KL10 / KLH10 Performance

[PoBe]

For those how have an interest in the kl10 / klh10 performance
I have compiled what I have observed when using the dfkfb benchmark test.
The dfkfb test is part of the KLH10 distribution.

In the KLH10 distribution (dfkfb.txt) there are some number from a real KL10
being provided by Kenneth L. Harrenstien himself.

The platforms being tested are

1. A real KL10B with MCA25 cache (DEC2065) executed in user mode
2. KLH10, Intel i7/3930K (slightly over clocked ~4.5 GHz) 16GB
3. Intel NUC DC3217IYE i3 (1.8 Ghz) 8GB


I also tried, the ARM Raspberry Model B (~900 MHz) 512MB with Raspberry Debian

The test does not run clean on the Raspberry "wheezy" platform (hard float). All instructions comes out with an excution time of around 504-507 nsec. Booting up and using KLH/TOPS20 however works fine, with network and everything. Possible the soft-float version "wheezy" should be used? At some point I might test it.


Instruction #1 #2 #3
2 - INDEXING 36 ns 1 ns 1 ns
3 - INDIRECT 269 ns 5 ns 11 ns
4 - INDEXING AND INDIRECT 336 ns 7 ns 12 ns
5 - MOVEI 269 ns 6 ns 13 ns
6 - MOVE FROM AC 374 ns 7 ns 15 ns
7 - MOVE FROM MEMORY 402 ns 8 ns 17 ns
8 - HRR FROM MEMORY 443 ns 9 ns 19 ns
9 - SETOM 0 571 ns 7 ns 15 ns
10 - JRST 303 ns 8 ns 17 ns
11 - JSR 642 ns 10 ns 21 ns
12 - PUSHJ 708 ns 11 ns 23 ns
13 - ADD FROM MEMORY 438 ns 15 ns 31 ns
14 - MUL (9 ADD/SUB - 18 SHIFTS) 2.42 us 51 ns 77 ns
15 - DIV 4.70 us 304 ns 663 ns
16 - FIX A FLOATING POINT ONE 874 ns 15 ns 28 ns
17 - FLTR AN INTERGER ONE 881 ns 41 ns 47 ns
18 - FAD (1 RIGHT SHIFT) 1.59 us 31 ns 62 ns
19 - FAD (8 SHIFT RIGHT - 3 LEFT) 1.81 us 48 ns 98 ns
20 - FMP (7 ADD/SUB - 14 SHIFTS) 2.56 us 62 ns 132 ns
21 - FDV 4.82 us 178 ns 376 ns
22 - DMOVE FROM MEMORY 608 ns 12 ns 24 ns
23 - DFAD (1 RIGHT SHIFT) 2.06 us 60 ns 130 ns
24 - DFAD (8 SHIFT RIGHT - 1 LEFT) 2.06 us 60 ns 130 ns
25 - DFMP (7 ADD/SUB - 32 SHIFTS) 4.25 us 131 ns 282 ns
26 - DFDV 8.71 us 2.01 us 4.53 us
27 - CONO PI ?? ns 11 ns 27 ns
28 - CONI PI ?? ns 10 ns 20 ns
29 - DATAO APR ?? ns 10 ns 22 ns
30 - DATAI APR ?? ns 10 ns 23 ns
31 - MOVE TO MEMORY 573 ns 8 ns 18 ns
32 - LOGICAL SHIFT (35 PLACES LEFT) 439 ns 12 ns 24 ns
33 - LOGICAL SHIFT (35 PLACES RIGHT) 470 ns 12 ns 24 ns
34 - LOGICAL SHIFT COMBINED (71 PLACES LEFT) 743 ns. 16 ns 34 ns
35 - LOGICAL SHIFT COMBINED (71 PLACES RIGHT)745 ns. 16 ns 34 ns
36 - INCREMENT BYTE POINTER 924 ns 18 ns 33 ns
37 - INCREMENT AND LOAD BYTE 1.21 us. 23 ns 46 ns
38 - INCREMENT AND DEPOSIT BYTE 1.60 us. 27 ns 54 ns
39 - JFCL 404 ns 5 ns 12 ns
40 - CAI 272 ns 5 ns 11 ns
41 - JUMP 342 ns 5 ns 11 ns
42 - CAM 376 ns 6 ns 12 ns
43 - EQV AC TO AC 402 ns 10 ns 18 ns
44 - EQV MEMORY TO AC 742 ns. 10 ns 22 ns
45 - SETOB 571 ns. 9 ns 18 ns
46 - AOS TO MEMORY 539 ns. 9 ns 19 ns
47 - EXCHANGE AN AC WITH AN AC 545 ns. 9 ns 20 ns
48 - EXCHANGE AN AC WITH MEMORY 706 ns 11 ns 22 ns
49 - EXECUTE 471 ns. 8 ns 14 ns
50 - BLT MEMORY TO MEMORY 1.58 us 16 ns 33 ns
51 - BLT AC TO MEMORY 1.48 us 16 ns 32 ns
52 - DATAI ?? ns 14 ns 28 ns
53 - DATAO ?? ns 12 ns 26 ns

[PoBe]

The dfkfb test on Raspberry "wheezy" (soft-float) has the same problem as the hardfloat, all instructions execution time is around 504-506 ns.

[David Todd]

On Sunday, March 24, 2013 2:52:05 PM UTC-4, PoBe wrote:
> The dfkfb test on Raspberry "wheezy" (soft-float) has the same problem as the hardfloat, all instructions execution time is around 504-506 ns.

I got the same results on my RPi with dfkfb. I wonder if the odd timings are due to the ITIME/ISYNC/CTY parameter definition changes required in Makefile.mk to create a working system, and I wonder if there are any other combinations of those parameters that would result in a working system WITHOUT the sporadic DTE errors but with timings that make sense. I do think the performance on the Raspberry Pi should be better than I'm seeing (executing FORTRAN programs on TOPS-20), and those parameters may have something to do with that.

To add to your collection of timings, here's what dfkfb gets on my Mac Mini, vintage summer 2013 (2.6GHz I7, 8GB memory):


1 - BASIC CLOCK CYCLE IS 0 NSEC.
2 - INDEXING TAKES 0 NSEC.
3 - INDIRECT TAKES 6 NSEC.
4 - INDEXING AND INDIRECT TAKES 6 NSEC.
5 - MOVEI TAKES 5 NSEC.
6 - MOVE FROM AC TAKES 6 NSEC.
7 - MOVE FROM MEMORY TAKES 7 NSEC.
8 - HRR FROM MEMORY TAKES 9 NSEC.
9 - SETOM 0 TAKES 6 NSEC.
10 - JRST TAKES 7 NSEC.
11 - JSR TAKES 8 NSEC.
12 - PUSHJ TAKES 9 NSEC.
13 - ADD FROM MEMORY TAKES 10 NSEC.
14 - MUL (9 ADD/SUB - 18 SHIFTS) TAKES 20 NSEC.
15 - DIV TAKES 111 NSEC.
16 - FIX A FLOATING POINT ONE TAKES 11 NSEC.
17 - FLTR AN INTERGER ONE TAKES 19 NSEC.
18 - FAD (1 RIGHT SHIFT) TAKES 19 NSEC.
19 - FAD (8 SHIFT RIGHT - 3 LEFT) TAKES 29 NSEC.
20 - FMP (7 ADD/SUB - 14 SHIFTS) TAKES 32 NSEC.
21 - FDV TAKES 76 NSEC.
22 - DMOVE FROM MEMORY TAKES 10 NSEC.
23 - DFAD (1 RIGHT SHIFT) TAKES 30 NSEC.
24 - DFAD (8 SHIFT RIGHT - 1 LEFT) TAKES 31 NSEC.
25 - DFMP (7 ADD/SUB - 32 SHIFTS) TAKES 57 NSEC.
26 - DFDV TAKES 275 NSEC.
27 - CONO PI TAKES 10 NSEC.
28 - CONI PI TAKES 8 NSEC.
29 - DATAO APR TAKES 9 NSEC.
30 - DATAI APR TAKES 8 NSEC.
31 - MOVE TO MEMORY TAKES 7 NSEC.
32 - LOGICAL SHIFT (35 PLACES LEFT) TAKES 6 NSEC.
33 - LOGICAL SHIFT (35 PLACES RIGHT) TAKES 7 NSEC.
34 - LOGICAL SHIFT COMBINED (71 PLACES LEFT) TAKES 9 NSEC.
35 - LOGICAL SHIFT COMBINED (71 PLACES RIGHT) TAKES 9 NSEC.
36 - INCREMENT BYTE POINTER TAKES 14 NSEC.
37 - INCREMENT AND LOAD BYTE TAKES 19 NSEC.
38 - INCREMENT AND DEPOSIT BYTE TAKES 22 NSEC.
39 - JFCL TAKES 5 NSEC.
40 - CAI TAKES 4 NSEC.
41 - JUMP TAKES 4 NSEC.
42 - CAM TAKES 5 NSEC.
43 - EQV AC TO AC TAKES 7 NSEC.
44 - EQV MEMORY TO AC TAKES 8 NSEC.
45 - SETOB TAKES 7 NSEC.
46 - AOS TO MEMORY TAKES 8 NSEC.
47 - EXCHANGE AN AC WITH AN AC TAKES 6 NSEC.
48 - EXCHANGE AN AC WITH MEMORY TAKES 7 NSEC.
49 - EXECUTE TAKES 6 NSEC.
50 - BLT MEMORY TO MEMORY TAKES 13 NSEC.
51 - BLT AC TO MEMORY TAKES 12 NSEC.
52 - DATAI TAKES 12 NSEC.
53 - DATAO TAKES 12 NSEC.

[Lars Brinkhoff]

PoBe wrote:
> For those how have an interest in the kl10 / klh10 performance
> I have compiled what I have observed when using the dfkfb benchmark test.
> The dfkfb test is part of the KLH10 distribution.

Numbers from a similar test run on a KA10. Text below this line is from a file on an ITS computer:


This output is from DM:SYS1;TS SPEEDY, running on ML.

PDP-10 INSTRUCTION TIMING TEST
60 HZ.

1 - MOVEI TAKES 2.24 USEC.
2 - INDEXING TAKES 296 NSEC.
3 - INDEXING AND INDIRECT TAKES 958 NSEC.
4 - JSR TAKES 4.27 USEC.
5 - JRST TAKES 2.24 USEC.
6 - MOVE FROM AC TAKES 2.73 USEC.
7 - MOVE FROM MEMORY TAKES 3.86 USEC.
8 - HRR FROM MEMORY TAKES 4.02 USEC.
9 - ADD FROM MEMORY TAKES 4.21 USEC.
10 - SETOM 0 TAKES 2.75 USEC.
11 - PUSHJ TAKES 4.98 USEC.
12 - SHIFTING TAKES 153 NSEC.
13 - FSC AN INTEGER ONE TAKES 7.32 USEC.
14 - FAD (1 RIGHT SHIFT) TAKES 6.27 USEC.
15 - FAD (8 SHIFT RIGHT-3 LEFT) TAKES 8.21 USEC.
16 - MUL (18 ADD/SUB - 36 SHIFTS) TAKES 12.44 USEC.
17 - FMP (14 ADD/SUB - 27 SHIFTS) TAKES 12.23 USEC.
18 - RELOCATION TAKES 0 NSEC.
19 - REPEATED I/O INSTRUCTIONS TAKE 77.01 USEC.
20 - SHIFT AND ADD TAKES 281 NSEC.

[m. thompson]

From a real KS10:

DECSYSTEM 2020 INSTRUCTION TIMING TEST (DSKFA)
VERSION 0.1, SV=0.3, CPU#=4097, MCV=117, MCO=0, HO=0, KASW=003740 000000

TTY SWITCH CONTROL ? - 0,S OR Y <CR> - 0
SWITCHES = 000000 000000

1 - BASIC CLOCK CYCLE IS 151 NSEC.
2 - INDEXING TAKES 303 NSEC.
3 - INDIRECT TAKES 910 NSEC.
4 - INDEXING AND INDIRECT TAKES 910 NSEC.
5 - MOVEI TAKES 1.22 USEC.
6 - MOVSI TAKES 1.52 USEC.
7 - MOVE FROM AC TAKES 1.52 USEC.
8 - MOVE FROM MEMORY TAKES 1.52 USEC.
9 - MOVE TO MEMORY TAKES 2.28 USEC.
10 - HRR FROM MEMORY TAKES 1.82 USEC.
11 - HRRZ FROM MEMORY TAKES 1.82 USEC.
12 - HLRZ FROM MEMORY TAKES 2.13 USEC.
13 - SETZ TAKES 1.52 USEC.
14 - SETZM TAKES 1.97 USEC.
15 - SETZB TAKES 2.28 USEC.
16 - SETO TAKES 1.52 USEC.
17 - SETOM TAKES 1.97 USEC.
18 - SETOB TAKES 2.28 USEC.
19 - JRST TAKES 914 NSEC.
20 - JRSTF @MEM TAKES 4.25 USEC.
21 - JUMPA TAKES 1.22 USEC.
22 - JUMP THAT DOESN'T TAKES 1.97 USEC.
23 - JUMP THAT DOES TAKES 1.97 USEC.
24 - JSR TAKES 2.74 USEC.
25 - JSP TAKES 1.82 USEC.
26 - AOBJN TAKES 1.97 USEC.
27 - SOJ TAKES 1.82 USEC.
28 - SOS TAKES 3.34 USEC.
29 - AOJ TAKES 1.82 USEC.
30 - AOS TAKES 3.34 USEC.
31 - JFFO (1B0) TAKES 3.95 USEC.
32 - JFFO (1B35) TAKES 14.55 USEC.
33 - PUSH TAKES 2.73 USEC.
34 - POP TAKES 3.79 USEC.
35 - MUUO TAKES 15.34 USEC.
36 - LUUO TAKES 4.07 USEC.
37 - PUSHJ TAKES 2.73 USEC.
38 - ADD IMMEDIATE TAKES 1.67 USEC.
39 - ADD FROM MEMORY TAKES 1.97 USEC.
40 - ADD TO MEMORY TAKES 4.15 USEC.
41 - MUL (9 ADD/SUB - 18 SHIFTS) TAKES 15.76 USEC.
42 - IMULI TAKES 14.70 USEC.
43 - DIV TAKES 22.81 USEC.
44 - IDIVI TAKES 19.40 USEC.
45 - FIX A FLOATING POINT ONE TAKES 13.04 USEC.
46 - FLTR AN INTERGER ONE TAKES 17.74 USEC.
47 - FAD (1 RIGHT SHIFT) TAKES 6.83 USEC.
48 - FAD (8 SHIFT RIGHT - 3 LEFT) TAKES 11.07 USEC.
49 - FMP (7 ADD/SUB - 14 SHIFTS) TAKES 18.34 USEC.
50 - FDV TAKES 22.74 USEC.
51 - DMOVE FROM MEMORY TAKES 2.73 USEC.
52 - DMOVEM TO MEMORY TAKES 3.50 USEC.
53 - DFAD (1 RIGHT SHIFT) TAKES 12.73 USEC.
54 - DFAD (8 SHIFT RIGHT - 1 LEFT) TAKES 14.86 USEC.
55 - DFMP (7 ADD/SUB - 32 SHIFTS) TAKES 63.35 USEC.
56 - DFDV TAKES 52.74 USEC.
57 - CONO PI TAKES 7.28 USEC.
58 - CONI PI TAKES 2.88 USEC.
59 - CONO PAG TAKES 109.13 USEC.
60 - DATAO PAG LOAD UBR TAKES 109.94 USEC.
61 - DATAO PAG LOAD AC BLK TAKES 3.65 USEC.
62 - WRIO TAKES 14.25 USEC.
63 - RDIO TAKES 15.46 USEC.
64 - LOGICAL SHIFT (35 PLACES LEFT) TAKES 7.43 USEC.
65 - LOGICAL SHIFT (35 PLACES RIGHT) TAKES 7.43 USEC.
66 - LOGICAL SHIFT COMBINED (71 PLACES LEFT) TAKES 25.46 USEC.
67 - LOGICAL SHIFT COMBINED (71 PLACES RIGHT) TAKES 25.76 USEC.
68 - INCREMENT BYTE POINTER TAKES 3.04 USEC.
69 - ILDB (7 BITS) TAKES 5.92 USEC.
70 - IDPB (7 BITS) TAKES 6.98 USEC.
71 - ILDB (6 BITS) TAKES 9.86 USEC.
72 - IDPB (6 BITS) TAKES 15.01 USEC.
73 - LDB (7 BITS - POS 6) TAKES 4.70 USEC.
74 - LDB (7 BITS - POS 13) TAKES 4.70 USEC.
75 - LDB (7 BITS - POS 20) TAKES 4.70 USEC.
76 - LDB (7 BITS - POS 27) TAKES 4.70 USEC.
77 - LDB (7 BITS - POS 34) TAKES 4.70 USEC.
78 - DPB (7 BITS - POS 6) TAKES 5.76 USEC.
79 - DPB (7 BITS - POS 13) TAKES 5.76 USEC.
80 - DPB (7 BITS - POS 20) TAKES 5.76 USEC.
81 - DPB (7 BITS - POS 27) TAKES 5.76 USEC.
82 - DPB (7 BITS - POS 34) TAKES 5.76 USEC.
83 - LDB (6 BITS - POS 5) TAKES 9.70 USEC.
84 - LDB (6 BITS - POS 11) TAKES 8.79 USEC.
85 - LDB (6 BITS - POS 17) TAKES 7.88 USEC.
86 - LDB (6 BITS - POS 23) TAKES 9.40 USEC.
87 - LDB (6 BITS - POS 29) TAKES 8.49 USEC.
88 - LDB (6 BITS - POS 35) TAKES 7.58 USEC.
89 - DPB (6 BITS - POS 5) TAKES 14.86 USEC.
90 - DPB (6 BITS - POS 11) TAKES 13.95 USEC.
91 - DPB (6 BITS - POS 17) TAKES 13.04 USEC.
92 - DPB (6 BITS - POS 23) TAKES 12.13 USEC.
93 - DPB (6 BITS - POS 29) TAKES 11.22 USEC.
94 - DPB (6 BITS - POS 35) TAKES 10.31 USEC.
95 - JFCL TAKES 1.67 USEC.
96 - CAI NO-OP TAKES 1.22 USEC.
97 - CAI THAT DOESN'T SKIP TAKES 2.13 USEC.
98 - CAI THAT SKIPS TAKES 2.13 USEC.
99 - CAM NO-OP TO AC TAKES 1.97 USEC.
100 - CAM TO AC THAT DOESN'T SKIP TAKES 2.43 USEC.
101 - CAM TO AC THAT SKIPS TAKES 2.43 USEC.
102 - CAM NO-OP TO MEMORY TAKES 1.97 USEC.
103 - CAM TO MEMORY THAT DOESN'T SKIP TAKES 2.43 USEC.
104 - CAM TO MEMORY THAT SKIPS TAKES 2.43 USEC.
105 - TLN NO SKIP TAKES 2.28 USEC.
106 - TLN SKIP TAKES 2.28 USEC.
107 - TRN NO SKIP TAKES 1.97 USEC.
108 - TRN SKIP TAKES 1.98 USEC.
109 - TDN NO SKIP TAKES 2.28 USEC.
110 - TDN SKIP TAKES 2.28 USEC.
111 - TSN NO SKIP TAKES 2.58 USEC.
112 - TSN SKIP TAKES 2.59 USEC.
113 - SKIP TO AC THAT DOESN'T TAKES 2.58 USEC.
114 - SKIP TO AC THAT DOES TAKES 2.59 USEC.
115 - SKIP TO MEMORY THAT DOESN'T TAKES 2.58 USEC.
116 - SKIP TO MEMORY THAT DOES TAKES 2.59 USEC.
117 - EQV AC TO AC TAKES 1.82 USEC.
118 - EQV MEMORY TO AC TAKES 1.82 USEC.
119 - EXCHANGE AN AC WITH AN AC TAKES 2.58 USEC.
120 - EXCHANGE AN AC WITH MEMORY TAKES 2.58 USEC.
121 - EXECUTE TAKES 2.12 USEC.
122 - BLT MEMORY TO MEMORY TAKES 7.43 USEC.
123 - BLT AC TO MEMORY TAKES 7.43 USEC.

[Paul Rubin]

"m. thompson" <michael.9...@gmail.com> writes:
>> For those how have an interest in the kl10 / klh10 performance

Thanks for posting that. The 2013 post that you replied to and included
isn't available through my nntp feed any more, and it was nice to see
those numbers.

Rather than timing individual instructions, are there any application
benchmarks comparing the different -10 implementations, including
emulators? How did the SC machines compare with the KL?

For no good reason I reformatted the data from the 2013 post into a
table, below. I can't promise to have not made any errors. Machines
per the original post are

1. A real KL10B with MCA25 cache (DEC2065) executed in user mode
2. KLH10, Intel i7/3930K (slightly over clocked ~4.5 GHz) 16GB
3. Intel NUC DC3217IYE i3 (1.8 Ghz) 8GB

|------------------------------------------+---------+---------+---------|
| instruction | KL-10 | i7 | NUC |
|------------------------------------------+---------+---------+---------|

| INDEXING | 36 ns | 1 ns | 1 ns |

| INDIRECT | 269 ns | 5 ns | 11 ns |

| INDEXING AND INDIRECT | 336 ns | 7 ns | 12 ns |

| MOVEI | 269 ns | 6 ns | 13 ns |

| MOVE FROM AC | 374 ns | 7 ns | 15 ns |

| MOVE FROM MEMORY | 402 ns | 8 ns | 17 ns |

| HRR FROM MEMORY | 443 ns | 9 ns | 19 ns |

| SETOM 0 | 571 ns | 7 ns | 15 ns |

| JRST | 303 ns | 8 ns | 17 ns |

| JSR | 642 ns | 10 ns | 21 ns |

| PUSHJ | 708 ns | 11 ns | 23 ns |

| ADD FROM MEMORY | 438 ns | 15 ns | 31 ns |

| MUL (9 ADD/SUB - 18 SHIFTS) | 2.42 us | 51 ns | 77 ns |

| DIV | 4.70 us | 304 ns | 663 ns |

| FIX A FLOATING POINT ONE | 874 ns | 15 ns | 28 ns |

| FLTR AN INTERGER ONE | 881 ns | 41 ns | 47 ns |

| FAD (1 RIGHT SHIFT) | 1.59 us | 31 ns | 62 ns |

| FAD (8 SHIFT RIGHT - 3 LEFT) | 1.81 us | 48 ns | 98 ns |

| FMP (7 ADD/SUB - 14 SHIFTS) | 2.56 us | 62 ns | 132 ns |

| FDV | 4.82 us | 178 ns | 376 ns |

| DMOVE FROM MEMORY | 608 ns | 12 ns | 24 ns |

| DFAD (1 RIGHT SHIFT) | 2.06 us | 60 ns | 130 ns |

| DFAD (8 SHIFT RIGHT - 1 LEFT) | 2.06 us | 60 ns | 130 ns |

| DFMP (7 ADD/SUB - 32 SHIFTS) | 4.25 us | 131 ns | 282 ns |

| DFDV | 8.71 us | 2.01 us | 4.53 us |

| CONO PI | ?? ns | 11 ns | 27 ns |

| CONI PI | ?? ns | 10 ns | 20 ns |

| DATAO APR | ?? ns | 10 ns | 22 ns |

| DATAI A{R | ?? ns | 10 ns | 23 ns |

| MOVE TO MEMORY | 573 ns | 8 ns | 18 ns |

| LOGICAL SHIFT (35 PLACES LEFT) | 439 ns | 12 ns | 24 ns |

| LOGICAL SHIFT (35 PLACES RIGHT) | 470 ns | 12 ns | 24 ns |

| LOGICAL SHIFT COMBINED (71 PLACES LEFT) | 743 ns | 16 ns | 34 ns |
| LOGICAL SHIFT COMBINED (71 PLACES RIGHT) | 745 ns | 16 ns | 34 ns |

| INCREMENT BYTE POINTER | 924 ns | 18 ns | 33 ns |

| INCREMENT AND LOAD BYTE | 1.21 us | 23 ns | 46 ns |
| INCREMENT AND DEPOSIT BYTE | 1.60 us | 27 ns | 54 ns |

| JFCL | 404 ns | 5 ns | 12 ns |

| CAI | 272 ns | 5 ns | 11 ns |

| JUMP | 342 ns | 5 ns | 11 ns |

| CAM | 376 ns | 6 ns | 12 ns |

| EQV AC TO AC | 402 ns | 10 ns | 18 ns |

| EQV MEMORY TO AC | 742 ns | 10 ns | 22 ns |
| SETOB | 571 ns | 9 ns | 18 ns |
| AOS TO MEMORY | 539 ns | 9 ns | 19 ns |
| EXCHANGE AN AC WITH AN AC | 545 ns | 9 ns | 20 ns |

| EXCHANGE AN AC WITH MEMORY | 706 ns | 11 ns | 22 ns |

| EXECUTE | 471 ns | 8 ns | 14 ns |

| BLT MEMORY TO MEMORY | 1.58 us | 16 ns | 33 ns |

| BLT AC TO MEMORY | 1.48 us | 16 ns | 32 ns |

| DATAI | ?? ns | 14 ns | 28 ns |

| DATAO | ?? ns | 12 ns | 26 ns |

|------------------------------------------+---------+---------+---------|