00:04 gfxstrand[d]: emufan4568[d]: Yeah, there isn't currently a limit in Vulkan that separates the maximum size of a VkBuffer from the maximum size of a buffer binding.
00:04 gfxstrand[d]: But in theory, if you don't care about robustness, you could create a giant sparse buffer and just use BDA on it all the time.
00:17 mhenning[d]: gfxstrand[d]: Really? I thought that maxStorageBufferRange and maxBufferSize were separate but I admit I haven't studied the spec too closely
00:17 gfxstrand[d]: gfxstrand[d]: Okay, I take that back a bit. Yeah, the ranges are separate from the buffer size limit. We could probably increase the later a lot.
00:18 gfxstrand[d]: It just took me a while to dig though spec from my phone over 4G. 🤳🏻
04:25 realopalsdeal: what intel and amd do is all correct, albeit with high entropy, so the military and secure systems would work different as a scale below of such lossy environments however in a way similar to the current, but to HdkR I still say you do not count the bits at all which is your dilemma, I show you how there is no need for this. intel instruction set is very rich but that’s not for
04:25 realopalsdeal: programming experts it’s for rookies. marketing tricks and real science is different. they target equipment also for incapable people. similarly real sports is different than shown in tv. cause all real athletes are in McDonalds beef and in mental institution harvested for male hormones so that illborn people can also feel worthy and capable. but unfortunately intel is more correct in its
04:25 realopalsdeal: commodity systems than I am. cause my code is too capable and dangerous to the whole world. but now quantum computers if they exist or not. well I tend to think yes, antimatter rocket yes, moon exploration yes etc. but I do not know for sure. there were news for non ambient condition and vice versa quantum computers too, but too wide era of discussion again.
04:26 HdkR: 🎉
07:41 emufan4568[d]: gfxstrand[d]: Would like to use descriptors for that yes and atomics
08:22 gfxstrand[d]: As long as the descriptors don't cover more than a 4GiB range, we can support that.
08:22 gfxstrand[d]: emufan4568[d]: As long as the descriptors don't cover more than a 4GiB range, we can support that.
08:46 emufan4568[d]: gfxstrand[d]: So that size check can be increased then, assuming descriptor ranges are already checked to fit inside 32bits
08:49 marysaka[d]: gfxstrand[d]: don't we need to also extends the address space to do so atm? :aki_thonk:
08:50 marysaka[d]: or wait no we are at 38-bit atm so should be fine
08:50 marysaka[d]: (was thinking of the whole DMA methods needing an update on Pascal+ to support 40-bit AS)
09:37 gfxstrand[d]: emufan4568[d]: Yeah, I think so. We should probably make sure we have descriptor range asserts various places but we should be able to increase the limit. We just have to make sure that the check and the physical device limit match.
22:58 emufan4568[d]: gfxstrand[d]: Awesome thanks