Skip to content

Commit d86311f

Browse files
[Flang-new][OpenMP] Add bitcode files for AMD GPU OpenMP (#96742)
Flang-new needs to add `mlink-builtin-bitcode` objects to properly support offload code generation for AMD GPUs (for example, math functions). Both Flang-new and Clang rely on `mlink-builtin-bitcode` flags. These flags are added by the `AMDGPUOpenMPToolchain::addClangTargetOptions` function. Now, both compilers reuse the same function. Flang-new tests for AMDGPU were updated by adding the `-nogpulib` flag. This flag allows running AMDGPU tests on machines without the ROCm stack.
1 parent e3a3397 commit d86311f

File tree

4 files changed

+29
-19
lines changed

4 files changed

+29
-19
lines changed

clang/lib/Driver/ToolChains/Flang.cpp

+3
Original file line numberDiff line numberDiff line change
@@ -342,6 +342,9 @@ void Flang::AddAMDGPUTargetArgs(const ArgList &Args,
342342
StringRef Val = A->getValue();
343343
CmdArgs.push_back(Args.MakeArgString("-mcode-object-version=" + Val));
344344
}
345+
346+
const ToolChain &TC = getToolChain();
347+
TC.addClangTargetOptions(Args, CmdArgs, Action::OffloadKind::OFK_OpenMP);
345348
}
346349

347350
void Flang::addTargetOptions(const ArgList &Args,

flang/test/Driver/omp-driver-offload.f90

+23-16
Original file line numberDiff line numberDiff line change
@@ -14,12 +14,12 @@
1414
! Test regular -fopenmp with offload, and invocation filtering options
1515
! RUN: %flang -S -### %s -o %t 2>&1 \
1616
! RUN: -fopenmp --offload-arch=gfx90a --offload-arch=sm_70 \
17-
! RUN: --target=aarch64-unknown-linux-gnu \
17+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
1818
! RUN: | FileCheck %s --check-prefix=OFFLOAD-HOST-AND-DEVICE
1919

2020
! RUN: %flang -S -### %s -o %t 2>&1 \
2121
! RUN: -fopenmp --offload-arch=gfx90a --offload-arch=sm_70 --offload-host-device \
22-
! RUN: --target=aarch64-unknown-linux-gnu \
22+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
2323
! RUN: | FileCheck %s --check-prefix=OFFLOAD-HOST-AND-DEVICE
2424

2525
! OFFLOAD-HOST-AND-DEVICE: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
@@ -29,7 +29,7 @@
2929

3030
! RUN: %flang -S -### %s -o %t 2>&1 \
3131
! RUN: -fopenmp --offload-arch=gfx90a --offload-arch=sm_70 --offload-host-only \
32-
! RUN: --target=aarch64-unknown-linux-gnu \
32+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
3333
! RUN: | FileCheck %s --check-prefix=OFFLOAD-HOST
3434

3535
! OFFLOAD-HOST: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
@@ -39,7 +39,7 @@
3939

4040
! RUN: %flang -S -### %s 2>&1 \
4141
! RUN: -fopenmp --offload-arch=gfx90a --offload-arch=sm_70 --offload-device-only \
42-
! RUN: --target=aarch64-unknown-linux-gnu \
42+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
4343
! RUN: | FileCheck %s --check-prefix=OFFLOAD-DEVICE
4444

4545
! OFFLOAD-DEVICE: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
@@ -48,13 +48,13 @@
4848
! OFFLOAD-DEVICE-NOT: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
4949

5050
! Test regular -fopenmp with offload for basic fopenmp-is-target-device flag addition and correct fopenmp
51-
! RUN: %flang -### -fopenmp --offload-arch=gfx90a -fopenmp-targets=amdgcn-amd-amdhsa %s 2>&1 | FileCheck --check-prefixes=CHECK-OPENMP-IS-TARGET-DEVICE %s
51+
! RUN: %flang -### -fopenmp --offload-arch=gfx90a -fopenmp-targets=amdgcn-amd-amdhsa -nogpulib %s 2>&1 | FileCheck --check-prefixes=CHECK-OPENMP-IS-TARGET-DEVICE %s
5252
! CHECK-OPENMP-IS-TARGET-DEVICE: "{{[^"]*}}flang-new" "-fc1" {{.*}} "-fopenmp" {{.*}} "-fopenmp-is-target-device" {{.*}}.f90"
5353

5454
! Testing appropriate flags are gnerated and appropriately assigned by the driver when offloading
5555
! RUN: %flang -S -### %s -o %t 2>&1 \
5656
! RUN: -fopenmp --offload-arch=gfx90a \
57-
! RUN: --target=aarch64-unknown-linux-gnu \
57+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
5858
! RUN: | FileCheck %s --check-prefix=OPENMP-OFFLOAD-ARGS
5959
! OPENMP-OFFLOAD-ARGS: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu" {{.*}} "-fopenmp" {{.*}}.f90"
6060
! OPENMP-OFFLOAD-ARGS-NEXT: "{{[^"]*}}flang-new" "-fc1" "-triple" "amdgcn-amd-amdhsa"
@@ -70,19 +70,19 @@
7070
! RUN: %flang -### %s -o %t 2>&1 \
7171
! RUN: -fopenmp --offload-arch=gfx90a \
7272
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
73-
! RUN: -fopenmp-assume-threads-oversubscription \
73+
! RUN: -fopenmp-assume-threads-oversubscription -nogpulib \
7474
! RUN: | FileCheck %s --check-prefixes=CHECK-THREADS-OVS
7575
! RUN: %flang -### %s -o %t 2>&1 \
7676
! RUN: -fopenmp --offload-arch=sm_70 \
7777
! RUN: -fopenmp-targets=nvptx64-nvidia-cuda \
78-
! RUN: -fopenmp-assume-threads-oversubscription \
78+
! RUN: -fopenmp-assume-threads-oversubscription \
7979
! RUN: | FileCheck %s --check-prefixes=CHECK-THREADS-OVS
8080
! CHECK-THREADS-OVS: "{{[^"]*}}flang-new" "-fc1" {{.*}} "-fopenmp" {{.*}} "-fopenmp-is-target-device" "-fopenmp-assume-threads-oversubscription" {{.*}}.f90"
8181

8282
! RUN: %flang -### %s -o %t 2>&1 \
8383
! RUN: -fopenmp --offload-arch=gfx90a \
8484
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
85-
! RUN: -fopenmp-assume-teams-oversubscription \
85+
! RUN: -fopenmp-assume-teams-oversubscription -nogpulib\
8686
! RUN: | FileCheck %s --check-prefixes=CHECK-TEAMS-OVS
8787
! RUN: %flang -### %s -o %t 2>&1 \
8888
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -94,7 +94,7 @@
9494
! RUN: %flang -### %s -o %t 2>&1 \
9595
! RUN: -fopenmp --offload-arch=gfx90a \
9696
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
97-
! RUN: -fopenmp-assume-no-nested-parallelism \
97+
! RUN: -fopenmp-assume-no-nested-parallelism -nogpulib\
9898
! RUN: | FileCheck %s --check-prefixes=CHECK-NEST-PAR
9999
! RUN: %flang -### %s -o %t 2>&1 \
100100
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -106,7 +106,7 @@
106106
! RUN: %flang -### %s -o %t 2>&1 \
107107
! RUN: -fopenmp --offload-arch=gfx90a \
108108
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
109-
! RUN: -fopenmp-assume-no-thread-state \
109+
! RUN: -fopenmp-assume-no-thread-state -nogpulib\
110110
! RUN: | FileCheck %s --check-prefixes=CHECK-THREAD-STATE
111111
! RUN: %flang -### %s -o %t 2>&1 \
112112
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -118,7 +118,7 @@
118118
! RUN: %flang -### %s -o %t 2>&1 \
119119
! RUN: -fopenmp --offload-arch=gfx90a \
120120
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
121-
! RUN: -fopenmp-target-debug \
121+
! RUN: -fopenmp-target-debug -nogpulib\
122122
! RUN: | FileCheck %s --check-prefixes=CHECK-TARGET-DEBUG
123123
! RUN: %flang -### %s -o %t 2>&1 \
124124
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -130,7 +130,7 @@
130130
! RUN: %flang -### %s -o %t 2>&1 \
131131
! RUN: -fopenmp --offload-arch=gfx90a \
132132
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
133-
! RUN: -fopenmp-target-debug \
133+
! RUN: -fopenmp-target-debug -nogpulib\
134134
! RUN: | FileCheck %s --check-prefixes=CHECK-TARGET-DEBUG
135135
! RUN: %flang -### %s -o %t 2>&1 \
136136
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -144,7 +144,7 @@
144144
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
145145
! RUN: -fopenmp-target-debug -fopenmp-assume-threads-oversubscription \
146146
! RUN: -fopenmp-assume-teams-oversubscription -fopenmp-assume-no-nested-parallelism \
147-
! RUN: -fopenmp-assume-no-thread-state \
147+
! RUN: -fopenmp-assume-no-thread-state -nogpulib\
148148
! RUN: | FileCheck %s --check-prefixes=CHECK-RTL-ALL
149149
! RUN: %flang -S -### %s -o %t 2>&1 \
150150
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -160,7 +160,7 @@
160160
! RUN: %flang -### %s -o %t 2>&1 \
161161
! RUN: -fopenmp --offload-arch=gfx90a \
162162
! RUN: -fopenmp-targets=amdgcn-amd-amdhsa \
163-
! RUN: -fopenmp-version=45 \
163+
! RUN: -fopenmp-version=45 -nogpulib\
164164
! RUN: | FileCheck %s --check-prefixes=CHECK-OPENMP-VERSION
165165
! RUN: %flang -### %s -o %t 2>&1 \
166166
! RUN: -fopenmp --offload-arch=sm_70 \
@@ -193,10 +193,17 @@
193193
! Test -fopenmp-force-usm option with offload
194194
! RUN: %flang -S -### %s -o %t 2>&1 \
195195
! RUN: -fopenmp -fopenmp-force-usm --offload-arch=gfx90a \
196-
! RUN: --target=aarch64-unknown-linux-gnu \
196+
! RUN: --target=aarch64-unknown-linux-gnu -nogpulib\
197197
! RUN: | FileCheck %s --check-prefix=FORCE-USM-OFFLOAD
198198

199199
! FORCE-USM-OFFLOAD: "{{[^"]*}}flang-new" "-fc1" "-triple" "aarch64-unknown-linux-gnu"
200200
! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
201201
! FORCE-USM-OFFLOAD-NEXT: "{{[^"]*}}flang-new" "-fc1" "-triple" "amdgcn-amd-amdhsa"
202202
! FORCE-USM-OFFLOAD-SAME: "-fopenmp" "-fopenmp-force-usm"
203+
204+
! RUN: %flang -### -v --target=x86_64-unknown-linux-gnu -fopenmp \
205+
! RUN: --offload-arch=gfx900 \
206+
! RUN: --rocm-path=%S/Inputs/rocm %s 2>&1 \
207+
! RUN: | FileCheck --check-prefix=MLINK-BUILTIN-BITCODE %s
208+
! MLINK-BUILTIN-BITCODE: "{{[^"]*}}flang-new" "-fc1" "-triple" "amdgcn-amd-amdhsa"
209+
! MLINK-BUILTIN-BITCODE-SAME: "-mlink-builtin-bitcode" {{.*Inputs.*rocm.*amdgcn.*bitcode.*}}oclc_isa_version_900.bc

flang/test/Driver/target-cpu-features.f90

+2-2
Original file line numberDiff line numberDiff line change
@@ -29,10 +29,10 @@
2929
! RUN: %flang --target=riscv64-linux-gnu -c %s -### 2>&1 \
3030
! RUN: | FileCheck %s -check-prefix=CHECK-RV64
3131

32-
! RUN: %flang --target=amdgcn-amd-amdhsa -mcpu=gfx908 -c %s -### 2>&1 \
32+
! RUN: %flang --target=amdgcn-amd-amdhsa -mcpu=gfx908 -nogpulib -c %s -### 2>&1 \
3333
! RUN: | FileCheck %s -check-prefix=CHECK-AMDGPU
3434

35-
! RUN: %flang --target=r600-unknown-unknown -mcpu=cayman -c %s -### 2>&1 \
35+
! RUN: %flang --target=r600-unknown-unknown -mcpu=cayman -nogpulib -c %s -### 2>&1 \
3636
! RUN: | FileCheck %s -check-prefix=CHECK-AMDGPU-R600
3737

3838
! CHECK-A57: "-fc1" "-triple" "aarch64-unknown-linux-gnu"

flang/test/Driver/target-gpu-features.f90

+1-1
Original file line numberDiff line numberDiff line change
@@ -3,7 +3,7 @@
33
! Test that -mcpu are used and that the -target-cpu and -target-features
44
! are also added to the fc1 command.
55

6-
! RUN: %flang --target=amdgcn-amd-amdhsa -mcpu=gfx902 -c %s -### 2>&1 \
6+
! RUN: %flang --target=amdgcn-amd-amdhsa -mcpu=gfx902 -nogpulib -c %s -### 2>&1 \
77
! RUN: | FileCheck %s -check-prefix=CHECK-AMDGCN
88

99
! CHECK-AMDGCN: "-fc1" "-triple" "amdgcn-amd-amdhsa"

0 commit comments

Comments
 (0)