TensorFlower Gardener <******@****.***> 1775126337 -0700

with JMH as of 2021-04-30.

The async profiler is neat because it does not suffer from the safepoint
bias problem. And because it makes pretty flame graphs!

Let user processes read performance stuff:
```
sudo bash
echo 0 > /proc/sys/kernel/kptr_restrict
echo 1 > /proc/sys/kernel/perf_event_paranoid
exit
```

// Get the DataType of a variable
TF_CAPI_EXPORT extern TF_DataType TF_CheckpointReaderGetVariableDataType(
    TF_CheckpointReader* reader, const char* name);
// Read the shape of a variable and write to `dims`
TF_CAPI_EXPORT extern void TF_CheckpointReaderGetVariableShape(
    TF_CheckpointReader* reader, const char* name, int64_t* dims, int num_dims,
    TF_Status* status);

	defer xioutil.ODirectPoolSmall.Put(bufp)

	for left > 0 {
		// Read expected hash...
		h.Reset()
		n, err := io.ReadFull(r, hashBuf)
		if err != nil {
			// Read's failed for object with right size, file is corrupt.
			return err
		}
		// Subtract hash length..
		left -= int64(n)
		if left < shardSize {
			shardSize = left
		}

		read, err := io.CopyBuffer(h, io.LimitReader(r, shardSize), *bufp)

Shinsuke Sugaya <******@****.***> 1763645653 +0900

Dawid Weiss <******@****.***> 1615370543 +0100

Jesse Wilson <******@****.***> 1761860399 -0400

Jesse Wilson <******@****.***> 1761860399 -0400

Yuri Schimke <******@****.***> 1752914935 +0100

    detail at that link.

4.  Serialized forms of ALL objects are subject to change unless noted
    otherwise. Do not persist these and assume they can be read by a future
    version of the library.

5.  Our classes are not designed to protect against a malicious caller. You
    should not use them for communication between trusted and untrusted code.