OpenFPGA/tutorials/fpga_flow/how2use.md

97 lines
5.6 KiB
Markdown
Raw Normal View History

2019-07-10 11:03:30 -05:00
# FPGA flow
2019-07-10 11:03:30 -05:00
This tutorial will help the user to understand how to use OpenFPGA flow.<br />
2019-07-15 22:16:15 -05:00
During this tutorial we consider the user start in the OpenFPGA folder and we'll use tips and informations provided in [tutorial index](https://github.com/LNIS-Projects/OpenFPGA/blob/master/tutorials/tutorial_index.md#tips-and-informations). Details on how the folder is organized are available [here](https://github.com/LNIS-Projects/OpenFPGA/blob/master/tutorials/fpga_flow/folder_organization.md).
2019-07-10 11:03:30 -05:00
## Running fpga_flow.pl
2019-07-10 11:03:30 -05:00
A script example can be found at OPENFPGAPATHKEYWORD/fpga_flow/tuto_fpga_flow.sh.
2019-07-10 11:37:05 -05:00
### Experiment
2019-07-10 11:03:30 -05:00
cd fpga_flow<br />
./tuto_fpga_flow.sh<br />
2019-07-10 11:37:05 -05:00
### Explanation
2019-07-10 11:03:30 -05:00
By running this script we took an architecture description file, generated its netlist, generated a bitstream to implement a benchmark on it and verified this implementation.<br />
When you open this file you can see that 2 scripts are called. The first one is **rewrite_path_in_file.pl** which allow us to make this tutorial generic by generating full path to dependancies.<br />
2019-07-15 22:16:15 -05:00
The second one is **fpga_flow.pl**. This script launch OpenFPGA flow and can be used with a lot of [options](https://github.com/LNIS-Projects/OpenFPGA/blob/master/tutorials/fpga_flow/options.md).<br />
2019-07-10 17:00:22 -05:00
There is 3 important things to see:
- All FPGA-Verilog options have been activated
- fpga_flow.pl calls a configuration file through "config_file" variable
- fpga_flow.pl calls a list of benchmark to implement and test through "bench_txt" variable
2019-07-16 12:52:24 -05:00
### Configuration file
2019-07-08 11:36:58 -05:00
In this file paths have to be full path. Relative path could lead to errors.<br />
The file is organized in 3 parts:
2019-07-08 17:18:08 -05:00
* **dir_path**: provides all the tools and repository path
* **flow_conf**: provides information on how the flow run
* **csv_tags**: *to complete*
While empty the file is as follow:
2019-07-08 11:36:58 -05:00
[dir_path]<br />
script_base = OPENFPGAPATHKEYWORD/fpga_flow/scripts<br />
benchmark_dir = *<Path to the folder containing all sources of the design>*<br />
2019-07-08 11:36:58 -05:00
yosys_path = OPENFPGAPATHKEYWORD/yosys<br />
odin2_path = not_used<br />
cirkit_path = not_used<br />
abc_path = OPENFPGAPATHKEYWORD/abc<br />
abc_mccl_path = OPENFPGAPATHKEYWORD/abc<br />
abc_with_bb_support_path = OPENFPGAPATHKEYWORD/abc<br />
mpack1_path = not_used<br />
m2net_path = not_used<br />
mpack2_path = not_used<br />
vpr_path = OPENFPGAPATHKEYWORD/vpr7_x2p/vpr<br />
rpt_dir = *<wherever you want logs to be saved>*<br />
2019-07-08 11:36:58 -05:00
ace_path = OPENFPGAPATHKEYWORD/ace2<br />
[flow_conf]<br />
flow_type = yosys_vpr *to use verilog input*<br />
vpr_arch = *<wherever the architecture file is saved>*<br />
2019-07-08 11:36:58 -05:00
mpack1_abc_stdlib = DRLC7T_SiNWFET.genlib # Use relative path under ABC folder is OK<br />
m2net_conf = not_used<br />
mpack2_arch = not_used<br />
power_tech_xml = *<wherever the xml tech file is saved>*<br />
2019-07-08 11:36:58 -05:00
2019-07-10 17:00:22 -05:00
[csv_tags]<br />
2019-07-08 11:36:58 -05:00
mpack1_tags = Global mapping efficiency:|efficiency:|occupancy wo buf:|efficiency wo buf:<br />
mpack2_tags = BLE Number:|BLE Fill Rate: <br />
vpr_tags = Netlist clb blocks:|Final critical path:|Total logic delay:|total net delay:|Total routing area:|Total used logic block area:|Total wirelength:|Packing took|Placement took|Routing took|Average net density:|Median net density:|Recommend no. of clock cycles:<br />
vpr_power_tags = PB Types|Routing|Switch Box|Connection Box|Primitives|Interc Structures|lut6|ff<br />
2019-07-10 17:00:22 -05:00
*This example file can be found at OPENFPGAPATHKEYWORD/fpga_flow/configs/tutorial/tuto.conf*
2019-07-16 12:52:24 -05:00
### Benchmark list
The benchmark folder contains 3 sub-folders:
2019-07-08 17:18:08 -05:00
* **Blif**: contains .blif and .act of benchmarks
* **List**: contains all benchmark list files
* **Verilog**: contains Verilog designs
2019-07-09 15:28:14 -05:00
Blif and Verilog folders are organized by folders with the name of projects. **Folder, top module and top module file must share the same name.**<br />
The benchmark list file can contain as many benchmarks as available in the same folder targetted by "benchmark_dir" variable from the configuration file. It's written as:<br />
top_module/*.v,<int_value>; where <int_value> is the number of channel/wire between each block.
2019-07-10 17:00:22 -05:00
*This example file can be found at OPENFPGAPATHKEYWORD/fpga_flow/benchmarks/List/tuto_benchmark.txt*
2019-07-16 12:52:24 -05:00
## Modifying flow
Once dependancies are understood, we can modify the flow by changing the architecture and the route channel width.
### Experiment
* cd OPENFPGAPATHKEYWORD/fpga_flow/configs/tutorial
* replace the architectures "k6_N10_sram_chain_HC_template.xml" and "k6_N10_sram_chain_HC.xml" respectively with "k8_N10_sram_chain_FC_template.xml" and "k8_N10_sram_chain_FC.xml" in tuto.conf
* cd OPENFPGAPATHKEYWORD/fpga_flow/benchmarks/List
* replace "200" with "300" in tuto_benchmark.txt
* cd OPENFPGAPATHKEYWORD/fpga_flow
* replace the architectures "k6_N10_sram_chain_HC_template.xml" and "k6_N10_sram_chain_HC.xml" respectively with "k8_N10_sram_chain_FC_template.xml" and "k8_N10_sram_chain_FC.xml" in tuto_fpga_flow.sh
* ./tuto_fpga_flow.sh
### Explanation
2019-07-16 14:13:58 -05:00
With this last experiment we replace the [**K6 architecture**](https://github.com/LNIS-Projects/OpenFPGA/blob/master/tutorials/images/architectures_schematics/frac_lut6.pdf) with a [**K8 architecture**](https://github.com/LNIS-Projects/OpenFPGA/blob/master/tutorials/images/architectures_schematics/frac_lut8.pdf), which means a 8-inputs fracturable LUT (implemented by LUT6 and LUT4 with 2 shared inputs). This architecture provides more modes for the CLB and the crossbar changed from a half-connected to a fully connected, implying bigger multiplexor between the CLB and LUT inputs. These requirement in term of interconnection will lead to the increase in routing channel width. Indeed, if the routing channel is to low, it could be impossible to route a benchmark or the FPGA output can be delayed.