To : Richard Miller SUVL NOV 12, 1991
From : Roy Stedman ATX
CC : Scott Matson,
John Chertude,
Ira Goldstein ATX
Combo Blitter Test Vector Evaluation - Updated
TASKS
-----
I. Evaluate current test vector
II. Verify performance of existing vector
III. Improve vector for better testing
Interpretation of the results in (II) was by referance to the statistics
of combo blitter failiures sent 8/14/91 from George Chang along with
dismounted defective parts. These statistics may not be entirely accurate
as several errors are reported that should not appear as they are listed.
Five complete boards with defective combo blitters were tested at ATX.
RESULTS
-------
I.
The vector I used was the input vector that is fed to the simulator,
the results of which are used to form the complete test vector for the part.
I did not look at the halftone RAM tests, as they were not
commented and appear to work reasonably well.
Blit 1: 1 word blit copied to set up the chip. Endmask 1 = $fffe
Blit 2: An 8x4 block blit copied.
Blit 3: Same as blit 2, but skew = 1
II.
Two of the blitters tested at ATX should fail this vector.
Many of the dismounted blitters should fail it (up to 37 out of 84).
Either the test vector resulting from simulator runs is incorrectly
formed, testing is not being done properly, or the vector's performance
does not correspond to its comments.
III. a)
Suggested improvements to the vector:
Add blit of a 40x1 block (corresponding to production test cartrige)
Change the source data for greater bit and pattern coverage.
Vary the source and destination addresses, use high word of addresses.
(if possible)
Add blits of varying operations and halftone operations.
Add blit of varying Source X & Y increments.
Add blit of varying Destination X & Y increments.
Add blits using varying endmasks.
Add blit using larger (> 8 bits) skew.
Add reverse blit.
Many of these could be combined into a few blit operations
III. b)
Vector improvements as of 11/12/91:
Two Blit operations were added, resulting in:
Blit of 32 x 1 block
Source Data now has better coverage
Multiple blit operations are now used
Reverse Blits added with varying source and destination increments
Endmask values provide better masking tests
Large skew blits added
Still Need To Be Done:
(require assistance or larger time investment)
Vary and enlarge Source and Destination addresses (use upper word)
Add blits using halftone ops
Provide more thorough testing of operations, halftones, endmasks, etc