###### tags: `Embedded`

# Soft Floating Point Performance on ARMv7M

Cycle time of various operations using GNU libgcc. Probably using `gcc-arm-none-eabi-8` or `gcc-arm-none-eabi-9` juding from the time posted, the original author didn't mention the version.

> The code has been generated for Cortex-M4 (Arm architecture v7EM) and runs on an NXP Kinetis K66.

## Float, Math

| Operation | Min | Max | Avg |
| -------- | -------- | -------- | -------- |
| `__aeabi_fadd` | 50 | 54 | 52.8 |
| `__aeabi_fsub` | 51 | 61 | 56.3 |
| `__aeabi_frsub` | 53 | 63 | 58.4 |
| `__aeabi_fmul` | 30 | 30 | 30.0 |
| `__aeabi_fdiv` | 144 | 144 | 144.0 |

## Float, Conversion

| Operation | Min | Max | Avg |
| -------- | -------- | -------- | -------- |
| `__aeabi_f2iz` | 13 | 13 | 13.0 |
| `__aeabi_f2uiz` | 11 | 11 | 11.0 |
| `__aeabi_f2lz` | 261 | 382 | 310.4 |
| `__aeabi_f2ulz` | 194 | 307 | 239.4 |
| `__aeabi_i2f` | 26 | 26 | 26.0 |
| `__aeabi_ui2f` | 26 | 26 | 26.0 |
| `__aeabi_l2f` | 28 | 29 | 28.5 |
| `__aeabi_ul2f` | 28 | 28 | 28.0 |
| `__aeabi_f2d` | 8 | 8 | 8.0 |

## My application

`__aeabi_ui2f`: 11
`__aeabi_fmul`: 16
`__aeabi_f2uiz`: 2
`__aeabi_l2f`: 1
`__aeabi_fadd`: 2
`__aeabi_f2lz`: 5
`__aeabi_fsub`: 2
`__aeabi_fdiv`: 2

$3 * 26 + 6 * 30 + 1* 6 1 + 1 * 144 + 2 * 382 = 1227$

## References
1. https://blog.segger.com/wp-content/uploads/2019/11/Bench_Comparison.pdf