[go: nahoru, domu]
Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

//tensorflow/compiler/xla/service/cpu:vectorized_reduce_with_no_vector_registers_test fails on s390x #44713

Open
skribm9 opened this issue Nov 9, 2020 · 7 comments
Open

//tensorflow/compiler/xla/service/cpu:vectorized_reduce_with_no_vector_registers_test fails on s390x #44713

skribm9 opened this issue Nov 9, 2020 · 7 comments
Assignees
@r4nt
Labels
comp:xla XLA stat:awaiting tensorflower Status - Awaiting response from tensorflower TF 2.9 Issues found in the TF 2.9 release (or RCs) type:bug Bug

Comments

@skribm9
Copy link
Contributor
skribm9 commented Nov 9, 2020 •
edited
Loading

System information

  • Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
  • OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ub18.04
  • Mobile device (e.g. iPhone 8, Pixel 2, Samsung Galaxy) if the issue happens on mobile device: N/A
  • TensorFlow installed from (source or binary): source
  • TensorFlow version (use command below): 2.3.1
  • Python version: 3.6.9
  • Bazel version (if compiling from source): 3.4.1
  • GCC/Compiler version (if compiling from source): Ubuntu 7.5.0-3ubuntu1~18.04
  • CUDA/cuDNN version: N/A
  • GPU model and memory: N/A

Describe the current behavior
I am running Tensorflow 2.3.1 on s390x and //tensorflow/compiler/xla/service/cpu:vectorized_reduce_with_no_vector_registers_test is failing on running bazel tests.
The TC was earlier failing with error: Internal: TargetRegistry::lookupTarget failed: No available targets are compatible with triple "x86_64-pc-linux"

To fix this, I added support for s390x in test_target_triple_helper.h
This is the code change I did:

diff --git a/tensorflow/compiler/xla/service/cpu/test_target_triple_helper.h b/tensorflow/compiler/xla/service/cpu/test_target_triple_helper.h
index 857de4a814..1fb04f821a 100644
--- a/tensorflow/compiler/xla/service/cpu/test_target_triple_helper.h
+++ b/tensorflow/compiler/xla/service/cpu/test_target_triple_helper.h
@@ -21,8 +21,13 @@ limitations under the License.
 static const char kTargetCpuForHost[] = "ppc";
 static const char kTargetTripleForHost[] = "ppc64le-ibm-linux-gnu";
 #else
+#if (defined(__s390x__) && (__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__))
+static const char kTargetCpuForHost[] = "";
+static const char kTargetTripleForHost[] = "systemz-none-linux-gnu"; 
+#else
 static const char kTargetCpuForHost[] = "";
 static const char kTargetTripleForHost[] = "x86_64-pc-linux";
 #endif
+#endif

 #endif

Even after making this change, the test case is still failing. The LLVM is identifying systemz-none-linux-gnu as target but not able to calculate vector_register_byte_size.
The error output looks like this:

==================== Test output for //tensorflow/compiler/xla/service/cpu:vectorized_reduce_with_no_vector_registers_test:
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from CodegenReduceOnArchWithNoVectorRegisters
[ RUN      ] CodegenReduceOnArchWithNoVectorRegisters.Test
2020-11-09 22:19:12.755885: I tensorflow/core/platform/profile_utils/cpu_utils.cc:104] CPU Frequency: 1555500000 Hz
tensorflow/compiler/xla/service/cpu/vectorized_reduce_with_no_vector_registers_test.cc:86: Failure
Expected equality of these values:
  vector_register_byte_size_for_x86_64
    Which is: 0
  16
[  FAILED  ] CodegenReduceOnArchWithNoVectorRegisters.Test (4 ms)
[----------] 1 test from CodegenReduceOnArchWithNoVectorRegisters (4 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (4 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] CodegenReduceOnArchWithNoVectorRegisters.Test

Describe the expected behavior
The LLVM function should return correct values for vector_register_byte_size and the test case should pass.

Other info / logs
I noticed the previous issue regarding XLA Testcases and have already taken the changes done in the PR 39912
@skribm9 skribm9 added the type:bug Bug label Nov 9, 2020
@Saduf2019 Saduf2019 added TF 2.3 Issues related to TF 2.3 comp:xla XLA labels Nov 10, 2020
@Saduf2019 Saduf2019 assigned ymodak and unassigned Saduf2019 Nov 10, 2020
@ymodak ymodak assigned r4nt and unassigned ymodak Nov 10, 2020
@ymodak ymodak added the stat:awaiting tensorflower Status - Awaiting response from tensorflower label Nov 10, 2020
@skribm9
Copy link
Contributor Author
skribm9 commented Jan 4, 2021

Hi @r4nt A very happy new year to you! Could you find anything on this issue?

@sushreebarsa
Copy link
Contributor

@skribm9 It looks like you are using an older Version of Tensorflow . Many bugs have been fixed in the latest version. Could you please execute your code using Latest Version 2.5 and let us know if the issue still persists? Thanks!

@sushreebarsa sushreebarsa added stat:awaiting response Status - Awaiting response from author and removed stat:awaiting tensorflower Status - Awaiting response from tensorflower labels Aug 11, 2021
@rposts
Copy link
Contributor
rposts commented Aug 11, 2021

@sushreebarsa These problems still exist on TF 2.5.0:

exec ${PAGER:-/usr/bin/less} "$0" || exit 1
Executing tests from //tensorflow/compiler/xla/service/cpu:vectorized_reduce_with_no_vector_registers_test
-----------------------------------------------------------------------------
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from CodegenReduceOnArchWithNoVectorRegisters
[ RUN      ] CodegenReduceOnArchWithNoVectorRegisters.Test
2021-08-11 14:25:59.351811: I tensorflow/core/platform/profile_utils/cpu_utils.cc:114] CPU Frequency: 1555500000 Hz
tensorflow/compiler/xla/service/cpu/vectorized_reduce_with_no_vector_registers_test.cc:81: Failure
Expected equality of these values:
  vector_register_byte_size_for_x86_64
    Which is: 0
  16
[  FAILED  ] CodegenReduceOnArchWithNoVectorRegisters.Test (64 ms)
[----------] 1 test from CodegenReduceOnArchWithNoVectorRegisters (64 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (65 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] CodegenReduceOnArchWithNoVectorRegisters.Test

 1 FAILED TEST

@sushreebarsa sushreebarsa added stat:awaiting tensorflower Status - Awaiting response from tensorflower TF 2.5 Issues related to TF 2.5 and removed TF 2.3 Issues related to TF 2.3 stat:awaiting response Status - Awaiting response from author labels Aug 11, 2021
@rposts
Copy link
Contributor
rposts commented Jun 29, 2022

@sushreebarsa Any update on the issue? Problem still exists in TF 2.9.1.

@kun-lu20
Copy link
Contributor

Hi All,

I've run this test case on TF v2.9.1 on s390x and found the cause of GetTargetVectorRegisterByteSize() returning 0 is that the code passes "" as the value of cpu string parameter to LLVM function createTargetMachine(). This works for Intel, but on s390x we need to pass a specific CPU name (such as "z13") to this argument, since only CPU models newer than z13 can support vector facility on s390x.

Also this test case uses triple i686-none-android for subsequent tests. Since this triple has the arch value of x86, the lookupTarget() function from LLVM code works on Intel but will return error on other platforms. For example, on s390x we can use kTargetTripleForHost instead to follow the original code logic and avoid this error.

Looks like this test case was primarily designed for Intel platform. I wonder if we should skip this test case on other platforms or expand it to adapt to all platforms? Any ideas or suggestions would be highly appreciated.

Thanks!

Hi @sushreebarsa,

Could you please take a look? Thank you!

@sushreebarsa sushreebarsa added TF 2.9 Issues found in the TF 2.9 release (or RCs) and removed TF 2.5 Issues related to TF 2.5 labels Aug 17, 2022
@kun-lu20
Copy link
Contributor
kun-lu20 commented Dec 9, 2022

Hi @sushreebarsa ,

Are there any updates on this issue? Thanks!

@rposts
Copy link
Contributor
rposts commented Apr 4, 2023

This can be closed @sushreebarsa as it is x86 specific.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@r4nt r4nt

Labels
comp:xla XLA stat:awaiting tensorflower Status - Awaiting response from tensorflower TF 2.9 Issues found in the TF 2.9 release (or RCs) type:bug Bug
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
7 participants
@r4nt @rposts @ymodak @Saduf2019 @skribm9 @kun-lu20 @sushreebarsa