Aras Pranckevicius aras_p
10 Followers · 0 Following
Repositories
2
 Projects Public Activity Starred Repositories
Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-17 15:16:52 +02:00
0f534a16db Use blender comment style
Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-17 15:15:00 +02:00
08c1b230e3 Use vget_lane instead of indexing for wider compiler support
Aras Pranckevicius commented on pull request blender/blender#127544 2024-09-17 14:56:26 +02:00
WM: improve thumbnail scaling performance (debug builds)

I suspect the difference in the Debug build performance is less of an "this is a faster algorithm", and more of a "your code uses raw C math instead of C++ math library".

In my tests…

Aras Pranckevicius commented on pull request blender/blender#127708 2024-09-17 13:17:45 +02:00
BLI: add float<->half conversion functions with correct math, use in Vulkan

The thing which is actually worrysome is the performance. Being correct in corner cases is fine, but not if it has a significant impact on the performance of typical cases. It is not very clear…

Aras Pranckevicius commented on pull request blender/blender#127708 2024-09-17 13:15:40 +02:00
BLI: add float<->half conversion functions with correct math, use in Vulkan

Indexing with a constant syntax is supported by both Clang and GCC (but not by MSVC), results in same assembly as using vget_lane. I can change it to vget_lane indeed (though our windows arm64…

Aras Pranckevicius commented on pull request blender/blender#127708 2024-09-17 11:01:12 +02:00
BLI: add float<->half conversion functions with correct math, use in Vulkan

@blender-bot build

Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-17 10:57:34 +02:00
48d2daae31 Merge branch 'main' into fp16_conv
b43ad2987f Remove half_to_float from BLI_math_bits.h, switch space_view3d to the new one
a2e543590e Use BLI_math_half.hh functions in vk_data_conversion.hh
d918c22fcc Revert "CMake: suppress warnings building with mold on Linux"
65d71c2672 Cleanup: spelling in comments
Compare 37 commits »
Aras Pranckevicius pushed to main at blender/blender 2024-09-17 06:06:18 +02:00
bab4f7a0cd Fix #127654: Video Deinterlace option does not work in some cases
Aras Pranckevicius deleted branch video_deinterlace_fix from aras_p/blender 2024-09-17 06:06:16 +02:00
Aras Pranckevicius closed issue blender/blender#127654 2024-09-17 06:06:16 +02:00
Regression: Deinterlace checkbox makes video invisible
Aras Pranckevicius merged pull request blender/blender#127689 2024-09-17 06:06:13 +02:00
Fix #127654: Video Deinterlace option does not work in some cases
Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-16 20:42:05 +02:00
a6fc624efb Fix Linux build
Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-16 20:35:16 +02:00
15f6f33dff Use hardware half<->float conversion on ARM when available (e.g. Mac)
Aras Pranckevicius created pull request blender/blender#127708 2024-09-16 17:13:01 +02:00
WIP: BLI: add float<->half conversion functions with actually correct math
Aras Pranckevicius pushed to fp16_conv at aras_p/blender 2024-09-16 17:12:29 +02:00
9c723fd6c4 BLI: add float<->half conversion functions with actually correct math
7180df49dd Fix: Incorrect tool brush check for grease pencil vertex paint mode
806b0e8379 BLI: improve 2/3/4d vector codegen for debug or asserts-enabled builds
65fdb3fa30 Grease Pencil: Properly support brushes with draw mode primitive tools
d52d70324d Fix (chat-reported) Py-defined operators id being added to translation files.
Compare 10 commits »
Aras Pranckevicius created branch fp16_conv in aras_p/blender 2024-09-16 17:12:29 +02:00
Aras Pranckevicius deleted branch math_vector_debug_perf from aras_p/blender 2024-09-16 13:06:37 +02:00
Aras Pranckevicius merged pull request blender/blender#127577 2024-09-16 13:06:35 +02:00
BLI: improve 2/3/4d vector codegen for debug or asserts-enabled builds
Aras Pranckevicius pushed to main at blender/blender 2024-09-16 13:06:34 +02:00
806b0e8379 BLI: improve 2/3/4d vector codegen for debug or asserts-enabled builds
Aras Pranckevicius commented on pull request blender/blender#127544 2024-09-16 12:44:19 +02:00
WM: improve thumbnail scaling performance (debug builds)

I suspect the difference in the Debug build performance is less of an "this is a faster algorithm", and more of a "your code uses raw C math instead of C++ math library". I guess if you tried code…