1cc35da9be
SWDEV-438790 - Remove DEBUG_HIP_7_PREVIEW env var keeping the hipGetLastError changes by default ( #337 )
...
[ROCm/clr commit: 664bf232dd
2025-05-21 22:12:45 +05:30
834d770ca4
SWDEV-508964 - Match hip launch kernel retuns with cuda ( #357 )
...
[ROCm/clr commit: 88f15877f6
2025-05-21 21:59:11 +05:30
720660d552
SWDEV-527851 SWDEV-527890 SWDEV-529456 - Make HIP vector simple struct ( #356 )
...
* Make HIP vector simple struct
Change-Id: I8442c2cc9af26b2a3c7d6719e3348df1593e83b3
* Update make_vector_type
Change-Id: Ic5060994a08baa4262c2a4b09fcbe6bc74276720
[ROCm/clr commit: 3cd3b3ffc5
2025-05-21 21:30:55 +05:30
39b9a46025
SWDEV-418904 - Remove hiprtc symbols from hip library ( #370 )
...
[ROCm/clr commit: e9dbd7c99d
2025-05-21 21:22:47 +05:30
6467351f4c
SWDEV-508962 - [6.4 Preview] Update hipPointerGetAttributes to match CUDA > 11.0 behavior ( #120 ) ( #363 )
...
[ROCm/clr commit: 0b64eec921
2025-05-21 21:18:05 +05:30
3ea732b38a
SWDEV-508965 - [6.4 Preview] Remove HIP_MEMSET_NODE_PARAMS struct ( #121 ) ( #364 )
...
[ROCm/clr commit: 968f5599a8
2025-05-21 21:16:42 +05:30
22da169b86
SWDEV-532868 - Disable implicit wait in hipFree for async allocations ( #323 )
...
* SWDEV-532868 - Disable implicit wait in hipFree for async allocations
* Fix compilation error
[ROCm/clr commit: 58df22546b
2025-05-21 21:14:38 +05:30
16aefe62f6
SWDEV-508979 - Match hipModuleLoad negative return with Cuda ( #326 )
...
[ROCm/clr commit: 7a357800dd
2025-05-21 21:09:07 +05:30
10d97b4b3c
SWDEV-508970 - Match hipBindTextureToArray negative return with Cuda ( #325 )
...
[ROCm/clr commit: a46ed60e2e
2025-05-21 20:57:09 +05:30
12b43826dd
SWDEV-519346 - Fix __clock64() compiling issue in SPIRV ( #207 )
...
[ROCm/clr commit: 78f3caba0b
2025-05-20 16:57:17 -04:00
a17e22a78c
SWDEV-529929 - hipMemGetHandleForAddressRange implementation. ( #245 )
...
[ROCm/clr commit: 12131de4a9
2025-05-20 15:56:04 -04:00
b59859c52e
SWDEV-457749 - Use size of handle for range instead of actual size for physmem. ( #342 )
...
[ROCm/clr commit: bed454caa1
2025-05-20 15:24:49 -04:00
ee7c5554db
SWDEV-523279 - Use preferred engine mask for SDMA ( #317 )
...
- ROCr now reports preferred engine for copy status. We can leverage
this for max bandwidth for inter-GPU copies
- Cleanup logging
[ROCm/clr commit: 1b0ea080e4
2025-05-19 16:04:51 -07:00
5fd5e846e2
SWDEV-531518 - Fix offset accumulation ( #333 )
...
srcAddress/dstAddress accumulation was cumulative, which shouldnt be
done if we increment offset.
[ROCm/clr commit: 5712944c7c
2025-05-19 18:03:06 +05:30
f4c9056cdc
Revert "SWDEV-522707 - Set phys_mem_handle type to sizeof(size_t) to avoid blocking address range. ( #105 )" ( #348 )
...
This reverts commit 0071d33754bb7750a946
2025-05-19 15:19:16 +05:30
4ea2d9a5ee
SWDEV-531711 - Report correct error code based on device failure. ( #286 )
...
[ROCm/clr commit: f5b8db33f1
2025-05-17 06:33:13 -04:00
dc39d67017
SWDEV-508742 - Make clCreatePipe spec compliant ( #80 )
...
[ROCm/clr commit: 7698d799ce
2025-05-16 15:18:35 +05:30
c04355bbe3
Add __syncwarp operation ( #160 )
...
Change-Id: I6a3783beafdbb9f11a3b37333f4ff3f5be27ea54
[ROCm/clr commit: c50610b44d
2025-05-15 14:20:13 +05:30
7b6e59cbea
SWDEV-530803 - User current device id while cloning graph node. ( #313 )
...
[ROCm/clr commit: 32eb6a5d89
2025-05-15 09:06:15 +05:30
b13eec7049
SWDEV-345024 - Retain the program on Fini kernel execution ( #307 )
...
Fini kernel is executed during the invocation of amd::Program destructor,
but the dispatch logic can retain/release the reference counter and
cause double free. Avoid double free with an extra retain() call
[ROCm/clr commit: bddb8f14d1
2025-05-14 21:21:26 +05:30
d0b5bee680
SWDEV-527781 - Remove Stream Validation in HIP APIs
...
[ROCm/clr commit: 0457b634f8
2025-05-13 13:45:27 -07:00
71f19d7017
SWDEV-529449 - Bug fix when retrieving a memobj from the IPC mem handle
...
[ROCm/clr commit: f7482ef0a6
2025-05-13 19:18:22 +01:00
f6e2ce709e
SWDEV-491314 - Re-enable cross-lane sync builtins ( #94 )
...
* Enables warp sync builtins by default
* Removes HIP_ENABLE_WARP_SYNC_BUILTINS; that macro will no longer have an effect. Instead, we will now be able to disable the builtins with the macro: HIP_DISABLE_WARP_SYNC_BUILTINS
[ROCm/clr commit: 5606debd8e
2025-05-13 16:35:58 +01:00
92b0f9d114
Update palamida.yml ( #266 )
...
* Add palamida.yml - removing url
[ROCm/clr commit: 29df3ae6e9
2025-05-12 21:39:21 -07:00
a128beeeb5
SWDEV-506467 - Fixing compilation issue seen on clang compilation for ASAN. ( #253 )
...
[ROCm/clr commit: 876de49b11
2025-05-12 17:16:56 -04:00
cc6bfe71fc
SWDEV-528683 - Hardcode valid wavefront compile time options ( #306 )
...
[ROCm/clr commit: f9199ac205
2025-05-12 19:29:39 +02:00
b9669ea266
SWDEV-531678 - Remove split path from the dispatch ( #283 )
...
The split path for blit kernels are no longer necessary, since the new blit kernels
don't use the copy size as the global workload
[ROCm/clr commit: da198ac5b2
2025-05-12 12:50:32 -04:00
5dff283f4e
SWDEV-526855 - Modify the SIMDPerCU calculation for gfx1250/1. ( #275 )
...
[ROCm/clr commit: acb1f7e8d5
2025-05-12 11:09:03 -04:00
3d862af4df
SWDEV-512344 - Unmap all subbuffers ( #214 )
...
[ROCm/clr commit: c5ced8c3a2
2025-05-12 16:56:10 +02:00
558a26cf98
SWDEV-511204 - Mapped virtual memory should use device instead of host context ( #213 )
...
Since the sub-buffer(virtual memory that is mapped to device memory) is associated with device memory, it should utilize the device context instead of the host context. The original implementation caused hipMemcpyPeer to not take the P2P path, as the memory object was treated as host memory.
[ROCm/clr commit: a7492c516d
2025-05-12 16:55:25 +02:00
5bd5f3aa6e
SWDEV-521135 - Make common way to set/parse UUID bytes from PAL props. ( #63 )
...
[ROCm/clr commit: 6858b0fca1
2025-05-12 17:00:30 +05:30
fe14cc349b
SWDEV-517078: Fix gfx11 trap handler ( #212 )
...
Fix incorrect edits done when porting the 2nd level trap handler from
the hsa-runtime.
Change-Id: I7bc5160be47b8f669efe05c4d194bc3c47fc0661
[ROCm/clr commit: c35e9643ec
2025-05-11 01:12:28 +01:00
142e347aac
SWDEV-528860 - reserve some memory in visible frame buffer ( #251 )
...
[ROCm/clr commit: faac50c77a
2025-05-09 20:08:23 -04:00
3decd4f975
SWDEV-527299 - Support HIP_POINTER_ATTRIBUTE_CONTEXT ( #180 )
...
* SWDEV-527299 - Support HIP_POINTER_ATTRIBUTE_CONTEXT
As HIP enables UVA by default, it seems we can simply expose the context to support this feature.
[ROCm/clr commit: b434fbe2bd
2025-05-09 17:34:16 +08:00
a8630e866d
SWDEV-525933 - add constexpr operators for fp16/bf16 ( #199 )
...
[ROCm/clr commit: 2f73e1385b
2025-05-09 09:53:58 +01:00
e1d2194b75
SWDEV-528913 - support gfx950 in rocsetting ( #217 )
...
* SWDEV-528913 - support gfx950 in rocsetting
---------
Co-authored-by: Jimbo Xie <jiabaxie@amd.com >
[ROCm/clr commit: a320a3f214
2025-05-07 15:44:49 -04:00
dc1c1e3199
SWDEV-518221 - Don't link against libamd_comgr.so at runtime
...
Convention is to always link against .so.* at runtime.
Having it link against .so will break on systems that package
the .so files in their dev/devel package.
This issue was found when building ROCm 6.4 for Fedora.
Commiting on behalf of GitHub user Mystro256
[ROCm/clr commit: 6b12154583
2025-05-07 11:56:41 -07:00
fbabd2b69d
SWDEV-528142 - add error check for KernelParameters::capture ( #276 )
...
* SWDEV-528142 - add error check for KernelParameters::capture
* Update kernel.cpp
---------
Co-authored-by: victzhan <victzhan@amd.com >
[ROCm/clr commit: f960433dcd
2025-05-07 09:52:09 -04:00
cde2a250ec
SWDEV-493805 - Cleaning up launch parameters arguments. ( #241 )
...
[ROCm/clr commit: fa55557f46
2025-05-06 15:06:13 -04:00
086a1c289a
SWDEV-529831 - Return error if the program is empty ( #257 )
...
[ROCm/clr commit: 607f8f26fd
2025-05-06 15:12:12 +05:30
b5f67d4804
SWDEV-529854 - __hmax/__hmin should handle nan's ( #246 )
...
[ROCm/clr commit: a71c6eb1a0
2025-05-06 09:42:15 +01:00
a12739ecd9
SWDEV-529927 - add missing operations for fp16/bf16 ( #238 )
...
[ROCm/clr commit: b1ebf33850
2025-05-06 09:41:21 +01:00
3ea758a2d4
SWDEV-528808 - Release all HW queues even if only one is idle ( #240 )
...
Pytorch may not explicitly idle each queue. Thus, some queues can be considered as busy,
but have idle state in reality
[ROCm/clr commit: 65a0181a7c
2025-05-05 19:09:01 -04:00
ee91a1e94a
Disable HIP_PLATFORM auto-detect if already defined ( #254 )
...
Co-authored-by: Stella Laurenzo <stellaraccident@gmail.com >
[ROCm/clr commit: 3775298655
2025-05-05 15:37:53 -04:00
13d8f9adae
SWDEV-1 - Stop using ocml rounding functions ( #228 )
...
Directly use the builtins. Use the elementwise versions since there's
no implied errno, regardless of -f[no]-math-errno.
I didn't change the cases unnecessarily casting. The bfloat and vector
cases should work directly.
[ROCm/clr commit: 1db9a7d48b
2025-05-05 19:35:12 +02:00
c512258e45
SWDEV-528808 - Disable dynamic queue by default ( #256 )
...
Dynamic queue management will be disabled by default and
the original sort logic is restored
[ROCm/clr commit: 9b018165ce
2025-05-05 10:56:35 -04:00
e480220c6a
Fix typos in warning msgs ( #231 )
...
[ROCm/clr commit: cd9bc61559
2025-05-02 14:31:42 -07:00
f619372ae6
SWDEV-514560 - add fp6 header implementation ( #54 )
...
Co-authored-by: rahul manocha <rmanocha_amdeng>
[ROCm/clr commit: 12febe6782
2025-05-01 15:17:38 +01:00
3d4ff304d7
SWDEV-521011 - Fix alignment in PalResource::CreateSvm
...
[ROCm/clr commit: 9d6a0d1a4d
2025-05-01 02:22:49 +01:00
13c7977d50
SWDEV-526836 - add PipelineStageBlt flag ( #229 )
...
CP sync requires PipelineStageBlt flag.
[ROCm/clr commit: 84a4f293f4
2025-04-30 14:27:41 -04:00
Dittakavi, Satyanvesh
Brzak, Branislav
Belton-Schure, Aidan
Assiouras, Ioannis
Andryeyev, German
Sang, Tao
Jayaprakash, Karthik
Kudchadker, Saleel
Patel, Jaydeepkumar
Xie, Pengda
Hernandez, Gerardo
Hila, Nino
Arandjelovic, Marko
Six, Lancelot
Xie, AlexBin
Huang, AnZhong
Chaudhary, Jatin Jaikishan
Xie, Jiabao(Jimbo)
Lambert, Jacob
Zhang, Victor
Guan, Zichuan
Arsenault, Matthew
Searles, Mark