Post

Replies

Boosts

Views

Activity

Reply to Metal performance compared to OpenCL
device const sphericalparams * params [[buffer(2)]] --- Metal ... xyout[index].x = params->x + params->scale * x; xy_out[index].y = params->y + params->scale * y; vs float scale, float x0, float y0 --- OpenCL xyout[i].even = x0 + scale * x; xyout[i].odd = y0 + scale * y; The compiler code optimizer most likely likes the OpenCL approach. The params -> indirection may be costly.
Topic: Graphics & Games SubTopic: General Tags:
Nov ’20
Reply to Big Sur OpenCL changes on AMD GPUs
Based on the info you have posted, this is a LuxRays bug. The referenced file is one of their own internal source files. I am the author of Fractal Architect for Mac and iOS. I am not having any issues with my own app. In fact, it looks like Apple has fixed issues I reported back in Catalina. I have had an OpenCL app on MacOS since 2009. And yes OpenCL has been broken 5 years ago when El Capitan was released, but since then it has been robust for single GPU renders. OpenCL has been problematic for multiple GPU rendering, whereas Metal works properly. I have not yet retested the multiple GPU kernel scheduling bug in Big Sur.
Topic: App & System Services SubTopic: Core OS Tags:
Nov ’20