master
Fix fused resnet low accuracy (#21122)
Add async GPU dependency Engine (#20331)
[FEATURE] Add tanh approximation for GeLU activation (#21034)
[v2.0][LICENSE] Port #20493 (#20608)
Port top-level-project updates from v1.x branch (#21162)
[BUGFIX] _npi_repeats with swap (#21112)
Add quantized batch norm operator fused with ReLU (#21137)
Improve bf16 support (#21002)
[FEATURE] Dnnl sum primitive path (#21132)
[Master] Clang-formatter: only src/ directory (#20571)
Get rid of warnings (#21099)
[master][clang-format] Re-format cc. .h. .cu files; cond. (#20704)
[BUGFIX] Reenable fwd conv engine 5 on test_group_conv2d_16c (#21104)
[BACKPORT][BUGFIX][FEATURE] Add oneDNN 1D and 3D deconvolution support and fix bias (#20292)
Port #20520 (#20609)
Automatic Layout Management (#20718)
[2.0] [BACKPORT] of [1.x][FEATURE] CUDA graphs support (#19142) (#20324)
[BUGFIX] Fix floor divide (#21096)
Support bfloat and float mixed input in elemwise broadcast ops (#20932)
oneDNN FullyConnected weight caching & refactor (#21047)
Restore quantized RNN to master (#20952)
Port website fixes (#20993, #20995, #21009) from v1.9.x branch (#21015)
Refactor SupportDNNL functions (#21032)
[v2.0][LICENSE] Port #20496 (#20610)
[FEATURE] Refactor SwapAxis operator. (#21024)