Skip to contentSkip to conversation

lib/lzo: performance improvements

PATCH Superseded Dave Rodgman 5 participants 23 messages Nov 30, 2018 – Jan 7, 2019

Subsystems (4)

v4 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v4 cover lib/lzo: performance improvements

This patch series introduces performance improvements for lzo.

The previous version of this patchset is here:
https://lkml.org/lkml/2018/11/30/807

This version of the patchset fixes a maybe-used-uninitialized warning (although the previous version was still safe).

Dave

#1 lore raw

PATCH 1/8 lib/lzo: tidy-up ifdefs

Modify the ifdefs in lzodefs.h to be more consistent with normal kernel macros (e.g., change __aarch64__ to CONFIG_ARM64).

Cc Herbert Xu <[email protected]>
Cc David S. Miller <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Matt Sealey <[email protected]>
---
 lib/lzo/lzodefs.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
lib/lzo/lzodefs.h +4 -4
index 4edefd2f540c..497f9c9f03a8 100644
@@ -15,7 +15,7 @@
 
 #define COPY4(dst, src)	\
 		put_unaligned(get_unaligned((const u32 *)(src)), (u32 *)(dst))
-#if defined(__x86_64__)
+#if defined(CONFIG_X86_64)
 #define COPY8(dst, src)	\
 		put_unaligned(get_unaligned((const u64 *)(src)), (u64 *)(dst))
 #else
@@ -25,12 +25,12 @@
 
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
-#elif defined(__x86_64__)
+#elif defined(CONFIG_X86_64)
 #define LZO_USE_CTZ64	1
 #define LZO_USE_CTZ32	1
-#elif defined(__i386__) || defined(__powerpc__)
+#elif defined(CONFIG_X86) || defined(CONFIG_PPC)
 #define LZO_USE_CTZ32	1
-#elif defined(__arm__) && (__LINUX_ARM_ARCH__ >= 5)
+#elif defined(CONFIG_ARM) && (__LINUX_ARM_ARCH__ >= 5)
 #define LZO_USE_CTZ32	1
 #endif
 
Signature · 1 line
2.17.1
Acked-by on v1 1/8
#2 lore raw

Re: [PATCH 1/8] lib/lzo: tidy-up ifdefs

Quoted 50 lines · Dave Rodgman

Modify the ifdefs in lzodefs.h to be more consistent with normal kernel macros (e.g., change __aarch64__ to CONFIG_ARM64).

Cc Herbert Xu <[email protected]>
Cc David S. Miller <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Matt Sealey <[email protected]>
---
 lib/lzo/lzodefs.h | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
lib/lzo/lzodefs.h +4 -4
index 4edefd2f540c..497f9c9f03a8 100644
@@ -15,7 +15,7 @@
 
 #define COPY4(dst, src)	\
 		put_unaligned(get_unaligned((const u32 *)(src)), (u32 *)(dst))
-#if defined(__x86_64__)
+#if defined(CONFIG_X86_64)
 #define COPY8(dst, src)	\
 		put_unaligned(get_unaligned((const u64 *)(src)), (u64 *)(dst))
 #else
@@ -25,12 +25,12 @@
 
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
-#elif defined(__x86_64__)
+#elif defined(CONFIG_X86_64)
 #define LZO_USE_CTZ64	1
 #define LZO_USE_CTZ32	1
-#elif defined(__i386__) || defined(__powerpc__)
+#elif defined(CONFIG_X86) || defined(CONFIG_PPC)
 #define LZO_USE_CTZ32	1
-#elif defined(__arm__) && (__LINUX_ARM_ARCH__ >= 5)
+#elif defined(CONFIG_ARM) && (__LINUX_ARM_ARCH__ >= 5)
 #define LZO_USE_CTZ32	1
 #endif
 
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#3 lore raw

PATCH 2/8 lib/lzo: clean-up by introducing COPY16

In reply to Dave Rodgman

From: Matt Sealey <[email protected]>

Most compilers should be able to merge adjacent loads/stores of sizes which are less than but effect a multiple of a machine word size (in effect a memcpy() of a constant amount). However the semantics of the macro are that it just does the copy, the pointer increment is in the code, hence we see

    *a = *b
    a += 8
    b += 8
    *a = *b
    a += 8
    b += 8

This introduces a dependency between the two groups of statements which seems to defeat said compiler optimizers and generate some very strange sequences of addition and subtraction of address offsets (i.e. it is overcomplicated).

Since COPY8 is only ever used to copy amounts of 16 bytes (in pairs), just define COPY16 as COPY8,COPY8. We leave the definition to preserve the need to do unaligned accesses to machine-sized words per the original code intent, we just don't use it in the code proper.

COPY16 then gives us code like:

    *a = *b
    *(a+8) = *(b+8)
    a += 16
    b += 16

This seems to allow compilers to generate much better code by using base register writeback or simply positively incrementing offsets which seems to positively affect performance. It is, at least, fewer instructions to do the same job.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzo1x_compress.c        |  9 +++------
 lib/lzo/lzo1x_decompress_safe.c | 18 ++++++------------
 lib/lzo/lzodefs.h               |  3 +++
 3 files changed, 12 insertions(+), 18 deletions(-)
lib/lzo/lzo1x_compress.c +3 -6
index 236eb21167b5..82fb5571ce5e 100644
@@ -60,8 +60,7 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 				op += t;
 			} else if (t <= 16) {
 				*op++ = (t - 3);
-				COPY8(op, ii);
-				COPY8(op + 8, ii + 8);
+				COPY16(op, ii);
 				op += t;
 			} else {
 				if (t <= 18) {
@@ -76,8 +75,7 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 					*op++ = tt;
 				}
 				do {
-					COPY8(op, ii);
-					COPY8(op + 8, ii + 8);
+					COPY16(op, ii);
 					op += 16;
 					ii += 16;
 					t -= 16;
@@ -255,8 +253,7 @@ int lzo1x_1_compress(const unsigned char *in, size_t in_len,
 			*op++ = tt;
 		}
 		if (t >= 16) do {
-			COPY8(op, ii);
-			COPY8(op + 8, ii + 8);
+			COPY16(op, ii);
 			op += 16;
 			ii += 16;
 			t -= 16;
lib/lzo/lzo1x_decompress_safe.c +6 -12
index a1c387f6afba..aa95d3066b7d 100644
@@ -86,12 +86,9 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 					const unsigned char *ie = ip + t;
 					unsigned char *oe = op + t;
 					do {
-						COPY8(op, ip);
-						op += 8;
-						ip += 8;
-						COPY8(op, ip);
-						op += 8;
-						ip += 8;
+						COPY16(op, ip);
+						op += 16;
+						ip += 16;
 					} while (ip < ie);
 					ip = ie;
 					op = oe;
@@ -187,12 +184,9 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 			unsigned char *oe = op + t;
 			if (likely(HAVE_OP(t + 15))) {
 				do {
-					COPY8(op, m_pos);
-					op += 8;
-					m_pos += 8;
-					COPY8(op, m_pos);
-					op += 8;
-					m_pos += 8;
+					COPY16(op, m_pos);
+					op += 16;
+					m_pos += 16;
 				} while (op < oe);
 				op = oe;
 				if (HAVE_IP(6)) {
lib/lzo/lzodefs.h +3 -0
index 497f9c9f03a8..e1b3cf6459a9 100644
@@ -23,6 +23,9 @@
 		COPY4(dst, src); COPY4((dst) + 4, (src) + 4)
 #endif
 
+#define COPY16(dst, src) \
+	do { COPY8(dst, src); COPY8((dst) + 8, (src) + 8); } while (0)
+
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
 #elif defined(CONFIG_X86_64)
Signature · 1 line
2.17.1
#4 lore raw

Re: [PATCH 2/8] lib/lzo: clean-up by introducing COPY16

*NOT* acked by Markus Oberhumer, original author of the LZO data compression library and author of the current Linux kernel LZO implementation.

Please see my email reply to the patch series header for more info.

Quoted 142 lines · Dave Rodgman

From: Matt Sealey <[email protected]>

Most compilers should be able to merge adjacent loads/stores of sizes which are less than but effect a multiple of a machine word size (in effect a memcpy() of a constant amount). However the semantics of the macro are that it just does the copy, the pointer increment is in the code, hence we see

    *a = *b
    a += 8
    b += 8
    *a = *b
    a += 8
    b += 8

This introduces a dependency between the two groups of statements which seems to defeat said compiler optimizers and generate some very strange sequences of addition and subtraction of address offsets (i.e. it is overcomplicated).

Since COPY8 is only ever used to copy amounts of 16 bytes (in pairs), just define COPY16 as COPY8,COPY8. We leave the definition to preserve the need to do unaligned accesses to machine-sized words per the original code intent, we just don't use it in the code proper.

COPY16 then gives us code like:

    *a = *b
    *(a+8) = *(b+8)
    a += 16
    b += 16

This seems to allow compilers to generate much better code by using base register writeback or simply positively incrementing offsets which seems to positively affect performance. It is, at least, fewer instructions to do the same job.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzo1x_compress.c        |  9 +++------
 lib/lzo/lzo1x_decompress_safe.c | 18 ++++++------------
 lib/lzo/lzodefs.h               |  3 +++
 3 files changed, 12 insertions(+), 18 deletions(-)
lib/lzo/lzo1x_compress.c +3 -6
index 236eb21167b5..82fb5571ce5e 100644
@@ -60,8 +60,7 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 				op += t;
 			} else if (t <= 16) {
 				*op++ = (t - 3);
-				COPY8(op, ii);
-				COPY8(op + 8, ii + 8);
+				COPY16(op, ii);
 				op += t;
 			} else {
 				if (t <= 18) {
@@ -76,8 +75,7 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 					*op++ = tt;
 				}
 				do {
-					COPY8(op, ii);
-					COPY8(op + 8, ii + 8);
+					COPY16(op, ii);
 					op += 16;
 					ii += 16;
 					t -= 16;
@@ -255,8 +253,7 @@ int lzo1x_1_compress(const unsigned char *in, size_t in_len,
 			*op++ = tt;
 		}
 		if (t >= 16) do {
-			COPY8(op, ii);
-			COPY8(op + 8, ii + 8);
+			COPY16(op, ii);
 			op += 16;
 			ii += 16;
 			t -= 16;
lib/lzo/lzo1x_decompress_safe.c +6 -12
index a1c387f6afba..aa95d3066b7d 100644
@@ -86,12 +86,9 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 					const unsigned char *ie = ip + t;
 					unsigned char *oe = op + t;
 					do {
-						COPY8(op, ip);
-						op += 8;
-						ip += 8;
-						COPY8(op, ip);
-						op += 8;
-						ip += 8;
+						COPY16(op, ip);
+						op += 16;
+						ip += 16;
 					} while (ip < ie);
 					ip = ie;
 					op = oe;
@@ -187,12 +184,9 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 			unsigned char *oe = op + t;
 			if (likely(HAVE_OP(t + 15))) {
 				do {
-					COPY8(op, m_pos);
-					op += 8;
-					m_pos += 8;
-					COPY8(op, m_pos);
-					op += 8;
-					m_pos += 8;
+					COPY16(op, m_pos);
+					op += 16;
+					m_pos += 16;
 				} while (op < oe);
 				op = oe;
 				if (HAVE_IP(6)) {
lib/lzo/lzodefs.h +3 -0
index 497f9c9f03a8..e1b3cf6459a9 100644
@@ -23,6 +23,9 @@
 		COPY4(dst, src); COPY4((dst) + 4, (src) + 4)
 #endif
 
+#define COPY16(dst, src) \
+	do { COPY8(dst, src); COPY8((dst) + 8, (src) + 8); } while (0)
+
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
 #elif defined(CONFIG_X86_64)
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#5 lore raw

PATCH 3/8 lib/lzo: enable 64-bit CTZ on Arm

In reply to Dave Rodgman

From: Matt Sealey <[email protected]>

ARMv6 Thumb state introduced an RBIT instruction which, combined with CLZ as present in ARMv5, introduces an extremely fast path for counting trailing zeroes.

Enable the use of the GCC builtin for this on ARMv6+ with CONFIG_THUMB2_KERNEL to ensure we get the 'new' instruction usage.

We do not bother enabling LZO_USE_CTZ64 support for ARMv5 as the builtin code path does the same thing as the LZO_USE_CTZ32 code, only with more register pressure.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
lib/lzo/lzodefs.h +6 -1
index e1b3cf6459a9..c0193f726db0 100644
@@ -33,9 +33,14 @@
 #define LZO_USE_CTZ32	1
 #elif defined(CONFIG_X86) || defined(CONFIG_PPC)
 #define LZO_USE_CTZ32	1
-#elif defined(CONFIG_ARM) && (__LINUX_ARM_ARCH__ >= 5)
+#elif defined(CONFIG_ARM)
+#if (__LINUX_ARM_ARCH__ >= 5)
 #define LZO_USE_CTZ32	1
 #endif
+#if (__LINUX_ARM_ARCH__ >= 6) && defined(CONFIG_THUMB2_KERNEL)
+#define LZO_USE_CTZ64	1
+#endif
+#endif
 
 #define M1_MAX_OFFSET	0x0400
 #define M2_MAX_OFFSET	0x0800
Signature · 1 line
2.17.1
#6 lore raw

Re: [PATCH 3/8] lib/lzo: enable 64-bit CTZ on Arm

I think that LZO_USE_CTZ64 should only be defined for ARM64 and not for 32-bit ARM.

Quoted 52 lines · Dave Rodgman

From: Matt Sealey <[email protected]>

ARMv6 Thumb state introduced an RBIT instruction which, combined with CLZ as present in ARMv5, introduces an extremely fast path for counting trailing zeroes.

Enable the use of the GCC builtin for this on ARMv6+ with CONFIG_THUMB2_KERNEL to ensure we get the 'new' instruction usage.

We do not bother enabling LZO_USE_CTZ64 support for ARMv5 as the builtin code path does the same thing as the LZO_USE_CTZ32 code, only with more register pressure.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
lib/lzo/lzodefs.h +6 -1
index e1b3cf6459a9..c0193f726db0 100644
@@ -33,9 +33,14 @@
 #define LZO_USE_CTZ32	1
 #elif defined(CONFIG_X86) || defined(CONFIG_PPC)
 #define LZO_USE_CTZ32	1
-#elif defined(CONFIG_ARM) && (__LINUX_ARM_ARCH__ >= 5)
+#elif defined(CONFIG_ARM)
+#if (__LINUX_ARM_ARCH__ >= 5)
 #define LZO_USE_CTZ32	1
 #endif
+#if (__LINUX_ARM_ARCH__ >= 6) && defined(CONFIG_THUMB2_KERNEL)
+#define LZO_USE_CTZ64	1
+#endif
+#endif
 
 #define M1_MAX_OFFSET	0x0400
 #define M2_MAX_OFFSET	0x0800
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#7 lore raw

PATCH 4/8 lib/lzo: 64-bit CTZ on arm64

In reply to Dave Rodgman

From: Matt Sealey <[email protected]>

LZO leaves some performance on the table by not realising that arm64 can optimize count-trailing-zeros bit operations.

Add CONFIG_ARM64 to the checked definitions alongside CONFIG_X86_64 to enable the use of rbit/clz instructions on full 64-bit quantities.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/lzo/lzodefs.h +1 -1
index c0193f726db0..c8965dc181df 100644
@@ -28,7 +28,7 @@
 
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
-#elif defined(CONFIG_X86_64)
+#elif defined(CONFIG_X86_64) || defined(CONFIG_ARM64)
 #define LZO_USE_CTZ64	1
 #define LZO_USE_CTZ32	1
 #elif defined(CONFIG_X86) || defined(CONFIG_PPC)
Signature · 1 line
2.17.1
Acked-by on v1 4/8
#8 lore raw

Re: [PATCH 4/8] lib/lzo: 64-bit CTZ on arm64

Quoted 40 lines · Dave Rodgman

From: Matt Sealey <[email protected]>

LZO leaves some performance on the table by not realising that arm64 can optimize count-trailing-zeros bit operations.

Add CONFIG_ARM64 to the checked definitions alongside CONFIG_X86_64 to enable the use of rbit/clz instructions on full 64-bit quantities.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/lzo/lzodefs.h +1 -1
index c0193f726db0..c8965dc181df 100644
@@ -28,7 +28,7 @@
 
 #if defined(__BIG_ENDIAN) && defined(__LITTLE_ENDIAN)
 #error "conflicting endian definitions"
-#elif defined(CONFIG_X86_64)
+#elif defined(CONFIG_X86_64) || defined(CONFIG_ARM64)
 #define LZO_USE_CTZ64	1
 #define LZO_USE_CTZ32	1
 #elif defined(CONFIG_X86) || defined(CONFIG_PPC)
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#9 lore raw

PATCH 5/8 lib/lzo: fast 8-byte copy on arm64

In reply to Dave Rodgman

From: Matt Sealey <[email protected]>

Enable faster 8-byte copies on arm64.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/lzo/lzodefs.h +1 -1
index c8965dc181df..06fa83a38e0a 100644
@@ -15,7 +15,7 @@
 
 #define COPY4(dst, src)	\
 		put_unaligned(get_unaligned((const u32 *)(src)), (u32 *)(dst))
-#if defined(CONFIG_X86_64)
+#if defined(CONFIG_X86_64) || defined(CONFIG_ARM64)
 #define COPY8(dst, src)	\
 		put_unaligned(get_unaligned((const u64 *)(src)), (u64 *)(dst))
 #else
Signature · 1 line
2.17.1
Acked-by on v1 5/8
#10 lore raw

Re: [PATCH 5/8] lib/lzo: fast 8-byte copy on arm64

Quoted 36 lines · Dave Rodgman

From: Matt Sealey <[email protected]>

Enable faster 8-byte copies on arm64.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 lib/lzo/lzodefs.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/lzo/lzodefs.h +1 -1
index c8965dc181df..06fa83a38e0a 100644
@@ -15,7 +15,7 @@
 
 #define COPY4(dst, src)	\
 		put_unaligned(get_unaligned((const u32 *)(src)), (u32 *)(dst))
-#if defined(CONFIG_X86_64)
+#if defined(CONFIG_X86_64) || defined(CONFIG_ARM64)
 #define COPY8(dst, src)	\
 		put_unaligned(get_unaligned((const u64 *)(src)), (u64 *)(dst))
 #else
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#11 lore raw

PATCH 6/8 lib/lzo: implement run-length encoding

In reply to Dave Rodgman

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).

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Matt Sealey <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 Documentation/lzo.txt           |  35 ++++++++---
 include/linux/lzo.h             |   2 +-
 lib/lzo/lzo1x_compress.c        | 100 ++++++++++++++++++++++++++++----
 lib/lzo/lzo1x_decompress_safe.c |  75 ++++++++++++++++--------
 lib/lzo/lzodefs.h               |  12 +++-
 5 files changed, 181 insertions(+), 43 deletions(-)
Documentation/lzo.txt +28 -7
index 6fa6a93d0949..306c60344ca7 100644
@@ -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 <[email protected]> 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 <[email protected]> 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.
include/linux/lzo.h +1 -1
index 2ae27cb89927..547a86c71e1b 100644
@@ -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,
lib/lzo/lzo1x_compress.c +89 -11

162 lines not on this page · raw message

lib/lzo/lzo1x_decompress_safe.c +52 -23

95 lines not on this page · raw message

lib/lzo/lzodefs.h +11 -1
index 06fa83a38e0a..682359058b3c 100644
@@ -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)
@@ -31,6 +37,7 @@
 #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)
@@ -45,7 +52,7 @@
 #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
@@ -61,6 +68,9 @@
 #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)
Signature · 1 line
2.17.1
#12 lore raw

Re: [PATCH 6/8] lib/lzo: implement run-length encoding

*NOT* acked by Markus Oberhumer, original author of the LZO data compression library and author of the current Linux kernel LZO implementation.

Please see my email reply to the patch series header for more info.

Quoted 432 lines · Dave Rodgman

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).

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Matt Sealey <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 Documentation/lzo.txt           |  35 ++++++++---
 include/linux/lzo.h             |   2 +-
 lib/lzo/lzo1x_compress.c        | 100 ++++++++++++++++++++++++++++----
 lib/lzo/lzo1x_decompress_safe.c |  75 ++++++++++++++++--------
 lib/lzo/lzodefs.h               |  12 +++-
 5 files changed, 181 insertions(+), 43 deletions(-)
Documentation/lzo.txt +28 -7
index 6fa6a93d0949..306c60344ca7 100644
@@ -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 <[email protected]> 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 <[email protected]> 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.
include/linux/lzo.h +1 -1
index 2ae27cb89927..547a86c71e1b 100644
@@ -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,
lib/lzo/lzo1x_compress.c +89 -11
index 82fb5571ce5e..03834b99a5a1 100644
@@ -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 *in, size_t in_len,
 	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) {
@@ -86,6 +144,17 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 			}
 		}
 
+		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)
@@ -168,7 +237,6 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 
 		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));
@@ -205,6 +273,9 @@ lzo1x_1_do_compress(const unsigned char *in, size_t in_len,
 			*op++ = (m_off << 2);
 			*op++ = (m_off >> 6);
 		}
+		*state_offset = -2;
+finished_writing_instruction:
+		ii = ip;
 		goto next;
 	}
 	*out_len = op - out;
@@ -219,6 +290,12 @@ int lzo1x_1_compress(const unsigned char *in, size_t in_len,
 	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);
@@ -227,7 +304,8 @@ int lzo1x_1_compress(const unsigned char *in, size_t in_len,
 			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;
@@ -240,7 +318,7 @@ int lzo1x_1_compress(const unsigned char *in, size_t in_len,
 		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 {
lib/lzo/lzo1x_decompress_safe.c +52 -23
index aa95d3066b7d..b8f88d5ea3ff 100644
@@ -46,11 +46,23 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 	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) {
@@ -151,32 +163,49 @@ int lzo1x_decompress_safe(const unsigned char *in, size_t in_len,
 			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)
lib/lzo/lzodefs.h +11 -1
index 06fa83a38e0a..682359058b3c 100644
@@ -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)
@@ -31,6 +37,7 @@
 #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)
@@ -45,7 +52,7 @@
 #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
@@ -61,6 +68,9 @@
 #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)
Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#13 lore raw

PATCH 7/8 lib/lzo: separate lzo-rle from lzo

In reply to Dave Rodgman

To prevent any issues with persistent data, separate lzo-rle from lzo so that it is treated as a separate algorithm, and lzo is still available.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Matt Sealey <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
---
 Documentation/lzo.txt      |  12 ++-
 crypto/Makefile            |   2 +-
 crypto/lzo-rle.c           | 175 +++++++++++++++++++++++++++++++++++++
 crypto/tcrypt.c            |   4 +-
 drivers/block/zram/zcomp.c |   1 +
 include/linux/lzo.h        |   4 +
 lib/lzo/lzo1x_compress.c   |  42 +++++++--
 lib/lzo/lzodefs.h          |   3 +-
 8 files changed, 226 insertions(+), 17 deletions(-)
 create mode 100644 crypto/lzo-rle.c
Documentation/lzo.txt +8 -4
index 306c60344ca7..f79934225d8d 100644
@@ -88,6 +88,10 @@ 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).
 
+For maximum compatibility, both versions are available under different names
+(lzo and lzo-rle). Differences in the encoding are noted in this document with
+e.g.: version 1 only.
+
 Byte sequences
 ==============
 
@@ -99,8 +103,8 @@ Byte sequences
                 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.
+                gives the bitstream version (version 1 only). 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 ]
@@ -154,8 +158,8 @@ 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 version 1 only, 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.
 
crypto/Makefile +1 -1
index 5e789dc2d4fd..23491b70e601 100644
@@ -127,7 +127,7 @@ obj-$(CONFIG_CRYPTO_CRC32C) += crc32c_generic.o
 obj-$(CONFIG_CRYPTO_CRC32) += crc32_generic.o
 obj-$(CONFIG_CRYPTO_CRCT10DIF) += crct10dif_common.o crct10dif_generic.o
 obj-$(CONFIG_CRYPTO_AUTHENC) += authenc.o authencesn.o
-obj-$(CONFIG_CRYPTO_LZO) += lzo.o
+obj-$(CONFIG_CRYPTO_LZO) += lzo.o lzo-rle.o
 obj-$(CONFIG_CRYPTO_LZ4) += lz4.o
 obj-$(CONFIG_CRYPTO_LZ4HC) += lz4hc.o
 obj-$(CONFIG_CRYPTO_842) += 842.o
crypto/lzo-rle.c new file +175 -0

175 lines not on this page · raw message

crypto/tcrypt.c +2 -2
index 0590a9204562..c1c56c9771cf 100644
@@ -76,8 +76,8 @@ static char *check[] = {
 	"cast6", "arc4", "michael_mic", "deflate", "crc32c", "tea", "xtea",
 	"khazad", "wp512", "wp384", "wp256", "tnepres", "xeta",  "fcrypt",
 	"camellia", "seed", "salsa20", "rmd128", "rmd160", "rmd256", "rmd320",
-	"lzo", "cts", "sha3-224", "sha3-256", "sha3-384", "sha3-512",
-	"streebog256", "streebog512",
+	"lzo", "lzo-rle", "cts", "sha3-224", "sha3-256", "sha3-384",
+	"sha3-512", "streebog256", "streebog512",
 	NULL
 };
 
drivers/block/zram/zcomp.c +1 -0
index 4ed0a78fdc09..4d9a38890965 100644
@@ -20,6 +20,7 @@
 
 static const char * const backends[] = {
 	"lzo",
+	"lzo-rle",
 #if IS_ENABLED(CONFIG_CRYPTO_LZ4)
 	"lz4",
 #endif
include/linux/lzo.h +4 -0
index 547a86c71e1b..e95c7d1092b2 100644
@@ -24,6 +24,10 @@
 int lzo1x_1_compress(const unsigned char *src, size_t src_len,
 		     unsigned char *dst, size_t *dst_len, void *wrkmem);
 
+/* This requires 'wrkmem' of size LZO1X_1_MEM_COMPRESS */
+int lzorle1x_1_compress(const unsigned char *src, size_t src_len,
+		     unsigned char *dst, size_t *dst_len, void *wrkmem);
+
 /* safe decompression with overrun testing */
 int lzo1x_decompress_safe(const unsigned char *src, size_t src_len,
 			  unsigned char *dst, size_t *dst_len);
lib/lzo/lzo1x_compress.c +33 -9

84 lines not on this page · raw message

lib/lzo/lzodefs.h +2 -1
index 682359058b3c..c0657441a35d 100644
@@ -52,7 +52,8 @@
 #define M1_MAX_OFFSET	0x0400
 #define M2_MAX_OFFSET	0x0800
 #define M3_MAX_OFFSET	0x4000
-#define M4_MAX_OFFSET	0xbffe
+#define M4_MAX_OFFSET_V0	0xbfff
+#define M4_MAX_OFFSET_V1	0xbffe
 
 #define M1_MIN_LEN	2
 #define M1_MAX_LEN	2
Signature · 1 line
2.17.1
maintainer Acked-by on v1 7/8
#14 lore raw

Re: [PATCH 7/8] lib/lzo: separate lzo-rle from lzo

Quoted 19 lines · Dave Rodgman

To prevent any issues with persistent data, separate lzo-rle from lzo so that it is treated as a separate algorithm, and lzo is still available.

Cc David S. Miller <[email protected]>
Cc Greg Kroah-Hartman <[email protected]>
Cc Herbert Xu <[email protected]>
Cc Markus F.X.J. Oberhumer <[email protected]>
Cc Matt Sealey <[email protected]>
Cc Minchan Kim <[email protected]>
Cc Nitin Gupta <[email protected]>
Cc Richard Purdie <[email protected]>
Cc Sergey Senozhatsky <[email protected]>
Cc Sonny Rao <[email protected]>
Quoted 11 lines
---
 Documentation/lzo.txt      |  12 ++-
 crypto/Makefile            |   2 +-
 crypto/lzo-rle.c           | 175 +++++++++++++++++++++++++++++++++++++
 crypto/tcrypt.c            |   4 +-
 drivers/block/zram/zcomp.c |   1 +
 include/linux/lzo.h        |   4 +
 lib/lzo/lzo1x_compress.c   |  42 +++++++--
 lib/lzo/lzodefs.h          |   3 +-
 8 files changed, 226 insertions(+), 17 deletions(-)
 create mode 100644 crypto/lzo-rle.c

Thanks,

Signature · 3 lines
Email: Herbert Xu <[email protected]>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt
#15 lore raw

PATCH 8/8 zram: default to lzo-rle instead of lzo

In reply to Dave Rodgman

lzo-rle gives higher performance and similar compression ratios to lzo.

Testing with 80 browser tabs showed a 27% reduction in total time spent (de)compressing data during swapping.

---
 drivers/block/zram/zram_drv.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/block/zram/zram_drv.c +1 -1
index 58b025c5c83e..b8dc0976e385 100644
@@ -41,7 +41,7 @@ static DEFINE_IDR(zram_index_idr);
 static DEFINE_MUTEX(zram_index_mutex);
 
 static int zram_major;
-static const char *default_compressor = "lzo";
+static const char *default_compressor = "lzo-rle";
 
 /* Module params (documentation at end) */
 static unsigned int num_devices = 1;
Signature · 1 line
2.17.1
#16 lore raw
In reply to Dave Rodgman
Quoted 8 lines · Dave Rodgman

This patch series introduces performance improvements for lzo.

The previous version of this patchset is here:
https://lkml.org/lkml/2018/11/30/807

This version of the patchset fixes a maybe-used-uninitialized warning (although the previous version was still safe).

Hi Dave,

Notices this warning today:

lib/lzo/lzo1x_compress.c: In function ‘lzo1x_1_do_compress’:
lib/lzo/lzo1x_compress.c:239:14: warning: ‘m_pos’ may be used uninitialized in this function [-Wmaybe-uninitialized]
239 | m_off = ip - m_pos;

Care to take a look? (could be false positive)

	-ss
#17 lore raw
Quoted 9 lines · Sergey Senozhatsky

Hi Dave,

Notices this warning today:

lib/lzo/lzo1x_compress.c: In function ‘lzo1x_1_do_compress’:
lib/lzo/lzo1x_compress.c:239:14: warning: ‘m_pos’ may be used uninitialized in this function [-Wmaybe-uninitialized]
239 | m_off = ip - m_pos;

Care to take a look? (could be false positive)

Hi Sergey,

This warning (it's a false positive only seen with gcc) was fixed by v4 of this patchset. I've just tested the latest linux-next/master (next-20181205) and it doesn't have this issue.

cheers

Dave

#18 lore raw
Quoted 7 lines · Dave Rodgman
Quoted 5 lines

lib/lzo/lzo1x_compress.c: In function ‘lzo1x_1_do_compress’:
lib/lzo/lzo1x_compress.c:239:14: warning: ‘m_pos’ may be used uninitialized in this function [-Wmaybe-uninitialized]
239 | m_off = ip - m_pos;

Care to take a look? (could be false positive)

Hi Sergey,

Hi,

Quoted 2 lines

This warning (it's a false positive only seen with gcc) was fixed by v4 of this patchset.

OK, cool.

Quoted 2 lines

I've just tested the latest linux-next/master (next-20181205) and it doesn't have this issue.

Indeed.

	-ss
#19 lore raw
In reply to Dave Rodgman
Quoted 9 lines · Dave Rodgman

This patch series introduces performance improvements for lzo.

The previous version of this patchset is here:
https://lkml.org/lkml/2018/11/30/807

This version of the patchset fixes a maybe-used-uninitialized warning (although the previous version was still safe).

Dave

Hi Dave,

as indicated in my previous mail please split your series into three distinct pull requests.

Request 1 - ARM64 improvements; acked by me

[PATCH 1/8] lib/lzo: tidy-up ifdefs
[PATCH 3/8] lib/lzo: enable 64-bit CTZ on Arm
[PATCH 4/8] lib/lzo: 64-bit CTZ on arm64
[PATCH 5/8] lib/lzo: fast 8-byte copy on arm64

are simple arch patches that give a nice speedup on ARM64 and should get merged ASAP.

Request 2 - add COPY16; *NOT* acked by me

[PATCH 2/8] lib/lzo: clean-up by introducing COPY16

is still not correct because of possible overlapping copies. I'll address this on the weekend.

Request 3 - add lzo-rle; *NOT* acked by me

[PATCH 6/8] lib/lzo: implement run-length encoding [PATCH 7/8] lib/lzo: separate lzo-rle from lzo [PATCH 8/8] zram: default to lzo-rle instead of lzo

This can *NOT* be applied in the current implementation.

It (1) silently changes the compressed data format, (2) crashes on MIPS, and (3) makes compression and decompression on typical data 10% slower on X86_64 with our internal benchmarks, and (4) has to be carefully checked for buffer overflows.

I understand that we want some optimizations for data with many zeros like in the typical ZRAM use case, but the implementation will clearly need some more work. I'll also have a look at the weekend - eg I have a nice idea how to deal with (1).

As a final comment, I question the quality your benchmarks - combining arch-related ARM64 improvements and algorithmic changes into one benchmark comparision is just unprofessional marketing.

Cheers,
Markus

Signature · 1 line
Markus Oberhumer, <[email protected]>, http://www.oberhumer.com/
#20 lore raw

Markus,

Quoted 6 lines

Request 2 - add COPY16; *NOT* acked by me

[PATCH 2/8] lib/lzo: clean-up by introducing COPY16

is still not correct because of possible overlapping copies. I'll address this on the weekend.

Can you give a syndrome as to why

{
	COPY8(op, ip);
	COPY8(op+8,ip+8);
	ip+=16;
	op+=16;
}

or

{ 
	COPY8(op, ip);
	ip+=8;
	op+=8;
	COPY8(op, ip);
	ip+=8;
	op+=8;
}

vs.

#define COPY16(dst,src) COPY8(dst,src); COPY8(dst+8,src+8)

{
	COPY16(op, ip);
	ip+=16;
	op+=16;
}

.. causes "overlapping copies"?

COPY8 was only ever used in pairs as above and the second method broke compiler optimizers since it adds an artificial barrier between the two groups. The only difference was that decompress and compress had the pointer increments spread out. If we need to fix that then that's a good reason, but your reasoning continues to elude me.

I can refactor the patch to align the second method with the first and make compress and decompress get the same codegen, which is functionally identical to the COPY16 patch, but that would seem to in your opinion be the whole problem..

I'll see what you've got after the weekend ;D

Ta
Matt Sealey

#21 lore raw
In reply to Markus F.X.J. Oberhumer

Hi Markus,

On 06/12/2018 3:47 pm, Markus F.X.J. Oberhumer wrote:> Request 3 - add lzo-rle; *NOT* acked by me

Quoted 6 lines

[PATCH 6/8] lib/lzo: implement run-length encoding [PATCH 7/8] lib/lzo: separate lzo-rle from lzo [PATCH 8/8] zram: default to lzo-rle instead of lzo

It (1) silently changes the compressed data format

I'm not sure this is relevant: as a separate algorithm, there's no reason to retain the same format (although backwards compatibility can help with migration). If you know of a way to improve the compatibility aspect though, that would be great!

Quoted 1 line

(2) crashes on MIPS,

Please could you provide more detail? I tested on x86-32, x86-64, arm, arm64 and big-endian MIPS64, but if there is an issue I missed I'd like to address it.

Quoted 2 lines

and (3) makes compression and decompression on typical data 10% slower on X86_64 with our internal benchmarks,

It is of course data-dependent. In my testing, as I mentioned previously, RLE without the other patches does regress slightly on high-entropy data, but offers a win on low-entropy data. For the right applications (e.g., zram), this makes it overall beneficial.

Quoted 1 line

and (4) has to be carefully checked for buffer overflows.

This has been reviewed prior to sharing on LKML, and of course tested, but further review is of course welcome.

Quoted 3 lines

As a final comment, I question the quality your benchmarks - combining arch-related ARM64 improvements and algorithmic changes into one benchmark comparision is just unprofessional marketing.

I felt it was helpful to show overall performance with the complete patchset:
this is what end-users experience. However, as you can see below, I also previously shared a summary of the two main components of the patchset to try and address this sort of concern:

Quoted 10 lines
Quoted 10 lines

As a quick summary of the impact of these patches on bigger chunks of data, I've compared the performance of four different variants of lzo on two large (~40 MB) files. The numbers show round-trip throughput in MB/s:

Variant         | Low-entropy | High-entropy
Current lzo     |  242        | 157
Arm opts        |  290        | 159
RLE             |  876        | 151
Arm opts + RLE  | 1150        | 181

cheers

Dave

#22 lore raw

Hi Markus,

What are your thoughts on how we should proceed with this patchset? You raised a few concerns in December - however, I'm not sure what further changes might be needed, if any. IMO this could be merged as it stands.

Regarding compatibility concerns: patchset v4 does not modify the behaviour of existing lzo. It introduces an independent algorithm (closely based on lzo); and also introduces some Arm performance benefits for existing lzo, without modifying the behaviour. So I don't see a compatibility risk.

You mentioned a crash on MIPS - do you have any details on this please? I have not seen any crashes in my testing so I'm not able to look into this without more data.

Quoted 34 lines · Dave Rodgman

Hi Markus,

On 06/12/2018 3:47 pm, Markus F.X.J. Oberhumer wrote:> Request 3 - add lzo-rle; *NOT* acked by me

Quoted 6 lines

     [PATCH 6/8] lib/lzo: implement run-length encoding [PATCH 7/8] lib/lzo: separate lzo-rle from lzo [PATCH 8/8] zram: default to lzo-rle instead of lzo

It (1) silently changes the compressed data format

I'm not sure this is relevant: as a separate algorithm, there's no reason to retain the same format (although backwards compatibility can help with migration). If you know of a way to improve the compatibility aspect though, that would be great!

Quoted 1 line

(2) crashes on MIPS,

Please could you provide more detail? I tested on x86-32, x86-64, arm, arm64 and big-endian MIPS64, but if there is an issue I missed I'd like to address it.

Quoted 2 lines

and (3) makes compression and decompression on typical data 10% slower on X86_64 with our internal benchmarks,

It is of course data-dependent. In my testing, as I mentioned previously, RLE without the other patches does regress slightly on high-entropy data, but offers a win on low-entropy data. For the right applications (e.g., zram), this makes it overall beneficial.

Quoted 1 line

and (4) has to be carefully checked for buffer overflows.

This has been reviewed prior to sharing on LKML, and of course tested, but further review is of course welcome.

Quoted 19 lines
Quoted 3 lines

As a final comment, I question the quality your benchmarks - combining arch-related ARM64 improvements and algorithmic changes into one benchmark comparision is just unprofessional marketing.

I felt it was helpful to show overall performance with the complete patchset:
this is what end-users experience. However, as you can see below, I also previously shared a summary of the two main components of the patchset to try and address this sort of concern:

Quoted 10 lines
Quoted 10 lines

As a quick summary of the impact of these patches on bigger chunks of data, I've compared the performance of four different variants of lzo on two large (~40 MB) files. The numbers show round-trip throughput in MB/s:

Variant         | Low-entropy | High-entropy
Current lzo     |  242        | 157
Arm opts        |  290        | 159
RLE             |  876        | 151
Arm opts + RLE  | 1150        | 181

Regards

Dave