devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add reciprocal packet op and fast specializations for float with SSE, AVX, and AVX512.

Browse Source
This commit is contained in:
Rasmus Munk Larsen
2022-01-21 23:49:18 +00:00
parent 4b0926f99b
commit ea2c02060c
14 changed files with 95 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
unsupported/Eigen/src/SpecialFunctions/SpecialFunctionsImpl.h
View File

@@ -601,13 +601,12 @@ EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T generic_ndtri_lt_exp_neg_two(
ScalarType(6.79019408009981274425e-9)
};
const T eight = pset1<T>(ScalarType(8.0));
const T one = pset1<T>(ScalarType(1));
const T neg_two = pset1<T>(ScalarType(-2));
T x, x0, x1, z;
x = psqrt(pmul(neg_two, plog(b)));
x0 = psub(x, pdiv(plog(x), x));
z = pdiv(one, x);
z = preciprocal(x);
x1 = pmul(
z, pselect(
pcmp_lt(x, eight),

4
unsupported/test/cxx11_tensor_expr.cpp
View File

@@ -130,7 +130,7 @@ static void test_3d()
Tensor<float, 3, RowMajor> mat4(2,3,7);
mat4 = mat2 * 3.14f;
Tensor<float, 3> mat5(2,3,7);
mat5 = mat1.inverse().log();
mat5 = (mat1 + mat1.constant(1)).inverse().log();
Tensor<float, 3, RowMajor> mat6(2,3,7);
mat6 = mat2.pow(0.5f) * 3.14f;
Tensor<float, 3> mat7(2,3,7);
@@ -150,7 +150,7 @@ static void test_3d()
for (int k = 0; k < 7; ++k) {
VERIFY_IS_APPROX(mat3(i,j,k), val + val);
VERIFY_IS_APPROX(mat4(i,j,k), val * 3.14f);
VERIFY_IS_APPROX(mat5(i,j,k), logf(1.0f/val));
VERIFY_IS_APPROX(mat5(i,j,k), logf(1.0f/(val + 1)));
VERIFY_IS_APPROX(mat6(i,j,k), sqrtf(val) * 3.14f);
VERIFY_IS_APPROX(mat7(i,j,k), expf((std::max)(val, mat5(i,j,k) * 2.0f)));
VERIFY_IS_APPROX(mat8(i,j,k), expf(-val) * 3.14f);