scchan created this revision.
scchan added a reviewer: yaxunl.
Herald added a project: All.
scchan requested review of this revision.
Herald added subscribers: cfe-commits, MaskRay.
Herald added a project: clang.
When compiling in device only mode (e.g. --offload-device-only), the
host linker phase would not happen and therefore, the driver should
ignore all the host linker flags.
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D154881
Files:
clang/lib/Driver/Driver.cpp
clang/test/Driver/hip-phases.hip
Index: clang/test/Driver/hip-phases.hip
===================================================================
--- clang/test/Driver/hip-phases.hip
+++ clang/test/Driver/hip-phases.hip
@@ -229,6 +229,25 @@
// DBIN-DAG: [[P7:[0-9]+]]: linker, {[[P6]]}, hip-fatbin, (device-hip, )
// DBIN-DAG: [[P8:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)"
{[[P7]]}, hip-fatbin
// DBIN-NOT: host
+
+//
+// Test single gpu architecture with complete compilation in device-only
+// compilation mode with an unused host linker flag.
+//
+// RUN: %clang -x hip --target=x86_64-unknown-linux-gnu -ccc-print-phases \
+// RUN: --cuda-gpu-arch=gfx803 %s --cuda-device-only -Wl,--disable-new-dtags
2>&1 \
+// RUN: | FileCheck -check-prefixes=HLFDBIN %s
+// HLFDBIN-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T:hip]],
(device-[[T]], [[ARCH:gfx803]])
+// HLFDBIN-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output,
(device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P2:[0-9]+]]: compiler, {[[P1]]}, ir, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P3:[0-9]+]]: backend, {[[P2]]}, assembler, (device-[[T]],
[[ARCH]])
+// HLFDBIN-DAG: [[P4:[0-9]+]]: assembler, {[[P3]]}, object, (device-[[T]],
[[ARCH]])
+// HLFDBIN-DAG: [[P5:[0-9]+]]: linker, {[[P4]]}, image, (device-[[T]],
[[ARCH]])
+// HLFDBIN-DAG: [[P6:[0-9]+]]: offload, "device-[[T]]
(amdgcn-amd-amdhsa:[[ARCH]])" {[[P5]]}, image
+// HLFDBIN-DAG: [[P7:[0-9]+]]: linker, {[[P6]]}, hip-fatbin, (device-hip, )
+// HLFDBIN-DAG: [[P8:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)"
{[[P7]]}, hip-fatbin
+// HLFDBIN-NOT: host
+
//
// Test single gpu architecture up to the assemble phase in device-only
// compilation mode.
@@ -305,6 +324,32 @@
// DBIN2-DAG: [[P14:[0-9]+]]: linker, {[[P6]], [[P13]]}, hip-fatbin,
(device-hip, )
// DBIN2-DAG: [[P15:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)"
{[[P14]]}, hip-fatbin
// DBIN2-NOT: host
+
+//
+// Test two gpu architectures with complete compilation in device-only
+// compilation mode with an unused host linker flag.
+//
+// RUN: %clang -x hip --target=x86_64-unknown-linux-gnu -ccc-print-phases \
+// RUN: --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s --cuda-device-only
-Wl,--disable-new-dtags \
+// RUN: 2>&1 | FileCheck -check-prefixes=HLFDBIN2 %s
+// HLFDBIN2-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T:hip]],
(device-[[T]], [[ARCH:gfx803]])
+// HLFDBIN2-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output,
(device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P2:[0-9]+]]: compiler, {[[P1]]}, ir, (device-[[T]],
[[ARCH]])
+// HLFDBIN2-DAG: [[P3:[0-9]+]]: backend, {[[P2]]}, assembler, (device-[[T]],
[[ARCH]])
+// HLFDBIN2-DAG: [[P4:[0-9]+]]: assembler, {[[P3]]}, object, (device-[[T]],
[[ARCH]])
+// HLFDBIN2-DAG: [[P5:[0-9]+]]: linker, {[[P4]]}, image, (device-[[T]],
[[ARCH]])
+// HLFDBIN2-DAG: [[P6:[0-9]+]]: offload, "device-[[T]]
(amdgcn-amd-amdhsa:[[ARCH]])" {[[P5]]}, image
+// HLFDBIN2-DAG: [[P7:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T]],
(device-[[T]], [[ARCH2:gfx900]])
+// HLFDBIN2-DAG: [[P8:[0-9]+]]: preprocessor, {[[P7]]}, [[T]]-cpp-output,
(device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P9:[0-9]+]]: compiler, {[[P8]]}, ir, (device-[[T]],
[[ARCH2]])
+// HLFDBIN2-DAG: [[P10:[0-9]+]]: backend, {[[P9]]}, assembler, (device-[[T]],
[[ARCH2]])
+// HLFDBIN2-DAG: [[P11:[0-9]+]]: assembler, {[[P10]]}, object, (device-[[T]],
[[ARCH2]])
+// HLFDBIN2-DAG: [[P12:[0-9]+]]: linker, {[[P11]]}, image, (device-[[T]],
[[ARCH2]])
+// HLFDBIN2-DAG: [[P13:[0-9]+]]: offload, "device-[[T]]
(amdgcn-amd-amdhsa:[[ARCH2]])" {[[P12]]}, image
+// HLFDBIN2-DAG: [[P14:[0-9]+]]: linker, {[[P6]], [[P13]]}, hip-fatbin,
(device-hip, )
+// HLFDBIN2-DAG: [[P15:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)"
{[[P14]]}, hip-fatbin
+// HLFDBIN2-NOT: host
+
//
// Test two gpu architectures up to the assemble phase in device-only
// compilation mode.
Index: clang/lib/Driver/Driver.cpp
===================================================================
--- clang/lib/Driver/Driver.cpp
+++ clang/lib/Driver/Driver.cpp
@@ -4144,8 +4144,10 @@
if (Phase == phases::Link) {
assert(Phase == PL.back() && "linking must be final compilation
step.");
// We don't need to generate additional link commands if emitting AMD
bitcode
+ // or compiling only for the offload device
if (!(C.getInputArgs().hasArg(options::OPT_hip_link) &&
- (C.getInputArgs().hasArg(options::OPT_emit_llvm))))
+ (C.getInputArgs().hasArg(options::OPT_emit_llvm))) &&
+ !C.getInputArgs().hasArg(options::OPT_offload_device_only))
LinkerInputs.push_back(Current);
Current = nullptr;
break;
Index: clang/test/Driver/hip-phases.hip
===================================================================
--- clang/test/Driver/hip-phases.hip
+++ clang/test/Driver/hip-phases.hip
@@ -229,6 +229,25 @@
// DBIN-DAG: [[P7:[0-9]+]]: linker, {[[P6]]}, hip-fatbin, (device-hip, )
// DBIN-DAG: [[P8:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)" {[[P7]]}, hip-fatbin
// DBIN-NOT: host
+
+//
+// Test single gpu architecture with complete compilation in device-only
+// compilation mode with an unused host linker flag.
+//
+// RUN: %clang -x hip --target=x86_64-unknown-linux-gnu -ccc-print-phases \
+// RUN: --cuda-gpu-arch=gfx803 %s --cuda-device-only -Wl,--disable-new-dtags 2>&1 \
+// RUN: | FileCheck -check-prefixes=HLFDBIN %s
+// HLFDBIN-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T:hip]], (device-[[T]], [[ARCH:gfx803]])
+// HLFDBIN-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P2:[0-9]+]]: compiler, {[[P1]]}, ir, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P3:[0-9]+]]: backend, {[[P2]]}, assembler, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P4:[0-9]+]]: assembler, {[[P3]]}, object, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P5:[0-9]+]]: linker, {[[P4]]}, image, (device-[[T]], [[ARCH]])
+// HLFDBIN-DAG: [[P6:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:[[ARCH]])" {[[P5]]}, image
+// HLFDBIN-DAG: [[P7:[0-9]+]]: linker, {[[P6]]}, hip-fatbin, (device-hip, )
+// HLFDBIN-DAG: [[P8:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)" {[[P7]]}, hip-fatbin
+// HLFDBIN-NOT: host
+
//
// Test single gpu architecture up to the assemble phase in device-only
// compilation mode.
@@ -305,6 +324,32 @@
// DBIN2-DAG: [[P14:[0-9]+]]: linker, {[[P6]], [[P13]]}, hip-fatbin, (device-hip, )
// DBIN2-DAG: [[P15:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)" {[[P14]]}, hip-fatbin
// DBIN2-NOT: host
+
+//
+// Test two gpu architectures with complete compilation in device-only
+// compilation mode with an unused host linker flag.
+//
+// RUN: %clang -x hip --target=x86_64-unknown-linux-gnu -ccc-print-phases \
+// RUN: --cuda-gpu-arch=gfx803 --cuda-gpu-arch=gfx900 %s --cuda-device-only -Wl,--disable-new-dtags \
+// RUN: 2>&1 | FileCheck -check-prefixes=HLFDBIN2 %s
+// HLFDBIN2-DAG: [[P0:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T:hip]], (device-[[T]], [[ARCH:gfx803]])
+// HLFDBIN2-DAG: [[P1:[0-9]+]]: preprocessor, {[[P0]]}, [[T]]-cpp-output, (device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P2:[0-9]+]]: compiler, {[[P1]]}, ir, (device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P3:[0-9]+]]: backend, {[[P2]]}, assembler, (device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P4:[0-9]+]]: assembler, {[[P3]]}, object, (device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P5:[0-9]+]]: linker, {[[P4]]}, image, (device-[[T]], [[ARCH]])
+// HLFDBIN2-DAG: [[P6:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:[[ARCH]])" {[[P5]]}, image
+// HLFDBIN2-DAG: [[P7:[0-9]+]]: input, "{{.*}}hip-phases.hip", [[T]], (device-[[T]], [[ARCH2:gfx900]])
+// HLFDBIN2-DAG: [[P8:[0-9]+]]: preprocessor, {[[P7]]}, [[T]]-cpp-output, (device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P9:[0-9]+]]: compiler, {[[P8]]}, ir, (device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P10:[0-9]+]]: backend, {[[P9]]}, assembler, (device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P11:[0-9]+]]: assembler, {[[P10]]}, object, (device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P12:[0-9]+]]: linker, {[[P11]]}, image, (device-[[T]], [[ARCH2]])
+// HLFDBIN2-DAG: [[P13:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:[[ARCH2]])" {[[P12]]}, image
+// HLFDBIN2-DAG: [[P14:[0-9]+]]: linker, {[[P6]], [[P13]]}, hip-fatbin, (device-hip, )
+// HLFDBIN2-DAG: [[P15:[0-9]+]]: offload, "device-[[T]] (amdgcn-amd-amdhsa:)" {[[P14]]}, hip-fatbin
+// HLFDBIN2-NOT: host
+
//
// Test two gpu architectures up to the assemble phase in device-only
// compilation mode.
Index: clang/lib/Driver/Driver.cpp
===================================================================
--- clang/lib/Driver/Driver.cpp
+++ clang/lib/Driver/Driver.cpp
@@ -4144,8 +4144,10 @@
if (Phase == phases::Link) {
assert(Phase == PL.back() && "linking must be final compilation step.");
// We don't need to generate additional link commands if emitting AMD bitcode
+ // or compiling only for the offload device
if (!(C.getInputArgs().hasArg(options::OPT_hip_link) &&
- (C.getInputArgs().hasArg(options::OPT_emit_llvm))))
+ (C.getInputArgs().hasArg(options::OPT_emit_llvm))) &&
+ !C.getInputArgs().hasArg(options::OPT_offload_device_only))
LinkerInputs.push_back(Current);
Current = nullptr;
break;
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
