{"entries":[{"name":".gitignore","mode":"0100644","size":7,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}},{"name":"CMakeLists.txt","mode":"0100644","size":1227,"lastCommit":{"hash":"ff27f80a74bbe5303acd511a6781a1de6d619b3c","oid":"ff27f80a74bbe5303acd511a6781a1de6d619b3c","parents":["d3248d9b6557c75d59954c594bb53cf517591e91"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-08-15T21:11:22+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-08-15T21:11:22+08:00"},"message":"ggml: initial IBM zDNN backend (#14975)\n\n* ggml-zdnn: inital backend impl\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: temp change z17 to arch15\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: fix build bugs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: tensor->extra logging check\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: add layout name mapping, ztensor information\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: separate logging into its own line\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: add shape comparison\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: add ggml_tensor shape log\n\nSigned-off-by: Aaron Teo \n\nggml-zdnn: fix incorrect shape logging\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add output buffer check\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: run compute and store into tensor->extra\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add set_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add more loggers\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update set_tensor logging to check only for matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: last working matmul version\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add comments to prevent accidentally deleting lines\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: support op out_prod\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update op out_prod to use tensor->extra\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rewrite the backend implementation\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: bugfix new impl\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix compiler warnings and bugfixes\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: test ztensor finding in init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: implement at least 1 op to test\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: assign tensor->extra to buffer\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add check for view tensors to prevent init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rework init_tensor to create new buffers\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to std vector instead of array\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch buffers back and set to arbitrary number\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: impl init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update supports_op matmul matrix\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix incorrect ztensor shape, reduce memory padding\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code clean up\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: impl matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix compiler error missing type\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix missing data transform call\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add bias init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: tighten memory usage, change string allocation\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add bias ztensor and data free\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add bias data transform\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add more debug info for extra buffer transform\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add logger to check if mat mul ops go through set_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: activate bias transform in matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move weights transform into mulmat\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add more safeguards in matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix sequencing of transforms\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: bugfix transform ztensor vs origtensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: figure out why sigtrap is happening\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix sigsegv\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move everything back to local declaration\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move bias data to local also\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: bring back working matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rewrite into mre\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix missing vector import\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix missing vector import in header\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt to fix sigsegv\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix missing load tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix invalid ztensor buffer release\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add logging to debug free buffer\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: remove free_buffer debug info\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add parmblkformat detections\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add nnpa installed detection\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add zdnn_init call for static libs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt at fixing invalid buffer\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to using deque to fix pointer deref problem\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add weights logging to check\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt to use unique ptr\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add tensor to pre_tfm_desc logging\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add inputs logging\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: disable op_none initialisation for testing\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix missing return from init_tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: load ztensors in cgraph exec\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: work on moving output ztensor as well\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: disable logging and breakpoints for full test\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt at manually changing the layout\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt at using default nwhc format instead\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: disable global load ztensor for now\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix errorenous output load tensor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: add guards to prevent loading ztensor if transformed\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: bring load ztensor back to init routine\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code clean up\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix ztensor deallocation abort\n\nstabilise ggml <-> zdnn api\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: clean up matmul selection\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: clean up project structure\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update documentation, prepare for upstream\n\nSigned-off-by: Aaron Teo \n\n* chore: add codeowners\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: disable batched matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: attempt at fixing tensor views during matmul\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: deny all view tensors directly\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix pr comments\n\nSigned-off-by: Aaron Teo \n\n* docs: update ops docs for zdnn\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: redo test-backend-ops for ops.md\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: fix typo in build-s390x.md\n\nSigned-off-by: Aaron Teo \n\n* codeowners: remove taronaeo for now\n\nSigned-off-by: Aaron Teo \n\n* Revert \"codeowners: remove taronaeo for now\"\n\nThis reverts commit 411ea4ed78d08778967bd0bd33a6538cfcbe082f.\n\n* ggml-zdnn: remove unused ggml_zdnn macro\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"75213ef13d16e318fda3a232871e40757b48dc55"}},{"name":"common.hpp","mode":"0100644","size":1194,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}},{"name":"ggml-zdnn.cpp","mode":"0100644","size":21393,"lastCommit":{"hash":"153d324bcf86d220b235ca010eeb11213f32b5d1","oid":"153d324bcf86d220b235ca010eeb11213f32b5d1","parents":["b3df57286c336255fd02c2162b0b5fe3adc2965e"],"author":{"name":"Ruben Ortlam","email":"rortlam@redhat.com","when":"2026-08-11T08:20:46+02:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2026-08-11T09:20:46+03:00"},"message":"llama: add default load-mode auto, which avoids mmap on iGPUs (#26081)\n\n* llama: add new default load-mode auto which picks mmap unless a non-Metal iGPU is used\n\n* Update ggml/src/ggml-hexagon/ggml-hexagon.cpp\n\nCo-authored-by: Max Krasnyansky \n\n* set mmap_support to false on OpenCL backend\n\n* fix order of load modes\n\n* use -1 for auto\n\n* resolve load mode auto earlier to correctly pick gpu host or cpu memory\n\n* add load mode auto to llama-bench\n\n* bump virtgpu api version, regenerate docs\n\n---------\n\nCo-authored-by: Piotr Wilkin (ilintar) \nCo-authored-by: Max Krasnyansky \nCo-authored-by: Georgi Gerganov ","tree":"f34c7980beeeddf5776f0c96b55e78f620b02ff4"}},{"name":"mmf.cpp","mode":"0100644","size":3636,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}},{"name":"mmf.hpp","mode":"0100644","size":307,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}},{"name":"utils.cpp","mode":"0100644","size":2830,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}},{"name":"utils.hpp","mode":"0100644","size":680,"lastCommit":{"hash":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","oid":"264f1b51872c125e23fa0ac1da5e2a1170de9a08","parents":["0bc7cc715472fe28822e57f036f0746592ed2c04"],"author":{"name":"Aaron Teo","email":"aaron.teo1@ibm.com","when":"2025-09-23T14:53:05+08:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2025-09-23T14:53:05+08:00"},"message":"zdnn: refactor codebase + add docs (#16178)\n\n* zdnn: initial matmul refactor\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm static from funcs\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: update ggml-zdnn.h\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: change header files to hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: switch to common.hpp\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: move mulmat forward around\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: rm inline from utils\n\nSigned-off-by: Aaron Teo \n\n* ggml-zdnn: code cleanup\n\nSigned-off-by: Aaron Teo \n\n* docs: add zDNN docs\n\nSigned-off-by: Aaron Teo \n\n---------\n\nSigned-off-by: Aaron Teo ","tree":"dfdb65842d2b78179004176d350bf1ce2e8ae1ca"}}],"lastCommit":{"hash":"153d324bcf86d220b235ca010eeb11213f32b5d1","oid":"153d324bcf86d220b235ca010eeb11213f32b5d1","parents":["b3df57286c336255fd02c2162b0b5fe3adc2965e"],"author":{"name":"Ruben Ortlam","email":"rortlam@redhat.com","when":"2026-08-11T08:20:46+02:00"},"committer":{"name":"GitHub","email":"noreply@github.com","when":"2026-08-11T09:20:46+03:00"},"message":"llama: add default load-mode auto, which avoids mmap on iGPUs (#26081)\n\n* llama: add new default load-mode auto which picks mmap unless a non-Metal iGPU is used\n\n* Update ggml/src/ggml-hexagon/ggml-hexagon.cpp\n\nCo-authored-by: Max Krasnyansky \n\n* set mmap_support to false on OpenCL backend\n\n* fix order of load modes\n\n* use -1 for auto\n\n* resolve load mode auto earlier to correctly pick gpu host or cpu memory\n\n* add load mode auto to llama-bench\n\n* bump virtgpu api version, regenerate docs\n\n---------\n\nCo-authored-by: Piotr Wilkin (ilintar) \nCo-authored-by: Max Krasnyansky \nCo-authored-by: Georgi Gerganov ","tree":"f34c7980beeeddf5776f0c96b55e78f620b02ff4"}}