This PR enabling lowering TE custom call kernels to command buffer through white list, this is a temporal support as 3rd library call should register command buffer support through FFI API, will remove this temporal support when TE has finished FFI API migration.
Copybara import of the project:
--
32c3d05c00d10698730e5adea8b06c2a02af39d9 by Shawn Wang shawnw@nvidia.com:
Add legayc custom call targets registry to command buffer
--
df2e5adcca71df12839284e9028087755785fee5 by Shawn Wang shawnw@nvidia.com:
fix unit test fail
--
88edf99cdc93be473686a63d51c32de47017cdc1 by Shawn Wang shawnw@nvidia.com:
PR #14266: [XLA:GPU] Add command buffer custom call targets recording for legacy custom call registry API
Imported from GitHub PR https://github.com/openxla/xla/pull/14266
This PR enabling lowering TE custom call kernels to command buffer through white list, this is a temporal support as 3rd library call should register command buffer support through FFI API, will remove this temporal support when TE has finished FFI API migration. Copybara import of the project:
-- 32c3d05c00d10698730e5adea8b06c2a02af39d9 by Shawn Wang shawnw@nvidia.com:
Add legayc custom call targets registry to command buffer
-- df2e5adcca71df12839284e9028087755785fee5 by Shawn Wang shawnw@nvidia.com:
fix unit test fail
-- 88edf99cdc93be473686a63d51c32de47017cdc1 by Shawn Wang shawnw@nvidia.com:
fix typos
Merging this change closes #14266