diff options
| author | Dave Parks <davep@lindenlab.com> | 2010-08-19 12:25:15 -0500 |
|---|---|---|
| committer | Dave Parks <davep@lindenlab.com> | 2010-08-19 12:25:15 -0500 |
| commit | 2fea1d5d33ec1b41a3cfa4307a1bfa58d8014f88 (patch) | |
| tree | 0438f2363b2a91a5ffe970a8130faa118f260e7e | |
| parent | bd0b3a2ddeafaf0d1669ede7ab5aee22d8da9af7 (diff) | |
Integrate SIMD API from oreh/server-trunk-oreh
57 files changed, 2731 insertions, 2602 deletions
diff --git a/indra/cmake/00-Common.cmake b/indra/cmake/00-Common.cmake index 2c7bb77758..f10a61e1e7 100644 --- a/indra/cmake/00-Common.cmake +++ b/indra/cmake/00-Common.cmake @@ -68,6 +68,7 @@ if (WINDOWS) add_definitions( /Zc:wchar_t- + /arch:SSE2
) endif (MSVC80 OR MSVC90) diff --git a/indra/llcharacter/llkeyframemotion.cpp b/indra/llcharacter/llkeyframemotion.cpp index 7bc9e97eb1..bce272082e 100644 --- a/indra/llcharacter/llkeyframemotion.cpp +++ b/indra/llcharacter/llkeyframemotion.cpp @@ -1151,7 +1151,7 @@ void LLKeyframeMotion::applyConstraint(JointConstraint* constraint, F32 time, U8 constraint->mPositions[joint_num] = new_pos; } constraint->mFixupDistanceRMS *= 1.f / (constraint->mTotalLength * (F32)(shared_data->mChainLength - 1)); - constraint->mFixupDistanceRMS = fsqrtf(constraint->mFixupDistanceRMS); + constraint->mFixupDistanceRMS = (F32) sqrt(constraint->mFixupDistanceRMS); //reset old joint rots for (joint_num = 0; joint_num <= shared_data->mChainLength; joint_num++) diff --git a/indra/llmath/CMakeLists.txt b/indra/llmath/CMakeLists.txt index dda07133d5..8d85765eb8 100644 --- a/indra/llmath/CMakeLists.txt +++ b/indra/llmath/CMakeLists.txt @@ -1,118 +1,128 @@ -# -*- cmake -*- - -project(llmath) - -include(00-Common) -include(LLCommon) - -include_directories( - ${LLCOMMON_INCLUDE_DIRS} - ) - -set(llmath_SOURCE_FILES - llbbox.cpp - llbboxlocal.cpp - llcamera.cpp - llcoordframe.cpp - llline.cpp - llmodularmath.cpp - llperlin.cpp - llquaternion.cpp - llrect.cpp - llsphere.cpp - llvolume.cpp - llvolumemgr.cpp - llvolumeoctree.cpp - llsdutil_math.cpp - m3math.cpp - m4math.cpp - raytrace.cpp - v2math.cpp - v3color.cpp - v3dmath.cpp - v3math.cpp - v4color.cpp - v4coloru.cpp - v4math.cpp - xform.cpp - ) - -set(llmath_HEADER_FILES - CMakeLists.txt - - camera.h - coordframe.h - llbbox.h - llbboxlocal.h - llcamera.h - llcoord.h - llcoordframe.h - llinterp.h - llline.h - llmath.h - llmodularmath.h - lloctree.h - llperlin.h - llplane.h - llquantize.h - llquaternion.h - llrect.h - llsphere.h - lltreenode.h - llv4math.h - llv4matrix3.h - llv4matrix4.h - llv4vector3.h - llvector4a.h - llmatrix4a.h - llvolume.h - llvolumemgr.h - llvolumeoctree.h - llsdutil_math.h - m3math.h - m4math.h - raytrace.h - v2math.h - v3color.h - v3dmath.h - v3math.h - v4color.h - v4coloru.h - v4math.h - xform.h - ) - -set_source_files_properties(${llmath_HEADER_FILES} - PROPERTIES HEADER_FILE_ONLY TRUE) - -list(APPEND llmath_SOURCE_FILES ${llmath_HEADER_FILES}) - -add_library (llmath ${llmath_SOURCE_FILES}) - -# Add tests -if (LL_TESTS) - include(LLAddBuildTest) - # UNIT TESTS - SET(llmath_TEST_SOURCE_FILES - llbboxlocal.cpp - llmodularmath.cpp - llrect.cpp - v2math.cpp - v3color.cpp - v4color.cpp - v4coloru.cpp - ) - LL_ADD_PROJECT_UNIT_TESTS(llmath "${llmath_TEST_SOURCE_FILES}") - - # INTEGRATION TESTS - set(test_libs llmath llcommon ${LLCOMMON_LIBRARIES} ${WINDOWS_LIBRARIES}) - # TODO: Some of these need refactoring to be proper Unit tests rather than Integration tests. - LL_ADD_INTEGRATION_TEST(llbbox llbbox.cpp "${test_libs}") - LL_ADD_INTEGRATION_TEST(llquaternion llquaternion.cpp "${test_libs}") - LL_ADD_INTEGRATION_TEST(mathmisc "" "${test_libs}") - LL_ADD_INTEGRATION_TEST(m3math "" "${test_libs}") - LL_ADD_INTEGRATION_TEST(v3dmath v3dmath.cpp "${test_libs}") - LL_ADD_INTEGRATION_TEST(v3math v3math.cpp "${test_libs}") - LL_ADD_INTEGRATION_TEST(v4math v4math.cpp "${test_libs}") - LL_ADD_INTEGRATION_TEST(xform xform.cpp "${test_libs}") -endif (LL_TESTS) +# -*- cmake -*-
+
+project(llmath)
+
+include(00-Common)
+include(LLCommon)
+
+include_directories(
+ ${LLCOMMON_INCLUDE_DIRS}
+ )
+
+set(llmath_SOURCE_FILES
+ llbbox.cpp
+ llbboxlocal.cpp
+ llcamera.cpp
+ llcoordframe.cpp
+ llline.cpp
+ llmatrix3a.cpp
+ llmodularmath.cpp
+ llperlin.cpp
+ llquaternion.cpp
+ llrect.cpp
+ llsphere.cpp
+ llvector4a.cpp
+ llvolume.cpp
+ llvolumemgr.cpp
+ llvolumeoctree.cpp
+ llsdutil_math.cpp
+ m3math.cpp
+ m4math.cpp
+ raytrace.cpp
+ v2math.cpp
+ v3color.cpp
+ v3dmath.cpp
+ v3math.cpp
+ v4color.cpp
+ v4coloru.cpp
+ v4math.cpp
+ xform.cpp
+ )
+
+set(llmath_HEADER_FILES
+ CMakeLists.txt
+
+ camera.h
+ coordframe.h
+ llbbox.h
+ llbboxlocal.h
+ llcamera.h
+ llcoord.h
+ llcoordframe.h
+ llinterp.h
+ llline.h
+ llmath.h
+ llmatrix3a.h
+ llmatrix3a.inl
+ llmodularmath.h
+ lloctree.h
+ llperlin.h
+ llplane.h
+ llquantize.h
+ llquaternion.h
+ llquaternion2.h
+ llquaternion2.inl
+ llrect.h
+ llsimdmath.h
+ llsimdtypes.h
+ llsimdtypes.inl
+ llsphere.h
+ lltreenode.h
+ llvector4a.h
+ llvector4a.inl
+ llvector4logical.h
+ llv4math.h
+ llv4matrix3.h
+ llv4matrix4.h
+ llv4vector3.h
+ llvolume.h
+ llvolumemgr.h
+ llvolumeoctree.h
+ llsdutil_math.h
+ m3math.h
+ m4math.h
+ raytrace.h
+ v2math.h
+ v3color.h
+ v3dmath.h
+ v3math.h
+ v4color.h
+ v4coloru.h
+ v4math.h
+ xform.h
+ )
+
+set_source_files_properties(${llmath_HEADER_FILES}
+ PROPERTIES HEADER_FILE_ONLY TRUE)
+
+list(APPEND llmath_SOURCE_FILES ${llmath_HEADER_FILES})
+
+add_library (llmath ${llmath_SOURCE_FILES})
+
+# Add tests
+if (LL_TESTS)
+ include(LLAddBuildTest)
+ # UNIT TESTS
+ SET(llmath_TEST_SOURCE_FILES
+ llbboxlocal.cpp
+ llmodularmath.cpp
+ llrect.cpp
+ v2math.cpp
+ v3color.cpp
+ v4color.cpp
+ v4coloru.cpp
+ )
+ LL_ADD_PROJECT_UNIT_TESTS(llmath "${llmath_TEST_SOURCE_FILES}")
+
+ # INTEGRATION TESTS
+ set(test_libs llmath llcommon ${LLCOMMON_LIBRARIES} ${WINDOWS_LIBRARIES})
+ # TODO: Some of these need refactoring to be proper Unit tests rather than Integration tests.
+ LL_ADD_INTEGRATION_TEST(llbbox llbbox.cpp "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(llquaternion llquaternion.cpp "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(mathmisc "" "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(m3math "" "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(v3dmath v3dmath.cpp "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(v3math v3math.cpp "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(v4math v4math.cpp "${test_libs}")
+ LL_ADD_INTEGRATION_TEST(xform xform.cpp "${test_libs}")
+endif (LL_TESTS)
diff --git a/indra/llmath/llcamera.cpp b/indra/llmath/llcamera.cpp index 6b56e4870e..beb5c48624 100644 --- a/indra/llmath/llcamera.cpp +++ b/indra/llmath/llcamera.cpp @@ -77,7 +77,7 @@ const LLCamera& LLCamera::operator=(const LLCamera& rhs) { memcpy(this, &rhs, sizeof(LLCamera)); alignPlanes(); - LLVector4a::memcpyNonAliased16((F32*) mAgentPlanes, (F32*) rhs.mAgentPlanes, 4*7); + LLVector4a::memcpyNonAliased16((F32*) mAgentPlanes, (F32*) rhs.mAgentPlanes, 4*7*sizeof(F32)); return *this; } diff --git a/indra/llmath/llmath.h b/indra/llmath/llmath.h index c3c15e1374..742bbc4751 100644 --- a/indra/llmath/llmath.h +++ b/indra/llmath/llmath.h @@ -1,525 +1,509 @@ -/** - * @file llmath.h - * @brief Useful math constants and macros. - * - * $LicenseInfo:firstyear=2000&license=viewergpl$ - * - * Copyright (c) 2000-2009, Linden Research, Inc. - * - * Second Life Viewer Source Code - * The source code in this file ("Source Code") is provided by Linden Lab - * to you under the terms of the GNU General Public License, version 2.0 - * ("GPL"), unless you have obtained a separate licensing agreement - * ("Other License"), formally executed by you and Linden Lab. Terms of - * the GPL can be found in doc/GPL-license.txt in this distribution, or - * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2 - * - * There are special exceptions to the terms and conditions of the GPL as - * it is applied to this Source Code. View the full text of the exception - * in the file doc/FLOSS-exception.txt in this software distribution, or - * online at - * http://secondlifegrid.net/programs/open_source/licensing/flossexception - * - * By copying, modifying or distributing this software, you acknowledge - * that you have read and understood your obligations described above, - * and agree to abide by those obligations. - * - * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO - * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY, - * COMPLETENESS OR PERFORMANCE. - * $/LicenseInfo$ - */ - -#ifndef LLMATH_H -#define LLMATH_H - -#include <cmath> -#include <cstdlib> -#include <complex> -#include "lldefs.h" -//#include "llstl.h" // *TODO: Remove when LLString is gone -//#include "llstring.h" // *TODO: Remove when LLString is gone -// lltut.h uses is_approx_equal_fraction(). This was moved to its own header -// file in llcommon so we can use lltut.h for llcommon tests without making -// llcommon depend on llmath. -#include "is_approx_equal_fraction.h" - -// work around for Windows & older gcc non-standard function names. -#if LL_WINDOWS -#include <float.h> -#define llisnan(val) _isnan(val) -#define llfinite(val) _finite(val) -#elif (LL_LINUX && __GNUC__ <= 2) -#define llisnan(val) isnan(val) -#define llfinite(val) isfinite(val) -#elif LL_SOLARIS -#define llisnan(val) isnan(val) -#define llfinite(val) (val <= std::numeric_limits<double>::max()) -#else -#define llisnan(val) std::isnan(val) -#define llfinite(val) std::isfinite(val) -#endif - -// Single Precision Floating Point Routines -#ifndef sqrtf -#define sqrtf(x) ((F32)sqrt((F64)(x))) -#endif -#ifndef fsqrtf -#define fsqrtf(x) sqrtf(x) -#endif - -#ifndef cosf -#define cosf(x) ((F32)cos((F64)(x))) -#endif -#ifndef sinf -#define sinf(x) ((F32)sin((F64)(x))) -#endif -#ifndef tanf -#define tanf(x) ((F32)tan((F64)(x))) -#endif -#ifndef acosf -#define acosf(x) ((F32)acos((F64)(x))) -#endif - -#ifndef powf -#define powf(x,y) ((F32)pow((F64)(x),(F64)(y))) -#endif -#ifndef expf -#define expf(x) ((F32)exp((F64)(x))) -#endif - -const F32 GRAVITY = -9.8f; - -// mathematical constants -const F32 F_PI = 3.1415926535897932384626433832795f; -const F32 F_TWO_PI = 6.283185307179586476925286766559f; -const F32 F_PI_BY_TWO = 1.5707963267948966192313216916398f; -const F32 F_SQRT_TWO_PI = 2.506628274631000502415765284811f; -const F32 F_E = 2.71828182845904523536f; -const F32 F_SQRT2 = 1.4142135623730950488016887242097f; -const F32 F_SQRT3 = 1.73205080756888288657986402541f; -const F32 OO_SQRT2 = 0.7071067811865475244008443621049f; -const F32 DEG_TO_RAD = 0.017453292519943295769236907684886f; -const F32 RAD_TO_DEG = 57.295779513082320876798154814105f; -const F32 F_APPROXIMATELY_ZERO = 0.00001f; -const F32 F_LN2 = 0.69314718056f; -const F32 OO_LN2 = 1.4426950408889634073599246810019f; - -const F32 F_ALMOST_ZERO = 0.0001f; -const F32 F_ALMOST_ONE = 1.0f - F_ALMOST_ZERO; - -// BUG: Eliminate in favor of F_APPROXIMATELY_ZERO above? -const F32 FP_MAG_THRESHOLD = 0.0000001f; - -// TODO: Replace with logic like is_approx_equal -inline BOOL is_approx_zero( F32 f ) { return (-F_APPROXIMATELY_ZERO < f) && (f < F_APPROXIMATELY_ZERO); } - -// These functions work by interpreting sign+exp+mantissa as an unsigned -// integer. -// For example: -// x = <sign>1 <exponent>00000010 <mantissa>00000000000000000000000 -// y = <sign>1 <exponent>00000001 <mantissa>11111111111111111111111 -// -// interpreted as ints = -// x = 10000001000000000000000000000000 -// y = 10000000111111111111111111111111 -// which is clearly a different of 1 in the least significant bit -// Values with the same exponent can be trivially shown to work. -// -// WARNING: Denormals of opposite sign do not work -// x = <sign>1 <exponent>00000000 <mantissa>00000000000000000000001 -// y = <sign>0 <exponent>00000000 <mantissa>00000000000000000000001 -// Although these values differ by 2 in the LSB, the sign bit makes -// the int comparison fail. -// -// WARNING: NaNs can compare equal -// There is no special treatment of exceptional values like NaNs -// -// WARNING: Infinity is comparable with F32_MAX and negative -// infinity is comparable with F32_MIN - -inline BOOL is_approx_equal(F32 x, F32 y) -{ - const S32 COMPARE_MANTISSA_UP_TO_BIT = 0x02; - return (std::abs((S32) ((U32&)x - (U32&)y) ) < COMPARE_MANTISSA_UP_TO_BIT); -} - -inline BOOL is_approx_equal(F64 x, F64 y) -{ - const S64 COMPARE_MANTISSA_UP_TO_BIT = 0x02; - return (std::abs((S32) ((U64&)x - (U64&)y) ) < COMPARE_MANTISSA_UP_TO_BIT); -} - -inline S32 llabs(const S32 a) -{ - return S32(std::labs(a)); -} - -inline F32 llabs(const F32 a) -{ - return F32(std::fabs(a)); -} - -inline F64 llabs(const F64 a) -{ - return F64(std::fabs(a)); -} - -inline S32 lltrunc( F32 f ) -{ -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) - // Avoids changing the floating point control word. - // Add or subtract 0.5 - epsilon and then round - const static U32 zpfp[] = { 0xBEFFFFFF, 0x3EFFFFFF }; - S32 result; - __asm { - fld f - mov eax, f - shr eax, 29 - and eax, 4 - fadd dword ptr [zpfp + eax] - fistp result - } - return result; -#else - return (S32)f; -#endif -} - -inline S32 lltrunc( F64 f ) -{ - return (S32)f; -} - -inline S32 llfloor( F32 f ) -{ -#if LL_WINDOWS && !defined( __INTEL_COMPILER ) - // Avoids changing the floating point control word. - // Accurate (unlike Stereopsis version) for all values between S32_MIN and S32_MAX and slightly faster than Stereopsis version. - // Add -(0.5 - epsilon) and then round - const U32 zpfp = 0xBEFFFFFF; - S32 result; - __asm { - fld f - fadd dword ptr [zpfp] - fistp result - } - return result; -#else - return (S32)floorf(f); -#endif -} - - -inline S32 llceil( F32 f ) -{ - // This could probably be optimized, but this works. - return (S32)ceil(f); -} - - -#ifndef BOGUS_ROUND -// Use this round. Does an arithmetic round (0.5 always rounds up) -inline S32 llround(const F32 val) -{ - return llfloor(val + 0.5f); -} - -#else // BOGUS_ROUND -// Old llround implementation - does banker's round (toward nearest even in the case of a 0.5. -// Not using this because we don't have a consistent implementation on both platforms, use -// llfloor(val + 0.5f), which is consistent on all platforms. -inline S32 llround(const F32 val) -{ - #if LL_WINDOWS - // Note: assumes that the floating point control word is set to rounding mode (the default) - S32 ret_val; - _asm fld val - _asm fistp ret_val; - return ret_val; - #elif LL_LINUX - // Note: assumes that the floating point control word is set - // to rounding mode (the default) - S32 ret_val; - __asm__ __volatile__( "flds %1 \n\t" - "fistpl %0 \n\t" - : "=m" (ret_val) - : "m" (val) ); - return ret_val; - #else - return llfloor(val + 0.5f); - #endif -} - -// A fast arithmentic round on intel, from Laurent de Soras http://ldesoras.free.fr -inline int round_int(double x) -{ - const float round_to_nearest = 0.5f; - int i; - __asm - { - fld x - fadd st, st (0) - fadd round_to_nearest - fistp i - sar i, 1 - } - return (i); -} -#endif // BOGUS_ROUND - -inline F32 llround( F32 val, F32 nearest ) -{ - return F32(floor(val * (1.0f / nearest) + 0.5f)) * nearest; -} - -inline F64 llround( F64 val, F64 nearest ) -{ - return F64(floor(val * (1.0 / nearest) + 0.5)) * nearest; -} - -// these provide minimum peak error -// -// avg error = -0.013049 -// peak error = -31.4 dB -// RMS error = -28.1 dB - -const F32 FAST_MAG_ALPHA = 0.960433870103f; -const F32 FAST_MAG_BETA = 0.397824734759f; - -// these provide minimum RMS error -// -// avg error = 0.000003 -// peak error = -32.6 dB -// RMS error = -25.7 dB -// -//const F32 FAST_MAG_ALPHA = 0.948059448969f; -//const F32 FAST_MAG_BETA = 0.392699081699f; - -inline F32 fastMagnitude(F32 a, F32 b) -{ - a = (a > 0) ? a : -a; - b = (b > 0) ? b : -b; - return(FAST_MAG_ALPHA * llmax(a,b) + FAST_MAG_BETA * llmin(a,b)); -} - - - -//////////////////// -// -// Fast F32/S32 conversions -// -// Culled from www.stereopsis.com/FPU.html - -const F64 LL_DOUBLE_TO_FIX_MAGIC = 68719476736.0*1.5; //2^36 * 1.5, (52-_shiftamt=36) uses limited precisicion to floor -const S32 LL_SHIFT_AMOUNT = 16; //16.16 fixed point representation, - -// Endian dependent code -#ifdef LL_LITTLE_ENDIAN - #define LL_EXP_INDEX 1 - #define LL_MAN_INDEX 0 -#else - #define LL_EXP_INDEX 0 - #define LL_MAN_INDEX 1 -#endif - -/* Deprecated: use llround(), lltrunc(), or llfloor() instead -// ================================================================================================ -// Real2Int -// ================================================================================================ -inline S32 F64toS32(F64 val) -{ - val = val + LL_DOUBLE_TO_FIX_MAGIC; - return ((S32*)&val)[LL_MAN_INDEX] >> LL_SHIFT_AMOUNT; -} - -// ================================================================================================ -// Real2Int -// ================================================================================================ -inline S32 F32toS32(F32 val) -{ - return F64toS32 ((F64)val); -} -*/ - -//////////////////////////////////////////////// -// -// Fast exp and log -// - -// Implementation of fast exp() approximation (from a paper by Nicol N. Schraudolph -// http://www.inf.ethz.ch/~schraudo/pubs/exp.pdf -static union -{ - double d; - struct - { -#ifdef LL_LITTLE_ENDIAN - S32 j, i; -#else - S32 i, j; -#endif - } n; -} LLECO; // not sure what the name means - -#define LL_EXP_A (1048576 * OO_LN2) // use 1512775 for integer -#define LL_EXP_C (60801) // this value of C good for -4 < y < 4 - -#define LL_FAST_EXP(y) (LLECO.n.i = llround(F32(LL_EXP_A*(y))) + (1072693248 - LL_EXP_C), LLECO.d) - - - -inline F32 llfastpow(const F32 x, const F32 y) -{ - return (F32)(LL_FAST_EXP(y * log(x))); -} - - -inline F32 snap_to_sig_figs(F32 foo, S32 sig_figs) -{ - // compute the power of ten - F32 bar = 1.f; - for (S32 i = 0; i < sig_figs; i++) - { - bar *= 10.f; - } - - foo = (F32)llround(foo * bar); - - // shift back - foo /= bar; - return foo; -} - -inline F32 lerp(F32 a, F32 b, F32 u) -{ - return a + ((b - a) * u); -} - -inline F32 lerp2d(F32 x00, F32 x01, F32 x10, F32 x11, F32 u, F32 v) -{ - F32 a = x00 + (x01-x00)*u; - F32 b = x10 + (x11-x10)*u; - F32 r = a + (b-a)*v; - return r; -} - -inline F32 ramp(F32 x, F32 a, F32 b) -{ - return (a == b) ? 0.0f : ((a - x) / (a - b)); -} - -inline F32 rescale(F32 x, F32 x1, F32 x2, F32 y1, F32 y2) -{ - return lerp(y1, y2, ramp(x, x1, x2)); -} - -inline F32 clamp_rescale(F32 x, F32 x1, F32 x2, F32 y1, F32 y2) -{ - if (y1 < y2) - { - return llclamp(rescale(x,x1,x2,y1,y2),y1,y2); - } - else - { - return llclamp(rescale(x,x1,x2,y1,y2),y2,y1); - } -} - - -inline F32 cubic_step( F32 x, F32 x0, F32 x1, F32 s0, F32 s1 ) -{ - if (x <= x0) - return s0; - - if (x >= x1) - return s1; - - F32 f = (x - x0) / (x1 - x0); - - return s0 + (s1 - s0) * (f * f) * (3.0f - 2.0f * f); -} - -inline F32 cubic_step( F32 x ) -{ - x = llclampf(x); - - return (x * x) * (3.0f - 2.0f * x); -} - -inline F32 quadratic_step( F32 x, F32 x0, F32 x1, F32 s0, F32 s1 ) -{ - if (x <= x0) - return s0; - - if (x >= x1) - return s1; - - F32 f = (x - x0) / (x1 - x0); - F32 f_squared = f * f; - - return (s0 * (1.f - f_squared)) + ((s1 - s0) * f_squared); -} - -inline F32 llsimple_angle(F32 angle) -{ - while(angle <= -F_PI) - angle += F_TWO_PI; - while(angle > F_PI) - angle -= F_TWO_PI; - return angle; -} - -//SDK - Renamed this to get_lower_power_two, since this is what this actually does. -inline U32 get_lower_power_two(U32 val, U32 max_power_two) -{ - if(!max_power_two) - { - max_power_two = 1 << 31 ; - } - if(max_power_two & (max_power_two - 1)) - { - return 0 ; - } - - for(; val < max_power_two ; max_power_two >>= 1) ; - - return max_power_two ; -} - -// calculate next highest power of two, limited by max_power_two -// This is taken from a brilliant little code snipped on http://acius2.blogspot.com/2007/11/calculating-next-power-of-2.html -// Basically we convert the binary to a solid string of 1's with the same -// number of digits, then add one. We subtract 1 initially to handle -// the case where the number passed in is actually a power of two. -// WARNING: this only works with 32 bit ints. -inline U32 get_next_power_two(U32 val, U32 max_power_two) -{ - if(!max_power_two) - { - max_power_two = 1 << 31 ; - } - - if(val >= max_power_two) - { - return max_power_two; - } - - val--; - val = (val >> 1) | val; - val = (val >> 2) | val; - val = (val >> 4) | val; - val = (val >> 8) | val; - val = (val >> 16) | val; - val++; - - return val; -} - -//get the gaussian value given the linear distance from axis x and guassian value o -inline F32 llgaussian(F32 x, F32 o) -{ - return 1.f/(F_SQRT_TWO_PI*o)*powf(F_E, -(x*x)/(2*o*o)); -} - -#endif +/**
+ * @file llmath.h
+ * @brief Useful math constants and macros.
+ *
+ * $LicenseInfo:firstyear=2000&license=viewergpl$
+ *
+ * Copyright (c) 2000-2009, Linden Research, Inc.
+ *
+ * Second Life Viewer Source Code
+ * The source code in this file ("Source Code") is provided by Linden Lab
+ * to you under the terms of the GNU General Public License, version 2.0
+ * ("GPL"), unless you have obtained a separate licensing agreement
+ * ("Other License"), formally executed by you and Linden Lab. Terms of
+ * the GPL can be found in doc/GPL-license.txt in this distribution, or
+ * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2
+ *
+ * There are special exceptions to the terms and conditions of the GPL as
+ * it is applied to this Source Code. View the full text of the exception
+ * in the file doc/FLOSS-exception.txt in this software distribution, or
+ * online at
+ * http://secondlifegrid.net/programs/open_source/licensing/flossexception
+ *
+ * By copying, modifying or distributing this software, you acknowledge
+ * that you have read and understood your obligations described above,
+ * and agree to abide by those obligations.
+ *
+ * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO
+ * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY,
+ * COMPLETENESS OR PERFORMANCE.
+ * $/LicenseInfo$
+ */
+
+#ifndef LLMATH_H
+#define LLMATH_H
+
+#include <cmath>
+#include <cstdlib>
+#include "lldefs.h"
+//#include "llstl.h" // *TODO: Remove when LLString is gone
+//#include "llstring.h" // *TODO: Remove when LLString is gone
+// lltut.h uses is_approx_equal_fraction(). This was moved to its own header
+// file in llcommon so we can use lltut.h for llcommon tests without making
+// llcommon depend on llmath.
+#include "is_approx_equal_fraction.h"
+
+// work around for Windows & older gcc non-standard function names.
+#if LL_WINDOWS
+#include <float.h>
+#define llisnan(val) _isnan(val)
+#define llfinite(val) _finite(val)
+#elif (LL_LINUX && __GNUC__ <= 2)
+#define llisnan(val) isnan(val)
+#define llfinite(val) isfinite(val)
+#elif LL_SOLARIS
+#define llisnan(val) isnan(val)
+#define llfinite(val) (val <= std::numeric_limits<double>::max())
+#else
+#define llisnan(val) std::isnan(val)
+#define llfinite(val) std::isfinite(val)
+#endif
+
+// Single Precision Floating Point Routines
+// (There used to be more defined here, but they appeared to be redundant and
+// were breaking some other includes. Removed by Falcon, reviewed by Andrew, 11/25/09)
+/*#ifndef tanf
+#define tanf(x) ((F32)tan((F64)(x)))
+#endif*/
+
+const F32 GRAVITY = -9.8f;
+
+// mathematical constants
+const F32 F_PI = 3.1415926535897932384626433832795f;
+const F32 F_TWO_PI = 6.283185307179586476925286766559f;
+const F32 F_PI_BY_TWO = 1.5707963267948966192313216916398f;
+const F32 F_SQRT_TWO_PI = 2.506628274631000502415765284811f;
+const F32 F_E = 2.71828182845904523536f;
+const F32 F_SQRT2 = 1.4142135623730950488016887242097f;
+const F32 F_SQRT3 = 1.73205080756888288657986402541f;
+const F32 OO_SQRT2 = 0.7071067811865475244008443621049f;
+const F32 DEG_TO_RAD = 0.017453292519943295769236907684886f;
+const F32 RAD_TO_DEG = 57.295779513082320876798154814105f;
+const F32 F_APPROXIMATELY_ZERO = 0.00001f;
+const F32 F_LN2 = 0.69314718056f;
+const F32 OO_LN2 = 1.4426950408889634073599246810019f;
+
+const F32 F_ALMOST_ZERO = 0.0001f;
+const F32 F_ALMOST_ONE = 1.0f - F_ALMOST_ZERO;
+
+// BUG: Eliminate in favor of F_APPROXIMATELY_ZERO above?
+const F32 FP_MAG_THRESHOLD = 0.0000001f;
+
+// TODO: Replace with logic like is_approx_equal
+inline BOOL is_approx_zero( F32 f ) { return (-F_APPROXIMATELY_ZERO < f) && (f < F_APPROXIMATELY_ZERO); }
+
+// These functions work by interpreting sign+exp+mantissa as an unsigned
+// integer.
+// For example:
+// x = <sign>1 <exponent>00000010 <mantissa>00000000000000000000000
+// y = <sign>1 <exponent>00000001 <mantissa>11111111111111111111111
+//
+// interpreted as ints =
+// x = 10000001000000000000000000000000
+// y = 10000000111111111111111111111111
+// which is clearly a different of 1 in the least significant bit
+// Values with the same exponent can be trivially shown to work.
+//
+// WARNING: Denormals of opposite sign do not work
+// x = <sign>1 <exponent>00000000 <mantissa>00000000000000000000001
+// y = <sign>0 <exponent>00000000 <mantissa>00000000000000000000001
+// Although these values differ by 2 in the LSB, the sign bit makes
+// the int comparison fail.
+//
+// WARNING: NaNs can compare equal
+// There is no special treatment of exceptional values like NaNs
+//
+// WARNING: Infinity is comparable with F32_MAX and negative
+// infinity is comparable with F32_MIN
+
+inline BOOL is_approx_equal(F32 x, F32 y)
+{
+ const S32 COMPARE_MANTISSA_UP_TO_BIT = 0x02;
+ return (std::abs((S32) ((U32&)x - (U32&)y) ) < COMPARE_MANTISSA_UP_TO_BIT);
+}
+
+inline BOOL is_approx_equal(F64 x, F64 y)
+{
+ const S64 COMPARE_MANTISSA_UP_TO_BIT = 0x02;
+ return (std::abs((S32) ((U64&)x - (U64&)y) ) < COMPARE_MANTISSA_UP_TO_BIT);
+}
+
+inline S32 llabs(const S32 a)
+{
+ return S32(std::labs(a));
+}
+
+inline F32 llabs(const F32 a)
+{
+ return F32(std::fabs(a));
+}
+
+inline F64 llabs(const F64 a)
+{
+ return F64(std::fabs(a));
+}
+
+inline S32 lltrunc( F32 f )
+{
+#if LL_WINDOWS && !defined( __INTEL_COMPILER )
+ // Avoids changing the floating point control word.
+ // Add or subtract 0.5 - epsilon and then round
+ const static U32 zpfp[] = { 0xBEFFFFFF, 0x3EFFFFFF };
+ S32 result;
+ __asm {
+ fld f
+ mov eax, f
+ shr eax, 29
+ and eax, 4
+ fadd dword ptr [zpfp + eax]
+ fistp result
+ }
+ return result;
+#else
+ return (S32)f;
+#endif
+}
+
+inline S32 lltrunc( F64 f )
+{
+ return (S32)f;
+}
+
+inline S32 llfloor( F32 f )
+{
+#if LL_WINDOWS && !defined( __INTEL_COMPILER )
+ // Avoids changing the floating point control word.
+ // Accurate (unlike Stereopsis version) for all values between S32_MIN and S32_MAX and slightly faster than Stereopsis version.
+ // Add -(0.5 - epsilon) and then round
+ const U32 zpfp = 0xBEFFFFFF;
+ S32 result;
+ __asm {
+ fld f
+ fadd dword ptr [zpfp]
+ fistp result
+ }
+ return result;
+#else
+ return (S32)floor(f);
+#endif
+}
+
+
+inline S32 llceil( F32 f )
+{
+ // This could probably be optimized, but this works.
+ return (S32)ceil(f);
+}
+
+
+#ifndef BOGUS_ROUND
+// Use this round. Does an arithmetic round (0.5 always rounds up)
+inline S32 llround(const F32 val)
+{
+ return llfloor(val + 0.5f);
+}
+
+#else // BOGUS_ROUND
+// Old llround implementation - does banker's round (toward nearest even in the case of a 0.5.
+// Not using this because we don't have a consistent implementation on both platforms, use
+// llfloor(val + 0.5f), which is consistent on all platforms.
+inline S32 llround(const F32 val)
+{
+ #if LL_WINDOWS
+ // Note: assumes that the floating point control word is set to rounding mode (the default)
+ S32 ret_val;
+ _asm fld val
+ _asm fistp ret_val;
+ return ret_val;
+ #elif LL_LINUX
+ // Note: assumes that the floating point control word is set
+ // to rounding mode (the default)
+ S32 ret_val;
+ __asm__ __volatile__( "flds %1 \n\t"
+ "fistpl %0 \n\t"
+ : "=m" (ret_val)
+ : "m" (val) );
+ return ret_val;
+ #else
+ return llfloor(val + 0.5f);
+ #endif
+}
+
+// A fast arithmentic round on intel, from Laurent de Soras http://ldesoras.free.fr
+inline int round_int(double x)
+{
+ const float round_to_nearest = 0.5f;
+ int i;
+ __asm
+ {
+ fld x
+ fadd st, st (0)
+ fadd round_to_nearest
+ fistp i
+ sar i, 1
+ }
+ return (i);
+}
+#endif // BOGUS_ROUND
+
+inline F32 llround( F32 val, F32 nearest )
+{
+ return F32(floor(val * (1.0f / nearest) + 0.5f)) * nearest;
+}
+
+inline F64 llround( F64 val, F64 nearest )
+{
+ return F64(floor(val * (1.0 / nearest) + 0.5)) * nearest;
+}
+
+// these provide minimum peak error
+//
+// avg error = -0.013049
+// peak error = -31.4 dB
+// RMS error = -28.1 dB
+
+const F32 FAST_MAG_ALPHA = 0.960433870103f;
+const F32 FAST_MAG_BETA = 0.397824734759f;
+
+// these provide minimum RMS error
+//
+// avg error = 0.000003
+// peak error = -32.6 dB
+// RMS error = -25.7 dB
+//
+//const F32 FAST_MAG_ALPHA = 0.948059448969f;
+//const F32 FAST_MAG_BETA = 0.392699081699f;
+
+inline F32 fastMagnitude(F32 a, F32 b)
+{
+ a = (a > 0) ? a : -a;
+ b = (b > 0) ? b : -b;
+ return(FAST_MAG_ALPHA * llmax(a,b) + FAST_MAG_BETA * llmin(a,b));
+}
+
+
+
+////////////////////
+//
+// Fast F32/S32 conversions
+//
+// Culled from www.stereopsis.com/FPU.html
+
+const F64 LL_DOUBLE_TO_FIX_MAGIC = 68719476736.0*1.5; //2^36 * 1.5, (52-_shiftamt=36) uses limited precisicion to floor
+const S32 LL_SHIFT_AMOUNT = 16; //16.16 fixed point representation,
+
+// Endian dependent code
+#ifdef LL_LITTLE_ENDIAN
+ #define LL_EXP_INDEX 1
+ #define LL_MAN_INDEX 0
+#else
+ #define LL_EXP_INDEX 0
+ #define LL_MAN_INDEX 1
+#endif
+
+/* Deprecated: use llround(), lltrunc(), or llfloor() instead
+// ================================================================================================
+// Real2Int
+// ================================================================================================
+inline S32 F64toS32(F64 val)
+{
+ val = val + LL_DOUBLE_TO_FIX_MAGIC;
+ return ((S32*)&val)[LL_MAN_INDEX] >> LL_SHIFT_AMOUNT;
+}
+
+// ================================================================================================
+// Real2Int
+// ================================================================================================
+inline S32 F32toS32(F32 val)
+{
+ return F64toS32 ((F64)val);
+}
+*/
+
+////////////////////////////////////////////////
+//
+// Fast exp and log
+//
+
+// Implementation of fast exp() approximation (from a paper by Nicol N. Schraudolph
+// http://www.inf.ethz.ch/~schraudo/pubs/exp.pdf
+static union
+{
+ double d;
+ struct
+ {
+#ifdef LL_LITTLE_ENDIAN
+ S32 j, i;
+#else
+ S32 i, j;
+#endif
+ } n;
+} LLECO; // not sure what the name means
+
+#define LL_EXP_A (1048576 * OO_LN2) // use 1512775 for integer
+#define LL_EXP_C (60801) // this value of C good for -4 < y < 4
+
+#define LL_FAST_EXP(y) (LLECO.n.i = llround(F32(LL_EXP_A*(y))) + (1072693248 - LL_EXP_C), LLECO.d)
+
+
+
+inline F32 llfastpow(const F32 x, const F32 y)
+{
+ return (F32)(LL_FAST_EXP(y * log(x)));
+}
+
+
+inline F32 snap_to_sig_figs(F32 foo, S32 sig_figs)
+{
+ // compute the power of ten
+ F32 bar = 1.f;
+ for (S32 i = 0; i < sig_figs; i++)
+ {
+ bar *= 10.f;
+ }
+
+ //F32 new_foo = (F32)llround(foo * bar);
+ // the llround() implementation sucks. Don't us it.
+
+ F32 sign = (foo > 0.f) ? 1.f : -1.f;
+ F32 new_foo = F32( S64(foo * bar + sign * 0.5f));
+ new_foo /= bar;
+
+ return new_foo;
+}
+
+inline F32 lerp(F32 a, F32 b, F32 u)
+{
+ return a + ((b - a) * u);
+}
+
+inline F32 lerp2d(F32 x00, F32 x01, F32 x10, F32 x11, F32 u, F32 v)
+{
+ F32 a = x00 + (x01-x00)*u;
+ F32 b = x10 + (x11-x10)*u;
+ F32 r = a + (b-a)*v;
+ return r;
+}
+
+inline F32 ramp(F32 x, F32 a, F32 b)
+{
+ return (a == b) ? 0.0f : ((a - x) / (a - b));
+}
+
+inline F32 rescale(F32 x, F32 x1, F32 x2, F32 y1, F32 y2)
+{
+ return lerp(y1, y2, ramp(x, x1, x2));
+}
+
+inline F32 clamp_rescale(F32 x, F32 x1, F32 x2, F32 y1, F32 y2)
+{
+ if (y1 < y2)
+ {
+ return llclamp(rescale(x,x1,x2,y1,y2),y1,y2);
+ }
+ else
+ {
+ return llclamp(rescale(x,x1,x2,y1,y2),y2,y1);
+ }
+}
+
+
+inline F32 cubic_step( F32 x, F32 x0, F32 x1, F32 s0, F32 s1 )
+{
+ if (x <= x0)
+ return s0;
+
+ if (x >= x1)
+ return s1;
+
+ F32 f = (x - x0) / (x1 - x0);
+
+ return s0 + (s1 - s0) * (f * f) * (3.0f - 2.0f * f);
+}
+
+inline F32 cubic_step( F32 x )
+{
+ x = llclampf(x);
+
+ return (x * x) * (3.0f - 2.0f * x);
+}
+
+inline F32 quadratic_step( F32 x, F32 x0, F32 x1, F32 s0, F32 s1 )
+{
+ if (x <= x0)
+ return s0;
+
+ if (x >= x1)
+ return s1;
+
+ F32 f = (x - x0) / (x1 - x0);
+ F32 f_squared = f * f;
+
+ return (s0 * (1.f - f_squared)) + ((s1 - s0) * f_squared);
+}
+
+inline F32 llsimple_angle(F32 angle)
+{
+ while(angle <= -F_PI)
+ angle += F_TWO_PI;
+ while(angle > F_PI)
+ angle -= F_TWO_PI;
+ return angle;
+}
+
+//SDK - Renamed this to get_lower_power_two, since this is what this actually does.
+inline U32 get_lower_power_two(U32 val, U32 max_power_two)
+{
+ if(!max_power_two)
+ {
+ max_power_two = 1 << 31 ;
+ }
+ if(max_power_two & (max_power_two - 1))
+ {
+ return 0 ;
+ }
+
+ for(; val < max_power_two ; max_power_two >>= 1) ;
+
+ return max_power_two ;
+}
+
+// calculate next highest power of two, limited by max_power_two
+// This is taken from a brilliant little code snipped on http://acius2.blogspot.com/2007/11/calculating-next-power-of-2.html
+// Basically we convert the binary to a solid string of 1's with the same
+// number of digits, then add one. We subtract 1 initially to handle
+// the case where the number passed in is actually a power of two.
+// WARNING: this only works with 32 bit ints.
+inline U32 get_next_power_two(U32 val, U32 max_power_two)
+{
+ if(!max_power_two)
+ {
+ max_power_two = 1 << 31 ;
+ }
+
+ if(val >= max_power_two)
+ {
+ return max_power_two;
+ }
+
+ val--;
+ val = (val >> 1) | val;
+ val = (val >> 2) | val;
+ val = (val >> 4) | val;
+ val = (val >> 8) | val;
+ val = (val >> 16) | val;
+ val++;
+
+ return val;
+}
+
+//get the gaussian value given the linear distance from axis x and guassian value o
+inline F32 llgaussian(F32 x, F32 o)
+{
+ return 1.f/(F_SQRT_TWO_PI*o)*powf(F_E, -(x*x)/(2*o*o));
+}
+
+// Include simd math header
+#include "llsimdmath.h"
+
+#endif
diff --git a/indra/llmath/lloctree.h b/indra/llmath/lloctree.h index 59828ae565..432e9fbcd8 100644 --- a/indra/llmath/lloctree.h +++ b/indra/llmath/lloctree.h @@ -142,7 +142,7 @@ public: S32 getOctant(const LLVector4a& pos) const //get the octant pos is in { - return pos.greaterThan4(mD[CENTER]).getComparisonMask() & 0x7; + return pos.greaterThan(mD[CENTER]).getGatheredBits() & 0x7; } inline bool isInside(const LLVector4a& pos, const F32& rad) const @@ -157,13 +157,13 @@ public: bool isInside(const LLVector4a& pos) const { - S32 gt = pos.greaterThan4(mD[MAX]).getComparisonMask() & 0x7; + S32 gt = pos.greaterThan(mD[MAX]).getGatheredBits() & 0x7; if (gt) { return false; } - S32 lt = pos.lessEqual4(mD[MIN]).getComparisonMask() & 0x7; + S32 lt = pos.lessEqual(mD[MIN]).getGatheredBits() & 0x7; if (lt) { return false; @@ -206,13 +206,13 @@ public: { const LLVector4a& pos = data->getPositionGroup(); - LLVector4a gt = pos.greaterThan4(center); + LLVector4a gt = pos.greaterThan(center); LLVector4a up; - up.mQ = _mm_and_ps(size.mQ, gt.mQ); + up = _mm_and_ps(size, gt); LLVector4a down; - down.mQ = _mm_andnot_ps(gt.mQ, size.mQ); + down = _mm_andnot_ps(gt, size); center.add(up); center.sub(down); @@ -326,9 +326,8 @@ public: LLVector4a val; val.setSub(center, getCenter()); val.setAbs(val); - LLVector4a app_zero; - app_zero.mQ = F_APPROXIMATELY_ZERO_4A; - S32 lt = val.lessThan4(app_zero).getComparisonMask() & 0x7; + + S32 lt = val.lessThan(LLVector4a::getEpsilon()).getGatheredBits() & 0x7; if( lt == 0x7 ) { @@ -642,7 +641,7 @@ public: LLVector4a val; val.setSub(v, BaseType::mD[BaseType::CENTER]); val.setAbs(val); - S32 lt = val.lessThan4(MAX_MAG).getComparisonMask() & 0x7; + S32 lt = val.lessThan(MAX_MAG).getGatheredBits() & 0x7; if (lt != 0x7) { diff --git a/indra/llmath/llquantize.h b/indra/llmath/llquantize.h index 2192427f07..000d8a060f 100644 --- a/indra/llmath/llquantize.h +++ b/indra/llmath/llquantize.h @@ -1,152 +1,158 @@ -/** - * @file llquantize.h - * @brief useful routines for quantizing floats to various length ints - * and back out again - * - * $LicenseInfo:firstyear=2001&license=viewergpl$ - * - * Copyright (c) 2001-2009, Linden Research, Inc. - * - * Second Life Viewer Source Code - * The source code in this file ("Source Code") is provided by Linden Lab - * to you under the terms of the GNU General Public License, version 2.0 - * ("GPL"), unless you have obtained a separate licensing agreement - * ("Other License"), formally executed by you and Linden Lab. Terms of - * the GPL can be found in doc/GPL-license.txt in this distribution, or - * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2 - * - * There are special exceptions to the terms and conditions of the GPL as - * it is applied to this Source Code. View the full text of the exception - * in the file doc/FLOSS-exception.txt in this software distribution, or - * online at - * http://secondlifegrid.net/programs/open_source/licensing/flossexception - * - * By copying, modifying or distributing this software, you acknowledge - * that you have read and understood your obligations described above, - * and agree to abide by those obligations. - * - * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO - * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY, - * COMPLETENESS OR PERFORMANCE. - * $/LicenseInfo$ - */ - -#ifndef LL_LLQUANTIZE_H -#define LL_LLQUANTIZE_H - -const U16 U16MAX = 65535; -const F32 OOU16MAX = 1.f/(F32)(U16MAX); - -const U8 U8MAX = 255; -const F32 OOU8MAX = 1.f/(F32)(U8MAX); - -const U8 FIRSTVALIDCHAR = 54; -const U8 MAXSTRINGVAL = U8MAX - FIRSTVALIDCHAR; //we don't allow newline or null - - -inline U16 F32_to_U16_ROUND(F32 val, F32 lower, F32 upper) -{ - val = llclamp(val, lower, upper); - // make sure that the value is positive and normalized to <0, 1> - val -= lower; - val /= (upper - lower); - - // round the value. Sreturn the U16 - return (U16)(llround(val*U16MAX)); -} - - -inline U16 F32_to_U16(F32 val, F32 lower, F32 upper) -{ - val = llclamp(val, lower, upper); - // make sure that the value is positive and normalized to <0, 1> - val -= lower; - val /= (upper - lower); - - // return the U16 - return (U16)(llfloor(val*U16MAX)); -} - -inline F32 U16_to_F32(U16 ival, F32 lower, F32 upper) -{ - F32 val = ival*OOU16MAX; - F32 delta = (upper - lower); - val *= delta; - val += lower; - - F32 max_error = delta*OOU16MAX; - - // make sure that zero's come through as zero - if (fabsf(val) < max_error) - val = 0.f; - - return val; -} - - -inline U8 F32_to_U8_ROUND(F32 val, F32 lower, F32 upper) -{ - val = llclamp(val, lower, upper); - // make sure that the value is positive and normalized to <0, 1> - val -= lower; - val /= (upper - lower); - - // return the rounded U8 - return (U8)(llround(val*U8MAX)); -} - - -inline U8 F32_to_U8(F32 val, F32 lower, F32 upper) -{ - val = llclamp(val, lower, upper); - // make sure that the value is positive and normalized to <0, 1> - val -= lower; - val /= (upper - lower); - - // return the U8 - return (U8)(llfloor(val*U8MAX)); -} - -inline F32 U8_to_F32(U8 ival, F32 lower, F32 upper) -{ - F32 val = ival*OOU8MAX; - F32 delta = (upper - lower); - val *= delta; - val += lower; - - F32 max_error = delta*OOU8MAX; - - // make sure that zero's come through as zero - if (fabsf(val) < max_error) - val = 0.f; - - return val; -} - -inline U8 F32_TO_STRING(F32 val, F32 lower, F32 upper) -{ - val = llclamp(val, lower, upper); //[lower, upper] - // make sure that the value is positive and normalized to <0, 1> - val -= lower; //[0, upper-lower] - val /= (upper - lower); //[0,1] - val = val * MAXSTRINGVAL; //[0, MAXSTRINGVAL] - val = floor(val + 0.5f); //[0, MAXSTRINGVAL] - - U8 stringVal = (U8)(val) + FIRSTVALIDCHAR; //[FIRSTVALIDCHAR, MAXSTRINGVAL + FIRSTVALIDCHAR] - return stringVal; -} - -inline F32 STRING_TO_F32(U8 ival, F32 lower, F32 upper) -{ - // remove empty space left for NULL, newline, etc. - ival -= FIRSTVALIDCHAR; //[0, MAXSTRINGVAL] - - F32 val = (F32)ival * (1.f / (F32)MAXSTRINGVAL); //[0, 1] - F32 delta = (upper - lower); - val *= delta; //[0, upper - lower] - val += lower; //[lower, upper] - - return val; -} - -#endif +/**
+ * @file llquantize.h
+ * @brief useful routines for quantizing floats to various length ints
+ * and back out again
+ *
+ * $LicenseInfo:firstyear=2001&license=viewergpl$
+ *
+ * Copyright (c) 2001-2009, Linden Research, Inc.
+ *
+ * Second Life Viewer Source Code
+ * The source code in this file ("Source Code") is provided by Linden Lab
+ * to you under the terms of the GNU General Public License, version 2.0
+ * ("GPL"), unless you have obtained a separate licensing agreement
+ * ("Other License"), formally executed by you and Linden Lab. Terms of
+ * the GPL can be found in doc/GPL-license.txt in this distribution, or
+ * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2
+ *
+ * There are special exceptions to the terms and conditions of the GPL as
+ * it is applied to this Source Code. View the full text of the exception
+ * in the file doc/FLOSS-exception.txt in this software distribution, or
+ * online at
+ * http://secondlifegrid.net/programs/open_source/licensing/flossexception
+ *
+ * By copying, modifying or distributing this software, you acknowledge
+ * that you have read and understood your obligations described above,
+ * and agree to abide by those obligations.
+ *
+ * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO
+ * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY,
+ * COMPLETENESS OR PERFORMANCE.
+ * $/LicenseInfo$
+ */
+
+#ifndef LL_LLQUANTIZE_H
+#define LL_LLQUANTIZE_H
+
+const U16 U16MAX = 65535;
+LL_ALIGN_16( const F32 F_U16MAX_4A[4] ) = { 65535.f, 65535.f, 65535.f, 65535.f };
+
+const F32 OOU16MAX = 1.f/(F32)(U16MAX);
+LL_ALIGN_16( const F32 F_OOU16MAX_4A[4] ) = { OOU16MAX, OOU16MAX, OOU16MAX, OOU16MAX };
+
+const U8 U8MAX = 255;
+LL_ALIGN_16( const F32 F_U8MAX_4A[4] ) = { 255.f, 255.f, 255.f, 255.f };
+
+const F32 OOU8MAX = 1.f/(F32)(U8MAX);
+LL_ALIGN_16( const F32 F_OOU8MAX_4A[4] ) = { OOU8MAX, OOU8MAX, OOU8MAX, OOU8MAX };
+
+const U8 FIRSTVALIDCHAR = 54;
+const U8 MAXSTRINGVAL = U8MAX - FIRSTVALIDCHAR; //we don't allow newline or null
+
+
+inline U16 F32_to_U16_ROUND(F32 val, F32 lower, F32 upper)
+{
+ val = llclamp(val, lower, upper);
+ // make sure that the value is positive and normalized to <0, 1>
+ val -= lower;
+ val /= (upper - lower);
+
+ // round the value. Sreturn the U16
+ return (U16)(llround(val*U16MAX));
+}
+
+
+inline U16 F32_to_U16(F32 val, F32 lower, F32 upper)
+{
+ val = llclamp(val, lower, upper);
+ // make sure that the value is positive and normalized to <0, 1>
+ val -= lower;
+ val /= (upper - lower);
+
+ // return the U16
+ return (U16)(llfloor(val*U16MAX));
+}
+
+inline F32 U16_to_F32(U16 ival, F32 lower, F32 upper)
+{
+ F32 val = ival*OOU16MAX;
+ F32 delta = (upper - lower);
+ val *= delta;
+ val += lower;
+
+ F32 max_error = delta*OOU16MAX;
+
+ // make sure that zero's come through as zero
+ if (fabsf(val) < max_error)
+ val = 0.f;
+
+ return val;
+}
+
+
+inline U8 F32_to_U8_ROUND(F32 val, F32 lower, F32 upper)
+{
+ val = llclamp(val, lower, upper);
+ // make sure that the value is positive and normalized to <0, 1>
+ val -= lower;
+ val /= (upper - lower);
+
+ // return the rounded U8
+ return (U8)(llround(val*U8MAX));
+}
+
+
+inline U8 F32_to_U8(F32 val, F32 lower, F32 upper)
+{
+ val = llclamp(val, lower, upper);
+ // make sure that the value is positive and normalized to <0, 1>
+ val -= lower;
+ val /= (upper - lower);
+
+ // return the U8
+ return (U8)(llfloor(val*U8MAX));
+}
+
+inline F32 U8_to_F32(U8 ival, F32 lower, F32 upper)
+{
+ F32 val = ival*OOU8MAX;
+ F32 delta = (upper - lower);
+ val *= delta;
+ val += lower;
+
+ F32 max_error = delta*OOU8MAX;
+
+ // make sure that zero's come through as zero
+ if (fabsf(val) < max_error)
+ val = 0.f;
+
+ return val;
+}
+
+inline U8 F32_TO_STRING(F32 val, F32 lower, F32 upper)
+{
+ val = llclamp(val, lower, upper); //[lower, upper]
+ // make sure that the value is positive and normalized to <0, 1>
+ val -= lower; //[0, upper-lower]
+ val /= (upper - lower); //[0,1]
+ val = val * MAXSTRINGVAL; //[0, MAXSTRINGVAL]
+ val = floor(val + 0.5f); //[0, MAXSTRINGVAL]
+
+ U8 stringVal = (U8)(val) + FIRSTVALIDCHAR; //[FIRSTVALIDCHAR, MAXSTRINGVAL + FIRSTVALIDCHAR]
+ return stringVal;
+}
+
+inline F32 STRING_TO_F32(U8 ival, F32 lower, F32 upper)
+{
+ // remove empty space left for NULL, newline, etc.
+ ival -= FIRSTVALIDCHAR; //[0, MAXSTRINGVAL]
+
+ F32 val = (F32)ival * (1.f / (F32)MAXSTRINGVAL); //[0, 1]
+ F32 delta = (upper - lower);
+ val *= delta; //[0, upper - lower]
+ val += lower; //[lower, upper]
+
+ return val;
+}
+
+#endif
diff --git a/indra/llmath/llquaternion.cpp b/indra/llmath/llquaternion.cpp index fdcc19d657..efdc10e2c6 100644 --- a/indra/llmath/llquaternion.cpp +++ b/indra/llmath/llquaternion.cpp @@ -1,960 +1,961 @@ -/** - * @file llquaternion.cpp - * @brief LLQuaternion class implementation. - * - * $LicenseInfo:firstyear=2000&license=viewergpl$ - * - * Copyright (c) 2000-2009, Linden Research, Inc. - * - * Second Life Viewer Source Code - * The source code in this file ("Source Code") is provided by Linden Lab - * to you under the terms of the GNU General Public License, version 2.0 - * ("GPL"), unless you have obtained a separate licensing agreement - * ("Other License"), formally executed by you and Linden Lab. Terms of - * the GPL can be found in doc/GPL-license.txt in this distribution, or - * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2 - * - * There are special exceptions to the terms and conditions of the GPL as - * it is applied to this Source Code. View the full text of the exception - * in the file doc/FLOSS-exception.txt in this software distribution, or - * online at - * http://secondlifegrid.net/programs/open_source/licensing/flossexception - * - * By copying, modifying or distributing this software, you acknowledge - * that you have read and understood your obligations described above, - * and agree to abide by those obligations. - * - * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO - * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY, - * COMPLETENESS OR PERFORMANCE. - * $/LicenseInfo$ - */ - -#include "linden_common.h" - -#include "llquaternion.h" - -#include "llmath.h" // for F_PI -//#include "vmath.h" -#include "v3math.h" -#include "v3dmath.h" -#include "v4math.h" -#include "m4math.h" -#include "m3math.h" -#include "llquantize.h" - -// WARNING: Don't use this for global const definitions! using this -// at the top of a *.cpp file might not give you what you think. -const LLQuaternion LLQuaternion::DEFAULT; - -// Constructors - -LLQuaternion::LLQuaternion(const LLMatrix4 &mat) -{ - *this = mat.quaternion(); - normalize(); -} - -LLQuaternion::LLQuaternion(const LLMatrix3 &mat) -{ - *this = mat.quaternion(); - normalize(); -} - -LLQuaternion::LLQuaternion(F32 angle, const LLVector4 &vec) -{ - LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]); - v.normalize(); - - F32 c, s; - c = cosf(angle*0.5f); - s = sinf(angle*0.5f); - - mQ[VX] = v.mV[VX] * s; - mQ[VY] = v.mV[VY] * s; - mQ[VZ] = v.mV[VZ] * s; - mQ[VW] = c; - normalize(); -} - -LLQuaternion::LLQuaternion(F32 angle, const LLVector3 &vec) -{ - LLVector3 v(vec); - v.normalize(); - - F32 c, s; - c = cosf(angle*0.5f); - s = sinf(angle*0.5f); - - mQ[VX] = v.mV[VX] * s; - mQ[VY] = v.mV[VY] * s; - mQ[VZ] = v.mV[VZ] * s; - mQ[VW] = c; - normalize(); -} - -LLQuaternion::LLQuaternion(const LLVector3 &x_axis, - const LLVector3 &y_axis, - const LLVector3 &z_axis) -{ - LLMatrix3 mat; - mat.setRows(x_axis, y_axis, z_axis); - *this = mat.quaternion(); - normalize(); -} - -// Quatizations -void LLQuaternion::quantize16(F32 lower, F32 upper) -{ - F32 x = mQ[VX]; - F32 y = mQ[VY]; - F32 z = mQ[VZ]; - F32 s = mQ[VS]; - - x = U16_to_F32(F32_to_U16_ROUND(x, lower, upper), lower, upper); - y = U16_to_F32(F32_to_U16_ROUND(y, lower, upper), lower, upper); - z = U16_to_F32(F32_to_U16_ROUND(z, lower, upper), lower, upper); - s = U16_to_F32(F32_to_U16_ROUND(s, lower, upper), lower, upper); - - mQ[VX] = x; - mQ[VY] = y; - mQ[VZ] = z; - mQ[VS] = s; - - normalize(); -} - -void LLQuaternion::quantize8(F32 lower, F32 upper) -{ - mQ[VX] = U8_to_F32(F32_to_U8_ROUND(mQ[VX], lower, upper), lower, upper); - mQ[VY] = U8_to_F32(F32_to_U8_ROUND(mQ[VY], lower, upper), lower, upper); - mQ[VZ] = U8_to_F32(F32_to_U8_ROUND(mQ[VZ], lower, upper), lower, upper); - mQ[VS] = U8_to_F32(F32_to_U8_ROUND(mQ[VS], lower, upper), lower, upper); - - normalize(); -} - -// LLVector3 Magnitude and Normalization Functions - - -// Set LLQuaternion routines - -const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, F32 x, F32 y, F32 z) -{ - LLVector3 vec(x, y, z); - vec.normalize(); - - angle *= 0.5f; - F32 c, s; - c = cosf(angle); - s = sinf(angle); - - mQ[VX] = vec.mV[VX]*s; - mQ[VY] = vec.mV[VY]*s; - mQ[VZ] = vec.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, const LLVector3 &vec) -{ - LLVector3 v(vec); - v.normalize(); - - angle *= 0.5f; - F32 c, s; - c = cosf(angle); - s = sinf(angle); - - mQ[VX] = v.mV[VX]*s; - mQ[VY] = v.mV[VY]*s; - mQ[VZ] = v.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, const LLVector4 &vec) -{ - LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]); - v.normalize(); - - F32 c, s; - c = cosf(angle*0.5f); - s = sinf(angle*0.5f); - - mQ[VX] = v.mV[VX]*s; - mQ[VY] = v.mV[VY]*s; - mQ[VZ] = v.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setEulerAngles(F32 roll, F32 pitch, F32 yaw) -{ - LLMatrix3 rot_mat(roll, pitch, yaw); - rot_mat.orthogonalize(); - *this = rot_mat.quaternion(); - - normalize(); - return (*this); -} - -// deprecated -const LLQuaternion& LLQuaternion::set(const LLMatrix3 &mat) -{ - *this = mat.quaternion(); - normalize(); - return (*this); -} - -// deprecated -const LLQuaternion& LLQuaternion::set(const LLMatrix4 &mat) -{ - *this = mat.quaternion(); - normalize(); - return (*this); -} - -// deprecated -const LLQuaternion& LLQuaternion::setQuat(F32 angle, F32 x, F32 y, F32 z) -{ - LLVector3 vec(x, y, z); - vec.normalize(); - - angle *= 0.5f; - F32 c, s; - c = cosf(angle); - s = sinf(angle); - - mQ[VX] = vec.mV[VX]*s; - mQ[VY] = vec.mV[VY]*s; - mQ[VZ] = vec.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -// deprecated -const LLQuaternion& LLQuaternion::setQuat(F32 angle, const LLVector3 &vec) -{ - LLVector3 v(vec); - v.normalize(); - - angle *= 0.5f; - F32 c, s; - c = cosf(angle); - s = sinf(angle); - - mQ[VX] = v.mV[VX]*s; - mQ[VY] = v.mV[VY]*s; - mQ[VZ] = v.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setQuat(F32 angle, const LLVector4 &vec) -{ - LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]); - v.normalize(); - - F32 c, s; - c = cosf(angle*0.5f); - s = sinf(angle*0.5f); - - mQ[VX] = v.mV[VX]*s; - mQ[VY] = v.mV[VY]*s; - mQ[VZ] = v.mV[VZ]*s; - mQ[VW] = c; - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setQuat(F32 roll, F32 pitch, F32 yaw) -{ - LLMatrix3 rot_mat(roll, pitch, yaw); - rot_mat.orthogonalize(); - *this = rot_mat.quaternion(); - - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setQuat(const LLMatrix3 &mat) -{ - *this = mat.quaternion(); - normalize(); - return (*this); -} - -const LLQuaternion& LLQuaternion::setQuat(const LLMatrix4 &mat) -{ - *this = mat.quaternion(); - normalize(); - return (*this); -//#if 1 -// // NOTE: LLQuaternion's are actually inverted with respect to -// // the matrices, so this code also assumes inverted quaternions -// // (-x, -y, -z, w). The result is that roll,pitch,yaw are applied -// // in reverse order (yaw,pitch,roll). -// F64 cosX = cos(roll); -// F64 cosY = cos(pitch); -// F64 cosZ = cos(yaw); -// -// F64 sinX = sin(roll); -// F64 sinY = sin(pitch); -// F64 sinZ = sin(yaw); -// -// mQ[VW] = (F32)sqrt(cosY*cosZ - sinX*sinY*sinZ + cosX*cosZ + cosX*cosY + 1.0)*.5; -// if (fabs(mQ[VW]) < F_APPROXIMATELY_ZERO) -// { -// // null rotation, any axis will do -// mQ[VX] = 0.0f; -// mQ[VY] = 1.0f; -// mQ[VZ] = 0.0f; -// } -// else -// { -// F32 inv_s = 1.0f / (4.0f * mQ[VW]); -// mQ[VX] = (F32)-(-sinX*cosY - cosX*sinY*sinZ - sinX*cosZ) * inv_s; -// mQ[VY] = (F32)-(-cosX*sinY*cosZ + sinX*sinZ - sinY) * inv_s; -// mQ[VZ] = (F32)-(-cosY*sinZ - sinX*sinY*cosZ - cosX*sinZ) * inv_s; -// } -// -//#else // This only works on a certain subset of roll/pitch/yaw -// -// F64 cosX = cosf(roll/2.0); -// F64 cosY = cosf(pitch/2.0); -// F64 cosZ = cosf(yaw/2.0); -// -// F64 sinX = sinf(roll/2.0); -// F64 sinY = sinf(pitch/2.0); -// F64 sinZ = sinf(yaw/2.0); -// -// mQ[VW] = (F32)(cosX*cosY*cosZ + sinX*sinY*sinZ); -// mQ[VX] = (F32)(sinX*cosY*cosZ - cosX*sinY*sinZ); -// mQ[VY] = (F32)(cosX*sinY*cosZ + sinX*cosY*sinZ); -// mQ[VZ] = (F32)(cosX*cosY*sinZ - sinX*sinY*cosZ); -//#endif -// -// normalize(); -// return (*this); -} - -// SJB: This code is correct for a logicly stored (non-transposed) matrix; -// Our matrices are stored transposed, OpenGL style, so this generates the -// INVERSE matrix, or the CORRECT matrix form an INVERSE quaternion. -// Because we use similar logic in LLMatrix3::quaternion(), -// we are internally consistant so everything works OK :) -LLMatrix3 LLQuaternion::getMatrix3(void) const -{ - LLMatrix3 mat; - F32 xx, xy, xz, xw, yy, yz, yw, zz, zw; - - xx = mQ[VX] * mQ[VX]; - xy = mQ[VX] * mQ[VY]; - xz = mQ[VX] * mQ[VZ]; - xw = mQ[VX] * mQ[VW]; - - yy = mQ[VY] * mQ[VY]; - yz = mQ[VY] * mQ[VZ]; - yw = mQ[VY] * mQ[VW]; - - zz = mQ[VZ] * mQ[VZ]; - zw = mQ[VZ] * mQ[VW]; - - mat.mMatrix[0][0] = 1.f - 2.f * ( yy + zz ); - mat.mMatrix[0][1] = 2.f * ( xy + zw ); - mat.mMatrix[0][2] = 2.f * ( xz - yw ); - - mat.mMatrix[1][0] = 2.f * ( xy - zw ); - mat.mMatrix[1][1] = 1.f - 2.f * ( xx + zz ); - mat.mMatrix[1][2] = 2.f * ( yz + xw ); - - mat.mMatrix[2][0] = 2.f * ( xz + yw ); - mat.mMatrix[2][1] = 2.f * ( yz - xw ); - mat.mMatrix[2][2] = 1.f - 2.f * ( xx + yy ); - - return mat; -} - -LLMatrix4 LLQuaternion::getMatrix4(void) const -{ - LLMatrix4 mat; - F32 xx, xy, xz, xw, yy, yz, yw, zz, zw; - - xx = mQ[VX] * mQ[VX]; - xy = mQ[VX] * mQ[VY]; - xz = mQ[VX] * mQ[VZ]; - xw = mQ[VX] * mQ[VW]; - - yy = mQ[VY] * mQ[VY]; - yz = mQ[VY] * mQ[VZ]; - yw = mQ[VY] * mQ[VW]; - - zz = mQ[VZ] * mQ[VZ]; - zw = mQ[VZ] * mQ[VW]; - - mat.mMatrix[0][0] = 1.f - 2.f * ( yy + zz ); - mat.mMatrix[0][1] = 2.f * ( xy + zw ); - mat.mMatrix[0][2] = 2.f * ( xz - yw ); - - mat.mMatrix[1][0] = 2.f * ( xy - zw ); - mat.mMatrix[1][1] = 1.f - 2.f * ( xx + zz ); - mat.mMatrix[1][2] = 2.f * ( yz + xw ); - - mat.mMatrix[2][0] = 2.f * ( xz + yw ); - mat.mMatrix[2][1] = 2.f * ( yz - xw ); - mat.mMatrix[2][2] = 1.f - 2.f * ( xx + yy ); - - // TODO -- should we set the translation portion to zero? - - return mat; -} - - - - -// Other useful methods - - -// calculate the shortest rotation from a to b -void LLQuaternion::shortestArc(const LLVector3 &a, const LLVector3 &b) -{ - // Make a local copy of both vectors. - LLVector3 vec_a = a; - LLVector3 vec_b = b; - - // Make sure neither vector is zero length. Also normalize - // the vectors while we are at it. - F32 vec_a_mag = vec_a.normalize(); - F32 vec_b_mag = vec_b.normalize(); - if (vec_a_mag < F_APPROXIMATELY_ZERO || - vec_b_mag < F_APPROXIMATELY_ZERO) - { - // Can't calculate a rotation from this. - // Just return ZERO_ROTATION instead. - loadIdentity(); - return; - } - - // Create an axis to rotate around, and the cos of the angle to rotate. - LLVector3 axis = vec_a % vec_b; - F32 cos_theta = vec_a * vec_b; - - // Check the angle between the vectors to see if they are parallel or anti-parallel. - if (cos_theta > 1.0 - F_APPROXIMATELY_ZERO) - { - // a and b are parallel. No rotation is necessary. - loadIdentity(); - } - else if (cos_theta < -1.0 + F_APPROXIMATELY_ZERO) - { - // a and b are anti-parallel. - // Rotate 180 degrees around some orthogonal axis. - // Find the projection of the x-axis onto a, and try - // using the vector between the projection and the x-axis - // as the orthogonal axis. - LLVector3 proj = vec_a.mV[VX] / (vec_a * vec_a) * vec_a; - LLVector3 ortho_axis(1.f, 0.f, 0.f); - ortho_axis -= proj; - - // Turn this into an orthonormal axis. - F32 ortho_length = ortho_axis.normalize(); - // If the axis' length is 0, then our guess at an orthogonal axis - // was wrong (a is parallel to the x-axis). - if (ortho_length < F_APPROXIMATELY_ZERO) - { - // Use the z-axis instead. - ortho_axis.setVec(0.f, 0.f, 1.f); - } - - // Construct a quaternion from this orthonormal axis. - mQ[VX] = ortho_axis.mV[VX]; - mQ[VY] = ortho_axis.mV[VY]; - mQ[VZ] = ortho_axis.mV[VZ]; - mQ[VW] = 0.f; - } - else - { - // a and b are NOT parallel or anti-parallel. - // Return the rotation between these vectors. - F32 theta = (F32)acos(cos_theta); - - setAngleAxis(theta, axis); - } -} - -// constrains rotation to a cone angle specified in radians -const LLQuaternion &LLQuaternion::constrain(F32 radians) -{ - const F32 cos_angle_lim = cosf( radians/2 ); // mQ[VW] limit - const F32 sin_angle_lim = sinf( radians/2 ); // rotation axis length limit - - if (mQ[VW] < 0.f) - { - mQ[VX] *= -1.f; - mQ[VY] *= -1.f; - mQ[VZ] *= -1.f; - mQ[VW] *= -1.f; - } - - // if rotation angle is greater than limit (cos is less than limit) - if( mQ[VW] < cos_angle_lim ) - { - mQ[VW] = cos_angle_lim; - F32 axis_len = sqrtf( mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] ); // sin(theta/2) - F32 axis_mult_fact = sin_angle_lim / axis_len; - mQ[VX] *= axis_mult_fact; - mQ[VY] *= axis_mult_fact; - mQ[VZ] *= axis_mult_fact; - } - - return *this; -} - -// Operators - -std::ostream& operator<<(std::ostream &s, const LLQuaternion &a) -{ - s << "{ " - << a.mQ[VX] << ", " << a.mQ[VY] << ", " << a.mQ[VZ] << ", " << a.mQ[VW] - << " }"; - return s; -} - - -// Does NOT renormalize the result -LLQuaternion operator*(const LLQuaternion &a, const LLQuaternion &b) -{ -// LLQuaternion::mMultCount++; - - LLQuaternion q( - b.mQ[3] * a.mQ[0] + b.mQ[0] * a.mQ[3] + b.mQ[1] * a.mQ[2] - b.mQ[2] * a.mQ[1], - b.mQ[3] * a.mQ[1] + b.mQ[1] * a.mQ[3] + b.mQ[2] * a.mQ[0] - b.mQ[0] * a.mQ[2], - b.mQ[3] * a.mQ[2] + b.mQ[2] * a.mQ[3] + b.mQ[0] * a.mQ[1] - b.mQ[1] * a.mQ[0], - b.mQ[3] * a.mQ[3] - b.mQ[0] * a.mQ[0] - b.mQ[1] * a.mQ[1] - b.mQ[2] * a.mQ[2] - ); - return q; -} - -/* -LLMatrix4 operator*(const LLMatrix4 &m, const LLQuaternion &q) -{ - LLMatrix4 qmat(q); - return (m*qmat); -} -*/ - - - -LLVector4 operator*(const LLVector4 &a, const LLQuaternion &rot) -{ - F32 rw = - rot.mQ[VX] * a.mV[VX] - rot.mQ[VY] * a.mV[VY] - rot.mQ[VZ] * a.mV[VZ]; - F32 rx = rot.mQ[VW] * a.mV[VX] + rot.mQ[VY] * a.mV[VZ] - rot.mQ[VZ] * a.mV[VY]; - F32 ry = rot.mQ[VW] * a.mV[VY] + rot.mQ[VZ] * a.mV[VX] - rot.mQ[VX] * a.mV[VZ]; - F32 rz = rot.mQ[VW] * a.mV[VZ] + rot.mQ[VX] * a.mV[VY] - rot.mQ[VY] * a.mV[VX]; - - F32 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY]; - F32 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ]; - F32 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX]; - - return LLVector4(nx, ny, nz, a.mV[VW]); -} - -LLVector3 operator*(const LLVector3 &a, const LLQuaternion &rot) -{ - F32 rw = - rot.mQ[VX] * a.mV[VX] - rot.mQ[VY] * a.mV[VY] - rot.mQ[VZ] * a.mV[VZ]; - F32 rx = rot.mQ[VW] * a.mV[VX] + rot.mQ[VY] * a.mV[VZ] - rot.mQ[VZ] * a.mV[VY]; - F32 ry = rot.mQ[VW] * a.mV[VY] + rot.mQ[VZ] * a.mV[VX] - rot.mQ[VX] * a.mV[VZ]; - F32 rz = rot.mQ[VW] * a.mV[VZ] + rot.mQ[VX] * a.mV[VY] - rot.mQ[VY] * a.mV[VX]; - - F32 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY]; - F32 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ]; - F32 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX]; - - return LLVector3(nx, ny, nz); -} - -LLVector3d operator*(const LLVector3d &a, const LLQuaternion &rot) -{ - F64 rw = - rot.mQ[VX] * a.mdV[VX] - rot.mQ[VY] * a.mdV[VY] - rot.mQ[VZ] * a.mdV[VZ]; - F64 rx = rot.mQ[VW] * a.mdV[VX] + rot.mQ[VY] * a.mdV[VZ] - rot.mQ[VZ] * a.mdV[VY]; - F64 ry = rot.mQ[VW] * a.mdV[VY] + rot.mQ[VZ] * a.mdV[VX] - rot.mQ[VX] * a.mdV[VZ]; - F64 rz = rot.mQ[VW] * a.mdV[VZ] + rot.mQ[VX] * a.mdV[VY] - rot.mQ[VY] * a.mdV[VX]; - - F64 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY]; - F64 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ]; - F64 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX]; - - return LLVector3d(nx, ny, nz); -} - -F32 dot(const LLQuaternion &a, const LLQuaternion &b) -{ - return a.mQ[VX] * b.mQ[VX] + - a.mQ[VY] * b.mQ[VY] + - a.mQ[VZ] * b.mQ[VZ] + - a.mQ[VW] * b.mQ[VW]; -} - -// DEMO HACK: This lerp is probably inocrrect now due intermediate normalization -// it should look more like the lerp below -#if 0 -// linear interpolation -LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q) -{ - LLQuaternion r; - r = t * (q - p) + p; - r.normalize(); - return r; -} -#endif - -// lerp from identity to q -LLQuaternion lerp(F32 t, const LLQuaternion &q) -{ - LLQuaternion r; - r.mQ[VX] = t * q.mQ[VX]; - r.mQ[VY] = t * q.mQ[VY]; - r.mQ[VZ] = t * q.mQ[VZ]; - r.mQ[VW] = t * (q.mQ[VZ] - 1.f) + 1.f; - r.normalize(); - return r; -} - -LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q) -{ - LLQuaternion r; - F32 inv_t; - - inv_t = 1.f - t; - - r.mQ[VX] = t * q.mQ[VX] + (inv_t * p.mQ[VX]); - r.mQ[VY] = t * q.mQ[VY] + (inv_t * p.mQ[VY]); - r.mQ[VZ] = t * q.mQ[VZ] + (inv_t * p.mQ[VZ]); - r.mQ[VW] = t * q.mQ[VW] + (inv_t * p.mQ[VW]); - r.normalize(); - return r; -} - - -// spherical linear interpolation -LLQuaternion slerp( F32 u, const LLQuaternion &a, const LLQuaternion &b ) -{ - // cosine theta = dot product of a and b - F32 cos_t = a.mQ[0]*b.mQ[0] + a.mQ[1]*b.mQ[1] + a.mQ[2]*b.mQ[2] + a.mQ[3]*b.mQ[3]; - - // if b is on opposite hemisphere from a, use -a instead - int bflip; - if (cos_t < 0.0f) - { - cos_t = -cos_t; - bflip = TRUE; - } - else - bflip = FALSE; - - // if B is (within precision limits) the same as A, - // just linear interpolate between A and B. - F32 alpha; // interpolant - F32 beta; // 1 - interpolant - if (1.0f - cos_t < 0.00001f) - { - beta = 1.0f - u; - alpha = u; - } - else - { - F32 theta = acosf(cos_t); - F32 sin_t = sinf(theta); - beta = sinf(theta - u*theta) / sin_t; - alpha = sinf(u*theta) / sin_t; - } - - if (bflip) - beta = -beta; - - // interpolate - LLQuaternion ret; - ret.mQ[0] = beta*a.mQ[0] + alpha*b.mQ[0]; - ret.mQ[1] = beta*a.mQ[1] + alpha*b.mQ[1]; - ret.mQ[2] = beta*a.mQ[2] + alpha*b.mQ[2]; - ret.mQ[3] = beta*a.mQ[3] + alpha*b.mQ[3]; - - return ret; -} - -// lerp whenever possible -LLQuaternion nlerp(F32 t, const LLQuaternion &a, const LLQuaternion &b) -{ - if (dot(a, b) < 0.f) - { - return slerp(t, a, b); - } - else - { - return lerp(t, a, b); - } -} - -LLQuaternion nlerp(F32 t, const LLQuaternion &q) -{ - if (q.mQ[VW] < 0.f) - { - return slerp(t, q); - } - else - { - return lerp(t, q); - } -} - -// slerp from identity quaternion to another quaternion -LLQuaternion slerp(F32 t, const LLQuaternion &q) -{ - F32 c = q.mQ[VW]; - if (1.0f == t || 1.0f == c) - { - // the trivial cases - return q; - } - - LLQuaternion r; - F32 s, angle, stq, stp; - - s = (F32) sqrt(1.f - c*c); - - if (c < 0.0f) - { - // when c < 0.0 then theta > PI/2 - // since quat and -quat are the same rotation we invert one of - // p or q to reduce unecessary spins - // A equivalent way to do it is to convert acos(c) as if it had - // been negative, and to negate stp - angle = (F32) acos(-c); - stp = -(F32) sin(angle * (1.f - t)); - stq = (F32) sin(angle * t); - } - else - { - angle = (F32) acos(c); - stp = (F32) sin(angle * (1.f - t)); - stq = (F32) sin(angle * t); - } - - r.mQ[VX] = (q.mQ[VX] * stq) / s; - r.mQ[VY] = (q.mQ[VY] * stq) / s; - r.mQ[VZ] = (q.mQ[VZ] * stq) / s; - r.mQ[VW] = (stp + q.mQ[VW] * stq) / s; - - return r; -} - -LLQuaternion mayaQ(F32 xRot, F32 yRot, F32 zRot, LLQuaternion::Order order) -{ - LLQuaternion xQ( xRot*DEG_TO_RAD, LLVector3(1.0f, 0.0f, 0.0f) ); - LLQuaternion yQ( yRot*DEG_TO_RAD, LLVector3(0.0f, 1.0f, 0.0f) ); - LLQuaternion zQ( zRot*DEG_TO_RAD, LLVector3(0.0f, 0.0f, 1.0f) ); - LLQuaternion ret; - switch( order ) - { - case LLQuaternion::XYZ: - ret = xQ * yQ * zQ; - break; - case LLQuaternion::YZX: - ret = yQ * zQ * xQ; - break; - case LLQuaternion::ZXY: - ret = zQ * xQ * yQ; - break; - case LLQuaternion::XZY: - ret = xQ * zQ * yQ; - break; - case LLQuaternion::YXZ: - ret = yQ * xQ * zQ; - break; - case LLQuaternion::ZYX: - ret = zQ * yQ * xQ; - break; - } - return ret; -} - -const char *OrderToString( const LLQuaternion::Order order ) -{ - const char *p = NULL; - switch( order ) - { - default: - case LLQuaternion::XYZ: - p = "XYZ"; - break; - case LLQuaternion::YZX: - p = "YZX"; - break; - case LLQuaternion::ZXY: - p = "ZXY"; - break; - case LLQuaternion::XZY: - p = "XZY"; - break; - case LLQuaternion::YXZ: - p = "YXZ"; - break; - case LLQuaternion::ZYX: - p = "ZYX"; - break; - } - return p; -} - -LLQuaternion::Order StringToOrder( const char *str ) -{ - if (strncmp(str, "XYZ", 3)==0 || strncmp(str, "xyz", 3)==0) - return LLQuaternion::XYZ; - - if (strncmp(str, "YZX", 3)==0 || strncmp(str, "yzx", 3)==0) - return LLQuaternion::YZX; - - if (strncmp(str, "ZXY", 3)==0 || strncmp(str, "zxy", 3)==0) - return LLQuaternion::ZXY; - - if (strncmp(str, "XZY", 3)==0 || strncmp(str, "xzy", 3)==0) - return LLQuaternion::XZY; - - if (strncmp(str, "YXZ", 3)==0 || strncmp(str, "yxz", 3)==0) - return LLQuaternion::YXZ; - - if (strncmp(str, "ZYX", 3)==0 || strncmp(str, "zyx", 3)==0) - return LLQuaternion::ZYX; - - return LLQuaternion::XYZ; -} - -void LLQuaternion::getAngleAxis(F32* angle, LLVector3 &vec) const -{ - F32 cos_a = mQ[VW]; - if (cos_a > 1.0f) cos_a = 1.0f; - if (cos_a < -1.0f) cos_a = -1.0f; - - F32 sin_a = (F32) sqrt( 1.0f - cos_a * cos_a ); - - if ( fabs( sin_a ) < 0.0005f ) - sin_a = 1.0f; - else - sin_a = 1.f/sin_a; - - F32 temp_angle = 2.0f * (F32) acos( cos_a ); - if (temp_angle > F_PI) - { - // The (angle,axis) pair should never have angles outside [PI, -PI] - // since we want the _shortest_ (angle,axis) solution. - // Since acos is defined for [0, PI], and we multiply by 2.0, we - // can push the angle outside the acceptible range. - // When this happens we set the angle to the other portion of a - // full 2PI rotation, and negate the axis, which reverses the - // direction of the rotation (by the right-hand rule). - *angle = 2.f * F_PI - temp_angle; - vec.mV[VX] = - mQ[VX] * sin_a; - vec.mV[VY] = - mQ[VY] * sin_a; - vec.mV[VZ] = - mQ[VZ] * sin_a; - } - else - { - *angle = temp_angle; - vec.mV[VX] = mQ[VX] * sin_a; - vec.mV[VY] = mQ[VY] * sin_a; - vec.mV[VZ] = mQ[VZ] * sin_a; - } -} - - -// quaternion does not need to be normalized -void LLQuaternion::getEulerAngles(F32 *roll, F32 *pitch, F32 *yaw) const -{ - LLMatrix3 rot_mat(*this); - rot_mat.orthogonalize(); - rot_mat.getEulerAngles(roll, pitch, yaw); - -// // NOTE: LLQuaternion's are actually inverted with respect to -// // the matrices, so this code also assumes inverted quaternions -// // (-x, -y, -z, w). The result is that roll,pitch,yaw are applied -// // in reverse order (yaw,pitch,roll). -// F32 x = -mQ[VX], y = -mQ[VY], z = -mQ[VZ], w = mQ[VW]; -// F64 m20 = 2.0*(x*z-y*w); -// if (1.0f - fabsf(m20) < F_APPROXIMATELY_ZERO) -// { -// *roll = 0.0f; -// *pitch = (F32)asin(m20); -// *yaw = (F32)atan2(2.0*(x*y-z*w), 1.0 - 2.0*(x*x+z*z)); -// } -// else -// { -// *roll = (F32)atan2(-2.0*(y*z+x*w), 1.0-2.0*(x*x+y*y)); -// *pitch = (F32)asin(m20); -// *yaw = (F32)atan2(-2.0*(x*y+z*w), 1.0-2.0*(y*y+z*z)); -// } -} - -// Saves space by using the fact that our quaternions are normalized -LLVector3 LLQuaternion::packToVector3() const -{ - if( mQ[VW] >= 0 ) - { - return LLVector3( mQ[VX], mQ[VY], mQ[VZ] ); - } - else - { - return LLVector3( -mQ[VX], -mQ[VY], -mQ[VZ] ); - } -} - -// Saves space by using the fact that our quaternions are normalized -void LLQuaternion::unpackFromVector3( const LLVector3& vec ) -{ - mQ[VX] = vec.mV[VX]; - mQ[VY] = vec.mV[VY]; - mQ[VZ] = vec.mV[VZ]; - F32 t = 1.f - vec.magVecSquared(); - if( t > 0 ) - { - mQ[VW] = sqrt( t ); - } - else - { - // Need this to avoid trying to find the square root of a negative number due - // to floating point error. - mQ[VW] = 0; - } -} - -BOOL LLQuaternion::parseQuat(const std::string& buf, LLQuaternion* value) -{ - if( buf.empty() || value == NULL) - { - return FALSE; - } - - LLQuaternion quat; - S32 count = sscanf( buf.c_str(), "%f %f %f %f", quat.mQ + 0, quat.mQ + 1, quat.mQ + 2, quat.mQ + 3 ); - if( 4 == count ) - { - value->set( quat ); - return TRUE; - } - - return FALSE; -} - - -// End +/**
+ * @file llquaternion.cpp
+ * @brief LLQuaternion class implementation.
+ *
+ * $LicenseInfo:firstyear=2000&license=viewergpl$
+ *
+ * Copyright (c) 2000-2009, Linden Research, Inc.
+ *
+ * Second Life Viewer Source Code
+ * The source code in this file ("Source Code") is provided by Linden Lab
+ * to you under the terms of the GNU General Public License, version 2.0
+ * ("GPL"), unless you have obtained a separate licensing agreement
+ * ("Other License"), formally executed by you and Linden Lab. Terms of
+ * the GPL can be found in doc/GPL-license.txt in this distribution, or
+ * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2
+ *
+ * There are special exceptions to the terms and conditions of the GPL as
+ * it is applied to this Source Code. View the full text of the exception
+ * in the file doc/FLOSS-exception.txt in this software distribution, or
+ * online at
+ * http://secondlifegrid.net/programs/open_source/licensing/flossexception
+ *
+ * By copying, modifying or distributing this software, you acknowledge
+ * that you have read and understood your obligations described above,
+ * and agree to abide by those obligations.
+ *
+ * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO
+ * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY,
+ * COMPLETENESS OR PERFORMANCE.
+ * $/LicenseInfo$
+ */
+
+#include "linden_common.h"
+
+#include "llmath.h" // for F_PI
+
+#include "llquaternion.h"
+
+//#include "vmath.h"
+#include "v3math.h"
+#include "v3dmath.h"
+#include "v4math.h"
+#include "m4math.h"
+#include "m3math.h"
+#include "llquantize.h"
+
+// WARNING: Don't use this for global const definitions! using this
+// at the top of a *.cpp file might not give you what you think.
+const LLQuaternion LLQuaternion::DEFAULT;
+
+// Constructors
+
+LLQuaternion::LLQuaternion(const LLMatrix4 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+}
+
+LLQuaternion::LLQuaternion(const LLMatrix3 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+}
+
+LLQuaternion::LLQuaternion(F32 angle, const LLVector4 &vec)
+{
+ LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]);
+ v.normalize();
+
+ F32 c, s;
+ c = cosf(angle*0.5f);
+ s = sinf(angle*0.5f);
+
+ mQ[VX] = v.mV[VX] * s;
+ mQ[VY] = v.mV[VY] * s;
+ mQ[VZ] = v.mV[VZ] * s;
+ mQ[VW] = c;
+ normalize();
+}
+
+LLQuaternion::LLQuaternion(F32 angle, const LLVector3 &vec)
+{
+ LLVector3 v(vec);
+ v.normalize();
+
+ F32 c, s;
+ c = cosf(angle*0.5f);
+ s = sinf(angle*0.5f);
+
+ mQ[VX] = v.mV[VX] * s;
+ mQ[VY] = v.mV[VY] * s;
+ mQ[VZ] = v.mV[VZ] * s;
+ mQ[VW] = c;
+ normalize();
+}
+
+LLQuaternion::LLQuaternion(const LLVector3 &x_axis,
+ const LLVector3 &y_axis,
+ const LLVector3 &z_axis)
+{
+ LLMatrix3 mat;
+ mat.setRows(x_axis, y_axis, z_axis);
+ *this = mat.quaternion();
+ normalize();
+}
+
+// Quatizations
+void LLQuaternion::quantize16(F32 lower, F32 upper)
+{
+ F32 x = mQ[VX];
+ F32 y = mQ[VY];
+ F32 z = mQ[VZ];
+ F32 s = mQ[VS];
+
+ x = U16_to_F32(F32_to_U16_ROUND(x, lower, upper), lower, upper);
+ y = U16_to_F32(F32_to_U16_ROUND(y, lower, upper), lower, upper);
+ z = U16_to_F32(F32_to_U16_ROUND(z, lower, upper), lower, upper);
+ s = U16_to_F32(F32_to_U16_ROUND(s, lower, upper), lower, upper);
+
+ mQ[VX] = x;
+ mQ[VY] = y;
+ mQ[VZ] = z;
+ mQ[VS] = s;
+
+ normalize();
+}
+
+void LLQuaternion::quantize8(F32 lower, F32 upper)
+{
+ mQ[VX] = U8_to_F32(F32_to_U8_ROUND(mQ[VX], lower, upper), lower, upper);
+ mQ[VY] = U8_to_F32(F32_to_U8_ROUND(mQ[VY], lower, upper), lower, upper);
+ mQ[VZ] = U8_to_F32(F32_to_U8_ROUND(mQ[VZ], lower, upper), lower, upper);
+ mQ[VS] = U8_to_F32(F32_to_U8_ROUND(mQ[VS], lower, upper), lower, upper);
+
+ normalize();
+}
+
+// LLVector3 Magnitude and Normalization Functions
+
+
+// Set LLQuaternion routines
+
+const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, F32 x, F32 y, F32 z)
+{
+ LLVector3 vec(x, y, z);
+ vec.normalize();
+
+ angle *= 0.5f;
+ F32 c, s;
+ c = cosf(angle);
+ s = sinf(angle);
+
+ mQ[VX] = vec.mV[VX]*s;
+ mQ[VY] = vec.mV[VY]*s;
+ mQ[VZ] = vec.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, const LLVector3 &vec)
+{
+ LLVector3 v(vec);
+ v.normalize();
+
+ angle *= 0.5f;
+ F32 c, s;
+ c = cosf(angle);
+ s = sinf(angle);
+
+ mQ[VX] = v.mV[VX]*s;
+ mQ[VY] = v.mV[VY]*s;
+ mQ[VZ] = v.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setAngleAxis(F32 angle, const LLVector4 &vec)
+{
+ LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]);
+ v.normalize();
+
+ F32 c, s;
+ c = cosf(angle*0.5f);
+ s = sinf(angle*0.5f);
+
+ mQ[VX] = v.mV[VX]*s;
+ mQ[VY] = v.mV[VY]*s;
+ mQ[VZ] = v.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setEulerAngles(F32 roll, F32 pitch, F32 yaw)
+{
+ LLMatrix3 rot_mat(roll, pitch, yaw);
+ rot_mat.orthogonalize();
+ *this = rot_mat.quaternion();
+
+ normalize();
+ return (*this);
+}
+
+// deprecated
+const LLQuaternion& LLQuaternion::set(const LLMatrix3 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+ return (*this);
+}
+
+// deprecated
+const LLQuaternion& LLQuaternion::set(const LLMatrix4 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+ return (*this);
+}
+
+// deprecated
+const LLQuaternion& LLQuaternion::setQuat(F32 angle, F32 x, F32 y, F32 z)
+{
+ LLVector3 vec(x, y, z);
+ vec.normalize();
+
+ angle *= 0.5f;
+ F32 c, s;
+ c = cosf(angle);
+ s = sinf(angle);
+
+ mQ[VX] = vec.mV[VX]*s;
+ mQ[VY] = vec.mV[VY]*s;
+ mQ[VZ] = vec.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+// deprecated
+const LLQuaternion& LLQuaternion::setQuat(F32 angle, const LLVector3 &vec)
+{
+ LLVector3 v(vec);
+ v.normalize();
+
+ angle *= 0.5f;
+ F32 c, s;
+ c = cosf(angle);
+ s = sinf(angle);
+
+ mQ[VX] = v.mV[VX]*s;
+ mQ[VY] = v.mV[VY]*s;
+ mQ[VZ] = v.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setQuat(F32 angle, const LLVector4 &vec)
+{
+ LLVector3 v(vec.mV[VX], vec.mV[VY], vec.mV[VZ]);
+ v.normalize();
+
+ F32 c, s;
+ c = cosf(angle*0.5f);
+ s = sinf(angle*0.5f);
+
+ mQ[VX] = v.mV[VX]*s;
+ mQ[VY] = v.mV[VY]*s;
+ mQ[VZ] = v.mV[VZ]*s;
+ mQ[VW] = c;
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setQuat(F32 roll, F32 pitch, F32 yaw)
+{
+ LLMatrix3 rot_mat(roll, pitch, yaw);
+ rot_mat.orthogonalize();
+ *this = rot_mat.quaternion();
+
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setQuat(const LLMatrix3 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+ return (*this);
+}
+
+const LLQuaternion& LLQuaternion::setQuat(const LLMatrix4 &mat)
+{
+ *this = mat.quaternion();
+ normalize();
+ return (*this);
+//#if 1
+// // NOTE: LLQuaternion's are actually inverted with respect to
+// // the matrices, so this code also assumes inverted quaternions
+// // (-x, -y, -z, w). The result is that roll,pitch,yaw are applied
+// // in reverse order (yaw,pitch,roll).
+// F64 cosX = cos(roll);
+// F64 cosY = cos(pitch);
+// F64 cosZ = cos(yaw);
+//
+// F64 sinX = sin(roll);
+// F64 sinY = sin(pitch);
+// F64 sinZ = sin(yaw);
+//
+// mQ[VW] = (F32)sqrt(cosY*cosZ - sinX*sinY*sinZ + cosX*cosZ + cosX*cosY + 1.0)*.5;
+// if (fabs(mQ[VW]) < F_APPROXIMATELY_ZERO)
+// {
+// // null rotation, any axis will do
+// mQ[VX] = 0.0f;
+// mQ[VY] = 1.0f;
+// mQ[VZ] = 0.0f;
+// }
+// else
+// {
+// F32 inv_s = 1.0f / (4.0f * mQ[VW]);
+// mQ[VX] = (F32)-(-sinX*cosY - cosX*sinY*sinZ - sinX*cosZ) * inv_s;
+// mQ[VY] = (F32)-(-cosX*sinY*cosZ + sinX*sinZ - sinY) * inv_s;
+// mQ[VZ] = (F32)-(-cosY*sinZ - sinX*sinY*cosZ - cosX*sinZ) * inv_s;
+// }
+//
+//#else // This only works on a certain subset of roll/pitch/yaw
+//
+// F64 cosX = cosf(roll/2.0);
+// F64 cosY = cosf(pitch/2.0);
+// F64 cosZ = cosf(yaw/2.0);
+//
+// F64 sinX = sinf(roll/2.0);
+// F64 sinY = sinf(pitch/2.0);
+// F64 sinZ = sinf(yaw/2.0);
+//
+// mQ[VW] = (F32)(cosX*cosY*cosZ + sinX*sinY*sinZ);
+// mQ[VX] = (F32)(sinX*cosY*cosZ - cosX*sinY*sinZ);
+// mQ[VY] = (F32)(cosX*sinY*cosZ + sinX*cosY*sinZ);
+// mQ[VZ] = (F32)(cosX*cosY*sinZ - sinX*sinY*cosZ);
+//#endif
+//
+// normalize();
+// return (*this);
+}
+
+// SJB: This code is correct for a logicly stored (non-transposed) matrix;
+// Our matrices are stored transposed, OpenGL style, so this generates the
+// INVERSE matrix, or the CORRECT matrix form an INVERSE quaternion.
+// Because we use similar logic in LLMatrix3::quaternion(),
+// we are internally consistant so everything works OK :)
+LLMatrix3 LLQuaternion::getMatrix3(void) const
+{
+ LLMatrix3 mat;
+ F32 xx, xy, xz, xw, yy, yz, yw, zz, zw;
+
+ xx = mQ[VX] * mQ[VX];
+ xy = mQ[VX] * mQ[VY];
+ xz = mQ[VX] * mQ[VZ];
+ xw = mQ[VX] * mQ[VW];
+
+ yy = mQ[VY] * mQ[VY];
+ yz = mQ[VY] * mQ[VZ];
+ yw = mQ[VY] * mQ[VW];
+
+ zz = mQ[VZ] * mQ[VZ];
+ zw = mQ[VZ] * mQ[VW];
+
+ mat.mMatrix[0][0] = 1.f - 2.f * ( yy + zz );
+ mat.mMatrix[0][1] = 2.f * ( xy + zw );
+ mat.mMatrix[0][2] = 2.f * ( xz - yw );
+
+ mat.mMatrix[1][0] = 2.f * ( xy - zw );
+ mat.mMatrix[1][1] = 1.f - 2.f * ( xx + zz );
+ mat.mMatrix[1][2] = 2.f * ( yz + xw );
+
+ mat.mMatrix[2][0] = 2.f * ( xz + yw );
+ mat.mMatrix[2][1] = 2.f * ( yz - xw );
+ mat.mMatrix[2][2] = 1.f - 2.f * ( xx + yy );
+
+ return mat;
+}
+
+LLMatrix4 LLQuaternion::getMatrix4(void) const
+{
+ LLMatrix4 mat;
+ F32 xx, xy, xz, xw, yy, yz, yw, zz, zw;
+
+ xx = mQ[VX] * mQ[VX];
+ xy = mQ[VX] * mQ[VY];
+ xz = mQ[VX] * mQ[VZ];
+ xw = mQ[VX] * mQ[VW];
+
+ yy = mQ[VY] * mQ[VY];
+ yz = mQ[VY] * mQ[VZ];
+ yw = mQ[VY] * mQ[VW];
+
+ zz = mQ[VZ] * mQ[VZ];
+ zw = mQ[VZ] * mQ[VW];
+
+ mat.mMatrix[0][0] = 1.f - 2.f * ( yy + zz );
+ mat.mMatrix[0][1] = 2.f * ( xy + zw );
+ mat.mMatrix[0][2] = 2.f * ( xz - yw );
+
+ mat.mMatrix[1][0] = 2.f * ( xy - zw );
+ mat.mMatrix[1][1] = 1.f - 2.f * ( xx + zz );
+ mat.mMatrix[1][2] = 2.f * ( yz + xw );
+
+ mat.mMatrix[2][0] = 2.f * ( xz + yw );
+ mat.mMatrix[2][1] = 2.f * ( yz - xw );
+ mat.mMatrix[2][2] = 1.f - 2.f * ( xx + yy );
+
+ // TODO -- should we set the translation portion to zero?
+
+ return mat;
+}
+
+
+
+
+// Other useful methods
+
+
+// calculate the shortest rotation from a to b
+void LLQuaternion::shortestArc(const LLVector3 &a, const LLVector3 &b)
+{
+ // Make a local copy of both vectors.
+ LLVector3 vec_a = a;
+ LLVector3 vec_b = b;
+
+ // Make sure neither vector is zero length. Also normalize
+ // the vectors while we are at it.
+ F32 vec_a_mag = vec_a.normalize();
+ F32 vec_b_mag = vec_b.normalize();
+ if (vec_a_mag < F_APPROXIMATELY_ZERO ||
+ vec_b_mag < F_APPROXIMATELY_ZERO)
+ {
+ // Can't calculate a rotation from this.
+ // Just return ZERO_ROTATION instead.
+ loadIdentity();
+ return;
+ }
+
+ // Create an axis to rotate around, and the cos of the angle to rotate.
+ LLVector3 axis = vec_a % vec_b;
+ F32 cos_theta = vec_a * vec_b;
+
+ // Check the angle between the vectors to see if they are parallel or anti-parallel.
+ if (cos_theta > 1.0 - F_APPROXIMATELY_ZERO)
+ {
+ // a and b are parallel. No rotation is necessary.
+ loadIdentity();
+ }
+ else if (cos_theta < -1.0 + F_APPROXIMATELY_ZERO)
+ {
+ // a and b are anti-parallel.
+ // Rotate 180 degrees around some orthogonal axis.
+ // Find the projection of the x-axis onto a, and try
+ // using the vector between the projection and the x-axis
+ // as the orthogonal axis.
+ LLVector3 proj = vec_a.mV[VX] / (vec_a * vec_a) * vec_a;
+ LLVector3 ortho_axis(1.f, 0.f, 0.f);
+ ortho_axis -= proj;
+
+ // Turn this into an orthonormal axis.
+ F32 ortho_length = ortho_axis.normalize();
+ // If the axis' length is 0, then our guess at an orthogonal axis
+ // was wrong (a is parallel to the x-axis).
+ if (ortho_length < F_APPROXIMATELY_ZERO)
+ {
+ // Use the z-axis instead.
+ ortho_axis.setVec(0.f, 0.f, 1.f);
+ }
+
+ // Construct a quaternion from this orthonormal axis.
+ mQ[VX] = ortho_axis.mV[VX];
+ mQ[VY] = ortho_axis.mV[VY];
+ mQ[VZ] = ortho_axis.mV[VZ];
+ mQ[VW] = 0.f;
+ }
+ else
+ {
+ // a and b are NOT parallel or anti-parallel.
+ // Return the rotation between these vectors.
+ F32 theta = (F32)acos(cos_theta);
+
+ setAngleAxis(theta, axis);
+ }
+}
+
+// constrains rotation to a cone angle specified in radians
+const LLQuaternion &LLQuaternion::constrain(F32 radians)
+{
+ const F32 cos_angle_lim = cosf( radians/2 ); // mQ[VW] limit
+ const F32 sin_angle_lim = sinf( radians/2 ); // rotation axis length limit
+
+ if (mQ[VW] < 0.f)
+ {
+ mQ[VX] *= -1.f;
+ mQ[VY] *= -1.f;
+ mQ[VZ] *= -1.f;
+ mQ[VW] *= -1.f;
+ }
+
+ // if rotation angle is greater than limit (cos is less than limit)
+ if( mQ[VW] < cos_angle_lim )
+ {
+ mQ[VW] = cos_angle_lim;
+ F32 axis_len = sqrtf( mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] ); // sin(theta/2)
+ F32 axis_mult_fact = sin_angle_lim / axis_len;
+ mQ[VX] *= axis_mult_fact;
+ mQ[VY] *= axis_mult_fact;
+ mQ[VZ] *= axis_mult_fact;
+ }
+
+ return *this;
+}
+
+// Operators
+
+std::ostream& operator<<(std::ostream &s, const LLQuaternion &a)
+{
+ s << "{ "
+ << a.mQ[VX] << ", " << a.mQ[VY] << ", " << a.mQ[VZ] << ", " << a.mQ[VW]
+ << " }";
+ return s;
+}
+
+
+// Does NOT renormalize the result
+LLQuaternion operator*(const LLQuaternion &a, const LLQuaternion &b)
+{
+// LLQuaternion::mMultCount++;
+
+ LLQuaternion q(
+ b.mQ[3] * a.mQ[0] + b.mQ[0] * a.mQ[3] + b.mQ[1] * a.mQ[2] - b.mQ[2] * a.mQ[1],
+ b.mQ[3] * a.mQ[1] + b.mQ[1] * a.mQ[3] + b.mQ[2] * a.mQ[0] - b.mQ[0] * a.mQ[2],
+ b.mQ[3] * a.mQ[2] + b.mQ[2] * a.mQ[3] + b.mQ[0] * a.mQ[1] - b.mQ[1] * a.mQ[0],
+ b.mQ[3] * a.mQ[3] - b.mQ[0] * a.mQ[0] - b.mQ[1] * a.mQ[1] - b.mQ[2] * a.mQ[2]
+ );
+ return q;
+}
+
+/*
+LLMatrix4 operator*(const LLMatrix4 &m, const LLQuaternion &q)
+{
+ LLMatrix4 qmat(q);
+ return (m*qmat);
+}
+*/
+
+
+
+LLVector4 operator*(const LLVector4 &a, const LLQuaternion &rot)
+{
+ F32 rw = - rot.mQ[VX] * a.mV[VX] - rot.mQ[VY] * a.mV[VY] - rot.mQ[VZ] * a.mV[VZ];
+ F32 rx = rot.mQ[VW] * a.mV[VX] + rot.mQ[VY] * a.mV[VZ] - rot.mQ[VZ] * a.mV[VY];
+ F32 ry = rot.mQ[VW] * a.mV[VY] + rot.mQ[VZ] * a.mV[VX] - rot.mQ[VX] * a.mV[VZ];
+ F32 rz = rot.mQ[VW] * a.mV[VZ] + rot.mQ[VX] * a.mV[VY] - rot.mQ[VY] * a.mV[VX];
+
+ F32 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY];
+ F32 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ];
+ F32 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX];
+
+ return LLVector4(nx, ny, nz, a.mV[VW]);
+}
+
+LLVector3 operator*(const LLVector3 &a, const LLQuaternion &rot)
+{
+ F32 rw = - rot.mQ[VX] * a.mV[VX] - rot.mQ[VY] * a.mV[VY] - rot.mQ[VZ] * a.mV[VZ];
+ F32 rx = rot.mQ[VW] * a.mV[VX] + rot.mQ[VY] * a.mV[VZ] - rot.mQ[VZ] * a.mV[VY];
+ F32 ry = rot.mQ[VW] * a.mV[VY] + rot.mQ[VZ] * a.mV[VX] - rot.mQ[VX] * a.mV[VZ];
+ F32 rz = rot.mQ[VW] * a.mV[VZ] + rot.mQ[VX] * a.mV[VY] - rot.mQ[VY] * a.mV[VX];
+
+ F32 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY];
+ F32 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ];
+ F32 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX];
+
+ return LLVector3(nx, ny, nz);
+}
+
+LLVector3d operator*(const LLVector3d &a, const LLQuaternion &rot)
+{
+ F64 rw = - rot.mQ[VX] * a.mdV[VX] - rot.mQ[VY] * a.mdV[VY] - rot.mQ[VZ] * a.mdV[VZ];
+ F64 rx = rot.mQ[VW] * a.mdV[VX] + rot.mQ[VY] * a.mdV[VZ] - rot.mQ[VZ] * a.mdV[VY];
+ F64 ry = rot.mQ[VW] * a.mdV[VY] + rot.mQ[VZ] * a.mdV[VX] - rot.mQ[VX] * a.mdV[VZ];
+ F64 rz = rot.mQ[VW] * a.mdV[VZ] + rot.mQ[VX] * a.mdV[VY] - rot.mQ[VY] * a.mdV[VX];
+
+ F64 nx = - rw * rot.mQ[VX] + rx * rot.mQ[VW] - ry * rot.mQ[VZ] + rz * rot.mQ[VY];
+ F64 ny = - rw * rot.mQ[VY] + ry * rot.mQ[VW] - rz * rot.mQ[VX] + rx * rot.mQ[VZ];
+ F64 nz = - rw * rot.mQ[VZ] + rz * rot.mQ[VW] - rx * rot.mQ[VY] + ry * rot.mQ[VX];
+
+ return LLVector3d(nx, ny, nz);
+}
+
+F32 dot(const LLQuaternion &a, const LLQuaternion &b)
+{
+ return a.mQ[VX] * b.mQ[VX] +
+ a.mQ[VY] * b.mQ[VY] +
+ a.mQ[VZ] * b.mQ[VZ] +
+ a.mQ[VW] * b.mQ[VW];
+}
+
+// DEMO HACK: This lerp is probably inocrrect now due intermediate normalization
+// it should look more like the lerp below
+#if 0
+// linear interpolation
+LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q)
+{
+ LLQuaternion r;
+ r = t * (q - p) + p;
+ r.normalize();
+ return r;
+}
+#endif
+
+// lerp from identity to q
+LLQuaternion lerp(F32 t, const LLQuaternion &q)
+{
+ LLQuaternion r;
+ r.mQ[VX] = t * q.mQ[VX];
+ r.mQ[VY] = t * q.mQ[VY];
+ r.mQ[VZ] = t * q.mQ[VZ];
+ r.mQ[VW] = t * (q.mQ[VZ] - 1.f) + 1.f;
+ r.normalize();
+ return r;
+}
+
+LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q)
+{
+ LLQuaternion r;
+ F32 inv_t;
+
+ inv_t = 1.f - t;
+
+ r.mQ[VX] = t * q.mQ[VX] + (inv_t * p.mQ[VX]);
+ r.mQ[VY] = t * q.mQ[VY] + (inv_t * p.mQ[VY]);
+ r.mQ[VZ] = t * q.mQ[VZ] + (inv_t * p.mQ[VZ]);
+ r.mQ[VW] = t * q.mQ[VW] + (inv_t * p.mQ[VW]);
+ r.normalize();
+ return r;
+}
+
+
+// spherical linear interpolation
+LLQuaternion slerp( F32 u, const LLQuaternion &a, const LLQuaternion &b )
+{
+ // cosine theta = dot product of a and b
+ F32 cos_t = a.mQ[0]*b.mQ[0] + a.mQ[1]*b.mQ[1] + a.mQ[2]*b.mQ[2] + a.mQ[3]*b.mQ[3];
+
+ // if b is on opposite hemisphere from a, use -a instead
+ int bflip;
+ if (cos_t < 0.0f)
+ {
+ cos_t = -cos_t;
+ bflip = TRUE;
+ }
+ else
+ bflip = FALSE;
+
+ // if B is (within precision limits) the same as A,
+ // just linear interpolate between A and B.
+ F32 alpha; // interpolant
+ F32 beta; // 1 - interpolant
+ if (1.0f - cos_t < 0.00001f)
+ {
+ beta = 1.0f - u;
+ alpha = u;
+ }
+ else
+ {
+ F32 theta = acosf(cos_t);
+ F32 sin_t = sinf(theta);
+ beta = sinf(theta - u*theta) / sin_t;
+ alpha = sinf(u*theta) / sin_t;
+ }
+
+ if (bflip)
+ beta = -beta;
+
+ // interpolate
+ LLQuaternion ret;
+ ret.mQ[0] = beta*a.mQ[0] + alpha*b.mQ[0];
+ ret.mQ[1] = beta*a.mQ[1] + alpha*b.mQ[1];
+ ret.mQ[2] = beta*a.mQ[2] + alpha*b.mQ[2];
+ ret.mQ[3] = beta*a.mQ[3] + alpha*b.mQ[3];
+
+ return ret;
+}
+
+// lerp whenever possible
+LLQuaternion nlerp(F32 t, const LLQuaternion &a, const LLQuaternion &b)
+{
+ if (dot(a, b) < 0.f)
+ {
+ return slerp(t, a, b);
+ }
+ else
+ {
+ return lerp(t, a, b);
+ }
+}
+
+LLQuaternion nlerp(F32 t, const LLQuaternion &q)
+{
+ if (q.mQ[VW] < 0.f)
+ {
+ return slerp(t, q);
+ }
+ else
+ {
+ return lerp(t, q);
+ }
+}
+
+// slerp from identity quaternion to another quaternion
+LLQuaternion slerp(F32 t, const LLQuaternion &q)
+{
+ F32 c = q.mQ[VW];
+ if (1.0f == t || 1.0f == c)
+ {
+ // the trivial cases
+ return q;
+ }
+
+ LLQuaternion r;
+ F32 s, angle, stq, stp;
+
+ s = (F32) sqrt(1.f - c*c);
+
+ if (c < 0.0f)
+ {
+ // when c < 0.0 then theta > PI/2
+ // since quat and -quat are the same rotation we invert one of
+ // p or q to reduce unecessary spins
+ // A equivalent way to do it is to convert acos(c) as if it had
+ // been negative, and to negate stp
+ angle = (F32) acos(-c);
+ stp = -(F32) sin(angle * (1.f - t));
+ stq = (F32) sin(angle * t);
+ }
+ else
+ {
+ angle = (F32) acos(c);
+ stp = (F32) sin(angle * (1.f - t));
+ stq = (F32) sin(angle * t);
+ }
+
+ r.mQ[VX] = (q.mQ[VX] * stq) / s;
+ r.mQ[VY] = (q.mQ[VY] * stq) / s;
+ r.mQ[VZ] = (q.mQ[VZ] * stq) / s;
+ r.mQ[VW] = (stp + q.mQ[VW] * stq) / s;
+
+ return r;
+}
+
+LLQuaternion mayaQ(F32 xRot, F32 yRot, F32 zRot, LLQuaternion::Order order)
+{
+ LLQuaternion xQ( xRot*DEG_TO_RAD, LLVector3(1.0f, 0.0f, 0.0f) );
+ LLQuaternion yQ( yRot*DEG_TO_RAD, LLVector3(0.0f, 1.0f, 0.0f) );
+ LLQuaternion zQ( zRot*DEG_TO_RAD, LLVector3(0.0f, 0.0f, 1.0f) );
+ LLQuaternion ret;
+ switch( order )
+ {
+ case LLQuaternion::XYZ:
+ ret = xQ * yQ * zQ;
+ break;
+ case LLQuaternion::YZX:
+ ret = yQ * zQ * xQ;
+ break;
+ case LLQuaternion::ZXY:
+ ret = zQ * xQ * yQ;
+ break;
+ case LLQuaternion::XZY:
+ ret = xQ * zQ * yQ;
+ break;
+ case LLQuaternion::YXZ:
+ ret = yQ * xQ * zQ;
+ break;
+ case LLQuaternion::ZYX:
+ ret = zQ * yQ * xQ;
+ break;
+ }
+ return ret;
+}
+
+const char *OrderToString( const LLQuaternion::Order order )
+{
+ const char *p = NULL;
+ switch( order )
+ {
+ default:
+ case LLQuaternion::XYZ:
+ p = "XYZ";
+ break;
+ case LLQuaternion::YZX:
+ p = "YZX";
+ break;
+ case LLQuaternion::ZXY:
+ p = "ZXY";
+ break;
+ case LLQuaternion::XZY:
+ p = "XZY";
+ break;
+ case LLQuaternion::YXZ:
+ p = "YXZ";
+ break;
+ case LLQuaternion::ZYX:
+ p = "ZYX";
+ break;
+ }
+ return p;
+}
+
+LLQuaternion::Order StringToOrder( const char *str )
+{
+ if (strncmp(str, "XYZ", 3)==0 || strncmp(str, "xyz", 3)==0)
+ return LLQuaternion::XYZ;
+
+ if (strncmp(str, "YZX", 3)==0 || strncmp(str, "yzx", 3)==0)
+ return LLQuaternion::YZX;
+
+ if (strncmp(str, "ZXY", 3)==0 || strncmp(str, "zxy", 3)==0)
+ return LLQuaternion::ZXY;
+
+ if (strncmp(str, "XZY", 3)==0 || strncmp(str, "xzy", 3)==0)
+ return LLQuaternion::XZY;
+
+ if (strncmp(str, "YXZ", 3)==0 || strncmp(str, "yxz", 3)==0)
+ return LLQuaternion::YXZ;
+
+ if (strncmp(str, "ZYX", 3)==0 || strncmp(str, "zyx", 3)==0)
+ return LLQuaternion::ZYX;
+
+ return LLQuaternion::XYZ;
+}
+
+void LLQuaternion::getAngleAxis(F32* angle, LLVector3 &vec) const
+{
+ F32 cos_a = mQ[VW];
+ if (cos_a > 1.0f) cos_a = 1.0f;
+ if (cos_a < -1.0f) cos_a = -1.0f;
+
+ F32 sin_a = (F32) sqrt( 1.0f - cos_a * cos_a );
+
+ if ( fabs( sin_a ) < 0.0005f )
+ sin_a = 1.0f;
+ else
+ sin_a = 1.f/sin_a;
+
+ F32 temp_angle = 2.0f * (F32) acos( cos_a );
+ if (temp_angle > F_PI)
+ {
+ // The (angle,axis) pair should never have angles outside [PI, -PI]
+ // since we want the _shortest_ (angle,axis) solution.
+ // Since acos is defined for [0, PI], and we multiply by 2.0, we
+ // can push the angle outside the acceptible range.
+ // When this happens we set the angle to the other portion of a
+ // full 2PI rotation, and negate the axis, which reverses the
+ // direction of the rotation (by the right-hand rule).
+ *angle = 2.f * F_PI - temp_angle;
+ vec.mV[VX] = - mQ[VX] * sin_a;
+ vec.mV[VY] = - mQ[VY] * sin_a;
+ vec.mV[VZ] = - mQ[VZ] * sin_a;
+ }
+ else
+ {
+ *angle = temp_angle;
+ vec.mV[VX] = mQ[VX] * sin_a;
+ vec.mV[VY] = mQ[VY] * sin_a;
+ vec.mV[VZ] = mQ[VZ] * sin_a;
+ }
+}
+
+
+// quaternion does not need to be normalized
+void LLQuaternion::getEulerAngles(F32 *roll, F32 *pitch, F32 *yaw) const
+{
+ LLMatrix3 rot_mat(*this);
+ rot_mat.orthogonalize();
+ rot_mat.getEulerAngles(roll, pitch, yaw);
+
+// // NOTE: LLQuaternion's are actually inverted with respect to
+// // the matrices, so this code also assumes inverted quaternions
+// // (-x, -y, -z, w). The result is that roll,pitch,yaw are applied
+// // in reverse order (yaw,pitch,roll).
+// F32 x = -mQ[VX], y = -mQ[VY], z = -mQ[VZ], w = mQ[VW];
+// F64 m20 = 2.0*(x*z-y*w);
+// if (1.0f - fabsf(m20) < F_APPROXIMATELY_ZERO)
+// {
+// *roll = 0.0f;
+// *pitch = (F32)asin(m20);
+// *yaw = (F32)atan2(2.0*(x*y-z*w), 1.0 - 2.0*(x*x+z*z));
+// }
+// else
+// {
+// *roll = (F32)atan2(-2.0*(y*z+x*w), 1.0-2.0*(x*x+y*y));
+// *pitch = (F32)asin(m20);
+// *yaw = (F32)atan2(-2.0*(x*y+z*w), 1.0-2.0*(y*y+z*z));
+// }
+}
+
+// Saves space by using the fact that our quaternions are normalized
+LLVector3 LLQuaternion::packToVector3() const
+{
+ if( mQ[VW] >= 0 )
+ {
+ return LLVector3( mQ[VX], mQ[VY], mQ[VZ] );
+ }
+ else
+ {
+ return LLVector3( -mQ[VX], -mQ[VY], -mQ[VZ] );
+ }
+}
+
+// Saves space by using the fact that our quaternions are normalized
+void LLQuaternion::unpackFromVector3( const LLVector3& vec )
+{
+ mQ[VX] = vec.mV[VX];
+ mQ[VY] = vec.mV[VY];
+ mQ[VZ] = vec.mV[VZ];
+ F32 t = 1.f - vec.magVecSquared();
+ if( t > 0 )
+ {
+ mQ[VW] = sqrt( t );
+ }
+ else
+ {
+ // Need this to avoid trying to find the square root of a negative number due
+ // to floating point error.
+ mQ[VW] = 0;
+ }
+}
+
+BOOL LLQuaternion::parseQuat(const std::string& buf, LLQuaternion* value)
+{
+ if( buf.empty() || value == NULL)
+ {
+ return FALSE;
+ }
+
+ LLQuaternion quat;
+ S32 count = sscanf( buf.c_str(), "%f %f %f %f", quat.mQ + 0, quat.mQ + 1, quat.mQ + 2, quat.mQ + 3 );
+ if( 4 == count )
+ {
+ value->set( quat );
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+
+// End
diff --git a/indra/llmath/llquaternion.h b/indra/llmath/llquaternion.h index 0769f29f23..bbd4326483 100644 --- a/indra/llmath/llquaternion.h +++ b/indra/llmath/llquaternion.h @@ -1,590 +1,594 @@ -/** - * @file llquaternion.h - * @brief LLQuaternion class header file. - * - * $LicenseInfo:firstyear=2000&license=viewergpl$ - * - * Copyright (c) 2000-2009, Linden Research, Inc. - * - * Second Life Viewer Source Code - * The source code in this file ("Source Code") is provided by Linden Lab - * to you under the terms of the GNU General Public License, version 2.0 - * ("GPL"), unless you have obtained a separate licensing agreement - * ("Other License"), formally executed by you and Linden Lab. Terms of - * the GPL can be found in doc/GPL-license.txt in this distribution, or - * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2 - * - * There are special exceptions to the terms and conditions of the GPL as - * it is applied to this Source Code. View the full text of the exception - * in the file doc/FLOSS-exception.txt in this software distribution, or - * online at - * http://secondlifegrid.net/programs/open_source/licensing/flossexception - * - * By copying, modifying or distributing this software, you acknowledge - * that you have read and understood your obligations described above, - * and agree to abide by those obligations. - * - * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO - * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY, - * COMPLETENESS OR PERFORMANCE. - * $/LicenseInfo$ - */ - -#ifndef LLQUATERNION_H -#define LLQUATERNION_H - -#include "llmath.h" - -class LLVector4; -class LLVector3; -class LLVector3d; -class LLMatrix4; -class LLMatrix3; - -// NOTA BENE: Quaternion code is written assuming Unit Quaternions!!!! -// Moreover, it is written assuming that all vectors and matricies -// passed as arguments are normalized and unitary respectively. -// VERY VERY VERY VERY BAD THINGS will happen if these assumptions fail. - -static const U32 LENGTHOFQUAT = 4; - -class LLQuaternion -{ -public: - F32 mQ[LENGTHOFQUAT]; - - static const LLQuaternion DEFAULT; - - LLQuaternion(); // Initializes Quaternion to (0,0,0,1) - explicit LLQuaternion(const LLMatrix4 &mat); // Initializes Quaternion from Matrix4 - explicit LLQuaternion(const LLMatrix3 &mat); // Initializes Quaternion from Matrix3 - LLQuaternion(F32 x, F32 y, F32 z, F32 w); // Initializes Quaternion to normalize(x, y, z, w) - LLQuaternion(F32 angle, const LLVector4 &vec); // Initializes Quaternion to axis_angle2quat(angle, vec) - LLQuaternion(F32 angle, const LLVector3 &vec); // Initializes Quaternion to axis_angle2quat(angle, vec) - LLQuaternion(const F32 *q); // Initializes Quaternion to normalize(x, y, z, w) - LLQuaternion(const LLVector3 &x_axis, - const LLVector3 &y_axis, - const LLVector3 &z_axis); // Initializes Quaternion from Matrix3 = [x_axis ; y_axis ; z_axis] - - BOOL isIdentity() const; - BOOL isNotIdentity() const; - BOOL isFinite() const; // checks to see if all values of LLQuaternion are finite - void quantize16(F32 lower, F32 upper); // changes the vector to reflect quatization - void quantize8(F32 lower, F32 upper); // changes the vector to reflect quatization - void loadIdentity(); // Loads the quaternion that represents the identity rotation - - const LLQuaternion& set(F32 x, F32 y, F32 z, F32 w); // Sets Quaternion to normalize(x, y, z, w) - const LLQuaternion& set(const LLQuaternion &quat); // Copies Quaternion - const LLQuaternion& set(const F32 *q); // Sets Quaternion to normalize(quat[VX], quat[VY], quat[VZ], quat[VW]) - const LLQuaternion& set(const LLMatrix3 &mat); // Sets Quaternion to mat2quat(mat) - const LLQuaternion& set(const LLMatrix4 &mat); // Sets Quaternion to mat2quat(mat) - - const LLQuaternion& setAngleAxis(F32 angle, F32 x, F32 y, F32 z); // Sets Quaternion to axis_angle2quat(angle, x, y, z) - const LLQuaternion& setAngleAxis(F32 angle, const LLVector3 &vec); // Sets Quaternion to axis_angle2quat(angle, vec) - const LLQuaternion& setAngleAxis(F32 angle, const LLVector4 &vec); // Sets Quaternion to axis_angle2quat(angle, vec) - const LLQuaternion& setEulerAngles(F32 roll, F32 pitch, F32 yaw); // Sets Quaternion to euler2quat(pitch, yaw, roll) - - const LLQuaternion& setQuatInit(F32 x, F32 y, F32 z, F32 w); // deprecated - const LLQuaternion& setQuat(const LLQuaternion &quat); // deprecated - const LLQuaternion& setQuat(const F32 *q); // deprecated - const LLQuaternion& setQuat(const LLMatrix3 &mat); // deprecated - const LLQuaternion& setQuat(const LLMatrix4 &mat); // deprecated - const LLQuaternion& setQuat(F32 angle, F32 x, F32 y, F32 z); // deprecated - const LLQuaternion& setQuat(F32 angle, const LLVector3 &vec); // deprecated - const LLQuaternion& setQuat(F32 angle, const LLVector4 &vec); // deprecated - const LLQuaternion& setQuat(F32 roll, F32 pitch, F32 yaw); // deprecated - - LLMatrix4 getMatrix4(void) const; // Returns the Matrix4 equivalent of Quaternion - LLMatrix3 getMatrix3(void) const; // Returns the Matrix3 equivalent of Quaternion - void getAngleAxis(F32* angle, F32* x, F32* y, F32* z) const; // returns rotation in radians about axis x,y,z - void getAngleAxis(F32* angle, LLVector3 &vec) const; - void getEulerAngles(F32 *roll, F32* pitch, F32 *yaw) const; - - F32 normalize(); // Normalizes Quaternion and returns magnitude - F32 normQuat(); // deprecated - - const LLQuaternion& conjugate(void); // Conjugates Quaternion and returns result - const LLQuaternion& conjQuat(void); // deprecated - - // Other useful methods - const LLQuaternion& transpose(); // transpose (same as conjugate) - const LLQuaternion& transQuat(); // deprecated - - void shortestArc(const LLVector3 &a, const LLVector3 &b); // shortest rotation from a to b - const LLQuaternion& constrain(F32 radians); // constrains rotation to a cone angle specified in radians - - // Standard operators - friend std::ostream& operator<<(std::ostream &s, const LLQuaternion &a); // Prints a - friend LLQuaternion operator+(const LLQuaternion &a, const LLQuaternion &b); // Addition - friend LLQuaternion operator-(const LLQuaternion &a, const LLQuaternion &b); // Subtraction - friend LLQuaternion operator-(const LLQuaternion &a); // Negation - friend LLQuaternion operator*(F32 a, const LLQuaternion &q); // Scale - friend LLQuaternion operator*(const LLQuaternion &q, F32 b); // Scale - friend LLQuaternion operator*(const LLQuaternion &a, const LLQuaternion &b); // Returns a * b - friend LLQuaternion operator~(const LLQuaternion &a); // Returns a* (Conjugate of a) - bool operator==(const LLQuaternion &b) const; // Returns a == b - bool operator!=(const LLQuaternion &b) const; // Returns a != b - - friend const LLQuaternion& operator*=(LLQuaternion &a, const LLQuaternion &b); // Returns a * b - - friend LLVector4 operator*(const LLVector4 &a, const LLQuaternion &rot); // Rotates a by rot - friend LLVector3 operator*(const LLVector3 &a, const LLQuaternion &rot); // Rotates a by rot - friend LLVector3d operator*(const LLVector3d &a, const LLQuaternion &rot); // Rotates a by rot - - // Non-standard operators - friend F32 dot(const LLQuaternion &a, const LLQuaternion &b); - friend LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // linear interpolation (t = 0 to 1) from p to q - friend LLQuaternion lerp(F32 t, const LLQuaternion &q); // linear interpolation (t = 0 to 1) from identity to q - friend LLQuaternion slerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // spherical linear interpolation from p to q - friend LLQuaternion slerp(F32 t, const LLQuaternion &q); // spherical linear interpolation from identity to q - friend LLQuaternion nlerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // normalized linear interpolation from p to q - friend LLQuaternion nlerp(F32 t, const LLQuaternion &q); // normalized linear interpolation from p to q - - LLVector3 packToVector3() const; // Saves space by using the fact that our quaternions are normalized - void unpackFromVector3(const LLVector3& vec); // Saves space by using the fact that our quaternions are normalized - - enum Order { - XYZ = 0, - YZX = 1, - ZXY = 2, - XZY = 3, - YXZ = 4, - ZYX = 5 - }; - // Creates a quaternions from maya's rotation representation, - // which is 3 rotations (in DEGREES) in the specified order - friend LLQuaternion mayaQ(F32 x, F32 y, F32 z, Order order); - - // Conversions between Order and strings like "xyz" or "ZYX" - friend const char *OrderToString( const Order order ); - friend Order StringToOrder( const char *str ); - - static BOOL parseQuat(const std::string& buf, LLQuaternion* value); - - // For debugging, only - //static U32 mMultCount; -}; - -// checker -inline BOOL LLQuaternion::isFinite() const -{ - return (llfinite(mQ[VX]) && llfinite(mQ[VY]) && llfinite(mQ[VZ]) && llfinite(mQ[VS])); -} - -inline BOOL LLQuaternion::isIdentity() const -{ - return - ( mQ[VX] == 0.f ) && - ( mQ[VY] == 0.f ) && - ( mQ[VZ] == 0.f ) && - ( mQ[VS] == 1.f ); -} - -inline BOOL LLQuaternion::isNotIdentity() const -{ - return - ( mQ[VX] != 0.f ) || - ( mQ[VY] != 0.f ) || - ( mQ[VZ] != 0.f ) || - ( mQ[VS] != 1.f ); -} - - - -inline LLQuaternion::LLQuaternion(void) -{ - mQ[VX] = 0.f; - mQ[VY] = 0.f; - mQ[VZ] = 0.f; - mQ[VS] = 1.f; -} - -inline LLQuaternion::LLQuaternion(F32 x, F32 y, F32 z, F32 w) -{ - mQ[VX] = x; - mQ[VY] = y; - mQ[VZ] = z; - mQ[VS] = w; - - //RN: don't normalize this case as its used mainly for temporaries during calculations - //normalize(); - /* - F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]); - mag -= 1.f; - mag = fabs(mag); - llassert(mag < 10.f*FP_MAG_THRESHOLD); - */ -} - -inline LLQuaternion::LLQuaternion(const F32 *q) -{ - mQ[VX] = q[VX]; - mQ[VY] = q[VY]; - mQ[VZ] = q[VZ]; - mQ[VS] = q[VW]; - - normalize(); - /* - F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]); - mag -= 1.f; - mag = fabs(mag); - llassert(mag < FP_MAG_THRESHOLD); - */ -} - - -inline void LLQuaternion::loadIdentity() -{ - mQ[VX] = 0.0f; - mQ[VY] = 0.0f; - mQ[VZ] = 0.0f; - mQ[VW] = 1.0f; -} - - -inline const LLQuaternion& LLQuaternion::set(F32 x, F32 y, F32 z, F32 w) -{ - mQ[VX] = x; - mQ[VY] = y; - mQ[VZ] = z; - mQ[VS] = w; - normalize(); - return (*this); -} - -inline const LLQuaternion& LLQuaternion::set(const LLQuaternion &quat) -{ - mQ[VX] = quat.mQ[VX]; - mQ[VY] = quat.mQ[VY]; - mQ[VZ] = quat.mQ[VZ]; - mQ[VW] = quat.mQ[VW]; - normalize(); - return (*this); -} - -inline const LLQuaternion& LLQuaternion::set(const F32 *q) -{ - mQ[VX] = q[VX]; - mQ[VY] = q[VY]; - mQ[VZ] = q[VZ]; - mQ[VS] = q[VW]; - normalize(); - return (*this); -} - - -// deprecated -inline const LLQuaternion& LLQuaternion::setQuatInit(F32 x, F32 y, F32 z, F32 w) -{ - mQ[VX] = x; - mQ[VY] = y; - mQ[VZ] = z; - mQ[VS] = w; - normalize(); - return (*this); -} - -// deprecated -inline const LLQuaternion& LLQuaternion::setQuat(const LLQuaternion &quat) -{ - mQ[VX] = quat.mQ[VX]; - mQ[VY] = quat.mQ[VY]; - mQ[VZ] = quat.mQ[VZ]; - mQ[VW] = quat.mQ[VW]; - normalize(); - return (*this); -} - -// deprecated -inline const LLQuaternion& LLQuaternion::setQuat(const F32 *q) -{ - mQ[VX] = q[VX]; - mQ[VY] = q[VY]; - mQ[VZ] = q[VZ]; - mQ[VS] = q[VW]; - normalize(); - return (*this); -} - -// There may be a cheaper way that avoids the sqrt. -// Does sin_a = VX*VX + VY*VY + VZ*VZ? -// Copied from Matrix and Quaternion FAQ 1.12 -inline void LLQuaternion::getAngleAxis(F32* angle, F32* x, F32* y, F32* z) const -{ - F32 cos_a = mQ[VW]; - if (cos_a > 1.0f) cos_a = 1.0f; - if (cos_a < -1.0f) cos_a = -1.0f; - - F32 sin_a = (F32) sqrt( 1.0f - cos_a * cos_a ); - - if ( fabs( sin_a ) < 0.0005f ) - sin_a = 1.0f; - else - sin_a = 1.f/sin_a; - - F32 temp_angle = 2.0f * (F32) acos( cos_a ); - if (temp_angle > F_PI) - { - // The (angle,axis) pair should never have angles outside [PI, -PI] - // since we want the _shortest_ (angle,axis) solution. - // Since acos is defined for [0, PI], and we multiply by 2.0, we - // can push the angle outside the acceptible range. - // When this happens we set the angle to the other portion of a - // full 2PI rotation, and negate the axis, which reverses the - // direction of the rotation (by the right-hand rule). - *angle = 2.f * F_PI - temp_angle; - *x = - mQ[VX] * sin_a; - *y = - mQ[VY] * sin_a; - *z = - mQ[VZ] * sin_a; - } - else - { - *angle = temp_angle; - *x = mQ[VX] * sin_a; - *y = mQ[VY] * sin_a; - *z = mQ[VZ] * sin_a; - } -} - -inline const LLQuaternion& LLQuaternion::conjugate() -{ - mQ[VX] *= -1.f; - mQ[VY] *= -1.f; - mQ[VZ] *= -1.f; - return (*this); -} - -inline const LLQuaternion& LLQuaternion::conjQuat() -{ - mQ[VX] *= -1.f; - mQ[VY] *= -1.f; - mQ[VZ] *= -1.f; - return (*this); -} - -// Transpose -inline const LLQuaternion& LLQuaternion::transpose() -{ - mQ[VX] *= -1.f; - mQ[VY] *= -1.f; - mQ[VZ] *= -1.f; - return (*this); -} - -// deprecated -inline const LLQuaternion& LLQuaternion::transQuat() -{ - mQ[VX] *= -1.f; - mQ[VY] *= -1.f; - mQ[VZ] *= -1.f; - return (*this); -} - - -inline LLQuaternion operator+(const LLQuaternion &a, const LLQuaternion &b) -{ - return LLQuaternion( - a.mQ[VX] + b.mQ[VX], - a.mQ[VY] + b.mQ[VY], - a.mQ[VZ] + b.mQ[VZ], - a.mQ[VW] + b.mQ[VW] ); -} - - -inline LLQuaternion operator-(const LLQuaternion &a, const LLQuaternion &b) -{ - return LLQuaternion( - a.mQ[VX] - b.mQ[VX], - a.mQ[VY] - b.mQ[VY], - a.mQ[VZ] - b.mQ[VZ], - a.mQ[VW] - b.mQ[VW] ); -} - - -inline LLQuaternion operator-(const LLQuaternion &a) -{ - return LLQuaternion( - -a.mQ[VX], - -a.mQ[VY], - -a.mQ[VZ], - -a.mQ[VW] ); -} - - -inline LLQuaternion operator*(F32 a, const LLQuaternion &q) -{ - return LLQuaternion( - a * q.mQ[VX], - a * q.mQ[VY], - a * q.mQ[VZ], - a * q.mQ[VW] ); -} - - -inline LLQuaternion operator*(const LLQuaternion &q, F32 a) -{ - return LLQuaternion( - a * q.mQ[VX], - a * q.mQ[VY], - a * q.mQ[VZ], - a * q.mQ[VW] ); -} - -inline LLQuaternion operator~(const LLQuaternion &a) -{ - LLQuaternion q(a); - q.conjQuat(); - return q; -} - -inline bool LLQuaternion::operator==(const LLQuaternion &b) const -{ - return ( (mQ[VX] == b.mQ[VX]) - &&(mQ[VY] == b.mQ[VY]) - &&(mQ[VZ] == b.mQ[VZ]) - &&(mQ[VS] == b.mQ[VS])); -} - -inline bool LLQuaternion::operator!=(const LLQuaternion &b) const -{ - return ( (mQ[VX] != b.mQ[VX]) - ||(mQ[VY] != b.mQ[VY]) - ||(mQ[VZ] != b.mQ[VZ]) - ||(mQ[VS] != b.mQ[VS])); -} - -inline const LLQuaternion& operator*=(LLQuaternion &a, const LLQuaternion &b) -{ -#if 1 - LLQuaternion q( - b.mQ[3] * a.mQ[0] + b.mQ[0] * a.mQ[3] + b.mQ[1] * a.mQ[2] - b.mQ[2] * a.mQ[1], - b.mQ[3] * a.mQ[1] + b.mQ[1] * a.mQ[3] + b.mQ[2] * a.mQ[0] - b.mQ[0] * a.mQ[2], - b.mQ[3] * a.mQ[2] + b.mQ[2] * a.mQ[3] + b.mQ[0] * a.mQ[1] - b.mQ[1] * a.mQ[0], - b.mQ[3] * a.mQ[3] - b.mQ[0] * a.mQ[0] - b.mQ[1] * a.mQ[1] - b.mQ[2] * a.mQ[2] - ); - a = q; -#else - a = a * b; -#endif - return a; -} - -const F32 ONE_PART_IN_A_MILLION = 0.000001f; - -inline F32 LLQuaternion::normalize() -{ - F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]); - - if (mag > FP_MAG_THRESHOLD) - { - // Floating point error can prevent some quaternions from achieving - // exact unity length. When trying to renormalize such quaternions we - // can oscillate between multiple quantized states. To prevent such - // drifts we only renomalize if the length is far enough from unity. - if (fabs(1.f - mag) > ONE_PART_IN_A_MILLION) - { - F32 oomag = 1.f/mag; - mQ[VX] *= oomag; - mQ[VY] *= oomag; - mQ[VZ] *= oomag; - mQ[VS] *= oomag; - } - } - else - { - // we were given a very bad quaternion so we set it to identity - mQ[VX] = 0.f; - mQ[VY] = 0.f; - mQ[VZ] = 0.f; - mQ[VS] = 1.f; - } - - return mag; -} - -// deprecated -inline F32 LLQuaternion::normQuat() -{ - F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]); - - if (mag > FP_MAG_THRESHOLD) - { - if (fabs(1.f - mag) > ONE_PART_IN_A_MILLION) - { - // only renormalize if length not close enough to 1.0 already - F32 oomag = 1.f/mag; - mQ[VX] *= oomag; - mQ[VY] *= oomag; - mQ[VZ] *= oomag; - mQ[VS] *= oomag; - } - } - else - { - mQ[VX] = 0.f; - mQ[VY] = 0.f; - mQ[VZ] = 0.f; - mQ[VS] = 1.f; - } - - return mag; -} - -LLQuaternion::Order StringToOrder( const char *str ); - -// Some notes about Quaternions - -// What is a Quaternion? -// --------------------- -// A quaternion is a point in 4-dimensional complex space. -// Q = { Qx, Qy, Qz, Qw } -// -// -// Why Quaternions? -// ---------------- -// The set of quaternions that make up the the 4-D unit sphere -// can be mapped to the set of all rotations in 3-D space. Sometimes -// it is easier to describe/manipulate rotations in quaternion space -// than rotation-matrix space. -// -// -// How Quaternions? -// ---------------- -// In order to take advantage of quaternions we need to know how to -// go from rotation-matricies to quaternions and back. We also have -// to agree what variety of rotations we're generating. -// -// Consider the equation... v' = v * R -// -// There are two ways to think about rotations of vectors. -// 1) v' is the same vector in a different reference frame -// 2) v' is a new vector in the same reference frame -// -// bookmark -- which way are we using? -// -// -// Quaternion from Angle-Axis: -// --------------------------- -// Suppose we wanted to represent a rotation of some angle (theta) -// about some axis ({Ax, Ay, Az})... -// -// axis of rotation = {Ax, Ay, Az} -// angle_of_rotation = theta -// -// s = sin(0.5 * theta) -// c = cos(0.5 * theta) -// Q = { s * Ax, s * Ay, s * Az, c } -// -// -// 3x3 Matrix from Quaternion -// -------------------------- -// -// | | -// | 1 - 2 * (y^2 + z^2) 2 * (x * y + z * w) 2 * (y * w - x * z) | -// | | -// M = | 2 * (x * y - z * w) 1 - 2 * (x^2 + z^2) 2 * (y * z + x * w) | -// | | -// | 2 * (x * z + y * w) 2 * (y * z - x * w) 1 - 2 * (x^2 + y^2) | -// | | - -#endif +/**
+ * @file llquaternion.h
+ * @brief LLQuaternion class header file.
+ *
+ * $LicenseInfo:firstyear=2000&license=viewergpl$
+ *
+ * Copyright (c) 2000-2009, Linden Research, Inc.
+ *
+ * Second Life Viewer Source Code
+ * The source code in this file ("Source Code") is provided by Linden Lab
+ * to you under the terms of the GNU General Public License, version 2.0
+ * ("GPL"), unless you have obtained a separate licensing agreement
+ * ("Other License"), formally executed by you and Linden Lab. Terms of
+ * the GPL can be found in doc/GPL-license.txt in this distribution, or
+ * online at http://secondlifegrid.net/programs/open_source/licensing/gplv2
+ *
+ * There are special exceptions to the terms and conditions of the GPL as
+ * it is applied to this Source Code. View the full text of the exception
+ * in the file doc/FLOSS-exception.txt in this software distribution, or
+ * online at
+ * http://secondlifegrid.net/programs/open_source/licensing/flossexception
+ *
+ * By copying, modifying or distributing this software, you acknowledge
+ * that you have read and understood your obligations described above,
+ * and agree to abide by those obligations.
+ *
+ * ALL LINDEN LAB SOURCE CODE IS PROVIDED "AS IS." LINDEN LAB MAKES NO
+ * WARRANTIES, EXPRESS, IMPLIED OR OTHERWISE, REGARDING ITS ACCURACY,
+ * COMPLETENESS OR PERFORMANCE.
+ * $/LicenseInfo$
+ */
+
+#ifndef LLQUATERNION_H
+#define LLQUATERNION_H
+
+#include <iostream>
+
+#ifndef LLMATH_H //enforce specific include order to avoid tangling inline dependencies
+#error "Please include llmath.h first."
+#endif
+
+class LLVector4;
+class LLVector3;
+class LLVector3d;
+class LLMatrix4;
+class LLMatrix3;
+
+// NOTA BENE: Quaternion code is written assuming Unit Quaternions!!!!
+// Moreover, it is written assuming that all vectors and matricies
+// passed as arguments are normalized and unitary respectively.
+// VERY VERY VERY VERY BAD THINGS will happen if these assumptions fail.
+
+static const U32 LENGTHOFQUAT = 4;
+
+class LLQuaternion
+{
+public:
+ F32 mQ[LENGTHOFQUAT];
+
+ static const LLQuaternion DEFAULT;
+
+ LLQuaternion(); // Initializes Quaternion to (0,0,0,1)
+ explicit LLQuaternion(const LLMatrix4 &mat); // Initializes Quaternion from Matrix4
+ explicit LLQuaternion(const LLMatrix3 &mat); // Initializes Quaternion from Matrix3
+ LLQuaternion(F32 x, F32 y, F32 z, F32 w); // Initializes Quaternion to normalize(x, y, z, w)
+ LLQuaternion(F32 angle, const LLVector4 &vec); // Initializes Quaternion to axis_angle2quat(angle, vec)
+ LLQuaternion(F32 angle, const LLVector3 &vec); // Initializes Quaternion to axis_angle2quat(angle, vec)
+ LLQuaternion(const F32 *q); // Initializes Quaternion to normalize(x, y, z, w)
+ LLQuaternion(const LLVector3 &x_axis,
+ const LLVector3 &y_axis,
+ const LLVector3 &z_axis); // Initializes Quaternion from Matrix3 = [x_axis ; y_axis ; z_axis]
+
+ BOOL isIdentity() const;
+ BOOL isNotIdentity() const;
+ BOOL isFinite() const; // checks to see if all values of LLQuaternion are finite
+ void quantize16(F32 lower, F32 upper); // changes the vector to reflect quatization
+ void quantize8(F32 lower, F32 upper); // changes the vector to reflect quatization
+ void loadIdentity(); // Loads the quaternion that represents the identity rotation
+
+ const LLQuaternion& set(F32 x, F32 y, F32 z, F32 w); // Sets Quaternion to normalize(x, y, z, w)
+ const LLQuaternion& set(const LLQuaternion &quat); // Copies Quaternion
+ const LLQuaternion& set(const F32 *q); // Sets Quaternion to normalize(quat[VX], quat[VY], quat[VZ], quat[VW])
+ const LLQuaternion& set(const LLMatrix3 &mat); // Sets Quaternion to mat2quat(mat)
+ const LLQuaternion& set(const LLMatrix4 &mat); // Sets Quaternion to mat2quat(mat)
+
+ const LLQuaternion& setAngleAxis(F32 angle, F32 x, F32 y, F32 z); // Sets Quaternion to axis_angle2quat(angle, x, y, z)
+ const LLQuaternion& setAngleAxis(F32 angle, const LLVector3 &vec); // Sets Quaternion to axis_angle2quat(angle, vec)
+ const LLQuaternion& setAngleAxis(F32 angle, const LLVector4 &vec); // Sets Quaternion to axis_angle2quat(angle, vec)
+ const LLQuaternion& setEulerAngles(F32 roll, F32 pitch, F32 yaw); // Sets Quaternion to euler2quat(pitch, yaw, roll)
+
+ const LLQuaternion& setQuatInit(F32 x, F32 y, F32 z, F32 w); // deprecated
+ const LLQuaternion& setQuat(const LLQuaternion &quat); // deprecated
+ const LLQuaternion& setQuat(const F32 *q); // deprecated
+ const LLQuaternion& setQuat(const LLMatrix3 &mat); // deprecated
+ const LLQuaternion& setQuat(const LLMatrix4 &mat); // deprecated
+ const LLQuaternion& setQuat(F32 angle, F32 x, F32 y, F32 z); // deprecated
+ const LLQuaternion& setQuat(F32 angle, const LLVector3 &vec); // deprecated
+ const LLQuaternion& setQuat(F32 angle, const LLVector4 &vec); // deprecated
+ const LLQuaternion& setQuat(F32 roll, F32 pitch, F32 yaw); // deprecated
+
+ LLMatrix4 getMatrix4(void) const; // Returns the Matrix4 equivalent of Quaternion
+ LLMatrix3 getMatrix3(void) const; // Returns the Matrix3 equivalent of Quaternion
+ void getAngleAxis(F32* angle, F32* x, F32* y, F32* z) const; // returns rotation in radians about axis x,y,z
+ void getAngleAxis(F32* angle, LLVector3 &vec) const;
+ void getEulerAngles(F32 *roll, F32* pitch, F32 *yaw) const;
+
+ F32 normalize(); // Normalizes Quaternion and returns magnitude
+ F32 normQuat(); // deprecated
+
+ const LLQuaternion& conjugate(void); // Conjugates Quaternion and returns result
+ const LLQuaternion& conjQuat(void); // deprecated
+
+ // Other useful methods
+ const LLQuaternion& transpose(); // transpose (same as conjugate)
+ const LLQuaternion& transQuat(); // deprecated
+
+ void shortestArc(const LLVector3 &a, const LLVector3 &b); // shortest rotation from a to b
+ const LLQuaternion& constrain(F32 radians); // constrains rotation to a cone angle specified in radians
+
+ // Standard operators
+ friend std::ostream& operator<<(std::ostream &s, const LLQuaternion &a); // Prints a
+ friend LLQuaternion operator+(const LLQuaternion &a, const LLQuaternion &b); // Addition
+ friend LLQuaternion operator-(const LLQuaternion &a, const LLQuaternion &b); // Subtraction
+ friend LLQuaternion operator-(const LLQuaternion &a); // Negation
+ friend LLQuaternion operator*(F32 a, const LLQuaternion &q); // Scale
+ friend LLQuaternion operator*(const LLQuaternion &q, F32 b); // Scale
+ friend LLQuaternion operator*(const LLQuaternion &a, const LLQuaternion &b); // Returns a * b
+ friend LLQuaternion operator~(const LLQuaternion &a); // Returns a* (Conjugate of a)
+ bool operator==(const LLQuaternion &b) const; // Returns a == b
+ bool operator!=(const LLQuaternion &b) const; // Returns a != b
+
+ friend const LLQuaternion& operator*=(LLQuaternion &a, const LLQuaternion &b); // Returns a * b
+
+ friend LLVector4 operator*(const LLVector4 &a, const LLQuaternion &rot); // Rotates a by rot
+ friend LLVector3 operator*(const LLVector3 &a, const LLQuaternion &rot); // Rotates a by rot
+ friend LLVector3d operator*(const LLVector3d &a, const LLQuaternion &rot); // Rotates a by rot
+
+ // Non-standard operators
+ friend F32 dot(const LLQuaternion &a, const LLQuaternion &b);
+ friend LLQuaternion lerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // linear interpolation (t = 0 to 1) from p to q
+ friend LLQuaternion lerp(F32 t, const LLQuaternion &q); // linear interpolation (t = 0 to 1) from identity to q
+ friend LLQuaternion slerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // spherical linear interpolation from p to q
+ friend LLQuaternion slerp(F32 t, const LLQuaternion &q); // spherical linear interpolation from identity to q
+ friend LLQuaternion nlerp(F32 t, const LLQuaternion &p, const LLQuaternion &q); // normalized linear interpolation from p to q
+ friend LLQuaternion nlerp(F32 t, const LLQuaternion &q); // normalized linear interpolation from p to q
+
+ LLVector3 packToVector3() const; // Saves space by using the fact that our quaternions are normalized
+ void unpackFromVector3(const LLVector3& vec); // Saves space by using the fact that our quaternions are normalized
+
+ enum Order {
+ XYZ = 0,
+ YZX = 1,
+ ZXY = 2,
+ XZY = 3,
+ YXZ = 4,
+ ZYX = 5
+ };
+ // Creates a quaternions from maya's rotation representation,
+ // which is 3 rotations (in DEGREES) in the specified order
+ friend LLQuaternion mayaQ(F32 x, F32 y, F32 z, Order order);
+
+ // Conversions between Order and strings like "xyz" or "ZYX"
+ friend const char *OrderToString( const Order order );
+ friend Order StringToOrder( const char *str );
+
+ static BOOL parseQuat(const std::string& buf, LLQuaternion* value);
+
+ // For debugging, only
+ //static U32 mMultCount;
+};
+
+// checker
+inline BOOL LLQuaternion::isFinite() const
+{
+ return (llfinite(mQ[VX]) && llfinite(mQ[VY]) && llfinite(mQ[VZ]) && llfinite(mQ[VS]));
+}
+
+inline BOOL LLQuaternion::isIdentity() const
+{
+ return
+ ( mQ[VX] == 0.f ) &&
+ ( mQ[VY] == 0.f ) &&
+ ( mQ[VZ] == 0.f ) &&
+ ( mQ[VS] == 1.f );
+}
+
+inline BOOL LLQuaternion::isNotIdentity() const
+{
+ return
+ ( mQ[VX] != 0.f ) ||
+ ( mQ[VY] != 0.f ) ||
+ ( mQ[VZ] != 0.f ) ||
+ ( mQ[VS] != 1.f );
+}
+
+
+
+inline LLQuaternion::LLQuaternion(void)
+{
+ mQ[VX] = 0.f;
+ mQ[VY] = 0.f;
+ mQ[VZ] = 0.f;
+ mQ[VS] = 1.f;
+}
+
+inline LLQuaternion::LLQuaternion(F32 x, F32 y, F32 z, F32 w)
+{
+ mQ[VX] = x;
+ mQ[VY] = y;
+ mQ[VZ] = z;
+ mQ[VS] = w;
+
+ //RN: don't normalize this case as its used mainly for temporaries during calculations
+ //normalize();
+ /*
+ F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]);
+ mag -= 1.f;
+ mag = fabs(mag);
+ llassert(mag < 10.f*FP_MAG_THRESHOLD);
+ */
+}
+
+inline LLQuaternion::LLQuaternion(const F32 *q)
+{
+ mQ[VX] = q[VX];
+ mQ[VY] = q[VY];
+ mQ[VZ] = q[VZ];
+ mQ[VS] = q[VW];
+
+ normalize();
+ /*
+ F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]);
+ mag -= 1.f;
+ mag = fabs(mag);
+ llassert(mag < FP_MAG_THRESHOLD);
+ */
+}
+
+
+inline void LLQuaternion::loadIdentity()
+{
+ mQ[VX] = 0.0f;
+ mQ[VY] = 0.0f;
+ mQ[VZ] = 0.0f;
+ mQ[VW] = 1.0f;
+}
+
+
+inline const LLQuaternion& LLQuaternion::set(F32 x, F32 y, F32 z, F32 w)
+{
+ mQ[VX] = x;
+ mQ[VY] = y;
+ mQ[VZ] = z;
+ mQ[VS] = w;
+ normalize();
+ return (*this);
+}
+
+inline const LLQuaternion& LLQuaternion::set(const LLQuaternion &quat)
+{
+ mQ[VX] = quat.mQ[VX];
+ mQ[VY] = quat.mQ[VY];
+ mQ[VZ] = quat.mQ[VZ];
+ mQ[VW] = quat.mQ[VW];
+ normalize();
+ return (*this);
+}
+
+inline const LLQuaternion& LLQuaternion::set(const F32 *q)
+{
+ mQ[VX] = q[VX];
+ mQ[VY] = q[VY];
+ mQ[VZ] = q[VZ];
+ mQ[VS] = q[VW];
+ normalize();
+ return (*this);
+}
+
+
+// deprecated
+inline const LLQuaternion& LLQuaternion::setQuatInit(F32 x, F32 y, F32 z, F32 w)
+{
+ mQ[VX] = x;
+ mQ[VY] = y;
+ mQ[VZ] = z;
+ mQ[VS] = w;
+ normalize();
+ return (*this);
+}
+
+// deprecated
+inline const LLQuaternion& LLQuaternion::setQuat(const LLQuaternion &quat)
+{
+ mQ[VX] = quat.mQ[VX];
+ mQ[VY] = quat.mQ[VY];
+ mQ[VZ] = quat.mQ[VZ];
+ mQ[VW] = quat.mQ[VW];
+ normalize();
+ return (*this);
+}
+
+// deprecated
+inline const LLQuaternion& LLQuaternion::setQuat(const F32 *q)
+{
+ mQ[VX] = q[VX];
+ mQ[VY] = q[VY];
+ mQ[VZ] = q[VZ];
+ mQ[VS] = q[VW];
+ normalize();
+ return (*this);
+}
+
+// There may be a cheaper way that avoids the sqrt.
+// Does sin_a = VX*VX + VY*VY + VZ*VZ?
+// Copied from Matrix and Quaternion FAQ 1.12
+inline void LLQuaternion::getAngleAxis(F32* angle, F32* x, F32* y, F32* z) const
+{
+ F32 cos_a = mQ[VW];
+ if (cos_a > 1.0f) cos_a = 1.0f;
+ if (cos_a < -1.0f) cos_a = -1.0f;
+
+ F32 sin_a = (F32) sqrt( 1.0f - cos_a * cos_a );
+
+ if ( fabs( sin_a ) < 0.0005f )
+ sin_a = 1.0f;
+ else
+ sin_a = 1.f/sin_a;
+
+ F32 temp_angle = 2.0f * (F32) acos( cos_a );
+ if (temp_angle > F_PI)
+ {
+ // The (angle,axis) pair should never have angles outside [PI, -PI]
+ // since we want the _shortest_ (angle,axis) solution.
+ // Since acos is defined for [0, PI], and we multiply by 2.0, we
+ // can push the angle outside the acceptible range.
+ // When this happens we set the angle to the other portion of a
+ // full 2PI rotation, and negate the axis, which reverses the
+ // direction of the rotation (by the right-hand rule).
+ *angle = 2.f * F_PI - temp_angle;
+ *x = - mQ[VX] * sin_a;
+ *y = - mQ[VY] * sin_a;
+ *z = - mQ[VZ] * sin_a;
+ }
+ else
+ {
+ *angle = temp_angle;
+ *x = mQ[VX] * sin_a;
+ *y = mQ[VY] * sin_a;
+ *z = mQ[VZ] * sin_a;
+ }
+}
+
+inline const LLQuaternion& LLQuaternion::conjugate()
+{
+ mQ[VX] *= -1.f;
+ mQ[VY] *= -1.f;
+ mQ[VZ] *= -1.f;
+ return (*this);
+}
+
+inline const LLQuaternion& LLQuaternion::conjQuat()
+{
+ mQ[VX] *= -1.f;
+ mQ[VY] *= -1.f;
+ mQ[VZ] *= -1.f;
+ return (*this);
+}
+
+// Transpose
+inline const LLQuaternion& LLQuaternion::transpose()
+{
+ mQ[VX] *= -1.f;
+ mQ[VY] *= -1.f;
+ mQ[VZ] *= -1.f;
+ return (*this);
+}
+
+// deprecated
+inline const LLQuaternion& LLQuaternion::transQuat()
+{
+ mQ[VX] *= -1.f;
+ mQ[VY] *= -1.f;
+ mQ[VZ] *= -1.f;
+ return (*this);
+}
+
+
+inline LLQuaternion operator+(const LLQuaternion &a, const LLQuaternion &b)
+{
+ return LLQuaternion(
+ a.mQ[VX] + b.mQ[VX],
+ a.mQ[VY] + b.mQ[VY],
+ a.mQ[VZ] + b.mQ[VZ],
+ a.mQ[VW] + b.mQ[VW] );
+}
+
+
+inline LLQuaternion operator-(const LLQuaternion &a, const LLQuaternion &b)
+{
+ return LLQuaternion(
+ a.mQ[VX] - b.mQ[VX],
+ a.mQ[VY] - b.mQ[VY],
+ a.mQ[VZ] - b.mQ[VZ],
+ a.mQ[VW] - b.mQ[VW] );
+}
+
+
+inline LLQuaternion operator-(const LLQuaternion &a)
+{
+ return LLQuaternion(
+ -a.mQ[VX],
+ -a.mQ[VY],
+ -a.mQ[VZ],
+ -a.mQ[VW] );
+}
+
+
+inline LLQuaternion operator*(F32 a, const LLQuaternion &q)
+{
+ return LLQuaternion(
+ a * q.mQ[VX],
+ a * q.mQ[VY],
+ a * q.mQ[VZ],
+ a * q.mQ[VW] );
+}
+
+
+inline LLQuaternion operator*(const LLQuaternion &q, F32 a)
+{
+ return LLQuaternion(
+ a * q.mQ[VX],
+ a * q.mQ[VY],
+ a * q.mQ[VZ],
+ a * q.mQ[VW] );
+}
+
+inline LLQuaternion operator~(const LLQuaternion &a)
+{
+ LLQuaternion q(a);
+ q.conjQuat();
+ return q;
+}
+
+inline bool LLQuaternion::operator==(const LLQuaternion &b) const
+{
+ return ( (mQ[VX] == b.mQ[VX])
+ &&(mQ[VY] == b.mQ[VY])
+ &&(mQ[VZ] == b.mQ[VZ])
+ &&(mQ[VS] == b.mQ[VS]));
+}
+
+inline bool LLQuaternion::operator!=(const LLQuaternion &b) const
+{
+ return ( (mQ[VX] != b.mQ[VX])
+ ||(mQ[VY] != b.mQ[VY])
+ ||(mQ[VZ] != b.mQ[VZ])
+ ||(mQ[VS] != b.mQ[VS]));
+}
+
+inline const LLQuaternion& operator*=(LLQuaternion &a, const LLQuaternion &b)
+{
+#if 1
+ LLQuaternion q(
+ b.mQ[3] * a.mQ[0] + b.mQ[0] * a.mQ[3] + b.mQ[1] * a.mQ[2] - b.mQ[2] * a.mQ[1],
+ b.mQ[3] * a.mQ[1] + b.mQ[1] * a.mQ[3] + b.mQ[2] * a.mQ[0] - b.mQ[0] * a.mQ[2],
+ b.mQ[3] * a.mQ[2] + b.mQ[2] * a.mQ[3] + b.mQ[0] * a.mQ[1] - b.mQ[1] * a.mQ[0],
+ b.mQ[3] * a.mQ[3] - b.mQ[0] * a.mQ[0] - b.mQ[1] * a.mQ[1] - b.mQ[2] * a.mQ[2]
+ );
+ a = q;
+#else
+ a = a * b;
+#endif
+ return a;
+}
+
+const F32 ONE_PART_IN_A_MILLION = 0.000001f;
+
+inline F32 LLQuaternion::normalize()
+{
+ F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]);
+
+ if (mag > FP_MAG_THRESHOLD)
+ {
+ // Floating point error can prevent some quaternions from achieving
+ // exact unity length. When trying to renormalize such quaternions we
+ // can oscillate between multiple quantized states. To prevent such
+ // drifts we only renomalize if the length is far enough from unity.
+ if (fabs(1.f - mag) > ONE_PART_IN_A_MILLION)
+ {
+ F32 oomag = 1.f/mag;
+ mQ[VX] *= oomag;
+ mQ[VY] *= oomag;
+ mQ[VZ] *= oomag;
+ mQ[VS] *= oomag;
+ }
+ }
+ else
+ {
+ // we were given a very bad quaternion so we set it to identity
+ mQ[VX] = 0.f;
+ mQ[VY] = 0.f;
+ mQ[VZ] = 0.f;
+ mQ[VS] = 1.f;
+ }
+
+ return mag;
+}
+
+// deprecated
+inline F32 LLQuaternion::normQuat()
+{
+ F32 mag = sqrtf(mQ[VX]*mQ[VX] + mQ[VY]*mQ[VY] + mQ[VZ]*mQ[VZ] + mQ[VS]*mQ[VS]);
+
+ if (mag > FP_MAG_THRESHOLD)
+ {
+ if (fabs(1.f - mag) > ONE_PART_IN_A_MILLION)
+ {
+ // only renormalize if length not close enough to 1.0 already
+ F32 oomag = 1.f/mag;
+ mQ[VX] *= oomag;
+ mQ[VY] *= oomag;
+ mQ[VZ] *= oomag;
+ mQ[VS] *= oomag;
+ }
+ }
+ else
+ {
+ mQ[VX] = 0.f;
+ mQ[VY] = 0.f;
+ mQ[VZ] = 0.f;
+ mQ[VS] = 1.f;
+ }
+
+ return mag;
+}
+
+LLQuaternion::Order StringToOrder( const char *str );
+
+// Some notes about Quaternions
+
+// What is a Quaternion?
+// ---------------------
+// A quaternion is a point in 4-dimensional complex space.
+// Q = { Qx, Qy, Qz, Qw }
+//
+//
+// Why Quaternions?
+// ----------------
+// The set of quaternions that make up the the 4-D unit sphere
+// can be mapped to the set of all rotations in 3-D space. Sometimes
+// it is easier to describe/manipulate rotations in quaternion space
+// than rotation-matrix space.
+//
+//
+// How Quaternions?
+// ----------------
+// In order to take advantage of quaternions we need to know how to
+// go from rotation-matricies to quaternions and back. We also have
+// to agree what variety of rotations we're generating.
+//
+// Consider the equation... v' = v * R
+//
+// There are two ways to think about rotations of vectors.
+// 1) v' is the same vector in a different reference frame
+// 2) v' is a new vector in the same reference frame
+//
+// bookmark -- which way are we using?
+//
+//
+// Quaternion from Angle-Axis:
+// ---------------------------
+// Suppose we wanted to represent a rotation of some angle (theta)
+// about some axis ({Ax, Ay, Az})...
+//
+// axis of rotation = {Ax, Ay, Az}
+// angle_of_rotation = theta
+//
+// s = sin(0.5 * theta)
+// c = cos(0.5 * theta)
+// Q = { s * Ax, s * Ay, s * Az, c }
+//
+//
+// 3x3 Matrix from Quaternion
+// --------------------------
+//
+// | |
+// | 1 - 2 * (y^2 + z^2) 2 * (x * y + z * w) 2 * (y * w - x * z) |
+// | |
+// M = | 2 * (x * y - z * w) 1 - 2 * (x^2 + z^2) 2 * (y * z + x * w) |
+// | |
+// | 2 * (x * z + y * w) 2 * (y * z - x * w) 1 - 2 * (x^2 + y^2) |
+// | |
+
+#endif
diff --git a/indra/llmath/llvolume.cpp b/indra/llmath/llvolume.cpp index bba0a6d089..ab9f8c4c24 100644 --- a/indra/llmath/llvolume.cpp +++ b/indra/llmath/llvolume.cpp @@ -45,7 +45,7 @@ #include "v4math.h" #include "m4math.h" #include "m3math.h" -#include "llmatrix4a.h" +#include "llmatrix3a.h" #include "lloctree.h" #include "lldarray.h" #include "llvolume.h" @@ -53,6 +53,7 @@ #include "llstl.h" #include "llsdserialize.h" #include "llvector4a.h" +#include "llmatrix4a.h" #define DEBUG_SILHOUETTE_BINORMALS 0 #define DEBUG_SILHOUETTE_NORMALS 0 // TomY: Use this to display normals using the silhouette @@ -161,7 +162,7 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a det; det.setAllDot3(edge1, pvec); - if (det.greaterEqual4(LLVector4a::getApproximatelyZero()).getComparisonMask() & 0x7) + if (det.greaterEqual(LLVector4a::getEpsilon()).getGatheredBits() & 0x7) { /* calculate distance from vert0 to ray origin */ LLVector4a tvec; @@ -171,8 +172,8 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a u; u.setAllDot3(tvec,pvec); - if ((u.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7) && - (u.lessEqual4(det).getComparisonMask() & 0x7)) + if ((u.greaterEqual(LLVector4a::getZero()).getGatheredBits() & 0x7) && + (u.lessEqual(det).getGatheredBits() & 0x7)) { /* prepare to test V parameter */ LLVector4a qvec; @@ -188,8 +189,8 @@ BOOL LLTriangleRayIntersect(const LLVector4a& vert0, const LLVector4a& vert1, co LLVector4a sum_uv; sum_uv.setAdd(u, v); - S32 v_gequal = v.greaterEqual4(LLVector4a::getZero()).getComparisonMask() & 0x7; - S32 sum_lequal = sum_uv.lessEqual4(det).getComparisonMask() & 0x7; + S32 v_gequal = v.greaterEqual(LLVector4a::getZero()).getGatheredBits() & 0x7; + S32 sum_lequal = sum_uv.lessEqual(det).getGatheredBits() & 0x7; if (v_gequal && sum_lequal) { @@ -230,7 +231,7 @@ BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& v pvec.setCross3(dir, edge2); /* if determinant is near zero, ray lies in plane of triangle */ - F32 det = edge1.dot3(pvec); + F32 det = edge1.dot3(pvec).getF32(); if (det > -F_APPROXIMATELY_ZERO && det < F_APPROXIMATELY_ZERO) @@ -245,7 +246,7 @@ BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& v tvec.setSub(orig, vert0); /* calculate U parameter and test bounds */ - u = (tvec.dot3(pvec)) * inv_det; + u = (tvec.dot3(pvec).getF32()) * inv_det; if (u < 0.f || u > 1.f) { return FALSE; @@ -255,7 +256,7 @@ BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& v tvec.sub(edge1); /* calculate V parameter and test bounds */ - v = (dir.dot3(tvec)) * inv_det; + v = (dir.dot3(tvec).getF32()) * inv_det; if (v < 0.f || u + v > 1.f) { @@ -263,7 +264,7 @@ BOOL LLTriangleRayIntersectTwoSided(const LLVector4a& vert0, const LLVector4a& v } /* calculate t, ray intersects triangle */ - t = (edge2.dot3(tvec)) * inv_det; + t = (edge2.dot3(tvec).getF32()) * inv_det; intersection_a = u; intersection_b = v; @@ -326,20 +327,20 @@ public: //stretch by triangles in node tri = *iter; - min.setMin(*tri->mV[0]); - min.setMin(*tri->mV[1]); - min.setMin(*tri->mV[2]); + min.setMin(min, *tri->mV[0]); + min.setMin(min, *tri->mV[1]); + min.setMin(min, *tri->mV[2]); - max.setMax(*tri->mV[0]); - max.setMax(*tri->mV[1]); - max.setMax(*tri->mV[2]); + max.setMax(max, *tri->mV[0]); + max.setMax(max, *tri->mV[1]); + max.setMax(max, *tri->mV[2]); } for (S32 i = 0; i < branch->getChildCount(); ++i) { //stretch by child extents LLVolumeOctreeListener* child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); - min.setMin(child->mExtents[0]); - max.setMax(child->mExtents[1]); + min.setMin(min, child->mExtents[0]); + max.setMax(min, child->mExtents[1]); } } else if (branch->getChildCount() != 0) @@ -352,8 +353,8 @@ public: for (S32 i = 1; i < branch->getChildCount(); ++i) { //stretch by child extents child = (LLVolumeOctreeListener*) branch->getChild(i)->getListener(0); - min.setMin(child->mExtents[0]); - max.setMax(child->mExtents[1]); + min.setMin(min, child->mExtents[0]); + max.setMax(max, child->mExtents[1]); } } else @@ -2011,7 +2012,7 @@ const LLVolumeFace::VertexData& LLVolumeFace::VertexData::operator=(const LLVolu if (this != &rhs) { init(); - LLVector4a::memcpyNonAliased16((F32*) mData, (F32*) rhs.mData, 8); + LLVector4a::memcpyNonAliased16((F32*) mData, (F32*) rhs.mData, 8*sizeof(F32)); mTexCoord = rhs.mTexCoord; } return *this; @@ -2055,8 +2056,8 @@ void LLVolumeFace::VertexData::setNormal(const LLVector4a& norm) bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)const { - const F32* lp = this->getPosition().getF32(); - const F32* rp = rhs.getPosition().getF32(); + const F32* lp = this->getPosition().getF32ptr(); + const F32* rp = rhs.getPosition().getF32ptr(); if (lp[0] != rp[0]) { @@ -2073,8 +2074,8 @@ bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)con return lp[2] < rp[2]; } - lp = getNormal().getF32(); - rp = rhs.getNormal().getF32(); + lp = getNormal().getF32ptr(); + rp = rhs.getNormal().getF32ptr(); if (lp[0] != rp[0]) { @@ -2101,23 +2102,23 @@ bool LLVolumeFace::VertexData::operator<(const LLVolumeFace::VertexData& rhs)con bool LLVolumeFace::VertexData::operator==(const LLVolumeFace::VertexData& rhs)const { - return mData[POSITION].equal3(rhs.getPosition()) && - mData[NORMAL].equal3(rhs.getNormal()) && + return mData[POSITION].equals3(rhs.getPosition()) && + mData[NORMAL].equals3(rhs.getNormal()) && mTexCoord == rhs.mTexCoord; } bool LLVolumeFace::VertexData::compareNormal(const LLVolumeFace::VertexData& rhs, F32 angle_cutoff) const { bool retval = false; - if (rhs.mData[POSITION].equal3(mData[POSITION]) && rhs.mTexCoord == mTexCoord) + if (rhs.mData[POSITION].equals3(mData[POSITION]) && rhs.mTexCoord == mTexCoord) { if (angle_cutoff > 1.f) { - retval = (mData[NORMAL].equal3(rhs.mData[NORMAL])); + retval = (mData[NORMAL].equals3(rhs.mData[NORMAL])); } else { - F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]); + F32 cur_angle = rhs.mData[NORMAL].dot3(mData[NORMAL]).getF32(); retval = cur_angle > angle_cutoff; } } @@ -2331,8 +2332,8 @@ bool LLVolume::unpackVolumeFaces(std::istream& is, S32 size) } else { - min.setMin(*pos_out); - max.setMax(*pos_out); + min.setMin(min, *pos_out); + max.setMax(max, *pos_out); } pos_out++; @@ -2944,7 +2945,7 @@ void sculpt_calc_mesh_resolution(U16 width, U16 height, U8 type, F32 detail, S32 ratio = (F32) width / (F32) height; - s = (S32)fsqrtf(((F32)vertices / ratio)); + s = (S32)(F32) sqrt(((F32)vertices / ratio)); s = llmax(s, 4); // no degenerate sizes, please t = vertices / s; @@ -5280,16 +5281,15 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) freeData(); - LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 12); + LLVector4a::memcpyNonAliased16((F32*) mExtents, (F32*) src.mExtents, 12*sizeof(F32)); resizeVertices(src.mNumVertices); resizeIndices(src.mNumIndices); if (mNumVertices) { - S32 vert_size = mNumVertices*4; + S32 vert_size = mNumVertices*4*sizeof(F32); S32 tc_size = (mNumVertices*8+0xF) & ~0xF; - tc_size /= 4; LLVector4a::memcpyNonAliased16((F32*) mPositions, (F32*) src.mPositions, vert_size); LLVector4a::memcpyNonAliased16((F32*) mNormals, (F32*) src.mNormals, vert_size); @@ -5322,8 +5322,7 @@ LLVolumeFace& LLVolumeFace::operator=(const LLVolumeFace& src) if (mNumIndices) { S32 idx_size = (mNumIndices*2+0xF) & ~0xF; - idx_size /= 4; - + LLVector4a::memcpyNonAliased16((F32*) mIndices, (F32*) src.mIndices, idx_size); } @@ -5388,9 +5387,9 @@ void LLVolumeFace::getVertexData(U16 index, LLVolumeFace::VertexData& cv) bool LLVolumeFace::VertexMapData::operator==(const LLVolumeFace::VertexData& rhs) const { - return getPosition().equal3(rhs.getPosition()) && + return getPosition().equals3(rhs.getPosition()) && mTexCoord == rhs.mTexCoord && - getNormal().equal3(rhs.getNormal()); + getNormal().equals3(rhs.getNormal()); } bool LLVolumeFace::VertexMapData::ComparePosition::operator()(const LLVector3& a, const LLVector3& b) const @@ -5423,7 +5422,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) getVertexData(index, cv); BOOL found = FALSE; - VertexMapData::PointMap::iterator point_iter = point_map.find(LLVector3(cv.getPosition().getF32())); + VertexMapData::PointMap::iterator point_iter = point_map.find(LLVector3(cv.getPosition().getF32ptr())); if (point_iter != point_map.end()) { //duplicate point might exist for (U32 j = 0; j < point_iter->second.size(); ++j) @@ -5455,7 +5454,7 @@ void LLVolumeFace::optimize(F32 angle_cutoff) } else { - point_map[LLVector3(d.getPosition().getF32())].push_back(d); + point_map[LLVector3(d.getPosition().getF32ptr())].push_back(d); } } } @@ -5491,12 +5490,12 @@ void LLVolumeFace::createOctree() tri->mIndex[2] = mIndices[i+2]; LLVector4a min = v0; - min.setMin(v1); - min.setMin(v2); + min.setMin(min, v1); + min.setMin(min, v2); LLVector4a max = v0; - max.setMax(v1); - max.setMax(v2); + max.setMax(max, v1); + max.setMax(max, v2); LLVector4a center; center.setAdd(min, max); @@ -5507,7 +5506,7 @@ void LLVolumeFace::createOctree() LLVector4a size; size.setSub(max,min); - tri->mRadius = size.length3() * 0.5f; + tri->mRadius = size.getLength3().getF32() * 0.5f; mOctree->insert(tri); } @@ -5655,12 +5654,13 @@ BOOL LLVolumeFace::createUnCutCubeCap(LLVolume* volume, BOOL partial_build) if (gx == 0 && gy == 0) { - min = max = newVert.getPosition(); + min = newVert.getPosition(); + max = min; } else { - min.setMin(newVert.getPosition()); - max.setMax(newVert.getPosition()); + min.setMin(min, newVert.getPosition()); + max.setMax(max, newVert.getPosition()); } } } @@ -5795,7 +5795,8 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) if (i == 0) { - min = max = pos[i]; + max = pos[i]; + min = max; min_uv = max_uv = tc[i]; } else @@ -5848,8 +5849,8 @@ BOOL LLVolumeFace::createCap(LLVolume* volume, BOOL partial_build) for (S32 i = 0; i < num_vertices; i++) { - binorm[i].load4a((F32*) &binormal.mQ); - norm[i].load4a((F32*) &normal.mQ); + binorm[i].load4a(binormal.getF32ptr()); + norm[i].load4a(normal.getF32ptr()); } if (partial_build) @@ -6186,7 +6187,7 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con LLVector4a* dst = (LLVector4a*) ll_aligned_malloc_16(new_size); if (mPositions) { - LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mPositions, old_size/4); + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mPositions, old_size); ll_aligned_free_16(mPositions); } mPositions = dst; @@ -6195,7 +6196,7 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con dst = (LLVector4a*) ll_aligned_malloc_16(new_size); if (mNormals) { - LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mNormals, old_size/4); + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mNormals, old_size); ll_aligned_free_16(mNormals); } mNormals = dst; @@ -6209,7 +6210,7 @@ void LLVolumeFace::pushVertex(const LLVector4a& pos, const LLVector4a& norm, con LLVector2* dst = (LLVector2*) ll_aligned_malloc_16(new_size); if (mTexCoords) { - LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mTexCoords, old_size/4); + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mTexCoords, old_size); ll_aligned_free_16(mTexCoords); } } @@ -6268,7 +6269,7 @@ void LLVolumeFace::pushIndex(const U16& idx) U16* dst = (U16*) ll_aligned_malloc_16(new_size); if (mIndices) { - LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mIndices, old_size/4); + LLVector4a::memcpyNonAliased16((F32*) dst, (F32*) mIndices, old_size); ll_aligned_free_16(mIndices); } mIndices = dst; @@ -6319,9 +6320,9 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat if (mNumVertices > 0) { //copy old buffers - LLVector4a::memcpyNonAliased16((F32*) new_pos, (F32*) mPositions, mNumVertices*4); - LLVector4a::memcpyNonAliased16((F32*) new_norm, (F32*) mNormals, mNumVertices*4); - LLVector4a::memcpyNonAliased16((F32*) new_tc, (F32*) mTexCoords, mNumVertices*2); + LLVector4a::memcpyNonAliased16((F32*) new_pos, (F32*) mPositions, mNumVertices*4*sizeof(F32)); + LLVector4a::memcpyNonAliased16((F32*) new_norm, (F32*) mNormals, mNumVertices*4*sizeof(F32)); + LLVector4a::memcpyNonAliased16((F32*) new_tc, (F32*) mTexCoords, mNumVertices*2*sizeof(F32)); } //free old buffer space @@ -6382,7 +6383,7 @@ void LLVolumeFace::appendFace(const LLVolumeFace& face, LLMatrix4& mat_in, LLMat if (mNumIndices > 0) { //copy old index buffer S32 old_size = (mNumIndices*2+0xF) & ~0xF; - LLVector4a::memcpyNonAliased16((F32*) new_indices, (F32*) mIndices, old_size/4); + LLVector4a::memcpyNonAliased16((F32*) new_indices, (F32*) mIndices, old_size); } //free old index buffer diff --git a/indra/llmath/tests/v2math_test.cpp b/indra/llmath/tests/v2math_test.cpp index 4660fcb955..c745b9989e 100644 --- a/indra/llmath/tests/v2math_test.cpp +++ b/indra/llmath/tests/v2math_test.cpp @@ -91,7 +91,7 @@ namespace tut F32 x = 2.2345f, y = 3.5678f ; LLVector2 vec2(x,y); ensure("magVecSquared:Fail ", is_approx_equal(vec2.magVecSquared(), (x*x + y*y))); - ensure("magVec:Fail ", is_approx_equal(vec2.magVec(), fsqrtf(x*x + y*y))); + ensure("magVec:Fail ", is_approx_equal(vec2.magVec(), (F32) sqrt(x*x + y*y))); } template<> template<> @@ -413,7 +413,7 @@ namespace tut ensure_equals("dist_vec_squared values are not equal",val2, val1); val1 = dist_vec(vec2, vec3); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2)); ensure_equals("dist_vec values are not equal",val2, val1); } @@ -437,7 +437,7 @@ namespace tut LLVector2 vec2(x1, y1); F32 vecMag = vec2.normVec(); - F32 mag = fsqrtf(x1*x1 + y1*y1); + F32 mag = (F32) sqrt(x1*x1 + y1*y1); F32 oomag = 1.f / mag; val1 = x1 * oomag; diff --git a/indra/llmath/tests/v3color_test.cpp b/indra/llmath/tests/v3color_test.cpp index 316b6e392f..0efba8e9f3 100644 --- a/indra/llmath/tests/v3color_test.cpp +++ b/indra/llmath/tests/v3color_test.cpp @@ -99,7 +99,7 @@ namespace tut F32 r = 2.3436212f, g = 1231.f, b = 4.7849321232f; LLColor3 llcolor3(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor3.magVecSquared(), (r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor3.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor3.magVec(), (F32) sqrt(r*r + g*g + b*b))); } template<> template<> @@ -109,7 +109,7 @@ namespace tut F32 val1, val2,val3; LLColor3 llcolor3(r,g,b); F32 vecMag = llcolor3.normVec(); - F32 mag = fsqrtf(r*r + g*g + b*b); + F32 mag = (F32) sqrt(r*r + g*g + b*b); F32 oomag = 1.f / mag; val1 = r * oomag; val2 = g * oomag; @@ -292,7 +292,7 @@ namespace tut F32 r1 =1.f, g1 = 2.f,b1 = 1.2f, r2 = -2.3f, g2 = 1.11f, b2 = 1234.234f; LLColor3 llcolor3(r1,g1,b1),llcolor3a(r2,g2,b2); F32 val = distVec(llcolor3,llcolor3a); - ensure("distVec failed ", is_approx_equal(fsqrtf((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val)); + ensure("distVec failed ", is_approx_equal((F32) sqrt((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val)); F32 val1 = distVec_squared(llcolor3,llcolor3a); ensure("distVec_squared failed ", is_approx_equal(((r1-r2)*(r1-r2) + (g1-g2)*(g1-g2) + (b1-b2)*(b1-b2)) ,val1)); diff --git a/indra/llmath/tests/v3dmath_test.cpp b/indra/llmath/tests/v3dmath_test.cpp index e7c949186c..894b6200f5 100644 --- a/indra/llmath/tests/v3dmath_test.cpp +++ b/indra/llmath/tests/v3dmath_test.cpp @@ -409,7 +409,7 @@ namespace tut LLVector3d vec3D(x,y,z); F64 res = (x*x + y*y + z*z) - vec3D.magVecSquared(); ensure("1:magVecSquared:Fail ", ((-F_APPROXIMATELY_ZERO <= res)&& (res <=F_APPROXIMATELY_ZERO))); - res = fsqrtf(x*x + y*y + z*z) - vec3D.magVec(); + res = (F32) sqrt(x*x + y*y + z*z) - vec3D.magVec(); ensure("2:magVec: Fail ", ((-F_APPROXIMATELY_ZERO <= res)&& (res <=F_APPROXIMATELY_ZERO))); } diff --git a/indra/llmath/tests/v3math_test.cpp b/indra/llmath/tests/v3math_test.cpp index 7faf076243..d5c8dd2f9c 100644 --- a/indra/llmath/tests/v3math_test.cpp +++ b/indra/llmath/tests/v3math_test.cpp @@ -155,7 +155,7 @@ namespace tut F32 x = 2.32f, y = 1.212f, z = -.12f; LLVector3 vec3(x,y,z); ensure("1:magVecSquared:Fail ", is_approx_equal(vec3.magVecSquared(), (x*x + y*y + z*z))); - ensure("2:magVec:Fail ", is_approx_equal(vec3.magVec(), fsqrtf(x*x + y*y + z*z))); + ensure("2:magVec:Fail ", is_approx_equal(vec3.magVec(), (F32) sqrt(x*x + y*y + z*z))); } template<> template<> @@ -515,7 +515,7 @@ namespace tut F32 val1,val2; LLVector3 vec3(x1,y1,z1),vec3a(x2,y2,z2); val1 = dist_vec(vec3,vec3a); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); ensure_equals("1:dist_vec: Fail ",val2, val1); val1 = dist_vec_squared(vec3,vec3a); val2 =((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); diff --git a/indra/llmath/tests/v4color_test.cpp b/indra/llmath/tests/v4color_test.cpp index 33921e0f0f..636446027a 100644 --- a/indra/llmath/tests/v4color_test.cpp +++ b/indra/llmath/tests/v4color_test.cpp @@ -161,7 +161,7 @@ namespace tut F32 r = 0x20, g = 0xFFFF, b = 0xFF; LLColor4 llcolor4(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor4.magVecSquared(), (r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor4.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor4.magVec(), (F32) sqrt(r*r + g*g + b*b))); } template<> template<> @@ -170,7 +170,7 @@ namespace tut F32 r = 0x20, g = 0xFFFF, b = 0xFF; LLColor4 llcolor4(r,g,b); F32 vecMag = llcolor4.normVec(); - F32 mag = fsqrtf(r*r + g*g + b*b); + F32 mag = (F32) sqrt(r*r + g*g + b*b); F32 oomag = 1.f / mag; F32 val1 = r * oomag, val2 = g * oomag, val3 = b * oomag; ensure("1:normVec failed ", (is_approx_equal(val1, llcolor4.mV[0]) && is_approx_equal(val2, llcolor4.mV[1]) && is_approx_equal(val3, llcolor4.mV[2]) && is_approx_equal(vecMag, mag))); diff --git a/indra/llmath/tests/v4coloru_test.cpp b/indra/llmath/tests/v4coloru_test.cpp index 9f71cfc8cc..b3dbfece34 100644 --- a/indra/llmath/tests/v4coloru_test.cpp +++ b/indra/llmath/tests/v4coloru_test.cpp @@ -141,7 +141,7 @@ namespace tut U8 r = 0x12, g = 0xFF, b = 0xAF; LLColor4U llcolor4u(r,g,b); ensure("magVecSquared:Fail ", is_approx_equal(llcolor4u.magVecSquared(), (F32)(r*r + g*g + b*b))); - ensure("magVec:Fail ", is_approx_equal(llcolor4u.magVec(), fsqrtf(r*r + g*g + b*b))); + ensure("magVec:Fail ", is_approx_equal(llcolor4u.magVec(), (F32) sqrt((F32) (r*r + g*g + b*b)))); } template<> template<> diff --git a/indra/llmath/tests/v4math_test.cpp b/indra/llmath/tests/v4math_test.cpp index fe051c27e9..e919c90efa 100644 --- a/indra/llmath/tests/v4math_test.cpp +++ b/indra/llmath/tests/v4math_test.cpp @@ -102,7 +102,7 @@ namespace tut { F32 x = 10.f, y = -2.3f, z = -.023f; LLVector4 vec4(x,y,z); - ensure("magVec:Fail ", is_approx_equal(vec4.magVec(), fsqrtf(x*x + y*y + z*z))); + ensure("magVec:Fail ", is_approx_equal(vec4.magVec(), (F32) sqrt(x*x + y*y + z*z))); ensure("magVecSquared:Fail ", is_approx_equal(vec4.magVecSquared(), (x*x + y*y + z*z))); } @@ -343,7 +343,7 @@ namespace tut F32 val1,val2; LLVector4 vec4(x1,y1,z1),vec4a(x2,y2,z2); val1 = dist_vec(vec4,vec4a); - val2 = fsqrtf((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); + val2 = (F32) sqrt((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); ensure_equals("dist_vec: Fail ",val2, val1); val1 = dist_vec_squared(vec4,vec4a); val2 =((x1 - x2)*(x1 - x2) + (y1 - y2)* (y1 - y2) + (z1 - z2)* (z1 -z2)); diff --git a/indra/llmath/v2math.cpp b/indra/llmath/v2math.cpp index 220336e0c2..2603127f75 100644 --- a/indra/llmath/v2math.cpp +++ b/indra/llmath/v2math.cpp @@ -92,7 +92,7 @@ F32 dist_vec(const LLVector2 &a, const LLVector2 &b) { F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; - return fsqrtf( x*x + y*y ); + return (F32) sqrt( x*x + y*y ); } F32 dist_vec_squared(const LLVector2 &a, const LLVector2 &b) diff --git a/indra/llmath/v2math.h b/indra/llmath/v2math.h index ae26c85ce4..35fd1b6048 100644 --- a/indra/llmath/v2math.h +++ b/indra/llmath/v2math.h @@ -225,7 +225,7 @@ inline void LLVector2::setVec(const F32 *vec) inline F32 LLVector2::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); } inline F32 LLVector2::lengthSquared(void) const @@ -235,7 +235,7 @@ inline F32 LLVector2::lengthSquared(void) const inline F32 LLVector2::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -262,7 +262,7 @@ inline bool LLVector2::isFinite() const // deprecated inline F32 LLVector2::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); } // deprecated @@ -274,7 +274,7 @@ inline F32 LLVector2::magVecSquared(void) const // deprecated inline F32 LLVector2::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1]); F32 oomag; if (mag > FP_MAG_THRESHOLD) diff --git a/indra/llmath/v3color.h b/indra/llmath/v3color.h index 1915d80502..95a3de8b62 100644 --- a/indra/llmath/v3color.h +++ b/indra/llmath/v3color.h @@ -284,7 +284,7 @@ inline F32 LLColor3::brightness(void) const inline F32 LLColor3::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLColor3::lengthSquared(void) const @@ -294,7 +294,7 @@ inline F32 LLColor3::lengthSquared(void) const inline F32 LLColor3::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag) @@ -310,7 +310,7 @@ inline F32 LLColor3::normalize(void) // deprecated inline F32 LLColor3::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } // deprecated @@ -322,7 +322,7 @@ inline F32 LLColor3::magVecSquared(void) const // deprecated inline F32 LLColor3::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag) @@ -444,7 +444,7 @@ inline F32 distVec(const LLColor3 &a, const LLColor3 &b) F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; F32 z = a.mV[2] - b.mV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F32 distVec_squared(const LLColor3 &a, const LLColor3 &b) diff --git a/indra/llmath/v3dmath.h b/indra/llmath/v3dmath.h index 6ab31e8a41..ab253de064 100644 --- a/indra/llmath/v3dmath.h +++ b/indra/llmath/v3dmath.h @@ -240,7 +240,7 @@ inline const LLVector3d& LLVector3d::setVec(const F64 *vec) inline F64 LLVector3d::normVec(void) { - F64 mag = fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + F64 mag = (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); F64 oomag; if (mag > FP_MAG_THRESHOLD) @@ -262,7 +262,7 @@ inline F64 LLVector3d::normVec(void) inline F64 LLVector3d::normalize(void) { - F64 mag = fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + F64 mag = (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); F64 oomag; if (mag > FP_MAG_THRESHOLD) @@ -286,7 +286,7 @@ inline F64 LLVector3d::normalize(void) inline F64 LLVector3d::magVec(void) const { - return fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + return (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); } inline F64 LLVector3d::magVecSquared(void) const @@ -296,7 +296,7 @@ inline F64 LLVector3d::magVecSquared(void) const inline F64 LLVector3d::length(void) const { - return fsqrtf(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); + return (F32) sqrt(mdV[0]*mdV[0] + mdV[1]*mdV[1] + mdV[2]*mdV[2]); } inline F64 LLVector3d::lengthSquared(void) const @@ -406,7 +406,7 @@ inline F64 dist_vec(const LLVector3d &a, const LLVector3d &b) F64 x = a.mdV[0] - b.mdV[0]; F64 y = a.mdV[1] - b.mdV[1]; F64 z = a.mdV[2] - b.mdV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F64 dist_vec_squared(const LLVector3d &a, const LLVector3d &b) diff --git a/indra/llmath/v3math.h b/indra/llmath/v3math.h index 75c860a91e..5d483a8753 100644 --- a/indra/llmath/v3math.h +++ b/indra/llmath/v3math.h @@ -282,7 +282,7 @@ inline void LLVector3::setVec(const F32 *vec) inline F32 LLVector3::normalize(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -305,7 +305,7 @@ inline F32 LLVector3::normalize(void) // deprecated inline F32 LLVector3::normVec(void) { - F32 mag = fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + F32 mag = (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -329,7 +329,7 @@ inline F32 LLVector3::normVec(void) inline F32 LLVector3::length(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLVector3::lengthSquared(void) const @@ -339,7 +339,7 @@ inline F32 LLVector3::lengthSquared(void) const inline F32 LLVector3::magVec(void) const { - return fsqrtf(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); + return (F32) sqrt(mV[0]*mV[0] + mV[1]*mV[1] + mV[2]*mV[2]); } inline F32 LLVector3::magVecSquared(void) const @@ -473,7 +473,7 @@ inline F32 dist_vec(const LLVector3 &a, const LLVector3 &b) F32 x = a.mV[0] - b.mV[0]; F32 y = a.mV[1] - b.mV[1]; F32 z = a.mV[2] - b.mV[2]; - return fsqrtf( x*x + y*y + z*z ); + return (F32) sqrt( x*x + y*y + z*z ); } inline F32 dist_vec_squared(const LLVector3 &a, const LLVector3 &b) diff --git a/indra/llmath/v4color.h b/indra/llmath/v4color.h index 6b63b976b0..dd92e1cc63 100644 --- a/indra/llmath/v4color.h +++ b/indra/llmath/v4color.h @@ -392,7 +392,7 @@ inline const LLColor4& LLColor4::setAlpha(F32 a) inline F32 LLColor4::length(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLColor4::lengthSquared(void) const @@ -402,7 +402,7 @@ inline F32 LLColor4::lengthSquared(void) const inline F32 LLColor4::normalize(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag) @@ -418,7 +418,7 @@ inline F32 LLColor4::normalize(void) // deprecated inline F32 LLColor4::magVec(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } // deprecated @@ -430,7 +430,7 @@ inline F32 LLColor4::magVecSquared(void) const // deprecated inline F32 LLColor4::normVec(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag) diff --git a/indra/llmath/v4coloru.h b/indra/llmath/v4coloru.h index 4ec5a345eb..08245403a1 100644 --- a/indra/llmath/v4coloru.h +++ b/indra/llmath/v4coloru.h @@ -300,7 +300,7 @@ inline const LLColor4U& LLColor4U::setAlpha(U8 a) inline F32 LLColor4U::length(void) const { - return fsqrtf( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); + return (F32) sqrt( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); } inline F32 LLColor4U::lengthSquared(void) const @@ -311,7 +311,7 @@ inline F32 LLColor4U::lengthSquared(void) const // deprecated inline F32 LLColor4U::magVec(void) const { - return fsqrtf( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); + return (F32) sqrt( ((F32)mV[VX]) * mV[VX] + ((F32)mV[VY]) * mV[VY] + ((F32)mV[VZ]) * mV[VZ] ); } // deprecated diff --git a/indra/llmath/v4math.h b/indra/llmath/v4math.h index 4c82e6b629..72a477ed20 100644 --- a/indra/llmath/v4math.h +++ b/indra/llmath/v4math.h @@ -321,7 +321,7 @@ inline void LLVector4::setVec(const F32 *vec) inline F32 LLVector4::length(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLVector4::lengthSquared(void) const @@ -331,7 +331,7 @@ inline F32 LLVector4::lengthSquared(void) const inline F32 LLVector4::magVec(void) const { - return fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + return (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); } inline F32 LLVector4::magVecSquared(void) const @@ -463,7 +463,7 @@ inline LLVector4 lerp(const LLVector4 &a, const LLVector4 &b, F32 u) inline F32 LLVector4::normalize(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag > FP_MAG_THRESHOLD) @@ -486,7 +486,7 @@ inline F32 LLVector4::normalize(void) // deprecated inline F32 LLVector4::normVec(void) { - F32 mag = fsqrtf(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); + F32 mag = (F32) sqrt(mV[VX]*mV[VX] + mV[VY]*mV[VY] + mV[VZ]*mV[VZ]); F32 oomag; if (mag > FP_MAG_THRESHOLD) diff --git a/indra/llmessage/llsdmessagebuilder.cpp b/indra/llmessage/llsdmessagebuilder.cpp index 6e41b03895..2680dd0b77 100755 --- a/indra/llmessage/llsdmessagebuilder.cpp +++ b/indra/llmessage/llsdmessagebuilder.cpp @@ -35,6 +35,7 @@ #include "llsdmessagebuilder.h" #include "llmessagetemplate.h" +#include "llmath.h" #include "llquaternion.h" #include "llsdutil.h" #include "llsdutil_math.h" diff --git a/indra/llmessage/lltemplatemessagebuilder.cpp b/indra/llmessage/lltemplatemessagebuilder.cpp index fa02456d90..6f9707ed57 100644 --- a/indra/llmessage/lltemplatemessagebuilder.cpp +++ b/indra/llmessage/lltemplatemessagebuilder.cpp @@ -35,6 +35,7 @@ #include "lltemplatemessagebuilder.h" #include "llmessagetemplate.h" +#include "llmath.h" #include "llquaternion.h" #include "u64.h" #include "v3dmath.h" diff --git a/indra/llmessage/lltemplatemessagereader.cpp b/indra/llmessage/lltemplatemessagereader.cpp index 8f56cf2521..2b67e9f6fa 100644 --- a/indra/llmessage/lltemplatemessagereader.cpp +++ b/indra/llmessage/lltemplatemessagereader.cpp @@ -36,6 +36,7 @@ #include "llfasttimer.h" #include "llmessagebuilder.h" #include "llmessagetemplate.h" +#include "llmath.h" #include "llquaternion.h" #include "message.h" #include "u64.h" diff --git a/indra/llrender/llrender.cpp b/indra/llrender/llrender.cpp index 70601663e6..2b1a9bbf94 100644 --- a/indra/llrender/llrender.cpp +++ b/indra/llrender/llrender.cpp @@ -924,7 +924,7 @@ LLVector3 LLRender::getUITranslation() { llerrs << "UI offset stack empty." << llendl; } - return LLVector3(mUIOffset[mUIStackDepth-1].getF32()); + return LLVector3(mUIOffset[mUIStackDepth-1].getF32ptr()); } LLVector3 LLRender::getUIScale() @@ -933,7 +933,7 @@ LLVector3 LLRender::getUIScale() { llerrs << "UI scale stack empty." << llendl; } - return LLVector3(mUIScale[mUIStackDepth-1].getF32()); + return LLVector3(mUIScale[mUIStackDepth-1].getF32ptr()); } diff --git a/indra/newview/llagent.cpp b/indra/newview/llagent.cpp index 72d51540ef..5fae5b893f 100644 --- a/indra/newview/llagent.cpp +++ b/indra/newview/llagent.cpp @@ -1210,7 +1210,7 @@ void LLAgent::startAutoPilotGlobal(const LLVector3d &target_global, const std::s else { // Guess at a reasonable stop distance. - mAutoPilotStopDistance = fsqrtf( distance ); + mAutoPilotStopDistance = (F32) sqrt( distance ); if (mAutoPilotStopDistance < 0.5f) { mAutoPilotStopDistance = 0.5f; diff --git a/indra/newview/lldrawable.cpp b/indra/newview/lldrawable.cpp index 8d0a06378f..68f52e04bc 100644 --- a/indra/newview/lldrawable.cpp +++ b/indra/newview/lldrawable.cpp @@ -743,7 +743,7 @@ void LLDrawable::updateDistance(LLCamera& camera, bool force_update) } else { - pos = LLVector3(getPositionGroup().getF32()); + pos = LLVector3(getPositionGroup().getF32ptr()); } pos -= camera.getOrigin(); @@ -824,7 +824,7 @@ void LLDrawable::shiftPos(const LLVector4a &shift_vector) for (S32 i = 0; i < getNumFaces(); i++) { LLFace *facep = getFace(i); - facep->mCenterAgent += LLVector3(shift_vector.getF32()); + facep->mCenterAgent += LLVector3(shift_vector.getF32ptr()); facep->mExtents[0].add(shift_vector); facep->mExtents[1].add(shift_vector); @@ -1153,13 +1153,13 @@ void LLSpatialBridge::updateSpatialExtents() LLVector4a max; max.setAdd(center, delta); - newMin.setMin(min); - newMax.setMax(max); + newMin.setMin(newMin, min); + newMax.setMax(newMax, max); } LLVector4a diagonal; diagonal.setSub(newMax, newMin); - mRadius = diagonal.length3() * 0.5f; + mRadius = diagonal.getLength3().getF32() * 0.5f; mPositionGroup->setAdd(newMin,newMax); mPositionGroup->mul(0.5f); diff --git a/indra/newview/lldrawpoolavatar.cpp b/indra/newview/lldrawpoolavatar.cpp index af082eea3e..d698624c15 100644 --- a/indra/newview/lldrawpoolavatar.cpp +++ b/indra/newview/lldrawpoolavatar.cpp @@ -37,6 +37,7 @@ #include "llvoavatar.h" #include "m3math.h" +#include "llmatrix4a.h" #include "llagent.h" //for gAgent.needsRenderAvatar() #include "lldrawable.h" @@ -495,7 +496,7 @@ S32 LLDrawPoolAvatar::getNumPasses() } else { - return 3; + return 4; } #else if (LLPipeline::sImpostorRender) @@ -814,26 +815,46 @@ void LLDrawPoolAvatar::endSkinned() #if LL_MESH_ENABLED void LLDrawPoolAvatar::beginRiggedSimple() { - if (LLPipeline::sUnderWaterRender) + if (sShaderLevel > 0) { - sVertexProgram = &gSkinnedObjectSimpleWaterProgram; + if (LLPipeline::sUnderWaterRender) + { + sVertexProgram = &gSkinnedObjectSimpleWaterProgram; + } + else + { + sVertexProgram = &gSkinnedObjectSimpleProgram; + } } else { - sVertexProgram = &gSkinnedObjectSimpleProgram; + if (LLPipeline::sUnderWaterRender) + { + sVertexProgram = &gObjectSimpleWaterProgram; + } + else + { + sVertexProgram = &gObjectSimpleProgram; + } } - sDiffuseChannel = 0; - sVertexProgram->bind(); - LLVertexBuffer::sWeight4Loc = sVertexProgram->getAttribLocation(LLViewerShaderMgr::OBJECT_WEIGHT); + if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) + { + sDiffuseChannel = 0; + sVertexProgram->bind(); + LLVertexBuffer::sWeight4Loc = sVertexProgram->getAttribLocation(LLViewerShaderMgr::OBJECT_WEIGHT); + } } void LLDrawPoolAvatar::endRiggedSimple() { LLVertexBuffer::unbind(); - sVertexProgram->unbind(); - sVertexProgram = NULL; - LLVertexBuffer::sWeight4Loc = -1; + if (sShaderLevel > 0 || gPipeline.canUseVertexShaders()) + { + sVertexProgram->unbind(); + sVertexProgram = NULL; + LLVertexBuffer::sWeight4Loc = -1; + } } void LLDrawPoolAvatar::beginRiggedAlpha() @@ -1281,8 +1302,10 @@ void LLDrawPoolAvatar::renderAvatars(LLVOAvatar* single_avatar, S32 pass) } #if LL_MESH_ENABLED -void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLFace* face, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face) +void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, LLFace* face, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face) { + LLVertexBuffer* buffer = face->mVertexBuffer; + U32 data_mask = 0; for (U32 i = 0; i < face->mRiggedIndex.size(); ++i) { @@ -1292,17 +1315,23 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLFace* face, const LLMeshSk } } - LLVertexBuffer* buff = face->mVertexBuffer; - - if (!buff || - buff->getTypeMask() != data_mask || - buff->getRequestedVerts() != vol_face.mNumVertices) + if (!buffer || + buffer->getTypeMask() != data_mask || + buffer->getRequestedVerts() != vol_face.mNumVertices) { face->setGeomIndex(0); face->setIndicesIndex(0); face->setSize(vol_face.mNumVertices, vol_face.mNumIndices, true); - face->mVertexBuffer = new LLVertexBuffer(data_mask, 0); + if (sShaderLevel > 0) + { + face->mVertexBuffer = new LLVertexBuffer(data_mask, GL_DYNAMIC_DRAW_ARB); + } + else + { + face->mVertexBuffer = new LLVertexBuffer(data_mask, GL_STREAM_DRAW_ARB); + } + face->mVertexBuffer->allocateBuffer(face->getGeomCount(), face->getIndicesCount(), true); U16 offset = 0; @@ -1319,7 +1348,83 @@ void LLDrawPoolAvatar::updateRiggedFaceVertexBuffer(LLFace* face, const LLMeshSk LLMatrix3 mat_normal(mat3); face->getGeometryVolume(*volume, face->getTEOffset(), mat_vert, mat_normal, offset, true); - buff = face->mVertexBuffer; + buffer = face->mVertexBuffer; + } + + if (sShaderLevel <= 0 && face->mLastSkinTime < avatar->getLastSkinTime()) + { //perform software vertex skinning for this face + LLStrider<LLVector3> position; + LLStrider<LLVector3> normal; + + buffer->getVertexStrider(position); + buffer->getNormalStrider(normal); + + LLVector4a* pos = (LLVector4a*) position.get(); + LLVector4a* norm = (LLVector4a*) normal.get(); + + //build matrix palette + LLMatrix4a mp[64]; + LLMatrix4* mat = (LLMatrix4*) mp; + + for (U32 j = 0; j < skin->mJointNames.size(); ++j) + { + LLJoint* joint = avatar->getJoint(skin->mJointNames[j]); + if (joint) + { + mat[j] = skin->mInvBindMatrix[j]; + mat[j] *= joint->getWorldMatrix(); + } + } + + LLVector4a* weight = vol_face.mWeights; + + LLMatrix4a bind_shape_matrix; + bind_shape_matrix.loadu(skin->mBindShapeMatrix); + + for (U32 j = 0; j < buffer->getRequestedVerts(); ++j) + { + LLMatrix4a final_mat; + final_mat.clear(); + + S32 idx[4]; + + LLVector4 wght; + + F32 scale = 0.f; + for (U32 k = 0; k < 4; k++) + { + F32 w = weight[j][k]; + + idx[k] = (S32) floorf(w); + wght[k] = w - floorf(w); + scale += wght[k]; + } + + wght *= 1.f/scale; + + for (U32 k = 0; k < 4; k++) + { + F32 w = wght[k]; + + LLMatrix4a src; + src.setMul(mp[idx[k]], w); + + final_mat.add(src); + } + + + LLVector4a& v = vol_face.mPositions[j]; + LLVector4a t; + LLVector4a dst; + bind_shape_matrix.affineTransform(v, t); + final_mat.affineTransform(t, dst); + pos[j] = dst; + + LLVector4a& n = vol_face.mNormals[j]; + bind_shape_matrix.rotate(n, t); + final_mat.rotate(t, dst); + norm[j] = dst; + } } } @@ -1371,7 +1476,7 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) stop_glerror(); const LLVolumeFace& vol_face = volume->getVolumeFace(te); - updateRiggedFaceVertexBuffer(face, skin, volume, vol_face); + updateRiggedFaceVertexBuffer(avatar, face, skin, volume, vol_face); stop_glerror(); @@ -1381,30 +1486,37 @@ void LLDrawPoolAvatar::renderRigged(LLVOAvatar* avatar, U32 type, bool glow) if (buff) { - LLMatrix4 mat[64]; + if (sShaderLevel > 0) + { //upload matrix palette to shader + LLMatrix4 mat[64]; - for (U32 i = 0; i < skin->mJointNames.size(); ++i) - { - LLJoint* joint = avatar->getJoint(skin->mJointNames[i]); - if (joint) + for (U32 i = 0; i < skin->mJointNames.size(); ++i) { - mat[i] = skin->mInvBindMatrix[i]; - mat[i] *= joint->getWorldMatrix(); + LLJoint* joint = avatar->getJoint(skin->mJointNames[i]); + if (joint) + { + mat[i] = skin->mInvBindMatrix[i]; + mat[i] *= joint->getWorldMatrix(); + } } + + stop_glerror(); + + LLDrawPoolAvatar::sVertexProgram->uniformMatrix4fv("matrixPalette", + skin->mJointNames.size(), + FALSE, + (GLfloat*) mat[0].mMatrix); + LLDrawPoolAvatar::sVertexProgram->uniformMatrix4fv("matrixPalette[0]", + skin->mJointNames.size(), + FALSE, + (GLfloat*) mat[0].mMatrix); + + stop_glerror(); + } + else + { + data_mask &= ~LLVertexBuffer::MAP_WEIGHT4; } - - stop_glerror(); - - LLDrawPoolAvatar::sVertexProgram->uniformMatrix4fv("matrixPalette", - skin->mJointNames.size(), - FALSE, - (GLfloat*) mat[0].mMatrix); - LLDrawPoolAvatar::sVertexProgram->uniformMatrix4fv("matrixPalette[0]", - skin->mJointNames.size(), - FALSE, - (GLfloat*) mat[0].mMatrix); - - stop_glerror(); buff->setBuffer(data_mask); diff --git a/indra/newview/lldrawpoolavatar.h b/indra/newview/lldrawpoolavatar.h index b01394534b..d08ae04516 100644 --- a/indra/newview/lldrawpoolavatar.h +++ b/indra/newview/lldrawpoolavatar.h @@ -138,7 +138,8 @@ public: void endDeferredRiggedSimple(); void endDeferredRiggedBump(); - void updateRiggedFaceVertexBuffer(LLFace* facep, + void updateRiggedFaceVertexBuffer(LLVOAvatar* avatar, + LLFace* facep, const LLMeshSkinInfo* skin, LLVolume* volume, const LLVolumeFace& vol_face); diff --git a/indra/newview/llface.cpp b/indra/newview/llface.cpp index 8533f9710c..5d7d3387a4 100644 --- a/indra/newview/llface.cpp +++ b/indra/newview/llface.cpp @@ -106,8 +106,8 @@ void planarProjection(LLVector2 &tc, const LLVector4a& normal, LLVector4a tangent; tangent.setCross3(binormal,normal); - tc.mV[1] = -((tangent.dot3(vec))*2 - 0.5f); - tc.mV[0] = 1.0f+((binormal.dot3(vec))*2 - 0.5f); + tc.mV[1] = -((tangent.dot3(vec).getF32())*2 - 0.5f); + tc.mV[0] = 1.0f+((binormal.dot3(vec).getF32())*2 - 0.5f); } void sphericalProjection(LLVector2 &tc, const LLVector4a& normal, @@ -156,6 +156,7 @@ void LLFace::init(LLDrawable* drawablep, LLViewerObject* objp) mLastUpdateTime = gFrameTimeSeconds; mLastMoveTime = 0.f; + mLastSkinTime = gFrameTimeSeconds; mVSize = 0.f; mPixelArea = 16.f; mState = GLOBAL; @@ -828,8 +829,8 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f, LLVector4a max; max.setAdd(center, delta); - newMin.setMin(min); - newMax.setMax(max); + newMin.setMin(newMin,min); + newMax.setMax(newMax,max); } if (!mDrawablep->isActive()) @@ -844,11 +845,11 @@ BOOL LLFace::genVolumeBBoxes(const LLVolume &volume, S32 f, t.mul(0.5f); //VECTORIZE THIS - mCenterLocal.set(t.getF32()); + mCenterLocal.set(t.getF32ptr()); t.setSub(newMax,newMin); t.mul(0.5f); - mBoundingSphereRadius = t.length3(); + mBoundingSphereRadius = t.getLength3().getF32(); updateCenterAgent(); } @@ -1313,7 +1314,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, { if (!do_xform) { - LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2); + LLVector4a::memcpyNonAliased16((F32*) tex_coords.get(), (F32*) vf.mTexCoords, num_vertices*2*sizeof(F32)); } else { @@ -1529,13 +1530,13 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (mDrawablep->isActive()) { LLVector3 t; - t.set(binormal.getF32()); + t.set(binormal.getF32ptr()); t *= bump_quat; binormal.load3(t.mV); } binormal.normalize3fast(); - tc += LLVector2( bump_s_primary_light_ray.dot3(tangent), bump_t_primary_light_ray.dot3(binormal) ); + tc += LLVector2( bump_s_primary_light_ray.dot3(tangent).getF32(), bump_t_primary_light_ray.dot3(binormal).getF32() ); *tex_coords2++ = tc; } @@ -1583,7 +1584,7 @@ BOOL LLFace::getGeometryVolume(const LLVolume& volume, if (rebuild_weights && vf.mWeights) { - LLVector4a::memcpyNonAliased16((F32*) weights, (F32*) vf.mWeights, num_vertices*4); + LLVector4a::memcpyNonAliased16((F32*) weights, (F32*) vf.mWeights, num_vertices*4*sizeof(F32)); } if (rebuild_color) @@ -1705,21 +1706,21 @@ BOOL LLFace::calcPixelArea(F32& cos_angle_to_view_dir, F32& radius) LLViewerCamera* camera = LLViewerCamera::getInstance(); - F32 size_squared = size.dot3(size); + F32 size_squared = size.dot3(size).getF32(); LLVector4a lookAt; LLVector4a t; t.load3(camera->getOrigin().mV); lookAt.setSub(center, t); - F32 dist = lookAt.length3(); + F32 dist = lookAt.getLength3().getF32(); lookAt.normalize3fast() ; //get area of circle around node - F32 app_angle = atanf(fsqrtf(size_squared) / dist); + F32 app_angle = atanf((F32) sqrt(size_squared) / dist); radius = app_angle*LLDrawable::sCurPixelAngle; mPixelArea = radius*radius * 3.14159f; LLVector4a x_axis; x_axis.load3(camera->getXAxis().mV); - cos_angle_to_view_dir = lookAt.dot3(x_axis); + cos_angle_to_view_dir = lookAt.dot3(x_axis).getF32(); //if has media, check if the face is out of the view frustum. if(hasMedia()) diff --git a/indra/newview/llface.h b/indra/newview/llface.h index b6a67c7fc4..a7d7889350 100644 --- a/indra/newview/llface.h +++ b/indra/newview/llface.h @@ -240,6 +240,7 @@ public: LLPointer<LLVertexBuffer> mVertexBuffer; LLPointer<LLVertexBuffer> mLastVertexBuffer; F32 mLastUpdateTime; + F32 mLastSkinTime; F32 mLastMoveTime; LLMatrix4* mTextureMatrix; LLDrawInfo* mDrawInfo; @@ -274,7 +275,7 @@ private: S32 mReferenceIndex; std::vector<S32> mRiggedIndex; - + F32 mVSize; F32 mPixelArea; diff --git a/indra/newview/llflexibleobject.cpp b/indra/newview/llflexibleobject.cpp index 8be4e34748..62336b03bf 100644 --- a/indra/newview/llflexibleobject.cpp +++ b/indra/newview/llflexibleobject.cpp @@ -100,7 +100,7 @@ void LLVolumeImplFlexible::onParameterChanged(U16 param_type, LLNetworkData *dat void LLVolumeImplFlexible::onShift(const LLVector4a &shift_vector) { //VECTORIZE THIS - LLVector3 shift(shift_vector.getF32()); + LLVector3 shift(shift_vector.getF32ptr()); for (int section = 0; section < (1<<FLEXIBLE_OBJECT_MAX_SECTIONS)+1; ++section) { mSection[section].mPosition += shift; diff --git a/indra/newview/llhudicon.cpp b/indra/newview/llhudicon.cpp index 63040904df..e5da871d03 100644 --- a/indra/newview/llhudicon.cpp +++ b/indra/newview/llhudicon.cpp @@ -302,7 +302,7 @@ BOOL LLHUDIcon::lineSegmentIntersect(const LLVector3& start, const LLVector3& en { dir.mul(t); starta.add(dir); - *intersection = LLVector3((F32*) &starta.mQ); + *intersection = LLVector3(starta.getF32ptr()); } return TRUE; } diff --git a/indra/newview/llpanelnearbymedia.cpp b/indra/newview/llpanelnearbymedia.cpp index 7f4609b83e..adeb401b77 100644 --- a/indra/newview/llpanelnearbymedia.cpp +++ b/indra/newview/llpanelnearbymedia.cpp @@ -362,7 +362,7 @@ void LLPanelNearByMedia::updateListItem(LLScrollListItem* item, LLViewerMediaImp debug_str += llformat("%g/", (float)impl->getInterest()); // proximity distance is actually distance squared -- display it as straight distance. - debug_str += llformat("%g/", fsqrtf(impl->getProximityDistance())); + debug_str += llformat("%g/", (F32) sqrt(impl->getProximityDistance())); // s += llformat("%g/", (float)impl->getCPUUsage()); // s += llformat("%g/", (float)impl->getApproximateTextureInterest()); diff --git a/indra/newview/llpanelprimmediacontrols.cpp b/indra/newview/llpanelprimmediacontrols.cpp index 98fbebbc5d..d84ac2e4c8 100644 --- a/indra/newview/llpanelprimmediacontrols.cpp +++ b/indra/newview/llpanelprimmediacontrols.cpp @@ -576,8 +576,8 @@ void LLPanelPrimMediaControls::updateShape() const LLVolumeFace& vf = volume->getVolumeFace(mTargetObjectFace); LLVector3 ext[2]; - ext[0].set(vf.mExtents[0].getF32()); - ext[1].set(vf.mExtents[1].getF32()); + ext[0].set(vf.mExtents[0].getF32ptr()); + ext[1].set(vf.mExtents[1].getF32ptr()); LLVector3 center = (ext[0]+ext[1])*0.5f; LLVector3 size = (ext[1]-ext[0])*0.5f; diff --git a/indra/newview/llselectmgr.cpp b/indra/newview/llselectmgr.cpp index 42f09f7396..92903a6aa9 100644 --- a/indra/newview/llselectmgr.cpp +++ b/indra/newview/llselectmgr.cpp @@ -1124,13 +1124,13 @@ void LLSelectMgr::getGrid(LLVector3& origin, LLQuaternion &rotation, LLVector3 & size.setSub(max_extents, min_extents); size.mul(0.5f); - mGridOrigin.set(center.getF32()); + mGridOrigin.set(center.getF32ptr()); LLDrawable* drawable = first_grid_object->mDrawable; if (drawable && drawable->isActive()) { mGridOrigin = mGridOrigin * first_grid_object->getRenderMatrix(); } - mGridScale.set(size.getF32()); + mGridScale.set(size.getF32ptr()); } } else // GRID_MODE_WORLD or just plain default @@ -3543,7 +3543,7 @@ void LLSelectMgr::deselectAllIfTooFar() { if (mDebugSelectMgr) { - llinfos << "Selection manager: auto-deselecting, select_dist = " << fsqrtf(select_dist_sq) << llendl; + llinfos << "Selection manager: auto-deselecting, select_dist = " << (F32) sqrt(select_dist_sq) << llendl; llinfos << "agent pos global = " << gAgent.getPositionGlobal() << llendl; llinfos << "selection pos global = " << selectionCenter << llendl; } diff --git a/indra/newview/llspatialpartition.cpp b/indra/newview/llspatialpartition.cpp index 2335c7bb8e..e55ec03356 100644 --- a/indra/newview/llspatialpartition.cpp +++ b/indra/newview/llspatialpartition.cpp @@ -231,7 +231,7 @@ U8* get_box_fan_indices(LLCamera* camera, const LLVector4a& center) LLVector4a origin; origin.load3(camera->getOrigin().mV); - S32 cypher = center.greaterThan4(origin).getComparisonMask() & 0x7; + S32 cypher = center.greaterThan(origin).getGatheredBits() & 0x7; return sOcclusionIndices+cypher*8; } @@ -253,7 +253,7 @@ void LLSpatialGroup::buildOcclusion() r2.splat(0.25f); r2.add(mBounds[1]); - r.setMin(r2); + r.setMin(r, r2); LLVector4a* v = mOcclusionVerts; const LLVector4a& c = mBounds[0]; @@ -775,8 +775,8 @@ BOOL LLSpatialGroup::boundObjects(BOOL empty, LLVector4a& minOut, LLVector4a& ma } else { - minOut.setMin(newMin); - maxOut.setMax(newMax); + minOut.setMin(minOut, newMin); + maxOut.setMax(maxOut, newMax); } return TRUE; @@ -1220,8 +1220,8 @@ void LLSpatialGroup::updateDistance(LLCamera &camera) #endif if (!getData().empty()) { - mRadius = mSpatialPartition->mRenderByGroup ? mObjectBounds[1].length3() : - (F32) mOctreeNode->getSize().length3(); + mRadius = mSpatialPartition->mRenderByGroup ? mObjectBounds[1].getLength3().getF32() : + (F32) mOctreeNode->getSize().getLength3().getF32(); mDistance = mSpatialPartition->calcDistance(this, camera); mPixelArea = mSpatialPartition->calcPixelArea(this, camera); } @@ -1241,7 +1241,7 @@ F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera) { LLVector4a v = eye; - dist = eye.length3(); + dist = eye.getLength3().getF32(); eye.normalize3fast(); if (!group->isState(LLSpatialGroup::ALPHA_DIRTY)) @@ -1253,7 +1253,7 @@ F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera) LLVector4a diff; diff.setSub(view_angle, *group->mLastUpdateViewAngle); - if (diff.length3() > 0.64f) + if (diff.getLength3().getF32() > 0.64f) { *group->mViewAngle = view_angle; *group->mLastUpdateViewAngle = view_angle; @@ -1279,11 +1279,11 @@ F32 LLSpatialPartition::calcDistance(LLSpatialGroup* group, LLCamera& camera) t.mul(group->mObjectBounds[1]); v.sub(t); - group->mDepth = v.dot3(ata); + group->mDepth = v.dot3(ata).getF32(); } else { - dist = eye.length3(); + dist = eye.getLength3().getF32(); } if (dist < 16.f) @@ -1497,8 +1497,8 @@ BOOL LLSpatialGroup::rebound() const LLVector4a& max = group->mExtents[1]; const LLVector4a& min = group->mExtents[0]; - newMax.setMax(max); - newMin.setMin(min); + newMax.setMax(newMax, max); + newMin.setMin(newMin, min); } boundObjects(FALSE, newMin, newMax); @@ -2196,8 +2196,8 @@ BOOL LLSpatialPartition::getVisibleExtents(LLCamera& camera, LLVector3& visMin, LLOctreeCullVisExtents vis(&camera, visMina, visMaxa); vis.traverse(mOctree); - visMin.set(visMina.getF32()); - visMax.set(visMaxa.getF32()); + visMin.set(visMina.getF32ptr()); + visMax.set(visMaxa.getF32ptr()); return vis.mEmpty; } @@ -2280,13 +2280,13 @@ BOOL earlyFail(LLCamera* camera, LLSpatialGroup* group) LLVector4a max; max.setAdd(c,r); - S32 lt = e.lessThan4(min).getComparisonMask() & 0x7; + S32 lt = e.lessThan(min).getGatheredBits() & 0x7; if (lt) { return FALSE; } - S32 gt = e.greaterThan4(max).getComparisonMask() & 0x7; + S32 gt = e.greaterThan(max).getGatheredBits() & 0x7; if (gt) { return FALSE; @@ -2745,8 +2745,8 @@ void renderNormals(LLDrawable* drawablep) p.setAdd(face.mPositions[j], n); gGL.color4f(1,1,1,1); - gGL.vertex3fv(face.mPositions[j].getF32()); - gGL.vertex3fv(p.getF32()); + gGL.vertex3fv(face.mPositions[j].getF32ptr()); + gGL.vertex3fv(p.getF32ptr()); if (face.mBinormals) { @@ -2754,8 +2754,8 @@ void renderNormals(LLDrawable* drawablep) p.setAdd(face.mPositions[j], n); gGL.color4f(0,1,1,1); - gGL.vertex3fv(face.mPositions[j].getF32()); - gGL.vertex3fv(p.getF32()); + gGL.vertex3fv(face.mPositions[j].getF32ptr()); + gGL.vertex3fv(p.getF32ptr()); } } @@ -3024,8 +3024,8 @@ public: LLVolumeOctreeListener* vl = (LLVolumeOctreeListener*) branch->getListener(0); LLVector3 center, size; - center.set(vl->mBounds[0].getF32()); - size.set(vl->mBounds[1].getF32()); + center.set(vl->mBounds[0].getF32ptr()); + size.set(vl->mBounds[1].getF32ptr()); drawBoxOutline(center, size); } diff --git a/indra/newview/llviewercamera.cpp b/indra/newview/llviewercamera.cpp index cef7c4abbb..917185ed04 100644 --- a/indra/newview/llviewercamera.cpp +++ b/indra/newview/llviewercamera.cpp @@ -810,7 +810,7 @@ BOOL LLViewerCamera::areVertsVisible(LLViewerObject* volumep, BOOL all_verts) render_mata.affineTransform(t, vec); } - BOOL in_frustum = pointInFrustum(LLVector3(vec.getF32())) > 0; + BOOL in_frustum = pointInFrustum(LLVector3(vec.getF32ptr())) > 0; if (( !in_frustum && all_verts) || (in_frustum && !all_verts)) diff --git a/indra/newview/llviewerjointmesh.cpp b/indra/newview/llviewerjointmesh.cpp index 9e35b5cc51..0a181e09bb 100644 --- a/indra/newview/llviewerjointmesh.cpp +++ b/indra/newview/llviewerjointmesh.cpp @@ -726,8 +726,8 @@ void LLViewerJointMesh::updateFaceData(LLFace *face, F32 pixel_area, BOOL damp_w U32 words = num_verts*4; - LLVector4a::memcpyNonAliased16(v, (F32*) mMesh->getCoords(), words); - LLVector4a::memcpyNonAliased16(n, (F32*) mMesh->getNormals(), words); + LLVector4a::memcpyNonAliased16(v, (F32*) mMesh->getCoords(), words*sizeof(F32)); + LLVector4a::memcpyNonAliased16(n, (F32*) mMesh->getNormals(), words*sizeof(F32)); if (!terse_update) @@ -740,9 +740,9 @@ void LLViewerJointMesh::updateFaceData(LLFace *face, F32 pixel_area, BOOL damp_w F32* vw = (F32*) vertex_weightsp.get(); F32* cw = (F32*) clothing_weightsp.get(); - LLVector4a::memcpyNonAliased16(tc, (F32*) mMesh->getTexCoords(), num_verts*2); - LLVector4a::memcpyNonAliased16(vw, (F32*) mMesh->getWeights(), num_verts); - LLVector4a::memcpyNonAliased16(cw, (F32*) mMesh->getClothingWeights(), num_verts*4); + LLVector4a::memcpyNonAliased16(tc, (F32*) mMesh->getTexCoords(), num_verts*2*sizeof(F32)); + LLVector4a::memcpyNonAliased16(vw, (F32*) mMesh->getWeights(), num_verts*sizeof(F32)); + LLVector4a::memcpyNonAliased16(cw, (F32*) mMesh->getClothingWeights(), num_verts*4*sizeof(F32)); } const U32 idx_count = mMesh->getNumFaces()*3; diff --git a/indra/newview/llviewerjoystick.cpp b/indra/newview/llviewerjoystick.cpp index 79d8fc7df9..16a6022b86 100644 --- a/indra/newview/llviewerjoystick.cpp +++ b/indra/newview/llviewerjoystick.cpp @@ -764,7 +764,7 @@ void LLViewerJoystick::moveAvatar(bool reset) sDelta[RX_I] += (cur_delta[RX_I] - sDelta[RX_I]) * time * feather; sDelta[RY_I] += (cur_delta[RY_I] - sDelta[RY_I]) * time * feather; - handleRun(fsqrtf(sDelta[Z_I]*sDelta[Z_I] + sDelta[X_I]*sDelta[X_I])); + handleRun((F32) sqrt(sDelta[Z_I]*sDelta[Z_I] + sDelta[X_I]*sDelta[X_I])); // Allow forward/backward movement some priority if (dom_axis == Z_I) diff --git a/indra/newview/llviewermedia.cpp b/indra/newview/llviewermedia.cpp index 178d928f57..303f339f7d 100644 --- a/indra/newview/llviewermedia.cpp +++ b/indra/newview/llviewermedia.cpp @@ -865,7 +865,7 @@ void LLViewerMedia::updateMedia(void *dummy_arg) // Set the low priority size for downsampling to approximately the size the texture is displayed at. { - F32 approximate_interest_dimension = fsqrtf(pimpl->getInterest()); + F32 approximate_interest_dimension = (F32) sqrt(pimpl->getInterest()); pimpl->setLowPrioritySizeLimit(llround(approximate_interest_dimension)); } diff --git a/indra/newview/llviewerobject.cpp b/indra/newview/llviewerobject.cpp index 2b89deaa53..9587fbafb1 100644 --- a/indra/newview/llviewerobject.cpp +++ b/indra/newview/llviewerobject.cpp @@ -2973,7 +2973,7 @@ F32 LLViewerObject::getBinRadius() const LLVector4a* ext = mDrawable->getSpatialExtents(); LLVector4a diff; diff.setSub(ext[1], ext[0]); - return diff.length3(); + return diff.getLength3().getF32(); } return getScale().magVec(); diff --git a/indra/newview/llviewerpartsim.cpp b/indra/newview/llviewerpartsim.cpp index 41848e8b7a..4759454ee7 100644 --- a/indra/newview/llviewerpartsim.cpp +++ b/indra/newview/llviewerpartsim.cpp @@ -161,8 +161,8 @@ LLViewerPartGroup::LLViewerPartGroup(const LLVector3 ¢er_agent, const F32 bo if (group != NULL) { - LLVector3 center(group->mOctreeNode->getCenter().getF32()); - LLVector3 size(group->mOctreeNode->getSize().getF32()); + LLVector3 center(group->mOctreeNode->getCenter().getF32ptr()); + LLVector3 size(group->mOctreeNode->getSize().getF32ptr()); size += LLVector3(0.01f, 0.01f, 0.01f); mMinObjPos = center - size; mMaxObjPos = center + size; diff --git a/indra/newview/llviewertexture.cpp b/indra/newview/llviewertexture.cpp index 719c5b0da5..d50efe89dd 100644 --- a/indra/newview/llviewertexture.cpp +++ b/indra/newview/llviewertexture.cpp @@ -1580,7 +1580,7 @@ F32 LLViewerFetchedTexture::calcDecodePriority() S32 cur_discard = getCurrentDiscardLevelForFetching(); bool have_all_data = (cur_discard >= 0 && (cur_discard <= mDesiredDiscardLevel)); - F32 pixel_priority = fsqrtf(mMaxVirtualSize); + F32 pixel_priority = (F32) sqrt(mMaxVirtualSize); F32 priority = 0.f; diff --git a/indra/newview/llvoavatar.cpp b/indra/newview/llvoavatar.cpp index abf22b5e5a..e50d0fce49 100644 --- a/indra/newview/llvoavatar.cpp +++ b/indra/newview/llvoavatar.cpp @@ -675,6 +675,7 @@ LLVOAvatar::LLVOAvatar(const LLUUID& id, mTexHairColor( NULL ), mTexEyeColor( NULL ), mNeedsSkin(FALSE), + mLastSkinTime(0.f), mUpdatePeriod(1), mFullyLoaded(FALSE), mPreviousFullyLoaded(FALSE), @@ -1356,7 +1357,7 @@ void LLVOAvatar::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax) if (isImpostor() && !needsImpostorUpdate()) { LLVector3 delta = getRenderPosition() - - ((LLVector3(mDrawable->getPositionGroup().getF32())-mImpostorOffset)); + ((LLVector3(mDrawable->getPositionGroup().getF32ptr())-mImpostorOffset)); newMin.load3( (mLastAnimExtents[0] + delta).mV); newMax.load3( (mLastAnimExtents[1] + delta).mV); @@ -1364,12 +1365,12 @@ void LLVOAvatar::updateSpatialExtents(LLVector4a& newMin, LLVector4a &newMax) else { getSpatialExtents(newMin,newMax); - mLastAnimExtents[0].set(newMin.getF32()); - mLastAnimExtents[1].set(newMax.getF32()); + mLastAnimExtents[0].set(newMin.getF32ptr()); + mLastAnimExtents[1].set(newMax.getF32ptr()); LLVector4a pos_group; pos_group.setAdd(newMin,newMax); pos_group.mul(0.5f); - mImpostorOffset = LLVector3(pos_group.getF32())-getRenderPosition(); + mImpostorOffset = LLVector3(pos_group.getF32ptr())-getRenderPosition(); mDrawable->setPositionGroup(pos_group); } } @@ -1435,7 +1436,7 @@ void LLVOAvatar::getSpatialExtents(LLVector4a& newMin, LLVector4a& newMax) distance.setSub(ext[1], ext[0]); LLVector4a max_span(max_attachment_span); - S32 lt = distance.lessThan4(max_span).getComparisonMask() & 0x7; + S32 lt = distance.lessThan(max_span).getGatheredBits() & 0x7; // Only add the prim to spatial extents calculations if it isn't a megaprim. // max_attachment_span calculated at the start of the function @@ -2533,14 +2534,14 @@ void LLVOAvatar::idleUpdateMisc(bool detailed_update) getSpatialExtents(ext[0], ext[1]); LLVector4a diff; diff.setSub(ext[1], mImpostorExtents[1]); - if (diff.length3() > 0.05f) + if (diff.getLength3().getF32() > 0.05f) { mNeedsImpostorUpdate = TRUE; } else { diff.setSub(ext[0], mImpostorExtents[0]); - if (diff.length3() > 0.05f) + if (diff.getLength3().getF32() > 0.05f) { mNeedsImpostorUpdate = TRUE; } @@ -3887,7 +3888,8 @@ U32 LLVOAvatar::renderSkinned(EAvatarRenderPass pass) mMeshLOD[MESH_ID_HAIR]->updateJointGeometry(); } mNeedsSkin = FALSE; - + mLastSkinTime = gFrameTimeSeconds; + LLVertexBuffer* vb = mDrawable->getFace(0)->mVertexBuffer; if (vb) { @@ -4231,7 +4233,7 @@ void LLVOAvatar::updateTextures() if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_TEXTURE_AREA)) { - setDebugText(llformat("%4.0f:%4.0f", fsqrtf(mMinPixelArea),fsqrtf(mMaxPixelArea))); + setDebugText(llformat("%4.0f:%4.0f", (F32) sqrt(mMinPixelArea),(F32) sqrt(mMaxPixelArea))); } } @@ -5443,7 +5445,7 @@ void LLVOAvatar::setPixelAreaAndAngle(LLAgent &agent) } else { - F32 radius = size.length3(); + F32 radius = size.getLength3().getF32(); mAppAngle = (F32) atan2( radius, range) * RAD_TO_DEG; } diff --git a/indra/newview/llvoavatar.h b/indra/newview/llvoavatar.h index 95b0665f7d..94b564fc8f 100644 --- a/indra/newview/llvoavatar.h +++ b/indra/newview/llvoavatar.h @@ -345,6 +345,7 @@ public: U32 renderImpostor(LLColor4U color = LLColor4U(255,255,255,255), S32 diffuse_channel = 0); U32 renderRigid(); U32 renderSkinned(EAvatarRenderPass pass); + F32 getLastSkinTime() { return mLastSkinTime; } U32 renderSkinnedAttachments(); U32 renderTransparent(BOOL first_pass); void renderCollisionVolumes(); @@ -357,6 +358,8 @@ private: bool shouldAlphaMask(); BOOL mNeedsSkin; // avatar has been animated and verts have not been updated + F32 mLastSkinTime; //value of gFrameTimeSeconds at last skin update + S32 mUpdatePeriod; S32 mNumInitFaces; //number of faces generated when creating the avatar drawable, does not inculde splitted faces due to long vertex buffer. diff --git a/indra/newview/llvograss.cpp b/indra/newview/llvograss.cpp index fe1e36cbe8..65829b213e 100644 --- a/indra/newview/llvograss.cpp +++ b/indra/newview/llvograss.cpp @@ -328,7 +328,7 @@ void LLVOGrass::updateTextures() { if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_TEXTURE_AREA)) { - setDebugText(llformat("%4.0f", fsqrtf(mPixelArea))); + setDebugText(llformat("%4.0f", (F32) sqrt(mPixelArea))); } getTEImage(0)->addTextureStats(mPixelArea); } diff --git a/indra/newview/llvosurfacepatch.cpp b/indra/newview/llvosurfacepatch.cpp index 63f662c5a1..c047758a59 100644 --- a/indra/newview/llvosurfacepatch.cpp +++ b/indra/newview/llvosurfacepatch.cpp @@ -999,8 +999,8 @@ BOOL LLVOSurfacePatch::lineSegmentIntersect(const LLVector3& start, const LLVect const LLVector4a* exta = mDrawable->getSpatialExtents(); LLVector3 ext[2]; - ext[0].set(exta[0].getF32()); - ext[1].set(exta[1].getF32()); + ext[0].set(exta[0].getF32ptr()); + ext[1].set(exta[1].getF32ptr()); F32 rad = (delta*tdelta).magVecSquared(); diff --git a/indra/newview/llvotextbubble.cpp b/indra/newview/llvotextbubble.cpp index 339da3c0bf..e790373d02 100644 --- a/indra/newview/llvotextbubble.cpp +++ b/indra/newview/llvotextbubble.cpp @@ -254,8 +254,8 @@ void LLVOTextBubble::getGeometry(S32 idx, LLVector2* dst_tc = (LLVector2*) texcoordsp.get(); LLVector2* src_tc = (LLVector2*) face.mTexCoords; - LLVector4a::memcpyNonAliased16((F32*) dst_norm, (F32*) src_norm, face.mNumVertices*4); - LLVector4a::memcpyNonAliased16((F32*) dst_tc, (F32*) src_tc, face.mNumVertices*2); + LLVector4a::memcpyNonAliased16((F32*) dst_norm, (F32*) src_norm, face.mNumVertices*4*sizeof(F32)); + LLVector4a::memcpyNonAliased16((F32*) dst_tc, (F32*) src_tc, face.mNumVertices*2*sizeof(F32)); for (U32 i = 0; i < face.mNumVertices; i++) diff --git a/indra/newview/llvotree.cpp b/indra/newview/llvotree.cpp index eb790b04cc..e3b4efb9dd 100644 --- a/indra/newview/llvotree.cpp +++ b/indra/newview/llvotree.cpp @@ -472,7 +472,7 @@ void LLVOTree::updateTextures() { if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_TEXTURE_AREA)) { - setDebugText(llformat("%4.0f", fsqrtf(mPixelArea))); + setDebugText(llformat("%4.0f", (F32) sqrt(mPixelArea))); } mTreeImagep->addTextureStats(mPixelArea); } @@ -1278,8 +1278,8 @@ BOOL LLVOTree::lineSegmentIntersect(const LLVector3& start, const LLVector3& end //VECTORIZE THIS LLVector3 ext[2]; - ext[0].set(exta[0].getF32()); - ext[1].set(exta[1].getF32()); + ext[0].set(exta[0].getF32ptr()); + ext[1].set(exta[1].getF32ptr()); LLVector3 center = (ext[1]+ext[0])*0.5f; LLVector3 size = (ext[1]-ext[0]); diff --git a/indra/newview/llvovolume.cpp b/indra/newview/llvovolume.cpp index 1397b64623..128fd15142 100644 --- a/indra/newview/llvovolume.cpp +++ b/indra/newview/llvovolume.cpp @@ -696,7 +696,7 @@ void LLVOVolume::updateTextureVirtualSize() const LLTextureEntry *te = face->getTextureEntry(); LLViewerTexture *imagep = face->getTexture(); if (!imagep || !te || - face->mExtents[0].equal3(face->mExtents[1])) + face->mExtents[0].equals3(face->mExtents[1])) { continue; } @@ -820,15 +820,15 @@ void LLVOVolume::updateTextureVirtualSize() if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_TEXTURE_AREA)) { - setDebugText(llformat("%.0f:%.0f", fsqrtf(min_vsize),fsqrtf(max_vsize))); + setDebugText(llformat("%.0f:%.0f", (F32) sqrt(min_vsize),(F32) sqrt(max_vsize))); } else if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_TEXTURE_PRIORITY)) { - setDebugText(llformat("%.0f:%.0f", fsqrtf(min_vsize),fsqrtf(max_vsize))); + setDebugText(llformat("%.0f:%.0f", (F32) sqrt(min_vsize),(F32) sqrt(max_vsize))); } else if (gPipeline.hasRenderDebugMask(LLPipeline::RENDER_DEBUG_FACE_AREA)) { - setDebugText(llformat("%.0f:%.0f", fsqrtf(min_vsize),fsqrtf(max_vsize))); + setDebugText(llformat("%.0f:%.0f", (F32) sqrt(min_vsize),(F32) sqrt(max_vsize))); } if (mPixelArea == 0) @@ -1355,8 +1355,8 @@ BOOL LLVOVolume::genBBoxes(BOOL force_global) } else { - min.setMin(face->mExtents[0]); - max.setMax(face->mExtents[1]); + min.setMin(min, face->mExtents[0]); + max.setMax(max, face->mExtents[1]); } } } @@ -1864,7 +1864,7 @@ LLVector3 LLVOVolume::getApproximateFaceNormal(U8 face_id) result.add(face.mNormals[i]); } - LLVector3 ret((F32*) &result.mQ); + LLVector3 ret(result.getF32ptr()); ret = volumeDirectionToAgent(ret); ret.normVec(); } @@ -3075,7 +3075,7 @@ F32 LLVOVolume::getBinRadius() LLVector4a rad; rad.setSub(ext[1], ext[0]); - radius = rad.length3()*0.5f; + radius = rad.getLength3().getF32()*0.5f; } else if (mDrawable->isStatic()) { diff --git a/indra/newview/llworld.cpp b/indra/newview/llworld.cpp index a1e4df8a66..53eca0d08e 100644 --- a/indra/newview/llworld.cpp +++ b/indra/newview/llworld.cpp @@ -591,7 +591,7 @@ void LLWorld::updateVisibilities() region_list_t::iterator curiter = iter++; LLViewerRegion* regionp = *curiter; F32 height = regionp->getLand().getMaxZ() - regionp->getLand().getMinZ(); - F32 radius = 0.5f*fsqrtf(height * height + diagonal_squared); + F32 radius = 0.5f*(F32) sqrt(height * height + diagonal_squared); if (!regionp->getLand().hasZData() || LLViewerCamera::getInstance()->sphereInFrustum(regionp->getCenterAgent(), radius)) { @@ -612,7 +612,7 @@ void LLWorld::updateVisibilities() } F32 height = regionp->getLand().getMaxZ() - regionp->getLand().getMinZ(); - F32 radius = 0.5f*fsqrtf(height * height + diagonal_squared); + F32 radius = 0.5f*(F32) sqrt(height * height + diagonal_squared); if (LLViewerCamera::getInstance()->sphereInFrustum(regionp->getCenterAgent(), radius)) { regionp->calculateCameraDistance(); diff --git a/indra/newview/pipeline.cpp b/indra/newview/pipeline.cpp index fec7da1dd0..b0a7b1ce83 100644 --- a/indra/newview/pipeline.cpp +++ b/indra/newview/pipeline.cpp @@ -1522,7 +1522,7 @@ F32 LLPipeline::calcPixelArea(const LLVector4a& center, const LLVector4a& size, LLVector4a lookAt; lookAt.setSub(center, origin); - F32 dist = lookAt.length3(); + F32 dist = lookAt.getLength3().getF32(); //ramp down distance for nearby objects //shrink dist by dist/16. @@ -1534,7 +1534,7 @@ F32 LLPipeline::calcPixelArea(const LLVector4a& center, const LLVector4a& size, } //get area of circle around node - F32 app_angle = atanf(size.length3()/dist); + F32 app_angle = atanf(size.getLength3().getF32()/dist); F32 radius = app_angle*LLDrawable::sCurPixelAngle; return radius*radius * F_PI; } @@ -4671,7 +4671,7 @@ static F32 calc_light_dist(LLVOVolume* light, const LLVector3& cam_pos, F32 max_ { return max_dist; } - F32 dist = fsqrtf(dist2); + F32 dist = (F32) sqrt(dist2); dist *= 1.f / inten; dist -= radius; if (selected) @@ -6980,7 +6980,7 @@ void LLPipeline::renderDeferredLighting() LLVector4a center; center.load3(drawablep->getPositionAgent().mV); - const F32* c = center.getF32(); + const F32* c = center.getF32ptr(); F32 s = volume->getLightRadius()*1.5f; LLColor3 col = volume->getLightColor(); @@ -7078,7 +7078,7 @@ void LLPipeline::renderDeferredLighting() LLVector4a center; center.load3(drawablep->getPositionAgent().mV); - const F32* c = center.getF32(); + const F32* c = center.getF32ptr(); F32 s = volume->getLightRadius()*1.5f; sVisibleLightCount++; @@ -9184,8 +9184,8 @@ void LLPipeline::generateImpostor(LLVOAvatar* avatar) up.mul(up); up.normalize3fast(); - tdim.mV[0] = fabsf(half_height.dot3(left)); - tdim.mV[1] = fabsf(half_height.dot3(up)); + tdim.mV[0] = fabsf(half_height.dot3(left).getF32()); + tdim.mV[1] = fabsf(half_height.dot3(up).getF32()); glMatrixMode(GL_PROJECTION); glPushMatrix(); |