Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance. Project price only 1 $
You can buy this project and download/modify it how often you want.
package org.saddle.ops.macroImpl
import scala.reflect.macros.blackbox.Context
import scala.language.experimental.macros
import org.saddle.Mat
import org.saddle.ops._
object BinOpMatMacros {
import BinOpMacros._
def matMatIP[A, B, OP]: BinOpInPlace[OP, Mat[A], Mat[B]] =
macro matMatIPImpl[A, B, OP]
def matMatIPImpl[
A: cxt.WeakTypeTag,
B: cxt.WeakTypeTag,
OP: cxt.WeakTypeTag
](cxt: Context) = {
import cxt.universe._
val a = weakTypeOf[A]
val b = weakTypeOf[B]
val opT = weakTypeOf[OP]
val (name1, name2, body) =
inline[A, B, A](cxt)(createOperationBody[A, B, A, OP](cxt))
val r = q"""
new _root_.org.saddle.ops.BinOpInPlace[$opT, _root_.org.saddle.Mat[$a], _root_.org.saddle.Mat[$b]] {
def apply(v1: _root_.org.saddle.Mat[$a], v2: _root_.org.saddle.Mat[$b]) =
if (v1.numRows == v2.numRows && v1.numCols == v2.numCols) {
val sz = v1.length
var i = 0
val v1a = v1.toArray
val v2a = v2.toArray
while (i < sz) {
val $name1 = v1a(i)
val $name2 = v2a(i)
v1a(i) = $body
i += 1
}
} else if ((v1.numRows == v2.numRows || v1.numRows == 1 || v2.numRows == 1) &&
(v1.numCols == v2.numCols || v1.numCols == 1 || v2.numCols == 1)) {
// Broadcasting
val nR = _root_.scala.math.max(v1.numRows, v2.numRows)
val nC = _root_.scala.math.max(v1.numCols, v2.numCols)
val nR1 = v1.numRows
val nR2 = v2.numRows
val nC1 = v1.numCols
val nC2 = v2.numCols
val v1a = v1.toArray
val v2a = v2.toArray
val sz = nR * nC
var i = 0
while (i < sz) {
val r = i / nC
val c = i % nC
val r1 = if (nR1 == 1) 0 else r
val r2 = if (nR2 == 1) 0 else r
val c1 = if (nC1 == 1) 0 else c
val c2 = if (nC2 == 1) 0 else c
val $name1 = v1a(r1 * nC1 + c1)
val $name2 = v2a(r2 * nC2 + c2)
v1a(i) = $body
i += 1
}
} else throw new RuntimeException("Mats must have compatible size!")
}
"""
r
}
def matMat[A, B, C, OP]: BinOp[OP, Mat[A], Mat[B], Mat[C]] =
macro matMatImpl[A, B, C, OP]
def matMatImpl[
A: cxt.WeakTypeTag,
B: cxt.WeakTypeTag,
C: cxt.WeakTypeTag,
OP: cxt.WeakTypeTag
](cxt: Context) = {
import cxt.universe._
val a = weakTypeOf[A]
val b = weakTypeOf[B]
val c = weakTypeOf[C]
val opT = weakTypeOf[OP]
val (name1, name2, body) =
inline[A, B, C](cxt)(createOperationBody[A, B, C, OP](cxt))
val r = q"""
new _root_.org.saddle.ops.BinOp[$opT, _root_.org.saddle.Mat[$a], _root_.org.saddle.Mat[$b], _root_.org.saddle.Mat[$c]] {
def apply(v1: _root_.org.saddle.Mat[$a], v2: _root_.org.saddle.Mat[$b]) =
if (v1.numRows == v2.numRows && v1.numCols == v2.numCols) {
val sz = v1.length
val ar = new Array[$c](sz)
var i = 0
val v1a = v1.toArray
val v2a = v2.toArray
while (i < sz) {
val $name1 = v1a(i)
val $name2 = v2a(i)
ar(i) = $body
i += 1
}
_root_.org.saddle.Mat(v1.numRows, v1.numCols, ar)
} else if ((v1.numRows == v2.numRows || v1.numRows == 1 || v2.numRows == 1) &&
(v1.numCols == v2.numCols || v1.numCols == 1 || v2.numCols == 1)) {
// Broadcasting
val nR = _root_.scala.math.max(v1.numRows, v2.numRows)
val nC = _root_.scala.math.max(v1.numCols, v2.numCols)
val nR1 = v1.numRows
val nR2 = v2.numRows
val nC1 = v1.numCols
val nC2 = v2.numCols
val v1a = v1.toArray
val v2a = v2.toArray
val sz = nR * nC
val ar = new Array[$c](sz)
var i = 0
while (i < sz) {
val r = i / nC
val c = i % nC
val r1 = if (nR1 == 1) 0 else r
val r2 = if (nR2 == 1) 0 else r
val c1 = if (nC1 == 1) 0 else c
val c2 = if (nC2 == 1) 0 else c
val $name1 = v1a(r1 * nC1 + c1)
val $name2 = v2a(r2 * nC2 + c2)
ar(i) = $body
i += 1
}
_root_.org.saddle.Mat(nR, nC, ar)
} else throw new RuntimeException("Mats must have compatible size!")
}
"""
r
}
def matScalarElementwise[A, B, C, OP]: BinOp[OP, Mat[A], B, Mat[C]] =
macro matScalarElementwiseImpl[A, B, C, OP]
def matScalarElementwiseImpl[
A: cxt.WeakTypeTag,
B: cxt.WeakTypeTag,
C: cxt.WeakTypeTag,
OP: cxt.WeakTypeTag
](cxt: Context) = {
import cxt.universe._
val a = weakTypeOf[A]
val b = weakTypeOf[B]
val c = weakTypeOf[C]
val opT = weakTypeOf[OP]
val (name1, name2, body) =
inline[A, B, C](cxt)(createOperationBody[A, B, C, OP](cxt))
val r = q"""
new _root_.org.saddle.ops.BinOp[$opT, _root_.org.saddle.Mat[$a], $b, _root_.org.saddle.Mat[$c]] {
def apply(v1: _root_.org.saddle.Mat[$a], v2: $b) = {
val sz = v1.length
val v1a = v1.toArray
val ar = new Array[$c](sz)
var i = 0
while (i < sz) {
val $name1 = v1a(i)
val $name2 = v2
ar(i) = $body
i += 1
}
_root_.org.saddle.Mat(v1.numRows, v1.numCols, ar)
}
}
"""
r
}
def matScalarElementwiseIP[A, B, OP]: BinOpInPlace[OP, Mat[A], B] =
macro matScalarElementwiseIPImpl[A, B, OP]
def matScalarElementwiseIPImpl[
A: cxt.WeakTypeTag,
B: cxt.WeakTypeTag,
OP: cxt.WeakTypeTag
](cxt: Context) = {
import cxt.universe._
val a = weakTypeOf[A]
val b = weakTypeOf[B]
val opT = weakTypeOf[OP]
val (name1, name2, body) =
inline[A, B, A](cxt)(createOperationBody[A, B, A, OP](cxt))
val r = q"""
new _root_.org.saddle.ops.BinOpInPlace[$opT, _root_.org.saddle.Mat[$a], $b] {
def apply(v1: _root_.org.saddle.Mat[$a], v2: $b) = {
val sz = v1.length
val v1a = v1.toArray
var i = 0
while (i < sz) {
val $name1 = v1a(i)
val $name2 = v2
v1a(i) = $body
i += 1
}
}
}
"""
r
}
}
