diff options
| author | Edward Liu <shiqiu1105@gmail.com> | 2022-11-14 12:08:01 -0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2022-11-14 12:08:01 -0800 |
| commit | 368ec3116ea0f10f44acbf76b5dc9e34d6ff3d32 (patch) | |
| tree | 3d9def111db278affb8413bddb5aab9ce3cf73a6 /source/slang/slang-emit.cpp | |
| parent | 623f5c36e0dc8190753aa5fa2e89f1010c367c67 (diff) | |
Minimum binary arithmetic reverse autodiff working. (#2514)
* Initial plumbing of backward autodiff in the frontend.
* More plumbing.
* Initial reverse autodiff working.
* Bug fixes.
* Misc.
* Remove redundant code.
* More clean up.
* Misc.
* Rebase and add backward diff test.
* Disable test.
* Clean up.
* Minor fix.
Co-authored-by: Yong He <yhe@nvidia.com>
Diffstat (limited to 'source/slang/slang-emit.cpp')
| -rw-r--r-- | source/slang/slang-emit.cpp | 6 |
1 files changed, 1 insertions, 5 deletions
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp index de9d23e97..2478eccc2 100644 --- a/source/slang/slang-emit.cpp +++ b/source/slang/slang-emit.cpp @@ -376,11 +376,7 @@ Result linkAndOptimizeIR( dumpIRIfEnabled(codeGenContext, irModule, "BEFORE-AUTODIFF"); // Process higher-order calles to auto-diff passes. - // 1. Generate JVP code wherever necessary. (Linearization or "forward-mode" pass) - processForwardDifferentiableFuncs(irModule, sink); - - // 2. Transpose JVP to VJP code wherever needed. (Transposition or "reverse-mode" pass) - // processVJPDerivativeMarkers(module); // Disabled currently. No impl yet. + processDifferentiableFuncs(irModule, sink); stripAutoDiffDecorations(irModule); |