Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

feat: add missing kernels to ndarray/base/unary-reduce-subarray #6421

Merged
merged 15 commits into from
Apr 1, 2025
Original file line number Diff line number Diff line change
@@ -0,0 +1,394 @@
/**
* @license Apache-2.0
*
* Copyright (c) 2025 The Stdlib Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/

/* eslint-disable max-depth, max-len */

'use strict';

// MODULES //

var loopOrder = require( '@stdlib/ndarray/base/unary-loop-interchange-order' );
var blockSize = require( '@stdlib/ndarray/base/unary-tiling-block-size' );
var takeIndexed = require( '@stdlib/array/base/take-indexed' );
var zeros = require( '@stdlib/array/base/zeros' );
var incrementOffsets = require( './increment_offsets.js' );
var setViewOffsets = require( './set_view_offsets.js' );
var offsets = require( './offsets.js' );


// MAIN //

/**
* Performs a reduction over an input ndarray and assigns results to a provided output ndarray via loop blocking.
*
* @private
* @param {Function} fcn - reduction function
* @param {Array<Object>} arrays - ndarrays
* @param {Array<Object>} views - initialized ndarray-like objects representing sub-array views
* @param {IntegerArray} strides - loop dimension strides for the input ndarray
* @param {Options} opts - function options
* @returns {void}
*
* @example
* var toAccessorArray = require( '@stdlib/array/base/to-accessor-array' );
* var accessors = require( '@stdlib/array/base/accessors' );
* var Float64Array = require( '@stdlib/array/float64' );
* var filled = require( '@stdlib/array/base/filled' );
* var ndarray2array = require( '@stdlib/ndarray/base/to-array' );
* var base = require( '@stdlib/ndarray/base/every' );
*
* // Create data buffers:
* var xbuf = toAccessorArray( new Float64Array( [ 1.0, 2.0, 3.0, 4.0, 5.0, 0.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0 ] ) );
* var ybuf = toAccessorArray( filled( false, 3 ) );
*
* // Define the array shapes:
* var xsh = [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3, 2, 2 ];
* var ysh = [ 1, 1, 1, 1, 1, 1, 1, 1, 1, 3 ];
*
* // Define the array strides:
* var sx = [ 12, 12, 12, 12, 12, 12, 12, 12, 12, 4, 2, 1 ];
* var sy = [ 3, 3, 3, 3, 3, 3, 3, 3, 3, 1 ];
*
* // Define the index offsets:
* var ox = 0;
* var oy = 0;
*
* // Create an input ndarray-like object:
* var x = {
* 'dtype': 'float64',
* 'data': xbuf,
* 'shape': xsh,
* 'strides': sx,
* 'offset': ox,
* 'order': 'row-major',
* 'accessors': accessors( xbuf ).accessors
* };
*
* // Create an output ndarray-like object:
* var y = {
* 'dtype': 'generic',
* 'data': ybuf,
* 'shape': ysh,
* 'strides': sy,
* 'offset': oy,
* 'order': 'row-major',
* 'accessors': accessors( ybuf ).accessors
* };
*
* // Initialize ndarray-like objects representing sub-array views:
* var views = [
* {
* 'dtype': x.dtype,
* 'data': x.data,
* 'shape': [ 2, 2 ],
* 'strides': [ 2, 1 ],
* 'offset': x.offset,
* 'order': x.order
* }
* ];
*
* // Perform a reduction:
* blockedunary10d( base, [ x, y ], views, [ 12, 12, 12, 12, 12, 12, 12, 12, 12, 4 ], {} );
*
* var arr = ndarray2array( y.data, y.shape, y.strides, y.offset, y.order );
* // returns [ [ [ [ [ [ [ [ [ [ true, false, true ] ] ] ] ] ] ] ] ] ]
*/
function blockedunary10d( fcn, arrays, views, strides, opts ) { // eslint-disable-line max-statements, max-lines-per-function
var bsize;
var ybuf;
var set;
var dv0;
var dv1;
var dv2;
var dv3;
var dv4;
var dv5;
var dv6;
var dv7;
var dv8;
var dv9;
var ov1;
var ov2;
var ov3;
var ov4;
var ov5;
var ov6;
var ov7;
var ov8;
var ov9;
var sh;
var s0;
var s1;
var s2;
var s3;
var s4;
var s5;
var s6;
var s7;
var s8;
var s9;
var sv;
var ov;
var iv;
var i0;
var i1;
var i2;
var i3;
var i4;
var i5;
var i6;
var i7;
var i8;
var i9;
var j0;
var j1;
var j2;
var j3;
var j4;
var j5;
var j6;
var j7;
var j8;
var j9;
var N;
var x;
var y;
var o;
var k;

// Note on variable naming convention: S#, dv#, i#, j# where # corresponds to the loop number, with `0` being the innermost loop...

N = arrays.length;
x = arrays[ 0 ];
y = arrays[ 1 ];

// Resolve the loop interchange order:
o = loopOrder( y.shape, strides, y.strides );
sh = o.sh;
sv = [ o.sx, o.sy ];
for ( k = 2; k < N; k++ ) {
sv.push( takeIndexed( arrays[k].strides, o.idx ) );
}
// Determine the block size:
bsize = blockSize( x.dtype, y.dtype );

// Resolve a list of pointers to the first indexed elements in the respective ndarrays:
ov = offsets( arrays );

// Cache a reference to the output ndarray buffer:
ybuf = y.data;

// Cache accessors:
set = y.accessors[ 1 ];

// Cache offset increments for the innermost loop...
dv0 = [];
for ( k = 0; k < N; k++ ) {
dv0.push( sv[k][0] );
}
// Initialize loop variables...
ov1 = zeros( N );
ov2 = zeros( N );
ov3 = zeros( N );
ov4 = zeros( N );
ov5 = zeros( N );
ov6 = zeros( N );
ov7 = zeros( N );
ov8 = zeros( N );
ov9 = zeros( N );
dv1 = zeros( N );
dv2 = zeros( N );
dv3 = zeros( N );
dv4 = zeros( N );
dv5 = zeros( N );
dv6 = zeros( N );
dv7 = zeros( N );
dv8 = zeros( N );
dv9 = zeros( N );
iv = zeros( N );

// Iterate over blocks...
for ( j9 = sh[9]; j9 > 0; ) {
if ( j9 < bsize ) {
s9 = j9;
j9 = 0;
} else {
s9 = bsize;
j9 -= bsize;
}
for ( k = 0; k < N; k++ ) {
ov9[ k ] = ov[k] + ( j9*sv[k][9] );
}
for ( j8 = sh[8]; j8 > 0; ) {
if ( j8 < bsize ) {
s8 = j8;
j8 = 0;
} else {
s8 = bsize;
j8 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv9 = sv[k][9] - ( s8*sv[k][8] );
ov8[ k ] = ov9[k] + ( j8*sv[k][8] );
}
for ( j7 = sh[7]; j7 > 0; ) {
if ( j7 < bsize ) {
s7 = j7;
j7 = 0;
} else {
s7 = bsize;
j7 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv8 = sv[k][8] - ( s7*sv[k][7] );
ov7[ k ] = ov8[k] + ( j7*sv[k][7] );
}
for ( j6 = sh[6]; j6 > 0; ) {
if ( j6 < bsize ) {
s6 = j6;
j6 = 0;
} else {
s6 = bsize;
j6 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv7 = sv[k][7] - ( s6*sv[k][6] );
ov6[ k ] = ov7[k] + ( j6*sv[k][6] );
}
for ( j5 = sh[5]; j5 > 0; ) {
if ( j5 < bsize ) {
s5 = j5;
j5 = 0;
} else {
s5 = bsize;
j5 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv6 = sv[k][6] - ( s5*sv[k][5] );
ov5[ k ] = ov6[k] + ( j5*sv[k][5] );
}
for ( j4 = sh[4]; j4 > 0; ) {
if ( j4 < bsize ) {
s4 = j4;
j4 = 0;
} else {
s4 = bsize;
j4 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv5 = sv[k][5] - ( s4*sv[k][4] );
ov4[ k ] = ov5[k] + ( j4*sv[k][4] );
}
for ( j3 = sh[3]; j3 > 0; ) {
if ( j3 < bsize ) {
s3 = j3;
j3 = 0;
} else {
s3 = bsize;
j3 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv4[ k ] = sv[k][4] - ( s3*sv[k][3] );
ov3[ k ] = ov4[k] + ( j3*sv[k][3] );
}
for ( j2 = sh[2]; j2 > 0; ) {
if ( j2 < bsize ) {
s2 = j2;
j2 = 0;
} else {
s2 = bsize;
j2 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv3[ k ] = sv[k][3] - ( s2*sv[k][2] );
ov2[ k ] = ov3[k] + ( j2*sv[k][2] );
}
for ( j1 = sh[1]; j1 > 0; ) {
if ( j1 < bsize ) {
s1 = j1;
j1 = 0;
} else {
s1 = bsize;
j1 -= bsize;
}
for ( k = 0; k < N; k++ ) {
dv2[ k ] = sv[k][2] - ( s1*sv[k][1] );
ov1[ k ] = ov2[k] + ( j1*sv[k][1] );
}
for ( j0 = sh[0]; j0 > 0; ) {
if ( j0 < bsize ) {
s0 = j0;
j0 = 0;
} else {
s0 = bsize;
j0 -= bsize;
}
// Compute index offsets and loop offset increments for the first ndarray elements in the current block...
for ( k = 0; k < N; k++ ) {
iv[ k ] = ov1[k] + ( j0*sv[k][0] );
dv1[ k ] = sv[k][1] - ( s0*sv[k][0] );
}
// Iterate over the non-reduced ndarray dimensions...
for ( i9 = 0; i9 < s9; i9++ ) {
for ( i8 = 0; i8 < s8; i8++ ) {
for ( i7 = 0; i7 < s7; i7++ ) {
for ( i6 = 0; i6 < s6; i6++ ) {
for ( i5 = 0; i5 < s5; i5++ ) {
for ( i4 = 0; i4 < s4; i4++ ) {
for ( i3 = 0; i3 < s3; i3++ ) {
for ( i2 = 0; i2 < s2; i2++ ) {
for ( i1 = 0; i1 < s1; i1++ ) {
for ( i0 = 0; i0 < s0; i0++ ) {
setViewOffsets( views, iv );
set( ybuf, iv[ 1 ], fcn( views, opts ) );
incrementOffsets( iv, dv0 );
}
incrementOffsets( iv, dv1 );
}
incrementOffsets( iv, dv2 );
}
incrementOffsets( iv, dv3 );
}
incrementOffsets( iv, dv4 );
}
incrementOffsets( iv, dv5 );
}
incrementOffsets( iv, dv6 );
}
incrementOffsets( iv, dv7 );
}
incrementOffsets( iv, dv8 );
}
incrementOffsets( iv, dv9 );
}
}
}
}
}
}
}
}
}
}
}
}


// EXPORTS //

module.exports = blockedunary10d;
Loading