Fix canReachLocInTime to account for register waits and refactor calculateAward

[guosran]

This PR addresses two improvements to the mapping algorithm:

1. Fix canReachLocInTime BFS to model register-based waiting

   • Previously: Only tracked tile-level visits without time awareness
   • Now: Uses (tile, time) state tracking to allow register waiting
   • Aligns with tryRouteDataMove's routing behavior
   • Both link traversal and register waiting are properly modeled

2. Refactor calculateAward with tunable parameters

   • Separated tile-based awards from time-based awards
   • Added constants for tuning: AWARD_PROXIMITY_SCALE, AWARD_BACKWARD_PROXIMITY_SCALE, AWARD_BASE_MULTIPLIER, AWARD_CRITICAL_BONUS_DIV (all set to 1 for now)
   • Improved proximity calculation for producers and backward users
   • Better handling of critical ops on recurrence paths

Test updates:

• Updated backtrack-config from 'simple' to 'customized' (5,3)
• Compiled II improvements:
  • nested_loop: 17 → 13 (23.5% improvement)
  • branch_for: 5 → 4 (20% improvement)
  • fusion: 14 → 13 and 15 → 12

[Copilot]

Pull request overview

This PR improves the mapping algorithm by fixing the canReachLocInTime BFS to properly model register-based waiting and refactoring calculateAward to separate tile-based from time-based scoring with tunable parameters. The changes result in significant compiled II improvements across multiple test cases.

Key changes:

• Fixed canReachLocInTime to use (tile, time) state tracking instead of tile-only tracking, allowing register waiting to be properly modeled
• Refactored calculateAward to separate proximity-based tile scoring from time-step scoring with tunable constants (all set to 1 for now)
• Test updates reflecting II improvements: nested_loop (17→13), branch_for (5→4), fusion (14→13, 15→12)

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
lib/NeuraDialect/Mapping/mapping_util.cpp	Core algorithm changes: BFS state tracking fix and award calculation refactoring with tunable constants
test/c2llvm2mlir/nested_loop/test.mlir	Updated expected compiled_ii from 17 to 13
test/neura/fusion/test.mlir	Updated expected compiled_ii values (14→13, 15→12)
test/neura/for_loop/relu_test.mlir	Updated expected mapping output formatting
test/neura/ctrl/branch_for.mlir	Updated expected compiled_ii from 5 to 4 and formatting
test/mapping_quality/branch_for.mlir	Updated expected compiled_ii from 5 to 4 with expanded output
test/e2e/relu/relu_kernel.mlir	Updated expected mapping locations and formatting
test/e2e/histogram/histogram_kernel.mlir	Updated expected mapping locations and formatting
test/e2e/fir/fir_kernel_vec.mlir	Updated expected mapping locations
test/e2e/fir/fir_kernel.mlir	Updated expected mapping locations
test/e2e/bicg/bicg_kernel.mlir	Updated expected YAML output with expanded mappings
test/code_gen/test_code_generate.mlir	Updated expected compiled_ii from 6 to 5

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[guosran]

This is a follow-up of PR #195 .

[guosran]

Pull request overview

This PR improves the mapping algorithm by fixing the canReachLocInTime BFS to properly model register-based waiting and refactoring calculateAward to separate tile-based from time-based scoring with tunable parameters. The changes result in significant compiled II improvements across multiple test cases.

Key changes:

• Fixed canReachLocInTime to use (tile, time) state tracking instead of tile-only tracking, allowing register waiting to be properly modeled
• Refactored calculateAward to separate proximity-based tile scoring from time-step scoring with tunable constants (all set to 1 for now)
• Test updates reflecting II improvements: nested_loop (17→13), branch_for (5→4), fusion (14→13, 15→12)

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated no comments.

Show a summary per file
File Description
lib/NeuraDialect/Mapping/mapping_util.cpp Core algorithm changes: BFS state tracking fix and award calculation refactoring with tunable constants
test/c2llvm2mlir/nested_loop/test.mlir Updated expected compiled_ii from 17 to 13
test/neura/fusion/test.mlir Updated expected compiled_ii values (14→13, 15→12)
test/neura/for_loop/relu_test.mlir Updated expected mapping output formatting
test/neura/ctrl/branch_for.mlir Updated expected compiled_ii from 5 to 4 and formatting
test/mapping_quality/branch_for.mlir Updated expected compiled_ii from 5 to 4 with expanded output
test/e2e/relu/relu_kernel.mlir Updated expected mapping locations and formatting
test/e2e/histogram/histogram_kernel.mlir Updated expected mapping locations and formatting
test/e2e/fir/fir_kernel_vec.mlir Updated expected mapping locations
test/e2e/fir/fir_kernel.mlir Updated expected mapping locations
test/e2e/bicg/bicg_kernel.mlir Updated expected YAML output with expanded mappings
test/code_gen/test_code_generate.mlir Updated expected compiled_ii from 6 to 5
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

Thank u Copilot T^T