Torq Compiler
main
Introduction
Quickstart
Model Preparation
Input Guide
Step-by-Step Model Deployment Examples
Supported Operators
Profiling
Handling Unsupported Ops
Compiling Models for Custom Synaptics SoCs
Debug tips and tricks
References
Glossary
Release Notes
Torq Compiler
Index
Edit on GitHub
Index
Symbols
|
A
|
B
|
C
|
D
|
H
|
I
|
K
|
L
|
M
|
N
|
P
|
S
|
T
|
V
Symbols
: IREE Virtual Machine executed at runtime to run the compiled VMFB\
: Virtual Machine FlatBuffer: this is the default output format of files generated
: Application Programming Interface to create binary code for the Torq HW.
: dialects are the mechanism used in MLIR to define new attributes, operations, and types
: in compiler technology to lower means to rewrite complex operations in term of simpler ones
: In MLIR passes represent the basic infrastructure for transformation and optimization of
: in this context the code that implements a given layer or operation in a model. The words
: Intermediate Representation
: Intermediate Representation Execution Environment, based on MLIR\
: Machine Learning
: Multi-Level Intermediate Representation project, based on LLVM\
: N-Dimensional Loop: an NDL agent is an HW block able to generate an address sequence
: originally an acronym for Low Level Virtual Machine, it has broadened in scope and
: Proprietary Synaptics NPU
: Static Single Assignment form, is a common representation used in compilers where each variable
: Tensor Operator Set Architecture: introduced by ARM, it is a standard specifying a set of
{term}IRs\
A
and allow to model different levels of abstractions in an uniform way in the same IR\
B
by the IREE compiler.
C
corresponding to multiple levels of nested for-loops, each specified with a counter and a stride.
D
dialect
H
https://en.wikipedia.org/wiki/LLVM
https://en.wikipedia.org/wiki/Static_single-assignment_form
https://iree.dev
https://iree.dev/developers/design-docs/vm/#virtual-machine-vm
https://mlir.llvm.org/
https://mlir.llvm.org/docs/LangRef/#dialects
https://mlir.llvm.org/docs/PassManagement/
https://www.mlplatform.org/tosa/
HW API
I
IR
IREE
is assigned exactly once\
is now a toolkit for building compilers\
K
kernel
kernel, layer and operation are often used as synonims.
L
LLVM
lower
M
ML
MLIR
N
NDL
P
pass
S
SSA
T
tensor operations commonly employed by Neural Networks\
This API provides a more abstract interface than accessing HW registers directly.
Torq
TOSA
V
VM
VMFB
VMFB files have .vmfb extension and can be executed on the IREE runtime VM
User Manual
main (latest)
Versions
main (latest)