Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

sw: Add initial version of script to explore tiled behaviour #38

Draft
wants to merge 19 commits into
base: main
Choose a base branch
from
Draft

sw: Add initial version of script to explore tiled behaviour #38

wants to merge 19 commits into from

Conversation

JosseVanDelm
Copy link

@JosseVanDelm JosseVanDelm commented Oct 25, 2023 •
edited
Loading

You run it like this (note the calling folder!) yay:

/repo/target/snitch_cluster# python3 ../../util/trace/perf_tiled.py

rgantonio and others added 15 commits September 14, 2023 14:21
@rgantonio
Modify Snitch integer core to have CSR decoding (#7)
5fc9f1f 
* hw: Add CSR address range for SNAX

* hw: Add SNAX CSR accelerator port address

* hw: Add CSR decoding for SNAX

* Revert "hw: Add CSR address range for SNAX"

This reverts commit afdd15c.

* hw: Decoupled CSR address range from opcode set

* hw: Update SNAX CSR address to use decoupled definition

* hw: Add new line and fix trailing spaces

* hw: Fix trailing spaces

* bender: Add csr_snax_def.sv to bender list

* lint: Exclude csr_snax_def.sv from license checker

* hw: Add CSR immediate decoding for opb MUX
@rgantonio
Add SNAX core cluster and cluster modifications and connections (#6)
a556e8b 
* hw: Add SNAX ports to core cluster

* hw: Add switchers and generators for SNAX

* hw: Add comment for TCDM generation with SNAX ports
@xiaoling-yi @rgantonio
Add HWPE-MAC for Bender (#10)
56aa3a4 
* bender: add hwpe-mac

* target: Add +define+SYNTHESIS to verilator flags to bypass MAC engine timescale

---------

Co-authored-by: rgantonio <[email protected]>
@xiaoling-yi @rgantonio
Add HWPE MAC files (#12)
280cb54 
* hw: Add HWPE MAC ctrl code

* hw: Add HWPE MAC reqrsp code

* hw: Add HWPE MAC top code

* bender: add HWPE mac files to bender list

* lint: Exclude HWPE MAC files from license checker

* hw: make sure the module name matches the fine name

* bender: change the file name of hwpe mac

* hw: rm the hwpe mac file with wrong filename

* bender: change the order of hwpe mac in bender list

* hw: edit comments

* hw: Clarify comments about the data addressing

---------

Co-authored-by: rgantonio <[email protected]>
@xiaoling-yi
ADD MAC Unit (#13)
a8a1a10 
* hw: add mac unit

* hw: remove extra comments in snitch cluster

* hw: change the name of main mac engine
@rgantonio
Add and modify templates for SNAX (#14)
2a25945 
* docs: Add SNAX definition to schema

* hw: Add SNAX to cluster wrapper template.

* target: Add SNAX configuration for custom use.

* docs: Fix typo for SNAX flag

* cfg: Add explicit SNAX declarations for other templates

* tests: Create separate SNAX run to exclude multicore Snitch tests

* ci: Separate SNAX test

* ci: Clean the HW build for SNAX

* ci: Break long line into multiple for lint

* ci: Fix cfg path for HW build

* ci: Make new line again for lint
@rgantonio
Add SW test for the MAC engine (#15)
c830042 
* sw: Add snax-mac.c test

* sw: Add data for snax-mac test

* Add Makefile for compiling snax-mac

* sw: Add snax-mac to list of to-compile builds

* sw: Re-align data for lint

* sw: Fix lint for snax-mac.c

* lint: Exclude SNAX tests from license checking

* sw: Swap snrt.h and data.h positions

* sw: Fix lint for data.h

* sw: Fix lint for snax-mac.c

* sw: Code cleaning and fix comments for clarity

* tests: Add snax-mac to snax-run.yaml

* Revert "tests: Add snax-mac to snax-run.yaml"

This reverts commit 150b91a.

* sw: Fix comments for snax-mac

* Add snax-rum.yaml for the application test list.

* ci: Modify test in CI to use snax-run.yaml for the apps side
@rgantonio
Fix HWPE MAC output streamer for double word writes (#20)
72b19f7 
* hw: Add special switcher for output streamer

* hw: Add switcher for double word alignment

* hw: Decouple input and output streams in SNAX MAC

* hw: Fix bulk changes to allow 64-bit input and output streaming of HWPE MAC

* sw: Move from 32 bit to 64 bit data

* Fix parameters and comments
@rgantonio
Add simple-mult test for streaming multiplicaton (#21)
edfb964 
* sw: Add snax-mac-simple main code

* sw: Add data for snax-mac-simple test

* sw: Add Makefile for snax-mac-simple

* lint: Exclude snax-test from licensing

* sw: Fix lint for data.h

* sw: Fix lint for snax-mac-simple.c

* sw: Fix lint again for data.h

* sw: Fix lint again x2 for data.h

* sw: Modify snax-mac-simple to use 64 bits

* sw: Take out unused variable

* sw: Add CPU checker to compare with MAC output

* sw: Add snax-mac-simple test to the test suite

* sw: Add new line for data.h

* Revert "sw: Add new line for data.h"

This reverts commit 7dc87fe.

* sw: Add proper newline to data.h
@rgantonio
Fix TCDM ports to take out unused TCDM ports during generation (#23)
06c3f34 
* hw: Fix snitch_cluster to have better modularity for SNAX tcdm ports

* schema: Add the default snax_tcdm_ports configuration

* tpl: Modify snitch cluster template for snax ports

* hw: Fix lint for trailing spaces

* tpl: Add the configuration for the snitch cluster

* cfg: Update snax.hjson config to include the SNAX TCDM ports
@rgantonio @xiaoling-yi
Add SNAX GEMM (#25)
a84b80b 
* bender: Add snax-gemm to bender

* bender: Update bender with latest snax-gemm repo

* hw: Add snax-gemm.sv

* hw: Modify snitch_cluster and add snax-gemm instance

* hw: Modify snitch cluster template

* docs: Modify snitch cluster schema

* lint: Exclude snax_gemm from license check

* hw: Lint snax_gemm

* cfg: Modify snax.hjson with new config

* improve gemm

* modify gemm.sv

* hw: modify status CSR

* cfg: Add snax-hjson configuration

* tests: Renamed snax-run.yaml to snax-mac-run.yaml

* cfg: Modify to use MAC generation

* cfg: Rename snax.hjson to snax-mac.hjson

* ci: Change build to SNAX-MAC runs

* cfg: Modify snax-gemm.hjson

* tests: Add snax-gemm-run.yaml

* ci: Add the SNAX-GEMM in CI

* ci: Fix lint indentation

* cfg: Take out original Snitch compute core templates

---------

Co-authored-by: xiaoling-yi <[email protected]>
@rgantonio
Add GEMM sw tests (#27)
6854212 
* sw: Add GEMM sw test

* sw: Lint codes

* lint: Add license

* sw: Add base test without the GEMM engine

* sw: Add engine and base test to apps Makefile

* sw: Add GEMM tests to test list

* lint: Exclude GEMM tests from license check

* sw: Add new lines at the end of each file
@rgantonio
Update HWPE MAC to use 32-bits only (#32)
b5497a7 
* hw: Update SNAX control for 32-bit support

* hw: Update SNAX hwpe to reqrsp to support 32-bits

* hw: Take out unused parameter

* sw: Move uint64_t to uint32_t

* sw: Take out unnecessary debug code

* sw: Add new line to data.h
@JosseVanDelm @rgantonio
Add tiled simple mult example (#33)
3103b97 
* sw: Copy simple mac example and refactor

* sw: Refactor accelerator setup

* sw: Add function for launching accelerator

* sw: Refactor correctness check

* sw: Working multiple SNAX runs

* sw: Added duplicate runs

* sw: Make duplicate runs in for loop

* sw: Fix lint for code

* sw: Move snax_mac_sw_clear to sw_barrier

* sw: Tile execution of MAC operation

* sw: Add data in to tiled implementation

* sw: Add output transfer to double buffering loop

* sw: Add license headers

* sw: Fix linting errors

* sw: Add tiled mac to makefile and runfile

* sw: Add datagen

* sw: Fix lint

* sw: Resolve minor comments

* sw: Add datagen.py to Makefile

* sw: Fix lint

* sw: Add performance coutners outside tiling loop

* sw: Fix lint

---------

Co-authored-by: rgantonio <[email protected]>
@JosseVanDelm
Add data generation for both simple vec examples (#36)
02d5f2a 
* sw: Move files into snax-mac-simple folder

* sw: Fix relative paths for CI after move

* sw: Rename files for move

* sw: Adapt relative paths for compilation

* sw: Rename snax-mac-simple to untiled

* sw: Use datagen to drive untiled example

* sw: Fix C linting
@rgantonio
Copy link

I will make a fix for this wait. It will be in a separate PR.

rgantonio
rgantonio approved these changes Oct 30, 2023
View reviewed changes
Copy link

@rgantonio rgantonio left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@JosseVanDelm I am okay with this.

This does not need to be part of the CI right? Because it doesn't make sense to me to generate .perf files during CI since we're not checking performance anyway. CI was meant to just see if it builds right?

Also, let's first rebase the PR I made for the signal fix.

util/trace/perf_tiled.py
2. run bin/snitch_cluster.vlt something.elf
3. make traces
4. Process logs
5. TODO store logs somewhere
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I suggest to dump it in the same logs directory that is generated when we build tests? The one in /repo/target/snitch_cluter/logs because it's also automatically part of the .gitignore that was previously present.

@JosseVanDelm JosseVanDelm mentioned this pull request Oct 31, 2023
Merged
@JosseVanDelm
Refactor simple mac examples (#40)
56a9660 
* sw: Refactor mac library to be outside of examples

* sw: Refactor untiled example to use utilities from library

* sw: Remove all performance counters

* sw: Refactor accelerator setup

* sw: Refactor correctness check

* sw: Build snax/mac as dependency of apps

* sw: Make lint happy

* sw: Add stdint to includes to make clang-format happy

Clang-format reorders includes to alphabetical order, which results in some includes being broken

* sw: Insert missing barrier in untiled example

* sw: Add KU Leuven to License checker

* sw: Make lint happy

* sw: Address Xiaoling's comments

* sw: Remove unnecessary mac.o dependency from makefile

* sw: Add SNAX as dependency to tests
@JosseVanDelm
sw: Add initial version of script to explore tiled behaviour
d8bcb6d
@JosseVanDelm
sw: Add post-processing to perf_tile.py
b709e30
@JosseVanDelm
Copy link
Author

@rgantonio CI is meant for continuous integration.
If we have performance tests than it would make sense to also keep track of performance with new versions of our codebase. In some scenario's performance regressions are just as important to track as functional behavior regressions.
Regardless, this script is definitely not meant for being run all the time, it is more for us to explore tiled behaviour in a personal setting. However, once we get a good grasp of a sensible tiling size etc. we might want to add performance tests for specific accelerators as well.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@rgantonio rgantonio rgantonio approved these changes

Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@JosseVanDelm @rgantonio @xiaoling-yi