10:45 <daniels> linkmauve: G310 is also v7 but substantially lower performance than the G52

10:45 <daniels> *G31

10:45 <daniels> HdkR: hmm, we do have the perf debug for exactly this, which is also iirc wired through to the EGL debug interface so apps/trace/etc can also capture that

12:37 <urja> ... #¤$@%& libclc @£€£¤!...

12:56 <urja> "The CLC compiler requires LLVM, but LLVM is disabled." ... i remember someone saying libclc (& thus llvm) is only a build-time dep, not a runtime dep... how do i configure mesa like that?

13:01 <daniels> urja: you can do it as two-pass, where you do one build of only the tools with llvm enabled, then a second pass where you disable llvm and also tell it to use the 'system' compiler

13:18 <urja> okay, now finding the set of options to build mesa-clc and nothing but it ... (

14:22 <urja> install-mesa-clc does not imply mesa-clc=enabled (and if no drivers or anything else is enabled, the default of auto means disabled). logical :D

15:23 <robmur01> bbrezillon: do you still have your old IOMMU_DEV_PRIVATE branch around anywhere?

15:25 <robmur01> ISTR you prepped a v3 that never made it to the list

15:29 <bbrezillon> robmur01: uh, I can check, but I'm pretty sure I don't

15:31 <bbrezillon> robmur01: nope, I must have trashed a whole bunch of inactive kernel branches a while back

15:32 <bbrezillon> OTOH, the code was pretty trivial IIRC

15:33 <bbrezillon> robmur01: BTW, if you're considering adding new IOMMU flags, it'd be great to also support the stuff we need to share pages memory between the CPU and FW/MCU

15:35 <bbrezillon> https://elixir.bootlin.com/linux/v6.16-rc2/source/drivers/gpu/drm/panthor/panthor_fw.c#L593

15:37 <urja> motionmark on the veyron/ASUS C201 went from a score of 15 (@ 45 fps) to a score of 48 with that patch (mind you, i updated mesa from 24-something to 25.1.3 at the same time, but still)

15:39 <urja> I'm just amused that the RK3588 score was 217 _before_ the patch... so, firefox on the RK3288 now way faster, but still very slow :P

16:52 <robmur01> bbrezillon: I'll have to find a source for what exactly those CSF_FW flags are supposed to mean, but I'd be rather surprised if they need anything significantly different from regular GPU BOs

17:45 <bbrezillon> robmur01: let me dig out the kbase source I based that on

17:47 <bbrezillon> robmur01: https://github.com/rockchip-linux/kernel/blob/develop-6.1/drivers/gpu/arm/bifrost/csf/mali_kbase_csf_firmware.c#L479

21:45 <robmur01> yeesh, well that code is certainly clear as mud... good thing I have this other browser tab with the CSF specs open to look at tomorrow :D