We define the number of bytes to be the number of *input* bytes. We define a 'char' to be a code point (between 1 and 4 bytes). =========================== Using ICU version 72.1 testcases: 4 input detected as UTF16 little-endian current system detected as haswell =========================== convert_utf32_to_latin1+haswell, input size: 328672, iterations: 30000, dataset: ./unicode_lipsum/wikipedia_mars/esperanto.utflatin32.txt 0.251 ins/byte, 0.062 cycle/byte, 77.135 GB/s (42.3 %), 4.798 GHz, 4.033 ins/cycle 1.004 ins/char, 0.249 cycle/char, 19.284 Gc/s (42.3 %) 4.00 byte/char 4261.0 ns WARNING: Measurements are noisy, try increasing iteration count (-I). input detected as UTF16 little-endian current system detected as haswell =========================== convert_utf32_to_latin1+haswell, input size: 1729220, iterations: 30000, dataset: ./unicode_lipsum/wikipedia_mars/french.utflatin32.txt 0.250 ins/byte, 0.058 cycle/byte, 75.259 GB/s (40.1 %), 4.399 GHz, 4.279 ins/cycle 1.001 ins/char, 0.234 cycle/char, 18.815 Gc/s (40.1 %) 4.00 byte/char 22977.0 ns WARNING: Measurements are noisy, try increasing iteration count (-I). input detected as UTF16 little-endian current system detected as haswell =========================== convert_utf32_to_latin1+haswell, input size: 797324, iterations: 30000, dataset: ./unicode_lipsum/wikipedia_mars/german.utflatin32.txt 0.250 ins/byte, 0.060 cycle/byte, 76.110 GB/s (37.5 %), 4.585 GHz, 4.155 ins/cycle 1.001 ins/char, 0.241 cycle/char, 19.027 Gc/s (37.5 %) 4.00 byte/char 10476.0 ns WARNING: Measurements are noisy, try increasing iteration count (-I). input detected as UTF16 little-endian current system detected as haswell =========================== convert_utf32_to_latin1+haswell, input size: 1086972, iterations: 30000, dataset: ./unicode_lipsum/wikipedia_mars/portuguese.utflatin32.txt 0.250 ins/byte, 0.060 cycle/byte, 84.124 GB/s (56.8 %), 5.040 GHz, 4.177 ins/cycle 1.001 ins/char, 0.240 cycle/char, 21.031 Gc/s (56.8 %) 4.00 byte/char 12921.0 ns WARNING: Measurements are noisy, try increasing iteration count (-I).