int Test_3x3mulM1M2(void)
{
// Init an array flanked by guard pages
btMatrix3x3 in1[ARRAY_SIZE];
btMatrix3x3 in2[ARRAY_SIZE];
btMatrix3x3 out[ARRAY_SIZE];
btMatrix3x3 out2[ARRAY_SIZE];
// Init the data
size_t i, j;
for( i = 0; i < ARRAY_SIZE; i++ )
{
in1[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4() );
in2[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4() );
out[i] = M3x3mulM1M2_ref(in1[i], in2[i]);
out2[i] = (in1[i] * in2[i]);
if( out[i] != out2[i] )
{
vlog( "Error - M3x3mulM1M2 result error! ");
vlog( "failure @ %ld\n", i);
btVector3 m0, m1, m2;
m0 = out[i].getRow(0);
m1 = out[i].getRow(1);
m2 = out[i].getRow(2);
vlog( "\ncorrect = (%10.4f, %10.4f, %10.4f, %10.4f) "
"\n (%10.4f, %10.4f, %10.4f, %10.4f) "
"\n (%10.4f, %10.4f, %10.4f, %10.4f) \n",
m0.m_floats[0], m0.m_floats[1], m0.m_floats[2], m0.m_floats[3],
m1.m_floats[0], m1.m_floats[1], m1.m_floats[2], m1.m_floats[3],
m2.m_floats[0], m2.m_floats[1], m2.m_floats[2], m2.m_floats[3]);
m0 = out2[i].getRow(0);
m1 = out2[i].getRow(1);
m2 = out2[i].getRow(2);
vlog( "\ntested = (%10.4f, %10.4f, %10.4f, %10.4f) "
"\n (%10.4f, %10.4f, %10.4f, %10.4f) "
"\n (%10.4f, %10.4f, %10.4f, %10.4f) \n",
m0.m_floats[0], m0.m_floats[1], m0.m_floats[2], m0.m_floats[3],
m1.m_floats[0], m1.m_floats[1], m1.m_floats[2], m1.m_floats[3],
m2.m_floats[0], m2.m_floats[1], m2.m_floats[2], m2.m_floats[3]);
return -1;
}
}
uint64_t scalarTime, vectorTime;
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < LOOPCOUNT; j++)
{
startTime = ReadTicks();
for( i = 0; i < ARRAY_SIZE; i++ )
out[i] = M3x3mulM1M2_ref(in1[i], in2[i]);
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= LOOPCOUNT;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < LOOPCOUNT; j++)
{
startTime = ReadTicks();
for( i = 0; i < ARRAY_SIZE; i++ )
out2[i] = (in1[i] * in2[i]);
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= LOOPCOUNT;
vlog( "Timing:\n" );
vlog( "\t scalar\t vector\n" );
vlog( "\t%10.2f\t%10.2f\n", TicksToCycles( scalarTime ) / ARRAY_SIZE, TicksToCycles( vectorTime ) / ARRAY_SIZE );
return 0;
}
int Test_3x3getRot(void)
{
// Init an array flanked by guard pages
btMatrix3x3 in1[ARRAY_SIZE];
btQuaternion out[ARRAY_SIZE];
btQuaternion out2[ARRAY_SIZE];
// Init the data
size_t i, j;
for (i = 0; i < ARRAY_SIZE; i++)
{
in1[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4());
out[i] = btQuaternion(qtNAN_f4());
out2[i] = btQuaternion(qtNAN_f4());
M3x3getRot_ref(in1[i], out[i]);
in1[i].getRotation(out2[i]);
if (out[i] != out2[i])
{
vlog("Error - M3x3getRot result error! ");
vlog("failure @ %ld\n", i);
vlog(
"\ncorrect = (%10.7f, %10.7f, %10.7f, %10.7f) "
"\ntested = (%10.7f, %10.7f, %10.7f, %10.7f) \n",
out[i].x(), out[i].y(), out[i].z(), out[i].w(),
out2[i].x(), out2[i].y(), out2[i].z(), out2[i].w());
return -1;
}
}
uint64_t scalarTime, vectorTime;
uint64_t startTime, bestTime, currentTime;
bestTime = ~(bestTime & 0); //-1ULL;
scalarTime = 0;
for (j = 0; j < LOOPCOUNT; j++)
{
startTime = ReadTicks();
for (i = 0; i < ARRAY_SIZE; i++)
M3x3getRot_ref(in1[i], out[i]);
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
scalarTime = bestTime;
else
scalarTime /= LOOPCOUNT;
bestTime = ~(bestTime & 0); //-1ULL;
vectorTime = 0;
for (j = 0; j < LOOPCOUNT; j++)
{
startTime = ReadTicks();
for (i = 0; i < ARRAY_SIZE; i++)
{
in1[i].getRotation(out2[i]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
vectorTime = bestTime;
else
vectorTime /= LOOPCOUNT;
vlog("Timing:\n");
vlog("\t scalar\t vector\n");
vlog("\t%10.2f\t%10.2f\n", TicksToCycles(scalarTime) / ARRAY_SIZE, TicksToCycles(vectorTime) / ARRAY_SIZE);
return 0;
}
int Test_v3norm(void)
{
btVector3 v1, v2;
float x,y,z,w;
// Init the data
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = BT_NAN; // w channel NaN
v1.setValue(x,y,z);
v1.setW(w);
v2 = v1;
btVector3 correct_res, test_res;
{
float vNaN = BT_NAN;
correct_res.setValue(vNaN, vNaN, vNaN);
test_res.setValue(vNaN, vNaN, vNaN);
correct_res = v3norm_ref(v1);
test_res = v2.normalize();
if( fabs(correct_res.m_floats[0] - test_res.m_floats[0]) +
fabs(correct_res.m_floats[1] - test_res.m_floats[1]) +
fabs(correct_res.m_floats[2] - test_res.m_floats[2]) > FLT_EPSILON * 4)
{
vlog( "Error - v3norm result error! "
"\ncorrect = (%10.4f, %10.4f, %10.4f) "
"\ntested = (%10.4f, %10.4f, %10.4f) \n",
correct_res.m_floats[0], correct_res.m_floats[1], correct_res.m_floats[2],
test_res.m_floats[0], test_res.m_floats[1], test_res.m_floats[2]);
return 1;
}
}
#define DATA_SIZE LOOPCOUNT
btVector3 vec3_arr0[DATA_SIZE];
btVector3 vec3_arr1[DATA_SIZE];
uint64_t scalarTime;
uint64_t vectorTime;
size_t j, k;
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr1[k].setValue(x,y,z);
vec3_arr1[k].setW(w);
}
startTime = ReadTicks();
for( k = 0; k+4 <= LOOPCOUNT; k+=4 )
{
vec3_arr0[k] = v3norm_ref(vec3_arr1[k]);
vec3_arr0[k+1] = v3norm_ref(vec3_arr1[k+1]);
vec3_arr0[k+2] = v3norm_ref(vec3_arr1[k+2]);
vec3_arr0[k+3] = v3norm_ref(vec3_arr1[k+3]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr1[k].setValue(x,y,z);
vec3_arr1[k].setW(w);
}
startTime = ReadTicks();
for( k = 0; k+4 <= LOOPCOUNT; k+=4 )
{
vec3_arr0[k] = vec3_arr1[k].normalize();
vec3_arr0[k+1] = vec3_arr1[k+1].normalize();
vec3_arr0[k+2] = vec3_arr1[k+2].normalize();
vec3_arr0[k+3] = vec3_arr1[k+3].normalize();
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT,
TicksToCycles( vectorTime ) / LOOPCOUNT );
return 0;
}
int Test_mindot(void)
{
// Init an array flanked by guard pages
btSimdFloat4 *data = (btSimdFloat4 *)GuardCalloc(1, MAX_SIZE * sizeof(btSimdFloat4), NULL);
float *fp = (float *)data;
long correct, test;
btVector3 localScaling(0.1f, 0.2f, 0.3f);
size_t size;
// Init the data
size_t i;
for (i = 0; i < MAX_SIZE; i++)
{
fp[4 * i] = (int32_t)RANDF_16;
fp[4 * i + 1] = (int32_t)RANDF_16;
fp[4 * i + 2] = (int32_t)RANDF_16;
fp[4 * i + 3] = BT_NAN; // w channel NaN
}
float correctDot, testDot;
fp = (float *)localScaling;
float maxRelativeError = 0.f;
for (size = 1; size <= MAX_SIZE; size++)
{
float *in = (float *)(data + MAX_SIZE - size);
size_t position;
for (position = 0; position < size; position++)
{
float *biggest = in + position * 4;
float old[4] = {biggest[0], biggest[1], biggest[2], biggest[3]};
biggest[0] -= LARGE_FLOAT17;
biggest[1] -= LARGE_FLOAT17;
biggest[2] -= LARGE_FLOAT17;
biggest[3] -= LARGE_FLOAT17;
correctDot = BT_NAN;
testDot = BT_NAN;
correct = mindot_ref((btSimdFloat4 *)in, (float *)&localScaling, size, &correctDot);
test = localScaling.minDot((btVector3 *)in, size, testDot);
if (test < 0 || test >= size)
{
vlog("Error @ %ld: index out of bounds! *%ld vs %ld \n", size, correct, test);
continue;
}
if (correct != test)
{
vlog("Error @ %ld: index misreported! *%ld vs %ld (*%f, %f)\n", size, correct, test,
fp[0] * in[4 * correct] + fp[1] * in[4 * correct + 1] + fp[2] * in[4 * correct + 2],
fp[0] * in[4 * test] + fp[1] * in[4 * test + 1] + fp[2] * in[4 * test + 2]);
return 1;
}
if (test != position)
{
vlog("Biggest not found where it is supposed to be: *%ld vs %ld (*%f, %f)\n", position, test,
fp[0] * in[4 * test] + fp[1] * in[4 * test + 1] + fp[2] * in[4 * test + 2],
fp[0] * in[4 * position] + fp[1] * in[4 * position + 1] + fp[2] * in[4 * position + 2]);
return 1;
}
if (correctDot != testDot)
{
float relativeError = btFabs((testDot - correctDot) / correctDot);
if (relativeError > 1e6)
{
vlog("Error @ %ld: dotpr misreported! *%f vs %f (*%f, %f)\n", size, correctDot, testDot,
fp[0] * in[4 * correct] + fp[1] * in[4 * correct + 1] + fp[2] * in[4 * correct + 2],
fp[0] * in[4 * test] + fp[1] * in[4 * test + 1] + fp[2] * in[4 * test + 2]);
return 1;
}
else
{
if (maxRelativeError < relativeError)
{
maxRelativeError = relativeError;
}
}
}
memcpy(biggest, old, 16);
}
}
if (maxRelativeError)
{
printf("Warning: relative error = %e\n", maxRelativeError);
}
uint64_t scalarTimes[33 + (MAX_LOG2_SIZE - 5)];
uint64_t vectorTimes[33 + (MAX_LOG2_SIZE - 5)];
size_t j, k;
float *in = (float *)data;
for (size = 1; size <= 32; size++)
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTimes[size] = 0;
for (j = 0; j < 100; j++)
{
startTime = ReadTicks();
for (k = 0; k < LOOPCOUNT; k++)
correct += mindot_ref((btSimdFloat4 *)in, (float *)&localScaling, size, &correctDot);
currentTime = ReadTicks() - startTime;
scalarTimes[size] += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
scalarTimes[size] = bestTime;
else
scalarTimes[size] /= 100;
}
uint64_t *timep = &scalarTimes[33];
for (size = 64; size <= MAX_SIZE; size *= 2)
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
timep[0] = 0;
for (j = 0; j < 100; j++)
{
startTime = ReadTicks();
for (k = 0; k < LOOPCOUNT; k++)
correct += mindot_ref((btSimdFloat4 *)in, (float *)&localScaling, size, &correctDot);
currentTime = ReadTicks() - startTime;
timep[0] += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
timep[0] = bestTime;
else
timep[0] /= 100;
timep++;
}
for (size = 1; size <= 32; size++)
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTimes[size] = 0;
for (j = 0; j < 100; j++)
{
startTime = ReadTicks();
for (k = 0; k < LOOPCOUNT; k++)
test += localScaling.minDot((btVector3 *)in, size, testDot);
currentTime = ReadTicks() - startTime;
vectorTimes[size] += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
vectorTimes[size] = bestTime;
else
vectorTimes[size] /= 100;
}
timep = &vectorTimes[33];
for (size = 64; size <= MAX_SIZE; size *= 2)
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
timep[0] = 0;
for (j = 0; j < 100; j++)
{
startTime = ReadTicks();
for (k = 0; k < LOOPCOUNT; k++)
test += localScaling.minDot((btVector3 *)in, size, testDot);
currentTime = ReadTicks() - startTime;
timep[0] += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
timep[0] = bestTime;
else
timep[0] /= 100;
timep++;
}
vlog("Timing:\n");
vlog(" size\t scalar\t vector\n");
for (size = 1; size <= 32; size++)
vlog("%5lu\t%10.2f\t%10.2f\n", size, TicksToCycles(scalarTimes[size]) / LOOPCOUNT, TicksToCycles(vectorTimes[size]) / LOOPCOUNT);
size_t index = 33;
for (size = 64; size <= MAX_SIZE; size *= 2)
{
vlog("%5lu\t%10.2f\t%10.2f\n", size, TicksToCycles(scalarTimes[index]) / LOOPCOUNT, TicksToCycles(vectorTimes[index]) / LOOPCOUNT);
index++;
}
// Useless check to make sure that the timing loops are not optimized away
if (test != correct)
vlog("Error: Test != correct: *%ld vs. %ld\n", correct, test);
GuardFree(data);
return 0;
}
int Test_v3triple(void)
{
btVector3 v1, v2, v3;
float x,y,z,w;
// Init the data
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = BT_NAN; // w channel NaN
v1.setValue(x,y,z);
v1.setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v2.setValue(x,y,z);
v2.setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v3.setValue(x,y,z);
v3.setW(w);
float correctTriple0, testTriple0;
{
correctTriple0 = w;
testTriple0 = w;
testTriple0 = v3triple_ref(v1,v2,v3);
correctTriple0 = v1.triple(v2, v3);
if( fabsf(correctTriple0 - testTriple0) > FLT_EPSILON * 4 )
{
vlog( "Error - v3triple result error! %f != %f \n", correctTriple0, testTriple0);
return 1;
}
}
#define DATA_SIZE 1024
btVector3 v3_arr1[DATA_SIZE];
btVector3 v3_arr2[DATA_SIZE];
btVector3 v3_arr3[DATA_SIZE];
btScalar res_arr[DATA_SIZE];
uint64_t scalarTime;
uint64_t vectorTime;
size_t j, k;
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v3_arr1[k].setValue(x,y,z);
v3_arr1[k].setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v3_arr2[k].setValue(x,y,z);
v3_arr2[k].setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v3_arr3[k].setValue(x,y,z);
v3_arr3[k].setW(w);
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = uint64_t(-1LL);
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for( k = 0; k+4 <= LOOPCOUNT; k+=4 )
{
size_t k32 = (k & (DATA_SIZE-1));
res_arr[k32] = v3triple_ref( v3_arr1[k32], v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3triple_ref( v3_arr1[k32], v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3triple_ref( v3_arr1[k32], v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3triple_ref( v3_arr1[k32], v3_arr2[k32], v3_arr3[k32]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = uint64_t(-1LL);
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for( k = 0; k+4 <= LOOPCOUNT; k+=4 )
{
size_t k32 = k & (DATA_SIZE -1);
res_arr[k32] = v3_arr1[k32].triple(v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3_arr1[k32].triple(v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3_arr1[k32].triple(v3_arr2[k32], v3_arr3[k32]); k32++;
res_arr[k32] = v3_arr1[k32].triple(v3_arr2[k32], v3_arr3[k32]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT, TicksToCycles( vectorTime ) / LOOPCOUNT );
return 0;
}
int Test_3x3timesTranspose(void)
{
// Init an array flanked by guard pages
btMatrix3x3 in1[ARRAY_SIZE];
btMatrix3x3 in2[ARRAY_SIZE];
btMatrix3x3 out[ARRAY_SIZE];
btMatrix3x3 out2[ARRAY_SIZE];
// Init the data
size_t i, j;
for( i = 0; i < ARRAY_SIZE; i++ )
{
in1[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4() );
in2[i] = btMatrix3x3(rand_f4(), rand_f4(), rand_f4() );
out[i] = timesTranspose(in1[i], in2[i]);
out2[i] = in1[i].timesTranspose(in2[i]);
if( out[i] != out2[i] )
{
printf( "failure @ %ld\n", i);
return -1;
}
}
uint64_t scalarTime, vectorTime;
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < LOOPCOUNT; j++) {
startTime = ReadTicks();
for( i = 0; i < ARRAY_SIZE; i++ )
out[i] = timesTranspose(in1[i], in2[i]);
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= LOOPCOUNT;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < LOOPCOUNT; j++) {
startTime = ReadTicks();
for( i = 0; i < ARRAY_SIZE; i++ )
out[i] = in1[i].timesTranspose(in2[i]);
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= LOOPCOUNT;
vlog( "Timing:\n" );
vlog( "\t scalar\t vector\n" );
vlog( "\t%10.2f\t%10.2f\n", TicksToCycles( scalarTime ) / ARRAY_SIZE, TicksToCycles( vectorTime ) / ARRAY_SIZE );
return 0;
}
int Test_btDbvt(void)
{
btDbvtAabbMm a[DATA_SIZE], b[DATA_SIZE], c[DATA_SIZE];
btDbvtAabbMm a_ref[DATA_SIZE], b_ref[DATA_SIZE], c_ref[DATA_SIZE];
int i;
bool Intersect_Test_Res[DATA_SIZE], Intersect_Ref_Res[DATA_SIZE];
int Select_Test_Res[DATA_SIZE], Select_Ref_Res[DATA_SIZE];
for (i = 0; i < DATA_SIZE; i++)
{
a[i].tMins().m_floats[0] = (float)rand() / (float)RAND_MAX;
a[i].tMins().m_floats[1] = (float)rand() / (float)RAND_MAX;
a[i].tMins().m_floats[2] = (float)rand() / (float)RAND_MAX;
a[i].tMins().m_floats[3] = (float)rand() / (float)RAND_MAX;
a[i].tMaxs().m_floats[0] = (float)rand() / (float)RAND_MAX;
a[i].tMaxs().m_floats[1] = (float)rand() / (float)RAND_MAX;
a[i].tMaxs().m_floats[2] = (float)rand() / (float)RAND_MAX;
a[i].tMaxs().m_floats[3] = (float)rand() / (float)RAND_MAX;
b[i].tMins().m_floats[0] = (float)rand() / (float)RAND_MAX;
b[i].tMins().m_floats[1] = (float)rand() / (float)RAND_MAX;
b[i].tMins().m_floats[2] = (float)rand() / (float)RAND_MAX;
b[i].tMins().m_floats[3] = (float)rand() / (float)RAND_MAX;
b[i].tMaxs().m_floats[0] = (float)rand() / (float)RAND_MAX;
b[i].tMaxs().m_floats[1] = (float)rand() / (float)RAND_MAX;
b[i].tMaxs().m_floats[2] = (float)rand() / (float)RAND_MAX;
b[i].tMaxs().m_floats[3] = (float)rand() / (float)RAND_MAX;
c[i].tMins().m_floats[0] = (float)rand() / (float)RAND_MAX;
c[i].tMins().m_floats[1] = (float)rand() / (float)RAND_MAX;
c[i].tMins().m_floats[2] = (float)rand() / (float)RAND_MAX;
c[i].tMins().m_floats[3] = (float)rand() / (float)RAND_MAX;
c[i].tMaxs().m_floats[0] = (float)rand() / (float)RAND_MAX;
c[i].tMaxs().m_floats[1] = (float)rand() / (float)RAND_MAX;
c[i].tMaxs().m_floats[2] = (float)rand() / (float)RAND_MAX;
c[i].tMaxs().m_floats[3] = (float)rand() / (float)RAND_MAX;
a_ref[i].tMins().m_floats[0] = a[i].tMins().m_floats[0];
a_ref[i].tMins().m_floats[1] = a[i].tMins().m_floats[1];
a_ref[i].tMins().m_floats[2] = a[i].tMins().m_floats[2];
a_ref[i].tMins().m_floats[3] = a[i].tMins().m_floats[3];
a_ref[i].tMaxs().m_floats[0] = a[i].tMaxs().m_floats[0];
a_ref[i].tMaxs().m_floats[1] = a[i].tMaxs().m_floats[1];
a_ref[i].tMaxs().m_floats[2] = a[i].tMaxs().m_floats[2];
a_ref[i].tMaxs().m_floats[3] = a[i].tMaxs().m_floats[3];
b_ref[i].tMins().m_floats[0] = b[i].tMins().m_floats[0];
b_ref[i].tMins().m_floats[1] = b[i].tMins().m_floats[1];
b_ref[i].tMins().m_floats[2] = b[i].tMins().m_floats[2];
b_ref[i].tMins().m_floats[3] = b[i].tMins().m_floats[3];
b_ref[i].tMaxs().m_floats[0] = b[i].tMaxs().m_floats[0];
b_ref[i].tMaxs().m_floats[1] = b[i].tMaxs().m_floats[1];
b_ref[i].tMaxs().m_floats[2] = b[i].tMaxs().m_floats[2];
b_ref[i].tMaxs().m_floats[3] = b[i].tMaxs().m_floats[3];
c_ref[i].tMins().m_floats[0] = c[i].tMins().m_floats[0];
c_ref[i].tMins().m_floats[1] = c[i].tMins().m_floats[1];
c_ref[i].tMins().m_floats[2] = c[i].tMins().m_floats[2];
c_ref[i].tMins().m_floats[3] = c[i].tMins().m_floats[3];
c_ref[i].tMaxs().m_floats[0] = c[i].tMaxs().m_floats[0];
c_ref[i].tMaxs().m_floats[1] = c[i].tMaxs().m_floats[1];
c_ref[i].tMaxs().m_floats[2] = c[i].tMaxs().m_floats[2];
c_ref[i].tMaxs().m_floats[3] = c[i].tMaxs().m_floats[3];
}
#if 1
for (i = 0; i < DATA_SIZE; i++)
{
Intersect_Test_Res[i] = Intersect(a[i], b[i]);
Intersect_Ref_Res[i] = Intersect_ref(a_ref[i], b_ref[i]);
if(Intersect_Test_Res[i] != Intersect_Ref_Res[i])
{
printf("Diff on %d\n", i);
printf("a_mx_f[0] = %.3f, a_mx_f[1] = %.3f, a_mx_f[2] = %.3f, a_mx_f[3] = %.3f\n", a[i].tMaxs().m_floats[0], a[i].tMaxs().m_floats[1], a[i].tMaxs().m_floats[2], a[i].tMaxs().m_floats[3]);
printf("a_mi_f[0] = %.3f, a_mi_f[1] = %.3f, a_mi_f[2] = %.3f, a_mi_f[3] = %.3f\n", a[i].tMins().m_floats[0], a[i].tMins().m_floats[1], a[i].tMins().m_floats[2], a[i].tMins().m_floats[3]);
printf("b_mx_f[0] = %.3f, b_mx_f[1] = %.3f, b_mx_f[2] = %.3f, b_mx_f[3] = %.3f\n", b[i].tMaxs().m_floats[0], b[i].tMaxs().m_floats[1], b[i].tMaxs().m_floats[2], b[i].tMaxs().m_floats[3]);
printf("b_mi_f[0] = %.3f, b_mi_f[1] = %.3f, b_mi_f[2] = %.3f, b_mi_f[3] = %.3f\n", b[i].tMins().m_floats[0], b[i].tMins().m_floats[1], b[i].tMins().m_floats[2], b[i].tMins().m_floats[3]);
printf("a_mx_f_ref[0] = %.3f, a_mx_f_ref[1] = %.3f, a_mx_f_ref[2] = %.3f, a_mx_f_ref[3] = %.3f\n", a_ref[i].tMaxs().m_floats[0], a_ref[i].tMaxs().m_floats[1], a_ref[i].tMaxs().m_floats[2], a_ref[i].tMaxs().m_floats[3]);
printf("a_mi_f_ref[0] = %.3f, a_mi_f_ref[1] = %.3f, a_mi_f_ref[2] = %.3f, a_mi_f_ref[3] = %.3f\n", a_ref[i].tMins().m_floats[0], a_ref[i].tMins().m_floats[1], a_ref[i].tMins().m_floats[2], a_ref[i].tMins().m_floats[3]);
printf("b_mx_f_ref[0] = %.3f, b_mx_f_ref[1] = %.3f, b_mx_f_ref[2] = %.3f, b_mx_f_ref[3] = %.3f\n", b_ref[i].tMaxs().m_floats[0], b_ref[i].tMaxs().m_floats[1], b_ref[i].tMaxs().m_floats[2], b_ref[i].tMaxs().m_floats[3]);
printf("b_mi_f_ref[0] = %.3f, b_mi_f_ref[1] = %.3f, b_mi_f_ref[2] = %.3f, b_mi_f_ref[3] = %.3f\n", b_ref[i].tMins().m_floats[0], b_ref[i].tMins().m_floats[1], b_ref[i].tMins().m_floats[2], b_ref[i].tMins().m_floats[3]);
}
}
#endif
uint64_t scalarTime;
uint64_t vectorTime;
size_t j;
////////////////////////////////////
//
// Time and Test Intersect
//
////////////////////////////////////
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Intersect_Ref_Res[i] = Intersect_ref(a_ref[i], b_ref[i]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Intersect_Test_Res[i] = Intersect(a[i], b[i]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Intersect Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT, TicksToCycles( vectorTime ) / LOOPCOUNT );
//printf("scalar = %llu, vector = %llu\n", scalarTime, vectorTime);
for (i = 0; i < DATA_SIZE; i++)
{
if(Intersect_Test_Res[i] != Intersect_Ref_Res[i])
{
printf("Intersect fail at %d\n", i);
return 1;
}
}
////////////////////////////////////
//
// Time and Test Merge
//
////////////////////////////////////
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Merge_ref(a_ref[i], b_ref[i], c_ref[i]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Merge(a[i], b[i], c[i]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Merge Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT, TicksToCycles( vectorTime ) / LOOPCOUNT );
//printf("scalar = %llu, vector = %llu\n", scalarTime, vectorTime);
/*
c [0] float32_t 0.00455523
[1] float32_t 0.559712
[2] float32_t 0.0795838
[3] float32_t 0.10182
c_ref
[0] float32_t 0.00455523
[1] float32_t 0.559712
[2] float32_t 0.0795838
[3] float32_t 0.552081
c [0] float32_t 0.829904
[1] float32_t 0.692891
[2] float32_t 0.961654
[3] float32_t 0.666956
c_ref
[0] float32_t 0.829904
[1] float32_t 0.692891
[2] float32_t 0.961654
[3] float32_t 0.522878
*/
for (i = 0; i < DATA_SIZE; i++)
{
//ignore 4th component because it is not computed in all code-paths
if( (fabs(c[i].tMaxs().m_floats[0] - c_ref[i].tMaxs().m_floats[0]) > 0.001) ||
(fabs(c[i].tMaxs().m_floats[1] - c_ref[i].tMaxs().m_floats[1]) > 0.001) ||
(fabs(c[i].tMaxs().m_floats[2] - c_ref[i].tMaxs().m_floats[2]) > 0.001) ||
// (fabs(c[i].tMaxs().m_floats[3] - c_ref[i].tMaxs().m_floats[3]) > 0.001) ||
(fabs(c[i].tMins().m_floats[0] - c_ref[i].tMins().m_floats[0]) > 0.001) ||
(fabs(c[i].tMins().m_floats[1] - c_ref[i].tMins().m_floats[1]) > 0.001) ||
(fabs(c[i].tMins().m_floats[2] - c_ref[i].tMins().m_floats[2]) > 0.001)
//|| (fabs(c[i].tMins().m_floats[3] - c_ref[i].tMins().m_floats[3]) > 0.001)
)
//if((c[i].tMaxs().m_floats[0] != c_ref[i].tMaxs().m_floats[0]) || (c[i].tMaxs().m_floats[1] != c_ref[i].tMaxs().m_floats[1]) || (c[i].tMaxs().m_floats[2] != c_ref[i].tMaxs().m_floats[2]) || (c[i].tMaxs().m_floats[3] != c_ref[i].tMaxs().m_floats[3]) || (c[i].tMins().m_floats[0] != c_ref[i].tMins().m_floats[0]) || (c[i].tMins().m_floats[1] != c_ref[i].tMins().m_floats[1]) || (c[i].tMins().m_floats[2] != c_ref[i].tMins().m_floats[2]) || (c[i].tMins().m_floats[3] != c_ref[i].tMins().m_floats[3]))
{
printf("Merge fail at %d with test = %d, ref = %d\n", i, Select_Test_Res[i], Select_Ref_Res[i]);
printf("a_mx_f[0] = %.3f, a_mx_f[1] = %.3f, a_mx_f[2] = %.3f, a_mx_f[3] = %.3f\n", a[i].tMaxs().m_floats[0], a[i].tMaxs().m_floats[1], a[i].tMaxs().m_floats[2], a[i].tMaxs().m_floats[3]);
printf("a_mi_f[0] = %.3f, a_mi_f[1] = %.3f, a_mi_f[2] = %.3f, a_mi_f[3] = %.3f\n", a[i].tMins().m_floats[0], a[i].tMins().m_floats[1], a[i].tMins().m_floats[2], a[i].tMins().m_floats[3]);
printf("b_mx_f[0] = %.3f, b_mx_f[1] = %.3f, b_mx_f[2] = %.3f, b_mx_f[3] = %.3f\n", b[i].tMaxs().m_floats[0], b[i].tMaxs().m_floats[1], b[i].tMaxs().m_floats[2], b[i].tMaxs().m_floats[3]);
printf("b_mi_f[0] = %.3f, b_mi_f[1] = %.3f, b_mi_f[2] = %.3f, b_mi_f[3] = %.3f\n", b[i].tMins().m_floats[0], b[i].tMins().m_floats[1], b[i].tMins().m_floats[2], b[i].tMins().m_floats[3]);
printf("c_mx_f[0] = %.3f, c_mx_f[1] = %.3f, c_mx_f[2] = %.3f, c_mx_f[3] = %.3f\n", c[i].tMaxs().m_floats[0], c[i].tMaxs().m_floats[1], c[i].tMaxs().m_floats[2], c[i].tMaxs().m_floats[3]);
printf("c_mi_f[0] = %.3f, c_mi_f[1] = %.3f, c_mi_f[2] = %.3f, c_mi_f[3] = %.3f\n", c[i].tMins().m_floats[0], c[i].tMins().m_floats[1], c[i].tMins().m_floats[2], c[i].tMins().m_floats[3]);
printf("a_mx_f_ref[0] = %.3f, a_mx_f_ref[1] = %.3f, a_mx_f_ref[2] = %.3f, a_mx_f_ref[3] = %.3f\n", a_ref[i].tMaxs().m_floats[0], a_ref[i].tMaxs().m_floats[1], a_ref[i].tMaxs().m_floats[2], a_ref[i].tMaxs().m_floats[3]);
printf("a_mi_f_ref[0] = %.3f, a_mi_f_ref[1] = %.3f, a_mi_f_ref[2] = %.3f, a_mi_f_ref[3] = %.3f\n", a_ref[i].tMins().m_floats[0], a_ref[i].tMins().m_floats[1], a_ref[i].tMins().m_floats[2], a_ref[i].tMins().m_floats[3]);
printf("b_mx_f_ref[0] = %.3f, b_mx_f_ref[1] = %.3f, b_mx_f_ref[2] = %.3f, b_mx_f_ref[3] = %.3f\n", b_ref[i].tMaxs().m_floats[0], b_ref[i].tMaxs().m_floats[1], b_ref[i].tMaxs().m_floats[2], b_ref[i].tMaxs().m_floats[3]);
printf("b_mi_f_ref[0] = %.3f, b_mi_f_ref[1] = %.3f, b_mi_f_ref[2] = %.3f, b_mi_f_ref[3] = %.3f\n", b_ref[i].tMins().m_floats[0], b_ref[i].tMins().m_floats[1], b_ref[i].tMins().m_floats[2], b_ref[i].tMins().m_floats[3]);
printf("c_mx_f_ref[0] = %.3f, c_mx_f_ref[1] = %.3f, c_mx_f_ref[2] = %.3f, c_mx_f_ref[3] = %.3f\n", c_ref[i].tMaxs().m_floats[0], c_ref[i].tMaxs().m_floats[1], c_ref[i].tMaxs().m_floats[2], c_ref[i].tMaxs().m_floats[3]);
printf("c_mi_f_ref[0] = %.3f, c_mi_f_ref[1] = %.3f, c_mi_f_ref[2] = %.3f, c_mi_f_ref[3] = %.3f\n", c_ref[i].tMins().m_floats[0], c_ref[i].tMins().m_floats[1], c_ref[i].tMins().m_floats[2], c_ref[i].tMins().m_floats[3]);
return 1;
}
}
////////////////////////////////////
//
// Time and Test Select
//
////////////////////////////////////
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Select_Ref_Res[i] = Select_ref(a_ref[i], b_ref[i], c_ref[i]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (i = 0; i < DATA_SIZE; i++)
{
Select_Test_Res[i] = Select(a[i], b[i], c[i]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Select Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT, TicksToCycles( vectorTime ) / LOOPCOUNT );
//printf("scalar = %llu, vector = %llu\n", scalarTime, vectorTime);
for (i = 0; i < DATA_SIZE; i++)
{
Select_Ref_Res[i] = Select_ref(a_ref[i], b_ref[i], c_ref[i]);
Select_Test_Res[i] = Select(a[i], b[i], c[i]);
if(Select_Test_Res[i] != Select_Ref_Res[i])
{
printf("Select fail at %d with test = %d, ref = %d\n", i, Select_Test_Res[i], Select_Ref_Res[i]);
printf("a_mx_f[0] = %.3f, a_mx_f[1] = %.3f, a_mx_f[2] = %.3f, a_mx_f[3] = %.3f\n", a[i].tMaxs().m_floats[0], a[i].tMaxs().m_floats[1], a[i].tMaxs().m_floats[2], a[i].tMaxs().m_floats[3]);
printf("a_mi_f[0] = %.3f, a_mi_f[1] = %.3f, a_mi_f[2] = %.3f, a_mi_f[3] = %.3f\n", a[i].tMins().m_floats[0], a[i].tMins().m_floats[1], a[i].tMins().m_floats[2], a[i].tMins().m_floats[3]);
printf("b_mx_f[0] = %.3f, b_mx_f[1] = %.3f, b_mx_f[2] = %.3f, b_mx_f[3] = %.3f\n", b[i].tMaxs().m_floats[0], b[i].tMaxs().m_floats[1], b[i].tMaxs().m_floats[2], b[i].tMaxs().m_floats[3]);
printf("b_mi_f[0] = %.3f, b_mi_f[1] = %.3f, b_mi_f[2] = %.3f, b_mi_f[3] = %.3f\n", b[i].tMins().m_floats[0], b[i].tMins().m_floats[1], b[i].tMins().m_floats[2], b[i].tMins().m_floats[3]);
printf("c_mx_f[0] = %.3f, c_mx_f[1] = %.3f, c_mx_f[2] = %.3f, c_mx_f[3] = %.3f\n", c[i].tMaxs().m_floats[0], c[i].tMaxs().m_floats[1], c[i].tMaxs().m_floats[2], c[i].tMaxs().m_floats[3]);
printf("c_mi_f[0] = %.3f, c_mi_f[1] = %.3f, c_mi_f[2] = %.3f, c_mi_f[3] = %.3f\n", c[i].tMins().m_floats[0], c[i].tMins().m_floats[1], c[i].tMins().m_floats[2], c[i].tMins().m_floats[3]);
printf("a_mx_f_ref[0] = %.3f, a_mx_f_ref[1] = %.3f, a_mx_f_ref[2] = %.3f, a_mx_f_ref[3] = %.3f\n", a_ref[i].tMaxs().m_floats[0], a_ref[i].tMaxs().m_floats[1], a_ref[i].tMaxs().m_floats[2], a_ref[i].tMaxs().m_floats[3]);
printf("a_mi_f_ref[0] = %.3f, a_mi_f_ref[1] = %.3f, a_mi_f_ref[2] = %.3f, a_mi_f_ref[3] = %.3f\n", a_ref[i].tMins().m_floats[0], a_ref[i].tMins().m_floats[1], a_ref[i].tMins().m_floats[2], a_ref[i].tMins().m_floats[3]);
printf("b_mx_f_ref[0] = %.3f, b_mx_f_ref[1] = %.3f, b_mx_f_ref[2] = %.3f, b_mx_f_ref[3] = %.3f\n", b_ref[i].tMaxs().m_floats[0], b_ref[i].tMaxs().m_floats[1], b_ref[i].tMaxs().m_floats[2], b_ref[i].tMaxs().m_floats[3]);
printf("b_mi_f_ref[0] = %.3f, b_mi_f_ref[1] = %.3f, b_mi_f_ref[2] = %.3f, b_mi_f_ref[3] = %.3f\n", b_ref[i].tMins().m_floats[0], b_ref[i].tMins().m_floats[1], b_ref[i].tMins().m_floats[2], b_ref[i].tMins().m_floats[3]);
printf("c_mx_f_ref[0] = %.3f, c_mx_f_ref[1] = %.3f, c_mx_f_ref[2] = %.3f, c_mx_f_ref[3] = %.3f\n", c_ref[i].tMaxs().m_floats[0], c_ref[i].tMaxs().m_floats[1], c_ref[i].tMaxs().m_floats[2], c_ref[i].tMaxs().m_floats[3]);
printf("c_mi_f_ref[0] = %.3f, c_mi_f_ref[1] = %.3f, c_mi_f_ref[2] = %.3f, c_mi_f_ref[3] = %.3f\n", c_ref[i].tMins().m_floats[0], c_ref[i].tMins().m_floats[1], c_ref[i].tMins().m_floats[2], c_ref[i].tMins().m_floats[3]);
return 1;
}
}
return 0;
}
int Test_v3skew(void)
{
btVector3 v, v1, v2, v3, vt1, vt2, vt3;
float x,y,z,w;
// Init the data
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = BT_NAN; // w channel NaN
v.setValue(x,y,z);
v.setW(w);
v1.setValue(w,w,w);
v1.setW(w);
vt3 = vt2 = vt1 = v3 = v2 = v1;
{
v3skew_ref(&v, &v1, &v2, &v3);
v.getSkewSymmetricMatrix(&vt1, &vt2, &vt3);
/*
if( v1.m_floats[0] != vt1.m_floats[0] ||
v1.m_floats[1] != vt1.m_floats[1] ||
v1.m_floats[2] != vt1.m_floats[2] )
*/
if(!(v1 == vt1))
{
vlog( "Error - v3skew result error! "
"\ncorrect v1 = (%10.4f, %10.4f, %10.4f) "
"\ntested v1 = (%10.4f, %10.4f, %10.4f) \n",
v1.m_floats[0], v1.m_floats[1], v1.m_floats[2],
vt1.m_floats[0], vt1.m_floats[1], vt1.m_floats[2]);
return 1;
}
/*
if( v2.m_floats[0] != vt2.m_floats[0] ||
v2.m_floats[1] != vt2.m_floats[1] ||
v2.m_floats[2] != vt2.m_floats[2] )
*/
if(!(v2 == vt2))
{
vlog( "Error - v3skew result error! "
"\ncorrect v2 = (%10.4f, %10.4f, %10.4f) "
"\ntested v2 = (%10.4f, %10.4f, %10.4f) \n",
v2.m_floats[0], v2.m_floats[1], v2.m_floats[2],
vt2.m_floats[0], vt2.m_floats[1], vt2.m_floats[2]);
return 1;
}
/*
if( v3.m_floats[0] != vt3.m_floats[0] ||
v3.m_floats[1] != vt3.m_floats[1] ||
v3.m_floats[2] != vt3.m_floats[2] )
*/
if(!(v3 == vt3))
{
vlog( "Error - v3skew result error! "
"\ncorrect v3 = (%10.4f, %10.4f, %10.4f) "
"\ntested v3 = (%10.4f, %10.4f, %10.4f) \n",
v3.m_floats[0], v3.m_floats[1], v3.m_floats[2],
vt3.m_floats[0], vt3.m_floats[1], vt3.m_floats[2]);
return 1;
}
}
#define DATA_SIZE 256
btVector3 v3_arr0[DATA_SIZE];
btVector3 v3_arr1[DATA_SIZE];
btVector3 v3_arr2[DATA_SIZE];
btVector3 v3_arr3[DATA_SIZE];
uint64_t scalarTime;
uint64_t vectorTime;
size_t j, k;
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v3_arr0[k].setValue(x,y,z);
v3_arr0[k].setW(w);
v3_arr1[k].setValue(w,w,w);
v3_arr1[k].setW(w);
v3_arr3[k] = v3_arr2[k] = v3_arr1[k];
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for( k = 0; k < LOOPCOUNT; k++ )
{
size_t k32 = (k & (DATA_SIZE-1));
v3skew_ref( &v3_arr0[k32], &v3_arr1[k32], &v3_arr2[k32], &v3_arr3[k32]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for( k = 0; k < LOOPCOUNT; k++ )
{
size_t k32 = (k & (DATA_SIZE -1));
v3_arr0[k32].getSkewSymmetricMatrix(&v3_arr1[k32], &v3_arr2[k32], &v3_arr3[k32]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT, TicksToCycles( vectorTime ) / LOOPCOUNT );
return 0;
}
int Test_v3lerp(void)
{
btVector3 v1, v2;
btScalar rt;
float x,y,z,w;
float vNaN =BT_NAN;
w =BT_NAN; // w channel NaN
btVector3 correct_res, test_res;
for (rt = 0.0f; rt <= 1.0f; rt += 0.1f)
{
correct_res.setValue(vNaN, vNaN, vNaN);
test_res.setValue(vNaN, vNaN, vNaN);
// Init the data
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v1.setValue(x,y,z);
v1.setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
v2.setValue(x,y,z);
v2.setW(w);
correct_res = v3lerp_ref(correct_res, v1, v2, rt);
test_res = v1.lerp(v2, rt);
if( fabs(correct_res.m_floats[0] - test_res.m_floats[0]) +
fabs(correct_res.m_floats[1] - test_res.m_floats[1]) +
fabs(correct_res.m_floats[2] - test_res.m_floats[2]) > FLT_EPSILON * 4)
{
vlog( "Error - v3lerp result error! "
"\ncorrect = (%10.4f, %10.4f, %10.4f) "
"\ntested = (%10.4f, %10.4f, %10.4f) \n"
"\n rt=%10.4f",
correct_res.m_floats[0], correct_res.m_floats[1], correct_res.m_floats[2],
test_res.m_floats[0], test_res.m_floats[1], test_res.m_floats[2], rt);
return 1;
}
}
#define DATA_SIZE LOOPCOUNT
btVector3 vec3_arr1[DATA_SIZE];
btVector3 vec3_arr2[DATA_SIZE];
btScalar rt_arr[DATA_SIZE];
uint64_t scalarTime;
uint64_t vectorTime;
size_t j, k;
{
uint64_t startTime, bestTime, currentTime;
w =BT_NAN; // w channel NaN
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr1[k].setValue(x,y,z);
vec3_arr1[k].setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr2[k].setValue(x,y,z);
vec3_arr2[k].setW(w);
rt_arr[k] = RANDF_01;
}
startTime = ReadTicks();
for( k = 0; k < LOOPCOUNT; k++ )
{
v3lerp_ref(vec3_arr1[k], vec3_arr1[k], vec3_arr2[k], rt_arr[k]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
for( k = 0; k < DATA_SIZE; k++ )
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr1[k].setValue(x,y,z);
vec3_arr1[k].setW(w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
vec3_arr2[k].setValue(x,y,z);
vec3_arr2[k].setW(w);
rt_arr[k] = RANDF_01;
}
startTime = ReadTicks();
for( k = 0; k < LOOPCOUNT; k++ )
{
vec3_arr1[k] = vec3_arr1[k].lerp(vec3_arr2[k], rt_arr[k]);
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if( currentTime < bestTime )
bestTime = currentTime;
}
if( 0 == gReportAverageTimes )
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog( "Timing:\n" );
vlog( " \t scalar\t vector\n" );
vlog( " \t%10.4f\t%10.4f\n", TicksToCycles( scalarTime ) / LOOPCOUNT,
TicksToCycles( vectorTime ) / LOOPCOUNT );
return 0;
}
int Test_qtdot(void)
{
btQuaternion q1, q2;
float x, y, z, w, vNaN;
vNaN = BT_NAN; // w channel NaN
// Init the data
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = RANDF_01;
q1.setValue(x, y, z, w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = RANDF_01;
q2.setValue(x, y, z, w);
btScalar correct_res, test_res;
{
correct_res = vNaN;
test_res = vNaN;
correct_res = qtdot_ref(q1, q2);
test_res = BT_OP(q1, q2);
if (fabsf(correct_res - test_res) > FLT_EPSILON * 4)
{
vlog(
"Error - qtdot result error! "
"\ncorrect = %10.4f "
"\ntested = %10.4f \n",
correct_res, test_res);
return 1;
}
}
#define DATA_SIZE LOOPCOUNT
btQuaternion qt_arr1[DATA_SIZE];
btQuaternion qt_arr2[DATA_SIZE];
btScalar res_arr[DATA_SIZE];
uint64_t scalarTime;
uint64_t vectorTime;
size_t j, k;
for (k = 0; k < DATA_SIZE; k++)
{
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = RANDF_01;
qt_arr1[k].setValue(x, y, z, w);
x = RANDF_01;
y = RANDF_01;
z = RANDF_01;
w = RANDF_01;
qt_arr2[k].setValue(x, y, z, w);
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
scalarTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (k = 0; k + 4 <= LOOPCOUNT; k += 4)
{
size_t km = (k & (DATA_SIZE - 1));
res_arr[km] = qtdot_ref(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = qtdot_ref(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = qtdot_ref(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = qtdot_ref(qt_arr1[km], qt_arr2[km]);
}
currentTime = ReadTicks() - startTime;
scalarTime += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
scalarTime = bestTime;
else
scalarTime /= NUM_CYCLES;
}
{
uint64_t startTime, bestTime, currentTime;
bestTime = -1LL;
vectorTime = 0;
for (j = 0; j < NUM_CYCLES; j++)
{
startTime = ReadTicks();
for (k = 0; k + 4 <= LOOPCOUNT; k += 4)
{
size_t km = (k & (DATA_SIZE - 1));
res_arr[km] = BT_OP(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = BT_OP(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = BT_OP(qt_arr1[km], qt_arr2[km]);
km++;
res_arr[km] = BT_OP(qt_arr1[km], qt_arr2[km]);
km++;
}
currentTime = ReadTicks() - startTime;
vectorTime += currentTime;
if (currentTime < bestTime)
bestTime = currentTime;
}
if (0 == gReportAverageTimes)
vectorTime = bestTime;
else
vectorTime /= NUM_CYCLES;
}
vlog("Timing:\n");
vlog(" \t scalar\t vector\n");
vlog(" \t%10.4f\t%10.4f\n", TicksToCycles(scalarTime) / LOOPCOUNT,
TicksToCycles(vectorTime) / LOOPCOUNT);
return 0;
}
