[HIP] device globals amdgcn-link error

[sherholz-intel]

Describe the bug

Trying to compile the Blender/Cycles SYCL backend with device globals support for AMD/HIP causes the build to fail with the following amdgcn-link error:

lld: /home/intel/Develop/sycl-demo/source/llvm-sycl/llvm/include/llvm/Support/Casting.h:578: decltype(auto) llvm::cast(From *) [To = llvm::Instruction, From = llvm::Value]: Assertion `isa<To>(Val) && "cast<Ty>() argument of incompatible type!"' failed.
PLEASE submit a bug report to https://github.com/intel/llvm/issues and include the crash backtrace.
Stack dump:
0.	Program arguments: /home/intel/Develop/sycl-demo/build/llvm-sycl/install/bin/lld -flavor gnu -m elf64_amdgpu --no-undefined -shared -plugin-opt=-amdgpu-internalize-symbols -plugin-opt=mcpu=gfx1030 -plugin-opt=O3 --lto-CGO3 --whole-archive -o /tmp/kernel-gfx1030-920f48-592629.out /tmp/kernel-gfx1030-0b121a-fee210.o --no-whole-archive
1.	Running pass 'Function Pass Manager' on module 'ld-temp.o'.
2.	Running pass 'FPBuiltin Function Selection' on function '@_ZTSZ41oneapi_kernel_integrator_init_from_cameraP16KernelGlobalsGPUmmRN4sycl3_V17handlerEP14KernelWorkTileiPfiEUlNS2_7nd_itemILi1EEEE_'
Stack dump without symbol names (ensure you have llvm-symbolizer in your PATH or set the environment var `LLVM_SYMBOLIZER_PATH` to point to it):
0  lld       0x00005e3c495481d8 llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) + 40
1  lld       0x00005e3c49545dae llvm::sys::RunSignalHandlers() + 238
2  lld       0x00005e3c49548b78
3  libc.so.6 0x000076bd8aa45330
4  libc.so.6 0x000076bd8aa9eb2c pthread_kill + 284
5  libc.so.6 0x000076bd8aa4527e gsignal + 30
6  libc.so.6 0x000076bd8aa288ff abort + 223
7  libc.so.6 0x000076bd8aa2881b
8  libc.so.6 0x000076bd8aa3b517
9  lld       0x00005e3c4b80fedb
10 lld       0x00005e3c4b80e382
11 lld       0x00005e3c4b80f7bc
12 lld       0x00005e3c4c61ed05 llvm::FPPassManager::runOnFunction(llvm::Function&) + 677
13 lld       0x00005e3c4c627132 llvm::FPPassManager::runOnModule(llvm::Module&) + 50
14 lld       0x00005e3c4c61f79c llvm::legacy::PassManagerImpl::run(llvm::Module&) + 1932
15 lld       0x00005e3c4a8a420c
16 lld       0x00005e3c4a8a308d llvm::lto::backend(llvm::lto::Config const&, std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>, unsigned int, llvm::Module&, llvm::ModuleSummaryIndex&) + 461
17 lld       0x00005e3c4a88d626 llvm::lto::LTO::runRegularLTO(std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>) + 1654
18 lld       0x00005e3c4a88cae0 llvm::lto::LTO::run(std::function<llvm::Expected<std::unique_ptr<llvm::CachedFileStream, std::default_delete<llvm::CachedFileStream>>> (unsigned int, llvm::Twine const&)>, llvm::FileCache) + 1168
19 lld       0x00005e3c4971c431 lld::elf::BitcodeCompiler::compile() + 1313
20 lld       0x00005e3c4969d67c void lld::elf::LinkerDriver::compileBitcodeFiles<llvm::object::ELFType<(llvm::endianness)1, true>>(bool) + 188
21 lld       0x00005e3c4967d89f void lld::elf::LinkerDriver::link<llvm::object::ELFType<(llvm::endianness)1, true>>(llvm::opt::InputArgList&) + 11023
22 lld       0x00005e3c49660ba6 lld::elf::LinkerDriver::linkerMain(llvm::ArrayRef<char const*>) + 15062
23 lld       0x00005e3c4965d050 lld::elf::link(llvm::ArrayRef<char const*>, llvm::raw_ostream&, llvm::raw_ostream&, bool, bool) + 800
24 lld       0x00005e3c49581956 lld::unsafeLldMain(llvm::ArrayRef<char const*>, llvm::raw_ostream&, llvm::raw_ostream&, llvm::ArrayRef<lld::DriverDef>, bool) + 2134
25 lld       0x00005e3c494d39fd lld_main(int, char**, llvm::ToolContext const&) + 301
26 lld       0x00005e3c494d3f47 main + 87
27 libc.so.6 0x000076bd8aa2a1ca
28 libc.so.6 0x000076bd8aa2a28b __libc_start_main + 139
29 lld       0x00005e3c494d3605 _start + 37
llvm-foreach: Aborted (core dumped)
clang++: error: amdgcn-link command failed with exit code 254 (use -v to see invocation)
clang version 20.0.0git (https://github.com/intel/llvm b512633e2de8547cecfce12aed2c62090282c6ed)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /home/intel/Develop/sycl-demo/build/llvm-sycl/install/bin
Build config: +assertions
clang++: note: diagnostic msg: Error generating preprocessed source(s).

To reproduce

1. Include code snippet as short as possible
2. Specify the command which should be used to compile the program

cmake -DWITH_CYCLES_DEBUG=OFF \
    -DWITH_CYCLES_DEVICE_CUDA=OFF \
    -DWITH_CYCLES_DEVICE_OPTIX=OFF \
    -DWITH_CYCLES_DEVICE_ONEAPI=ON \
    -DWITH_CYCLES_DEVICE_HIP=OFF \
    -DWITH_CYCLES_DEVICE_HIPRT=OFF \
    -DWITH_CYCLES_ONEAPI_BINARIES=ON \
    -DWITH_CYCLES_EMBREE=OFF \
	-DSYCL_OFFLINE_COMPILER_PARALLEL_JOBS=8 \
    -DSYCL_ROOT_DIR=${DPCPP_INSTALL_DIR} \
    -DCYCLES_ONEAPI_SYCL_TARGETS="amdgcn-amd-amdhsa" \
    -DCYCLES_ONEAPI_SYCL_OPTIONS_amdgcn-amd-amdhsa="--offload-arch=gfx1030" \
    -DWITH_X11_XINPUT=OFF \
    -DCMAKE_INSTALL_PREFIX=../install/blender-${BUILDTYPE} \
    ${PREFIX_PATH}/source/blender

make -j 20
make install -j

Environment

• OS: Linux
• Target device and vendor: AMD GPU
• DPC++ version: [e.g. commit hash or output of clang++ --version]
• Dependencies version: [e.g. the output of rocm-smi and sycl-ls --verbose]

SYCL-LS:

[level_zero:gpu][level_zero:0] Intel(R) oneAPI Unified Runtime over Level-Zero, Intel(R) Arc(TM) A770 Graphics 12.55.8 [1.6.32224+14]
[opencl:gpu][opencl:0] Intel(R) OpenCL Graphics, Intel(R) Arc(TM) A770 Graphics OpenCL 3.0 NEO  [24.52.32224]
[cuda:gpu][cuda:0] NVIDIA CUDA BACKEND, NVIDIA GeForce RTX 4070 8.9 [CUDA 12.4]
[hip:gpu][hip:0] AMD HIP BACKEND, AMD Radeon RX 6800 gfx1030 [HIP 60342.13]

Additional context

No response

[npmiller]

Is this with static device globals? It may be related to: #15329

[sherholz-intel]

@npmiller It ii probably not the same error since the device globals are not static:

SYCL_EXTERNAL extern device_global<const KernelData> kernel_global_oneapi_data;
SYCL_EXTERNAL extern device_global<const IntegratorStateGPU> kernel_global_oneapi_integrator_state;

But it might be related to the solution of the mentioned problem, since with an older compiler version v6.0.0-rc1 (5f25e25675b9393831eb53afef483c06ad483f2b) the code builds but throws a similar error as reported above:

<HIP>[ERROR]: 
UR HIP ERROR:
	Value:           500
	Name:            hipErrorNotFound
	Description:     named symbol not found
	Function:        getGlobalVariablePointer
	Source Location: /home/intel/Develop/sycl-demo/build/llvm-v6.0.0-rc1/_deps/unified-runtime-src/source/adapters/hip/program.cpp:254

<HIP>[ERROR]: 
UR HIP ERROR:
	Value:           UR_RESULT_ERROR_UNKNOWN
	Function:        deviceGlobalCopyHelper
	Source Location: /home/intel/Develop/sycl-demo/build/llvm-v6.0.0-rc1/_deps/unified-runtime-src/source/adapters/hip/enqueue.cpp:1691

oneAPI test kernel execution: got a runtime exception "Native API failed. Native API returns: 2147483646 (UR_RESULT_ERROR_UNKNOWN)"

But with the current compiler version (b512633e2de8547cecfce12aed2c62090282c6ed) it does not even compile/link anymore.

Maybe it is related to the solution for this #2475 , which probably are #15224 or #15148