refactor: reduce code duplication in blas/base level 1 routines

PR-URL: #2517 Reviewed-by: Athan Reines <[email protected]>
stdlib-js · Jul 6, 2024 · 271f5d5 · 271f5d5 · stdlib-bot · Jul 6, 2024
1 parent 243ab4d
commit 271f5d5
Show file tree

Hide file tree

Showing 13 changed files with 87 additions and 645 deletions.
diff --git a/lib/node_modules/@stdlib/blas/base/daxpy/lib/daxpy.js b/lib/node_modules/@stdlib/blas/base/daxpy/lib/daxpy.js
@@ -18,9 +18,10 @@
 
 'use strict';
 
-// VARIABLES //
+// MODULES //
 
-var M = 4;
+var stride2offset = require( '@stdlib/strided/base/stride2offset' );
+var ndarray = require('./ndarray.js');
 
 
 // MAIN //
@@ -49,48 +50,12 @@ var M = 4;
 function daxpy( N, alpha, x, strideX, y, strideY ) {
 	var ix;
 	var iy;
-	var m;
-	var i;
 	if ( N <= 0 || alpha === 0.0 ) {
 		return y;
 	}
-	// Use unrolled loops if both strides are equal to `1`...
-	if ( strideX === 1 && strideY === 1 ) {
-		m = N % M;
-
-		// If we have a remainder, run a clean-up loop...
-		if ( m > 0 ) {
-			for ( i = 0; i < m; i++ ) {
-				y[ i ] += alpha * x[ i ];
-			}
-		}
-		if ( N < M ) {
-			return y;
-		}
-		for ( i = m; i < N; i += M ) {
-			y[ i ] += alpha * x[ i ];
-			y[ i+1 ] += alpha * x[ i+1 ];
-			y[ i+2 ] += alpha * x[ i+2 ];
-			y[ i+3 ] += alpha * x[ i+3 ];
-		}
-		return y;
-	}
-	if ( strideX < 0 ) {
-		ix = (1-N) * strideX;
-	} else {
-		ix = 0;
-	}
-	if ( strideY < 0 ) {
-		iy = (1-N) * strideY;
-	} else {
-		iy = 0;
-	}
-	for ( i = 0; i < N; i++ ) {
-		y[ iy ] += alpha * x[ ix ];
-		ix += strideX;
-		iy += strideY;
-	}
-	return y;
+	ix = stride2offset( N, strideX );
+	iy = stride2offset( N, strideY );
+	return ndarray( N, alpha, x, strideX, ix, y, strideY, iy );
 }
 
 

diff --git a/lib/node_modules/@stdlib/blas/base/ddot/lib/ddot.js b/lib/node_modules/@stdlib/blas/base/ddot/lib/ddot.js
@@ -18,9 +18,10 @@
 
 'use strict';
 
-// VARIABLES //
+// MODULES //
 
-var M = 5;
+var stride2offset = require( '@stdlib/strided/base/stride2offset' );
+var ndarray = require( './ndarray.js' );
 
 
 // MAIN //
@@ -45,50 +46,14 @@ var M = 5;
 * // returns -5.0
 */
 function ddot( N, x, strideX, y, strideY ) {
-	var dot;
 	var ix;
 	var iy;
-	var m;
-	var i;
-
-	dot = 0.0;
 	if ( N <= 0 ) {
-		return dot;
-	}
-	// Use unrolled loops if both strides are equal to `1`...
-	if ( strideX === 1 && strideY === 1 ) {
-		m = N % M;
-
-		// If we have a remainder, run a clean-up loop...
-		if ( m > 0 ) {
-			for ( i = 0; i < m; i++ ) {
-				dot += x[ i ] * y[ i ];
-			}
-		}
-		if ( N < M ) {
-			return dot;
-		}
-		for ( i = m; i < N; i += M ) {
-			dot += ( x[ i ] * y[ i ] ) + ( x[ i+1 ] * y[ i+1 ] ) + ( x[ i+2 ] * y[ i+2 ] ) + ( x[ i+3 ] * y[ i+3 ] ) + ( x[ i+4 ] * y[ i+4 ] ); // eslint-disable-line max-len
-		}
-		return dot;
-	}
-	if ( strideX < 0 ) {
-		ix = ( 1-N ) * strideX;
-	} else {
-		ix = 0;
-	}
-	if ( strideY < 0 ) {
-		iy = ( 1-N ) * strideY;
-	} else {
-		iy = 0;
-	}
-	for ( i = 0; i < N; i++ ) {
-		dot += ( x[ ix ] * y[ iy ] );
-		ix += strideX;
-		iy += strideY;
+		return 0.0;
 	}
-	return dot;
+	ix = stride2offset( N, strideX );
+	iy = stride2offset( N, strideY );
+	return ndarray( N, x, strideX, ix, y, strideY, iy );
 }
 
 

diff --git a/lib/node_modules/@stdlib/blas/base/drot/lib/drot.js b/lib/node_modules/@stdlib/blas/base/drot/lib/drot.js
@@ -18,6 +18,12 @@
 
 'use strict';
 
+// MODULES //
+
+var stride2offset = require( '@stdlib/strided/base/stride2offset' );
+var ndarray = require( './ndarray.js' );
+
+
 // MAIN //
 
 /**
@@ -43,42 +49,15 @@
 * // y => <Float64Array>[ ~4.2, 4.4, 4.6, 4.8, 5.0 ]
 */
 function drot( N, x, strideX, y, strideY, c, s ) {
-	var tmp;
 	var ix;
 	var iy;
-	var i;
 
 	if ( N <= 0 ) {
 		return y;
 	}
-	// If both strides are equal to `1`...
-	if ( strideX === 1 && strideY === 1 ) {
-		for ( i = 0; i < N; i++ ) {
-			tmp = ( c * x[ i ] ) + ( s * y[ i ] );
-			y[ i ] = ( c * y[ i ] ) - ( s * x[ i ] );
-			x[ i ] = tmp;
-		}
-		return y;
-	}
-	// If both strides are not equal to `1`...
-	if ( strideX < 0 ) {
-		ix = ( 1 - N ) * strideX;
-	} else {
-		ix = 0;
-	}
-	if ( strideY < 0 ) {
-		iy = ( 1 - N ) * strideY;
-	} else {
-		iy = 0;
-	}
-	for ( i = 0; i < N; i++ ) {
-		tmp = ( c * x[ ix ] ) + ( s * y[ iy ] );
-		y[ iy ] = ( c * y[ iy ] ) - ( s * x[ ix ] );
-		x[ ix ] = tmp;
-		ix += strideX;
-		iy += strideY;
-	}
-	return y;
+	ix = stride2offset( N, strideX );
+	iy = stride2offset( N, strideY );
+	return ndarray( N, x, strideX, ix, y, strideY, iy, c, s );
 }
 
 

diff --git a/lib/node_modules/@stdlib/blas/base/drotm/lib/drotm.js b/lib/node_modules/@stdlib/blas/base/drotm/lib/drotm.js
@@ -18,6 +18,12 @@
 
 'use strict';
 
+// MODULES //
+
+var stride2offset = require( '@stdlib/strided/base/stride2offset' );
+var ndarray = require( './ndarray.js' );
+
+
 // MAIN //
 
 /**
@@ -44,108 +50,16 @@
 */
 function drotm( N, x, strideX, y, strideY, param ) {
 	var dflag;
-	var dh11;
-	var dh12;
-	var dh21;
-	var dh22;
 	var ix;
 	var iy;
-	var i;
-	var w;
-	var z;
 
 	dflag = param[ 0 ];
 	if ( N <= 0 || dflag === -2.0 ) {
 		return y;
 	}
-	if ( strideX === strideY && strideX > 0 ) {
-		ix = 0;
-		if ( dflag < 0.0 ) {
-			dh11 = param[ 1 ];
-			dh12 = param[ 3 ];
-			dh21 = param[ 2 ];
-			dh22 = param[ 4 ];
-			for ( i = 0; i < N; i++ ) {
-				w = x[ ix ];
-				z = y[ ix ];
-				x[ ix ] = ( w * dh11 ) + ( z * dh12 );
-				y[ ix ] = ( w * dh21 ) + ( z * dh22 );
-				ix += strideX;
-			}
-			return y;
-		}
-		if ( dflag === 0.0 ) {
-			dh12 = param[ 3 ];
-			dh21 = param[ 2 ];
-			for ( i = 0; i < N; i++ ) {
-				w = x[ ix ];
-				z = y[ ix ];
-				x[ ix ] = w + ( z * dh12 );
-				y[ ix ] = ( w * dh21 ) + z;
-				ix += strideX;
-			}
-			return y;
-		}
-		dh11 = param[ 1 ];
-		dh22 = param[ 4 ];
-		for ( i = 0; i < N; i++ ) {
-			w = x[ ix ];
-			z = y[ ix ];
-			x[ ix ] = ( w * dh11 ) + z;
-			y[ ix ] = -w + ( z * dh22 );
-			ix += strideX;
-		}
-		return y;
-	}
-	if ( strideX < 0 ) {
-		ix = ( 1 - N ) * strideX;
-	} else {
-		ix = 0;
-	}
-	if ( strideY < 0 ) {
-		iy = ( 1 - N ) * strideY;
-	} else {
-		iy = 0;
-	}
-	if ( dflag < 0.0 ) {
-		dh11 = param[ 1 ];
-		dh12 = param[ 3 ];
-		dh21 = param[ 2 ];
-		dh22 = param[ 4 ];
-		for ( i = 0; i < N; i++ ) {
-			w = x[ ix ];
-			z = y[ iy ];
-			x[ ix ] = ( w * dh11 ) + ( z * dh12 );
-			y[ iy ] = ( w * dh21 ) + ( z * dh22 );
-			ix += strideX;
-			iy += strideY;
-		}
-		return y;
-	}
-	if ( dflag === 0.0 ) {
-		dh12 = param[ 3 ];
-		dh21 = param[ 2 ];
-		for ( i = 0; i < N; i++ ) {
-			w = x[ ix ];
-			z = y[ iy ];
-			x[ ix ] = w + ( z * dh12 );
-			y[ iy ] = ( w * dh21 ) + z;
-			ix += strideX;
-			iy += strideY;
-		}
-		return y;
-	}
-	dh11 = param[ 1 ];
-	dh22 = param[ 4 ];
-	for ( i = 0; i < N; i++ ) {
-		w = x[ ix ];
-		z = y[ iy ];
-		x[ ix ] = ( w * dh11 ) + z;
-		y[ iy ] = -w + ( z * dh22 );
-		ix += strideX;
-		iy += strideY;
-	}
-	return y;
+	ix = stride2offset( N, strideX );
+	iy = stride2offset( N, strideY );
+	return ndarray( N, x, strideX, ix, y, strideY, iy, param );
 }
 
 

diff --git a/lib/node_modules/@stdlib/blas/base/dsdot/lib/dsdot.js b/lib/node_modules/@stdlib/blas/base/dsdot/lib/dsdot.js
@@ -18,9 +18,10 @@
 
 'use strict';
 
-// VARIABLES //
+// MODULES //
 
-var M = 5;
+var stride2offset = require( '@stdlib/strided/base/stride2offset' );
+var ndarray = require( './ndarray.js' );
 
 
 // MAIN //
@@ -45,50 +46,14 @@ var M = 5;
 * // returns -5.0
 */
 function dsdot( N, x, strideX, y, strideY ) {
-	var dot;
 	var ix;
 	var iy;
-	var m;
-	var i;
-
-	dot = 0.0;
 	if ( N <= 0 ) {
-		return dot;
-	}
-	// Use unrolled loops if both strides are equal to `1`...
-	if ( strideX === 1 && strideY === 1 ) {
-		m = N % M;
-
-		// If we have a remainder, run a clean-up loop...
-		if ( m > 0 ) {
-			for ( i = 0; i < m; i++ ) {
-				dot += x[ i ] * y[ i ];
-			}
-		}
-		if ( N < M ) {
-			return dot;
-		}
-		for ( i = m; i < N; i += M ) {
-			dot += ( x[i]*y[i] ) + ( x[i+1]*y[i+1] ) + ( x[i+2]*y[i+2] ) + ( x[i+3]*y[i+3] ) + ( x[i+4]*y[i+4] ); // eslint-disable-line max-len
-		}
-		return dot;
-	}
-	if ( strideX < 0 ) {
-		ix = ( 1-N ) * strideX;
-	} else {
-		ix = 0;
-	}
-	if ( strideY < 0 ) {
-		iy = ( 1-N ) * strideY;
-	} else {
-		iy = 0;
-	}
-	for ( i = 0; i < N; i++ ) {
-		dot += x[ ix ] * y[ iy ];
-		ix += strideX;
-		iy += strideY;
+		return 0.0;
 	}
-	return dot;
+	ix = stride2offset( N, strideX );
+	iy = stride2offset( N, strideY );
+	return ndarray( N, x, strideX, ix, y, strideY, iy );
 }
Package	Statements	Branches	Functions	Lines
blas/base/daxpy	$\color{green}431/431$ $\color{green}+100.00\%$	$\color{green}30/30$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}431/431$ $\color{green}+100.00\%$
blas/base/ddot	$\color{green}419/419$ $\color{green}+100.00\%$	$\color{green}28/28$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}419/419$ $\color{green}+100.00\%$
blas/base/drot	$\color{green}409/409$ $\color{green}+100.00\%$	$\color{green}19/19$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}409/409$ $\color{green}+100.00\%$
blas/base/drotm	$\color{green}488/488$ $\color{green}+100.00\%$	$\color{green}38/38$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}488/488$ $\color{green}+100.00\%$
blas/base/dsdot	$\color{green}419/419$ $\color{green}+100.00\%$	$\color{green}28/28$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}419/419$ $\color{green}+100.00\%$
blas/base/dswap	$\color{green}440/440$ $\color{green}+100.00\%$	$\color{green}28/28$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}440/440$ $\color{green}+100.00\%$
blas/base/saxpy	$\color{green}436/436$ $\color{green}+100.00\%$	$\color{green}30/30$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}436/436$ $\color{green}+100.00\%$
blas/base/scopy	$\color{green}425/425$ $\color{green}+100.00\%$	$\color{green}29/29$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}425/425$ $\color{green}+100.00\%$
blas/base/sdot	$\color{green}395/395$ $\color{green}+100.00\%$	$\color{green}19/19$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}395/395$ $\color{green}+100.00\%$
blas/base/sdsdot	$\color{green}429/429$ $\color{green}+100.00\%$	$\color{green}28/28$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}429/429$ $\color{green}+100.00\%$
blas/base/srot	$\color{green}409/409$ $\color{green}+100.00\%$	$\color{green}19/19$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}409/409$ $\color{green}+100.00\%$
blas/base/srotm	$\color{green}493/493$ $\color{green}+100.00\%$	$\color{green}38/38$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}493/493$ $\color{green}+100.00\%$
blas/base/sswap	$\color{green}440/440$ $\color{green}+100.00\%$	$\color{green}28/28$ $\color{green}+100.00\%$	$\color{green}4/4$ $\color{green}+100.00\%$	$\color{green}440/440$ $\color{green}+100.00\%$