-
Notifications
You must be signed in to change notification settings - Fork 0
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
[SparseTIR] Index Lowering #8
[SparseTIR] Index Lowering #8
Conversation
2d011e1
to
deb46df
Compare
@yzh119 I guess the main code of index lowering is done. You can take some time to review when getting time. Note that the lowering of SparseTIR is not all about index lowering - other IR components such as blocks or loops are probably needed to be transformed as well. I'll think and write these transformations after the code of Block and For are completely pushed. Also, some unit tests will be taken into consideration once Block and For are ready. |
deb46df
to
6d44fc2
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
This PR has an assumption that each sparse iterator was only used in a single PR, which we should fix later.
33a7dcb
to
9660c40
Compare
6d44fc2
to
e4aa7e3
Compare
c00f885
to
2328dc4
Compare
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
LGTM.
Let's fix remaining issues later.
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
* Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF
[SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd
[SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd
[SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd
[SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd
Format and Buffer data structure (#1) [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd
Format and Buffer data structure (#1) [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> Fix AxisTree (#3) * fix axis tree * upd [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr [BugFix] Fix binary search & SpIterVar (#7) [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting [SparseTIR] Index Lowering (#8) * Add StmtFunctor/ExprFunctor for SparseBufferStore/Load * Add basic index lowering * Finish index lowering (maybe) * Address comments * Convert CRLF to LF Frontend update, demo scripts. (#10) * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * fix axis tree * upd * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * Format and Buffer data structure (#1) * [SparseTIR] Constructors and Python Interface for `Axis` and `SparseBuffer` (#2) * add methods for Object * axis constructors * methods for SparseBuffer * put into registry * python interface * [CherryPick][Intrinsic] lower_bound and upper_bound for binary search in Sparse TIR. (apache#483) (#4) * upd * upd * fix * upd * upd * upd * upd * upd * fix * upd * upd * upd * upd * upd * upd * upd * codegen-rule * upd * upd * test * upd * fix * two arguments Co-authored-by: Zihao Ye <expye@outlook.com> * Fix AxisTree (#3) * fix axis tree * upd * [SparseTIR] Add SparseBufferLoad/SparseBufferStore (#5) * Add dtype for SparseBuffer * Add name for SparseBuffer. Remove `ndim` * Remove namespace sparse * Add SparseBufferLoad/Store * Add method `ndim()` * [SparseTIR] Introduce SpIterVar (#6) * [SparseTIR] Introduce SpIterVar * Add conversion to PrimExpr * [BugFix] Fix binary search & SpIterVar (#7) * [BugFix] Add field `is_reduction` for SpIterVar (#9) * [BugFix] Add field `is_reduction` for SpIterVar * Formatting * upd * upd Co-authored-by: Ruihang Lai <lairuihangdongdong@qq.com> [SparseTIR] SparseBlock on C++/Python side (#11) * Fix a bug in the last commit * SparseBlock on C++ & Python side [BugFix][SparseTIR] TVMScript Parser for Axis & SpIterVar (#12) * Update `cord` and `pos` * Fix `idtype` * Formatting.. * Bug fix 1 * Move new special stmts * Parser for Axis and SpIterVar * Fix context_maintainer.py [SparseTIR] Enhance SparseBlock to contain enough PrimFunc information (#13) * Enhance SparseBlock to have enough PrimFunc info * Remove `func_sparse_buffer_map_` * Don't print the map uh-huh [SparseTIR] Parser, Printer, Roundtrip (#14) * SparseBlock scope handler (part 1) * SparseBlock scope handler (part 2) * SparseBlock scope handler (part 3) * SparseBlock scope handler (fix 1) * Add SparseBufferLoad/Store on Python side * Parser for SparseBufferLoad/Store * Add SparseBlock to Python __init__ * StmtFunctor for SparseBlock * Ensure at least one dimension for SparseBuffer * Make `axis` field of SpIterVar mandatory * SparseBlock scope handler (fix 2) * Update Axis syntax by removing `name` parameter * Move to intrin.py * Add filed `from_sparse` to DenseFixedAxis * SparseTIR script printer * Roundtrip test * `update_symbol` bug fix * Fix attr visit in SparseBuffer * Define then compare in SparseBlock * Fix printer bug for SparseBuffer * Enable graph match for Axis and SparseBuffer * Complete HashReduce and EqualReduce for AxisTree and SparseBuffer * Fix typo * Rename test * Bug fix 1 * Bug fix 2 * Add more tests Move tests (#15) [SparseTIR] ReprPrinter for Axis and SpIterVar (#16) upd (#17) flatten (#18) ELL and BSR correctness test scripts (#19) [SparseTIR] SparseTIR Lowering (#20) * Fix a previous bug of sparse-fixed SpIterVar creation * Fix a previous bug in `GetDenseValue` * Refactor Collector and IndexTransformer * Construct block and loops * Fix a previous bug which rejects DV iters in collector * Update buffer map * Create root block * Fix bug of sparse-fixed SpIterVar creation * Fix bug on SpIterVar conversion (with refactor) * Fix bug when getting dependent SpIterVars * Fix bug on dependency map and index lowering * Full block read/write region * Test version 1 * Fix bug of loop order * Fix bug of batch-mm iterator ordering * Update PrimFunc args to use symbolic params * Fix bug of test "csr_element_wise" * Fix bug of index accumulation for sparse-fixed axis * Update correctness test * Test structural equality * Refactor and use Array fix nnz cols Add docstring for sparse tir lowering (#21) * add docstring * upd Add more examples part 1 (sddmm) (#22) * upd * upd * upd [SparseTIR][Schedule] SparseBlockRV, GetSparseBlock, SparseReorder (#23) * Test initialization * Fix a stupid bug of ReprPrinter * Add SparseBlockRV * Schedule: GetSparseBlock * Schedule: Reorder [SparseTIR][Schedule] GetSpIters (#24) remove hybrid script for successful compilation Add atomic intrinsic for output nonzero inference. (#25) * upd * upd Add "sparse" block attribute. (#26) Revert "remove hybrid script for successful compilation" This reverts commit eebd7c1. [SparseTIR] Hack `IsAffineBinding` check (#27) * [TensorIR][Schedule] Inherit block anotation upon creating new blocks * Fix SDDMM test * Hack IsAffineBinding for sparse blocks Axis Dependency Tree aware code-gen and bmm example (#28) * upd * upd * upd * upd * upd * upd * upd * upd * remove redundancy * fix * upd * upd Re-design Indices lowering (#29) * upd * upd * upd * upd * upd * init * format * fix * revise coding-style * format Complete indices lowering (#30) * upd * upd * upd * done * upd * passed test * upd Add more docstrings and depress warnings for new lowering algorithm. (#31) Refactor derived axis, frontend support of fusion. (#32) * upd * upd * fix Fatal bugfix and change the signature of DenseVariableAxis. (#33) Syntax simplification (#34) Change the order of generated blocks for block isolation. (#35) * upd * upd * upd Syntax of AttachAxis for BMM (#36) * upd * upd * upd [SparseTIR] Add "square sum" lowering test (#37) * Add square sum test * Remove pylint comment [BugFix] Fix offset caching in lowering (#38) * Hack compact dataflow check in a dirty way * Add two-K square sum test * Mark skipped tests * Fix offset saving in lowering Fusion syntax fix + SDDMM example. (#39) Some structure change on update offsets. (#40) [Refactor] SparseTIR Lowering (#41) * Take out methods in Scope * Refactor * Refactor "match" * Tweak scope contents * Refactor ViewIndexInAxis * Refactor Scope * SDDMM tests under implementation * Refactor block stack * Use Map for var_map * Extract NeedCreateNewBlock * Simplify SpIterVarToIterVar via GetIterExtent * Refactor NeedCreateNewBlock * Add docstring * Use "auto" correctly * Minor refactor and use some move Remove redundant analyzers (#42) Support indices lowering for attach and fuse. (#43) * upd * upd * upd Fix irregular BMM example. (#44) * upd * upd * upd * upd RGCN forward and butterfly pattern example. (#45) Fused SDDMM example. (#46) * upd * wip * fix Fix sparse reorder after refactor (#47) [Refactor] Refactor Unittest (#48) * upd * remove redundancy [Unittest] Correctness test for benchmarking scripts (#49) Bugfix and more test for axis fusion, new workload (#50) * upd * upd upd upd upd upd upd upd
Work in progress...Waiting for sparse blocks and for loops.