sigmoid_kernels_bf16.c

Go to the documentation of this file.

/**
 * @file sigmoid_kernels_bf16.c
 * @brief Sigmoid activation kernels for BF16 tensors
 *
 * CK-ENGINE KERNEL RULES:
 * =======================
 * 1. NO malloc/free - memory via bump allocator, pointers passed in
 * 2. NO OpenMP - parallelization at orchestrator/codegen layer
 * 3. API must define: inputs, outputs, workspace, and memory layouts
 * 4. Pure computation - deterministic, no side effects
 *
 * After changes: make test && make llamacpp-parity-full
 *
 * Sigmoid: y = 1 / (1 + exp(-x))
 */
 
#include <stddef.h>
#include <stdint.h>
 
#include "bf16_utils.h"
#include "ckernel_engine.h"
 
/*
 * BF16 sigmoid forward with caller-provided scratch buffers.
 * scratch_input, scratch_output: each [n] floats
 */
void sigmoid_forward_bf16(const uint16_t *input,
                          uint16_t *output,
                          size_t n,
                          float *scratch_input,
                          float *scratch_output)
{
    if (!input || !output || n == 0) return;
    if (!scratch_input || !scratch_output) return;
 
    bf16_tensor_to_float(input, scratch_input, n);
    sigmoid_forward(scratch_input, scratch_output, n);
    float_tensor_to_bf16(scratch_output, output, n);
}
 
/*
 * BF16 sigmoid backward with caller-provided scratch buffers.
 * scratch_input, scratch_d_output, scratch_d_input: each [n] floats
 */
void sigmoid_backward_bf16(const uint16_t *input,
                           const uint16_t *d_output,
                           uint16_t *d_input,
                           size_t n,
                           float *scratch_input,
                           float *scratch_d_output,
                           float *scratch_d_input)
{
    if (!input || !d_output || !d_input || n == 0) return;
    if (!scratch_input || !scratch_d_output || !scratch_d_input) return;
 
    bf16_tensor_to_float(input, scratch_input, n);
    bf16_tensor_to_float(d_output, scratch_d_output, n);
    sigmoid_backward(scratch_input, scratch_d_output, scratch_d_input, n);
    float_tensor_to_bf16(scratch_d_input, d_input, n);
}