max_pool1d_backward_f32

void max_pool1d_backward_f32(const float * g, const int * argmax, float * dx, int B, int C, int L, int OL)

Single-precision 1-D max-pool backward — scatters each upstream gradient onto the input slot that won its pool window.

The caller must zero dx first because windows may overlap (when stride_l < KL) and the kernel uses += to accumulate contributions.

gconst float*

Upstream gradient of shape (B, C, OL).

argmaxconst int32_t*

Argmax indices from the matching forward call; shape (B, C, OL).

dxfloat*

Output gradient w.r.t. the original input; shape (B, C, L); accumulated — caller must zero before invocation.

B, C, L, OLint

Layout extents.

void max_pool1d_backward_f32(const float * g, const int * argmax, float * dx, int B, int C, int L, int OL)