Also add flag: HIP_FORCE_SPIRV_CODEOBJECT to allow override to force use
SPIRV.
* use cache for already compiled code objects
* address review comments and use the two spirv isa names
Fix pytorch 2.5 issues, by defining reduce sync operations for type __half in amd_hip_fp16.h and not in
amd_warp_sync_functions.h which is problematic in case __half does not get included before that header.
Only define types not supported by cuda if HIP_ENABLE_EXTRA_WARP_SYNC_TYPES is defined, to avoid portability issues
Explicitly nulling the pointer causes us to report the error below
instead of keeping a dangling pointer around that will most likely lead
to a subsequent segfault.
Currently, we check if there's enough system RAM even if we don't allocate on host device. This is incorrect logic.
We should not check for this size on windows because PAL checks for memory allocation. See SWDEV-467263.
Co-authored-by: Jimbo Xie <jiabaxie@amd.com>
This reverts commit 57df1b348f.
Reason for revert: 6.4 Preview changes need not be merged to amd-staging as of now
Change-Id: I86452adfed14655f72d90440a486089743cc6587
This reverts commit c07468e53c.
Reason for revert: 6.4 Preview changes need not be merged to amd-staging as of now
Change-Id: Ifba0c8a248bc40deaa9c59b7f2901531300e5ea4
This reverts commit 9faaf20aae.
Reason for revert: 6.4 Preview changes need not be merged to amd-staging as of now
Change-Id: I04af8603053338f08c396e78ff8a6715e641ca19
Using target_link_libraries does not properly link the hipRTC-header.o
into libhiprtc for static build. Change to use target_sources instead.
This does not affect the linkage in the shared build.
Change-Id: I626f9eacc1637b792a50e7ddddb5db09e704ac4a
- hipStreamWaitEvent may not resolve streams
- Correct usage of flag passed to streamWait function
Change-Id: I2ee163615d303b98937c1035d60da283cce6f677
- This change tries to save extra synchronization packets we may insert
as we didnt track the completion signals for every command. We track
the current enqueued command until it exits the enqueue stage. We also
record the exit scope to know if we flushed the caches
- Handle correct release scopes and store completion signal as HW events
- Use a new finishCommand implementation to only wait for the command
passed as the argument
Change-Id: Ie4350c5dd24f5d48dfa6ccbabd892f0544caadcc
Since hipMemMap can be called for multiple device handles on the same virtual memory, the same is true for hipMemUnmap, meaning that virtual memory can be "partially unmapped".
This means that the unmap function can be called for a specific part of the reserved address, meaning that only the designated subbuffer should be released. If unmap is called on the entire reserved memory, then all subbuffers should be released.
The main point is that for every hsa_amd_vmem_map, there should be a corresponding hsa_amd_vmem_unmap. Otherwise, if entire memory is unmapped by a single unmap call, then HSA will report the memory as "in use" if an attempt is made to delete it.
Change-Id: I039308eafb820decfb1c09f60347f26cdad1a362
1) Add Linker APIs to runtime to support SPIRV linking
2) Migrate Internal implementations to runtime and share with rtc
3) Add Support to bundled and unbundled SPIRV Code object linking.
Change-Id: Ic1fd4431f842a208a2468e8aec54a65b5fa6b0e3
This change removes the stream callback from hipStreamWaitEvent and
uses a stream memory wait operation instead. This allows the
hipStreamWaitEvent to be non-blocking on the host.
Change-Id: Ie5530febda5a5bcb5daa0db8a01249d6b137fd43
Add initial implementation of virtual memory heap with
dynamic virtual memory mapping support for memory pools.
DEBUG_HIP_MEM_POOL_VMHEAP controls the new method.
Change-Id: I8dc5be2e0f34ab472f1800f43bb6243639a5e500
- Also add a cache, which allows compiled code objects to be reused
instead of compiling again. This should improve performance on
multigpu systems.
Change-Id: Ib135d616c076b77f8aaf28de275d408b38021d89
There are 2 functional changes to this patch:
* Use GPU timing for internal markers for HIP.
* Measure CPU time closer to GPU timer, to reduce delta between GPU/CPU timestamp measurements.
There are some smaller non-functional updates:
* waifForFence -> waitForFence typo
* Remove unused drmProfiling
Change-Id: I4c5fa600a842ab60e454888779edcac8449a902a
Resolved an issue where hipEventSynchronize and hipStreamWaitEvent APIs
did not function correctly for events created with the hipEventInterprocess flag.
The bug caused the event to be incorrectly marked as "recorded,"
leading to these APIs failing to wait for the event as expected.
Change-Id: Ic9fdfaab2393beb93d6e0b83661545e902a63499
hipGetLastError should return the error by any of the previous APIs
in the same host thread to match the CUDA behavior, whereas
hipExtGetLastError will return the error by the immediate previous API.
This Ext API was added earlier to facilitate the existing HIP apps which
are following the current behavior of hipGetLastError
Change-Id: I61e95b1fc136cc761e2434e02187b7ed2598b733
Xie, Pengda
Chaudhary, Jatin Jaikishan
Hernandez, Gerardo
Haehnle, Nicolai
Xie, Jiabao(Jimbo)
Patel, Jaydeepkumar
Stojiljkovic, Vladana
Brzak, Branislav
Andryeyev, German
Betigeri, Sourabh
Sang, Tao
Belton-Schure, Aidan
Arandjelovic, Marko
Godavarthy Surya, Anusha
Rakesh Roy
Ioannis Assiouras
Jacob Lambert
Julia Jiang
Saleel Kudchadker
Rahul Manocha
Li, Todd tiantuo
Jimbo Xie
Satyanvesh Dittakavi