potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-21 19:33:03 +04:00
Code Packages Releases Wiki Activity

Merge pull request #28999 from Teddy-Yangjiale:rvv-sigmoid-opt

Browse Source 
dnn: vectorize SigmoidFunctor using universal intrinsics
This commit is contained in:
Alexander Smorkalov
2026-05-12 11:29:12 +03:00
committed by GitHub
parent b7ae67e8ec 4b1c861ab7
commit 594cd4204a
1 changed files with 49 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/dnn/src/layers/elementwise_layers.cpp
+49
View File
@@ -1287,6 +1287,16 @@ struct SigmoidFunctor : public BaseDefaultFunctor<SigmoidFunctor>
{
typedef SigmoidLayer Layer;
int vlanes;
explicit SigmoidFunctor() {
#if (CV_SIMD || CV_SIMD_SCALABLE)
vlanes = VTraits<v_float32>::vlanes();
#else
vlanes = 1;
#endif
}
bool supportBackend(int backendId, int)
{
#ifdef HAVE_INF_ENGINE
@@ -1311,6 +1321,45 @@ struct SigmoidFunctor : public BaseDefaultFunctor<SigmoidFunctor>
return y;
}
void apply(const float* srcptr, float* dstptr, int stripeStart, int len, size_t planeSize, int cn0, int cn1) const {
CV_UNUSED(stripeStart);
for (int cn = cn0; cn < cn1; cn++, srcptr += planeSize, dstptr += planeSize) {
int i = 0;
#if (CV_SIMD || CV_SIMD_SCALABLE)
// Using identity: 1 / (1 + exp(-x)) === 1 - (1 / (1 + exp(x)))
// Clamping x to [-80.f, 88.f] prevents both v_exp overflow (inf) and subnormal underflow.
v_float32 one = vx_setall_f32(1.0f);
v_float32 min_val = vx_setall_f32(-80.0f), max_val = vx_setall_f32(88.0f);
int step2 = vlanes * 2;
// 2-way unrolled loop optimized for minimal register dependency
for (; i <= len - step2; i += step2) {
v_float32 x0 = vx_load(srcptr + i);
v_float32 x1 = vx_load(srcptr + i + vlanes);
x0 = v_min(v_max(x0, min_val), max_val);
x1 = v_min(v_max(x1, min_val), max_val);
// y = 1.0f - (1.0f / (1.0f + exp(x)))
v_float32 dst0 = v_sub(one, v_div(one, v_add(one, v_exp(x0))));
v_float32 dst1 = v_sub(one, v_div(one, v_add(one, v_exp(x1))));
vx_store(dstptr + i, dst0);
vx_store(dstptr + i + vlanes, dst1);
}
for (; i <= len - vlanes; i += vlanes) {
v_float32 x = vx_load(srcptr + i);
x = v_min(v_max(x, min_val), max_val);
vx_store(dstptr + i, v_sub(one, v_div(one, v_add(one, v_exp(x)))));
}
#endif
for (; i < len; i++) {
dstptr[i] = calculate(srcptr[i]);
}
}
}
#ifdef HAVE_CUDA
Ptr<BackendNode> initCUDA(int target, csl::Stream stream)
{