xref: /aosp_15_r20/external/llvm/test/CodeGen/AMDGPU/ctlz.ll (revision 9880d6810fe72a1726cb53787c6711e909410d58)
1*9880d681SAndroid Build Coastguard Worker; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
2*9880d681SAndroid Build Coastguard Worker; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s
3*9880d681SAndroid Build Coastguard Worker; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s
4*9880d681SAndroid Build Coastguard Worker
5*9880d681SAndroid Build Coastguard Workerdeclare i7 @llvm.ctlz.i7(i7, i1) nounwind readnone
6*9880d681SAndroid Build Coastguard Workerdeclare i8 @llvm.ctlz.i8(i8, i1) nounwind readnone
7*9880d681SAndroid Build Coastguard Workerdeclare i16 @llvm.ctlz.i16(i16, i1) nounwind readnone
8*9880d681SAndroid Build Coastguard Worker
9*9880d681SAndroid Build Coastguard Workerdeclare i32 @llvm.ctlz.i32(i32, i1) nounwind readnone
10*9880d681SAndroid Build Coastguard Workerdeclare <2 x i32> @llvm.ctlz.v2i32(<2 x i32>, i1) nounwind readnone
11*9880d681SAndroid Build Coastguard Workerdeclare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1) nounwind readnone
12*9880d681SAndroid Build Coastguard Worker
13*9880d681SAndroid Build Coastguard Workerdeclare i64 @llvm.ctlz.i64(i64, i1) nounwind readnone
14*9880d681SAndroid Build Coastguard Workerdeclare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1) nounwind readnone
15*9880d681SAndroid Build Coastguard Workerdeclare <4 x i64> @llvm.ctlz.v4i64(<4 x i64>, i1) nounwind readnone
16*9880d681SAndroid Build Coastguard Worker
17*9880d681SAndroid Build Coastguard Workerdeclare i32 @llvm.r600.read.tidig.x() nounwind readnone
18*9880d681SAndroid Build Coastguard Worker
19*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}s_ctlz_i32:
20*9880d681SAndroid Build Coastguard Worker; SI: s_load_dword [[VAL:s[0-9]+]], s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
21*9880d681SAndroid Build Coastguard Worker; SI-DAG: s_flbit_i32_b32 [[CTLZ:s[0-9]+]], [[VAL]]
22*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e64 [[CMPZ:s\[[0-9]+:[0-9]+\]]], 0, [[VAL]]
23*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_mov_b32_e32 [[VCTLZ:v[0-9]+]], [[CTLZ]]
24*9880d681SAndroid Build Coastguard Worker; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[VCTLZ]], 32, [[CMPZ]]
25*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dword [[RESULT]]
26*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
27*9880d681SAndroid Build Coastguard Worker
28*9880d681SAndroid Build Coastguard Worker; EG: FFBH_UINT
29*9880d681SAndroid Build Coastguard Worker; EG: CNDE_INT
30*9880d681SAndroid Build Coastguard Workerdefine void @s_ctlz_i32(i32 addrspace(1)* noalias %out, i32 %val) nounwind {
31*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
32*9880d681SAndroid Build Coastguard Worker  store i32 %ctlz, i32 addrspace(1)* %out, align 4
33*9880d681SAndroid Build Coastguard Worker  ret void
34*9880d681SAndroid Build Coastguard Worker}
35*9880d681SAndroid Build Coastguard Worker
36*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i32:
37*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dword [[VAL:v[0-9]+]],
38*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_ffbh_u32_e32 [[CTLZ:v[0-9]+]], [[VAL]]
39*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
40*9880d681SAndroid Build Coastguard Worker; SI: v_cndmask_b32_e64 [[RESULT:v[0-9]+]], [[CTLZ]], 32, vcc
41*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dword [[RESULT]],
42*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
43*9880d681SAndroid Build Coastguard Worker
44*9880d681SAndroid Build Coastguard Worker; EG: FFBH_UINT
45*9880d681SAndroid Build Coastguard Worker; EG: CNDE_INT
46*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i32(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
47*9880d681SAndroid Build Coastguard Worker  %val = load i32, i32 addrspace(1)* %valptr, align 4
48*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
49*9880d681SAndroid Build Coastguard Worker  store i32 %ctlz, i32 addrspace(1)* %out, align 4
50*9880d681SAndroid Build Coastguard Worker  ret void
51*9880d681SAndroid Build Coastguard Worker}
52*9880d681SAndroid Build Coastguard Worker
53*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_v2i32:
54*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dwordx2
55*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
56*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
57*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dwordx2
58*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
59*9880d681SAndroid Build Coastguard Worker
60*9880d681SAndroid Build Coastguard Worker; EG: FFBH_UINT
61*9880d681SAndroid Build Coastguard Worker; EG: CNDE_INT
62*9880d681SAndroid Build Coastguard Worker; EG: FFBH_UINT
63*9880d681SAndroid Build Coastguard Worker; EG: CNDE_INT
64*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_v2i32(<2 x i32> addrspace(1)* noalias %out, <2 x i32> addrspace(1)* noalias %valptr) nounwind {
65*9880d681SAndroid Build Coastguard Worker  %val = load <2 x i32>, <2 x i32> addrspace(1)* %valptr, align 8
66*9880d681SAndroid Build Coastguard Worker  %ctlz = call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %val, i1 false) nounwind readnone
67*9880d681SAndroid Build Coastguard Worker  store <2 x i32> %ctlz, <2 x i32> addrspace(1)* %out, align 8
68*9880d681SAndroid Build Coastguard Worker  ret void
69*9880d681SAndroid Build Coastguard Worker}
70*9880d681SAndroid Build Coastguard Worker
71*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_v4i32:
72*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dwordx4
73*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
74*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
75*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
76*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
77*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dwordx4
78*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
79*9880d681SAndroid Build Coastguard Worker
80*9880d681SAndroid Build Coastguard Worker
81*9880d681SAndroid Build Coastguard Worker; EG-DAG: FFBH_UINT
82*9880d681SAndroid Build Coastguard Worker; EG-DAG: CNDE_INT
83*9880d681SAndroid Build Coastguard Worker
84*9880d681SAndroid Build Coastguard Worker; EG-DAG: FFBH_UINT
85*9880d681SAndroid Build Coastguard Worker; EG-DAG: CNDE_INT
86*9880d681SAndroid Build Coastguard Worker
87*9880d681SAndroid Build Coastguard Worker; EG-DAG: FFBH_UINT
88*9880d681SAndroid Build Coastguard Worker; EG-DAG: CNDE_INT
89*9880d681SAndroid Build Coastguard Worker
90*9880d681SAndroid Build Coastguard Worker; EG-DAG: FFBH_UINT
91*9880d681SAndroid Build Coastguard Worker; EG-DAG: CNDE_INT
92*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_v4i32(<4 x i32> addrspace(1)* noalias %out, <4 x i32> addrspace(1)* noalias %valptr) nounwind {
93*9880d681SAndroid Build Coastguard Worker  %val = load <4 x i32>, <4 x i32> addrspace(1)* %valptr, align 16
94*9880d681SAndroid Build Coastguard Worker  %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %val, i1 false) nounwind readnone
95*9880d681SAndroid Build Coastguard Worker  store <4 x i32> %ctlz, <4 x i32> addrspace(1)* %out, align 16
96*9880d681SAndroid Build Coastguard Worker  ret void
97*9880d681SAndroid Build Coastguard Worker}
98*9880d681SAndroid Build Coastguard Worker
99*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i8:
100*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
101*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
102*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[CTLZ]]
103*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cndmask_b32_e64 [[CORRECTED_FFBH:v[0-9]+]], [[FFBH]], 32, vcc
104*9880d681SAndroid Build Coastguard Worker; SI: v_add_i32_e32 [[RESULT:v[0-9]+]], vcc, 0xffffffe8, [[CORRECTED_FFBH]]
105*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_byte [[RESULT]],
106*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i8(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
107*9880d681SAndroid Build Coastguard Worker  %val = load i8, i8 addrspace(1)* %valptr
108*9880d681SAndroid Build Coastguard Worker  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 false) nounwind readnone
109*9880d681SAndroid Build Coastguard Worker  store i8 %ctlz, i8 addrspace(1)* %out
110*9880d681SAndroid Build Coastguard Worker  ret void
111*9880d681SAndroid Build Coastguard Worker}
112*9880d681SAndroid Build Coastguard Worker
113*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}s_ctlz_i64:
114*9880d681SAndroid Build Coastguard Worker; SI: s_load_dwordx2 s{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}, s{{\[[0-9]+:[0-9]+\]}}, {{0xb|0x2c}}
115*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e64 vcc, 0, s[[HI]]
116*9880d681SAndroid Build Coastguard Worker; SI-DAG: s_flbit_i32_b32 [[FFBH_LO:s[0-9]+]], s[[LO]]
117*9880d681SAndroid Build Coastguard Worker; SI-DAG: s_add_i32 [[ADD:s[0-9]+]], [[FFBH_LO]], 32
118*9880d681SAndroid Build Coastguard Worker; SI-DAG: s_flbit_i32_b32 [[FFBH_HI:s[0-9]+]], s[[HI]]
119*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_mov_b32_e32 [[VFFBH_LO:v[0-9]+]], [[ADD]]
120*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_mov_b32_e32 [[VFFBH_HI:v[0-9]+]], [[FFBH_HI]]
121*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cndmask_b32_e32 v[[CTLZ:[0-9]+]], [[VFFBH_HI]], [[VFFBH_LO]]
122*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
123*9880d681SAndroid Build Coastguard Worker; SI: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CTLZ]]:[[CTLZ_HI]]{{\]}}
124*9880d681SAndroid Build Coastguard Workerdefine void @s_ctlz_i64(i64 addrspace(1)* noalias %out, i64 %val) nounwind {
125*9880d681SAndroid Build Coastguard Worker  %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
126*9880d681SAndroid Build Coastguard Worker  store i64 %ctlz, i64 addrspace(1)* %out
127*9880d681SAndroid Build Coastguard Worker  ret void
128*9880d681SAndroid Build Coastguard Worker}
129*9880d681SAndroid Build Coastguard Worker
130*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}s_ctlz_i64_trunc:
131*9880d681SAndroid Build Coastguard Workerdefine void @s_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 %val) nounwind {
132*9880d681SAndroid Build Coastguard Worker  %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
133*9880d681SAndroid Build Coastguard Worker  %trunc = trunc i64 %ctlz to i32
134*9880d681SAndroid Build Coastguard Worker  store i32 %trunc, i32 addrspace(1)* %out
135*9880d681SAndroid Build Coastguard Worker  ret void
136*9880d681SAndroid Build Coastguard Worker}
137*9880d681SAndroid Build Coastguard Worker
138*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i64:
139*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_mov_b32_e32 v[[CTLZ_HI:[0-9]+]], 0{{$}}
140*9880d681SAndroid Build Coastguard Worker; SI-DAG: {{buffer|flat}}_load_dwordx2 v{{\[}}[[LO:[0-9]+]]:[[HI:[0-9]+]]{{\]}}
141*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e64 [[CMPHI:s\[[0-9]+:[0-9]+\]]], 0, v[[HI]]
142*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_ffbh_u32_e32 [[FFBH_LO:v[0-9]+]], v[[LO]]
143*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_add_i32_e32 [[ADD:v[0-9]+]], vcc, 32, [[FFBH_LO]]
144*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_ffbh_u32_e32 [[FFBH_HI:v[0-9]+]], v[[HI]]
145*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cndmask_b32_e64 v[[CTLZ:[0-9]+]], [[FFBH_HI]], [[ADD]], [[CMPHI]]
146*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_or_b32_e32 [[OR:v[0-9]+]], v[[LO]], v[[HI]]
147*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cmp_eq_i32_e32 vcc, 0, [[OR]]
148*9880d681SAndroid Build Coastguard Worker; SI-DAG: v_cndmask_b32_e64 v[[CLTZ_LO:[0-9]+]], v[[CTLZ:[0-9]+]], 64, vcc
149*9880d681SAndroid Build Coastguard Worker; SI: {{buffer|flat}}_store_dwordx2 {{.*}}v{{\[}}[[CLTZ_LO]]:[[CTLZ_HI]]{{\]}}
150*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
151*9880d681SAndroid Build Coastguard Worker  %tid = call i32 @llvm.r600.read.tidig.x()
152*9880d681SAndroid Build Coastguard Worker  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
153*9880d681SAndroid Build Coastguard Worker  %out.gep = getelementptr i64, i64 addrspace(1)* %out, i32 %tid
154*9880d681SAndroid Build Coastguard Worker  %val = load i64, i64 addrspace(1)* %in.gep
155*9880d681SAndroid Build Coastguard Worker  %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
156*9880d681SAndroid Build Coastguard Worker  store i64 %ctlz, i64 addrspace(1)* %out.gep
157*9880d681SAndroid Build Coastguard Worker  ret void
158*9880d681SAndroid Build Coastguard Worker}
159*9880d681SAndroid Build Coastguard Worker
160*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i64_trunc:
161*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i64_trunc(i32 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %in) nounwind {
162*9880d681SAndroid Build Coastguard Worker  %tid = call i32 @llvm.r600.read.tidig.x()
163*9880d681SAndroid Build Coastguard Worker  %in.gep = getelementptr i64, i64 addrspace(1)* %in, i32 %tid
164*9880d681SAndroid Build Coastguard Worker  %out.gep = getelementptr i32, i32 addrspace(1)* %out, i32 %tid
165*9880d681SAndroid Build Coastguard Worker  %val = load i64, i64 addrspace(1)* %in.gep
166*9880d681SAndroid Build Coastguard Worker  %ctlz = call i64 @llvm.ctlz.i64(i64 %val, i1 false)
167*9880d681SAndroid Build Coastguard Worker  %trunc = trunc i64 %ctlz to i32
168*9880d681SAndroid Build Coastguard Worker  store i32 %trunc, i32 addrspace(1)* %out.gep
169*9880d681SAndroid Build Coastguard Worker  ret void
170*9880d681SAndroid Build Coastguard Worker}
171*9880d681SAndroid Build Coastguard Worker
172*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i32_sel_eq_neg1:
173*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dword [[VAL:v[0-9]+]],
174*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
175*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dword [[RESULT]],
176*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
177*9880d681SAndroid Build Coastguard Worker define void @v_ctlz_i32_sel_eq_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
178*9880d681SAndroid Build Coastguard Worker  %val = load i32, i32 addrspace(1)* %valptr
179*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
180*9880d681SAndroid Build Coastguard Worker  %cmp = icmp eq i32 %val, 0
181*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i32 -1, i32 %ctlz
182*9880d681SAndroid Build Coastguard Worker  store i32 %sel, i32 addrspace(1)* %out
183*9880d681SAndroid Build Coastguard Worker  ret void
184*9880d681SAndroid Build Coastguard Worker}
185*9880d681SAndroid Build Coastguard Worker
186*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i32_sel_ne_neg1:
187*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dword [[VAL:v[0-9]+]],
188*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32 [[RESULT:v[0-9]+]], [[VAL]]
189*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_dword [[RESULT]],
190*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
191*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i32_sel_ne_neg1(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
192*9880d681SAndroid Build Coastguard Worker  %val = load i32, i32 addrspace(1)* %valptr
193*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
194*9880d681SAndroid Build Coastguard Worker  %cmp = icmp ne i32 %val, 0
195*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i32 %ctlz, i32 -1
196*9880d681SAndroid Build Coastguard Worker  store i32 %sel, i32 addrspace(1)* %out
197*9880d681SAndroid Build Coastguard Worker  ret void
198*9880d681SAndroid Build Coastguard Worker}
199*9880d681SAndroid Build Coastguard Worker
200*9880d681SAndroid Build Coastguard Worker; TODO: Should be able to eliminate select here as well.
201*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i32_sel_eq_bitwidth:
202*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dword
203*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
204*9880d681SAndroid Build Coastguard Worker; SI: v_cmp
205*9880d681SAndroid Build Coastguard Worker; SI: v_cndmask
206*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
207*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i32_sel_eq_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
208*9880d681SAndroid Build Coastguard Worker  %val = load i32, i32 addrspace(1)* %valptr
209*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
210*9880d681SAndroid Build Coastguard Worker  %cmp = icmp eq i32 %ctlz, 32
211*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i32 -1, i32 %ctlz
212*9880d681SAndroid Build Coastguard Worker  store i32 %sel, i32 addrspace(1)* %out
213*9880d681SAndroid Build Coastguard Worker  ret void
214*9880d681SAndroid Build Coastguard Worker}
215*9880d681SAndroid Build Coastguard Worker
216*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i32_sel_ne_bitwidth:
217*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_dword
218*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32
219*9880d681SAndroid Build Coastguard Worker; SI: v_cmp
220*9880d681SAndroid Build Coastguard Worker; SI: v_cndmask
221*9880d681SAndroid Build Coastguard Worker; SI: s_endpgm
222*9880d681SAndroid Build Coastguard Workerdefine void @v_ctlz_i32_sel_ne_bitwidth(i32 addrspace(1)* noalias %out, i32 addrspace(1)* noalias %valptr) nounwind {
223*9880d681SAndroid Build Coastguard Worker  %val = load i32, i32 addrspace(1)* %valptr
224*9880d681SAndroid Build Coastguard Worker  %ctlz = call i32 @llvm.ctlz.i32(i32 %val, i1 false) nounwind readnone
225*9880d681SAndroid Build Coastguard Worker  %cmp = icmp ne i32 %ctlz, 32
226*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i32 %ctlz, i32 -1
227*9880d681SAndroid Build Coastguard Worker  store i32 %sel, i32 addrspace(1)* %out
228*9880d681SAndroid Build Coastguard Worker  ret void
229*9880d681SAndroid Build Coastguard Worker}
230*9880d681SAndroid Build Coastguard Worker
231*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i8_sel_eq_neg1:
232*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
233*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
234*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_byte [[FFBH]],
235*9880d681SAndroid Build Coastguard Worker define void @v_ctlz_i8_sel_eq_neg1(i8 addrspace(1)* noalias %out, i8 addrspace(1)* noalias %valptr) nounwind {
236*9880d681SAndroid Build Coastguard Worker  %val = load i8, i8 addrspace(1)* %valptr
237*9880d681SAndroid Build Coastguard Worker  %ctlz = call i8 @llvm.ctlz.i8(i8 %val, i1 false) nounwind readnone
238*9880d681SAndroid Build Coastguard Worker  %cmp = icmp eq i8 %val, 0
239*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i8 -1, i8 %ctlz
240*9880d681SAndroid Build Coastguard Worker  store i8 %sel, i8 addrspace(1)* %out
241*9880d681SAndroid Build Coastguard Worker  ret void
242*9880d681SAndroid Build Coastguard Worker}
243*9880d681SAndroid Build Coastguard Worker
244*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i16_sel_eq_neg1:
245*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_ushort [[VAL:v[0-9]+]],
246*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
247*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_short [[FFBH]],
248*9880d681SAndroid Build Coastguard Worker define void @v_ctlz_i16_sel_eq_neg1(i16 addrspace(1)* noalias %out, i16 addrspace(1)* noalias %valptr) nounwind {
249*9880d681SAndroid Build Coastguard Worker  %val = load i16, i16 addrspace(1)* %valptr
250*9880d681SAndroid Build Coastguard Worker  %ctlz = call i16 @llvm.ctlz.i16(i16 %val, i1 false) nounwind readnone
251*9880d681SAndroid Build Coastguard Worker  %cmp = icmp eq i16 %val, 0
252*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i16 -1, i16 %ctlz
253*9880d681SAndroid Build Coastguard Worker  store i16 %sel, i16 addrspace(1)* %out
254*9880d681SAndroid Build Coastguard Worker  ret void
255*9880d681SAndroid Build Coastguard Worker}
256*9880d681SAndroid Build Coastguard Worker
257*9880d681SAndroid Build Coastguard Worker; FUNC-LABEL: {{^}}v_ctlz_i7_sel_eq_neg1:
258*9880d681SAndroid Build Coastguard Worker; SI: buffer_load_ubyte [[VAL:v[0-9]+]],
259*9880d681SAndroid Build Coastguard Worker; SI: v_ffbh_u32_e32 [[FFBH:v[0-9]+]], [[VAL]]
260*9880d681SAndroid Build Coastguard Worker; SI: v_and_b32_e32 [[TRUNC:v[0-9]+]], 0x7f, [[FFBH]]
261*9880d681SAndroid Build Coastguard Worker; SI: buffer_store_byte [[TRUNC]],
262*9880d681SAndroid Build Coastguard Worker define void @v_ctlz_i7_sel_eq_neg1(i7 addrspace(1)* noalias %out, i7 addrspace(1)* noalias %valptr) nounwind {
263*9880d681SAndroid Build Coastguard Worker  %val = load i7, i7 addrspace(1)* %valptr
264*9880d681SAndroid Build Coastguard Worker  %ctlz = call i7 @llvm.ctlz.i7(i7 %val, i1 false) nounwind readnone
265*9880d681SAndroid Build Coastguard Worker  %cmp = icmp eq i7 %val, 0
266*9880d681SAndroid Build Coastguard Worker  %sel = select i1 %cmp, i7 -1, i7 %ctlz
267*9880d681SAndroid Build Coastguard Worker  store i7 %sel, i7 addrspace(1)* %out
268*9880d681SAndroid Build Coastguard Worker  ret void
269*9880d681SAndroid Build Coastguard Worker}
270