Skip to content
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

Jump to bottom

RecoLocalTracker/SiPixelRecHits: changes to vectorize loops in VVIObjF::VVIObjF constructor #39218

Merged
merged 6 commits into from
Aug 31, 2022
Merged

RecoLocalTracker/SiPixelRecHits: changes to vectorize loops in VVIObjF::VVIObjF constructor #39218

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
4fe0004
first attempt to vectorize loop with vdf::fast_logf in VVIObjF::VVIObjF
gartung Aug 26, 2022
551164a
Aply change to both versions of VVIObjF::VVIObjF constructor.
gartung Aug 26, 2022
61ef59b
code format
gartung Aug 26, 2022
7b8f8cf
Update array variable names to avoid confusion with member variables
gartung Aug 26, 2022
2fd301c
missed c1_
gartung Aug 26, 2022
d042492
Fix c1V[k] assignment. Use q and q2 as in original loop
gartung Aug 26, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
157 changes: 121 additions & 36 deletions RecoLocalTracker/SiPixelRecHits/src/VVIObjF.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,7 @@ VVIObjF::VVIObjF(float kappa, float beta2, int mode) : mode_(mode) {
const float xp[9] = {9.29, 2.47, .89, .36, .15, .07, .03, .02, 0.0};
const float xq[7] = {.012, .03, .08, .26, .87, 3.83, 11.0};
float h_[7];
float q, u, x, c1, c2, c3, c4, d1, h4, h5, h6, q2, x1, d, ll, ul, xf1, xf2, rv;
float q, u, h4, h5, h6, q2, d, ll, ul, rv;
int lp, lq, k, l, n;

// Make sure that the inputs are reasonable
Expand Down Expand Up @@ -107,31 +107,76 @@ VVIObjF::VVIObjF(float kappa, float beta2, int mode) : mode_(mode) {
}
q = -1.;
q2 = 2.;

float x[n];
x[0] = 0.f;
float x1[n];
x1[0] = 0.f;
float c1[n];
c1[0] = 0.f;
float c2[n];
c2[0] = 0.f;
float c3[n];
c3[0] = 0.f;
float c4[n];
c4[0] = 0.f;
float s[n];
s[0] = 0.f;
float c[n];
c[0] = 0.f;
float xf1[n];
xf1[0] = 0.f;
float xf2[n];
xf2[0] = 0.f;
for (k = 1; k < n; ++k) {
l = n - k;
x = omega_ * k;
x1 = h6 * x;
VVIObjFDetails::sincosint(x1, c2, c1);
c1 = vdt::fast_logf(x) - c1;
vdt::fast_sincosf(x1, c3, c4);
xf1 = kappa * (beta2 * c1 - c4) - x * c2;
xf2 = x * c1 + kappa * (c3 + beta2 * c2) + t0_ * x;
float s, c;
vdt::fast_sincosf(xf2, s, c);
if (mode_ == 0) {
d1 = q * d * omega_ * vdt::fast_expf(xf1);
a_[l - 1] = d1 * c;
b_[l - 1] = -d1 * s;
} else {
d1 = q * d * vdt::fast_expf(xf1) / k;
a_[l - 1] = d1 * s;
b_[l - 1] = d1 * c;
x[k] = omega_ * k;
x1[k] = h6 * x[k];
}
for (k = 1; k < n; ++k) {
VVIObjFDetails::sincosint(x1[k], c2[k], c1[k]);
}
for (k = 1; k < n; ++k) {
c1[k] = vdt::fast_logf(x[k]) - c1[k];
}
for (k = 1; k < n; ++k) {
vdt::fast_sincosf(x1[k], c3[k], c4[k]);
xf1[k] = kappa * (beta2 * c1[k] - c4[k]) - x[k] * c2[k];
xf2[k] = x[k] * c1[k] + kappa * (c3[k] + beta2 * c2[k]) + t0_ * x[k];
}
for (k = 1; k < n; ++k) {
vdt::fast_sincosf(xf2[k], s[k], c[k]);
}
float d1[n];
d1[0] = 0.f;
if (mode_ == 0) {
for (k = 1; k < n; ++k) {
d1[k] = q * d * omega_ * vdt::fast_expf(xf1[k]);
q = -q;
}
for (k = 1; k < n; ++k) {
l = n - k;
a_[l - 1] = d1[k] * c[k];
b_[l - 1] = -d1[k] * s[k];
}
} else {
for (k = 1; k < n; ++k) {
d1[k] = q * d * vdt::fast_expf(xf1[k]) / k;
q = -q;
}
for (k = 1; k < n; ++k) {
l = n - k;
a_[l - 1] = d1[k] * s[k];
}
for (k = 1; k < n; ++k) {
l = n - k;
b_[l - 1] = d1[k] * c[k];
}
for (k = 1; k < n; ++k) {
l = n - k;
a_[n - 1] += q2 * a_[l - 1];
q2 = -q2;
}
q = -q;
q2 = -q2;
}

} // VVIObjF

// ***************************************************************************************************************************************
Expand All @@ -145,7 +190,7 @@ VVIObjF::VVIObjF(float kappa) : mode_(1) {
const float xp[9] = {9.29, 2.47, .89, .36, .15, .07, .03, .02, 0.0};
const float xq[7] = {.012, .03, .08, .26, .87, 3.83, 11.0};
float h_[5];
float q, u, x, c1, c2, c3, c4, d1, h4, h5, h6, q2, x1, d, ll, ul, xf1, xf2, rv;
float q, u, h4, h5, h6, q2, d, ll, ul, rv;
int lp, lq, k, l, n;

// Make sure that the inputs are reasonable
Expand Down Expand Up @@ -196,22 +241,62 @@ VVIObjF::VVIObjF(float kappa) : mode_(1) {
a_[n - 1] = 0.f;
q = -1.;
q2 = 2.;

float x[n];
x[0] = 0.f;
float x1[n];
x1[0] = 0.f;
float c1[n];
c1[0] = 0.f;
float c2[n];
c2[0] = 0.f;
float c3[n];
c3[0] = 0.f;
float c4[n];
c4[0] = 0.f;
float s[n];
s[0] = 0.f;
float c[n];
c[0] = 0.f;
float xf1[n];
xf1[0] = 0.f;
float xf2[n];
xf2[0] = 0.f;
for (k = 1; k < n; ++k) {
x[k] = omega_ * k;
x1[k] = h6 * x[k];
}
for (k = 1; k < n; ++k) {
VVIObjFDetails::sincosint(x1[k], c2[k], c1[k]);
}
for (k = 1; k < n; ++k) {
c1[k] = vdt::fast_logf(x[k]) - c1[k];
}
for (k = 1; k < n; ++k) {
vdt::fast_sincosf(x1[k], c3[k], c4[k]);
xf1[k] = kappa * (c1[k] - c4[k]) - x[k] * c2[k];
xf2[k] = x[k] * c1[k] + kappa * (c3[k] + c2[k]) + t0_ * x[k];
}
for (k = 1; k < n; ++k) {
vdt::fast_sincosf(xf2[k], s[k], c[k]);
}
float d1[n];
d1[0] = 0.f;
for (k = 1; k < n; ++k) {
d1[k] = q * d * vdt::fast_expf(xf1[k]) / k;
q = -q;
}
for (k = 1; k < n; ++k) {
l = n - k;
a_[l - 1] = d1[k] * s[k];
}
for (k = 1; k < n; ++k) {
l = n - k;
b_[l - 1] = d1[k] * c[k];
}
for (k = 1; k < n; ++k) {
l = n - k;
x = omega_ * k;
x1 = h6 * x;
VVIObjFDetails::sincosint(x1, c2, c1);
c1 = vdt::fast_logf(x) - c1;
vdt::fast_sincosf(x1, c3, c4);
xf1 = kappa * (c1 - c4) - x * c2;
xf2 = x * c1 + kappa * (c3 + c2) + t0_ * x;
float s, c;
vdt::fast_sincosf(xf2, s, c);
d1 = q * d * vdt::fast_expf(xf1) / k;
a_[l - 1] = d1 * s;
b_[l - 1] = d1 * c;
a_[n - 1] += q2 * a_[l - 1];
q = -q;
q2 = -q2;
}

Expand Down