Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
//
|
|
|
|
|
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
|
|
|
|
|
// See https://llvm.org/LICENSE.txt for license information.
|
|
|
|
|
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
|
2021-09-30 00:03:40 +08:00
|
|
|
// Also available under a BSD-style license. See LICENSE.
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
//
|
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
|
|
#include "PassDetail.h"
|
2021-09-24 10:18:56 +08:00
|
|
|
|
2022-04-27 03:27:51 +08:00
|
|
|
#include "mlir/Dialect/Affine/IR/AffineOps.h"
|
|
|
|
|
#include "mlir/Dialect/Arithmetic/IR/Arithmetic.h"
|
2022-02-13 02:47:12 +08:00
|
|
|
#include "mlir/Dialect/ControlFlow/IR/ControlFlowOps.h"
|
2022-03-16 18:44:23 +08:00
|
|
|
#include "mlir/Dialect/Func/IR/FuncOps.h"
|
2022-01-05 00:00:48 +08:00
|
|
|
#include "mlir/Dialect/Linalg/IR/Linalg.h"
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
#include "mlir/Dialect/Math/IR/Math.h"
|
2022-03-31 22:24:44 +08:00
|
|
|
#include "mlir/Dialect/SCF/SCF.h"
|
2021-07-08 02:55:31 +08:00
|
|
|
#include "mlir/Dialect/Tensor/IR/Tensor.h"
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
#include "mlir/IR/OpDefinition.h"
|
|
|
|
|
#include "mlir/Transforms/DialectConversion.h"
|
2022-03-03 00:48:15 +08:00
|
|
|
#include "torch-mlir-dialects/Dialect/TMTensor/IR/TMTensorDialect.h"
|
|
|
|
|
#include "torch-mlir-dialects/Dialect/TMTensor/IR/TMTensorOps.h"
|
2022-01-12 09:59:42 +08:00
|
|
|
#include "torch-mlir/Dialect/TorchConversion/IR/TorchConversionOps.h"
|
2021-09-24 11:59:12 +08:00
|
|
|
#include "torch-mlir/Dialect/TorchConversion/Transforms/Passes.h"
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
|
|
|
|
#include "mlir/IR/BuiltinOps.h"
|
|
|
|
|
|
|
|
|
|
using namespace mlir;
|
2021-09-24 11:59:12 +08:00
|
|
|
using namespace mlir::torch;
|
|
|
|
|
using namespace mlir::torch::TorchConversion;
|
2022-03-03 00:48:15 +08:00
|
|
|
using namespace TMTensor;
|
|
|
|
|
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
|
|
|
|
namespace {
|
2021-09-24 10:18:56 +08:00
|
|
|
class VerifyLinalgOnTensorsBackendContractPass
|
|
|
|
|
: public VerifyLinalgOnTensorsBackendContractBase<
|
|
|
|
|
VerifyLinalgOnTensorsBackendContractPass> {
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
void runOnOperation() override {
|
|
|
|
|
MLIRContext *context = &getContext();
|
|
|
|
|
auto module = getOperation();
|
|
|
|
|
TypeConverter converter;
|
|
|
|
|
converter.addConversion([](RankedTensorType type) -> Type {
|
|
|
|
|
if (BaseMemRefType::isValidElementType(type.getElementType()))
|
|
|
|
|
return type;
|
|
|
|
|
return nullptr;
|
|
|
|
|
});
|
|
|
|
|
TypeConverter scalarConverter;
|
|
|
|
|
for (TypeConverter *c : {&converter, &scalarConverter}) {
|
|
|
|
|
c->addConversion([](FloatType type) { return type; });
|
|
|
|
|
c->addConversion([](IntegerType type) { return type; });
|
|
|
|
|
c->addConversion([](IndexType type) { return type; });
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
auto opHasLegalTypes = [&](Operation *op) { return converter.isLegal(op); };
|
|
|
|
|
auto isLegalScalarOp = [&](Operation *op) {
|
|
|
|
|
// We recognize basic scalar ops by them having the trait "Elementwise",
|
|
|
|
|
// even though we don't expect them to operate on tensors.
|
|
|
|
|
return scalarConverter.isLegal(op) &&
|
|
|
|
|
op->hasTrait<OpTrait::Elementwise>();
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
ConversionTarget target(*context);
|
|
|
|
|
|
|
|
|
|
// Structural operations.
|
2022-04-27 03:27:51 +08:00
|
|
|
target.addDynamicallyLegalOp<ModuleOp, func::FuncOp, func::ReturnOp>(
|
2022-03-16 18:44:23 +08:00
|
|
|
opHasLegalTypes);
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
2022-01-12 09:59:42 +08:00
|
|
|
target.addDynamicallyLegalOp<GetNextSeedOp>(opHasLegalTypes);
|
|
|
|
|
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
// Basic scalar operations.
|
2022-03-16 18:44:23 +08:00
|
|
|
target.addDynamicallyLegalDialect<func::FuncDialect>(isLegalScalarOp);
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
target.addDynamicallyLegalDialect<math::MathDialect>(isLegalScalarOp);
|
2021-10-16 02:34:29 +08:00
|
|
|
target.addDynamicallyLegalDialect<arith::ArithmeticDialect>(
|
|
|
|
|
isLegalScalarOp);
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
2021-04-22 06:07:15 +08:00
|
|
|
// Tensor operations should go through linalg and the tensor dialect.
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
target.addDynamicallyLegalDialect<linalg::LinalgDialect>(opHasLegalTypes);
|
2021-04-22 06:07:15 +08:00
|
|
|
target.addDynamicallyLegalDialect<tensor::TensorDialect>(opHasLegalTypes);
|
2021-12-03 12:13:00 +08:00
|
|
|
target.addDynamicallyLegalDialect<AffineDialect>(opHasLegalTypes);
|
2022-02-13 02:47:12 +08:00
|
|
|
target.addDynamicallyLegalDialect<cf::ControlFlowDialect>(opHasLegalTypes);
|
2022-03-03 00:48:15 +08:00
|
|
|
target.addDynamicallyLegalDialect<TMTensorDialect>(opHasLegalTypes);
|
2022-03-31 22:24:44 +08:00
|
|
|
target.addDynamicallyLegalDialect<scf::SCFDialect>(opHasLegalTypes);
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
|
|
|
|
// ConstantOp is used for tensors and for scalars.
|
2021-10-16 02:34:29 +08:00
|
|
|
target.addDynamicallyLegalOp<arith::ConstantOp>(opHasLegalTypes);
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
|
|
|
|
|
RewritePatternSet patterns(context);
|
|
|
|
|
if (failed(applyFullConversion(module, target, std::move(patterns)))) {
|
2021-05-20 08:36:00 +08:00
|
|
|
// We avoid `module.emitError()` so that mlir-print-op-on-diagnostics
|
|
|
|
|
// doesn't unnecessarily spew out the entire module.
|
|
|
|
|
emitError(module.getLoc())
|
2021-09-28 02:36:44 +08:00
|
|
|
<< "Module does not conform to the linalg-on-tensors backend contract. "
|
|
|
|
|
"See dialect conversion legality information above.";
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
return signalPassFailure();
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
} // namespace
|
|
|
|
|
|
|
|
|
|
std::unique_ptr<OperationPass<ModuleOp>>
|
2021-09-24 11:59:12 +08:00
|
|
|
mlir::torch::TorchConversion::createVerifyLinalgOnTensorsBackendContractPass() {
|
2021-09-24 10:18:56 +08:00
|
|
|
return std::make_unique<VerifyLinalgOnTensorsBackendContractPass>();
|
Add npcomp-verify-backend-contract pass.
This pass verifies that a given module satisfies the contract that we
have for backends. This is phrased as an "allowlist", because we want to
keep this interface tight. Also, this gives much better diagnostics than
a backend randomly crashing or failing to compile would (though they
could still be improved).
This was especially painful because if we had
`tensor<?x!numpy.any_dtype>` slip through, at some point RefBackend
would convert it to a memref type and trip the "verify type invariants"
assertion which gives no location or anything and crashed the process,
which was very unpleasant.
We implement this with the dialect conversion framework, which works
reasonably well and was quick to put together and familiar, but is still
very "op oriented". We probably want to make this hand-rolled
eventually, especially the error reporting (the most useful kind of
error for a dialect conversion user is not necessarily the best for this
use case). Also, in production, these error will go to users, and need
to be surfaced carefully such as "the compiler needs a type annotation
on this function parameter" which in general requires some special
analysis, wordsmithing, and overall awareness of the e2e use case (such
as how much we can lean into certain source locations) to provide a
meaningful user-level diagnostic.
Also, add `inline` to the current frontend lowering pass pipeline to
allow slightly more complicated programs that otherwise would fail on
shape inference.
2021-04-13 09:39:53 +08:00
|
|
|
}
|
