[SYCL][CUDA] Correctly free managed memory #4181
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Depending on the host system's CUDA implementation, CU_POINTER_ATTRIBUTE_MEMORY_TYPE queries may report different values for managed pointers. Managed memory must however always be freed using cuMemFree. This patch adds an additional case for the cuMemFree path in cuda_piextUSMFree. Signed-off-by: Steffen Larsen <[email protected]>
romanovvlad
approved these changes
Jul 26, 2021
alexbatashev
added a commit
to alexbatashev/llvm
that referenced
this pull request
Jul 27, 2021
…ackend_plugin * upstream/sycl: (755 commits) [SYCL] Add operator= to atomic_ref specializations (intel#4183) [SYCL] Make spelling of Debug value for CMAKE_BUILD_TYPE variable case insensitive (intel#4069) [SYCL][LIBCLC] Add atan and cbrt for amdgcn-amdhsa (intel#4180) [SYCL][CUDA] Correctly free managed memory (intel#4181) [SYCL] Revert barrier deprecation note (intel#4162) [SYCL][FPGA] Refactor of statement attributes (intel#4136) [Driver][SYCL] Enable way to emit int-footer source to a specific dir (intel#4167) [Driver] Fix default MSVC version setting for -fms-compatibilty-version (intel#4165) [BuildBot] Add llvm-enable-projects flag to configure.py (intel#4169) [Driver][SYCL][FPGA] Improve aocx archive processing for FPGA (intel#4160) [SYCL] Correct int-header emission with type aliases [SYCL] Fix name collisions in SYCL enums (intel#4154) [SYCL] Return the correct status info for host_task event (intel#4161) [ESIMD][NFC] Added tests for simd class type traits (intel#4146) [SYCL][ROCm] Fix missing parameter in ROCm plugin (intel#4166) [SYCL][L0] Add temporary option to allow user to use copy engine for device to device copy (intel#4127) Remove check for AMD HIP to fix Driver/cuda-arch-translation.cu Reapply after conflict resolution 418a6d6 "Fix nvptx_target_teams_distribute_parallel_for_simd_codegen failure" Revert "[SYCL] Removes redefinitions of macros in libclc (intel#3505)" [PGO] Change test-run line to check NewPM pass behavior ...
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Depending on the host system's CUDA implementation,
CU_POINTER_ATTRIBUTE_MEMORY_TYPEqueries may report different values for managed pointers. Managed memory must however always be freed usingcuMemFree. This patch adds an additional case for thecuMemFreepath incuda_piextUSMFree.