Turn on suggestions

Auto-suggest helps you quickly narrow down your search results by suggesting possible matches as you type.

Showing results for

- Intel Community
- FPGAs and Programmable Solutions
- FPGA Intellectual Property
- ALTFP_MATRIX_MULT problem in Quartus 9.1

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Mute
- Printer Friendly Page

Highlighted
##

Altera_Forum

Valued Contributor III

- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Email to a Friend
- Report Inappropriate Content

11-11-2009
01:16 AM

698 Views

ALTFP_MATRIX_MULT problem in Quartus 9.1

We are not able to generate a single precision floating point matrix multiply under 9.1 in the same way that we did under 9.0sp2.

Under Quartus 9.0sp2, MegaWizard Plug-In Manager allows Matrix AA to be 56x56 and Matrix BB to be 56x1, with a vector size of 8 and a block size of 2, for a Stratix III dev board with a EP3SL150 chip. Quartus 9.1 modified the ALTFP_MATRIX_MULT to allow double precision floating point, and single precision complex. It has clearly been reworked. However, we can't find any updated documentation thus far - only documentation from March 2009. Under 9.1, in the MegaWizard Plug-In manager, selecting 56 columns for matrix AA succeeds. Then selecting 56 rows for matrix AA succeeds. Finally, selecting "1" for the columns in matrix BB generates the following errors in the feedback window: -------------- Error: of object VECTORSIZE with settings BLOCKS(2), VECTORSIZE(8) did not evaluate to be a boolean expression, but evaluates to Error: Condition (VECTORSIZE/(2*BLOCKS) lte 6) Error: Illegal integer values 8, true in divide/mod operation in expression (VECTORSIZE/(2*BLOCKS) lte 6) Error: of object VECTORSIZE with settings BLOCKS(2), VECTORSIZE(8) did not evaluate to be a boolean expression, but evaluates to Error: Condition (VECTORSIZE/(2*BLOCKS) gte 2) Error: Illegal integer values 8, true in divide/mod operation in expression (VECTORSIZE/(2*BLOCKS) gte 2) Error: Parameter COLUMNSBB has been assigned the value 1. The legal range is 8..64 ------------------ As far as I can tell, these aren't coherent error messages. It's not possible to switch the matrices - trying to select matrix AA rows=1 is not allowed. Is this a fundamental design change, or is it maybe just an error in the user input validation rules that the MegaWizard is using? For now we will continue to use the multiplier generated with 90sp2 tools. Thanks, Mike
0 Replies

For more complete information about compiler optimizations, see our Optimization Notice.