#
✕ nick:i509VCB
#asahi-gpu · 2024-01-04 21:20
i509vcbhmm no swizzle needed
#asahi-gpu · 2024-01-04 20:37
i509vcbI'll need to check (won't I need a swizzle for that), but I guess for that case it would make sense to have getting the copy format for snorm and sfloat formats be driver specific
#asahi-gpu · 2024-01-04 20:06
i509vcbAh right compressed formats, let me see how many failures I get again if I turn those back on
#asahi-gpu · 2024-01-04 20:00
i509vcbM2 Air with 16GB of ram
#asahi-gpu · 2024-01-04 19:57
i509vcbIt will probably take several hours to run when agxv is done enough to try to do a full cts run
#asahi-gpu · 2024-01-04 19:55
i509vcbgrepped out image_to_image in the api subgroup
#asahi-gpu · 2024-01-04 19:53
i509vcb(3d and stencil/depth are the largest groups I need to get fixed still)
#asahi-gpu · 2024-01-04 19:53
i509vcbFor dEQP-VK.api.copy_and_blit.core.image_to_image.*
#asahi-gpu · 2024-01-04 19:53
i509vcbPass: 8754, Fail: 372, Skip: 41386, Flake: 2, Duration: 3:10, Remaining: 0
#asahi-gpu · 2024-01-04 19:39
i509vcbstupid typo for r32g32b32a32
#asahi-gpu · 2024-01-04 18:47
i509vcbhttps://pixfmtdb.emersion.fr/VK_FORMAT_B8G8R8A8_SNORM seems to show it's not much and Zink docs doesn't seem to that format is needed
#asahi-gpu · 2024-01-04 18:44
i509vcbAlthough r32g32b32a32_sfloat has a hardware equvilant
#asahi-gpu · 2024-01-04 18:44
i509vcbr32g32b32a32_sfloat still fails for some weird reason
#asahi-gpu · 2024-01-04 18:43
i509vcbb8g8r8a8_snorm can't be copied via meta-copy
#asahi-gpu · 2024-01-04 18:42
i509vcbSadly a few formats always fail, I think the hardware has no SINT/UINT equivalent for these
#asahi-gpu · 2024-01-04 18:42
i509vcbalyssa: yep I forgot that and that fixed nearly every failure
#asahi-gpu · 2024-01-04 03:26
i509vcbalyssa: ^
#asahi-gpu · 2024-01-04 03:24
i509vcbhttps://usercontent.irccloud-cdn.com/file/oktOpvbI/agxdecode-dEQP-VK.api.copy_and_blit.core.image_to_image.all_formats.color.2d_to_2d.b8g8r8a8_snorm.r16g16_sfloat.general_genera.zip
#asahi-gpu · 2024-01-04 01:41
i509vcbI'll grab that in a bit, gotta do something
#asahi-gpu · 2024-01-04 01:40
i509vcbAll four combinations of casts seem to do this
#asahi-gpu · 2024-01-04 01:40
i509vcbWhat's odd here is I am casting both sides of the copy
#asahi-gpu · 2024-01-03 20:19
i509vcbIs the hardware producing different NaN values from what Vulkan expects?
#asahi-gpu · 2024-01-03 20:19
i509vcbWhere NAN is the GCC definition of NAN
#asahi-gpu · 2024-01-03 20:19
i509vcbassert(deFloat32To16(NAN) == 0x7e00); is true
#asahi-gpu · 2024-01-03 07:55
i509vcbdifferent NaN values... https://usercontent.irccloud-cdn.com/file/BLImn2ED/image.png
#asahi-gpu · 2024-01-03 06:39
i509vcb(The result is R16G16)
#asahi-gpu · 2024-01-03 06:39
i509vcbI guess what I suspected is what's happened: https://usercontent.irccloud-cdn.com/file/L7hyff5s/image.png
#asahi-gpu · 2024-01-03 02:04
i509vcbFrom what I can see in the tests, the tests are pure bitwise comparisons
#asahi-gpu · 2024-01-03 02:00
i509vcbIs there some way to ask the vulkan dEQP to dump the raw bytes of the result and reference?
#asahi-gpu · 2024-01-02 23:43
i509vcbAnd uint attempts did the same
#asahi-gpu · 2024-01-02 23:43
i509vcbI did try casting both ends to sint and nothing changed
#asahi-gpu · 2024-01-02 23:43
i509vcb /s/norm/snorm
#asahi-gpu · 2024-01-02 23:42
i509vcbI think this is the a8b8g8r8_norm to r16g16_sfloat copy test
#asahi-gpu · 2024-01-02 23:42
i509vcb(checking with a color picker, these are the same)
#asahi-gpu · 2024-01-02 23:31
i509vcbfun https://usercontent.irccloud-cdn.com/file/8wD2AH8q/image.png
#asahi-gpu · 2024-01-02 22:32
i509vcbAny ideas why R16 doesn't work with the casting but R32 does? (Yes the code is horrible) https://gitlab.freedesktop.org/i509VCB/mesa/-/blob/i5-vkCmdCopy/src/vulkan/runtime/vk_meta_copy.c#L510
#asahi-gpu · 2024-01-02 06:33
i509vcbsrc_format == VK_FORMAT_B8G8R8A8_UNORM && dst_format == VK_FORMAT_R32_SFLOAT works though with the hackery (with UINT instead in the dst change)
#asahi-gpu · 2024-01-02 06:30
i509vcbentirely different formats I've yet to figure out a pattern
#asahi-gpu · 2024-01-02 06:30
i509vcbHuh the cast works with R32 but not R16?: https://gist.github.com/i509VCB/219ecfa91e39fb64db47d970b36caae6
#asahi-gpu · 2024-01-02 06:23
i509vcbAlthough I might not have it figured out yet
#asahi-gpu · 2024-01-02 06:20
i509vcbI guess I'll be writing new helpers so that I have a way to get a format with a different numeric type
#asahi-gpu · 2024-01-02 06:19
i509vcbalyssa: I think the sfloat issue is what you mentioned, now to generalize a solution for it...
#asahi-gpu · 2024-01-02 04:58
i509vcbI'm sure I've written far worse code when I was 12
#asahi-gpu · 2024-01-02 04:48
i509vcbYeah oof sorry my temp disable of compressed formats is not good
#asahi-gpu · 2024-01-01 21:40
i509vcbI've tried not to make a mess with what I was given: https://gitlab.freedesktop.org/i509VCB/mesa/-/blob/i5-vkCmdCopy/src/vulkan/runtime/vk_meta_copy.c?ref_type=heads#L104
#asahi-gpu · 2024-01-01 21:40
i509vcbI'm not 100% on how I'd do that casting. nir_bitcast_vector seems wrong since it wants the bitsize (probably for f32 to f64 casts or something like that) rather than a transmute of sorts
#asahi-gpu · 2024-01-01 19:56
i509vcbI did kind of vanish for a month lol
#asahi-gpu · 2024-01-01 19:56
i509vcbyeah I guess I'll try that first
#asahi-gpu · 2024-01-01 19:32
i509vcb(I know there is stuff you pointed me to in the past, just need to try a few things)
#asahi-gpu · 2024-01-01 19:31
i509vcbI guess I am forced to figure out layered rendering for 3d copies

plus ›