Skip to content

Replace AVX2/FMA with SIMDe (SIMD everywhere) #237

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
rwk-unil wants to merge 4 commits into
base: master
Choose a base branch
from
Open

Replace AVX2/FMA with SIMDe (SIMD everywhere) #237

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
f14177d
phase: add SIMDe
rwk-unil Nov 11, 2024
3a9458e
phase: models: phasing_hmm: Replace AVX2 and FMA with SIMDe
rwk-unil Nov 8, 2024
ee77135
phase: models: imputation_hmm: Replace AVX2 and FMA with SIMDe
rwk-unil Nov 8, 2024
ea1ab10
phase: makefile: include SIMDe
rwk-unil Nov 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
4 changes: 3 additions & 1 deletion phase/makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,6 +87,8 @@ ifeq ($(BGEN_SUPPORT),YES)
CXXFLAG+= -D__BGEN__ -I$(BGEN_INC)
endif

SIMDE_INC=./src/simde

#CONDITIONAL PATH DEFINITON

system: DYN_LIBS=-lz -lpthread -lbz2 -llzma
Expand Down Expand Up @@ -156,7 +158,7 @@ $(EXEFILE): $(OFILE)
$(CXX) $(LDFLAG) -static -static-libgcc -static-libstdc++ -pthread -o $(EXEFILE) $^ $(HTSLIB_LIB) $(BOOST_LIB_IO) $(BOOST_LIB_PO) $(BOOST_LIB_SE) -Wl,-Bstatic $(DYN_LIBS)

obj/%.o: %.cpp $(HFILE)
$(CXX) $(CXXFLAG) -c $< -o $@ -Isrc -I$(HTSLIB_INC) -I$(BOOST_INC)
$(CXX) $(CXXFLAG) -c $< -o $@ -Isrc -I$(HTSLIB_INC) -I$(BOOST_INC) -I$(SIMDE_INC)

clean:
rm -f obj/*.o $(BFILE)
186 changes: 93 additions & 93 deletions phase/src/models/imputation_hmm.cpp
View file
Open in desktop

Large diffs are not rendered by default.

5 changes: 4 additions & 1 deletion phase/src/models/imputation_hmm.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,10 @@

#include <utils/otools.h>
#include <containers/conditioning_set.h>
#include <immintrin.h>
//#include <immintrin.h>
//#define SIMDE_ENABLE_NATIVE_ALIASES
#include <simde/x86/avx2.h>
#include <simde/x86/fma.h>
#include <boost/align/aligned_allocator.hpp>

template <typename T>
Expand Down
181 changes: 92 additions & 89 deletions phase/src/models/phasing_hmm.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,10 @@

#include <utils/otools.h>
#include <containers/conditioning_set.h>
#include <immintrin.h>
//#include <immintrin.h>
//#define SIMDE_ENABLE_NATIVE_ALIASES
#include <simde/x86/avx2.h>
#include <simde/x86/fma.h>
#include <boost/align/aligned_allocator.hpp>

template <typename T>
Expand All @@ -43,16 +46,16 @@ using aligned_vector32 = std::vector<T, boost::alignment::aligned_allocator < T,
#define ALLELE(hap , pos) ((hap) & (1<<(pos)))

inline
float horizontal_add (const __m256& a) //@simo: implemented horizontal add with instructions in the 4B minimum
float horizontal_add (const simde__m256& a) //@simo: implemented horizontal add with instructions in the 4B minimum
{
__m128 vlow = _mm256_castps256_ps128(a);
__m128 vhigh = _mm256_extractf128_ps(a, 1); // high 128
vlow = _mm_add_ps(vlow, vhigh); // add the low 128
__m128 shuf = _mm_movehdup_ps(vlow); // broadcast elements 3,1 to 2,0
__m128 sums = _mm_add_ps(vlow, shuf);
shuf = _mm_movehl_ps(shuf, sums); // high half -> low half
sums = _mm_add_ss(sums, shuf);
return _mm_cvtss_f32(sums);
simde__m128 vlow = simde_mm256_castps256_ps128(a);
simde__m128 vhigh = simde_mm256_extractf128_ps(a, 1); // high 128
vlow = simde_mm_add_ps(vlow, vhigh); // add the low 128
simde__m128 shuf = simde_mm_movehdup_ps(vlow); // broadcast elements 3,1 to 2,0
simde__m128 sums = simde_mm_add_ps(vlow, shuf);
shuf = simde_mm_movehl_ps(shuf, sums); // high half -> low half
sums = simde_mm_add_ss(sums, shuf);
return simde_mm_cvtss_f32(sums);
}

class phasing_hmm {
Expand Down Expand Up @@ -137,30 +140,30 @@ class phasing_hmm {
inline
void phasing_hmm::INIT_PEAK_HET(int curr_het)
{
const std::array <__m256, 2 > emits = {_mm256_load_ps(&EMIT0[curr_het][0]),_mm256_load_ps(&EMIT1[curr_het][0])};
__m256 _sum = _mm256_set1_ps(0.0f);
const std::array <simde__m256, 2 > emits = {simde_mm256_load_ps(&EMIT0[curr_het][0]),simde_mm256_load_ps(&EMIT1[curr_het][0])};
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ah = C->Hvar.get(curr_rel_locus, k);
_sum = _mm256_add_ps(_sum, emits[ah]);
_mm256_store_ps(&prob[i], emits[ah]);
_sum = simde_mm256_add_ps(_sum, emits[ah]);
simde_mm256_store_ps(&prob[i], emits[ah]);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::INIT_PEAK_HOM(bool ag)
{
const std::array <__m256, 2 > emits = {_mm256_set1_ps(1.0f),_mm256_set1_ps(C->ed_phs/C->ee_phs)};
__m256 _sum = _mm256_set1_ps(0.0f);
const std::array <simde__m256, 2 > emits = {simde_mm256_set1_ps(1.0f),simde_mm256_set1_ps(C->ed_phs/C->ee_phs)};
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ag_ah = C->Hvar.get(curr_rel_locus, k)!=ag;
_sum = _mm256_add_ps(_sum, emits[ag_ah]);
_mm256_store_ps(&prob[i], emits[ag_ah]);
_sum = simde_mm256_add_ps(_sum, emits[ag_ah]);
simde_mm256_store_ps(&prob[i], emits[ag_ah]);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

Expand All @@ -174,116 +177,116 @@ void phasing_hmm::INIT_FLAT_HET() {
inline
void phasing_hmm::RUN_PEAK_HET(int curr_het)
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
const std::array <__m256, 2 > emits = {_mm256_load_ps(&EMIT0[curr_het][0]),_mm256_load_ps(&EMIT1[curr_het][0])};
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
const std::array <simde__m256, 2 > emits = {simde_mm256_load_ps(&EMIT0[curr_het][0]),simde_mm256_load_ps(&EMIT1[curr_het][0])};
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ah = C->Hvar.get(curr_rel_locus, k);
const __m256 _prob_curr = _mm256_mul_ps(_mm256_fmadd_ps(_mm256_load_ps(&prob[i]), _nt, _tFreq), emits[ah]);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
const simde__m256 _prob_curr = simde_mm256_mul_ps(simde_mm256_fmadd_ps(simde_mm256_load_ps(&prob[i]), _nt, _tFreq), emits[ah]);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::RUN_PEAK_HOM(bool ag)
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
const __m256 _mism = _mm256_set1_ps(C->ed_phs/C->ee_phs);
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
const simde__m256 _mism = simde_mm256_set1_ps(C->ed_phs/C->ee_phs);
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ah = C->Hvar.get(curr_rel_locus, k);
const __m256 _prob_prev = _mm256_load_ps(&prob[i]);
__m256 _prob_curr = _mm256_fmadd_ps(_prob_prev, _nt, _tFreq);
if (ag!=ah) _prob_curr = _mm256_mul_ps(_prob_curr, _mism);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
const simde__m256 _prob_prev = simde_mm256_load_ps(&prob[i]);
simde__m256 _prob_curr = simde_mm256_fmadd_ps(_prob_prev, _nt, _tFreq);
if (ag!=ah) _prob_curr = simde_mm256_mul_ps(_prob_curr, _mism);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::RUN_FLAT_HET()
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const __m256 _prob_curr = _mm256_fmadd_ps(_mm256_load_ps(&prob[i]), _nt, _tFreq);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
const simde__m256 _prob_curr = simde_mm256_fmadd_ps(simde_mm256_load_ps(&prob[i]), _nt, _tFreq);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::COLLAPSE_PEAK_HET(int curr_het)
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
const std::array <__m256, 2 > emits = {_mm256_load_ps(&EMIT0[curr_het][0]),_mm256_load_ps(&EMIT1[curr_het][0])};
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
const std::array <simde__m256, 2 > emits = {simde_mm256_load_ps(&EMIT0[curr_het][0]),simde_mm256_load_ps(&EMIT1[curr_het][0])};
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ah = C->Hvar.get(curr_rel_locus, k);
const __m256 _prob_curr = _mm256_mul_ps(_mm256_fmadd_ps(_mm256_set1_ps(probSumK[k]), _nt, _tFreq), emits[ah]);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
const simde__m256 _prob_curr = simde_mm256_mul_ps(simde_mm256_fmadd_ps(simde_mm256_set1_ps(probSumK[k]), _nt, _tFreq), emits[ah]);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::COLLAPSE_PEAK_HOM(bool ag)
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
__m256 _sum = _mm256_set1_ps(0.0f);
const __m256 _mism = _mm256_set1_ps(C->ed_phs/C->ee_phs);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
const simde__m256 _mism = simde_mm256_set1_ps(C->ed_phs/C->ee_phs);

for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const bool ah = C->Hvar.get(curr_rel_locus, k);
__m256 _prob_curr = _mm256_fmadd_ps(_mm256_set1_ps(probSumK[k]), _nt, _tFreq);
if (ag!=ah) _prob_curr = _mm256_mul_ps(_prob_curr, _mism);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
simde__m256 _prob_curr = simde_mm256_fmadd_ps(simde_mm256_set1_ps(probSumK[k]), _nt, _tFreq);
if (ag!=ah) _prob_curr = simde_mm256_mul_ps(_prob_curr, _mism);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}

inline
void phasing_hmm::COLLAPSE_FLAT_HET ()
{
const __m256 _tFreq = _mm256_mul_ps(_mm256_load_ps(&probSumH[0]), _mm256_set1_ps(yt / (C->n_states * probSumT)));
const __m256 _nt = _mm256_set1_ps(nt / probSumT);
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _tFreq = simde_mm256_mul_ps(simde_mm256_load_ps(&probSumH[0]), simde_mm256_set1_ps(yt / (C->n_states * probSumT)));
const simde__m256 _nt = simde_mm256_set1_ps(nt / probSumT);
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
{
__m256 _prob_curr = _mm256_fmadd_ps(_mm256_set1_ps(probSumK[k]), _nt, _tFreq);
_sum = _mm256_add_ps(_sum, _prob_curr);
_mm256_store_ps(&prob[i], _prob_curr);
simde__m256 _prob_curr = simde_mm256_fmadd_ps(simde_mm256_set1_ps(probSumK[k]), _nt, _tFreq);
_sum = simde_mm256_add_ps(_sum, _prob_curr);
simde_mm256_store_ps(&prob[i], _prob_curr);
}
_mm256_store_ps(&probSumH[0], _sum);
simde_mm256_store_ps(&probSumH[0], _sum);
probSumT = horizontal_add(_sum);
}


inline
void phasing_hmm::SUMK() {
for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER)
probSumK[k] = horizontal_add(_mm256_load_ps(&prob[i]));
probSumK[k] = horizontal_add(simde_mm256_load_ps(&prob[i]));
}

inline
Expand All @@ -293,18 +296,18 @@ bool phasing_hmm::TRANS_HAP()
sumHProbs = 0.0f;
yt = C->getTransition(VAR_ABS[curr_idx_locus], VAR_ABS[curr_idx_locus+1]);
nt = 1.0f - yt;
const __m256 _fact2 = _mm256_set1_ps( nt / probSumT);
const simde__m256 _fact2 = simde_mm256_set1_ps( nt / probSumT);
int h1 = 0, j=0;
for (; h1 < HAP_NUMBER ; h1++, j += HAP_NUMBER)
{
const __m256 _fact1 = _mm256_set1_ps((probSumH[h1]/probSumT) * yt / C->n_states);
__m256 _sum = _mm256_set1_ps(0.0f);
const simde__m256 _fact1 = simde_mm256_set1_ps((probSumH[h1]/probSumT) * yt / C->n_states);
simde__m256 _sum = simde_mm256_set1_ps(0.0f);
for(int k=0, i=0; k != C->n_states ; ++k, i += HAP_NUMBER)
{
const __m256 _prob0 = _mm256_fmadd_ps(_mm256_set1_ps(prob[i+h1]), _fact2, _fact1);
_sum = _mm256_add_ps(_sum, _mm256_mul_ps(_prob0, _mm256_load_ps(&phasingProb[(curr_segment_index+1)*states_haps+i])));
const simde__m256 _prob0 = simde_mm256_fmadd_ps(simde_mm256_set1_ps(prob[i+h1]), _fact2, _fact1);
_sum = simde_mm256_add_ps(_sum, simde_mm256_mul_ps(_prob0, simde_mm256_load_ps(&phasingProb[(curr_segment_index+1)*states_haps+i])));
}
_mm256_store_ps(&HProbs[j], _sum);
simde_mm256_store_ps(&HProbs[j], _sum);
sumHProbs += horizontal_add(_sum);
}
return (std::isnan(sumHProbs) || std::isinf(sumHProbs) || sumHProbs < std::numeric_limits<float>::min());
Expand All @@ -328,23 +331,23 @@ inline
void phasing_hmm::IMPUTE_FLAT_HET()
{
const int states_haps = C->n_states*HAP_NUMBER;
const __m256 _one = _mm256_set1_ps(1.0f);
const __m256 _zero = _mm256_set1_ps(0.0f);
__m256 _scaleR = _mm256_load_ps(&imputeProbSum[curr_missing_locus*HAP_NUMBER]);
__m256 _scaleL = _mm256_load_ps(&probSumH[0]);
_scaleR = _mm256_div_ps(_one, _scaleR);
_scaleL = _mm256_div_ps(_one, _scaleL);
std::array <__m256, 2 > sums = {_mm256_set1_ps(0.0f),_mm256_set1_ps(0.0f)};
const simde__m256 _one = simde_mm256_set1_ps(1.0f);
const simde__m256 _zero = simde_mm256_set1_ps(0.0f);
simde__m256 _scaleR = simde_mm256_load_ps(&imputeProbSum[curr_missing_locus*HAP_NUMBER]);
simde__m256 _scaleL = simde_mm256_load_ps(&probSumH[0]);
_scaleR = simde_mm256_div_ps(_one, _scaleR);
_scaleL = simde_mm256_div_ps(_one, _scaleL);
std::array <simde__m256, 2 > sums = {simde_mm256_set1_ps(0.0f),simde_mm256_set1_ps(0.0f)};

for(int k = 0, i = 0 ; k != C->n_states ; ++k, i += HAP_NUMBER) {
const bool ah = C->Hvar.get(curr_rel_locus, k);
const __m256 _p1 = _mm256_mul_ps(_mm256_load_ps(&imputeProb[curr_missing_locus*states_haps + i]), _scaleR);
const __m256 _p2 = _mm256_mul_ps(_mm256_load_ps(&prob[i]), _scaleL);
sums[ah] = _mm256_add_ps(sums[ah], _mm256_mul_ps(_p1,_p2));
const simde__m256 _p1 = simde_mm256_mul_ps(simde_mm256_load_ps(&imputeProb[curr_missing_locus*states_haps + i]), _scaleR);
const simde__m256 _p2 = simde_mm256_mul_ps(simde_mm256_load_ps(&prob[i]), _scaleL);
sums[ah] = simde_mm256_add_ps(sums[ah], simde_mm256_mul_ps(_p1,_p2));
}
const __m256 _norm_sum = _mm256_div_ps(sums[1], _mm256_add_ps(sums[0], sums[1]));
const __m256 _clamp = _mm256_max_ps(_zero, _mm256_min_ps(_one, _norm_sum));
_mm256_store_ps(&imputeProbOf1s[curr_missing_locus*HAP_NUMBER], _clamp);
const simde__m256 _norm_sum = simde_mm256_div_ps(sums[1], simde_mm256_add_ps(sums[0], sums[1]));
const simde__m256 _clamp = simde_mm256_max_ps(_zero, simde_mm256_min_ps(_one, _norm_sum));
simde_mm256_store_ps(&imputeProbOf1s[curr_missing_locus*HAP_NUMBER], _clamp);
}

#endif
Expand Down
20 changes: 20 additions & 0 deletions phase/src/simde/COPYING
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,20 @@
Copyright (c) 2017 Evan Nemerson <evan@nemerson.com>

Permission is hereby granted, free of charge, to any person obtaining
a copy of this software and associated documentation files (the
"Software"), to deal in the Software without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Software, and to
permit persons to whom the Software is furnished to do so, subject to
the following conditions:

The above copyright notice and this permission notice shall be
included in all copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE
LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Loading