Code

Activation unit#1

Code

Merged

gamzeisl merged 210 commits intomainpulp-platform/ITA:mainfrom activation-unitpulp-platform/ITA:activation-unitCopy head branch name to clipboardSep 24, 2024

+1,888-327

Commits

Commits on Sep 24, 2024

• build(requirements): add pytorch

  gamzeisl

  committed
  c087fb8

• build(requirements): add pytest

  gamzeisl

  committed
  a17a7ee

• test(py-ita): add feedforward size parameter

  

  TimonFercho

  authored and

  gamzeisl

  committed
  840f9f7

• test(test-generator): add feedforward parameter and adjust output path

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5b273e4

• feat(py-ita): extend print_properties with feedforward size

  

  TimonFercho

  authored and

  gamzeisl

  committed
  43671fe

• feat(py-ita). implement floating point i_poly

  

  TimonFercho

  authored and

  gamzeisl

  committed
  cb9838d

• feat(py-ita): add i_poly_wrapper function

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ac44139

• build(requirements): add pytest-check

  

  TimonFercho

  authored and

  gamzeisl

  committed
  57a2520

• feat(py-ita): add symmetric (de-)quantization functions

  

  TimonFercho

  authored and

  gamzeisl

  committed
  6babe4b

• feat(py-ita): add functions for floating point poly and erf

  

  TimonFercho

  authored and

  gamzeisl

  committed
  963995b

• test(py-ita): add test for quantization, poly and erf functions

  

  TimonFercho

  authored and

  gamzeisl

  committed
  7718585

• fix(py-ita): fix q_c typo from definition

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c7b7cde

• feat(py-ita): implement floating point i_gelu

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9eacf08

• test(py-ita): add i_gelu tests

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1a89be3

• fix(py-ita): fix ipoly types to avoid overflow

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0362520

• refactor(py-ita): type i-gelu and i-erf

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2e2fd90

• test(py-ita): clean up and pretty print

  

  TimonFercho

  authored and

  gamzeisl

  committed
  f9701e2

• refactor(py-ita): fix typo

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1e460c6

• test(py-ita): make scaling factor output more readable

  

  TimonFercho

  authored and

  gamzeisl

  committed
  13d7e4e

• test(py-ita): verify i-gelu domain interval [b, -b]

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b8298a8

• test(py-ita): add error plotting

  

  TimonFercho

  authored and

  gamzeisl

  committed
  bd3d588

• refactor(py-ita): rename variable

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0555a27

• feat(py-ita): add missing imports

  

  TimonFercho

  authored and

  gamzeisl

  committed
  bd005fe

• feat(py-ita): add gelu golden model based on random preactivation values

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e5588a7

• feat(ita-tb): adjustments for feedforward parameter

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ab46f67

• feat(return-status): adjust parameters and check number of outputs

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2cd0b5f

• build(makefile): add feedforward parameter

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0c96449

• build(gitlab-ci): adjust for new parameters

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1ced88d

• fix(test-generator): typo in simvector path

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2651df8

• fix(test-generator): fix and refactor simvector paths

  

  TimonFercho

  authored and

  gamzeisl

  committed
  8680d6a

• feat(py-ita): write out gelu constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  51fd82c

• feat(gelu-tb): add testbench skeleton

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e8d1d5e

• feat(accel-pkg): add gelu bit width parameters

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9fd8fa8

• feat(gelu): add initial scalar gelu

  

  TimonFercho

  authored and

  gamzeisl

  committed
  505d9b2

• build(bender): add gelu and gelu testbench

  

  TimonFercho

  authored and

  gamzeisl

  committed
  d2be389

• feat(modelsim): add gelu simulation script

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b7874df

• build(modelsim): add gelu simulation target

  

  TimonFercho

  authored and

  gamzeisl

  committed
  112f69c

• test(gelu-tb): validate scalar gelu for all activations from precomputed stimuli files

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5d427ba

• chore(gitignore): ignore vscode settings.json

  

  TimonFercho

  authored and

  gamzeisl

  committed
  d53fa63

• fix(py-ita): clip i-gelu within valid range -127, 127

  

  TimonFercho

  authored and

  gamzeisl

  committed
  fd07bb6

• test(py-ita): add tests for i-gelu edge cases

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0a4ffbd

• refactor(gelu): explicitly sign-extend input to output bit width

  

  TimonFercho

  authored and

  gamzeisl

  committed
  56ec4fe

• fix(gelu): clip input within valid bounds

  

  TimonFercho

  authored and

  gamzeisl

  committed
  bd18157

• stlye(gelu): auto-format

  gamzeisl

  committed
  35c970b

• feat(py-ita): implement i-gelu with requantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4135e6f

• test(py-ita): add tests for i-gelu with requantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9df9fd0

• feat(py-ita): use 8 bit eps_mul for requantizer i-gelu

  

  TimonFercho

  authored and

  gamzeisl

  committed
  41ca067

• test(py-ita): auto-format

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ab2ce89

• feat(py-ita): write out i-gelu requantization constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  828d4c4

• test(gelu): read and apply gelu requantization constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a617613

• feat(gelu): implement requantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c50f82e

• test(gelu): remove redundant print statements

  

  TimonFercho

  authored and

  gamzeisl

  committed
  69dcae9

• refactor(gelu): rename types

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ce6b980

• refactor(gelu): use correct type for add

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4223674

• test(gelu): refactor to test higher-level activation module instead of just GELU

  gamzeisl

  committed
  ff027cd

• feat(relu): add RELU activation module

  

  TimonFercho

  authored and

  gamzeisl

  committed
  3e54818

• feat(activation): add module which lifts GELU and RELU to match output dimension of requantizer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  6e08608

• feat(accel-pkg): add enum to select activation function

  

  TimonFercho

  authored and

  gamzeisl

  committed
  812dc73

• build: adjust simulation and build files for activation testbench

  gamzeisl

  committed
  268196c

• test(activation): read activations in blocks of N_PE size

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5823ca1

• test(activation): also verify ReLU and identity activations

  

  TimonFercho

  authored and

  gamzeisl

  committed
  12c2581

• test(activation-tb): extract validation function to reduce redandancies

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1555fe8

• test(activation-tb): rename variables holding gelu constants for clarity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  00e92ef

• refactor(py-ita): rename files for GELU constants for clarity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c0c3177

• test(activation-tb): extract function which reads all GELU constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a78a55b

• test(activation-tb): fix equality check + refactor reading of GELU constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b96681b

• test(activation-tb): parametrize function for reading GELU constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  00b9986

• refactor(activation): reorder inputs

  

  TimonFercho

  authored and

  gamzeisl

  committed
  cfba15d

• feat(accel-pkg): extend controller interface with constants for GELU and activation mode

  

  TimonFercho

  authored and

  gamzeisl

  committed
  da4dc6c

• feat(ita): insert activation module in datapath between requantizer and output FIFO

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0010200

• test(ita-tb): supply ITA with additional activation-related control signals and constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5797220

• feat(py-ita): implement almost symmetric quantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e4a78fe

• test(py-ita): add simple test for almost symmetric quantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b9e0dcc

• test(py-ita): fix last test case of quantize

  

  TimonFercho

  authored and

  gamzeisl

  committed
  d07f5fe

• test(py-ita): use almost symmetric quantization instead of symmetric quantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  414c18c

• fix(ita): compute scaling factor based on almost symmetric quantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1265712

• fix(gelu): remove edge case treatment of q=-128, no longer necessary with almost symmetric quantization

  

  TimonFercho

  authored and

  gamzeisl

  committed
  04249e7

• refactor(activation): rename requant variables and extract typedefs

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e09fa65

• refactor(gelu): extract type for gelu output

  

  TimonFercho

  authored and

  gamzeisl

  committed
  728890f

• fix(py-ita): make sure to round before clipping and properly apply half-way rounding from zero

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0af5eb8

• feat(py-ita): ensure requantization constants are unsigned and compensate for eps_mul sign flip

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a3e4790

• test(py-ita): reduce error tolerances

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9236371

• feat(activation): apply requantization using existing vector module instead of inside GELU

  gamzeisl

  committed
  1cca3a8

• test(activation-tb): account for requantization latency

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2504a98

• refactor(activation): reorder condition for clarity

  gamzeisl

  committed
  550cc4b

• test(activation-tb): make sure to use the correct input file when expected values for RELU

  

  TimonFercho

  authored and

  gamzeisl

  committed
  94682d1

• refactor(fifo-controller): rename port for clarity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  7d41f51

• fix(fifo-controller): delay insertion by two cycles to account for activation requantizer latency

  

  TimonFercho

  authored and

  gamzeisl

  committed
  38a5ae9

• feat(activation): don't introduce unnecessary delay for RELU and IDENTITY

  gamzeisl

  committed
  fd9f510

• test(activation-tb): extend and refactor for requantized GELU

  

  TimonFercho

  authored and

  gamzeisl

  committed
  00a589b

• feat(fifo-controller): let fifo insertion condition depend on activation latency

  

  TimonFercho

  authored and

  gamzeisl

  committed
  99cf753

• refactor(ita): rename fifo controller port

  

  TimonFercho

  authored and

  gamzeisl

  committed
  68898bd

• refactor(relu): use typedef for clarity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  11b6a4e

• feat(py-ita): allow choosing activation function for step 6

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2566f6e

• feat(ita): delay activation signal to keep in sync with activation input

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ed5a9d6

• test(ita-tb): make value mismatch info more verbose

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0548c8c

• test(ita-tb): use RELU activation by default

  

  TimonFercho

  authored and

  gamzeisl

  committed
  abf0dbf

• feat(activation): fix activation latency to two cycles

  gamzeisl

  committed
  f3abd65

• fix(fifo-controller): account for fixed latency of activation module

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4e44cc2

• fix(ita): account for fixed latency of activation module

  

  TimonFercho

  authored and

  gamzeisl

  committed
  205e44e

• test(ita-tb): use GELU activation by default

  

  TimonFercho

  authored and

  gamzeisl

  committed
  131e871

• feat(py-ita): use GELU activation by default

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c8c13b1

• test(ita-tb): introduce activation parameter and set to identity by default

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1f38b33

• feat(py-ita): allow setting activation function using command line arguments

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4325925

• test(ita-tb): expose activation function as parameter

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2a57107

• build(makefile): allow configuring activation function

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5566813

• test(activation-tb): fix latency

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ed2e0e4

• refactor(accel_pkg): clean up typedefs

  

  TimonFercho

  authored and

  gamzeisl

  committed
  53b397c

• refactor(ita): remove unused first_inner_tile signal

  

  TimonFercho

  authored and

  gamzeisl

  committed
  856d400

• feat(activation): requantize relu output

  

  TimonFercho

  authored and

  gamzeisl

  committed
  f38296c

• feat(py-ita): requantize relu output

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1291b23

• test(activation-tb): read requantized relu output from generated file

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4453b2e

• refactor(test-ITA): move plot files into subdir

  

  TimonFercho

  authored and

  gamzeisl

  committed
  07104d0

• chore(gitignore): ignore pytest cache

  

  TimonFercho

  authored and

  gamzeisl

  committed
  fe22029

• test(activation-tb): rewrite debug msg

  

  TimonFercho

  authored and

  gamzeisl

  committed
  74d933c

• refactor(ita): rename gelu requant constants to more general activation requant constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b22f883

• build(requirements): set up pre-commit hooks for auto-formatting python files using yapf

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9fd5301

• style: auto-format files with yapf

  gamzeisl

  committed
  80c978b

• build(pre-commit-config): add some linters like checking for trailing whitespace and verifying python ast correctness

  

  TimonFercho

  authored and

  gamzeisl

  committed
  4d742d5

• style: fix trailing whitespace

  gamzeisl

  committed
  8727522

• test(py-ita): fix plot dir

  

  TimonFercho

  authored and

  gamzeisl

  committed
  cf8faca

• refactor(ita): remove unused signals

  

  TimonFercho

  authored and

  gamzeisl

  committed
  6d9a208

• refactor(accel-pkg): reorganize for clarity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1dbc91a

• feat(ita): add requantization controller which decouples state value from constants index using indirection

  gamzeisl

  committed
  aeae700

• refactor(requantizer): incorporate new requant_mode type

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9836319

• test(activation-tb): use new requant_mode type

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2d2e640

• refactor: make enums camel case

  

  TimonFercho

  authored and

  gamzeisl

  committed
  35225e2

• feat(accel-pkg): add feedforward layer type

  

  TimonFercho

  authored and

  gamzeisl

  committed
  92331b2

• refactor(py-ita): extract apply_activation function

  

  TimonFercho

  authored and

  gamzeisl

  committed
  7b92775

• refactor(py-ita): vectorize apply_activation

  

  TimonFercho

  authored and

  gamzeisl

  committed
  75fdc1f

• test(activation-tb): update file names of activation requant constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2f322c2

• feat(py-ita): update file names of activation requant constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a8d7db4

• test(ita-tb): rename files of activation and gelu constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1892a95

• test(ita-tb): extend testbench to run a single feedforward layer with GELU activation after attention layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5d97930

• feat(py-ita): generate testvectors for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  dae3de1

• feat(requantization-controller): reuse requantization constants at index 0 for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  f134049

• feat(accel-pkg): extend interface with layer mode and number of feedforward tiles

  

  TimonFercho

  authored and

  gamzeisl

  committed
  3169fad

• feat(controller): extend FSM for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  8d72902

• feat(py-ita): execute arbitrary activation in feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  75098f5

• test(ita-tb): allow executing arbitrary activation function for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  24d9a64

• feat(py-ita): add second stage with identity activation to feedforward layer according to transformer architecture

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e00531f

• test(ita-tb): execute second stage with identity activation in feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  b3d5532

• fix(ita): use correct imported function

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2b79d2a

• fix(py-ita): use instance field instead of removed local

  

  TimonFercho

  authored and

  gamzeisl

  committed
  80f7522

• fix(py-ita): call correct functions

  

  TimonFercho

  authored and

  gamzeisl

  committed
  aeaf93e

• refactor(makefile): remove duplicate variable definitions

  

  TimonFercho

  authored and

  gamzeisl

  committed
  aba51ef

• refactor(ita-tb): rename projection size to projection space

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ea19197

• fix(py-ita): use RQS constants at index 0 for 2nd FF layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  876b25c

• fix(ita-tb): add missing required finish_number argument for $fatal call

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5a89a08

• fix(hwpe-ita-tb): add missing required finish_number argument for $fatal call

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a5fbba8

• refactor(accel-pkg): rename layer typedef

  

  TimonFercho

  authored and

  gamzeisl

  committed
  cba4498

• refactor(accel-pkg): reorder ctrl fields

  

  TimonFercho

  authored and

  gamzeisl

  committed
  813bfbf

• test(ita-tb): use matching types

  

  TimonFercho

  authored and

  gamzeisl

  committed
  cf6a2e6

• refactor(ita-tb): rename activation constant variables

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9924f64

• refactor(ita-tb): rename index variable since it has no relationship to phase

  

  TimonFercho

  authored and

  gamzeisl

  committed
  3369e98

• fix(hwpe-ita-tb): fix simdir path

  

  TimonFercho

  authored and

  gamzeisl

  committed
  6f9df58

• fix(ita-package): update control engine to match control struct in accel_pkg

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0bdfeed

• test(hwpe-ita-tb): print correct base ptrs

  

  TimonFercho

  authored and

  gamzeisl

  committed
  129c3b7

• fix(hwpe-ita-tb): create arrays of correct size so that indexing with the step state does not go out-of-bounds when number of states changes

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5a1e5d6

• feat(ita-package): reorder control engine type

  

  TimonFercho

  authored and

  gamzeisl

  committed
  dac05b2

• refactor(ita-package): remove redundant definition of control engine structure

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1705931

• feat(ita-package): extend register mapping for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  e5a7cb2

• refactor(accel-pkg): explicitly state enum type to allow direct assignment without cast from implicit int

  

  TimonFercho

  authored and

  gamzeisl

  committed
  23e4179

• style(ita-package): formatting

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c005bad

• feat(hwpe-ita-tb): load constants and prepare registers for feedforward layer

  

  TimonFercho

  authored and

  gamzeisl

  committed
  a6ff617

• feat(ita-ctrl): pass constants and control signals for feedforward layer to control engine

  

  TimonFercho

  authored and

  gamzeisl

  committed
  43a0a37

• refactor(accel-pkg): explicitely compute N_STATES parameter

  

  TimonFercho

  authored and

  gamzeisl

  committed
  d0944c8

• refactor(py-ita): extract memfile constant

  

  TimonFercho

  authored and

  gamzeisl

  committed
  be561b9

• fix(py-ita): include all files in file list for hwpe

  

  TimonFercho

  authored and

  gamzeisl

  committed
  315d6ad

• perf(gelu): apply strength reduction to remove signed mult when computing gelu_erf

  

  TimonFercho

  authored and

  gamzeisl

  committed
  87577f5

• perf(gelu): reduce intermediate bitwidths

  

  TimonFercho

  authored and

  gamzeisl

  committed
  c7eebf1

• perf(gelu): use lower bitwidth for poly_d

  

  TimonFercho

  authored and

  gamzeisl

  committed
  1615d6b

• feat(py-ita): export mempool and snitch cluster on demand

  

  TimonFercho

  authored and

  gamzeisl

  committed
  17c1a21

• chore(gitlab-ci): use gelu activation for FF layer by default

  

  TimonFercho

  authored and

  gamzeisl

  committed
  283e749

• feat(activation): pipeline gelu

  gamzeisl

  committed
  26c55ff

• feat(accel-pkg): increase FIFO depth to account for gelu pipelining

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5a35b46

• test(activation-tb): adjust for increased latency

  

  TimonFercho

  authored and

  gamzeisl

  committed
  69ed6da

• test(ita-tb): fix input, weight and output timing

  

  TimonFercho

  authored and

  gamzeisl

  committed
  3eabd07

• test(activation-tb): fix latency for identity

  

  TimonFercho

  authored and

  gamzeisl

  committed
  467e6b7

• refactor(gelu): split up combinational block by stages

  

  TimonFercho

  authored and

  gamzeisl

  committed
  5dfc862

• perf(activation): reduce flop resources for RELU buffering by 70% by performing sign extension when reading out buffers

  

  TimonFercho

  authored and

  gamzeisl

  committed
  520cadd

• fix(ita): delay activation control signal until end of activation computations

  

  TimonFercho

  authored and

  gamzeisl

  committed
  2e23829

• perf(gelu): use calc_en signal to only compute during valid cycles

  

  TimonFercho

  authored and

  gamzeisl

  committed
  0ec18b9

• test(activation): add extra calc_en input

  

  TimonFercho

  authored and

  gamzeisl

  committed
  46120f2

• refactor(accel-pkg): removed unused fields in control_t

  

  TimonFercho

  authored and

  gamzeisl

  committed
  ea4d762

• test(ita-tb): don't reference unused control_t fields

  

  TimonFercho

  authored and

  gamzeisl

  committed
  866db19

• refactor(gelu): merge gelu one and gelu b constants

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9c0421f

• test(activation): removed unused signal

  

  TimonFercho

  authored and

  gamzeisl

  committed
  22686fe

• fix(py-ita): correctly compute L2 error

  gamzeisl

  committed
  b264e91

• refactor(ita-package): remove unused gelu one constant

  

  TimonFercho

  authored and

  gamzeisl

  committed
  9573935

• chore(accel-pkg): increase fifo depth

  

  TimonFercho

  authored and

  gamzeisl

  committed
  fcda25d

• build(ci): fix mismatch of generated testvectors

  

  TimonFercho

  authored and

  gamzeisl

  committed
  beb4cfc

• feat(return-status): add ff checks

  gamzeisl

  committed
  90e4d13

• change(hwpe-pkg): do not reuse regs for activations

  gamzeisl

  committed
  774ab10

• fix(ita_tb): lower input valid signal after handshake

  gamzeisl

  committed
  f4c85d3

• feat: add support for two layer ffn

  gamzeisl

  committed
  00555ef

• fix(PyITA): correct random vector gen for ffn

  gamzeisl

  committed
  29dce4c

• feat(PyITA): write hwpe files for ffn

  gamzeisl

  committed
  6b18f84

• feat(hwpe_tb): extend to test ffn

  gamzeisl

  committed
  e1941c5

• fix(PyITA): correct typecast in gelu

  gamzeisl

  committed
  82f56d9

• change(ci): add activation to hwpe sim

  gamzeisl

  committed
  a7a6e79

• feat(PyITA): add separate requantization params for ffn

  gamzeisl

  committed
  18bee9d

• feat(hw): add separate ffn requant params

  gamzeisl

  committed
  ac33aa5

• [PyITA] Move GELU functions

  gamzeisl

  committed
  2ab2114

• [ci] Add tests with relu

  gamzeisl

  committed
  7a22227

• [PyITA] Modify license headers

  gamzeisl

  committed
  8ebfda6

• Remove config yaml file

  gamzeisl

  committed
  f54f25c

• Add header

  gamzeisl

  committed
  dc5d411

• Fix python format

  gamzeisl

  committed
  b72dcbf

• Add relu test vectors to ci

  gamzeisl

  committed
  d5b724a