Java tutorial
/* * Licensed to the Apache Software Foundation (ASF) under one * or more contributor license agreements. See the NOTICE file * distributed with this work for additional information * regarding copyright ownership. The ASF licenses this file * to you 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. */ package org.apache.sysml.runtime.codegen; import java.util.Arrays; import org.apache.commons.math3.util.FastMath; import org.apache.sysml.runtime.functionobjects.BitwAnd; import org.apache.sysml.runtime.functionobjects.IntegerDivide; import org.apache.sysml.runtime.functionobjects.Modulus; import org.apache.sysml.runtime.matrix.data.LibMatrixDNN; import org.apache.sysml.runtime.matrix.data.LibMatrixMult; /** * This library contains all vector primitives that are used in * generated source code for fused operators. For primitives that * exist in LibMatrixMult, these calls are simply forwarded to * ensure consistency in performance and result correctness. * */ public class LibSpoofPrimitives { private static IntegerDivide intDiv = IntegerDivide.getFnObject(); private static Modulus mod = Modulus.getFnObject(); private static BitwAnd bwAnd = BitwAnd.getBitwAndFnObject(); //global pool of reusable vectors, individual operations set up their own thread-local //ring buffers of reusable vectors with specific number of vectors and vector sizes private static ThreadLocal<VectorBuffer> memPool = new ThreadLocal<VectorBuffer>() { @Override protected VectorBuffer initialValue() { return new VectorBuffer(0, 0, 0); } }; // forwarded calls to LibMatrixMult public static double dotProduct(double[] a, double[] b, int ai, int bi, int len) { if (a == null || b == null) return 0; return LibMatrixMult.dotProduct(a, b, ai, bi, len); } public static double dotProduct(double[] a, double[] b, int[] aix, int ai, int bi, int len) { if (a == null || b == null) return 0; return LibMatrixMult.dotProduct(a, b, aix, ai, bi, len); } public static double[] vectMatrixMult(double[] a, double[] b, int ai, int bi, int len) { //note: assumption b is already transposed for efficient dot products int m2clen = b.length / len; double[] c = allocVector(m2clen, false); for (int j = 0, bix = bi; j < m2clen; j++, bix += len) c[j] = LibMatrixMult.dotProduct(a, b, ai, bix, len); return c; } public static double[] vectMatrixMult(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { //note: assumption b is already transposed for efficient dot products int m2clen = b.length / len; double[] c = allocVector(m2clen, false); for (int j = 0, bix = bi; j < m2clen; j++, bix += len) c[j] = LibMatrixMult.dotProduct(a, b, aix, ai, bix, alen); return c; } public static void vectOuterMultAdd(double[] a, double[] b, double[] c, int ai, int bi, int ci, int len1, int len2) { if (isFlipOuter(len1, len2)) { for (int i = 0, cix = ci; i < len2; i++, cix += len1) { final double val = b[bi + i]; if (val != 0) LibMatrixMult.vectMultiplyAdd(val, a, c, ai, cix, len1); } } else { //rest, not aligned to 4-blocks final int bn = len1 % 4; for (int i = 0, cix = ci; i < bn; i++, cix += len2) if (a[ai + i] != 0) LibMatrixMult.vectMultiplyAdd(a[ai + i], b, c, bi, cix, len2); //unrolled 4-block (for fewer L1-dcache loads) for (int i = bn, cix = ci + bn * len2; i < len1; i += 4, cix += 4 * len2) { final int cix1 = cix, cix2 = cix + len2, cix3 = cix + 2 * len2, cix4 = cix + 3 * len2; final double aval1 = a[ai + i], aval2 = a[ai + i + 1], aval3 = a[ai + i + 2], aval4 = a[ai + i + 3]; for (int j = 0; j < len2; j++) { final double bval = b[bi + j]; c[cix1 + j] += aval1 * bval; c[cix2 + j] += aval2 * bval; c[cix3 + j] += aval3 * bval; c[cix4 + j] += aval4 * bval; } } } } public static void vectOuterMultAdd(double[] a, double[] b, double[] c, int[] aix, int ai, int bi, int ci, int alen, int len1, int len2) { if (isFlipOuter(len1, len2)) { for (int i = 0, cix = ci; i < len2; i++, cix += len1) { final double val = b[bi + i]; if (val != 0) LibMatrixMult.vectMultiplyAdd(val, a, c, aix, ai, cix, alen); } } else { for (int i = 0; i < alen; i++) LibMatrixMult.vectMultiplyAdd(a[ai + i], b, c, bi, ci + aix[ai + i] * len2, len2); } } public static void vectOuterMultAdd(double[] a, double[] b, double[] c, int ai, int[] bix, int bi, int ci, int blen, int len1, int len2) { if (isFlipOuter(len1, len2)) { for (int i = bi; i < bi + blen; i++) { final int cix = ci + bix[i] * len1; LibMatrixMult.vectMultiplyAdd(b[i], a, c, ai, cix, len1); } } else { for (int i = 0, cix = ci; i < len1; i++, cix += len2) LibMatrixMult.vectMultiplyAdd(a[ai + i], b, c, bix, bi, cix, blen); } } public static void vectMultAdd(double[] a, double bval, double[] c, int bi, int ci, int len) { if (a == null || bval == 0) return; LibMatrixMult.vectMultiplyAdd(bval, a, c, bi, ci, len); } public static void vectMultAdd(double bval, double[] a, double[] c, int bi, int ci, int len) { if (a == null || bval == 0) return; LibMatrixMult.vectMultiplyAdd(bval, a, c, bi, ci, len); } public static void vectMultAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (a == null || bval == 0) return; LibMatrixMult.vectMultiplyAdd(bval, a, c, aix, ai, ci, alen); } public static void vectMultAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (a == null || bval == 0) return; LibMatrixMult.vectMultiplyAdd(bval, a, c, aix, ai, ci, alen); } public static void vectMultAdd(double[] a, double[] b, double[] c, int bi, int ci, int len) { if (a == null || b == null) return; double[] tmp = vectMultWrite(a, b, 0, bi, len); LibMatrixMult.vectAdd(tmp, c, 0, ci, len); } public static double[] vectMultWrite(double[] a, double bval, int bi, int len) { if (a == null || bval == 0) return allocVector(len, true); double[] c = allocVector(len, false); LibMatrixMult.vectMultiplyWrite(bval, a, c, bi, 0, len); return c; } public static double[] vectMultWrite(double bval, double[] a, int bi, int len) { return vectMultWrite(a, bval, bi, len); } public static double[] vectMultWrite(double[] a, double[] b, int ai, int bi, int len) { if (a == null || b == null) return allocVector(len, true); double[] c = allocVector(len, false); LibMatrixMult.vectMultiplyWrite(a, b, c, ai, bi, 0, len); return c; } public static double[] vectMultWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); if (a == null) return c; LibMatrixMult.vectMultiplyAdd(bval, a, c, aix, ai, 0, alen); return c; } public static double[] vectMultWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectMultWrite(a, bval, aix, ai, alen, len); } public static double[] vectMultWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, true); if (a == null || b == null) return c; for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] * b[bi + aix[j]]; return c; } public static double[] vectMultWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectMultWrite(b, a, bix, ai, bi, blen, len); } public static void vectWrite(double[] a, double[] c, int ci, int len) { if (a == null) return; System.arraycopy(a, 0, c, ci, len); } public static void vectWrite(double[] a, double[] c, int ai, int ci, int len) { if (a == null) return; System.arraycopy(a, ai, c, ci, len); } public static void vectWrite(boolean[] a, boolean[] c, int[] aix) { if (a == null) return; for (int i = 0; i < aix.length; i++) c[aix[i]] = a[i]; } public static void vectWrite(boolean[] a, boolean[] c, int[] aix, int ai, int ci, int alen) { if (a == null) return; for (int i = ai; i < ai + alen; i++) c[ci + aix[i]] = a[i]; } // cbind handling public static double[] vectCbindAdd(double[] a, double b, double[] c, int ai, int ci, int len) { LibMatrixMult.vectAdd(a, c, ai, ci, len); c[ci + len] += b; return c; } public static double[] vectCbindAdd(double[] a, double b, double[] c, int[] aix, int ai, int ci, int alen, int len) { LibMatrixMult.vectAdd(a, c, aix, ai, ci, alen); c[ci + len] += b; return c; } public static double[] vectCbindWrite(double a, double b) { double[] c = allocVector(2, false); c[0] = a; c[1] = b; return c; } public static double[] vectCbindWrite(double[] a, double b, int aix, int len) { double[] c = allocVector(len + 1, false); System.arraycopy(a, aix, c, 0, len); c[len] = b; return c; } public static double[] vectCbindWrite(double[] a, double b, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len + 1, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j]; c[len] = b; return c; } // custom vector sums, mins, maxs /** * Computes c = sum(A), where A is a dense vectors. * * @param a dense input vector A * @param ai start position in A * @param len number of processed elements * @return sum value */ public static double vectSum(double[] a, int ai, int len) { double val = 0; final int bn = len % 8; //compute rest for (int i = ai; i < ai + bn; i++) val += a[i]; //unrolled 8-block (for better instruction-level parallelism) for (int i = ai + bn; i < ai + len; i += 8) { //read 64B cacheline of a, compute cval' = sum(a) + cval val += a[i + 0] + a[i + 1] + a[i + 2] + a[i + 3] + a[i + 4] + a[i + 5] + a[i + 6] + a[i + 7]; } //scalar result return val; } public static double vectSum(double[] avals, int[] aix, int ai, int alen, int len) { //forward to dense as column indexes not required here return vectSum(avals, ai, alen); } public static double vectSumsq(double[] a, int ai, int len) { return LibMatrixMult.dotProduct(a, a, ai, ai, len); } public static double vectSumsq(double[] avals, int[] aix, int ai, int alen, int len) { return LibMatrixMult.dotProduct(avals, avals, ai, ai, alen); } public static double vectMin(double[] a, int ai, int len) { double val = Double.POSITIVE_INFINITY; for (int i = ai; i < ai + len; i++) val = Math.min(a[i], val); return val; } public static double vectMin(double[] avals, int[] aix, int ai, int alen, int len) { double val = vectMin(avals, ai, alen); return (alen < len) ? Math.min(val, 0) : val; } public static double vectMax(double[] a, int ai, int len) { double val = Double.NEGATIVE_INFINITY; for (int i = ai; i < ai + len; i++) val = Math.max(a[i], val); return val; } public static double vectMax(double[] avals, int[] aix, int ai, int alen, int len) { double val = vectMax(avals, ai, alen); return (alen < len) ? Math.max(val, 0) : val; } public static double vectCountnnz(double[] a, int ai, int len) { int count = 0; for (int i = ai; i < ai + len; i++) count += (a[i] != 0) ? 1 : 0; return count; } public static double vectCountnnz(double[] avals, int[] aix, int ai, int alen, int len) { //pure meta data operation return alen; } //custom vector div public static void vectDivAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += a[j] / bval; } public static void vectDivAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += bval / a[j]; } public static void vectDivAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j] / bval; } public static void vectDivAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += bval / a[j]; } public static double[] vectDivWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] / bval; return c; } public static double[] vectDivWrite(double bval, double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = bval / a[ai + j]; return c; } public static double[] vectDivWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] / b[bi + j]; return c; } public static double[] vectDivWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval != 0) ? 0 : Double.NaN; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] / bval; return c; } public static double[] vectDivWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { double init = (bval != 0) ? Double.POSITIVE_INFINITY : Double.NaN; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = bval / a[j]; return c; } public static double[] vectDivWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) if (b[bi + j] == 0) //prep 0/0=NaN c[j] = Double.NaN; for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] / b[bi + aix[j]]; return c; } public static double[] vectDivWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) { double aval = a[bi + j]; c[j] = (aval == 0) ? Double.NaN : (aval > 0) ? Double.POSITIVE_INFINITY : Double.NEGATIVE_INFINITY; } for (int j = bi; j < bi + blen; j++) c[bix[j]] = a[ai + bix[j]] / b[j]; return c; } //custom vector minus public static void vectMinusAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += a[j] - bval; } public static void vectMinusAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += bval - a[j]; } public static void vectMinusAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval != 0) //subtract bval if necessary for (int j = ci; j < ci + len; j++) c[j] -= bval; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j]; } public static void vectMinusAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval != 0) //add bval is necessary for (int j = ci; j < ci + len; j++) c[j] += bval; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] -= a[j]; } public static double[] vectMinusWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] - bval; return c; } public static double[] vectMinusWrite(double bval, double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = bval - a[ai + j]; return c; } public static double[] vectMinusWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] - b[bi + j]; return c; } public static double[] vectMinusWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, -bval); for (int j = ai; j < ai + alen; j++) c[aix[j]] += a[j]; return c; } public static double[] vectMinusWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, bval); for (int j = ai; j < ai + alen; j++) c[aix[j]] -= a[j]; return c; } public static double[] vectMinusWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = -b[bi + j]; for (int j = ai; j < ai + alen; j++) c[aix[j]] += a[j]; return c; } public static double[] vectMinusWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { double[] c = allocVector(len, false); System.arraycopy(a, ai, c, 0, len); for (int j = bi; j < bi + blen; j++) c[bix[j]] -= b[j]; return c; } //custom vector plus public static void vectPlusAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { LibMatrixMult.vectAdd(a, bval, c, ai, ci, len); } public static void vectPlusAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { LibMatrixMult.vectAdd(a, bval, c, ai, ci, len); } public static void vectPlusAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ci; j < ci + len; j++) c[j] += bval; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j]; } public static void vectPlusAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectPlusAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectPlusWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] + bval; return c; } public static double[] vectPlusWrite(double bval, double[] a, int ai, int len) { return vectPlusWrite(a, bval, ai, len); } public static double[] vectPlusWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = a[ai + j] + b[bi + j]; return c; } public static double[] vectPlusWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, bval); for (int j = ai; j < ai + alen; j++) c[aix[j]] += a[j]; return c; } public static double[] vectPlusWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectPlusWrite(a, bval, aix, ai, alen, len); } public static double[] vectPlusWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); System.arraycopy(b, bi, c, 0, len); for (int j = ai; j < ai + alen; j++) c[aix[j]] += a[j]; return c; } public static double[] vectPlusWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectPlusWrite(b, a, bix, bi, ai, blen, len); } //custom vector xor /** * Computes c = xor(A,B) * * @param a dense input vector A * @param bval scalar value * @param c resultant vector * @param ai start position in A * @param ci index of c * @param len number of processed elements */ public static void vectXorAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += ((a[j] != 0) != (bval != 0)) ? 1 : 0; } public static void vectXorAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectXorAdd(a, bval, c, ai, ci, len); } public static void vectXorAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] != 0) != (bval != 0)) ? 1 : 0; } public static void vectXorAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectXorAdd(a, bval, c, aix, ai, ci, alen, len); } //1. scalar vs. dense vector public static double[] vectXorWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = ((a[ai + j] != 0) != (bval != 0)) ? 1 : 0; return c; } //2. dense vector vs. scalar public static double[] vectXorWrite(double bval, double[] a, int ai, int len) { return vectXorWrite(a, bval, ai, len); } //3. dense vector vs. dense vector public static double[] vectXorWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = ((a[ai + j] != 0) != (b[bi + j] != 0)) ? 1 : 0; return c; } //4. sparse vector vs scalar public static double[] vectXorWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval != 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] != 0) ? 0 : 1; return c; } //5. scalar vs. sparse vector public static double[] vectXorWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectXorWrite(a, bval, aix, ai, alen, len); } //6. sparse vector vs. dense vector public static double[] vectXorWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (b[bi + j] != 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = ((a[j] != 0) != (c[aix[j]] != 0)) ? 1 : 0; return c; } //6. sparse vector vs. dense vector public static double[] vectXorWrite(double[] a, double[] b, int ai, int[] aix, int bi, int alen, int len) { return vectXorWrite(a, b, aix, ai, bi, alen, len); } //custom vector pow public static void vectPowAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.pow(a[j], bval); } public static void vectPowAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.pow(bval, a[j]); } public static void vectPowAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval == 0) //handle 0^0=1 & a^0=1 for (int j = 0; j < len; j++) c[ci + j] += 1; else //handle 0^b=0 & a^b for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.pow(a[j], bval); } public static void vectPowAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = 0; j < len; j++) //handle 0^0=1 & b^0=1 c[ci + j] += 1; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.pow(bval, a[j]) - 1; } public static double[] vectPowWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.pow(a[ai], bval); return c; } public static double[] vectPowWrite(double bval, double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.pow(bval, a[ai]); return c; } public static double[] vectPowWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = Math.pow(a[ai], b[bi]); return c; } public static double[] vectPowWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval == 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.pow(a[j], bval) - init; return c; } public static double[] vectPowWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, 1); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.pow(bval, a[j]); return c; } //custom vector min public static void vectMinAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.min(a[j], bval); } public static void vectMinAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectMinAdd(a, bval, c, ai, ci, len); } public static void vectMinAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval < 0) for (int j = 0; j < len; j++) c[ci + j] += bval; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += (bval >= 0) ? Math.min(a[j], bval) : 0; } public static void vectMinAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectMinAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectMinWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.min(a[ai], bval); return c; } public static double[] vectMinWrite(double bval, double[] a, int ai, int len) { return vectMinWrite(a, bval, ai, len); } public static double[] vectMinWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = Math.min(a[ai], b[bi]); return c; } public static double[] vectMinWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval < 0) ? bval : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.min(a[j], bval); return c; } public static double[] vectMinWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectMinWrite(a, bval, aix, ai, alen, len); } public static double[] vectMinWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = Math.min(b[bi + j], 0); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.min(a[j], b[bi + aix[j]]); return c; } public static double[] vectMinWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectMinWrite(b, a, bix, bi, ai, blen, len); } //custom vector max public static void vectMaxAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.max(a[j], bval); } public static void vectMaxAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectMaxAdd(a, bval, c, ai, ci, len); } public static void vectMaxAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval > 0) for (int j = 0; j < len; j++) c[ci + j] += bval; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += (bval <= 0) ? Math.max(a[j], bval) : 0; } public static void vectMaxAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectMaxAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectMaxWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.max(a[ai], bval); return c; } public static double[] vectMaxWrite(double bval, double[] a, int ai, int len) { return vectMaxWrite(a, bval, ai, len); } public static double[] vectMaxWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = Math.max(a[ai], b[bi]); return c; } public static double[] vectMaxWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval > 0) ? bval : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.max(a[j], bval); return c; } public static double[] vectMaxWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectMaxWrite(a, bval, aix, ai, alen, len); } public static double[] vectMaxWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = Math.max(b[bi + j], 0); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.max(a[j], b[bi + aix[j]]); return c; } public static double[] vectMaxWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectMaxWrite(b, a, bix, bi, ai, blen, len); } //custom exp public static void vectExpAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = 0; j < len; j++) c[ci + j] += FastMath.exp(a[ai + j]); } public static void vectExpAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ci; j < ci + len; j++) //exp(0)=1 c[j] += 1; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.exp(a[j]) - 1; } public static double[] vectExpWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = FastMath.exp(a[ai + j]); return c; } public static double[] vectExpWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, 1); //exp(0)=1 for (int j = ai; j < ai + alen; j++) //overwrite c[aix[j]] = FastMath.exp(a[j]); return c; } //custom cumsum public static void vectCumsumAdd(double[] a, double[] c, int ai, int ci, int len) { double val = 0; for (int j = 0; j < len; j++) { val += a[ai * j]; c[ci + j] += val; } } public static void vectCumsumAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes for (int j2 = lastIx + 1; j2 < aix[j]; j2++) c[j2] += val; //update value and add current index val += a[j]; c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes for (int j2 = lastIx + 1; j2 < len; j2++) c[j2] += val; } public static double[] vectCumsumWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); double val = 0; for (int j = 0; j < len; j++) { val += a[ai + j]; c[j] = val; } return c; } public static double[] vectCumsumWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, false); double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes Arrays.fill(c, lastIx + 1, aix[j], val); //update value and add current index val += a[j]; c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes Arrays.fill(c, lastIx + 1, len, val); return c; } //custom cummin public static void vectCumminAdd(double[] a, double[] c, int ai, int ci, int len) { double val = 0; for (int j = 0; j < len; j++) { val = Math.min(val, a[ai * j]); c[ci + j] += val; } } public static void vectCumminAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes for (int j2 = lastIx + 1; j2 < aix[j]; j2++) c[j2] += val; //update value and add current index val = Math.min(val, a[j]); c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes for (int j2 = lastIx + 1; j2 < len; j2++) c[j2] += val; } public static double[] vectCumminWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); double val = 0; for (int j = 0; j < len; j++) { val = Math.min(val, a[ai + j]); c[j] = val; } return c; } public static double[] vectCumminWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, false); double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes Arrays.fill(c, lastIx + 1, aix[j], val); //update value and add current index val = Math.min(val, a[j]); c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes Arrays.fill(c, lastIx + 1, len, val); return c; } //custom cummax public static void vectCummaxAdd(double[] a, double[] c, int ai, int ci, int len) { double val = 0; for (int j = 0; j < len; j++) { val = Math.max(val, a[ai * j]); c[ci + j] += val; } } public static void vectCummaxAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes for (int j2 = lastIx + 1; j2 < aix[j]; j2++) c[j2] += val; //update value and add current index val = Math.max(val, a[j]); c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes for (int j2 = lastIx + 1; j2 < len; j2++) c[j2] += val; } public static double[] vectCummaxWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); double val = 0; for (int j = 0; j < len; j++) { val = Math.max(val, a[ai + j]); c[j] = val; } return c; } public static double[] vectCummaxWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, false); double val = 0; int lastIx = -1; for (int j = ai; j < ai + alen; j++) { //add non-existing indexes Arrays.fill(c, lastIx + 1, aix[j], val); //update value and add current index val = Math.max(val, a[j]); c[aix[j]] = val; lastIx = aix[j]; } //add non-existing indexes Arrays.fill(c, lastIx + 1, len, val); return c; } //custom log public static void vectLogAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.log(a[j]); } public static void vectLogAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.log(a[j]); } public static double[] vectLogWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.log(a[ai]); return c; } public static double[] vectLogWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, Double.NEGATIVE_INFINITY); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.log(a[j]); return c; } //custom abs public static void vectAbsAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.abs(a[j]); } public static void vectAbsAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.abs(a[j]); } public static double[] vectAbsWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.abs(a[ai]); return c; } public static double[] vectAbsWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.abs(a[j]); return c; } //custom round public static void vectRoundAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.round(a[j]); } public static void vectRoundAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.round(a[j]); } public static double[] vectRoundWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.round(a[ai]); return c; } public static double[] vectRoundWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.round(a[j]); return c; } //custom ceil public static void vectCeilAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.ceil(a[j]); } public static void vectCeilAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.ceil(a[j]); } public static double[] vectCeilWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.ceil(a[ai]); return c; } public static double[] vectCeilWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.ceil(a[j]); return c; } //custom floor public static void vectFloorAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.floor(a[j]); } public static void vectFloorAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.floor(a[j]); } public static double[] vectFloorWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.floor(a[ai]); return c; } public static double[] vectFloorWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.floor(a[j]); return c; } //custom sin public static void vectSinAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.sin(a[j]); } public static void vectSinAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.sin(a[j]); } public static double[] vectSinWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.sin(a[ai]); return c; } public static double[] vectSinWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.sin(a[j]); return c; } //custom cos public static void vectCosAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.cos(a[j]); } public static void vectCosAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.cos(a[j]); } public static double[] vectCosWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.cos(a[ai]); return c; } public static double[] vectCosWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, 1); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.cos(a[j]); return c; } //custom tan public static void vectTanAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.tan(a[j]); } public static void vectTanAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.tan(a[j]); } public static double[] vectTanWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.tan(a[ai]); return c; } public static double[] vectTanWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.tan(a[j]); return c; } //custom asin public static void vectAsinAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.asin(a[j]); } public static void vectAsinAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.asin(a[j]); } public static double[] vectAsinWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.asin(a[ai]); return c; } public static double[] vectAsinWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.asin(a[j]); return c; } //custom acos public static void vectAcosAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.acos(a[j]); } public static void vectAcosAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.acos(a[j]); } public static double[] vectAcosWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.acos(a[ai]); return c; } public static double[] vectAcosWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, Math.PI / 2); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.acos(a[j]); return c; } //custom atan public static void vectAtanAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.atan(a[j]); } public static void vectAtanAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.atan(a[j]); } public static double[] vectAtanWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.atan(a[ai]); return c; } public static double[] vectAtanWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.atan(a[j]); return c; } //custom sinh public static void vectSinhAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.sinh(a[j]); } public static void vectSinhAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.sinh(a[j]); } public static double[] vectSinhWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.sinh(a[ai]); return c; } public static double[] vectSinhWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.sinh(a[j]); return c; } //custom cosh public static void vectCoshAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.cosh(a[j]); } public static void vectCoshAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.cosh(a[j]); } public static double[] vectCoshWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.cosh(a[ai]); return c; } public static double[] vectCoshWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, 1); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.cosh(a[j]); return c; } //custom tanh public static void vectTanhAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.tanh(a[j]); } public static void vectTanhAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.tanh(a[j]); } public static double[] vectTanhWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.tanh(a[ai]); return c; } public static double[] vectTanhWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.tanh(a[j]); return c; } //custom sign public static void vectSignAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += FastMath.signum(a[j]); } public static void vectSignAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += FastMath.signum(a[j]); } public static double[] vectSignWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = FastMath.signum(a[ai]); return c; } public static double[] vectSignWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = FastMath.signum(a[j]); return c; } //custom pow2 public static void vectPow2Add(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += a[j] * a[j]; } public static void vectPow2Add(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j] * a[j]; } public static double[] vectPow2Write(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = a[ai] * a[ai]; return c; } public static double[] vectPow2Write(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] * a[j]; return c; } //custom mult2 public static void vectMult2Add(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += a[j] + a[j]; } public static void vectMult2Add(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j] + a[j]; } public static double[] vectMult2Write(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = a[ai] + a[ai]; return c; } public static double[] vectMult2Write(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] + a[j]; return c; } //custom sqrt public static void vectSqrtAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += Math.sqrt(a[j]); } public static void vectSqrtAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += Math.sqrt(a[j]); } public static double[] vectSqrtWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = Math.sqrt(a[ai]); return c; } public static double[] vectSqrtWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = Math.sqrt(a[j]); return c; } //custom sprop public static void vectSpropAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += a[j] * (1 - a[j]); } public static void vectSpropAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += a[j] * (1 - a[j]); } public static double[] vectSpropWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = a[j] * (1 - a[j]); return c; } public static double[] vectSpropWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = a[j] * (1 - a[j]); return c; } //custom sigmoid public static void vectSigmoidAdd(double[] a, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += 1 / (1 + FastMath.exp(-a[j])); } public static void vectSigmoidAdd(double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += 1 / (1 + FastMath.exp(-a[j])); } public static double[] vectSigmoidWrite(double[] a, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = 1 / (1 + FastMath.exp(-a[j])); return c; } public static double[] vectSigmoidWrite(double[] a, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true, 0.5); //sigmoid(0) = 0.5 for (int j = ai; j < ai + alen; j++) c[aix[j]] = 1 / (1 + FastMath.exp(-a[j])); return c; } //custom vector equal public static void vectEqualAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] == bval) ? 1 : 0; } public static void vectEqualAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectEqualAdd(a, bval, c, ai, ci, len); } public static void vectEqualAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval == 0) for (int j = 0; j < len; j++) c[j] += 1; else for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += (a[j] == bval) ? 1 : 0; } public static void vectEqualAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectEqualAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectEqualWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] == bval) ? 1 : 0; return c; } public static double[] vectEqualWrite(double bval, double[] a, int ai, int len) { return vectEqualWrite(a, bval, ai, len); } public static double[] vectEqualWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] == b[bi]) ? 1 : 0; return c; } public static double[] vectEqualWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval == 0) ? 1 : 0; double[] c = allocVector(len, true, init); if (bval != 0) for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] == bval) ? 1 : 0; return c; } public static double[] vectEqualWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectEqualWrite(a, bval, aix, ai, alen, len); } public static double[] vectEqualWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (b[bi + j] == 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] == b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectEqualWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectEqualWrite(b, a, bix, bi, ai, blen, len); } //custom vector not equal public static void vectNotequalAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] != bval) ? 1 : 0; } public static void vectNotequalAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectNotequalAdd(a, bval, c, ai, ci, len); } public static void vectNotequalAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval != 0) for (int j = 0; j < len; j++) c[j] += 1; double init = (bval != 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] != bval) ? 1 : 0) - init; } public static void vectNotequalAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectNotequalAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectNotequalWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] != bval) ? 1 : 0; return c; } public static double[] vectNotequalWrite(double bval, double[] a, int ai, int len) { return vectNotequalWrite(a, bval, ai, len); } public static double[] vectNotequalWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] != b[bi]) ? 1 : 0; return c; } public static double[] vectNotequalWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval != 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = ((a[j] != bval) ? 1 : 0); return c; } public static double[] vectNotequalWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectNotequalWrite(a, bval, aix, ai, alen, len); } public static double[] vectNotequalWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (b[bi + j] != 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] != b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectNotequalWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectNotequalWrite(b, a, bix, bi, ai, blen, len); } //custom vector less public static void vectLessAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] < bval) ? 1 : 0; } public static void vectLessAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectGreaterequalAdd(a, bval, c, ai, ci, len); } public static void vectLessAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval > 0) for (int j = 0; j < len; j++) c[j] += 1; double init = (bval > 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] < bval) ? 1 : 0) - init; } public static void vectLessAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectGreaterequalAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectLessWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] < bval) ? 1 : 0; return c; } public static double[] vectLessWrite(double bval, double[] a, int ai, int len) { return vectGreaterequalWrite(a, bval, ai, len); } public static double[] vectLessWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] < b[bi]) ? 1 : 0; return c; } public static double[] vectLessWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval > 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] < bval) ? 1 : 0; return c; } public static double[] vectLessWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectGreaterequalWrite(a, bval, aix, ai, alen, len); } public static double[] vectLessWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (0 < b[bi + j]) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] < b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectLessWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectGreaterequalWrite(b, a, bix, bi, ai, blen, len); } //custom vector less equal public static void vectLessequalAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] <= bval) ? 1 : 0; } public static void vectLessequalAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectGreaterAdd(a, bval, c, ai, ci, len); } public static void vectLessequalAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval >= 0) for (int j = 0; j < len; j++) c[j] += 1; double init = (bval >= 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] <= bval) ? 1 : 0) - init; } public static void vectLessequalAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectGreaterAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectLessequalWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] <= bval) ? 1 : 0; return c; } public static double[] vectLessequalWrite(double bval, double[] a, int ai, int len) { return vectGreaterWrite(a, bval, ai, len); } public static double[] vectLessequalWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] <= b[bi]) ? 1 : 0; return c; } public static double[] vectLessequalWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval >= 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = ((a[j] <= bval) ? 1 : 0); return c; } public static double[] vectLessequalWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectGreaterWrite(a, bval, aix, ai, alen, len); } public static double[] vectLessequalWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (0 <= b[bi + j]) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] <= b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectLessequalWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectGreaterWrite(b, a, bix, bi, ai, blen, len); } //custom vector greater public static void vectGreaterAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] > bval) ? 1 : 0; } public static void vectGreaterAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectLessequalAdd(a, bval, c, ai, ci, len); } public static void vectGreaterAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval < 0) for (int j = 0; j < len; j++) c[j] += 1; double init = (bval < 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] > bval) ? 1 : 0) - init; } public static void vectGreaterAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectLessequalAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectGreaterWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] > bval) ? 1 : 0; return c; } public static double[] vectGreaterWrite(double bval, double[] a, int ai, int len) { return vectLessWrite(a, bval, ai, len); } public static double[] vectGreaterWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] > b[bi]) ? 1 : 0; return c; } public static double[] vectGreaterWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval < 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = ((a[j] > bval) ? 1 : 0); return c; } public static double[] vectGreaterWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectLessequalWrite(a, bval, aix, ai, alen, len); } public static double[] vectGreaterWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (0 > b[bi + j]) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] > b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectGreaterWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectLessequalWrite(b, a, bix, bi, ai, blen, len); } //custom vector greater equal public static void vectGreaterequalAdd(double[] a, double bval, double[] c, int ai, int ci, int len) { for (int j = ai; j < ai + len; j++, ci++) c[ci] += (a[j] >= bval) ? 1 : 0; } public static void vectGreaterequalAdd(double bval, double[] a, double[] c, int ai, int ci, int len) { vectLessAdd(a, bval, c, ai, ci, len); } public static void vectGreaterequalAdd(double[] a, double bval, double[] c, int[] aix, int ai, int ci, int alen, int len) { if (bval <= 0) for (int j = 0; j < len; j++) c[j] += 1; double init = (bval <= 0) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[ci + aix[j]] += ((a[j] >= bval) ? 1 : 0) - init; } public static void vectGreaterequalAdd(double bval, double[] a, double[] c, int[] aix, int ai, int ci, int alen, int len) { vectLessAdd(a, bval, c, aix, ai, ci, alen, len); } public static double[] vectGreaterequalWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++) c[j] = (a[ai] >= bval) ? 1 : 0; return c; } public static double[] vectGreaterequalWrite(double bval, double[] a, int ai, int len) { return vectLessWrite(a, bval, ai, len); } public static double[] vectGreaterequalWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++, ai++, bi++) c[j] = (a[ai] >= b[bi]) ? 1 : 0; return c; } public static double[] vectGreaterequalWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double init = (bval < 0) ? 1 : 0; double[] c = allocVector(len, true, init); for (int j = ai; j < ai + alen; j++) c[aix[j]] = ((a[j] >= bval) ? 1 : 0) - init; return c; } public static double[] vectGreaterequalWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectLessWrite(a, bval, aix, ai, alen, len); } public static double[] vectGreaterequalWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = (0 >= b[bi + j]) ? 1 : 0; for (int j = ai; j < ai + alen; j++) c[aix[j]] = (a[j] >= b[bi + aix[j]]) ? 1 : 0; return c; } public static double[] vectGreaterequalWrite(double[] a, double[] b, int ai, int[] bix, int bi, int blen, int len) { //invariant to the ordering of inputs return vectLessWrite(b, a, bix, bi, ai, blen, len); } //bitwise and //1. dense vector vs. scalar public static double[] vectBitwandWrite(double[] a, double bval, int ai, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = bwAnd(a[ai + j], bval); return c; } //2. scalar vs. dense vector public static double[] vectBitwandWrite(double bval, double[] a, int ai, int len) { return vectBitwandWrite(a, bval, ai, len); } //3. dense vector vs. dense vector public static double[] vectBitwandWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); for (int j = 0; j < len; j++) c[j] = bwAnd(a[ai + j], b[bi + j]); return c; } //4. sparse vector vs. scalar. public static double[] vectBitwandWrite(double[] a, double bval, int[] aix, int ai, int alen, int len) { double[] c = allocVector(len, true); int bval1 = (int) bval; for (int j = ai; j < ai + alen; j++) c[aix[j]] = bwAnd(a[j], bval1); return c; } //5. scalar vs. sparse vector public static double[] vectBitwandWrite(double bval, double[] a, int[] aix, int ai, int alen, int len) { return vectBitwandWrite(a, bval, aix, ai, alen, len); } //6. sparse vector vs. dense vector public static double[] vectBitwandWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, true); for (int j = ai; j < ai + alen; j++) c[aix[j]] = bwAnd(a[j], b[bi + aix[j]]); return c; } //6. sparse vector vs. dense vector public static double[] vectBitwandWrite(double[] a, double[] b, int ai, int[] aix, int bi, int alen, int len) { return vectBitwandWrite(a, b, aix, ai, bi, alen, len); } // bias add public static double[] vectBiasaddWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); System.arraycopy(a, ai, c, 0, len); LibMatrixDNN.addBias(c, b, 1, 1, b.length, len / b.length); return c; } public static double[] vectBiasaddWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, true); for (int k = ai; k < ai + alen; k++) c[aix[k]] = a[k]; LibMatrixDNN.addBias(c, b, 1, 1, b.length, len / b.length); return c; } // bias mult public static double[] vectBiasmultWrite(double[] a, double[] b, int ai, int bi, int len) { double[] c = allocVector(len, false); System.arraycopy(a, ai, c, 0, len); LibMatrixDNN.multBias(c, b, 1, b.length, len / b.length); return c; } public static double[] vectBiasmultWrite(double[] a, double[] b, int[] aix, int ai, int bi, int alen, int len) { double[] c = allocVector(len, true); for (int k = ai; k < ai + alen; k++) c[aix[k]] = a[k]; LibMatrixDNN.multBias(c, b, 1, b.length, len / b.length); return c; } //complex builtin functions that are not directly generated //(included here in order to reduce the number of imports) public static double intDiv(double in1, double in2) { return intDiv.execute(in1, in2); } public static double mod(double in1, double in2) { return mod.execute(in1, in2); } public static double bwAnd(double in1, double in2) { return bwAnd.execute(in1, in2); } public static boolean isFlipOuter(int len1, int len2) { return (len1 > 64 * len2); } //dynamic memory management public static void setupThreadLocalMemory(int numVectors, int len) { if (numVectors > 0) setupThreadLocalMemory(numVectors, len, -1); } public static void setupThreadLocalMemory(int numVectors, int len, int len2) { if (numVectors > 0) memPool.set(new VectorBuffer(numVectors, len, len2)); } public static void cleanupThreadLocalMemory() { memPool.remove(); } public static double[] allocVector(int len, boolean reset) { return allocVector(len, reset, 0); } protected static double[] allocVector(int len, boolean reset, double resetVal) { VectorBuffer buff = memPool.get(); //find next matching vector in ring buffer or //allocate new vector if required double[] vect = buff.next(len); if (vect == null) vect = new double[len]; //reset vector if required if (reset) Arrays.fill(vect, resetVal); return vect; } /** * Simple ring buffer of allocated vectors, where * vectors of different sizes are interspersed. */ private static class VectorBuffer { private static final int MAX_SIZE = 512 * 1024; //4MB private final double[][] _data; private int _pos; private int _len1; private int _len2; public VectorBuffer(int num, int len1, int len2) { //best effort size restriction since large intermediates //not necessarily used (num refers to the total number) len1 = Math.min(len1, MAX_SIZE); len2 = Math.min(len2, MAX_SIZE); //pre-allocate ring buffer int lnum = (len2 > 0 && len1 != len2) ? 2 * num : num; _data = new double[lnum][]; for (int i = 0; i < num; i++) { if (lnum > num) { _data[2 * i] = new double[len1]; _data[2 * i + 1] = new double[len2]; } else { _data[i] = new double[len1]; } } _pos = -1; _len1 = len1; _len2 = len2; } public double[] next(int len) { if (_len1 != len && _len2 != len) return null; do { _pos = (_pos + 1 >= _data.length) ? 0 : _pos + 1; } while (_data[_pos].length != len); return _data[_pos]; } @SuppressWarnings("unused") public boolean isReusable(int num, int len1, int len2) { int lnum = (len2 > 0 && len1 != len2) ? 2 * num : num; return (_len1 == len1 && _len2 == len2 && _data.length == lnum); } } }