Ticket UUID: | 926164 | |||
Title: | Unacceptable slowdown in fix for 729692 | |||
Type: | Bug | Version: | obsolete: 8.5a2 | |
Submitter: | msofer | Created on: | 2004-03-30 18:09:24 | |
Subsystem: | 47. Bytecode Compiler | Assigned To: | msofer | |
Priority: | 5 Medium | Severity: | ||
Status: | Closed | Last Modified: | 2007-09-13 08:10:05 | |
Resolution: | Fixed | Closed By: | msofer | |
Closed on: | 2007-09-13 01:10:05 | |||
Description: |
the fix for bug 729692 introduced the new INST_START_CMD, causing an unacceptable core slowdown. Tclbench results attached; summary of slowdowns: * 3% in completing the whole suite * 5-10% bytecode intensive (loops, base64, klist, list iterate, ...) * 15-20% worst cases (sha1, heapsort). See some comments on improving performance in the original bug ticket. | |||
User Comments: |
hobbs added on 2005-10-13 07:21:58:
Logged In: YES user_id=72656 Passing to dgp, since he's having so much fun optimizing the head lately. :) msofer added on 2005-03-23 01:15:51: File Added - 126751: start msofer added on 2005-03-23 01:15:49: Logged In: YES user_id=148712 Jeff: can you please test if this patch is good, bad or indifferent wrt to this issue? I have a very bad platform for reliable timings. msofer added on 2004-05-13 00:44:02: Logged In: YES user_id=148712 More careful emission of INST_START_CMD and a runtime peephole optimisation commited. Effect not measured yet. msofer added on 2004-03-31 02:32:25: Logged In: YES user_id=148712 The too frequent calls to Tcl_AsyncReady() are responsible for 60-70% of the slowdown. Reverting to the previous model (check now every 64th instruction, instead of every INST_START_CMD). Note that this also fixes [Bug 926118]. msofer added on 2004-03-31 01:09:25: File Added - 81971: tclbench |