/
2025-02-11 Ordinary Meeting Notes

2025-02-11 Ordinary Meeting Notes

Date

Oct 14, 2024

Disclosures

RVI_Disclosures_Female_Short.mp4

Participants

Name

Affiliation

Name

Affiliation

Paul Kirth (Google)

Google

Christoph Müllner (VRULL)

VRULL

Roberto Avanzi (Qualcomm, Inc.)

Qualcomm

Samuel Holland (SiFive)

SiFive

Yann Loisel (SiFive)

SiFive

Makeljana Shkurti (VRULL)

VRULL

Nick Kossifidis (Foundation for Research and Technology - Hellas (FORTH))

Foundation for Research and Technology - Hellas (FORTH)

Shao-Chung Wang (Andes Technology Corporation)

Andes Technology Corporation

Thurston Dang (Google LLC)

Google

Alexey Baturo

Individual

Prashanth Mundkur (RISC-V International)

Individual

Florian Mayer (Google LLC)

Google LLC

deepak gupta (Rivos Inc)

Rivos Inc

Luís Fiolhais (Individual)

Individual

David Schrammel

Individual

Adam US (NVIDIA Corporation)

NVIDIA Corporation

Radim Krčmář (Ventana Micro Systems)

Individual

Konrad Hohentanner

-

Agenda

Presentations

Title

Presenter

File

Title

Presenter

File

1

QEMU analysis of glibc for riscv zimte

Christoph Müllner

https://drive.google.com/drive/folders/1UM0eWtmUsZYCO1BFOZ25bWShpYJ-2EbZ

2

 

 

 

Notes & Action Items

Key outcomes
Allocator need to access chunk headers which must be with a constant tag value (tag == 0) or elide tag checks on those checks. Current zimte spec uses pointer_tag == 0 as elision for tag checks. It is going to change as it provides bypass. If TG ends up defining a elide mechanism, that will be useful for these needs as well.
Focus on 4 bit tags but keep ISA (and thus allowing software as well) tagwidth agnostic.
ldtag/gettag usage is still not clear
special invalid tag makes generation of tag and addition of tag flows a bit cumbersome due to the fact hw/sw need to work together to avoid that special invalid encoding. We added this to make good use of one extra bit (in mctag) we have with 7 bit tagging. Although to make things homogenous with 4 bit tagging, we ended up using an entire encoding. If we focus only on 4 bit tagging, we can get rid of notion of invalid tag and let it get introduced whenever a new extension comes for 7 bit tags

 


Standard_2.png

RISC-V International