First draft of ANSIC++ Vector code. Needs to be tested.
This commit is contained in:
Tristan Campbell
parent
5725ff9c66
commit
b5fab4fcc6
|
|
@ -0,0 +1,332 @@
|
|||
|
||||
//#include <iostream>
|
||||
#include <jni.h>
|
||||
#include "MatrixOpCommon.h"
|
||||
|
||||
bool Matrix::identicalDataSpaces(Matrix & other)
|
||||
{
|
||||
if (address != other.address)
|
||||
return JNI_FALSE;
|
||||
if (stride != other.stride)
|
||||
return JNI_FALSE;
|
||||
if ((width * height) != (other.width * other.height))
|
||||
return JNI_FALSE;
|
||||
return JNI_TRUE;
|
||||
}
|
||||
|
||||
bool Matrix::intersectingDataSpaces(Matrix & other)
|
||||
{
|
||||
char * my_max_address = &address[ stride * elements ];
|
||||
char * other_max_address = &other.address[ other.stride * other.elements];
|
||||
|
||||
if (address >= other.address || address <= other_max_address) return JNI_TRUE;
|
||||
if (other.address >= address || other.address <= my_max_address) return JNI_TRUE;
|
||||
return JNI_FALSE;
|
||||
}
|
||||
|
||||
void Matrix::transposeMatrix(float * src, float * dst, int src_width, int src_height)
|
||||
{
|
||||
// square matrix transpose
|
||||
if (src_width == src_height)
|
||||
{
|
||||
for (int i = 0; i < src_width; i++)
|
||||
for (int j = 0; j < src_width; j++)
|
||||
dst[i + src_width * j] = src[j + i * src_width];
|
||||
}
|
||||
// non square matrix transpose
|
||||
else
|
||||
{
|
||||
for (int i = 0; i < src_width; i ++)
|
||||
for (int j = 0; j < src_height; j++)
|
||||
dst[i + src_height * j] = src[j + i * src_height];
|
||||
}
|
||||
}
|
||||
|
||||
void Matrix::transposeMatrix(float * mat, int src_width, int src_height)
|
||||
{
|
||||
float temp;
|
||||
|
||||
// square matrix transpose
|
||||
if (src_width == src_height)
|
||||
{
|
||||
for (int col = 0; col < src_width; col++)
|
||||
{
|
||||
for (int row = col+1; row < src_height; row++)
|
||||
{
|
||||
// swap the two elements
|
||||
temp = mat [col * src_height + row];
|
||||
mat[col * src_height + row] = mat[row * src_width + col];
|
||||
mat[row * src_width + col] = temp;
|
||||
}
|
||||
}
|
||||
}
|
||||
// non square matrix transpose
|
||||
else
|
||||
{
|
||||
transposeMatrix(mat, transpose_record, src_width, src_height);
|
||||
memcpy(mat, transpose_record, src_width * src_height * sizeof(float));
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
||||
SrcMatrix::SrcMatrix ( jint addr, jint s,
|
||||
jint w, jint h,
|
||||
jint e, jboolean t):
|
||||
Matrix(addr, s, e),
|
||||
record_offset((char *) addr),
|
||||
record_size (w*h)
|
||||
{
|
||||
if (t) {
|
||||
width = h;
|
||||
height = w;
|
||||
}
|
||||
else {
|
||||
width = w;
|
||||
height = h;
|
||||
}
|
||||
|
||||
elements = e;
|
||||
record = new float[width * height];
|
||||
|
||||
// vectors do not need to be transposed
|
||||
transpose = (t == JNI_TRUE)
|
||||
&& (w != 1)
|
||||
&& (h != 1);
|
||||
|
||||
if (transpose && (width != height))
|
||||
// only need temp storage for transpose if the matrix is not square
|
||||
transpose_record = new float[width*height];
|
||||
else
|
||||
transpose_record = 0;
|
||||
|
||||
if (elements == 1)
|
||||
{
|
||||
// fool the nextRecord function into returning a value
|
||||
elements = 2;
|
||||
nextRecord();
|
||||
elements = 1;
|
||||
}
|
||||
}
|
||||
|
||||
SrcMatrix::~SrcMatrix()
|
||||
{
|
||||
//cout << "SrcMatrix destructor \n";
|
||||
|
||||
delete [] record;
|
||||
|
||||
if (transpose_record != 0)
|
||||
delete [] transpose_record;
|
||||
}
|
||||
|
||||
float * SrcMatrix::nextRecord()
|
||||
{
|
||||
if (elements > 1)
|
||||
{
|
||||
//cout << "Elements: " << elements << "\n";
|
||||
//cout << "Address: " << (unsigned int) (record_offset) << "\n";
|
||||
|
||||
// the record is not properly aligned
|
||||
if ((unsigned int) (record_offset) & FLOAT_ALIGNMENT)
|
||||
{
|
||||
// copy the floats into a buffer so that they are aligned
|
||||
// on 4 byte margins (not necessary on intel, but a good thing)
|
||||
|
||||
memcpy (record, record_offset, record_size * sizeof(float));
|
||||
|
||||
if (transpose)
|
||||
transposeMatrix (record, height, width);
|
||||
|
||||
record_offset = &record_offset[stride];
|
||||
current_record_ptr = record;
|
||||
}
|
||||
// the record is aligned but it has to be transposed
|
||||
else if (transpose)
|
||||
{
|
||||
transposeMatrix ((float *) (record_offset), record, height, width);
|
||||
record_offset = &record_offset[stride];
|
||||
current_record_ptr = record;
|
||||
}
|
||||
// nothing has to be done to the record
|
||||
else
|
||||
{
|
||||
// the floats are aligned in memory
|
||||
current_record_ptr = (float *) record_offset;
|
||||
record_offset = &record_offset[stride];
|
||||
}
|
||||
}
|
||||
|
||||
return current_record_ptr;
|
||||
}
|
||||
|
||||
DstMatrix::DstMatrix (jint addr, jint s, jint w, jint h, jint e, jboolean t):
|
||||
Matrix(addr, s, e)
|
||||
{
|
||||
width = w;
|
||||
height = h;
|
||||
record_size = width * height;
|
||||
record = new float[record_size];
|
||||
|
||||
// vectors do not need to be transposed
|
||||
transpose = (t) && (w != 1) && (h != 1);
|
||||
|
||||
if (transpose)
|
||||
transpose_record = new float[width*height];
|
||||
else
|
||||
transpose_record = 0;
|
||||
|
||||
data_buffered = JNI_FALSE;
|
||||
record_buffered = JNI_FALSE;
|
||||
|
||||
record_offset = address - stride;
|
||||
}
|
||||
|
||||
DstMatrix::~DstMatrix()
|
||||
{
|
||||
//cout << "DstMatrix destructor \n";
|
||||
|
||||
delete [] record;
|
||||
if (transpose_record != 0)
|
||||
delete [] transpose_record;
|
||||
|
||||
// copy back any buffered data
|
||||
if (data_buffered)
|
||||
{
|
||||
char * src = buffer;
|
||||
char * dest = address;
|
||||
|
||||
for (int i = 0; i < elements; i++)
|
||||
{
|
||||
memcpy(dest, src, record_size * sizeof(float));
|
||||
src += stride;
|
||||
dest += stride;
|
||||
}
|
||||
|
||||
delete [] buffer;
|
||||
}
|
||||
}
|
||||
|
||||
void DstMatrix::configureBuffer(SrcMatrix & a, SrcMatrix & b)
|
||||
{
|
||||
|
||||
|
||||
if (!a.intersectingDataSpaces(b))
|
||||
{
|
||||
// as long as the output only overlays 1 of the sources, and the other
|
||||
// source only has 1 matrix in it, only a record_buffer is required
|
||||
if (a.elements == 1 && identicalDataSpaces(b))
|
||||
record_buffered = JNI_TRUE;
|
||||
else if (b.elements == 1 && identicalDataSpaces(a))
|
||||
record_buffered = JNI_TRUE;
|
||||
else
|
||||
// otherwise all of the output has to be buffered
|
||||
createBuffer();
|
||||
}
|
||||
else
|
||||
createBuffer();
|
||||
}
|
||||
|
||||
void DstMatrix::configureBuffer(SrcMatrix & a)
|
||||
{
|
||||
if (identicalDataSpaces(a))
|
||||
record_buffered = JNI_TRUE;
|
||||
else if (intersectingDataSpaces(a))
|
||||
createBuffer();
|
||||
}
|
||||
|
||||
void DstMatrix::createBuffer()
|
||||
{
|
||||
data_buffered = JNI_TRUE;
|
||||
buffer = new char[ elements * stride ];
|
||||
record_offset = buffer - stride;
|
||||
}
|
||||
|
||||
float * DstMatrix::nextRecord()
|
||||
{
|
||||
record_offset = &record_offset[stride];
|
||||
|
||||
if ((((unsigned int)(record_offset)) & FLOAT_ALIGNMENT) || transpose || record_buffered)
|
||||
{
|
||||
last_record_in_temp = JNI_TRUE;
|
||||
return record;
|
||||
}
|
||||
else
|
||||
{
|
||||
last_record_in_temp = JNI_FALSE;
|
||||
return (float *) record_offset;
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
void DstMatrix::writeRecord()
|
||||
{
|
||||
if (last_record_in_temp)
|
||||
{
|
||||
// 3 reasons why the record would be in temp
|
||||
//
|
||||
// 1. The record is not aligned
|
||||
// 2. The result will need to be transposed
|
||||
// 3. Direct Mode where result would overlay an operand
|
||||
|
||||
if (((unsigned int)(record_offset)) & FLOAT_ALIGNMENT)
|
||||
{
|
||||
if (transpose)
|
||||
transposeMatrix(record, width, height);
|
||||
memcpy (record, record_offset, record_size * sizeof(jfloat));
|
||||
}
|
||||
else if (transpose)
|
||||
{
|
||||
transposeMatrix(record, (float *) record_offset, width, height);
|
||||
}
|
||||
else
|
||||
memcpy (record_offset, record, record_size * sizeof(jfloat));
|
||||
}
|
||||
}
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////
|
||||
|
||||
void subMatrix (const float * src, int side, float * dst , int col_omit, int row_omit)
|
||||
{
|
||||
int index = 0;
|
||||
|
||||
for (int c = 0; c < side; c++)
|
||||
{
|
||||
if (c == col_omit) continue;
|
||||
for (int r = 0; r < side; r++)
|
||||
{
|
||||
if (r == row_omit) continue;
|
||||
dst[index++] = src[r + c * side];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
float determinant (const float * matrix , int side)
|
||||
{
|
||||
// we are assuming for this case that the data is in column major format
|
||||
|
||||
float det = 0;
|
||||
|
||||
if (side == 2)
|
||||
// your basic cross product
|
||||
det = matrix[0] * matrix[3] - matrix[2] * matrix[1];
|
||||
else
|
||||
{
|
||||
int temp_side = side - 1;
|
||||
float temp_matrix [temp_side * temp_side];
|
||||
float sign = 1;
|
||||
|
||||
for (int i = 0; i < side; i++)
|
||||
{
|
||||
// get a sub matrix by eliminating the ith row and the 0th column
|
||||
subMatrix(matrix, side, temp_matrix, 0, i);
|
||||
|
||||
// add to the determinant sign * [a]i0 * [M]i0
|
||||
det += sign * matrix[i] * determinant (temp_matrix, temp_side);
|
||||
|
||||
// alternate the sign
|
||||
sign = (sign == 1) ? -1 : 1;
|
||||
}
|
||||
}
|
||||
|
||||
return det;
|
||||
}
|
||||
|
|
@ -0,0 +1,91 @@
|
|||
/*
|
||||
* Matrix.h
|
||||
*
|
||||
*
|
||||
* Created by tristan on Sat Aug 24 2002.
|
||||
* Copyright (c) 2001 __MyCompanyName__. All rights reserved.
|
||||
*
|
||||
*/
|
||||
|
||||
|
||||
|
||||
#define FLOAT_ALIGNMENT 0x00000003
|
||||
|
||||
float determinant (const float * matrix , int side);
|
||||
void subMatrix (const float * src, int side, float * dst , int col_omit, int row_omit);
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////
|
||||
// Matrix
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
class Matrix
|
||||
{
|
||||
protected:
|
||||
float * transpose_record; // to use while transposing the record
|
||||
|
||||
public:
|
||||
char * address; // the start of the data
|
||||
jint stride; // the distance between each record
|
||||
jint width, // the width of the matrix
|
||||
height, // the height of the matrix
|
||||
elements; // the number of matricies
|
||||
jboolean transpose; // whether this matrix is or will be transposed
|
||||
|
||||
Matrix (jint a, jint s, jint e):
|
||||
address((char *)a), stride(s), elements(e) {}
|
||||
bool identicalDataSpaces (Matrix & other);
|
||||
bool intersectingDataSpaces(Matrix & other);
|
||||
void transposeMatrix(float * src, float * dst, int src_width, int src_height);
|
||||
void transposeMatrix(float * mat, int src_width, int src_height);
|
||||
};
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////
|
||||
// Src Matrix
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
class SrcMatrix: public Matrix
|
||||
{
|
||||
private:
|
||||
char * record_offset; // the offset of this record in memory
|
||||
|
||||
float * record; // temporary storage to store a fully aligned and transposed
|
||||
// copy of the record, if the one in memory is not so
|
||||
float * current_record_ptr; // the address of the memory containing the record last
|
||||
// returned by the nextRecord() function
|
||||
jint record_size; // the total floats in each record
|
||||
|
||||
public:
|
||||
SrcMatrix ( jint address, jint stride, jint width, jint height, jint elements, jboolean transpose);
|
||||
void rewind() { record_offset = address; }
|
||||
float * nextRecord();
|
||||
~SrcMatrix();
|
||||
};
|
||||
|
||||
///////////////////////////////////////////////////////////////////////////////////////
|
||||
// Dst Matrix
|
||||
//////////////////////////////////////////////////////////////////////////////////////
|
||||
|
||||
class DstMatrix: public Matrix
|
||||
{
|
||||
char * record_offset; // the offset of the record in memory
|
||||
|
||||
jboolean data_buffered; // if all of the data has to be buffered
|
||||
char * buffer; // a buffer used when data_buffered
|
||||
|
||||
jboolean last_record_in_temp;
|
||||
jboolean record_buffered; // if only a single record is buffered
|
||||
float * record; // to store data if source is unaligned
|
||||
|
||||
jint record_size;
|
||||
|
||||
public:
|
||||
DstMatrix (jint address, jint stride, jint width, jint height, jint elements, jboolean transpose);
|
||||
void configureBuffer(SrcMatrix & a, SrcMatrix & b);
|
||||
void configureBuffer(SrcMatrix & a);
|
||||
void createBuffer();
|
||||
float * nextRecord();
|
||||
void writeRecord();
|
||||
~DstMatrix();
|
||||
};
|
||||
|
||||
|
||||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpAdd_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpAdd_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +68,34 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpAdd_00024MatrixOpDirect_
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
left.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(left, right);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < left.elements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
|
||||
right.rewind();
|
||||
for (int j = 0; j < right.elements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (int k = (dest.width * dest.height) - 1; k >= 0; k--)
|
||||
destRecord[k] = leftRecord[k] + rightRecord[k];
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpAdd_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpAdd_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +68,33 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpAdd_00024MatrixOpSafe_ex
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
left.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < leftElements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
right.rewind();
|
||||
|
||||
for (int j = 0; j < rightElements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (int k = (leftSourceWidth * rightSourceWidth) - 1; k >= 0; k--)
|
||||
destRecord[k] = leftRecord[k] + rightRecord[k];
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpCopy_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpCopy_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -61,8 +62,29 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpCopy_00024MatrixOpDirect
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
if (transposeSource == transposeDest)
|
||||
{
|
||||
transposeSource = false;
|
||||
transposeDest = false;
|
||||
}
|
||||
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(source);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
// just do a straight memory copy
|
||||
memcpy(destRecord, sourceRecord, source.width*source.height*sizeof(jfloat));
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpCopy_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpCopy_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +62,25 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpCopy_00024MatrixOpSafe_e
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
// remove any unnecessary copying
|
||||
if (transposeSource == transposeDest)
|
||||
{
|
||||
transposeSource = false;
|
||||
transposeDest = false;
|
||||
}
|
||||
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
// just do a straight memory copy
|
||||
memcpy(destRecord, sourceRecord, source.width * source.height * sizeof(jfloat));
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpInvert_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpInvert_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +62,47 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpInvert_00024MatrixOpDire
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
// We are under the assumption that sourceWidth == sourceHeight and the matrix
|
||||
// defined within is invertable
|
||||
|
||||
SrcMatrix source (sourceAddress, sourceStride,
|
||||
sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(source);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
|
||||
int temp_side = source.width-1;
|
||||
float temp_matrix [temp_side*temp_side];
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
float det = determinant(sourceRecord, sourceWidth);
|
||||
float sign;
|
||||
|
||||
for (int c = 0; c < source.width; c++)
|
||||
{
|
||||
sign = (c & 1) ? 1.0f : -1.0f;
|
||||
|
||||
for (int r = 0; r < source.width; r++)
|
||||
{
|
||||
// get the sub matrix
|
||||
subMatrix(sourceRecord, source.width, temp_matrix, c, r);
|
||||
|
||||
// transpose the result
|
||||
destRecord[r + c * source.width]
|
||||
= (sign / det) * determinant(temp_matrix, temp_side);
|
||||
|
||||
// swap signs
|
||||
sign = (sign == 1) ? -1.0f : 1.0f;
|
||||
}
|
||||
}
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,9 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpInvert_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
|
||||
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpInvert_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +64,45 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpInvert_00024MatrixOpSafe
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
// We are under the assumption that sourceWidth == sourceHeight and the matrix
|
||||
// defined within is invertable
|
||||
|
||||
SrcMatrix source (sourceAddress, sourceStride,
|
||||
sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
|
||||
int temp_side = source.width-1;
|
||||
float temp_matrix [temp_side*temp_side];
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
float det = determinant(sourceRecord, sourceWidth);
|
||||
float sign;
|
||||
|
||||
for (int c = 0; c < source.width; c++)
|
||||
{
|
||||
sign = (c & 1) ? 1.0f : -1.0f;
|
||||
|
||||
for (int r = 0; r < source.width; r++)
|
||||
{
|
||||
// get the sub matrix
|
||||
subMatrix(sourceRecord, source.width, temp_matrix, c, r);
|
||||
|
||||
// transpose the result
|
||||
destRecord[r + c * source.width]
|
||||
= (sign / det) * determinant(temp_matrix, temp_side);
|
||||
|
||||
// swap signs
|
||||
sign = (sign == 1) ? -1.0f : 1.0f;
|
||||
}
|
||||
}
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,8 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpMultiply_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
#include <cstring>
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpMultiply_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +69,55 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpMultiply_00024MatrixOpDi
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
if (transposeLeftSource && transposeRightSource)
|
||||
{
|
||||
transposeLeftSource = false;
|
||||
transposeRightSource = false;
|
||||
transposeDest = !transposeDest;
|
||||
}
|
||||
|
||||
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
right.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(left, right);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
// check out discussions envolving ordering
|
||||
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < leftElements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
|
||||
right.rewind();
|
||||
for (int j = 0; j < rightElements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
memset(destRecord, 0, dest.width * dest.height * sizeof(jfloat));
|
||||
|
||||
for (int rightCol = 0; rightCol < right.width; rightCol++)
|
||||
{
|
||||
for (int leftIndex = 0; leftIndex < left.width*left.height; leftIndex++)
|
||||
{
|
||||
destRecord[i % dest.height] += leftRecord[i] * rightRecord[i / leftSourceHeight];
|
||||
}
|
||||
|
||||
rightRecord = &rightRecord[right.height];
|
||||
destRecord = &destRecord[dest.height];
|
||||
|
||||
}
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,9 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpMultiply_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
#include <cstring>
|
||||
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpMultiply_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +70,54 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpMultiply_00024MatrixOpSa
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
if (transposeLeftSource && transposeRightSource)
|
||||
{
|
||||
transposeLeftSource = false;
|
||||
transposeRightSource = false;
|
||||
transposeDest = !transposeDest;
|
||||
}
|
||||
|
||||
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
right.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
// check out discussions envolving ordering
|
||||
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < leftElements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
|
||||
right.rewind();
|
||||
for (int j = 0; j < rightElements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
memset(destRecord, 0, dest.width * dest.height * sizeof(jfloat));
|
||||
|
||||
for (int rightCol = 0; rightCol < right.width; rightCol++)
|
||||
{
|
||||
for (int leftIndex = 0; leftIndex < left.width*left.height; leftIndex++)
|
||||
{
|
||||
destRecord[i % dest.height] += leftRecord[i] * rightRecord[i / leftSourceHeight];
|
||||
}
|
||||
|
||||
rightRecord = &rightRecord[right.height];
|
||||
destRecord = &destRecord[dest.height];
|
||||
|
||||
}
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpNegate_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpNegate_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +62,25 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpNegate_00024MatrixOpDire
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, numElements, transposeDest);
|
||||
|
||||
dest.configureBuffer(source);
|
||||
|
||||
int * sourceRecord, * destRecord;
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = (int *) source.nextRecord();
|
||||
destRecord = (int *) dest.nextRecord();
|
||||
|
||||
// we can cheat and use the less expensive xor
|
||||
// to switch the sign bit of the float
|
||||
// single precision format 1 - sign 8 - exponent (excess 127) 23 - mantisa
|
||||
|
||||
for (int j = 0; j < sourceWidth*sourceHeight; j++)
|
||||
destRecord[j] = sourceRecord[j] ^ 0x80000000;
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpNegate_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpNegate_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +62,23 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpNegate_00024MatrixOpSafe
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, numElements, transposeDest);
|
||||
|
||||
int * sourceRecord, * destRecord;
|
||||
|
||||
for (int i = 0; i < source.elements; i++)
|
||||
{
|
||||
sourceRecord = (int *) source.nextRecord();
|
||||
destRecord = (int *) dest.nextRecord();
|
||||
|
||||
// we can cheat and use the less expensive xor
|
||||
// to switch the sign bit of the float
|
||||
// single precision format 1 - sign 8 - exponent (excess 127) 23 - mantisa
|
||||
|
||||
for (int j = 0; j < sourceWidth*sourceHeight; j++)
|
||||
destRecord[j] = sourceRecord[j] ^ 0x80000000;
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,11 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpNormalise_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
#include <cmath>
|
||||
|
||||
using namespace std;
|
||||
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpNormalise_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +66,32 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpNormalise_00024MatrixOpD
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(source);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
float magnitude, magnitude_squared;
|
||||
|
||||
int i;
|
||||
register int j;
|
||||
|
||||
for (i = 0; i < source.elements; i++)
|
||||
{
|
||||
|
||||
magnitude_squared = 0;
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (j = 0 ; j < sourceWidth*sourceHeight; i++)
|
||||
magnitude_squared += sourceRecord[j] * sourceRecord[j];
|
||||
|
||||
magnitude = (float) sqrt((double) magnitude_squared);
|
||||
|
||||
for (j = 0; j < sourceWidth*sourceHeight; i++)
|
||||
destRecord[j] = sourceRecord[j] / magnitude;
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,11 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpSubtract_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
#include <cmath>
|
||||
|
||||
using namespace std;
|
||||
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpNormalise_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -61,6 +66,30 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpNormalise_00024MatrixOpS
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * source = (float *) sourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
SrcMatrix source (sourceAddress, sourceStride, sourceWidth, sourceHeight, numElements, transposeSource);
|
||||
DstMatrix dest (destAddress, destStride, source.width, source.height, source.elements, transposeDest);
|
||||
|
||||
float * sourceRecord, * destRecord;
|
||||
float magnitude, magnitude_squared;
|
||||
|
||||
int i;
|
||||
register int j;
|
||||
|
||||
for (i = 0; i < source.elements; i++)
|
||||
{
|
||||
|
||||
magnitude_squared = 0;
|
||||
sourceRecord = source.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (j = 0 ; j < sourceWidth*sourceHeight; i++)
|
||||
magnitude_squared += sourceRecord[j] * sourceRecord[j];
|
||||
|
||||
magnitude = (float) sqrt((double) magnitude_squared);
|
||||
|
||||
for (j = 0; j < sourceWidth*sourceHeight; i++)
|
||||
destRecord[j] = sourceRecord[j] / magnitude;
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
|
|
|
|||
|
|
@ -41,6 +41,7 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpSubtract_MatrixOpDirect.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpSubtract_MatrixOpDirect
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +68,41 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpSubtract_00024MatrixOpDi
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
if (transposeLeftSource && transposeRightSource)
|
||||
{
|
||||
transposeLeftSource = false;
|
||||
transposeRightSource = false;
|
||||
transposeDest = !transposeDest;
|
||||
}
|
||||
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
left.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
dest.configureBuffer(left, right);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < left.elements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
|
||||
right.rewind();
|
||||
for (int j = 0; j < right.elements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (int k = (left.width * left.height) - 1; k >= 0; k--)
|
||||
destRecord[k] = leftRecord[k] - rightRecord[k];
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
|
|
|||
|
|
@ -41,6 +41,8 @@
|
|||
|
||||
#include <windows.h>
|
||||
#include "org_lwjgl_Math_MatrixOpSubtract_MatrixOpSafe.h"
|
||||
#include "MatrixOpCommon.h"
|
||||
|
||||
/*
|
||||
* Class: org_lwjgl_Math_MatrixOpSubtract_MatrixOpSafe
|
||||
* Method: execute
|
||||
|
|
@ -67,9 +69,40 @@ JNIEXPORT void JNICALL Java_org_lwjgl_Math_00024MatrixOpSubtract_00024MatrixOpSa
|
|||
jboolean transposeDest
|
||||
)
|
||||
{
|
||||
float * leftSource = (float *) leftSourceAddress;
|
||||
float * rightSource = (float *) rightSourceAddress;
|
||||
float * dest = (float *) destAddress;
|
||||
if (transposeLeftSource && transposeRightSource)
|
||||
{
|
||||
transposeLeftSource = false;
|
||||
transposeRightSource = false;
|
||||
transposeDest = !transposeDest;
|
||||
}
|
||||
|
||||
SrcMatrix left (leftSourceAddress, leftSourceStride,
|
||||
leftSourceWidth, leftSourceHeight, leftElements, transposeLeftSource);
|
||||
SrcMatrix right (rightSourceAddress, leftSourceStride,
|
||||
rightSourceWidth, rightSourceHeight, rightElements, transposeRightSource);
|
||||
DstMatrix dest (destAddress, destStride,
|
||||
left.width, left.height, left.elements * right.elements, transposeDest);
|
||||
|
||||
float * leftRecord, * rightRecord, * destRecord;
|
||||
|
||||
left.rewind();
|
||||
for (int i = 0; i < left.elements; i++)
|
||||
{
|
||||
leftRecord = left.nextRecord();
|
||||
|
||||
right.rewind();
|
||||
for (int j = 0; j < right.elements; j++)
|
||||
{
|
||||
rightRecord = right.nextRecord();
|
||||
destRecord = dest.nextRecord();
|
||||
|
||||
for (int k = (left.width * left.height) - 1; k >= 0; k--)
|
||||
destRecord[k] = leftRecord[k] - rightRecord[k];
|
||||
|
||||
dest.writeRecord();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
|
||||
|
|
|
|||
Loading…
Reference in New Issue