This is a finger exercise in SIMD-optimization. The operation implemented is:
- We have a bitonal bitmap (=one bit per pixel).
- In a second bitmap, we want to set each pixel which is a one (or a zero) in the bitonal bitmap to a specific value.
The bitonal bitmap here is organized as follows: the most significant bit corresponds to the leftmost pixel. Here is an example for "FillFromBitonalFromOnes" with a "valueForOnes" of 0x23:
The repository contains implementations leveraging AVX-instructions (Intel/AMD) and Neon (ARM).
cd build
mkdir build
cmake .. -DCMAKE_BUILD_TYPE=Release
cmake --build . --config Release
Performance for the operation "FillFromOnes", 1MB=1,000,000 bytes.
i9-12900K, VisualStudio2022, x64
| version | performance in MB/s |
|---|---|
| gray8_C | 320.7 |
| gray8_AVX | 8297.7 |
| gray16_C | 648.7 |
| gray16_AVX | 8407.2 |
| BGR24_C | 1014.9 |
| BGR24_AVX | 8089.6 |
| BGR48_C | 1887.1 |
| BGR48_AVX | 7780.4 |
| float32_C | 1336.1 |
| float32_AVX | 7576.2 |
Raspberry Pi 4, arm64, gcc
| version | performance in MB/s |
|---|---|
| gray8_C | 160.6 |
| gray8_Neon | 1592.7 |
| gray16_C | 320.0 |
| gray16_Neon | 1561.4 |
| BGR24_C | 461.5 |
| BGR24_Neon | 1533.6 |
| BGR48_C | 921.2 |
| BGR48_Neon | 1449.1 |
| float32_C | 636.6 |
| float32_Neon | 1514.6 |
Galaxy Book Go (Snapdragon 7c Gen2), arm64, VisualStudio2022
| version | performance in MB/s |
|---|---|
| gray8_C | 256.3 |
| gray8_Neon | 4932.1 |
| gray16_C | 531.0 |
| gray16_Neon | 6325.8 |
| BGR24_C | 754.6 |
| BGR24_Neon | 5315.1 |
| BGR48_C | 1446.7 |
| BGR48_Neon | 6339.4 |
| float32_C | 1009.3 |
| float32_Neon | 6302.6 |
AMD Ryzen 5 1600X (Linux), x64, gcc9.3
| version | performance in MB/s |
|---|---|
| gray8_C | 334.6 |
| gray8_Neon | 1271.8 |
| gray16_C | 678.2 |
| gray16_Neon | 2435.5 |
| BGR24_C | 986.3 |
| BGR24_Neon | 2818.2 |
| BGR48_C | 1974.8 |
| BGR48_Neon | 3026.3 |
| float32_C | 1358.1 |
| float32_Neon | 3026.3 |
AMD Ryzen 7 4700U (Windows), x64, VisualStudio2022
| version | performance in MB/s |
|---|---|
| gray8_C | 284.7 |
| gray8_Neon | 2595.5 |
| gray16_C | 549.8 |
| gray16_Neon | 2745.7 |
| BGR24_C | 847.0 |
| BGR24_Neon | 2793.8 |
| BGR48_C | 1674.5 |
| BGR48_Neon | 2818.1 |
| float32_C | 1136.1 |
| float32_Neon | 2770.2 |
MediaTek Helio P60 (MT6771V/C), aarch64, gcc8.3
| version | performance in MB/s |
|---|---|
| gray8_C | 262.0 |
| gray8_Neon | 3198.5 |
| gray16_C | 522.0 |
| gray16_Neon | 3107.4 |
| BGR24_C | 743.0 |
| BGR24_Neon | 2671.3 |
| BGR48_C | 1485.6 |
| BGR48_Neon | 3088.7 |
| float32_C | 1044.1 |
| float32_Neon | 2892.6 |
i7-8700K (Windows), x64, VisualStudio2022
| version | performance in MB/s |
|---|---|
| gray8_C | 306.0 |
| gray8_Neon | 4850.9 |
| gray16_C | 625.4 |
| gray16_Neon | 5022.7 |
| BGR24_C | 968.2 |
| BGR24_Neon | 5006.9 |
| BGR48_C | 1816.9 |
| BGR48_Neon | 4877.4 |
| float32_C | 1232.1 |
| float32_Neon | 4944.1 |
The code is licensed under the unlicense license. Please see the file license.txt in this repository.