xuweijia-buaa / cpufp

A CPU tool for benchmarking the peak of floating points

cpufp

This is a cpu tool for benchmarking the floating-points peak performance. Now it supports linux and x86-64 platform. It can automatically sense the local SIMD ISAs while compiling.

How to use

build:

./build.sh

benchmark:

./cpufp --thread_pool=[xxx]

clean:

./clean.sh

xxx indicates that all the cores defined by xxx will be used for benchmarking(by affinity setting). For example, [0,3,5-8,13-15].

Support x86-64 SIMD ISA

ISA	Data Type	Description
SSE	fp32/fp64	Before Sandy Bridge
AVX	fp32/fp64	From Sandy Bridge
FMA	fp32/fp64	From Haswell/Zen
AVX512f	fp32/fp64	From Skylake X/Zen4
AVX512_VNNI	int8/int16	From IceLake
AVX_VNNI	int8/int16	From Alder Lake

Some benchmark results

Intel Alder Lake i7-1280p

For single Golden Cove(big) core:

$ ./cpufp --thread_pool=[0]
Number Threads: 1
Thread Pool Binding: 0
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| AVX_VNNI        | INT8      | 590.31 GOPS      |
| AVX_VNNI        | INT16     | 295.06 GOPS      |
| FMA             | FP32      | 149.87 GFLOPS    |
| FMA             | FP64      | 74.931 GFLOPS    |
| AVX             | FP32      | 112.39 GFLOPS    |
| AVX             | FP64      | 56.203 GFLOPS    |
| SSE             | FP32      | 56.054 GFLOPS    |
| SSE             | FP64      | 28.001 GFLOPS    |
--------------------------------------------------

For multiple big cores:

$ ./cpufp --thread_pool=[0,2,4,6,8,10]
Number Threads: 6
Thread Pool Binding: 0 2 4 6 8 10
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| AVX_VNNI        | INT8      | 2636.8 GOPS      |
| AVX_VNNI        | INT16     | 1319.1 GOPS      |
| FMA             | FP32      | 670.05 GFLOPS    |
| FMA             | FP64      | 335 GFLOPS       |
| AVX             | FP32      | 502.4 GFLOPS     |
| AVX             | FP64      | 251.2 GFLOPS     |
| SSE             | FP32      | 250.42 GFLOPS    |
| SSE             | FP64      | 125.16 GFLOPS    |
--------------------------------------------------

For single Gracemont(little) core:

$ ./cpufp --thread_pool=[12]
Number Threads: 1
Thread Pool Binding: 12
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| AVX_VNNI        | INT8      | 114.89 GOPS      |
| AVX_VNNI        | INT16     | 57.445 GOPS      |
| FMA             | FP32      | 57.444 GFLOPS    |
| FMA             | FP64      | 28.723 GFLOPS    |
| AVX             | FP32      | 28.723 GFLOPS    |
| AVX             | FP64      | 14.362 GFLOPS    |
| SSE             | FP32      | 28.312 GFLOPS    |
| SSE             | FP64      | 14.361 GFLOPS    |
--------------------------------------------------

For multiple little cores:

$ ./cpufp --thread_pool=[12-19]
Number Threads: 8
Thread Pool Binding: 12 13 14 15 16 17 18 19
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| AVX_VNNI        | INT8      | 867.99 GOPS      |
| AVX_VNNI        | INT16     | 434 GOPS         |
| FMA             | FP32      | 434 GFLOPS       |
| FMA             | FP64      | 217 GFLOPS       |
| AVX             | FP32      | 217.01 GFLOPS    |
| AVX             | FP64      | 108.5 GFLOPS     |
| SSE             | FP32      | 216.39 GFLOPS    |
| SSE             | FP64      | 108.5 GFLOPS     |
--------------------------------------------------

AMD Ryzen9 6900HX(Zen3+)

For single core:

$ ./cpufp --thread_pool=[0]
Number Threads: 1
Thread Pool Binding: 0
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| FMA             | FP32      | 156.18 GFLOPS    |
| FMA             | FP64      | 78.371 GFLOPS    |
| AVX             | FP32      | 156.55 GFLOPS    |
| AVX             | FP64      | 78.256 GFLOPS    |
| SSE             | FP32      | 78.219 GFLOPS    |
| SSE             | FP64      | 38.99 GFLOPS     |
--------------------------------------------------

For multi-cores:

$ ./cpufp --thread_pool=[0,2,4,6,8,10,12,14]
Number Threads: 8
Thread Pool Binding: 0 2 4 6 8 10 12 14
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| FMA             | FP32      | 1151.2 GFLOPS    |
| FMA             | FP64      | 569.89 GFLOPS    |
| AVX             | FP32      | 1088.7 GFLOPS    |
| AVX             | FP64      | 536.37 GFLOPS    |
| SSE             | FP32      | 541.35 GFLOPS    |
| SSE             | FP64      | 269.56 GFLOPS    |
--------------------------------------------------

Intel Celeron N5105(Jasper Lake)

For single core:

$ ./cpufp --thread_pool=[0]
Number Threads: 1
Thread Pool Binding: 0
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| SSE             | FP32      | 23.102 GFLOPS    |
| SSE             | FP64      | 11.564 GFLOPS    |
--------------------------------------------------

For multi_cores:

$ ./cpufp --thread_pool=[0-3]
Number Threads: 4
Thread Pool Binding: 0 1 2 3
--------------------------------------------------
| Instruction Set | Data Type | Peak Performance |
| SSE             | FP32      | 89.327 GFLOPS    |
| SSE             | FP64      | 44.664 GFLOPS    |
--------------------------------------------------

TODO

The next version may support ARMv7 and ARMv8 architectures.

Welcome for bug reporting.