[clang][sycl][nvlink] Share static library linking in Frontend/Offloading (#201253)

Move clang-nvlink-wrapper's archive member selection engine into a new
shared library in llvm/lib/Frontend/Offloading (ArchiveLinker.h/.cpp)
and use it from both clang-nvlink-wrapper and clang-sycl-linker, adding
static library (.a) and -l support to the SYCL linker.

The shared llvm::offloading::resolveArchiveMembers() API:
- Searches -L paths for -l library names (lib<name>.a or :<name>)
- Expands archives, honouring --whole-archive/--no-whole-archive
- Runs a symbol-driven fixed-point loop to extract only the archive
  members that resolve undefined symbols
- Returns the resolved MemoryBuffers and symbol table; the symbol table
  is consumed by clang-nvlink-wrapper's LTO resolution pass

clang-sycl-linker gains -l, --whole-archive/--no-whole-archive, and -u
options (added to SYCLLinkOpts.td). The previous --bc-library option
has been removed in favor of the standard -l mechanism.

Bug fixes included:

* Fix dangling StringRef UB: Args.getAllArgValues() returns a temporary
  vector; retain it in ForcedUndefStorage so the StringRefs
  remain valid through the resolveArchiveMembers call (both tools).
* Fix assert crash in clang-sycl-linker when all positional inputs are
  non-existent: return a proper error instead of propagating an empty
  buffer vector to linkInputs.
* Fixed forced undefined symbol handling: corrected -u option 
  processing to properly handle symbols.

Co-Authored-By: Claude
This commit is contained in:
Alexey BaderandClaude authored and GitHub committed 2026-06-05 15:41:04 -07:00
1 parent 1dae5dee50
commit 8aafa50c7a
12 files changed
+804 -406

No files matched your search

+77 -3
View File
@@ -50,7 +50,10 @@ be passed down to downstream AOT compilation tools like 'ocloc' and 'opencl-aot'
-help-hidden Display all available options
-help Display available options (--help-hidden for more)
-L <dir> Add <dir> to the library search path
--bc-library <name> Add LLVM bitcode library <name> (with extension) to the link. A relative <name> is resolved against -L paths; an absolute path is taken as-is.
-l <libname> Search for library <libname>
--whole-archive Include all archive members in the link
--no-whole-archive Only include archive members that resolve undefined symbols (default)
-u <symbol> Force undefined symbol during linking
--module-split-mode=<mode> Module split mode: 'source' (default), 'kernel', or 'none'
--ocloc-options=<value> Options passed to ocloc for Intel GPU AOT compilation
--opencl-aot-options=<value> Options passed to opencl-aot for Intel CPU AOT compilation
@@ -61,8 +64,58 @@ be passed down to downstream AOT compilation tools like 'ocloc' and 'opencl-aot'
-v Print verbose information
-spirv-dump-device-code=<dir> Directory to dump SPIR-V IR code into
Example
=======
Library Linking
===============
Device bitcode libraries can be packaged into archive libraries (``.a`` files)
using ``llvm-ar`` and linked using the ``-l`` option:
.. code-block:: console
llvm-ar rc libdevice.a func1.bc func2.bc func3.bc
clang-sycl-linker input.bc -l device -L /path/to/libs
The linker supports standard archive library search semantics:
* ``-l <name>`` searches for ``lib<name>.a`` in the directories specified by ``-L``
* ``-l :<exact-name>`` searches for the exact filename in the ``-L`` paths
* Absolute paths can be passed as positional arguments: ``clang-sycl-linker input.bc /path/to/libdevice.a``
By default, archive linking is **lazy** - only archive members (individual ``.bc`` files)
that resolve undefined symbols are extracted and linked. This happens at file
granularity: if any symbol in a ``.bc`` file is needed, all symbols in that file
are included. The linker uses a symbol-driven fixed-point algorithm: it
repeatedly scans archives to extract members that resolve currently undefined
symbols until no more extractions occur.
To force extraction of all archive members regardless of symbol resolution, use
``--whole-archive``:
.. code-block:: console
clang-sycl-linker input.bc --whole-archive -l device --no-whole-archive -l other
The ``-u <symbol>`` option can be used to force a symbol to be undefined, which
can trigger extraction of archive members that define that symbol:
.. code-block:: console
clang-sycl-linker input.bc -u my_init_function -l device
Implementation
--------------
Archive linking in ``clang-sycl-linker`` is implemented using the shared
``llvm::offloading::resolveArchiveMembers()`` API from
``llvm/lib/Frontend/Offloading/ArchiveLinker.cpp``. This same infrastructure is
also used by ``clang-nvlink-wrapper``, ensuring consistent archive linking
semantics across offloading tools.
Examples
========
Basic Usage
-----------
This tool is intended to be invoked when targeting any of the target offloading
toolchains. When the --sycl-link option is passed to the clang driver, the
@@ -74,3 +127,24 @@ generate the final executable.
.. code-block:: console
clang-sycl-linker --triple spirv64 --arch bmg_g21 input.bc
Linking with Device Libraries
------------------------------
To link device bitcode libraries, first package them into archive files:
.. code-block:: console
# Create device library archives
llvm-ar rc libmath.a sin.bc cos.bc tan.bc
llvm-ar rc libutils.a helper1.bc helper2.bc
# Link with lazy loading (only needed members extracted)
clang-sycl-linker --triple spirv64 kernel.bc -l math -l utils -L /path/to/libs -o kernel.spv
# Force all members to be included from libmath.a
clang-sycl-linker --triple spirv64 kernel.bc --whole-archive -l math --no-whole-archive -l utils -L /path/to/libs -o kernel.spv
# Use exact archive filename or absolute path
clang-sycl-linker --triple spirv64 kernel.bc -l :libmath.a -L /path/to/libs -o kernel.spv
clang-sycl-linker --triple spirv64 kernel.bc /absolute/path/libmath.a -o kernel.spv
@@ -22,76 +22,100 @@
;
; Test non-existent input file
; RUN: not clang-sycl-linker %t-missing.bc -o %t.out 2>&1 | FileCheck %s --check-prefix=MISSING
; MISSING: Input file '{{.*}}-missing.bc' does not exist
; MISSING: input file not found: '{{.*}}-missing.bc'
;
; Test the dry run of a simple case to link two input files.
; Test that IMG_SPIRV image kind is set for non-AOT compilation.
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc %t/input2.bc -o %t/spirv.out 2>&1 \
; RUN: | FileCheck %s --check-prefix=SIMPLE-FO
; SIMPLE-FO: link: inputs: {{.*}}.bc, {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; SIMPLE-FO: link: inputs: {{.*}}.bc, {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; SIMPLE-FO-NEXT: LLVM backend: input: [[LLVMLINKOUT]].bc, output: {{.*}}_0.spv
; SIMPLE-FO-NEXT: sycl-bundle: image kind: spv, triple: spirv64, arch: {{$}}
; SIMPLE-FO-NOT: {{.+}}
;
; Test the dry run of a simple case with device library files specified.
; Test the dry run of a simple case with device library archive specified using --whole-archive.
; RUN: mkdir -p %t/libs
; RUN: touch %t/libs/lib1.bc
; RUN: touch %t/libs/lib2.bc
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc %t/input2.bc --library-path=%t/libs --bc-library lib1.bc --bc-library lib2.bc -o a.spv 2>&1 \
; RUN: llvm-as %t/lib1.ll -o %t/libs/lib1.bc
; RUN: llvm-as %t/lib2.ll -o %t/libs/lib2.bc
; RUN: rm -f %t/libs/libdevice.a
; RUN: llvm-ar rc %t/libs/libdevice.a %t/libs/lib1.bc %t/libs/lib2.bc
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc %t/input2.bc --library-path=%t/libs --whole-archive -l device -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=DEVLIBS
; DEVLIBS: link: inputs: {{.*}}.bc libfiles: {{.*}}lib1.bc, {{.*}}lib2.bc output: [[LLVMLINKOUT:.*]].bc
; DEVLIBS: link: inputs: {{.*}}.bc, {{.*}}.bc, {{.*}}libdevice.a(lib1.bc), {{.*}}libdevice.a(lib2.bc) output: [[LLVMLINKOUT:.*]].bc
; DEVLIBS-NEXT: LLVM backend: input: [[LLVMLINKOUT]].bc, output: a_0.spv
; DEVLIBS-NEXT: sycl-bundle: image kind: spv, triple: spirv64, arch: {{$}}
; DEVLIBS-NOT: {{.+}}
;
; Test -L short form (joined) and --bc-library= joined form.
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc -L%t/libs --bc-library=lib1.bc -o a.spv 2>&1 \
; Test -L short form (joined) and -l with archive using --whole-archive.
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc -L%t/libs --whole-archive -l device -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=DEVLIBS-SHORT
; DEVLIBS-SHORT: link: inputs: {{.*}}.bc libfiles: {{.*}}libs{{[/\\]}}lib1.bc output: {{.*}}.bc
; DEVLIBS-SHORT: link: inputs: {{.*}}.bc, {{.*}}libdevice.a(lib1.bc), {{.*}}libdevice.a(lib2.bc) output: {{.*}}.bc
;
; Test that search continues past the first -L when the library is not found there. lib1.bc exists only in %t/libs (the second -L).
; Test that search continues past the first -L when the library is not found there. libdevice.a exists only in %t/libs (the second -L).
; RUN: mkdir -p %t/empty
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc -L %t/empty -L %t/libs --bc-library lib1.bc -o a.spv 2>&1 \
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc -L %t/empty -L %t/libs --whole-archive -l device -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=DEVLIBS-FALLTHROUGH
; DEVLIBS-FALLTHROUGH: link: inputs: {{.*}}.bc libfiles: {{.*}}libs{{[/\\]}}lib1.bc output: {{.*}}.bc
;
; Test that -L paths are searched in order: when the same name exists in multiple -L dirs, the first one wins.
; RUN: mkdir -p %t/libs2
; RUN: touch %t/libs/shadow.bc %t/libs2/shadow.bc
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t/input1.bc -L %t/libs2 -L %t/libs --bc-library shadow.bc -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=DEVLIBS-ORDER
; DEVLIBS-ORDER: link: inputs: {{.*}}.bc libfiles: {{.*}}libs2{{[/\\]}}shadow.bc output: {{.*}}.bc
; DEVLIBS-FALLTHROUGH: link: inputs: {{.*}}.bc, {{.*}}libdevice.a(lib1.bc), {{.*}}libdevice.a(lib2.bc) output: {{.*}}.bc
;
; Test a simple case with a random file (not bitcode) as input.
; RUN: touch %t/dummy.o
; RUN: not clang-sycl-linker %t/dummy.o -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=FILETYPEERROR
; FILETYPEERROR: Unsupported file type
; FILETYPEERROR: Unsupported file type: '{{.*}}dummy.o'
;
; Test that unsupported file type error includes buffer identifier when found inside an archive.
; Create an archive containing an unsupported file (text file instead of bitcode).
; RUN: echo "not bitcode" > %t/invalid.txt
; RUN: rm -f %t/libinvalid.a
; RUN: llvm-ar rc %t/libinvalid.a %t/invalid.txt
; RUN: not clang-sycl-linker --dry-run %t/input1.bc -L %t --whole-archive -l invalid -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=ARCHIVE-INVALID-MEMBER
; ARCHIVE-INVALID-MEMBER: Unsupported file type: '{{.*}}libinvalid.a(invalid.txt)'
;
; Test mixed archive: valid bitcode member + invalid member.
; The error should clearly identify which member is invalid.
; RUN: rm -f %t/libmixed.a
; RUN: llvm-ar rc %t/libmixed.a %t/libs/lib1.bc %t/invalid.txt
; RUN: not clang-sycl-linker --dry-run %t/input1.bc -L %t --whole-archive -l mixed -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=ARCHIVE-MIXED-INVALID
; ARCHIVE-MIXED-INVALID: Unsupported file type: '{{.*}}libmixed.a(invalid.txt)'
;
; Test to see if device library related errors are emitted.
; RUN: not clang-sycl-linker --dry-run %t/input1.bc %t/input2.bc --library-path=%t/libs --bc-library lib1.bc --bc-library lib2.bc --bc-library lib3.bc -o a.spv 2>&1 \
; RUN: not clang-sycl-linker --dry-run %t/input1.bc %t/input2.bc --library-path=%t/libs -l device -l nonexistent -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=DEVLIBSERR
; DEVLIBSERR: '{{.*}}lib3.bc' library file not found
; DEVLIBSERR: unable to find library -lnonexistent
;
; Test that there is no implicit CWD search: a bare bitcode name without any -L
; Test that there is no implicit CWD search: a bare library name without any -L
; must fail to resolve, even if a same-named file exists in the CWD.
; RUN: cd %t && not clang-sycl-linker --dry-run input1.bc --bc-library input1.bc -o a.spv 2>&1 \
; RUN: cd %t && not clang-sycl-linker --dry-run input1.bc -l mixed -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=NO-CWD-SEARCH
; NO-CWD-SEARCH: 'input1.bc' library file not found
; NO-CWD-SEARCH: unable to find library -lmixed
;
; Test that a directory matching the requested name is not accepted as a library:
; %t/libs is a directory created above; resolving --bc-library libs against -L %t
; would otherwise pick it up and fail later with a confusing bitcode-reader error.
; RUN: not clang-sycl-linker --dry-run %t/input1.bc -L %t --bc-library libs -o a.spv 2>&1 \
; %t/libs is a directory created above; resolving -l:libs against -L %t
; would detect it's a directory and error with the filename in the message.
; RUN: not clang-sycl-linker --dry-run %t/input1.bc -L %t -l :libs -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=NO-DIR-AS-LIB
; NO-DIR-AS-LIB: 'libs' library file not found
; NO-DIR-AS-LIB: '{{.*}}libs': Is a directory
;
; Test that providing only an empty archive results in "No input files could be resolved" error
; RUN: rm -f %t/empty.a
; RUN: llvm-ar rc %t/empty.a
; RUN: not clang-sycl-linker --dry-run --whole-archive %t/empty.a -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=NO-RESOLVED-INPUT
; NO-RESOLVED-INPUT: No input files could be resolved
;
; Test that providing only a lazy archive with no extracted members results in "No input files could be resolved" error
; RUN: not clang-sycl-linker --dry-run %t/libs/libdevice.a -o a.spv 2>&1 \
; RUN: | FileCheck %s --check-prefix=NO-RESOLVED-LAZY
; NO-RESOLVED-LAZY: No input files could be resolved
;
; Test AOT compilation for an Intel GPU.
; Test that IMG_Object image kind is set for AOT compilation (Intel GPU).
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none -arch=bmg_g21 %t/input1.bc %t/input2.bc -o %t/aot-gpu.out 2>&1 \
; RUN: --ocloc-options="-a -b" \
; RUN: | FileCheck %s --check-prefix=AOT-INTEL-GPU
; AOT-INTEL-GPU: link: inputs: {{.*}}.bc, {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; AOT-INTEL-GPU: link: inputs: {{.*}}.bc, {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; AOT-INTEL-GPU-NEXT: LLVM backend: input: [[LLVMLINKOUT]].bc, output: [[SPIRVTRANSLATIONOUT:.*]]_0.spv
; AOT-INTEL-GPU-NEXT: "{{.*}}ocloc{{.*}}" {{.*}}-device bmg_g21 -a -b {{.*}}-output [[SPIRVTRANSLATIONOUT]]_0.out -file [[SPIRVTRANSLATIONOUT]]_0.spv
; AOT-INTEL-GPU-NEXT: sycl-bundle: image kind: o, triple: spirv64, arch: bmg_g21
@@ -102,7 +126,7 @@
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none -arch=graniterapids %t/input1.bc %t/input2.bc -o %t/aot-cpu.out 2>&1 \
; RUN: --opencl-aot-options="-a -b" \
; RUN: | FileCheck %s --check-prefix=AOT-INTEL-CPU
; AOT-INTEL-CPU: link: inputs: {{.*}}.bc, {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; AOT-INTEL-CPU: link: inputs: {{.*}}.bc, {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; AOT-INTEL-CPU-NEXT: LLVM backend: input: [[LLVMLINKOUT]].bc, output: [[SPIRVTRANSLATIONOUT:.*]]_0.spv
; AOT-INTEL-CPU-NEXT: "{{.*}}opencl-aot{{.*}}" {{.*}}--device=cpu -a -b {{.*}}-o [[SPIRVTRANSLATIONOUT]]_0.out [[SPIRVTRANSLATIONOUT]]_0.spv
; AOT-INTEL-CPU-NEXT: sycl-bundle: image kind: o, triple: spirv64, arch: graniterapids
@@ -164,3 +188,19 @@ target triple = "spirv64"
define spir_func i32 @helper() {
ret i32 0
}
;--- lib1.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
target triple = "spirv64"
define spir_func i32 @lib1_func() {
ret i32 1
}
;--- lib2.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
target triple = "spirv64"
define spir_func i32 @lib2_func() {
ret i32 2
}
@@ -7,6 +7,11 @@
; RUN: llvm-as %t/bar.ll -o %t/bar.bc
; RUN: llvm-as %t/baz.ll -o %t/baz.bc
; RUN: llvm-as %t/libfoo.ll -o %t/libfoo.bc
; RUN: llvm-as %t/addFive.ll -o %t/addFive.bc
; RUN: llvm-as %t/unusedFunc.ll -o %t/unusedFunc.bc
; RUN: rm -f %t/libfoo.a %t/libdevice.a
; RUN: llvm-ar rc %t/libfoo.a %t/libfoo.bc
; RUN: llvm-ar rc %t/libdevice.a %t/addFive.bc %t/unusedFunc.bc
;
; Test linking two input files.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc --dry-run -o a.spv --print-linked-module 2>&1 \
@@ -22,18 +27,101 @@
; RUN: | FileCheck %s --check-prefix=CHECK-MULTIPLE-DEFS
; CHECK-MULTIPLE-DEFS: error: Linking globals named {{.*}}bar_func1{{.*}} symbol multiply defined!
;
; Test linking with a BC library file resolved through -L.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc --bc-library libfoo.bc -L %t --dry-run -o a.spv --print-linked-module 2>&1 \
; Test lazy linking with an archive library: only needed members are extracted.
; foo.bc references addFive, so addFive.bc is extracted from libdevice.a.
; unusedFunc.bc is not needed, so it should NOT be extracted.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc -l device -L %t --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-LAZY-LINK
; CHECK-LAZY-LINK: define {{.*}}foo_func1{{.*}}
; CHECK-LAZY-LINK: define {{.*}}foo_func2{{.*}}
; CHECK-LAZY-LINK: define {{.*}}bar_func1{{.*}}
; CHECK-LAZY-LINK: define {{.*}}addFive{{.*}}
; CHECK-LAZY-LINK-NOT: define {{.*}}unusedFunc{{.*}}
;
; Test linking with an archive library file using -l:libname.a syntax.
; Archive linking extracts members at file granularity, so all functions in libfoo.bc are included.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc -l :libfoo.a -L %t --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-DEVICE-LIB
; CHECK-DEVICE-LIB: define {{.*}}foo_func1{{.*}}
; CHECK-DEVICE-LIB: define {{.*}}foo_func2{{.*}}
; CHECK-DEVICE-LIB: define {{.*}}bar_func1{{.*}}
; CHECK-DEVICE-LIB: define {{.*}}addFive{{.*}}
; CHECK-DEVICE-LIB-NOT: define {{.*}}unusedFunc{{.*}}
; CHECK-DEVICE-LIB: define {{.*}}unusedFunc{{.*}}
;
; Test that an absolute path to --bc-library is taken as-is, with no -L required.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc --bc-library %t/libfoo.bc --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-DEVICE-LIB
; Test that an absolute path as a positional argument still performs lazy member extraction.
; libdevice.a has two members (addFive.bc and unusedFunc.bc).
; Since foo.bc needs addFive, only addFive.bc member is extracted; unusedFunc.bc is not.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc %t/libdevice.a --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-DEVICE-LIB-POS
; CHECK-DEVICE-LIB-POS: define {{.*}}foo_func1{{.*}}
; CHECK-DEVICE-LIB-POS: define {{.*}}foo_func2{{.*}}
; CHECK-DEVICE-LIB-POS: define {{.*}}bar_func1{{.*}}
; CHECK-DEVICE-LIB-POS: define {{.*}}addFive{{.*}}
; CHECK-DEVICE-LIB-POS-NOT: define {{.*}}unusedFunc{{.*}}
;
; Test that -L paths are searched in order: when the same name exists in multiple -L dirs, the first one wins.
; RUN: mkdir -p %t/libs1 %t/libs2
; RUN: rm -f %t/libs1/libshadow.a %t/libs2/libshadow.a
; RUN: llvm-ar rc %t/libs1/libshadow.a %t/addFive.bc
; RUN: llvm-ar rc %t/libs2/libshadow.a %t/unusedFunc.bc
; RUN: clang-sycl-linker %t/foo.bc -L %t/libs2 -L %t/libs1 --whole-archive -l shadow --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-LIB-ORDER
; CHECK-LIB-ORDER: define {{.*}}unusedFunc
; CHECK-LIB-ORDER-NOT: define {{.*}}addFive
;
; Test that -u forces extraction of an otherwise-unreferenced archive member.
; Without -u, unusedFunc is not extracted. With -u unusedFunc, it is pulled in.
; RUN: clang-sycl-linker %t/bar.bc %t/libdevice.a --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-NO-FORCE-UNDEF
; CHECK-NO-FORCE-UNDEF: define {{.*}}bar_func1{{.*}}
; CHECK-NO-FORCE-UNDEF-NOT: define {{.*}}unusedFunc{{.*}}
; CHECK-NO-FORCE-UNDEF-NOT: define {{.*}}addFive{{.*}}
;
; RUN: clang-sycl-linker %t/bar.bc %t/libdevice.a -u unusedFunc --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-FORCE-UNDEF
; CHECK-FORCE-UNDEF: define {{.*}}bar_func1{{.*}}
; CHECK-FORCE-UNDEF: define {{.*}}unusedFunc{{.*}}
; CHECK-FORCE-UNDEF-NOT: define {{.*}}addFive{{.*}}
;
; Test that multiple -u flags work correctly and extract all specified members.
; RUN: clang-sycl-linker %t/bar.bc %t/libdevice.a -u unusedFunc -u addFive --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-MULTI-UNDEF
; CHECK-MULTI-UNDEF: define {{.*}}bar_func1{{.*}}
; CHECK-MULTI-UNDEF: define {{.*}}addFive{{.*}}
; CHECK-MULTI-UNDEF: define {{.*}}unusedFunc{{.*}}
;
; Test that -u works correctly with -l library syntax (not just positional archives).
; RUN: clang-sycl-linker %t/bar.bc -l device -L %t -u unusedFunc --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-UNDEF-WITH-L
; CHECK-UNDEF-WITH-L: define {{.*}}bar_func1{{.*}}
; CHECK-UNDEF-WITH-L: define {{.*}}unusedFunc{{.*}}
; CHECK-UNDEF-WITH-L-NOT: define {{.*}}addFive{{.*}}
;
; Test that -u combined with actual references works correctly (both should be extracted).
; foo.bc references addFive, and -u forces unusedFunc.
; RUN: clang-sycl-linker %t/foo.bc %t/bar.bc %t/libdevice.a -u unusedFunc --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-UNDEF-PLUS-REF
; CHECK-UNDEF-PLUS-REF: define {{.*}}foo_func1{{.*}}
; CHECK-UNDEF-PLUS-REF: define {{.*}}bar_func1{{.*}}
; CHECK-UNDEF-PLUS-REF: define {{.*}}addFive{{.*}}
; CHECK-UNDEF-PLUS-REF: define {{.*}}unusedFunc{{.*}}
;
; Regression test: -u symbol should remain undefined until resolved by archive member.
; This test verifies the fix for the bug where forced-undefined entries were overwritten
; before ResolvesReference was computed, making -u ineffective.
; RUN: clang-sycl-linker %t/bar.bc -u addFive %t/libdevice.a --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-UNDEF-REMAINS
; CHECK-UNDEF-REMAINS: define {{.*}}bar_func1{{.*}}
; CHECK-UNDEF-REMAINS: define {{.*}}addFive{{.*}}
; CHECK-UNDEF-REMAINS-NOT: define {{.*}}unusedFunc{{.*}}
;
; Test -u with archive processed BEFORE the symbol table has been populated by regular inputs.
; This specifically tests that the forced-undefined placeholder survives initial processing.
; RUN: clang-sycl-linker -u addFive %t/libdevice.a %t/bar.bc --dry-run -o a.spv --print-linked-module 2>&1 \
; RUN: | FileCheck %s --check-prefix=CHECK-UNDEF-FIRST
; CHECK-UNDEF-FIRST: define {{.*}}addFive{{.*}}
; CHECK-UNDEF-FIRST: define {{.*}}bar_func1{{.*}}
; CHECK-UNDEF-FIRST-NOT: define {{.*}}unusedFunc{{.*}}
;--- foo.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
@@ -93,3 +181,23 @@ entry:
%res = mul nsw i32 %a, 5
ret i32 %res
}
;--- addFive.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
target triple = "spirv64"
define spir_func i32 @addFive(i32 %a) {
entry:
%res = add nsw i32 %a, 5
ret i32 %res
}
;--- unusedFunc.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
target triple = "spirv64"
define spir_func i32 @unusedFunc(i32 %a) {
entry:
%res = mul nsw i32 %a, 5
ret i32 %res
}
@@ -12,7 +12,7 @@
; Test the split mode ("none"): kernels from different TUs are not split into separate images.
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=none %t.bc -o %t-none.out 2>&1 \
; RUN: | FileCheck %s --check-prefix=SPLIT-NONE
; SPLIT-NONE: link: inputs: {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; SPLIT-NONE: link: inputs: {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; SPLIT-NONE-NEXT: LLVM backend: input: [[LLVMLINKOUT]].bc, output: {{.*}}_0.spv
; SPLIT-NONE-NEXT: sycl-bundle: image kind: spv, triple: spirv64, arch: {{$}}
; SPLIT-NONE-NOT: {{.+}}
@@ -20,7 +20,7 @@
; Test the split mode ("kernel"): each SPIR_KERNEL function produces its own device image.
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=kernel %t.bc -o %t-split-kernel.out 2>&1 \
; RUN: | FileCheck %s --check-prefix=SPLIT-KERNEL
; SPLIT-KERNEL: link: inputs: {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; SPLIT-KERNEL: link: inputs: {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; SPLIT-KERNEL-NEXT: sycl-module-split: input: [[LLVMLINKOUT]].bc, mode: kernel
; SPLIT-KERNEL-NEXT: [[SPLIT0:.*]].bc [kernel_c ]
; SPLIT-KERNEL-NEXT: [[SPLIT1:.*]].bc [kernel_b ]
@@ -43,7 +43,7 @@
; Test per-TU split ('source' explicitly provided)
; RUN: clang-sycl-linker --dry-run -v --module-split-mode=source %t.bc -o %t-src.out 2>&1 \
; RUN: | FileCheck %s --check-prefix=SPLIT-SRC
; SPLIT-SRC: link: inputs: {{.*}}.bc libfiles: output: [[LLVMLINKOUT:.*]].bc
; SPLIT-SRC: link: inputs: {{.*}}.bc output: [[LLVMLINKOUT:.*]].bc
; SPLIT-SRC-NEXT: sycl-module-split: input: [[LLVMLINKOUT]].bc, mode: source
; SPLIT-SRC-NEXT: [[S0:.*]].bc [kernel_b kernel_c ]
; SPLIT-SRC-NEXT: [[S1:.*]].bc [kernel_a ]
@@ -63,6 +63,8 @@ define spir_kernel void @kernel_c() #0 {
attributes #0 = { "sycl-module-id"="TU3.cpp" }
;--- no-triple.ll
target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
define spir_kernel void @kernel_d() #0 {
ret void
}
@@ -3,6 +3,7 @@ set(LLVM_LINK_COMPONENTS
BitWriter
Core
BinaryFormat
FrontendOffloading
MC
Target
TransformUtils
@@ -20,6 +20,7 @@
#include "llvm/BinaryFormat/Magic.h"
#include "llvm/Bitcode/BitcodeWriter.h"
#include "llvm/CodeGen/CommandFlags.h"
#include "llvm/Frontend/Offloading/ArchiveLinker.h"
#include "llvm/IR/DiagnosticPrinter.h"
#include "llvm/LTO/LTO.h"
#include "llvm/Object/Archive.h"
@@ -206,47 +207,6 @@ Expected<std::string> findProgram(const ArgList &Args, StringRef Name,
return *Path;
}
std::optional<std::string> findFile(StringRef Dir, StringRef Root,
const Twine &Name) {
SmallString<128> Path;
if (Dir.starts_with("="))
sys::path::append(Path, Root, Dir.substr(1), Name);
else
sys::path::append(Path, Dir, Name);
if (sys::fs::exists(Path))
return static_cast<std::string>(Path);
return std::nullopt;
}
std::optional<std::string>
findFromSearchPaths(StringRef Name, StringRef Root,
ArrayRef<StringRef> SearchPaths) {
for (StringRef Dir : SearchPaths)
if (std::optional<std::string> File = findFile(Dir, Root, Name))
return File;
return std::nullopt;
}
std::optional<std::string>
searchLibraryBaseName(StringRef Name, StringRef Root,
ArrayRef<StringRef> SearchPaths) {
for (StringRef Dir : SearchPaths)
if (std::optional<std::string> File =
findFile(Dir, Root, "lib" + Name + ".a"))
return File;
return std::nullopt;
}
/// Search for static libraries in the linker's library path given input like
/// `-lfoo` or `-l:libfoo.a`.
std::optional<std::string> searchLibrary(StringRef Input, StringRef Root,
ArrayRef<StringRef> SearchPaths) {
if (Input.starts_with(":"))
return findFromSearchPaths(Input.drop_front(), Root, SearchPaths);
return searchLibraryBaseName(Input, Root, SearchPaths);
}
void printCommands(ArrayRef<StringRef> CmdArgs) {
if (CmdArgs.empty())
return;
@@ -255,49 +215,6 @@ void printCommands(ArrayRef<StringRef> CmdArgs) {
errs() << join(std::next(CmdArgs.begin()), CmdArgs.end(), " ") << "\n";
}
/// A minimum symbol interface that provides the necessary information to
/// extract archive members and resolve LTO symbols.
struct Symbol {
enum Flags {
None = 0,
Undefined = 1 << 0,
Weak = 1 << 1,
};
Symbol() : File(), Flags(None), UsedInRegularObj(false) {}
Symbol(Symbol::Flags Flags) : File(), Flags(Flags), UsedInRegularObj(true) {}
Symbol(MemoryBufferRef File, const irsymtab::Reader::SymbolRef Sym)
: File(File), Flags(0), UsedInRegularObj(false) {
if (Sym.isUndefined())
Flags |= Undefined;
if (Sym.isWeak())
Flags |= Weak;
}
Symbol(MemoryBufferRef File, const SymbolRef Sym)
: File(File), Flags(0), UsedInRegularObj(false) {
auto FlagsOrErr = Sym.getFlags();
if (!FlagsOrErr)
reportError(FlagsOrErr.takeError());
if (*FlagsOrErr & SymbolRef::SF_Undefined)
Flags |= Undefined;
if (*FlagsOrErr & SymbolRef::SF_Weak)
Flags |= Weak;
auto NameOrErr = Sym.getName();
if (!NameOrErr)
reportError(NameOrErr.takeError());
}
bool isWeak() const { return Flags & Weak; }
bool isUndefined() const { return Flags & Undefined; }
MemoryBufferRef File;
uint32_t Flags;
bool UsedInRegularObj;
};
Expected<StringRef> runPTXAs(StringRef File, const ArgList &Args) {
SmallVector<StringRef, 1> SearchPaths;
if (Arg *A = Args.getLastArg(OPT_cuda_path_EQ))
@@ -413,97 +330,10 @@ Expected<std::unique_ptr<lto::LTO>> createLTO(const ArgList &Args) {
return std::make_unique<lto::LTO>(std::move(Conf), Backend, Partitions, Kind);
}
Expected<bool> getSymbolsFromBitcode(MemoryBufferRef Buffer,
StringMap<Symbol> &SymTab, bool IsLazy) {
Expected<IRSymtabFile> IRSymtabOrErr = readIRSymtab(Buffer);
if (!IRSymtabOrErr)
return IRSymtabOrErr.takeError();
bool Extracted = !IsLazy;
StringMap<Symbol> PendingSymbols;
for (unsigned I = 0; I != IRSymtabOrErr->Mods.size(); ++I) {
for (const auto &IRSym : IRSymtabOrErr->TheReader.module_symbols(I)) {
if (IRSym.isFormatSpecific() || !IRSym.isGlobal())
continue;
Symbol &OldSym = !SymTab.count(IRSym.getName()) && IsLazy
? PendingSymbols[IRSym.getName()]
: SymTab[IRSym.getName()];
Symbol Sym = Symbol(Buffer, IRSym);
if (OldSym.File.getBuffer().empty())
OldSym = Sym;
bool ResolvesReference =
!Sym.isUndefined() &&
(OldSym.isUndefined() || (OldSym.isWeak() && !Sym.isWeak())) &&
!(OldSym.isWeak() && OldSym.isUndefined() && IsLazy);
Extracted |= ResolvesReference;
Sym.UsedInRegularObj = OldSym.UsedInRegularObj;
if (ResolvesReference)
OldSym = Sym;
}
}
if (Extracted)
for (const auto &[Name, Symbol] : PendingSymbols)
SymTab[Name] = Symbol;
return Extracted;
}
Expected<bool> getSymbolsFromObject(ObjectFile &ObjFile,
StringMap<Symbol> &SymTab, bool IsLazy) {
bool Extracted = !IsLazy;
StringMap<Symbol> PendingSymbols;
for (SymbolRef ObjSym : ObjFile.symbols()) {
auto NameOrErr = ObjSym.getName();
if (!NameOrErr)
return NameOrErr.takeError();
Symbol &OldSym = !SymTab.count(*NameOrErr) && IsLazy
? PendingSymbols[*NameOrErr]
: SymTab[*NameOrErr];
Symbol Sym = Symbol(ObjFile.getMemoryBufferRef(), ObjSym);
if (OldSym.File.getBuffer().empty())
OldSym = Sym;
bool ResolvesReference = OldSym.isUndefined() && !Sym.isUndefined() &&
(!OldSym.isWeak() || !IsLazy);
Extracted |= ResolvesReference;
if (ResolvesReference)
OldSym = Sym;
OldSym.UsedInRegularObj = true;
}
if (Extracted)
for (const auto &[Name, Symbol] : PendingSymbols)
SymTab[Name] = Symbol;
return Extracted;
}
Expected<bool> getSymbols(MemoryBufferRef Buffer, StringMap<Symbol> &SymTab,
bool IsLazy) {
switch (identify_magic(Buffer.getBuffer())) {
case file_magic::bitcode: {
return getSymbolsFromBitcode(Buffer, SymTab, IsLazy);
}
case file_magic::elf_relocatable: {
Expected<std::unique_ptr<ObjectFile>> ObjFile =
ObjectFile::createObjectFile(Buffer);
if (!ObjFile)
return ObjFile.takeError();
return getSymbolsFromObject(**ObjFile, SymTab, IsLazy);
}
default:
return createStringError("Unsupported file type");
}
}
Expected<SmallVector<StringRef>> getInput(const ArgList &Args) {
SmallVector<StringRef> LibraryPaths;
for (const opt::Arg *Arg : Args.filtered(OPT_library_path))
LibraryPaths.push_back(Arg->getValue());
// Build input descriptors for the archive resolver
SmallVector<offloading::InputDesc> InputDescs;
bool WholeArchive = false;
SmallVector<std::pair<std::unique_ptr<MemoryBuffer>, bool>> InputFiles;
for (const opt::Arg *Arg : Args.filtered(
OPT_INPUT, OPT_library, OPT_whole_archive, OPT_no_whole_archive)) {
if (Arg->getOption().matches(OPT_whole_archive) ||
@@ -512,84 +342,43 @@ Expected<SmallVector<StringRef>> getInput(const ArgList &Args) {
continue;
}
std::optional<std::string> Filename =
Arg->getOption().matches(OPT_library)
? searchLibrary(Arg->getValue(), /*Root=*/"", LibraryPaths)
: std::string(Arg->getValue());
if (!Filename && Arg->getOption().matches(OPT_library))
return createStringError("unable to find library -l%s", Arg->getValue());
if (!Filename || !sys::fs::exists(*Filename) ||
sys::fs::is_directory(*Filename))
continue;
ErrorOr<std::unique_ptr<MemoryBuffer>> BufferOrErr =
MemoryBuffer::getFileOrSTDIN(*Filename);
if (std::error_code EC = BufferOrErr.getError())
return createFileError(*Filename, EC);
MemoryBufferRef Buffer = **BufferOrErr;
switch (identify_magic(Buffer.getBuffer())) {
case file_magic::bitcode:
case file_magic::elf_relocatable:
InputFiles.emplace_back(std::move(*BufferOrErr), /*IsLazy=*/false);
break;
case file_magic::archive: {
Expected<std::unique_ptr<object::Archive>> LibFile =
object::Archive::create(Buffer);
if (!LibFile)
return LibFile.takeError();
Error Err = Error::success();
for (auto Child : (*LibFile)->children(Err)) {
auto ChildBufferOrErr = Child.getMemoryBufferRef();
if (!ChildBufferOrErr)
return ChildBufferOrErr.takeError();
std::unique_ptr<MemoryBuffer> ChildBuffer =
MemoryBuffer::getMemBufferCopy(
ChildBufferOrErr->getBuffer(),
ChildBufferOrErr->getBufferIdentifier());
InputFiles.emplace_back(std::move(ChildBuffer), !WholeArchive);
}
if (Err)
return Err;
break;
}
default:
return createStringError("Unsupported file type");
}
offloading::InputDesc Desc;
Desc.Value = Arg->getValue();
Desc.InputKind = Arg->getOption().matches(OPT_library)
? offloading::InputDesc::Kind::Library
: offloading::InputDesc::Kind::File;
Desc.WholeArchive = WholeArchive;
InputDescs.push_back(Desc);
}
bool Extracted = true;
StringMap<Symbol> SymTab;
for (auto &Sym : Args.getAllArgValues(OPT_u))
SymTab[Sym] = Symbol(Symbol::Undefined);
SmallVector<std::unique_ptr<MemoryBuffer>> LinkerInput;
while (Extracted) {
Extracted = false;
for (auto &[Input, IsLazy] : InputFiles) {
if (!Input)
continue;
// Gather search paths and forced undefined symbols
SmallVector<StringRef> LibraryPaths;
for (const opt::Arg *Arg : Args.filtered(OPT_library_path))
LibraryPaths.push_back(Arg->getValue());
if (hasFatBinary(Args, *Input)) {
LinkerInput.emplace_back(std::move(Input));
continue;
}
std::vector<std::string> ForcedUndefStorage = Args.getAllArgValues(OPT_u);
SmallVector<StringRef> ForcedUndefs(ForcedUndefStorage.begin(),
ForcedUndefStorage.end());
// Archive members only extract if they define needed symbols. We will
// re-scan all the inputs if any files were extracted for the link job.
Expected<bool> ExtractOrErr = getSymbols(*Input, SymTab, IsLazy);
if (!ExtractOrErr)
return ExtractOrErr.takeError();
// The device code we are linking targets NVPTX. Any other ELF object is a
// host "fat binary" that should be forwarded without symbol scanning. The
// --assume-device-object flag (under --dry-run) overrides this and treats
// every input as device code, so disable detection by passing no archs.
SmallVector<Triple::ArchType> DeviceArchs;
if (!(Args.hasArg(OPT_dry_run) && Args.hasArg(OPT_assume_device_object)))
DeviceArchs = {Triple::nvptx, Triple::nvptx64};
Extracted |= *ExtractOrErr;
if (!*ExtractOrErr)
continue;
// Resolve archive members.
Expected<offloading::ResolvedInputs> ResolvedOrErr =
offloading::resolveArchiveMembers(InputDescs, LibraryPaths, ForcedUndefs,
"", DeviceArchs);
if (!ResolvedOrErr)
return ResolvedOrErr.takeError();
LinkerInput.emplace_back(std::move(Input));
}
}
InputFiles.clear();
offloading::ResolvedInputs &Resolved = *ResolvedOrErr;
SmallVector<std::unique_ptr<MemoryBuffer>> LinkerInput =
std::move(Resolved.Buffers);
StringMap<offloading::Symbol> &SymTab = Resolved.SymTab;
// Extract any bitcode files to be passed to the LTO pipeline.
SmallVector<std::unique_ptr<MemoryBuffer>> BitcodeFiles;
@@ -616,7 +405,7 @@ Expected<SmallVector<StringRef>> getInput(const ArgList &Args) {
size_t Idx = 0;
for (auto &Sym : Symbols) {
lto::SymbolResolution &Res = Resolutions[Idx++];
Symbol ObjSym = SymTab[Sym.getName()];
offloading::Symbol ObjSym = SymTab[Sym.getName()];
// We will use this as the prevailing symbol in LTO if it is not
// undefined and it is from the file that contained the canonical
// definition.
+67 -110
View File
@@ -18,11 +18,13 @@
#include "clang/Basic/OffloadArch.h"
#include "clang/Basic/Version.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringSwitch.h"
#include "llvm/BinaryFormat/Magic.h"
#include "llvm/Bitcode/BitcodeWriter.h"
#include "llvm/CodeGen/CommandFlags.h"
#include "llvm/Frontend/Offloading/ArchiveLinker.h"
#include "llvm/Frontend/Offloading/Utility.h"
#include "llvm/IR/DiagnosticPrinter.h"
#include "llvm/IR/LLVMContext.h"
@@ -187,84 +189,49 @@ static Error executeCommands(StringRef ExecutablePath,
return Error::success();
}
static Expected<SmallVector<std::string>> getInput(const ArgList &Args) {
// Collect all input bitcode files to be passed to the linking stage.
SmallVector<std::string> BitcodeFiles;
auto Inputs = Args.filtered(OPT_INPUT);
if (Inputs.empty())
static Expected<SmallVector<std::unique_ptr<MemoryBuffer>>>
getInput(const ArgList &Args) {
// Build input descriptors for the shared archive resolver
SmallVector<offloading::InputDesc> InputDescs;
bool WholeArchive = false;
for (const opt::Arg *Arg : Args.filtered(
OPT_INPUT, OPT_library, OPT_whole_archive, OPT_no_whole_archive)) {
if (Arg->getOption().matches(OPT_whole_archive) ||
Arg->getOption().matches(OPT_no_whole_archive)) {
WholeArchive = Arg->getOption().matches(OPT_whole_archive);
continue;
}
offloading::InputDesc Desc;
Desc.Value = Arg->getValue();
Desc.InputKind = Arg->getOption().matches(OPT_library)
? offloading::InputDesc::Kind::Library
: offloading::InputDesc::Kind::File;
Desc.WholeArchive = WholeArchive;
InputDescs.push_back(Desc);
}
if (InputDescs.empty())
return createStringError("No input files provided");
for (const opt::Arg *Arg : Inputs) {
StringRef Filename = Arg->getValue();
if (!sys::fs::exists(Filename) || sys::fs::is_directory(Filename))
return createStringError("Input file '" + Filename + "' does not exist");
file_magic Magic;
if (auto EC = identify_magic(Filename, Magic))
return createStringError("Failed to open file " + Filename);
// TODO: Current use case involves LLVM IR bitcode files as input.
// This will be extended to support SPIR-V IR files.
if (Magic != file_magic::bitcode)
return createStringError("Unsupported file type for '" + Filename + "'");
BitcodeFiles.push_back(std::string(Filename));
}
return BitcodeFiles;
}
/// Handle cases where input file is a LLVM IR bitcode file.
/// When clang-sycl-linker is called via clang-linker-wrapper tool, input files
/// are LLVM IR bitcode files.
// TODO: Support SPIR-V IR files.
static Expected<std::unique_ptr<Module>> getBitcodeModule(StringRef File,
LLVMContext &C) {
SMDiagnostic Err;
auto M = getLazyIRFileModule(File, Err, C);
if (M)
return std::move(M);
return createStringError(Err.getMessage());
}
static std::optional<std::string> findFile(StringRef Dir, const Twine &Name) {
SmallString<128> Path(Dir);
llvm::sys::path::append(Path, Name);
if (sys::fs::exists(Path) && !sys::fs::is_directory(Path))
return std::string(Path);
return std::nullopt;
}
static std::optional<std::string>
searchLibrary(StringRef Name, ArrayRef<StringRef> SearchPaths) {
// An absolute path is taken as-is; -L paths are only consulted for relative
// names.
if (sys::path::is_absolute(Name)) {
if (sys::fs::exists(Name) && !sys::fs::is_directory(Name))
return std::string(Name);
return std::nullopt;
}
for (StringRef Dir : SearchPaths)
if (std::optional<std::string> File = findFile(Dir, Name))
return File;
return std::nullopt;
}
/// Gather all library files. The list of files and its location are passed from
/// driver.
static Expected<SmallVector<std::string>>
getBCLibraryNames(const ArgList &Args) {
// Gather search paths and forced undefined symbols
SmallVector<StringRef> LibraryPaths;
for (const opt::Arg *Arg : Args.filtered(OPT_library_path))
LibraryPaths.push_back(Arg->getValue());
SmallVector<std::string> LibraryFiles;
for (const opt::Arg *Arg : Args.filtered(OPT_bc_library)) {
std::optional<std::string> LibName =
searchLibrary(Arg->getValue(), LibraryPaths);
if (!LibName)
return createStringError("'" + Twine(Arg->getValue()) +
"' library file not found");
LibraryFiles.push_back(std::move(*LibName));
}
std::vector<std::string> ForcedUndefStorage = Args.getAllArgValues(OPT_u);
SmallVector<StringRef> ForcedUndefs(ForcedUndefStorage.begin(),
ForcedUndefStorage.end());
return LibraryFiles;
Expected<offloading::ResolvedInputs> ResolvedOrErr =
offloading::resolveArchiveMembers(InputDescs, LibraryPaths, ForcedUndefs);
if (!ResolvedOrErr)
return ResolvedOrErr.takeError();
if (ResolvedOrErr->Buffers.empty())
return createStringError("No input files could be resolved");
return std::move(ResolvedOrErr->Buffers);
}
namespace {
@@ -280,19 +247,12 @@ struct LinkResult {
/// first input that supplies a triple as canonical. Issue an error if any
/// triple inputs disagree.
/// 2. Link all input bitcode images into one image using the linkInModule API.
/// 3. Gather all library bitcode images.
/// 4. Link all the images gathered in Step 3 with the output of Step 2 using
/// linkInModule API. LinkOnlyNeeded flag is used.
static Expected<LinkResult> linkInputs(ArrayRef<std::string> InputFiles,
const ArgList &Args, LLVMContext &C) {
static Expected<LinkResult>
linkInputs(ArrayRef<std::unique_ptr<MemoryBuffer>> InputBuffers,
const ArgList &Args, LLVMContext &C) {
llvm::TimeTraceScope TimeScope("Link code");
assert(InputFiles.size() && "No inputs to link");
// Get all library files.
Expected<SmallVector<std::string>> BCLibFiles = getBCLibraryNames(Args);
if (!BCLibFiles)
return BCLibFiles.takeError();
assert(InputBuffers.size() && "No inputs to link");
// Create a new file to write the linked file to.
auto BitcodeOutput =
@@ -301,11 +261,12 @@ static Expected<LinkResult> linkInputs(ArrayRef<std::string> InputFiles,
return BitcodeOutput.takeError();
if (Verbose) {
std::string Inputs = llvm::join(InputFiles.begin(), InputFiles.end(), ", ");
std::string LibInputs =
llvm::join((*BCLibFiles).begin(), (*BCLibFiles).end(), ", ");
errs() << formatv("link: inputs: {0} libfiles: {1} output: {2}\n", Inputs,
LibInputs, *BitcodeOutput);
std::string Inputs = llvm::join(
llvm::map_range(InputBuffers,
[](const auto &B) { return B->getBufferIdentifier(); }),
", ");
errs() << formatv("link: inputs: {0} output: {1}\n", Inputs,
*BitcodeOutput);
}
// Link input files. Resolve the target triple.
@@ -314,8 +275,14 @@ static Expected<LinkResult> linkInputs(ArrayRef<std::string> InputFiles,
auto LinkerOutput = std::make_unique<Module>("linker-output", C);
Linker L(*LinkerOutput);
for (auto &File : InputFiles) {
auto ModOrErr = getBitcodeModule(File, C);
for (const auto &Buffer : InputBuffers) {
// Check file type before attempting to parse as bitcode
file_magic Magic = identify_magic(Buffer->getBuffer());
if (Magic != file_magic::bitcode)
return createStringError("Unsupported file type: '" +
Buffer->getBufferIdentifier() + "'");
auto ModOrErr = parseBitcodeFile(Buffer->getMemBufferRef(), C);
if (!ModOrErr)
return ModOrErr.takeError();
@@ -323,11 +290,12 @@ static Expected<LinkResult> linkInputs(ArrayRef<std::string> InputFiles,
if (!T.empty() && T != TargetTriple) {
if (TargetTriple.empty()) {
TargetTriple = T;
TripleSource = File;
TripleSource = Buffer->getBufferIdentifier();
} else {
return createStringError(
"conflicting target triples: '" + TargetTriple.str() + "' (from " +
TripleSource + ") vs '" + T.str() + "' (from " + File + ")");
TripleSource + ") vs '" + T.str() + "' (from " +
Buffer->getBufferIdentifier() + ")");
}
}
@@ -339,18 +307,6 @@ static Expected<LinkResult> linkInputs(ArrayRef<std::string> InputFiles,
return createStringError(
"Target triple must be specified or inferable from inputs");
// Link in library files.
for (auto &File : *BCLibFiles) {
auto LibMod = getBitcodeModule(File, C);
if (!LibMod)
return LibMod.takeError();
if ((*LibMod)->getTargetTriple() == TargetTriple) {
unsigned Flags = Linker::Flags::LinkOnlyNeeded;
if (L.linkInModule(std::move(*LibMod), Flags))
return createStringError("Could not link IR");
}
}
// Dump linked output for testing.
if (Args.hasArg(OPT_print_linked_module))
outs() << *LinkerOutput;
@@ -693,13 +649,14 @@ static bool canSkipModuleSplit(IRSplitMode Mode, const Module &M,
/// 4. Optionally run AOT compilation when targeting an Intel HW arch.
/// 5. Pack the resulting images into a single OffloadBinary written to the
/// output file.
static Error runSYCLLink(ArrayRef<std::string> Files, const ArgList &Args) {
static Error runSYCLLink(ArrayRef<std::unique_ptr<MemoryBuffer>> Buffers,
const ArgList &Args) {
llvm::TimeTraceScope TimeScope("SYCL linking");
LLVMContext C;
// Link all input bitcode files and library files.
Expected<LinkResult> LinkedOrErr = linkInputs(Files, Args, C);
Expected<LinkResult> LinkedOrErr = linkInputs(Buffers, Args, C);
if (!LinkedOrErr)
return LinkedOrErr.takeError();
LinkResult &Result = *LinkedOrErr;
@@ -849,10 +806,10 @@ int main(int argc, char **argv) {
reportError(createStringError("Output file must be specified"));
OutputFile = Args.getLastArgValue(OPT_o);
// Get the input files to pass to the linking stage.
auto FilesOrErr = getInput(Args);
if (!FilesOrErr)
reportError(FilesOrErr.takeError());
// Get the input buffers to pass to the linking stage.
auto BuffersOrErr = getInput(Args);
if (!BuffersOrErr)
reportError(BuffersOrErr.takeError());
if (auto *A = Args.getLastArg(OPT_spirv_dump_device_code_EQ)) {
StringRef V = A->getValue();
@@ -871,7 +828,7 @@ int main(int argc, char **argv) {
}
// Run SYCL linking process on the generated inputs.
if (Error Err = runSYCLLink(*FilesOrErr, Args))
if (Error Err = runSYCLLink(*BuffersOrErr, Args))
reportError(std::move(Err));
// Remove the temporary files created.
+15 -6
View File
@@ -24,12 +24,21 @@ def library_path_S : Separate<["--", "-"], "library-path">, Flags<[HelpHidden]>,
def library_path_EQ : Joined<["--", "-"], "library-path=">, Flags<[HelpHidden]>,
Alias<library_path>;
def bc_library : Separate<["--", "-"], "bc-library">, MetaVarName<"<name>">,
HelpText<"Add LLVM bitcode library <name> (with extension) to the link. A "
"relative <name> is resolved against -L paths; an absolute path is "
"taken as-is.">;
def bc_library_EQ : Joined<["--", "-"], "bc-library=">, Flags<[HelpHidden]>,
Alias<bc_library>;
def library : JoinedOrSeparate<["-"], "l">, MetaVarName<"<libname>">,
HelpText<"Search for library <libname>">;
def library_S : Separate<["--", "-"], "library">, Flags<[HelpHidden]>,
Alias<library>;
def library_EQ : Joined<["--", "-"], "library=">, Flags<[HelpHidden]>,
Alias<library>;
def whole_archive : Flag<["--", "-"], "whole-archive">,
HelpText<"Include all archive members in the link">;
def no_whole_archive : Flag<["--", "-"], "no-whole-archive">,
HelpText<"Only include archive members that resolve undefined symbols (default)">;
def u : JoinedOrSeparate<["-"], "u">, MetaVarName<"<symbol>">,
HelpText<"Force undefined symbol during linking">;
def undefined : JoinedOrSeparate<["--"], "undefined">, Alias<u>;
def arch_EQ : Joined<["--", "-"], "arch=">,
Flags<[LinkerOnlyOption]>,
@@ -0,0 +1,116 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file declares shared functionality for linking static libraries
// (archives) in offloading tools. It provides a symbol-driven fixed-point
// archive member selection algorithm used by both clang-nvlink-wrapper and
// clang-sycl-linker.
//
//===----------------------------------------------------------------------===//
#ifndef LLVM_FRONTEND_OFFLOADING_ARCHIVELINKER_H
#define LLVM_FRONTEND_OFFLOADING_ARCHIVELINKER_H
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/Object/IRSymtab.h"
#include "llvm/Object/SymbolicFile.h"
#include "llvm/Support/Error.h"
#include "llvm/Support/MemoryBufferRef.h"
#include "llvm/TargetParser/Triple.h"
#include <cstdint>
#include <memory>
namespace llvm {
class MemoryBuffer;
namespace object {
class SymbolRef;
} // namespace object
namespace offloading {
/// A minimum symbol interface that provides the necessary information to
/// extract archive members and resolve LTO symbols.
struct Symbol {
enum Flags {
None = 0,
Undefined = 1 << 0,
Weak = 1 << 1,
};
Symbol() : File(), SymFlags(None), UsedInRegularObj(false) {}
Symbol(Symbol::Flags F) : File(), SymFlags(F), UsedInRegularObj(true) {}
Symbol(MemoryBufferRef File, const irsymtab::Reader::SymbolRef Sym)
: File(File), SymFlags(0), UsedInRegularObj(false) {
if (Sym.isUndefined())
SymFlags |= Undefined;
if (Sym.isWeak())
SymFlags |= Weak;
}
/// Create a Symbol from an object file symbol reference.
/// Returns an error if symbol flags cannot be retrieved.
static Expected<Symbol> createFromObject(MemoryBufferRef File,
const object::SymbolRef &Sym);
bool isWeak() const { return SymFlags & Weak; }
bool isUndefined() const { return SymFlags & Undefined; }
MemoryBufferRef File;
uint32_t SymFlags;
bool UsedInRegularObj;
};
/// Description of a single input (file or library).
struct InputDesc {
enum class Kind { File, Library };
StringRef Value; // File path, or library name for -l (the value after -l).
Kind InputKind = Kind::File;
bool WholeArchive = false; // --whole-archive state in effect at this input.
};
/// Result of archive member resolution.
struct ResolvedInputs {
SmallVector<std::unique_ptr<MemoryBuffer>>
Buffers; // Members to link, in order.
StringMap<Symbol> SymTab; // Symbol table (for LTO resolution).
};
/// Resolve archive members from the given inputs using a symbol-driven
/// fixed-point algorithm. For each input:
/// - If it's a Library, search for lib<name>.a or :<name> in SearchPaths
/// - If it's a File, use the path directly
/// - Archives are expanded and members are lazily extracted based on symbol
/// references unless WholeArchive is true
/// - Non-archive inputs (bitcode, ELF objects) are always included
///
/// Returns the buffers to link and the symbol table for LTO resolution.
///
/// \param Order Positional inputs + -l libraries in order.
/// \param SearchPaths -L paths for library search.
/// \param ForcedUndefs -u symbols (may be empty).
/// \param Root Sysroot for "=" prefixed paths ("" if none).
/// \param DeviceArchs Architectures of the device code being linked. When
/// non-empty, any ELF input whose architecture is not in this list (or
/// which cannot be parsed as an object) is treated as a "fat binary"
/// and passed through without symbol scanning (e.g., nvlink's cubin
/// detection). When empty, all inputs are scanned normally.
Expected<ResolvedInputs> resolveArchiveMembers(
ArrayRef<InputDesc> Order, ArrayRef<StringRef> SearchPaths,
ArrayRef<StringRef> ForcedUndefs = {}, StringRef Root = "",
ArrayRef<Triple::ArchType> DeviceArchs = {});
} // namespace offloading
} // namespace llvm
#endif // LLVM_FRONTEND_OFFLOADING_ARCHIVELINKER_H
@@ -0,0 +1,301 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file implements shared functionality for linking static libraries
// (archives) in offloading tools.
//
//===----------------------------------------------------------------------===//
#include "llvm/Frontend/Offloading/ArchiveLinker.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/BinaryFormat/Magic.h"
#include "llvm/Object/Archive.h"
#include "llvm/Object/IRObjectFile.h"
#include "llvm/Object/ObjectFile.h"
#include "llvm/Support/Error.h"
#include "llvm/Support/FileSystem.h"
#include "llvm/Support/MemoryBuffer.h"
#include "llvm/Support/Path.h"
#include <optional>
#include <string>
using namespace llvm;
using namespace llvm::object;
namespace llvm {
namespace offloading {
Expected<Symbol> Symbol::createFromObject(MemoryBufferRef File,
const SymbolRef &Sym) {
Symbol Result;
Result.File = File;
auto FlagsOrErr = Sym.getFlags();
if (!FlagsOrErr)
return FlagsOrErr.takeError();
if (*FlagsOrErr & SymbolRef::SF_Undefined)
Result.SymFlags |= Undefined;
if (*FlagsOrErr & SymbolRef::SF_Weak)
Result.SymFlags |= Weak;
return Result;
}
static std::optional<std::string> findFile(StringRef Dir, StringRef Root,
const Twine &Name) {
SmallString<128> Path;
if (Dir.starts_with("="))
sys::path::append(Path, Root, Dir.substr(1), Name);
else
sys::path::append(Path, Dir, Name);
if (sys::fs::exists(Path))
return static_cast<std::string>(Path);
return std::nullopt;
}
static std::optional<std::string>
findFromSearchPaths(StringRef Name, StringRef Root,
ArrayRef<StringRef> SearchPaths) {
for (StringRef Dir : SearchPaths)
if (std::optional<std::string> File = findFile(Dir, Root, Name))
return File;
return std::nullopt;
}
/// Search for static libraries in the linker's library path given input like
/// `-lfoo` or `-l:libfoo.a`.
static std::optional<std::string>
searchLibrary(StringRef Input, StringRef Root,
ArrayRef<StringRef> SearchPaths) {
if (Input.starts_with(":"))
return findFromSearchPaths(Input.drop_front(), Root, SearchPaths);
SmallString<128> LibName;
("lib" + Input + ".a").toVector(LibName);
return findFromSearchPaths(LibName, Root, SearchPaths);
}
static Expected<bool> getSymbolsFromBitcode(MemoryBufferRef Buffer,
StringMap<Symbol> &SymTab,
bool IsLazy) {
Expected<IRSymtabFile> IRSymtabOrErr = readIRSymtab(Buffer);
if (!IRSymtabOrErr)
return IRSymtabOrErr.takeError();
bool Extracted = !IsLazy;
StringMap<Symbol> PendingSymbols;
for (unsigned I = 0; I != IRSymtabOrErr->Mods.size(); ++I) {
for (const auto &IRSym : IRSymtabOrErr->TheReader.module_symbols(I)) {
if (IRSym.isFormatSpecific() || !IRSym.isGlobal())
continue;
StringMap<Symbol> &Target =
(IsLazy && !SymTab.count(IRSym.getName())) ? PendingSymbols : SymTab;
Symbol &OldSym = Target[IRSym.getName()];
Symbol Sym = Symbol(Buffer, IRSym);
if (OldSym.SymFlags == Symbol::None)
OldSym = Sym;
bool ResolvesReference =
!Sym.isUndefined() &&
(OldSym.isUndefined() || (OldSym.isWeak() && !Sym.isWeak())) &&
!(OldSym.isWeak() && OldSym.isUndefined() && IsLazy);
Extracted |= ResolvesReference;
Sym.UsedInRegularObj = OldSym.UsedInRegularObj;
if (ResolvesReference)
OldSym = Sym;
}
}
if (Extracted)
for (const auto &[Name, Symbol] : PendingSymbols)
SymTab[Name] = Symbol;
return Extracted;
}
static Expected<bool> getSymbolsFromObject(ObjectFile &ObjFile,
StringMap<Symbol> &SymTab,
bool IsLazy) {
bool Extracted = !IsLazy;
StringMap<Symbol> PendingSymbols;
for (SymbolRef ObjSym : ObjFile.symbols()) {
auto NameOrErr = ObjSym.getName();
if (!NameOrErr)
return NameOrErr.takeError();
StringMap<Symbol> &Target =
(IsLazy && !SymTab.count(*NameOrErr)) ? PendingSymbols : SymTab;
Symbol &OldSym = Target[*NameOrErr];
auto SymOrErr =
Symbol::createFromObject(ObjFile.getMemoryBufferRef(), ObjSym);
if (!SymOrErr)
return SymOrErr.takeError();
Symbol Sym = *SymOrErr;
if (OldSym.SymFlags == Symbol::None)
OldSym = Sym;
bool ResolvesReference = OldSym.isUndefined() && !Sym.isUndefined() &&
(!OldSym.isWeak() || !IsLazy);
Extracted |= ResolvesReference;
if (ResolvesReference)
OldSym = Sym;
OldSym.UsedInRegularObj = true;
}
if (Extracted)
for (const auto &[Name, Symbol] : PendingSymbols)
SymTab[Name] = Symbol;
return Extracted;
}
/// Identify "fat binary" inputs that should be passed through to the linker
/// without symbol-driven extraction. An input is a fat binary if \p DeviceArchs
/// is non-empty and the input is an ELF object whose architecture is not one of
/// the device architectures (or which fails to parse as an object file).
static bool isFatBinary(MemoryBufferRef Buffer,
ArrayRef<Triple::ArchType> DeviceArchs) {
if (DeviceArchs.empty())
return false;
if (identify_magic(Buffer.getBuffer()) != file_magic::elf_relocatable)
return false;
Expected<std::unique_ptr<ObjectFile>> ObjFile =
ObjectFile::createObjectFile(Buffer);
if (!ObjFile) {
// Assume fat binary if the object creation fails.
consumeError(ObjFile.takeError());
return true;
}
return !llvm::is_contained(DeviceArchs, (*ObjFile)->getArch());
}
static Expected<bool> getSymbols(MemoryBufferRef Buffer,
StringMap<Symbol> &SymTab, bool IsLazy) {
switch (identify_magic(Buffer.getBuffer())) {
case file_magic::bitcode: {
return getSymbolsFromBitcode(Buffer, SymTab, IsLazy);
}
case file_magic::elf_relocatable: {
Expected<std::unique_ptr<ObjectFile>> ObjFile =
ObjectFile::createObjectFile(Buffer);
if (!ObjFile)
return ObjFile.takeError();
return getSymbolsFromObject(**ObjFile, SymTab, IsLazy);
}
default:
return createStringError("Unsupported file type: '" +
Buffer.getBufferIdentifier() + "'");
}
}
Expected<ResolvedInputs>
resolveArchiveMembers(ArrayRef<InputDesc> Order,
ArrayRef<StringRef> SearchPaths,
ArrayRef<StringRef> ForcedUndefs, StringRef Root,
ArrayRef<Triple::ArchType> DeviceArchs) {
ResolvedInputs Result;
SmallVector<std::pair<std::unique_ptr<MemoryBuffer>, bool>> InputFiles;
// Process each input descriptor.
for (const InputDesc &Desc : Order) {
std::optional<std::string> Filename;
if (Desc.InputKind == InputDesc::Kind::Library) {
Filename = searchLibrary(Desc.Value, Root, SearchPaths);
if (!Filename)
return createStringError("unable to find library -l%s",
Desc.Value.str().c_str());
if (sys::fs::is_directory(*Filename))
return createStringError("'%s': Is a directory", Filename->c_str());
} else {
if (!sys::fs::exists(Desc.Value))
return createStringError("input file not found: '" + Desc.Value + "'");
if (sys::fs::is_directory(Desc.Value))
return createStringError("'" + Desc.Value + "': Is a directory");
Filename = Desc.Value.str();
}
if (!Filename)
continue;
auto BufferOrErr =
errorOrToExpected(MemoryBuffer::getFileOrSTDIN(*Filename));
if (!BufferOrErr)
return createFileError(*Filename, BufferOrErr.takeError());
MemoryBufferRef Buffer = (*BufferOrErr)->getMemBufferRef();
switch (identify_magic(Buffer.getBuffer())) {
case file_magic::bitcode:
case file_magic::elf_relocatable:
InputFiles.emplace_back(std::move(*BufferOrErr), /*IsLazy=*/false);
break;
case file_magic::archive: {
Expected<std::unique_ptr<object::Archive>> LibFile =
object::Archive::create(Buffer);
if (!LibFile)
return LibFile.takeError();
Error Err = Error::success();
for (auto Child : (*LibFile)->children(Err)) {
auto ChildBufferOrErr = Child.getMemoryBufferRef();
if (!ChildBufferOrErr)
return ChildBufferOrErr.takeError();
// Include archive name in buffer identifier for better diagnostics.
std::string BufferIdentifier =
(*Filename + "(" + ChildBufferOrErr->getBufferIdentifier() + ")")
.str();
std::unique_ptr<MemoryBuffer> ChildBuffer =
MemoryBuffer::getMemBufferCopy(ChildBufferOrErr->getBuffer(),
BufferIdentifier);
InputFiles.emplace_back(std::move(ChildBuffer), !Desc.WholeArchive);
}
if (Err)
return Err;
break;
}
default:
return createStringError("Unsupported file type: '" + *Filename + "'");
}
}
// Seed symbol table with forced undefined symbols.
for (StringRef Sym : ForcedUndefs)
Result.SymTab[Sym] = Symbol(Symbol::Undefined);
// Fixed-point loop to extract archive members.
bool Extracted = true;
while (Extracted) {
Extracted = false;
for (auto &[Input, IsLazy] : InputFiles) {
if (!Input)
continue;
// Check if this is a fat binary that should be passed through.
if (isFatBinary(*Input, DeviceArchs)) {
Result.Buffers.emplace_back(std::move(Input));
continue;
}
// Archive members only extract if they define needed symbols.
Expected<bool> ExtractOrErr = getSymbols(*Input, Result.SymTab, IsLazy);
if (!ExtractOrErr)
return ExtractOrErr.takeError();
Extracted |= *ExtractOrErr;
if (!*ExtractOrErr)
continue;
Result.Buffers.emplace_back(std::move(Input));
}
}
return Result;
}
} // namespace offloading
} // namespace llvm
@@ -1,4 +1,5 @@
add_llvm_component_library(LLVMFrontendOffloading
ArchiveLinker.cpp
Utility.cpp
OffloadWrapper.cpp
PropertySet.cpp