Commit Graph

23 Commits

Author SHA1 Message Date
Stefan Weil
a61d7ac2ee Add / fix namespace tesseract for unittest
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2020-12-27 10:54:43 +01:00
Stefan Weil
92b6c652f3 Use std::vector for scales_
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2020-10-29 08:00:11 +01:00
Stefan Weil
c15dd26b84 Don't pass scales_ to IntSimdMatrix::Init
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2020-10-28 20:35:53 +01:00
Stefan Weil
fe76142a3d Remove GenericVector::scale() again
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2020-10-28 16:24:59 +01:00
Robin Watts
872816897a Rejig intsimdmatrix to reduce FP ops.
Avoid 1) floating point division by 127, 2) conversion of
bias to double, 3) FP addition, in favour of 1) integer
multiplication by 127, and 2) integer addition.

(Also costs extra work in the serialisation/deserialisation of
the scale values, and conversion of weights to int formats, but
these are all one offs).
2020-10-12 04:30:46 -07:00
Robin Watts
9dfdac51c6 Tweak scales array for intSimdMatrix case.
Currently, the size of the scales array is not rounded up
in the same way as the weights are. This blocks us pushing
the scale calculations into the SIMD, as when we "overread"
the end of the scale array, we potentially get errors.

Here, we adjust the intSimdMatrix stuff to ensure that the
scales array reserves enough entries to allow such overreads
to work.

This doesn't make any difference for now, but opens the way
for future optimisations.
2020-10-12 11:47:16 +01:00
Stefan Weil
dfdc2abef0 unittest: Improve logging for intsimdmatrix_test
Use GTEST_SKIP if AVX2 or SSE tests are skipped.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-11-28 17:51:37 +01:00
Stefan Weil
a1a139cbd2 Replace AVX_OPT, ..., AVX macros by HAVE_AVX, ... and clean related code
- Replace AVX_OPT, AVX2_OPT, FMA_OPT, SSE41_OPT
- Replace AVX, AVX2, FMA, SSE4_1
- Write new HAVE_AVX, HAVE_AVX2, HAVE_FMA, HAVE_SSE4_1 into config_auto.h
- Put related conditionals in Makefile.am in one place

This makes the code clearer and fixes a log message in
IntSimdMatrixTest.AVX2.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-11-28 17:51:37 +01:00
Stefan Weil
e3e7a9bf33 Use #include <tesseract/*.h> for unittest
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-10-29 18:01:18 +01:00
Stefan Weil
26ba7e2f81 Fix #include path of public headers for unittest
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-10-29 08:41:47 +01:00
Stefan Weil
8e7b1119b5 Run more unittests with the user's locale
Hopefully this improves the test coverage.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-05-16 18:12:55 +02:00
Stefan Weil
502bb624c2 More optimisations for IntSimdMatrix
* Move IntDotProductSSE. That allows inlining of the code.
* Improve IntDotProductSSE by moving some instructions.
* Remove unused num_input_groups_ from IntSimdMatrix.
* Re-order elements in IntSimdMatrix to avoid padding.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
95606398f5 Clean code for IntSimdMatrix
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
7fc7d28dd0 Compile files for AVX, AVX2 or SSE only when needed
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
a9a1035e55 Move IntSimdMatrixNative from IntSimdMatrix to unittest
It is only used for the unit test.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
605b4d66c7 Replace dynamically allocated IntSimdMatrix instances by constants
Two header files are no longer needed and could be removed.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
26be7c5d2e Use constructor with parameters for IntSimdMatrix
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
7c70147701 Move shaped weights from IntSimMatrix to WeightMatrix
Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 21:34:37 +01:00
Stefan Weil
a5283f293d Add test for the C++ implementation of MatrixDotVector
Check also whether the sum of all results matches the expected value.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2019-01-14 17:56:35 +01:00
Stefan Weil
7768f9b336 Clean more include files and include statements
The changes are based on an analysis done with include-what-you-use.

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2018-06-24 19:45:12 +02:00
Stefan Weil
023e1b340e Use POSIX data types and macros (#878)
* api: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccmain: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccstruct: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* classify: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* cutil: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* dict: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* textord: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* training: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* wordrec: Replace Tesseract data types by POSIX data types

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccutil: Replace Tesseract data types by POSIX data types

Now all Tesseract data types which are no longer needed can be removed
from ccutil/host.h.

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccmain: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccstruct: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* classify: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* dict: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* lstm: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* textord: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* wordrec: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* ccutil: Replace Tesseract's MIN_*INT, MAX_*INT* by POSIX *INT*_MIN, *INT*_MAX

Remove the macros which are now unused from ccutil/host.h.
Remove also the obsolete history comments.

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* Fix build error caused by ambiguous ClipToRange

Error message vom Appveyor CI:

    C:\projects\tesseract\ccstruct\coutln.cpp(818): error C2672: 'ClipToRange': no matching overloaded function found [C:\projects\tesseract\build\libtesseract.vcxproj]
    C:\projects\tesseract\ccstruct\coutln.cpp(818): error C2782: 'T ClipToRange(const T &,const T &,const T &)': template parameter 'T' is ambiguous [C:\projects\tesseract\build\libtesseract.vcxproj]
      c:\projects\tesseract\ccutil\helpers.h(122): note: see declaration of 'ClipToRange'
      C:\projects\tesseract\ccstruct\coutln.cpp(818): note: could be 'char'
      C:\projects\tesseract\ccstruct\coutln.cpp(818): note: or       'int'

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* unittest: Replace Tesseract's MAX_INT8 by POSIX INT8_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>

* arch: Replace Tesseract's MAX_INT8 by POSIX INT8_MAX

Signed-off-by: Stefan Weil <sw@weilnetz.de>
2018-03-13 21:36:30 +01:00
Ray Smith
3493785f7d Fixed apiexample and intsimdmatrix tests and prepared Makefile.am for more tests 2017-09-08 17:34:31 +01:00
Ray Smith
fc6a390c6c Added intsimdmatrix as a generic integer matrixdotvector function with AVX2 and SSE specializations 2017-09-08 15:06:19 +01:00