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

Arith: Implement efficient SIMD trigonometic functions #308

Merged
merged 8 commits into from
Oct 7, 2024
Merged

Arith: Implement efficient SIMD trigonometic functions #308

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
f26576f
Arith: Implement efficient SIMD trigonometic functions
eschnett Sep 12, 2024
ed4293d
Arith: Correct SIMD tests
eschnett Sep 12, 2024
d412c15
Arith: Correct SIMD tests
eschnett Sep 12, 2024
1cb3459
Arith: Remove `constexpr` that some compilers don't like
eschnett Sep 12, 2024
351b248
Arith: Remove `constexpr` that some compilers don't like
eschnett Sep 12, 2024
bdd9e46
TestDerivs: Correct C++ calls to trigonometic functions
eschnett Sep 12, 2024
4dfd658
CI: Disable SIMD the new way
eschnett Sep 12, 2024
8ebc67b
Merge branch 'main' into eschnett/simd-cbrt
eschnett Oct 7, 2024
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
6 changes: 6 additions & 0 deletions Arith/src/defs.hxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -206,6 +206,12 @@ constexpr ARITH_INLINE ARITH_DEVICE ARITH_HOST T flipsign(const T &x,
return copysign(T(1), y) * x;
}

// Return 1/x
template <typename T>
constexpr ARITH_INLINE ARITH_DEVICE ARITH_HOST T inv(const T &x) {
return T(1) / x;
}

// A max function that returns nan when any argument is nan
template <typename T>
constexpr ARITH_INLINE ARITH_DEVICE ARITH_HOST T max1(const T &x, const T &y) {
Expand Down
153 changes: 146 additions & 7 deletions Arith/src/simd.cxx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,7 @@
#include <cctk.h>

#include <cassert>
#include <cmath>

namespace Arith {

Expand All @@ -25,17 +26,34 @@ std::size_t get_memop_count() {
return count;
}

template <typename T> constexpr bool isequal(const simd<T> x, const T a) {
return all(x == a);
}
template <typename T> constexpr bool isequal(const simdl<T> x, const bool a) {
return all(x == a);
}
template <typename T> constexpr bool isapprox(const simd<T> x, const T a) {
return all(fabs(x - a) < 1.0e-14);
}

void check(const bool isgood) {
if (isgood)
return;
CCTK_ERROR("Test failure");
}

void TestSIMD() {
// nvcc V11.1.74 doesn't accept this as "constexpr" values
#ifndef __CUDACC__
typedef simd<CCTK_REAL> realv;
using real = CCTK_REAL;
using realv = simd<real>;

realv x;
realv y = 0;
realv z = zero<realv>();

assert(all(y == 0));
assert(all(y == z));
check(all(y == 0));
check(all(y == z));

realv a = 2;
realv b = 3;
Expand All @@ -45,10 +63,131 @@ void TestSIMD() {
realv r1 = mulsub(a, b, c);
realv r2 = negmuladd(a, b, c);
realv r3 = negmulsub(a, b, c);
assert(all(r0 == muladd(2, 3, 4)));
assert(all(r1 == mulsub(2, 3, 4)));
assert(all(r2 == negmuladd(2, 3, 4)));
assert(all(r3 == negmulsub(2, 3, 4)));
check(all(r0 == muladd(2, 3, 4)));
check(all(r1 == mulsub(2, 3, 4)));
check(all(r2 == negmuladd(2, 3, 4)));
check(all(r3 == negmulsub(2, 3, 4)));

real s = 2;
real t = 2;
real u = 4;

check(isequal(+a, +s));
check(isequal(-a, -s));

check(isequal(a + b, s + t));
check(isequal(a - b, s - t));
check(isequal(a * b, s * t));
check(isequal(a / b, s / t));
check(isequal(s + b, s + t));
check(isequal(s - b, s - t));
check(isequal(s * b, s * t));
check(isequal(s / b, s / t));
check(isequal(a + t, s + t));
check(isequal(a - t, s - t));
check(isequal(a * t, s * t));
check(isequal(a / t, s / t));

check(isequal(a == b, s == t));
check(isequal(a != b, s != t));
check(isequal(a < b, s < t));
check(isequal(a > b, s > t));
check(isequal(a <= b, s <= t));
check(isequal(a >= b, s >= t));
check(isequal(s == b, s == t));
check(isequal(s != b, s != t));
check(isequal(s < b, s < t));
check(isequal(s > b, s > t));
check(isequal(s <= b, s <= t));
check(isequal(s >= b, s >= t));
check(isequal(a == t, s == t));
check(isequal(a != t, s != t));
check(isequal(a < t, s < t));
check(isequal(a > t, s > t));
check(isequal(a <= t, s <= t));
check(isequal(a >= t, s >= t));

check(isapprox(abs(a), abs(s)));
check(isapprox(acos(a), acos(s)));
check(isapprox(acosh(a), acosh(s)));
check(allisfinite(a) == allisfinite(s));
check(anyisnan(a) == anyisnan(s));
check(isapprox(asin(a), asin(s)));
check(isapprox(asinh(a), asinh(s)));
check(isapprox(atan(a), atan(s)));
check(isapprox(atanh(a), atanh(s)));
check(isapprox(cbrt(a), cbrt(s)));
// check(isapprox(cis(a), cis(s)));
// check(isapprox(cispi(a), cispi(s)));
check(isapprox(copysign(a, b), copysign(s, t)));
check(isapprox(cos(a), cos(s)));
check(isapprox(cosh(a), cosh(s)));
check(isapprox(cospi(a), cos(CCTK_REAL(M_PI) * s)));
check(isapprox(exp(a), exp(s)));
check(isapprox(exp10(a), pow(CCTK_REAL(10), s)));
check(isapprox(exp2(a), exp2(s)));
check(isapprox(fabs(a), fabs(s)));
check(isapprox(flipsign(a, b), flipsign(s, t)));
check(isapprox(fmax(a, b), fmax(s, t)));
check(isapprox(fmax(a, t), fmax(s, t)));
check(isapprox(fmax(s, b), fmax(s, t)));
check(isapprox(fmin(a, b), fmin(s, t)));
check(isapprox(fmin(a, t), fmin(s, t)));
check(isapprox(fmin(s, b), fmin(s, t)));
check(isapprox(hypot(a, b), hypot(s, t)));
check(isapprox(inv(a), inv(s)));
check(isequal(isfinite(a), isfinite(s)));
check(isequal(isinf(a), isinf(s)));
check(isequal(isnan(a), isnan(s)));
check(isapprox(log(a), log(s)));
check(isapprox(log10(a), log10(s)));
check(isapprox(log2(a), log2(s)));
check(isapprox(max(a, b), max(s, t)));
check(isapprox(max(a, t), max(s, t)));
check(isapprox(max(s, b), max(s, t)));
check(isapprox(min(a, b), min(s, t)));
check(isapprox(min(a, t), min(s, t)));
check(isapprox(min(s, b), min(s, t)));
check(isapprox(muladd(a, b, c), muladd(s, t, u)));
check(isapprox(muladd(a, b, u), muladd(s, t, u)));
check(isapprox(muladd(a, t, c), muladd(s, t, u)));
check(isapprox(muladd(a, t, u), muladd(s, t, u)));
check(isapprox(muladd(s, b, c), muladd(s, t, u)));
check(isapprox(muladd(s, b, u), muladd(s, t, u)));
check(isapprox(muladd(s, t, c), muladd(s, t, u)));
check(isapprox(mulsub(a, b, c), mulsub(s, t, u)));
check(isapprox(mulsub(a, b, u), mulsub(s, t, u)));
check(isapprox(mulsub(a, t, c), mulsub(s, t, u)));
check(isapprox(mulsub(a, t, u), mulsub(s, t, u)));
check(isapprox(mulsub(s, b, c), mulsub(s, t, u)));
check(isapprox(mulsub(s, b, u), mulsub(s, t, u)));
check(isapprox(mulsub(s, t, c), mulsub(s, t, u)));
check(isapprox(negmuladd(a, b, c), negmuladd(s, t, u)));
check(isapprox(negmuladd(a, b, u), negmuladd(s, t, u)));
check(isapprox(negmuladd(a, t, c), negmuladd(s, t, u)));
check(isapprox(negmuladd(a, t, u), negmuladd(s, t, u)));
check(isapprox(negmuladd(s, b, c), negmuladd(s, t, u)));
check(isapprox(negmuladd(s, b, u), negmuladd(s, t, u)));
check(isapprox(negmuladd(s, t, c), negmuladd(s, t, u)));
check(isapprox(negmulsub(a, b, c), negmulsub(s, t, u)));
check(isapprox(negmulsub(a, b, u), negmulsub(s, t, u)));
check(isapprox(negmulsub(a, t, c), negmulsub(s, t, u)));
check(isapprox(negmulsub(a, t, u), negmulsub(s, t, u)));
check(isapprox(negmulsub(s, b, c), negmulsub(s, t, u)));
check(isapprox(negmulsub(s, b, u), negmulsub(s, t, u)));
check(isapprox(negmulsub(s, t, c), negmulsub(s, t, u)));
check(isapprox(pow(a, b), pow(s, t)));
check(isapprox(pow(a, t), pow(s, t)));
check(isapprox(pow(s, b), pow(s, t)));
check(isequal(signbit(a), signbit(s)));
check(isapprox(sin(a), sin(s)));
check(isapprox(sinh(a), sinh(s)));
check(isapprox(sinpi(a), sin(CCTK_REAL(M_PI) * s)));
check(isapprox(sqrt(a), sqrt(s)));
check(isapprox(tan(a), tan(s)));
check(isapprox(tanh(a), tanh(s)));
// check(isapprox(tanpi(a), tanpi(s)));

#endif
}

Expand Down
Loading