00:53 cmarcelo: Leftmost: yeah, I think we are missing the case in the body handling function, and might be missing something else from there. (a) what GPU are you using? (b) do you have a test case I can look at?
00:54 cmarcelo: Leftmost: for dumping instructions, if this is intel driver, INTEL_DEBUG=shaders might do the dump you are looking for. For dump of NIR stuff, NIR_DEBUG=print. You also can use nir_print_shader() directly in the end of spirv_to_nir() function.
01:30 Leftmost: cmarcelo, Radeon RX 7800 GT using the radeonsi driver. Can check zink in a moment here. The test case I'm using is test_basic from the OpenCL conformance suite (https://github.com/KhronosGroup/OpenCL-CTS) with https://gitlab.freedesktop.org/leftmostcat/mesa/-/commit/c74f5d82eb1cd2989d40bd8aaf8ae515fe49269a applied.
01:30 Leftmost: Err, that patch applied to mesa.
01:31 cmarcelo: I'll see if I can find a vulkan test I can try here.
01:33 Leftmost: zink seems to also repro, though fortunately it doesn't kill my machine entirely.
01:33 cmarcelo: Leftmost: can you share the SPIR-V source that is used by the test?
01:34 cmarcelo: (i.e. whatever the translator is producing)
01:36 Leftmost: Yep, just a sec, rebuilding mesa.
01:48 cmarcelo: Leftmost: (I'll have to go soon, if you'd like, create a Mesa issue with any other details and tag me, I can get back to this on tuesday)
01:49 Leftmost: Great, sounds good. I'll file a bug and be around Tuesday; happy to do footwork on this.
01:50 Leftmost: The spvasm is pretty large, so I'll just make it an attachment to the issue.
01:51 cmarcelo: ok tks. if I got the footwork comment right, two suggestions: would be nice to have a vulkan vkrunner test (-: (it does allow using shaders with SPIR-V assembly -- so we just need a recent spirv-as to play). or tweak the VK-GL-CTS existing tests to include a case with UntypedPointerVariables that are private (if you're familiar with CTS, this might be easier, otherwise take vkrunner route)
01:53 Leftmost: I'll try the vkrunner test. And thanks for the suggestion. :)
13:51 lumag: robclark: I'm sorry for a double spam, could you please a-b the msm_gpuvm fix? dim requests an ack to apply that fix.
14:16 robclark: lumag: done
14:17 lumag: robclark: thanks!