Skip to content

src: improve base64 encoding performance #39701

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.

Sign up for GitHub

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

Closed
wants to merge 1 commit into from
Closed

src: improve base64 encoding performance #39701

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
69 changes: 58 additions & 11 deletions src/base64-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,23 @@
#include "base64.h"
#include "util.h"

#if (__BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) || \
defined(__LITTLE_ENDIAN__) || \
_MSC_VER
// Little Endian
#ifdef _MSC_VER
#define NODE_BSWAP32(x) _byteswap_ulong(x)
#define NODE_BSWAP64(x) _byteswap_uint64(x)
#else
#define NODE_BSWAP32(x) __builtin_bswap32(x)
#define NODE_BSWAP64(x) __builtin_bswap64(x)
#endif
#else
// Big Endian
#define NODE_BSWAP32(x) (x)
#define NODE_BSWAP64(x) (x)
#endif

namespace node {

extern const int8_t unbase64_table[256];
Expand Down Expand Up @@ -136,29 +153,50 @@ inline size_t base64_encode(const char* src,
unsigned c;
unsigned i;
unsigned k;
unsigned n;
unsigned n32;
unsigned n64;

const char* table = base64_select_table(mode);

i = 0;
k = 0;
n = slen / 3 * 3;
n32 = (slen >= 4 ? ((slen - 1) / 3) * 3 : 0);
n64 = (slen >= 8 ? ((slen - 2) / 6) * 3 : 0);

// Read in chunks of 8 bytes for as long as possible
while (i < n64) {
const uint64_t dword = *reinterpret_cast<const uint64_t*>(src + i);
Copy link
Member

Choose a reason for hiding this comment

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

Some architectures don't support unaligned 64-bit access, so this could end up being somewhat slower on those. But still an optimization on most systems!

Copy link
Contributor Author

Choose a reason for hiding this comment

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

That's why I was curious to check on ARM. However from what I'm seeing ARMv7 and newer support unaligned reads, with ARMv7 supposedly not supporting them for only 2 instructions?

Copy link
Member

Choose a reason for hiding this comment

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

Have you tried to start reading 1/4 bytes until 8 byte alignment is reached?

const uint64_t swapped = NODE_BSWAP64(dword);

dst[k + 0] = table[(swapped >> 58) & 0x3f];
dst[k + 1] = table[(swapped >> 52) & 0x3f];
dst[k + 2] = table[(swapped >> 46) & 0x3f];
dst[k + 3] = table[(swapped >> 40) & 0x3f];
dst[k + 4] = table[(swapped >> 34) & 0x3f];
dst[k + 5] = table[(swapped >> 28) & 0x3f];
dst[k + 6] = table[(swapped >> 22) & 0x3f];
dst[k + 7] = table[(swapped >> 16) & 0x3f];

i += 6;
k += 8;
}

while (i < n) {
a = src[i + 0] & 0xff;
b = src[i + 1] & 0xff;
c = src[i + 2] & 0xff;
// Read in chunks of 4 bytes for as long as possible
while (i < n32) {
const uint32_t dword = *reinterpret_cast<const uint32_t*>(src + i);
const uint32_t swapped = NODE_BSWAP32(dword);

dst[k + 0] = table[a >> 2];
dst[k + 1] = table[((a & 3) << 4) | (b >> 4)];
dst[k + 2] = table[((b & 0x0f) << 2) | (c >> 6)];
dst[k + 3] = table[c & 0x3f];
dst[k + 0] = table[(swapped >> 26) & 0x3f];
dst[k + 1] = table[(swapped >> 20) & 0x3f];
dst[k + 2] = table[(swapped >> 14) & 0x3f];
dst[k + 3] = table[(swapped >> 8) & 0x3f];

i += 3;
k += 4;
}

switch (slen - n) {
// Deal with leftover bytes
switch (slen - n32) {
case 1:
a = src[i + 0] & 0xff;
dst[k + 0] = table[a >> 2];
Expand All @@ -177,6 +215,15 @@ inline size_t base64_encode(const char* src,
if (mode == Base64Mode::NORMAL)
dst[k + 3] = '=';
break;
case 3:
a = src[i + 0] & 0xff;
b = src[i + 1] & 0xff;
c = src[i + 2] & 0xff;
dst[k + 0] = table[a >> 2];
dst[k + 1] = table[((a & 3) << 4) | (b >> 4)];
dst[k + 2] = table[((b & 0x0f) << 2) | (c >> 6)];
dst[k + 3] = table[c & 0x3f];
break;
}

return dlen;
Expand Down