| Commit message (Collapse) | Author | Age | Files | Lines |
... | |
|
|
|
|
|
|
|
|
|
|
|
| |
- dropping what looks like outdated code post some of the previous
updates
Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>
Closes tensorflow/mlir#179
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/179 from bondhugula:llfix 2a72ea441fe1b3924802273ffbe9870afeb90f91
PiperOrigin-RevId: 274158273
|
|
|
|
| |
PiperOrigin-RevId: 274154655
|
|
|
|
| |
PiperOrigin-RevId: 274152154
|
|
|
|
|
|
| |
On failure, the IR is likely to be in an invalid state, meaning the custom printer for some operations may now crash. Using the generic op form prevents this from happening.
PiperOrigin-RevId: 274104146
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This cl adds support for generating a .mlir file containing a reproducer for crashes and failures that happen during pass execution. The reproducer contains a comment detailing the configuration of the pass manager(e.g. the textual description of the pass pipeline that the pass manager was executing), along with the original input module.
Example Output:
// configuration: -pass-pipeline='func(cse, canonicalize), inline'
// note: verifyPasses=false
module {
...
}
PiperOrigin-RevId: 274088134
|
|
|
|
|
|
| |
constructor.
PiperOrigin-RevId: 274087577
|
|
|
|
|
|
|
|
|
|
|
|
| |
In Standard to LLVM dialect conversion, the binary op conversion pattern
implicitly assumed some operands were of LLVM IR dialect type. This is not
necessarily true, for example if the Ops that produce those operands did not
match the existing convresion patterns. Check if all operands are of LLVM IR
dialect type and if not, fail to patch the binary op pattern.
Closes tensorflow/mlir#168
PiperOrigin-RevId: 274063207
|
|
|
|
|
|
|
|
|
|
|
| |
Translation to LLVM expects the entry module to have only specific types of ops
that correspond to LLVM IR entities allowed in a module. Currently those are
restricted to functions and globals. Introduce an additional check at the
module level. Inside individual functions, the check for supported Ops is
already performed, but it accepts all LLVM dialect Ops and wouldn't be
immediately applicable at the module level.
PiperOrigin-RevId: 274058651
|
|
|
|
|
|
|
|
|
|
|
| |
The lowering is specified as a pattern and is done only if the result
is a SPIR-V scalar type or vector type.
Handling ConstantOp with index return type needs special handling
since SPIR-V dialect does not have index types. Based on the bitwidth
of the attribute value, either i32 or i64 is chosen.
Other constant lowerings are left as a TODO.
PiperOrigin-RevId: 274056805
|
|
|
|
|
|
| |
This will allow for inlining newly devirtualized calls, as well as give a more accurate cost model(when we have one). Currently canonicalization will only run for nodes that have no child edges, as the child nodes may be erased during canonicalization. We can support this in the future, but it requires more intricate deletion tracking.
PiperOrigin-RevId: 274011386
|
|
|
|
|
|
|
|
| |
been replaced.
When an operation with regions gets replaced, we currently require that all of the remaining nested operations are still converted even though they are going to be replaced when the rewrite is finished. This cl adds a tracking for a minimal set of operations that are known to be "dead". This allows for ignoring the legalization of operations that are won't survive after conversion.
PiperOrigin-RevId: 274009003
|
|
|
|
| |
PiperOrigin-RevId: 273948293
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This function-like operation allows one to define functions that have wrapped
LLVM IR function type, in particular variadic functions. The operation was
added in parallel to the existing lowering flow, this commit only switches the
flow to use it.
Using a custom function type makes the LLVM IR dialect type system more
consistent and avoids complex conversion rules for functions that previously
had to use the built-in function type instead of a wrapped LLVM IR dialect type
and perform conversions during the analysis.
PiperOrigin-RevId: 273910855
|
|
|
|
|
|
|
| |
Closes tensorflow/mlir#172
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/172 from kiszk:quantops e27b57eac8f4c6ef7ee6a6f7b497d3e2f56f6798
PiperOrigin-RevId: 273879164
|
|
|
|
|
|
|
| |
That space is 4 StringRefs, not 3, because element 0 of the match always
contains the entire source string.
PiperOrigin-RevId: 273875606
|
|
|
|
|
|
|
|
| |
Allow printing out pipelines in a format that is as close as possible to the
textual pass pipeline format. Individual passes can override the print function
in order to format any options that may have been used to construct that pass.
PiperOrigin-RevId: 273813627
|
|
|
|
|
|
| |
Avoid unexpected side effect in rewriter insertion point.
PiperOrigin-RevId: 273785794
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The lowering infrastructure needs to be enhanced to lower into a
spv.Module that is consistent with the SPIR-V spec. The following
changes are needed
1) The Vulkan/SPIR-V validation rules dictates entry functions to have
signature of void(void). This requires changes to the function
signature conversion infrastructure within the dialect conversion
framework. When an argument is dropped from the original function
signature, a function can be specified that when invoked will return
the value to use as a replacement for the argument from the original
function.
2) Some changes to the type converter to make the converted type
consistent with the Vulkan/SPIR-V validation rules,
a) Add support for converting dynamically shaped tensors to
spv.rtarray type.
b) Make the global variable of type !spv.ptr<!spv.struct<...>>
3) Generate the entry point operation for the kernel functions and
automatically compute all the interface variables needed
PiperOrigin-RevId: 273784229
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This PR is a stepping stone towards supporting generic multi-store
source loop nests in affine loop fusion. It extends the algorithm to
support fusion of multi-store loop nests that:
1. have only one store that writes to a function-local live out, and
2. the remaining stores are involved in loop nest self dependences
or no dependences within the function.
Closes tensorflow/mlir#162
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/162 from dcaballe:dcaballe/multi-output-fusion 7fb7dec6fe8b45f5ce176f018bfe37b256420c45
PiperOrigin-RevId: 273773907
|
|
|
|
| |
PiperOrigin-RevId: 273707610
|
|
|
|
| |
PiperOrigin-RevId: 273707291
|
|
|
|
|
|
|
|
|
|
| |
SameOperandsAndResultType traits
Currently SameOperandsAndResultShape trait allows operands to have tensor<*xf32> and tensor<2xf32> but doesn't allow tensor<?xf32> and tensor<10xf32>.
Also, use the updated shape compatibility helper function in TensorCastOp::areCastCompatible method.
PiperOrigin-RevId: 273658336
|
|
|
|
|
|
| |
This enhances the symbol table utility methods to handle the case where an unknown operation may define a symbol table. When walking symbols, we now collect all symbol uses before allowing the user to iterate. This prevents the user from assuming that all symbols are actually known before performing a transformation.
PiperOrigin-RevId: 273651963
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This allows individual passes to define options structs and for these options to be parsed per instance of the pass while building the pass pipeline from the command line provided textual specification.
The user can specify these per-instance pipeline options like so:
```
struct MyPassOptions : public PassOptions<MyPassOptions> {
Option<int> exampleOption{*this, "flag-name", llvm::cl::desc("...")};
List<int> exampleListOption{*this, "list-flag-name", llvm::cl::desc("...")};
};
static PassRegistration<MyPass, MyPassOptions> pass("my-pass", "description");
```
PiperOrigin-RevId: 273650140
|
|
|
|
|
|
| |
The restriction that symbols can only have identifier names is arbitrary, and artificially limits the names that a symbol may have. This change adds support for parsing and printing symbols that don't fit in the 'bare-identifier' grammar by printing the reference in quotes, e.g. @"0_my_reference" can now be used as a symbol name.
PiperOrigin-RevId: 273644768
|
|
|
|
|
|
|
|
|
| |
This is matching what the runtime library is expecting.
Closes tensorflow/mlir#171
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/171 from deven-amd:deven-rocdl-device-func-i64 80762629a8c34e844ebdc542b34dd783990db9db
PiperOrigin-RevId: 273640767
|
|
|
|
|
|
|
|
|
|
|
| |
Add a pass to decorate the composite types used by
composite objects in the StorageBuffer, PhysicalStorageBuffer,
Uniform, and PushConstant storage classes with layout information.
Closes tensorflow/mlir#156
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/156 from denis0x0D:sandbox/layout_info_decoration 7c50840fd38ca169a2da7ce9886b52b50c868b84
PiperOrigin-RevId: 273634140
|
|
|
|
|
|
| |
This is similar to the `inlineRegionBefore` hook, except the original blocks are unchanged. The region to be cloned *must* not have been modified during the conversion process at the point of cloning, i.e. it must belong an operation that has yet to be converted, or the operation that is currently being converted.
PiperOrigin-RevId: 273622533
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- bodies would earlier appear in the order (i, i+3, i+2, i+1) instead of
(i, i+1, i+2, i+3) for example for factor 4.
- clean up hardcoded test cases
Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>
Closes tensorflow/mlir#170
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/170 from bondhugula:ujam b66b405b2b1894a03b376952e32a9d0292042665
PiperOrigin-RevId: 273613131
|
|
|
|
|
|
| |
MLIR uses symbol references to model references to many global entities, such as functions/variables/etc. Before this change, there is no way to actually reason about the uses of such entities. This change provides a walker for symbol references(via SymbolTable::walkSymbolUses), as well as 'use_empty' support(via SymbolTable::symbol_use_empty). It also resolves some deficiencies in the LangRef definition of SymbolRefAttr, namely the restrictions on where a SymbolRefAttr can be stored, ArrayAttr and DictionaryAttr, and the relationship with operations containing the SymbolTable trait.
PiperOrigin-RevId: 273549331
|
|
|
|
|
|
| |
The default value is never used as the value of the elide option is only used if it has an occurrence.
PiperOrigin-RevId: 273545143
|
|
|
|
|
|
|
|
|
|
|
| |
During the conversion, both the original and the converted function may coexist
in the module and have the same symbol name. There is no guarantee which of the
two will be found by the symbol lookup. Avoid returning the result of the
library function lookup when lowering Linalg to Standard or LLVM. Use the
symbol reference instead. After the conversion completes, only one symbol will
remain and the Ops using SymbolRefAttrs will be referring to the correct one.
PiperOrigin-RevId: 273510079
|
|
|
|
|
|
|
|
|
|
|
| |
Originally, we were attaching attributes containing CUBIN blobs to the kernel
function called by `gpu.launch_func`. This kernel is now contained in a nested
module that is used as a compilation unit. Attach compiled CUBIN blobs to the
module rather than to the function since we were compiling the module. This
also avoids duplication of the attribute on multiple kernels within the same
module.
PiperOrigin-RevId: 273497303
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Originally, the CUBIN getter function was introduced as a mechanism to
circumvent the absence of globals in the LLVM dialect. It would allocate memory
and populate it with the CUBIN data. LLVM dialect now supports globals and they
are already used to store CUBIN data, making the getter function a trivial
address computation of a global. Emit the address computation directly at the
place of `gpu.launch_func` instead of putting it in a function and calling it.
This simplifies the conversion flow and prepares it for using the
DialectConversion infrastructure.
PiperOrigin-RevId: 273496221
|
|
|
|
|
|
|
|
|
|
|
| |
Now that the accessor function is a trivial getter of the global variable, it
makes less sense to have the getter generation as a separate pass. Move the
getter generation into the lowering of `gpu.launch_func` to CUDA calls. This
change is mostly code motion, but the process can be simplified further by
generating the addressof inplace instead of using a call. This is will be done
in a follow-up.
PiperOrigin-RevId: 273492517
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The kernel function called by gpu.launch_func is now placed into an isolated
nested module during the outlining stage to simplify separate compilation.
Until recently, modules did not have names and could not be referenced. This
limitation was circumvented by introducing a stub kernel at the same name at
the same nesting level as the module containing the actual kernel. This
relation is only effective in one direction: from actual kernel function to its
launch_func "caller".
Leverage the recently introduced symbol name attributes on modules to refer to
a specific nested module from `gpu.launch_func`. This removes the implicit
connection between the identically named stub and kernel functions. It also
enables support for `gpu.launch_func`s to call different kernels located in the
same module.
PiperOrigin-RevId: 273491891
|
|
|
|
|
|
| |
directly.
PiperOrigin-RevId: 273446814
|
|
|
|
|
|
| |
Some modules may have extremely large ElementsAttrs, which makes debugging involving IR dumping extremely slow and painful. This change adds a flag that will elide ElementsAttrs with a "large"(as defined by the user) number of elements by printing "..." instead of the element data.
PiperOrigin-RevId: 273413100
|
|
|
|
| |
PiperOrigin-RevId: 273406833
|
|
|
|
| |
PiperOrigin-RevId: 273384063
|
|
|
|
|
|
|
|
|
| |
Since MLIR integer types don't make a distinction between signed vs
unsigned integers, during deserialization of SPIR-V binaries, the
OpBitcast might result in a cast from/to the same type. Do not add a
spv.Bitcast operation to the spv.module in these cases.
PiperOrigin-RevId: 273381887
|
|
|
|
|
|
|
|
| |
Operation::print behavior.
This allows for controlling the behavior of the AsmPrinter programmatically, instead of relying exclusively on cl::opt flags. This will also allow for more fine-tuned control of printing behavior per callsite, instead of being applied globally.
PiperOrigin-RevId: 273368361
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The SPIR-V spec recommends all OpUndef instructions be generated at
module level. For the SPIR-V dialect its better for UndefOp to produce
an SSA value for use with other instructions. If UndefOp is to be used
at module level, it cannot produce an SSA value (use of this SSA value
within FuncOp would need implicit capture). To satisfy needs of the
SPIR-V spec while making it simpler to represent UndefOp in the SPIR-V
dialect, the serialization is updated to create OpUndef instruction
at module scope.
PiperOrigin-RevId: 273355526
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The structured selection/loop's entry block does not have arguments.
If the function's header block is also part of the structured control
flow, we cannot just simply erase it because it may contain arguments
matching the function signature and used by the cloned blocks. Instead,
turn it into a block only containing a spv.Branch op.
Also, we can directly emit instructions for the spv.selection header
block to the block containing the spv.selection op. This eliminates
unnecessary branches in the SPIR-V blob.
Added a test for nested spv.loop.
PiperOrigin-RevId: 273351424
|
|
|
|
|
|
|
|
|
| |
Signed-off-by: Uday Bondhugula <uday@polymagelabs.com>
Closes tensorflow/mlir#157
COPYBARA_INTEGRATE_REVIEW=https://github.com/tensorflow/mlir/pull/157 from bondhugula:quickfix bd1fcd79825fc0bd5b4a3e688153fa0993ab703d
PiperOrigin-RevId: 273316498
|
|
|
|
|
|
| |
because ilist_node_with_parent specifically requires a 'getParent() const' method. If/When ilist_node removes this constraint we should drop the const to fit the rest of the MLIR const model.
PiperOrigin-RevId: 273316153
|
|
|
|
|
|
|
|
| |
Now that linalg.view and strided memrefs are unified, there is no reason to
disallow AllocOp in alias analysis. This CLs adds support for AllocOp which allows writing shorter tests that do not require explicitly creating a view for
each operation.
PiperOrigin-RevId: 273303060
|
|
|
|
|
|
| |
Add new `typeDescription` (description was already used by base constraint class) field to type to allow writing longer descriptions about a type being defined. This allows for providing additional information/rationale for a defined type. This currently uses `description` as the heading/name for the type in the generated documentation.
PiperOrigin-RevId: 273299332
|
|
|
|
|
|
|
|
|
| |
See RFC: https://groups.google.com/a/tensorflow.org/forum/#!topic/mlir/xE2IzfhE3Wg.
Opaque location stores two pointers, one of them points to some data structure that is external to MLIR, and the other one is unique for each type and represents type id of that data structure. OpaqueLoc also stores an optional location that can be used if the first one is not suitable.
OpaqueLoc is managed similar to FileLineColLoc. It is passed around by MLIR transformations and can be used in compound locations like CallSiteLoc.
PiperOrigin-RevId: 273266510
|
|
|
|
|
|
| |
gpu.all_reduce now supports block sizes that are not multiple of 32.
PiperOrigin-RevId: 273255204
|