Skip to content

[AArch64] Optimize ZSTD_isRLE() with 64-byte NEON vector unrolling #4736

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

Open
wittkung wants to merge 1 commit into
base: dev
Choose a base branch
from
+30 −7
Open

[AArch64] Optimize ZSTD_isRLE() with 64-byte NEON vector unrolling #4736

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
37 changes: 30 additions & 7 deletions lib/compress/zstd_compress.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3595,18 +3595,41 @@ size_t ZSTD_mergeBlockDelimiters(ZSTD_Sequence* sequences, size_t seqsSize) {
/* Unrolled loop to read four size_ts of input at a time. Returns 1 if is RLE, 0 if not. */
static int ZSTD_isRLE(const BYTE* src, size_t length) {
const BYTE* ip = src;
const BYTE value = ip[0];
const size_t valueST = (size_t)((U64)value * 0x0101010101010101ULL);
const size_t unrollSize = sizeof(size_t) * 4;
const size_t unrollMask = unrollSize - 1;
const size_t prefixLength = length & unrollMask;
BYTE value;
size_t valueST;
size_t unrollSize;
size_t unrollMask;
size_t prefixLength;
size_t i;
if (length == 1) return 1;
if (length <= 1) return (int)length;
value = ip[0];
valueST = (size_t)((U64)value * 0x0101010101010101ULL);
unrollSize = sizeof(size_t) * 4;
unrollMask = unrollSize - 1;
prefixLength = length & unrollMask;
i = prefixLength;
/* Check if prefix is RLE first before using unrolled loop */
if (prefixLength && ZSTD_count(ip+1, ip, ip+prefixLength) != prefixLength-1) {
return 0;
}
for (i = prefixLength; i != length; i += unrollSize) {
#if defined(ZSTD_ARCH_ARM_NEON) && (defined(__aarch64__) || defined(_M_ARM64))
{
const uint8x16_t vval = vdupq_n_u8(value);
for (; i + 64 <= length; i += 64) {
const uint8x16_t q0 = veorq_u8(vld1q_u8(ip + i), vval);
const uint8x16_t q1 = veorq_u8(vld1q_u8(ip + i + 16), vval);
const uint8x16_t q2 = veorq_u8(vld1q_u8(ip + i + 32), vval);
const uint8x16_t q3 = veorq_u8(vld1q_u8(ip + i + 48), vval);
const uint8x16_t q_accum = vorrq_u8(vorrq_u8(q0, q1), vorrq_u8(q2, q3));
const U64 d0 = vgetq_lane_u64(vreinterpretq_u64_u8(q_accum), 0);
const U64 d1 = vgetq_lane_u64(vreinterpretq_u64_u8(q_accum), 1);
if ((d0 | d1) != 0) {
return 0;
}
}
}
#endif
for (; i != length; i += unrollSize) {
size_t u;
for (u = 0; u < unrollSize; u += sizeof(size_t)) {
if (MEM_readST(ip + i + u) != valueST) {
Expand Down