Skip to content
Permalink
Browse files

Fix rendering IQM models between model frames

For lerped frames (refEntity_t frame not equal oldframe) IQM joint
matrices may have incorrect axis scale. This can cause significant model
distortion. The matrix lerp is linear causing each vector to move in a
straight line between frames instead of arcing like a circle. Each joint
frame can have a different scale so can't just normalize the joint
matrix.

Store joints as quaternions and spherical lerp between them and then
convert to a matrix. For my test model, setting up the skeleton is four
times slower now but it still seems to be fast enough to be usable.
  • Loading branch information...
zturtleman committed Apr 29, 2019
1 parent d13d064 commit d404519cce565402aa98c3f9943221ed6ddb2790
Showing with 296 additions and 166 deletions.
  1. +4 −0 code/qcommon/q_shared.h
  2. +9 −2 code/renderergl1/tr_local.h
  3. +137 −81 code/renderergl1/tr_model_iqm.c
  4. +9 −2 code/renderergl2/tr_local.h
  5. +137 −81 code/renderergl2/tr_model_iqm.c
@@ -368,6 +368,8 @@ typedef vec_t vec3_t[3];
typedef vec_t vec4_t[4];
typedef vec_t vec5_t[5];

typedef vec_t quat_t[4];

typedef int fixed4_t;
typedef int fixed8_t;
typedef int fixed16_t;
@@ -578,6 +580,8 @@ typedef struct {

#define Byte4Copy(a,b) ((b)[0]=(a)[0],(b)[1]=(a)[1],(b)[2]=(a)[2],(b)[3]=(a)[3])

#define QuatCopy(a,b) ((b)[0]=(a)[0],(b)[1]=(a)[1],(b)[2]=(a)[2],(b)[3]=(a)[3])

#define SnapVector(v) {v[0]=((int)(v[0]));v[1]=((int)(v[1]));v[2]=((int)(v[2]));}
// just in case you don't want to use the macros
vec_t _DotProduct( const vec3_t v1, const vec3_t v2 );
@@ -589,6 +589,12 @@ typedef struct {
drawVert_t *verts;
} srfTriangles_t;

typedef struct {
vec3_t translate;
quat_t rotate;
vec3_t scale;
} iqmTransform_t;

// inter-quake-model
typedef struct {
int num_vertexes;
@@ -623,8 +629,9 @@ typedef struct {

char *jointNames;
int *jointParents;
float *jointMats;
float *poseMats;
float *bindJoints; // [num_joints * 12]
float *invBindJoints; // [num_joints * 12]
iqmTransform_t *poses; // [num_frames * num_poses]
float *bounds;
} iqmData_t;

@@ -2,6 +2,7 @@
===========================================================================
Copyright (C) 2011 Thilo Schulz <thilo@tjps.eu>
Copyright (C) 2011 Matthias Bentrup <matthias.bentrup@googlemail.com>
Copyright (C) 2011-2019 Zack Middleton <zturtleman@gmail.com>
This file is part of Quake III Arena source code.
@@ -44,7 +45,7 @@ static qboolean IQM_CheckRange( iqmHeader_t *header, int offset,
}
// "multiply" 3x4 matrices, these are assumed to be the top 3 rows
// of a 4x4 matrix with the last row = (0 0 0 1)
static void Matrix34Multiply( float *a, float *b, float *out ) {
static void Matrix34Multiply( const float *a, const float *b, float *out ) {
out[ 0] = a[0] * b[0] + a[1] * b[4] + a[ 2] * b[ 8];
out[ 1] = a[0] * b[1] + a[1] * b[5] + a[ 2] * b[ 9];
out[ 2] = a[0] * b[2] + a[1] * b[6] + a[ 2] * b[10];
@@ -58,23 +59,7 @@ static void Matrix34Multiply( float *a, float *b, float *out ) {
out[10] = a[8] * b[2] + a[9] * b[6] + a[10] * b[10];
out[11] = a[8] * b[3] + a[9] * b[7] + a[10] * b[11] + a[11];
}
static void InterpolateMatrix( float *a, float *b, float lerp, float *mat ) {
float unLerp = 1.0f - lerp;

mat[ 0] = a[ 0] * unLerp + b[ 0] * lerp;
mat[ 1] = a[ 1] * unLerp + b[ 1] * lerp;
mat[ 2] = a[ 2] * unLerp + b[ 2] * lerp;
mat[ 3] = a[ 3] * unLerp + b[ 3] * lerp;
mat[ 4] = a[ 4] * unLerp + b[ 4] * lerp;
mat[ 5] = a[ 5] * unLerp + b[ 5] * lerp;
mat[ 6] = a[ 6] * unLerp + b[ 6] * lerp;
mat[ 7] = a[ 7] * unLerp + b[ 7] * lerp;
mat[ 8] = a[ 8] * unLerp + b[ 8] * lerp;
mat[ 9] = a[ 9] * unLerp + b[ 9] * lerp;
mat[10] = a[10] * unLerp + b[10] * lerp;
mat[11] = a[11] * unLerp + b[11] * lerp;
}
static void JointToMatrix( vec4_t rot, vec3_t scale, vec3_t trans,
static void JointToMatrix( const quat_t rot, const vec3_t scale, const vec3_t trans,
float *mat ) {
float xx = 2.0f * rot[0] * rot[0];
float yy = 2.0f * rot[1] * rot[1];
@@ -99,8 +84,7 @@ static void JointToMatrix( vec4_t rot, vec3_t scale, vec3_t trans,
mat[10] = scale[2] * (1.0f - (xx + yy));
mat[11] = trans[2];
}
static void Matrix34Invert( float *inMat, float *outMat )
{
static void Matrix34Invert( const float *inMat, float *outMat ) {
vec3_t trans;
float invSqrLen, *v;

@@ -120,6 +104,61 @@ static void Matrix34Invert( float *inMat, float *outMat )
outMat[ 7] = -DotProduct(outMat + 4, trans);
outMat[11] = -DotProduct(outMat + 8, trans);
}
static void QuatSlerp(const quat_t from, const quat_t _to, float fraction, quat_t out) {
float angle, cosAngle, sinAngle, backlerp, lerp;
quat_t to;

// cos() of angle
cosAngle = from[0] * _to[0] + from[1] * _to[1] + from[2] * _to[2] + from[3] * _to[3];

// negative handling is needed for taking shortest path (required for model joints)
if ( cosAngle < 0.0f ) {
cosAngle = -cosAngle;
to[0] = - _to[0];
to[1] = - _to[1];
to[2] = - _to[2];
to[3] = - _to[3];
} else {
QuatCopy( _to, to );
}

if ( cosAngle < 0.999999f ) {
// spherical lerp (slerp)
angle = acosf( cosAngle );
sinAngle = sinf( angle );
backlerp = sinf( ( 1.0f - fraction ) * angle ) / sinAngle;
lerp = sinf( fraction * angle ) / sinAngle;
} else {
// linear lerp
backlerp = 1.0f - fraction;
lerp = fraction;
}

out[0] = from[0] * backlerp + to[0] * lerp;
out[1] = from[1] * backlerp + to[1] * lerp;
out[2] = from[2] * backlerp + to[2] * lerp;
out[3] = from[3] * backlerp + to[3] * lerp;
}
static vec_t QuatNormalize2( const quat_t v, quat_t out) {
float length, ilength;

length = v[0]*v[0] + v[1]*v[1] + v[2]*v[2] + v[3]*v[3];

if (length) {
/* writing it this way allows gcc to recognize that rsqrt can be used */
ilength = 1/(float)sqrt (length);
/* sqrt(length) = length * (1 / sqrt(length)) */
length *= ilength;
out[0] = v[0]*ilength;
out[1] = v[1]*ilength;
out[2] = v[2]*ilength;
out[3] = v[3]*ilength;
} else {
out[0] = out[1] = out[2] = out[3] = 0;
}

return length;
}

/*
=================
@@ -139,7 +178,7 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na
unsigned short *framedata;
char *str;
int i, j, k;
float jointInvMats[IQM_MAX_JOINTS * 12] = {0.0f};
iqmTransform_t *transform;
float *mat, *matInv;
size_t size, joint_names;
byte *dataPtr;
@@ -559,10 +598,11 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na
if( header->num_joints ) {
size += joint_names; // joint names
size += header->num_joints * sizeof(int); // joint parents
size += header->num_joints * 12 * sizeof( float ); // joint mats
size += header->num_joints * 12 * sizeof(float); // bind joint matricies
size += header->num_joints * 12 * sizeof(float); // inverse bind joint matricies
}
if( header->num_poses ) {
size += header->num_poses * header->num_frames * 12 * sizeof( float ); // pose mats
size += header->num_poses * header->num_frames * sizeof(iqmTransform_t); // pose transforms
}
if( header->ofs_bounds ) {
size += header->num_frames * 6 * sizeof(float); // model bounds
@@ -633,12 +673,15 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na
iqmData->jointParents = (int*)dataPtr;
dataPtr += header->num_joints * sizeof(int); // joint parents

iqmData->jointMats = (float*)dataPtr;
dataPtr += header->num_joints * 12 * sizeof( float ); // joint mats
iqmData->bindJoints = (float*)dataPtr;
dataPtr += header->num_joints * 12 * sizeof(float); // bind joint matricies

iqmData->invBindJoints = (float*)dataPtr;
dataPtr += header->num_joints * 12 * sizeof(float); // inverse bind joint matricies
}
if( header->num_poses ) {
iqmData->poseMats = (float*)dataPtr;
dataPtr += header->num_poses * header->num_frames * 12 * sizeof( float ); // pose mats
iqmData->poses = (iqmTransform_t*)dataPtr;
dataPtr += header->num_poses * header->num_frames * sizeof(iqmTransform_t); // pose transforms
}
if( header->ofs_bounds ) {
iqmData->bounds = (float*)dataPtr;
@@ -804,22 +847,23 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na
iqmData->jointParents[i] = joint->parent;
}

// calculate joint matrices and their inverses
// joint inverses are needed only until the pose matrices are calculated
mat = iqmData->jointMats;
matInv = jointInvMats;
// calculate bind joint matrices and their inverses
mat = iqmData->bindJoints;
matInv = iqmData->invBindJoints;
joint = (iqmJoint_t *)((byte *)header + header->ofs_joints);
for( i = 0; i < header->num_joints; i++, joint++ ) {
float baseFrame[12], invBaseFrame[12];

QuatNormalize2( joint->rotate, joint->rotate );

JointToMatrix( joint->rotate, joint->scale, joint->translate, baseFrame );
Matrix34Invert( baseFrame, invBaseFrame );

if ( joint->parent >= 0 )
{
Matrix34Multiply( iqmData->jointMats + 12 * joint->parent, baseFrame, mat );
Matrix34Multiply( iqmData->bindJoints + 12 * joint->parent, baseFrame, mat );
mat += 12;
Matrix34Multiply( invBaseFrame, jointInvMats + 12 * joint->parent, matInv );
Matrix34Multiply( invBaseFrame, iqmData->invBindJoints + 12 * joint->parent, matInv );
matInv += 12;
}
else
@@ -834,16 +878,15 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na

if( header->num_poses )
{
// calculate pose matrices
// calculate pose transforms
transform = iqmData->poses;
framedata = (unsigned short *)((byte *)header + header->ofs_frames);
mat = iqmData->poseMats;
for( i = 0; i < header->num_frames; i++ ) {
pose = (iqmPose_t *)((byte *)header + header->ofs_poses);
for( j = 0; j < header->num_poses; j++, pose++ ) {
for( j = 0; j < header->num_poses; j++, pose++, transform++ ) {
vec3_t translate;
vec4_t rotate;
quat_t rotate;
vec3_t scale;
float mat1[12], mat2[12];

translate[0] = pose->channeloffset[0];
if( pose->mask & 0x001)
@@ -878,18 +921,9 @@ qboolean R_LoadIQM( model_t *mod, void *buffer, int filesize, const char *mod_na
if( pose->mask & 0x200)
scale[2] += *framedata++ * pose->channelscale[9];

// construct transformation matrix
JointToMatrix( rotate, scale, translate, mat1 );

if( pose->parent >= 0 ) {
Matrix34Multiply( iqmData->jointMats + 12 * pose->parent,
mat1, mat2 );
} else {
Com_Memcpy( mat2, mat1, sizeof(mat1) );
}

Matrix34Multiply( mat2, jointInvMats + 12 * j, mat );
mat += 12;
VectorCopy( translate, transform->translate );
QuatNormalize2( rotate, transform->rotate );
VectorCopy( scale, transform->scale );
}
}
}
@@ -1128,37 +1162,59 @@ void R_AddIQMSurfaces( trRefEntity_t *ent ) {


static void ComputePoseMats( iqmData_t *data, int frame, int oldframe,
float backlerp, float *mat ) {
float *mat1, *mat2;
int *joint = data->jointParents;
int i;

float backlerp, float *poseMats ) {
iqmTransform_t relativeJoints[IQM_MAX_JOINTS];
iqmTransform_t *relativeJoint;
const iqmTransform_t *pose;
const iqmTransform_t *oldpose;
const int *jointParent;
const float *invBindMat;
float *poseMat, lerp;
int i;

relativeJoint = relativeJoints;

// copy or lerp animation frame pose
if ( oldframe == frame ) {
mat1 = data->poseMats + 12 * data->num_poses * frame;
for( i = 0; i < data->num_poses; i++, joint++ ) {
if( *joint >= 0 ) {
Matrix34Multiply( mat + 12 * *joint,
mat1 + 12*i, mat + 12*i );
} else {
Com_Memcpy( mat + 12*i, mat1 + 12*i, 12 * sizeof(float) );
}
pose = &data->poses[frame * data->num_poses];
for ( i = 0; i < data->num_poses; i++, pose++, relativeJoint++ ) {
VectorCopy( pose->translate, relativeJoint->translate );
QuatCopy( pose->rotate, relativeJoint->rotate );
VectorCopy( pose->scale, relativeJoint->scale );
}
} else {
mat1 = data->poseMats + 12 * data->num_poses * frame;
mat2 = data->poseMats + 12 * data->num_poses * oldframe;

for( i = 0; i < data->num_poses; i++, joint++ ) {
if( *joint >= 0 ) {
float tmpMat[12];
InterpolateMatrix( mat1 + 12*i, mat2 + 12*i,
backlerp, tmpMat );
Matrix34Multiply( mat + 12 * *joint,
tmpMat, mat + 12*i );

} else {
InterpolateMatrix( mat1 + 12*i, mat2 + 12*i,
backlerp, mat + 12*i );
}
} else {
lerp = 1.0f - backlerp;
pose = &data->poses[frame * data->num_poses];
oldpose = &data->poses[oldframe * data->num_poses];
for ( i = 0; i < data->num_poses; i++, oldpose++, pose++, relativeJoint++ ) {
relativeJoint->translate[0] = oldpose->translate[0] * backlerp + pose->translate[0] * lerp;
relativeJoint->translate[1] = oldpose->translate[1] * backlerp + pose->translate[1] * lerp;
relativeJoint->translate[2] = oldpose->translate[2] * backlerp + pose->translate[2] * lerp;

relativeJoint->scale[0] = oldpose->scale[0] * backlerp + pose->scale[0] * lerp;
relativeJoint->scale[1] = oldpose->scale[1] * backlerp + pose->scale[1] * lerp;
relativeJoint->scale[2] = oldpose->scale[2] * backlerp + pose->scale[2] * lerp;

QuatSlerp( oldpose->rotate, pose->rotate, lerp, relativeJoint->rotate );
}
}

// multiply by inverse of bind pose and parent 'pose mat' (bind pose transform matrix)
relativeJoint = relativeJoints;
jointParent = data->jointParents;
invBindMat = data->invBindJoints;
poseMat = poseMats;
for ( i = 0; i < data->num_poses; i++, relativeJoint++, jointParent++, invBindMat += 12, poseMat += 12 ) {
float mat1[12], mat2[12];

JointToMatrix( relativeJoint->rotate, relativeJoint->scale, relativeJoint->translate, mat1 );

if ( *jointParent >= 0 ) {
Matrix34Multiply( &data->bindJoints[(*jointParent)*12], mat1, mat2 );
Matrix34Multiply( mat2, invBindMat, mat1 );
Matrix34Multiply( &poseMats[(*jointParent)*12], mat1, poseMat );
} else {
Matrix34Multiply( mat1, invBindMat, poseMat );
}
}
}
@@ -1169,7 +1225,7 @@ static void ComputeJointMats( iqmData_t *data, int frame, int oldframe,
int i;

if ( data->num_poses == 0 ) {
Com_Memcpy( mat, data->jointMats, data->num_joints * 12 * sizeof(float) );
Com_Memcpy( mat, data->bindJoints, data->num_joints * 12 * sizeof(float) );
return;
}

@@ -1181,7 +1237,7 @@ static void ComputeJointMats( iqmData_t *data, int frame, int oldframe,

Com_Memcpy(outmat, mat1, sizeof(outmat));

Matrix34Multiply( outmat, data->jointMats + 12*i, mat1 );
Matrix34Multiply( outmat, data->bindJoints + 12*i, mat1 );
}
}

Oops, something went wrong.

0 comments on commit d404519

Please sign in to comment.
You can’t perform that action at this time.