Tags: amd/ZenDNN
Tags
[ZENDNN] Unify ZenDNN and zentorch framework cmake file (#531) - ZENAI-3320 - Single master fwk/ZenDnnlFwkIntegrate.cmake (ZenTorch includes this; no duplicate logic in plugin) - Static import target: whole-archive for heavy .a deps; shared import: normal INTERFACE links - Other minor changes Signed-off-by: Rakul Chauhan <rakul.chauhan@amd.com> Change-Id: I3e32bc9b260ad5b0044c2fca3175933f1f4d399e
[ZENDNNL] Immutable Caller Params (#566)
ZENAI-4203
- matmul_direct: copy params before reorder, kernel select, and dispatch
- group_matmul_direct: share exec_params across sequential and parallel paths
- fused_moe: route Op1/Op2 through exec_params working copies
- group quant: document and test caller params stay unchanged at API boundary
Signed-off-by: Navaneeth Krishna S <Navaneeth Krishna S>
Update PR notification * Fix PR notification: send legacy + robust jq card build ZENAI-3777 Change-Id: Ic772b9866b20a3fd3f1220ff95aabe5145673bb6 Signed-off-by: Kiriti Pendyala <kiriti.pendyala@amd.com>
[ZENDNNL] Gtest fix (#564) ZENAI-3967 Change-Id: I908325e280fcd7955c4a43b0f82e6a4fb9c1debe Signed-off-by: JITEN <Jiten.Parmar@amd.com>
Changing the version file name for zendnn (#550) ZENAI-3660 Changing the version file name for zendnn from zendnnl-version.cmake to zendnnl-config-version.cmake Change-Id: Ie2a67ca05ce362b96fac6b362d4b0f4001c9f57f Signed-off-by: Megha Sangtani <megha.sangtani@amd.com>
[ZenDNN]Add FP16 support for softmax operator (#540) * [ZenDNN]Add FP16 support for softmax operator - ZENAI-3951 Signed-off-by: Tirumala Reddy, Venkata Chetan <VenkataChetan.TirumalaReddy@amd.com> Change-Id: Iceea041d86599b076b93a6ee90738bef2b895de8
[ZENDNNL] Fix RC3 AIGtest (#541) LWPZENDNN-1034 - BMM: fix boundary fill/validation in AI gtests - Group quant: align INT8_DYNAMIC_GEMM reference with kernel path - GEMV: fall back to DLP when native BRGEMM alpha is zero Change-Id: Idb4bef232c68356a04424f5cdfff6963e5bb37e8 Signed-off-by: Navaneeth Krishna S <Navaneeth Krishna S> Co-authored-by: Navaneeth Krishna S <Navaneeth Krishna S>
[ZENDNNL] Fix RC3 AIGtest (#541) LWPZENDNN-1034 - BMM: fix boundary fill/validation in AI gtests - Group quant: align INT8_DYNAMIC_GEMM reference with kernel path - GEMV: fall back to DLP when native BRGEMM alpha is zero Change-Id: Idb4bef232c68356a04424f5cdfff6963e5bb37e8 Signed-off-by: Navaneeth Krishna S <Navaneeth Krishna S> Co-authored-by: Navaneeth Krishna S <Navaneeth Krishna S>
PreviousNext