Implement a new tensorview instruction #158
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
nadavrot
approved these changes
Dec 19, 2017
| @@ -70,7 +70,10 @@ struct IRGenVisitor : NodeWalker { | |||
| } | |||
| assert(!generatedNodeDest_.count(N) && | |||
| "Already generated code for this node"); | |||
| assert(isa<AllocActivationInst>(v) && "The value must be an activation"); | |||
| auto *dest = v; | |||
| if (auto *zn = dyn_cast<ZeroInst>(v)) | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| @@ -59,6 +59,10 @@ class IRBuilder { | |||
|
|
|||
| ReshapeInst *createReshapeOp(Value *input, llvm::ArrayRef<size_t> shape); | |||
|
|
|||
| TensorViewInst *createTensorView(ElemKind elemKind, | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| @@ -70,6 +70,24 @@ Tensor *Interpreter::getOrCreateTensor(const Value *v) { | |||
| return it->second; | |||
| } | |||
|
|
|||
| Tensor *Interpreter::getOrCreateUnownedTensor(const Value *v, | |||
| const Value *src) { | |||
| assert(llvm::isa<TensorViewInst>(v) && "Expected a tensor view"); | |||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| @@ -60,6 +60,7 @@ class Interpreter final : public Backend { | |||
| /// tensor is already allocated for \p v. | |||
| /// \returns a tensor for \p v. | |||
| Tensor *getOrCreateTensor(const Value *v); | |||
| Tensor *getOrCreateUnownedTensor(const Value *v, const Value *src); | |||
|
|
|||
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| auto odim = flattenCdr(outW.dims()); | ||
| auto idim = flattenCdr(inW.dims()); | ||
| assert(odim.first == idim.first && "Mismatch batch size"); | ||
| // outW and inW are 2-dimensional. |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
| assert(inW.dims().size() == 2); | ||
| assert(inG.dims().size() == 2); | ||
|
|
||
| // outG and inW are 2-dimensional. |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
opti-mix
commented
Dec 19, 2017
src/glow/IR/IRGen.cpp
Outdated
| Value *InGview = InG; | ||
| if (InGview->dims().size() != 2) { | ||
| auto idim = flattenCdr(InG->dims()); | ||
| builder_.createZeroInst("zero", InG); |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
nadavrot
reviewed
Dec 19, 2017
| auto idim = flattenCdr(inW.dims()); | ||
| assert(odim.first == idim.first && "Mismatch batch size"); | ||
| // outW and inW are 2-dimensional. | ||
| // Dimensions are batch size x activation. Weights are Y x Z. |
This comment was marked as off-topic.
This comment was marked as off-topic.
Sorry, something went wrong.
This instruction is used to create an unowned tensor with a required dimensionality from an existing tensor. The result is just a view of an existing tensor and does not allocate any new memory. Tensorview is essentially a typecast applied to a tensor.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This instruction is used to create an unowned tensor with a required dimensionality from an existing tensor. The result is just a view of an existing tensor and does not allocate any new memory. Tensorview is essentially a typecast applied to a tensor.