Skip to content

Skip more fast paths in memmove for shrink level >= 2 #1072

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

Merged
dcodeIO merged 2 commits into from
Jan 24, 2020
Merged

Skip more fast paths in memmove for shrink level >= 2 #1072

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6b33bf8
skip more fast paths. Remove incosistencies
MaxGraey Jan 24, 2020
e92a186
up level to 2
MaxGraey Jan 24, 2020
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
305 changes: 151 additions & 154 deletions std/assembly/util/memory.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,150 +1,143 @@
export function memcpy(dest: usize, src: usize, n: usize): void { // see: musl/src/string/memcpy.c
if (ASC_SHRINK_LEVEL > 1) {
while (n) {
store<u8>(dest++, load<u8>(src++));
--n;
}
} else {
let w: u32, x: u32;
var w: u32, x: u32;

// copy 1 byte each until src is aligned to 4 bytes
while (n && (src & 3)) {
// copy 1 byte each until src is aligned to 4 bytes
while (n && (src & 3)) {
store<u8>(dest++, load<u8>(src++));
n--;
}

// if dst is aligned to 4 bytes as well, copy 4 bytes each
if ((dest & 3) == 0) {
while (n >= 16) {
store<u32>(dest , load<u32>(src ));
store<u32>(dest + 4, load<u32>(src + 4));
store<u32>(dest + 8, load<u32>(src + 8));
store<u32>(dest + 12, load<u32>(src + 12));
src += 16; dest += 16; n -= 16;
}
if (n & 8) {
store<u32>(dest , load<u32>(src ));
store<u32>(dest + 4, load<u32>(src + 4));
dest += 8; src += 8;
}
if (n & 4) {
store<u32>(dest, load<u32>(src));
dest += 4; src += 4;
}
if (n & 2) { // drop to 2 bytes each
store<u16>(dest, load<u16>(src));
dest += 2; src += 2;
}
if (n & 1) { // drop to 1 byte
store<u8>(dest++, load<u8>(src++));
n--;
}
return;
}

// if dst is aligned to 4 bytes as well, copy 4 bytes each
if ((dest & 3) == 0) {
while (n >= 16) {
store<u32>(dest , load<u32>(src ));
store<u32>(dest + 4, load<u32>(src + 4));
store<u32>(dest + 8, load<u32>(src + 8));
store<u32>(dest + 12, load<u32>(src + 12));
src += 16; dest += 16; n -= 16;
}
if (n & 8) {
store<u32>(dest , load<u32>(src ));
store<u32>(dest + 4, load<u32>(src + 4));
dest += 8; src += 8;
}
if (n & 4) {
store<u32>(dest, load<u32>(src));
dest += 4; src += 4;
}
if (n & 2) { // drop to 2 bytes each
store<u16>(dest, load<u16>(src));
dest += 2; src += 2;
}
if (n & 1) { // drop to 1 byte
// if dst is not aligned to 4 bytes, use alternating shifts to copy 4 bytes each
// doing shifts if faster when copying enough bytes (here: 32 or more)
if (n >= 32) {
switch (dest & 3) {
// known to be != 0
case 1: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
}
return;
}

// if dst is not aligned to 4 bytes, use alternating shifts to copy 4 bytes each
// doing shifts if faster when copying enough bytes (here: 32 or more)
if (n >= 32) {
switch (dest & 3) {
// known to be != 0
case 1: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
n -= 3;
while (n >= 17) {
x = load<u32>(src + 1);
store<u32>(dest, w >> 24 | x << 8);
w = load<u32>(src + 5);
store<u32>(dest + 4, x >> 24 | w << 8);
x = load<u32>(src + 9);
store<u32>(dest + 8, w >> 24 | x << 8);
w = load<u32>(src + 13);
store<u32>(dest + 12, x >> 24 | w << 8);
src += 16; dest += 16; n -= 16;
}
break;
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
n -= 3;
while (n >= 17) {
x = load<u32>(src + 1);
store<u32>(dest, w >> 24 | x << 8);
w = load<u32>(src + 5);
store<u32>(dest + 4, x >> 24 | w << 8);
x = load<u32>(src + 9);
store<u32>(dest + 8, w >> 24 | x << 8);
w = load<u32>(src + 13);
store<u32>(dest + 12, x >> 24 | w << 8);
src += 16; dest += 16; n -= 16;
}
case 2: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
n -= 2;
while (n >= 18) {
x = load<u32>(src + 2);
store<u32>(dest, w >> 16 | x << 16);
w = load<u32>(src + 6);
store<u32>(dest + 4, x >> 16 | w << 16);
x = load<u32>(src + 10);
store<u32>(dest + 8, w >> 16 | x << 16);
w = load<u32>(src + 14);
store<u32>(dest + 12, x >> 16 | w << 16);
src += 16; dest += 16; n -= 16;
}
break;
break;
}
case 2: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
n -= 2;
while (n >= 18) {
x = load<u32>(src + 2);
store<u32>(dest, w >> 16 | x << 16);
w = load<u32>(src + 6);
store<u32>(dest + 4, x >> 16 | w << 16);
x = load<u32>(src + 10);
store<u32>(dest + 8, w >> 16 | x << 16);
w = load<u32>(src + 14);
store<u32>(dest + 12, x >> 16 | w << 16);
src += 16; dest += 16; n -= 16;
}
case 3: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
n -= 1;
while (n >= 19) {
x = load<u32>(src + 3);
store<u32>(dest, w >> 8 | x << 24);
w = load<u32>(src + 7);
store<u32>(dest + 4, x >> 8 | w << 24);
x = load<u32>(src + 11);
store<u32>(dest + 8, w >> 8 | x << 24);
w = load<u32>(src + 15);
store<u32>(dest + 12, x >> 8 | w << 24);
src += 16; dest += 16; n -= 16;
}
break;
break;
}
case 3: {
w = load<u32>(src);
store<u8>(dest++, load<u8>(src++));
n -= 1;
while (n >= 19) {
x = load<u32>(src + 3);
store<u32>(dest, w >> 8 | x << 24);
w = load<u32>(src + 7);
store<u32>(dest + 4, x >> 8 | w << 24);
x = load<u32>(src + 11);
store<u32>(dest + 8, w >> 8 | x << 24);
w = load<u32>(src + 15);
store<u32>(dest + 12, x >> 8 | w << 24);
src += 16; dest += 16; n -= 16;
}
break;
}
}
}

// copy remaining bytes one by one
if (n & 16) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 8) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 4) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 2) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 1) {
store<u8>(dest++, load<u8>(src++));
}
// copy remaining bytes one by one
if (n & 16) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 8) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 4) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 2) {
store<u8>(dest++, load<u8>(src++));
store<u8>(dest++, load<u8>(src++));
}
if (n & 1) {
store<u8>(dest++, load<u8>(src++));
}
}

Expand All @@ -159,32 +152,36 @@ export function memmove(dest: usize, src: usize, n: usize): void { // see: musl/
}
}
if (dest < src) {
if ((src & 7) == (dest & 7)) {
while (dest & 7) {
if (!n) return;
--n;
store<u8>(dest++, load<u8>(src++));
}
while (n >= 8) {
store<u64>(dest, load<u64>(src));
n -= 8;
dest += 8;
src += 8;
if (ASC_SHRINK_LEVEL < 2) {
if ((src & 7) == (dest & 7)) {
while (dest & 7) {
if (!n) return;
--n;
store<u8>(dest++, load<u8>(src++));
}
while (n >= 8) {
store<u64>(dest, load<u64>(src));
n -= 8;
dest += 8;
src += 8;
}
}
}
while (n) {
store<u8>(dest++, load<u8>(src++));
--n;
}
} else {
if ((src & 7) == (dest & 7)) {
while ((dest + n) & 7) {
if (!n) return;
store<u8>(dest + --n, load<u8>(src + n));
}
while (n >= 8) {
n -= 8;
store<u64>(dest + n, load<u64>(src + n));
if (ASC_SHRINK_LEVEL < 2) {
if ((src & 7) == (dest & 7)) {
while ((dest + n) & 7) {
if (!n) return;
store<u8>(dest + --n, load<u8>(src + n));
}
while (n >= 8) {
n -= 8;
store<u64>(dest + n, load<u64>(src + n));
}
}
}
while (n) {
Expand Down