Skip to content

Commit 665a0a2

Browse files
npmillernpmiller
authored
[Driver][SYCL] Fix offload-bundler and offload-deps triples (#4616)
The patch 9838076 changed triple processing so gpu arch could be deducted even without extra `-` so we no longer need to add padding, it also seems like SYCL was inadvertently removed from the branch adding in the bound arch to the triple. So this patch fixes adding the bound arch in the triple when using SYCL, removes leftover triple padding code in the offload deps command. The test was also updated accordingly and it now also checks the triple used for `clang-offload-deps` so that we can hopefully catch mismatch between the two earlier in the future. This fixes #4307
1 parent 42a347d commit 665a0a2

File tree

2 files changed

+15
-18
lines changed

2 files changed

+15
-18
lines changed

‎clang/lib/Driver/ToolChains/Clang.cpp

+4-9
Original file line numberDiff line numberDiff line change
@@ -8207,7 +8207,7 @@ void OffloadBundler::ConstructJob(Compilation &C, const JobAction &JA,
82078207
Triples += '-';
82088208
Triples += CurTC->getTriple().normalize();
82098209
if ((CurKind == Action::OFK_HIP || CurKind == Action::OFK_OpenMP ||
8210-
CurKind == Action::OFK_Cuda) &&
8210+
CurKind == Action::OFK_Cuda || CurKind == Action::OFK_SYCL) &&
82118211
CurDep->getOffloadingArch()) {
82128212
Triples += '-';
82138213
Triples += CurDep->getOffloadingArch();
@@ -8365,7 +8365,8 @@ void OffloadBundler::ConstructJobMultipleOutputs(
83658365
Triples += Dep.DependentToolChain->getTriple().normalize();
83668366
if ((Dep.DependentOffloadKind == Action::OFK_HIP ||
83678367
Dep.DependentOffloadKind == Action::OFK_OpenMP ||
8368-
Dep.DependentOffloadKind == Action::OFK_Cuda) &&
8368+
Dep.DependentOffloadKind == Action::OFK_Cuda ||
8369+
Dep.DependentOffloadKind == Action::OFK_SYCL) &&
83698370
!Dep.DependentBoundArch.empty()) {
83708371
Triples += '-';
83718372
Triples += Dep.DependentBoundArch;
@@ -8654,13 +8655,7 @@ void OffloadDeps::constructJob(Compilation &C, const JobAction &JA,
86548655
if ((Dep.DependentOffloadKind == Action::OFK_HIP ||
86558656
Dep.DependentOffloadKind == Action::OFK_SYCL) &&
86568657
!Dep.DependentBoundArch.empty()) {
8657-
// If OffloadArch is present it can only appear as the 6th hyphen
8658-
// separated field of Bundle Entry ID. So, pad required number of
8659-
// hyphens in Triple.
8660-
// e.g. if NormalizedTriple is nvptx64-nvidia-cuda, 2 more - to
8661-
// generate nvptx64-nvidia-cuda--
8662-
for (int i = 4 - StringRef(NormalizedTriple).count("-"); i > 0; i--)
8663-
Targets += '-';
8658+
Targets += '-';
86648659
Targets += Dep.DependentBoundArch;
86658660
}
86668661
}

‎clang/test/Driver/sycl-offload-static-lib-2.cpp

+11-9
Original file line numberDiff line numberDiff line change
@@ -22,10 +22,11 @@
2222
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir %t_lib.lo -### %t_obj.o 2>&1 \
2323
// RUN: | FileCheck %s -check-prefix=STATIC_LIB -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown
2424
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -L/dummy/dir %t_lib.a -### %t_obj.o 2>&1 \
25-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
25+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
2626
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -L/dummy/dir %t_lib.lo -### %t_obj.o 2>&1 \
27-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
27+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
2828
// STATIC_LIB: clang-offload-bundler{{.*}} "-type=o" "-targets={{.*}},[[BUNDLE_TRIPLE]]" "-inputs=[[INPUTO:.+\.o]]" "-outputs=[[HOSTOBJ:.+\.o]],{{.+\.o}}"
29+
// STATIC_LIB: clang-offload-deps{{.*}} "-targets=[[BUNDLE_TRIPLE]]"
2930
// STATIC_LIB: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]" "-inputs={{.*}}" "-outputs=[[OUTFILE:.+\.a]]"
3031
// STATIC_LIB: llvm-link{{.*}} "[[OUTFILE]]"
3132
// STATIC_LIB: ld{{.*}} "{{.*}}_lib.{{(a|lo)}}" "[[HOSTOBJ]]"
@@ -40,10 +41,11 @@
4041
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl %t_lib.a -### %t-1.o %t-2.o %t-3.o 2>&1 \
4142
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_MULTI_O -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown
4243
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda %t_lib.a -### %t-1.o %t-2.o %t-3.o 2>&1 \
43-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_MULTI_O -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
44+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_MULTI_O -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
4445
// STATIC_LIB_MULTI_O: clang-offload-bundler{{.*}} "-type=o" "-targets={{.*}},[[BUNDLE_TRIPLE]]" "-inputs={{.+}}-1.o"
4546
// STATIC_LIB_MULTI_O: clang-offload-bundler{{.*}} "-type=o" "-targets={{.*}},[[BUNDLE_TRIPLE]]" "-inputs={{.+}}-2.o"
4647
// STATIC_LIB_MULTI_O: clang-offload-bundler{{.*}} "-type=o" "-targets={{.*}},[[BUNDLE_TRIPLE]]" "-inputs={{.+}}-3.o"
48+
// STATIC_LIB_MULTI_O: clang-offload-deps{{.*}} "-targets=[[BUNDLE_TRIPLE]]"
4749
// STATIC_LIB_MULTI_O: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]" {{.*}} "-outputs=[[OUTFILE:.+\.a]]"
4850
// STATIC_LIB_MULTI_O: llvm-link{{.*}} "[[OUTFILE]]"
4951

@@ -112,7 +114,7 @@
112114
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl %t_lib.a -o output_name -lOpenCL -### %s 2>&1 \
113115
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC2 -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown -DDEPS_TRIPLE=sycl-spir64-unknown-unknown
114116
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda %t_lib.a -o output_name -lOpenCL -### %s 2>&1 \
115-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC2 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda -DDEPS_TRIPLE=sycl-nvptx64-nvidia-cuda--sm_50
117+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC2 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50 -DDEPS_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
116118
// STATIC_LIB_SRC2: clang{{.*}} "-emit-obj" {{.*}} "-o" "[[HOSTOBJ:.+\.o]]"
117119
// STATIC_LIB_SRC2: ld{{(.exe)?}}" {{.*}} "-o" "[[HOSTEXE:.+\.out]]"
118120
// STATIC_LIB_SRC2: clang-offload-deps{{.*}} "-targets=[[DEPS_TRIPLE]]" "-outputs=[[OUTDEPS:.+\.bc]]" "[[HOSTEXE]]"
@@ -127,7 +129,7 @@
127129
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl %t_lib.a -o output_name -lstdc++ -z relro -### %s 2>&1 \
128130
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC3 -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown
129131
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda %t_lib.a -o output_name -lstdc++ -z relro -### %s 2>&1 \
130-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC3 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
132+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_SRC3 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
131133
// STATIC_LIB_SRC3: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]"
132134
// STATIC_LIB_SRC3: llvm-link{{.*}} "{{.*}}"
133135
// STATIC_LIB_SRC3: ld{{(.exe)?}}" {{.*}} "-o" "output_name" {{.*}} "-lstdc++" "-z" "relro"
@@ -145,9 +147,9 @@
145147
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -L/dummy/dir %t_obj.o -Wl,@%/t_arg.arg -### 2>&1 \
146148
// RUN: | FileCheck %s -check-prefix=WHOLE_STATIC_LIB -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown
147149
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -L/dummy/dir %t_obj.o -Wl,--whole-archive %t_lib.a %t_lib_2.a -Wl,--no-whole-archive -### 2>&1 \
148-
// RUN: | FileCheck %s -check-prefixes=WHOLE_STATIC_LIB,WHOLE_STATIC_LIB_1 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
150+
// RUN: | FileCheck %s -check-prefixes=WHOLE_STATIC_LIB,WHOLE_STATIC_LIB_1 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
149151
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -L/dummy/dir %t_obj.o -Wl,@%/t_arg.arg -### 2>&1 \
150-
// RUN: | FileCheck %s -check-prefix=WHOLE_STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
152+
// RUN: | FileCheck %s -check-prefix=WHOLE_STATIC_LIB -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
151153
// WHOLE_STATIC_LIB: clang-offload-bundler{{.*}} "-type=o" "-targets={{.*}},[[BUNDLE_TRIPLE]]"
152154
// WHOLE_STATIC_LIB: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]" "-inputs=[[INPUTA:.+\.a]]" "-outputs=[[OUTPUTA:.+\.a]]"
153155
// WHOLE_STATIC_LIB: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]" "-inputs=[[INPUTB:.+\.a]]" "-outputs=[[OUTPUTB:.+\.a]]"
@@ -172,9 +174,9 @@
172174
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fno-sycl-device-lib=all -L/dummy/dir %t_lib.lo -### 2>&1 \
173175
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -check-prefix=STATIC_LIB_NOSRC-SPIR -DTARGET=spir64 -DBUNDLE_TRIPLE=sycl-spir64-unknown-unknown
174176
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -fno-sycl-device-lib=all -L/dummy/dir %t_lib.a -### 2>&1 \
175-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -check-prefix=STATIC_LIB_NOSRC-CUDA -DTARGET=nvptx64 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
177+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -check-prefix=STATIC_LIB_NOSRC-CUDA -DTARGET=nvptx64 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
176178
// RUN: %clangxx -target x86_64-unknown-linux-gnu -fsycl -fsycl-targets=nvptx64-nvidia-cuda -fno-sycl-device-lib=all -L/dummy/dir %t_lib.lo -### 2>&1 \
177-
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -check-prefix=STATIC_LIB_NOSRC-CUDA -DTARGET=nvptx64 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda
179+
// RUN: | FileCheck %s -check-prefix=STATIC_LIB_NOSRC -check-prefix=STATIC_LIB_NOSRC-CUDA -DTARGET=nvptx64 -DBUNDLE_TRIPLE=sycl-nvptx64-nvidia-cuda-sm_50
178180
// STATIC_LIB_NOSRC: clang-offload-bundler{{.*}} "-type=a" "-targets=[[BUNDLE_TRIPLE]]" "-inputs={{.*}}_lib.{{(a|lo)}}" "-outputs=[[DEVICELIB:.+\.a]]" "-unbundle"
179181
// STATIC_LIB_NOSRC: llvm-link{{.*}} "[[DEVICELIB]]" "-o" "[[BCFILE:.+\.bc]]"
180182
// STATIC_LIB_NOSRC: sycl-post-link{{.*}} "-o" "[[TABLE:.+]]" "[[BCFILE]]"

0 commit comments

Comments
 (0)