[phylum]<CTamarin>TestVecs>MpyTestVecs.tioga!7

MpyTestVecs.tioga

by Ross May 8, 1987 4:30:33 pm PDT

Last Edited by: Ross May 14, 1987 3:14:09 pm PDT

Vectors to check out the entire multiplier. Try FirstLevelOnly to get the overall picture of what tests are performed. The control assignments are listed below for reference:

EUControl[0:2] = 110 => Load Mpy

EUControl[0:2] = 111 => UnLoad Mpy

EUControl[3:6] = x0xx => Load 32 bit number

EUControl[3:6] = x1xx => Load floating pt. number

Note: the difference between signed and unsigned multipliers when doingUpper is controlled by whether a single cycle of doingLower follows the 8 cycles (of CLK2 not CLK). It makes no sense to consider an unsigned multiply when only doingLower. See the example below for the timing of this operation

state[0:1] = 00 => done

state[0:1] = 01 => doingLower

state[0:1] = 10 => doingUpper

state[0:1] = 11 => Invalid

For testing purposes, the D1 and FloatD2 inputs are tied together as with D2 and FloatD1.

Vectors which have been commented out have all passed but not deleted from the file.

d1 d2 state euctl neuctl | R

—————————————————————————————————————

Initialization:

0 0 0 60 1f | xxxxxxxx

Basics with early completion and waiting after done:

1 1 0 60 1f | xxxxxxxx -- shows early completion

0 0 1 00 7f | xxxxxxxx -- and not removing the product

0 0 1 00 7f | xxxxxxxx -- immediately.

0 0 1 00 7f | xxxxxxxx

0 0 0 00 7f | xxxxxxxx

0 0 0 70 0f | 1

2 3 0 60 1f | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 00 7f | xxxxxxxx

0 0 0 70 0f | 6

Floating Point load:

The interesting thing about these next two is trying out whether the loading of the floating point works. If it does, the numbers should swap and it should take more cycles to do the booth encoding.

78 3245 0 60 1f | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 70 0f | 179058

78 3245 0 64 1b | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 70 0f | 179058

More Serious Examples:

61000000 3486107a 0 60 1f | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 70 0f | 3a000000

12345678 2457afeb 0 60 1f | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 70 0f | bfef6828

Check out the multiply that returns the upper 32 bits:

69000000 19000000 0 60 1f | xxxxxxxx

0 0 2 00 7f | xxxxxxxx

0 0 0 70 0f | 0a410000

61000000 3486107a 0 60 1f | xxxxxxxx

0 0 2 00 7f | xxxxxxxx

0 0 0 70 0f | 13e6cc3e

12345678 2457afeb 0 60 1f | xxxxxxxx

0 0 2 00 7f | xxxxxxxx

0 0 0 70 0f | 02959874

Demonstration of signed vs. unsigned mpy upper:

signed:

80000000 80000000 0 60 1f | xxxxxxxx

0 0 2 00 7f | xxxxxxxx

0 0 0 70 0f | c0000000

Now for the signed (notice the cycle with state 1 after the cycles with state 2):

80000000 80000000 0 60 1f | xxxxxxxx

0 0 2 00 7f | xxxxxxxx

0 0 1 00 7f | xxxxxxxx

0 0 0 70 0f | 40000000

.