Skip to content

nvk,nak: Implement VK_KHR_shader_subgroup_rotate

Robin Kertels requested to merge K0bin/mesa:nvk-subgroup-rotate into main

Closes #10685 (closed)

Using shfle directly doesn't seem possible because according to the CUDA docs, it does not support wrap-around. So instead it is implemented using the lower-to-shuffle pass.

CTS results:

./deqp-vk -n dEQP-VK.subgroups.shuffle..subgrouprotate_

Test run totals: Passed: 528/1100 (48.0%) Failed: 0/1100 (0.0%) Not supported: 572/1100 (52.0%) Warnings: 0/1100 (0.0%) Waived: 0/1100 (0.0%)

./deqp-vk -n dEQP-VK.subgroups.shuffle..subgroupclusteredrotate_

Test run totals: Passed: 528/1100 (48.0%) Failed: 0/1100 (0.0%) Not supported: 572/1100 (52.0%) Warnings: 0/1100 (0.0%) Waived: 0/1100 (0.0%)

Edited by Robin Kertels

Merge request reports