Cards
eugene created this presentation simply by filling in the following story cards:1 style:designer
Doing High-level Synthesis ?
[image1]
Is RTL *coverage* a requirement?
2 style:narrative
A case study of using JEDAcc for HLS
3 style:smallPicture
8051 Processor
[image1]- 800+ lines of C code
- 8-bit uP
- 4k program memory,
- 128 byte data memory
- up to 64k external memory
4 style:typography
8051 Testbench
[image1]sort, fib, xram, all
5 style:highlight
Measure code coverage at HL using JEDA
[image1]
6 style:designer
Measure coverage using other tools: 88%
[image1]
*False High*
7 style:designer
Using JEDA HW-Aware coverage product JEDAcc: Only 25%
[image1]JEDA uncovers coverage holes others can't
8 style:highlight
Measure code coverage for RTL model
[image1]9 style:bullet
Measure RTL Coverage
- Line coverage at 85%
- Decision coverage at 78%
10 style:highlight
Correlation between HL & RTL
[image1]RTL coverage grows with HL as TB improves and new tests added
11 style:typography
(Excluded from the show)12 style:designer
Conclusions
- *HW-aware* Coverage generates similar results to RTL
- Remove *Verification* weak link in HLS flow
- RTL coverage is redundant for HLS
- Move verification to high level, 10x-100x faster
