[Bug c++/104951] New: avx512fintrin.h(9146): error: identifier "__builtin_ia32_rndscaless_round" is undefined

[Bug c++/104951] New: avx512fintrin.h(9146): error: identifier "__builtin_ia32_rndscaless_round" is undefined

[tansheng at spacesoftwares dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104951

            Bug ID: 104951
           Summary: avx512fintrin.h(9146): error: identifier
                    "__builtin_ia32_rndscaless_round" is undefined
           Product: gcc
           Version: 9.4.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tansheng at spacesoftwares dot com
  Target Milestone: ---

Compiler ERROR, I was trying to compile paddlepaddle (both v2.2 and 2.1) with
gcc9.4.0, the following log give all the process and information.

It seems this is a problem from the compiler, not from the definitions or flags
of the project.

-----------------
Basic information:
   1）PaddlePaddle2.2.2
   2）CPU：i5-11400H
   3）GPU：NVIDIA GeForce RTX 3050 Laptop GPU, cuda_11.4.3_470.82.01_linux,
cudnn-11.4-linux-x64-v8.2.4.15
   4）system environment AsusLaptop TUF Gaming (FX506HCB) + Ubuntu20.04 +
Python3.8.10

-----------------
GCC + make log:
https://paddle-inference.readthedocs.io/en/latest/user_guides/source_compile.html#ubuntu-18-04
(1) down load and configure
git clone https://github.com/PaddlePaddle/Paddle.git
cd Paddle
git checkout release/2.2
mkdir build_cuda && cd build_cuda
cmake .. -DPY_VERSION=3
-DWITH_TESTING=OFF
-DWITH_MKL=ON
-DWITH_GPU=ON
-DON_INFER=ON
..
(2) make
make -j12
[ 0%] Built target extern_gflags
[ 1%] Built target extern_zlib
[ 1%] Built target extern_lapack
[ 1%] Built target extern_utf8proc
[ 1%] Built target extern_warpctc
[ 1%] Built target extern_dlpack
[ 1%] Built target extern_boost
[ 2%] Built target extern_threadpool
[ 3%] Built target extern_eigen3
[ 3%] copy_if_different
/home/mc/ocr/Paddle/build_cuda/paddle/fluid/operators/jit/kernels.h
[ 3%] Built target copy_kernels_command
[ 3%] copy_if_different
/home/mc/ocr/Paddle/build_cuda/paddle/fluid/inference/api/paddle_inference_pass.h
[ 3%] copy_if_different
/home/mc/ocr/Paddle/build_cuda/paddle/fluid/pybind/pybind.h
[ 3%] Built target download_externalError
[ 3%] Built target extern_xbyak
[ 3%] Built target extern_gloo
[ 3%] Built target copy_paddle_inference_pass_command
[ 3%] Built target copy_pybind_command
[ 3%] Built target extern_cryptopp
[ 3%] Built target extern_pocketfft
[ 3%] Built target extern_protobuf
[ 3%] Built target extern_pybind
[ 4%] Built target extern_mkldnn
[ 4%] Built target profiler_py_proto_init
[ 4%] Built target extern_glog
[ 4%] Built target extern_mklml
Consolidate compiler generated dependencies of target heter_service_proto
Consolidate compiler generated dependencies of target error_codes_proto
[ 4%] Built target framework_py_proto_init
Consolidate compiler generated dependencies of target data_feed_proto
Consolidate compiler generated dependencies of target framework_proto
Consolidate compiler generated dependencies of target external_error_proto
Copy generated python proto into directory paddle/fluid/proto/profiler.
[ 4%] Built target mkldnn_cmd
[ 4%] Built target trainer_py_proto
[ 5%] Built target distributed_strategy_py_proto
[ 5%] Built target pass_desc_py_proto
Consolidate compiler generated dependencies of target cudnn_workspace_helper
Consolidate compiler generated dependencies of target mkldnn
[ 5%] Built target mkldnn
[ 5%] Built target profiler_py_proto
[ 5%] Built target fleet_proto_init
Consolidate compiler generated dependencies of target monitor
Consolidate compiler generated dependencies of target version
Consolidate compiler generated dependencies of target denormal
Consolidate compiler generated dependencies of target shape_range_info_proto
[ 5%] Built target cudnn_workspace_helper
[ 5%] Built target version
[ 5%] Built target denormal
[ 5%] Built target error_codes_proto
[ 5%] Built target heter_service_proto
[ 5%] Built target framework_proto
[ 5%] Built target data_feed_proto
[ 5%] Built target external_error_proto
Copy generated python proto into directory paddle/fluid/proto.
[ 5%] Built target shape_range_info_proto
Consolidate compiler generated dependencies of target timer
Consolidate compiler generated dependencies of target pass_desc_proto
[ 5%] Built target pass_desc_proto
Consolidate compiler generated dependencies of target flags
Consolidate compiler generated dependencies of target string_array
Consolidate compiler generated dependencies of target op_def_proto
[ 5%] Built target op_def_proto
[ 5%] Built target monitor
[ 5%] Built target timer
Consolidate compiler generated dependencies of target zero_copy_tensor_dummy
[ 6%] Built target zero_copy_tensor_dummy
Consolidate compiler generated dependencies of target profiler_proto
[ 6%] Built target profiler_proto
Consolidate compiler generated dependencies of target trainer_desc_proto
[ 6%] Built target trainer_desc_proto
Consolidate compiler generated dependencies of target errors
Consolidate compiler generated dependencies of target op_version_proto
[ 6%] Built target errors
[ 6%] Built target op_version_proto
Consolidate compiler generated dependencies of target table_printer
[ 7%] Built target table_printer
[ 7%] Built target string_array
[ 7%] Built target extern_xxhash
Consolidate compiler generated dependencies of target paddle_pass_builder
Consolidate compiler generated dependencies of target op_def_api
[ 7%] Built target third_party
[ 7%] Built target op_def_api
[ 7%] Built target framework_py_proto
Consolidate compiler generated dependencies of target prune
Consolidate compiler generated dependencies of target op_version_registry
Consolidate compiler generated dependencies of target activation_functions
[ 7%] Built target paddle_pass_builder
[ 7%] Built target activation_functions
[ 7%] Built target flags
Consolidate compiler generated dependencies of target enforce
[ 7%] Built target enforce
Consolidate compiler generated dependencies of target imperative_profiler
[ 7%] Built target imperative_profiler
Consolidate compiler generated dependencies of target cuda_stream
Consolidate compiler generated dependencies of target imperative_flag
[ 7%] Built target imperative_flag
Consolidate compiler generated dependencies of target place
Consolidate compiler generated dependencies of target cpu_info
Consolidate compiler generated dependencies of target dynamic_loader
Consolidate compiler generated dependencies of target string_helper
[ 7%] Built target string_helper
Consolidate compiler generated dependencies of target stream_callback_manager
[ 7%] Built target cpu_info
Consolidate compiler generated dependencies of target ddim
Consolidate compiler generated dependencies of target pretty_log
[ 7%] Built target pretty_log
Consolidate compiler generated dependencies of target threadpool
Consolidate compiler generated dependencies of target stringpiece
[ 7%] Built target stringpiece
Consolidate compiler generated dependencies of target eigen_function
Consolidate compiler generated dependencies of target fs
Consolidate compiler generated dependencies of target shell
Consolidate compiler generated dependencies of target attribute
[ 7%] Built target prune
[ 7%] Built target attribute
[ 7%] Built target fs
[ 7%] Built target ddim
[ 7%] Built target threadpool
[ 7%] Built target place
[ 7%] Built target dynamic_loader
[ 7%] Built target stream_callback_manager
[ 7%] Built target shell
[ 7%] Built target op_version_registry
Consolidate compiler generated dependencies of target cuda_profiler
[ 7%] Built target cuda_profiler
Consolidate compiler generated dependencies of target workqueue
[ 7%] Built target workqueue
Consolidate compiler generated dependencies of target benchmark
Consolidate compiler generated dependencies of target dynload_lapack
[ 7%] Built target dynload_lapack
[ 7%] Built target benchmark
Consolidate compiler generated dependencies of target dynload_mklml
Consolidate compiler generated dependencies of target generator
Consolidate compiler generated dependencies of target data_loader
[ 7%] Built target dynload_mklml
[ 7%] Built target generator
[ 7%] Built target data_loader
Consolidate compiler generated dependencies of target memory_block
Consolidate compiler generated dependencies of target cblas
[ 7%] Built target cblas
Consolidate compiler generated dependencies of target dynload_warpctc
[ 7%] Built target memory_block
Consolidate compiler generated dependencies of target cpu_helper
Consolidate compiler generated dependencies of target allocator
Consolidate compiler generated dependencies of target paddle_crypto
[ 7%] Built target cpu_helper
[ 7%] Built target allocator
[ 7%] Built target dynload_warpctc
[ 7%] Built target paddle_crypto
Consolidate compiler generated dependencies of target best_fit_allocator
Consolidate compiler generated dependencies of target retry_allocator
[ 7%] Built target best_fit_allocator
[ 7%] Built target retry_allocator
Consolidate compiler generated dependencies of target op_proto_maker
[ 7%] Built target op_proto_maker
Consolidate compiler generated dependencies of target locked_allocator
Consolidate compiler generated dependencies of target cpu_allocator
Consolidate compiler generated dependencies of target buffered_allocator
Consolidate compiler generated dependencies of target pinned_allocator
Consolidate compiler generated dependencies of target thread_local_allocator
[ 7%] Built target locked_allocator
[ 7%] Built target cpu_allocator
[ 7%] Built target buffered_allocator
[ 7%] Built target pinned_allocator
Consolidate compiler generated dependencies of target aligned_allocator
Consolidate compiler generated dependencies of target mmap_allocator
[ 7%] Built target thread_local_allocator
[ 7%] Built target mmap_allocator
[ 7%] Built target aligned_allocator
Consolidate compiler generated dependencies of target
auto_growth_best_fit_allocator
Consolidate compiler generated dependencies of target op_call_stack
[ 7%] Built target auto_growth_best_fit_allocator
[ 7%] Built target op_call_stack
Consolidate compiler generated dependencies of target dynload_cuda
[ 7%] Built target cuda_stream
[ 7%] Built target dynload_cuda
[ 7%] Built target eigen_function
Consolidate compiler generated dependencies of target jit_kernel_base
[ 7%] Built target jit_kernel_base
Consolidate compiler generated dependencies of target gpu_info
Consolidate compiler generated dependencies of target device_tracer
[ 7%] Built target gpu_info
[ 7%] Built target device_tracer
Consolidate compiler generated dependencies of target jit_kernel_refer
Consolidate compiler generated dependencies of target jit_kernel_mkl
[ 7%] Built target jit_kernel_refer
[ 7%] Built target jit_kernel_mkl
Consolidate compiler generated dependencies of target jit_kernel_intrinsic
Consolidate compiler generated dependencies of target device_memory_aligment
Consolidate compiler generated dependencies of target cuda_resource_pool
[ 7%] Built target jit_kernel_intrinsic
[ 7%] Built target device_memory_aligment
[ 7%] Built target cuda_resource_pool
Consolidate compiler generated dependencies of target system_allocator
[ 7%] Built target system_allocator
Consolidate compiler generated dependencies of target cuda_device_guard
[ 7%] Built target cuda_device_guard
Consolidate compiler generated dependencies of target jit_kernel_mix
[ 7%] Built target jit_kernel_mix
Consolidate compiler generated dependencies of target buddy_allocator
[ 7%] Built target buddy_allocator
Consolidate compiler generated dependencies of target cuda_allocator
[ 7%] Built target cuda_allocator
Consolidate compiler generated dependencies of target profiler
[ 8%] Built target profiler
Consolidate compiler generated dependencies of target naive_best_fit_allocator
[ 8%] Built target naive_best_fit_allocator
Consolidate compiler generated dependencies of target allocator_strategy
[ 8%] Built target allocator_strategy
Consolidate compiler generated dependencies of target allocator_facade
Consolidate compiler generated dependencies of target jit_kernel_jitcode
[ 8%] Built target allocator_facade
Consolidate compiler generated dependencies of target cuda_graph
[ 9%] Built target jit_kernel_jitcode
[ 9%] Built target cuda_graph
Consolidate compiler generated dependencies of target jit_kernel_helper
[ 9%] Built target jit_kernel_helper
Consolidate compiler generated dependencies of target malloc
[ 9%] Built target malloc
Consolidate compiler generated dependencies of target device_context
[ 10%] Built target device_context
Consolidate compiler generated dependencies of target data_type
Consolidate compiler generated dependencies of target op_kernel_type
[ 10%] Built target data_type
Consolidate compiler generated dependencies of target device_code
[ 10%] Built target op_kernel_type
[ 10%] Built target device_code
Consolidate compiler generated dependencies of target
no_need_buffer_vars_inference
Consolidate compiler generated dependencies of target memcpy
[ 10%] Built target no_need_buffer_vars_inference
[ 10%] Built target memcpy
Consolidate compiler generated dependencies of target shape_inference
[ 10%] Built target shape_inference
Consolidate compiler generated dependencies of target mkldnn_axpy_handler
[ 10%] Built target mkldnn_axpy_handler
Consolidate compiler generated dependencies of target sequence_padding
Consolidate compiler generated dependencies of target blas
Consolidate compiler generated dependencies of target cos_sim_functor
Consolidate compiler generated dependencies of target im2col
[ 10%] Built target blas
[ 10%] Built target sequence_padding
Consolidate compiler generated dependencies of target sequence_scale
[ 10%] Built target cos_sim_functor
[ 10%] Built target im2col
Consolidate compiler generated dependencies of target cross_entropy
Consolidate compiler generated dependencies of target depthwise_conv
[ 10%] Built target sequence_scale
[ 10%] Built target cross_entropy
Consolidate compiler generated dependencies of target concat_and_split
Consolidate compiler generated dependencies of target lstm_compute
[ 10%] Built target concat_and_split
[ 10%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/lstm_compute.dir/lstm_compute.cu.o
Consolidate compiler generated dependencies of target maxouting
Consolidate compiler generated dependencies of target sample_prob
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/fc.dir/fc.cc.o
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/lapack_function.dir/lapack_function.cc.o
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/matrix_bit_code.dir/matrix_bit_code.cc.o
[ 10%] Built target depthwise_conv
[ 10%] Built target maxouting
Consolidate compiler generated dependencies of target sampler
[ 10%] Built target sampler
Consolidate compiler generated dependencies of target heter_wrapper
[ 10%] Built target heter_wrapper
[ 10%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/prelu.dir/prelu.cu.o
Consolidate compiler generated dependencies of target collective_helper
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/unpooling.dir/unpooling.cc.o
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/vol2col.dir/vol2col.cc.o
Consolidate compiler generated dependencies of target pooling
Consolidate compiler generated dependencies of target sequence2batch
[ 10%] Built target pooling
[ 10%] Built target sequence2batch
[ 10%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/vol2col.dir/vol2col.cu.o
[ 10%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/fc.dir/fc.cu.o
[ 10%] Built target sample_prob
[ 10%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/bert_encoder_functor.dir/bert_encoder_functor.cu.o
[ 10%] Built target collective_helper
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/matrix_inverse.dir/matrix_inverse.cc.o
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/segment_pooling.dir/segment_pooling.cc.o
[ 10%] Linking CXX static library liblapack_function.a
[ 10%] Built target lapack_function
[ 10%] Building CXX object
paddle/fluid/operators/math/CMakeFiles/matrix_inverse.dir/matrix_inverse.cu.cc.o
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(9146): error:
identifier "__builtin_ia32_rndscaless_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(9155): error:
identifier "__builtin_ia32_rndscalesd_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(14797): error:
identifier "__builtin_ia32_rndscaless_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(14806): error:
identifier "__builtin_ia32_rndscalesd_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512dqintrin.h(1365): error:
identifier "__builtin_ia32_fpclassss" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512dqintrin.h(1372): error:
identifier "__builtin_ia32_fpclasssd" is undefined
6 errors detected in the compilation of
"/home/mc/ocr/Paddle/paddle/fluid/operators/math/lstm_compute.cu".
make[2]: ***
[paddle/fluid/operators/math/CMakeFiles/lstm_compute.dir/build.make:90：paddle/fluid/operators/math/CMakeFiles/lstm_compute.dir/lstm_compute.cu.o]
错误 1
make[1]: ***
[CMakeFiles/Makefile2:77899：paddle/fluid/operators/math/CMakeFiles/lstm_compute.dir/all]
错误 2
make[1]: *** 正在等待未完成的任务....
[ 11%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/segment_pooling.dir/segment_pooling.cu.o
[ 11%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/unpooling.dir/unpooling.cu.o
[ 11%] Linking CXX static library libmatrix_bit_code.a
[ 11%] Built target matrix_bit_code
[ 11%] Linking CXX static library libmatrix_inverse.a
[ 11%] Built target matrix_inverse
[ 11%] Linking CXX static library libvol2col.a
[ 11%] Built target vol2col
[ 11%] Linking CXX static library libprelu.a
[ 11%] Built target prelu
[ 11%] Linking CXX static library libunpooling.a
[ 11%] Built target unpooling
[ 11%] Linking CXX static library libfc.a
[ 11%] Linking CXX static library libbert_encoder_functor.a
[ 11%] Built target bert_encoder_functor
[ 11%] Built target fc
[ 11%] Linking CXX static library libsegment_pooling.a
[ 11%] Built target segment_pooling
make: *** [Makefile:136：all] 错误 2

-----------------------------------------
below info is from
paddlepaddle2.1
-----------------------------------------
同样的配置，我试图使用不同的paddle版本，发现没有用；Paddle2.1也出现了类似的报错，主要都在fluid/operators/math下面的cuda算子函数，不清楚是否因为cuda版本问题引起，
[ 13%] Built target blas
[ 13%] Building CUDA object
paddle/fluid/operators/math/CMakeFiles/depthwise_conv.dir/depthwise_conv.cu.o
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(9146): error:
identifier "__builtin_ia32_rndscaless_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(9155): error:
identifier "__builtin_ia32_rndscalesd_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(14797): error:
identifier "__builtin_ia32_rndscaless_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512fintrin.h(14806): error:
identifier "__builtin_ia32_rndscalesd_round" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512dqintrin.h(1365): error:
identifier "__builtin_ia32_fpclassss" is undefined
/usr/lib/gcc/x86_64-linux-gnu/9/include/avx512dqintrin.h(1372): error:
identifier "__builtin_ia32_fpclasssd" is undefined

[Bug c++/104951] avx512fintrin.h(9146): error: identifier "__builtin_ia32_rndscaless_round" is undefined

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104951

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
That is not a diagnostic from GCC, most likely it comes from nvcc.
That means it isn't a GCC bug but either nvcc or user's fault in using gcc
intrinsics headers with a compiler that isn't able to deal with those.

[Bug c++/104951] avx512fintrin.h(9146): error: identifier "__builtin_ia32_rndscaless_round" is undefined

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104951

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
https://github.com/scale-snu/ckks-gpu-core/issues/1

https://stackoverflow.com/questions/60824242/suddenly-getting-builtin-ia32-sqrtsd-round-is-undefined-with-nvcc-gcc


I think this is a bug in PaddlePaddle build system.

[Bug target/104951] avx512fintrin.h(9146): error: identifier "__builtin_ia32_rndscaless_round" is undefined

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104951

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |MOVED

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Not a GCC bug so closing.