https://github.com/zchrissirhcz/neon_sim

NEON intrinsics reference implementation in naive C
https://github.com/zchrissirhcz/neon_sim

arm intrinsics neon simulation

Last synced: 2 months ago
JSON representation

NEON intrinsics reference implementation in naive C

• Host: GitHub
• URL: https://github.com/zchrissirhcz/neon_sim
• Owner: zchrissirhcz
• License: mit
• Created: 2022-03-20T16:13:01.000Z (almost 3 years ago)
• Default Branch: main
• Last Pushed: 2024-04-06T09:21:42.000Z (10 months ago)
• Last Synced: 2024-04-06T10:25:14.126Z (10 months ago)
• Topics: arm, intrinsics, neon, simulation
• Language: C
• Homepage:
• Size: 517 KB
• Stars: 20
• Watchers: 4
• Forks: 4
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# neon_sim

ARM NEON Intrinsics implementation in C, for accurate understanding of each "neon function".

## Usage

Replace

```cpp

#if __ARM_NEON

#include 

#endif // __ARM_NEON

```

with

```cpp

#include "neon_sim.h"

```

## Features

- Real cross-platform

    - 100% in C/C++, no SSE stuffs

    - "Debuggable inside the intrinsics"

- Closer to native neon usage

    - Support initializer list

        ```cpp

        int8x8_t a = { 1, -2, 3, 4, 5, 6, 7, 8 };

        ```

    - Support compile time vector type checking

        ```cpp

        uint8_t b[4] = { 5, 6, 7, 8 };

        uint16x4_t v_b = vld1_u16(b);

        ```

    - Support same-length-different-type conversion(require `-flax-vector-conversions` sometimes)

## Known issues

1. The correctness of the simulation implementation is not guaranteed. 

2. However, the accuracy can be improved by adding examples and continuously verifying.

3. Not support inline assembly, i.e. `asm volatile("...")` stuffs.

4. `vgetq_lane_f32(sum_vec, 4)` failed to check-and-report index out of bounds of `[0, 3]` in compile time.