in gemmology.h [1116:1126]
void Engine<Arch>::QuantizeU(const float *input, uint8_t *output,
float quant_mult, size_t size) {
using batch8 = xsimd::batch<int8_t, Arch>;
xsimd::batch<float, Arch> q(quant_mult);
const float *end = input + size;
for (; input != end; input += batch8::size, output += batch8::size) {
auto tile = QuantizeTile8::ConsecutiveU(q, input);
tile.store_aligned(output);
}
}