xref: /llvm-project/llvm/test/CodeGen/AMDGPU/GlobalISel/minmaxabs.ll (revision deb4bb2b3abd216aceaa109a5304f025a234540f)
1*deb4bb2bSNikita Popov; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2*deb4bb2bSNikita Popov; RUN: llc -global-isel -mtriple=amdgcn-unknown-amdhsa < %s | FileCheck %s
3*deb4bb2bSNikita Popov
4*deb4bb2bSNikita Popovdeclare i32 @llvm.umin.i32(i32, i32)
5*deb4bb2bSNikita Popovdeclare i32 @llvm.umax.i32(i32, i32)
6*deb4bb2bSNikita Popovdeclare i32 @llvm.smin.i32(i32, i32)
7*deb4bb2bSNikita Popovdeclare i32 @llvm.smax.i32(i32, i32)
8*deb4bb2bSNikita Popov
9*deb4bb2bSNikita Popovdeclare <4 x i32> @llvm.umin.v4i32(<4 x i32>, <4 x i32>)
10*deb4bb2bSNikita Popovdeclare <4 x i32> @llvm.umax.v4i32(<4 x i32>, <4 x i32>)
11*deb4bb2bSNikita Popovdeclare <4 x i32> @llvm.smin.v4i32(<4 x i32>, <4 x i32>)
12*deb4bb2bSNikita Popovdeclare <4 x i32> @llvm.smax.v4i32(<4 x i32>, <4 x i32>)
13*deb4bb2bSNikita Popov
14*deb4bb2bSNikita Popovdefine i32 @test_umin_i32(i32 %a, i32 %b) {
15*deb4bb2bSNikita Popov; CHECK-LABEL: test_umin_i32:
16*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
17*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
18*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_u32_e32 v0, v0, v1
19*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
20*deb4bb2bSNikita Popov  %r = call i32 @llvm.umin.i32(i32 %a, i32 %b)
21*deb4bb2bSNikita Popov  ret i32 %r
22*deb4bb2bSNikita Popov}
23*deb4bb2bSNikita Popov
24*deb4bb2bSNikita Popovdefine i32 @test_umax_i32(i32 %a, i32 %b) {
25*deb4bb2bSNikita Popov; CHECK-LABEL: test_umax_i32:
26*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
27*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
28*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_u32_e32 v0, v0, v1
29*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
30*deb4bb2bSNikita Popov  %r = call i32 @llvm.umax.i32(i32 %a, i32 %b)
31*deb4bb2bSNikita Popov  ret i32 %r
32*deb4bb2bSNikita Popov}
33*deb4bb2bSNikita Popov
34*deb4bb2bSNikita Popovdefine i32 @test_smin_i32(i32 %a, i32 %b) {
35*deb4bb2bSNikita Popov; CHECK-LABEL: test_smin_i32:
36*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
37*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
38*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_i32_e32 v0, v0, v1
39*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
40*deb4bb2bSNikita Popov  %r = call i32 @llvm.smin.i32(i32 %a, i32 %b)
41*deb4bb2bSNikita Popov  ret i32 %r
42*deb4bb2bSNikita Popov}
43*deb4bb2bSNikita Popov
44*deb4bb2bSNikita Popovdefine i32 @test_smax_i32(i32 %a, i32 %b) {
45*deb4bb2bSNikita Popov; CHECK-LABEL: test_smax_i32:
46*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
47*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
48*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_i32_e32 v0, v0, v1
49*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
50*deb4bb2bSNikita Popov  %r = call i32 @llvm.smax.i32(i32 %a, i32 %b)
51*deb4bb2bSNikita Popov  ret i32 %r
52*deb4bb2bSNikita Popov}
53*deb4bb2bSNikita Popov
54*deb4bb2bSNikita Popovdefine <4 x i32> @test_umin_v4i32(<4 x i32> %a, <4 x i32> %b) {
55*deb4bb2bSNikita Popov; CHECK-LABEL: test_umin_v4i32:
56*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
57*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
58*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_u32_e32 v0, v0, v4
59*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_u32_e32 v1, v1, v5
60*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_u32_e32 v2, v2, v6
61*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_u32_e32 v3, v3, v7
62*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
63*deb4bb2bSNikita Popov  %r = call <4 x i32> @llvm.umin.v4i32(<4 x i32> %a, <4 x i32> %b)
64*deb4bb2bSNikita Popov  ret <4 x i32> %r
65*deb4bb2bSNikita Popov}
66*deb4bb2bSNikita Popov
67*deb4bb2bSNikita Popovdefine <4 x i32> @test_umax_v4i32(<4 x i32> %a, <4 x i32> %b) {
68*deb4bb2bSNikita Popov; CHECK-LABEL: test_umax_v4i32:
69*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
70*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
71*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_u32_e32 v0, v0, v4
72*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_u32_e32 v1, v1, v5
73*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_u32_e32 v2, v2, v6
74*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_u32_e32 v3, v3, v7
75*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
76*deb4bb2bSNikita Popov  %r = call <4 x i32> @llvm.umax.v4i32(<4 x i32> %a, <4 x i32> %b)
77*deb4bb2bSNikita Popov  ret <4 x i32> %r
78*deb4bb2bSNikita Popov}
79*deb4bb2bSNikita Popov
80*deb4bb2bSNikita Popovdefine <4 x i32> @test_smin_v4i32(<4 x i32> %a, <4 x i32> %b) {
81*deb4bb2bSNikita Popov; CHECK-LABEL: test_smin_v4i32:
82*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
83*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
84*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_i32_e32 v0, v0, v4
85*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_i32_e32 v1, v1, v5
86*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_i32_e32 v2, v2, v6
87*deb4bb2bSNikita Popov; CHECK-NEXT:    v_min_i32_e32 v3, v3, v7
88*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
89*deb4bb2bSNikita Popov  %r = call <4 x i32> @llvm.smin.v4i32(<4 x i32> %a, <4 x i32> %b)
90*deb4bb2bSNikita Popov  ret <4 x i32> %r
91*deb4bb2bSNikita Popov}
92*deb4bb2bSNikita Popov
93*deb4bb2bSNikita Popovdefine <4 x i32> @test_smax_v4i32(<4 x i32> %a, <4 x i32> %b) {
94*deb4bb2bSNikita Popov; CHECK-LABEL: test_smax_v4i32:
95*deb4bb2bSNikita Popov; CHECK:       ; %bb.0:
96*deb4bb2bSNikita Popov; CHECK-NEXT:    s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
97*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_i32_e32 v0, v0, v4
98*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_i32_e32 v1, v1, v5
99*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_i32_e32 v2, v2, v6
100*deb4bb2bSNikita Popov; CHECK-NEXT:    v_max_i32_e32 v3, v3, v7
101*deb4bb2bSNikita Popov; CHECK-NEXT:    s_setpc_b64 s[30:31]
102*deb4bb2bSNikita Popov  %r = call <4 x i32> @llvm.smax.v4i32(<4 x i32> %a, <4 x i32> %b)
103*deb4bb2bSNikita Popov  ret <4 x i32> %r
104*deb4bb2bSNikita Popov}
105