This patch updates AVX512 support for xbzrle
v6 cover This patch updates AVX512 support for xbzrle
This patch updates code of avx512 support for xbzrle_encode_buffer
function. We modified code of algorithm and check of avx512. Besides, we provide benchmark in xbzrle-bench.c for performance comparison.
ling xu (2):
Update AVX512 support for xbzrle_encode_buffer
Unit test code and benchmark code
meson.build | 16 ++ meson_options.txt | 2 + migration/ram.c | 34 ++- migration/xbzrle.c | 124 ++++++++++ migration/xbzrle.h | 4 + tests/bench/meson.build | 4 + tests/bench/xbzrle-bench.c | 465 +++++++++++++++++++++++++++++++++++++ tests/unit/test-xbzrle.c | 39 +++- 8 files changed, 680 insertions(+), 8 deletions(-) create mode 100644 tests/bench/xbzrle-bench.c
Signature · 1 line
2.25.1
PATCH v6 1/2 Update AVX512 support for xbzrle_encode_buffer
This commit updates code of avx512 support for xbzrle_encode_buffer function to accelerate xbzrle encoding speed. Runtime check of avx512 support and benchmark for this feature are added. Compared with C version of xbzrle_encode_buffer function, avx512 version can achieve 50%-70% performance improvement on benchmarking. In addition, if dirty data is randomly located in 4K page, the avx512 version can achieve almost 140% performance gain.
--- meson.build | 16 ++++++ meson_options.txt | 2 + migration/ram.c | 34 +++++++++++-- migration/xbzrle.c | 124 +++++++++++++++++++++++++++++++++++++++++++++ migration/xbzrle.h | 4 ++ 5 files changed, 177 insertions(+), 3 deletions(-)
meson.build +16 -0
@@ -2264,6 +2264,22 @@ config_host_data.set('CONFIG_AVX512F_OPT', get_option('avx512f') \ int main(int argc, char *argv[]) { return bar(argv[0]); } '''), error_message: 'AVX512F not available').allowed()) +config_host_data.set('CONFIG_AVX512BW_OPT', get_option('avx512bw') \ + .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot enable AVX512BW') \ + .require(cc.links(''' + #pragma GCC push_options + #pragma GCC target("avx512bw") + #include <cpuid.h> + #include <immintrin.h> + static int bar(void *a) { + + __m512i *x = a; + __m512i res= _mm512_abs_epi8(*x); + return res[1]; + } + int main(int argc, char *argv[]) { return bar(argv[0]); } + '''), error_message: 'AVX512BW not available').allowed()) + have_pvrdma = get_option('pvrdma') \ .require(rdma.found(), error_message: 'PVRDMA requires OpenFabrics libraries') \ .require(cc.compiles(gnu_source_prefix + '''
meson_options.txt +2 -0
@@ -104,6 +104,8 @@ option('avx2', type: 'feature', value: 'auto', description: 'AVX2 optimizations') option('avx512f', type: 'feature', value: 'disabled', description: 'AVX512F optimizations') +option('avx512bw', type: 'feature', value: 'auto', + description: 'AVX512BW optimizations') option('keyring', type: 'feature', value: 'auto', description: 'Linux keyring support')
migration/ram.c +31 -3
@@ -83,6 +83,34 @@/* 0x80 is reserved in migration.h start with 0x100 next */ #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 +int (*xbzrle_encode_buffer_func)(uint8_t *,uint8_t *,int , +uint8_t *,int ) = xbzrle_encode_buffer; +#if defined(CONFIG_AVX512BW_OPT ) +#include "qemu/cpuid.h" +static void __attribute__((constructor))init_cpu_flag (void ) +{ +unsigned max =__get_cpuid_max (0 ,NULL ); +int a, b, c, d; +if (max >=1 ) { +__cpuid (1 , a, b, c, d); +/* We must check that AVX is not just available, but usable. */ +if ((c & bit_OSXSAVE) && (c & bit_AVX) && max >=7 ) { +int bv; + __asm("xgetbv" :"=a" (bv),"=d" (d) :"c" (0 )); +__cpuid_count (7 ,0 , a, b, c, d); +/* 0xe6: +* XCR0[7:5] = 111b (OPMASK state, upper 256-bit of ZMM0-ZMM15 +* and ZMM16-ZMM31 state are enabled by OS) +* XCR0[2:1] = 11b (XMM state and YMM state are enabled by OS) +*/ +if ((bv &0xe6 ) ==0xe6 && (b & bit_AVX512BW)) { + xbzrle_encode_buffer_func = xbzrle_encode_buffer_avx512; + } + } + } +} +#endif +XBZRLECacheStats xbzrle_counters;/* struct contains XBZRLE cache and a static page
@@ -802,9 +830,9 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,memcpy (XBZRLE .current_buf, *current_data,TARGET_PAGE_SIZE );/* XBZRLE encoding (if there is no overflow) */ - encoded_len =xbzrle_encode_buffer (prev_cached_page,XBZRLE .current_buf, -TARGET_PAGE_SIZE ,XBZRLE .encoded_buf, -TARGET_PAGE_SIZE ); + encoded_len =xbzrle_encode_buffer_func (prev_cached_page,XBZRLE .current_buf, +TARGET_PAGE_SIZE ,XBZRLE .encoded_buf, +TARGET_PAGE_SIZE );/* * Update the cache contents, so that it corresponds to the data
migration/xbzrle.c +124 -0
127 lines not on this page · raw message
migration/xbzrle.h +4 -0
@@ -18,4 +18,8 @@ int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen, uint8_t *dst, int dlen);int xbzrle_decode_buffer (uint8_t *src,int slen,uint8_t *dst,int dlen); +#if defined(CONFIG_AVX512BW_OPT ) +int xbzrle_encode_buffer_avx512 (uint8_t *old_buf,uint8_t *new_buf,int slen, +uint8_t *dst,int dlen); +#endif #endif
Signature · 1 line
2.25.1
RE: [PATCH v6 1/2] Update AVX512 support for xbzrle_encode_buffer
Hi, All,
This is a "ping" email~.
It seems that my patch has been ignored. So I "ping" this patchset.
Link for the patch: https://lore.kernel.org/qemu-devel/[email protected]/
Best Regards
Ling
-----Original Message-----
From: Xu, Ling1 <[email protected]>
Sent: Friday, August 26, 2022 5:57 PM
To: [email protected]
Cc: [email protected]; [email protected]; Xu, Ling1 <[email protected]>; Zhao, Zhou <[email protected]>; Jin, Jun I <[email protected]>
Subject: [PATCH v6 1/2] Update AVX512 support for xbzrle_encode_buffer
This commit updates code of avx512 support for xbzrle_encode_buffer function to accelerate xbzrle encoding speed. Runtime check of avx512 support and benchmark for this feature are added. Compared with C version of xbzrle_encode_buffer function, avx512 version can achieve 50%-70% performance improvement on benchmarking. In addition, if dirty data is randomly located in 4K page, the avx512 version can achieve almost 140% performance gain.
--- meson.build | 16 ++++++ meson_options.txt | 2 + migration/ram.c | 34 +++++++++++-- migration/xbzrle.c | 124 +++++++++++++++++++++++++++++++++++++++++++++ migration/xbzrle.h | 4 ++ 5 files changed, 177 insertions(+), 3 deletions(-)
meson.build +17 -0
@@ -2264,6 +2264,22 @@ config_host_data.set('CONFIG_AVX512F_OPT', get_option('avx512f') \ int main(int argc, char *argv[]) { return bar(argv[0]); } '''), error_message: 'AVX512F not available').allowed()) +config_host_data.set('CONFIG_AVX512BW_OPT', get_option('avx512bw') \ + .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot +enable AVX512BW') \ + .require(cc.links(''' + #pragma GCC push_options + #pragma GCC target("avx512bw") + #include <cpuid.h> + #include <immintrin.h> + static int bar(void *a) { + + __m512i *x = a; + __m512i res= _mm512_abs_epi8(*x); + return res[1]; + } + int main(int argc, char *argv[]) { return bar(argv[0]); } '''), + error_message: 'AVX512BW not available').allowed()) + have_pvrdma = get_option('pvrdma') \ .require(rdma.found(), error_message: 'PVRDMA requires OpenFabrics libraries') \
.require(cc.compiles(gnu_source_prefix + '''
meson_options.txt +2 -0
@@ -104,6 +104,8 @@ option('avx2', type: 'feature', value: 'auto', description: 'AVX2 optimizations') option('avx512f', type: 'feature', value: 'disabled', description: 'AVX512F optimizations') +option('avx512bw', type: 'feature', value: 'auto', + description: 'AVX512BW optimizations') option('keyring', type: 'feature', value: 'auto', description: 'Linux keyring support')
migration/ram.c +26 -0
@@ -83,6 +83,34 @@/* 0x80 is reserved in migration.h start with 0x100 next */ #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 +int (*xbzrle_encode_buffer_func)(uint8_t *,uint8_t *,int , + uint8_t *, int) = xbzrle_encode_buffer; #if +defined(CONFIG_AVX512BW_OPT) #include "qemu/cpuid.h" +static void __attribute__((constructor))init_cpu_flag (void ) { +unsigned max =__get_cpuid_max (0 ,NULL ); +int a, b, c, d; +if (max >=1 ) { +__cpuid (1 , a, b, c, d); +/* We must check that AVX is not just available, but usable. */ +if ((c & bit_OSXSAVE) && (c & bit_AVX) && max >=7 ) { +int bv; + __asm("xgetbv" :"=a" (bv),"=d" (d) :"c" (0 )); +__cpuid_count (7 ,0 , a, b, c, d); +/* 0xe6: +* XCR0[7:5] = 111b (OPMASK state, upper 256-bit of ZMM0-ZMM15 +* and ZMM16-ZMM31 state are enabled by OS) +* XCR0[2:1] = 11b (XMM state and YMM state are enabled by OS) +*/ +if ((bv &0xe6 ) ==0xe6 && (b & bit_AVX512BW)) { + xbzrle_encode_buffer_func = xbzrle_encode_buffer_avx512; + } + } + } +} +#endif +XBZRLECacheStats xbzrle_counters;/* struct contains XBZRLE cache and a static page @@ -802,9 +830,9 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE);
/* XBZRLE encoding (if there is no overflow) */- encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
- TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
- TARGET_PAGE_SIZE);
+ encoded_len = xbzrle_encode_buffer_func(prev_cached_page, XBZRLE.current_buf,
+ TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
+ TARGET_PAGE_SIZE);
/*
* Update the cache contents, so that it corresponds to the data diff --git a/migration/xbzrle.c b/migration/xbzrle.c index 1ba482ded9..05366e86c0 100644
migration/xbzrle.c +123 -0
126 lines not on this page · raw message
migration/xbzrle.h +3 -0
@@ -18,4 +18,8 @@ int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen, uint8_t *dst, int dlen);int xbzrle_decode_buffer (uint8_t *src,int slen,uint8_t *dst,int dlen); +#if defined(CONFIG_AVX512BW_OPT ) +int xbzrle_encode_buffer_avx512 (uint8_t *old_buf,uint8_t *new_buf,int slen, +uint8_t *dst,int dlen);#endif #endif
--
2.25.1
RE: [PATCH v6 1/2] Update AVX512 support for xbzrle_encode_buffer
In reply to ling xuHi, All,
This is a "ping" email~.
It seems that the newest version of our patch has been ignored. So I "ping" this patchset again. All comments and suggestions have been revised and updated in this V6 version patch, and link for the patch is below:
https://lore.kernel.org/qemu-devel/[email protected]/
Looking forward to hearing from you!
Best Regards
Ling
-----Original Message-----
From: Xu, Ling1 <[email protected]>
Sent: Friday, August 26, 2022 5:57 PM
To: [email protected]
Cc: [email protected]; [email protected]; Xu, Ling1 <[email protected]>; Zhao, Zhou <[email protected]>; Jin, Jun I <[email protected]>
Subject: [PATCH v6 1/2] Update AVX512 support for xbzrle_encode_buffer
This commit updates code of avx512 support for xbzrle_encode_buffer function to accelerate xbzrle encoding speed. Runtime check of avx512 support and benchmark for this feature are added. Compared with C version of xbzrle_encode_buffer function, avx512 version can achieve 50%-70% performance improvement on benchmarking. In addition, if dirty data is randomly located in 4K page, the avx512 version can achieve almost 140% performance gain.
--- meson.build | 16 ++++++ meson_options.txt | 2 + migration/ram.c | 34 +++++++++++-- migration/xbzrle.c | 124 +++++++++++++++++++++++++++++++++++++++++++++ migration/xbzrle.h | 4 ++ 5 files changed, 177 insertions(+), 3 deletions(-)
meson.build +17 -0
@@ -2264,6 +2264,22 @@ config_host_data.set('CONFIG_AVX512F_OPT', get_option('avx512f') \ int main(int argc, char *argv[]) { return bar(argv[0]); } '''), error_message: 'AVX512F not available').allowed()) +config_host_data.set('CONFIG_AVX512BW_OPT', get_option('avx512bw') \ + .require(have_cpuid_h, error_message: 'cpuid.h not available, cannot +enable AVX512BW') \ + .require(cc.links(''' + #pragma GCC push_options + #pragma GCC target("avx512bw") + #include <cpuid.h> + #include <immintrin.h> + static int bar(void *a) { + + __m512i *x = a; + __m512i res= _mm512_abs_epi8(*x); + return res[1]; + } + int main(int argc, char *argv[]) { return bar(argv[0]); } '''), + error_message: 'AVX512BW not available').allowed()) + have_pvrdma = get_option('pvrdma') \ .require(rdma.found(), error_message: 'PVRDMA requires OpenFabrics libraries') \
.require(cc.compiles(gnu_source_prefix + '''
meson_options.txt +2 -0
@@ -104,6 +104,8 @@ option('avx2', type: 'feature', value: 'auto', description: 'AVX2 optimizations') option('avx512f', type: 'feature', value: 'disabled', description: 'AVX512F optimizations') +option('avx512bw', type: 'feature', value: 'auto', + description: 'AVX512BW optimizations') option('keyring', type: 'feature', value: 'auto', description: 'Linux keyring support')
migration/ram.c +26 -0
@@ -83,6 +83,34 @@/* 0x80 is reserved in migration.h start with 0x100 next */ #define RAM_SAVE_FLAG_COMPRESS_PAGE 0x100 +int (*xbzrle_encode_buffer_func)(uint8_t *,uint8_t *,int , + uint8_t *, int) = xbzrle_encode_buffer; #if +defined(CONFIG_AVX512BW_OPT) #include "qemu/cpuid.h" +static void __attribute__((constructor))init_cpu_flag (void ) { +unsigned max =__get_cpuid_max (0 ,NULL ); +int a, b, c, d; +if (max >=1 ) { +__cpuid (1 , a, b, c, d); +/* We must check that AVX is not just available, but usable. */ +if ((c & bit_OSXSAVE) && (c & bit_AVX) && max >=7 ) { +int bv; + __asm("xgetbv" :"=a" (bv),"=d" (d) :"c" (0 )); +__cpuid_count (7 ,0 , a, b, c, d); +/* 0xe6: +* XCR0[7:5] = 111b (OPMASK state, upper 256-bit of ZMM0-ZMM15 +* and ZMM16-ZMM31 state are enabled by OS) +* XCR0[2:1] = 11b (XMM state and YMM state are enabled by OS) +*/ +if ((bv &0xe6 ) ==0xe6 && (b & bit_AVX512BW)) { + xbzrle_encode_buffer_func = xbzrle_encode_buffer_avx512; + } + } + } +} +#endif +XBZRLECacheStats xbzrle_counters;/* struct contains XBZRLE cache and a static page @@ -802,9 +830,9 @@ static int save_xbzrle_page(RAMState *rs, uint8_t **current_data,
memcpy(XBZRLE.current_buf, *current_data, TARGET_PAGE_SIZE);
/* XBZRLE encoding (if there is no overflow) */- encoded_len = xbzrle_encode_buffer(prev_cached_page, XBZRLE.current_buf,
- TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
- TARGET_PAGE_SIZE);
+ encoded_len = xbzrle_encode_buffer_func(prev_cached_page, XBZRLE.current_buf,
+ TARGET_PAGE_SIZE, XBZRLE.encoded_buf,
+ TARGET_PAGE_SIZE);
/*
* Update the cache contents, so that it corresponds to the data diff --git a/migration/xbzrle.c b/migration/xbzrle.c index 1ba482ded9..05366e86c0 100644
migration/xbzrle.c +123 -0
126 lines not on this page · raw message
migration/xbzrle.h +3 -0
@@ -18,4 +18,8 @@ int xbzrle_encode_buffer(uint8_t *old_buf, uint8_t *new_buf, int slen, uint8_t *dst, int dlen);int xbzrle_decode_buffer (uint8_t *src,int slen,uint8_t *dst,int dlen); +#if defined(CONFIG_AVX512BW_OPT ) +int xbzrle_encode_buffer_avx512 (uint8_t *old_buf,uint8_t *new_buf,int slen, +uint8_t *dst,int dlen);#endif #endif
--
2.25.1
Re: [PATCH v6 1/2] Update AVX512 support for xbzrle_encode_buffer
In reply to ling xuQuoted 11 lines · ling xu
This commit updates code of avx512 support for
xbzrle_encode_bufferfunction to accelerate xbzrle encoding speed. Runtime check of avx512 support and benchmark for this feature are added. Compared with C version ofxbzrle_encode_bufferfunction, avx512 version can achieve 50%-70% performance improvement on benchmarking. In addition, if dirty data is randomly located in 4K page, the avx512 version can achieve almost 140% performance gain.Signed-off-by ling xu <[email protected]>Co-authored-by Zhou Zhao <[email protected]>Co-authored-by Jun Jin <[email protected]>
queued.
PATCH v6 2/2 Unit test code and benchmark code
In reply to ling xuUnit test code is in test-xbzrle.c, and benchmark code is in xbzrle-bench.c for performance benchmarking.
--- tests/bench/meson.build | 4 + tests/bench/xbzrle-bench.c | 465 +++++++++++++++++++++++++++++++++++++ tests/unit/test-xbzrle.c | 39 +++- 3 files changed, 503 insertions(+), 5 deletions(-) create mode 100644 tests/bench/xbzrle-bench.c
tests/bench/meson.build +4 -0
@@ -3,6 +3,10 @@ qht_bench = executable('qht-bench', sources: 'qht-bench.c', dependencies: [qemuutil]) +xbzrle_bench = executable('xbzrle-bench', + sources: 'xbzrle-bench.c', + dependencies: [qemuutil,migration]) + executable('atomic_add-bench', sources: files('atomic_add-bench.c'), dependencies: [qemuutil],
tests/bench/xbzrle-bench.c new file +465 -0
465 lines not on this page · raw message
tests/unit/test-xbzrle.c +34 -5
@@ -16,6 +16,35 @@#define XBZRLE_PAGE_SIZE 4096 +int (*xbzrle_encode_buffer_func)(uint8_t *,uint8_t *,int , +uint8_t *,int ) = xbzrle_encode_buffer; +#if defined(CONFIG_AVX512BW_OPT ) +#include "qemu/cpuid.h" +static void __attribute__((constructor))init_cpu_flag (void ) +{ +unsigned max =__get_cpuid_max (0 ,NULL ); +int a, b, c, d; +if (max >=1 ) { +__cpuid (1 , a, b, c, d); +/* We must check that AVX is not just available, but usable. */ +if ((c & bit_OSXSAVE) && (c & bit_AVX) && max >=7 ) { +int bv; + __asm("xgetbv" :"=a" (bv),"=d" (d) :"c" (0 )); +__cpuid_count (7 ,0 , a, b, c, d); +/* 0xe6: +* XCR0[7:5] = 111b (OPMASK state, upper 256-bit of ZMM0-ZMM15 +* and ZMM16-ZMM31 state are enabled by OS) +* XCR0[2:1] = 11b (XMM state and YMM state are enabled by OS) +*/ +if ((bv &0xe6 ) ==0xe6 && (b & bit_AVX512BW)) { + xbzrle_encode_buffer_func = xbzrle_encode_buffer_avx512; + } + } + } +return ; +} +#endif +static void test_uleb (void ) {uint32_t i, val;
@@ -54,7 +83,7 @@ static void test_encode_decode_zero(void) buffer[1000 + diff_len +5 ] =105 ;/* encode zero page */ - dlen =xbzrle_encode_buffer (buffer, buffer,XBZRLE_PAGE_SIZE , compressed, + dlen =xbzrle_encode_buffer_func (buffer, buffer,XBZRLE_PAGE_SIZE , compressed,XBZRLE_PAGE_SIZE );g_assert (dlen ==0 );
@@ -78,7 +107,7 @@ static void test_encode_decode_unchanged(void) test[1000 + diff_len +5 ] =109 ;/* test unchanged buffer */ - dlen =xbzrle_encode_buffer (test, test,XBZRLE_PAGE_SIZE , compressed, + dlen =xbzrle_encode_buffer_func (test, test,XBZRLE_PAGE_SIZE , compressed,XBZRLE_PAGE_SIZE );g_assert (dlen ==0 );
@@ -96,7 +125,7 @@ static void test_encode_decode_1_byte(void) test[XBZRLE_PAGE_SIZE -1 ] =1 ; - dlen =xbzrle_encode_buffer (buffer, test,XBZRLE_PAGE_SIZE , compressed, + dlen =xbzrle_encode_buffer_func (buffer, test,XBZRLE_PAGE_SIZE , compressed,XBZRLE_PAGE_SIZE );g_assert (dlen == (uleb128_encode_small (&buf[0 ],4095 ) +2 ));
@@ -121,7 +150,7 @@ static void test_encode_decode_overflow(void) }/* encode overflow */ - rc =xbzrle_encode_buffer (buffer, test,XBZRLE_PAGE_SIZE , compressed, + rc =xbzrle_encode_buffer_func (buffer, test,XBZRLE_PAGE_SIZE , compressed,XBZRLE_PAGE_SIZE );g_assert (rc ==-1 );
@@ -152,7 +181,7 @@ static void encode_decode_range(void) test[1000 + diff_len +5 ] =109 ;/* test encode/decode */ - dlen =xbzrle_encode_buffer (test, buffer,XBZRLE_PAGE_SIZE , compressed, + dlen =xbzrle_encode_buffer_func (test, buffer,XBZRLE_PAGE_SIZE , compressed,XBZRLE_PAGE_SIZE ); rc =xbzrle_decode_buffer (compressed, dlen, test,XBZRLE_PAGE_SIZE );
Signature · 1 line
2.25.1
Re: [PATCH v6 2/2] Unit test code and benchmark code
Quoted 6 lines · ling xu
Unit test code is in
test-xbzrle.c, and benchmark code is inxbzrle-bench.cfor performance benchmarking.Signed-off-by ling xu <[email protected]>Co-authored-by Zhou Zhao <[email protected]>Co-authored-by Jun Jin <[email protected]>
queued.