https://github.com/sashi0034/amx-basic-example
Intel Advanced Matrix Extensions Basic Example
https://github.com/sashi0034/amx-basic-example
amx intel
Last synced: about 2 months ago
JSON representation
Intel Advanced Matrix Extensions Basic Example
- Host: GitHub
- URL: https://github.com/sashi0034/amx-basic-example
- Owner: sashi0034
- Created: 2025-01-27T06:49:40.000Z (5 months ago)
- Default Branch: main
- Last Pushed: 2025-02-04T07:12:08.000Z (4 months ago)
- Last Synced: 2025-04-30T10:24:24.979Z (about 2 months ago)
- Topics: amx, intel
- Language: C
- Homepage:
- Size: 15.6 KB
- Stars: 2
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# amx-basic-example
This repository contains unofficial example code using [Intel AMX](https://www.intel.com/content/www/us/en/products/docs/accelerator-engines/advanced-matrix-extensions/overview.html).
# Matrix product operation using AMX
Pay attention to the memory layout of B in the multiplication instruction A * B.
- `int8_mut`: int8 matrix product operation
```
cd int8_mul
icc main.c -o int8_mul
```
- `bf16_mul`: BF16 matrix product operation
```
cd bf16_mul
icc main.c -o bf16_mul
```
# Convolutional operation using AMX
Convolutional operations using AMX require unique handling.
The approach is similar to im2col, which is commonly used on GPUs.
Note that AMX multiply instructions are accumulative, not overwritten.
- `int8_conv`: int8 convolution operation
```
cd int8_conv
icc int8_conv -o int8_conv
```
# References
- [Intel Intrinsics Guide](https://www.intel.com/content/www/us/en/docs/intrinsics-guide/index.html#!=undefined&techs=AMX)