Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

buffer: optimize readDouble and readFloat methods #17775

Closed
wants to merge 2 commits into from
Closed

buffer: optimize readDouble and readFloat methods #17775

wants to merge 2 commits into from

Conversation

bnoordhuis
Copy link
Member

@bnoordhuis bnoordhuis commented Dec 20, 2017 •
edited

Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime. The improvements are not insubstantial:

                                                         improvement confidence      p.value
value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08

CI: https://ci.nodejs.org/job/node-test-commit/14968/

@bnoordhuis
buffer: optimize readDouble and readFloat methods
49817d3 
Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime.  The improvements are not insubstantial:

                                                             improvement confidence      p.value
    value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
    value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
    value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
    value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
    value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
    value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
    value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
    value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
    value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
    value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
    value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
    value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
    value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
    value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
    value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
    value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
    value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
    value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
    value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
    value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
    value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
    value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
    value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
    value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
    value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
    value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
    value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
    value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
    value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
    value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
    value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
    value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
    value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
    value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
    value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
    value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
    value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
    value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
    value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
    value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08
@nodejs-github-bot nodejs-github-bot added buffer Issues and PRs related to the buffer subsystem. c++ Issues and PRs that require attention from people who are familiar with C++. labels Dec 20, 2017
@bnoordhuis bnoordhuis mentioned this pull request Dec 20, 2017
Closed
Trott
Trott approved these changes Dec 20, 2017
View reviewed changes
Copy link
Member

@Trott Trott left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM if CI is green.

(Might be good to run make test-coverage to make sure we aren't reducing test coverage in buffer.js with this change, but even if we are, that doesn't necessarily need to be fixed in this PR. Could just open an issue and tag it good first contribution or something.)

@targos
Copy link
Member

targos commented Dec 20, 2017

Couldn't we somehow use DataView.prototype.getFloat32/64() ?

@andrasq
Copy link

andrasq commented Dec 20, 2017 •
edited

do these need a comprehensive set of tests now that they can't rely on the cpu test suite? I don't know what's there already.

@bnoordhuis
Copy link
Member Author

Couldn't we somehow use DataView.prototype.getFloat32/64() ?

Could? Yes. Should? Probably not.

I didn't benchmark it but getFloat32() and getFloat64() are not very optimized (run-time calls, not inline-able) so I guesstimate it's on the same order as the existing C++ code, and quite possibly slower because you have to construct a DataView for each call.

do these need a comprehensive set of tests now that they can't rely on the cpu test suite?

We have pretty good test coverage (NaN, positive/negative zero and infinity, denormals, the works) but I'll check whether all branches are exercised.

cjihrig
cjihrig approved these changes Dec 20, 2017
View reviewed changes
Copy link
Contributor

@cjihrig cjihrig left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice

BridgeAR
BridgeAR approved these changes Dec 20, 2017
View reviewed changes
Copy link
Member

@BridgeAR BridgeAR left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM with two suggestions.

lib/buffer.js
const sign = (x1 >>> 31) ? -1 : 1;
if (expt === 0) {
if (frac === 0) return sign * 0;
return sign * frac * 2 ** -1074;
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It would be nice to use brackets here for better readability. The same applies to all places where the exponential operator is used.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I kind of disagree since it's unambiguous: exponentiation has higher precedence than multiplication. Binary expressions with superfluous parens grinds my gears because it suggests its author didn't know the precedence rules.

I would have written it as sign * frac * 2**-1074 if eslint didn't complain about it, but it does.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm good with this without brackets as well.

benchmark/buffers/buffer-read-float.js
for (var i = 0; i !== ${len}; i++) {
buff.${fn}(0, ${JSON.stringify(noAssert)});
}
`);
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I do not really see the necessity for new Function here. It would be nice to just set a var to the specific function as in

for (var i = 0; i !== len; i++) {
  fn(0, noAssert);
}

(with a few other changes necessary for this)

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

That doesn't work. buf[fn](0, noAssert) could work but takes a different, non-default code path in V8 (keyed vs. named load.)

That said, I didn't invent this, the other buffer benchmarks do the same thing.

jasnell
jasnell reviewed Dec 20, 2017
View reviewed changes
lib/buffer.js
checkOffset(offset, 4, this.length);
return _readFloatLE(this, offset);
// For the casual reader who has not at the current time memorized the
// IEEE-754 standard in full detail: floating point numbers consist of
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah yes, IEEE-754... Lite-afternoon reading that goes perfect with coffee and scones.

jasnell
jasnell reviewed Dec 20, 2017
View reviewed changes
lib/buffer.js
//
// An all-bits-one exponent is either a positive or negative infinity, if
// the fraction is zero, or NaN when it is non-zero. The standard allows
// both quiet and signalling NaNs but since NaN is a canonical value in
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

extreme nit s/signalling/signaling if we're favouring the use of the American spellings ;-)

jasnell
jasnell approved these changes Dec 20, 2017
View reviewed changes
Copy link
Member

@jasnell jasnell left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This is great. Definite ++

@maclover7 maclover7 added the author ready PRs that have at least one approval, no pending requests for changes, and a CI started. label Dec 23, 2017
addaleax pushed a commit that referenced this pull request Dec 27, 2017
@bnoordhuis @addaleax
buffer: optimize readDouble and readFloat methods
df30fd5 
Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime.  The improvements are not insubstantial:

                                                             improvement confidence      p.value
    value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
    value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
    value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
    value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
    value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
    value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
    value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
    value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
    value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
    value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
    value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
    value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
    value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
    value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
    value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
    value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
    value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
    value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
    value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
    value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
    value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
    value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
    value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
    value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
    value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
    value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
    value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
    value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
    value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
    value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
    value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
    value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
    value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
    value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
    value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
    value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
    value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
    value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
    value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
    value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08

PR-URL: #17775
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
@addaleax
Copy link
Member

Landed in df30fd5

@addaleax addaleax closed this Dec 27, 2017
@addaleax addaleax added performance Issues and PRs related to the performance of Node.js. and removed author ready PRs that have at least one approval, no pending requests for changes, and a CI started. labels Dec 27, 2017
MylesBorins pushed a commit that referenced this pull request Jan 8, 2018
@bnoordhuis @MylesBorins
buffer: optimize readDouble and readFloat methods
a203f0c 
Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime.  The improvements are not insubstantial:

                                                             improvement confidence      p.value
    value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
    value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
    value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
    value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
    value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
    value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
    value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
    value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
    value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
    value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
    value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
    value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
    value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
    value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
    value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
    value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
    value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
    value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
    value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
    value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
    value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
    value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
    value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
    value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
    value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
    value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
    value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
    value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
    value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
    value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
    value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
    value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
    value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
    value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
    value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
    value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
    value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
    value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
    value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
    value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08

PR-URL: #17775
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
MylesBorins pushed a commit that referenced this pull request Jan 9, 2018
@bnoordhuis @MylesBorins
buffer: optimize readDouble and readFloat methods
ac68656 
Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime.  The improvements are not insubstantial:

                                                             improvement confidence      p.value
    value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
    value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
    value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
    value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
    value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
    value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
    value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
    value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
    value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
    value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
    value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
    value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
    value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
    value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
    value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
    value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
    value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
    value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
    value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
    value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
    value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
    value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
    value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
    value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
    value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
    value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
    value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
    value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
    value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
    value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
    value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
    value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
    value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
    value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
    value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
    value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
    value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
    value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
    value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
    value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08

PR-URL: #17775
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
MylesBorins pushed a commit that referenced this pull request Jan 9, 2018
@bnoordhuis @MylesBorins
buffer: optimize readDouble and readFloat methods
0aa403b 
Compute the floating point number in JavaScript to avoid having to call
out to the C++ runtime.  The improvements are not insubstantial:

                                                             improvement confidence      p.value
    value="big" endian="BE" type="Double" noAssert="false"      292.86 %        *** 1.688367e-08
    value="big" endian="BE" type="Double" noAssert="true"       353.19 %        *** 6.079414e-10
    value="big" endian="BE" type="Float" noAssert="false"       406.21 %        *** 1.730122e-07
    value="big" endian="BE" type="Float" noAssert="true"        450.81 %        *** 6.909242e-07
    value="big" endian="LE" type="Double" noAssert="false"      268.39 %        *** 8.625486e-09
    value="big" endian="LE" type="Double" noAssert="true"       310.66 %        *** 2.798332e-15
    value="big" endian="LE" type="Float" noAssert="false"       382.99 %        *** 3.412057e-07
    value="big" endian="LE" type="Float" noAssert="true"        394.60 %        *** 1.406742e-07
    value="inf" endian="BE" type="Double" noAssert="false"      312.91 %        *** 7.407943e-12
    value="inf" endian="BE" type="Double" noAssert="true"       392.47 %        *** 3.821179e-08
    value="inf" endian="BE" type="Float" noAssert="false"       466.01 %        *** 8.953363e-08
    value="inf" endian="BE" type="Float" noAssert="true"        460.76 %        *** 5.381256e-09
    value="inf" endian="LE" type="Double" noAssert="false"      279.50 %        *** 2.390682e-09
    value="inf" endian="LE" type="Double" noAssert="true"       335.30 %        *** 3.587173e-09
    value="inf" endian="LE" type="Float" noAssert="false"       439.77 %        *** 1.057133e-07
    value="inf" endian="LE" type="Float" noAssert="true"        426.72 %        *** 4.353408e-09
    value="nan" endian="BE" type="Double" noAssert="false"      271.18 %        *** 2.281526e-05
    value="nan" endian="BE" type="Double" noAssert="true"       312.63 %        *** 1.974975e-07
    value="nan" endian="BE" type="Float" noAssert="false"       429.17 %        *** 2.416228e-07
    value="nan" endian="BE" type="Float" noAssert="true"        461.39 %        *** 1.956714e-08
    value="nan" endian="LE" type="Double" noAssert="false"      267.03 %        *** 9.938479e-12
    value="nan" endian="LE" type="Double" noAssert="true"       276.93 %        *** 7.842481e-08
    value="nan" endian="LE" type="Float" noAssert="false"       415.97 %        *** 8.082710e-07
    value="nan" endian="LE" type="Float" noAssert="true"        433.68 %        *** 1.030200e-07
    value="small" endian="BE" type="Double" noAssert="false"    273.20 %        *** 9.071652e-11
    value="small" endian="BE" type="Double" noAssert="true"     326.25 %        *** 3.120167e-08
    value="small" endian="BE" type="Float" noAssert="false"     845.61 %        *** 8.044170e-08
    value="small" endian="BE" type="Float" noAssert="true"      868.61 %        *** 2.944539e-08
    value="small" endian="LE" type="Double" noAssert="false"    251.29 %        *** 5.613930e-09
    value="small" endian="LE" type="Double" noAssert="true"     286.82 %        *** 8.149603e-10
    value="small" endian="LE" type="Float" noAssert="false"     824.87 %        *** 1.199729e-08
    value="small" endian="LE" type="Float" noAssert="true"      834.35 %        *** 4.799620e-08
    value="zero" endian="BE" type="Double" noAssert="false"     216.70 %        *** 3.872293e-12
    value="zero" endian="BE" type="Double" noAssert="true"      239.31 %        *** 6.439601e-09
    value="zero" endian="BE" type="Float" noAssert="false"      353.75 %        *** 3.639974e-07
    value="zero" endian="BE" type="Float" noAssert="true"       388.86 %        *** 7.074318e-10
    value="zero" endian="LE" type="Double" noAssert="false"     179.34 %        *** 5.230763e-06
    value="zero" endian="LE" type="Double" noAssert="true"      199.66 %        *** 2.177589e-11
    value="zero" endian="LE" type="Float" noAssert="false"      299.55 %        *** 9.961978e-08
    value="zero" endian="LE" type="Float" noAssert="true"       333.30 %        *** 2.470764e-08

PR-URL: #17775
Reviewed-By: Rich Trott <rtrott@gmail.com>
Reviewed-By: Colin Ihrig <cjihrig@gmail.com>
Reviewed-By: Ruben Bridgewater <ruben@bridgewater.de>
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Anna Henningsen <anna@addaleax.net>
@MylesBorins MylesBorins mentioned this pull request Jan 10, 2018
Merged
@bnoordhuis bnoordhuis mentioned this pull request Jan 12, 2018
Closed
@gibfahn
Copy link
Member

gibfahn commented Jan 24, 2018

If someone would like to check that this provides a similar perf boost on 6.x and 8.x, we can backport this.

@arthurschreiber arthurschreiber mentioned this pull request Jan 28, 2018
Closed
@arthurschreiber
Copy link

I think these changes broke readDoubleLE: #18419

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@jasnell jasnell jasnell approved these changes

@Trott Trott Trott approved these changes

@addaleax addaleax addaleax approved these changes

@cjihrig cjihrig cjihrig approved these changes

@BridgeAR BridgeAR BridgeAR approved these changes

Assignees
No one assigned
Labels
buffer Issues and PRs related to the buffer subsystem. c++ Issues and PRs that require attention from people who are familiar with C++. performance Issues and PRs related to the performance of Node.js.
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet