| Age | Commit message (Collapse) | Author |
|
* Support high order diff pattern: `bwd_diff(fwd_diff(f))`.
* Fix.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Update restrictions section in autodiff doc.
* Update restrictions sections in autodiff doc.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Add support for `[PrimalSubstitute]` and `[PrimalSubstituteOf]`.
* Fix
* Fix.
* Cleanup.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* WIP: Fix for do-while loops
* Added a somewhat hacky fix for do-while loops
* Redid the indexed region map builder step to fix issue with the nested loops test
* rename
* Used managed pointers
|
|
(#2681)
|
|
* Multiple fixes to get various loop tests to pass.
* Create reverse-nested-loop.slang
* Fix for variables becoming inaccessible during cfg normalization
* Removed comments and moved break-branch-normalization to eliminateMultiLevelBreaks
* Fix.
* Override liveness tests
|
|
* Detect and deduplicate read-only resource access.
* Fix tests.
* Fix tests.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
Also added legalization for loops
|
|
* Fix crash when applying autodiff to functions with no arguments
* Fixes for loops where the break region is non-trivial
* Minor fix
* Implement array legalization correctly.
* Fix array legalization.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Modify control-flow test case
* Update reverse-control-flow-3.slang
* Fix `transposeCall`.
* Fix.
---------
Co-authored-by: Sai Praveen Bangaru <31557731+saipraveenb25@users.noreply.github.com>
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Added support for simple while loops
* Fix support for while loops by changing logic to grab the loop update block
|
|
* Fix differentiable type registration
* Fix use of non-differentiable return value in a differentiable func.
* Fix use of primal inst that does not dominate the diff block.
* Fix primal inst hoisting, and add missing type legalization logic.
* Make `detach` defined on all differentiable T.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
unzipping loops (#2663)
|
|
* WIP: Remove all legacy type checking
* Fixed issue with user-defined backward derivatives not bypassing the AD process
---------
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Fixed crash when lowering IR for no_diff struct member.
* Improve `setInsertBeforeOrdinaryInst` and `setInsertAfterOrdinaryInst`.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
* Full loop support now working. MaxItersAttr in progress
* Lookup table updates?
* Fixed the max iters decoration
* Minox fixes & remove superfluous code
* fixup warnings
* Revert "Lookup table updates?"
This reverts commit 7d9b0793fb5239f31d1155776e846dcf1892d8d9.
* Update 07-autodiff.md
* Change maxiters to MaxIters
* Added asserts
* Update 07-autodiff.md
|
|
|
|
* WIP: start adding backward derivatives
* Overhaul `transposeParameterBlock` to support `inout` params.
* Small bug fixes.
* Bug fix on differentiable intrinsic specialization.
* Fixes.
* Run autodiff tests on CPU.
* Clean up.
* Overhaul `transposeParameterBlock` to support `inout` params.
* Small bug fixes.
* Bug fix on differentiable intrinsic specialization.
* Fixes.
* Run autodiff tests on CPU.
* Clean up.
* More bug fixes.,
* WIP: working on detach
* Arithmetic simplifications and more IR clean up logic.
* WIP: adding detach and abs
* Fix detach and abs
* Fix.
* Add IR transform pass for cleaner code emit.
* Fix test cases.
* Fix type system logic for reference type.
* Add backward derivatives for functions that already have forward derivatives
* Fix changes
---------
Co-authored-by: Yong He <yhe@nvidia.com>
Co-authored-by: Lifan Wu <lifanw@nvidia.com>
|
|
|
|
* Overhaul `transposeParameterBlock` to support `inout` params.
* Small bug fixes.
* Bug fix on differentiable intrinsic specialization.
* Fixes.
* Run autodiff tests on CPU.
* Clean up.
* More bug fixes.,
* Add test coverage on inout param.
* Fix language server hinting for transcribed mutable params.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Support `out` parameters in backward differentiation.
* Fixes.
* Fix cleanup.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
(#2617)
|
|
* Add transposition logic for constructor opcodes.
* Fix.
* Add language server regression test.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
stdlib. (#2615)
* Allow array parameters in forward diff.
* Use type canonicalization instead of coersion.
* Reimplement array type.
* Fix.
* Update test case.
---------
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Added switch-case support; fixed non-diff parameter transposition
* Made region propagation much more robust. Partial loop unzip implementation
* WIP: Added most loop handling code, and a test. Still untested
* Added CFG Normalization pass + CFG Reversal Pass + Loop Unzipping + most loop transcription
* Add single-iter-loop test.
* proj files
* removed comments
* Update reverse-loop.slang
* Removed out-of-date code
* Disabled IR validation during constructSSA phase of normalizeCFG. constructSSA now reuses sharedBuilder
* Moved normalizeCFG() call to prepareFuncForBackwardDiff()
|
|
* Unify UpdateField and UpdateElement with access chain.
* Fix warnings.
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Reimplement address elimination pass.
* Fix error.
* Update test references.
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
flow works now (#2595)
|
|
|
|
* Make backward differentiation work with generics.
* Fix.
* Another fix.
* More fix.
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Initial multi-block implementation
* Implemented multi-block reverse-mode (without loops)
* Added logic to remove block-level decorations to avoid confusing IR simplification passes
* Fixed issues with block-level decorations during IR simplification by removing them prior to simplification.
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
* Added initial support for nested calls
* removed comments
Co-authored-by: Yong He <yonghe@outlook.com>
|
|
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
* Add vector arithmetic test. Make gradient accumulation work for any IRLoad
* Added support for general vector types, and split transposition into transpose & materialize to allow emitting the fully accumulated gradient for complex types.
* Several bug fixes + finished up support for vector & struct types + removed prop pass
* minor fixes (int/uint casts)
* Removed IRConstruct
* Added some type casts to prevent warnings
* minor fix for unused variable
|
|
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
|
|
* Make differentiable data-flow pass recognize interface methods.
* Make existing test to work with `[TreatAsDifferentiable]`.
Co-authored-by: Yong He <yhe@nvidia.com>
|
|
Co-authored-by: Yong He <yhe@nvidia.com>
|