[HARMONY-1564] [drlvm][jit] patch for possible ABCD bug

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Component/s: DRLVM
• Patch Info: Patch Available
• Estimated Complexity: Advanced
• Assignee: Alexey Varlamov
• Reporter: Naveen Neelakantam
• Created: 24/Sep/06 19:59
• Updated: 02/May/13 02:29
• Resolved: 06/Jun/07 22:18

Description

I've been reading through the ABCD implementation in jitrino, and if I understand it correctly, I found a bug. I've attached a patch to fix it. Someone who actually understands the code should verify this.

Also, did anyone ever test this ABCD pass? I ask because I've tried running it on a bidirectional bubble sort as mentioned in the original paper. The paper mentions that the pass should be able to prove all of the bounds checks in the sort method as redundant/unnecessary. However, when I try running the abcd pass on a bidirectional bubble sort (attached), none of the bounds checks are eliminated.

The bidirectional bubble sort program is based on code I found here: http://www.cs.ubc.ca/~harrison/Java/sorting-demo.html

Issue Links

• is depended upon by: HARMONY-1788 [drlvm][jit][abcd] Inequality graph generation from HIR