AFAIK it’s the copy cost for the memory. GPU makes sense only when the hardware allows this copy to go away. Generally, desktop PCs don’t have such specialized hardware.
I don’t see why you’d have to copy all that much. Depending on the rendering architecture, once all the glyphs are there you’d only need to send the relevant text data to be rendered. I don’t see that being much of a problem even when using SDFs.
It’s an extremely small amount of data by today’s standards and it can be updated on demand, but even if it couldn’t it would still be extremely fast to send over every frame.
If games do it, so can text editors. Real time text rendering on the GPU is a fairly common practice nowadays, unfortunately not in most GUI applications…
At this point I’m not expert enough to explain more details. You can check font renderers.
Below is what’s in my mind but it’s just a guess.
In typical PC architectures you have IO between the storage and the RAM, and then there’s the copying from the RAM to the VRAM, and editors maybe also want copying from the VRAM to RAM for decoration purposes etc.
I am familiar with the current PC and GPU architectures.
IO is a non issue. Even a massive file can be trivially memory mapped and parsed without much hassle, and in the case of a text editor you’d have to deal with IO only when opening or saving said file, not during rendering.
As for the rendering side, again, the amount of memory you’d have to transfer between RAM and VRAM would be minimal. The issue is latency, not speed, but that can be mitigated though asynchonous transfer operations, so if done properly stutters are unlikely.
Rendering monospaced fonts (with decorators and control characters) at thousands of frames a second nowadays is computationally trivial, take a look at refterm for an example. I suspect non-monospaced fonts would require more effort, but it’s doable.
As I said at the beginning, it’s not impossible, just a pain. But so is font rendering in general honestly :/
AFAIK it’s the copy cost for the memory. GPU makes sense only when the hardware allows this copy to go away. Generally, desktop PCs don’t have such specialized hardware.
I don’t see why you’d have to copy all that much. Depending on the rendering architecture, once all the glyphs are there you’d only need to send the relevant text data to be rendered. I don’t see that being much of a problem even when using SDFs. It’s an extremely small amount of data by today’s standards and it can be updated on demand, but even if it couldn’t it would still be extremely fast to send over every frame. If games do it, so can text editors. Real time text rendering on the GPU is a fairly common practice nowadays, unfortunately not in most GUI applications…
At this point I’m not expert enough to explain more details. You can check font renderers.
Below is what’s in my mind but it’s just a guess.
In typical PC architectures you have IO between the storage and the RAM, and then there’s the copying from the RAM to the VRAM, and editors maybe also want copying from the VRAM to RAM for decoration purposes etc.
I am familiar with the current PC and GPU architectures.
IO is a non issue. Even a massive file can be trivially memory mapped and parsed without much hassle, and in the case of a text editor you’d have to deal with IO only when opening or saving said file, not during rendering.
As for the rendering side, again, the amount of memory you’d have to transfer between RAM and VRAM would be minimal. The issue is latency, not speed, but that can be mitigated though asynchonous transfer operations, so if done properly stutters are unlikely.
Rendering monospaced fonts (with decorators and control characters) at thousands of frames a second nowadays is computationally trivial, take a look at refterm for an example. I suspect non-monospaced fonts would require more effort, but it’s doable.
As I said at the beginning, it’s not impossible, just a pain. But so is font rendering in general honestly :/
As I indicated, please check (articles and the documentations of) font renderers at this point.