Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 8, 2024

• added 'count(for .... {predicate})' comprehension

  vpisarev

  committed
  786b446

  Copy full SHA for 786b446

Commits on Oct 17, 2024

• * slightly updated onnx parser (started adding support for separate files with weights) and onnx sample (dump list of used operations)

  Show description for beb72f3
  vpisarev

  committed
  beb72f3

  Copy full SHA for beb72f3

Commits on Sep 6, 2024

• Update documentation with Source Serif 4 font (#24)

  Show description for 6f79438
  dmitrys99

  authored
  6f79438

  Copy full SHA for 6f79438

Commits on Aug 19, 2023

• fixed wrong symbol

  vpisarev

  committed
  3c9a8b7

  Copy full SHA for 3c9a8b7

• * added complex numbers: "val c = complex(re, im)"

  Show description for aed544e
  vpisarev

  committed
  aed544e

  Copy full SHA for aed544e

Commits on Aug 2, 2023

• mention the new long type

  vpisarev

  committed
  154a602

  Copy full SHA for 154a602

Commits on Jul 30, 2023

• * undone using -ffast-math for now: compiler crashes for some reason.

  Show description for 43e86b3
  vpisarev

  committed
  43e86b3

  Copy full SHA for 43e86b3

• * various little fixes and optimizations

  Show description for a8fb70c
  vpisarev

  committed
  a8fb70c

  Copy full SHA for a8fb70c

Commits on Jun 4, 2023

• * made a couple of fixes in compiler:

  Show description for c33950b
  vpisarev

  committed
  c33950b

  Copy full SHA for c33950b

Commits on May 16, 2023

• a few more fixes

  vpisarev

  committed
  c81a1cc

  Copy full SHA for c81a1cc

• fixed a couple of bugs in long's sub & mul

  vpisarev

  committed
  94bf47d

  Copy full SHA for 94bf47d

Commits on May 15, 2023

• updated precompiled ficus compiler

  vpisarev

  committed
  11034d4

  Copy full SHA for 11034d4

• * added preliminary support for long (aka bigint) at runtime level:

  Show description for 55b421e
  vpisarev

  committed
  55b421e

  Copy full SHA for 55b421e

Commits on May 13, 2023

• added Date.fx, the module to handle dates & calendars, example and tests.

  vpisarev

  committed
  4dd2db3

  Copy full SHA for 4dd2db3

Commits on May 5, 2023

• updated compiler sources to recognize "long"

  vpisarev

  committed
  617c3bf

  Copy full SHA for 617c3bf

• started adding "long": arbitrary-precision integer type. Since now on "L" suffix can only be used to denote "long" literals. For 64-bit signed/unsigned integers "i64"/"u64" should be used.

  vpisarev

  committed
  309f040

  Copy full SHA for 309f040

Commits on Dec 23, 2022

• Integration with Loops. Depthwise convolution. (#21)

  Show description for 45a1b98
  4ekmah

  authored
  45a1b98

  Copy full SHA for 45a1b98

Commits on Oct 20, 2022

• added some more include directories

  vpisarev

  committed
  bdd14eb

  Copy full SHA for bdd14eb

Commits on Oct 9, 2022

• fixed compile errors on Linux

  vpisarev

  committed
  dcd06a5

  Copy full SHA for dcd06a5

• print separator between iterations in the case of trace or detailed_profile options

  vpisarev

  committed
  247eac3

  Copy full SHA for 247eac3

• added detailed profile option

  vpisarev

  committed
  a43e752

  Copy full SHA for a43e752

Commits on Oct 8, 2022

• Fix the bug when group > 1. (#20)

  zihaomu

  authored
  8b6952e

  Copy full SHA for 8b6952e

Commits on Oct 7, 2022

• ok, with NC4HW4 layout Ficus NN processes correctly resnet50-v1-12.onnx, mobilenetv2-12-int8.onnx, efficientnet-lite4-11-int8.onnx and googlenet-12-int8.onnx. Detection models are to be added later…

  Show description for 2a930a3
  vpisarev

  committed
  2a930a3

  Copy full SHA for 2a930a3

Commits on Oct 6, 2022

• block layout (NC4HW4) for 8-bit inference is partially implemented.

  vpisarev

  committed
  90e0a87

  Copy full SHA for 90e0a87

Commits on Sep 14, 2022

• hopefully, improved 1x1 convolution performance (need more testing)

  vpisarev

  committed
  0ab4c21

  Copy full SHA for 0ab4c21

• accelerated depthwise convolution (FP32 branch); Efficient net: 18ms => 13ms.

  vpisarev

  committed
  18444bf

  Copy full SHA for 18444bf

Commits on Sep 3, 2022

• accelerated depthwise convolution by a lot (FP16 for now; FP32 - TBD). It's especially noticeable with efficientnet-lite: 14ms => 8.3ms (FP16)

  vpisarev

  committed
  4369cb5

  Copy full SHA for 4369cb5

• faster but probably less accurate FP16 path in im2col convolution

  vpisarev

  committed
  dc1b6f3

  Copy full SHA for dc1b6f3

• fixed compile bug

  vpisarev

  committed
  89031d0

  Copy full SHA for 89031d0

• temporarily (or not) reverted im2col-conv to the previous version

  vpisarev

  committed
  f3018eb

  Copy full SHA for f3018eb

• improved cache utilization in the latest version of im2col-based convolution

  vpisarev

  committed
  639c09d

  Copy full SHA for 639c09d

Commits on Sep 2, 2022

• alternative implementation of im2row-based convolution

  vpisarev

  committed
  f8813c1

  Copy full SHA for f8813c1

• * made Winograd initialization loop parallel, since it may take a while.

  Show description for bb99678
  vpisarev

  committed
  bb99678

  Copy full SHA for bb99678

Commits on Sep 1, 2022

• quick attempt to further reduce yolov4 running time by parallelizing add

  vpisarev

  committed
  d4ff635

  Copy full SHA for d4ff635

• fixed compile errors with GCC

  vpisarev

  committed
  95b5ccd

  Copy full SHA for 95b5ccd

Pagination

Previous

Next