Unable to compile hls model

[Sam-Nielsen-Dot]

When I try to run the first tutorial notebook from hls4ml-tutorial

The code stops at the hls_model.compile() line and shows the error

Writing HLS project
Done
build_lib.sh: line 14: g++: command not found
build_lib.sh: line 15: g++: command not found
build_lib.sh: line 16: g++: command not found
---------------------------------------------------------------------------
OSError                                   Traceback (most recent call last)
Input In [13], in <cell line: 1>()
----> 1 hls_model.compile()
      2 X_test = np.ascontiguousarray(X_test)
      3 y_hls = hls_model.predict(X_test)

File ~/anaconda3/envs/hls4ml-tutorial/lib/python3.8/site-packages/hls4ml/model/hls_model.py:546, in HLSModel.compile(self)
    544         dlclose_func.restype = ctypes.c_int
    545         dlclose_func(self._top_function_lib._handle)
--> 546     self._top_function_lib = ctypes.cdll.LoadLibrary(lib_name)
    547 finally:
    548     os.chdir(curr_dir)

File ~/anaconda3/envs/hls4ml-tutorial/lib/python3.8/ctypes/__init__.py:451, in LibraryLoader.LoadLibrary(self, name)
    450 def LoadLibrary(self, name):
--> 451     return self._dlltype(name)

File ~/anaconda3/envs/hls4ml-tutorial/lib/python3.8/ctypes/__init__.py:373, in CDLL.__init__(self, name, mode, handle, use_errno, use_last_error, winmode)
    370 self._FuncPtr = _FuncPtr
    372 if handle is None:
--> 373     self._handle = _dlopen(self._name, mode)
    374 else:
    375     self._handle = handle

OSError: firmware/myproject-478C5A15.so: cannot open shared object file: No such file or directory

This same error occurs when I try to run any of the other tutorial notebooks or compile a HLS4ML model in any other piece of code

Furthermore,

when I try to just build the model without compiling it I get the error message

Project myproject_prj does not exist. Rerun "hls4ml build -p my-hls-test".

Do you know how to rectify this?

Thankyou

[vloncar]

Some tips on solving that can be found here.

[AnthonyAportela]

What ends up happening when you give a reply like this is that it shows up as a google result.

[adityaroy10]

Hello, I have been facing the same error. Did you figure out the solution?

[Govarshini99]

[jmitrevs]

In this case hls4ml is reporting that it can't open the library, most likely because it didn't exist. Can you look further back when the library was being compiled to see if there are errors?

[Govarshini99]

I didn't have any errors before this

[Govarshini99]

I tensorflow/core/util/port.cc:110] oneDNN custom operations are on. You may see slightly different numerical results due to floating-point round-off errors from different computation orders. To turn them off, set the environment variable TF_ENABLE_ONEDNN_OPTS=0.
2024-09-27 22:13:53.108362: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.
To enable the following instructions: AVX2 AVX_VNNI FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.
2024-09-27 22:13:53.542386: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT
2024-09-27 22:14:06.907216: I tensorflow/compiler/xla/stream_executor/cuda/cuda_gpu_executor.cc:995] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero. See more at https://github.com/torvalds/linux/blob/v6.0/Documentation/ABI/testing/sysfs-bus-pci#L344-L355
2024-09-27 22:14:06.930246: W tensorflow/core/common_runtime/gpu/gpu_device.cc:1960] Cannot dlopen some GPU libraries. Please make sure the missing libraries mentioned above are installed properly if you would like to use GPU. Follow the guide at https://www.tensorflow.org/install/gpu for how to download and setup the required libraries for your platform.
Skipping registering GPU devices...
2024-09-27 22:14:15.650756: I tensorflow/compiler/xla/service/service.cc:168] XLA service 0x71ec680086d0 initialized for platform Host (this does not guarantee that XLA will be used). Devices:
2024-09-27 22:14:15.650783: I tensorflow/compiler/xla/service/service.cc:176] StreamExecutor device (0): Host, Default Version
2024-09-27 22:14:15.680605: I ./tensorflow/compiler/jit/device_compiler.h:186] Compiled cluster using XLA! This line is logged at most once for the lifetime of the process.

got these set of warnings during during evaluate step before using hls converter

[Govarshini99]

but this is related to gpu usage

[eason0307]

I also have the same problem
It is the first example code
My OS is RHEL9.4

Does it have the solution?

[Govarshini99]

g++: fatal error: Killed signal terminated program cc1plus
compilation terminated.
/usr/bin/ld: cannot find myproject.o: No such file or directory
collect2: error: ld returned 1 exit status

My terminal is showing this error

[vloncar]

@eason0307 and @Govarshini99 Google this error for a list of potential causes and solutions. There are a few discussions on StackOverflow and elsewhere. It is related to the OS, compiler installation and overall the available resources (memory), not hls4ml. You can check the build_lib.sh in the output directory to see how it is compiled and you can run it manually while looking for a fix (no need for Jupyter environment).

[Govarshini99]

Will check it up , thanks for the reply.

[Govarshini99]

I ran build_lib.sh and got this error
g++: internal compiler error: Segmentation fault signal terminated program cclplus Please submit a full bug report, with preprocessed source if appropriate. See file:///usr/share/doc/gcc-11/README.Bugs for instructions. / collect2: error: ld returned 1 exit status usr/bin/ld: cannot find myproject.o: No such file or directory

Also checked the memory usage when the compilation was run seems to be full RAM memory is utilized , i tried to increase the swap memory size and reduce the complexity of my model , but still this issue persists...

def inverted_residual_block(x, expansion_factor, filters, stride):
    in_channels = x.shape[-1]
    expanded_channels = in_channels * expansion_factor
    # Expansion (1x1 Convolution)
    if expansion_factor != 1:
        x = layers.Conv2D(expanded_channels, kernel_size=1, padding='same', use_bias=False)(x)
        x = layers.BatchNormalization()(x)
        x = layers.ReLU(max_value=6.0)(x)
    
    # 3x3 Convolution (replacing Depthwise Conv2D)
    x = layers.Conv2D(expanded_channels, kernel_size=3, strides=stride, padding='same', use_bias=False)(x)
    x = layers.BatchNormalization()(x)
    x = layers.ReLU(max_value=6.0)(x)
    
    # Pointwise Convolution (1x1)
    x = layers.Conv2D(filters, kernel_size=1, padding='same', use_bias=False)(x)
    x = layers.BatchNormalization()(x)
    
    return x


def build_reduced_mobilenetv2(input_shape, num_classes):
    inputs = layers.Input(shape=input_shape)
    # Initial Convolution Layer 
    x = layers.Conv2D(12, kernel_size=3, strides=2, padding='same', use_bias=False)(inputs)
    x = layers.BatchNormalization()(x)
    x = layers.ReLU(max_value=6.0)(x)
    
    # Fewer and simpler inverted residual blocks
    x = inverted_residual_block(x, expansion_factor=1, filters=12, stride=1)  # Lower expansion
    x = inverted_residual_block(x, expansion_factor=2, filters=16, stride=2)  # Lower filters
    x = inverted_residual_block(x, expansion_factor=2, filters=24, stride=2)  # Final block
    
    # Final Convolution Layer 
    x = layers.Conv2D(320, kernel_size=1, padding='same', use_bias=False)(x)
    x = layers.BatchNormalization()(x)
    x = layers.ReLU(max_value=6.0)(x)
    
    # Global Average Pooling
    x = layers.GlobalAveragePooling2D()(x)
    
    # Fully connected output layer
    outputs = layers.Dense(num_classes, activation='softmax')(x)
    
    # Build the model
    model = models.Model(inputs, outputs)
    
    return model

this is the model i'm using

[Govarshini99]

Is anybody able to compile this kind of model?? I tried reducing model complexity even more but still facing this issue.