1 ; RUN: opt -mtriple=amdgcn-amd-amdhsa -basicaa -load-store-vectorizer -S -o - %s | FileCheck %s
3 target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5"
5 declare i32 @llvm.amdgcn.workitem.id.x() #1
7 ; CHECK-LABEL: @merge_v2p1i8(
8 ; CHECK: load <2 x i64>
9 ; CHECK: inttoptr i64 %{{[^ ]+}} to i8 addrspace(1)*
10 ; CHECK: inttoptr i64 %{{[^ ]+}} to i8 addrspace(1)*
11 ; CHECK: store <2 x i64> zeroinitializer
12 define amdgpu_kernel void @merge_v2p1i8(i8 addrspace(1)* addrspace(1)* nocapture %a, i8 addrspace(1)* addrspace(1)* nocapture readonly %b) #0 {
14 %a.1 = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a, i64 1
15 %b.1 = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %b, i64 1
17 %ld.c = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %b, align 4
18 %ld.c.idx.1 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %b.1, align 4
20 store i8 addrspace(1)* null, i8 addrspace(1)* addrspace(1)* %a, align 4
21 store i8 addrspace(1)* null, i8 addrspace(1)* addrspace(1)* %a.1, align 4
26 ; CHECK-LABEL: @merge_v2p3i8(
27 ; CHECK: load <2 x i32>
28 ; CHECK: inttoptr i32 %{{[^ ]+}} to i8 addrspace(3)*
29 ; CHECK: inttoptr i32 %{{[^ ]+}} to i8 addrspace(3)*
30 ; CHECK: store <2 x i32> zeroinitializer
31 define amdgpu_kernel void @merge_v2p3i8(i8 addrspace(3)* addrspace(3)* nocapture %a, i8 addrspace(3)* addrspace(3)* nocapture readonly %b) #0 {
33 %a.1 = getelementptr inbounds i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %a, i64 1
34 %b.1 = getelementptr inbounds i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %b, i64 1
36 %ld.c = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %b, align 4
37 %ld.c.idx.1 = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %b.1, align 4
39 store i8 addrspace(3)* null, i8 addrspace(3)* addrspace(3)* %a, align 4
40 store i8 addrspace(3)* null, i8 addrspace(3)* addrspace(3)* %a.1, align 4
45 ; CHECK-LABEL: @merge_load_i64_ptr64(
46 ; CHECK: load <2 x i64>
47 ; CHECK: [[ELT1:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 1
48 ; CHECK: inttoptr i64 [[ELT1]] to i8 addrspace(1)*
49 define amdgpu_kernel void @merge_load_i64_ptr64(i64 addrspace(1)* nocapture %a) #0 {
51 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
52 %a.1.cast = bitcast i64 addrspace(1)* %a.1 to i8 addrspace(1)* addrspace(1)*
54 %ld.0 = load i64, i64 addrspace(1)* %a
55 %ld.1 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a.1.cast
60 ; CHECK-LABEL: @merge_load_ptr64_i64(
61 ; CHECK: load <2 x i64>
62 ; CHECK: [[ELT0:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 0
63 ; CHECK: inttoptr i64 [[ELT0]] to i8 addrspace(1)*
64 define amdgpu_kernel void @merge_load_ptr64_i64(i64 addrspace(1)* nocapture %a) #0 {
66 %a.cast = bitcast i64 addrspace(1)* %a to i8 addrspace(1)* addrspace(1)*
67 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
69 %ld.0 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a.cast
70 %ld.1 = load i64, i64 addrspace(1)* %a.1
75 ; CHECK-LABEL: @merge_store_ptr64_i64(
76 ; CHECK: [[ELT0:%[^ ]+]] = ptrtoint i8 addrspace(1)* %ptr0 to i64
77 ; CHECK: insertelement <2 x i64> undef, i64 [[ELT0]], i32 0
78 ; CHECK: store <2 x i64>
79 define amdgpu_kernel void @merge_store_ptr64_i64(i64 addrspace(1)* nocapture %a, i8 addrspace(1)* %ptr0, i64 %val1) #0 {
81 %a.cast = bitcast i64 addrspace(1)* %a to i8 addrspace(1)* addrspace(1)*
82 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
85 store i8 addrspace(1)* %ptr0, i8 addrspace(1)* addrspace(1)* %a.cast
86 store i64 %val1, i64 addrspace(1)* %a.1
91 ; CHECK-LABEL: @merge_store_i64_ptr64(
92 ; CHECK: [[ELT1:%[^ ]+]] = ptrtoint i8 addrspace(1)* %ptr1 to i64
93 ; CHECK: insertelement <2 x i64> %{{[^ ]+}}, i64 [[ELT1]], i32 1
94 ; CHECK: store <2 x i64>
95 define amdgpu_kernel void @merge_store_i64_ptr64(i8 addrspace(1)* addrspace(1)* nocapture %a, i64 %val0, i8 addrspace(1)* %ptr1) #0 {
97 %a.1 = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a, i64 1
98 %a.cast = bitcast i8 addrspace(1)* addrspace(1)* %a to i64 addrspace(1)*
100 store i64 %val0, i64 addrspace(1)* %a.cast
101 store i8 addrspace(1)* %ptr1, i8 addrspace(1)* addrspace(1)* %a.1
106 ; CHECK-LABEL: @merge_load_i32_ptr32(
107 ; CHECK: load <2 x i32>
108 ; CHECK: [[ELT1:%[^ ]+]] = extractelement <2 x i32> %{{[^ ]+}}, i32 1
109 ; CHECK: inttoptr i32 [[ELT1]] to i8 addrspace(3)*
110 define amdgpu_kernel void @merge_load_i32_ptr32(i32 addrspace(3)* nocapture %a) #0 {
112 %a.1 = getelementptr inbounds i32, i32 addrspace(3)* %a, i32 1
113 %a.1.cast = bitcast i32 addrspace(3)* %a.1 to i8 addrspace(3)* addrspace(3)*
115 %ld.0 = load i32, i32 addrspace(3)* %a
116 %ld.1 = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %a.1.cast
121 ; CHECK-LABEL: @merge_load_ptr32_i32(
122 ; CHECK: load <2 x i32>
123 ; CHECK: [[ELT0:%[^ ]+]] = extractelement <2 x i32> %{{[^ ]+}}, i32 0
124 ; CHECK: inttoptr i32 [[ELT0]] to i8 addrspace(3)*
125 define amdgpu_kernel void @merge_load_ptr32_i32(i32 addrspace(3)* nocapture %a) #0 {
127 %a.cast = bitcast i32 addrspace(3)* %a to i8 addrspace(3)* addrspace(3)*
128 %a.1 = getelementptr inbounds i32, i32 addrspace(3)* %a, i32 1
130 %ld.0 = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %a.cast
131 %ld.1 = load i32, i32 addrspace(3)* %a.1
136 ; CHECK-LABEL: @merge_store_ptr32_i32(
137 ; CHECK: [[ELT0:%[^ ]+]] = ptrtoint i8 addrspace(3)* %ptr0 to i32
138 ; CHECK: insertelement <2 x i32> undef, i32 [[ELT0]], i32 0
139 ; CHECK: store <2 x i32>
140 define amdgpu_kernel void @merge_store_ptr32_i32(i32 addrspace(3)* nocapture %a, i8 addrspace(3)* %ptr0, i32 %val1) #0 {
142 %a.cast = bitcast i32 addrspace(3)* %a to i8 addrspace(3)* addrspace(3)*
143 %a.1 = getelementptr inbounds i32, i32 addrspace(3)* %a, i32 1
145 store i8 addrspace(3)* %ptr0, i8 addrspace(3)* addrspace(3)* %a.cast
146 store i32 %val1, i32 addrspace(3)* %a.1
151 ; CHECK-LABEL: @merge_store_i32_ptr32(
152 ; CHECK: [[ELT1:%[^ ]+]] = ptrtoint i8 addrspace(3)* %ptr1 to i32
153 ; CHECK: insertelement <2 x i32> %{{[^ ]+}}, i32 [[ELT1]], i32 1
154 ; CHECK: store <2 x i32>
155 define amdgpu_kernel void @merge_store_i32_ptr32(i8 addrspace(3)* addrspace(3)* nocapture %a, i32 %val0, i8 addrspace(3)* %ptr1) #0 {
157 %a.1 = getelementptr inbounds i8 addrspace(3)*, i8 addrspace(3)* addrspace(3)* %a, i32 1
158 %a.cast = bitcast i8 addrspace(3)* addrspace(3)* %a to i32 addrspace(3)*
160 store i32 %val0, i32 addrspace(3)* %a.cast
161 store i8 addrspace(3)* %ptr1, i8 addrspace(3)* addrspace(3)* %a.1
166 ; CHECK-LABEL: @no_merge_store_ptr32_i64(
167 ; CHECK: store i8 addrspace(3)*
169 define amdgpu_kernel void @no_merge_store_ptr32_i64(i64 addrspace(1)* nocapture %a, i8 addrspace(3)* %ptr0, i64 %val1) #0 {
171 %a.cast = bitcast i64 addrspace(1)* %a to i8 addrspace(3)* addrspace(1)*
172 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
175 store i8 addrspace(3)* %ptr0, i8 addrspace(3)* addrspace(1)* %a.cast
176 store i64 %val1, i64 addrspace(1)* %a.1
181 ; CHECK-LABEL: @no_merge_store_i64_ptr32(
183 ; CHECK: store i8 addrspace(3)*
184 define amdgpu_kernel void @no_merge_store_i64_ptr32(i8 addrspace(3)* addrspace(1)* nocapture %a, i64 %val0, i8 addrspace(3)* %ptr1) #0 {
186 %a.1 = getelementptr inbounds i8 addrspace(3)*, i8 addrspace(3)* addrspace(1)* %a, i64 1
187 %a.cast = bitcast i8 addrspace(3)* addrspace(1)* %a to i64 addrspace(1)*
189 store i64 %val0, i64 addrspace(1)* %a.cast
190 store i8 addrspace(3)* %ptr1, i8 addrspace(3)* addrspace(1)* %a.1
195 ; CHECK-LABEL: @no_merge_load_i64_ptr32(
197 ; CHECK: load i8 addrspace(3)*,
198 define amdgpu_kernel void @no_merge_load_i64_ptr32(i64 addrspace(1)* nocapture %a) #0 {
200 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
201 %a.1.cast = bitcast i64 addrspace(1)* %a.1 to i8 addrspace(3)* addrspace(1)*
203 %ld.0 = load i64, i64 addrspace(1)* %a
204 %ld.1 = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(1)* %a.1.cast
209 ; CHECK-LABEL: @no_merge_load_ptr32_i64(
210 ; CHECK: load i8 addrspace(3)*,
212 define amdgpu_kernel void @no_merge_load_ptr32_i64(i64 addrspace(1)* nocapture %a) #0 {
214 %a.cast = bitcast i64 addrspace(1)* %a to i8 addrspace(3)* addrspace(1)*
215 %a.1 = getelementptr inbounds i64, i64 addrspace(1)* %a, i64 1
217 %ld.0 = load i8 addrspace(3)*, i8 addrspace(3)* addrspace(1)* %a.cast
218 %ld.1 = load i64, i64 addrspace(1)* %a.1
223 ; XXX - This isn't merged for some reason
224 ; CHECK-LABEL: @merge_v2p1i8_v2p1i8(
225 ; CHECK: load <2 x i8 addrspace(1)*>
226 ; CHECK: load <2 x i8 addrspace(1)*>
227 ; CHECK: store <2 x i8 addrspace(1)*>
228 ; CHECK: store <2 x i8 addrspace(1)*>
229 define amdgpu_kernel void @merge_v2p1i8_v2p1i8(<2 x i8 addrspace(1)*> addrspace(1)* nocapture noalias %a, <2 x i8 addrspace(1)*> addrspace(1)* nocapture readonly noalias %b) #0 {
231 %a.1 = getelementptr inbounds <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %a, i64 1
232 %b.1 = getelementptr inbounds <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %b, i64 1
234 %ld.c = load <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %b, align 4
235 %ld.c.idx.1 = load <2 x i8 addrspace(1)*>, <2 x i8 addrspace(1)*> addrspace(1)* %b.1, align 4
237 store <2 x i8 addrspace(1)*> zeroinitializer, <2 x i8 addrspace(1)*> addrspace(1)* %a, align 4
238 store <2 x i8 addrspace(1)*> zeroinitializer, <2 x i8 addrspace(1)*> addrspace(1)* %a.1, align 4
242 ; CHECK-LABEL: @merge_load_ptr64_f64(
243 ; CHECK: load <2 x i64>
244 ; CHECK: [[ELT0:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 0
245 ; CHECK: [[ELT0_INT:%[^ ]+]] = inttoptr i64 [[ELT0]] to i8 addrspace(1)*
246 ; CHECK: [[ELT1_INT:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 1
247 ; CHECK: bitcast i64 [[ELT1_INT]] to double
248 define amdgpu_kernel void @merge_load_ptr64_f64(double addrspace(1)* nocapture %a) #0 {
250 %a.cast = bitcast double addrspace(1)* %a to i8 addrspace(1)* addrspace(1)*
251 %a.1 = getelementptr inbounds double, double addrspace(1)* %a, i64 1
253 %ld.0 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a.cast
254 %ld.1 = load double, double addrspace(1)* %a.1
259 ; CHECK-LABEL: @merge_load_f64_ptr64(
260 ; CHECK: load <2 x i64>
261 ; CHECK: [[ELT0:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 0
262 ; CHECK: bitcast i64 [[ELT0]] to double
263 ; CHECK: [[ELT1:%[^ ]+]] = extractelement <2 x i64> %{{[^ ]+}}, i32 1
264 ; CHECK: inttoptr i64 [[ELT1]] to i8 addrspace(1)*
265 define amdgpu_kernel void @merge_load_f64_ptr64(double addrspace(1)* nocapture %a) #0 {
267 %a.1 = getelementptr inbounds double, double addrspace(1)* %a, i64 1
268 %a.1.cast = bitcast double addrspace(1)* %a.1 to i8 addrspace(1)* addrspace(1)*
270 %ld.0 = load double, double addrspace(1)* %a
271 %ld.1 = load i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a.1.cast
276 ; CHECK-LABEL: @merge_store_ptr64_f64(
277 ; CHECK: [[ELT0_INT:%[^ ]+]] = ptrtoint i8 addrspace(1)* %ptr0 to i64
278 ; CHECK: insertelement <2 x i64> undef, i64 [[ELT0_INT]], i32 0
279 ; CHECK: [[ELT1_INT:%[^ ]+]] = bitcast double %val1 to i64
280 ; CHECK: insertelement <2 x i64> %{{[^ ]+}}, i64 [[ELT1_INT]], i32 1
281 ; CHECK: store <2 x i64>
282 define amdgpu_kernel void @merge_store_ptr64_f64(double addrspace(1)* nocapture %a, i8 addrspace(1)* %ptr0, double %val1) #0 {
284 %a.cast = bitcast double addrspace(1)* %a to i8 addrspace(1)* addrspace(1)*
285 %a.1 = getelementptr inbounds double, double addrspace(1)* %a, i64 1
287 store i8 addrspace(1)* %ptr0, i8 addrspace(1)* addrspace(1)* %a.cast
288 store double %val1, double addrspace(1)* %a.1
293 ; CHECK-LABEL: @merge_store_f64_ptr64(
294 ; CHECK: [[ELT0_INT:%[^ ]+]] = bitcast double %val0 to i64
295 ; CHECK: insertelement <2 x i64> undef, i64 [[ELT0_INT]], i32 0
296 ; CHECK: [[ELT1_INT:%[^ ]+]] = ptrtoint i8 addrspace(1)* %ptr1 to i64
297 ; CHECK: insertelement <2 x i64> %{{[^ ]+}}, i64 [[ELT1_INT]], i32 1
298 ; CHECK: store <2 x i64>
299 define amdgpu_kernel void @merge_store_f64_ptr64(i8 addrspace(1)* addrspace(1)* nocapture %a, double %val0, i8 addrspace(1)* %ptr1) #0 {
301 %a.1 = getelementptr inbounds i8 addrspace(1)*, i8 addrspace(1)* addrspace(1)* %a, i64 1
302 %a.cast = bitcast i8 addrspace(1)* addrspace(1)* %a to double addrspace(1)*
304 store double %val0, double addrspace(1)* %a.cast
305 store i8 addrspace(1)* %ptr1, i8 addrspace(1)* addrspace(1)* %a.1
310 attributes #0 = { nounwind }
311 attributes #1 = { nounwind readnone }