From 3a423fc610b00450c9c0f7307aeaf2c7ad366bf0 Mon Sep 17 00:00:00 2001
From: "se.gorbunov" <se.gorbunov@gsi.de>
Date: Sat, 3 Sep 2022 21:23:12 +0000
Subject: [PATCH] L1: prepare for Vc
---
reco/L1/L1Algo/L1Def.h | 7 +-
reco/L1/L1Algo/L1Filtration.h | 6 +-
reco/L1/L1Algo/L1FitMaterial.cxx | 8 +--
reco/L1/L1Algo/L1TrackFitter.cxx | 30 ++++-----
reco/L1/L1Algo/L1TrackParFit.cxx | 20 +++---
reco/L1/L1Algo/L1TrackParFit.h | 64 +++++++++----------
.../CbmL1RichENNRingFinderParallel.cxx | 12 ++--
.../CbmL1RichENNRingFinderParallel.h | 4 +-
reco/L1/ParticleFinder/CbmL1PFFitter.cxx | 19 +++---
reco/L1/vectors/P4_F32vec4.h | 30 ++++++---
10 files changed, 104 insertions(+), 96 deletions(-)
diff --git a/reco/L1/L1Algo/L1Def.h b/reco/L1/L1Algo/L1Def.h
index 141fba1c6d..7178406074 100644
--- a/reco/L1/L1Algo/L1Def.h
+++ b/reco/L1/L1Algo/L1Def.h
@@ -13,12 +13,7 @@
#include <assert.h>
-#if defined(__CLING__) && defined(__arm64__)
-#include "vectors/PSEUDO_F32vec4.h"
-#else
-#include "vectors/P4_F32vec4.h"
-#endif
-
+#include "vectors/L1vec.h"
#ifdef FAST_CODE
diff --git a/reco/L1/L1Algo/L1Filtration.h b/reco/L1/L1Algo/L1Filtration.h
index 8193d11382..4aecad1227 100644
--- a/reco/L1/L1Algo/L1Filtration.h
+++ b/reco/L1/L1Algo/L1Filtration.h
@@ -37,7 +37,7 @@ inline void FilterTime(L1TrackPar& T, fvec t, fvec dt, fvec timeInfo = fvec::One
// it helps to keep the initial time errors reasonably small
// the calculations in the covariance matrix are not affected
- const fvec maskDoFilter = (HCH < dt2 * 16.f);
+ const fmask maskDoFilter = (HCH < dt2 * 16.f);
//const fvec maskDoFilter = _f32vec4_true;
fvec wi = w / (dt2 + 1.0000001f * HCH);
@@ -106,7 +106,7 @@ inline void L1Filter(L1TrackPar& T, const L1UMeasurementInfo& info, fvec u, fvec
// it helps to keep the initial track errors reasonably small
// the calculations in the covariance matrix are not affected
- const fvec maskDoFilter = (HCH < info.sigma2 * 16.f);
+ const fmask maskDoFilter = (HCH < info.sigma2 * 16.f);
//const fvec maskDoFilter = _f32vec4_true;
// correction to HCH is needed for the case when sigma2 is so small
@@ -176,7 +176,7 @@ inline void L1FilterNoField(L1TrackPar& T, const L1UMeasurementInfo& info, fvec
F5 = info.cos_phi * T.C50 + info.sin_phi * T.C51;
//const fmask maskDoFilter = (HCH < info.sigma2 * 16.f);
- const fvec maskDoFilter(fvec::MaskOne());
+ const fmask maskDoFilter(MaskOne());
//TODO: SG: try this
//fvec wi = w / (info.sigma2 + 1.0000001f * HCH);
diff --git a/reco/L1/L1Algo/L1FitMaterial.cxx b/reco/L1/L1Algo/L1FitMaterial.cxx
index 6619648eeb..09eeefaa48 100644
--- a/reco/L1/L1Algo/L1FitMaterial.cxx
+++ b/reco/L1/L1Algo/L1FitMaterial.cxx
@@ -148,7 +148,7 @@ fvec L1Fit::ApproximateBetheBloch(const fvec& bg2)
const fvec x = 0.5f * log(bg2);
const fvec lhwI = log(28.816f * 1e-9f * sqrt(rho * mZA) / mI);
- fvec init = x > x1;
+ fmask init = x > x1;
d2 = masked(lhwI + x - 0.5f, init);
const fvec r = (x1 - x) / (x1 - x0);
init = (x > x0) & (x1 > x);
@@ -195,7 +195,7 @@ fvec L1Fit::ApproximateBetheBloch(const fvec& bg2, const fvec& kp0, const fvec&
const fvec x = 0.5f * log(bg2);
const fvec lhwI = log(28.816f * 1e-9f * sqrt(rho * mZA) / mI);
- fvec init = x > x1;
+ fmask init = x > x1;
d2 = masked(lhwI + x - 0.5f, init);
const fvec r = (x1 - x) / (x1 - x0);
init = (x > x0) & (x1 > x);
@@ -228,7 +228,7 @@ void L1Fit::EnergyLossCorrection(L1TrackPar& T, const fvec& radThick, fvec& qp0,
const fvec E2Corrected = (sqrt(E2) + direction * dE) * (sqrt(E2) + direction * dE);
fvec corr = sqrt(p2 / (E2Corrected - fMass2));
- fvec ok = (corr == corr) & (fvec::Zero() < w);
+ fmask ok = (corr == corr) & (fvec::Zero() < w);
corr = if3(ok, corr, fvec::One());
qp0 *= corr;
@@ -270,7 +270,7 @@ void L1Fit::EnergyLossCorrection(float atomicA, float rho, float radLen, L1Track
const fvec E2Corrected = (sqrt(E2) + direction * dE) * (sqrt(E2) + direction * dE);
fvec corr = sqrt(p2 / (E2Corrected - fMass2));
- fvec ok = (corr == corr) & (fvec::Zero() < w);
+ fmask ok = (corr == corr) & (fvec::Zero() < w);
corr = if3(ok, corr, fvec::One());
qp0 *= corr;
diff --git a/reco/L1/L1Algo/L1TrackFitter.cxx b/reco/L1/L1Algo/L1TrackFitter.cxx
index 1bee3607e0..2e10696e8b 100644
--- a/reco/L1/L1Algo/L1TrackFitter.cxx
+++ b/reco/L1/L1Algo/L1TrackFitter.cxx
@@ -530,10 +530,10 @@ void L1Algo::L1KFTrackFitter()
fldB0.Combine(fB[i], w[i]);
fld.Set(fldB0, fldZ0, fldB1, fldZ1, fldB2, fldZ2);
- fvec initialised = (z[i] < z_end) & (z_start <= z[i]);
- fvec w1 = masked(w[i], initialised);
- fvec w1_time = masked(w_time[i], initialised);
- fvec wIn = masked(fvec::One(), initialised);
+ fmask initialised = (z[i] < z_end) & (z_start <= z[i]);
+ fvec w1 = masked(w[i], initialised);
+ fvec w1_time = masked(w_time[i], initialised);
+ fvec wIn = masked(fvec::One(), initialised);
fld1 = fld;
@@ -692,10 +692,10 @@ void L1Algo::L1KFTrackFitter()
fldB0.Combine(fB[i], w[i]);
fld.Set(fldB0, fldZ0, fldB1, fldZ1, fldB2, fldZ2);
- fvec initialised = (z[i] <= z_end) & (z_start < z[i]);
- fvec w1 = masked(w[i], initialised);
- fvec w1_time = masked(w_time[i], initialised);
- fvec wIn = masked(fvec::One(), initialised);
+ fmask initialised = (z[i] <= z_end) & (z_start < z[i]);
+ fvec w1 = masked(w[i], initialised);
+ fvec w1_time = masked(w_time[i], initialised);
+ fvec wIn = masked(fvec::One(), initialised);
L1Extrapolate(T, z[i], qp0, fld, &w1);
@@ -1002,9 +1002,9 @@ void L1Algo::L1KFTrackFitterMuch()
fld.Set(fldB2, fldZ2, fldB1, fldZ1, fldB0, fldZ0);
for (++i; i < nHits; i++) {
- fvec initialised = (z[i] <= z_end) & (z_start < z[i]);
- fvec w1 = masked(w[i], initialised);
- fvec wIn = masked(fvec::One(), initialised);
+ fmask initialised = (z[i] <= z_end) & (z_start < z[i]);
+ fvec w1 = masked(w[i], initialised);
+ fvec wIn = masked(fvec::One(), initialised);
fldZ0 = z[i];
dz = (fldZ1 - fldZ0);
@@ -1088,8 +1088,8 @@ void L1Algo::L1KFTrackFitterMuch()
for (int iStep = 0; iStep < max_steps + 1; iStep++) {
- const fvec maskLastStep = (nofSteps == nofSteps1);
- z_cur = if3(maskLastStep, z_last, T1.fz + stepSize);
+ const fmask maskLastStep = (nofSteps == nofSteps1);
+ z_cur = if3(maskLastStep, z_last, T1.fz + stepSize);
// fvec v_mc = fabs(1/qp01)/sqrt(mass2+fabs(1/qp01)*fabs(1/qp01));
// T1.ExtrapolateLine1( z, &w2, v_mc);
@@ -1198,7 +1198,7 @@ void L1Algo::L1KFTrackFitterMuch()
for (--i; i >= 0; i--) {
- fvec initialised = (z[i] < z_end) & (z_start <= z[i]);
+ fmask initialised = (z[i] < z_end) & (z_start <= z[i]);
fvec w1 = masked(w[i], initialised);
fvec wIn = masked(fvec::One(), initialised);
@@ -1228,7 +1228,7 @@ void L1Algo::L1KFTrackFitterMuch()
for (int iStep = 0; iStep < max_steps + 1; iStep++) {
- const fvec maskLastStep = (nofSteps == nofSteps1);
+ const fmask maskLastStep = (nofSteps == nofSteps1);
z_cur = if3(maskLastStep, z_last, T1.fz - stepSize);
// fvec v_mc = fabs(1/qp01)/sqrt(mass2+fabs(1/qp01)*fabs(1/qp01));
diff --git a/reco/L1/L1Algo/L1TrackParFit.cxx b/reco/L1/L1Algo/L1TrackParFit.cxx
index e7f98ae5a7..0c80a50761 100644
--- a/reco/L1/L1Algo/L1TrackParFit.cxx
+++ b/reco/L1/L1Algo/L1TrackParFit.cxx
@@ -32,7 +32,7 @@ void L1TrackParFit::Filter(L1UMeasurementInfo& info, fvec u, fvec w)
F5 = info.cos_phi * C50 + info.sin_phi * C51;
#if 0 // use mask
- const fvec mask = (HCH < info.sigma2 * 16.);
+ const fmask mask = (HCH < info.sigma2 * 16.);
wi = w/( (mask & info.sigma2) +HCH );
zetawi = zeta *wi;
chi2 += mask & (zeta * zetawi);
@@ -99,7 +99,7 @@ void L1TrackParFit::FilterNoP(L1UMeasurementInfo& info, fvec u, fvec w)
F5 = info.cos_phi * C50 + info.sin_phi * C51;
#if 0 // use mask
- const fvec mask = (HCH < info.sigma2 * 16.);
+ const fmask mask = (HCH < info.sigma2 * 16.);
wi = w/( (mask & info.sigma2) +HCH );
zetawi = zeta *wi;
chi2 += mask & (zeta * zetawi);
@@ -166,10 +166,10 @@ void L1TrackParFit::Filter(fvec t0, fvec dt0, fvec w, fvec timeInfo)
F5 = C55;
#if 1 // use mask
- const fvec mask = (timeInfo > 0);
- wi = mask & w / (dt0 * dt0 + HCH);
- zetawi = zeta * wi;
- chi2 += mask & (zeta * zetawi);
+ const fmask mask = (timeInfo > 0.f);
+ wi = masked(w / (dt0 * dt0 + HCH), mask);
+ zetawi = zeta * wi;
+ chi2 += masked(zeta * zetawi, mask);
#else
wi = w / (dt0 * dt0 + HCH);
zetawi = zeta * wi;
@@ -220,7 +220,7 @@ void L1TrackParFit::ExtrapolateLine(fvec z_out, fvec* w)
cnst c_light = 29.9792458;
fvec dz = (z_out - fz);
- if (w) { dz = dz & (fvec(0.f) < *w); }
+ if (w) { dz = masked(dz, (fvec(0.f) < *w)); }
fx += dz * ftx;
fy += dz * fty;
@@ -266,7 +266,7 @@ void L1TrackParFit::ExtrapolateLine1(fvec z_out, fvec* w, fvec v)
cnst c_light = 29.9792458;
fvec dz = (z_out - fz);
- if (w) { dz = dz & (fvec(0.f) < *w); }
+ if (w) { dz = masked(dz, (fvec(0.f) < *w)); }
fx += dz * ftx;
fy += dz * fty;
@@ -806,7 +806,7 @@ void L1TrackParFit::EnergyLossCorrection(const fvec& radThick, fvec& qp0, fvec d
const fvec E2Corrected = (sqrt(E2) + direction * dE) * (sqrt(E2) + direction * dE);
fvec corr = sqrt(p2 / (E2Corrected - fMass2));
- fvec ok = (corr == corr) & (fvec::Zero() < w);
+ fmask ok = (corr == corr) & (fvec::Zero() < w);
corr = if3(ok, corr, fvec::One());
qp0 *= corr;
@@ -839,7 +839,7 @@ void L1TrackParFit::EnergyLossCorrection(float atomicA, float rho, float radLen,
const fvec E2Corrected = (sqrt(E2) + direction * dE) * (sqrt(E2) + direction * dE);
fvec corr = sqrt(p2 / (E2Corrected - fMass2));
- fvec ok = (corr == corr) & (fvec::Zero() < w);
+ fmask ok = (corr == corr) & (fvec::Zero() < w);
corr = if3(ok, corr, fvec::One());
qp0 *= corr;
diff --git a/reco/L1/L1Algo/L1TrackParFit.h b/reco/L1/L1Algo/L1TrackParFit.h
index 2712c44af5..433a2ee141 100644
--- a/reco/L1/L1Algo/L1TrackParFit.h
+++ b/reco/L1/L1Algo/L1TrackParFit.h
@@ -22,36 +22,36 @@ public:
fvec fMass2 = fMass * fMass; // mass squared
L1TrackParFit()
- : fx(0)
- , fy(0)
- , ftx(0)
- , fty(0)
- , fqp(0)
- , fz(0)
- , ft(0)
- , C00(0)
- , C10(0)
- , C11(0)
- , C20(0)
- , C21(0)
- , C22(0)
- , C30(0)
- , C31(0)
- , C32(0)
- , C33(0)
- , C40(0)
- , C41(0)
- , C42(0)
- , C43(0)
- , C44(0)
- , C50(0)
- , C51(0)
- , C52(0)
- , C53(0)
- , C54(0)
- , C55(0)
- , chi2(0)
- , NDF(0) {};
+ : fx(0.)
+ , fy(0.)
+ , ftx(0.)
+ , fty(0.)
+ , fqp(0.)
+ , fz(0.)
+ , ft(0.)
+ , C00(0.)
+ , C10(0.)
+ , C11(0.)
+ , C20(0.)
+ , C21(0.)
+ , C22(0.)
+ , C30(0.)
+ , C31(0.)
+ , C32(0.)
+ , C33(0.)
+ , C40(0.)
+ , C41(0.)
+ , C42(0.)
+ , C43(0.)
+ , C44(0.)
+ , C50(0.)
+ , C51(0.)
+ , C52(0.)
+ , C53(0.)
+ , C54(0.)
+ , C55(0.)
+ , chi2(0.)
+ , NDF(0.) {};
L1TrackParFit(double* T, double* C)
: fx(T[0])
, fy(T[1])
@@ -83,8 +83,8 @@ public:
, C53(C[18])
, C54(C[19])
, C55(C[20])
- , chi2(0)
- , NDF(0) {};
+ , chi2(0.)
+ , NDF(0.) {};
void SetOneEntry(const int i0, const L1TrackParFit& T1, const int i1);
diff --git a/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.cxx b/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.cxx
index a357adb554..a6909fb11f 100644
--- a/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.cxx
+++ b/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.cxx
@@ -305,7 +305,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
fvec S0, S1, S2, S3, S4, S5, S6, S7;
fvec X = 0, Y = 0, R = 0, R2 = 0;
- fvec validRing(fvec::MaskOne()); // mask of the valid rings
+ fmask validRing(MaskOne()); // mask of the valid rings
fvec SearchAreaSize = 0; // number of hits to fit and search ring
fvec PickUpAreaSize = 0;
@@ -375,7 +375,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
#endif
for (int isa = 0; isa < MaxSearchAreaSize; isa++) { // TODO don't work w\o this because of nan in wights
ENNSearchHitV& sHit = SearchArea[isa];
- const fvec validHit = (fvec(isa) < SearchAreaSize) & validRing;
+ const fmask validHit = (fvec(isa) < SearchAreaSize) & validRing;
sHit.lx = if3(validHit, sHit.lx, 0);
sHit.ly = if3(validHit, sHit.ly, 0);
sHit.lr2 = if3(validHit, sHit.lr2, 0);
@@ -386,7 +386,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
S0 = S1 = S2 = S3 = S4 = 0.;
for (int ih = 0; ih < MaxSearchAreaSize; ih++) {
ENNSearchHitV& sHit = SearchArea[ih];
- const fvec validHit = (fvec(ih) < SearchAreaSize) & validRing;
+ const fmask validHit = (fvec(ih) < SearchAreaSize) & validRing;
fvec& lr2 = sHit.lr2;
fvec lr = sqrt(lr2);
@@ -435,7 +435,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
Dmax = -1.;
for (THitIndex ih = 0; ih < MaxSearchAreaSize; ih++) {
- const fvec validHit = (fvec(ih) < SearchAreaSize) & validRing;
+ const fmask validHit = (fvec(ih) < SearchAreaSize) & validRing;
// ENNHit *j = &(SearchArea[ih]);
ENNSearchHitV& sHit = SearchArea[ih];
const fvec dx = sHit.lx - X;
@@ -455,7 +455,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
S4 += w * sHit.S4;
}
- } while (NotEmptyFvec(Dmax > fvec(0.)));
+ } while (NotEmptyFmask(Dmax > fvec(0.)));
#ifdef PRINT_TIMING
@@ -505,7 +505,7 @@ void CbmL1RichENNRingFinderParallel::ENNRingFinder(const int NHits, nsL1vector<E
#ifdef PRINT_TIMING
GetTimer("Ring finding: Store ring").Start(0);
#endif // PRINT_TIMING
- if (ISUNLIKELY(EmptyFvec(validRing))) continue;
+ if (ISUNLIKELY(EmptyFmask(validRing))) continue;
///////////
#if 0 // TODO 1
diff --git a/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.h b/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.h
index e6a3889ff8..9a52d27f56 100644
--- a/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.h
+++ b/reco/L1/OffLineInterface/CbmL1RichENNRingFinderParallel.h
@@ -114,13 +114,13 @@ class CbmL1RichENNRingFinderParallel : public CbmRichRingFinder {
, Cx(0)
, Cy(0)
, // coefficients for the parameter space
- on_ring(fvec::MaskOne()) {};
+ on_ring(MaskOne()) {};
// variables for local search:
fvec lx, ly, lr2; // local coordinates
fvec S0, S1, S2, S3, S4; // coefficients for calculation of E
fvec C, Cx, Cy; // coefficients for the parameter space
- fvec on_ring; // is the hit close to the current ring
+ fmask on_ring; // is the hit close to the current ring
};
diff --git a/reco/L1/ParticleFinder/CbmL1PFFitter.cxx b/reco/L1/ParticleFinder/CbmL1PFFitter.cxx
index 98873e494c..5865d7fd58 100644
--- a/reco/L1/ParticleFinder/CbmL1PFFitter.cxx
+++ b/reco/L1/ParticleFinder/CbmL1PFFitter.cxx
@@ -252,9 +252,9 @@ void CbmL1PFFitter::Fit(vector<CbmStsTrack>& Tracks, vector<int>& pidHypo)
b0.Combine(fB[i], w[i]);
fld.Set(b0, z0, b1, z1, b2, z2);
- fvec initialised = (z[i] <= z_end) & (z_start < z[i]);
- fvec w1 = masked(w[i], initialised);
- fvec wIn = mask2int(initialised);
+ fmask initialised = (z[i] <= z_end) & (z_start < z[i]);
+ fvec w1 = masked(w[i], initialised);
+ fvec wIn = mask2int(initialised);
L1Extrapolate(T, z[i], qp0, fld, &w1);
if (i == NMvdStations) { // TODO: How a hit can be also a station? (S.Zharko)
@@ -329,9 +329,9 @@ void CbmL1PFFitter::Fit(vector<CbmStsTrack>& Tracks, vector<int>& pidHypo)
b0.Combine(fB[i], w[i]);
fld.Set(b0, z0, b1, z1, b2, z2);
- fvec initialised = (z[i] < z_end) & (z_start <= z[i]);
- fvec w1 = masked(w[i], initialised);
- fvec wIn = mask2int(initialised);
+ fmask initialised = (z[i] < z_end) & (z_start <= z[i]);
+ fvec w1 = masked(w[i], initialised);
+ fvec wIn = mask2int(initialised);
L1Extrapolate(T, z[i], qp0, fld, &w1);
if (i == NMvdStations - 1) {
@@ -410,8 +410,9 @@ void CbmL1PFFitter::GetChiToVertex(vector<CbmStsTrack>& Tracks, vector<L1FieldRe
int NMvdStations = CbmL1::Instance()->fpAlgo->GetNstationsBeforePipe();
const L1Station* sta = CbmL1::Instance()->fpAlgo->GetParameters()->GetStations().begin();
fvec* zSta = new fvec[nStations];
- for (int iSta = 0; iSta < nStations; iSta++)
+ for (int iSta = 0; iSta < nStations; iSta++) {
zSta[iSta] = sta[iSta].z;
+ }
field.reserve(Tracks.size());
@@ -426,7 +427,7 @@ void CbmL1PFFitter::GetChiToVertex(vector<CbmStsTrack>& Tracks, vector<L1FieldRe
unsigned short N_vTracks = Tracks.size();
int ista;
for (unsigned short itrack = 0; itrack < N_vTracks; itrack += fvecLen) {
- if (N_vTracks - itrack < static_cast<unsigned short>(fvecLen)) nTracks_SIMD = N_vTracks - itrack;
+ if (N_vTracks - itrack < static_cast<unsigned short>(fvecLen)) { nTracks_SIMD = N_vTracks - itrack; }
fvec mass2;
for (int iVec = 0; iVec < nTracks_SIMD; iVec++) {
@@ -525,7 +526,7 @@ void CbmL1PFFitter::GetChiToVertex(vector<CbmStsTrack>& Tracks, vector<L1FieldRe
c[2] += fvec(Cv[2]);
fvec d = c[0] * c[2] - c[1] * c[1];
fvec chi = sqrt(fabs(0.5 * (dx * dx * c[0] - 2 * dx * dy * c[1] + dy * dy * c[2]) / d));
- chi = masked(chi, fabs(d) >= 1.e-20);
+ chi = masked(chi, fabs(d) >= fvec(1.e-20));
for (int iVec = 0; iVec < nTracks_SIMD; iVec++) {
chiToVtx.push_back(chi[iVec]);
diff --git a/reco/L1/vectors/P4_F32vec4.h b/reco/L1/vectors/P4_F32vec4.h
index b1ac7fce33..6f06883c2b 100644
--- a/reco/L1/vectors/P4_F32vec4.h
+++ b/reco/L1/vectors/P4_F32vec4.h
@@ -132,8 +132,6 @@ public:
static F32vec4 One() { return F32vec4(1.); }
static F32vec4 Zero() { return F32vec4(0.); }
- static F32vec4 MaskOne() { return _f32vec4_true; }
- static F32vec4 MaskZero() { return _f32vec4_false; }
/* Define all operators for consistensy */
@@ -175,12 +173,10 @@ public:
typedef F32vec4 fvec;
typedef float fscal;
+typedef fvec fmask;
const int fvecLen = 4;
//#define fvec_true _f32vec4_true
//#define fvec_false _f32vec4_false
-#define _fvecalignment __attribute__((aligned(16)))
-
-inline fvec fabs(const fvec& a) { return abs(a); }
inline fvec if3(const fvec& mask, const fvec& b, const fvec& c)
{
@@ -188,9 +184,18 @@ inline fvec if3(const fvec& mask, const fvec& b, const fvec& c)
return (b & mask) + (c & (!mask));
}
-inline fvec masked(const fvec& a, const fvec& mask) { return if3(mask, a, fvec::Zero()); }
+inline fmask MaskOne() { return _f32vec4_true; }
+inline fmask MaskZero() { return _f32vec4_false; }
+
+
+#define _fvecalignment __attribute__((aligned(16)))
+
+
+inline fvec fabs(const fvec& a) { return abs(a); }
+
+inline fvec masked(const fvec& a, const fmask& mask) { return if3(mask, a, fvec::Zero()); }
-inline fvec mask2int(const fvec& mask)
+inline fvec mask2int(const fmask& mask)
{ // mask returned
return if3(mask, fvec::One(), fvec::Zero());
}
@@ -218,9 +223,16 @@ inline bool IsNanAny(const fvec& v)
return ret;
}
-inline bool NotEmptyFvec(const F32vec4& a) { return bool(a[0]) || bool(a[1]) || bool(a[2]) || bool(a[3]); }
+inline bool EmptyFmask(const fmask& a)
+{
+ bool ret = true;
+ for (int i = 0; i < fvecLen; i++) {
+ ret = ret && (!bool(a[i]));
+ }
+ return ret;
+}
-inline bool EmptyFvec(const F32vec4& a) { return !(bool(a[0]) || bool(a[1]) || bool(a[2]) || bool(a[3])); }
+inline bool NotEmptyFmask(const fmask& a) { return !EmptyFmask(a); }
#include "std_alloc.h"
--
GitLab