1 // REQUIRES: x86-registered-target
2 // REQUIRES: nvptx-registered-target
3 // REQUIRES: amdgpu-registered-target
4 
5 // RUN: clang-offload-packager -o %t.out --image=file=%S/Inputs/dummy-elf.o,kind=openmp,triple=nvptx64-nvidia-cuda,arch=sm_70
6 // RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o \
7 // RUN:   -fembed-offload-object=%t.out
8 // RUN: clang-linker-wrapper --print-wrapped-module --dry-run --host-triple=x86_64-unknown-linux-gnu \
9 // RUN:   --linker-path=/usr/bin/ld -- %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=OPENMP
10 
11 //      OPENMP: @__start_omp_offloading_entries = external hidden constant %__tgt_offload_entry
12 // OPENMP-NEXT: @__stop_omp_offloading_entries = external hidden constant %__tgt_offload_entry
13 // OPENMP-NEXT: @__dummy.omp_offloading.entry = hidden constant [0 x %__tgt_offload_entry] zeroinitializer, section "omp_offloading_entries"
14 // OPENMP-NEXT: @.omp_offloading.device_image = internal unnamed_addr constant [[[SIZE:[0-9]+]] x i8] c"\10\FF\10\AD{{.*}}"
15 // OPENMP-NEXT: @.omp_offloading.device_images = internal unnamed_addr constant [1 x %__tgt_device_image] [%__tgt_device_image { ptr @.omp_offloading.device_image, ptr getelementptr inbounds ([[[SIZE]] x i8], ptr @.omp_offloading.device_image, i64 1, i64 0), ptr @__start_omp_offloading_entries, ptr @__stop_omp_offloading_entries }]
16 // OPENMP-NEXT: @.omp_offloading.descriptor = internal constant %__tgt_bin_desc { i32 1, ptr @.omp_offloading.device_images, ptr @__start_omp_offloading_entries, ptr @__stop_omp_offloading_entries }
17 // OPENMP-NEXT: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @.omp_offloading.descriptor_reg, ptr null }]
18 // OPENMP-NEXT: @llvm.global_dtors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @.omp_offloading.descriptor_unreg, ptr null }]
19 
20 //      OPENMP: define internal void @.omp_offloading.descriptor_reg() section ".text.startup" {
21 // OPENMP-NEXT: entry:
22 // OPENMP-NEXT:   call void @__tgt_register_lib(ptr @.omp_offloading.descriptor)
23 // OPENMP-NEXT:   ret void
24 // OPENMP-NEXT: }
25 
26 //      OPENMP: define internal void @.omp_offloading.descriptor_unreg() section ".text.startup" {
27 // OPENMP-NEXT: entry:
28 // OPENMP-NEXT:   call void @__tgt_unregister_lib(ptr @.omp_offloading.descriptor)
29 // OPENMP-NEXT:   ret void
30 // OPENMP-NEXT: }
31 
32 // RUN: clang-offload-packager -o %t.out --image=file=%S/Inputs/dummy-elf.o,kind=cuda,triple=nvptx64-nvidia-cuda,arch=sm_70
33 // RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o \
34 // RUN:   -fembed-offload-object=%t.out
35 // RUN: clang-linker-wrapper --print-wrapped-module --dry-run --host-triple=x86_64-unknown-linux-gnu \
36 // RUN:   --linker-path=/usr/bin/ld -- %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=CUDA
37 
38 //      CUDA: @.fatbin_image = internal constant [0 x i8] zeroinitializer, section ".nv_fatbin"
39 // CUDA-NEXT: @.fatbin_wrapper = internal constant %fatbin_wrapper { i32 1180844977, i32 1, ptr @.fatbin_image, ptr null }, section ".nvFatBinSegment", align 8
40 // CUDA-NEXT: @__dummy.cuda_offloading.entry = hidden constant [0 x %__tgt_offload_entry] zeroinitializer, section "cuda_offloading_entries"
41 // CUDA-NEXT: @.cuda.binary_handle = internal global ptr null
42 // CUDA-NEXT: @__start_cuda_offloading_entries = external hidden constant [0 x %__tgt_offload_entry]
43 // CUDA-NEXT: @__stop_cuda_offloading_entries = external hidden constant [0 x %__tgt_offload_entry]
44 // CUDA-NEXT: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @.cuda.fatbin_reg, ptr null }]
45 
46 //      CUDA: define internal void @.cuda.fatbin_reg() section ".text.startup" {
47 // CUDA-NEXT: entry:
48 // CUDA-NEXT:   %0 = call ptr @__cudaRegisterFatBinary(ptr @.fatbin_wrapper)
49 // CUDA-NEXT:   store ptr %0, ptr @.cuda.binary_handle, align 8
50 // CUDA-NEXT:   call void @.cuda.globals_reg(ptr %0)
51 // CUDA-NEXT:   call void @__cudaRegisterFatBinaryEnd(ptr %0)
52 // CUDA-NEXT:   %1 = call i32 @atexit(ptr @.cuda.fatbin_unreg)
53 // CUDA-NEXT:   ret void
54 // CUDA-NEXT: }
55 
56 //      CUDA: define internal void @.cuda.fatbin_unreg() section ".text.startup" {
57 // CUDA-NEXT: entry:
58 // CUDA-NEXT:   %0 = load ptr, ptr @.cuda.binary_handle, align 8
59 // CUDA-NEXT:   call void @__cudaUnregisterFatBinary(ptr %0)
60 // CUDA-NEXT:   ret void
61 // CUDA-NEXT: }
62 
63 //      CUDA: define internal void @.cuda.globals_reg(ptr %0) section ".text.startup" {
64 // CUDA-NEXT: entry:
65 // CUDA-NEXT:   br i1 icmp ne (ptr @__start_cuda_offloading_entries, ptr @__stop_cuda_offloading_entries), label %while.entry, label %while.end
66 
67 //      CUDA: while.entry:
68 // CUDA-NEXT:  %entry1 = phi ptr [ @__start_cuda_offloading_entries, %entry ], [ %7, %if.end ]
69 // CUDA-NEXT:  %1 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 0
70 // CUDA-NEXT:  %addr = load ptr, ptr %1, align 8
71 // CUDA-NEXT:  %2 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 1
72 // CUDA-NEXT:  %name = load ptr, ptr %2, align 8
73 // CUDA-NEXT:  %3 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 2
74 // CUDA-NEXT:  %size = load i64, ptr %3, align 4
75 // CUDA-NEXT:  %4 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 3
76 // CUDA-NEXT:  %flag = load i32, ptr %4, align 4
77 // CUDA-NEXT:  %5 = icmp eq i64 %size, 0
78 // CUDA-NEXT:  br i1 %5, label %if.then, label %if.else
79 
80 //      CUDA: if.then:
81 // CUDA-NEXT:   %6 = call i32 @__cudaRegisterFunction(ptr %0, ptr %addr, ptr %name, ptr %name, i32 -1, ptr null, ptr null, ptr null, ptr null, ptr null)
82 // CUDA-NEXT:   br label %if.end
83 
84 //      CUDA: if.else:
85 // CUDA-NEXT:   switch i32 %flag, label %if.end [
86 // CUDA-NEXT:     i32 0, label %sw.global
87 // CUDA-NEXT:     i32 1, label %sw.managed
88 // CUDA-NEXT:     i32 2, label %sw.surface
89 // CUDA-NEXT:     i32 3, label %sw.texture
90 // CUDA-NEXT:   ]
91 
92 //      CUDA: sw.global:
93 // CUDA-NEXT:   call void @__cudaRegisterVar(ptr %0, ptr %addr, ptr %name, ptr %name, i32 0, i64 %size, i32 0, i32 0)
94 // CUDA-NEXT:   br label %if.end
95 
96 //      CUDA: sw.managed:
97 // CUDA-NEXT:   br label %if.end
98 
99 //      CUDA: sw.surface:
100 // CUDA-NEXT:   br label %if.end
101 
102 //      CUDA: sw.texture:
103 // CUDA-NEXT:   br label %if.end
104 
105 //      CUDA: if.end:
106 // CUDA-NEXT:   %7 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 1
107 // CUDA-NEXT:   %8 = icmp eq ptr %7, @__stop_cuda_offloading_entries
108 // CUDA-NEXT:   br i1 %8, label %while.end, label %while.entry
109 
110 //      CUDA: while.end:
111 // CUDA-NEXT:   ret void
112 // CUDA-NEXT: }
113 
114 // RUN: clang-offload-packager -o %t.out --image=file=%S/Inputs/dummy-elf.o,kind=hip,triple=amdgcn-amd-amdhsa,arch=gfx908
115 // RUN: %clang -cc1 %s -triple x86_64-unknown-linux-gnu -emit-obj -o %t.o \
116 // RUN:   -fembed-offload-object=%t.out
117 // RUN: clang-linker-wrapper --print-wrapped-module --dry-run --host-triple=x86_64-unknown-linux-gnu \
118 // RUN:   -linker-path /usr/bin/ld -- %t.o -o a.out 2>&1 | FileCheck %s --check-prefix=HIP
119 
120 //      HIP: @.fatbin_image = internal constant [0 x i8] zeroinitializer, section ".hip_fatbin"
121 // HIP-NEXT: @.fatbin_wrapper = internal constant %fatbin_wrapper { i32 1212764230, i32 1, ptr @.fatbin_image, ptr null }, section ".hipFatBinSegment", align 8
122 // HIP-NEXT: @__dummy.hip_offloading.entry = hidden constant [0 x %__tgt_offload_entry] zeroinitializer, section "hip_offloading_entries"
123 // HIP-NEXT: @.hip.binary_handle = internal global ptr null
124 // HIP-NEXT: @__start_hip_offloading_entries = external hidden constant [0 x %__tgt_offload_entry]
125 // HIP-NEXT: @__stop_hip_offloading_entries = external hidden constant [0 x %__tgt_offload_entry]
126 // HIP-NEXT: @llvm.global_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 1, ptr @.hip.fatbin_reg, ptr null }]
127 
128 //      HIP: define internal void @.hip.fatbin_reg() section ".text.startup" {
129 // HIP-NEXT: entry:
130 // HIP-NEXT:   %0 = call ptr @__hipRegisterFatBinary(ptr @.fatbin_wrapper)
131 // HIP-NEXT:   store ptr %0, ptr @.hip.binary_handle, align 8
132 // HIP-NEXT:   call void @.hip.globals_reg(ptr %0)
133 // HIP-NEXT:   %1 = call i32 @atexit(ptr @.hip.fatbin_unreg)
134 // HIP-NEXT:   ret void
135 // HIP-NEXT: }
136 
137 //      HIP: define internal void @.hip.fatbin_unreg() section ".text.startup" {
138 // HIP-NEXT: entry:
139 // HIP-NEXT:   %0 = load ptr, ptr @.hip.binary_handle, align 8
140 // HIP-NEXT:   call void @__hipUnregisterFatBinary(ptr %0)
141 // HIP-NEXT:   ret void
142 // HIP-NEXT: }
143 
144 //      HIP: define internal void @.hip.globals_reg(ptr %0) section ".text.startup" {
145 // HIP-NEXT: entry:
146 // HIP-NEXT:   br i1 icmp ne (ptr @__start_hip_offloading_entries, ptr @__stop_hip_offloading_entries), label %while.entry, label %while.end
147 
148 //      HIP: while.entry:
149 // HIP-NEXT:   %entry1 = phi ptr [ @__start_hip_offloading_entries, %entry ], [ %7, %if.end ]
150 // HIP-NEXT:   %1 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 0
151 // HIP-NEXT:   %addr = load ptr, ptr %1, align 8
152 // HIP-NEXT:   %2 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 1
153 // HIP-NEXT:   %name = load ptr, ptr %2, align 8
154 // HIP-NEXT:   %3 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 2
155 // HIP-NEXT:   %size = load i64, ptr %3, align 4
156 // HIP-NEXT:   %4 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 0, i32 3
157 // HIP-NEXT:   %flag = load i32, ptr %4, align 4
158 // HIP-NEXT:   %5 = icmp eq i64 %size, 0
159 // HIP-NEXT:   br i1 %5, label %if.then, label %if.else
160 
161 //      HIP: if.then:
162 // HIP-NEXT:   %6 = call i32 @__hipRegisterFunction(ptr %0, ptr %addr, ptr %name, ptr %name, i32 -1, ptr null, ptr null, ptr null, ptr null, ptr null)
163 // HIP-NEXT:   br label %if.end
164 
165 //      HIP: if.else:
166 // HIP-NEXT:   switch i32 %flag, label %if.end [
167 // HIP-NEXT:     i32 0, label %sw.global
168 // HIP-NEXT:     i32 1, label %sw.managed
169 // HIP-NEXT:     i32 2, label %sw.surface
170 // HIP-NEXT:     i32 3, label %sw.texture
171 // HIP-NEXT:   ]
172 
173 //      HIP: sw.global:
174 // HIP-NEXT:   call void @__hipRegisterVar(ptr %0, ptr %addr, ptr %name, ptr %name, i32 0, i64 %size, i32 0, i32 0)
175 // HIP-NEXT:   br label %if.end
176 
177 //      HIP: sw.managed:
178 // HIP-NEXT:   br label %if.end
179 
180 //      HIP: sw.surface:
181 // HIP-NEXT:   br label %if.end
182 
183 //      HIP: sw.texture:
184 // HIP-NEXT:   br label %if.end
185 
186 //      HIP: if.end:
187 // HIP-NEXT:   %7 = getelementptr inbounds %__tgt_offload_entry, ptr %entry1, i64 1
188 // HIP-NEXT:   %8 = icmp eq ptr %7, @__stop_hip_offloading_entries
189 // HIP-NEXT:   br i1 %8, label %while.end, label %while.entry
190 
191 //      HIP: while.end:
192 // HIP-NEXT:   ret void
193 // HIP-NEXT: }
194