Ideally this is upstreamed as an std.list or something.

…dims.

Cleanup as part of iree-org#4675 should move this pass to a shared place so it
can be run both in flow and hal.

This lives under a new iree/compiler/Bindings/ path that we can use
for all binding-specific transforms/ops/etc. This will make it easier to
cross-target for example HLO->tflite bindings or tflite->some future IREE
bindings. If the SIP stuff sticks around it could move to Bindings/SIP/ or
something that can be conditionally enabled for only when it's needed.
Progress on iree-org#3975 (still needs SIP workaround, but fns generated).

This creates the entry point functions exported for the tflite bindings
to use as their main function.
Fixes iree-org#3974.
Progress on iree-org#3978 and iree-org#3972 (plumbing in place, just needs attr checks).

Dynamic shapes are blocked on iree-org#3975 - SIP is munging the function
signatures and needs a bypass.
Fixes iree-org#3948.

- Move HLO -> HLO & Legalize workflow before LinalgOnTensors -> HLO
- Run canonicalization after HLO -> HLO.
- Create xla_ops tests.

…anup

Only the timeline semaphore emulation uses this so I'm moving it to a
place where it will be easy to delete (and attribute binary size to).

It'll come back as part of iree-org#3909.

This cleans up a long time #yolo that was reusing the base type across both
systems. The type can still use some cleanup/be setup on
iree_vm_ref_object_t, etc but this at least splits the dependency.

The last user of it; iree-org#4369 is tracking replacing DynamicSymbols, the
handle types could be ported to unique_ptr semantics, and then the rest is
the timeline semaphores fallback.

This will allow iree/hal/local/arena.h to be swapped in its place.

There's some goo here for internal support that I forget if is still used.

…ompiling the Android Java bindings. (iree-org#4705)

Add Gradle tasks to configure and make host tools with CMake before compiling the Android Java bindings.

Progress on iree-org#4617

Fixes iree-org#4690

Progress on iree-org#3974. Unblocks iree-org#3972/iree-org#3975/iree-org#3978.

…ree-org#4699)

This verifies that the tool is functioning - we in general shouldn't be
checking in any tflite flatbuffers (ew) but we need something as a
reference.

Following iree-org#4702

Should just port these to pure C, or delete them. I like them though D:

*  bde47d5 Synchronize submodules with LLVM at llvm/llvm-project@ea5b75de4999
* 6200d0b Merge pull request iree-org#4707 from not-jenni:main-to-google
* 5f63a33 Integrate LLVM at llvm/llvm-project@bbb755540385
* a118dda Synchronize submodules with LLVM at llvm/llvm-project@e635feb15a91
* 149caee Integrate LLVM at llvm/llvm-project@e635feb15a91
* 7f8aeff Integrate LLVM at llvm/llvm-project@54afcade3bbc
* fe858c4 Integrate LLVM at llvm/llvm-project@fae6d129dac2
* 2fdd400 Integrate LLVM at llvm/llvm-project@2668714747c5

…sPass (iree-org#4714)

InvalidArgumentErrorBuilder is all that remains.

We'll expose the concept through dedicated APIs where it makes sense
(like importing/exporting platform handles/etc) vs. allowing users to
poke right into these things.

This moves casting from the bytecode dispatcher to the shared
implementation which now only consumes and return int32_t. This
simplifies the printer in the CModuleTarget (as it eliminates the need
to support unsigned int) and avoids implicits casts.

Additional arguments were added to `translateFromMLIRToVM()`, but not to the call within `translateFromMLIRToVMCModule()`.

This should be usable by build cops as-is, but the next step would be
integrating it to run on commits to the google branch to automatically
fix them up.