[RFC][libc++][format] Implements compile-time caching.

[mordante]

This is a proof-of-concept patch to implement caching in basic_format_string. This is only intended for this Discourse discussion.

The real patch needs to do things cleaner.

lit -a -Doptimization=speed libcxx/test/benchmarks/format/parsing.bench.cpp BEFORE

| ---------------------------------------------------------------------------- | Benchmark Time CPU Iterations | ---------------------------------------------------------------------------- | BM_empty 6.18 ns 6.17 ns 87897649 | BM_curly_open 11.2 ns 11.2 ns 62229450 | BM_curly_close 10.7 ns 10.7 ns 65368952 | BM_pipe 12.9 ns 12.9 ns 54477724 | BM_int 47.6 ns 47.6 ns 14718165 | BM_int_formatted 97.4 ns 97.4 ns 6953144 | BM_3_ints 141 ns 141 ns 4944337 | BM_3_ints_formatted 283 ns 283 ns 2464184 | BM_prefix_5_and_empty 16.5 ns 16.5 ns 42569346 | BM_prefix_5_and_curly_open 17.8 ns 17.8 ns 39273542 | BM_prefix_5_and_curly_close 17.8 ns 17.8 ns 39140130 | BM_prefix_5_and_pipe 19.5 ns 19.5 ns 35803584 | BM_prefix_5_and_int 51.1 ns 51.0 ns 13436274 | BM_prefix_5_and_int_formatted 102 ns 102 ns 6793303 | BM_prefix_5_and_3_ints 151 ns 151 ns 4573472 | BM_prefix_5_and_3_ints_formatted 291 ns 291 ns 2351850 | BM_prefix_10_and_empty 24.9 ns 24.9 ns 28109036 | BM_prefix_10_and_curly_open 27.2 ns 27.2 ns 25713490 | BM_prefix_10_and_curly_close 30.2 ns 30.2 ns 23176332 | BM_prefix_10_and_pipe 32.7 ns 32.6 ns 21420861 | BM_prefix_10_and_int 61.6 ns 61.6 ns 11034101 | BM_prefix_10_and_int_formatted 120 ns 120 ns 5610531 | BM_prefix_10_and_3_ints 156 ns 156 ns 4470738 | BM_prefix_10_and_3_ints_formatted 312 ns 311 ns 2217207 | BM_prefix_20_and_empty 52.7 ns 52.7 ns 13138425 | BM_prefix_20_and_curly_open 58.8 ns 58.8 ns 11863712 | BM_prefix_20_and_curly_close 56.6 ns 56.6 ns 12407764 | BM_prefix_20_and_pipe 56.5 ns 56.4 ns 12411003 | BM_prefix_20_and_int 86.9 ns 86.8 ns 8089463 | BM_prefix_20_and_int_formatted 153 ns 153 ns 4582443 | BM_prefix_20_and_3_ints 204 ns 204 ns 3354539 | BM_prefix_20_and_3_ints_formatted 349 ns 349 ns 2002457 | BM_prefix_40_and_empty 116 ns 116 ns 6018618 | BM_prefix_40_and_curly_open 122 ns 122 ns 5655627 | BM_prefix_40_and_curly_close 121 ns 121 ns 5722255 | BM_prefix_40_and_pipe 120 ns 120 ns 5823239 | BM_prefix_40_and_int 158 ns 158 ns 4476432 | BM_prefix_40_and_int_formatted 207 ns 207 ns 3306182 | BM_prefix_40_and_3_ints 255 ns 255 ns 2759277 | BM_prefix_40_and_3_ints_formatted 371 ns 370 ns 1884154 

AFTER

| ---------------------------------------------------------------------------- | Benchmark Time CPU Iterations | ---------------------------------------------------------------------------- | BM_empty 21.6 ns 21.6 ns 32391462 | BM_curly_open 22.9 ns 22.9 ns 30588441 | BM_curly_close 22.9 ns 22.9 ns 30592394 | BM_pipe 26.0 ns 26.0 ns 26953859 | BM_int 60.5 ns 60.5 ns 11577625 | BM_int_formatted 81.3 ns 81.2 ns 8594596 | BM_3_ints 156 ns 156 ns 4409940 | BM_3_ints_formatted 220 ns 220 ns 3187392 | BM_prefix_5_and_empty 26.0 ns 26.0 ns 26936925 | BM_prefix_5_and_curly_open 26.0 ns 26.0 ns 26965917 | BM_prefix_5_and_curly_close 26.0 ns 26.0 ns 26961356 | BM_prefix_5_and_pipe 26.0 ns 26.0 ns 26959641 | BM_prefix_5_and_int 64.1 ns 64.1 ns 10920953 | BM_prefix_5_and_int_formatted 84.4 ns 84.4 ns 8037453 | BM_prefix_5_and_3_ints 165 ns 165 ns 4217532 | BM_prefix_5_and_3_ints_formatted 247 ns 247 ns 2823553 | BM_prefix_10_and_empty 26.0 ns 26.0 ns 26949246 | BM_prefix_10_and_curly_open 26.0 ns 26.0 ns 26955524 | BM_prefix_10_and_curly_close 26.0 ns 26.0 ns 26956279 | BM_prefix_10_and_pipe 26.0 ns 26.0 ns 26947963 | BM_prefix_10_and_int 59.8 ns 59.8 ns 11674226 | BM_prefix_10_and_int_formatted 87.1 ns 87.0 ns 7814714 | BM_prefix_10_and_3_ints 165 ns 165 ns 4149378 | BM_prefix_10_and_3_ints_formatted 245 ns 245 ns 2884163 | BM_prefix_20_and_empty 29.3 ns 29.3 ns 23873593 | BM_prefix_20_and_curly_open 28.5 ns 28.4 ns 24625298 | BM_prefix_20_and_curly_close 27.5 ns 27.5 ns 25438870 | BM_prefix_20_and_pipe 29.4 ns 29.3 ns 23863557 | BM_prefix_20_and_int 57.6 ns 57.6 ns 12247413 | BM_prefix_20_and_int_formatted 104 ns 104 ns 6690343 | BM_prefix_20_and_3_ints 183 ns 183 ns 3828112 | BM_prefix_20_and_3_ints_formatted 244 ns 244 ns 2833578 | BM_prefix_40_and_empty 37.5 ns 37.5 ns 18650804 | BM_prefix_40_and_curly_open 36.8 ns 36.8 ns 19094990 | BM_prefix_40_and_curly_close 37.1 ns 37.0 ns 18904266 | BM_prefix_40_and_pipe 36.5 ns 36.5 ns 19147265 | BM_prefix_40_and_int 81.1 ns 81.0 ns 8611919 | BM_prefix_40_and_int_formatted 102 ns 102 ns 6854642 | BM_prefix_40_and_3_ints 181 ns 181 ns 3828217 | BM_prefix_40_and_3_ints_formatted 243 ns 243 ns 2814977 

[github-actions]

⚠️ C/C++ code formatter, clang-format found issues in your code. ⚠️

You can test this locally with the following command:

git-clang-format --diff dd97324f119af515944b7ec79fe627cfa70d4a2a e02e5bac0bbdd2e62d8cd9d07c2f0e7810b393b9 --extensions cpp,h -- libcxx/test/benchmarks/format/parsing.bench.cpp libcxx/test/std/utilities/format/format.functions/cache.pass.cpp libcxx/include/__format/format_functions.h

View the diff from clang-format here.

diff --git a/libcxx/include/__format/format_functions.h b/libcxx/include/__format/format_functions.h index 98e2d97718..d092561743 100644 --- a/libcxx/include/__format/format_functions.h +++ b/libcxx/include/__format/format_functions.h @@ -725,8 +725,7 @@ private: break; switch (__element.__type) { - case __element_type::__formatter_std: - { + case __element_type::__formatter_std: { std::__visit_format_arg( [&](auto __arg) { if constexpr (same_as<decltype(__arg), monostate>) diff --git a/libcxx/test/std/utilities/format/format.functions/cache.pass.cpp b/libcxx/test/std/utilities/format/format.functions/cache.pass.cpp index a1d8ced4ae..bd6e07a726 100644 --- a/libcxx/test/std/utilities/format/format.functions/cache.pass.cpp +++ b/libcxx/test/std/utilities/format/format.functions/cache.pass.cpp @@ -8,7 +8,6 @@ // UNSUPPORTED: c++03, c++11, c++14, c++17 // UNSUPPORTED: GCC-ALWAYS_INLINE-FIXME - // TODO FMT This test should not require std::to_chars(floating-point) // XFAIL: availability-fp_to_chars-missing @@ -16,7 +15,6 @@ // <format> - // This is a test for the new caching mechanism. #include <format> @@ -84,8 +82,8 @@ static void test() { check(SV("FULL:0x0042"), SV("{:#06x}"), 0x42); check(SV("FULL:0x0042=answer"), SV("{:#06x}={}"), 0x42, SV("answer")); -// check(SV("FULL:hello world"), SV("{} world"), String<char>{"hello"}); -// check(SV("FULL:hello world"), SV("{0:} world"), String<char>{"hello"}); + // check(SV("FULL:hello world"), SV("{} world"), String<char>{"hello"}); + // check(SV("FULL:hello world"), SV("{0:} world"), String<char>{"hello"}); // TODO TEST WITH ARG EATER }