experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix for unit test failures introduced by weekly sync - 201026

Browse Source 
The following tests started failing with the weekly merge

```
//tensorflow/python:stateful_random_ops_test                             FAILED in 3 out of 3 in 6.1s
//tensorflow/python:stateful_random_ops_test_gpu                         FAILED in 3 out of 3 in 6.1s
```

The error message for those failures indicates that the failures are due to lack of support for the `Bitcast` operator on the ROCm platform, and this commit adds that missing support. The support was already available for the CUDA platform, so enabling it for ROCm is easy.

example error message

```
======================================================================
ERROR: testSameAsOldRandomOpsGPU (__main__.StatefulRandomOpsTest)
testSameAsOldRandomOpsGPU (__main__.StatefulRandomOpsTest)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/framework/test_util.py", line 1436, in decorated
    return f(self, *args, **kwargs)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/framework/test_util.py", line 1502, in decorated
    return f(self, *args, **kwargs)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops_test.py", line 451, in testSameAsOldRandomOpsGPU
    self._sameAsOldRandomOps(test_util.gpu_device_name(), GPU_FLOATS)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops_test.py", line 430, in _sameAsOldRandomOps
    compare(dtype, old_normal, new_normal)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops_test.py", line 398, in compare
    self.assertAllEqual(run_old(), run_new())
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops_test.py", line 395, in run_new
    return new(dtype, gen)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops_test.py", line 410, in new_normal
    return gen._standard_normal(shape, dtype=dtype)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops.py", line 576, in _standard_normal
    key, counter = self._prepare_key_counter(shape)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/stateful_random_ops.py", line 630, in _prepare_key_counter
    counter = array_ops.bitcast(counter_key[:counter_size], dtypes.uint64)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/ops/gen_array_ops.py", line 598, in bitcast
    _ops.raise_from_not_ok_status(e, name)
  File "/root/.cache/bazel/_bazel_root/efb88f6336d9c4a18216fb94287b8d97/execroot/org_tensorflow/bazel-out/k8-opt/bin/tensorflow/python/stateful_random_ops_test_gpu.runfiles/org_tensorflow/tensorflow/python/framework/ops.py", line 6862, in raise_from_not_ok_status
    six.raise_from(core._status_to_exception(e.code, message), None)
  File "<string>", line 3, in raise_from
tensorflow.python.framework.errors_impl.InvalidArgumentError: Could not satisfy device specification '/job:localhost/replica:0/task:0/device:GPU:0'. enable_soft_placement=0. Supported device types [CPU]. All available devices [/job:localhost/replica:0/task:0/device:GPU:0, /job:localhost/replica:0/task:0/device:CPU:0, /job:localhost/replica:0/task:0/device:CPU:1]. [Op:Bitcast]
```
This commit is contained in:
Deven Desai 2020-10-26 20:22:20 +00:00
parent e91327a5b5
commit 16cab3e8ce
2 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
tensorflow/c/kernels/bitcast_op.cc
View File

@ -148,7 +148,7 @@ void RegisterBitcastOpKernel() {
<< "Error while registering bitcast kernel";
}
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
{
auto* builder = TF_NewKernelBuilder("Bitcast", tensorflow::DEVICE_GPU,
&BitcastOp_Create, &BitcastOp_Compute,

12
tensorflow/c/kernels_test.cc
View File

@ -352,7 +352,7 @@ class DeviceKernelOpTest : public OpsTestBase {
EXPECT_EQ(TF_OK, TF_GetCode(status));
TF_DeleteStatus(status);
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
std::unique_ptr<Device> device(
DeviceFactory::NewDevice(device_name_, {}, "/job:a/replica:0/task:0"));
OpsTestBase::SetDevice(DEVICE_GPU, std::move(device));
@ -361,7 +361,7 @@ class DeviceKernelOpTest : public OpsTestBase {
TF_ASSERT_OK(InitOp());
}
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
const char* device_name_ = tensorflow::DEVICE_GPU;
#else
const char* device_name_ = tensorflow::DEVICE_CPU;
@ -468,7 +468,7 @@ TEST_F(DeviceKernelOpTest, TestAllocateTempSizeOne) {
int64_t dim = 1;
TF_AllocatorAttributes alloc_attrs;
alloc_attrs.struct_size = TF_ALLOCATOR_ATTRIBUTES_STRUCT_SIZE;
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
alloc_attrs.on_host = 0;
#else
alloc_attrs.on_host = 1;
@ -505,7 +505,7 @@ TEST_F(DeviceKernelOpTest, TestAllocateTempEmpty) {
int64_t dim = 0;
TF_AllocatorAttributes alloc_attrs;
alloc_attrs.struct_size = TF_ALLOCATOR_ATTRIBUTES_STRUCT_SIZE;
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
alloc_attrs.on_host = 0;
#else
alloc_attrs.on_host = 1;
@ -538,7 +538,7 @@ TEST_F(DeviceKernelOpTest, TestAllocateTempSize2x3) {
int64_t dim[2] = {2, 3};
TF_AllocatorAttributes alloc_attrs;
alloc_attrs.struct_size = TF_ALLOCATOR_ATTRIBUTES_STRUCT_SIZE;
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
alloc_attrs.on_host = 0;
#else
alloc_attrs.on_host = 1;
@ -646,7 +646,7 @@ template <typename T>
void set_tensor_data(TF_Tensor* tensor, T* values, size_t tensor_size_bytes,
TF_OpKernelContext* ctx) {
T* data = reinterpret_cast<T*>(TF_TensorData(tensor));
#if GOOGLE_CUDA
#if GOOGLE_CUDA || TENSORFLOW_USE_ROCM
OpKernelContext* cc_ctx = reinterpret_cast<OpKernelContext*>(ctx);
cc_ctx->eigen_gpu_device().memcpyHostToDevice(data, values,
tensor_size_bytes);