void
foo ()
{
x = y;
_mm256_zeroall ();
_mm256_zeroupper ();
_mm256_zeroupper ();
_mm256_zeroupper ();
}
int main(int, char**)
{
/* AVX */
_mm256_zeroall();
__m256i a = _mm256_setzero_si256();
/* AVX2 */
__m256i b = _mm256_and_si256(a, a);
__m256i result = _mm256_add_epi8(a, b);
(void)result;
return 0;
}
void ImageAdd::_RunInt()
{
const size_t cszIters = static_cast< size_t >( TEST_RUN_COUNT );
printf("\n Running \"_RunInt()\" - Image-Size = %4d x %4d\n", ciImageWidth, ciImageHeight);
HandlerList lstHandlers;
AppendHandlerList( lstHandlers, ImageAddInt_GetScalarHandlers() );
AppendHandlerList( lstHandlers, ImageAddInt_GetSSEHandlers() );
AppendHandlerList( lstHandlers, ImageAddInt_GetAVXHandlers() );
CREATE_IMAGE( InputImageType1, spInputImage1 );
CREATE_IMAGE( InputImageType2, spInputImage2 );
CREATE_IMAGE( OutputImageType, spOutputImage );
ImageIO::LoadTestImage( *spInputImage1 );
ImageIO::LoadTestImage( *spInputImage2 );
for (auto itHandler : lstHandlers)
{
printf( "\n Launch handler \"%10s\" -> ", itHandler->GetName().c_str() );
double dTime = omp_get_wtime();
for (size_t szIter = static_cast<size_t>(0); szIter < cszIters; ++szIter)
{
_mm256_zeroall();
itHandler->Launch( spOutputImage, spInputImage1, spInputImage2 );
}
dTime = omp_get_wtime() - dTime;
dTime = (dTime * 1000.) / static_cast< double >( cszIters );
printf( "%10.6f ms", dTime );
}
#ifdef CONFIG_CHECK_OUTPUT
IMAGE_ADD_CHECK_OUTPUT( Int );
#endif
printf("\n");
}
void TopologicalErosion::_RunFloat(unsigned int uiKernelSize)
{
const size_t cszIters = static_cast< size_t >( TEST_RUN_COUNT );
printf("\n Running \"_RunFloat()\" - Image-Size = %4d x %4d\n", ciImageWidth, ciImageHeight);
HandlerList lstHandlers;
AppendHandlerList( lstHandlers, TopologicalErosionFloat_GetScalarHandlers() );
AppendHandlerList( lstHandlers, TopologicalErosionFloat_GetSSEHandlers() );
AppendHandlerList( lstHandlers, TopologicalErosionFloat_GetAVXHandlers() );
CREATE_IMAGE( InputImageType, spInputImage );
CREATE_IMAGE( OutputImageType, spOutputImage );
ImageIO::LoadTestImage( *spInputImage );
for (auto itHandler : lstHandlers)
{
printf( "\n Launch handler \"%10s\" -> ", itHandler->GetName().c_str() );
double dTime = omp_get_wtime();
for (size_t szIter = static_cast<size_t>(0); szIter < cszIters; ++szIter)
{
_mm256_zeroall();
itHandler->Launch( spOutputImage, spInputImage, uiKernelSize );
}
dTime = omp_get_wtime() - dTime;
dTime = (dTime * 1000.) / static_cast< double >( cszIters );
printf( "%10.6f ms", dTime );
}
#ifdef CONFIG_CHECK_OUTPUT
TOPOLOGICAL_EROSION_CHECK_OUTPUT( Float );
#endif
printf("\n");
}
void _Run(OutputPixelType aaOutput[ciHeight][ciWidth], InputPixelType_1 aaInput1[ciHeight][ciWidth], InputPixelType_2 aaInput2[ciHeight][ciWidth])
{
for (int iY = 0; iY < ciHeight; ++iY)
{
_mm256_zeroall();
OutputPixelType *pOutput = aaOutput[iY];
InputPixelType_1 *pInput1 = aaInput1[iY];
InputPixelType_2 *pInput2 = aaInput2[iY];
for (int iX = 0; iX < ciWidth; iX += VectorWidth)
{
__m256 mmIn1 = _mm256_loadu_ps( pInput1 + iX );
__m256 mmIn2 = _mm256_loadu_ps( pInput2 + iX );
_mm256_storeu_ps( pOutput + iX, _mm256_add_ps(mmIn1, mmIn2) );
}
_mm256_zeroupper();
}
}
void
foo ()
{
x = y;
_mm256_zeroall ();
}
void animate()
{
float mx;
float my;
if(ManualControl)
{
POINT pos;
GetCursorPos(&pos);
RECT rc;
GetClientRect(hMainWnd, &rc);
ScreenToClient(hMainWnd, &pos);
mx = pos.x;
my = pos.y;
}
else
{
UpdatePosition(mx, my);
}
const auto size = partCount;
VertexData *pVertexBuffer;
pVertexObject->Lock(0, 0, (void**)&pVertexBuffer, D3DLOCK_DISCARD);
_mm256_zeroall();
#pragma omp parallel \
shared(pVertexBuffer, particlesCoord, particlesVel, mx, my, size)
{
#pragma omp for nowait
for(int i = 0; i < size; i += 4)
{
float mouseCoordVec[8] = { mx, my, mx, my, mx, my, mx, my };
float *particleCoordsVec = (float*)particlesCoord + i;
float *velocityVec = (float*)particlesVel + i;
auto xyCoord = _mm256_loadu_ps(particleCoordsVec);
auto hwTempData = _mm256_sub_ps(xyCoord, _mm256_loadu_ps(mouseCoordVec));
auto squares = _mm256_mul_ps(hwTempData, hwTempData);
auto distSquare = _mm256_hadd_ps(squares, squares);
distSquare = _mm256_shuffle_ps(distSquare, distSquare, 0x50);
auto theForce = _mm256_div_ps(_mm256_set1_ps(G), distSquare);
if(distSquare.m256_f32[0] < 400)
{
theForce.m256_f32[0] = 0;
theForce.m256_f32[1] = 0;
}
if(distSquare.m256_f32[2] < 400)
{
theForce.m256_f32[2] = 0;
theForce.m256_f32[3] = 0;
}
if(distSquare.m256_f32[4] < 400)
{
theForce.m256_f32[4] = 0;
theForce.m256_f32[5] = 0;
}
if(distSquare.m256_f32[6] < 400)
{
theForce.m256_f32[6] = 0;
theForce.m256_f32[7] = 0;
}
auto xyForces = _mm256_mul_ps(_mm256_xor_ps(hwTempData, _mm256_set1_ps(-0.f)), theForce);
auto xyVelocities = _mm256_loadu_ps(velocityVec);
xyVelocities = _mm256_mul_ps(xyVelocities, _mm256_set1_ps(Resistance));
xyVelocities = _mm256_add_ps(xyVelocities, xyForces);
xyCoord = _mm256_add_ps(xyCoord, xyVelocities);
_mm256_storeu_ps(velocityVec, xyVelocities);
_mm256_storeu_ps(particleCoordsVec, xyCoord);
processIfOutOfBounds(((ParticleCoord*)particleCoordsVec)[0], ((ParticleVel*)velocityVec)[0]);
processIfOutOfBounds(((ParticleCoord*)particleCoordsVec)[1], ((ParticleVel*)velocityVec)[1]);
processIfOutOfBounds(((ParticleCoord*)particleCoordsVec)[2], ((ParticleVel*)velocityVec)[2]);
processIfOutOfBounds(((ParticleCoord*)particleCoordsVec)[3], ((ParticleVel*)velocityVec)[3]);
pVertexBuffer[i].x = ((ParticleCoord*)particleCoordsVec)[0].x;
pVertexBuffer[i].y = ((ParticleCoord*)particleCoordsVec)[0].y;
pVertexBuffer[i + 1].x = ((ParticleCoord*)particleCoordsVec)[1].x;
pVertexBuffer[i + 1].y = ((ParticleCoord*)particleCoordsVec)[1].y;
pVertexBuffer[i + 2].x = ((ParticleCoord*)particleCoordsVec)[2].x;
pVertexBuffer[i + 2].y = ((ParticleCoord*)particleCoordsVec)[2].y;
pVertexBuffer[i + 3].x = ((ParticleCoord*)particleCoordsVec)[3].x;
pVertexBuffer[i + 3].y = ((ParticleCoord*)particleCoordsVec)[3].y;
}
}
pVertexObject->Unlock();
_mm256_zeroall();
}
