100
|
1 ; RUN: llc -march=r600 -mcpu=redwood < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
|
|
2 ; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
|
3 ; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
4
|
100
|
5 ; FUNC-LABEL: {{^}}udiv_i32:
|
|
6 ; EG-NOT: SETGE_INT
|
|
7 ; EG: CF_END
|
|
8 define void @udiv_i32(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
9 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
10 %a = load i32, i32 addrspace(1) * %in
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
11 %b = load i32, i32 addrspace(1) * %b_ptr
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
12 %result = udiv i32 %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
13 store i32 %result, i32 addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
14 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
15 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
16
|
100
|
17 ; FUNC-LABEL: {{^}}s_udiv_i32:
|
|
18
|
|
19 define void @s_udiv_i32(i32 addrspace(1)* %out, i32 %a, i32 %b) {
|
|
20 %result = udiv i32 %a, %b
|
|
21 store i32 %result, i32 addrspace(1)* %out
|
|
22 ret void
|
|
23 }
|
|
24
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
25
|
100
|
26 ; The code generated by udiv is long and complex and may frequently
|
|
27 ; change. The goal of this test is to make sure the ISel doesn't fail
|
|
28 ; when it gets a v4i32 udiv
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
29
|
100
|
30 ; FUNC-LABEL: {{^}}udiv_v2i32:
|
|
31 ; EG: CF_END
|
|
32
|
|
33 ; SI: s_endpgm
|
|
34 define void @udiv_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
35 %b_ptr = getelementptr <2 x i32>, <2 x i32> addrspace(1)* %in, i32 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
36 %a = load <2 x i32>, <2 x i32> addrspace(1) * %in
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
37 %b = load <2 x i32>, <2 x i32> addrspace(1) * %b_ptr
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
38 %result = udiv <2 x i32> %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
39 store <2 x i32> %result, <2 x i32> addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
40 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
41 }
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
42
|
100
|
43 ; FUNC-LABEL: {{^}}udiv_v4i32:
|
|
44 ; EG: CF_END
|
|
45 ; SI: s_endpgm
|
|
46 define void @udiv_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
|
95
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
47 %b_ptr = getelementptr <4 x i32>, <4 x i32> addrspace(1)* %in, i32 1
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
48 %a = load <4 x i32>, <4 x i32> addrspace(1) * %in
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
49 %b = load <4 x i32>, <4 x i32> addrspace(1) * %b_ptr
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
50 %result = udiv <4 x i32> %a, %b
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
51 store <4 x i32> %result, <4 x i32> addrspace(1)* %out
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
52 ret void
|
Kaito Tokumori <e105711@ie.u-ryukyu.ac.jp>
parents:
diff
changeset
|
53 }
|
100
|
54
|
|
55 ; FUNC-LABEL: {{^}}udiv_i32_div_pow2:
|
|
56 ; SI: buffer_load_dword [[VAL:v[0-9]+]]
|
|
57 ; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 4, [[VAL]]
|
|
58 ; SI: buffer_store_dword [[RESULT]]
|
|
59 define void @udiv_i32_div_pow2(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
|
|
60 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
|
|
61 %a = load i32, i32 addrspace(1)* %in
|
|
62 %result = udiv i32 %a, 16
|
|
63 store i32 %result, i32 addrspace(1)* %out
|
|
64 ret void
|
|
65 }
|
|
66
|
|
67 ; FUNC-LABEL: {{^}}udiv_i32_div_k_even:
|
|
68 ; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]]
|
|
69 ; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0xfabbd9c1
|
|
70 ; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]]
|
|
71 ; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 25, [[MULHI]]
|
|
72 ; SI: buffer_store_dword [[RESULT]]
|
|
73 define void @udiv_i32_div_k_even(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
|
|
74 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
|
|
75 %a = load i32, i32 addrspace(1)* %in
|
|
76 %result = udiv i32 %a, 34259182
|
|
77 store i32 %result, i32 addrspace(1)* %out
|
|
78 ret void
|
|
79 }
|
|
80
|
|
81 ; FUNC-LABEL: {{^}}udiv_i32_div_k_odd:
|
|
82 ; SI-DAG: buffer_load_dword [[VAL:v[0-9]+]]
|
|
83 ; SI-DAG: v_mov_b32_e32 [[K:v[0-9]+]], 0x7d5deca3
|
|
84 ; SI: v_mul_hi_u32 [[MULHI:v[0-9]+]], [[K]], [[VAL]]
|
|
85 ; SI: v_lshrrev_b32_e32 [[RESULT:v[0-9]+]], 24, [[MULHI]]
|
|
86 ; SI: buffer_store_dword [[RESULT]]
|
|
87 define void @udiv_i32_div_k_odd(i32 addrspace(1)* %out, i32 addrspace(1)* %in) {
|
|
88 %b_ptr = getelementptr i32, i32 addrspace(1)* %in, i32 1
|
|
89 %a = load i32, i32 addrspace(1)* %in
|
|
90 %result = udiv i32 %a, 34259183
|
|
91 store i32 %result, i32 addrspace(1)* %out
|
|
92 ret void
|
|
93 }
|