Investigate replacing double_conversion library with fast_float #2772

Constellation · 2022-07-27T06:44:58Z

`178e072`

Investigate replacing double_conversion library with fast_float
https://bugs.webkit.org/show_bug.cgi?id=221178

Reviewed by Sam Weinig.

Parsing double numbers becomes 2x faster.

                                  ToT                     Patched

    json-parse-double      230.0022+-0.0838     ^    111.1297+-0.0983        ^ definitely 2.0697x faster

Apple OSS approval for fast_float is OSS-4890.

* Source/WTF/WTF.xcodeproj/project.pbxproj:
* Source/WTF/wtf/CMakeLists.txt:
* Source/WTF/wtf/FastFloat.cpp: Added.
(WTF::parseDouble):
* Source/WTF/wtf/FastFloat.h: Added.
* Source/WTF/wtf/dtoa.h:
* Source/WTF/wtf/fast_float/LICENSE: Added.
* Source/WTF/wtf/fast_float/ascii_number.h: Added.
(fast_float::byteswap):
(fast_float::read_u64):
(fast_float::write_u64):
(fast_float::parse_eight_digits_unrolled):
* Source/WTF/wtf/fast_float/bigint.h: Added.
(fast_float::stackvec::stackvec):
(fast_float::bigint::bigint):
* Source/WTF/wtf/fast_float/decimal_to_binary.h: Added.
(fast_float::compute_product_approximation):
* Source/WTF/wtf/fast_float/digit_comparison.h: Added.
* Source/WTF/wtf/fast_float/fast_float.h: Added.
(fast_float::parse_options::parse_options):
* Source/WTF/wtf/fast_float/fast_table.h: Added.
* Source/WTF/wtf/fast_float/float_common.h: Added.
(fast_float::fastfloat_strncasecmp):
(fast_float::span::span):
(fast_float::value128::value128):
(fast_float::leading_zeroes):
(fast_float::emulu):
(fast_float::_umul128):
(fast_float::full_multiplication):
(fast_float::adjusted_mantissa::operator== const):
(fast_float::adjusted_mantissa::operator!= const):
(fast_float::binary_format<double>::mantissa_explicit_bits):
(fast_float::binary_format<float>::mantissa_explicit_bits):
(fast_float::binary_format<double>::max_exponent_round_to_even):
(fast_float::binary_format<float>::max_exponent_round_to_even):
(fast_float::binary_format<double>::min_exponent_round_to_even):
(fast_float::binary_format<float>::min_exponent_round_to_even):
(fast_float::binary_format<double>::minimum_exponent):
(fast_float::binary_format<float>::minimum_exponent):
(fast_float::binary_format<double>::infinite_power):
(fast_float::binary_format<float>::infinite_power):
(fast_float::binary_format<double>::sign_index):
(fast_float::binary_format<float>::sign_index):
(fast_float::binary_format<double>::min_exponent_fast_path):
(fast_float::binary_format<float>::min_exponent_fast_path):
(fast_float::binary_format<double>::max_exponent_fast_path):
(fast_float::binary_format<float>::max_exponent_fast_path):
(fast_float::binary_format<double>::max_mantissa_fast_path):
(fast_float::binary_format<float>::max_mantissa_fast_path):
(fast_float::binary_format<double>::exact_power_of_ten):
(fast_float::binary_format<float>::exact_power_of_ten):
(fast_float::binary_format<double>::largest_power_of_ten):
(fast_float::binary_format<float>::largest_power_of_ten):
(fast_float::binary_format<double>::smallest_power_of_ten):
(fast_float::binary_format<float>::smallest_power_of_ten):
(fast_float::binary_format<double>::max_digits):
(fast_float::binary_format<float>::max_digits):
(fast_float::binary_format<float>::exponent_mask):
(fast_float::binary_format<double>::exponent_mask):
(fast_float::binary_format<float>::mantissa_mask):
(fast_float::binary_format<double>::mantissa_mask):
(fast_float::binary_format<float>::hidden_bit_mask):
(fast_float::binary_format<double>::hidden_bit_mask):
(fast_float::to_float):
* Source/WTF/wtf/fast_float/parse_number.h: Added.
* Source/WTF/wtf/fast_float/simple_decimal_conversion.h: Added.
(fast_float::detail::trim):
(fast_float::detail::number_of_digits_decimal_left_shift):
(fast_float::detail::round):
(fast_float::detail::decimal_left_shift):
(fast_float::detail::decimal_right_shift):
(fast_float::compute_float):
(fast_float::parse_long_mantissa):

Canonical link: https://commits.webkit.org/257675@main

900ddd2

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 mac-wk1	❌ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	✅ 🧪 mac-wk2	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	✅ 🧪 mac-AS-debug-wk2	✅ 🧪 jsc-armv7-tests
✅ 🛠 🧪 merge	✅ 🛠 watch	✅ 🧪 mac-wk2-stress	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim		✅ 🧪 jsc-mips-tests

weinig

Thanks for following up on this!

weinig · 2022-07-31T20:28:24Z

Source/WTF/wtf/FastFloat.h

+
+namespace WTF {
+
+WTF_EXPORT_PRIVATE double parseDouble(const LChar* string, size_t length, size_t& parsedLength);


I I think the way I would structured things would be to rename this file to ParseDouble.h and add all the ParseDouble header bits from dtoa.h here.

I would then name the file you have called FastFloat.cpp to ParseDoubleFastFloat.cpp.

And then, probably in a different change, rename dtoa.h to NumberToString.h and dtoa.cpp to NumberToStringDoubleConversion.cpp

weinig · 2022-07-31T20:31:36Z

Source/WTF/wtf/fast_float/ascii_number.h

+
+#include <wtf/fast_float/float_common.h>
+
+namespace fast_float {


For whatever reason, we wrapped all the double_conversion namespaces in an our namespace WTF. Not sure it makes a difference given this is all only included in one cpp file, but maybe for mangling reasons it would be a good idea?

weinig · 2022-08-01T01:17:52Z

Another good follow up to this would be removing the now unneeded parsing code from our copy of double_conversion.

webkit-early-warning-system · 2022-12-02T04:40:01Z

EWS run on previous version of this PR (hash 498737e)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	~~🛠 ios~~	~~🛠 mac~~	~~🛠 wpe~~	⏳ 🛠 🧪 win
✅ 🧪 bindings	~~🛠 ios-sim~~	~~🛠 mac-debug~~	⏳ 🛠 gtk	⏳ 🛠 wincairo
✅ 🧪 webkitperl	~~🧪 ios-wk2~~	~~🛠 mac-AS-debug~~	⏳ 🧪 gtk-wk2
	~~🧪 api-ios~~	~~🧪 api-mac~~	⏳ 🧪 api-gtk
⏳ 🛠 🧪 jsc	~~🛠 tv~~	~~🧪 mac-wk1~~	✅ 🛠 jsc-armv7
	⏳ 🛠 tv-sim	~~🧪 mac-wk2~~	~~🧪 jsc-armv7-tests~~
	~~🛠 watch~~	~~🧪 mac-AS-debug-wk2~~	✅ 🛠 jsc-mips
	~~🛠 watch-sim~~	~~🧪 mac-wk2-stress~~	~~🧪 jsc-mips-tests~~

webkit-early-warning-system · 2022-12-02T04:40:02Z

EWS run on previous version of this PR (hash 4d7604d)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
	✅ 🛠 ios-sim	✅ 🛠 mac-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🛠 mac-AS-debug	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 api-mac	✅ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	✅ 🧪 mac-wk1	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2	✅ 🧪 jsc-armv7-tests
	✅ 🛠 watch	✅ 🧪 mac-AS-debug-wk2	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim	✅ 🧪 mac-wk2-stress	✅ 🧪 jsc-mips-tests

webkit-early-warning-system · 2022-12-02T04:47:17Z

EWS run on previous version of this PR (hash b5e041d)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	❌ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	❌ 🧪 mac-wk1	❌ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	❌ 🧪 mac-wk2	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	❌ 🧪 mac-AS-debug-wk2	✅ 🧪 jsc-armv7-tests
	✅ 🛠 watch	✅ 🧪 mac-wk2-stress	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim		✅ 🧪 jsc-mips-tests

Constellation · 2022-12-02T18:49:42Z

ios-wk2 failures are flakiness ones.

webkit-early-warning-system · 2022-12-07T06:01:46Z

EWS run on current version of this PR (hash 900ddd2)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
❌ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 🧪 win
	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🛠 gtk	✅ 🛠 wincairo
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 gtk-wk2
	✅ 🧪 api-ios	✅ 🧪 mac-wk1	❌ 🧪 api-gtk
✅ 🛠 🧪 jsc	✅ 🛠 tv	✅ 🧪 mac-wk2	✅ 🛠 jsc-armv7
	✅ 🛠 tv-sim	✅ 🧪 mac-AS-debug-wk2	✅ 🧪 jsc-armv7-tests
✅ 🛠 🧪 merge	✅ 🛠 watch	✅ 🧪 mac-wk2-stress	✅ 🛠 jsc-mips
	✅ 🛠 watch-sim		✅ 🧪 jsc-mips-tests

https://bugs.webkit.org/show_bug.cgi?id=221178 Reviewed by Sam Weinig. Parsing double numbers becomes 2x faster. ToT Patched json-parse-double 230.0022+-0.0838 ^ 111.1297+-0.0983 ^ definitely 2.0697x faster Apple OSS approval for fast_float is OSS-4890. * Source/WTF/WTF.xcodeproj/project.pbxproj: * Source/WTF/wtf/CMakeLists.txt: * Source/WTF/wtf/FastFloat.cpp: Added. (WTF::parseDouble): * Source/WTF/wtf/FastFloat.h: Added. * Source/WTF/wtf/dtoa.h: * Source/WTF/wtf/fast_float/LICENSE: Added. * Source/WTF/wtf/fast_float/ascii_number.h: Added. (fast_float::byteswap): (fast_float::read_u64): (fast_float::write_u64): (fast_float::parse_eight_digits_unrolled): * Source/WTF/wtf/fast_float/bigint.h: Added. (fast_float::stackvec::stackvec): (fast_float::bigint::bigint): * Source/WTF/wtf/fast_float/decimal_to_binary.h: Added. (fast_float::compute_product_approximation): * Source/WTF/wtf/fast_float/digit_comparison.h: Added. * Source/WTF/wtf/fast_float/fast_float.h: Added. (fast_float::parse_options::parse_options): * Source/WTF/wtf/fast_float/fast_table.h: Added. * Source/WTF/wtf/fast_float/float_common.h: Added. (fast_float::fastfloat_strncasecmp): (fast_float::span::span): (fast_float::value128::value128): (fast_float::leading_zeroes): (fast_float::emulu): (fast_float::_umul128): (fast_float::full_multiplication): (fast_float::adjusted_mantissa::operator== const): (fast_float::adjusted_mantissa::operator!= const): (fast_float::binary_format<double>::mantissa_explicit_bits): (fast_float::binary_format<float>::mantissa_explicit_bits): (fast_float::binary_format<double>::max_exponent_round_to_even): (fast_float::binary_format<float>::max_exponent_round_to_even): (fast_float::binary_format<double>::min_exponent_round_to_even): (fast_float::binary_format<float>::min_exponent_round_to_even): (fast_float::binary_format<double>::minimum_exponent): (fast_float::binary_format<float>::minimum_exponent): (fast_float::binary_format<double>::infinite_power): (fast_float::binary_format<float>::infinite_power): (fast_float::binary_format<double>::sign_index): (fast_float::binary_format<float>::sign_index): (fast_float::binary_format<double>::min_exponent_fast_path): (fast_float::binary_format<float>::min_exponent_fast_path): (fast_float::binary_format<double>::max_exponent_fast_path): (fast_float::binary_format<float>::max_exponent_fast_path): (fast_float::binary_format<double>::max_mantissa_fast_path): (fast_float::binary_format<float>::max_mantissa_fast_path): (fast_float::binary_format<double>::exact_power_of_ten): (fast_float::binary_format<float>::exact_power_of_ten): (fast_float::binary_format<double>::largest_power_of_ten): (fast_float::binary_format<float>::largest_power_of_ten): (fast_float::binary_format<double>::smallest_power_of_ten): (fast_float::binary_format<float>::smallest_power_of_ten): (fast_float::binary_format<double>::max_digits): (fast_float::binary_format<float>::max_digits): (fast_float::binary_format<float>::exponent_mask): (fast_float::binary_format<double>::exponent_mask): (fast_float::binary_format<float>::mantissa_mask): (fast_float::binary_format<double>::mantissa_mask): (fast_float::binary_format<float>::hidden_bit_mask): (fast_float::binary_format<double>::hidden_bit_mask): (fast_float::to_float): * Source/WTF/wtf/fast_float/parse_number.h: Added. * Source/WTF/wtf/fast_float/simple_decimal_conversion.h: Added. (fast_float::detail::trim): (fast_float::detail::number_of_digits_decimal_left_shift): (fast_float::detail::round): (fast_float::detail::decimal_left_shift): (fast_float::detail::decimal_right_shift): (fast_float::compute_float): (fast_float::parse_long_mantissa): Canonical link: https://commits.webkit.org/257675@main

webkit-commit-queue · 2022-12-10T06:56:33Z

Committed 257675@main (178e072): https://commits.webkit.org/257675@main

Reviewed commits have been landed. Closing PR #2772 and removing active labels.

Constellation self-assigned this Jul 27, 2022

Constellation added the WebKit Nightly Build label Jul 27, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from 3dd9123 to 9e41db1 Compare July 27, 2022 06:54

webkit-early-warning-system added the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from 9e41db1 to e1d5f9f Compare July 27, 2022 07:00

webkit-early-warning-system added the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from e1d5f9f to d1ebe50 Compare July 27, 2022 07:27

webkit-early-warning-system added the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch 2 times, most recently from 16c8fe0 to 07b0cb9 Compare July 27, 2022 09:30

webkit-early-warning-system added the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Jul 27, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch 2 times, most recently from 32a2d64 to 4d7604d Compare July 29, 2022 21:20

weinig approved these changes Jul 31, 2022

View reviewed changes

xeenon added WebCore Misc. For miscellaneous bugs in the WebCore framework (and not JavaScriptCore or WebKit). and removed WebKit Nightly Build labels Nov 21, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from 4d7604d to 498737e Compare December 2, 2022 04:38

Constellation marked this pull request as ready for review December 2, 2022 04:40

Constellation requested a review from a team as a code owner December 2, 2022 04:40

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from 498737e to b5e041d Compare December 2, 2022 04:45

webkit-ews-buildbot added the merging-blocked Applied to prevent a change from being merged label Dec 2, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Dec 2, 2022

webkit-ews-buildbot added the merging-blocked Applied to prevent a change from being merged label Dec 3, 2022

Constellation removed the merging-blocked Applied to prevent a change from being merged label Dec 7, 2022

Constellation force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from b5e041d to 900ddd2 Compare December 7, 2022 06:01

Constellation added the merge-queue Applied to send a pull request to merge-queue label Dec 10, 2022

webkit-early-warning-system force-pushed the eng/Investigate-replacing-double_conversion-library-with-fast_float branch from 900ddd2 to 178e072 Compare December 10, 2022 06:56

webkit-commit-queue removed the merge-queue Applied to send a pull request to merge-queue label Dec 10, 2022

webkit-early-warning-system merged commit 178e072 into WebKit:main Dec 10, 2022

Constellation deleted the eng/Investigate-replacing-double_conversion-library-with-fast_float branch December 12, 2022 00:50

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Investigate replacing double_conversion library with fast_float #2772

Investigate replacing double_conversion library with fast_float #2772

Constellation commented Jul 27, 2022 •

edited by webkit-early-warning-system

weinig left a comment

weinig Jul 31, 2022

weinig Jul 31, 2022

weinig commented Aug 1, 2022

webkit-early-warning-system commented Dec 2, 2022 •

edited

webkit-early-warning-system commented Dec 2, 2022

webkit-early-warning-system commented Dec 2, 2022 •

edited

Constellation commented Dec 2, 2022

webkit-early-warning-system commented Dec 7, 2022 •

edited

webkit-commit-queue commented Dec 10, 2022


		namespace WTF {

		WTF_EXPORT_PRIVATE double parseDouble(const LChar* string, size_t length, size_t& parsedLength);


		#include <wtf/fast_float/float_common.h>

		namespace fast_float {

Investigate replacing double_conversion library with fast_float #2772

Investigate replacing double_conversion library with fast_float #2772

Conversation

Constellation commented Jul 27, 2022 • edited by webkit-early-warning-system

178e072

weinig left a comment

Choose a reason for hiding this comment

weinig Jul 31, 2022

Choose a reason for hiding this comment

weinig Jul 31, 2022

Choose a reason for hiding this comment

weinig commented Aug 1, 2022

webkit-early-warning-system commented Dec 2, 2022 • edited

webkit-early-warning-system commented Dec 2, 2022

webkit-early-warning-system commented Dec 2, 2022 • edited

Constellation commented Dec 2, 2022

webkit-early-warning-system commented Dec 7, 2022 • edited

webkit-commit-queue commented Dec 10, 2022

Constellation commented Jul 27, 2022 •

edited by webkit-early-warning-system

`178e072`

webkit-early-warning-system commented Dec 2, 2022 •

edited

webkit-early-warning-system commented Dec 2, 2022 •

edited

webkit-early-warning-system commented Dec 7, 2022 •

edited