Low bit support for ARM Cortex-M export

• Add aidge types (type not compressed & type compressed)
  • int4 (. . . . v v v v) & dual_int4 (v v v v v v v v)
  • int3 (. . . . . v v v) & dual_int3 (. v v v . v v v)
  • int2 & quad_int2
  • uint4 & dual_uint4
  • uint3 & dual_uint3
  • uint2 & quad_uint2
  • binary & octo_binary
• Each aidge type uses a std::int8_t or std::uint8_t (for aide tensor, tensor binding to numpy,...)
• Add operator WeightInterleaving that compact tensor data on the last dimension of the Tensor. It requires input tensor format NHWC because the implemented kernels in ARM Cortex-M are HWC.
• Python binding of Operator WeightInterleaving and ApplyWeightInterleaving recipe
• Adapt data_convertion to a generic function taking a data conversionmap as input and integrate it into nodeExport. Each nodeExport instanciation in the export operator registry can specify a different data conversion map (from aidge to export type).

Closes : aidge_export_arm_cortexm#22 (closed)

assigned to @thibaultallenet

changed target branch from main to dev

changed title from Draft: Low bit support form ARM Cortex-M export to Draft: Low bit support for ARM Cortex-M export

added StatusWork in Progress label

changed the description

marked the checklist item Add aidge types int4, int3, int2, uint4, uint3, uint2 as completed

marked the checklist item Each aidge type use a std::int8_t or std::uint8_t (for aide tensor, tensor binding to numpy,...) as completed

marked the checklist item Operator WeightInterleaving that compact tensor data on the last dimension of the Tensor. It requires input tensor format NHWC because the implemented kernels in ARM Cortex-M are HWC. as completed

added 111 commits

• cfcf64fe...41e3501d - 103 commits from branch dev
• f4d3d101 - First version of compactData for custom datatype in export arm
• b60ad289 - Update pybind_core with compact data binding
• e5fa197b - Add type aidge::int4 with tensor implementation int8_t
• 57592ed7 - Remove Compact data from utils - functions are integrated into WeightInterleaving Operator
• b0cff912 - Add WeightInterleaving Operator
• 61b9a632 - Add support of UInt4, Int3, UInt3, Int2, UInt2
• f5c02edf - Export when Aidge Datatype isn't convertible into cpp type then set cpp export type to None
• 8dba3c09 - Add ImplecSpec constructor binding for export ARM

Compare with previous version

added 70 commits

• 8dba3c09...e7869ad3 - 55 commits from branch dev
• e7869ad3...c5f707cb - 5 earlier commits
• b43080da - Add support of UInt4, Int3, UInt3, Int2, UInt2
• 9f24fe6e - Export when Aidge Datatype isn't convertible into cpp type then set cpp export type to None
• cfcf64fe - Add ImplecSpec constructor binding for export ARM
• 793f92e1 - Merge branch 'dev' into low_bit_support_arm
• 69c15912 - Merge branch 'low_bit_support_arm' of...
• 3013b54b - Merge branch 'dev' into low_bit_support_arm
• 170c5a23 - Added possibility to create a GenericOp from any Operator
• 653773d5 - Merge branch 'dev' into low_bit_support_arm
• 3eafaa31 - Merge branch 'genericop_from_op' into low_bit_support_arm
• 54630773 - Add datatypes for compacted low bits integers

Compare with previous version

added 12 commits

• 54630773...a677e993 - 2 earlier commits
• 6a677044 - Merge branch 'genericop_from_op' into low_bit_support_arm
• b5ed3f15 - Fixed unsupported std::void_t
• e7a13ff5 - Merge branch 'genericop_from_op' into low_bit_support_arm
• 5fd06284 - Added missing hash specialization for std::pair
• 7050788e - Add integer datatypes for weightInterleaving
• 764c7f08 - Merge branch 'genericop_from_op' into low_bit_support_arm
• 15bbe6ea - Add set_dataformat binding for operator
• 006e8f4b - Add python binding for WeightInterleaving Operator
• 3ab0a780 - Add recipe apply weightInterleaving
• 309ab5a7 - Add binding for applyWeightInterleaving

Compare with previous version

added 1 commit

• 88558942 - Update applyWeightInterleaving recipe to handle 2D weight tensor (FC)

Compare with previous version

requested review from @cmoineau and @olivierbichler

added 36 commits

• 88558942...10143f53 - 34 commits from branch dev
• e7b038f2 - Merge branch 'dev' into low_bit_support_arm
• 651b8db0 - Upgrade the data conversion to a generic function and integrate it to nodeExport class constructor

Compare with previous version

added 3 commits

• 651b8db0...c312252d - 2 commits from branch dev
• 18c36ba4 - Merge branch 'dev' into low_bit_support_arm

Compare with previous version

marked this merge request as ready

changed the description

changed the description

removed StatusWork in Progress label