benvanik
Follow
Pinned
1,614 contributions in the last year
Activity overview
Contribution activity
August 2022
Created 31 commits in 1 repository
Created a pull request in iree-org/iree that received 4 comments
Adding interfaces for global ops used by the util dialect and fixing SinkGlobalBufferLoads.
The interfaces allow us to share more code between dialects using globals and generalizes the Explorer infrastructure to support global ops in any …
+1,235
−865
•
4
comments
Opened 13 other pull requests in 1 repository
iree-org/iree
13
merged
- Gracefully ignore failed driver inits when listing all devices.
- Fixing HAL executable benchmarks after workgroup count ops.
- Fixing util.buffer.size -> vm.buffer.size result type conversion.
- Making executable layouts optional for local executables.
- Small tweaks found during stream->hal->vm lowering.
- Fixing flag argv truncation.
-
Supporting printing/parsing of
(null)in tools for null refs. - Adding tooling utils and cleaning up VM context/module creation.
- Adding VM module dependency versioning support.
- Ensuring rodata names and mime types can go from util->vm.
- Fixing iree-run-mlir command line parsing on Windows.
- Fixing iree_cc_unified_library debug output file name aliasing.
-
Using
llvm::Optional::value_orinstead of deprecatedgetValueOr.
Reviewed 30 pull requests in 2 repositories
iree-org/iree
25 pull requests
- Fix "cpu" to "llvm-cpu" in one last Colab notebook.
- Adds 2d x32 binary and unary elementwise microkernels.
- Adding interfaces for global ops used by the util dialect and fixing SinkGlobalBufferLoads.
- Gracefully ignore failed driver inits when listing all devices.
- Used prefixed form for accessors in WebGPUTarget.
- Summarize dispatch op names using linalg interfaces and heuristics.
- Fix relative path used to search for linker tools.
- Fix executable names for web samples.
- Consider aliasing in branch op conversions
- Remove pydm. So long little buddy.
-
NFC: Use
OpFoldResultforflow.dispatch.tensor.load/storefolders. - Remove legacy and abbreviated llvm-cpu target name registrations.
- Stripping debug info before generating LLVM target
- Fix samples after VM changes.
- [EmitC] Cleanup analysis struct
-
Canonicalize offset, sizes and strides for
flow.dispatch.tensor.store. - Adding tooling utils and cleaning up VM context/module creation.
- Remove unnecessary visibility
- Move memref->buffer conversion out of VM conversion.
- Fixing iree-run-mlir command line parsing on Windows.
- Show executable sources via source locs in Tracy if available.
- Integrate llvm-project at 955cc56af448301e2d353df90ca1f0c6a74118f2
- Rework strided buffer descriptor to be op/pattern based instead of IR…
- Integrate llvm-project at 809855b56f06dd7182685f88fbbc64111df9339a
- Early inc to avoid invalidation
- Some pull request reviews not shown.