RISC V Vector Extension Proposal
summary
Proposes a compact RISC-V vector ISA with 32 vector registers, a configurable VL register that controls how many lanes execute (examples use VL=5), per-register element types and three shapes (scalar/vector/matrix), strided loads, indexed gather/scatter, vector atomics, and a single convert opcode. Implementation mandates zeroing of inactive elements to allow out-of-order designs, stores write only VL elements, memory ordering follows loop semantics, and Esperanto reports an early RTL prototype.
AI-written summary · about
chapters (21)
- 0:00Introduction and motivation
- 1:23Vector ISA features and benefits
- 2:25Registers and vector state
- 3:53Vector instruction set basics
- 5:41Simple vector add example
- 7:03SIMD vs vector execution
- 8:49Scalar operations in vectors
- 10:33Masked execution explained
- 12:37Vector memory loads
- 14:42Indexed vector loads
- 16:05Vector memory stores
- 17:08Memory ordering guarantees
- 18:32Typed vector registers
- 22:07Variable vector length support
- 23:10Register file organization
- 27:05Software usage and set VL
- 28:29Encoding and instruction format
- 29:11Exceptions and future plans
- 30:35Speaker’s vector ISA history
- 31:38Software support and next steps
- 32:44Q&A on page faults
related talks
13mUnleashing ML Processing Power Through RISC-V Vectors: Applications, Implementation and...
01h 45mPorting SIMD to RISC-V Vectors
01h 45mPorting SIMD to RISC-V Vectors | Corrected Audio
11mSimulate, trace, and evaluate a RISC-V system leveraging very long vectors
12mDemo: Boosting AI on Semidynamics RISC-V Cores with Custom Tensor Instructions
24mMoving to RISC-V Vector: A Practical Journey of AI Operator Optimization
23mSemidynamics New Family of High Bandwidth Vector-Capable Cores
22mMatrix Extensions for RISC-VDelivering on the Promise
