Verilog Implementation of TAGE based predictor by Andre Seznec and Pierre Michaud The implementation is based on this paper http://www.irisa.fr/caps/people/seznec/JILP-COTTAGE.pdf
I have done some modification with respect to the tag length, this was to limit the hardware bugdet by varying the tag- length and number of indexes
The TAGE predictor has 1 Bimodal Table and 4 TAGGED predictor Component with different geometric lengths respectively 7, 14, 44, 130
Table 1 . Tag-length of Different tag component in different models
| Model No | T1 | T2 | T3 | T4 | No of Entries | Total hardware budget in Bits |
|---|---|---|---|---|---|---|
| 2 | 8 | 8 | 9 | 9 | 512 | 29174 |
| 3 | 5 | 5 | 5 | 5 | 1024 | 44032 |
The source directory has model 2(only the index_Tag.Generator.v file is changed for different models). The Model 3 is there in directory Different_TAG_Lengths.
I have provided a sample test benchmark. The trace was obtained from the 2011 championship branh predictor tournament.They are in Zip file Branch_Benchmark.zip
The implementation is basic, you can optimize the verilog code as per your requirement based on the many TAGE papers published.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent