device: iphone 11 os: ios 15.6
I have a metal applicaton on IOS where a series of computer shaders are encoded, then disptached and comiited together at last. When I capture a GPU trace of my application, however I noticed there are these gaps between each computer shader invocation. And these gaps seem to take up a big part of the GPU time.
I'm wondering what are these gaps and what are causing them. Since all compute dispatch commands are commiited toghether at once, these gaps shouldn't be synchronizations between cpu and GPU
PS: In my application, later compute commands mostly depend on former ones and would use the result buffer from former invocations. But as shown in the picture, bandwith and read/write buffer limiter are not high as far as I'm concerned.
