static void ref_norm_transform_normalize( const GLmatrix *mat,
GLfloat scale,
const GLvector4f *in,
const GLfloat *lengths,
GLvector4f *dest )
{
GLuint i;
const GLfloat *s = in->start;
const GLfloat *m = mat->inv;
GLfloat (*out)[4] = (GLfloat (*)[4]) dest->start;
for ( i = 0 ; i < in->count ; i++ ) {
GLfloat t[3];
TRANSFORM_NORMAL( t, s, m );
if ( !lengths ) {
GLfloat len = LEN_SQUARED_3FV( t );
if ( len > 1e-20 ) {
/* Hmmm, don't know how we could test the precalculated
* length case...
*/
scale = 1.0 / SQRTF( len );
SCALE_SCALAR_3V( out[i], scale, t );
} else {
out[i][0] = out[i][1] = out[i][2] = 0;
}
} else {
scale = lengths[i];;
SCALE_SCALAR_3V( out[i], scale, t );
}
s = (GLfloat *)((char *)s + in->stride);
}
}
static int test_norm_function( normal_func func, int mtype, long *cycles )
{
GLvector4f source[1], dest[1], dest2[1], ref[1], ref2[1];
GLmatrix mat[1];
GLfloat s[TEST_COUNT][5], d[TEST_COUNT][4], r[TEST_COUNT][4];
GLfloat d2[TEST_COUNT][4], r2[TEST_COUNT][4], length[TEST_COUNT];
GLfloat scale;
GLfloat *m;
int i, j;
#ifdef RUN_DEBUG_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
mat->m = (GLfloat *) _mesa_align_malloc( 16 * sizeof(GLfloat), 16 );
mat->inv = m = mat->m;
init_matrix( m );
scale = 1.0F + rnd () * norm_scale_types[mtype];
for ( i = 0 ; i < 4 ; i++ ) {
for ( j = 0 ; j < 4 ; j++ ) {
switch ( norm_templates[mtype][i * 4 + j] ) {
case NIL:
m[j * 4 + i] = 0.0;
break;
case ONE:
m[j * 4 + i] = 1.0;
break;
case NEG:
m[j * 4 + i] = -1.0;
break;
case VAR:
break;
default:
exit(1);
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
ASSIGN_3V( d[i], 0.0, 0.0, 0.0 );
ASSIGN_3V( s[i], 0.0, 0.0, 0.0 );
ASSIGN_3V( d2[i], 0.0, 0.0, 0.0 );
for ( j = 0 ; j < 3 ; j++ )
s[i][j] = rnd();
length[i] = 1 / SQRTF( LEN_SQUARED_3FV( s[i] ) );
}
source->data = (GLfloat(*)[4]) s;
source->start = (GLfloat *) s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->flags = 0;
dest->data = d;
dest->start = (GLfloat *) d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[4]);
dest->flags = 0;
dest2->data = d2;
dest2->start = (GLfloat *) d2;
dest2->count = TEST_COUNT;
dest2->stride = sizeof(float[4]);
dest2->flags = 0;
ref->data = r;
ref->start = (GLfloat *) r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[4]);
ref->flags = 0;
ref2->data = r2;
ref2->start = (GLfloat *) r2;
ref2->count = TEST_COUNT;
ref2->stride = sizeof(float[4]);
ref2->flags = 0;
if ( norm_normalize_types[mtype] == 0 ) {
ref_norm_transform_rescale( mat, scale, source, NULL, ref );
} else {
ref_norm_transform_normalize( mat, scale, source, NULL, ref );
ref_norm_transform_normalize( mat, scale, source, length, ref2 );
}
if ( mesa_profile ) {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, dest );
END_RACE( *cycles );
func( mat, scale, source, length, dest2 );
} else {
func( mat, scale, source, NULL, dest );
func( mat, scale, source, length, dest2 );
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
for ( j = 0 ; j < 3 ; j++ ) {
if ( significand_match( d[i][j], r[i][j] ) < REQUIRED_PRECISION ) {
printf( "-----------------------------\n" );
printf( "(i = %i, j = %i)\n", i, j );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][0], r[i][0], r[i][0]/d[i][0],
MAX_PRECISION - significand_match( d[i][0], r[i][0] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][1], r[i][1], r[i][1]/d[i][1],
MAX_PRECISION - significand_match( d[i][1], r[i][1] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d[i][2], r[i][2], r[i][2]/d[i][2],
MAX_PRECISION - significand_match( d[i][2], r[i][2] ) );
return 0;
}
if ( norm_normalize_types[mtype] != 0 ) {
if ( significand_match( d2[i][j], r2[i][j] ) < REQUIRED_PRECISION ) {
printf( "------------------- precalculated length case ------\n" );
printf( "(i = %i, j = %i)\n", i, j );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][0], r2[i][0], r2[i][0]/d2[i][0],
MAX_PRECISION - significand_match( d2[i][0], r2[i][0] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][1], r2[i][1], r2[i][1]/d2[i][1],
MAX_PRECISION - significand_match( d2[i][1], r2[i][1] ) );
printf( "%f \t %f \t [ratio = %e - %i bit missed]\n",
d2[i][2], r2[i][2], r2[i][2]/d2[i][2],
MAX_PRECISION - significand_match( d2[i][2], r2[i][2] ) );
return 0;
}
}
}
}
_mesa_align_free( mat->m );
return 1;
}
/**
* Determine type and flags from scratch.
*
* \param mat matrix.
*
* This is expensive enough to only want to do it once.
*/
static void analyse_from_scratch( GLmatrix *mat )
{
const GLfloat *m = mat->m;
GLuint mask = 0;
GLuint i;
for (i = 0 ; i < 16 ; i++) {
if (m[i] == 0.0) mask |= (1<<i);
}
if (m[0] == 1.0F) mask |= (1<<16);
if (m[5] == 1.0F) mask |= (1<<21);
if (m[10] == 1.0F) mask |= (1<<26);
if (m[15] == 1.0F) mask |= (1<<31);
mat->flags &= ~MAT_FLAGS_GEOMETRY;
/* Check for translation - no-one really cares
*/
if ((mask & MASK_NO_TRX) != MASK_NO_TRX)
mat->flags |= MAT_FLAG_TRANSLATION;
/* Do the real work
*/
if (mask == (GLuint) MASK_IDENTITY) {
mat->type = MATRIX_IDENTITY;
}
else if ((mask & MASK_2D_NO_ROT) == (GLuint) MASK_2D_NO_ROT) {
mat->type = MATRIX_2D_NO_ROT;
if ((mask & MASK_NO_2D_SCALE) != MASK_NO_2D_SCALE)
mat->flags |= MAT_FLAG_GENERAL_SCALE;
}
else if ((mask & MASK_2D) == (GLuint) MASK_2D) {
GLfloat mm = DOT2(m, m);
GLfloat m4m4 = DOT2(m+4,m+4);
GLfloat mm4 = DOT2(m,m+4);
mat->type = MATRIX_2D;
/* Check for scale */
if (SQ(mm-1) > SQ(1e-6) ||
SQ(m4m4-1) > SQ(1e-6))
mat->flags |= MAT_FLAG_GENERAL_SCALE;
/* Check for rotation */
if (SQ(mm4) > SQ(1e-6))
mat->flags |= MAT_FLAG_GENERAL_3D;
else
mat->flags |= MAT_FLAG_ROTATION;
}
else if ((mask & MASK_3D_NO_ROT) == (GLuint) MASK_3D_NO_ROT) {
mat->type = MATRIX_3D_NO_ROT;
/* Check for scale */
if (SQ(m[0]-m[5]) < SQ(1e-6) &&
SQ(m[0]-m[10]) < SQ(1e-6)) {
if (SQ(m[0]-1.0) > SQ(1e-6)) {
mat->flags |= MAT_FLAG_UNIFORM_SCALE;
}
}
else {
mat->flags |= MAT_FLAG_GENERAL_SCALE;
}
}
else if ((mask & MASK_3D) == (GLuint) MASK_3D) {
GLfloat c1 = DOT3(m,m);
GLfloat c2 = DOT3(m+4,m+4);
GLfloat c3 = DOT3(m+8,m+8);
GLfloat d1 = DOT3(m, m+4);
GLfloat cp[3];
mat->type = MATRIX_3D;
/* Check for scale */
if (SQ(c1-c2) < SQ(1e-6) && SQ(c1-c3) < SQ(1e-6)) {
if (SQ(c1-1.0) > SQ(1e-6))
mat->flags |= MAT_FLAG_UNIFORM_SCALE;
/* else no scale at all */
}
else {
mat->flags |= MAT_FLAG_GENERAL_SCALE;
}
/* Check for rotation */
if (SQ(d1) < SQ(1e-6)) {
CROSS3( cp, m, m+4 );
SUB_3V( cp, cp, (m+8) );
if (LEN_SQUARED_3FV(cp) < SQ(1e-6))
mat->flags |= MAT_FLAG_ROTATION;
else
mat->flags |= MAT_FLAG_GENERAL_3D;
}
else {
mat->flags |= MAT_FLAG_GENERAL_3D; /* shear, etc */
}
}
else if ((mask & MASK_PERSPECTIVE) == MASK_PERSPECTIVE && m[11]==-1.0F) {
mat->type = MATRIX_PERSPECTIVE;
mat->flags |= MAT_FLAG_GENERAL;
}
else {
mat->type = MATRIX_GENERAL;
mat->flags |= MAT_FLAG_GENERAL;
}
}
