| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 214853860
|
|
|
|
|
|
|
|
|
|
|
| |
This change switches `tf.contrib.data.Optional` to use a `Structure` class to represent
the structure of its value, instead of `output_types`, `output_shapes`, and `output_classes` properties. It adds support for nesting `Optional` objects and representing their structure.
This change also makes a modification to the `Structure` class: `Structure.is_compatible_with(x)` now takes another `Structure` as the `x` argument, instead of a value. This makes it easier to work with nested structures (where we might not have a value readily available), and better matches the interface of other `is_compatible_with()` methods (e.g. in `tf.TensorShape` and `tf.DType`).
Finally, in the process of making this change, I observed possible crash-failures when a DT_VARIANT tensor containing another DT_VARIANT tensor is copied between CPU and GPU. This change "fixes" the immediate problem by raising an UnimplementedError, but more work will be necessary to support the full range of use cases.
PiperOrigin-RevId: 214198993
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
1. Change Variant Decode to accept VariantTensorData (non-ref).
This should allow some optimization in the future.
In the meantime it means removing the variant.h include from tensor.h, since
variant_encode_decode.h now relies on tensor.h and variant.h now relies on that.
It also means we found a bunch of places where tensor.proto.h, variant.h, and
mutex.h were being imported through tensor.h (along with a bunch of other crap);
so now we directly import them in order to compile.
2. Move Variant registry to use TypeIndex instead of a TypeName string; this should
speed up registry lookups.
PiperOrigin-RevId: 212478896
|
|
|
|
| |
PiperOrigin-RevId: 207490563
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
more than one device-to-device copy stream per GPU device.
This is an experimental feature that will have no effect unless
copy operations explicitly request a stream other than 0, which
currently does not occur anywhere in a standard build.
Eventually it may be of benefit in the presence of multiple
bi-directional concurrent data copies.
PiperOrigin-RevId: 202354513
|
|
|
|
| |
PiperOrigin-RevId: 194768567
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The conditional expression only has ONE fixed value category. Because in the code as written the two operands are of different value categories, the result is in fact a prvalue, i.e. a copy. This seems unintended, and we should simply preserve the existing lvalue.
If we do want to allow moving, we need multiple statements:
if (num == 1) {
f(std::move(copier));
} else {
f(copier);
}
PiperOrigin-RevId: 176414503
|
|
|
|
| |
PiperOrigin-RevId: 175546097
|
|
|
|
|
|
|
|
|
|
| |
Adds a test in the variant_op_copy_test.
Modifies the base GPUDevice to use this registry if it sees a singleton variant.
Modifies the rendezvous manager to do the same.
PiperOrigin-RevId: 170908757
|
|
|
|
| |
Change: 153861629
|
|
|
|
| |
Change: 136164533
|
|
|
|
| |
Change: 135518672
|
|
|
|
| |
Change: 131891101
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
done by models distributed across many devices. A small
microbenchmark model that runs two banks (A and B) of 30 nodes with a
30x30 full shuffle between them, where each of the nodes in A and in B
run with one node on each of the 30 devices (so 30*29+30+30, or ~930
separate RPCs) was showing ~111,000 allocations per iteration of the graph.
With the changes here, this is now down to ~64,300 allocations per iteration.
Changes include:
o DeviceContext::CopyDeviceTensorToCPU and related helper routines:
use StringPiece instead of const string& for the tensor name (avoids
creating a string in some cases where the caller only has a
StringPiece available).
o Change some Rendezvous and BaseRemoteRendezvous interfaces to
take a 'const Rendezvous::ParsedKey& key', rather than 'const string& key'.
In many cases, the callers were already having to parse the key
into a ParsedKey, and so we were doing the parsing multiple times at
different levels as we processed receiving or sending of a tensor. This
reduces the number of times that we parse a key as it flows from a Send
node through to a Recv node on another worker.
o Changed Rendezvous::ParsedKey so that it makes a copy of the underlying
full key, and then uses StringPiece objects to point into this copy for
the src_device, dst_device, and edge_name pieces. This turns 3 string
allocations into 1 per Rendezvous::ParseKey call.
o Added new StringPiece Rendezvous::ParsedKey::FullKey() accessor to
return a StringPiece for the underlying full key, and used that in a
few places (mostly logging) where that is useful.
o In many places, used std::move(function_variable) when assigning to
an instance variable. This eliminates a very large number of excess
std::function allocations/initializations (~56000 of the baseline
allocations were related to std::function setup or cloning, and this
is now down to ~11000 after this cl).
o In the RPC-based remote workers (StubbyRemoteWorker and
GrpcRemoteWorker), changed the code path in RecvTensorAsync to avoid
creation of a std::function with 6 arguments unless necessary. There
are three cases now handled separately:
(a) We're not logging, and we didn't make a copy of the request that we
need to free: just use the passed in 'StatusCallback done' object
directly, without creating a wrapper std::function object at all
(b) We're not logging, but we made a copy of the request that we
need to free: we create a simple wrapper std::function that
invokes the passed in 'done' callback, and then frees the
req_copy request copy object.
(c) We're logging: we create the std::function object with all the
necessary state to log when the recv has finished.
o Changed DeviceMgr::LookupDevice to take a StringPiece, rather than a
const string&, and changed the hash table to use StringPiece keys.
This allows clients that just have a StringPiece device name in their
hand to avoid a string creation to lookup the Device* object.
o Changed ExecutorState to use a specialized TaggedNodeReadyQueue that
internally uses a gtl::InlinedVector<TaggedNode, 16>, rather than
using a std::deque<TaggedNode> for keeping track of nodes ready to
execute. This is faster because it avoids allocations entirely if the
ready node queue doesn't get bigger than 16, and inlined vectors are
generally faster than std::deque, at a minor risk of using more memory
if this queue grows to very large numbers of ready nodes (mostly imaginable
only in pathological graphs).
o In ExecutorState::Process, allocated a single ExecutorState::AsyncState
object to keep track of all the state we need to preserve for an asynchronously
executed node, rather than keeping this state implicitly via a very large
number of arguments to a lamda function.
o Added new atomic std::atomic<bool> status_is_ok_ in
BaseRemoteRendezvous. This allows us to avoid acquiring the lock when
we just want to check if the status is non-OK in
BaseRemoteRendezvous::Send and BaseRemoteRendezvous::ValidateDevices.
o In GraphMgr::RunAllDone, changed assignment of args.runner to avoid
one extra level of std::function indirection (binding the function directly
to the ThreadPool::Schedule routine, rather than creating an intermediate
lambda function that invokes this inside the body of the lambda.
o Added freelist of RpcRecvTensorCall objects in
third_party/tensorflow/core/distributed_runtime/rpc/rpc_rendezvous_mgr.cc
o Changed third_party/tensorflow/core/framework/rendezvous.cc to keep the
hashtable of Item* objects keyed by uint64 (hash of the tensor name), rather
than the full-string tensor name. Collisions in the 64-bit hash space
should basically never happen.
o Sped up DeviceNameUtils::ParseFullName by optimizing for the common
ordering of parts of /job, /replica, /task, /device. The parsing code
was general enough to handle any order, but did so by comparing the
prefixes 4, 3, 2, and 1 times, respectively, rather than 1, 1, 1, and 1 times.
o Sped up DeviceNameUtils::SplitDeviceName to avoid extra string copies.
Change: 125991891
|
|
|
|
| |
Change: 123900938
|
|
|
|
| |
Change: 123860431
|
|
|
|
|
|
| |
from 5.5mb to 3.2mb (compressed).
Change: 113369407
|
|
|
|
| |
Change: 112903292
|
|
* The "core_cpu_internal" build target no longer includes files from the
common_runtime/gpu/ directory.
* tensorflow/core internal targets instead can get access to those headers via
the "gpu_runtime" target.
* The class "CopyTensor" is introduced. It lives in common_runtime/
but supports registration of copy functions so the "gpu_runtime"
target can add a GPU->GPU copy ability if it is linked in.
This registration should make it easier to add more device types
in the future.
* The "core_cpu" and "core_cpu_internal" build targets no longer
reference GPUUtil::CopyViaDMA; rendezvous_mgr uses CopyTensor
instead.
Also the "copy_tensor" build target was not needed.
Change: 112821119
|