Number: 57

Date: 17-Mar-84  0':04':03

Submitter: Sannella.PA

Source: MASINTER.PA

Subject: want BIN in DLion 12K Control Store UCode

Assigned To: 

Attn: Charnley

Status: Open

In/By: 

Problem Type: Performance

Impact: Moderate

Difficulty: Moderate

Frequency: Everytime

Priority: Hopefully

System: Language Support

Subsystem: Microcode

Machine: 

Disk: 

Lisp Version: 

Source Files: 

Microcode Version: 

Memory Size: 

File Server: 

Server Software Version: 

Disposition: [I took some of the ideas in this and submitted separate ARs. lmm - 17Aug 84] 

Description: '
Date': 13 MAR 84 22':15 PST'
From': MASINTER.PA'
Subject': DLion microcode priorities'
To':   charnley'
cc':   LispSupport'
'
The next set of opcodes which seem'
to be relevant are':'
'
a) more GCRECLAIMCELL'
b) EVAL (for interpreter)'
c) BIN'
'
I think those are the relevant ones for current benchmarks.'
'
A separate AR is to please automate and document benchmarks. '
'
-----'
'
Date': 27 Mar 84 23':06 PST'
From': Masinter.pa'
Subject': Lisp': microcode enhancements'
To': LispSupport.pa'
cc': sheil (fyi)'
Lisp-System-Date': 27-Mar-84 08':53':34'
Machine-Type': Dorado'
'
This is one of several ARs on rewriting ALL microcodes to include new opcodes. Looking at the function inside Rosie that is taking up the most time, I believe it would be helped substantially by (not necessarily in this order)':'
'
(a) microcoded ELT'
(b) microcoded LLSH'
(c) microcde combining CAR/CDR chains to length 2 or 3'
(e.g., CAAR, CDDR, CADR, although CDDR is most prevelant)'
(d) JEQ (if faster than EQ/TJUMP)'
(e) stack relative addressing, if it enabled the stack optimization I had partially implemented'
(f) microcoded FASSOC'
(g) special (CONS x NIL) opcode, maybe (LIST X Y) opcode.'
'
[big memory helps if memory/compute time is large; compute time can dominate with bad compiled code]'
'
-----'
'
'


Workaround: 

Test Case: 

Edit-By: Masinter

Edit-Date: 17-Aug-84 12':28':24