v1 = load i32, i32* %p.1, align 4
15 %sum = add i32 %v0, %v1
19 define half @fh(half* %p) {
20 %p.1 = getelementptr half, half* %p, i32 1
21 %p.2 = getelementptr half, half* %p, i32 2
22 %p.3 = getelementptr half, half* %p, i32 3
23 %p.4 = getelementptr half, half* %p, i32 4
24 %v0 = load half, half* %p, align 64
25 %v1 = load half, half* %p.1, align 4
26 %v2 = load half, half* %p.2, align 4
27 %v3 = load half, half* %p.3, align 4
28 %v4 = load half, half* %p.4, align 4
29 %sum1 = fadd half %v0, %v1
30 %sum2 = fadd half %v2, %v3
31 %sum3 = fadd half %sum1, %sum2
32 %sum = fadd half %sum3, %v4
36 define float @ff(float* %p) {
37 %p.1 = getelementptr float, float* %p, i32 1
38 %p.2 = getelementptr float, float* %p, i32 2
39 %p.3 = getelementptr float, float* %p, i32 3
40 %p.4 = getelementptr float, float* %p, i32 4
41 %v0 = load float, float* %p, align 64
42 %v1 = load float, float* %p.1, align 4
43 %v2 = load float, float* %p.2, align 4
44 %v3 = load float, float* %p.3, align 4
45 %v4 = load float, float* %p.4, align 4
46 %sum1 = fadd float %v0, %v1
47 %sum2 = fadd float %v2, %v3
48 %sum3 = fadd float %sum1, %sum2
49 %sum = fadd float %sum3, %v4