| Commit message (Collapse) | Author | Age |
|
|
|
| |
PiperOrigin-RevId: 204225504
|
|
|
|
| |
PiperOrigin-RevId: 204223077
|
|
|
|
| |
PiperOrigin-RevId: 204210338
|
|
|
|
| |
PiperOrigin-RevId: 204204001
|
|
|
|
| |
PiperOrigin-RevId: 204199604
|
|\
| |
| |
| | |
PiperOrigin-RevId: 204199229
|
| |
| |
| |
| |
| |
| | |
NeonVectorBatchVectorCwiseProductAccumulate.
PiperOrigin-RevId: 204196998
|
|\ \
| | |
| | |
| | | |
PiperOrigin-RevId: 204196525
|
| | |
| | |
| | |
| | | |
PiperOrigin-RevId: 204194591
|
| | |
| | |
| | |
| | |
| | |
| | | |
implementation-less skeleton.
PiperOrigin-RevId: 204188173
|
|\ \ \
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204186827
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204183931
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204180483
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204177704
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204176713
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204164915
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204162855
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204160548
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
thresholds.
PiperOrigin-RevId: 204159841
|
|\ \ \ \
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 204158338
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
accept complex types.
PiperOrigin-RevId: 204156712
|
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | |
| | | | | |
This makes --copt=-w silence warnings both in the host compiler and in
the CUDA compiler.
PiperOrigin-RevId: 204156508
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 204154837
|
| |_|/ /
|/| | | |
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
unnecessary transitive dependencies.
PiperOrigin-RevId: 204154526
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204153252
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
- Break dependency on tensorflow platform and logging and absl
- Propagate exceptions that capture the TensorFlow lite errors
in a buffer.
PiperOrigin-RevId: 204148724
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204148516
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Add support to XlaOpKernelContext for accessing inputs by name .
PiperOrigin-RevId: 204148428
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204146100
|
| | | |
| | | |
| | | |
| | | | |
PiperOrigin-RevId: 204144214
|
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | |
| | | | |
Parameters to while nodes, for instance, do alias with temps. I haven't
directly encountered this bug on HEAD, but a proper fix for b/111116907 exposes
this.
PiperOrigin-RevId: 204143492
|
|\ \ \ \
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 204142076
|
| | | | |
| | | | |
| | | | |
| | | | | |
PiperOrigin-RevId: 204140674
|
|\ \ \ \ \
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204139191
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
Less visual noise.
PiperOrigin-RevId: 204139183
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204137901
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
- Fix typos.
- Clarify comments.
- Reduce nesting in a few places.
- Add asserts that this code is dealing with specifically a loop fusion.
- Rename some functions. In particular, it's confusing to have a
function with a generic name like EmitCodeWithBoundCheck that actually
is specialized to a tiled implementation.
- Remove statement expression (GCC language extension), replacing it
with an IIFE.
- Don't refer to shared-memory tile space as "buffer" without other
qualifying words, since that's ambiguous with what XLA refers to as a
"buffer".
- Use llvm::cast instead of static_cast.
- Comply with style guide naming rules for compile-time constants
(kFoo).
- Use c_accumulate instead of std::accumulate.
- Put std::function parameter at the end of the param list. This lets
us cleanly embed the lambda into the call because of how clang-format
formats such calls. (I think this one is possibly the most helpful
change in this patch, as it suddenly makes clear to me the way that we
use two calls to emit_tiled_elemental_code_with_bounds_check to emit
the code.)
PiperOrigin-RevId: 204134102
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204132868
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204131441
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204128363
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204114884
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204114708
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
We should use DCHECK_GE instead of DCHECK_GT.
PiperOrigin-RevId: 204107900
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
After this change, to build TensorFlow with GPU support on Windows, you just need to follow the same procedure as Linux.
Also re-enable remote cache since the bug in b/111106393 has been fixed by the new wrapper script.
The clean-up of the wrapper scripts also made the build around 5 mins faster than using Bazel's msvc wrapper script.
PiperOrigin-RevId: 204105368
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204105366
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
So far, the behavior for out of bounds cases is undefined as
long as it does not crash. Clamping is cheaper because we don't
need a URem operation, and for DynamicSlice we already do that.
PiperOrigin-RevId: 204096995
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
being non-numeric types such as strings.
PiperOrigin-RevId: 204087922
|
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
TopKV2 is implemented by using XlaSort, and XlaSort is not implemented yet on CPU and GPU.
PiperOrigin-RevId: 204082873
|
| | | | | |
| | | | | |
| | | | | |
| | | | | | |
PiperOrigin-RevId: 204071680
|