From: Dave Rodgman <dave.rodgman@xxxxxxx> Subject: lib/lzo: implement run-length encoding Patch series "lib/lzo: run-length encoding support", v5. Following on from the previous lzo-rle patchset: https://lkml.org/lkml/2018/11/30/972 This patchset contains only the RLE patches, and should be applied on top of the non-RLE patches ( https://lkml.org/lkml/2019/2/5/366 ). Previously, some questions were raised around the RLE patches. I've done some additional benchmarking to answer these questions. In short: - RLE offers significant additional performance (data-dependent) - I didn't measure any regressions that were clearly outside the noise One concern with this patchset was around performance - specifically, measuring RLE impact separately from Matt Sealey's patches (CTZ & fast copy). I have done some additional benchmarking which I hope clarifies the benefits of each part of the patchset. Firstly, I've captured some memory via /dev/fmem from a Chromebook with many tabs open which is starting to swap, and then split this into 4178 4k pages. I've excluded the all-zero pages (as zram does), and also the no-zero pages (which won't tell us anything about RLE performance). This should give a realistic test dataset for zram. What I found was that the data is VERY bimodal: 44% of pages in this dataset contain 5% or fewer zeros, and 44% contain over 90% zeros (30% if you include the no-zero pages). This supports the idea of special-casing zeros in zram. Next, I've benchmarked four variants of lzo on these pages (on 64-bit Arm at max frequency): baseline LZO; baseline + Matt Sealey's patches (aka MS); baseline + RLE only; baseline + MS + RLE. Numbers are for weighted roundtrip throughput (the weighting reflects that zram does more compression than decompression). https://drive.google.com/file/d/1VLtLjRVxgUNuWFOxaGPwJYhl_hMQXpHe/view?usp=sharing Matt's patches help in all cases for Arm (and no effect on Intel), as expected. RLE also behaves as expected: with few zeros present, it makes no difference; above ~75%, it gives a good improvement (50 - 300 MB/s on top of the benefit from Matt's patches). Best performance is seen with both MS and RLE patches. Finally, I have benchmarked the same dataset on an x86-64 device. Here, the MS patches make no difference (as expected); RLE helps, similarly as on Arm. There were no definite regressions; allowing for observational error, 0.1% (3/4178) of cases had a regression > 1 standard deviation, of which the largest was 4.6% (1.2 standard deviations). I think this is probably within the noise. https://drive.google.com/file/d/1xCUVwmiGD0heEMx5gcVEmLBI4eLaageV/view?usp=sharing One point to note is that the graphs show RLE appears to help very slightly with no zeros present! This is because the extra code causes the clang optimiser to change code layout in a way that happens to have a significant benefit. Taking baseline LZO and adding a do-nothing line like "__builtin_prefetch(out_len);" immediately before the "goto next" has the same effect. So this is a real, but basically spurious effect - it's small enough not to upset the overall findings. This patch (of 3): When using zram, we frequently encounter long runs of zero bytes. This adds a special case which identifies runs of zeros and encodes them using run-length encoding. This is faster for both compression and decompresion. For high-entropy data which doesn't hit this case, impact is minimal. Compression ratio is within a few percent in all cases. This modifies the bitstream in a way which is backwards compatible (i.e., we can decompress old bitstreams, but old versions of lzo cannot decompress new bitstreams). Link: http://lkml.kernel.org/r/20190205155944.16007-2-dave.rodgman@xxxxxxx Signed-off-by: Dave Rodgman <dave.rodgman@xxxxxxx> Cc: David S. Miller <davem@xxxxxxxxxxxxx> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> Cc: Herbert Xu <herbert@xxxxxxxxxxxxxxxxxxx> Cc: Markus F.X.J. Oberhumer <markus@xxxxxxxxxxxxx> Cc: Matt Sealey <matt.sealey@xxxxxxx> Cc: Minchan Kim <minchan@xxxxxxxxxx> Cc: Nitin Gupta <nitingupta910@xxxxxxxxx> Cc: Richard Purdie <rpurdie@xxxxxxxxxxxxxx> Cc: Sergey Senozhatsky <sergey.senozhatsky.work@xxxxxxxxx> Cc: Sonny Rao <sonnyrao@xxxxxxxxxx> Signed-off-by: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> --- --- a/Documentation/lzo.txt~lib-lzo-implement-run-length-encoding +++ a/Documentation/lzo.txt @@ -78,16 +78,30 @@ Description is an implementation design choice independent on the algorithm or encoding. +Versions + +0: Original version +1: LZO-RLE + +Version 1 of LZO implements an extension to encode runs of zeros using run +length encoding. This improves speed for data with many zeros, which is a +common case for zram. This modifies the bitstream in a backwards compatible way +(v1 can correctly decompress v0 compressed data, but v0 cannot read v1 data). + Byte sequences ============== First byte encoding:: - 0..17 : follow regular instruction encoding, see below. It is worth - noting that codes 16 and 17 will represent a block copy from - the dictionary which is empty, and that they will always be + 0..16 : follow regular instruction encoding, see below. It is worth + noting that code 16 will represent a block copy from the + dictionary which is empty, and that it will always be invalid at this place. + 17 : bitstream version. If the first byte is 17, the next byte + gives the bitstream version. If the first byte is not 17, + the bitstream version is 0. + 18..21 : copy 0..3 literals state = (byte - 17) = 0..3 [ copy <state> literals ] skip byte @@ -140,6 +154,11 @@ Byte sequences state = S (copy S literals after this block) End of stream is reached if distance == 16384 + In version 1, this instruction is also used to encode a run of zeros if + distance = 0xbfff, i.e. H = 1 and the D bits are all 1. + In this case, it is followed by a fourth byte, X. + run length = ((X << 3) | (0 0 0 0 0 L L L)) + 4. + 0 0 1 L L L L L (32..63) Copy of small block within 16kB distance (preferably less than 34B) length = 2 + (L ?: 31 + (zero_bytes * 255) + non_zero_byte) @@ -165,7 +184,9 @@ Authors ======= This document was written by Willy Tarreau <w@xxxxxx> on 2014/07/19 during an - analysis of the decompression code available in Linux 3.16-rc5. The code is - tricky, it is possible that this document contains mistakes or that a few - corner cases were overlooked. In any case, please report any doubt, fix, or - proposed updates to the author(s) so that the document can be updated. + analysis of the decompression code available in Linux 3.16-rc5, and updated + by Dave Rodgman <dave.rodgman@xxxxxxx> on 2018/10/30 to introduce run-length + encoding. The code is tricky, it is possible that this document contains + mistakes or that a few corner cases were overlooked. In any case, please + report any doubt, fix, or proposed updates to the author(s) so that the + document can be updated. --- a/include/linux/lzo.h~lib-lzo-implement-run-length-encoding +++ a/include/linux/lzo.h @@ -18,7 +18,7 @@ #define LZO1X_1_MEM_COMPRESS (8192 * sizeof(unsigned short)) #define LZO1X_MEM_COMPRESS LZO1X_1_MEM_COMPRESS -#define lzo1x_worst_compress(x) ((x) + ((x) / 16) + 64 + 3) +#define lzo1x_worst_compress(x) ((x) + ((x) / 16) + 64 + 3 + 2) /* This requires 'wrkmem' of size LZO1X_1_MEM_COMPRESS */ int lzo1x_1_compress(const unsigned char *src, size_t src_len, --- a/lib/lzo/lzo1x_compress.c~lib-lzo-implement-run-length-encoding +++ a/lib/lzo/lzo1x_compress.c @@ -20,7 +20,7 @@ static noinline size_t lzo1x_1_do_compress(const unsigned char *in, size_t in_len, unsigned char *out, size_t *out_len, - size_t ti, void *wrkmem) + size_t ti, void *wrkmem, signed char *state_offset) { const unsigned char *ip; unsigned char *op; @@ -35,27 +35,85 @@ lzo1x_1_do_compress(const unsigned char ip += ti < 4 ? 4 - ti : 0; for (;;) { - const unsigned char *m_pos; + const unsigned char *m_pos = NULL; size_t t, m_len, m_off; u32 dv; + u32 run_length = 0; literal: ip += 1 + ((ip - ii) >> 5); next: if (unlikely(ip >= ip_end)) break; dv = get_unaligned_le32(ip); - t = ((dv * 0x1824429d) >> (32 - D_BITS)) & D_MASK; - m_pos = in + dict[t]; - dict[t] = (lzo_dict_t) (ip - in); - if (unlikely(dv != get_unaligned_le32(m_pos))) - goto literal; + + if (dv == 0) { + const unsigned char *ir = ip + 4; + const unsigned char *limit = ip_end + < (ip + MAX_ZERO_RUN_LENGTH + 1) + ? ip_end : ip + MAX_ZERO_RUN_LENGTH + 1; +#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && \ + defined(LZO_FAST_64BIT_MEMORY_ACCESS) + u64 dv64; + + for (; (ir + 32) <= limit; ir += 32) { + dv64 = get_unaligned((u64 *)ir); + dv64 |= get_unaligned((u64 *)ir + 1); + dv64 |= get_unaligned((u64 *)ir + 2); + dv64 |= get_unaligned((u64 *)ir + 3); + if (dv64) + break; + } + for (; (ir + 8) <= limit; ir += 8) { + dv64 = get_unaligned((u64 *)ir); + if (dv64) { +# if defined(__LITTLE_ENDIAN) + ir += __builtin_ctzll(dv64) >> 3; +# elif defined(__BIG_ENDIAN) + ir += __builtin_clzll(dv64) >> 3; +# else +# error "missing endian definition" +# endif + break; + } + } +#else + while ((ir < (const unsigned char *) + ALIGN((uintptr_t)ir, 4)) && + (ir < limit) && (*ir == 0)) + ir++; + for (; (ir + 4) <= limit; ir += 4) { + dv = *((u32 *)ir); + if (dv) { +# if defined(__LITTLE_ENDIAN) + ir += __builtin_ctz(dv) >> 3; +# elif defined(__BIG_ENDIAN) + ir += __builtin_clz(dv) >> 3; +# else +# error "missing endian definition" +# endif + break; + } + } +#endif + while (likely(ir < limit) && unlikely(*ir == 0)) + ir++; + run_length = ir - ip; + if (run_length > MAX_ZERO_RUN_LENGTH) + run_length = MAX_ZERO_RUN_LENGTH; + } else { + t = ((dv * 0x1824429d) >> (32 - D_BITS)) & D_MASK; + m_pos = in + dict[t]; + dict[t] = (lzo_dict_t) (ip - in); + if (unlikely(dv != get_unaligned_le32(m_pos))) + goto literal; + } ii -= ti; ti = 0; t = ip - ii; if (t != 0) { if (t <= 3) { - op[-2] |= t; + op[*state_offset] |= t; COPY4(op, ii); op += t; } else if (t <= 16) { @@ -88,6 +146,17 @@ next: } } + if (unlikely(run_length)) { + ip += run_length; + run_length -= MIN_ZERO_RUN_LENGTH; + put_unaligned_le32((run_length << 21) | 0xfffc18 + | (run_length & 0x7), op); + op += 4; + run_length = 0; + *state_offset = -3; + goto finished_writing_instruction; + } + m_len = 4; { #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(LZO_USE_CTZ64) @@ -170,7 +239,6 @@ m_len_done: m_off = ip - m_pos; ip += m_len; - ii = ip; if (m_len <= M2_MAX_LEN && m_off <= M2_MAX_OFFSET) { m_off -= 1; *op++ = (((m_len - 1) << 5) | ((m_off & 7) << 2)); @@ -207,6 +275,9 @@ m_len_done: *op++ = (m_off << 2); *op++ = (m_off >> 6); } + *state_offset = -2; +finished_writing_instruction: + ii = ip; goto next; } *out_len = op - out; @@ -221,6 +292,12 @@ int lzo1x_1_compress(const unsigned char unsigned char *op = out; size_t l = in_len; size_t t = 0; + signed char state_offset = -2; + + // LZO v0 will never write 17 as first byte, + // so this is used to version the bitstream + *op++ = 17; + *op++ = LZO_VERSION; while (l > 20) { size_t ll = l <= (M4_MAX_OFFSET + 1) ? l : (M4_MAX_OFFSET + 1); @@ -229,7 +306,8 @@ int lzo1x_1_compress(const unsigned char break; BUILD_BUG_ON(D_SIZE * sizeof(lzo_dict_t) > LZO1X_1_MEM_COMPRESS); memset(wrkmem, 0, D_SIZE * sizeof(lzo_dict_t)); - t = lzo1x_1_do_compress(ip, ll, op, out_len, t, wrkmem); + t = lzo1x_1_do_compress(ip, ll, op, out_len, + t, wrkmem, &state_offset); ip += ll; op += *out_len; l -= ll; @@ -242,7 +320,7 @@ int lzo1x_1_compress(const unsigned char if (op == out && t <= 238) { *op++ = (17 + t); } else if (t <= 3) { - op[-2] |= t; + op[state_offset] |= t; } else if (t <= 18) { *op++ = (t - 3); } else { --- a/lib/lzo/lzo1x_decompress_safe.c~lib-lzo-implement-run-length-encoding +++ a/lib/lzo/lzo1x_decompress_safe.c @@ -46,11 +46,23 @@ int lzo1x_decompress_safe(const unsigned const unsigned char * const ip_end = in + in_len; unsigned char * const op_end = out + *out_len; + unsigned char bitstream_version; + op = out; ip = in; if (unlikely(in_len < 3)) goto input_overrun; + + if (likely(*ip == 17)) { + bitstream_version = ip[1]; + ip += 2; + if (unlikely(in_len < 5)) + goto input_overrun; + } else { + bitstream_version = 0; + } + if (*ip > 17) { t = *ip++ - 17; if (t < 4) { @@ -154,32 +166,49 @@ copy_literal_run: m_pos -= next >> 2; next &= 3; } else { - m_pos = op; - m_pos -= (t & 8) << 11; - t = (t & 7) + (3 - 1); - if (unlikely(t == 2)) { - size_t offset; - const unsigned char *ip_last = ip; + NEED_IP(2); + next = get_unaligned_le16(ip); + if (((next & 0xfffc) == 0xfffc) && + ((t & 0xf8) == 0x18) && + likely(bitstream_version)) { + NEED_IP(3); + t &= 7; + t |= ip[2] << 3; + t += MIN_ZERO_RUN_LENGTH; + NEED_OP(t); + memset(op, 0, t); + op += t; + next &= 3; + ip += 3; + goto match_next; + } else { + m_pos = op; + m_pos -= (t & 8) << 11; + t = (t & 7) + (3 - 1); + if (unlikely(t == 2)) { + size_t offset; + const unsigned char *ip_last = ip; - while (unlikely(*ip == 0)) { - ip++; - NEED_IP(1); - } - offset = ip - ip_last; - if (unlikely(offset > MAX_255_COUNT)) - return LZO_E_ERROR; + while (unlikely(*ip == 0)) { + ip++; + NEED_IP(1); + } + offset = ip - ip_last; + if (unlikely(offset > MAX_255_COUNT)) + return LZO_E_ERROR; - offset = (offset << 8) - offset; - t += offset + 7 + *ip++; - NEED_IP(2); + offset = (offset << 8) - offset; + t += offset + 7 + *ip++; + NEED_IP(2); + next = get_unaligned_le16(ip); + } + ip += 2; + m_pos -= next >> 2; + next &= 3; + if (m_pos == op) + goto eof_found; + m_pos -= 0x4000; } - next = get_unaligned_le16(ip); - ip += 2; - m_pos -= next >> 2; - next &= 3; - if (m_pos == op) - goto eof_found; - m_pos -= 0x4000; } TEST_LB(m_pos); #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) --- a/lib/lzo/lzodefs.h~lib-lzo-implement-run-length-encoding +++ a/lib/lzo/lzodefs.h @@ -13,6 +13,12 @@ */ +/* Version + * 0: original lzo version + * 1: lzo with support for RLE + */ +#define LZO_VERSION 1 + #define COPY4(dst, src) \ put_unaligned(get_unaligned((const u32 *)(src)), (u32 *)(dst)) #if defined(CONFIG_X86_64) || defined(CONFIG_ARM64) @@ -28,6 +34,7 @@ #elif defined(CONFIG_X86_64) || defined(CONFIG_ARM64) #define LZO_USE_CTZ64 1 #define LZO_USE_CTZ32 1 +#define LZO_FAST_64BIT_MEMORY_ACCESS #elif defined(CONFIG_X86) || defined(CONFIG_PPC) #define LZO_USE_CTZ32 1 #elif defined(CONFIG_ARM) && (__LINUX_ARM_ARCH__ >= 5) @@ -37,7 +44,7 @@ #define M1_MAX_OFFSET 0x0400 #define M2_MAX_OFFSET 0x0800 #define M3_MAX_OFFSET 0x4000 -#define M4_MAX_OFFSET 0xbfff +#define M4_MAX_OFFSET 0xbffe #define M1_MIN_LEN 2 #define M1_MAX_LEN 2 @@ -53,6 +60,9 @@ #define M3_MARKER 32 #define M4_MARKER 16 +#define MIN_ZERO_RUN_LENGTH 4 +#define MAX_ZERO_RUN_LENGTH (2047 + MIN_ZERO_RUN_LENGTH) + #define lzo_dict_t unsigned short #define D_BITS 13 #define D_SIZE (1u << D_BITS) _
