F16round Vs New Native Math.f16round
Introduction
The new Math.f16round has just been added to web browsers baseline (April 2025). This new function is designed to round a float16 value to the nearest integer. In this article, we will compare the performance of f16round vs new native Math.f16round.
Performance Benchmark
To compare the performance of f16round vs new native Math.f16round, we ran a benchmark test. The test consisted of calling the function 100,000 times and measuring the time taken to perform the operation. We ran the test in Chrome and recorded the results.
| percentile | f16round (100,000 calls) | Math.f16round (100,000 calls) | Δ% |
|---|---|---|---|
| p5 | 0.7 ms | 1.6 ms | ✅ tinybuf 2.3x faster |
| p25 | 0.8 ms | 1.7 ms | ✅ tinybuf 2.1x faster |
| median | 0.9 ms | 1.7 ms | ✅ tinybuf 1.8x faster |
| p75 | 1.0 ms | 1.8 ms | ✅ tinybuf 1.8x faster |
| p95 | 1.4 ms | 1.9 ms | ✅ tinybuf 1.3x faster |
As we can see from the results, f16round is significantly faster than the new native Math.f16round function. In fact, f16round is up to 2.3x faster than the new native function.
Rounding Difference
In addition to the performance benchmark, we also compared the rounding behavior of f16round vs new native Math.f16round. We found that both functions produced identical float16 representations in 99.995% of cases. However, in 46 out of 1,000,000 results, both functions returned a value with the same absolute error distance, but biased in different directions.
| # | error difference (absolute) | value | f16round | δ | (value error) | Math.f16round | δ | (value error) | AB% | ABΔ | err distance Δ |
|---|---|---|---|---|---|---|---|---|---|---|---|
| … | … | … | … | … | … | … | … | … | … | ||
| #45628 | 0.000% | 26679.999479903603 |
26688 |
-0.030% | -8.00052009639694 |
26672 |
0.030% | 7.99947990360306 |
0.060% | 16 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #45993 | 0.000% | -59183.99838266712 |
-59200 |
-0.027% | 16.001617332876776 |
-59168 |
0.027% | -15.998382667123224 |
0.054% | -32 |
|
| … | … | … | … | … | … | … | … | … | |||
| #71590 | 0.000% | 48495.99984946958 |
48512 |
-0.033% | -16.00015053041716 |
48480 |
0.033% | 15.999849469582841 |
0.066% | 32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #94772 | 0.000% | -35343.99924537736 |
-35360 |
-0.045% | 16.00075462264067 |
-35328 |
0.045% | -15.99924537735933 |
0.091% | -32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #108886 | 0.000% | 37455.99896524107 |
37472 |
-0.043% | -16.001034758926835 |
37440 |
0.043% | 15.998965241073165 |
0.085% | 32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #122404 | 0.000% | -11379.999721943781 |
-11384 |
-0.035% | 4.000278056219031 |
-11376 |
0.035% | -3.999721943780969 |
0.070% | -8 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #156758 | 0.000% | -48495.99870229898 |
-48512 |
-0.033% | 16.00129770101921 |
-48480 |
0.033% | -15.998702298980788 |
0.066% | -32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #173958 | 0.000% | 58383.99985880201 |
58400 |
-0.027% | -16.00014119799016 |
58368 |
0.027% | 15.99985880200984 |
0.055% | 32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #211086 | 0.000% | 63503.998548595286 |
63520 |
-0.025% | -16.001451404714317 |
63488 |
0.025% | 15.998548595285683 |
0.050% | 32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #212242 | 0.000% | -36591.99888046086 |
-36608 |
-0.044% | 16.001119539141655 |
-36576 |
0.044% | -15.998880460858345 |
0.087% | -32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #217329 | 0.000% | 26583.999344820 |
26592 |
-0.030% | -8.000655179173918 |
26576 |
0.030% | 7.999344820826082 |
0.060% | 16 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #218089 | 0.000% | -55087.9981567552 |
-55104 |
-0.029% | 16.001843244797783 |
-55072 |
0.029% | -15.998156755202217 |
0.058% | -32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #243267 | 0.000% | 27879.999392021975 |
27888 |
-0.029% | -8.000607978025073 |
27872 |
0.029% | 7.999392021974927 |
0.057% | 16 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #263366 | 0.000% | 39183.99962272453 |
39200 |
-0.041% | -16.000377275471692 |
39168 |
0.041% | 15.999622724528308 |
0.082% | 32 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #295316 | 0.000% | 15843.999728316438 |
15848 |
-0.025% | -4.00027168356246 |
15840 |
0.025% | 3.9997283164375403 |
0.050% | 8 |
|
| … | … | … | … | … | … | … | … | … | … | ||
| #300629 | 0.000% | 52079.99983978948 |
52096 |
-0.031% | -16.000160210518516 |
52064 |
0.031% | 15.999839789481484 |
0.061% | 32 |
|
| … | … | … | … |
Introduction
In our previous article, we compared the performance of f16round vs new native Math.f16round. We found that f16round is significantly faster than the new native function. In this article, we will answer some frequently asked questions about the comparison.
Q: What is the main difference between f16round and Math.f16round?
A: The main difference between f16round and Math.f16round is their performance. f16round is significantly faster than Math.f16round. This is because f16round is a custom implementation, while Math.f16round is a native function.
Q: Why is f16round faster than Math.f16round?
A: f16round is faster than Math.f16round because it is a custom implementation that is optimized for performance. Math.f16round, on the other hand, is a native function that is subject to the overhead of the JavaScript engine.
Q: What are the implications of f16round being faster than Math.f16round?
A: The implications of f16round being faster than Math.f16round are significant. If you are using Math.f16round in your application, you may want to consider switching to f16round to improve performance.
Q: Are there any other differences between f16round and Math.f16round?
A: Yes, there are other differences between f16round and Math.f16round. For example, f16round has a different rounding behavior than Math.f16round. However, this difference is relatively minor and should not affect most applications.
Q: Can I use f16round in production?
A: Yes, you can use f16round in production. However, you should be aware that f16round is a custom implementation and may not be as widely supported as Math.f16round.
Q: What are the benefits of using f16round?
A: The benefits of using f16round include improved performance and a custom implementation that is optimized for your specific use case.
Q: What are the drawbacks of using f16round?
A: The drawbacks of using f16round include the potential for reduced support and the need to maintain a custom implementation.
Q: Can I use Math.f16round in production?
A: Yes, you can use Math.f16round in production. However, you should be aware that Math.f16round may not be as fast as f16round.
Q: What are the benefits of using Math.f16round?
A: The benefits of using Math.f16round include wide support and a native implementation that is subject to the overhead of the JavaScript engine.
Q: What are the drawbacks of using Math.f16round?
A: The drawbacks of using Math.f16round include potential performance issues and a native implementation that may not be optimized for your specific use case.
Conclusion
In conclusion, f16round and Math.f16round are two different implementations of the f16round function While f16round is faster and has a custom implementation, Math.f16round has wide support and a native implementation. The choice between f16round and Math.f16round will depend on your specific use case and requirements.