https://github.com/junka/vkcompbench
benchmark for gpu compute capability using vulkan
https://github.com/junka/vkcompbench
Last synced: 9 months ago
JSON representation
benchmark for gpu compute capability using vulkan
- Host: GitHub
- URL: https://github.com/junka/vkcompbench
- Owner: junka
- Created: 2025-03-12T04:55:25.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-03-31T07:00:45.000Z (over 1 year ago)
- Last Synced: 2025-03-31T07:24:33.975Z (over 1 year ago)
- Language: C++
- Homepage:
- Size: 138 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# vkcompbench
Do benchmark for vulkan compute pipeline
We will need ```glslc``` from shaderc projec.
- For Macos, It can be installed by: ```brew install shaderc```
- For Linux, It can be installed by: ```sudo apt install glslc```
Then compile and run:
```
cmake -S . -B build
cmake --build build
./build/vkcompbench
```
Note for ```fp64``` benchmark:
- _Float64 did not get supported in clang, use double instead
Note for ```fp16``` benchmark:
- _Float16 need gcc-12 at least
C23 features will support ```std::float16_t``` and ```std::float64_t```, but not all current active compilers will work with new features.
## testcases cover
- fp64: float point 64
- int64: integer 64
- fp32: float point 32
- int32: integer 32
- fp16: float point 16
- int16: integer 16
- int8: integer 8
- int8dot: integer 8 dot product
- int8dotaccsat: integer 8 dot product with accumulate and saturation
- int8dot4x8packed: integer 8 dot product with 4x8 packed