.cpp

#include <vector>
#include <algorithm>
#include <cstdint>

void scale(std::vector<uint8_t>& data, uint8_t in_lo, uint8_t in_hi, uint8_t out_lo, uint8_t out_hi) {
    int in_range = in_hi - in_lo;
    int out_range = out_hi - out_lo;
    // Should all fit in L1.
    uint8_t lutnorm[256];
    uint64_t lut0[256]; uint64_t lut1[256]; uint64_t lut2[256]; uint64_t lut3[256];
    uint64_t lut4[256]; uint64_t lut5[256]; uint64_t lut6[256]; uint64_t lut7[256];
    for (int i = 0; i < 256; ++i) {
        uint8_t x = std::min(std::max<uint8_t>(i, in_lo), in_hi);
        uint64_t o = out_lo + (x - in_lo) * out_range / in_range;
        lutnorm[i] = o;
        lut0[i] = o << (8*0);
        lut1[i] = o << (8*1);
        lut2[i] = o << (8*2);
        lut3[i] = o << (8*3);
        lut4[i] = o << (8*4);
        lut5[i] = o << (8*5);
        lut6[i] = o << (8*6);
        lut7[i] = o << (8*7);
    }

    size_t i = 0;
    size_t sz = data.size();
    while (i < (sz - 8)) {
        uint64_t x = *reinterpret_cast<uint64_t*>(&data[i]);
        uint64_t o = 0;
        o ^= lut0[(x >> (8*0)) & 0xff];
        o ^= lut1[(x >> (8*1)) & 0xff];
        o ^= lut2[(x >> (8*2)) & 0xff];
        o ^= lut3[(x >> (8*3)) & 0xff];
        o ^= lut4[(x >> (8*4)) & 0xff];
        o ^= lut5[(x >> (8*5)) & 0xff];
        o ^= lut6[(x >> (8*6)) & 0xff];
        o ^= lut7[(x >> (8*7)) & 0xff];
        *reinterpret_cast<uint64_t*>(&data[i]) = o;
        i += 8;
    }

    while (i < sz) {
        data[i] = lutnorm[data[i]];
        ++i;
    }
}