uint8_t ti89_get_byte(uint32_t adr)
{
// RAM access
if(IN_BOUNDS(0x000000, adr, 0x1fffff))
{
return get_b(tihw.ram, adr, RAM_SIZE_TI89 - 1);
}
// FLASH access
else if(IN_BOUNDS(0x200000, adr, 0x5fffff))
{
return get_b(tihw.rom, adr, ROM_SIZE_TI89 - 1) | wsm.ret_or;
}
// memory-mapped I/O
else if(IN_BOUNDS(0x600000, adr, 0x6fffff))
{
return io_get_byte(adr);
}
// memory-mapped I/O (hw2)
else if(IN_RANGE(adr, 0x700000, IO2_SIZE_TI89))
{
return io2_get_byte(adr);
}
return 0x14;
}
double Mask::involution(BYTE* image,int widthBytes,int type,int x,int y){
//type indicates which color tube to use
double sum=0;
if (type==1)//r
{
for (int i = -1; i <= 1; ++i)
for (int j = -1;j<=1;j++)
sum+=get_r(image,widthBytes,x-i,y+j)*mask[(i+1)*3+j+1];
// sum/=16;
}
if (type==2)//g
{
for (int i = -1; i <= 1; ++i)
for (int j = -1;j<=1;j++)
sum+=get_g(image,widthBytes,x-i,y+j)*mask[(i+1)*3+j+1];
// sum/=16;
}
if (type==3)//b
{
for (int i = -1; i <= 1; ++i)
for (int j = -1;j<=1;j++)
sum+=get_b(image,widthBytes,x-i,y+j)*mask[(i+1)*3+j+1];
// sum/=16;
}
return sum;
}
uint8_t ti89_get_byte(uint32_t adr)
{
// RAM access
if(IN_BOUNDS(0x000000, adr, 0x1fffff))
{
return get_b(ram, adr, RAM_SIZE_TI89 - 1);
}
// FLASH access
else if(IN_BOUNDS(0x200000, adr, 0x5fffff))
{
return get_b(rom, adr, ROM_SIZE_TI89 - 1);
}
return 0x14;
}
void get_coefs(double *a, double *b, double *c)
{
*a = get_a() ;
*b = get_b() ;
*c = get_c() ;
}
/* this performs the discrete transform */
CVC_trans discrete_transform(CVC *YPbPr)
{
CVC_trans pack_to_return;
pack_to_return.Pb = YPbPr->avg_Pb;
pack_to_return.Pr = YPbPr->avg_Pr;
pack_to_return.a = get_a(YPbPr);
pack_to_return.b = get_b(YPbPr);
pack_to_return.c = get_c(YPbPr);
pack_to_return.d = get_d(YPbPr);
return pack_to_return;
}
void SIAM::get_Sigma()
{
if (!SymmetricCase)
{ printf("going through asymmetric\n");
double b = get_b();
for (int i=0; i<N; i++)
Sigma[i] = U*n + SOCSigma[i]
/ ( (double)1.0 - complex<double>(b) * SOCSigma[i] );
}
else
for (int i=0; i<N; i++)
Sigma[i] = U*n + SOCSigma[i];
}
void Analyze::writeGrid () {
std::ofstream fout ("/Users/fred.christensen/Dropbox/school/Parallel/genetic/result.ppm");
fout << "P3\n640 420\n255\n";
for (int y = 0; y < height; ++y) {
for (int x = 0; x < width; ++x) {
double bias = (double)grid[x][y] / (double)max;
fout << get_r(bias) << " " << get_g(bias) << " " << get_b(bias);
if (x < width -1) {
fout << "\t";
} else {
fout << std::endl;
}
}
}
}
void IASIAM::get_G_f_CHM()
{
get_b();
get_Sigma_f();
for (int i=0; i<N; i++) G_f[i] = 1.0/(mu + ii*omega[i] - Sigma_f[i] - Delta[i]);
/*for (int i=0; i<N; i++)
{ complex<double>* g= new complex<double>[N_CHM];
for (int j=0; j<N_CHM; j++)
g[j] = dos[j] / ( mu + ii*omega[i] - omega_CHM[j] - Sigma_f[i]);
G_f[i] = TrapezIntegral(N_CHM, g, omega_CHM);
delete [] g;
}*/
}
void IASIAM::get_Sigma()
{
if (not PHSymmetricCase)
{ printf("going through asymmetric\n");
double b = get_b();
for (int i=0; i<N; i++)
r->Sigma[i] = U*r->n + r->SOCSigma[i]
/ ( 1.0 - b * r->SOCSigma[i] );
}
else
{ printf("going through symmetric\n");
for (int i=0; i<N; i++)
r->Sigma[i] = U * r->n + r->SOCSigma[i];
}
if (PatchTailWithAtomicLimit) r->PatchAtomicLimitSigma(AtomicCutoff, U);
}
void SIAM::get_Sigma()
{
if (!SymmetricCase)
{ //printf("going through asymmetric\n");
double b = get_b();
#pragma omp parallel for
for (int i=0; i<N; i++)
r->Sigma[i] = U*r->n + r->SOCSigma[i]
/ ( 1.0 - b * r->SOCSigma[i] );
}
else
#pragma omp parallel for
for (int i=0; i<N; i++)
r->Sigma[i] = U * r->n + r->SOCSigma[i];
}
double HeMoTKMaterial :: computeCapacityCoeff(MatResponseMode mode, GaussPoint *gp, TimeStep *tStep)
{
if ( mode == Capacity_ww ) {
return 1.0 * rho;
} else if ( mode == Capacity_wh ) {
return 0.0;
} else if ( mode == Capacity_hw ) {
TransportMaterialStatus *status = static_cast< TransportMaterialStatus * >( this->giveStatus(gp) );
FloatArray s;
double w, t;
s = status->giveTempField();
if ( s.isEmpty() ) {
OOFEM_ERROR("undefined state vector");
}
w = s.at(2);
t = s.at(1);
return get_b(w, t) * get_latent(w, t);
} else if ( mode == Capacity_hh ) {
TransportMaterialStatus *status = static_cast< TransportMaterialStatus * >( this->giveStatus(gp) );
FloatArray s;
double w, t;
s = status->giveTempField();
if ( s.isEmpty() ) {
OOFEM_ERROR("undefined state vector");
}
w = s.at(2);
t = s.at(1);
return get_ceff(w, t);
} else {
OOFEM_ERROR("Unknown MatResponseMode");
}
return 0.0; // to make compiler happy
}
double HeMoTKMaterial :: computeCapacityCoeff(MatResponseMode mode, GaussPoint *gp, TimeStep *atTime)
{
if ( mode == Capacity_ww ) {
return 1.0 * rho;
} else if ( mode == Capacity_wh ) {
return 0.0;
} else if ( mode == Capacity_hw ) {
TransportMaterialStatus *status = ( TransportMaterialStatus * ) this->giveStatus(gp);
FloatArray s;
double w, t;
s = status->giveTempStateVector();
if ( s.isEmpty() ) {
_error("computeCapacityCoeff: undefined state vector");
}
w = s.at(2);
t = s.at(1);
return get_b(w, t) * get_latent(w, t);
} else if ( mode == Capacity_hh ) {
TransportMaterialStatus *status = ( TransportMaterialStatus * ) this->giveStatus(gp);
FloatArray s;
double w, t;
s = status->giveTempStateVector();
if ( s.isEmpty() ) {
_error("computeCapacityCoeff: undefined state vector");
}
w = s.at(2);
t = s.at(1);
return get_ceff(w, t);
} else {
_error("Unknown MatResponseMode");
}
return 0.0; // to make compiler happy
}
uint8_t ti89t_get_byte(uint32_t adr)
{
// RAM access
if(IN_BOUNDS(0x000000, adr, 0x03ffff) ||
IN_BOUNDS(0x200000, adr, 0x23ffff) ||
IN_BOUNDS(0x400000, adr, 0x43ffff))
{
return get_b(tihw.ram, adr, 0x03ffff);
}
// FLASH access
else if(IN_BOUNDS(0x800000, adr, 0xbfffff))
{
return FlashReadByte(adr);
}
// memory-mapped I/O
else if(IN_BOUNDS(0x600000, adr, 0x6fffff))
{
return io_get_byte(adr);
}
// memory-mapped I/O (hw2)
else if(IN_RANGE(adr, 0x700000, IO2_SIZE_TI89T))
{
return io2_get_byte(adr);
}
// memory-mapped I/O (hw3)
else if(IN_RANGE(adr, 0x710000, IO3_SIZE_TI89T))
{
return io3_get_byte(adr);
}
return 0x14;
}
std::list<Point3D> const LineIntersectionFinder::find_intersections(std::list<Line3D>& lines) const {
EventStructure event_structure;
StatusStructure status_structure;
std::list<Point3D> intersections;
for (auto line = lines.begin(); line != lines.end(); ++line) {
event_structure.add_event(new StartEvent(line->get_a(), &(*line)));
event_structure.add_event(new EndEvent(line->get_b(), &(*line)));
}
while (!event_structure.empty()) {
// std::cout << "Event structure:\n" << event_structure << std::endl;
Event* event(event_structure.get_top());
event_structure.delete_top();
Point3D intersection(status_structure.process_event(event, event_structure));
if (intersection.get_z() != -1)
intersections.push_back(intersection);
// std::cout << "Status structure:\n" << status_structure << std::endl;
}
return intersections;
}
uint calc(int x, int y){
// r,g,b color values around center 'c11'
// 00 10 20
// 01 11 21
// 02 12 22
uint c00=get(x-1,y-1);
uint c01=get(x-1,y);
uint c02=get(x-1,y+1);
uint c10=get(x,y-1);
uint c11=get(x,y);
uint c12=get(x,y+1);
uint c20=get(x+1,y-1);
uint c21=get(x+1,y);
uint c22=get(x+1,y+1);
// red
uint r00=get_r(x-1,y-1);
uint r01=get_r(x-1,y) ;
uint r02=get_r(x-1,y+1);
uint r10=get_r(x,y-1) ;
uint r11=get_r(x,y) ;
uint r12=get_r(x,y+1) ;
uint r20=get_r(x+1,y-1);
uint r21=get_r(x+1,y) ;
uint r22=get_r(x+1,y+1);
// green
uint g00=get_g(x-1,y-1);
uint g01=get_g(x-1,y) ;
uint g02=get_g(x-1,y+1);
uint g10=get_g(x,y-1) ;
uint g11=get_g(x,y) ;
uint g12=get_g(x,y+1) ;
uint g20=get_g(x+1,y-1);
uint g21=get_g(x+1,y) ;
uint g22=get_g(x+1,y+1);
// blue
uint b00=get_b(x-1,y-1);
uint b01=get_b(x-1,y) ;
uint b02=get_b(x-1,y+1);
uint b10=get_b(x,y-1) ;
uint b11=get_b(x,y) ;
uint b12=get_b(x,y+1) ;
uint b20=get_b(x+1,y-1);
uint b21=get_b(x+1,y) ;
uint b22=get_b(x+1,y+1);
// center colors of THIS dot
uint r=r11;
uint g=g11;
uint b=b11;
// random colors
uint zr=rand()%256;
uint zg=rand()%256;
uint zb=rand()%256;
// mean color value of surrounding
uint r0=((r10+r01+r21+r12)/4);
uint g0=((g10+g01+g21+g12)/4);
uint b0=(b10+b01+b21+b12)/4;
// Including this dot
// uint r0=((r10+r01+r11+r21+r12)/5);
// uint g0=((g10+g01+g11+g21+g12)/5);
// uint b0=(b10+b01+b11+b21+b12)/5;
float fr=r0/256.0;
float fg=g0/256.0;
float fb=b0/256.0;
uint k=128;
float h=100.0;
//////////
// done with initialization,
// now updated the dot's color based on some crazy experimental whatever function
// feel free to wildly experiment with this algorithm!
// POPULATION UPDATE
// inspired by Conway's game of life:
// if there is a sufficient population, then grow in numbers:
if(b0>=100)b=b0*1.01;
// if the population is small then shrink
if(b0<100)b=b0*(0.99-fr/100);
// if the population is too big then collapse
if(b0>240)b=0;
// repopulate collapsed populations
if(b0<3)b=120;//zb*((1.5-f)+fr*f);
if(r0>=100)r=r0*1.01;
if(r0<100)r=r0*(0.99-fg/100);
if(r11>240)r=0;
if(r11<3)r=120*(1.2-fr*fg);//zb*((1.5-f)+fr*f);
if(g0>=100)g=g0*1.01;
if(g0<100)g=g0*(0.99-f*fb/100);
if(g11>240){g=0;b=b/2;}
if(g11<3)g=120*(1-fb*fr);//zb*((1.5-f)+fr*f);
// END OF POPULATION UPDATE
if(r>255)r=255;
if(g>255)g=255;
if(b>255)b=255;
return (r<<16)+(g<<8)+b;
}
bool BinAsm::finds_labels()
{
unsigned int lc = 0;
unsigned int errc = 0;
unsigned int offset = 0;
uint16_t opd = 0;
std::string err = "";
std::vector<std::string> lines=split_text(_src);
std::vector<std::string>::iterator lit = lines.begin();
for (;lit!=lines.end();lit++)
{
lc++;
std::vector<std::string> w=split_line(*lit);
if (!w.size()) continue;
if (get_op(w[0]))
{
offset++;
if (w.size() < 3) continue;
opd=0;
get_b(w[1],opd,err);
if (opd) offset++;
opd=0;
get_a(w[2],opd,err);
if (opd) offset++;
}
else if (get_sop(w[0],err) && !err.size())
{
offset++;
if (w.size() < 2) continue;
opd=0;
get_a(w[1],opd,err);
if (opd) offset++;
}
else
{
std::vector<std::string>::iterator wit = w.begin();
Label l;
l.line = lc;
l.offset = offset;
l.name = std::string();
for (;wit!=w.end();wit++)
{
if (wit->size() < 2) continue;
if ((*wit)[wit->size()-1] == ':')
{
l.name = wit->substr(0,wit->size()-1);
break;
}
else if ((*wit)[0] == ':')
{
l.name = wit->substr(1,wit->size());
break;
}
}
if (_labels.find(l.name)!=_labels.end())
{
err="label " + l.name + " redefined";
print_error(lc, false,err);
errc++;
}
else if (l.name.size())
{
_labels[l.name]=l;
}
}
uint16_t* data = new uint16_t[lit->size()];
offset += get_data(*lit, data,err);
delete data;
}
return !errc;
}
void BinAsm::save(const std::string& filename)
{
uint32_t allocsize=0x10000;
uint32_t filesize=0;
int error_count=0;
uint16_t* buff = new uint16_t[allocsize];
unsigned int lc = 0;
uint16_t opcode = 0;
std::vector<std::string> lines=split_text(_src);
std::vector<std::string>::iterator lit = lines.begin();
for (;lit!=lines.end();lit++)
{
lc++;
std::string err = std::string();
std::vector<std::string> w=split_line(*lit);
if (!w.size()) continue;
opcode = 0;
if (filesize > allocsize - lit->size() - 3)
{
std::cerr << "fatal error: assembling file is too big";
std::cerr << "cannot be used on dcpu-16" << std::endl;
break;
}
if (w[0].size() && (w[0][w[0].size()-1] == ':' || w[0][0] == ':'))
{
continue;
}
if ((opcode = get_op(w[0])))
{
if (w.size() != 3)
{
err = "instruction " + w[0];
err += " need 2 arguments";
}
else
{
uint16_t a_word=0, b_word=0;
opcode |= ((get_b(w[1],b_word,err) & 0x1F) << 5);
if (err.size())
{
print_error(lc,false,err);
error_count++;
}
opcode |= (get_a(w[2],a_word,err) << 10);
buff[filesize]=opcode;
filesize++;
if (a_word)
{
buff[filesize]=a_word;
filesize++;
}
if (b_word)
{
buff[filesize]=b_word;
filesize++;
}
}
}
else if ((opcode=get_sop(w[0],err)) && !err.size())
{
if (w.size() != 2)
{
err = "special instruction " + w[0];
err += " need 1 argument";
}
else
{
uint16_t a_word=0;
opcode = ((opcode & 0x1F) << 5);
opcode |= (get_a(w[1],a_word,err) << 10);
buff[filesize]=opcode;
filesize++;
if (a_word)
{
buff[filesize]=a_word;
filesize++;
}
}
}
else {
if (w[0]=="dat"||w[0]==".dat"||w[0]=="DAT"||w[0]==".DAT")
err="";
filesize += get_data(*lit, &(buff[filesize]),err);
}
if (err.size())
{
print_error(lc,false,err);
error_count++;
}
}
if (error_count)
{
std::cerr << "assembling terminated with " << error_count;
std::cerr << " error(s)" << std::endl;
}
else
{
FILE* f = fopen(filename.c_str(), "wb");
if (!f)
std::cerr << "error: cannot open output file " << filename << std::endl;
fswitchendian(buff, filesize);
fwrite(buff,2,filesize,f);
fclose(f);
std::cout << "assembling " << filename;
std::cout << " terminated final size " << filesize*2;
std::cout << " bytes" << std::endl;
}
delete buff;
}
///////////////////////////////////////////////////////////////////////
// Class : CWinDisplay
// Method : data
// Description : Draw the data onto the screen
// Return Value : -
// Comments :
int CWinDisplay::data(int x,int y,int w,int h,float *d) {
int i,j;
clampData(w,h,d);
for (i=0;i<h;i++) {
const float *src = &d[i*w*numSamples];
unsigned int *dest = &imageData[((height-(i+y)-1)*width+x)];
switch(numSamples) {
case 0:
break;
case 1:
for (j=0;j<w;j++) {
unsigned char d = (unsigned char) (src[0]*255);
*dest++ = color(d,d,d,d);
src++;
}
break;
case 2:
for (j=0;j<w;j++) {
const float r = src[0]*src[1]*255 + (1-src[1])*get_r(dest[0]);
const float a = src[1]*255 + (1-src[1])*get_a(dest[0]);
unsigned char dr = (unsigned char) r;
unsigned char da = (unsigned char) a;
*dest++ = color(dr,dr,dr,da);
src += 2;
}
break;
case 3:
for (j=0;j<w;j++) {
unsigned char dr = (unsigned char) (src[0]*255);
unsigned char dg = (unsigned char) (src[1]*255);
unsigned char db = (unsigned char) (src[2]*255);
*dest++ = color(dr,dg,db,(unsigned char) 255);
src += 3;
}
break;
case 4:
for (j=0;j<w;j++) {
const float r = src[0]*src[3]*255 + (1-src[3])*get_r(dest[0]);
const float g = src[1]*src[3]*255 + (1-src[3])*get_g(dest[0]);
const float b = src[2]*src[3]*255 + (1-src[3])*get_b(dest[0]);
const float a = src[3]*255 + (1-src[3])*get_a(dest[0]);
unsigned char dr = (unsigned char) r;
unsigned char dg = (unsigned char) g;
unsigned char db = (unsigned char) b;
unsigned char da = (unsigned char) a;
*dest++ = color(dr,dg,db,da);
src += 4;
}
break;
default:
for (j=0;j<w;j++) {
float r = src[0]*src[3]*255 + (1-src[3])*get_r(*dest);
float g = src[1]*src[3]*255 + (1-src[3])*get_g(*dest);
float b = src[2]*src[3]*255 + (1-src[3])*get_b(*dest);
float a = src[3]*255 + (1-src[3])*get_a(*dest);
unsigned char dr = (unsigned char) r;
unsigned char dg = (unsigned char) g;
unsigned char db = (unsigned char) b;
unsigned char da = (unsigned char) a;
*dest++ = color(dr,dg,db,da);
src += numSamples;
}
break;
}
}
if (active) {
if (willRedraw == FALSE) {
// Pump messages
willRedraw = TRUE;
PostMessage(hWnd,WM_PAINT,(WPARAM) this,0);
}
}
return active;
}
int main()
{
float *arr = get_arr(); // [4, 3, 2, 1]
float *uarr = get_uarr(); // [5, 4, 3, 2]
float *arr2 = get_arr2(); // [4, 3, 2, 1]
float *uarr2 = get_uarr2(); // [5, 4, 3, 2]
__m128 a = get_a(); // [8, 6, 4, 2]
__m128 b = get_b(); // [1, 2, 3, 4]
// Check that test data is like expected.
Assert(((uintptr_t)arr & 0xF) == 0); // arr must be aligned by 16.
Assert(((uintptr_t)uarr & 0xF) != 0); // uarr must be unaligned.
Assert(((uintptr_t)arr2 & 0xF) == 0); // arr must be aligned by 16.
Assert(((uintptr_t)uarr2 & 0xF) != 0); // uarr must be unaligned.
// Test that aeq itself works and does not trivially return true on everything.
Assert(aeq_("",_mm_load_ps(arr), 4.f, 3.f, 2.f, 0.f, false) == false);
#ifdef TEST_M64
Assert(aeq64(u64castm64(0x22446688AACCEEFFULL), 0xABABABABABABABABULL, false) == false);
#endif
// SSE1 Load instructions:
aeq(_mm_load_ps(arr), 4.f, 3.f, 2.f, 1.f); // 4-wide load from aligned address.
aeq(_mm_load_ps1(uarr), 2.f, 2.f, 2.f, 2.f); // Load scalar from unaligned address and populate 4-wide.
aeq(_mm_load_ss(uarr), 0.f, 0.f, 0.f, 2.f); // Load scalar from unaligned address to lowest, and zero all highest.
aeq(_mm_load1_ps(uarr), 2.f, 2.f, 2.f, 2.f); // _mm_load1_ps == _mm_load_ps1
aeq(_mm_loadh_pi(a, (__m64*)uarr), 3.f, 2.f, 4.f, 2.f); // Load two highest addresses, preserve two lowest.
aeq(_mm_loadl_pi(a, (__m64*)uarr), 8.f, 6.f, 3.f, 2.f); // Load two lowest addresses, preserve two highest.
aeq(_mm_loadr_ps(arr), 1.f, 2.f, 3.f, 4.f); // 4-wide load from an aligned address, but reverse order.
aeq(_mm_loadu_ps(uarr), 5.f, 4.f, 3.f, 2.f); // 4-wide load from an unaligned address.
// SSE1 Set instructions:
aeq(_mm_set_ps(uarr[3], 2.f, 3.f, 4.f), 5.f, 2.f, 3.f, 4.f); // 4-wide set by specifying four immediate or memory operands.
aeq(_mm_set_ps1(uarr[3]), 5.f, 5.f, 5.f, 5.f); // 4-wide set by specifying one scalar that is expanded.
aeq(_mm_set_ss(uarr[3]), 0.f, 0.f, 0.f, 5.f); // Set scalar at lowest index, zero all higher.
aeq(_mm_set1_ps(uarr[3]), 5.f, 5.f, 5.f, 5.f); // _mm_set1_ps == _mm_set_ps1
aeq(_mm_setr_ps(uarr[3], 2.f, 3.f, 4.f), 4.f, 3.f, 2.f, 5.f); // 4-wide set by specifying four immediate or memory operands, but reverse order.
aeq(_mm_setzero_ps(), 0.f, 0.f, 0.f, 0.f); // Returns a new zero register.
// SSE1 Move instructions:
aeq(_mm_move_ss(a, b), 8.f, 6.f, 4.f, 4.f); // Copy three highest elements from a, and lowest from b.
aeq(_mm_movehl_ps(a, b), 8.f, 6.f, 1.f, 2.f); // Copy two highest elements from a, and take two highest from b and place them to the two lowest in output.
aeq(_mm_movelh_ps(a, b), 3.f, 4.f, 4.f, 2.f); // Copy two lowest elements from a, and take two lowest from b and place them to the two highest in output.
// SSE1 Store instructions:
#ifdef TEST_M64
/*M64*/*(uint64_t*)uarr = 0xCDCDCDCDCDCDCDCDULL; _mm_maskmove_si64(u64castm64(0x00EEDDCCBBAA9988ULL), u64castm64(0x0080FF7F01FEFF40ULL), (char*)uarr); Assert(*(uint64_t*)uarr == 0xCDEEDDCDCDAA99CDULL); // _mm_maskmove_si64: Conditionally store bytes of a 64-bit value.
/*M64*/*(uint64_t*)uarr = 0xABABABABABABABABULL; _m_maskmovq(u64castm64(0x00EEDDCCBBAA9988ULL), u64castm64(0x0080FF7F01FEFF40ULL), (char*)uarr); Assert(*(uint64_t*)uarr == 0xABEEDDABABAA99ABULL); // _m_maskmovq is an alias to _mm_maskmove_si64.
#endif
_mm_store_ps(arr2, a); aeq(_mm_load_ps(arr2), 8.f, 6.f, 4.f, 2.f); // _mm_store_ps: 4-wide store to aligned memory address.
_mm_store_ps1(arr2, a); aeq(_mm_load_ps(arr2), 2.f, 2.f, 2.f, 2.f); // _mm_store_ps1: Store lowest scalar to aligned address, duplicating the element 4 times.
_mm_storeu_ps(uarr2, _mm_set1_ps(100.f)); _mm_store_ss(uarr2, b); aeq(_mm_loadu_ps(uarr2), 100.f, 100.f, 100.f, 4.f); // _mm_store_ss: Store lowest scalar to unaligned address. Don't adjust higher addresses in memory.
_mm_store_ps(arr2, _mm_set1_ps(100.f)); _mm_store1_ps(arr2, a); aeq(_mm_load_ps(arr2), 2.f, 2.f, 2.f, 2.f); // _mm_store1_ps == _mm_store_ps1
_mm_storeu_ps(uarr2, _mm_set1_ps(100.f)); _mm_storeh_pi((__m64*)uarr2, a); aeq(_mm_loadu_ps(uarr2), 100.f, 100.f, 8.f, 6.f); // _mm_storeh_pi: Store two highest elements to memory.
_mm_storeu_ps(uarr2, _mm_set1_ps(100.f)); _mm_storel_pi((__m64*)uarr2, a); aeq(_mm_loadu_ps(uarr2), 100.f, 100.f, 4.f, 2.f); // _mm_storel_pi: Store two lowest elements to memory.
_mm_storer_ps(arr2, a); aeq(_mm_load_ps(arr2), 2.f, 4.f, 6.f, 8.f); // _mm_storer_ps: 4-wide store to aligned memory address, but reverse the elements on output.
_mm_storeu_ps(uarr2, a); aeq(_mm_loadu_ps(uarr2), 8.f, 6.f, 4.f, 2.f); // _mm_storeu_ps: 4-wide store to unaligned memory address.
#ifdef TEST_M64
/*M64*/_mm_stream_pi((__m64*)uarr, u64castm64(0x0080FF7F01FEFF40ULL)); Assert(*(uint64_t*)uarr == 0x0080FF7F01FEFF40ULL); // _mm_stream_pi: 2-wide store, but with a non-temporal memory cache hint.
#endif
_mm_store_ps(arr2, _mm_set1_ps(100.f)); _mm_stream_ps(arr2, a); aeq(_mm_load_ps(arr2), 8.f, 6.f, 4.f, 2.f); // _mm_stream_ps: 4-wide store, but with a non-temporal memory cache hint.
// SSE1 Arithmetic instructions:
aeq(_mm_add_ps(a, b), 9.f, 8.f, 7.f, 6.f); // 4-wide add.
aeq(_mm_add_ss(a, b), 8.f, 6.f, 4.f, 6.f); // Add lowest element, preserve three highest unchanged from a.
aeq(_mm_div_ps(a, _mm_set_ps(2.f, 3.f, 8.f, 2.f)), 4.f, 2.f, 0.5f, 1.f); // 4-wide div.
aeq(_mm_div_ss(a, _mm_set_ps(2.f, 3.f, 8.f, 8.f)), 8.f, 6.f, 4.f, 0.25f); // Div lowest element, preserve three highest unchanged from a.
aeq(_mm_mul_ps(a, b), 8.f, 12.f, 12.f, 8.f); // 4-wide mul.
aeq(_mm_mul_ss(a, b), 8.f, 6.f, 4.f, 8.f); // Mul lowest element, preserve three highest unchanged from a.
#ifdef TEST_M64
__m64 m1 = get_m1();
/*M64*/aeq64(_mm_mulhi_pu16(m1, u64castm64(0x22446688AACCEEFFULL)), 0x002233440B4C33CFULL); // Multiply u16 channels, and store high parts.
/*M64*/aeq64( _m_pmulhuw(m1, u64castm64(0x22446688AACCEEFFULL)), 0x002233440B4C33CFULL); // _m_pmulhuw is an alias to _mm_mulhi_pu16.
__m64 m2 = get_m2();
/*M64*/aeq64(_mm_sad_pu8(m1, m2), 0x368ULL); // Compute abs. differences of u8 channels, and sum those up to a single 16-bit scalar.
/*M64*/aeq64( _m_psadbw(m1, m2), 0x368ULL); // _m_psadbw is an alias to _mm_sad_pu8.
#endif
aeq(_mm_sub_ps(a, b), 7.f, 4.f, 1.f, -2.f); // 4-wide sub.
aeq(_mm_sub_ss(a, b), 8.f, 6.f, 4.f, -2.f); // Sub lowest element, preserve three highest unchanged from a.
// SSE1 Elementary Math functions:
#ifndef __EMSCRIPTEN__ // TODO: Enable support for this to pass.
aeq(_mm_rcp_ps(a), 0.124969f, 0.166626f, 0.249939f, 0.499878f); // Compute 4-wide 1/x.
aeq(_mm_rcp_ss(a), 8.f, 6.f, 4.f, 0.499878f); // Compute 1/x of lowest element, pass higher elements unchanged.
aeq(_mm_rsqrt_ps(a), 0.353455f, 0.408203f, 0.499878f, 0.706909f); // Compute 4-wide 1/sqrt(x).
aeq(_mm_rsqrt_ss(a), 8.f, 6.f, 4.f, 0.706909f); // Compute 1/sqrt(x) of lowest element, pass higher elements unchanged.
#endif
aeq(_mm_sqrt_ps(a), 2.82843f, 2.44949f, 2.f, 1.41421f); // Compute 4-wide sqrt(x).
aeq(_mm_sqrt_ss(a), 8.f, 6.f, 4.f, 1.41421f); // Compute sqrt(x) of lowest element, pass higher elements unchanged.
__m128 i1 = get_i1();
__m128 i2 = get_i2();
// SSE1 Logical instructions:
#ifndef __EMSCRIPTEN__ // TODO: The polyfill currently does NaN canonicalization and breaks these.
aeqi(_mm_and_ps(i1, i2), 0x83200100, 0x0fecc988, 0x80244021, 0x13458a88); // 4-wide binary AND
aeqi(_mm_andnot_ps(i1, i2), 0x388a9888, 0xf0021444, 0x7000289c, 0x00121046); // 4-wide binary (!i1) & i2
aeqi(_mm_or_ps(i1, i2), 0xbfefdba9, 0xffefdfed, 0xf7656bbd, 0xffffdbef); // 4-wide binary OR
aeqi(_mm_xor_ps(i1, i2), 0x3ccfdaa9, 0xf0031665, 0x77412b9c, 0xecba5167); // 4-wide binary XOR
#endif
// SSE1 Compare instructions:
// a = [8, 6, 4, 2], b = [1, 2, 3, 4]
aeqi(_mm_cmpeq_ps(a, _mm_set_ps(8.f, 0.f, 4.f, 0.f)), 0xFFFFFFFF, 0, 0xFFFFFFFF, 0); // 4-wide cmp ==
aeqi(_mm_cmpeq_ss(a, _mm_set_ps(8.f, 0.f, 4.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0xFFFFFFFF); // scalar cmp ==, pass three highest unchanged.
aeqi(_mm_cmpge_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0xFFFFFFFF, 0, 0xFFFFFFFF, 0); // 4-wide cmp >=
aeqi(_mm_cmpge_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 0.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0xFFFFFFFF); // scalar cmp >=, pass three highest unchanged.
aeqi(_mm_cmpgt_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0, 0, 0xFFFFFFFF, 0); // 4-wide cmp >
aeqi(_mm_cmpgt_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0); // scalar cmp >, pass three highest unchanged.
aeqi(_mm_cmple_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0xFFFFFFFF, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide cmp <=
aeqi(_mm_cmple_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 0.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0); // scalar cmp <=, pass three highest unchanged.
aeqi(_mm_cmplt_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide cmp <
aeqi(_mm_cmplt_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0); // scalar cmp <, pass three highest unchanged.
aeqi(_mm_cmpneq_ps(a, _mm_set_ps(8.f, 0.f, 4.f, 0.f)), 0, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide cmp !=
aeqi(_mm_cmpneq_ss(a, _mm_set_ps(8.f, 0.f, 4.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0); // scalar cmp !=, pass three highest unchanged.
aeqi(_mm_cmpnge_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide cmp not >=
aeqi(_mm_cmpnge_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 0.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0); // scalar cmp not >=, pass three highest unchanged.
aeqi(_mm_cmpngt_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0xFFFFFFFF, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide cmp not >
aeqi(_mm_cmpngt_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0xFFFFFFFF); // scalar cmp not >, pass three highest unchanged.
aeqi(_mm_cmpnle_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0, 0, 0xFFFFFFFF, 0); // 4-wide cmp not <=
aeqi(_mm_cmpnle_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 0.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0xFFFFFFFF); // scalar cmp not <=, pass three highest unchanged.
aeqi(_mm_cmpnlt_ps(a, _mm_set_ps(8.f, 7.f, 3.f, 5.f)), 0xFFFFFFFF, 0, 0xFFFFFFFF, 0); // 4-wide cmp not <
aeqi(_mm_cmpnlt_ss(a, _mm_set_ps(8.f, 7.f, 3.f, 2.f)), fcastu(8.f), fcastu(6.f), fcastu(4.f), 0xFFFFFFFF); // scalar cmp not <, pass three highest unchanged.
__m128 nan1 = get_nan1(); // [NAN, 0, 0, NAN]
__m128 nan2 = get_nan2(); // [NAN, NAN, 0, 0]
aeqi(_mm_cmpord_ps(nan1, nan2), 0, 0, 0xFFFFFFFF, 0); // 4-wide test if both operands are not nan.
aeqi(_mm_cmpord_ss(nan1, nan2), fcastu(NAN), 0, 0, 0); // scalar test if both operands are not nan, pass three highest unchanged.
// Intel Intrinsics Guide documentation is wrong on _mm_cmpunord_ps and _mm_cmpunord_ss. MSDN is right: http://msdn.microsoft.com/en-us/library/khy6fk1t(v=vs.90).aspx
aeqi(_mm_cmpunord_ps(nan1, nan2), 0xFFFFFFFF, 0xFFFFFFFF, 0, 0xFFFFFFFF); // 4-wide test if one of the operands is nan.
#ifndef __EMSCRIPTEN__ // TODO: The polyfill currently does NaN canonicalization and breaks these.
aeqi(_mm_cmpunord_ss(nan1, nan2), fcastu(NAN), 0, 0, 0xFFFFFFFF); // scalar test if one of the operands is nan, pass three highest unchanged.
#endif
Assert(_mm_comieq_ss(a, b) == 0); Assert(_mm_comieq_ss(a, a) == 1); // Scalar cmp == of lowest element, return int.
Assert(_mm_comige_ss(a, b) == 0); Assert(_mm_comige_ss(a, a) == 1); // Scalar cmp >= of lowest element, return int.
Assert(_mm_comigt_ss(b, a) == 1); Assert(_mm_comigt_ss(a, a) == 0); // Scalar cmp > of lowest element, return int.
Assert(_mm_comile_ss(b, a) == 0); Assert(_mm_comile_ss(a, a) == 1); // Scalar cmp <= of lowest element, return int.
Assert(_mm_comilt_ss(a, b) == 1); Assert(_mm_comilt_ss(a, a) == 0); // Scalar cmp < of lowest element, return int.
Assert(_mm_comineq_ss(a, b) == 1); Assert(_mm_comineq_ss(a, a) == 0); // Scalar cmp != of lowest element, return int.
// The ucomi versions are identical to comi, except that ucomi signal a FP exception only if one of the input operands is a SNaN, whereas the comi versions signal a FP
// exception when one of the input operands is either a QNaN or a SNaN.
#ifndef __EMSCRIPTEN__ // TODO: Fix ucomi support in SSE to treat NaNs properly.
Assert(_mm_ucomieq_ss(a, b) == 0); Assert(_mm_ucomieq_ss(a, a) == 1); Assert(_mm_ucomieq_ss(a, nan1) == 1);
#endif
Assert(_mm_ucomige_ss(a, b) == 0); Assert(_mm_ucomige_ss(a, a) == 1); Assert(_mm_ucomige_ss(a, nan1) == 0);
Assert(_mm_ucomigt_ss(b, a) == 1); Assert(_mm_ucomigt_ss(a, a) == 0); Assert(_mm_ucomigt_ss(a, nan1) == 0);
Assert(_mm_ucomile_ss(b, a) == 0); Assert(_mm_ucomile_ss(a, a) == 1); Assert(_mm_ucomile_ss(a, nan1) == 1);
Assert(_mm_ucomilt_ss(a, b) == 1); Assert(_mm_ucomilt_ss(a, a) == 0); Assert(_mm_ucomilt_ss(a, nan1) == 1);
#ifndef __EMSCRIPTEN__ // TODO: Fix ucomi support in SSE to treat NaNs properly.
Assert(_mm_ucomineq_ss(a, b) == 1); Assert(_mm_ucomineq_ss(a, a) == 0); Assert(_mm_ucomineq_ss(a, nan1) == 0);
#endif
// SSE1 Convert instructions:
__m128 c = get_c(); // [1.5, 2.5, 3.5, 4.5]
__m128 e = get_e(); // [INF, -INF, 2.5, 3.5]
__m128 f = get_f(); // [-1.5, 1.5, -2.5, -9223372036854775808]
#ifdef TEST_M64
/*M64*/aeq(_mm_cvt_pi2ps(a, m2), 8.f, 6.f, -19088744.f, 1985229312.f); // 2-way int32 to float conversion to two lowest channels of m128.
/*M64*/aeq64(_mm_cvt_ps2pi(c), 0x400000004ULL); // 2-way two lowest floats from m128 to integer, return as m64.
#endif
aeq(_mm_cvtsi32_ss(c, -16777215), 1.5f, 2.5f, 3.5f, -16777215.f); // Convert int to float, store in lowest channel of m128.
aeq( _mm_cvt_si2ss(c, -16777215), 1.5f, 2.5f, 3.5f, -16777215.f); // _mm_cvt_si2ss is an alias to _mm_cvtsi32_ss.
#ifndef __EMSCRIPTEN__ // TODO: Fix banker's rounding in cvt functions.
Assert(_mm_cvtss_si32(c) == 4); Assert(_mm_cvtss_si32(e) == 4); // Convert lowest channel of m128 from float to int.
Assert( _mm_cvt_ss2si(c) == 4); Assert( _mm_cvt_ss2si(e) == 4); // _mm_cvt_ss2si is an alias to _mm_cvtss_si32.
#endif
#ifdef TEST_M64
/*M64*/aeq(_mm_cvtpi16_ps(m1), 255.f , -32767.f, 4336.f, 14207.f); // 4-way convert int16s to floats, return in a m128.
/*M64*/aeq(_mm_cvtpi32_ps(a, m1), 8.f, 6.f, 16744449.f, 284178304.f); // 2-way convert int32s to floats, return in two lowest channels of m128, pass two highest unchanged.
/*M64*/aeq(_mm_cvtpi32x2_ps(m1, m2), -19088744.f, 1985229312.f, 16744449.f, 284178304.f); // 4-way convert int32s from two different m64s to float.
/*M64*/aeq(_mm_cvtpi8_ps(m1), 16.f, -16.f, 55.f, 127.f); // 4-way convert int8s from lowest end of m64 to float in a m128.
/*M64*/aeq64(_mm_cvtps_pi16(c), 0x0002000200040004ULL); // 4-way convert floats to int16s in a m64.
/*M64*/aeq64(_mm_cvtps_pi32(c), 0x0000000400000004ULL); // 2-way convert two lowest floats to int32s in a m64.
/*M64*/aeq64(_mm_cvtps_pi8(c), 0x0000000002020404ULL); // 4-way convert floats to int8s in a m64, zero higher half of the returned m64.
/*M64*/aeq(_mm_cvtpu16_ps(m1), 255.f , 32769.f, 4336.f, 14207.f); // 4-way convert uint16s to floats, return in a m128.
/*M64*/aeq(_mm_cvtpu8_ps(m1), 16.f, 240.f, 55.f, 127.f); // 4-way convert uint8s from lowest end of m64 to float in a m128.
#endif
aeq(_mm_cvtsi64_ss(c, -9223372036854775808ULL), 1.5f, 2.5f, 3.5f, -9223372036854775808.f); // Convert single int64 to float, store in lowest channel of m128, and pass three higher channel unchanged.
Assert(_mm_cvtss_f32(c) == 4.5f); // Extract lowest channel of m128 to a plain old float.
Assert(_mm_cvtss_si64(f) == -9223372036854775808ULL); // Convert lowest channel of m128 from float to int64.
#ifdef TEST_M64
/*M64*/aeq64(_mm_cvtt_ps2pi(e), 0x0000000200000003ULL); aeq64(_mm_cvtt_ps2pi(f), 0xfffffffe80000000ULL); // Truncating conversion from two lowest floats of m128 to int32s, return in a m64.
#endif
Assert(_mm_cvttss_si32(e) == 3); // Truncating conversion from the lowest float of a m128 to int32.
Assert( _mm_cvtt_ss2si(e) == 3); // _mm_cvtt_ss2si is an alias to _mm_cvttss_si32.
#ifdef TEST_M64
/*M64*/aeq64(_mm_cvttps_pi32(c), 0x0000000300000004ULL); // Truncating conversion from two lowest floats of m128 to m64.
#endif
Assert(_mm_cvttss_si64(f) == -9223372036854775808ULL); // Truncating conversion from lowest channel of m128 from float to int64.
#ifndef __EMSCRIPTEN__ // TODO: Not implemented.
// SSE1 General support:
unsigned int mask = _MM_GET_EXCEPTION_MASK();
_MM_SET_EXCEPTION_MASK(mask);
unsigned int flushZeroMode = _MM_GET_FLUSH_ZERO_MODE();
_MM_SET_FLUSH_ZERO_MODE(flushZeroMode);
unsigned int roundingMode = _MM_GET_ROUNDING_MODE();
_MM_SET_ROUNDING_MODE(roundingMode);
unsigned int csr = _mm_getcsr();
_mm_setcsr(csr);
unsigned char dummyData[4096];
_mm_prefetch(dummyData, _MM_HINT_T0);
_mm_prefetch(dummyData, _MM_HINT_T1);
_mm_prefetch(dummyData, _MM_HINT_T2);
_mm_prefetch(dummyData, _MM_HINT_NTA);
_mm_sfence();
#endif
// SSE1 Misc instructions:
#ifdef TEST_M64
/*M64*/Assert(_mm_movemask_pi8(m1) == 100); // Return int with eight lowest bits set depending on the highest bits of the 8 uint8 input channels of the m64.
/*M64*/Assert( _m_pmovmskb(m1) == 100); // _m_pmovmskb is an alias to _mm_movemask_pi8.
#endif
Assert(_mm_movemask_ps(_mm_set_ps(-1.f, 0.f, 1.f, NAN)) == 8); Assert(_mm_movemask_ps(_mm_set_ps(-INFINITY, -0.f, INFINITY, -INFINITY)) == 13); // Return int with four lowest bits set depending on the highest bits of the 4 m128 input channels.
// SSE1 Probability/Statistics instructions:
#ifdef TEST_M64
/*M64*/aeq64(_mm_avg_pu16(m1, m2), 0x7FEE9D4D43A234C8ULL); // 4-way average uint16s.
/*M64*/aeq64( _m_pavgw(m1, m2), 0x7FEE9D4D43A234C8ULL); // _m_pavgw is an alias to _mm_avg_pu16.
/*M64*/aeq64(_mm_avg_pu8(m1, m2), 0x7FEE9D4D43A23548ULL); // 8-way average uint8s.
/*M64*/aeq64( _m_pavgb(m1, m2), 0x7FEE9D4D43A23548ULL); // _m_pavgb is an alias to _mm_avg_pu8.
// SSE1 Special Math instructions:
/*M64*/aeq64(_mm_max_pi16(m1, m2), 0xFFBA987654377FULL); // 4-way average uint16s.
/*M64*/aeq64( _m_pmaxsw(m1, m2), 0xFFBA987654377FULL); // _m_pmaxsw is an alias to _mm_max_pi16.
/*M64*/aeq64(_mm_max_pu8(m1, m2), 0xFEFFBA9876F0377FULL); // 4-way average uint16s.
/*M64*/aeq64( _m_pmaxub(m1, m2), 0xFEFFBA9876F0377FULL); // _m_pmaxub is an alias to _mm_max_pu8.
/*M64*/aeq64(_mm_min_pi16(m1, m2), 0xFEDC800110F03210ULL); // 4-way average uint16s.
/*M64*/aeq64( _m_pminsw(m1, m2), 0xFEDC800110F03210ULL); // is an alias to _mm_min_pi16.
/*M64*/aeq64(_mm_min_pu8(m1, m2), 0xDC800110543210ULL); // 4-way average uint16s.
/*M64*/aeq64( _m_pminub(m1, m2), 0xDC800110543210ULL); // is an alias to _mm_min_pu8.
#endif
// a = [8, 6, 4, 2], b = [1, 2, 3, 4]
aeq(_mm_max_ps(a, b), 8.f, 6.f, 4.f, 4.f); // 4-wide max.
aeq(_mm_max_ss(a, _mm_set1_ps(100.f)), 8.f, 6.f, 4.f, 100.f); // Scalar max, pass three highest unchanged.
aeq(_mm_min_ps(a, b), 1.f, 2.f, 3.f, 2.f); // 4-wide min.
aeq(_mm_min_ss(a, _mm_set1_ps(-100.f)), 8.f, 6.f, 4.f, -100.f); // Scalar min, pass three highest unchanged.
// SSE1 Swizzle instructions:
#ifdef TEST_M64
/*M64*/Assert(_mm_extract_pi16(m1, 1) == 4336); // Extract the given int16 channel from a m64.
/*M64*/Assert( _m_pextrw(m1, 1) == 4336); // _m_pextrw is an alias to _mm_extract_pi16.
/*M64*/aeq64(_mm_insert_pi16(m1, 0xABCD, 1), 0xFF8001ABCD377FULL); // Insert a int16 to a specific channel of a m64.
/*M64*/aeq64( _m_pinsrw(m1, 0xABCD, 1), 0xFF8001ABCD377FULL); // _m_pinsrw is an alias to _mm_insert_pi16.
/*M64*/aeq64(_mm_shuffle_pi16(m1, _MM_SHUFFLE(1, 0, 3, 2)), 0x10F0377F00FF8001ULL); // Shuffle int16s around in the 4 channels of the m64.
/*M64*/aeq64( _m_pshufw(m1, _MM_SHUFFLE(1, 0, 3, 2)), 0x10F0377F00FF8001ULL); // _m_pshufw is an alias to _mm_shuffle_pi16.
#endif
aeq(_mm_shuffle_ps(a, b, _MM_SHUFFLE(1, 0, 3, 2)), 3.f, 4.f, 8.f, 6.f);
aeq(_mm_unpackhi_ps(a, b), 1.f , 8.f, 2.f, 6.f);
aeq(_mm_unpacklo_ps(a, b), 3.f , 4.f, 4.f, 2.f);
// Transposing a matrix via the xmmintrin.h-provided intrinsic.
__m128 c0 = a; // [8, 6, 4, 2]
__m128 c1 = b; // [1, 2, 3, 4]
__m128 c2 = get_c(); // [1.5, 2.5, 3.5, 4.5]
__m128 c3 = get_d(); // [8.5, 6.5, 4.5, 2.5]
_MM_TRANSPOSE4_PS(c0, c1, c2, c3);
aeq(c0, 2.5f, 4.5f, 4.f, 2.f);
aeq(c1, 4.5f, 3.5f, 3.f, 4.f);
aeq(c2, 6.5f, 2.5f, 2.f, 6.f);
aeq(c3, 8.5f, 1.5f, 1.f, 8.f);
// All done!
if (numFailures == 0)
printf("Success!\n");
else
printf("%d tests failed!\n", numFailures);
}
