os/graphics/m3g/m3gcore11/src/m3g_math.c
changeset 0 bde4ae8d615e
     1.1 --- /dev/null	Thu Jan 01 00:00:00 1970 +0000
     1.2 +++ b/os/graphics/m3g/m3gcore11/src/m3g_math.c	Fri Jun 15 03:10:57 2012 +0200
     1.3 @@ -0,0 +1,3132 @@
     1.4 +/*
     1.5 +* Copyright (c) 2003 Nokia Corporation and/or its subsidiary(-ies).
     1.6 +* All rights reserved.
     1.7 +* This component and the accompanying materials are made available
     1.8 +* under the terms of the License "Eclipse Public License v1.0"
     1.9 +* which accompanies this distribution, and is available
    1.10 +* at the URL "http://www.eclipse.org/legal/epl-v10.html".
    1.11 +*
    1.12 +* Initial Contributors:
    1.13 +* Nokia Corporation - initial contribution.
    1.14 +*
    1.15 +* Contributors:
    1.16 +*
    1.17 +* Description: Internal math function implementations
    1.18 +*
    1.19 +*/
    1.20 +
    1.21 +
    1.22 +/*!
    1.23 + * \internal
    1.24 + * \file
    1.25 + * \brief Internal math function implementations
    1.26 + *
    1.27 + */
    1.28 +
    1.29 +#ifndef M3G_CORE_INCLUDE
    1.30 +#   error included by m3g_core.c; do not compile separately.
    1.31 +#endif
    1.32 +
    1.33 +#include "m3g_defs.h"
    1.34 +#include "m3g_memory.h"
    1.35 +
    1.36 +#if defined(M3G_SOFT_FLOAT)
    1.37 +#include <math.h>
    1.38 +#endif
    1.39 +
    1.40 +/*----------------------------------------------------------------------
    1.41 + * Private types and definitions
    1.42 + *--------------------------------------------------------------------*/
    1.43 +
    1.44 +/* Enumerated common matrix classifications */
    1.45 +#define MC_IDENTITY             0x40100401 // 01000000 00010000 00000100 00000001
    1.46 +#define MC_FRUSTUM              0x30BF0C03 // 00110000 10111111 00001100 00000011
    1.47 +#define MC_PERSPECTIVE          0x30B00C03 // 00110000 10110000 00001100 00000011
    1.48 +#define MC_ORTHO                0x7F300C03 // 01111111 00110000 00001100 00000011
    1.49 +#define MC_PARALLEL             0x70300C03 // 01110000 00110000 00001100 00000011
    1.50 +#define MC_SCALING_ROTATION     0x403F3F3F // 01000000 00111111 00111111 00111111
    1.51 +#define MC_SCALING              0x40300C03 // 01000000 00110000 00001100 00000011
    1.52 +#define MC_TRANSLATION          0x7F100401 // 01111111 00010000 00000100 00000001
    1.53 +#define MC_X_ROTATION           0x403C3C01 // 01000000 00111100 00111100 00000001
    1.54 +#define MC_Y_ROTATION           0x40330433 // 01000000 00110011 00000100 00110011
    1.55 +#define MC_Z_ROTATION           0x40100F0F // 01000000 00010000 00001111 00001111
    1.56 +#define MC_W_UNITY              0x7F3F3F3F // 01111111 00111111 00111111 00111111
    1.57 +#define MC_GENERIC              0xFFFFFFFF
    1.58 +
    1.59 +/* Partial masks for individual matrix components */
    1.60 +#define MC_TRANSLATION_PART     0x3F000000 // 00111111 00000000 00000000 00000000
    1.61 +#define MC_SCALE_PART           0x00300C03 // 00000000 00110000 00001100 00000011
    1.62 +#define MC_SCALE_ROTATION_PART  0x003F3F3F // 00000000 00111111 00111111 00111111
    1.63 +
    1.64 +/* Matrix element classification masks */
    1.65 +#define ELEM_ZERO       0x00
    1.66 +#define ELEM_ONE        0x01
    1.67 +#define ELEM_MINUS_ONE  0x02
    1.68 +#define ELEM_ANY        0x03
    1.69 +
    1.70 +/*!
    1.71 + * \internal
    1.72 + * \brief calculates the offset of a 4x4 matrix element in a linear
    1.73 + * array
    1.74 + *
    1.75 + * \notes The current convention is column-major, as in OpenGL ES
    1.76 + */
    1.77 +#define MELEM(row, col) ((row) + ((col) << 2))
    1.78 +
    1.79 +/*!
    1.80 + * \internal
    1.81 + * \brief Macro for accessing 4x4 float matrix elements
    1.82 + *
    1.83 + * \param mtx pointer to the first element of the matrix
    1.84 + * \param row matrix row
    1.85 + * \param col matrix column
    1.86 + */
    1.87 +#define M44F(mtx, row, col)     ((mtx)->elem[MELEM((row), (col))])
    1.88 +
    1.89 +/*--------------------------------------------------------------------*/
    1.90 +
    1.91 +/*----------------------------------------------------------------------
    1.92 + * Private functions
    1.93 + *--------------------------------------------------------------------*/
    1.94 +
    1.95 +
    1.96 +/*!
    1.97 + * \internal
    1.98 + * \brief ARM VFPv2 implementation of 4x4 matrix multiplication.
    1.99 + *
   1.100 + * \param dst	multiplication result
   1.101 + * \param left	left-hand matrix
   1.102 + * \param right right-hand matrix
   1.103 + */
   1.104 +#if defined(M3G_HW_FLOAT_VFPV2)
   1.105 +
   1.106 +__asm void _m3gGenericMatrixProduct(Matrix *dst,
   1.107 +                                    const Matrix *left, const Matrix *right)
   1.108 +{
   1.109 +// r0 = *dst
   1.110 +// r1 = *left
   1.111 +// r2 = *right
   1.112 +
   1.113 +		CODE32
   1.114 +
   1.115 +// save the VFP registers and set the vector STRIDE = 1, LENGTH = 4
   1.116 +
   1.117 +		FSTMFDD	sp!, {d8-d15}
   1.118 +
   1.119 +		FMRX	r3, FPSCR		 
   1.120 +		BIC		r12, r3, #0x00370000
   1.121 +		ORR		r12, #0x00030000
   1.122 +		FMXR	FPSCR, r12
   1.123 +
   1.124 +// left = [a0  a1  a2  a3	right = [b0  b1  b2  b3
   1.125 +//	 	   a4  a5  a6  a7		     b4  b5  b6  b7
   1.126 +//		   a8  a9 a10 a11		     b8  b9 b10 b11
   1.127 +//		  a12 a13 a14 a15]		    b12 b13 b14 b15]
   1.128 +//
   1.129 +// dst = [a0*b0+a4*b1+a8*b2+a12*b3  a1*b0+a5*b1+a9*b2+a13*b3  ..
   1.130 +//		  a0*b4+a4*b5+a8*b6+a12*b7  a1*b4+a5*b5+a9*b6+a13*b7  ..
   1.131 +//					.							.
   1.132 +//					.							.
   1.133 +
   1.134 +		FLDMIAS	r1!, {s8-s23}		// load the left matrix [a0-a15] to registers s8-s23
   1.135 +		FLDMIAS	r2!, {s0-s7}		// load [b0-b7] of right matrix to registers s0-s7
   1.136 +		FMULS	s24, s8, s0			// [s24-s27]  = [a0*b0  a1*b0  a2*b0  a3*b0]
   1.137 +		FMULS	s28, s8, s4			// [s28-s31]  = [a0*b4  a1*b4  a2*b4  a3*b4]
   1.138 +		FMACS	s24, s12, s1		// [s24-s27] += [a4*b1  a5*b1  a6*b1  a7*b1]
   1.139 +		FMACS	s28, s12, s5		// [s28-s31] += [a4*b5  a5*b5  a6*b5  a7*b5]
   1.140 +		FMACS	s24, s16, s2		// [s24-s27] += [a8*b2  a9*b2  a10*b2 a11*b2]
   1.141 +		FMACS	s28, s16, s6		// [s28-s31] += [a8*b6  a9*b6  a10*b6 a11*b6]
   1.142 +		FMACS	s24, s20, s3		// [s24-s27] += [a12*b3 a13*b3 a14*b3 a15*b3]
   1.143 +		FMACS	s28, s20, s7		// [s28-s31] += [a12*b7 a13*b37a14*b7 a15*b7]
   1.144 +		FLDMIAS	r2!, {s0-s7}		// load [b8-b15]
   1.145 +		FSTMIAS	r0!, {s24-s31}		// write [dst0-dst7]
   1.146 +		FMULS	s24, s8, s0			
   1.147 +		FMULS	s28, s8, s4			
   1.148 +		FMACS	s24, s12, s1		
   1.149 +		FMACS	s28, s12, s5		
   1.150 +		FMACS	s24, s16, s2		
   1.151 +		FMACS	s28, s16, s6		
   1.152 +		FMACS	s24, s20, s3		
   1.153 +		FMACS	s28, s20, s7		
   1.154 +		FSTMIAS	r0!, {s24-s31}
   1.155 +
   1.156 +// Restore the VFP registers and return.
   1.157 +
   1.158 +		FMXR	FPSCR, r3
   1.159 +
   1.160 +		FLDMFDD	sp!, {d8-d15}	
   1.161 +
   1.162 +		BX		lr
   1.163 +
   1.164 +}
   1.165 +#endif /* #if defined(M3G_HW_FLOAT_VFPV2) */
   1.166 +
   1.167 +
   1.168 +/*------------------ Elementary float ------------------*/
   1.169 +
   1.170 +#if defined(M3G_SOFT_FLOAT)
   1.171 +
   1.172 +#if defined (M3G_BUILD_ARM)
   1.173 +
   1.174 +/*!
   1.175 + * \internal
   1.176 + * \brief Floating point multiplication implementation for ARM.
   1.177 + */
   1.178 +__asm M3Gfloat m3gMul(const M3Gfloat a,
   1.179 +                      const M3Gfloat b)
   1.180 +{
   1.181 +    /**
   1.182 +     * Extract the exponents of the multiplicands and add them
   1.183 +     * together. Flush to zero if either exponent or their sum
   1.184 +     * is zero.
   1.185 +     */
   1.186 +
   1.187 +    mov     r12, #0xff;
   1.188 +    ands    r2, r0, r12, lsl #23;   // exit if e1 == 0
   1.189 +    andnes  r3, r1, r12, lsl #23;   // exit if e2 == 0
   1.190 +    subne   r2, r2, #(127 << 23);
   1.191 +    addnes  r12, r2, r3;            // exit if e1+e2-127 <= 0
   1.192 +    movle   r0, #0;
   1.193 +    bxle    lr;
   1.194 +
   1.195 +    /**
   1.196 +     * Determine the sign of the result. Note that the exponent
   1.197 +     * may have overflowed to the sign bit, and thus the result
   1.198 +     * may be an arbitrary negative value when it really should
   1.199 +     * be +Inf or -Inf.
   1.200 +     */
   1.201 +
   1.202 +    teq     r0, r1;
   1.203 +    orrmi   r12, r12, #0x80000000;
   1.204 +
   1.205 +    /**
   1.206 +     * Multiply the mantissas. First shift the mantissas up to
   1.207 +     * unsigned 1.31 fixed point, adding the leading "1" bit at
   1.208 +     * the same time, and finally do a 32x32 -> 64 bit unsigned
   1.209 +     * multiplication. The result is in unsigned 2.62 fixed point,
   1.210 +     * representing the interval [1.0, 4.0).
   1.211 +     */
   1.212 +
   1.213 +    mov     r2, #0x80000000;
   1.214 +    orr     r0, r2, r0, lsl #8;
   1.215 +    orr     r1, r2, r1, lsl #8;
   1.216 +    umulls  r2, r3, r0, r1;
   1.217 +
   1.218 +    /**
   1.219 +     * If the highest bit of the 64-bit result is set, then the
   1.220 +     * mantissa lies in [2.0, 4.0) and needs to be renormalized.
   1.221 +     * That is, the mantissa is shifted one bit to the right and
   1.222 +     * the exponent correspondingly increased by 1. Note that we
   1.223 +     * lose the leading "1" bit from the mantissa by adding it up
   1.224 +     * with the exponent.
   1.225 +     */
   1.226 +
   1.227 +    subpl   r12, r12, #(1 << 23);    // no overflow: exponent -= 1
   1.228 +    addpl   r0, r12, r3, lsr #7;     // no overflow: exponent += 1
   1.229 +    addmi   r0, r12, r3, lsr #8;     // overflow: exponent += 1
   1.230 +    bx      lr;
   1.231 +}
   1.232 +
   1.233 +/*!
   1.234 + * \internal
   1.235 + * \brief Floating point addition implementation for ARM.
   1.236 + */
   1.237 +__asm M3Gfloat m3gAdd(const M3Gfloat a,
   1.238 +                      const M3Gfloat b)
   1.239 +{
   1.240 +    /**
   1.241 +     * If the operands have opposite signs then this is not really
   1.242 +     * an addition but a subtraction. Subtraction is much slower,
   1.243 +     * so we have a separate code path for it, rather than trying
   1.244 +     * to save space by handling both in the same place.
   1.245 +     */
   1.246 +    
   1.247 +    teq     r0, r1;
   1.248 +    bmi     _m3gSub;
   1.249 +
   1.250 +    /**
   1.251 +     * Sort the operands such that the larger operand is in R0 and
   1.252 +     * the smaller in R1. The sign bits do not affect the ordering,
   1.253 +     * since they are known to be equal.
   1.254 +     */
   1.255 +    
   1.256 +    subs    r2, r0, r1;
   1.257 +    submi   r0, r0, r2;
   1.258 +    addmi   r1, r1, r2;
   1.259 +
   1.260 +    /**
   1.261 +     * Extract the exponent of the smaller operand into R2 and compute
   1.262 +     * the difference between the larger and smaller exponent into R3.
   1.263 +     * (Note that the sign bits cancel out in the subtraction.) The
   1.264 +     * exponent delta tells how many bits the mantissa of the smaller
   1.265 +     * operand must be shifted to the right in order to bring the
   1.266 +     * operands into equal scale.
   1.267 +     */
   1.268 +
   1.269 +    mov     r2, r1, lsr #23;
   1.270 +    rsb     r3, r2, r0, lsr #23;
   1.271 +
   1.272 +    /**
   1.273 +     * Check if the exponent delta is bigger than 23 bits, or if the
   1.274 +     * smaller exponent is zero. In either case, exit the routine and
   1.275 +     * return the larger operand (which is already in R0). Note that
   1.276 +     * this means that subnormals are treated as zero.
   1.277 +     */
   1.278 +
   1.279 +    rsbs    r12, r3, #23;               // N set, V clear if R3 > 23
   1.280 +    tstpl   r2, #0xff;                  // execute only if R3 <= 23
   1.281 +    bxle    lr;                         // exit if Z set or N != V
   1.282 +
   1.283 +    /**
   1.284 +     * Extract the mantissas and shift them up to unsigned 1.31 fixed
   1.285 +     * point, inserting the implied leading "1" bit at the same time.
   1.286 +     * Finally, align the decimal points and add up the mantissas.
   1.287 +     */
   1.288 +    
   1.289 +    mov     r12, #0x80000000;
   1.290 +    orr     r0, r12, r0, lsl #8;
   1.291 +    orr     r1, r12, r1, lsl #8;
   1.292 +    adds    r0, r0, r1, lsr r3;
   1.293 +
   1.294 +    /**
   1.295 +     * Compute the final exponent by adding up the smaller exponent
   1.296 +     * (R2), the exponent delta (R3), and the possible overflow bit
   1.297 +     * (carry flag). Note that in case of overflow, the leading "1"
   1.298 +     * has ended up in the carry flag and thus needs not be explicitly
   1.299 +     * discarded. Finally, put the mantissa together with the sign and
   1.300 +     * exponent.
   1.301 +     */
   1.302 +
   1.303 +    adc     r2, r2, r3;                 // r2 = smallExp + deltaExp + overflow
   1.304 +    movcc   r0, r0, lsl #1;             // no overflow: discard leading 1
   1.305 +    mov     r0, r0, lsr #9;
   1.306 +    orr     r0, r0, r2, lsl #23;
   1.307 +    bx      lr;
   1.308 +    
   1.309 +_m3gSub
   1.310 +
   1.311 +    /**
   1.312 +     * Sort the operands such that the one with larger magnitude is
   1.313 +     * in R0 and has the correct sign (the sign of the final result),
   1.314 +     * while the smaller operand is in R1 with an inverted sign bit.
   1.315 +     */
   1.316 +    
   1.317 +    eor     r1, r1, #0x80000000;
   1.318 +    subs    r2, r0, r1;
   1.319 +    eormi   r2, r2, #0x80000000;
   1.320 +    submi   r0, r0, r2;
   1.321 +    addmi   r1, r1, r2;
   1.322 +
   1.323 +    /**
   1.324 +     * Extract the exponent of the smaller operand into R2 and compute
   1.325 +     * the difference between the larger and smaller exponent into R3.
   1.326 +     * (Note that the sign bits cancel out in the subtraction.) The
   1.327 +     * exponent delta tells how many bits the mantissa of the smaller
   1.328 +     * operand must be shifted to the right in order to bring the
   1.329 +     * operands into equal scale.
   1.330 +     */
   1.331 +
   1.332 +    mov     r2, r1, lsr #23;
   1.333 +    rsbs    r3, r2, r0, lsr #23;
   1.334 +
   1.335 +    /**
   1.336 +     * Check if the exponent delta is bigger than 31 bits, or if the
   1.337 +     * smaller exponent is zero. In either case, exit the routine and
   1.338 +     * return the larger operand (which is already in R0). Note that
   1.339 +     * this means that subnormals are treated as zero.
   1.340 +     */
   1.341 +
   1.342 +    rsbs    r12, r3, #31;               // N set, V clear if R3 > 31
   1.343 +    tstpl   r2, #0xff;                  // execute only if R3 <= 31
   1.344 +    bxle    lr;                         // exit if Z set or N != V
   1.345 +
   1.346 +    /**
   1.347 +     * Extract the mantissas and shift them up to unsigned 1.31 fixed
   1.348 +     * point, inserting the implied leading "1" bit at the same time.
   1.349 +     * Then align the decimal points and subtract the smaller mantissa
   1.350 +     * from the larger one.
   1.351 +     */
   1.352 +    
   1.353 +    mov     r12, #0x80000000;
   1.354 +    orr     r0, r12, r0, lsl #8;
   1.355 +    orr     r1, r12, r1, lsl #8;
   1.356 +    subs    r0, r0, r1, lsr r3;
   1.357 +
   1.358 +    /**
   1.359 +     * We split the range of possible results into three categories:
   1.360 +     * 
   1.361 +     *   1. [1.0, 2.0) ==> no renormalization needed (highest bit set)
   1.362 +     *   2. [0.5, 1.0) ==> only one left-shift needed
   1.363 +     *   3. (0.0, 0.5) ==> multiple left-shifts needed
   1.364 +     *   4. zero       ==> just return
   1.365 +     *   
   1.366 +     * Cases 1 and 2 are handled in the main code path. Cases 3 and 4
   1.367 +     * are less common by far, so we branch to a separate code fragment
   1.368 +     * for those.
   1.369 +     */
   1.370 +    
   1.371 +    movpls  r0, r0, lsl #1;             // Cases 2,3,4: shift left
   1.372 +    bpl     _m3gSubRenormalize;         // Cases 3 & 4: branch out
   1.373 +    
   1.374 +    /**
   1.375 +     * Now we have normalized the mantissa such that the highest bit
   1.376 +     * is set. Here we only need to adjust the exponent, if necessary,
   1.377 +     * and put the pieces together. Note that we lose the leading "1"
   1.378 +     * bit from the mantissa by adding it up with the exponent. We can
   1.379 +     * also do proper rounding (towards nearest) instead of truncation
   1.380 +     * (towards zero) at no extra cost!
   1.381 +     */
   1.382 +
   1.383 +    sbc     r3, r3, #1;                 // deltaExp -= 1 (Case 1) or 2 (Case 2)
   1.384 +    add     r2, r2, r3;                 // resultExp = smallExp + deltaExp
   1.385 +    movs    r0, r0, lsr #8;             // shift mantissa, keep leading "1"
   1.386 +    adc     r0, r0, r2, lsl #23;        // resultExp += 1, mantissa += carry
   1.387 +    bx      lr;
   1.388 +
   1.389 +    /**
   1.390 +     * Separate code path for cases 3 and 4 (see above). The mantissa
   1.391 +     * has already been shifted up by one, but the exponent has not
   1.392 +     * been correspondingly decreased. We also know that the highest
   1.393 +     * bit is still not set, and that the carry flag is clear.
   1.394 +     */
   1.395 +
   1.396 +_m3gSubRenormalize
   1.397 +    bxeq    lr;
   1.398 +    subcc   r3, r3, #2;
   1.399 +    movccs  r0, r0, lsl #2;
   1.400 +
   1.401 +    /**
   1.402 +     * If the carry flag is still not set, i.e. there were more than
   1.403 +     * two leading zeros in the mantissa initially, loop until we
   1.404 +     * find the highest set bit.
   1.405 +     */
   1.406 +    
   1.407 +_m3gSubRenormalizeLoop
   1.408 +    subcc   r3, r3, #1;
   1.409 +    movccs  r0, r0, lsl #1;
   1.410 +    bcc     _m3gSubRenormalizeLoop;
   1.411 +
   1.412 +    /**
   1.413 +     * Now the leading "1" is in the carry flag, so we can just add up
   1.414 +     * the exponent and mantissa as usual, doing proper rounding at
   1.415 +     * the same time. However, cases where the exponent goes negative
   1.416 +     * (that is, underflows) must be detected and flushed to zero.
   1.417 +     */
   1.418 +    
   1.419 +    add     r3, r2, r3;
   1.420 +    movs    r0, r0, lsr #9;
   1.421 +    adc     r0, r0, r3, lsl #23;
   1.422 +    teq     r0, r2, lsl #23;
   1.423 +    movmi   r0, #0;
   1.424 +    bx      lr;
   1.425 +}
   1.426 +
   1.427 +#else /* M3G_BUILD_ARM */
   1.428 +
   1.429 +/*!
   1.430 + * \internal
   1.431 + * \brief Floating point addition implementation
   1.432 + *
   1.433 + */
   1.434 +static M3G_INLINE M3Gfloat m3gFloatAdd(const M3Guint aBits,
   1.435 +                                       const M3Guint bBits)
   1.436 +{
   1.437 +    M3Guint large, small, signMask;
   1.438 +    
   1.439 +    /* Early exits for underflow cases */
   1.440 +
   1.441 +    large = (M3Gint)(aBits & ~SIGN_MASK);
   1.442 +    if (large <= 0x00800000) {
   1.443 +        return INT_AS_FLOAT(bBits);
   1.444 +    }
   1.445 +    small = (M3Gint)(bBits & ~SIGN_MASK);
   1.446 +    if (small <= 0x00800000) {
   1.447 +        return INT_AS_FLOAT(aBits);
   1.448 +    }
   1.449 +
   1.450 +    /* Swap the numbers so that "large" really is larger; the unsigned
   1.451 +     * (or de-signed) bitmasks for floats are nicely monotonous, so we
   1.452 +     * can compare directly.  Also store the sign of the larger number
   1.453 +     * for future reference. */
   1.454 +
   1.455 +    if (small > large) {
   1.456 +        M3Gint temp = small;
   1.457 +        small = large;
   1.458 +        large = temp;
   1.459 +        signMask = (bBits & SIGN_MASK);
   1.460 +    }
   1.461 +    else {
   1.462 +        signMask = (aBits & SIGN_MASK);
   1.463 +    }
   1.464 +    
   1.465 +    {
   1.466 +        M3Guint res, overflow;
   1.467 +        M3Guint resExp, expDelta;
   1.468 +        
   1.469 +        /* Store the larger exponent as our candidate result exponent,
   1.470 +         * and compute the difference between the exponents */
   1.471 +
   1.472 +        resExp = (large >> 23);
   1.473 +        expDelta = resExp - (small >> 23);
   1.474 +
   1.475 +        /* Take an early exit if the change would be insignificant;
   1.476 +         * this also guards against odd results from shifting by more
   1.477 +         * than 31 (undefined in C) */
   1.478 +
   1.479 +        if (expDelta >= 24) {
   1.480 +            res = large | signMask;
   1.481 +            return INT_AS_FLOAT(res);
   1.482 +        }
   1.483 +
   1.484 +        /* Convert the mantissas into shifted integers, and shift the
   1.485 +         * smaller number to the same scale with the larger one. */
   1.486 +
   1.487 +        large = (large & MANTISSA_MASK) | LEADING_ONE;
   1.488 +        small = (small & MANTISSA_MASK) | LEADING_ONE;
   1.489 +        small >>= expDelta;
   1.490 +        M3G_ASSERT(large >= small);
   1.491 +        
   1.492 +        /* Check whether we're really adding or subtracting the
   1.493 +         * smaller number, and branch to slightly different routines
   1.494 +         * respectively */
   1.495 +
   1.496 +        if (((aBits ^ bBits) & SIGN_MASK) == 0) {
   1.497 +
   1.498 +            /* Matching signs; just add the numbers and check for
   1.499 +             * overflow, shifting to compensate as necessary. */
   1.500 +
   1.501 +            res = large + small;
   1.502 +            
   1.503 +            overflow = (res >> 24);
   1.504 +            res >>= overflow;
   1.505 +            resExp += overflow;
   1.506 +        }
   1.507 +        else {
   1.508 +
   1.509 +            /* Different signs, so let's subtract the smaller value;
   1.510 +             * also check that we're not subtracting a number from
   1.511 +             * itself (so we don't have to normalize a zero below) */
   1.512 +            
   1.513 +            if (small == large) {
   1.514 +                return 0.0f; /* x - x = 0 */
   1.515 +            }
   1.516 +
   1.517 +            res = (large << 8) - (small << 8);
   1.518 +
   1.519 +            /* Renormalize the number by shifting until we've got the
   1.520 +             * high bit in place */
   1.521 +
   1.522 +            while ((res >> 24) == 0) {
   1.523 +                res <<= 8;
   1.524 +                resExp -= 8;
   1.525 +            }
   1.526 +            while ((res >> 31) == 0) {
   1.527 +                res <<= 1;
   1.528 +                --resExp;
   1.529 +            }
   1.530 +            res >>= 8;
   1.531 +        }
   1.532 +
   1.533 +        /* Flush to zero in case of over/underflow of the exponent */
   1.534 +
   1.535 +        if (resExp >= 255) {
   1.536 +            return 0.0f;
   1.537 +        }
   1.538 +        
   1.539 +        /* Compose the final number into "res"; note that we pull in
   1.540 +         * the sign of the original larger number, which should still
   1.541 +         * be valid */
   1.542 +
   1.543 +        res &= MANTISSA_MASK;
   1.544 +        res |= (resExp << 23);
   1.545 +        res |= signMask;
   1.546 +
   1.547 +        return INT_AS_FLOAT(res);
   1.548 +    }
   1.549 +}
   1.550 +
   1.551 +/*!
   1.552 + * \internal
   1.553 + * \brief Floating point multiplication implementation
   1.554 + */
   1.555 +static M3G_INLINE M3Gfloat m3gFloatMul(const M3Guint aBits,
   1.556 +                                       const M3Guint bBits)
   1.557 +{
   1.558 +    M3Guint a, b;
   1.559 +
   1.560 +    /* Early exits for underflow and multiplication by zero */
   1.561 +
   1.562 +    a = (aBits & ~SIGN_MASK);
   1.563 +    if (a <= 0x00800000) {
   1.564 +        return 0.0f;
   1.565 +    }
   1.566 +    
   1.567 +    b = (bBits & ~SIGN_MASK);
   1.568 +    if (b <= 0x00800000) {
   1.569 +        return 0.0f;
   1.570 +    }
   1.571 +    
   1.572 +    {
   1.573 +        M3Guint res, exponent, overflow;
   1.574 +        
   1.575 +        /* Compute the exponent of the result, assuming the mantissas
   1.576 +         * don't overflow; then mask out the original exponents */
   1.577 +        
   1.578 +        exponent = (a >> 23) + (b >> 23) - 127;
   1.579 +        a &= MANTISSA_MASK;
   1.580 +        b &= MANTISSA_MASK;
   1.581 +
   1.582 +        /* Compute the new mantissa from:
   1.583 +         *
   1.584 +         *   (1 + a)(1 + b) = ab + a + b + 1
   1.585 +         *   
   1.586 +         * First shift the mantissas from 0.23 down to 0.16 for the
   1.587 +         * multiplication, then shift back to 0.23 for adding in the
   1.588 +         * "a + b + 1" part of the equation.  */
   1.589 +        
   1.590 +        res = (a >> 7) * (b >> 7);              /* "ab" at 0.32 */
   1.591 +        res = (res >> 9) + a + b + LEADING_ONE;
   1.592 +
   1.593 +        /* Add the leading one, then normalize the result by checking
   1.594 +         * the overflow bit and dividing by two if necessary */
   1.595 +
   1.596 +        overflow = (res >> 24);
   1.597 +        res >>= overflow;
   1.598 +        exponent += overflow;
   1.599 +
   1.600 +        /* Flush to zero in case of over/underflow of the exponent */
   1.601 +
   1.602 +        if (exponent >= 255) {
   1.603 +            return 0.0f;
   1.604 +        }
   1.605 +
   1.606 +        /* Compose the final number into "res" */
   1.607 +
   1.608 +        res &= MANTISSA_MASK;
   1.609 +        res |= (exponent << 23);
   1.610 +        res |= (aBits ^ bBits) & SIGN_MASK;
   1.611 +
   1.612 +        return INT_AS_FLOAT(res);
   1.613 +    }
   1.614 +}
   1.615 +
   1.616 +#endif /* M3G_BUILD_ARM */
   1.617 +
   1.618 +/*!
   1.619 + * \internal
   1.620 + * \brief Computes the signed fractional part of a floating point
   1.621 + * number
   1.622 + *
   1.623 + * \param x  floating point value
   1.624 + * \return x signed fraction of x in ]-1, 1[
   1.625 + */
   1.626 +static M3Gfloat m3gFrac(M3Gfloat x)
   1.627 +{
   1.628 +    /* Quick exit for -1 < x < 1 */
   1.629 +    
   1.630 +    if (m3gAbs(x) < 1.0f) {
   1.631 +        return x;
   1.632 +    }
   1.633 +
   1.634 +    /* Shift the mantissa to the proper place, mask out the integer
   1.635 +     * part, and renormalize */
   1.636 +    {
   1.637 +        M3Guint ix = FLOAT_AS_UINT(x);
   1.638 +        M3Gint expo = ((ix >> 23) & 0xFF) - 127;
   1.639 +        M3G_ASSERT(expo >= 0);
   1.640 +
   1.641 +        /* The decimal part will always be zero for large values with
   1.642 +         * exponents over 24 */
   1.643 +        
   1.644 +        if (expo >= 24) {
   1.645 +            return 0.f;
   1.646 +        }
   1.647 +        else {
   1.648 +            
   1.649 +            /* Shift the integer part out of the mantissa and see what
   1.650 +             * we have left */
   1.651 +            
   1.652 +            M3Guint base = (ix & MANTISSA_MASK) | LEADING_ONE;
   1.653 +            base = (base << expo) & MANTISSA_MASK;
   1.654 +
   1.655 +            /* Quick exit (and guard against infinite looping) for
   1.656 +             * zero */
   1.657 +
   1.658 +            if (base == 0) {
   1.659 +                return 0.f;
   1.660 +            }
   1.661 +
   1.662 +            /* We now have an exponent of 0 (i.e. no shifting), but
   1.663 +             * must renormalize to get a set bit in place of the
   1.664 +             * hidden (implicit one) bit */
   1.665 +            
   1.666 +            expo = 0;
   1.667 +            
   1.668 +            while ((base >> 19) == 0) {
   1.669 +                base <<= 4;
   1.670 +                expo -= 4;
   1.671 +            }
   1.672 +            while ((base >> 23) == 0) {
   1.673 +                base <<= 1;
   1.674 +                --expo;
   1.675 +            }
   1.676 +
   1.677 +            /* Compose the final number */
   1.678 +
   1.679 +            ix =
   1.680 +                (base & MANTISSA_MASK) |
   1.681 +                ((expo + 127) << 23) |
   1.682 +                (ix & SIGN_MASK);
   1.683 +            return INT_AS_FLOAT(ix);
   1.684 +        }
   1.685 +    }
   1.686 +}
   1.687 +
   1.688 +#endif /* M3G_SOFT_FLOAT */
   1.689 +
   1.690 +#if defined(M3G_DEBUG)
   1.691 +/*!
   1.692 + * \internal
   1.693 + * \brief Checks for NaN or infinity in a floating point input
   1.694 + */
   1.695 +static void m3gValidateFloats(int n, float *p)
   1.696 +{
   1.697 +    while (n-- > 0) {
   1.698 +        M3G_ASSERT(EXPONENT(*p) < 120);
   1.699 +        ++p;
   1.700 +    }
   1.701 +}
   1.702 +#else
   1.703 +#   define m3gValidateFloats(n, p)
   1.704 +#endif
   1.705 +    
   1.706 +/*------------------ Trigonometry and exp ----------*/
   1.707 +
   1.708 +
   1.709 +#if defined(M3G_SOFT_FLOAT)
   1.710 +/*!
   1.711 + * \internal
   1.712 + * \brief Sine for the first quadrant
   1.713 + *
   1.714 + * \param x floating point value in [0, PI/2]
   1.715 + * \return sine of \c x
   1.716 + */
   1.717 +static M3Gfloat m3gSinFirstQuadrant(M3Gfloat x)
   1.718 +{
   1.719 +    M3Guint bits = FLOAT_AS_UINT(x);
   1.720 +    
   1.721 +    if (bits <= 0x3ba3d70au)    /* 0.005f */
   1.722 +        return x;
   1.723 +    else {
   1.724 +        static const M3Gfloat sinTermLut[4] = {
   1.725 +            -1.0f / (2*3),
   1.726 +            -1.0f / (4*5),
   1.727 +            -1.0f / (6*7),
   1.728 +            -1.0f / (8*9)
   1.729 +        };
   1.730 +
   1.731 +        M3Gfloat xx = m3gSquare(x);
   1.732 +        M3Gfloat sinx = x;
   1.733 +        int i;
   1.734 +
   1.735 +        for (i = 0; i < 4; ++i) {
   1.736 +            x    = m3gMul(x, m3gMul(xx, sinTermLut[i]));
   1.737 +            sinx = m3gAdd(sinx, x);
   1.738 +        }
   1.739 +
   1.740 +        return sinx;
   1.741 +    }
   1.742 +}
   1.743 +#endif /* M3G_SOFT_FLOAT */
   1.744 +
   1.745 +#if defined(M3G_SOFT_FLOAT)
   1.746 +/*!
   1.747 + * \internal
   1.748 + * \brief Computes sine for the first period
   1.749 + *
   1.750 + * \param x floating point value in [0, 2PI]
   1.751 + * \return sine of x
   1.752 + */
   1.753 +static M3Gfloat m3gSinFirstPeriod(const M3Gfloat x)
   1.754 +{
   1.755 +    M3G_ASSERT(x >= 0 && x <= TWO_PI);
   1.756 +    
   1.757 +    if (x >= PI) {
   1.758 +        return m3gNegate(m3gSinFirstQuadrant(x >= ONE_AND_HALF_PI ?
   1.759 +                                             m3gSub(TWO_PI, x) :
   1.760 +                                             m3gSub(x, PI)));
   1.761 +    }
   1.762 +    return m3gSinFirstQuadrant((x >= HALF_PI) ? m3gSub(PI, x) : x);
   1.763 +}
   1.764 +#endif /* M3G_SOFT_FLOAT */
   1.765 +
   1.766 +/*------------- Float vs. int conversion helpers -------------*/
   1.767 +
   1.768 +/*!
   1.769 + * \internal
   1.770 + * \brief Scales and clamps a float to unsigned byte range
   1.771 + */
   1.772 +static M3G_INLINE M3Gint m3gFloatToByte(const M3Gfloat a)
   1.773 +{
   1.774 +    return m3gRoundToInt(m3gMul(255.f, m3gClampFloat(a, 0.f, 1.f)));
   1.775 +}
   1.776 +
   1.777 +/*------------------ Vector helpers ------------------*/
   1.778 +
   1.779 +/*!
   1.780 + * \internal
   1.781 + * \brief Computes the norm of a floating-point 3-vector
   1.782 + */
   1.783 +static M3G_INLINE M3Gfloat m3gNorm3(const M3Gfloat *v)
   1.784 +{
   1.785 +    return m3gAdd(m3gAdd(m3gSquare(v[0]), m3gSquare(v[1])),
   1.786 +                  m3gSquare(v[2]));
   1.787 +}
   1.788 +
   1.789 +/*!
   1.790 + * \internal
   1.791 + * \brief Computes the norm of a floating-point 4-vector
   1.792 + */
   1.793 +static M3G_INLINE M3Gfloat m3gNorm4(const M3Gfloat *v)
   1.794 +{
   1.795 +    return m3gAdd(m3gAdd(m3gSquare(v[0]), m3gSquare(v[1])),
   1.796 +                  m3gAdd(m3gSquare(v[2]), m3gSquare(v[3])));
   1.797 +}
   1.798 +
   1.799 +/*!
   1.800 + * \internal
   1.801 + * \brief Scales a floating-point 3-vector
   1.802 + */
   1.803 +static M3G_INLINE void m3gScale3(M3Gfloat *v, M3Gfloat s)
   1.804 +{
   1.805 +    v[0] = m3gMul(v[0], s);
   1.806 +    v[1] = m3gMul(v[1], s);
   1.807 +    v[2] = m3gMul(v[2], s);
   1.808 +}
   1.809 +
   1.810 +/*!
   1.811 + * \internal
   1.812 + * \brief Scales a floating-point 4-vector
   1.813 + */
   1.814 +static M3G_INLINE void m3gScale4(M3Gfloat *v, M3Gfloat s)
   1.815 +{
   1.816 +    v[0] = m3gMul(v[0], s);
   1.817 +    v[1] = m3gMul(v[1], s);
   1.818 +    v[2] = m3gMul(v[2], s);
   1.819 +    v[3] = m3gMul(v[3], s);
   1.820 +}
   1.821 +
   1.822 +
   1.823 +/*------------------ Matrices ------------------*/
   1.824 +
   1.825 +/*!
   1.826 + * \internal
   1.827 + */
   1.828 +static M3G_INLINE M3Gbool m3gIsClassified(const Matrix *mtx)
   1.829 +{
   1.830 +    M3G_ASSERT(mtx != NULL);
   1.831 +    return (M3Gbool) mtx->classified;
   1.832 +}
   1.833 +
   1.834 +/*!
   1.835 + * \internal
   1.836 + * \brief Returns a classification for a single floating point number
   1.837 + */
   1.838 +static M3G_INLINE M3Guint m3gElementClass(const M3Gfloat x)
   1.839 +{
   1.840 +    if (IS_ZERO(x)) {
   1.841 +        return ELEM_ZERO;
   1.842 +    }
   1.843 +    else if (IS_ONE(x)) {
   1.844 +        return ELEM_ONE;
   1.845 +    }
   1.846 +    else if (IS_MINUS_ONE(x)) {
   1.847 +        return ELEM_MINUS_ONE;
   1.848 +    }
   1.849 +    return ELEM_ANY;
   1.850 +}
   1.851 +
   1.852 +/*!
   1.853 + * \internal
   1.854 + * \brief Computes the classification mask of a matrix
   1.855 + *
   1.856 + * The mask is constructed from two bits per elements, with the lowest
   1.857 + * two bits corresponding to the first element in the \c elem array of
   1.858 + * the matrix.
   1.859 + */
   1.860 +static void m3gClassify(Matrix *mtx)
   1.861 +{
   1.862 +    M3Guint mask = 0;
   1.863 +    const M3Gfloat *p;
   1.864 +    int i;
   1.865 +
   1.866 +    M3G_ASSERT(mtx != NULL);
   1.867 +    M3G_ASSERT(!m3gIsClassified(mtx));
   1.868 +
   1.869 +    p = mtx->elem;
   1.870 +    for (i = 0; i < 16; ++i) {
   1.871 +        M3Gfloat elem = *p++;
   1.872 +        mask |= (m3gElementClass(elem) << (i*2));
   1.873 +    }
   1.874 +    mtx->mask = mask;
   1.875 +    mtx->classified = M3G_TRUE;
   1.876 +}
   1.877 +
   1.878 +/*!
   1.879 + * \internal
   1.880 + * \brief Sets matrix classification directly
   1.881 + */
   1.882 +static M3G_INLINE void m3gClassifyAs(M3Guint mask, Matrix *mtx)
   1.883 +{
   1.884 +    M3G_ASSERT(mtx != NULL);
   1.885 +    mtx->mask = mask;
   1.886 +    mtx->classified = M3G_TRUE;
   1.887 +    mtx->complete = M3G_FALSE;
   1.888 +}
   1.889 +
   1.890 +/*!
   1.891 + * \internal
   1.892 + * \brief Attempts to classify a matrix more precisely
   1.893 + *
   1.894 + * Tries to classify all "free" elements of a matrix into one of the
   1.895 + * predefined constants to improve precision and performance in
   1.896 + * subsequent calculations.
   1.897 + */
   1.898 +static void m3gSubClassify(Matrix *mtx)
   1.899 +{
   1.900 +    M3G_ASSERT_PTR(mtx);
   1.901 +    M3G_ASSERT(m3gIsClassified(mtx));
   1.902 +    {
   1.903 +        const M3Gfloat *p = mtx->elem;
   1.904 +        M3Guint inMask = mtx->mask;
   1.905 +        M3Guint outMask = inMask;
   1.906 +        int i;
   1.907 +
   1.908 +        for (i = 0; i < 16; ++i, inMask >>= 2) {
   1.909 +            M3Gfloat elem = *p++;
   1.910 +            if ((inMask & 0x03) == ELEM_ANY) {
   1.911 +                outMask &= ~(0x03u << (i*2));
   1.912 +                outMask |= (m3gElementClass(elem) << (i*2));
   1.913 +            }
   1.914 +        }
   1.915 +        mtx->mask = outMask;
   1.916 +    }
   1.917 +}
   1.918 +
   1.919 +/*!
   1.920 + * \internal
   1.921 + * \brief Fills in the implicit values for a classified matrix
   1.922 + */
   1.923 +static void m3gFillClassifiedMatrix(Matrix *mtx)
   1.924 +{
   1.925 +    int i;
   1.926 +    M3Guint mask;
   1.927 +    M3Gfloat *p;
   1.928 +
   1.929 +    M3G_ASSERT(mtx != NULL);
   1.930 +    M3G_ASSERT(mtx->classified);
   1.931 +    M3G_ASSERT(!mtx->complete);
   1.932 +
   1.933 +    mask = mtx->mask;
   1.934 +    p = mtx->elem;
   1.935 +
   1.936 +    for (i = 0; i < 16; ++i, mask >>= 2) {
   1.937 +        unsigned elem = (mask & 0x03);
   1.938 +        switch (elem) {
   1.939 +        case ELEM_ZERO:         *p++ =  0.0f; break;
   1.940 +        case ELEM_ONE:          *p++ =  1.0f; break;
   1.941 +        case ELEM_MINUS_ONE:    *p++ = -1.0f; break;
   1.942 +        default:                ++p;
   1.943 +        }
   1.944 +    }
   1.945 +    mtx->complete = M3G_TRUE;
   1.946 +}
   1.947 +
   1.948 +
   1.949 +#if !defined(M3G_HW_FLOAT)
   1.950 +/*!
   1.951 + * \internal
   1.952 + * \brief Performs one multiply-add of classified matrix elements
   1.953 + *
   1.954 + * \param amask element class of the first multiplicand
   1.955 + * \param a     float value of the first multiplicand
   1.956 + * \param bmask element class of the second multiplicand
   1.957 + * \param b     float value of the second multiplicand
   1.958 + * \param c     float value to add
   1.959 + * \return a * b + c
   1.960 + * 
   1.961 + * \notes inline, as only called from the matrix product function
   1.962 + */
   1.963 +static M3G_INLINE M3Gfloat m3gClassifiedMadd(
   1.964 +    const M3Gbitmask amask, const M3Gfloat *pa,
   1.965 +    const M3Gbitmask bmask, const M3Gfloat *pb,
   1.966 +    const M3Gfloat c)
   1.967 +{
   1.968 +    /* Check for zero product to reduce the switch cases below */
   1.969 +    
   1.970 +    if (amask == ELEM_ZERO || bmask == ELEM_ZERO) {
   1.971 +        return c;    
   1.972 +    }
   1.973 +
   1.974 +    /* Branch based on the classification of a */
   1.975 +    
   1.976 +    switch (amask) {
   1.977 +        
   1.978 +    case ELEM_ANY:
   1.979 +        if (bmask == ELEM_ONE) {
   1.980 +            return m3gAdd(*pa, c);      /*  a * 1 + c  =  a + c  */
   1.981 +        }
   1.982 +        if (bmask == ELEM_MINUS_ONE) {
   1.983 +            return m3gSub(c, *pa);      /*  a * -1 + c = -a + c  */
   1.984 +        }
   1.985 +        return m3gMadd(*pa, *pb, c);    /*  a * b + c            */
   1.986 +        
   1.987 +    case ELEM_ONE:
   1.988 +        if (bmask == ELEM_ONE) {
   1.989 +            return m3gAdd(c, 1.f);      /*  1 * 1 + c  = 1 + c   */
   1.990 +        }
   1.991 +        if (bmask == ELEM_MINUS_ONE) {
   1.992 +            return m3gSub(c, 1.f);      /*  1 * -1 + c = -1 + c  */
   1.993 +        }
   1.994 +        return m3gAdd(*pb, c);          /*  1 * b + c  =  b + c  */
   1.995 +        
   1.996 +    case ELEM_MINUS_ONE:
   1.997 +        if (bmask == ELEM_ONE) {
   1.998 +            return m3gSub(c, 1.f);      /* -1 * 1 + c  = -1 + c  */
   1.999 +        }
  1.1000 +        if (bmask == ELEM_MINUS_ONE) {
  1.1001 +            return m3gAdd(c, 1.f);      /* -1 * -1 + c =  1 + c  */
  1.1002 +        }
  1.1003 +        return m3gSub(c, *pb);          /* -1 * b + c  = -b + c  */
  1.1004 +        
  1.1005 +    default:
  1.1006 +        M3G_ASSERT(M3G_FALSE);
  1.1007 +        return 0.0f;
  1.1008 +    }
  1.1009 +}
  1.1010 +#endif /*!defined(M3G_HW_FLOAT)*/
  1.1011 +
  1.1012 +/*!
  1.1013 + * \internal
  1.1014 + * \brief Computes a generic 4x4 matrix product
  1.1015 + */
  1.1016 +static void m3gGenericMatrixProduct(Matrix *dst,
  1.1017 +                                    const Matrix *left, const Matrix *right)
  1.1018 +{
  1.1019 +    M3G_ASSERT(dst != NULL && left != NULL && right != NULL);
  1.1020 +
  1.1021 +    {
  1.1022 +#       if defined(M3G_HW_FLOAT)
  1.1023 +        if (!left->complete) {
  1.1024 +            m3gFillClassifiedMatrix((Matrix*)left);
  1.1025 +        }
  1.1026 +        if (!right->complete) {
  1.1027 +            m3gFillClassifiedMatrix((Matrix*)right);
  1.1028 +        }
  1.1029 +#       else
  1.1030 +        const unsigned lmask = left->mask;
  1.1031 +        const unsigned rmask = right->mask;
  1.1032 +#       endif
  1.1033 +        
  1.1034 +#if defined(M3G_HW_FLOAT_VFPV2)
  1.1035 +		_m3gGenericMatrixProduct(dst, left, right);
  1.1036 +#else	
  1.1037 +        {
  1.1038 +            int row;
  1.1039 +
  1.1040 +            for (row = 0; row < 4; ++row) {
  1.1041 +                int col;
  1.1042 +                for (col = 0; col < 4; ++col) {
  1.1043 +                    int k;
  1.1044 +                    M3Gfloat a = 0;
  1.1045 +                    for (k = 0; k < 4; ++k) {
  1.1046 +                        M3Gint lidx = MELEM(row, k);
  1.1047 +                        M3Gint ridx = MELEM(k, col);
  1.1048 +#                       if defined(M3G_HW_FLOAT)
  1.1049 +                        a = m3gMadd(left->elem[lidx], right->elem[ridx], a);
  1.1050 +#                       else
  1.1051 +                        a = m3gClassifiedMadd((lmask >> (2 * lidx)) & 3,
  1.1052 +                                              &left->elem[lidx],
  1.1053 +                                              (rmask >> (2 * ridx)) & 3,
  1.1054 +                                              &right->elem[ridx],
  1.1055 +                                              a);
  1.1056 +#                       endif /*!M3G_HW_FLOAT*/
  1.1057 +                    }
  1.1058 +                    M44F(dst, row, col) = a;
  1.1059 +                }
  1.1060 +            }
  1.1061 +        }
  1.1062 +#endif /*!M3G_HW_FLOAT_VFPV2*/
  1.1063 +    }
  1.1064 +    dst->complete = M3G_TRUE;
  1.1065 +    dst->classified = M3G_FALSE;
  1.1066 +}
  1.1067 +
  1.1068 +/*!
  1.1069 + * \internal
  1.1070 + * \brief Converts a float vector to 16-bit integers
  1.1071 + *
  1.1072 + * \param size   vector length
  1.1073 + * \param vec    pointer to float vector
  1.1074 + * \param scale  scale of output values, as the number of bits to
  1.1075 + *               shift left to get actual values
  1.1076 + * \param outVec output value vector
  1.1077 + */
  1.1078 +static void m3gFloatVecToShort(M3Gint size, const M3Gfloat *vec,
  1.1079 +                               M3Gint scale, M3Gshort *outVec)
  1.1080 +{
  1.1081 +    const M3Guint *vecInt = (const M3Guint*) vec;
  1.1082 +    M3Gint i;
  1.1083 +    
  1.1084 +    for (i = 0; i < size; ++i) {
  1.1085 +        M3Guint a = vecInt[i];
  1.1086 +        if ((a & ~SIGN_MASK) < (1 << 23)) {
  1.1087 +            *outVec++ = 0;
  1.1088 +        }
  1.1089 +        else {
  1.1090 +            M3Gint shift =
  1.1091 +                scale - ((M3Gint)((vecInt[i] >> 23) & 0xFFu) - (127 + 23));
  1.1092 +            M3G_ASSERT(shift > 8); /* or the high bits will overflow */
  1.1093 +            
  1.1094 +            if (shift > 23) {
  1.1095 +                *outVec++ = 0;
  1.1096 +            }
  1.1097 +            else {
  1.1098 +                M3Gint out =
  1.1099 +                    (M3Gint) (((a & MANTISSA_MASK) | LEADING_ONE) >> shift);
  1.1100 +                if (a >> 31) {
  1.1101 +                    out = -out;
  1.1102 +                }
  1.1103 +                M3G_ASSERT(m3gInRange(out, -32767, 32767));
  1.1104 +                *outVec++ = (M3Gshort) out;
  1.1105 +            }
  1.1106 +        }
  1.1107 +    }
  1.1108 +}
  1.1109 +
  1.1110 +/*----------------------------------------------------------------------
  1.1111 + * Internal functions
  1.1112 + *--------------------------------------------------------------------*/
  1.1113 +
  1.1114 +/*--------------------------------------------------------------------*/
  1.1115 +
  1.1116 +#if defined(M3G_SOFT_FLOAT)
  1.1117 +
  1.1118 +#if !defined (M3G_BUILD_ARM)
  1.1119 +
  1.1120 +static M3Gfloat m3gAdd(const M3Gfloat a, const M3Gfloat b)
  1.1121 +{
  1.1122 +    return m3gFloatAdd(FLOAT_AS_UINT(a), FLOAT_AS_UINT(b));
  1.1123 +}
  1.1124 +
  1.1125 +static M3Gfloat m3gMul(const M3Gfloat a, const M3Gfloat b)
  1.1126 +{
  1.1127 +    return m3gFloatMul(FLOAT_AS_UINT(a), FLOAT_AS_UINT(b));
  1.1128 +}
  1.1129 +
  1.1130 +#endif /* M3G_BUILD_ARM */
  1.1131 +
  1.1132 +/*!
  1.1133 + * \internal
  1.1134 + * \brief Computes the reciprocal of the square root
  1.1135 + *
  1.1136 + * \param x a floating point value
  1.1137 + * \return 1 / square root of \c x
  1.1138 + */
  1.1139 +static M3Gfloat m3gRcpSqrt(const M3Gfloat x)
  1.1140 +{
  1.1141 +    /* Approximation followed by Newton-Raphson iteration a'la
  1.1142 +     * "Floating-point tricks" by J. Blinn, but we iterate several
  1.1143 +     * times to improve precision */
  1.1144 +    
  1.1145 +    M3Gint i = (M3G_FLOAT_ONE + (M3G_FLOAT_ONE >> 1))
  1.1146 +        - (FLOAT_AS_UINT(x) >> 1);
  1.1147 +    M3Gfloat y = INT_AS_FLOAT(i);
  1.1148 +    for (i = 0; i < 3; ++i) {
  1.1149 +        y = m3gMul(y, m3gSub(1.5f, m3gHalf(m3gMul(x, m3gSquare(y)))));
  1.1150 +    }
  1.1151 +    return y;
  1.1152 +}
  1.1153 +
  1.1154 +/*!
  1.1155 + * \internal
  1.1156 + * \brief Computes the square root
  1.1157 + *
  1.1158 + * \param x a floating point value
  1.1159 + * \return square root of \c x
  1.1160 + */
  1.1161 +static M3Gfloat m3gSqrt(const M3Gfloat x)
  1.1162 +{
  1.1163 +    /* Approximation followed by Newton-Raphson iteration a'la
  1.1164 +     * "Floating-point tricks" by J. Blinn, but we iterate several
  1.1165 +     * times to improve precision */
  1.1166 +
  1.1167 +    M3Gint i = (FLOAT_AS_UINT(x) >> 1) + (M3G_FLOAT_ONE >> 1);
  1.1168 +    M3Gfloat y = INT_AS_FLOAT(i);
  1.1169 +    for (i = 0; i < 2; ++i) {
  1.1170 +        y = m3gDiv(m3gAdd(m3gSquare(y), x), m3gDouble(y));
  1.1171 +    }
  1.1172 +    return y;
  1.1173 +}
  1.1174 +
  1.1175 +#endif /* M3G_SOFT_FLOAT */
  1.1176 +
  1.1177 +/*--------------------------------------------------------------------*/
  1.1178 +
  1.1179 +#if defined(M3G_SOFT_FLOAT)
  1.1180 +/*!
  1.1181 + * \internal
  1.1182 + */
  1.1183 +static M3Gfloat m3gArcCos(const M3Gfloat x)
  1.1184 +{
  1.1185 +    return (M3Gfloat) acos(x);
  1.1186 +}
  1.1187 +
  1.1188 +/*!
  1.1189 + * \internal
  1.1190 + */
  1.1191 +static M3Gfloat m3gArcTan(const M3Gfloat y, const M3Gfloat x)
  1.1192 +{
  1.1193 +    return (M3Gfloat) atan2(y, x);
  1.1194 +}
  1.1195 +
  1.1196 +/*!
  1.1197 + * \internal
  1.1198 + */
  1.1199 +static M3Gfloat m3gCos(const M3Gfloat x)
  1.1200 +{
  1.1201 +    return m3gSin(m3gAdd(x, HALF_PI));
  1.1202 +}
  1.1203 +
  1.1204 +/*!
  1.1205 + * \internal
  1.1206 + * \brief
  1.1207 + */
  1.1208 +static M3Gfloat m3gSin(const M3Gfloat x)
  1.1209 +{
  1.1210 +    M3Gfloat f = x;
  1.1211 +    
  1.1212 +    /* If x is greater than two pi, do a modulo operation to bring it
  1.1213 +     * back in range for the internal sine function */
  1.1214 +    
  1.1215 +    if (m3gAbs(f) >= TWO_PI) {
  1.1216 +        f = m3gMul (f, (1.f / TWO_PI));
  1.1217 +        f = m3gFrac(f);
  1.1218 +        f = m3gMul (f, TWO_PI);
  1.1219 +    }
  1.1220 +
  1.1221 +    /* Compute the result, negating both the input value and the
  1.1222 +     * result if x was negative */
  1.1223 +    {
  1.1224 +        M3Guint i = FLOAT_AS_UINT(f);
  1.1225 +        M3Guint neg = (i & SIGN_MASK);
  1.1226 +        i ^= neg;
  1.1227 +        f = m3gSinFirstPeriod(INT_AS_FLOAT(i));
  1.1228 +        i = FLOAT_AS_UINT(f) ^ neg;
  1.1229 +        return INT_AS_FLOAT(i);
  1.1230 +    }
  1.1231 +}
  1.1232 +
  1.1233 +/*!
  1.1234 + * \internal
  1.1235 + */
  1.1236 +static M3Gfloat m3gTan(const M3Gfloat x)
  1.1237 +{
  1.1238 +    return (M3Gfloat) tan(x);
  1.1239 +}
  1.1240 +
  1.1241 +/*!
  1.1242 + * \internal
  1.1243 + */
  1.1244 +static M3Gfloat m3gExp(const M3Gfloat a)
  1.1245 +{
  1.1246 +    return (M3Gfloat) exp(a);
  1.1247 +}
  1.1248 +#endif /* M3G_SOFT_FLOAT */
  1.1249 +
  1.1250 +/*!
  1.1251 + * \brief Checks whether the bottom row of a matrix is 0 0 0 1
  1.1252 + */
  1.1253 +static M3Gbool m3gIsWUnity(const Matrix *mtx)
  1.1254 +{
  1.1255 +    M3G_ASSERT_PTR(mtx);
  1.1256 +
  1.1257 +    if (!m3gIsClassified(mtx)) {
  1.1258 +        return (IS_ZERO(M44F(mtx, 3, 0)) &&
  1.1259 +                IS_ZERO(M44F(mtx, 3, 1)) &&
  1.1260 +                IS_ZERO(M44F(mtx, 3, 2)) &&
  1.1261 +                IS_ONE (M44F(mtx, 3, 3)));
  1.1262 +    }
  1.1263 +    else {
  1.1264 +        return ((mtx->mask & 0xC0C0C0C0) == (ELEM_ONE << 30));
  1.1265 +    }
  1.1266 +}
  1.1267 +
  1.1268 +/*!
  1.1269 + * \brief Makes a quaternion by exponentiating a quaternion logarithm
  1.1270 + */
  1.1271 +static void m3gExpQuat(Quat *quat, const Vec3 *qExp)
  1.1272 +{
  1.1273 +    M3Gfloat theta;
  1.1274 +
  1.1275 +    M3G_ASSERT_PTR(quat);
  1.1276 +    M3G_ASSERT_PTR(qExp);
  1.1277 +
  1.1278 +    theta = m3gSqrt(m3gAdd(m3gAdd(m3gSquare(qExp->x),
  1.1279 +                                  m3gSquare(qExp->y)),
  1.1280 +                           m3gSquare(qExp->z)));
  1.1281 +
  1.1282 +    if (theta > EPSILON) {
  1.1283 +        M3Gfloat s = m3gMul(m3gSin(theta), m3gRcp(theta));
  1.1284 +        quat->x = m3gMul(qExp->x, s);
  1.1285 +        quat->y = m3gMul(qExp->y, s);
  1.1286 +        quat->z = m3gMul(qExp->z, s);
  1.1287 +        quat->w = m3gCos(theta);
  1.1288 +    }
  1.1289 +    else {
  1.1290 +        quat->x = quat->y = quat->z = 0.0f;
  1.1291 +        quat->w = 1.0f;
  1.1292 +    }
  1.1293 +}
  1.1294 +
  1.1295 +/*!
  1.1296 + * \brief Natural logarithm of a unit quaternion.
  1.1297 + */
  1.1298 +static void m3gLogQuat(Vec3 *qLog, const Quat *quat)
  1.1299 +{
  1.1300 +    M3Gfloat sinTheta = m3gSqrt(m3gNorm3((const float *) quat));
  1.1301 +
  1.1302 +    if (sinTheta > EPSILON) {
  1.1303 +        M3Gfloat s = m3gArcTan(sinTheta, quat->w) / sinTheta;
  1.1304 +        qLog->x = m3gMul(s, quat->x);
  1.1305 +        qLog->y = m3gMul(s, quat->y);
  1.1306 +        qLog->z = m3gMul(s, quat->z);
  1.1307 +    }
  1.1308 +    else {
  1.1309 +        qLog->x = qLog->y = qLog->z = 0.0f;
  1.1310 +    }
  1.1311 +}
  1.1312 +
  1.1313 +/*!
  1.1314 + * \brief Make quaternion the "logarithmic difference" between two
  1.1315 + * other quaternions.
  1.1316 + */
  1.1317 +static void m3gLogDiffQuat(Vec3 *logDiff,
  1.1318 +                           const Quat *from, const Quat *to)
  1.1319 +{
  1.1320 +    Quat temp;
  1.1321 +    temp.x = m3gNegate(from->x);
  1.1322 +    temp.y = m3gNegate(from->y);
  1.1323 +    temp.z = m3gNegate(from->z);
  1.1324 +    temp.w =           from->w;
  1.1325 +    m3gMulQuat(&temp, to);
  1.1326 +    m3gLogQuat(logDiff, &temp);
  1.1327 +}
  1.1328 +
  1.1329 +/*!
  1.1330 + * \brief Rounds a float to the nearest integer
  1.1331 + *
  1.1332 + * Overflows are clamped to the maximum or minimum representable
  1.1333 + * value.
  1.1334 + */
  1.1335 +static M3Gint m3gRoundToInt(const M3Gfloat a)
  1.1336 +{
  1.1337 +    M3Guint base = FLOAT_AS_UINT(a);
  1.1338 +    M3Gint signMask, expo;
  1.1339 +
  1.1340 +    /* Decompose the number into sign, exponent, and base number */
  1.1341 +    
  1.1342 +    signMask = ((M3Gint) base >> 31);   /* -> 0 or 0xFFFFFFFF */
  1.1343 +    expo = (M3Gint)((base >> 23) & 0xFF) - 127;
  1.1344 +    
  1.1345 +    /* First check for large values and return either the negative or
  1.1346 +     * the positive maximum integer in case of overflow.  The overflow
  1.1347 +     * check can be made on the exponent alone, as large floats are
  1.1348 +     * spaced several integer values apart so that nothing will
  1.1349 +     * overflow because of rounding later on */
  1.1350 +    
  1.1351 +    if (expo >= 31) {
  1.1352 +        return (M3Gint)((1U << 31) - 1 + (((M3Guint) signMask) & 1));
  1.1353 +    }
  1.1354 +
  1.1355 +    /* Also check for underflow to avoid problems with shifting by
  1.1356 +     * more than 31 */
  1.1357 +
  1.1358 +    if (expo < -1) {
  1.1359 +        return 0;
  1.1360 +    }
  1.1361 +    
  1.1362 +    /* Mask out the sign and exponent bits, shift the base number so
  1.1363 +     * that the lowest bit corresponds to one half, then add one
  1.1364 +     * (half) and shift to round to the closest integer. */
  1.1365 +
  1.1366 +    base = (base | LEADING_ONE) << 8;   /* shift mantissa to 1.31 */
  1.1367 +    base =  base >> (30 - expo);        /* integer value as 31.1 */
  1.1368 +    base = (base + 1) >> 1;             /* round to nearest 32.0 */
  1.1369 +    
  1.1370 +    /* Factor in the sign (negate if originally negative) and
  1.1371 +     * return */
  1.1372 +
  1.1373 +    return ((M3Gint) base ^ signMask) - signMask;
  1.1374 +}
  1.1375 +
  1.1376 +/*!
  1.1377 + * \brief Calculates ray-triangle intersection.
  1.1378 + *
  1.1379 + * http://www.acm.org/jgt/papers/MollerTrumbore97
  1.1380 + */
  1.1381 +static M3Gbool m3gIntersectTriangle(const Vec3 *orig, const Vec3 *dir,
  1.1382 +                                    const Vec3 *vert0, const Vec3 *vert1, const Vec3 *vert2,
  1.1383 +                                    Vec3 *tuv, M3Gint cullMode)
  1.1384 +{
  1.1385 +    Vec3 edge1, edge2, tvec, pvec, qvec;
  1.1386 +    M3Gfloat det,inv_det;
  1.1387 +
  1.1388 +    /* find vectors for two edges sharing vert0 */
  1.1389 +    edge1 = *vert1;
  1.1390 +    edge2 = *vert2;
  1.1391 +    m3gSubVec3(&edge1, vert0);
  1.1392 +    m3gSubVec3(&edge2, vert0);
  1.1393 +
  1.1394 +    /* begin calculating determinant - also used to calculate U parameter */
  1.1395 +    m3gCross(&pvec, dir, &edge2);
  1.1396 +
  1.1397 +    /* if determinant is near zero, ray lies in plane of triangle */
  1.1398 +    det = m3gDot3(&edge1, &pvec);
  1.1399 +
  1.1400 +    if (cullMode == 0 && det <= 0) return M3G_FALSE;
  1.1401 +    if (cullMode == 1 && det >= 0) return M3G_FALSE;
  1.1402 +
  1.1403 +    if (det > -EPSILON && det < EPSILON)
  1.1404 +        return M3G_FALSE;
  1.1405 +    inv_det = m3gRcp(det);
  1.1406 +
  1.1407 +    /* calculate distance from vert0 to ray origin */
  1.1408 +    tvec = *orig;
  1.1409 +    m3gSubVec3(&tvec, vert0);
  1.1410 +
  1.1411 +    /* calculate U parameter and test bounds */
  1.1412 +    tuv->y = m3gMul(m3gDot3(&tvec, &pvec), inv_det);
  1.1413 +    if (tuv->y < 0.0f || tuv->y > 1.0f)
  1.1414 +        return M3G_FALSE;
  1.1415 +
  1.1416 +    /* prepare to test V parameter */
  1.1417 +    m3gCross(&qvec, &tvec, &edge1);
  1.1418 +
  1.1419 +    /* calculate V parameter and test bounds */
  1.1420 +    tuv->z = m3gMul(m3gDot3(dir, &qvec), inv_det);
  1.1421 +    if (tuv->z < 0.0f || m3gAdd(tuv->y, tuv->z) > 1.0f)
  1.1422 +        return M3G_FALSE;
  1.1423 +
  1.1424 +    /* calculate t, ray intersects triangle */
  1.1425 +    tuv->x = m3gMul(m3gDot3(&edge2, &qvec), inv_det);
  1.1426 +
  1.1427 +    return M3G_TRUE;
  1.1428 +}
  1.1429 +
  1.1430 +/*!
  1.1431 + * \brief Calculates ray-box intersection.
  1.1432 + *
  1.1433 + * http://www.siggraph.org/education/materials/HyperGraph/raytrace/rtinter3.htm
  1.1434 + */
  1.1435 +
  1.1436 +#define XO	orig->x
  1.1437 +#define YO	orig->y
  1.1438 +#define ZO	orig->z
  1.1439 +#define XD	dir->x
  1.1440 +#define YD	dir->y
  1.1441 +#define ZD	dir->z
  1.1442 +
  1.1443 +#define XL	box->min[0]
  1.1444 +#define YL	box->min[1]
  1.1445 +#define ZL	box->min[2]
  1.1446 +#define XH	box->max[0]
  1.1447 +#define YH	box->max[1]
  1.1448 +#define ZH	box->max[2]
  1.1449 +
  1.1450 +/*!
  1.1451 + * \internal
  1.1452 + * \brief Ray - bounding box intersection
  1.1453 + *
  1.1454 + */
  1.1455 +static M3Gbool m3gIntersectBox(const Vec3 *orig, const Vec3 *dir, const AABB *box)
  1.1456 +{
  1.1457 +	M3Gfloat tnear = M3G_MIN_NEGATIVE_FLOAT;
  1.1458 +	M3Gfloat tfar  = M3G_MAX_POSITIVE_FLOAT;
  1.1459 +	M3Gfloat t1, t2, temp;
  1.1460 +
  1.1461 +	/* X slab */
  1.1462 +	if(XD != 0) {
  1.1463 +		t1 = m3gSub(XL, XO) / XD;
  1.1464 +		t2 = m3gSub(XH, XO) / XD;
  1.1465 +
  1.1466 +		if(t1 > t2) {
  1.1467 +			temp = t1;
  1.1468 +			t1 = t2;
  1.1469 +			t2 = temp;
  1.1470 +		}
  1.1471 +
  1.1472 +		if(t1 > tnear) tnear = t1;
  1.1473 +		if(t2 < tfar) tfar = t2;
  1.1474 +
  1.1475 +		if(tnear > tfar) return M3G_FALSE;
  1.1476 +		if(tfar < 0) return M3G_FALSE;
  1.1477 +	}
  1.1478 +	else {
  1.1479 +		if(XO > XH || XO < XL) return M3G_FALSE;
  1.1480 +	}
  1.1481 +
  1.1482 +	/* Y slab */
  1.1483 +	if(YD != 0) {
  1.1484 +		t1 = m3gSub(YL, YO) / YD;
  1.1485 +		t2 = m3gSub(YH, YO) / YD;
  1.1486 +
  1.1487 +		if(t1 > t2) {
  1.1488 +			temp = t1;
  1.1489 +			t1 = t2;
  1.1490 +			t2 = temp;
  1.1491 +		}
  1.1492 +
  1.1493 +		if(t1 > tnear) tnear = t1;
  1.1494 +		if(t2 < tfar) tfar = t2;
  1.1495 +
  1.1496 +		if(tnear > tfar) return M3G_FALSE;
  1.1497 +		if(tfar < 0) return M3G_FALSE;
  1.1498 +	}
  1.1499 +	else {
  1.1500 +		if(YO > YH || YO < YL) return M3G_FALSE;
  1.1501 +	}
  1.1502 +
  1.1503 +	/* Z slab */
  1.1504 +	if(ZD != 0) {
  1.1505 +		t1 = m3gSub(ZL, ZO) / ZD;
  1.1506 +		t2 = m3gSub(ZH, ZO) / ZD;
  1.1507 +
  1.1508 +		if(t1 > t2) {
  1.1509 +			temp = t1;
  1.1510 +			t1 = t2;
  1.1511 +			t2 = temp;
  1.1512 +		}
  1.1513 +
  1.1514 +		if(t1 > tnear) tnear = t1;
  1.1515 +		if(t2 < tfar) tfar = t2;
  1.1516 +
  1.1517 +		if(tnear > tfar) return M3G_FALSE;
  1.1518 +		if(tfar < 0) return M3G_FALSE;
  1.1519 +	}
  1.1520 +	else {
  1.1521 +		if(ZO > ZH || ZO < ZL) return M3G_FALSE;
  1.1522 +	}
  1.1523 +
  1.1524 +	return M3G_TRUE;
  1.1525 +}
  1.1526 +
  1.1527 +/*!
  1.1528 + * \brief Calculates the intersection of two rectangles. Always fills
  1.1529 + * the intersection result.
  1.1530 + *
  1.1531 + * \param dst   result of the intersection
  1.1532 + * \param r1    rectangle 1
  1.1533 + * \param r2    rectangle 2
  1.1534 + */
  1.1535 +static M3Gbool m3gIntersectRectangle(M3GRectangle *dst, M3GRectangle *r1, M3GRectangle *r2)
  1.1536 +{
  1.1537 +    M3Gbool intersects = M3G_TRUE;
  1.1538 +    M3Gint min, max;
  1.1539 +
  1.1540 +    max = (r1->x) >= (r2->x) ? (r1->x) : (r2->x);
  1.1541 +    min = (r1->x + r1->width) <= (r2->x + r2->width) ? (r1->x + r1->width) : (r2->x + r2->width);
  1.1542 +    if ((min - max) < 0) intersects = M3G_FALSE;
  1.1543 +    dst->x = max;
  1.1544 +    dst->width = min - max;
  1.1545 +
  1.1546 +    max = (r1->y) >= (r2->y) ? (r1->y) : (r2->y);
  1.1547 +    min = (r1->y + r1->height) <= (r2->y + r2->height) ? (r1->y + r1->height) : (r2->y + r2->height);
  1.1548 +    if ((min - max) < 0) intersects = M3G_FALSE;
  1.1549 +    dst->y = max;
  1.1550 +    dst->height = min - max;
  1.1551 +
  1.1552 +    return intersects;
  1.1553 +}
  1.1554 +
  1.1555 +/*-------- float-to-int color conversions --------*/
  1.1556 +
  1.1557 +static M3Guint m3gAlpha1f(M3Gfloat a)
  1.1558 +{
  1.1559 +    M3Guint alpha = (M3Guint) m3gFloatToByte(a);
  1.1560 +    return (alpha << 24) | M3G_RGB_MASK;
  1.1561 +}
  1.1562 +
  1.1563 +static M3Guint m3gColor3f(M3Gfloat r, M3Gfloat g, M3Gfloat b)
  1.1564 +{
  1.1565 +    return ((M3Guint) m3gFloatToByte(r) << 16)
  1.1566 +        |  ((M3Guint) m3gFloatToByte(g) <<  8)
  1.1567 +        |   (M3Guint) m3gFloatToByte(b)
  1.1568 +        |   M3G_ALPHA_MASK;
  1.1569 +}
  1.1570 +
  1.1571 +static M3Guint m3gColor4f(M3Gfloat r, M3Gfloat g, M3Gfloat b, M3Gfloat a)
  1.1572 +{
  1.1573 +    return ((M3Guint) m3gFloatToByte(r) << 16)
  1.1574 +        |  ((M3Guint) m3gFloatToByte(g) <<  8)
  1.1575 +        |   (M3Guint) m3gFloatToByte(b)
  1.1576 +        |  ((M3Guint) m3gFloatToByte(a) << 24);
  1.1577 +}
  1.1578 +
  1.1579 +static void m3gFloatColor(M3Gint argb, M3Gfloat intensity, M3Gfloat *rgba)
  1.1580 +{
  1.1581 +    /* NOTE we intentionally aim a bit high here -- some GL
  1.1582 +     * implementations may round down instead of closest */
  1.1583 +    
  1.1584 +    const M3Gfloat oneOver255 = (M3Gfloat)(1.0001 / 255.0);
  1.1585 +    
  1.1586 +	rgba[0] = (M3Gfloat)((argb >> 16) & 0xFF);
  1.1587 +	rgba[1] = (M3Gfloat)((argb >>  8) & 0xFF);
  1.1588 +	rgba[2] = (M3Gfloat)((argb      ) & 0xFF);
  1.1589 +	rgba[3] = (M3Gfloat)((argb >> 24) & 0xFF);
  1.1590 +    
  1.1591 +    m3gScale4(rgba, m3gMul(oneOver255, intensity));
  1.1592 +}
  1.1593 +
  1.1594 +/*!
  1.1595 + * \brief Converts the 3x3 submatrix of a matrix into fixed point
  1.1596 + *
  1.1597 + * The input matrix must be an affine one, i.e. the bottom row must be
  1.1598 + * 0 0 0 1.  The output matrix is guaranteed to be such that it can be
  1.1599 + * multiplied with a 16-bit 3-vector without overflowing, while using
  1.1600 + * the 32-bit range optimally.
  1.1601 + *
  1.1602 + * \param mtx  the original matrix
  1.1603 + * \param elem 9 shorts to hold the fixed point base numbers
  1.1604 + * \return floating point exponent for the values in \c elem
  1.1605 + *         (number of bits to shift left for actual values)
  1.1606 + */
  1.1607 +static M3Gint m3gGetFixedPoint3x3Basis(const Matrix *mtx, M3Gshort *elem)
  1.1608 +{
  1.1609 +    M3Gint outExp;
  1.1610 +    M3Gint row, col;
  1.1611 +    const M3Guint *m;
  1.1612 +    
  1.1613 +    if (!mtx->complete) {
  1.1614 +        m3gFillClassifiedMatrix((Matrix*) mtx);
  1.1615 +    }
  1.1616 +    m = (const M3Guint*) mtx->elem;
  1.1617 +    
  1.1618 +    /* First, find the maximum exponent value in the whole matrix */
  1.1619 +
  1.1620 +    outExp = 0;
  1.1621 +    for (col = 0; col < 3; ++col) {
  1.1622 +        for (row = 0; row < 3; ++row) {
  1.1623 +            M3Gint element = (M3Gint)(m[MELEM(row, col)] & ~SIGN_MASK);
  1.1624 +            outExp = M3G_MAX(outExp, element);
  1.1625 +        }
  1.1626 +    }
  1.1627 +    outExp >>= 23;
  1.1628 +
  1.1629 +    /* Our candidate exponent is the maximum found plus 9, which is
  1.1630 +     * guaranteed to shift the maximum unsigned 24-bit mantissa (which
  1.1631 +     * always will have the high bit set) down to the signed 16-bit
  1.1632 +     * range */
  1.1633 +
  1.1634 +    outExp += 9;
  1.1635 +    
  1.1636 +    /* Now proceed to sum each row and see what's the actual smallest
  1.1637 +     * exponent we can safely use without overflowing in a 16+16
  1.1638 +     * matrix-vector multiplication; this will win us one bit
  1.1639 +     * (doubling the precision) compared to the conservative approach
  1.1640 +     * of just shifting everything down by 10 bits */
  1.1641 +
  1.1642 +    for (row = 0; row < 3; ++row) {
  1.1643 +
  1.1644 +        /* Sum the absolute values on this row */
  1.1645 +            
  1.1646 +        M3Gint rowSum = 0;
  1.1647 +        for (col = 0; col < 3; ++col) {
  1.1648 +            M3Gint a = (M3Gint)(m[MELEM(row, col)] & ~SIGN_MASK);
  1.1649 +            M3Gint shift = outExp - (a >> 23);
  1.1650 +            M3G_ASSERT(shift < 265);
  1.1651 +                
  1.1652 +            if (shift < 24) {
  1.1653 +                rowSum += ((a & MANTISSA_MASK) | LEADING_ONE) >> shift;
  1.1654 +            }
  1.1655 +        }
  1.1656 +
  1.1657 +        /* Now we have a 26-bit sum of the absolute values on this
  1.1658 +         * row, and shift that down until we fit the target range of
  1.1659 +         * [0, 65535].  Note that this still leaves *exactly* enough
  1.1660 +         * space for adding in an arbitrary 16-bit translation vector
  1.1661 +         * after multiplying with the matrix! */
  1.1662 +            
  1.1663 +        while (rowSum >= (1 << 16)) {
  1.1664 +            rowSum >>= 1;
  1.1665 +            ++outExp;
  1.1666 +        }
  1.1667 +    }
  1.1668 +
  1.1669 +    /* De-bias the exponent, but add in an extra 23 to account for the
  1.1670 +     * decimal bits in the floating point mantissa values we started
  1.1671 +     * with (we're returning the exponent as "bits to shift left to
  1.1672 +     * get integers", so we're off by 23 from IEEE notation) */
  1.1673 +    
  1.1674 +    outExp = (outExp - 127) - 23;
  1.1675 +    
  1.1676 +    /* Finally, shift all the matrix elements to our final output
  1.1677 +     * precision */
  1.1678 +    
  1.1679 +    for (col = 0; col < 3; ++col) {
  1.1680 +        m3gFloatVecToShort(3, mtx->elem + MELEM(0, col), outExp, elem);
  1.1681 +        elem += 3;
  1.1682 +    }
  1.1683 +    return outExp;
  1.1684 +}
  1.1685 +
  1.1686 +/*!
  1.1687 + * \brief Gets the translation component of a matrix as fixed point
  1.1688 + *
  1.1689 + * \param mtx  the matrix
  1.1690 + * \param elem 3 shorts to write the vector into
  1.1691 + * \return floating point exponent for the values in \c elem
  1.1692 + *         (number of bits to shift left for actual values)
  1.1693 + */
  1.1694 +static M3Gint m3gGetFixedPointTranslation(const Matrix *mtx, M3Gshort *elem)
  1.1695 +{
  1.1696 +    const M3Guint *m;
  1.1697 +    
  1.1698 +    M3G_ASSERT(m3gIsWUnity(mtx));
  1.1699 +    if (!mtx->complete) {
  1.1700 +        m3gFillClassifiedMatrix((Matrix*) mtx);
  1.1701 +    }
  1.1702 +    m = (const M3Guint*) &mtx->elem[MELEM(0, 3)];
  1.1703 +
  1.1704 +    /* Find the maximum exponent, then scale down by 9 bits from that
  1.1705 +     * to shift the unsigned 24-bit mantissas to the signed 16-bit
  1.1706 +     * range */
  1.1707 +    {
  1.1708 +        M3Gint outExp;
  1.1709 +        M3Guint maxElem = m[0] & ~SIGN_MASK;
  1.1710 +        maxElem = M3G_MAX(maxElem, m[1] & ~SIGN_MASK);
  1.1711 +        maxElem = M3G_MAX(maxElem, m[2] & ~SIGN_MASK);
  1.1712 +        
  1.1713 +        outExp = (M3Gint)(maxElem >> 23) - (127 + 23) + 9;
  1.1714 +        m3gFloatVecToShort(3, mtx->elem + MELEM(0, 3), outExp, elem);
  1.1715 +        return outExp;
  1.1716 +    }
  1.1717 +}
  1.1718 +
  1.1719 +/*!
  1.1720 + * \internal
  1.1721 + * \brief Compute a bounding box enclosing two other boxes
  1.1722 + *
  1.1723 + * \param box   box to fit
  1.1724 + * \param a     first box to enclose or NULL
  1.1725 + * \param b     second box to enclose or NULL
  1.1726 + * 
  1.1727 + * \note If both input boxes are NULL, the box is not modified.
  1.1728 + */
  1.1729 +static void m3gFitAABB(AABB *box, const AABB *a, const AABB *b)
  1.1730 +{
  1.1731 +    int i;
  1.1732 +
  1.1733 +    M3G_ASSERT_PTR(box);
  1.1734 +    
  1.1735 +    if (a) {
  1.1736 +        m3gValidateAABB(a);
  1.1737 +    }
  1.1738 +    if (b) {
  1.1739 +        m3gValidateAABB(b);
  1.1740 +    }
  1.1741 +
  1.1742 +    if (a && b) {
  1.1743 +        for (i = 0; i < 3; ++i) {
  1.1744 +            box->min[i] = M3G_MIN(a->min[i], b->min[i]);
  1.1745 +            box->max[i] = M3G_MAX(a->max[i], b->max[i]);
  1.1746 +        }
  1.1747 +    }
  1.1748 +    else if (a) {
  1.1749 +        *box = *a;
  1.1750 +    }
  1.1751 +    else if (b) {
  1.1752 +        *box = *b;
  1.1753 +    }
  1.1754 +}
  1.1755 +
  1.1756 +/*
  1.1757 + * \internal
  1.1758 + * \brief Transform an axis-aligned bounding box with a matrix
  1.1759 + *
  1.1760 + * This results in a box that encloses the transformed original box.
  1.1761 + * 
  1.1762 + * Based on "Transforming Axis-Aligned Bounding Boxes" by Jim Arvo
  1.1763 + * from Graphics Gems.
  1.1764 + * 
  1.1765 + * \note The bottom row of the matrix is ignored in the transformation.
  1.1766 + */
  1.1767 +static void m3gTransformAABB(AABB *box, const Matrix *mtx)
  1.1768 +{
  1.1769 +    M3Gfloat boxMin[3], boxMax[3];
  1.1770 +    M3Gfloat newMin[3], newMax[3];
  1.1771 +
  1.1772 +    m3gValidateAABB(box);
  1.1773 +    
  1.1774 +    if (!mtx->complete) {
  1.1775 +        m3gFillClassifiedMatrix((Matrix*) mtx);
  1.1776 +    }
  1.1777 +
  1.1778 +    /* Get the original minimum and maximum extents as floats, and add
  1.1779 +     * the translation as the base for the transformed box */
  1.1780 +    {
  1.1781 +        int i;
  1.1782 +        for (i = 0; i < 3; ++i) {
  1.1783 +            boxMin[i] = box->min[i];
  1.1784 +            boxMax[i] = box->max[i];
  1.1785 +            newMin[i] = newMax[i] = M44F(mtx, i, 3);
  1.1786 +        }
  1.1787 +    }
  1.1788 +
  1.1789 +    /* Transform into the new minimum and maximum coordinates using
  1.1790 +     * the upper left 3x3 part of the matrix */
  1.1791 +    {
  1.1792 +        M3Gint row, col;
  1.1793 +        
  1.1794 +        for (row = 0; row < 3; ++row) {
  1.1795 +            for (col = 0; col < 3; ++col) {
  1.1796 +                M3Gfloat a = m3gMul(M44F(mtx, row, col), boxMin[col]);
  1.1797 +                M3Gfloat b = m3gMul(M44F(mtx, row, col), boxMax[col]);
  1.1798 +                
  1.1799 +                if (a < b) { 
  1.1800 +                    newMin[row] = m3gAdd(newMin[row], a);
  1.1801 +                    newMax[row] = m3gAdd(newMax[row], b);
  1.1802 +                }
  1.1803 +                else { 
  1.1804 +                    newMin[row] = m3gAdd(newMin[row], b);
  1.1805 +                    newMax[row] = m3gAdd(newMax[row], a);
  1.1806 +                }
  1.1807 +            }
  1.1808 +        }
  1.1809 +    }
  1.1810 +
  1.1811 +    /* Write back into the bounding box */
  1.1812 +    {
  1.1813 +        int i;
  1.1814 +        for (i = 0; i < 3; ++i) {
  1.1815 +            box->min[i] = newMin[i];
  1.1816 +            box->max[i] = newMax[i];
  1.1817 +        }
  1.1818 +    }
  1.1819 +    
  1.1820 +    m3gValidateAABB(box);
  1.1821 +}
  1.1822 +
  1.1823 +#if defined(M3G_DEBUG)
  1.1824 +/*!
  1.1825 + * \brief
  1.1826 + */
  1.1827 +static void m3gValidateAABB(const AABB *aabb)
  1.1828 +{
  1.1829 +    m3gValidateFloats(6, (float*) aabb);
  1.1830 +}
  1.1831 +#endif
  1.1832 +
  1.1833 +/*----------------------------------------------------------------------
  1.1834 + * Public functions
  1.1835 + *--------------------------------------------------------------------*/
  1.1836 +
  1.1837 +/*!
  1.1838 + * \brief Linear interpolation of vectors
  1.1839 + *
  1.1840 + * \param size     number of components
  1.1841 + * \param vec      output vector
  1.1842 + * \param s        interpolation factor
  1.1843 + * \param start    initial value
  1.1844 + * \param end      target value
  1.1845 + */
  1.1846 +#if defined(M3G_HW_FLOAT_VFPV2)
  1.1847 +
  1.1848 +__weak __asm void m3gLerp(M3Gint size,
  1.1849 +				   M3Gfloat *vec,
  1.1850 +				   M3Gfloat s,
  1.1851 +				   const M3Gfloat *start, const M3Gfloat *end)
  1.1852 +{
  1.1853 +// r0 = size
  1.1854 +// r1 = *vec
  1.1855 +// r2 = s
  1.1856 +// r3 = *start
  1.1857 +// sp[0] = *end
  1.1858 +
  1.1859 +		EXPORT	m3gLerp[DYNAMIC]
  1.1860 +		CODE32
  1.1861 +/*
  1.1862 +    M3Gfloat sCompl = 1.0 - s;
  1.1863 +    for (i = 0; i < size; ++i) {
  1.1864 +        vec[i] = sCompl*start[i] + s*end[i];
  1.1865 +    }
  1.1866 +*/
  1.1867 +//
  1.1868 +// if size = 0, return
  1.1869 +//
  1.1870 +		CMP		r0, #0
  1.1871 +		BXEQ	lr
  1.1872 +
  1.1873 +		FMSR	s0, r2
  1.1874 +		MOV		r2, r3
  1.1875 +		LDR		r3, [sp]
  1.1876 +
  1.1877 +		FLDS	s1,=1.0
  1.1878 +		STMFD	sp!, {r4-r5}
  1.1879 +		FSUBS	s2, s1, s0			// sCompl = 1 - s
  1.1880 +
  1.1881 +		FMRX	r4, FPSCR
  1.1882 +		CMP		r0, #4
  1.1883 +		BGT		_m3gLerp_over4Loop
  1.1884 +
  1.1885 +//
  1.1886 +// if 1 < size <= 4
  1.1887 +//
  1.1888 +// set vector STRIDE = 1, LENGTH = 4
  1.1889 +		BIC		r12, r4, #0x00370000
  1.1890 +		ORR		r12, #(3<<16)
  1.1891 +		FMXR	FPSCR, r12
  1.1892 +
  1.1893 +		FLDMIAS	r2!, {s4-s7}		// load the start[i] values
  1.1894 +		FLDMIAS	r3!, {s8-s11}		// load the end[i] values
  1.1895 +		FMULS	s12, s4, s2			// [s12-s15]  = [sCompl*start[0] .. sCompl*start[3]]
  1.1896 +		FMACS	s12, s8, s0			// [s12-s15] += [    	s*end[0] ..        s*end[3]]
  1.1897 +		CMP		r0, #1
  1.1898 +		FSTS	s12, [r1]
  1.1899 +		FSTSGT	s13, [r1, #4]
  1.1900 +		CMP		r0, #3
  1.1901 +		FSTSGE	s14, [r1, #8]
  1.1902 +		FSTSGT	s15, [r1, #12]
  1.1903 +
  1.1904 +		FMXR	FPSCR, r4
  1.1905 +
  1.1906 +		LDMFD	sp!, {r4-r5}
  1.1907 +
  1.1908 +		BX		lr
  1.1909 +//
  1.1910 +// if size > 4, interpolate 8 values in one loop
  1.1911 +//
  1.1912 +_m3gLerp_over4Loop
  1.1913 +
  1.1914 +		FSTMFDD	sp!, {d8-d9}
  1.1915 +		MOVS	r5, r0, ASR #3			// size/8
  1.1916 +		SUB		r0, r0, r5, LSL #3		// tail length
  1.1917 +
  1.1918 +// set vector STRIDE = 1, LENGTH = 8
  1.1919 +		BIC		r12, r4, #0x00370000
  1.1920 +		ORR		r12, #(7<<16)
  1.1921 +		FMXR	FPSCR, r12
  1.1922 +
  1.1923 +
  1.1924 +_m3gLerp_alignedLoop
  1.1925 +
  1.1926 +		FLDMIASNE	r2!, {s8-s15}		// load the start[i] values
  1.1927 +		FLDMIASNE	r3!, {s16-s23}		// load the end[i] values
  1.1928 +		FMULSNE		s24, s8, s2			// [s16-s23]  = [ sCompl*start[0] sCompl*start[1] .. sCompl*start[7]]
  1.1929 +		FMACSNE		s24, s16, s0		// [s16-s23] += [		 s*end[0]        s*end[1] ..		s*end[7]]
  1.1930 +		FSTMIASNE	r1!, {s24-s31}
  1.1931 +		SUBSNE		r5, #1
  1.1932 +
  1.1933 +		BNE			_m3gLerp_alignedLoop
  1.1934 +
  1.1935 +// process the 0-7 remaining values in the tail
  1.1936 +
  1.1937 +		CMP			r0, #1
  1.1938 +		FLDMIASGE	r2!, {s8-s15}		
  1.1939 +		FLDMIASGE	r3!, {s16-s23}		
  1.1940 +		FMULSGE		s24, s8, s2			
  1.1941 +		FMACSGE		s24, s16, s0		
  1.1942 +		FSTSGE		s24, [r1]
  1.1943 +		FSTSGT		s25, [r1, #4]
  1.1944 +		CMP			r0, #3
  1.1945 +		FSTSGE		s26, [r1, #8]
  1.1946 +		FSTSGT		s27, [r1, #12]
  1.1947 +		CMP			r0, #5
  1.1948 +		FSTSGE		s28, [r1, #16]
  1.1949 +		FSTSGT		s29, [r1, #20]
  1.1950 +		CMP			r0, #7
  1.1951 +		FSTSEQ		s30, [r1, #24]
  1.1952 +
  1.1953 +		FMXR	FPSCR, r4
  1.1954 +
  1.1955 +		FLDMFDD	sp!, {d8-d9}
  1.1956 +		LDMFD	sp!, {r4-r5}
  1.1957 +
  1.1958 +		BX		lr
  1.1959 +
  1.1960 +}
  1.1961 +#else /* #if defined(M3G_HW_FLOAT_VFPV2) */
  1.1962 +
  1.1963 +M3G_API void m3gLerp(M3Gint size,
  1.1964 +             M3Gfloat *vec,
  1.1965 +             M3Gfloat s,
  1.1966 +             const M3Gfloat *start, const M3Gfloat *end)
  1.1967 +{
  1.1968 +    int i;
  1.1969 +
  1.1970 +    M3Gfloat sCompl = m3gSub(1.f, s);
  1.1971 +    for (i = 0; i < size; ++i) {
  1.1972 +        vec[i] = m3gAdd(m3gMul(sCompl, start[i]), m3gMul(s, end[i]));
  1.1973 +    }
  1.1974 +}
  1.1975 +#endif /* #if defined(M3G_HW_FLOAT_VFPV2) */
  1.1976 +
  1.1977 +/*!
  1.1978 + * \brief Hermite spline interpolation of vectors
  1.1979 + *
  1.1980 + * \param size      number of components
  1.1981 + * \param vec       output vector
  1.1982 + * \param s         interpolation factor
  1.1983 + * \param start     start value vector
  1.1984 + * \param end       end value vector
  1.1985 + * \param tStart    start tangent vector
  1.1986 + * \param tEnd      end tangent vector
  1.1987 + */
  1.1988 +M3G_API void m3gHermite(M3Gint size,
  1.1989 +                        M3Gfloat *vec,
  1.1990 +                        M3Gfloat s,
  1.1991 +                        const M3Gfloat *start, const M3Gfloat *end,
  1.1992 +                        const M3Gfloat *tStart, const M3Gfloat *tEnd)
  1.1993 +{
  1.1994 +    M3Gfloat s2 = m3gSquare(s);
  1.1995 +    M3Gfloat s3 = m3gMul(s2, s);
  1.1996 +    int i;
  1.1997 +    
  1.1998 +    for (i = 0; i < size; ++i) {
  1.1999 +        vec[i] =
  1.2000 +            m3gMadd(start[i],
  1.2001 +                    m3gAdd(m3gSub(m3gDouble(s3), m3gMul(3.f, s2)), 1.f),
  1.2002 +                    m3gMadd(end[i],
  1.2003 +                            m3gSub(m3gMul(3.f, s2), m3gDouble(s3)),
  1.2004 +                            m3gMadd(tStart[i],
  1.2005 +                                    m3gAdd(m3gSub(s3, m3gDouble(s2)), s),
  1.2006 +                                    m3gMul(tEnd[i],
  1.2007 +                                           m3gSub(s3, s2)))));
  1.2008 +
  1.2009 +    }
  1.2010 +    
  1.2011 +    /*  vec = ( 2*s3 - 3*s2 + 1) * start
  1.2012 +            + (-2*s3 + 3*s2    ) * end
  1.2013 +            + (   s3 - 2*s2 + s) * tStart
  1.2014 +            + (   s3 -   s2    ) * tEnd;    */
  1.2015 +}
  1.2016 +
  1.2017 +/*--------------------------------------------------------------------*/
  1.2018 +
  1.2019 +/*!
  1.2020 + * \brief Sets a matrix to a copy of another matrix
  1.2021 + */
  1.2022 +M3G_API void m3gCopyMatrix(Matrix *dst, const Matrix *src)
  1.2023 +{
  1.2024 +    M3G_ASSERT(dst != NULL && src != NULL);
  1.2025 +    *dst = *src;
  1.2026 +}
  1.2027 +
  1.2028 +/*!
  1.2029 + * \brief Vector addition
  1.2030 + */
  1.2031 +M3G_API void m3gAddVec3(Vec3 *vec, const Vec3 *other)
  1.2032 +{
  1.2033 +    vec->x = m3gAdd(vec->x, other->x);
  1.2034 +    vec->y = m3gAdd(vec->y, other->y);
  1.2035 +    vec->z = m3gAdd(vec->z, other->z);
  1.2036 +}
  1.2037 +
  1.2038 +/*!
  1.2039 + * \brief Vector addition
  1.2040 + */
  1.2041 +M3G_API void m3gAddVec4(Vec4 *vec, const Vec4 *other)
  1.2042 +{
  1.2043 +    vec->x = m3gAdd(vec->x, other->x);
  1.2044 +    vec->y = m3gAdd(vec->y, other->y);
  1.2045 +    vec->z = m3gAdd(vec->z, other->z);
  1.2046 +    vec->w = m3gAdd(vec->w, other->w);
  1.2047 +}
  1.2048 +
  1.2049 +/*!
  1.2050 + * \brief Cross product of two 3D vectors expressed as 4D vectors
  1.2051 + */
  1.2052 +M3G_API void m3gCross(Vec3 *dst, const Vec3 *a, const Vec3 *b)
  1.2053 +{
  1.2054 +    dst->x = m3gSub(m3gMul(a->y, b->z), m3gMul(a->z, b->y));
  1.2055 +    dst->y = m3gSub(m3gMul(a->z, b->x), m3gMul(a->x, b->z));
  1.2056 +    dst->z = m3gSub(m3gMul(a->x, b->y), m3gMul(a->y, b->x));
  1.2057 +}
  1.2058 +
  1.2059 +/*!
  1.2060 + * \brief Dot product of two vectors
  1.2061 + */
  1.2062 +M3G_API M3Gfloat m3gDot3(const Vec3 *a, const Vec3 *b)
  1.2063 +{
  1.2064 +    M3Gfloat d;
  1.2065 +    d = m3gMul(a->x, b->x);
  1.2066 +    d = m3gMadd(a->y, b->y, d);
  1.2067 +    d = m3gMadd(a->z, b->z, d);
  1.2068 +    return d;
  1.2069 +}
  1.2070 +
  1.2071 +/*!
  1.2072 + * \brief Dot product of two vectors
  1.2073 + */
  1.2074 +M3G_API M3Gfloat m3gDot4(const Vec4 *a, const Vec4 *b)
  1.2075 +{
  1.2076 +    M3Gfloat d;
  1.2077 +    d = m3gMul(a->x, b->x);
  1.2078 +    d = m3gMadd(a->y, b->y, d);
  1.2079 +    d = m3gMadd(a->z, b->z, d);
  1.2080 +    d = m3gMadd(a->w, b->w, d);
  1.2081 +    return d;
  1.2082 +}
  1.2083 +
  1.2084 +/*!
  1.2085 + * \brief
  1.2086 + */
  1.2087 +M3G_API void m3gSetVec3(Vec3 *v, M3Gfloat x, M3Gfloat y, M3Gfloat z)
  1.2088 +{
  1.2089 +    M3G_ASSERT_PTR(v);
  1.2090 +    v->x = x;
  1.2091 +    v->y = y;
  1.2092 +    v->z = z;
  1.2093 +}
  1.2094 +
  1.2095 +/*!
  1.2096 + * \brief
  1.2097 + */
  1.2098 +M3G_API void m3gSetVec4(Vec4 *v, M3Gfloat x, M3Gfloat y, M3Gfloat z, M3Gfloat w)
  1.2099 +{
  1.2100 +    M3G_ASSERT_PTR(v);
  1.2101 +    v->x = x;
  1.2102 +    v->y = y;
  1.2103 +    v->z = z;
  1.2104 +    v->w = w;
  1.2105 +}
  1.2106 +
  1.2107 +/*!
  1.2108 + * \brief Vector subtraction
  1.2109 + */
  1.2110 +M3G_API void m3gSubVec3(Vec3 *vec, const Vec3 *other)
  1.2111 +{
  1.2112 +    vec->x = m3gSub(vec->x, other->x);
  1.2113 +    vec->y = m3gSub(vec->y, other->y);
  1.2114 +    vec->z = m3gSub(vec->z, other->z);
  1.2115 +}
  1.2116 +
  1.2117 +/*!
  1.2118 + * \brief Vector subtraction
  1.2119 + */
  1.2120 +M3G_API void m3gSubVec4(Vec4 *vec, const Vec4 *other)
  1.2121 +{
  1.2122 +    vec->x = m3gSub(vec->x, other->x);
  1.2123 +    vec->y = m3gSub(vec->y, other->y);
  1.2124 +    vec->z = m3gSub(vec->z, other->z);
  1.2125 +    vec->w = m3gSub(vec->w, other->w);
  1.2126 +}
  1.2127 +
  1.2128 +/*!
  1.2129 + * \brief Vector length
  1.2130 + */
  1.2131 +M3G_API M3Gfloat m3gLengthVec3(const Vec3 *vec)
  1.2132 +{
  1.2133 +    return m3gSqrt(m3gAdd(m3gAdd(m3gSquare(vec->x),
  1.2134 +                                 m3gSquare(vec->y)),
  1.2135 +                          m3gSquare(vec->z)));
  1.2136 +}
  1.2137 +
  1.2138 +/*!
  1.2139 + * \brief Vector scaling
  1.2140 + */
  1.2141 +M3G_API void m3gScaleVec3(Vec3 *vec, const M3Gfloat s)
  1.2142 +{
  1.2143 +    vec->x = m3gMul(vec->x, s);
  1.2144 +    vec->y = m3gMul(vec->y, s);
  1.2145 +    vec->z = m3gMul(vec->z, s);
  1.2146 +}
  1.2147 +
  1.2148 +/*!
  1.2149 + * \brief Vector scaling
  1.2150 + */
  1.2151 +M3G_API void m3gScaleVec4(Vec4 *vec, const M3Gfloat s)
  1.2152 +{
  1.2153 +    vec->x = m3gMul(vec->x, s);
  1.2154 +    vec->y = m3gMul(vec->y, s);
  1.2155 +    vec->z = m3gMul(vec->z, s);
  1.2156 +    vec->w = m3gMul(vec->w, s);
  1.2157 +}
  1.2158 +
  1.2159 +/*!
  1.2160 + * \brief Returns an angle-axis representation for a quaternion
  1.2161 + *
  1.2162 + * \note There are many, and this is not guaranteed to return a
  1.2163 + * particular one
  1.2164 + */
  1.2165 +M3G_API void m3gGetAngleAxis(const Quat *quat, M3Gfloat *angle, Vec3 *axis)
  1.2166 +{
  1.2167 +    M3Gfloat x, y, z, sinTheta;
  1.2168 +
  1.2169 +    M3G_ASSERT_PTR(quat);
  1.2170 +    M3G_ASSERT_PTR(angle);
  1.2171 +    M3G_ASSERT_PTR(axis);
  1.2172 +
  1.2173 +    x = quat->x;
  1.2174 +    y = quat->y;
  1.2175 +    z = quat->z;
  1.2176 +
  1.2177 +    sinTheta = m3gSqrt(m3gAdd(m3gAdd(m3gSquare(x), m3gSquare(y)),
  1.2178 +                              m3gSquare(z)));
  1.2179 +
  1.2180 +    if (sinTheta > EPSILON) {
  1.2181 +        M3Gfloat ooSinTheta = m3gRcp(sinTheta);
  1.2182 +        axis->x = m3gMul(x, ooSinTheta);
  1.2183 +        axis->y = m3gMul(y, ooSinTheta);
  1.2184 +        axis->z = m3gMul(z, ooSinTheta);
  1.2185 +    }
  1.2186 +    else {
  1.2187 +        /* return a valid axis even for no rotation */
  1.2188 +        axis->x = axis->y = 0.0f;
  1.2189 +        axis->z = 1.0f;
  1.2190 +    }
  1.2191 +    *angle = m3gMul(2.0f * RAD2DEG, m3gArcCos(quat->w));
  1.2192 +}
  1.2193 +
  1.2194 +/*!
  1.2195 + * \brief Gets a single matrix column
  1.2196 + */
  1.2197 +M3G_API void m3gGetMatrixColumn(const Matrix *mtx, M3Gint col, Vec4 *dst)
  1.2198 +{
  1.2199 +    if ((col & ~3) == 0) {
  1.2200 +        if (!mtx->complete) {
  1.2201 +            m3gFillClassifiedMatrix((Matrix*)mtx);
  1.2202 +        }
  1.2203 +        dst->x = M44F(mtx, 0, col);
  1.2204 +        dst->y = M44F(mtx, 1, col);
  1.2205 +        dst->z = M44F(mtx, 2, col);
  1.2206 +        dst->w = M44F(mtx, 3, col);
  1.2207 +    }
  1.2208 +    else {
  1.2209 +        M3G_ASSERT(M3G_FALSE);
  1.2210 +    }
  1.2211 +}
  1.2212 +
  1.2213 +/*!
  1.2214 + * \brief Returns the floating point values of a matrix as consecutive
  1.2215 + * columns
  1.2216 + */
  1.2217 +M3G_API void m3gGetMatrixColumns(const Matrix *mtx, M3Gfloat *dst)
  1.2218 +{
  1.2219 +    M3G_ASSERT(mtx != NULL && dst != NULL);
  1.2220 +
  1.2221 +    if (!mtx->complete) {
  1.2222 +        m3gFillClassifiedMatrix((Matrix*)mtx);
  1.2223 +    }
  1.2224 +    m3gCopy(dst, mtx->elem, sizeof(mtx->elem));
  1.2225 +}
  1.2226 +
  1.2227 +/*!
  1.2228 + * \brief Gets a single matrix row
  1.2229 + */
  1.2230 +M3G_API void m3gGetMatrixRow(const Matrix *mtx, M3Gint row, Vec4 *dst)
  1.2231 +{
  1.2232 +    if ((row & ~3) == 0) {
  1.2233 +        if (!mtx->complete) {
  1.2234 +            m3gFillClassifiedMatrix((Matrix*)mtx);
  1.2235 +        }
  1.2236 +        dst->x = M44F(mtx, row, 0);
  1.2237 +        dst->y = M44F(mtx, row, 1);
  1.2238 +        dst->z = M44F(mtx, row, 2);
  1.2239 +        dst->w = M44F(mtx, row, 3);
  1.2240 +    }
  1.2241 +    else {
  1.2242 +        M3G_ASSERT(M3G_FALSE);
  1.2243 +    }
  1.2244 +}
  1.2245 +
  1.2246 +/*!
  1.2247 + * \brief Returns the floating point values of a matrix as consecutive
  1.2248 + * rows
  1.2249 + */
  1.2250 +M3G_API void m3gGetMatrixRows(const Matrix *mtx, M3Gfloat *dst)
  1.2251 +{
  1.2252 +    M3G_ASSERT(mtx != NULL && dst != NULL);
  1.2253 +
  1.2254 +    if (!mtx->complete) {
  1.2255 +        m3gFillClassifiedMatrix((Matrix*)mtx);
  1.2256 +    }
  1.2257 +    {
  1.2258 +        int row;
  1.2259 +        for (row = 0; row < 4; ++row) {
  1.2260 +            *dst++ = mtx->elem[ 0 + row];
  1.2261 +            *dst++ = mtx->elem[ 4 + row];
  1.2262 +            *dst++ = mtx->elem[ 8 + row];
  1.2263 +            *dst++ = mtx->elem[12 + row];
  1.2264 +        }
  1.2265 +    }
  1.2266 +}
  1.2267 +
  1.2268 +/*!
  1.2269 + * \brief Sets a matrix to identity
  1.2270 + */
  1.2271 +M3G_API void m3gIdentityMatrix(Matrix *mtx)
  1.2272 +{
  1.2273 +    M3G_ASSERT(mtx != NULL);
  1.2274 +    m3gClassifyAs(MC_IDENTITY, mtx);
  1.2275 +}
  1.2276 +
  1.2277 +/*!
  1.2278 + * \brief Sets a quaternion to identity
  1.2279 + */
  1.2280 +M3G_API void m3gIdentityQuat(Quat *quat)
  1.2281 +{
  1.2282 +    M3G_ASSERT(quat != NULL);
  1.2283 +    quat->x = quat->y = quat->z = 0.0f;
  1.2284 +    quat->w = 1.0f;
  1.2285 +}
  1.2286 +
  1.2287 +/*!
  1.2288 + * \brief Inverts a matrix
  1.2289 + */
  1.2290 +M3G_API M3Gbool m3gInvertMatrix(Matrix *mtx)
  1.2291 +{
  1.2292 +    M3Gfloat *matrix;
  1.2293 +    M3Gint i;
  1.2294 +    M3Gfloat tmp[12];
  1.2295 +    M3Gfloat src[16];
  1.2296 +    M3Gfloat det;
  1.2297 +
  1.2298 +    M3G_ASSERT(mtx != NULL);
  1.2299 +
  1.2300 +    if (!m3gIsClassified(mtx)) {
  1.2301 +        m3gClassify(mtx);
  1.2302 +    }
  1.2303 +
  1.2304 +    /* Quick exit for identity */
  1.2305 +    
  1.2306 +    if (mtx->mask == MC_IDENTITY) {
  1.2307 +        return M3G_TRUE;
  1.2308 +    }
  1.2309 +
  1.2310 +    /* Look for other common cases; these require that we have valid
  1.2311 +     * values in all the elements, so fill in the values first */
  1.2312 +    {
  1.2313 +        M3Guint mask = mtx->mask;
  1.2314 +        
  1.2315 +        if (!mtx->complete) {
  1.2316 +            m3gFillClassifiedMatrix(mtx);
  1.2317 +        }
  1.2318 +        
  1.2319 +        if ((mask | (0x3F << 24)) == MC_TRANSLATION) {
  1.2320 +            M44F(mtx, 0, 3) = m3gNegate(M44F(mtx, 0, 3));
  1.2321 +            M44F(mtx, 1, 3) = m3gNegate(M44F(mtx, 1, 3));
  1.2322 +            M44F(mtx, 2, 3) = m3gNegate(M44F(mtx, 2, 3));
  1.2323 +            mtx->mask = MC_TRANSLATION;
  1.2324 +            return M3G_TRUE;
  1.2325 +        }
  1.2326 +        if ((mask | 0x300C03) == MC_SCALING) {
  1.2327 +            if ((mask &  3       ) == 0 ||
  1.2328 +                (mask & (3 << 10)) == 0 ||
  1.2329 +                (mask & (3 << 20)) == 0) {
  1.2330 +                return M3G_FALSE; /* zero scale for at least one axis */
  1.2331 +            }
  1.2332 +            M44F(mtx, 0, 0) = m3gRcp(M44F(mtx, 0, 0));
  1.2333 +            M44F(mtx, 1, 1) = m3gRcp(M44F(mtx, 1, 1));
  1.2334 +            M44F(mtx, 2, 2) = m3gRcp(M44F(mtx, 2, 2));
  1.2335 +            return M3G_TRUE;
  1.2336 +        }
  1.2337 +    }
  1.2338 +
  1.2339 +    /* Do a full 4x4 inversion as a last resort */
  1.2340 +        
  1.2341 +	matrix = mtx->elem;
  1.2342 +
  1.2343 +    /* transpose matrix */
  1.2344 +    for (i = 0; i < 4; i++) {
  1.2345 +        src[i] = matrix[i*4];
  1.2346 +        src[i+4] = matrix[i*4+1];
  1.2347 +        src[i+8] = matrix[i*4+2];
  1.2348 +        src[i+12] = matrix[i*4+3];
  1.2349 +    }
  1.2350 +
  1.2351 +    /* calculate pairs for first 8 elements (cofactors) */
  1.2352 +    tmp[0] = src[10]*src[15];
  1.2353 +    tmp[1] = src[11]*src[14];
  1.2354 +    tmp[2] = src[9]*src[15];
  1.2355 +    tmp[3] = src[11]*src[13];
  1.2356 +    tmp[4] = src[9]*src[14];
  1.2357 +    tmp[5] = src[10]*src[13];
  1.2358 +    tmp[6] = src[8]*src[15];
  1.2359 +    tmp[7] = src[11]*src[12];
  1.2360 +    tmp[8] = src[8]*src[14];
  1.2361 +    tmp[9] = src[10]*src[12];
  1.2362 +    tmp[10] = src[8]*src[13];
  1.2363 +    tmp[11] = src[9]*src[12];
  1.2364 +
  1.2365 +    /* calculate first 8 elements (cofactors) */
  1.2366 +    matrix[0] = tmp[0]*src[5] + tmp[3]*src[6] + tmp[4]*src[7];
  1.2367 +    matrix[0] -= tmp[1]*src[5] + tmp[2]*src[6] + tmp[5]*src[7];
  1.2368 +    matrix[1] = tmp[1]*src[4] + tmp[6]*src[6] + tmp[9]*src[7];
  1.2369 +    matrix[1] -= tmp[0]*src[4] + tmp[7]*src[6] + tmp[8]*src[7];
  1.2370 +    matrix[2] = tmp[2]*src[4] + tmp[7]*src[5] + tmp[10]*src[7];
  1.2371 +    matrix[2] -= tmp[3]*src[4] + tmp[6]*src[5] + tmp[11]*src[7];
  1.2372 +    matrix[3] = tmp[5]*src[4] + tmp[8]*src[5] + tmp[11]*src[6];
  1.2373 +    matrix[3] -= tmp[4]*src[4] + tmp[9]*src[5] + tmp[10]*src[6];
  1.2374 +    matrix[4] = tmp[1]*src[1] + tmp[2]*src[2] + tmp[5]*src[3];
  1.2375 +    matrix[4] -= tmp[0]*src[1] + tmp[3]*src[2] + tmp[4]*src[3];
  1.2376 +    matrix[5] = tmp[0]*src[0] + tmp[7]*src[2] + tmp[8]*src[3];
  1.2377 +    matrix[5] -= tmp[1]*src[0] + tmp[6]*src[2] + tmp[9]*src[3];
  1.2378 +    matrix[6] = tmp[3]*src[0] + tmp[6]*src[1] + tmp[11]*src[3];
  1.2379 +    matrix[6] -= tmp[2]*src[0] + tmp[7]*src[1] + tmp[10]*src[3];
  1.2380 +    matrix[7] = tmp[4]*src[0] + tmp[9]*src[1] + tmp[10]*src[2];
  1.2381 +    matrix[7] -= tmp[5]*src[0] + tmp[8]*src[1] + tmp[11]*src[2];
  1.2382 +
  1.2383 +    /* calculate pairs for second 8 elements (cofactors) */
  1.2384 +    tmp[0] = src[2]*src[7];
  1.2385 +    tmp[1] = src[3]*src[6];
  1.2386 +    tmp[2] = src[1]*src[7];
  1.2387 +    tmp[3] = src[3]*src[5];
  1.2388 +    tmp[4] = src[1]*src[6];
  1.2389 +    tmp[5] = src[2]*src[5];
  1.2390 +    tmp[6] = src[0]*src[7];
  1.2391 +    tmp[7] = src[3]*src[4];
  1.2392 +    tmp[8] = src[0]*src[6];
  1.2393 +    tmp[9] = src[2]*src[4];
  1.2394 +    tmp[10] = src[0]*src[5];
  1.2395 +    tmp[11] = src[1]*src[4];
  1.2396 +
  1.2397 +    /* calculate second 8 elements (cofactors) */
  1.2398 +    matrix[8] = tmp[0]*src[13] + tmp[3]*src[14] + tmp[4]*src[15];
  1.2399 +    matrix[8] -= tmp[1]*src[13] + tmp[2]*src[14] + tmp[5]*src[15];
  1.2400 +    matrix[9] = tmp[1]*src[12] + tmp[6]*src[14] + tmp[9]*src[15];
  1.2401 +    matrix[9] -= tmp[0]*src[12] + tmp[7]*src[14] + tmp[8]*src[15];
  1.2402 +    matrix[10] = tmp[2]*src[12] + tmp[7]*src[13] + tmp[10]*src[15];
  1.2403 +    matrix[10] -= tmp[3]*src[12] + tmp[6]*src[13] + tmp[11]*src[15];
  1.2404 +    matrix[11] = tmp[5]*src[12] + tmp[8]*src[13] + tmp[11]*src[14];
  1.2405 +    matrix[11] -= tmp[4]*src[12] + tmp[9]*src[13] + tmp[10]*src[14];
  1.2406 +    matrix[12] = tmp[2]*src[10] + tmp[5]*src[11] + tmp[1]*src[9];
  1.2407 +    matrix[12] -= tmp[4]*src[11] + tmp[0]*src[9] + tmp[3]*src[10];
  1.2408 +    matrix[13] = tmp[8]*src[11] + tmp[0]*src[8] + tmp[7]*src[10];
  1.2409 +    matrix[13] -= tmp[6]*src[10] + tmp[9]*src[11] + tmp[1]*src[8];
  1.2410 +    matrix[14] = tmp[6]*src[9] + tmp[11]*src[11] + tmp[3]*src[8];
  1.2411 +    matrix[14] -= tmp[10]*src[11] + tmp[2]*src[8] + tmp[7]*src[9];
  1.2412 +    matrix[15] = tmp[10]*src[10] + tmp[4]*src[8] + tmp[9]*src[9];
  1.2413 +    matrix[15] -= tmp[8]*src[9] + tmp[11]*src[10] + tmp[5]*src[8];
  1.2414 +
  1.2415 +    /* calculate determinant */
  1.2416 +    det = src[0]*matrix[0]+src[1]*matrix[1]+src[2]*matrix[2]+src[3]*matrix[3];
  1.2417 +
  1.2418 +    /* matrix has no inverse */
  1.2419 +    if (det == 0.0f) {
  1.2420 +        return M3G_FALSE;
  1.2421 +    }
  1.2422 +
  1.2423 +    /* calculate matrix inverse */
  1.2424 +    det = 1/det;
  1.2425 +    for (i = 0; i < 16; i++) {
  1.2426 +        matrix[i] *= det;
  1.2427 +    }
  1.2428 +
  1.2429 +    mtx->classified = M3G_FALSE;
  1.2430 +	return M3G_TRUE;
  1.2431 +}
  1.2432 +
  1.2433 +/*!
  1.2434 + * \brief Sets a matrix to the inverse of another matrix
  1.2435 + */
  1.2436 +M3G_API M3Gbool m3gMatrixInverse(Matrix *mtx, const Matrix *other)
  1.2437 +{
  1.2438 +    M3G_ASSERT(mtx != NULL && other != NULL);
  1.2439 +
  1.2440 +    if (!m3gIsClassified(other)) {
  1.2441 +        m3gClassify((Matrix*)other);
  1.2442 +    }
  1.2443 +
  1.2444 +	m3gCopyMatrix(mtx, other);
  1.2445 +	return m3gInvertMatrix(mtx);
  1.2446 +}
  1.2447 +
  1.2448 +/*!
  1.2449 + * \brief Sets a matrix to the transpose of another matrix
  1.2450 + */
  1.2451 +M3G_API void m3gMatrixTranspose(Matrix *mtx, const Matrix *other)
  1.2452 +{
  1.2453 +    M3Gbyte i;
  1.2454 +    M3G_ASSERT(mtx != NULL && other != NULL);
  1.2455 +
  1.2456 +    if (!other->complete) {
  1.2457 +        m3gFillClassifiedMatrix((Matrix *)other);
  1.2458 +    }
  1.2459 +
  1.2460 +    for (i = 0; i < 4; i++) {
  1.2461 +        mtx->elem[i] = other->elem[i*4];
  1.2462 +        mtx->elem[i+4] = other->elem[i*4+1];
  1.2463 +        mtx->elem[i+8] = other->elem[i*4+2];
  1.2464 +        mtx->elem[i+12] = other->elem[i*4+3];
  1.2465 +    }
  1.2466 +    mtx->classified = M3G_FALSE;
  1.2467 +    mtx->complete = M3G_TRUE;
  1.2468 +}
  1.2469 +
  1.2470 +M3G_API M3Gbool m3gInverseTranspose(Matrix *mtx, const Matrix *other)
  1.2471 +{
  1.2472 +    Matrix temp;
  1.2473 +    if (!m3gMatrixInverse(&temp, other)) {
  1.2474 +        return M3G_FALSE;
  1.2475 +    }
  1.2476 +    m3gMatrixTranspose(mtx, &temp);
  1.2477 +    return M3G_TRUE;
  1.2478 +}
  1.2479 +
  1.2480 +/*!
  1.2481 + * \brief Sets a matrix to the product of two other matrices
  1.2482 + *
  1.2483 + * \note \c dst can not be either of \c left or \c right; if it is,
  1.2484 + * the results are undefined
  1.2485 + */
  1.2486 +M3G_API void m3gMatrixProduct(Matrix *dst, const Matrix *left, const Matrix *right)
  1.2487 +{
  1.2488 +    M3G_ASSERT_PTR(dst);
  1.2489 +    M3G_ASSERT_PTR(left);
  1.2490 +    M3G_ASSERT_PTR(right);
  1.2491 +    M3G_ASSERT(dst != left && dst != right);
  1.2492 +
  1.2493 +    /* Classify input matrices and take early exits for identities */
  1.2494 +
  1.2495 +    if (!m3gIsClassified(left)) {
  1.2496 +        m3gClassify((Matrix*)left);
  1.2497 +    }
  1.2498 +    if (left->mask == MC_IDENTITY) {
  1.2499 +        m3gCopyMatrix(dst, right);
  1.2500 +        return;
  1.2501 +    }
  1.2502 +
  1.2503 +    if (!m3gIsClassified(right)) {
  1.2504 +        m3gClassify((Matrix*)right);
  1.2505 +    }
  1.2506 +    if (right->mask == MC_IDENTITY) {
  1.2507 +        m3gCopyMatrix(dst, left);
  1.2508 +        return;
  1.2509 +    }
  1.2510 +
  1.2511 +    /* Special quick paths for 3x4 matrices */
  1.2512 +    
  1.2513 +    if (m3gIsWUnity(left) && m3gIsWUnity(right)) {
  1.2514 +
  1.2515 +        /* Translation? */
  1.2516 +        
  1.2517 +        if ((left->mask & ~MC_TRANSLATION_PART) == MC_IDENTITY) {
  1.2518 +            
  1.2519 +            if (left->mask != MC_TRANSLATION && !left->complete) {
  1.2520 +                m3gFillClassifiedMatrix((Matrix*)left);
  1.2521 +            }
  1.2522 +            if (right->mask != MC_TRANSLATION && !right->complete) {
  1.2523 +                m3gFillClassifiedMatrix((Matrix*)right);
  1.2524 +            }
  1.2525 +
  1.2526 +            m3gCopyMatrix(dst, right);
  1.2527 +            
  1.2528 +            M44F(dst, 0, 3) = m3gAdd(M44F(left, 0, 3), M44F(dst, 0, 3));
  1.2529 +            M44F(dst, 1, 3) = m3gAdd(M44F(left, 1, 3), M44F(dst, 1, 3));
  1.2530 +            M44F(dst, 2, 3) = m3gAdd(M44F(left, 2, 3), M44F(dst, 2, 3));
  1.2531 +            
  1.2532 +            dst->mask |= MC_TRANSLATION_PART;
  1.2533 +            return;
  1.2534 +        }
  1.2535 +
  1.2536 +        if ((right->mask & ~MC_TRANSLATION_PART) == MC_IDENTITY) {
  1.2537 +            Vec4 tvec;
  1.2538 +
  1.2539 +            if (left->mask != MC_TRANSLATION && !left->complete) {
  1.2540 +                m3gFillClassifiedMatrix((Matrix*)left);
  1.2541 +            }
  1.2542 +            if (right->mask != MC_TRANSLATION && !right->complete) {
  1.2543 +                m3gFillClassifiedMatrix((Matrix*)right);
  1.2544 +            }
  1.2545 +
  1.2546 +            m3gCopyMatrix(dst, left);
  1.2547 +            
  1.2548 +            m3gGetMatrixColumn(right, 3, &tvec);
  1.2549 +            m3gTransformVec4(dst, &tvec);
  1.2550 +            
  1.2551 +            M44F(dst, 0, 3) = tvec.x;
  1.2552 +            M44F(dst, 1, 3) = tvec.y;
  1.2553 +            M44F(dst, 2, 3) = tvec.z;
  1.2554 +            
  1.2555 +            dst->mask |= MC_TRANSLATION_PART;
  1.2556 +            return;
  1.2557 +        }
  1.2558 +
  1.2559 +    }
  1.2560 +        
  1.2561 +    /* Compute product and set output classification */
  1.2562 +
  1.2563 +    m3gGenericMatrixProduct(dst, left, right);
  1.2564 +}
  1.2565 +
  1.2566 +/*!
  1.2567 + * \brief Normalizes a quaternion
  1.2568 + */
  1.2569 +M3G_API void m3gNormalizeQuat(Quat *q)
  1.2570 +{
  1.2571 +    M3Gfloat norm;
  1.2572 +    M3G_ASSERT_PTR(q);
  1.2573 +    
  1.2574 +    norm = m3gNorm4(&q->x);
  1.2575 +    
  1.2576 +    if (norm > EPSILON) {
  1.2577 +        norm = m3gRcpSqrt(norm);
  1.2578 +        m3gScale4(&q->x, norm);
  1.2579 +    }
  1.2580 +    else {
  1.2581 +        m3gIdentityQuat(q);
  1.2582 +    }
  1.2583 +}
  1.2584 +
  1.2585 +/*!
  1.2586 + * \brief Normalizes a three-vector
  1.2587 + */
  1.2588 +M3G_API void m3gNormalizeVec3(Vec3 *v)
  1.2589 +{
  1.2590 +    M3Gfloat norm;
  1.2591 +    M3G_ASSERT_PTR(v);
  1.2592 +    
  1.2593 +    norm = m3gNorm3(&v->x);
  1.2594 +    
  1.2595 +    if (norm > EPSILON) {
  1.2596 +        norm = m3gRcpSqrt(norm);
  1.2597 +        m3gScale3(&v->x, norm);
  1.2598 +    }
  1.2599 +    else {
  1.2600 +        m3gZero(v, sizeof(Vec3));
  1.2601 +    }
  1.2602 +}
  1.2603 +
  1.2604 +/*!
  1.2605 + * \brief Normalizes a four-vector
  1.2606 + */
  1.2607 +M3G_API void m3gNormalizeVec4(Vec4 *v)
  1.2608 +{
  1.2609 +    M3Gfloat norm;
  1.2610 +    M3G_ASSERT_PTR(v);
  1.2611 +    
  1.2612 +    norm = m3gNorm4(&v->x);
  1.2613 +    
  1.2614 +    if (norm > EPSILON) {
  1.2615 +        norm = m3gRcpSqrt(norm);
  1.2616 +        m3gScale4(&v->x, norm);
  1.2617 +    }
  1.2618 +    else {
  1.2619 +        m3gZero(v, sizeof(Vec4));
  1.2620 +    }
  1.2621 +}
  1.2622 +
  1.2623 +/*!
  1.2624 + * \brief Multiplies a matrix from the right with another matrix
  1.2625 + */
  1.2626 +M3G_API void m3gPostMultiplyMatrix(Matrix *mtx, const Matrix *other)
  1.2627 +{
  1.2628 +    Matrix temp;
  1.2629 +    M3G_ASSERT_PTR(mtx);
  1.2630 +    M3G_ASSERT_PTR(other);
  1.2631 +
  1.2632 +    m3gCopyMatrix(&temp, mtx);
  1.2633 +    m3gMatrixProduct(mtx, &temp, other);
  1.2634 +}
  1.2635 +
  1.2636 +/*!
  1.2637 + * \brief Multiplies a matrix from the left with another matrix
  1.2638 + */
  1.2639 +M3G_API void m3gPreMultiplyMatrix(Matrix *mtx, const Matrix *other)
  1.2640 +{
  1.2641 +    Matrix temp;
  1.2642 +    M3G_ASSERT_PTR(mtx);
  1.2643 +    M3G_ASSERT_PTR(other);
  1.2644 +
  1.2645 +    m3gCopyMatrix(&temp, mtx);
  1.2646 +    m3gMatrixProduct(mtx, other, &temp);
  1.2647 +}
  1.2648 +
  1.2649 +/*!
  1.2650 + * \brief Multiplies a matrix with a rotation matrix.
  1.2651 + */
  1.2652 +M3G_API void m3gPostRotateMatrix(Matrix *mtx,
  1.2653 +                         M3Gfloat angle,
  1.2654 +                         M3Gfloat ax, M3Gfloat ay, M3Gfloat az)
  1.2655 +{
  1.2656 +    Quat q;
  1.2657 +    m3gSetAngleAxis(&q, angle, ax, ay, az);
  1.2658 +    m3gPostRotateMatrixQuat(mtx, &q);
  1.2659 +}
  1.2660 +
  1.2661 +/*!
  1.2662 + * \brief Multiplies a matrix with a translation matrix.
  1.2663 + */
  1.2664 +M3G_API void m3gPostRotateMatrixQuat(Matrix *mtx, const Quat *quat)
  1.2665 +{
  1.2666 +    Matrix temp;
  1.2667 +    m3gQuatMatrix(&temp, quat);
  1.2668 +    m3gPostMultiplyMatrix(mtx, &temp);
  1.2669 +}
  1.2670 +
  1.2671 +/*!
  1.2672 + * \brief Multiplies a matrix with a scale matrix.
  1.2673 + */
  1.2674 +M3G_API void m3gPostScaleMatrix(Matrix *mtx, M3Gfloat sx, M3Gfloat sy, M3Gfloat sz)
  1.2675 +{
  1.2676 +    Matrix temp;
  1.2677 +    m3gScalingMatrix(&temp, sx, sy, sz);
  1.2678 +    m3gPostMultiplyMatrix(mtx, &temp);
  1.2679 +}
  1.2680 +
  1.2681 +/*!
  1.2682 + * \brief Multiplies a matrix with a translation (matrix).
  1.2683 + */
  1.2684 +M3G_API void m3gPostTranslateMatrix(Matrix *mtx,
  1.2685 +                            M3Gfloat tx, M3Gfloat ty, M3Gfloat tz)
  1.2686 +{
  1.2687 +    Matrix temp;
  1.2688 +    m3gTranslationMatrix(&temp, tx, ty, tz);
  1.2689 +    m3gPostMultiplyMatrix(mtx, &temp);
  1.2690 +}
  1.2691 +
  1.2692 +/*!
  1.2693 + * \brief Multiplies a matrix with a rotation matrix
  1.2694 + */
  1.2695 +M3G_API void m3gPreRotateMatrix(Matrix *mtx,
  1.2696 +                        M3Gfloat angle,
  1.2697 +                        M3Gfloat ax, M3Gfloat ay, M3Gfloat az)
  1.2698 +{
  1.2699 +    Quat q;
  1.2700 +    m3gSetAngleAxis(&q, angle, ax, ay, az);
  1.2701 +    m3gPreRotateMatrixQuat(mtx, &q);
  1.2702 +}
  1.2703 +
  1.2704 +/*!
  1.2705 + * \brief Multiplies a matrix with a quaternion rotation
  1.2706 + */
  1.2707 +M3G_API void m3gPreRotateMatrixQuat(Matrix *mtx, const Quat *quat)
  1.2708 +{
  1.2709 +    Matrix temp;
  1.2710 +    m3gQuatMatrix(&temp, quat);
  1.2711 +    m3gPreMultiplyMatrix(mtx, &temp);
  1.2712 +}
  1.2713 +
  1.2714 +/*!
  1.2715 + * \brief Multiplies a matrix with a scale matrix.
  1.2716 + */
  1.2717 +M3G_API void m3gPreScaleMatrix(Matrix *mtx, M3Gfloat sx, M3Gfloat sy, M3Gfloat sz)
  1.2718 +{
  1.2719 +    Matrix temp;
  1.2720 +    m3gScalingMatrix(&temp, sx, sy, sz);
  1.2721 +    m3gPreMultiplyMatrix(mtx, &temp);
  1.2722 +}
  1.2723 +
  1.2724 +/*!
  1.2725 + * \brief Multiplies a matrix with a translation (matrix).
  1.2726 + */
  1.2727 +M3G_API void m3gPreTranslateMatrix(Matrix *mtx,
  1.2728 +                           M3Gfloat tx, M3Gfloat ty, M3Gfloat tz)
  1.2729 +{
  1.2730 +    Matrix temp;
  1.2731 +    m3gTranslationMatrix(&temp, tx, ty, tz);
  1.2732 +    m3gPreMultiplyMatrix(mtx, &temp);
  1.2733 +}
  1.2734 +
  1.2735 +/*!
  1.2736 + * \brief Converts a quaternion into a matrix
  1.2737 + *
  1.2738 + * The output is a matrix effecting the same rotation as the
  1.2739 + * quaternion passed as input
  1.2740 + */
  1.2741 +M3G_API void m3gQuatMatrix(Matrix *mtx, const Quat *quat)
  1.2742 +{
  1.2743 +    M3Gfloat qx = quat->x;
  1.2744 +    M3Gfloat qy = quat->y;
  1.2745 +    M3Gfloat qz = quat->z;
  1.2746 +    M3Gfloat qw = quat->w;
  1.2747 +
  1.2748 +    /* Quick exit for identity rotations */
  1.2749 +
  1.2750 +    if (IS_ZERO(qx) && IS_ZERO(qy) && IS_ZERO(qz)) {
  1.2751 +        m3gIdentityMatrix(mtx);
  1.2752 +        return;
  1.2753 +    }
  1.2754 +
  1.2755 +    {
  1.2756 +        /* Determine the rough type of the output matrix */
  1.2757 +
  1.2758 +        M3Guint type = MC_SCALING_ROTATION;
  1.2759 +        if (IS_ZERO(qz) && IS_ZERO(qy)) {
  1.2760 +            type = MC_X_ROTATION;
  1.2761 +        }
  1.2762 +        else if (IS_ZERO(qz) && IS_ZERO(qx)) {
  1.2763 +            type = MC_Y_ROTATION;
  1.2764 +        }
  1.2765 +        else if (IS_ZERO(qx) && IS_ZERO(qy)) {
  1.2766 +            type = MC_Z_ROTATION;
  1.2767 +        }
  1.2768 +        m3gClassifyAs(type, mtx);
  1.2769 +
  1.2770 +        /* Generate the non-constant parts of the matrix */
  1.2771 +        {
  1.2772 +            M3Gfloat wx, wy, wz, xx, yy, yz, xy, xz, zz;
  1.2773 +
  1.2774 +            xx = m3gMul(qx, qx);
  1.2775 +            xy = m3gMul(qx, qy);
  1.2776 +            xz = m3gMul(qx, qz);
  1.2777 +            yy = m3gMul(qy, qy);
  1.2778 +            yz = m3gMul(qy, qz);
  1.2779 +            zz = m3gMul(qz, qz);
  1.2780 +            wx = m3gMul(qw, qx);
  1.2781 +            wy = m3gMul(qw, qy);
  1.2782 +            wz = m3gMul(qw, qz);
  1.2783 +
  1.2784 +            if (type != MC_X_ROTATION) {
  1.2785 +                M44F(mtx, 0, 0) = m3gSub(1.f, m3gDouble(m3gAdd(yy, zz)));
  1.2786 +                M44F(mtx, 0, 1) =             m3gDouble(m3gSub(xy, wz));
  1.2787 +                M44F(mtx, 0, 2) =             m3gDouble(m3gAdd(xz, wy));
  1.2788 +            }
  1.2789 +
  1.2790 +            if (type != MC_Y_ROTATION) {
  1.2791 +                M44F(mtx, 1, 0) =             m3gDouble(m3gAdd(xy, wz));
  1.2792 +                M44F(mtx, 1, 1) = m3gSub(1.f, m3gDouble(m3gAdd(xx, zz)));
  1.2793 +                M44F(mtx, 1, 2) =             m3gDouble(m3gSub(yz, wx));
  1.2794 +            }
  1.2795 +
  1.2796 +            if (type != MC_Z_ROTATION) {
  1.2797 +                M44F(mtx, 2, 0) =             m3gDouble(m3gSub(xz, wy));
  1.2798 +                M44F(mtx, 2, 1) =             m3gDouble(m3gAdd(yz, wx));
  1.2799 +                M44F(mtx, 2, 2) = m3gSub(1.f, m3gDouble(m3gAdd(xx, yy)));
  1.2800 +            }
  1.2801 +        }
  1.2802 +        m3gSubClassify(mtx);
  1.2803 +    }
  1.2804 +}
  1.2805 +
  1.2806 +/*!
  1.2807 + * \brief Generates a scaling matrix
  1.2808 + */
  1.2809 +M3G_API void m3gScalingMatrix(
  1.2810 +    Matrix *mtx,
  1.2811 +    const M3Gfloat sx, const M3Gfloat sy, const M3Gfloat sz)
  1.2812 +{
  1.2813 +    M3G_ASSERT_PTR(mtx);
  1.2814 +    M44F(mtx, 0, 0) = sx;
  1.2815 +    M44F(mtx, 1, 1) = sy;
  1.2816 +    M44F(mtx, 2, 2) = sz;
  1.2817 +    m3gClassifyAs(MC_SCALING, mtx);
  1.2818 +    m3gSubClassify(mtx);
  1.2819 +}
  1.2820 +
  1.2821 +/*!
  1.2822 + * \brief Sets a quaternion to represent an angle-axis rotation
  1.2823 + */
  1.2824 +M3G_API void m3gSetAngleAxis(Quat *quat,
  1.2825 +                     M3Gfloat angle,
  1.2826 +                     M3Gfloat ax, M3Gfloat ay, M3Gfloat az)
  1.2827 +{
  1.2828 +    m3gSetAngleAxisRad(quat, m3gMul(angle, M3G_DEG2RAD), ax, ay, az);
  1.2829 +}
  1.2830 +
  1.2831 +/*!
  1.2832 + * \brief Sets a quaternion to represent an angle-axis rotation
  1.2833 + */
  1.2834 +M3G_API void m3gSetAngleAxisRad(Quat *quat,
  1.2835 +                        M3Gfloat angleRad,
  1.2836 +                        M3Gfloat ax, M3Gfloat ay, M3Gfloat az)
  1.2837 +{
  1.2838 +    M3G_ASSERT_PTR(quat);
  1.2839 +    
  1.2840 +    if (!IS_ZERO(angleRad)) {
  1.2841 +        M3Gfloat s;
  1.2842 +        M3Gfloat halfAngle = m3gHalf(angleRad);
  1.2843 +
  1.2844 +        s = m3gSin(halfAngle);
  1.2845 +        
  1.2846 +        {
  1.2847 +            M3Gfloat sqrNorm = m3gMadd(ax, ax, m3gMadd(ay, ay, m3gMul(az, az)));
  1.2848 +            if (sqrNorm < 0.995f || sqrNorm > 1.005f) {
  1.2849 +                if (sqrNorm > EPSILON) {
  1.2850 +                    M3Gfloat ooNorm = m3gRcpSqrt(sqrNorm);
  1.2851 +                    ax = m3gMul(ax, ooNorm);
  1.2852 +                    ay = m3gMul(ay, ooNorm);
  1.2853 +                    az = m3gMul(az, ooNorm);
  1.2854 +                }
  1.2855 +                else {
  1.2856 +                    ax = ay = az = 0.0f;
  1.2857 +                }
  1.2858 +            }
  1.2859 +        }
  1.2860 +
  1.2861 +        quat->x = m3gMul(s, ax);
  1.2862 +        quat->y = m3gMul(s, ay);
  1.2863 +        quat->z = m3gMul(s, az);
  1.2864 +        quat->w = m3gCos(halfAngle);
  1.2865 +    }
  1.2866 +    else {
  1.2867 +        m3gIdentityQuat(quat);
  1.2868 +    }
  1.2869 +}
  1.2870 +
  1.2871 +/*!
  1.2872 + * \brief Quaternion multiplication.
  1.2873 + */
  1.2874 +M3G_API void m3gMulQuat(Quat *quat, const Quat *other)
  1.2875 +{
  1.2876 +    Quat q;
  1.2877 +    q = *quat;
  1.2878 +
  1.2879 +    quat->w = m3gMul(q.w, other->w) - m3gMul(q.x, other->x) - m3gMul(q.y, other->y) - m3gMul(q.z, other->z);
  1.2880 +    quat->x = m3gMul(q.w, other->x) + m3gMul(q.x, other->w) + m3gMul(q.y, other->z) - m3gMul(q.z, other->y);
  1.2881 +    quat->y = m3gMul(q.w, other->y) - m3gMul(q.x, other->z) + m3gMul(q.y, other->w) + m3gMul(q.z, other->x);
  1.2882 +    quat->z = m3gMul(q.w, other->z) + m3gMul(q.x, other->y) - m3gMul(q.y, other->x) + m3gMul(q.z, other->w);
  1.2883 +}
  1.2884 +
  1.2885 +/*!
  1.2886 + * \brief Makes this quaternion represent the rotation from one 3D
  1.2887 + * vector to another
  1.2888 + */
  1.2889 +M3G_API void m3gSetQuatRotation(Quat *quat,
  1.2890 +                                const Vec3 *from, const Vec3 *to)
  1.2891 +{
  1.2892 +    M3Gfloat cosAngle;
  1.2893 +
  1.2894 +    M3G_ASSERT_PTR(quat);
  1.2895 +    M3G_ASSERT_PTR(from);
  1.2896 +    M3G_ASSERT_PTR(to);
  1.2897 +
  1.2898 +    cosAngle = m3gDot3(from, to);
  1.2899 +
  1.2900 +    if (cosAngle > (1.0f - EPSILON)) {  /* zero angle */
  1.2901 +        m3gIdentityQuat(quat);
  1.2902 +        return;
  1.2903 +    }
  1.2904 +    else if (cosAngle > (1.0e-3f - 1.0f)) { /* normal case */
  1.2905 +        Vec3 axis;
  1.2906 +        m3gCross(&axis, from, to);
  1.2907 +        m3gSetAngleAxisRad(quat, m3gArcCos(cosAngle), axis.x, axis.y, axis.z);
  1.2908 +    }
  1.2909 +    else {
  1.2910 +
  1.2911 +        /* Opposite vectors; must generate an arbitrary perpendicular
  1.2912 +         * vector and use that as the rotation axis. Here, we try the
  1.2913 +         * Z axis first, and if that seems too parallel to the
  1.2914 +         * vectors, project the Y axis instead: Z is the only good
  1.2915 +         * choice for Z-constrained rotations, and Y by definition
  1.2916 +         * must be perpendicular to that. */
  1.2917 +
  1.2918 +        Vec3 axis, temp;
  1.2919 +        M3Gfloat s;
  1.2920 +
  1.2921 +        axis.x = axis.y = axis.z = 0.0f;
  1.2922 +        if (m3gAbs(from->z) < (1.0f - EPSILON)) {
  1.2923 +            axis.z = 1.0f;
  1.2924 +        }
  1.2925 +        else {
  1.2926 +            axis.y = 1.0f;
  1.2927 +        }
  1.2928 +
  1.2929 +        s = m3gDot3(&axis, from);
  1.2930 +        temp = *from;
  1.2931 +        m3gScaleVec3(&temp, s);
  1.2932 +        m3gSubVec3(&axis, &temp);
  1.2933 +
  1.2934 +        m3gSetAngleAxis(quat, 180.f, axis.x, axis.y, axis.z);
  1.2935 +    }
  1.2936 +}
  1.2937 +
  1.2938 +/*!
  1.2939 + * \brief Sets the values of a matrix
  1.2940 + *
  1.2941 + */
  1.2942 +M3G_API void m3gSetMatrixColumns(Matrix *mtx, const M3Gfloat *src)
  1.2943 +{
  1.2944 +    M3G_ASSERT(mtx != NULL && src != NULL);
  1.2945 +
  1.2946 +    m3gCopy(mtx->elem, src, sizeof(mtx->elem));
  1.2947 +    mtx->classified = M3G_FALSE;
  1.2948 +    mtx->complete = M3G_TRUE;
  1.2949 +}
  1.2950 +
  1.2951 +/*!
  1.2952 + * \brief Sets the values of a matrix
  1.2953 + *
  1.2954 + */
  1.2955 +M3G_API void m3gSetMatrixRows(Matrix *mtx, const M3Gfloat *src)
  1.2956 +{
  1.2957 +    M3G_ASSERT(mtx != NULL && src != NULL);
  1.2958 +    {
  1.2959 +        int row;
  1.2960 +        for (row = 0; row < 4; ++row) {
  1.2961 +            mtx->elem[ 0 + row] = *src++;
  1.2962 +            mtx->elem[ 4 + row] = *src++;
  1.2963 +            mtx->elem[ 8 + row] = *src++;
  1.2964 +            mtx->elem[12 + row] = *src++;
  1.2965 +        }
  1.2966 +    }
  1.2967 +    mtx->classified = M3G_FALSE;
  1.2968 +    mtx->complete = M3G_TRUE;
  1.2969 +}
  1.2970 +
  1.2971 +/*!
  1.2972 + * \brief Transforms a 4-vector with a matrix
  1.2973 + */
  1.2974 +#if defined(M3G_HW_FLOAT_VFPV2)
  1.2975 +
  1.2976 +__asm void _m3gTransformVec4(const Matrix *mtx, Vec4 *vec, M3Gint n)
  1.2977 +{
  1.2978 +
  1.2979 +		CODE32
  1.2980 +
  1.2981 +		FSTMFDD	sp!, {d8-d11}
  1.2982 +
  1.2983 +		FMRX	r3, FPSCR		 
  1.2984 +		BIC		r12, r3, #0x00370000
  1.2985 +		ORR		r12, #(3<<16)
  1.2986 +		FMXR	FPSCR, r12
  1.2987 +		CMP		r2, #4
  1.2988 +
  1.2989 +		FLDMIAS	r0, {s4-s19}		// [mtx0  mtx1 ..]
  1.2990 +		FLDMIAS	r1, {s0-s3}			// [vec.x  vec.y  vec.z  vec.w]
  1.2991 +		FMULS	s20, s4, s0			// [s20-s23]  = [v.x*mtx0  v.x*mtx1  v.x*mtx2  v.x*mtx3 ]
  1.2992 +		FMACS	s20, s8, s1			// [s20-s23] += [v.y*mtx4  v.y*mtx5  v.y*mtx6  v.y*mtx7 ]
  1.2993 +		FMACS	s20, s12, s2		// [s20-s23] += [v.z*mtx8  v.z*mtx9  v.z*mtx10 v.z*mtx11]
  1.2994 +		FMACS	s20, s16, s3		// [s20-s23] += [v.w*mtx12 v.w*mtx13 v.w*mtx14 v.w*mtx15]
  1.2995 +		FSTMIAS		r1!, {s20-s22}
  1.2996 +		FSTMIASEQ	r1, {s23}
  1.2997 +
  1.2998 +		FMXR	FPSCR, r3
  1.2999 +		FLDMFDD	sp!, {d8-d11}
  1.3000 +
  1.3001 +		BX		lr
  1.3002 +}
  1.3003 +#endif /* #if defined(M3G_HW_FLOAT_VFPV2) */
  1.3004 +
  1.3005 +M3G_API void m3gTransformVec4(const Matrix *mtx, Vec4 *vec)
  1.3006 +{
  1.3007 +    M3Guint type;
  1.3008 +    M3G_ASSERT(mtx != NULL && vec != NULL);
  1.3009 +
  1.3010 +    if (!m3gIsClassified(mtx)) {
  1.3011 +        m3gClassify((Matrix*)mtx);
  1.3012 +    }
  1.3013 +
  1.3014 +    type = mtx->mask;
  1.3015 +
  1.3016 +    if (type == MC_IDENTITY) {
  1.3017 +        return;
  1.3018 +    }
  1.3019 +    else {
  1.3020 +        int n = m3gIsWUnity(mtx) ? 3 : 4;
  1.3021 +
  1.3022 +        if (!mtx->complete) {
  1.3023 +            m3gFillClassifiedMatrix((Matrix*)mtx);
  1.3024 +        }
  1.3025 +#if	defined(M3G_HW_FLOAT_VFPV2)
  1.3026 +		_m3gTransformVec4(mtx, vec, n);
  1.3027 +#else
  1.3028 +        {
  1.3029 +            Vec4 v = *vec;
  1.3030 +            int i;
  1.3031 +
  1.3032 +            for (i = 0; i < n; ++i) {
  1.3033 +                M3Gfloat d = m3gMul(M44F(mtx, i, 0), v.x);
  1.3034 +                d = m3gMadd(M44F(mtx, i, 1), v.y, d);
  1.3035 +                d = m3gMadd(M44F(mtx, i, 2), v.z, d);
  1.3036 +                d = m3gMadd(M44F(mtx, i, 3), v.w, d);
  1.3037 +                (&vec->x)[i] = d;
  1.3038 +            }
  1.3039 +        }
  1.3040 +#endif
  1.3041 +    }
  1.3042 +}
  1.3043 +
  1.3044 +/*!
  1.3045 + * \brief Generates a translation matrix
  1.3046 + */
  1.3047 +M3G_API void m3gTranslationMatrix(
  1.3048 +    Matrix *mtx,
  1.3049 +    const M3Gfloat tx, const M3Gfloat ty, const M3Gfloat tz)
  1.3050 +{
  1.3051 +    M3G_ASSERT_PTR(mtx);
  1.3052 +    M44F(mtx, 0, 3) = tx;
  1.3053 +    M44F(mtx, 1, 3) = ty;
  1.3054 +    M44F(mtx, 2, 3) = tz;
  1.3055 +    m3gClassifyAs(MC_TRANSLATION, mtx);
  1.3056 +    m3gSubClassify(mtx);
  1.3057 +}
  1.3058 +
  1.3059 +/*!
  1.3060 + * \brief Float vector assignment.
  1.3061 + */
  1.3062 +M3G_API void m3gSetQuat(Quat *quat, const M3Gfloat *vec)
  1.3063 +{
  1.3064 +    quat->x = vec[0];
  1.3065 +    quat->y = vec[1];
  1.3066 +    quat->z = vec[2];
  1.3067 +    quat->w = vec[3];
  1.3068 +}
  1.3069 +
  1.3070 +/*!
  1.3071 + * \brief Slerp between quaternions q0 and q1, storing the result in quat.
  1.3072 + */
  1.3073 +M3G_API void m3gSlerpQuat(Quat *quat,
  1.3074 +                  M3Gfloat s,
  1.3075 +                  const Quat *q0, const Quat *q1)
  1.3076 +{
  1.3077 +    M3Gfloat s0, s1;
  1.3078 +    M3Gfloat cosTheta = m3gDot4((const Vec4 *)q0, (const Vec4 *)q1);
  1.3079 +    M3Gfloat oneMinusS = m3gSub(1.0f, s);
  1.3080 +
  1.3081 +    if (cosTheta > EPSILON - 1.0f) {
  1.3082 +        if (cosTheta < 1.0f - EPSILON) {
  1.3083 +            M3Gfloat theta    = m3gArcCos(cosTheta);
  1.3084 +            M3Gfloat sinTheta = m3gSin(theta);
  1.3085 +            s0 = m3gSin(m3gMul(oneMinusS, theta)) / sinTheta;
  1.3086 +            s1 = m3gSin(m3gMul(s, theta)) / sinTheta;
  1.3087 +        }
  1.3088 +        else {
  1.3089 +            /* For quaternions very close to each other, use plain
  1.3090 +               linear interpolation for numerical stability. */
  1.3091 +            s0 = oneMinusS;
  1.3092 +            s1 = s;
  1.3093 +        }
  1.3094 +        quat->x = m3gMadd(s0, q0->x, m3gMul(s1, q1->x));
  1.3095 +        quat->y = m3gMadd(s0, q0->y, m3gMul(s1, q1->y));
  1.3096 +        quat->z = m3gMadd(s0, q0->z, m3gMul(s1, q1->z));
  1.3097 +        quat->w = m3gMadd(s0, q0->w, m3gMul(s1, q1->w));
  1.3098 +    }
  1.3099 +    else {
  1.3100 +        /* Slerp is undefined if the two quaternions are (nearly)
  1.3101 +           opposite, so we just pick an arbitrary plane of
  1.3102 +           rotation here. */
  1.3103 +
  1.3104 +        quat->x = -q0->y;
  1.3105 +        quat->y =  q0->x;
  1.3106 +        quat->z = -q0->w;
  1.3107 +        quat->w =  q0->z;
  1.3108 +
  1.3109 +        s0 = m3gSin(m3gMul(oneMinusS, HALF_PI));
  1.3110 +        s1 = m3gSin(m3gMul(s, HALF_PI));
  1.3111 +
  1.3112 +        quat->x = m3gMadd(s0, q0->x, m3gMul(s1, quat->x));
  1.3113 +        quat->y = m3gMadd(s0, q0->y, m3gMul(s1, quat->y));
  1.3114 +        quat->z = m3gMadd(s0, q0->z, m3gMul(s1, quat->z));
  1.3115 +    }
  1.3116 +}
  1.3117 +
  1.3118 +/*!
  1.3119 + * \brief Interpolate quaternions using spline, or "squad" interpolation.
  1.3120 + */
  1.3121 +M3G_API void m3gSquadQuat(Quat *quat,
  1.3122 +                  M3Gfloat s,
  1.3123 +                  const Quat *q0, const Quat *a, const Quat *b, const Quat *q1)
  1.3124 +{
  1.3125 +
  1.3126 +    Quat temp0;
  1.3127 +    Quat temp1;
  1.3128 +    m3gSlerpQuat(&temp0, s, q0, q1);
  1.3129 +    m3gSlerpQuat(&temp1, s, a, b);
  1.3130 +
  1.3131 +    m3gSlerpQuat(quat, m3gDouble(m3gMul(s, m3gSub(1.0f, s))), &temp0, &temp1);
  1.3132 +}
  1.3133 +
  1.3134 +
  1.3135 +