Runtime error: Check failed: IsAligned()

jarrellmark commented 7 years ago

Thanks for the great work on tensorflow-opencl. It's really great.

Summary

I'm getting a runtime error for almost all tensorflow programs:

2017-03-18 12:52:52.241954: F ./tensorflow/core/framework/tensor.cc:488] Check failed: IsAligned() Aborted (core dumped)

Environment Description

I have an Intel HD Graphics 5500 GPU with the Intel Broadwell i5 CPU x64. I'm using Intel's OpenCL drivers from here: https://software.intel.com/en-us/articles/opencl-drivers .

The OS is Ubuntu 16.04 LTS.

Python version is 3.5. It's running in a conda environment using Anaconda's versions of python, numpy, scipy, pyyaml, h5py, pandas, and jupyter.

However, the tensorflow pip package was compiled using Ubuntu's version of everything as per the compile from source instructions. I disabled Anaconda by removing it from ~/.bashrc, compiled the pip package, re-enabled Anaconda, activated the conda environment, and installed the pip package into the conda environment.

Steps to Reproduce

Here's the only tensorflow program I tried that did not fail:

import random
import sys
import tensorflow as tf
import time

random_number_generator = random.SystemRandom()

NUM_ROWS = 1024
NUM_COLUMNS = 1024

a_array = []
for i in range(1, (NUM_ROWS * NUM_COLUMNS) - 1):
    a_array.append(random_number_generator.random())

b_array = []
for i in range(1, (NUM_ROWS * NUM_COLUMNS) - 1):
    b_array.append(random_number_generator.random())

# Creates a graph.
with tf.device('/device:SYCL:0'):
    a = tf.constant(a_array, shape=[NUM_ROWS, NUM_COLUMNS], name='a', dtype=tf.float64)
    b = tf.constant(b_array, shape=[NUM_COLUMNS, NUM_ROWS], name='b', dtype=tf.float64)
    c = tf.matmul(a, b)

sess = tf.Session()

start = time.time()
sess.run(c)

Changing NUM_ROWS and NUM_COLUMNS to even 1200 resulted in the error above.

I also installed keras into the same conda environment using pip install keras and ran this script: https://github.com/fchollet/keras/blob/master/examples/lstm_text_generation.py . This resulted in the same error: Check failed: IsAligned(). The error is displayed after Build model... is outputted to the console.

Commit Hash (`git rev-parse HEAD`)

dda6b4ee253ca3016841ff60b16df4be40b5b052

Bazel Version

...........
Build label: 0.4.5
Build target: bazel-out/local-fastbuild/bin/src/main/java/com/google/devtools/build/lib/bazel/BazelServer_deploy.jar
Build time: Thu Mar 16 12:19:38 2017 (1489666778)
Build timestamp: 1489666778
Build timestamp as int: 1489666778

clinfo

Number of platforms                               1
  Platform Name                                   Intel(R) OpenCL
  Platform Vendor                                 Intel(R) Corporation
  Platform Version                                OpenCL 2.0 
  Platform Profile                                FULL_PROFILE
  Platform Extensions                             cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_icd cl_khr_image2d_from_buffer cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_spir
  Platform Extensions function suffix             INTEL

  Platform Name                                   Intel(R) OpenCL
Number of devices                                 2
  Device Name                                     Intel(R) HD Graphics
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.0 
  Driver Version                                  r4.0.59481
  Device OpenCL C Version                         OpenCL C 2.0 
  Device Type                                     GPU
  Device Profile                                  FULL_PROFILE
  Max compute units                               24
  Max clock frequency                             900MHz
  Device Partition                                (core)
    Max number of sub-devices                     0
    Supported partition types                     by <unknown> (0x7FF200000000)
  Max work item dimensions                        3
  Max work item sizes                             256x256x256
  Max work group size                             256
  Preferred work group size multiple              32
  Preferred / native vector sizes                 
    char                                                16 / 16      
    short                                                8 / 8       
    int                                                  4 / 4       
    long                                                 1 / 1       
    half                                                 8 / 8        (cl_khr_fp16)
    float                                                1 / 1       
    double                                               1 / 1        (cl_khr_fp64)
  Half-precision Floating-point support           (cl_khr_fp16)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Single-precision Floating-point support         (core)
    Denormals                                     No
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  Yes
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Address bits                                    64, Little-Endian
  Global memory size                              13231777383 (12.32GiB)
  Error Correction support                        No
  Max memory allocation                           4294959103 (4GiB)
  Unified memory for Host and Device              Yes
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   No
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           64 bytes
    Global                                        64 bytes
    Local                                         64 bytes
  Max size for global variable                    65536 (64KiB)
  Preferred total size of global vars             4294959103 (4GiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        589824
  Global Memory cache line                        64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             16
    Max size for 1D images from buffer            268434943 pixels
    Max 1D or 2D image array size                 2048 images
    Base address alignment for 2D image buffers   4 bytes
    Pitch alignment for 2D image buffers          4 bytes
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             2048x2048x2048 pixels
    Max number of read image args                 128
    Max number of write image args                128
    Max number of read/write image args           128
  Max number of pipe args                         16
  Max active pipe reservations                    1
  Max pipe packet size                            1024
  Local memory type                               Local
  Local memory size                               65536 (64KiB)
  Max constant buffer size                        4294959103 (4GiB)
  Max number of constant args                     8
  Max size of kernel argument                     1024
  Queue properties (on host)                      
    Out-of-order execution                        Yes
    Profiling                                     Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                131072 (128KiB)
    Max size                                      67108864 (64MiB)
  Max queues on device                            1
  Max events on device                            1024
  Prefer user sync for interop                    Yes
  Profiling timer resolution                      80ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            No
    SPIR versions                                 1.2 
  printf() buffer size                            4194304 (4MiB)
  Built-in kernels                                block_motion_estimate_intel;block_advanced_motion_estimate_check_intel;block_advanced_motion_estimate_bidirectional_check_intel
  Motion Estimation accelerator version (Intel)   2
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Device Extensions                               cl_intel_accelerator cl_intel_advanced_motion_estimation cl_intel_device_side_avc_motion_estimation cl_intel_driver_diagnostics cl_intel_media_block_io cl_intel_motion_estimation cl_intel_planar_yuv cl_intel_packed_yuv cl_intel_required_subgroup_size cl_intel_subgroups cl_intel_va_api_media_sharing cl_khr_3d_image_writes cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_fp16 cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_icd cl_khr_image2d_from_buffer cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_mipmap_image cl_khr_mipmap_image_writes cl_khr_spir 

  Device Name                                     Intel(R) Core(TM) i5-5200U CPU @ 2.20GHz
  Device Vendor                                   Intel(R) Corporation
  Device Vendor ID                                0x8086
  Device Version                                  OpenCL 2.0 (Build 400)
  Driver Version                                  1.2.0.400
  Device OpenCL C Version                         OpenCL C 2.0 
  Device Type                                     CPU
  Device Profile                                  FULL_PROFILE
  Max compute units                               4
  Max clock frequency                             2200MHz
  Device Partition                                (core)
    Max number of sub-devices                     4
    Supported partition types                     by counts, equally, by names (Intel)
  Max work item dimensions                        3
  Max work item sizes                             8192x8192x8192
  Max work group size                             8192
  Preferred work group size multiple              128
  Preferred / native vector sizes                 
    char                                                 1 / 32      
    short                                                1 / 16      
    int                                                  1 / 8       
    long                                                 1 / 4       
    half                                                 0 / 0        (n/a)
    float                                                1 / 8       
    double                                               1 / 4        (cl_khr_fp64)
  Half-precision Floating-point support           (n/a)
  Single-precision Floating-point support         (core)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 No
    Round to infinity                             No
    IEEE754-2008 fused multiply-add               No
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Double-precision Floating-point support         (cl_khr_fp64)
    Denormals                                     Yes
    Infinity and NANs                             Yes
    Round to nearest                              Yes
    Round to zero                                 Yes
    Round to infinity                             Yes
    IEEE754-2008 fused multiply-add               Yes
    Support is emulated in software               No
    Correctly-rounded divide and sqrt operations  No
  Address bits                                    64, Little-Endian
  Global memory size                              16550207488 (15.41GiB)
  Error Correction support                        No
  Max memory allocation                           4137551872 (3.853GiB)
  Unified memory for Host and Device              Yes
  Shared Virtual Memory (SVM) capabilities        (core)
    Coarse-grained buffer sharing                 Yes
    Fine-grained buffer sharing                   No
    Fine-grained system sharing                   No
    Atomics                                       No
  Minimum alignment for any data type             128 bytes
  Alignment of base address                       1024 bits (128 bytes)
  Preferred alignment for atomics                 
    SVM                                           64 bytes
    Global                                        64 bytes
    Local                                         0 bytes
  Max size for global variable                    65536 (64KiB)
  Preferred total size of global vars             65536 (64KiB)
  Global Memory cache type                        Read/Write
  Global Memory cache size                        262144
  Global Memory cache line                        64 bytes
  Image support                                   Yes
    Max number of samplers per kernel             480
    Max size for 1D images from buffer            258596992 pixels
    Max 1D or 2D image array size                 2048 images
    Base address alignment for 2D image buffers   64 bytes
    Pitch alignment for 2D image buffers          64 bytes
    Max 2D image size                             16384x16384 pixels
    Max 3D image size                             2048x2048x2048 pixels
    Max number of read image args                 480
    Max number of write image args                480
    Max number of read/write image args           480
  Max number of pipe args                         16
  Max active pipe reservations                    65535
  Max pipe packet size                            1024
  Local memory type                               Global
  Local memory size                               32768 (32KiB)
  Max constant buffer size                        131072 (128KiB)
  Max number of constant args                     480
  Max size of kernel argument                     3840 (3.75KiB)
  Queue properties (on host)                      
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Local thread execution (Intel)                Yes
  Queue properties (on device)                    
    Out-of-order execution                        Yes
    Profiling                                     Yes
    Preferred size                                4294967295 (4GiB)
    Max size                                      4294967295 (4GiB)
  Max queues on device                            4294967295
  Max events on device                            4294967295
  Prefer user sync for interop                    No
  Profiling timer resolution                      1ns
  Execution capabilities                          
    Run OpenCL kernels                            Yes
    Run native kernels                            Yes
    SPIR versions                                 1.2
  printf() buffer size                            1048576 (1024KiB)
  Built-in kernels                                
  Device Available                                Yes
  Compiler Available                              Yes
  Linker Available                                Yes
  Device Extensions                               cl_khr_icd cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_depth_images cl_khr_3d_image_writes cl_intel_exec_by_local_thread cl_khr_spir cl_khr_fp64 cl_khr_image2d_from_buffer 

NULL platform behavior
  clGetPlatformInfo(NULL, CL_PLATFORM_NAME, ...)  No platform
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_ALL, ...)   No platform
  clCreateContext(NULL, ...) [default]            No platform
  clCreateContext(NULL, ...) [other]              Success [INTEL]
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CPU)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_GPU)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ACCELERATOR)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_CUSTOM)  No platform
  clCreateContextFromType(NULL, CL_DEVICE_TYPE_ALL)  No platform

computecpp_info

********************************************************************************

ComputeCpp Info (CE 0.1.2)

********************************************************************************

Toolchain information:

GLIBCXX: 20150426
This version of libstdc++ is supported.

********************************************************************************

Device Info:

Discovered 1 devices matching:
  platform    : <any>
  device type : <any>

--------------------------------------------------------------------------------
Device 0:

  Device is supported                     : UNTESTED - Device not tested on this OS
  CL_DEVICE_NAME                          : Intel(R) HD Graphics
  CL_DEVICE_VENDOR                        : Intel(R) Corporation
  CL_DRIVER_VERSION                       : r4.0.59481
  CL_DEVICE_TYPE                          : CL_DEVICE_TYPE_GPU 
********************************************************************************

********************************************************************************

********************************************************************************

lukeiwanski commented 7 years ago

Hi @jarrellmark

Thanks for reporting this!

That has been addressed in 5cc8cdd58f3324c81eeab3e9a0af47754716e7fc. Could you please try it out?

SYCL default allocator did not take alignment into consideration. That now has been addressed in Eigen, where we are passing the required alignment to the custom allocator. C++ is great!

Thanks,

jarrellmark commented 7 years ago

Hey @lukeiwanski,

The IsAligned() message went away, but I'm getting this message now:

2017-03-20 21:30:14.765017: W ./tensorflow/core/common_runtime/sycl/sycl_util.h:44] No OpenCL GPU found that is supported by ComputeCpp, trying OpenCL CPU

Is there a way to force the GPU?

lukeiwanski commented 7 years ago

Currently we have an issue with memory alignment on Intel GPUs and have set the Intel GPU as "blacklisted" in Eigen. This means Eigen will not try to target Intel GPUs at the moment. We are working on a resolution for this and will update you when we have a fix available.

jarrellmark commented 7 years ago

Thanks, Luke.

I appreciate it and am excited about the progress that tensorflow-opencl is making.

nicholaslarusstone commented 7 years ago

Hi @lukeiwanski,

I'm having the same issue and was wondering if you have added Eigen support for Intel GPUs yet. If not, is there some way I can un-blacklist the Intel GPU?

Thanks for your hard work on this project!

lukeiwanski commented 7 years ago

Can you give it a spin on this branch: https://github.com/lukeiwanski/tensorflow/tree/dev/eigen_mehdi ?

nicholaslarusstone commented 7 years ago

That fixed the error, thanks a lot!

An unrelated question, tensorflow keeps telling me I'm running on a SYCL device, but then it calls that device a CPU. When I run sess = tf.Session(config=tf.ConfigProto(log_device_placement=True)), I get the following output:

/job:localhost/replica:0/task:0/device:SYCL:0 -> id: 0, type: CPU, name: Intel(R) Core(TM) i5-5300U CPU @ 2.30GHz, vendor: Intel(R) Corporation, profile: FULL_PROFILE

Running tensorflow.python.client.device_lib.list_local_devices() gives me the following:

[name: "/cpu:0" device_type: "CPU" memory_limit: 268435456 locality { } incarnation: 177593382533810523, name: "/device:SYCL:0" device_type: "SYCL" memory_limit: 268435456 locality { } incarnation: 1258559034356206920 physical_device_desc: "id: 0, type: CPU, name: Intel(R) Core(TM) i5-5300U CPU @ 2.30GHz, vendor: Intel(R) Corporation, profile: FULL_PROFILE"]

However, this device is NOT my GPU, as can be seen from when I run clinfo:

Platform Name Intel(R) OpenCL Number of devices 2 Device Name Intel(R) HD Graphics Device Vendor Intel(R) Corporation Device Vendor ID 0x8086 Device Version OpenCL 2.0 Driver Version r5.0.63503 Device OpenCL C Version OpenCL C 2.0 Device Type GPU Device Profile FULL_PROFILE

. . .

Device Name Intel(R) Core(TM) i5-5300U CPU @ 2.30GHz Device Vendor Intel(R) Corporation Device Vendor ID 0x8086 Device Version OpenCL 2.0 (Build 475) Driver Version 1.2.0.475 Device OpenCL C Version OpenCL C 2.0 Device Type CPU Device Profile FULL_PROFILE

Thanks for all your help already!

nicholaslarusstone commented 7 years ago

However, I am later getting this error when I try to run a simple keras model (just 2 dense layers): InternalError: Unknown error detected on device /job:localhost/replica:0/task:0/device:SYCL:0

lukeiwanski commented 7 years ago

That's interesting.. could you provide code to reproduce that issue?

nicholaslarusstone commented 7 years ago

I'm having trouble reproducing this issue because the code seems to just be hanging (I'm getting a lot of these messages: ./tensorflow/core/common_runtime/executor.cc:1556] Process node: 48 step 2 mul_3 = Mul[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:SYCL:0"](beta_1/read, Variable/read) is dead: 0

But here's my code: from keras.models import Sequential from keras.layers import Dense from keras import optimizers model = Sequential() model.add(Dense(32, input_shape=(timesteps, D_in))) model.add(Dense(D_out)) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) model.fit(X_train, y_train, batch_size=N, epochs=5, validation_data=(X_test, y_test))

nicholaslarusstone commented 7 years ago

Ah, ok I've reproduced the earlier error by using LSTM layers. It may be unreasonable for me to expect LSTM layers to work, but I am also having trouble with just dense layers (see above). Here's my code:

from keras.models import Sequential from keras.layers import LSTM, Dense from keras.layers.wrappers import TimeDistributed from keras import optimizers model = Sequential() model.add(LSTM(32, return_sequences=True, input_dim=D_in, input_length=timesteps)) model.add(LSTM(32, return_sequences=True)) model.add(TimeDistributed(Dense(D_out, activation='softmax'))) model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) model.fit(X_train, y_train, batch_size=N, epochs=1, validation_data=(X_test, y_test))

And here's a trace of the error message: InternalError Traceback (most recent call last)

in () ----> 1 model.fit(X_train, y_train, batch_size=N, epochs=1, validation_data=(X_test, y_test)) /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/keras/models.pyc in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, **kwargs) 861 class_weight=class_weight, 862 sample_weight=sample_weight, --> 863 initial_epoch=initial_epoch) 864 865 def evaluate(self, x, y, batch_size=32, verbose=1, /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/keras/engine/training.pyc in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, **kwargs) 1428 val_f=val_f, val_ins=val_ins, shuffle=shuffle, 1429 callback_metrics=callback_metrics, -> 1430 initial_epoch=initial_epoch) 1431 1432 def evaluate(self, x, y, batch_size=32, verbose=1, sample_weight=None): /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/keras/engine/training.pyc in _fit_loop(self, f, ins, out_labels, batch_size, epochs, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics, initial_epoch) 1077 batch_logs['size'] = len(batch_ids) 1078 callbacks.on_batch_begin(batch_index, batch_logs) -> 1079 outs = f(ins_batch) 1080 if not isinstance(outs, list): 1081 outs = [outs] /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/keras/backend/tensorflow_backend.pyc in __call__(self, inputs) 2266 updated = session.run(self.outputs + [self.updates_op], 2267 feed_dict=feed_dict, -> 2268 **self.session_kwargs) 2269 return updated[:len(self.outputs)] 2270 /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in run(self, fetches, feed_dict, options, run_metadata) 887 try: 888 result = self._run(None, fetches, feed_dict, options_ptr, --> 889 run_metadata_ptr) 890 if run_metadata: 891 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr) /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in _run(self, handle, fetches, feed_dict, options, run_metadata) 1116 if final_fetches or final_targets or (handle and feed_dict_tensor): 1117 results = self._do_run(handle, final_targets, final_fetches, -> 1118 feed_dict_tensor, options, run_metadata) 1119 else: 1120 results = [] /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata) 1313 if handle is None: 1314 return self._do_call(_run_fn, self._session, feeds, fetches, targets, -> 1315 options, run_metadata) 1316 else: 1317 return self._do_call(_prun_fn, self._session, handle, feeds, fetches) /home/nicholas/.virtualenvs/tensorflow-luke/local/lib/python2.7/site-packages/tensorflow/python/client/session.pyc in _do_call(self, fn, *args) 1332 except KeyError: 1333 pass -> 1334 raise type(e)(node_def, op, message) 1335 1336 def _extend_graph(self): InternalError: Unknown error detected on device /job:localhost/replica:0/task:0/device:SYCL:0

mihailescu2m commented 7 years ago

Hi @lukeiwanski

I am having the same issue (Check failed: IsAligned()) with tf-coriander (https://github.com/hughperkins/tf-coriander) using a Mali T-728 GPU.

Do you have a patch I could try to fix this? Or an advice on how to go about fixing it? Thanks!

DeadZen commented 5 years ago

Ping on this issue

benoitsteiner / tensorflow-opencl