[OpenCL] pwg_baker_static_0.4 fail

[yyqwert4]

• 标题：[OpenCL] pwg_baker_static_0.4 fail
• 版本、预测库信息：
     1）Paddle Lite 版本：branch develop
     2）Host 环境：Ubuntu 22.04
     3）运行设备环境：X86
     4）预测后端信息：OpenCL AMD Radeon RX 6900 XT
• 预测信息
     1）预测 API：API：C++
     2）预测选项信息：benchmark_bin，

cmd="gdb -args ./build.lite.linux.x86.gcc.opencl/lite/api/tools/benchmark/benchmark_bin --model_file=./models/pwg_baker_static_0.4/pwgan_csmsc.pdmodel --param_file=./models/pwg_baker_static_0.4/pwgan_csmsc.pdiparams -input_shape=1,80 --warmup=10 --repeats=20 --backend=opencl,x86 --gpu_precision=fp32"

-Model link
https://paddlespeech.bj.bcebos.com/Parakeet/released_models/pwgan/pwg_baker_static_0.4.zip

======= Opt Info =======
Load paddle model from ./models/pwg_baker_static_0.4/pwgan_csmsc.pdmodel and ./models/pwg_baker_static_0.4/pwgan_csmsc.pdiparams
Save optimized model to ./models/pwg_baker_static_0.4/opt.nb
I1017 10:41:20.336551 3691052 paddle_api.cc:50] need to check fp16 valid:0
I1017 10:41:20.336556 3691052 paddle_api.cc:57] Found opencl library:1
I1017 10:41:20.336557 3691052 paddle_api.cc:63] dlsym_success:1
I1017 10:41:20.336560 3691052 cl_runtime.h:93] need to check fp16 valid:0
I1017 10:41:20.336561 3691052 paddle_api.cc:70] opencl_valid:1
I1017 10:41:20.336565 3691052 paddle_api.cc:323] opencl binary path and file name:/paddle_lite_opencl_kernel.bin
I1017 10:41:20.336566 3691052 paddle_api.cc:50] need to check fp16 valid:0
I1017 10:41:20.336567 3691052 paddle_api.cc:57] Found opencl library:1
I1017 10:41:20.336568 3691052 paddle_api.cc:63] dlsym_success:1
I1017 10:41:20.336570 3691052 cl_runtime.h:93] need to check fp16 valid:0
I1017 10:41:20.336571 3691052 paddle_api.cc:70] opencl_valid:1
I1017 10:41:20.336578 3691052 cl_runtime.cc:863] tuned_file:/paddle_lite_opencl_tuned.params
W1017 10:41:20.336585 3691052 cl_runtime.cc:872] Not found tuned file:/paddle_lite_opencl_tuned.params
[New Thread 0x7ffff751f640 (LWP 3691100)]
I1017 10:41:20.345726 3691052 paddle_api.cc:341] set opencl_tune_mode: CL_TUNE_NORMAL, lws_repeats:4
I1017 10:41:20.345738 3691052 paddle_api.cc:344] tuned file path & name:/paddle_lite_opencl_tuned.params
I1017 10:41:20.345741 3691052 paddle_api.cc:50] need to check fp16 valid:0
I1017 10:41:20.345742 3691052 paddle_api.cc:57] Found opencl library:1
I1017 10:41:20.345743 3691052 paddle_api.cc:63] dlsym_success:1
I1017 10:41:20.345744 3691052 cl_runtime.h:93] need to check fp16 valid:0
I1017 10:41:20.345747 3691052 paddle_api.cc:70] opencl_valid:1
I1017 10:41:20.345748 3691052 paddle_api.cc:356] set opencl precision: CL_PRECISION_FP32
[Thread 0x7ffff75d0640 (LWP 3691057) exited]
I1017 10:41:20.388281 3691052 cl_runtime.h:93] need to check fp16 valid:0
W1017 10:41:20.388857 3691052 cl_runtime.cc:220] There is no precompiled OpenCL binary[/paddle_lite_opencl_kernel.bin] in the given OpenCL binary path. Also please make sure the storage directory exist and you have Write&Read permission. Jump to build program from source.
I1017 10:41:20.474515 3691052 elementwise_image_compute.cc:100] with y->persistable
I1017 10:41:20.522027 3691052 elementwise_image_compute.cc:100] with y->persistable

Thread 1 "benchmark_bin" received signal SIGSEGV, Segmentation fault.
0x00007ffff7663a6e in clEnqueueCopyBuffer () from /opt/rocm-5.2.3/lib/libamdocl64.so

(gdb) bt
#0 0x00007ffff7663a6e in clEnqueueCopyBuffer () from /opt/rocm-5.2.3/lib/libamdocl64.so
#1 0x00007ffff7fb3f42 in clEnqueueCopyBuffer (command_queue=0x555559946820, src_buffer=0xbdc131e43daa7a32, dst_buffer=0x555559fc3000, src_offset=0,
dst_offset=0, cb=128000, num_events_in_wait_list=0, event_wait_list=0x0, event=0x0)
at /home/qingchuan/work/ROCm/ROCm-OpenCL-Runtime/khronos/icd/loader/icd_dispatch.c:975
#2 0x00005555567d22fb in clEnqueueCopyBuffer (command_queue=0x555559946820, src_buffer=0xbdc131e43daa7a32, dst_buffer=0x555559fc3000, src_offset=0,
dst_offset=0, size=128000, num_events_in_wait_list=0, event_wait_list=0x0, event=0x0)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/backends/opencl/cl_wrapper.cc:759
#3 0x0000555557564b23 in cl::CommandQueue::enqueueCopyBuffer (this=0x7fffffffd608, src=..., dst=..., src_offset=0, dst_offset=0, size=128000,
events=0x0, event=0x0) at /home/yaoyao/work/gitlab/fortest/paddle-lite/third-party/opencl/include/CL/cl2.hpp:7090
#4 0x0000555557562c6a in paddle::lite::TargetWrapper<(paddle::lite_api::TargetType)5, cl::CommandQueue, cl::Event>::MemcpySync (dst=0x555559696690,
src=0x555558eccfc0, size=128000, dir=paddle::lite::IoDirection::DtoD)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/backends/opencl/target_wrapper.cc:250
#5 0x00005555569a05a7 in paddle::lite::kernels::opencl::CopyFromDeviceToDeviceSync (target=0x555559696690, source=0x555558eccfc0, size=128000)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/kernels/opencl/io_copy_buffer_compute.cc:66
#6 0x000055555697e85d in paddle::lite::kernels::opencl::SqueezeUnsqueezeCompute::Run (this=0x55555925bc20)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/kernels/opencl/squeeze_unsqueeze_buffer_compute.cc:63
#7 0x00005555567f8a08 in paddle::lite::KernelBase::Launch (this=0x55555925bc20) at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/core/kernel.h:116
#8 0x00005555578e1145 in paddle::lite::Instruction::Run (this=0x555558b93130) at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/core/program.cc:843
#9 0x00005555578df449 in paddle::lite::RuntimeProgram::Run (this=0x5555591fd3a0)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/core/program.cc:643
#10 0x000055555688577a in paddle::lite::LightPredictor::Run (this=0x55555902e9b0)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/light_api.h:71
#11 0x00005555568820a8 in paddle::lite::LightPredictorImpl::Run (this=0x555559286ec0)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/light_api_impl.cc:139
#12 0x00005555566da0e4 in paddle::lite_api::RunImpl (predictor=std::shared_ptrpaddle::lite_api::PaddlePredictor (use count 2, weak count 0) = {...},
perf_data=0x7fffffffdbb0) at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/tools/benchmark/benchmark.cc:82
#13 0x00005555566da677 in paddle::lite_api::Run (model_file="./models/pwg_baker_static_0.4/opt.nb",
input_shapes=std::vector of length 1, capacity 1 = {...}) at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/tools/benchmark/benchmark.cc:193
#14 0x00005555566d9e7f in paddle::lite_api::Benchmark (argc=1, argv=0x7fffffffe230)
at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/tools/benchmark/benchmark.cc:60
#15 0x00005555566d9d87 in main (argc=8, argv=0x7fffffffe1f8) at /home/yaoyao/work/gitlab/fortest/paddle-lite/lite/api/tools/benchmark/benchmark.cc:36
(gdb)

[paddle-bot]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网文档、常见问题、历史Issue来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ and Github Issue to get the answer.Have a nice day!

[sprouteer]

正在跟进

[sprouteer]

可以测试下别的平台，比如英伟达，或者高通gpu，mali gpu是否会报同样的问题，怀疑是驱动不支持设备间拷贝