kumi 0.0.18 → 0.0.19
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/.rubocop.yml +4 -1
- data/.ruby-version +1 -0
- data/CHANGELOG.md +25 -0
- data/CLAUDE.md +26 -139
- data/README.md +143 -372
- data/data/functions/agg/boolean.yaml +14 -0
- data/data/functions/agg/numeric.yaml +80 -0
- data/data/functions/agg/string.yaml +8 -0
- data/data/functions/core/arithmetic.yaml +50 -0
- data/data/functions/core/boolean.yaml +18 -0
- data/data/functions/core/comparison.yaml +42 -0
- data/data/functions/core/constructor.yaml +32 -0
- data/data/functions/core/select.yaml +7 -0
- data/data/functions/core/stencil.yaml +21 -0
- data/data/functions/core/string.yaml +19 -0
- data/data/kernels/javascript/agg/boolean.yaml +12 -0
- data/data/kernels/javascript/agg/numeric.yaml +27 -0
- data/data/kernels/javascript/agg/string.yaml +5 -0
- data/data/kernels/javascript/core/arithmetic.yaml +32 -0
- data/data/kernels/javascript/core/boolean.yaml +12 -0
- data/data/kernels/javascript/core/comparison.yaml +24 -0
- data/data/kernels/javascript/core/constructor.yaml +15 -0
- data/data/kernels/javascript/core/select.yaml +7 -0
- data/data/kernels/javascript/core/string.yaml +12 -0
- data/data/kernels/ruby/agg/boolean.yaml +18 -0
- data/data/kernels/ruby/agg/numeric.yaml +29 -0
- data/data/kernels/ruby/agg/string.yaml +5 -0
- data/data/kernels/ruby/core/arithmetic.yaml +38 -0
- data/data/kernels/ruby/core/boolean.yaml +15 -0
- data/data/kernels/ruby/core/comparison.yaml +30 -0
- data/data/kernels/ruby/core/constructor.yaml +17 -0
- data/data/kernels/ruby/core/select.yaml +4 -0
- data/data/kernels/ruby/core/string.yaml +15 -0
- data/debug_ordering.rb +52 -0
- data/docs/FORM_SCHEMA.md +85 -0
- data/docs/OUTPUT_SCHEMA.md +69 -0
- data/docs/SYNTAX.md +626 -373
- data/golden/array_element/expected/ast.txt +21 -0
- data/golden/array_element/expected/input_plan.txt +5 -0
- data/golden/array_element/expected/lir_00_unoptimized.txt +13 -0
- data/golden/array_element/expected/lir_01_hoist_scalar_references.txt +13 -0
- data/golden/array_element/expected/lir_02_inlined.txt +13 -0
- data/golden/array_element/expected/lir_03_cse.txt +13 -0
- data/golden/array_element/expected/lir_04_1_loop_fusion.txt +13 -0
- data/golden/array_element/expected/lir_04_loop_invcm.txt +13 -0
- data/golden/array_element/expected/lir_06_const_prop.txt +13 -0
- data/golden/array_element/expected/nast.txt +7 -0
- data/golden/array_element/expected/schema_javascript.mjs +13 -0
- data/golden/array_element/expected/schema_ruby.rb +41 -0
- data/golden/array_element/expected/snast.txt +7 -0
- data/golden/array_element/expected.json +5 -0
- data/golden/array_element/input.json +9 -0
- data/golden/array_element/schema.kumi +11 -0
- data/golden/array_index/expected/ast.txt +59 -0
- data/golden/array_index/expected/input_plan.txt +5 -0
- data/golden/array_index/expected/lir_00_unoptimized.txt +41 -0
- data/golden/array_index/expected/lir_01_hoist_scalar_references.txt +41 -0
- data/golden/array_index/expected/lir_02_inlined.txt +42 -0
- data/golden/array_index/expected/lir_03_cse.txt +40 -0
- data/golden/array_index/expected/lir_04_1_loop_fusion.txt +40 -0
- data/golden/array_index/expected/lir_04_loop_invcm.txt +40 -0
- data/golden/array_index/expected/lir_06_const_prop.txt +40 -0
- data/golden/array_index/expected/nast.txt +33 -0
- data/golden/array_index/expected/schema_javascript.mjs +46 -0
- data/golden/array_index/expected/schema_ruby.rb +75 -0
- data/golden/array_index/expected/snast.txt +33 -0
- data/golden/array_index/expected.json +5 -0
- data/golden/array_index/input.json +1 -0
- data/golden/array_index/schema.kumi +19 -0
- data/golden/array_operations/expected/ast.txt +50 -0
- data/golden/array_operations/expected/input_plan.txt +10 -0
- data/golden/array_operations/expected/lir_00_unoptimized.txt +47 -0
- data/golden/array_operations/expected/lir_01_hoist_scalar_references.txt +47 -0
- data/golden/array_operations/expected/lir_02_inlined.txt +47 -0
- data/golden/array_operations/expected/lir_03_cse.txt +47 -0
- data/golden/array_operations/expected/lir_04_1_loop_fusion.txt +47 -0
- data/golden/array_operations/expected/lir_04_loop_invcm.txt +47 -0
- data/golden/array_operations/expected/lir_06_const_prop.txt +47 -0
- data/golden/array_operations/expected/nast.txt +32 -0
- data/golden/array_operations/expected/schema_javascript.mjs +60 -0
- data/golden/array_operations/expected/schema_ruby.rb +91 -0
- data/golden/array_operations/expected/snast.txt +32 -0
- data/golden/array_operations/expected.json +27 -0
- data/golden/array_operations/input.json +8 -0
- data/golden/array_operations/schema.kumi +12 -10
- data/golden/cascade_logic/expected/ast.txt +37 -0
- data/golden/cascade_logic/expected/input_plan.txt +2 -0
- data/golden/cascade_logic/expected/lir_00_unoptimized.txt +29 -0
- data/golden/cascade_logic/expected/lir_01_hoist_scalar_references.txt +29 -0
- data/golden/cascade_logic/expected/lir_02_inlined.txt +37 -0
- data/golden/cascade_logic/expected/lir_03_cse.txt +30 -0
- data/golden/cascade_logic/expected/lir_04_1_loop_fusion.txt +30 -0
- data/golden/cascade_logic/expected/lir_04_loop_invcm.txt +30 -0
- data/golden/cascade_logic/expected/lir_06_const_prop.txt +30 -0
- data/golden/cascade_logic/expected/nast.txt +32 -0
- data/golden/cascade_logic/expected/schema_javascript.mjs +31 -0
- data/golden/cascade_logic/expected/schema_ruby.rb +57 -0
- data/golden/cascade_logic/expected/snast.txt +32 -0
- data/golden/cascade_logic/expected.json +5 -0
- data/golden/cascade_logic/input.json +4 -0
- data/golden/cascade_logic/schema.kumi +1 -3
- data/golden/chained_fusion/expected/ast.txt +57 -0
- data/golden/chained_fusion/expected/input_plan.txt +13 -0
- data/golden/chained_fusion/expected/lir_00_unoptimized.txt +76 -0
- data/golden/chained_fusion/expected/lir_01_hoist_scalar_references.txt +76 -0
- data/golden/chained_fusion/expected/lir_02_inlined.txt +114 -0
- data/golden/chained_fusion/expected/lir_03_cse.txt +97 -0
- data/golden/chained_fusion/expected/lir_04_1_loop_fusion.txt +99 -0
- data/golden/chained_fusion/expected/lir_04_loop_invcm.txt +97 -0
- data/golden/chained_fusion/expected/lir_06_const_prop.txt +97 -0
- data/golden/chained_fusion/expected/nast.txt +55 -0
- data/golden/chained_fusion/expected/schema_javascript.mjs +116 -0
- data/golden/chained_fusion/expected/schema_ruby.rb +149 -0
- data/golden/chained_fusion/expected/snast.txt +55 -0
- data/golden/chained_fusion/expected.json +45 -0
- data/golden/chained_fusion/input.json +51 -0
- data/golden/chained_fusion/schema.kumi +52 -0
- data/golden/element_arrays/expected/ast.txt +55 -0
- data/golden/element_arrays/expected/input_plan.txt +7 -0
- data/golden/element_arrays/expected/lir_00_unoptimized.txt +81 -0
- data/golden/element_arrays/expected/lir_01_hoist_scalar_references.txt +81 -0
- data/golden/element_arrays/expected/lir_02_inlined.txt +85 -0
- data/golden/element_arrays/expected/lir_03_cse.txt +83 -0
- data/golden/element_arrays/expected/lir_04_1_loop_fusion.txt +83 -0
- data/golden/element_arrays/expected/lir_04_loop_invcm.txt +83 -0
- data/golden/element_arrays/expected/lir_06_const_prop.txt +83 -0
- data/golden/element_arrays/expected/nast.txt +42 -0
- data/golden/element_arrays/expected/schema_javascript.mjs +106 -0
- data/golden/element_arrays/expected/schema_ruby.rb +141 -0
- data/golden/element_arrays/expected/snast.txt +42 -0
- data/golden/element_arrays/expected.json +55 -0
- data/golden/element_arrays/input.json +12 -0
- data/golden/element_arrays/schema.kumi +21 -0
- data/golden/empty_and_null_inputs/expected/ast.txt +42 -0
- data/golden/empty_and_null_inputs/expected/input_plan.txt +12 -0
- data/golden/empty_and_null_inputs/expected/lir_00_unoptimized.txt +31 -0
- data/golden/empty_and_null_inputs/expected/lir_01_hoist_scalar_references.txt +31 -0
- data/golden/empty_and_null_inputs/expected/lir_02_inlined.txt +51 -0
- data/golden/empty_and_null_inputs/expected/lir_03_cse.txt +49 -0
- data/golden/empty_and_null_inputs/expected/lir_04_1_loop_fusion.txt +49 -0
- data/golden/empty_and_null_inputs/expected/lir_04_loop_invcm.txt +49 -0
- data/golden/empty_and_null_inputs/expected/lir_06_const_prop.txt +49 -0
- data/golden/empty_and_null_inputs/expected/nast.txt +18 -0
- data/golden/empty_and_null_inputs/expected/schema_javascript.mjs +47 -0
- data/golden/empty_and_null_inputs/expected/schema_ruby.rb +79 -0
- data/golden/empty_and_null_inputs/expected/snast.txt +18 -0
- data/golden/empty_and_null_inputs/expected.json +8 -0
- data/golden/empty_and_null_inputs/input.json +16 -0
- data/golden/empty_and_null_inputs/schema.kumi +30 -0
- data/golden/game_of_life/expected/ast.txt +118 -0
- data/golden/game_of_life/expected/input_plan.txt +5 -0
- data/golden/game_of_life/expected/lir_00_unoptimized.txt +343 -0
- data/golden/game_of_life/expected/lir_01_hoist_scalar_references.txt +343 -0
- data/golden/game_of_life/expected/lir_02_inlined.txt +1918 -0
- data/golden/game_of_life/expected/lir_03_cse.txt +766 -0
- data/golden/game_of_life/expected/lir_04_1_loop_fusion.txt +766 -0
- data/golden/game_of_life/expected/lir_04_loop_invcm.txt +766 -0
- data/golden/game_of_life/expected/lir_06_const_prop.txt +766 -0
- data/golden/game_of_life/expected/nast.txt +104 -0
- data/golden/game_of_life/expected/schema_javascript.mjs +98 -0
- data/golden/game_of_life/expected/schema_ruby.rb +125 -0
- data/golden/game_of_life/expected/snast.txt +104 -0
- data/golden/game_of_life/expected.json +3 -0
- data/golden/game_of_life/input.json +8 -0
- data/golden/game_of_life/schema.kumi +33 -0
- data/golden/hash_keys/expected/ast.txt +29 -0
- data/golden/hash_keys/expected/input_plan.txt +2 -0
- data/golden/hash_keys/expected/lir_00_unoptimized.txt +18 -0
- data/golden/hash_keys/expected/lir_01_hoist_scalar_references.txt +18 -0
- data/golden/hash_keys/expected/lir_02_inlined.txt +18 -0
- data/golden/hash_keys/expected/lir_03_cse.txt +18 -0
- data/golden/hash_keys/expected/lir_04_1_loop_fusion.txt +18 -0
- data/golden/hash_keys/expected/lir_04_loop_invcm.txt +18 -0
- data/golden/hash_keys/expected/lir_06_const_prop.txt +18 -0
- data/golden/hash_keys/expected/nast.txt +46 -0
- data/golden/hash_keys/expected/schema_javascript.mjs +34 -0
- data/golden/hash_keys/expected/schema_ruby.rb +61 -0
- data/golden/hash_keys/expected/snast.txt +46 -0
- data/golden/hash_keys/expected.json +20 -0
- data/golden/hash_keys/input.json +4 -0
- data/golden/hash_keys/schema.kumi +14 -0
- data/golden/hash_value/expected/ast.txt +37 -0
- data/golden/hash_value/expected/input_plan.txt +7 -0
- data/golden/hash_value/expected/lir_00_unoptimized.txt +30 -0
- data/golden/hash_value/expected/lir_01_hoist_scalar_references.txt +30 -0
- data/golden/hash_value/expected/lir_02_inlined.txt +36 -0
- data/golden/hash_value/expected/lir_03_cse.txt +33 -0
- data/golden/hash_value/expected/lir_04_1_loop_fusion.txt +33 -0
- data/golden/hash_value/expected/lir_04_loop_invcm.txt +33 -0
- data/golden/hash_value/expected/lir_06_const_prop.txt +33 -0
- data/golden/hash_value/expected/nast.txt +25 -0
- data/golden/hash_value/expected/schema_javascript.mjs +46 -0
- data/golden/hash_value/expected/schema_ruby.rb +75 -0
- data/golden/hash_value/expected/snast.txt +25 -0
- data/golden/hash_value/expected.json +19 -0
- data/golden/hash_value/input.json +12 -0
- data/golden/hash_value/schema.kumi +19 -0
- data/golden/hierarchical_complex/expected/ast.txt +85 -0
- data/golden/hierarchical_complex/expected/input_plan.txt +23 -0
- data/golden/hierarchical_complex/expected/lir_00_unoptimized.txt +87 -0
- data/golden/hierarchical_complex/expected/lir_01_hoist_scalar_references.txt +87 -0
- data/golden/hierarchical_complex/expected/lir_02_inlined.txt +115 -0
- data/golden/hierarchical_complex/expected/lir_03_cse.txt +89 -0
- data/golden/hierarchical_complex/expected/lir_04_1_loop_fusion.txt +89 -0
- data/golden/hierarchical_complex/expected/lir_04_loop_invcm.txt +89 -0
- data/golden/hierarchical_complex/expected/lir_06_const_prop.txt +89 -0
- data/golden/hierarchical_complex/expected/nast.txt +49 -0
- data/golden/hierarchical_complex/expected/schema_javascript.mjs +121 -0
- data/golden/hierarchical_complex/expected/schema_ruby.rb +151 -0
- data/golden/hierarchical_complex/expected/snast.txt +49 -0
- data/golden/hierarchical_complex/expected.json +34 -0
- data/golden/hierarchical_complex/input.json +26 -0
- data/golden/hierarchical_complex/schema.kumi +38 -0
- data/golden/input_reference/expected/ast.txt +46 -0
- data/golden/input_reference/expected/input_plan.txt +15 -0
- data/golden/input_reference/expected/lir_00_unoptimized.txt +39 -0
- data/golden/input_reference/expected/lir_01_hoist_scalar_references.txt +39 -0
- data/golden/input_reference/expected/lir_02_inlined.txt +39 -0
- data/golden/input_reference/expected/lir_03_cse.txt +39 -0
- data/golden/input_reference/expected/lir_04_1_loop_fusion.txt +39 -0
- data/golden/input_reference/expected/lir_04_loop_invcm.txt +39 -0
- data/golden/input_reference/expected/lir_06_const_prop.txt +39 -0
- data/golden/input_reference/expected/nast.txt +21 -0
- data/golden/input_reference/expected/schema_javascript.mjs +45 -0
- data/golden/input_reference/expected/schema_ruby.rb +74 -0
- data/golden/input_reference/expected/snast.txt +21 -0
- data/golden/input_reference/expected.json +7 -0
- data/golden/input_reference/input.json +11 -0
- data/golden/input_reference/schema.kumi +22 -0
- data/golden/interleaved_fusion/expected/ast.txt +51 -0
- data/golden/interleaved_fusion/expected/input_plan.txt +13 -0
- data/golden/interleaved_fusion/expected/lir_00_unoptimized.txt +53 -0
- data/golden/interleaved_fusion/expected/lir_01_hoist_scalar_references.txt +53 -0
- data/golden/interleaved_fusion/expected/lir_02_inlined.txt +89 -0
- data/golden/interleaved_fusion/expected/lir_03_cse.txt +77 -0
- data/golden/interleaved_fusion/expected/lir_04_1_loop_fusion.txt +77 -0
- data/golden/interleaved_fusion/expected/lir_04_loop_invcm.txt +77 -0
- data/golden/interleaved_fusion/expected/lir_06_const_prop.txt +77 -0
- data/golden/interleaved_fusion/expected/nast.txt +41 -0
- data/golden/interleaved_fusion/expected/schema_javascript.mjs +86 -0
- data/golden/interleaved_fusion/expected/schema_ruby.rb +122 -0
- data/golden/interleaved_fusion/expected/snast.txt +41 -0
- data/golden/interleaved_fusion/expected.json +37 -0
- data/golden/interleaved_fusion/input.json +26 -0
- data/golden/interleaved_fusion/schema.kumi +57 -0
- data/golden/let_inline/expected/ast.txt +33 -0
- data/golden/let_inline/expected/input_plan.txt +2 -0
- data/golden/let_inline/expected/lir_00_unoptimized.txt +26 -0
- data/golden/let_inline/expected/lir_01_hoist_scalar_references.txt +26 -0
- data/golden/let_inline/expected/lir_02_inlined.txt +36 -0
- data/golden/let_inline/expected/lir_03_cse.txt +30 -0
- data/golden/let_inline/expected/lir_04_1_loop_fusion.txt +30 -0
- data/golden/let_inline/expected/lir_04_loop_invcm.txt +30 -0
- data/golden/let_inline/expected/lir_06_const_prop.txt +30 -0
- data/golden/let_inline/expected/nast.txt +26 -0
- data/golden/let_inline/expected/schema_javascript.mjs +11 -0
- data/golden/let_inline/expected/schema_ruby.rb +37 -0
- data/golden/let_inline/expected/snast.txt +26 -0
- data/golden/let_inline/expected.json +1 -0
- data/golden/let_inline/input.json +1 -0
- data/golden/let_inline/schema.kumi +11 -0
- data/golden/loop_fusion/expected/ast.txt +44 -0
- data/golden/loop_fusion/expected/input_plan.txt +13 -0
- data/golden/loop_fusion/expected/lir_00_unoptimized.txt +43 -0
- data/golden/loop_fusion/expected/lir_01_hoist_scalar_references.txt +43 -0
- data/golden/loop_fusion/expected/lir_02_inlined.txt +62 -0
- data/golden/loop_fusion/expected/lir_03_cse.txt +57 -0
- data/golden/loop_fusion/expected/lir_04_1_loop_fusion.txt +57 -0
- data/golden/loop_fusion/expected/lir_04_loop_invcm.txt +57 -0
- data/golden/loop_fusion/expected/lir_06_const_prop.txt +57 -0
- data/golden/loop_fusion/expected/nast.txt +32 -0
- data/golden/loop_fusion/expected/schema_javascript.mjs +64 -0
- data/golden/loop_fusion/expected/schema_ruby.rb +97 -0
- data/golden/loop_fusion/expected/snast.txt +32 -0
- data/golden/loop_fusion/expected.json +30 -0
- data/golden/loop_fusion/input.json +28 -0
- data/golden/loop_fusion/schema.kumi +32 -0
- data/golden/min_reduce_scope/expected/ast.txt +49 -0
- data/golden/min_reduce_scope/expected/input_plan.txt +9 -0
- data/golden/min_reduce_scope/expected/lir_00_unoptimized.txt +59 -0
- data/golden/min_reduce_scope/expected/lir_01_hoist_scalar_references.txt +59 -0
- data/golden/min_reduce_scope/expected/lir_02_inlined.txt +63 -0
- data/golden/min_reduce_scope/expected/lir_03_cse.txt +60 -0
- data/golden/min_reduce_scope/expected/lir_04_1_loop_fusion.txt +60 -0
- data/golden/min_reduce_scope/expected/lir_04_loop_invcm.txt +60 -0
- data/golden/min_reduce_scope/expected/lir_06_const_prop.txt +60 -0
- data/golden/min_reduce_scope/expected/nast.txt +29 -0
- data/golden/min_reduce_scope/expected/schema_javascript.mjs +66 -0
- data/golden/min_reduce_scope/expected/schema_ruby.rb +99 -0
- data/golden/min_reduce_scope/expected/snast.txt +29 -0
- data/golden/min_reduce_scope/expected.json +9 -0
- data/golden/min_reduce_scope/input.json +18 -0
- data/golden/min_reduce_scope/schema.kumi +25 -0
- data/golden/mixed_dimensions/expected/ast.txt +54 -0
- data/golden/mixed_dimensions/expected/input_plan.txt +15 -0
- data/golden/mixed_dimensions/expected/lir_00_unoptimized.txt +42 -0
- data/golden/mixed_dimensions/expected/lir_01_hoist_scalar_references.txt +42 -0
- data/golden/mixed_dimensions/expected/lir_02_inlined.txt +48 -0
- data/golden/mixed_dimensions/expected/lir_03_cse.txt +48 -0
- data/golden/mixed_dimensions/expected/lir_04_1_loop_fusion.txt +48 -0
- data/golden/mixed_dimensions/expected/lir_04_loop_invcm.txt +48 -0
- data/golden/mixed_dimensions/expected/lir_06_const_prop.txt +48 -0
- data/golden/mixed_dimensions/expected/nast.txt +22 -0
- data/golden/mixed_dimensions/expected/schema_javascript.mjs +53 -0
- data/golden/mixed_dimensions/expected/schema_ruby.rb +84 -0
- data/golden/mixed_dimensions/expected/snast.txt +22 -0
- data/golden/mixed_dimensions/expected.json +6 -0
- data/golden/mixed_dimensions/input.json +22 -0
- data/golden/mixed_dimensions/schema.kumi +35 -0
- data/golden/multirank_hoisting/expected/ast.txt +72 -0
- data/golden/multirank_hoisting/expected/input_plan.txt +18 -0
- data/golden/multirank_hoisting/expected/lir_00_unoptimized.txt +75 -0
- data/golden/multirank_hoisting/expected/lir_01_hoist_scalar_references.txt +75 -0
- data/golden/multirank_hoisting/expected/lir_02_inlined.txt +126 -0
- data/golden/multirank_hoisting/expected/lir_03_cse.txt +109 -0
- data/golden/multirank_hoisting/expected/lir_04_1_loop_fusion.txt +109 -0
- data/golden/multirank_hoisting/expected/lir_04_loop_invcm.txt +109 -0
- data/golden/multirank_hoisting/expected/lir_06_const_prop.txt +109 -0
- data/golden/multirank_hoisting/expected/nast.txt +39 -0
- data/golden/multirank_hoisting/expected/schema_javascript.mjs +128 -0
- data/golden/multirank_hoisting/expected/schema_ruby.rb +162 -0
- data/golden/multirank_hoisting/expected/snast.txt +39 -0
- data/golden/multirank_hoisting/expected.json +15 -0
- data/golden/multirank_hoisting/input.json +19 -0
- data/golden/multirank_hoisting/schema.kumi +38 -0
- data/golden/nested_hash/expected/ast.txt +22 -0
- data/golden/nested_hash/expected/input_plan.txt +5 -0
- data/golden/nested_hash/expected/lir_00_unoptimized.txt +10 -0
- data/golden/nested_hash/expected/lir_01_hoist_scalar_references.txt +10 -0
- data/golden/nested_hash/expected/lir_02_inlined.txt +10 -0
- data/golden/nested_hash/expected/lir_03_cse.txt +10 -0
- data/golden/nested_hash/expected/lir_04_1_loop_fusion.txt +10 -0
- data/golden/nested_hash/expected/lir_04_loop_invcm.txt +10 -0
- data/golden/nested_hash/expected/lir_06_const_prop.txt +10 -0
- data/golden/nested_hash/expected/nast.txt +8 -0
- data/golden/nested_hash/expected/schema_javascript.mjs +9 -0
- data/golden/nested_hash/expected/schema_ruby.rb +35 -0
- data/golden/nested_hash/expected/snast.txt +8 -0
- data/golden/nested_hash/expected.json +3 -0
- data/golden/nested_hash/input.json +7 -0
- data/golden/nested_hash/schema.kumi +11 -0
- data/golden/reduction_broadcast/expected/ast.txt +49 -0
- data/golden/reduction_broadcast/expected/input_plan.txt +13 -0
- data/golden/reduction_broadcast/expected/lir_00_unoptimized.txt +49 -0
- data/golden/reduction_broadcast/expected/lir_01_hoist_scalar_references.txt +49 -0
- data/golden/reduction_broadcast/expected/lir_02_inlined.txt +80 -0
- data/golden/reduction_broadcast/expected/lir_03_cse.txt +68 -0
- data/golden/reduction_broadcast/expected/lir_04_1_loop_fusion.txt +68 -0
- data/golden/reduction_broadcast/expected/lir_04_loop_invcm.txt +68 -0
- data/golden/reduction_broadcast/expected/lir_06_const_prop.txt +68 -0
- data/golden/reduction_broadcast/expected/nast.txt +24 -0
- data/golden/reduction_broadcast/expected/schema_javascript.mjs +74 -0
- data/golden/reduction_broadcast/expected/schema_ruby.rb +110 -0
- data/golden/reduction_broadcast/expected/snast.txt +24 -0
- data/golden/reduction_broadcast/expected.json +25 -0
- data/golden/reduction_broadcast/input.json +19 -0
- data/golden/reduction_broadcast/schema.kumi +34 -0
- data/golden/roll/expected/ast.txt +36 -0
- data/golden/roll/expected/input_plan.txt +3 -0
- data/golden/roll/expected/lir_00_unoptimized.txt +56 -0
- data/golden/roll/expected/lir_01_hoist_scalar_references.txt +56 -0
- data/golden/roll/expected/lir_02_inlined.txt +56 -0
- data/golden/roll/expected/lir_03_cse.txt +55 -0
- data/golden/roll/expected/lir_04_1_loop_fusion.txt +55 -0
- data/golden/roll/expected/lir_04_loop_invcm.txt +55 -0
- data/golden/roll/expected/lir_06_const_prop.txt +55 -0
- data/golden/roll/expected/nast.txt +26 -0
- data/golden/roll/expected/schema_javascript.mjs +65 -0
- data/golden/roll/expected/schema_ruby.rb +95 -0
- data/golden/roll/expected/snast.txt +26 -0
- data/golden/roll/expected.json +6 -0
- data/golden/roll/input.json +1 -0
- data/golden/roll/schema.kumi +13 -0
- data/golden/shift/expected/ast.txt +48 -0
- data/golden/shift/expected/input_plan.txt +3 -0
- data/golden/shift/expected/lir_00_unoptimized.txt +96 -0
- data/golden/shift/expected/lir_01_hoist_scalar_references.txt +96 -0
- data/golden/shift/expected/lir_02_inlined.txt +96 -0
- data/golden/shift/expected/lir_03_cse.txt +90 -0
- data/golden/shift/expected/lir_04_1_loop_fusion.txt +90 -0
- data/golden/shift/expected/lir_04_loop_invcm.txt +90 -0
- data/golden/shift/expected/lir_06_const_prop.txt +90 -0
- data/golden/shift/expected/nast.txt +38 -0
- data/golden/shift/expected/schema_javascript.mjs +106 -0
- data/golden/shift/expected/schema_ruby.rb +138 -0
- data/golden/shift/expected/snast.txt +38 -0
- data/golden/shift/expected.json +8 -0
- data/golden/shift/input.json +1 -0
- data/golden/shift/schema.kumi +15 -0
- data/golden/shift_2d/expected/ast.txt +88 -0
- data/golden/shift_2d/expected/input_plan.txt +5 -0
- data/golden/shift_2d/expected/lir_00_unoptimized.txt +274 -0
- data/golden/shift_2d/expected/lir_01_hoist_scalar_references.txt +274 -0
- data/golden/shift_2d/expected/lir_02_inlined.txt +274 -0
- data/golden/shift_2d/expected/lir_03_cse.txt +262 -0
- data/golden/shift_2d/expected/lir_04_1_loop_fusion.txt +262 -0
- data/golden/shift_2d/expected/lir_04_loop_invcm.txt +262 -0
- data/golden/shift_2d/expected/lir_06_const_prop.txt +262 -0
- data/golden/shift_2d/expected/nast.txt +74 -0
- data/golden/shift_2d/expected/schema_javascript.mjs +320 -0
- data/golden/shift_2d/expected/schema_ruby.rb +358 -0
- data/golden/shift_2d/expected/snast.txt +74 -0
- data/golden/shift_2d/expected.json +15 -0
- data/golden/shift_2d/input.json +14 -0
- data/golden/shift_2d/schema.kumi +25 -0
- data/golden/simple_math/expected/ast.txt +40 -0
- data/golden/simple_math/expected/input_plan.txt +2 -0
- data/golden/simple_math/expected/lir_00_unoptimized.txt +31 -0
- data/golden/simple_math/expected/lir_01_hoist_scalar_references.txt +31 -0
- data/golden/simple_math/expected/lir_02_inlined.txt +31 -0
- data/golden/simple_math/expected/lir_03_cse.txt +31 -0
- data/golden/simple_math/expected/lir_04_1_loop_fusion.txt +31 -0
- data/golden/simple_math/expected/lir_04_loop_invcm.txt +31 -0
- data/golden/simple_math/expected/lir_06_const_prop.txt +31 -0
- data/golden/simple_math/expected/nast.txt +33 -0
- data/golden/simple_math/expected/schema_javascript.mjs +33 -0
- data/golden/simple_math/expected/schema_ruby.rb +59 -0
- data/golden/simple_math/expected/snast.txt +33 -0
- data/golden/simple_math/expected.json +1 -0
- data/golden/simple_math/input.json +4 -0
- data/golden/simple_math/schema.kumi +2 -1
- data/golden/streaming_basics/expected/ast.txt +64 -0
- data/golden/streaming_basics/expected/input_plan.txt +8 -0
- data/golden/streaming_basics/expected/lir_00_unoptimized.txt +73 -0
- data/golden/streaming_basics/expected/lir_01_hoist_scalar_references.txt +73 -0
- data/golden/streaming_basics/expected/lir_02_inlined.txt +100 -0
- data/golden/streaming_basics/expected/lir_03_cse.txt +84 -0
- data/golden/streaming_basics/expected/lir_04_1_loop_fusion.txt +84 -0
- data/golden/streaming_basics/expected/lir_04_loop_invcm.txt +84 -0
- data/golden/streaming_basics/expected/lir_06_const_prop.txt +84 -0
- data/golden/streaming_basics/expected/nast.txt +48 -0
- data/golden/streaming_basics/expected/schema_javascript.mjs +94 -0
- data/golden/streaming_basics/expected/schema_ruby.rb +127 -0
- data/golden/streaming_basics/expected/snast.txt +48 -0
- data/golden/streaming_basics/expected.json +10 -0
- data/golden/streaming_basics/input.json +8 -0
- data/golden/streaming_basics/schema.kumi +24 -0
- data/golden/tuples/expected/ast.txt +48 -0
- data/golden/tuples/expected/input_plan.txt +1 -0
- data/golden/tuples/expected/lir_00_unoptimized.txt +40 -0
- data/golden/tuples/expected/lir_01_hoist_scalar_references.txt +40 -0
- data/golden/tuples/expected/lir_02_inlined.txt +48 -0
- data/golden/tuples/expected/lir_03_cse.txt +48 -0
- data/golden/tuples/expected/lir_04_1_loop_fusion.txt +48 -0
- data/golden/tuples/expected/lir_04_loop_invcm.txt +48 -0
- data/golden/tuples/expected/lir_06_const_prop.txt +48 -0
- data/golden/tuples/expected/nast.txt +42 -0
- data/golden/tuples/expected/schema_javascript.mjs +51 -0
- data/golden/tuples/expected/schema_ruby.rb +77 -0
- data/golden/tuples/expected/snast.txt +42 -0
- data/golden/tuples/expected.json +7 -0
- data/golden/tuples/input.json +3 -0
- data/golden/tuples/schema.kumi +11 -0
- data/golden/tuples_and_arrays/expected/ast.txt +44 -0
- data/golden/tuples_and_arrays/expected/input_plan.txt +7 -0
- data/golden/tuples_and_arrays/expected/lir_00_unoptimized.txt +41 -0
- data/golden/tuples_and_arrays/expected/lir_01_hoist_scalar_references.txt +41 -0
- data/golden/tuples_and_arrays/expected/lir_02_inlined.txt +62 -0
- data/golden/tuples_and_arrays/expected/lir_03_cse.txt +51 -0
- data/golden/tuples_and_arrays/expected/lir_04_1_loop_fusion.txt +51 -0
- data/golden/tuples_and_arrays/expected/lir_04_loop_invcm.txt +51 -0
- data/golden/tuples_and_arrays/expected/lir_06_const_prop.txt +51 -0
- data/golden/tuples_and_arrays/expected/nast.txt +28 -0
- data/golden/tuples_and_arrays/expected/schema_javascript.mjs +58 -0
- data/golden/tuples_and_arrays/expected/schema_ruby.rb +88 -0
- data/golden/tuples_and_arrays/expected/snast.txt +28 -0
- data/golden/tuples_and_arrays/expected.json +18 -0
- data/golden/tuples_and_arrays/input.json +7 -0
- data/golden/tuples_and_arrays/schema.kumi +38 -0
- data/golden/with_constants/expected/ast.txt +28 -0
- data/golden/with_constants/expected/input_plan.txt +2 -0
- data/golden/with_constants/expected/lir_00_unoptimized.txt +17 -0
- data/golden/with_constants/expected/lir_01_hoist_scalar_references.txt +17 -0
- data/golden/with_constants/expected/lir_02_inlined.txt +17 -0
- data/golden/with_constants/expected/lir_03_cse.txt +17 -0
- data/golden/with_constants/expected/lir_04_1_loop_fusion.txt +17 -0
- data/golden/with_constants/expected/lir_04_loop_invcm.txt +17 -0
- data/golden/with_constants/expected/lir_06_const_prop.txt +17 -0
- data/golden/with_constants/expected/nast.txt +21 -0
- data/golden/with_constants/expected/schema_javascript.mjs +18 -0
- data/golden/with_constants/expected/schema_ruby.rb +44 -0
- data/golden/with_constants/expected/snast.txt +15 -0
- data/golden/with_constants/schema.kumi +10 -0
- data/lib/kumi/analyzer.rb +76 -24
- data/lib/kumi/configuration.rb +60 -0
- data/lib/kumi/core/analyzer/binder.rb +121 -0
- data/lib/kumi/core/analyzer/checkpoint.rb +15 -9
- data/lib/kumi/core/analyzer/constant_evaluator.rb +32 -37
- data/lib/kumi/core/analyzer/constant_folding_helpers.rb +55 -0
- data/lib/kumi/core/analyzer/debug.rb +14 -16
- data/lib/kumi/core/analyzer/fn_aliases.rb +46 -0
- data/lib/kumi/core/analyzer/folder.rb +94 -0
- data/lib/kumi/core/analyzer/macro_expander.rb +69 -0
- data/lib/kumi/core/analyzer/passes/assemble_irv2_pass.rb +130 -0
- data/lib/kumi/core/analyzer/passes/attach_anchors_pass.rb +71 -0
- data/lib/kumi/core/analyzer/passes/attach_terminal_info_pass.rb +181 -0
- data/lib/kumi/core/analyzer/passes/codegen/js/declaration_emitter.rb +317 -0
- data/lib/kumi/core/analyzer/passes/codegen/js/emitter.rb +75 -0
- data/lib/kumi/core/analyzer/passes/codegen/js/output_buffer.rb +103 -0
- data/lib/kumi/core/analyzer/passes/codegen/js_pass.rb +27 -0
- data/lib/kumi/core/analyzer/passes/codegen/ruby/declaration_emitter.rb +321 -0
- data/lib/kumi/core/analyzer/passes/codegen/ruby/emitter.rb +85 -0
- data/lib/kumi/core/analyzer/passes/codegen/ruby/output_buffer.rb +111 -0
- data/lib/kumi/core/analyzer/passes/codegen/ruby_pass.rb +32 -0
- data/lib/kumi/core/analyzer/passes/constant_folding_pass.rb +33 -0
- data/lib/kumi/core/analyzer/passes/contract_checker_pass.rb +220 -0
- data/lib/kumi/core/analyzer/passes/dependency_resolver.rb +5 -6
- data/lib/kumi/core/analyzer/passes/input_access_planner_pass.rb +7 -3
- data/lib/kumi/core/analyzer/passes/input_collector.rb +74 -122
- data/lib/kumi/core/analyzer/passes/input_form_schema_pass.rb +43 -0
- data/lib/kumi/core/analyzer/passes/ir_dependency_pass.rb +1 -1
- data/lib/kumi/core/analyzer/passes/ir_execution_schedule_pass.rb +1 -1
- data/lib/kumi/core/analyzer/passes/join_reduce_planning_pass.rb +21 -23
- data/lib/kumi/core/analyzer/passes/lir/constant_propagation_pass.rb +84 -0
- data/lib/kumi/core/analyzer/passes/lir/dead_code_elimination_pass.rb +93 -0
- data/lib/kumi/core/analyzer/passes/lir/hoist_scalar_references_pass.rb +115 -0
- data/lib/kumi/core/analyzer/passes/lir/inline_declarations_pass.rb +274 -0
- data/lib/kumi/core/analyzer/passes/lir/instruction_scheduling_pass.rb +198 -0
- data/lib/kumi/core/analyzer/passes/lir/kernel_binding_pass.rb +30 -0
- data/lib/kumi/core/analyzer/passes/lir/local_cse_pass.rb +121 -0
- data/lib/kumi/core/analyzer/passes/lir/loop_fusion_pass.rb +156 -0
- data/lib/kumi/core/analyzer/passes/lir/loop_invariant_code_motion_pass.rb +148 -0
- data/lib/kumi/core/analyzer/passes/lir/lower_pass.rb +407 -0
- data/lib/kumi/core/analyzer/passes/lir/stencil_emitter.rb +243 -0
- data/lib/kumi/core/analyzer/passes/lir/validation_pass.rb +83 -0
- data/lib/kumi/core/analyzer/passes/load_input_cse.rb +17 -16
- data/lib/kumi/core/analyzer/passes/lower_to_ir_pass.rb +4 -2
- data/lib/kumi/core/analyzer/passes/lower_to_irv2_pass.rb +197 -0
- data/lib/kumi/core/analyzer/passes/name_indexer.rb +4 -1
- data/lib/kumi/core/analyzer/passes/nast_dimensional_analyzer_pass.rb +237 -0
- data/lib/kumi/core/analyzer/passes/normalize_to_nast_pass.rb +156 -0
- data/lib/kumi/core/analyzer/passes/output_schema_pass.rb +38 -0
- data/lib/kumi/core/analyzer/passes/pass_base.rb +30 -4
- data/lib/kumi/core/analyzer/passes/precompute_access_paths_pass.rb +93 -0
- data/lib/kumi/core/analyzer/passes/scope_resolution_pass.rb +33 -30
- data/lib/kumi/core/analyzer/passes/semantic_constraint_validator.rb +5 -26
- data/lib/kumi/core/analyzer/passes/snast_pass.rb +229 -0
- data/lib/kumi/core/analyzer/passes/toposorter.rb +1 -0
- data/lib/kumi/core/analyzer/passes/type_checker.rb +13 -0
- data/lib/kumi/core/analyzer/passes/type_inferencer_pass.rb +14 -16
- data/lib/kumi/core/analyzer/passes/unsat_detector.rb +2 -1
- data/lib/kumi/core/analyzer/plans.rb +12 -0
- data/lib/kumi/core/analyzer/state_serde.rb +4 -4
- data/lib/kumi/core/analyzer/structs/access_plan.rb +2 -2
- data/lib/kumi/core/analyzer/unsat_constant_evaluator.rb +59 -0
- data/lib/kumi/core/compiler/access_codegen.rb +1 -1
- data/lib/kumi/core/compiler/access_emit/base.rb +9 -5
- data/lib/kumi/core/compiler/access_emit/each_indexed.rb +10 -3
- data/lib/kumi/core/compiler/access_emit/materialize.rb +12 -5
- data/lib/kumi/core/compiler/access_emit/ravel.rb +10 -3
- data/lib/kumi/core/compiler/access_emit/read.rb +4 -1
- data/lib/kumi/core/compiler/access_planner.rb +42 -4
- data/lib/kumi/core/compiler/access_planner_v2.rb +164 -0
- data/lib/kumi/core/explain.rb +18 -11
- data/lib/kumi/core/functions/loader.rb +47 -0
- data/lib/kumi/core/functions/model.rb +10 -0
- data/lib/kumi/core/functions/type_rules.rb +108 -0
- data/lib/kumi/core/ir/execution_engine/interpreter.rb +1 -1
- data/lib/kumi/core/ir/execution_engine/profiler.rb +107 -97
- data/lib/kumi/core/ir.rb +6 -2
- data/lib/kumi/core/irv2/builder.rb +48 -0
- data/lib/kumi/core/irv2/declaration.rb +28 -0
- data/lib/kumi/core/irv2/module.rb +108 -0
- data/lib/kumi/core/irv2/value.rb +28 -0
- data/lib/kumi/core/lir/analyze.rb +64 -0
- data/lib/kumi/core/lir/build.rb +363 -0
- data/lib/kumi/core/lir/emit.rb +62 -0
- data/lib/kumi/core/lir/structs/instruction.rb +44 -0
- data/lib/kumi/core/lir/structs/literal.rb +14 -0
- data/lib/kumi/core/lir/structs/stamp.rb +13 -0
- data/lib/kumi/core/lir/support/error.rb +9 -0
- data/lib/kumi/core/lir/support/ids.rb +34 -0
- data/lib/kumi/core/lir/validate.rb +74 -0
- data/lib/kumi/core/lir.rb +34 -0
- data/lib/kumi/core/nast.rb +216 -0
- data/lib/kumi/core/ruby_parser/dsl.rb +2 -2
- data/lib/kumi/core/ruby_parser/dsl_cascade_builder.rb +1 -1
- data/lib/kumi/core/ruby_parser/expression_converter.rb +10 -0
- data/lib/kumi/core/ruby_parser/input_builder.rb +16 -17
- data/lib/kumi/core/ruby_parser/schema_builder.rb +36 -10
- data/lib/kumi/core/ruby_parser/sugar.rb +64 -0
- data/lib/kumi/core/types/builder.rb +2 -2
- data/lib/kumi/core/types/normalizer.rb +2 -0
- data/lib/kumi/core/types/validator.rb +10 -3
- data/lib/kumi/core/types.rb +7 -0
- data/lib/kumi/dev/codegen.rb +194 -0
- data/lib/kumi/dev/golden/generator.rb +105 -0
- data/lib/kumi/dev/golden/reporter.rb +160 -0
- data/lib/kumi/dev/golden/representation.rb +45 -0
- data/lib/kumi/dev/golden/result.rb +98 -0
- data/lib/kumi/dev/golden/runtime_test.rb +93 -0
- data/lib/kumi/dev/golden/suite.rb +131 -0
- data/lib/kumi/dev/golden/verifier.rb +76 -0
- data/lib/kumi/dev/golden.rb +75 -0
- data/lib/kumi/dev/ir.rb +4 -4
- data/lib/kumi/dev/parse.rb +1 -1
- data/lib/kumi/dev/pretty_printer.rb +229 -0
- data/lib/kumi/dev/printer/irv2_formatter.rb +163 -0
- data/lib/kumi/dev/printer/width_aware_json.rb +44 -0
- data/lib/kumi/dev/profile_aggregator.rb +36 -38
- data/lib/kumi/dev/profile_runner.rb +19 -23
- data/lib/kumi/dev/runner.rb +12 -22
- data/lib/kumi/dev/support/kumi_runner.mjs +39 -0
- data/lib/kumi/dev.rb +3 -3
- data/lib/kumi/frontends/ruby.rb +12 -12
- data/lib/kumi/frontends/text.rb +26 -21
- data/lib/kumi/frontends.rb +6 -8
- data/lib/kumi/kernel_registry.rb +59 -0
- data/lib/kumi/pack/builder.rb +229 -0
- data/lib/kumi/pack.rb +15 -0
- data/lib/kumi/registry_v2/loader.rb +81 -0
- data/lib/kumi/registry_v2.rb +118 -0
- data/lib/kumi/schema.rb +83 -43
- data/lib/kumi/support/diff.rb +3 -2
- data/lib/kumi/support/ir_render.rb +1 -1
- data/lib/kumi/support/lir_printer.rb +143 -0
- data/lib/kumi/support/nast_printer.rb +93 -0
- data/lib/kumi/support/s_expression_printer.rb +5 -4
- data/lib/kumi/support/snast_printer.rb +111 -0
- data/lib/kumi/syntax/call_expression.rb +1 -1
- data/lib/kumi/syntax/input_declaration.rb +2 -2
- data/lib/kumi/syntax/node.rb +4 -12
- data/lib/kumi/syntax/root.rb +10 -0
- data/lib/kumi/syntax/trait_declaration.rb +1 -0
- data/lib/kumi/syntax/value_declaration.rb +1 -0
- data/lib/kumi/version.rb +1 -1
- data/lib/kumi.rb +47 -5
- metadata +571 -54
- data/docs/AST.md +0 -133
- data/docs/DSL.md +0 -154
- data/docs/FUNCTIONS.md +0 -176
- data/docs/VECTOR_SEMANTICS.md +0 -286
- data/docs/compiler_design_principles.md +0 -86
- data/docs/dev/analyzer-debug.md +0 -52
- data/docs/dev/parse-command.md +0 -64
- data/docs/dev/vm-profiling.md +0 -95
- data/docs/development/README.md +0 -120
- data/docs/development/error-reporting.md +0 -361
- data/docs/features/README.md +0 -58
- data/docs/features/analysis-type-inference.md +0 -42
- data/docs/features/analysis-unsat-detection.md +0 -71
- data/docs/features/hierarchical-broadcasting.md +0 -415
- data/docs/features/input-declaration-system.md +0 -58
- data/docs/features/performance.md +0 -14
- data/docs/features/s-expression-printer.md +0 -77
- data/docs/schema_metadata/broadcasts.md +0 -53
- data/docs/schema_metadata/cascades.md +0 -45
- data/docs/schema_metadata/declarations.md +0 -54
- data/docs/schema_metadata/dependencies.md +0 -57
- data/docs/schema_metadata/evaluation_order.md +0 -29
- data/docs/schema_metadata/examples.md +0 -95
- data/docs/schema_metadata/inferred_types.md +0 -46
- data/docs/schema_metadata/inputs.md +0 -86
- data/docs/schema_metadata.md +0 -108
- data/golden/mixed_nesting/schema.kumi +0 -42
- data/lib/kumi/core/function_registry/collection_functions.rb +0 -298
- data/lib/kumi/core/function_registry/comparison_functions.rb +0 -33
- data/lib/kumi/core/function_registry/conditional_functions.rb +0 -48
- data/lib/kumi/core/function_registry/function_builder.rb +0 -184
- data/lib/kumi/core/function_registry/logical_functions.rb +0 -214
- data/lib/kumi/core/function_registry/math_functions.rb +0 -74
- data/lib/kumi/core/function_registry/stat_functions.rb +0 -156
- data/lib/kumi/core/function_registry/string_functions.rb +0 -57
- data/lib/kumi/core/function_registry/type_functions.rb +0 -53
- data/lib/kumi/core/function_registry.rb +0 -176
- data/lib/kumi/registry.rb +0 -32
- data/lib/kumi/runtime/executable.rb +0 -135
- data/lib/kumi/runtime/run.rb +0 -105
- data/performance_results.txt +0 -63
- data/scripts/analyze_broadcast_methods.rb +0 -68
- data/scripts/analyze_cascade_methods.rb +0 -74
- data/scripts/check_broadcasting_coverage.rb +0 -51
- data/scripts/find_dead_code.rb +0 -114
- data/scripts/generate_function_docs.rb +0 -71
- data/scripts/test_mixed_nesting_performance.rb +0 -206
@@ -0,0 +1,407 @@
|
|
1
|
+
module Kumi
|
2
|
+
module Core
|
3
|
+
module Analyzer
|
4
|
+
module Passes
|
5
|
+
module LIR
|
6
|
+
class LowerPass < PassBase
|
7
|
+
include StencilEmitter
|
8
|
+
|
9
|
+
NAST = Kumi::Core::NAST
|
10
|
+
Ids = Kumi::Core::LIR::Ids
|
11
|
+
Build = Kumi::Core::LIR::Build
|
12
|
+
Emit = Kumi::Core::LIR::Emit
|
13
|
+
|
14
|
+
def run(_errors)
|
15
|
+
@snast = get_state(:snast_module, required: true)
|
16
|
+
@registry = get_state(:registry, required: true)
|
17
|
+
@ids = Ids.new
|
18
|
+
@anchors = get_state(:anchor_by_decl, required: true)
|
19
|
+
@pre = get_state(:precomputed_plan_by_fqn, required: true)
|
20
|
+
plans = get_state(:input_table, required: true)
|
21
|
+
@plans_by_fqn = plans.each_with_object({}) { |p, h| h[p.path_fqn.to_s] = p }
|
22
|
+
|
23
|
+
ops_by_decl = {}
|
24
|
+
@snast.decls.each do |name, decl|
|
25
|
+
@ops = []
|
26
|
+
@env = Env.new
|
27
|
+
@current_decl = decl
|
28
|
+
lower_declaration(decl)
|
29
|
+
ops_by_decl[name] = { operations: @ops }
|
30
|
+
end
|
31
|
+
|
32
|
+
ops_by_decl.freeze
|
33
|
+
state.with(:lir_module, ops_by_decl)
|
34
|
+
.with(:lir_00_unoptimized, ops_by_decl)
|
35
|
+
.with(:id_generator, @ids)
|
36
|
+
end
|
37
|
+
|
38
|
+
private
|
39
|
+
|
40
|
+
Env = Struct.new(:frames, :memo, keyword_init: true) do
|
41
|
+
def initialize(**) = super(frames: [], memo: Hash.new { |h, k| h[k] = {} })
|
42
|
+
def axes = frames.map { _1[:axis] }
|
43
|
+
def loop_ids = frames.map { _1[:id] }
|
44
|
+
def element_reg_for(axis) = frames.reverse.find { _1[:axis] == axis }&.dig(:as_element)
|
45
|
+
def index_reg_for(axis) = frames.reverse.find { _1[:axis] == axis }&.dig(:as_index)
|
46
|
+
def collection_reg_for(axis) = frames.reverse.find { _1[:axis] == axis }&.dig(:collection)
|
47
|
+
def push(frame) = frames << frame
|
48
|
+
def pop = frames.pop
|
49
|
+
def depth = frames.length
|
50
|
+
def memo_get(cat, key) = memo[cat][key]
|
51
|
+
def memo_set(cat, key, val) = memo[cat][key] = val
|
52
|
+
def invalidate_after_depth!(_d); end
|
53
|
+
end
|
54
|
+
|
55
|
+
# ---------- declarations ----------
|
56
|
+
|
57
|
+
def lower_declaration(decl)
|
58
|
+
wanted = axes_of(decl)
|
59
|
+
wanted = axes_of(decl.body) if wanted.empty?
|
60
|
+
|
61
|
+
if wanted.empty?
|
62
|
+
close_loops_to_depth(0)
|
63
|
+
else
|
64
|
+
ensure_context_for!(wanted, anchor_fqn: anchor_fqn_from_node!(decl.body, need_prefix: wanted))
|
65
|
+
end
|
66
|
+
|
67
|
+
@emit = Emit.new(registry: @registry, ids: @ids, ops: @ops)
|
68
|
+
res = lower_expr(decl.body)
|
69
|
+
@ops << Build.yield(result_register: res)
|
70
|
+
ensure
|
71
|
+
close_loops_to_depth(0)
|
72
|
+
end
|
73
|
+
|
74
|
+
# ---------- expressions ----------
|
75
|
+
|
76
|
+
def lower_expr(node)
|
77
|
+
case node
|
78
|
+
when NAST::Const then emit_const(node)
|
79
|
+
when NAST::InputRef then emit_input_ref(node)
|
80
|
+
when NAST::Ref then emit_ref(node)
|
81
|
+
when NAST::Tuple then emit_tuple(node)
|
82
|
+
when NAST::Select then emit_select(node)
|
83
|
+
when NAST::Fold then emit_fold(node)
|
84
|
+
when NAST::Reduce then emit_reduce(node)
|
85
|
+
when NAST::Call then call_emit_selection(node)
|
86
|
+
when NAST::Hash then emit_hash(node)
|
87
|
+
when NAST::IndexRef then emit_index_ref(node)
|
88
|
+
else raise "unknown node #{node.class}"
|
89
|
+
end
|
90
|
+
end
|
91
|
+
|
92
|
+
def call_emit_selection(n)
|
93
|
+
return emit_roll(n) if n.fn == :roll
|
94
|
+
return emit_shift(n) if n.fn == :shift
|
95
|
+
|
96
|
+
emit_call(n)
|
97
|
+
end
|
98
|
+
|
99
|
+
def emit_const(n)
|
100
|
+
ins = Build.constant(value: n.value, dtype: dtype_of(n), ids: @ids)
|
101
|
+
@ops << ins
|
102
|
+
ins.result_register
|
103
|
+
end
|
104
|
+
|
105
|
+
def emit_ref(n)
|
106
|
+
ins = Build.load_declaration(name: n.name, dtype: dtype_of(n), axes: axes_of(n), ids: @ids)
|
107
|
+
@ops << ins
|
108
|
+
ins.result_register
|
109
|
+
end
|
110
|
+
|
111
|
+
def emit_tuple(n)
|
112
|
+
regs = n.args.map { lower_expr(_1) }
|
113
|
+
ins = Build.make_tuple(elements: regs, out_dtype: dtype_of(n), ids: @ids)
|
114
|
+
@ops << ins
|
115
|
+
ins.result_register
|
116
|
+
end
|
117
|
+
|
118
|
+
def emit_hash(n)
|
119
|
+
keys = []
|
120
|
+
vals = []
|
121
|
+
n.pairs.each do |p|
|
122
|
+
keys << p.key
|
123
|
+
vals << lower_expr(p.value)
|
124
|
+
end
|
125
|
+
ins = Build.make_object(keys:, values: vals, ids: @ids)
|
126
|
+
@ops << ins
|
127
|
+
ins.result_register
|
128
|
+
end
|
129
|
+
|
130
|
+
def emit_call(n)
|
131
|
+
regs = n.args.map { lower_expr(_1) }
|
132
|
+
ins = Build.kernel_call(
|
133
|
+
function: @registry.resolve_function(n.fn),
|
134
|
+
args: regs,
|
135
|
+
out_dtype: dtype_of(n),
|
136
|
+
ids: @ids
|
137
|
+
)
|
138
|
+
@ops << ins
|
139
|
+
ins.result_register
|
140
|
+
end
|
141
|
+
|
142
|
+
def emit_select(n)
|
143
|
+
ax = axes_of(n)
|
144
|
+
ensure_context_for!(ax, anchor_fqn: anchor_fqn_from_node!(n, need_prefix: ax)) unless ax.empty?
|
145
|
+
c = lower_expr(n.cond)
|
146
|
+
t = lower_expr(n.on_true)
|
147
|
+
f = lower_expr(n.on_false)
|
148
|
+
ins = Build.select(cond: c, on_true: t, on_false: f, out_dtype: dtype_of(n), ids: @ids)
|
149
|
+
@ops << ins
|
150
|
+
ins.result_register
|
151
|
+
end
|
152
|
+
|
153
|
+
def emit_fold(n)
|
154
|
+
arg = lower_expr(n.arg)
|
155
|
+
ins = Build.fold(arg:, function: @registry.resolve_function(n.fn), out_dtype: dtype_of(n), ids: @ids)
|
156
|
+
@ops << ins
|
157
|
+
ins.result_register
|
158
|
+
end
|
159
|
+
|
160
|
+
def emit_reduce(n)
|
161
|
+
out_axes = axes_of(n)
|
162
|
+
in_axes = axes_of(n.arg)
|
163
|
+
function = @registry.resolve_function(n.fn)
|
164
|
+
raise "reduce: scalar input" if in_axes.empty?
|
165
|
+
raise "reduce: axes(arg)=#{in_axes} must equal out+over" unless in_axes == out_axes + Array(n.over)
|
166
|
+
|
167
|
+
ensure_context_for!(out_axes, anchor_fqn: anchor_fqn_from_node!(n.arg, need_prefix: out_axes))
|
168
|
+
|
169
|
+
dtype = dtype_of(n)
|
170
|
+
acc_name = @ids.generate_temp(prefix: :acc_)
|
171
|
+
@ops << Build.declare_accumulator(name: acc_name, dtype: dtype, ids: @ids)
|
172
|
+
|
173
|
+
open_suffix_loops!(over_axes: Array(n.over), anchor_fqn: anchor_fqn_from_node!(n.arg, need_prefix: in_axes))
|
174
|
+
|
175
|
+
val = lower_expr(n.arg)
|
176
|
+
@ops << Build.accumulate(accumulator: acc_name, dtype: dtype, function: function, value_register: val)
|
177
|
+
|
178
|
+
close_loops_to_depth(out_axes.length)
|
179
|
+
ins = Build.load_accumulator(accumulator: acc_name, dtype: dtype, ids: @ids)
|
180
|
+
@ops << ins
|
181
|
+
ins.result_register
|
182
|
+
end
|
183
|
+
|
184
|
+
# ---------- InputRef ----------
|
185
|
+
|
186
|
+
def emit_input_ref(n)
|
187
|
+
axes = axes_of(n)
|
188
|
+
keys = ir_key_chain(n)
|
189
|
+
|
190
|
+
if axes.empty?
|
191
|
+
# root access by explicit keys
|
192
|
+
raise "root access needs key_chain" if keys.empty?
|
193
|
+
|
194
|
+
head_dt = (keys.length == 1 ? dtype_of(n) : :any)
|
195
|
+
cur = Build.load_input(key: keys.first.to_sym, dtype: head_dt, ids: @ids).tap { @ops << _1 }.result_register
|
196
|
+
keys.drop(1).each_with_index do |k, i|
|
197
|
+
last = (i == keys.length - 2)
|
198
|
+
dt = last ? dtype_of(n) : :any
|
199
|
+
cur = Build.load_field(object_register: cur, key: k.to_sym, dtype: dt, ids: @ids).tap { @ops << _1 }.result_register
|
200
|
+
end
|
201
|
+
return cur
|
202
|
+
end
|
203
|
+
|
204
|
+
# inside loops: start from current element for innermost axis
|
205
|
+
cur = @env.element_reg_for(axes.last) or raise "no open element for axis #{axes.last.inspect}"
|
206
|
+
keys.each_with_index do |k, i|
|
207
|
+
last = (i == keys.length - 1)
|
208
|
+
dt = last ? dtype_of(n) : :any
|
209
|
+
cur = Build.load_field(object_register: cur, key: k.to_sym, dtype: dt, ids: @ids).tap { @ops << _1 }.result_register
|
210
|
+
end
|
211
|
+
cur
|
212
|
+
end
|
213
|
+
|
214
|
+
def emit_index_ref(n)
|
215
|
+
ax = axes_of(n)
|
216
|
+
raise "index ref without axes" if ax.empty?
|
217
|
+
|
218
|
+
debug "emit_index_ref: name=#{n.name}, axes=#{ax.inspect}, input_fqn=#{n.input_fqn}, env.axes=#{@env.axes.inspect}"
|
219
|
+
|
220
|
+
# IndexRef nodes reference an index that should already be in scope.
|
221
|
+
# The index refers to the LAST axis in the IndexRef's own stamp.
|
222
|
+
# We do NOT open loops for broadcast axes - those loops should already be open
|
223
|
+
# from other parts of the expression.
|
224
|
+
target_axis = ax.last
|
225
|
+
|
226
|
+
# The index variable is defined by the array that introduced this axis.
|
227
|
+
idx = @env.index_reg_for(target_axis) or raise "no index register for axis #{target_axis.inspect}"
|
228
|
+
|
229
|
+
debug "emit_index_ref: returning index register #{idx.inspect} for axis #{target_axis}"
|
230
|
+
|
231
|
+
# Index is an integer scalar elementwise over the current axes ⇒ just return the register.
|
232
|
+
idx
|
233
|
+
end
|
234
|
+
|
235
|
+
# ---------- context management ----------
|
236
|
+
|
237
|
+
def ensure_context_for!(target_axes, anchor_fqn:)
|
238
|
+
l = lcp(@env.axes, target_axes).length
|
239
|
+
close_loops_to_depth(l)
|
240
|
+
missing = target_axes[l..] || []
|
241
|
+
return if missing.empty?
|
242
|
+
|
243
|
+
open_suffix_loops!(over_axes: missing, anchor_fqn: anchor_fqn)
|
244
|
+
end
|
245
|
+
|
246
|
+
def open_suffix_loops!(over_axes:, anchor_fqn:)
|
247
|
+
return if over_axes.empty?
|
248
|
+
|
249
|
+
target_axes = @env.axes + over_axes
|
250
|
+
pre = @pre.fetch(anchor_fqn) { raise "no precomputed plan for #{anchor_fqn}" }
|
251
|
+
|
252
|
+
axis_to_loop = pre[:axis_to_loop] || {}
|
253
|
+
idxs = target_axes.map { |ax| axis_to_loop.fetch(ax) { raise "plan #{anchor_fqn} lacks axis #{ax.inspect}" } }
|
254
|
+
|
255
|
+
base = @env.axes.length
|
256
|
+
(base...target_axes.length).each do |i|
|
257
|
+
axis = target_axes[i]
|
258
|
+
li = idxs[i]
|
259
|
+
|
260
|
+
coll =
|
261
|
+
if i == 0 && base == 0
|
262
|
+
head_collection(pre, li)
|
263
|
+
else
|
264
|
+
prev_axis = target_axes[i - 1]
|
265
|
+
prev_el = @env.element_reg_for(prev_axis) or raise "no element for #{prev_axis}"
|
266
|
+
prev_li = idxs[i - 1]
|
267
|
+
between_loops(pre, prev_li, li, prev_el)
|
268
|
+
end
|
269
|
+
|
270
|
+
el = @ids.generate_temp(prefix: :"#{axis}_el_")
|
271
|
+
ix = @ids.generate_temp(prefix: :"#{axis}_i_")
|
272
|
+
lid = @ids.generate_loop_id
|
273
|
+
@ops << Build.loop_start(collection_register: coll, axis: axis, as_element: el, as_index: ix, id: lid)
|
274
|
+
@env.push(axis: axis, as_element: el, as_index: ix, id: lid, collection: coll)
|
275
|
+
end
|
276
|
+
end
|
277
|
+
|
278
|
+
def close_loops_to_depth(depth)
|
279
|
+
while @env.depth > depth
|
280
|
+
@env.pop
|
281
|
+
@ops << Build.loop_end
|
282
|
+
end
|
283
|
+
@env.invalidate_after_depth!(depth)
|
284
|
+
end
|
285
|
+
|
286
|
+
# ---------- small utils ----------
|
287
|
+
|
288
|
+
def axes_of(n) = Array(n.meta[:stamp]&.dig(:axes))
|
289
|
+
def dtype_of(n) = n.meta[:stamp]&.dig(:dtype)
|
290
|
+
|
291
|
+
def lcp(a, b)
|
292
|
+
i = 0
|
293
|
+
i += 1 while i < a.size && i < b.size && a[i] == b[i]
|
294
|
+
a[0...i]
|
295
|
+
end
|
296
|
+
|
297
|
+
def ir_fqn(n) = n.instance_variable_get(:@fqn) || n.path.join(".")
|
298
|
+
def ir_key_chain(n) = Array(n.instance_variable_get(:@key_chain))
|
299
|
+
def plan_for_fqn(fqn) = @plans_by_fqn.fetch(fqn) { raise "no InputPlan for #{fqn}" }
|
300
|
+
|
301
|
+
def anchor_fqn_from_node!(node, need_prefix:)
|
302
|
+
ir = find_anchor_inputref(node, need_prefix: need_prefix)
|
303
|
+
return ir if ir.is_a? String
|
304
|
+
|
305
|
+
ir_fqn(ir)
|
306
|
+
end
|
307
|
+
|
308
|
+
def find_anchor_inputref(node, need_prefix:)
|
309
|
+
found = nil
|
310
|
+
walk = lambda do |x|
|
311
|
+
case x
|
312
|
+
when NAST::InputRef
|
313
|
+
ax = axes_of(x)
|
314
|
+
found ||= x if need_prefix.each_with_index.all? { |tok, i| ax[i] == tok }
|
315
|
+
when NAST::Ref
|
316
|
+
decl = @snast.decls.fetch(x.name) { raise "unknown declaration #{x.name}" }
|
317
|
+
walk.call(decl.body)
|
318
|
+
when NAST::Select
|
319
|
+
walk.call(x.cond)
|
320
|
+
walk.call(x.on_true)
|
321
|
+
walk.call(x.on_false)
|
322
|
+
when NAST::Reduce, NAST::Fold
|
323
|
+
walk.call(x.arg)
|
324
|
+
when NAST::Call, NAST::Tuple
|
325
|
+
x.args.each { walk.call(_1) }
|
326
|
+
when NAST::Hash
|
327
|
+
x.pairs.each { walk.call(_1) }
|
328
|
+
when NAST::Pair
|
329
|
+
walk.call(x.value)
|
330
|
+
when NAST::IndexRef
|
331
|
+
ax = axes_of(x)
|
332
|
+
found ||= x.input_fqn if need_prefix.each_with_index.all? { |tok, i| ax[i] == tok }
|
333
|
+
end
|
334
|
+
end
|
335
|
+
walk.call(node)
|
336
|
+
found or raise "no anchor InputRef covering axes #{need_prefix.inspect}"
|
337
|
+
end
|
338
|
+
|
339
|
+
# ---------- path caches (no dtype checks) ----------
|
340
|
+
|
341
|
+
# --- replace head_collection entirely ---
|
342
|
+
def head_collection(pre, li)
|
343
|
+
key = [pre.object_id, li]
|
344
|
+
@env.memo_get(:head, key) || begin
|
345
|
+
reg = nil
|
346
|
+
path = Array(pre[:head_path_by_loop][li])
|
347
|
+
|
348
|
+
if path.empty?
|
349
|
+
# Defensive fallback: loop step carries the root key in pre[:steps][li]
|
350
|
+
step = pre[:steps][li] or raise "no step at loop index #{li}"
|
351
|
+
root_key = (step[:key] || step[:axis]).to_sym
|
352
|
+
reg = Build.load_input(key: root_key, dtype: :array, ids: @ids).tap { @ops << _1 }.result_register
|
353
|
+
@env.memo_set(:head, key, reg)
|
354
|
+
return reg
|
355
|
+
end
|
356
|
+
|
357
|
+
path.each do |kind, ksym|
|
358
|
+
case kind
|
359
|
+
when :input
|
360
|
+
reg = Build.load_input(key: ksym, dtype: :array, ids: @ids).tap { @ops << _1 }.result_register
|
361
|
+
when :field
|
362
|
+
raise "head path field before input" if reg.nil?
|
363
|
+
|
364
|
+
reg = Build.load_field(object_register: reg, key: ksym, dtype: :any, ids: @ids).tap do
|
365
|
+
@ops << _1
|
366
|
+
end.result_register
|
367
|
+
else
|
368
|
+
raise "unknown head hop #{kind.inspect}"
|
369
|
+
end
|
370
|
+
end
|
371
|
+
|
372
|
+
@env.memo_set(:head, key, reg)
|
373
|
+
end
|
374
|
+
end
|
375
|
+
|
376
|
+
def between_loops(pre, li_from, li_to, start_el_reg)
|
377
|
+
keys = pre[:between_loops].fetch([li_from, li_to], [])
|
378
|
+
return start_el_reg if keys.empty?
|
379
|
+
|
380
|
+
k = [start_el_reg.object_id, pre.object_id, li_from, li_to]
|
381
|
+
@env.memo_get(:between, k) || begin
|
382
|
+
cur = start_el_reg
|
383
|
+
keys.each do |sym|
|
384
|
+
cur = Build.load_field(object_register: cur, key: sym, dtype: :any, ids: @ids).tap { @ops << _1 }.result_register
|
385
|
+
end
|
386
|
+
@env.memo_set(:between, k, cur)
|
387
|
+
end
|
388
|
+
end
|
389
|
+
|
390
|
+
def length_of(reg)
|
391
|
+
k = reg.object_id
|
392
|
+
@env.memo_get(:len, k) || @env.memo_set(:len, k, @emit.length(reg))
|
393
|
+
end
|
394
|
+
|
395
|
+
def clamped_index(idx_reg, coll_reg)
|
396
|
+
k = [idx_reg.object_id, coll_reg.object_id]
|
397
|
+
@env.memo_get(:clamp_idx, k) || begin
|
398
|
+
hi = @emit.sub_i(length_of(coll_reg), @emit.iconst(1))
|
399
|
+
@env.memo_set(:clamp_idx, k, @emit.clamp(idx_reg, @emit.iconst(0), hi, out: :integer))
|
400
|
+
end
|
401
|
+
end
|
402
|
+
end
|
403
|
+
end
|
404
|
+
end
|
405
|
+
end
|
406
|
+
end
|
407
|
+
end
|
@@ -0,0 +1,243 @@
|
|
1
|
+
# frozen_string_literal: true
|
2
|
+
|
3
|
+
module Kumi
|
4
|
+
module Core
|
5
|
+
module Analyzer
|
6
|
+
module Passes
|
7
|
+
module LIR
|
8
|
+
module StencilEmitter
|
9
|
+
NAST = Kumi::Core::NAST
|
10
|
+
|
11
|
+
def emit_roll(n) = emit_stencil(:roll, n)
|
12
|
+
def emit_shift(n) = emit_stencil(:shift, n)
|
13
|
+
|
14
|
+
private
|
15
|
+
|
16
|
+
def emit_stencil(kind, n)
|
17
|
+
base_ir, offsets, policies = resolve_shift_chain(n)
|
18
|
+
axes = axes_of(base_ir)
|
19
|
+
raise "#{kind}: source has no axes" if axes.empty?
|
20
|
+
|
21
|
+
anchor_fqn = anchor_fqn_from_node!(base_ir, need_prefix: axes)
|
22
|
+
pre = @pre.fetch(anchor_fqn) { raise "no precomputed plan for #{anchor_fqn}" }
|
23
|
+
|
24
|
+
gather_with_offsets(base_ir, pre, offsets, policies)
|
25
|
+
end
|
26
|
+
|
27
|
+
def resolve_shift_chain(node)
|
28
|
+
offsets = Hash.new(0)
|
29
|
+
policies = {}
|
30
|
+
cur = node
|
31
|
+
|
32
|
+
loop do
|
33
|
+
case cur
|
34
|
+
when NAST::Call
|
35
|
+
break unless %i[shift roll].include?(cur.fn)
|
36
|
+
|
37
|
+
src, off_node = cur.args
|
38
|
+
axes = axes_of(src)
|
39
|
+
defaults = @registry.function(cur.fn)[:options] || {}
|
40
|
+
opts = merge_call_opts(cur, defaults)
|
41
|
+
explicit = opts[:policy]
|
42
|
+
policy = (if explicit
|
43
|
+
explicit.to_sym
|
44
|
+
else
|
45
|
+
(cur.fn == :roll ? :wrap : :wrap)
|
46
|
+
end)
|
47
|
+
aoff = Integer(opts.fetch(:axis_offset, 0))
|
48
|
+
axis = axes.fetch(axes.length - 1 - aoff) { raise "stencil: axis_offset out of range" }
|
49
|
+
offsets[axis] += literal_offset!(off_node)
|
50
|
+
policies[axis] = policy
|
51
|
+
cur = src
|
52
|
+
when NAST::Ref
|
53
|
+
decl = @snast.decls.fetch(cur.name) { raise "unknown decl #{cur.name}" }
|
54
|
+
cur = decl.body
|
55
|
+
else
|
56
|
+
break
|
57
|
+
end
|
58
|
+
end
|
59
|
+
|
60
|
+
base_ir =
|
61
|
+
case cur
|
62
|
+
when NAST::InputRef then cur
|
63
|
+
when NAST::Call, NAST::Ref then raise "stencil base must be InputRef after collapsing"
|
64
|
+
else raise "unsupported stencil base #{cur.class}"
|
65
|
+
end
|
66
|
+
|
67
|
+
[base_ir, offsets, policies]
|
68
|
+
end
|
69
|
+
|
70
|
+
def gather_with_offsets(src_ir, pre, offsets, policies)
|
71
|
+
src_axes = axes_of(src_ir)
|
72
|
+
steps = pre[:steps]
|
73
|
+
loop_ixs = pre[:loop_ixs]
|
74
|
+
loop_idx = lambda { |ax|
|
75
|
+
begin
|
76
|
+
loop_ixs[loop_ixs.index { |i| steps[i][:axis].to_sym == ax } ]
|
77
|
+
rescue StandardError
|
78
|
+
(raise "plan lacks axis #{ax.inspect}")
|
79
|
+
end
|
80
|
+
}
|
81
|
+
|
82
|
+
# start at head collection for first src axis
|
83
|
+
first_ax = src_axes.first
|
84
|
+
li = loop_idx.call(first_ax)
|
85
|
+
coll = @env.collection_reg_for(first_ax) || head_collection(pre, li)
|
86
|
+
|
87
|
+
ok_mask = nil
|
88
|
+
prev_li = li
|
89
|
+
cur = nil
|
90
|
+
|
91
|
+
src_axes.each_with_index do |ax, k|
|
92
|
+
li = loop_idx.call(ax)
|
93
|
+
# walk between loops from the previous axis to this axis on current object
|
94
|
+
coll = between_loops(pre, prev_li, li, coll) if k > 0
|
95
|
+
prev_li = li
|
96
|
+
|
97
|
+
# choose index and policy for this axis
|
98
|
+
i0 = @env.index_reg_for(ax) or raise "no index for #{ax}"
|
99
|
+
off = offsets.fetch(ax, 0)
|
100
|
+
n = length_of(coll)
|
101
|
+
j = off.zero? ? i0 : @emit.sub_i(i0, @emit.iconst(off))
|
102
|
+
|
103
|
+
pol = policies[ax] || :wrap # roll defaults to wrap
|
104
|
+
jg =
|
105
|
+
case pol
|
106
|
+
when :wrap
|
107
|
+
@emit.mod_i(@emit.add_i(@emit.mod_i(j, n), n), n)
|
108
|
+
when :clamp
|
109
|
+
hi = @emit.sub_i(n, @emit.iconst(1))
|
110
|
+
@emit.clamp(j, @emit.iconst(0), hi, out: :integer)
|
111
|
+
when :zero
|
112
|
+
ge0 = @emit.ge(j, @emit.iconst(0))
|
113
|
+
ltN = @emit.lt(j, n)
|
114
|
+
ok = @emit.and_(ge0, ltN)
|
115
|
+
ok_mask = ok_mask ? @emit.and_(ok_mask, ok) : ok
|
116
|
+
hi = @emit.sub_i(n, @emit.iconst(1))
|
117
|
+
@emit.clamp(j, @emit.iconst(0), hi, out: :integer)
|
118
|
+
else
|
119
|
+
raise "unknown stencil policy #{pol.inspect}"
|
120
|
+
end
|
121
|
+
|
122
|
+
# gather for this axis, then proceed
|
123
|
+
last = (k == src_axes.length - 1)
|
124
|
+
out_dt = last ? dtype_of(src_ir) : :any
|
125
|
+
cur = @emit.gather(coll, jg, out_dt)
|
126
|
+
|
127
|
+
# next axis will start from the object we just gathered
|
128
|
+
coll = cur
|
129
|
+
end
|
130
|
+
|
131
|
+
if ok_mask
|
132
|
+
z = @emit.const(0, dtype_of(src_ir))
|
133
|
+
@emit.select(ok_mask, cur, z, dtype_of(src_ir))
|
134
|
+
else
|
135
|
+
cur
|
136
|
+
end
|
137
|
+
end
|
138
|
+
|
139
|
+
# def neighbor_leaf_at(src:, src_axes:, at_axis:, pre:, liA:, collA:, jA:, safe_first_hop: false)
|
140
|
+
# mkey = [at_axis, jA.object_id, safe_first_hop, collA.object_id]
|
141
|
+
# if (cached = @env.memo_get(:neighbor, mkey))
|
142
|
+
# return cached[:leaf]
|
143
|
+
# end
|
144
|
+
|
145
|
+
# a_pos = src_axes.index(at_axis) or raise "axis #{at_axis} not in #{src_axes.inspect}"
|
146
|
+
# first_dtype = a_pos == src_axes.length - 1 ? dtype_of(src) : :any
|
147
|
+
|
148
|
+
# j0 = safe_first_hop ? jA : clamped_index(jA, collA)
|
149
|
+
# cur = @emit.gather(collA, j0, first_dtype)
|
150
|
+
|
151
|
+
# prev_li = liA
|
152
|
+
# (a_pos + 1).upto(src_axes.length - 1) do |k|
|
153
|
+
# ax = src_axes[k]
|
154
|
+
# li = loop_index_for_axis(pre, ax)
|
155
|
+
# cur = between_loops(pre, prev_li, li, cur)
|
156
|
+
|
157
|
+
# idx = @env.index_reg_for(ax) or raise "no index for #{ax}"
|
158
|
+
# cur = @emit.gather(cur, clamped_index(idx, cur), k == src_axes.length - 1 ? dtype_of(src) : :any)
|
159
|
+
|
160
|
+
# prev_li = li
|
161
|
+
# end
|
162
|
+
|
163
|
+
# @env.memo_set(:neighbor, mkey, { leaf: cur })
|
164
|
+
# cur
|
165
|
+
# end
|
166
|
+
|
167
|
+
# def loop_index_for_axis(pre, axis)
|
168
|
+
# steps = pre[:steps]
|
169
|
+
# loop_ixs = pre[:loop_ixs]
|
170
|
+
# loop_axes = loop_ixs.map { |i| steps[i][:axis].to_sym }
|
171
|
+
# j = loop_axes.index(axis) or raise "plan lacks axis #{axis.inspect}"
|
172
|
+
# loop_ixs[j]
|
173
|
+
# end
|
174
|
+
|
175
|
+
# def last_prior_env_axis_in_plan(pre, axis)
|
176
|
+
# in_plan = pre[:loop_ixs].map { |i| pre[:steps][i][:axis].to_sym }
|
177
|
+
# target_pos = in_plan.index(axis)
|
178
|
+
# return nil unless target_pos
|
179
|
+
|
180
|
+
# @env.axes.reverse.find { |ax| (pos = in_plan.index(ax)) && pos < target_pos }
|
181
|
+
# end
|
182
|
+
|
183
|
+
# # ---- math helpers ----
|
184
|
+
|
185
|
+
# def shifted_index(kind, policy, idx, off, nlen)
|
186
|
+
# case policy
|
187
|
+
# when :wrap
|
188
|
+
# i1 = @emit.add_i(idx, @emit.iconst(off))
|
189
|
+
# m1 = @emit.mod_i(i1, nlen)
|
190
|
+
# @emit.mod_i(@emit.add_i(m1, nlen), nlen)
|
191
|
+
# when :clamp
|
192
|
+
# j = @emit.add_i(idx, @emit.iconst(off))
|
193
|
+
# hi = @emit.sub_i(nlen, @emit.iconst(1))
|
194
|
+
# @emit.clamp(j, @emit.iconst(0), hi, out: :integer)
|
195
|
+
# when :zero
|
196
|
+
# @emit.add_i(idx, @emit.iconst(off))
|
197
|
+
# else
|
198
|
+
# kind == :roll ? shifted_index(kind, :wrap, idx, off, nlen) : raise("#{kind}: unknown policy #{policy.inspect}")
|
199
|
+
# end
|
200
|
+
# end
|
201
|
+
|
202
|
+
# TODO : fix call node.opts being {opts: {}} or just flat keys, make always flat keys
|
203
|
+
def merge_call_opts(call_node, defaults)
|
204
|
+
raw = call_node.opts || {}
|
205
|
+
# unwrap common nesting: {opts: {...}}
|
206
|
+
raw = raw[:opts] if raw.key?(:opts)
|
207
|
+
|
208
|
+
# normalize keys
|
209
|
+
co = raw.each_with_object({}) { |(k, v), h| h[(k.respond_to?(:to_sym) ? k.to_sym : k)] = v }
|
210
|
+
|
211
|
+
pol = (co.key?(:policy) ? co[:policy] : defaults[:policy])
|
212
|
+
aof = (co.key?(:axis_offset) ? co[:axis_offset] : (defaults[:axis_offset] || 0))
|
213
|
+
|
214
|
+
{
|
215
|
+
policy: pol.to_sym,
|
216
|
+
axis_offset: Integer(aof)
|
217
|
+
}
|
218
|
+
end
|
219
|
+
|
220
|
+
def literal_offset!(node)
|
221
|
+
v = node.is_a?(NAST::Const) ? node.value : nil
|
222
|
+
raise "stencil offset must be integer literal" unless v.is_a?(Integer)
|
223
|
+
|
224
|
+
v
|
225
|
+
end
|
226
|
+
|
227
|
+
def resolve_call(node)
|
228
|
+
case node
|
229
|
+
when NAST::Call
|
230
|
+
node
|
231
|
+
when NAST::Ref
|
232
|
+
decl = @snast.decls.fetch(node.name) { return nil }
|
233
|
+
resolve_call(decl.body)
|
234
|
+
else
|
235
|
+
nil
|
236
|
+
end
|
237
|
+
end
|
238
|
+
end
|
239
|
+
end
|
240
|
+
end
|
241
|
+
end
|
242
|
+
end
|
243
|
+
end
|