brl/mutter
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
7949cfe30a
mutter/clutter/clutter-fixed.c
Tomas Frydrych 5b92cc21cb fixed endianness issue in fast fp conversions
2007-07-25 18:34:05 +00:00

925 lines
26 KiB
C
Raw Blame History

/*
* Clutter.
*
* An OpenGL based 'interactive canvas' library.
*
* Authored By Tomas Frydrych <tf@openedhand.com>
*
* Copyright (C) 2006, 2007 OpenedHand
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, write to the
* Free Software Foundation, Inc., 59 Temple Place - Suite 330,
* Boston, MA 02111-1307, USA.
*/
#include "config.h"
#include <clutter-fixed.h>
/**
* SECTION:clutter-fixed
* @short_description: Fixed Point API
*
* Clutter has a fixed point API targeted at platforms without a floating
* point unit, such as embedded devices. This API should be preferred to
* the floating point one as it does not trigger the slow path of software
* emulation, relying on integer math for fixed-to-floating and
* floating-to-fixed conversion.
*
* Basic rules of Fixed Point arithmethic:
*
* <itemizedlist>
* <listitem>
* <para>Two fixed point numbers can be directly added and
* subtracted.</para>
* </listitem>
* <listitem>
* <para>To add other numerical type to a fixed point number it has to
* be first converted to fixed point.</para>
* </listitem>
* <listitem>
* <para>A fixed point number can be directly multiplied or divided by
* an integer.</para>
* </listitem>
* <listitem>
* <para>Two fixed point numbers can only be multiplied and divided by the
* provided #CLUTTER_FIXED_MUL and #CLUTTER_FIXED_DIV macros.</para>
* </listitem>
* </itemizedlist>
*/
/* pre-computed sin table for 1st quadrant
*
* Currently contains 257 entries.
*
* The current error (compared to system sin) is about
* 0.5% for values near the start of the table where the
* curve is steep, but improving rapidly. If this precission
* is not enough, we can increase the size of the table
*/
static ClutterFixed sin_tbl [] =
{
0x00000000L, 0x00000192L, 0x00000324L, 0x000004B6L,
0x00000648L, 0x000007DAL, 0x0000096CL, 0x00000AFEL,
0x00000C90L, 0x00000E21L, 0x00000FB3L, 0x00001144L,
0x000012D5L, 0x00001466L, 0x000015F7L, 0x00001787L,
0x00001918L, 0x00001AA8L, 0x00001C38L, 0x00001DC7L,
0x00001F56L, 0x000020E5L, 0x00002274L, 0x00002402L,
0x00002590L, 0x0000271EL, 0x000028ABL, 0x00002A38L,
0x00002BC4L, 0x00002D50L, 0x00002EDCL, 0x00003067L,
0x000031F1L, 0x0000337CL, 0x00003505L, 0x0000368EL,
0x00003817L, 0x0000399FL, 0x00003B27L, 0x00003CAEL,
0x00003E34L, 0x00003FBAL, 0x0000413FL, 0x000042C3L,
0x00004447L, 0x000045CBL, 0x0000474DL, 0x000048CFL,
0x00004A50L, 0x00004BD1L, 0x00004D50L, 0x00004ECFL,
0x0000504DL, 0x000051CBL, 0x00005348L, 0x000054C3L,
0x0000563EL, 0x000057B9L, 0x00005932L, 0x00005AAAL,
0x00005C22L, 0x00005D99L, 0x00005F0FL, 0x00006084L,
0x000061F8L, 0x0000636BL, 0x000064DDL, 0x0000664EL,
0x000067BEL, 0x0000692DL, 0x00006A9BL, 0x00006C08L,
0x00006D74L, 0x00006EDFL, 0x00007049L, 0x000071B2L,
0x0000731AL, 0x00007480L, 0x000075E6L, 0x0000774AL,
0x000078ADL, 0x00007A10L, 0x00007B70L, 0x00007CD0L,
0x00007E2FL, 0x00007F8CL, 0x000080E8L, 0x00008243L,
0x0000839CL, 0x000084F5L, 0x0000864CL, 0x000087A1L,
0x000088F6L, 0x00008A49L, 0x00008B9AL, 0x00008CEBL,
0x00008E3AL, 0x00008F88L, 0x000090D4L, 0x0000921FL,
0x00009368L, 0x000094B0L, 0x000095F7L, 0x0000973CL,
0x00009880L, 0x000099C2L, 0x00009B03L, 0x00009C42L,
0x00009D80L, 0x00009EBCL, 0x00009FF7L, 0x0000A130L,
0x0000A268L, 0x0000A39EL, 0x0000A4D2L, 0x0000A605L,
0x0000A736L, 0x0000A866L, 0x0000A994L, 0x0000AAC1L,
0x0000ABEBL, 0x0000AD14L, 0x0000AE3CL, 0x0000AF62L,
0x0000B086L, 0x0000B1A8L, 0x0000B2C9L, 0x0000B3E8L,
0x0000B505L, 0x0000B620L, 0x0000B73AL, 0x0000B852L,
0x0000B968L, 0x0000BA7DL, 0x0000BB8FL, 0x0000BCA0L,
0x0000BDAFL, 0x0000BEBCL, 0x0000BFC7L, 0x0000C0D1L,
0x0000C1D8L, 0x0000C2DEL, 0x0000C3E2L, 0x0000C4E4L,
0x0000C5E4L, 0x0000C6E2L, 0x0000C7DEL, 0x0000C8D9L,
0x0000C9D1L, 0x0000CAC7L, 0x0000CBBCL, 0x0000CCAEL,
0x0000CD9FL, 0x0000CE8EL, 0x0000CF7AL, 0x0000D065L,
0x0000D14DL, 0x0000D234L, 0x0000D318L, 0x0000D3FBL,
0x0000D4DBL, 0x0000D5BAL, 0x0000D696L, 0x0000D770L,
0x0000D848L, 0x0000D91EL, 0x0000D9F2L, 0x0000DAC4L,
0x0000DB94L, 0x0000DC62L, 0x0000DD2DL, 0x0000DDF7L,
0x0000DEBEL, 0x0000DF83L, 0x0000E046L, 0x0000E107L,
0x0000E1C6L, 0x0000E282L, 0x0000E33CL, 0x0000E3F4L,
0x0000E4AAL, 0x0000E55EL, 0x0000E610L, 0x0000E6BFL,
0x0000E76CL, 0x0000E817L, 0x0000E8BFL, 0x0000E966L,
0x0000EA0AL, 0x0000EAABL, 0x0000EB4BL, 0x0000EBE8L,
0x0000EC83L, 0x0000ED1CL, 0x0000EDB3L, 0x0000EE47L,
0x0000EED9L, 0x0000EF68L, 0x0000EFF5L, 0x0000F080L,
0x0000F109L, 0x0000F18FL, 0x0000F213L, 0x0000F295L,
0x0000F314L, 0x0000F391L, 0x0000F40CL, 0x0000F484L,
0x0000F4FAL, 0x0000F56EL, 0x0000F5DFL, 0x0000F64EL,
0x0000F6BAL, 0x0000F724L, 0x0000F78CL, 0x0000F7F1L,
0x0000F854L, 0x0000F8B4L, 0x0000F913L, 0x0000F96EL,
0x0000F9C8L, 0x0000FA1FL, 0x0000FA73L, 0x0000FAC5L,
0x0000FB15L, 0x0000FB62L, 0x0000FBADL, 0x0000FBF5L,
0x0000FC3BL, 0x0000FC7FL, 0x0000FCC0L, 0x0000FCFEL,
0x0000FD3BL, 0x0000FD74L, 0x0000FDACL, 0x0000FDE1L,
0x0000FE13L, 0x0000FE43L, 0x0000FE71L, 0x0000FE9CL,
0x0000FEC4L, 0x0000FEEBL, 0x0000FF0EL, 0x0000FF30L,
0x0000FF4EL, 0x0000FF6BL, 0x0000FF85L, 0x0000FF9CL,
0x0000FFB1L, 0x0000FFC4L, 0x0000FFD4L, 0x0000FFE1L,
0x0000FFECL, 0x0000FFF5L, 0x0000FFFBL, 0x0000FFFFL,
0x00010000L,
};
/* the difference of the angle for two adjacent values in the table
* expressed as ClutterFixed number
*/
#define CFX_SIN_STEP 0x00000192
/**
* clutter_sinx:
* @angle: a #ClutterFixed angle in radians
*
* Fixed point implementation of sine function
*
* Return value: #ClutterFixed sine value.
*
* Since: 0.2
*/
ClutterFixed
clutter_sinx (ClutterFixed angle)
{
int sign = 1, indx1, indx2;
ClutterFixed low, high, d1, d2;
/* convert negative angle to positive + sign */
if ((int)angle < 0)
{
sign = 1 + ~sign;
angle = 1 + ~angle;
}
/* reduce to <0, 2*pi) */
if (angle >= CFX_2PI)
{
ClutterFixed f = CLUTTER_FIXED_DIV (angle, CFX_2PI);
angle = angle - f;
}
/* reduce to first quadrant and sign */
if (angle > CFX_PI)
{
sign = 1 + ~sign;
if (angle > CFX_PI + CFX_PI_2)
{
/* fourth qudrant */
angle = CFX_2PI - angle;
}
else
{
/* third quadrant */
angle -= CFX_PI;
}
}
else
{
if (angle > CFX_PI_2)
{
/* second quadrant */
angle = CFX_PI - angle;
}
}
/* Calculate indices of the two nearest values in our table
* and return weighted average
*
* Handle the end of the table gracefully
*/
indx1 = CLUTTER_FIXED_DIV (angle, CFX_SIN_STEP);
indx1 = CLUTTER_FIXED_INT (indx1);
if (indx1 == sizeof (sin_tbl)/sizeof (ClutterFixed) - 1)
{
indx2 = indx1;
indx1 = indx2 - 1;
}
else
{
indx2 = indx1 + 1;
}
low = sin_tbl[indx1];
high = sin_tbl[indx2];
d1 = angle - indx1 * CFX_SIN_STEP;
d2 = indx2 * CFX_SIN_STEP - angle;
angle = ((low * d2 + high * d1) / (CFX_SIN_STEP));
if (sign < 0)
angle = (1 + ~angle);
return angle;
}
/**
* clutter_sini:
* @angle: a #ClutterAngle
*
* Very fast fixed point implementation of sine function.
*
* ClutterAngle is an integer such that 1024 represents
* full circle.
*
* Return value: #ClutterFixed sine value.
*
* Since: 0.2
*/
ClutterFixed
clutter_sini (ClutterAngle angle)
{
int sign = 1;
ClutterFixed result;
/* reduce negative angle to positive + sign */
if (angle < 0)
{
sign = 1 + ~sign;
angle = 1 + ~angle;
}
/* reduce to <0, 2*pi) */
angle &= 0x3ff;
/* reduce to first quadrant and sign */
if (angle > 512)
{
sign = 1 + ~sign;
if (angle > 768)
{
/* fourth qudrant */
angle = 1024 - angle;
}
else
{
/* third quadrant */
angle -= 512;
}
}
else
{
if (angle > 256)
{
/* second quadrant */
angle = 512 - angle;
}
}
result = sin_tbl[angle];
if (sign < 0)
result = (1 + ~result);
return result;
}
/* pre-computed tan table for 1st quadrant
*
* Currently contains 257 entries.
*
*/
static ClutterFixed tan_tbl [] =
{
0x00000000L, 0x00000192L, 0x00000324L, 0x000004b7L,
0x00000649L, 0x000007dbL, 0x0000096eL, 0x00000b01L,
0x00000c94L, 0x00000e27L, 0x00000fbaL, 0x0000114eL,
0x000012e2L, 0x00001477L, 0x0000160cL, 0x000017a1L,
0x00001937L, 0x00001acdL, 0x00001c64L, 0x00001dfbL,
0x00001f93L, 0x0000212cL, 0x000022c5L, 0x0000245fL,
0x000025f9L, 0x00002795L, 0x00002931L, 0x00002aceL,
0x00002c6cL, 0x00002e0aL, 0x00002faaL, 0x0000314aL,
0x000032ecL, 0x0000348eL, 0x00003632L, 0x000037d7L,
0x0000397dL, 0x00003b24L, 0x00003cccL, 0x00003e75L,
0x00004020L, 0x000041ccL, 0x00004379L, 0x00004528L,
0x000046d8L, 0x0000488aL, 0x00004a3dL, 0x00004bf2L,
0x00004da8L, 0x00004f60L, 0x0000511aL, 0x000052d5L,
0x00005492L, 0x00005651L, 0x00005812L, 0x000059d5L,
0x00005b99L, 0x00005d60L, 0x00005f28L, 0x000060f3L,
0x000062c0L, 0x0000648fL, 0x00006660L, 0x00006834L,
0x00006a0aL, 0x00006be2L, 0x00006dbdL, 0x00006f9aL,
0x0000717aL, 0x0000735dL, 0x00007542L, 0x0000772aL,
0x00007914L, 0x00007b02L, 0x00007cf2L, 0x00007ee6L,
0x000080dcL, 0x000082d6L, 0x000084d2L, 0x000086d2L,
0x000088d6L, 0x00008adcL, 0x00008ce7L, 0x00008ef4L,
0x00009106L, 0x0000931bL, 0x00009534L, 0x00009750L,
0x00009971L, 0x00009b95L, 0x00009dbeL, 0x00009febL,
0x0000a21cL, 0x0000a452L, 0x0000a68cL, 0x0000a8caL,
0x0000ab0eL, 0x0000ad56L, 0x0000afa3L, 0x0000b1f5L,
0x0000b44cL, 0x0000b6a8L, 0x0000b909L, 0x0000bb70L,
0x0000bdddL, 0x0000c04fL, 0x0000c2c7L, 0x0000c545L,
0x0000c7c9L, 0x0000ca53L, 0x0000cce3L, 0x0000cf7aL,
0x0000d218L, 0x0000d4bcL, 0x0000d768L, 0x0000da1aL,
0x0000dcd4L, 0x0000df95L, 0x0000e25eL, 0x0000e52eL,
0x0000e806L, 0x0000eae7L, 0x0000edd0L, 0x0000f0c1L,
0x0000f3bbL, 0x0000f6bfL, 0x0000f9cbL, 0x0000fce1L,
0x00010000L, 0x00010329L, 0x0001065dL, 0x0001099aL,
0x00010ce3L, 0x00011036L, 0x00011394L, 0x000116feL,
0x00011a74L, 0x00011df6L, 0x00012184L, 0x0001251fL,
0x000128c6L, 0x00012c7cL, 0x0001303fL, 0x00013410L,
0x000137f0L, 0x00013bdfL, 0x00013fddL, 0x000143ebL,
0x00014809L, 0x00014c37L, 0x00015077L, 0x000154c9L,
0x0001592dL, 0x00015da4L, 0x0001622eL, 0x000166ccL,
0x00016b7eL, 0x00017045L, 0x00017523L, 0x00017a17L,
0x00017f22L, 0x00018444L, 0x00018980L, 0x00018ed5L,
0x00019445L, 0x000199cfL, 0x00019f76L, 0x0001a53aL,
0x0001ab1cL, 0x0001b11dL, 0x0001b73fL, 0x0001bd82L,
0x0001c3e7L, 0x0001ca71L, 0x0001d11fL, 0x0001d7f4L,
0x0001def1L, 0x0001e618L, 0x0001ed6aL, 0x0001f4e8L,
0x0001fc96L, 0x00020473L, 0x00020c84L, 0x000214c9L,
0x00021d44L, 0x000225f9L, 0x00022ee9L, 0x00023818L,
0x00024187L, 0x00024b3aL, 0x00025534L, 0x00025f78L,
0x00026a0aL, 0x000274edL, 0x00028026L, 0x00028bb8L,
0x000297a8L, 0x0002a3fbL, 0x0002b0b5L, 0x0002bdddL,
0x0002cb79L, 0x0002d98eL, 0x0002e823L, 0x0002f740L,
0x000306ecL, 0x00031730L, 0x00032816L, 0x000339a6L,
0x00034bebL, 0x00035ef2L, 0x000372c6L, 0x00038776L,
0x00039d11L, 0x0003b3a6L, 0x0003cb48L, 0x0003e40aL,
0x0003fe02L, 0x00041949L, 0x000435f7L, 0x0004542bL,
0x00047405L, 0x000495a9L, 0x0004b940L, 0x0004def6L,
0x00050700L, 0x00053196L, 0x00055ef9L, 0x00058f75L,
0x0005c35dL, 0x0005fb14L, 0x00063709L, 0x000677c0L,
0x0006bdd0L, 0x000709ecL, 0x00075ce6L, 0x0007b7bbL,
0x00081b98L, 0x000889e9L, 0x0009046eL, 0x00098d4dL,
0x000a2736L, 0x000ad593L, 0x000b9cc6L, 0x000c828aL,
0x000d8e82L, 0x000ecb1bL, 0x001046eaL, 0x00121703L,
0x00145b00L, 0x0017448dL, 0x001b2672L, 0x002095afL,
0x0028bc49L, 0x0036519aL, 0x00517bb6L, 0x00a2f8fdL,
0x46d3eab2L,
};
/**
* clutter_tani:
* @angle: a #ClutterAngle
*
* Very fast fixed point implementation of tan function.
*
* ClutterAngle is an integer such that 1024 represents
* full circle.
*
* Return value: #ClutterFixed sine value.
*
* Since: 0.3
*/
ClutterFixed
clutter_tani (ClutterAngle angle)
{
int sign = 1;
ClutterFixed result;
/* reduce negative angle to positive + sign */
if (angle < 0)
{
sign = 1 + ~sign;
angle = 1 + ~angle;
}
/* reduce to <0, pi) */
angle &= 0x1ff;
/* reduce to first quadrant and sign */
if (angle > 256)
{
sign = 1 + ~sign;
angle = 512 - angle;
}
result = tan_tbl[angle];
if (sign < 0)
result = (1 + ~result);
return result;
}
ClutterFixed sqrt_tbl [] =
{
0x00000000L, 0x00010000L, 0x00016A0AL, 0x0001BB68L,
0x00020000L, 0x00023C6FL, 0x00027312L, 0x0002A550L,
0x0002D414L, 0x00030000L, 0x0003298BL, 0x0003510EL,
0x000376CFL, 0x00039B05L, 0x0003BDDDL, 0x0003DF7CL,
0x00040000L, 0x00041F84L, 0x00043E1EL, 0x00045BE1L,
0x000478DEL, 0x00049524L, 0x0004B0BFL, 0x0004CBBCL,
0x0004E624L, 0x00050000L, 0x00051959L, 0x00053237L,
0x00054AA0L, 0x0005629AL, 0x00057A2BL, 0x00059159L,
0x0005A828L, 0x0005BE9CL, 0x0005D4B9L, 0x0005EA84L,
0x00060000L, 0x00061530L, 0x00062A17L, 0x00063EB8L,
0x00065316L, 0x00066733L, 0x00067B12L, 0x00068EB4L,
0x0006A21DL, 0x0006B54DL, 0x0006C847L, 0x0006DB0CL,
0x0006ED9FL, 0x00070000L, 0x00071232L, 0x00072435L,
0x0007360BL, 0x000747B5L, 0x00075935L, 0x00076A8CL,
0x00077BBBL, 0x00078CC2L, 0x00079DA3L, 0x0007AE60L,
0x0007BEF8L, 0x0007CF6DL, 0x0007DFBFL, 0x0007EFF0L,
0x00080000L, 0x00080FF0L, 0x00081FC1L, 0x00082F73L,
0x00083F08L, 0x00084E7FL, 0x00085DDAL, 0x00086D18L,
0x00087C3BL, 0x00088B44L, 0x00089A32L, 0x0008A906L,
0x0008B7C2L, 0x0008C664L, 0x0008D4EEL, 0x0008E361L,
0x0008F1BCL, 0x00090000L, 0x00090E2EL, 0x00091C45L,
0x00092A47L, 0x00093834L, 0x0009460CL, 0x000953CFL,
0x0009617EL, 0x00096F19L, 0x00097CA1L, 0x00098A16L,
0x00099777L, 0x0009A4C6L, 0x0009B203L, 0x0009BF2EL,
0x0009CC47L, 0x0009D94FL, 0x0009E645L, 0x0009F32BL,
0x000A0000L, 0x000A0CC5L, 0x000A1979L, 0x000A261EL,
0x000A32B3L, 0x000A3F38L, 0x000A4BAEL, 0x000A5816L,
0x000A646EL, 0x000A70B8L, 0x000A7CF3L, 0x000A8921L,
0x000A9540L, 0x000AA151L, 0x000AAD55L, 0x000AB94BL,
0x000AC534L, 0x000AD110L, 0x000ADCDFL, 0x000AE8A1L,
0x000AF457L, 0x000B0000L, 0x000B0B9DL, 0x000B172DL,
0x000B22B2L, 0x000B2E2BL, 0x000B3998L, 0x000B44F9L,
0x000B504FL, 0x000B5B9AL, 0x000B66D9L, 0x000B720EL,
0x000B7D37L, 0x000B8856L, 0x000B936AL, 0x000B9E74L,
0x000BA973L, 0x000BB467L, 0x000BBF52L, 0x000BCA32L,
0x000BD508L, 0x000BDFD5L, 0x000BEA98L, 0x000BF551L,
0x000C0000L, 0x000C0AA6L, 0x000C1543L, 0x000C1FD6L,
0x000C2A60L, 0x000C34E1L, 0x000C3F59L, 0x000C49C8L,
0x000C542EL, 0x000C5E8CL, 0x000C68E0L, 0x000C732DL,
0x000C7D70L, 0x000C87ACL, 0x000C91DFL, 0x000C9C0AL,
0x000CA62CL, 0x000CB047L, 0x000CBA59L, 0x000CC464L,
0x000CCE66L, 0x000CD861L, 0x000CE254L, 0x000CEC40L,
0x000CF624L, 0x000D0000L, 0x000D09D5L, 0x000D13A2L,
0x000D1D69L, 0x000D2727L, 0x000D30DFL, 0x000D3A90L,
0x000D4439L, 0x000D4DDCL, 0x000D5777L, 0x000D610CL,
0x000D6A9AL, 0x000D7421L, 0x000D7DA1L, 0x000D871BL,
0x000D908EL, 0x000D99FAL, 0x000DA360L, 0x000DACBFL,
0x000DB618L, 0x000DBF6BL, 0x000DC8B7L, 0x000DD1FEL,
0x000DDB3DL, 0x000DE477L, 0x000DEDABL, 0x000DF6D8L,
0x000E0000L, 0x000E0922L, 0x000E123DL, 0x000E1B53L,
0x000E2463L, 0x000E2D6DL, 0x000E3672L, 0x000E3F70L,
0x000E4869L, 0x000E515DL, 0x000E5A4BL, 0x000E6333L,
0x000E6C16L, 0x000E74F3L, 0x000E7DCBL, 0x000E869DL,
0x000E8F6BL, 0x000E9832L, 0x000EA0F5L, 0x000EA9B2L,
0x000EB26BL, 0x000EBB1EL, 0x000EC3CBL, 0x000ECC74L,
0x000ED518L, 0x000EDDB7L, 0x000EE650L, 0x000EEEE5L,
0x000EF775L, 0x000F0000L, 0x000F0886L, 0x000F1107L,
0x000F1984L, 0x000F21FCL, 0x000F2A6FL, 0x000F32DDL,
0x000F3B47L, 0x000F43ACL, 0x000F4C0CL, 0x000F5468L,
0x000F5CBFL, 0x000F6512L, 0x000F6D60L, 0x000F75AAL,
0x000F7DEFL, 0x000F8630L, 0x000F8E6DL, 0x000F96A5L,
0x000F9ED9L, 0x000FA709L, 0x000FAF34L, 0x000FB75BL,
0x000FBF7EL, 0x000FC79DL, 0x000FCFB7L, 0x000FD7CEL,
0x000FDFE0L, 0x000FE7EEL, 0x000FEFF8L, 0x000FF7FEL,
0x00100000L,
};
/**
* clutter_sqrtx:
* @x: a #ClutterFixed
*
* A fixed point implementation of squre root
*
* Return value: #ClutterFixed square root.
*
* Since: 0.2
*/
ClutterFixed
clutter_sqrtx (ClutterFixed x)
{
/* The idea for this comes from the Alegro library, exploiting the
* fact that,
* sqrt (x) = sqrt (x/d) * sqrt (d);
*
* For d == 2^(n):
*
* sqrt (x) = sqrt (x/2^(2n)) * 2^n
*
* By locating suitable n for given x such that x >> 2n is in <0,255>
* we can use a LUT of precomputed values.
*
* This algorithm provides both good performance and precission;
* on ARM this function is about 5 times faster than c-lib sqrt, whilst
* producing errors < 1%.
*
*/
int t = 0;
int sh = 0;
unsigned int mask = 0x40000000;
unsigned fract = x & 0x0000ffff;
unsigned int d1, d2;
if (x <= 0)
return 0;
if (x > CFX_255 || x < CFX_ONE)
{
/*
* Find the highest bit set
*/
#if __arm__
/* This actually requires at least arm v5, but gcc does not seem
* to set the architecture defines correctly, and it is I think
* very unlikely that anyone will want to use clutter on anything
* less than v5.
*/
int bit;
__asm__ ("clz %0, %1\n"
"rsb %0, %0, #31\n"
:"=r"(bit)
:"r" (x));
/* make even (2n) */
bit &= 0xfffffffe;
#else
/* TODO -- add i386 branch using bshr */
int bit = 30;
while (bit >= 0)
{
if (x & mask)
break;
mask = (mask >> 1 | mask >> 2);
bit -= 2;
}
#endif
/* now bit indicates the highest bit set; there are two scenarios
*
* 1) bit < 23: Our number is smaller so we shift it left to maximase
* precision (< 16 really, since <16,23> never goes
* through here.
*
* 2) bit > 23: our number is above the table, so we shift right
*/
sh = ((bit - 22) >> 1);
if (bit >= 8)
t = (x >> (16 - 22 + bit));
else
t = (x << (22 - 16 - bit));
}
else
{
t = CLUTTER_FIXED_INT (x);
}
/* Do a weighted average of the two nearest values */
ClutterFixed v1 = sqrt_tbl[t];
ClutterFixed v2 = sqrt_tbl[t+1];
/*
* 12 is fairly arbitrary -- we want integer that is not too big to cost
* us precission
*/
d1 = (unsigned)(fract) >> 12;
d2 = ((unsigned)CFX_ONE >> 12) - d1;
x = ((v1*d2) + (v2*d1))/(CFX_ONE >> 12);
if (sh > 0)
x = x << sh;
else if (sh < 0)
x = (x >> (1 + ~sh));
return x;
}
/**
* clutter_sqrti:
* @x: integer value
*
* Very fast fixed point implementation of square root for integers.
*
* This function is about 10x faster than clib sqrt() on x86, and (this is
* not a typo!) more than 800x faster on ARM without FPU. It's error is < 5%
* for arguments < 132 and < 10% for arguments < 5591.
*
* Return value: integer square root.
*
*
* Since: 0.2
*/
gint
clutter_sqrti (gint number)
{
/* This is a fixed point implementation of the Quake III sqrt algorithm,
* described, for example, at
* http://www.codemaestro.com/reviews/review00000105.html
*
* While the original QIII is extremely fast, the use of floating division
* and multiplication makes it perform very on arm processors without FPU.
*
* The key to successfully replacing the floating point operations with
* fixed point is in the choice of the fixed point format. The QIII
* algorithm does not calculate the square root, but its reciprocal ('y'
* below), which is only at the end turned to the inverse value. In order
* for the algorithm to produce satisfactory results, the reciprocal value
* must be represented with sufficient precission; the 16.16 we use
* elsewhere in clutter is not good enough, and 10.22 is used instead.
*/
ClutterFixed x;
unsigned long y1; /* 10.22 fixed point */
unsigned long f = 0x600000; /* '1.5' as 10.22 fixed */
union
{
float f;
unsigned long i;
} flt, flt2;
flt.f = number;
x = CLUTTER_INT_TO_FIXED (number) / 2;
/* The QIII initial estimate */
flt.i = 0x5f3759df - ( flt.i >> 1 );
/* Now, we convert the float to 10.22 fixed. We exploit the mechanism
* described at http://www.d6.com/users/checker/pdfs/gdmfp.pdf.
*
* We want 22 bit fraction; a single precission float uses 23 bit
* mantisa, so we only need to add 2^(23-22) (no need for the 1.5
* multiplier as we are only dealing with positive numbers).
*
* Note: we have to use two separate variables here -- for some reason,
* if we try to use just the flt variable, gcc on ARM optimises the whole
* addition out, and it all goes pear shape, since without it, the bits
* in the float will not be correctly aligned.
*/
flt2.f = flt.f + 2.0;
flt2.i &= 0x7FFFFF;
/* Now we correct the estimate, only single iterration is needed */
y1 = (flt2.i >> 11) * (flt2.i >> 11);
y1 = (y1 >> 8) * (x >> 8);
y1 = f - y1;
flt2.i = (flt2.i >> 11) * (y1 >> 11);
/* Invert, round and convert from 10.22 to an integer
* 0x1e3c68 is a magical rounding constant that produces slightly
* better results than 0x200000.
*/
return (number * flt2.i + 0x1e3c68) >> 22;
}
/**
* clutter_fixed_qmulx:
* @op1: #ClutterFixed
* @op2: #ClutterFixed
*
* Return value: #ClutterFixed.
*
* Multiplies two fixed values using 64bit arithmetic; this provides
* significantly better precission than the #CLUTTER_FIXED_MUL macro,
* but at performance cost (about 2.7 times slowdown on ARMv5e, and 2 times
* on x86).
*
* Since: 0.3
*/
ClutterFixed
clutter_qmulx (ClutterFixed op1, ClutterFixed op2)
{
#ifdef __arm__
/* This provides about 12% speedeup on the gcc -O2 optimised
* C version
*
* Based on code found in the following thread:
* http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2006-August/014405.html
*/
int res_low, res_hi;
__asm__ ("smull %0, %1, %2, %3 \n"
"mov %0, %0, lsr %4 \n"
"add %1, %0, %1, lsl %5 \n"
: "=r"(res_hi), "=r"(res_low)\
: "r"(op1), "r"(op2), "i"(CFX_Q), "i"(32-CFX_Q));
return (ClutterFixed) res_low;
#else
long long r = (long long) op1 * (long long) op2;
return (unsigned int)(r >> CFX_Q);
#endif
}
/*
* The log2x() and pow2x() functions
*
* The implementation of the log2x() and pow2x() exploits the well-documented
* fact that the exponent part of IEEE floating number provides a good estimate
* of log2 of that number, while the mantisa serves as a good error-correction.
*
* The implemenation here uses a quadratic error correction as described by
* Ian Stephenson at http://www.dctsystems.co.uk/Software/power.html.
*/
/**
* clutter_log2x :
* @x: value to calculate base 2 logarithm from
*
* Calculates base 2 logarithm.
*
* This function is some 2.5 times faster on x86, and over 12 times faster on
* fpu-less arm, than using libc log().
*
* Return value: base 2 logarithm.
*
* Since: 0.4
*/
ClutterFixed
clutter_log2x (guint x)
{
/* Note: we could easily have a version for ClutterFixed x, but the int
* precission is enough for the current purposes.
*/
union
{
float f;
ClutterFixed i;
} flt;
ClutterFixed magic = 0x58bb;
ClutterFixed y;
/*
* Convert x to float, then extract exponent.
*
* We want the result to be 16.16 fixed, so we shift (23-16) bits only
*/
flt.f = x;
flt.i >>= 7;
flt.i -= CLUTTER_INT_TO_FIXED (127);
y = CLUTTER_FIXED_FRACTION (flt.i);
y = CFX_MUL ((y - CFX_MUL (y, y)), magic);
return flt.i + y;
}
/**
* clutter_pow2x :
* @x: exponent
*
* Calculates 2 to x power.
*
* This function is around 11 times faster on x86, and around 22 times faster
* on fpu-less arm than libc pow(2, x).
*
* Return value: 2 in x power.
*
* Since: 0.4
*/
guint
clutter_pow2x (ClutterFixed x)
{
/* Note: we could easily have a version that produces ClutterFixed result,
* but the the range would be limited to x < 15, and the int precission
* is enough for the current purposes.
*/
union
{
float f;
guint32 i;
} flt;
ClutterFixed magic = 0x56f7;
ClutterFixed y;
flt.i = x;
/*
* Reverse of the log2x function -- convert the fixed value to a suitable
* floating point exponent, and mantisa adjusted with quadratic error
* correction y.
*/
y = CLUTTER_FIXED_FRACTION (x);
y = CFX_MUL ((y - CFX_MUL (y, y)), magic);
/* Shift the exponent into it's position in the floating point
* representation; as our number is not int but 16.16 fixed, shift only
* by (23 - 16)
*/
flt.i += (CLUTTER_INT_TO_FIXED (127) - y);
flt.i <<= 7;
return CLUTTER_FLOAT_TO_UINT (flt.f);
}
/**
* clutter_powx :
* @x: base
* @y: #ClutterFixed exponent
*
* Calculates x to y power. (Note, if x is a constant it will be faster to
* calculate the power as clutter_pow2x (CLUTTER_FIXED_MUL(y, log2 (x)))
*
* Return value: x in y power.
*
* Since: 0.4
*/
guint
clutter_powx (guint x, ClutterFixed y)
{
return clutter_pow2x (CFX_MUL (y, clutter_log2x (x)));
}
/* <private> */
const double _magic = 68719476736.0*1.5;
/* Where in the 64 bits of double is the mantisa */
#if (__FLOAT_WORD_ORDER == 1234)
#define _CFX_MAN 0
#elif (__FLOAT_WORD_ORDER == 4321)
#define _CFX_MAN 1
#else
#define CFX_NO_FAST_CONVERSIONS
#endif
/*
* clutter_double_to_fixed :
* @value: value to be converted
*
* A fast conversion from double precision floating to fixed point
*
* Return value: Fixed point representation of the value
*
* Since: 0.2
*/
ClutterFixed
_clutter_double_to_fixed (double val)
{
#ifdef CFX_NO_FAST_CONVERSIONS
return (ClutterFixed)(val * (double)CFX_ONE);
#else
union
{
double d;
unsigned int i[2];
} dbl;
dbl.d = val;
dbl.d = dbl.d + _magic;
return dbl.i[_CFX_MAN];
#endif
}
/*
* clutter_double_to_int :
* @value: value to be converted
*
* A fast conversion from doulbe precision floatint point to int;
* used this instead of casting double/float to int.
*
* Return value: Integer part of the double
*
* Since: 0.2
*/
gint
_clutter_double_to_int (double val)
{
#ifdef CFX_NO_FAST_CONVERSIONS
return (gint)(val);
#else
union
{
double d;
unsigned int i[2];
} dbl;
dbl.d = val;
dbl.d = dbl.d + _magic;
return ((int)dbl.i[_CFX_MAN]) >> 16;
#endif
}
guint
_clutter_double_to_uint (double val)
{
#ifdef CFX_NO_FAST_CONVERSIONS
return (guint)(val);
#else
union
{
double d;
unsigned int i[2];
} dbl;
dbl.d = val;
dbl.d = dbl.d + _magic;
return (dbl.i[_CFX_MAN]) >> 16;
#endif
}
#undef _CFX_MAN
Reference in New Issue View Git Blame Copy Permalink