static int test_norm_function( normal_func func, int mtype,
int masked, long *cycles )
{
GLvector3f source[1], dest[1], dest2[1], ref[1], ref2[1];
GLmatrix mat[1];
GLfloat s[TEST_COUNT][5], d[TEST_COUNT][3], r[TEST_COUNT][3];
GLfloat d2[TEST_COUNT][3], r2[TEST_COUNT][3], length[TEST_COUNT];
GLfloat scale;
GLfloat *m;
GLubyte mask[TEST_COUNT];
int i, j;
#ifdef RUN_XFORM_BENCHMARK
int cycle_i; /* the counter for the benchmarks we run */
#endif
(void) cycles;
mat->m = (GLfloat *) ALIGN_MALLOC( 16 * sizeof(GLfloat), 16 );
mat->inv = m = mat->m;
m[0] = 63.0; m[4] = 43.0; m[ 8] = 29.0; m[12] = 43.0;
m[1] = 55.0; m[5] = 17.0; m[ 9] = 31.0; m[13] = 7.0;
m[2] = 44.0; m[6] = 9.0; m[10] = 7.0; m[14] = 3.0;
m[3] = 11.0; m[7] = 23.0; m[11] = 91.0; m[15] = 9.0;
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:
abort();
}
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
mask[i] = i % 2; /* mask every 2nd element */
d[i][0] = s[i][0] = d2[i][0] = 0.0;
d[i][1] = s[i][1] = d2[i][1] = 0.0;
d[i][2] = s[i][2] = d2[i][2] = 0.0;
for ( j = 0 ; j < 3 ; j++ )
s[i][j] = rnd();
length[i] = 1 / sqrt( s[i][0]*s[i][0] +
s[i][1]*s[i][1] +
s[i][2]*s[i][2] );
}
source->data = (GLfloat(*)[3])s;
source->start = (GLfloat *)s;
source->count = TEST_COUNT;
source->stride = sizeof(s[0]);
source->flags = 0;
dest->data = (GLfloat(*)[3])d;
dest->start = (GLfloat *)d;
dest->count = TEST_COUNT;
dest->stride = sizeof(float[3]);
dest->flags = 0;
dest2->data = (GLfloat(*)[3])d2;
dest2->start = (GLfloat *)d2;
dest2->count = TEST_COUNT;
dest2->stride = sizeof(float[3]);
dest2->flags = 0;
ref->data = (GLfloat(*)[3])r;
ref->start = (GLfloat *)r;
ref->count = TEST_COUNT;
ref->stride = sizeof(float[3]);
ref->flags = 0;
ref2->data = (GLfloat(*)[3])r2;
ref2->start = (GLfloat *)r2;
ref2->count = TEST_COUNT;
ref2->stride = sizeof(float[3]);
ref2->flags = 0;
if ( norm_normalize_types[mtype] == 0 ) {
ref_norm_transform_rescale( mat, scale, source, NULL, NULL, ref );
} else {
ref_norm_transform_normalize( mat, scale, source, NULL, NULL, ref );
ref_norm_transform_normalize( mat, scale, source, length, NULL, ref2 );
}
if ( mesa_profile ) {
if ( masked ) {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, mask, dest );
END_RACE( *cycles );
func( mat, scale, source, length, mask, dest2 );
} else {
BEGIN_RACE( *cycles );
func( mat, scale, source, NULL, NULL, dest );
END_RACE( *cycles );
func( mat, scale, source, length, NULL, dest2 );
}
} else {
if ( masked ) {
func( mat, scale, source, NULL, mask, dest );
func( mat, scale, source, length, mask, dest2 );
} else {
func( mat, scale, source, NULL, NULL, dest );
func( mat, scale, source, length, NULL, dest2 );
}
}
for ( i = 0 ; i < TEST_COUNT ; i++ ) {
if ( masked && !(mask[i] & 1) )
continue;
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;
}
}
}
}
ALIGN_FREE( mat->m );
return 1;
}
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;
}
