Commit Graph

13188 Commits

Author SHA1 Message Date
SaleelK 10635483ad clr: Fix packet batch write logic (#2236)
* When writing bulk packets always invalidate packet headers, Its
  possible that the CP fetcher can have multiple packets in flight. In
such cases we may end up with a malformed packet because the writes are
not complete yet CP finds a valid header.
2025-12-11 04:26:41 -08:00
Matt Arsenault a495d1137e SWDEV-548892 - Make declaration of __ockl_fdot2 always available (#2229) 2025-12-11 11:53:11 +01:00
German Andryeyev 3895aadba6 SWDEV-558849 - Make ROCR path in Windows more stable (#2181) 2025-12-10 12:37:10 -05:00
Pengda Xie 1d6b26f829 SWDEV-556684 - HSAIL Cleanup re-apply commit 4abdfe5: (#2024)
Removed some options

-xnack, -force-wgp-mode, -force-wave-size-32, -round-trip-spirv,
-fe-gen-spirv, -lower-pipe-builtins=0|1, -lower-atomics=0|1,
-set-lds=<value>, -set-scalar-registers=<value>,
-set-vector-registers=<value>, -limit-scalar-registers=<value>,
-limit-vector-registers=<value>, -sc-xnack-iommu,
-faa-for-barrier/-fno-a-for-barrier, -sc-dev-format, -verify-lwspir,
-verify-hwspir, -ffma-enable/-fno-fma-enable,
-fmad-enable/-fno-mad-enable, -fdisable-avx/-fno-disable-avx,
-fforce-llvm/-fno-force-llvm, -print-compile-phases,
-kernel-cache-enforce-miss, -kernel-cache-wipe, -kernel-cache,
-sc[=<filename>]/--load-sc-dll[=<filename>],
-be[=<filename>]/--load-be-dll[=<filename>],
-cg[=<filename>]/--load-cg-dll[=<filename>],
-link[=<filename>]/--load-link-dll[=<filename>],
-opt[=<filename>]/--load-opt-dll[=<filename>],
-fe[=<filename>]/--load-fe-dll[=<filename>],
-cl[=<filename>]/--load-cl-dll[=<filename>], -just-kernel=<kernel-name>,
-use-debugil, -fmulti-level-call/-fno-multi-level-call,
-fdebug-call/-fno-debug-call, -fmacro-call/-fno-macro-call,
-fstack-uav/-fno-stack-uav, -fdef-res-id/-fno-def-res-id,
-wokth=int/--waves-opt-kernel-threshold,
-ilkth=int/--inline-kernel-size-threshold,
-ilsth=int/--inline-size-threshold, -ilcth=int/--inline-cost-threshold,
-scopt=int/--sc-opt-level, -flib-no-inline/-fno-lib-no-inline,
-fuser-no-inline/-fno-user-no-inline,
-scras=int/--sc-si-opt-reg-alloc-strategy, -fsc-post-ra-sched,
-fsc-live-sched/-fno-sc-live-sched, -fsc-use-buffer-for-hsa-global,
-fsc-schedule-no-reorder, -fsc-min-reg-schedule,
-fsc-bias-schedule-to-minimize-insts,
-fsc-bias-schedule-to-minimize-regs, -fsc-disable-merge-memory,
-fsc-disable-loop-unroll, -fsc-use-mubuf/-fno-sc-use-mubuf,
-fsc-selective-inline/-fno-sc-selective-inline,
-fsc-keep-calls/-fno-sc-keep-calls, -slc=0|1/--simplifylibcall,
-stack-alignment=<n>, -fdiv2fmul=0|1, -prt-opt-liveness=0|1,
-liveness=0|1, -SRAE-threshold=<value>, -memcombine-max-vec-gen=<value>,
-small-global-objects, -fast-fmaf, -fast-fma, -bfo=0|1, -ebb=0|1, -aa,
-mem2reg=0|1, -licm=0|1, -unroll-allow-partial,
-unroll-threshold=<positive integer>, -unroll-count=<positive integer>,
-apt/--ap-threshold=<positive integer>, -srt/--sr-threshold=<positive
integer>, -fdebug-linker/-fno-debug-linker, -fbin-gpu64/-fno-bin-gpu64,
-fbin-disasm/-fno-bin-disasm, -fbin-bif30, -fbin-hsail/-fno-bin-hsail,
-fbin-amdil/-fno-bin-amdil, -fbin-spir/-fno-bin-spir, -fonly-bin-source,
-fper-pointer-uav/-fno-per-pointer-uav

Co-authored-by: Konstantin Zhuravlyov <kzhuravl_dev@outlook.com>
2025-12-10 09:09:12 -08:00
Dominic Widdows 75bea883e1 Remove redirect notice from redirect target (#2104)
README is copied from https://github.com/ROCm/clr which redirects to target
https://github.com/ROCm/rocm-systems/tree/develop/projects/clr

This is correct for https://github.com/ROCm/clr I think, but unnecessary for https://github.com/ROCm/rocm-systems/edit/develop/projects/clr/README.md which is already the correct redirect target.
2025-12-09 10:47:51 -08:00
Victor Zhang aaecffa50b SWDEV-568847 - prevent UAF when registering callbacks on completed events (#2066)
* SWDEV-568847 - prevent UAF when registering callbacks on completed events

* cache the status() of event earlier

* Update command.cpp

* revert cl_event.cpp

* Update cl_event.cpp

---------

Co-authored-by: cadolphe-amd <chris.adolphe@amd.com>
2025-12-09 11:38:45 -05:00
Jatin Chaudhary eea93d58a2 SWDEV-554626 - return correct error code (#1107)
* SWDEV-554626 - return hipErrorInvalidDeviceFunction when we can not load module
Return correct error code when modules are empty

* Match the error codes

* Revert the error code
2025-12-09 16:10:25 +01:00
SaleelK acc236fd89 clr: Avoid saving all ProfilingSignals at once (#2108)
* While reusing signals, its possible we can come across a timestamp
  that can contain several signals, like when profiling a graph. Reading
timestamps from all signals can make the call severely CPU bound.
Instead cache only that signal so as to avoid the overhead for critical
path.
2025-12-08 11:32:16 -08:00
Jin Jung deaf8ab38a SWDEV-567119 - Windows GL Interop Support (#1892) 2025-12-08 11:03:59 -05:00
Shadi Dashmiz 4812d8e78b SWDEV-566783 - clean up cmgr helper (#1864)
Signed-off-by: sdashmiz <shadi.dashmiz@amd.com>
2025-12-08 10:37:03 -05:00
Ioannis Assiouras 3faf36fb25 Fix Unit_hipStreamBeginCaptureToGraph_CapturePartialInThreads (#2072)
https://mlsejenkinsvm.amd.com/job/rocm-systems/job/hip/view/change-requests/job/PR-2072/6/
The last windowsCI has passed successfully
2025-12-08 13:30:23 +01:00
Lancelot Six 659737c824 clr: Bump _amdgpu_r_debug.r_version to 11 (#2063) 2025-12-05 16:01:08 -05:00
Rahul Manocha 9dd3c2fa70 SWDEV-563271 - return error when pal cmd submission fails (#1585) 2025-12-05 14:25:01 -05:00
Ajay GunaShekar d6f6435b88 SWDEV-526504 - Remove perl dependency in hip/clr build (#964)
* SWDEV-1 - Remove perl dependency in hip/clr build
* SWDEV-1 - use python3 inplace of perl for formatting date,time
2025-12-05 08:42:15 -08:00
Julia Jiang 272f06506f SWDEV-549696 - Fix HIP catch sub-test failure for MipmappedArray (#1198)
Co-authored-by: JeniferC99 <150404595+JeniferC99@users.noreply.github.com>
2025-12-05 11:00:06 -05:00
systems-assistant[bot] 06a3a5ca10 SWDEV-546110 - Fix encoding for certain types (#446) 2025-12-05 13:16:14 +00:00
harkgill-amd 8f622de972 Add gfx1152 support to PAL (#2077) 2025-12-03 10:39:22 -08:00
Matt Arsenault d75d0bc1c9 SWDEV-548892 - Stop using ocml exp and exp2 functions (#2032) 2025-12-02 13:39:09 -05:00
Ioannis Assiouras 65b769ee16 SWDEV-569101 - increase signal list size to at least DEBUG_HIP_GRAPH_BATCH_SIZE (#2084) 2025-12-01 18:52:51 -08:00
SaleelK c105dcd05b clr: Use graph segment scheduling to process HIP Graphs (#1372)
* clr: Use graph segment scheduling to process HIP Graphs

* Add a broader path to use capture packet capture for all topologies
* Refactor code
* Use DEBUG_HIP_GRAPH_SEGMENT_SCHEDULING to toggle new vs classic path,
  Enabled by default

* clr: Few fixes and improvements

* clr: Detect complex graphs to take classic path

* Use DEBUG_HIP_GRAPH_SEGMENT_SCHEDULING=2 to force segment scheduling
  path

* clr: Fix a cornercase stack corruption

* clr: Track commands of segments instead of snapshots

* clr: Fix Batch dispatch logic

* Track fence_dirty_ flag for command of other streams
* Dependency resolution markers can now accomodate dirty fence on cross
  streams

---------

Co-authored-by: Ioannis Assiouras <Ioannis.Assiouras@amd.com>
Co-authored-by: Godavarthy Surya, Anusha <agodavar@amd.com>
2025-12-01 12:49:26 -08:00
vstojilj 77f58ceb9f SWDEV-558557 - Remove duplicate nodes when capturing hipMemcpyAsync (#1226) 2025-12-01 11:25:13 +01:00
vstojilj 1c09c87cc7 SWDEV-564927 - Allow sizeBytes to be 0 when hipMemsetAsync is captured (#1849) 2025-11-27 17:13:33 +01:00
Godavarthy Surya, Anusha 2e1c37a926 SWDEV-490861 - Remove recursion and extra loop in hipGraphLaunch (#1792) 2025-11-27 10:25:08 +00:00
Shadi Dashmiz 962b99f925 SWDEV-567514: Remove default stream wait (#1977)
- when virtual map command is called

- can create deadlock

Signed-off-by: sdashmiz <shadi.dashmiz@amd.com>
2025-11-26 15:11:52 -05:00
AidanBeltonS d849b88aef SWDEV-558080 - Add recommended granularity (#1176)
* Add recommended granularity

* Improve granularity testing

* Update based on feedback
2025-11-26 16:10:58 +00:00
Matt Arsenault f089217e6a SWDEV-548892 - Stop using ockl steadyctr function (#1882)
Directly use the builtin
2025-11-26 09:29:06 -05:00
Todd tiantuo Li ee48f6221d SWDEV-562708 - change default maximum SVM size to 256GB (#1731) 2025-11-25 23:59:39 -08:00
Matt Arsenault 9fbb062505 SWDEV-548892 - Stop using ocml isinf wrapper (#1854) 2025-11-25 22:21:37 -05:00
Karthik Jayaprakash 740a06d567 SWDEV-559267 - Use CLPrint to DevLogPrintf with Log Level - detail debug. (#1160) 2025-11-25 19:25:32 -05:00
German Andryeyev 93682f2f75 SWDEV-567852 - Clean-up hip::init() (#1948) 2025-11-25 19:05:41 -05:00
cadolphe-amd cce94f6ee0 SWDEV-557412 - Incorporate proper chunk offset when remapping virtual memory (#1848)
* SWDEV-557412 - Incorporate proper offset when remapping virtual memory

* Fix condition to check if VMHeap allocation address matches a chunk address

* Move offset calculation outside if/else block

---------

Co-authored-by: JeniferC99 <150404595+JeniferC99@users.noreply.github.com>
2025-11-25 18:05:25 -05:00
Victor Zhang ede71ca3b0 SWDEV-567829 - populateFormatStringHashMap: relax printf hash collisi… (#1944)
* SWDEV-567829 - populateFormatStringHashMap: relax printf hash collision check for duplicate format strings

* function optimized by ai
2025-11-25 17:19:27 -05:00
German Andryeyev 2c5754844f SWDEV-465041 - Enable direct dispatch under Linux by default. (#1934) 2025-11-25 11:30:32 -05:00
Pengda Xie 6c31785eaf SWDEV-562761 - Cleanup static fatbin on runtime teardown (#1873) 2025-11-24 21:57:46 -08:00
Marius Brehler 2dc32d645b Explicitly load versioned libamdhip64.so (#1872)
* Explicitly load versioned libamdhip64.so

* Fix syntax errors

* Fix when patching happens in Windows workflow

---------

Co-authored-by: Joseph Macaranas <145489236+jayhawk-commits@users.noreply.github.com>
Co-authored-by: ammallya <ameyakeshava.mallya@amd.com>
2025-11-24 10:05:05 -08:00
sluzynsk-amd 2cf9faa93f SWDEV-563777 - fix warnings related to inconsistent overrides (#1625)
This patch adds missing override keywords. Fixes this class of warnings.

Signed-off-by: Sebastian Luzynski <Sebastian.Luzynski@amd.com>
2025-11-24 18:50:07 +01:00
AidanBeltonS 0580e2053c SWDEV-533546, SWDEV-540027 - Add e8m0 conversions and testing (#987)
* SWDEV-533546 - Add conversion functions for e8m0

* SWDEV-533546 - remove whitespace

* Add testing

* Update based on feedback

* Copilot suggestions

---------

Co-authored-by: systems-assistant[bot] <systems-assistant[bot]@users.noreply.github.com>
2025-11-24 09:14:03 +00:00
Ioannis Assiouras 36029ea1a8 SWDEV-559166 - Fix race condition in getDemangledName (#1868) 2025-11-23 08:45:45 +00:00
Ioannis Assiouras 7313c3752a SWDEV-567475 - Fix failures in graph tests due to GraphExec destroy h… (#1917) 2025-11-22 23:01:47 +00:00
Julia Jiang 78a9d9ff70 [clr] SWDEV-566950 - Adding changelog for 7.2 (#1891)
* [clr]SWDEV-566950 - Adding changelog for 7.2

* Update CHANGELOG.md

* Update CHANGELOG.md
2025-11-19 09:10:14 -08:00
raramakr eddd4c3601 SWDEV-505204 - Update libamdocl.so installation path to avoid exposing all ROCm libraries via ldconfig (#1914)
ldconfig is run during rocm-opencl package installation.
Installing libamdocl.so in /opt/rocm-xxx/lib exposes all ROCm libraries when /opt/rocm/lib is added to ldconfig.
To prevent this, libamdocl.so is now installed in /opt/rocm-xxx/lib/opencl.
ldconfig will use the updated path, limiting exposure to only libamdocl.so library.

Co-authored-by: raramakr <raramakr@amd.com>
2025-11-19 21:14:28 +05:30
German Andryeyev ff4782620e SWDEV-547108 - Fix PAL build with HSA backend (#1850)
When hip is built with HSA backend then the headers from ROCR will be used, but
scratch_backing_memory_byte_size is a part of amd_queue_v2_t structure
2025-11-14 12:28:03 -05:00
Matt Arsenault 4830979f0e SWDEV-548892 - Stop using ocml fma wrappers (#1702)
Directly use elementwise builtin
2025-11-13 16:20:27 -08:00
Matt Arsenault 42e91b8934 SWDEV-548892 - Stop using ocml sqrt wrappers (#1716) 2025-11-13 16:19:44 -08:00
Julia Jiang 5599e8b1de SWDEV-561500 - Update change log and port 7.1.1 to develop branch (#1688)
* SWDEV-561500 - Porting changelog(up to 7.1.1) to develop branch

* Update CHANGELOG.md

* Update CHANGELOG.md

* Update CHANGELOG.md
2025-11-13 12:22:34 -08:00
pcritchl-amd 60cd210dac Reapply "SWDEV-562996 - Build fix: Ubertrace callback calling convention mismatch on x86 (#1587)" (#1717) (#1754) 2025-11-12 13:47:24 -05:00
Ioannis Assiouras 4f91b68988 SWDEV-559166 - Remove obsolete member execInfoOffset from KernelParameters (#1790) 2025-11-12 17:20:36 +00:00
Satyanvesh Dittakavi 07dd4c85e7 SWDEV-546308 - Implement hipKernelGetParamInfo API (#1783) 2025-11-12 14:09:26 +05:30
jofrn 8f9da259ac Fix memory leak in hip_fatbin.cpp UncompressAndPopulateCodeObject (#1692)
Wrap amd_comgr_data_t item returned from action_data_get_data() in
ComgrDataUniqueHandle to ensure it gets released.
2025-11-11 16:48:06 -05:00
systems-assistant[bot] a66ca8809b SWDEV-511239 - Remove and and use && for preprocessors (#506)
This shows up as warning in msvc.

Co-authored-by: Jatin Chaudhary <JatinJaikishan.Chaudhary@amd.com>
2025-11-11 09:43:57 -08:00