Code

Cadence fusiong3 operators m2#7490

Code

Merged

facebook-github-bot merged 15 commits intopytorch:mainpytorch/executorch:mainfrom ckmadhira:cadence_fusiong3_operators_M2ckmadhira/fusion_g3_port:cadence_fusiong3_operators_M2Copy head branch name to clipboardJan 10, 2025

+2,184-308

Commits

Commits on Jan 7, 2025

• Added Fusion G3 NN library with kernels related to add, mul, quantize, dequantize, cat, layer norm, softmax to backends/cadence folder. Added operators to backends/cadence folder

  

  [email protected]

  authored and

  [email protected]

  committed
  2221f8a

  Copy full SHA for 2221f8a

• Updated name space of the operators by appending cadence

  Show description for a41efc8
  

  [email protected]

  authored and

  [email protected]

  committed
  a41efc8

  Copy full SHA for a41efc8

• Added nnlib-FusionG3 submodule from FOSS-xtensa git space

  Show description for ce89bf1
  

  [email protected]

  authored and

  [email protected]

  committed
  ce89bf1

  Copy full SHA for ce89bf1

• Resolved Linter errors

  Show description for 9011471
  

  [email protected]

  authored and

  [email protected]

  committed
  9011471

  Copy full SHA for 9011471

• added operators for sub, slice, permute, exp, mean, div. Disabled argument checks in the operators using macro

  

  [email protected]

  committed
  5bdd7eb

  Copy full SHA for 5bdd7eb

• updated fusion g3 kernel git repo

  

  [email protected]

  committed
  0da167b

  Copy full SHA for 0da167b

• For div operator and for generic case, added get_compute type function to support integer inputs and float output

  

  [email protected]

  committed
  6802158

  Copy full SHA for 6802158

• added unary_ufunc_realhbbf16_to_floathbf16.cpp to the cmakelist in backends/cadence/fusion_g3/operators to solve a comilation error

  

  [email protected]

  committed
  318f558

  Copy full SHA for 318f558

• resolved linter errors

  

  [email protected]

  committed
  dd1d629

  Copy full SHA for dd1d629

• resolved linter errors

  

  [email protected]

  committed
  c964d27

  Copy full SHA for c964d27

• copied XT_KERNEL_CHECK to a header file and included the header file in the operators at backends\cadence\fusion_f3\operators folder

  

  [email protected]

  committed
  b626b7f

  Copy full SHA for b626b7f

• Removed conflicts in op_quantize.cpp

  

  [email protected]

  committed
  444958c

  Copy full SHA for 444958c

• Name space usage is changed in the operators

  

  [email protected]

  committed
  e79b466

  Copy full SHA for e79b466

Commits on Jan 8, 2025

• Ordered the header file inclusion. Used allocate_temp instead of malloc in mean operator

  

  [email protected]

  committed
  6ace5a3

  Copy full SHA for 6ace5a3

Commits on Jan 10, 2025

• resovled all name space issues. Changed Bits16 to Uint16 in quantize and dequantize operators. Reduced the size of scratch memory in mean operator

  

  [email protected]

  committed
  a0bdd97

  Copy full SHA for a0bdd97