18:05fdobridge: <marysaka> On Maxwell Gen 2/Pascal GPUs what kind of falcon firmware are loaded? Is there one present handling GPFIFO?
18:19fdobridge: <karolherbst🐧🦀> yeah, you can find those in linux-firmware
18:51fdobridge: <gouz> Hello! do we have an mme alu operation for division?
18:52fdobridge: <karolherbst🐧🦀> I'm sure we don't
18:52fdobridge: <karolherbst🐧🦀> too expensive
18:53fdobridge: <karolherbst🐧🦀> in most cases you don't need a div anyway
18:53fdobridge: <gouz> i am trying to support vkCmdDrawIndirectByteCountEXT
18:53fdobridge: <gouz> and for the vertex count i need to divide by the stride
18:53fdobridge: <karolherbst🐧🦀> is the divisor a pot?
18:53fdobridge: <karolherbst🐧🦀> ehhh
18:53fdobridge: <gouz> 😄
18:54fdobridge: <karolherbst🐧🦀> convert it to some weird shift 😛
18:54fdobridge: <gouz> thanks @karolherbst🐧 i will try to figure something out!
18:55fdobridge: <karolherbst🐧🦀> check godbolt or the internet on how to lower shift by known constants
18:55fdobridge: <karolherbst🐧🦀> check godbolt or the internet on how to lower divisions by known constants (edited)
18:55fdobridge: <karolherbst🐧🦀> still messed up tho
18:56fdobridge: <karolherbst🐧🦀> but...
18:56fdobridge: <karolherbst🐧🦀> do you have to deal with multiple strides or just one?
18:56fdobridge: <karolherbst🐧🦀> could always do some calcs on thoe CPU side
18:57fdobridge: <karolherbst🐧🦀> could always do some calcs on the CPU side (edited)
19:01fdobridge: <gouz> its a user provided stride, yea i agree i probably need to do some pre-processing on the CPU side
19:01fdobridge: <gouz> its a user provided stride, yeah i agree i probably need to do some pre-processing on the CPU side (edited)
19:05fdobridge: <🌺 ¿butterflies? 🌸> Preferably avoid going the CPU processing route
19:08fdobridge: <gouz> https://rubenvannieuwpoort.nl/posts/division-by-constant-unsigned-integers
19:09fdobridge: <gouz> this seems quite helpful
19:18fdobridge: <karolherbst🐧🦀> yeah.. but even that is painful if you can simply do it on the host
19:28fdobridge: <phomes> @gouz cool 🙂 since you are in that area can I ask you to look at the code in !153? It is causing a hang that I cannot reproduce
19:30fdobridge: <gouz> @phomes i will try it too to check if i get a hang
20:38fdobridge: <gouz> @phomes hmm yes
20:39fdobridge: <gouz> i get a freeze in dEQP-VK.draw.renderpass.indirect_draw.sequential.indirect_draw_param_count.triangle_list
20:39fdobridge: <gouz> and then all the other tests in draw.txt are reporting device loss
20:44fdobridge: <gouz> ahh
20:45fdobridge: <gouz> The test that fails says in its description: "Draws geometry with VK_KHR_draw_indirect_count extension and limit draws count with call parameter"
20:46fdobridge: <gouz> I do not want to spoil the solution 😄
20:47fdobridge: <gouz> https://registry.khronos.org/vulkan/specs/1.2-extensions/html/chap21.html#vkCmdDrawIndirectCountKHR