Retry landing https://github.com/llvm/llvm-project/pull/153373
## Major changes from previous attempt
- remove the test in CAPI because no existing tests in CAPI deal with
sanitizer exemptions
- update `mlir/docs/Dialects/GPU.md` to reflect the new behavior: load
GPU binary in global ctors, instead of loading them at call site.
- skip the test on Aarch64 since we have an issue with initialization there
---------
Co-authored-by: Mehdi Amini <joker.eph@gmail.com>
This PR implements "automatic" location inference in the bindings. The
way it works is it walks the frame stack collecting source locations
(Python captures these in the frame itself). It is inspired by JAX's
[implementation](523ddcfbca/jax/_src/interpreters/mlir.py (L462))
but moves the frame stack traversal into the bindings for better
performance.
The system supports registering "included" and "excluded" filenames;
frames originating from functions in included filenames **will not** be
filtered and frames originating from functions in excluded filenames
**will** be filtered (in that order). This allows excluding all the
generated `*_ops_gen.py` files.
The system is also "toggleable" and off by default to save people who
have their own systems (such as JAX) from the added cost.
Note, the system stores the entire stacktrace (subject to
`locTracebackFramesLimit`) in the `Location` using specifically a
`CallSiteLoc`. This can be useful for profiling tools (flamegraphs
etc.).
Shoutout to the folks at JAX for coming up with a good system.
---------
Co-authored-by: Jacques Pienaar <jpienaar@google.com>
This patch specializes the Python bindings for ForallOp and
InParallelOp, similar to the existing one for ForOp. These bindings
create the regions and blocks properly and expose some additional
helpers.
- Introduces a `large_resource_limit` parameter across Python bindings,
enabling the eliding of resource strings exceeding a specified character
limit during IR printing.
- To maintain backward compatibilty, when using `operation.print()` API,
if `large_resource_limit` is None and the `large_elements_limit` is set,
the later will be used to elide the resource string as well. This change
was introduced by https://github.com/llvm/llvm-project/pull/125738.
- For printing using pass manager, the `large_resource_limit` and
`large_elements_limit` are completely independent of each other.
A new transform op to represent that an attribute is to be chosen from a
set of alternatives and that this choice is made available as a
`!transform.param`. When a `selected` argument is provided, the op's
`apply()` semantics is that of just making this selected attribute
available as the result. When `selected` is not provided, `apply()`
complains that nothing has resolved the non-determinism that the op is
representing.
The motivation is to avoid having to negate `isDynamic*` checks, avoid
double negations, and allow for `ShapedType::isStaticDim` to be used in
ADT functions without having to wrap it in a lambda performing the
negation.
Also add the new functions to C and Python bindings.
RFC:
https://discourse.llvm.org/t/rfc-deprecate-linalg-elemwise-unary-and-elemwise-binary/87144
Remove the two operations and fix the tests by:
* Cleaning simple operation tests of the old ops
* Changing `linalg.elemwise_{u|bi}nary` with `linalg.{exp|add}` on
transform tests
* Changing some of the tests with `linalg.elementwise` instead, to
broaden test coverage
* Surgically removing the `elemwise_*` part in the Python tests
* Update MLIR transform examples (text and tests) with
`linalg.elementwise` instead
Nothing else changed.
Removes the Debug... prefix on the ops in tablegen, in line with pretty
much all other Transform-dialect extension ops. This means that the ops
in Python look like
`debug.EmitParamAsRemarkOp`/`debug.emit_param_as_remark` instead of
`debug.DebugEmitParamAsRemarkOp`/`debug.debug_emit_param_as_remark`.
Interpret an option value with multiple values, either in the form of an
`ArrayAttr` (either static or passed through a param) or as the multiple
attrs associated to a param, as a comma-separated list, i.e. as a
ListOption on a pass.
Improve ApplyRegisteredPassOp's support for taking options by taking
them as a dict (vs a list of string-valued key-value pairs).
Values of options are provided as either static attributes or as params
(which pass in attributes at interpreter runtime). In either case, the
keys and value attributes are converted to strings and a single
options-string, in the format used on the commandline, is constructed to
pass to the `addToPipeline`-pass API.
This PR output debug message to assertion to help debug user python
code. Will print out more friendly information
```
> assert isinstance(arg, _cext.ir.Value), f"expects Value, got {type(arg)}"
E AssertionError: expected Value, got <class 'UserDefinedClass'>
```
…_reduce_matmul.
This patch exposes broadcast and transpose semantics on
'batch_reduce_matmul'. This is the last one in continuation of other two
variant of matmul ops.
The broadcast and transpose semantic are as follows:
Broadcast and Transpose semantics can be appiled by specifying the
explicit attribute 'indexing_maps' as shown below. This is a list
attribute, so must include maps for all arguments if specified.
Example Transpose:
```
linalg.batch_reduce_matmul indexing_maps = [
affine_map<(d0, d1, d2, d3) -> (d0, d3, d1)>, // transpose
affine_map<(d0, d1, d2, d3) -> (d0, d3, d2)>,
affine_map<(d0, d1, d2, d3) -> (d1, d2)>
]
ins(%arg0, %arg1 : memref<2x5x3xf32>,memref<2x5x7xf32>)
outs(%arg2: memref<3x7xf32>)
```
Example Broadcast:
```
linalg.batch_reduce_matmul indexing_maps = [
affine_map<(d0, d1, d2, d3) -> (d3)>, // broadcast
affine_map<(d0, d1, d2, d3) -> (d0, d3, d2)>,
affine_map<(d0, d1, d2, d3) -> (d1, d2)>
]
ins(%arg0, %arg1 : memref<5xf32>, memref<2x5x7xf32>)
outs(%arg2: memref<3x7xf32>)
```
Example Broadcast and Transpose:
```
linalg.batch_reduce_matmul indexing_maps = [
affine_map<(d0, d1, d2, d3) -> (d1, d3)>, // broadcast
affine_map<(d0, d1, d2, d3) -> (d0, d2, d3)>, // transpose
affine_map<(d0, d1, d2, d3) -> (d1, d2)>
]
ins(%arg0, %arg1 : memref<3x5xf32>, memref<2x7x5xf32>)
outs(%arg2: memref<3x7xf32>)
```
RFCs and related PR:
https://discourse.llvm.org/t/rfc-linalg-opdsl-constant-list-attribute-definition/80149https://discourse.llvm.org/t/rfc-op-explosion-in-linalg/82863https://discourse.llvm.org/t/rfc-mlir-linalg-operation-tree/83586https://github.com/llvm/llvm-project/pull/115319https://github.com/llvm/llvm-project/pull/122275
This is an implementation for [RFC: Supporting Sub-Channel Quantization
in
MLIR](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694).
In order to make the review process easier, the PR has been divided into
the following commit labels:
1. **Add implementation for sub-channel type:** Includes the class
design for `UniformQuantizedSubChannelType`, printer/parser and bytecode
read/write support. The existing types (per-tensor and per-axis) are
unaltered.
2. **Add implementation for sub-channel type:** Lowering of
`quant.qcast` and `quant.dcast` operations to Linalg operations.
3. **Adding C/Python Apis:** We first define he C-APIs and build the
Python-APIs on top of those.
4. **Add pass to normalize generic ....:** This pass normalizes
sub-channel quantized types to per-tensor per-axis types, if possible.
A design note:
- **Explicitly storing the `quantized_dimensions`, even when they can be
derived for ranked tensor.**
While it's possible to infer quantized dimensions from the static shape
of the scales (or zero-points) tensor for ranked
data tensors
([ref](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694/3)
for background), there are cases where this can lead to ambiguity and
issues with round-tripping.
```
Consider the example: tensor<2x4x!quant.uniform<i8:f32:{0:2, 0:2}, {{s00:z00, s01:z01}}>>
```
The shape of the scales tensor is [1, 2], which might suggest that only
axis 1 is quantized. While this inference is technically correct, as the
block size for axis 0 is a degenerate case (equal to the dimension
size), it can cause problems with round-tripping. Therefore, even for
ranked tensors, we are explicitly storing the quantized dimensions.
Suggestions welcome!
PS: I understand that the upcoming holidays may impact your schedule, so
please take your time with the review. There's no rush.
* `PyRegionList` is now sliceable. The dialect bindings generator seems
to assume it is sliceable already (!), yet accessing e.g. `cases` on
`scf.IndexedSwitchOp` raises a `TypeError` at runtime.
* `PyBlockList` and `PyOperationList` support negative indexing. It is
common for containers to do that in Python, and most container in the
MLIR Python bindings already allow the index to be negative.
In some projects like JAX ir.Context are used with disabled multi-threading to avoid
caching multiple threading pools:
623865fe95/jax/_src/interpreters/mlir.py (L606-L611)
However, when context has enabled multithreading it also uses locks on
the StorageUniquers and this can be helpful to avoid data races in the
multi-threaded execution (for example with free-threaded cpython,
https://github.com/jax-ml/jax/issues/26272).
With this PR user can enable the multi-threading: 1) enables additional
locking and 2) set a shared threading pool such that cached contexts can
have one global pool.
It is hoped that the Ops, Types, and Attribute of the NVGPU dialect can
be defined in separate files.If downstream projects extend NVGPU and
define other Ops, the types and attributes will be used.This PR was
raised to avoid including the definition of NVGPU Ops.
The current `write_bytecode` implementation necessarily requires the
serialized module to be duplicated in memory when the python `bytes`
object is created and sent over the binding. For modules with large
resources, we may want to avoid this in-memory copy by serializing
directly to a file instead of sending bytes across the boundary.
This PR https://github.com/llvm/llvm-project/pull/123902 broke python
bindings for `tensor.pack`/`unpack`. This PR fixes that. It also
1. adds convenience wrappers for pack/unpack
2. cleans up matmul-like ops in the linalg bindings
3. fixes linalg docs missing pack/unpack
As linalg.batch_matmul has been moved into tablegen from OpDSL, its
derived python wrapper no longer exist.This patch adds the required
python wrapper.
Also refactors the BatchmatmulOp printer to make it consistent with its
parser.
For extremely large models, it may be inefficient to load the model into
memory in Python prior to passing it to the MLIR C APIs for
deserialization. This change adds an API to parse a ModuleOp directly
from a file path.
Re-lands
[4e14b8a](4e14b8afb4).
Now that linalg.matmul is in tablegen, "hand write" the Python wrapper
that OpDSL used to derive. Similarly, add a Python wrapper for the new
linalg.contract op.
Required following misc. fixes:
1) make linalg.matmul's parsing and printing consistent w.r.t. whether
indexing_maps occurs before or after operands, i.e. per the tests cases
it comes _before_.
2) tablegen for linalg.contract did not state it accepted an optional
cast attr.
3) In ODS's C++-generating code, expand partial support for `$_builder`
access in `Attr::defaultValue` to full support. This enables access to
the current `MlirContext` when constructing the default value (as is
required when the default value consists of affine maps).
Goals:
1. To add syntax and semantic to 'batch_matmul' without changing any of
the existing syntax expectations for current usage. batch_matmul is
still just batch_matmul.
2. Move the definition of batch_matmul from linalg OpDsl to tablegen ODS
infra.
Scope of this patch:
To expose broadcast and transpose semantics on the 'batch_matmul'.
The broadcast and transpose semantic are as follows:
By default, 'linalg.batch_matmul' behavior will remain as is. Broadcast
and Transpose semantics can be applied by specifying the explicit
attribute 'indexing_maps' as shown below. This is a list attribute, so
the list must include all the maps if specified.
Example Transpose:
```
linalg.batch_matmul indexing_maps = [
affine_map< (d0, d1, d2, d3) -> (d0, d3, d1)>, //transpose
affine_map< (d0, d1, d2, d3) -> (d0, d3, d2)>,
affine_map< (d0, d1, d2, d3) -> (d0, d1, d2)>
]
ins (%arg0, %arg1: memref<2x5x3xf32>,memref<2x5x7xf32>)
outs (%arg2: memref<2x3x7xf32>)
```
Example Broadcast:
```
linalg.batch_matmul indexing_maps = [
affine_map< (d0, d1, d2, d3) -> (d3)>, //broadcast
affine_map< (d0, d1, d2, d3) -> (d0, d3, d2)>,
affine_map< (d0, d1, d2, d3) -> (d0, d1, d2)>
]
ins (%arg0, %arg1: memref<5xf32>,memref<2x5x7xf32>)
outs (%arg2: memref<2x3x7xf32>)
```
Example Broadcast and transpose:
```
linalg.batch_matmul indexing_maps = [
affine_map< (d0, d1, d2, d3) -> (d1, d3)>, //broadcast
affine_map< (d0, d1, d2, d3) -> (d0, d2, d3)>, //transpose
affine_map< (d0, d1, d2, d3) -> (d0, d1, d2)>
]
ins (%arg0, %arg1: memref<3x5xf32>, memref<2x7x5xf32>)
outs (%arg2: memref<2x3x7xf32>)
```
RFCs and related PR:
https://discourse.llvm.org/t/rfc-linalg-opdsl-constant-list-attribute-definition/80149https://discourse.llvm.org/t/rfc-op-explosion-in-linalg/82863https://discourse.llvm.org/t/rfc-mlir-linalg-operation-tree/83586https://github.com/llvm/llvm-project/pull/115319
For extremely large models, it may be inefficient to load the model into
memory in Python prior to passing it to the MLIR C APIs for
deserialization. This change adds an API to parse a ModuleOp directly
from a file path.
This logic is in the critical path for constructing an operation from
Python. It is faster to compute this in C++ than it is in Python, and it
is a minor change to do this.
This change also alters the API contract of
_ods_common.get_op_results_or_values to avoid calling
get_op_result_or_value on each element of a sequence, since the C++ code
will now do this.
Most of the diff here is simply reordering the code in IRCore.cpp.
* We can call .results without figuring out whether we have an Operation
or an OpView, and that's likely the common case anyway.
* If we have one or more results, we can return them directly, with no
need for a call to get_op_result_or_value. We're guaranteed that
.results returns a PyOpResultList, so we have either an OpResult or
sequence of OpResults, just as the API expects.
This saves a few 100ms during IR construction in an LLM JAX benchmark.
Gives option post as global list as well as arg to control which
dialects are loaded during context creation. This enables setting either
a good base set or skipping in individual cases.
This is a companion to #118583, although it can be landed independently
because since #117922 dialects do not have to use the same Python
binding framework as the Python core code.
This PR ports all of the in-tree dialect and pass extensions to
nanobind, with the exception of those that remain for testing pybind11
support.
This PR also:
* removes CollectDiagnosticsToStringScope from NanobindAdaptors.h. This
was overlooked in a previous PR and it is duplicated in Diagnostics.h.
---------
Co-authored-by: Jacques Pienaar <jpienaar@google.com>
Relands #118583, with a fix for Python 3.8 compatibility. It was not
possible to set the buffer protocol accessers via slots in Python 3.8.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::` to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
Why? https://nanobind.readthedocs.io/en/latest/why.html says it better
than I can, but my primary motivation for this change is to improve MLIR
IR construction time from JAX.
For a complicated Google-internal LLM model in JAX, this change improves
the MLIR
lowering time by around 5s (out of around 30s), which is a significant
speedup for simply switching binding frameworks.
To a large extent, this is a mechanical change, for instance changing
`pybind11::`
to `nanobind::`.
Notes:
* this PR needs Nanobind 2.4.0, because it needs a bug fix
(https://github.com/wjakob/nanobind/pull/806) that landed in that
release.
* this PR does not port the in-tree dialect extension modules. They can
be ported in a future PR.
* I removed the py::sibling() annotations from def_static and def_class
in `PybindAdapters.h`. These ask pybind11 to try to form an overload
with an existing method, but it's not possible to form mixed
pybind11/nanobind overloads this ways and the parent class is now
defined in nanobind. Better solutions may be possible here.
* nanobind does not contain an exact equivalent of pybind11's buffer
protocol support. It was not hard to add a nanobind implementation of a
similar API.
* nanobind is pickier about casting to std::vector<bool>, expecting that
the input is a sequence of bool types, not truthy values. In a couple of
places I added code to support truthy values during casting.
* nanobind distinguishes bytes (`nb::bytes`) from strings (e.g.,
`std::string`). This required nb::bytes overloads in a few places.
