reduce buffer creation for ctr mode #48
Conversation
| self._cache.writeUInt32BE(out[0], offset + 0) | ||
| self._cache.writeUInt32BE(out[1], offset + 4) | ||
| self._cache.writeUInt32BE(out[2], offset + 8) | ||
| self._cache.writeUInt32BE(out[3], offset + 12) |
There was a problem hiding this comment.
Is this really better than a copy?
There was a problem hiding this comment.
The writeUint is not better, but the reason I am doing this is that it avoids creating intermediary buffers. Before encryptBlock creates a new buffer, fills it with those four values, just so that the buffer can be copied in here. By using the array directly there is now one less buffer allocation in the loop and together with the preallocation, no buffer allocations are in the loop.
There was a problem hiding this comment.
Nevermind, my mistake, I thought out was a Buffer.
ACK :)
|
Why are you renaming the method?
…On Thu, Aug 17, 2017, 5:25 AM Daniel Cousens ***@***.***> wrote:
***@***.**** commented on this pull request.
------------------------------
In modes/ctr.js
<#48 (comment)>
:
> exports.encrypt = function (self, chunk) {
- while (self._cache.length < chunk.length) {
- self._cache = Buffer.concat([self._cache, getBlock(self)])
+ var chunkNum = Math.ceil(chunk.length / blockSize)
+ var start = self._cache.length
+ self._cache = Buffer.concat([
+ self._cache,
+ Buffer.allocUnsafe(chunkNum * blockSize)
+ ])
+ for (var i = 0; i < chunkNum; i++) {
+ var out = getBlock(self)
+ var offset = start + i * blockSize
+ self._cache.writeUInt32BE(out[0], offset + 0)
+ self._cache.writeUInt32BE(out[1], offset + 4)
+ self._cache.writeUInt32BE(out[2], offset + 8)
+ self._cache.writeUInt32BE(out[3], offset + 12)
Nevermind, my mistake, I thought out was a Buffer.
ACK :)
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#48 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/ABE4nxIilA_HN1mmMna5KyGEAqdGWuR9ks5sZAckgaJpZM4O5kwc>
.
|
|
@calvinmetcalf that's just githubs diff I pulled part of the function into it's own function, as I needed that functionality |
|
oh my bad |
|
@calvinmetcalf could you release? |
|
hey @calvinmetcalf @dcousens , When do you think this can be released ? |
|
@calvinmetcalf @dcousens yes please! :) |
|
Can't release yet, my local tests throw:
Investingating within 24 hours |
When using ctr mode, I saw a large amount of buffer creation + gc when running perf analysis. This avoids generating most of the temporary buffers when running update in ctr mode improving perf in the browser nicely.