Commit Graph

2863 Commits (0ca88028cd607eaba138704eb14b57b8a9efc3f0)
 

Author SHA1 Message Date
Sean Silva 1b48d0d80b Remove the present tcp.island.
The idea was half-baked and after some deep thought felt like a solution
looking for a problem. What we had here (and is removed in this patch)
just wasn't pulling its weight.

I cannot think of anything we would want to do with tcp.island as it is
removed here beyond just sinking and merging them within a basic block,
such that the witness argument is kind of pointless (only matters for
hoisting).

TCP compute ops like tcp.add and tcp.broadcast_to have the strong
invariant of "pure or undefined behavior", which means they are always
safe to sink. The island concept as removed here conferred no benefit.

Also, I'll note that "islands" are a trick you can only play once in a
system (unless they strictly nest). I have some early-stage thoughs on
having an island concept that helps with modeling tensor shapes
robustly which seems promising (the island would serve a similar role as
tie_shape).
2020-05-14 15:19:37 -07:00
Sean Silva 98a38c3527 Add some anonymous namespaces.
This brings this code in line with the LLVM style guide and avoids
potential ODR issues.
2020-05-14 15:02:46 -07:00
Sean Silva 889fe0d6c2 Tidy up test/E2E
- Make rank1.mlir be the new "basic.mlir", as it is really the simplest
  case.
- Move basic.mlir to mixed-ranks.mlir
- Delete starting-from-linalg.mlir, it wasn't really useful anymore.
2020-05-14 14:59:55 -07:00
Sean Silva eaeb4011e6 Lower !shape.shape to SSA values.
This uses an approach inspired by what is done in IREE. See comments on
LowerRankedShapes.cpp for how it works.

The basic gist is that we have an op that creates a !shape.shape from a
set of SSA values representing the extents, and then iteratively replace
any op producing a !shape.shape with instances of that op.
2020-05-13 17:20:23 -07:00
Sean Silva ef25428fe3 Add lowering from linalg to loops.
This also adds a small pass to clean up the `dim` ops that linalg
introduces. For now, it only has a trivial pattern that looks for a
`tcp.alloc_memref(%shape)` op to get the shape as we currently have an
invariant that all memrefs are the result of such ops.

But eventually this will need to look through view ops and any other
shape-ish stuff that linalg introduces as it lowers to loops, along with
any slicing ops introduced by buffer allocation.
2020-05-11 18:54:52 -07:00
Sean Silva f525d4dbcf Add custom assembly format for tcp.alloc_memref/tcp.get_extent
This makes the IR a bit easier to scan.
2020-05-11 15:28:34 -07:00
Sean Silva 174ab19c5f Add some cleanup passes.
This makes the IR more presentable before going to the next phase of
lowering (ops on memref/buffers -> LLVM ops).
2020-05-11 15:28:21 -07:00
Sean Silva 83db558db9 Update llvm-project to 310d32cb80a611e6384a921e85607fea05841f26 2020-05-11 15:12:47 -07:00
Sean Silva 53c17dbed9 "Finish" tensor -> memref conversion.
There's a lot of details to flesh out here, but the basic approach seems
promising (see comments in createE2ELoweringPipeline).

This approach will be put to the test when we try to do our first
fusions since that tickles some of the nasty phase ordering issues
involved here.

But we're not there yet.
2020-05-11 15:00:12 -07:00
Sean Silva fec2ee0072 Avoid introducing DimOp's in LowerBroadcastToToLoops.
This makes sure we stay resonably canonically using the shape machinery.
(In fact, DimOp should probably be in the shape dialect since it hides a
`shape.shape_of` call)
2020-05-11 13:12:16 -07:00
Sean Silva 1ef8b91a95 Add -DCMAKE_EXPORT_COMPILE_COMMANDS=TRUE
This allows VSCode (and presumably other IDE's) to pick up a
`tools/cmake_configure.sh`'ed build directory and have intellisense.
2020-05-11 12:58:42 -07:00
Stella Laurenzo 950ba12426 Bump llvm-project to 3af85fa8f06220b43f03f26de216a67be4568fe7. 2020-05-08 20:42:40 -07:00
Sean Silva e29aef855b Initial TCF/TCP E2E seed.
Very much WIP.

This is enough to get tcf.add down to approximately the "linalg.generic
on buffers" level of abstraction. (but there are nuances)
2020-05-08 20:20:41 -07:00
Sean Silva f394e12d86 Add tools/test_all.sh
We should look into having a `ninja check-npcomp` that runs everything
with lit so that we get decent test multithreading.
We can look to how LLVM does its gtest tests ("unittests") for
inspiration.
2020-05-08 18:13:19 -07:00
Stella Laurenzo f2985e0901 Add implicit constant capture.
We want more sophisticated capture later, but this allows basics to function.
2020-05-08 17:55:02 -07:00
Stella Laurenzo 8ae71a9551 Add MLIRContext.dense_elements_attr to create an attribute from a python buffer/array. 2020-05-08 17:36:07 -07:00
Stella Laurenzo a91b0bfbe1 Add numpy.get_slice op and wire it up to the tracer. 2020-05-08 16:04:58 -07:00
Stella Laurenzo db0b0ef1b2 Switch sample tracer to emit builtin_ufunc globals instead of the impl versions. 2020-05-08 14:35:13 -07:00
Stella Laurenzo 497f1de5a2 Add env var to enable experimental numpy array function handling. 2020-05-08 14:21:43 -07:00
Stella Laurenzo 0092b912ab Update all python imports to be absolute and use a .env file to set the path correctly.
This makes things just work for debugging in VSCode.
2020-05-06 23:25:04 -07:00
Stella Laurenzo 3611958b11 Move python native library to python_native/_npcomp...so.
This allows binary and source packages to exist at different physical paths.
2020-05-06 22:44:12 -07:00
Stella Laurenzo 680e11ae62 Rename basicpy_None_type to basicpy_NoneType to match native spelling.
* Also adds Basicpy.py to tests.
2020-05-06 19:07:50 -07:00
Stella Laurenzo 644d9fb0d3 Remove spammy warnings and filecheck info in run_tests.
* This suppresses a warning that arises from using "-m" to launch a module contained in a package that arranges modules via __init__.py. It seems irrelevant to the use case of running doctests.
2020-05-06 18:48:12 -07:00
Stella Laurenzo 6b7c913e0b Add DialectHelper for Basicpy dialect.
* Involved native code for the types and slot_object_get ops.
2020-05-06 18:26:03 -07:00
Stella Laurenzo 4ebf972503 Merge ir.Ops and ir.Types into ir.DialectHelper.
This will aid in managing hierarchies of custom dialect helpers.
2020-05-06 18:26:03 -07:00
Sean Silva aa9ffc3a11 Delete npcomp.edsc_test from python/run_tests.py 2020-05-06 18:24:44 -07:00
Stella Laurenzo 714bc01c02 Bump llvm version to 0c4aab27b3da05dd1b0c0c39472525325fda5e23.
* Fixes some api-change breakages.
* Deletes edsc module now that the IR bindings are working (and since it broke with this update).
2020-05-05 21:00:49 -07:00
Stella Laurenzo bc5ef81d68 Add basicpy.SlotObject type and ops to create/index into it.
* This is intended to provide low-level modeling for built-in objects.
* It is now possible to trace slice tuples (which are tuples of NoneType|EllipsisType|SlotObjectType<slice, ...>).
2020-05-05 18:16:01 -07:00
Stella Laurenzo bfd5fedba7 Add central registration for type ranges. 2020-05-05 14:16:39 -07:00
Stella Laurenzo f8cedd8656 Add known good llvm-project commit hash to readme. 2020-05-05 14:00:06 -07:00
Stella Laurenzo 9a0f498774 Add new gist to the design notes. 2020-05-05 13:49:05 -07:00
Stella Laurenzo 502ef8f195 Create skeleton for 'Basicpy' dialect.
* It is time to start adding more python mechanisms.
* Running into this for materializing slice() objects.
2020-05-04 17:48:02 -07:00
Stella Laurenzo ebb5bcf6af Handle np.transpose() and ndarray.T shortcut.
* Just the form without explicit permutation for now.
2020-05-04 16:20:36 -07:00
Stella Laurenzo a5f755d406 Implement __array_func__ hook and use it to trace np.dot.
* Creates an abstraction/registry around emitters (intended to generalize to AST compilation as well).
* Reworks ufuncs to use the same mechanism as array funcs.
* Adds the numpy.dot op.
2020-05-04 15:47:01 -07:00
Stella Laurenzo 1f54838d2e Add hook for __array_function__ and (failing) np.dot sample. 2020-05-03 13:39:30 -07:00
Stella Laurenzo a38a1e2850 Cleanup python namespace a bit for standalone use. 2020-05-02 21:54:13 -07:00
Stella Laurenzo 1f136f9dba Update readme 2020-05-02 19:54:49 -07:00
Stella Laurenzo c89a35f97f Rework the poc tracer to be structured how intended. 2020-05-02 19:52:21 -07:00
Stella Laurenzo 0805013716 Rename ufunc_call op builder for consistency. 2020-05-01 19:06:10 -07:00
Stella Laurenzo 78a8e6ec9e Add enough python bindings to build functions and ufunc calls. 2020-05-01 18:44:06 -07:00
Stella Laurenzo ba0c96b51a Add python side Numpy dialect wrapper. 2020-05-01 10:38:52 -07:00
Stella Laurenzo 23a9ffaabe Add wrappers for block and operation iteration.
I don't technically need this now but adding while the train of thought is fresh.
2020-05-01 10:16:19 -07:00
Stella Laurenzo c8740fd866 Start splitting Py* types into a header so that further C++ interop can be built. 2020-04-30 19:23:18 -07:00
Stella Laurenzo ec0f6b4b22 Add MLIRContext and ModuleOp python bindings with asm parse/print and diagnostics. 2020-04-30 17:14:03 -07:00
Stella Laurenzo 67d38db1e2 Start defining new IR bindings and cleanup python init. 2020-04-30 16:00:00 -07:00
Stella Laurenzo d3632af675 Add !numpy.any_dtype dialect type. 2020-04-29 18:20:42 -07:00
Stella Laurenzo b4425fe1d2 Add numpy.ufunc_call op. 2020-04-29 17:49:56 -07:00
Stella Laurenzo c4a192d5c9 Rename from npcomp::NUMPY to NPCOMP::numpy to follow IREE convention. 2020-04-29 17:10:10 -07:00
Stella Laurenzo f174c0557b Cleanup build instructions/scripts 2020-04-29 17:05:45 -07:00
Stella Laurenzo 03cf785bae Remove linker detection 2020-04-29 16:33:48 -07:00