riscv: implement accelerated crc using zbc
Subsystems (2)
cover riscv: implement accelerated crc using zbc
The RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patchset contains two new implementations:
- one in lib/hash/rte_crc_riscv64.h that uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one in lib/net/net_crc_zbc.c that uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implement rte_crc16_ccitt_zbc_handler and rte_crc32_eth_zbc_handler
My approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/
These implementations are behind a new flag, RTE_RISCV_ZBC. Due to use of bitmanip compiler intrinsics, a modern version of GCC (14+) or Clang (18+) is required to compile with this flag enabled.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:
|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |
These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
Daniel Gregory (5):
config/riscv: add flag for using Zbc extension
hash: implement crc using riscv carryless multiply
net: implement crc using riscv carryless multiply examples/l3fwd: use accelerated crc on riscv
ipfrag: use accelerated crc on riscv
MAINTAINERS | 2 + app/test/test_crc.c | 9 ++ app/test/test_hash.c | 7 ++ config/riscv/meson.build | 7 ++ examples/l3fwd/l3fwd_em.c | 2 +- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 +++++++++++++++ lib/hash/rte_hash_crc.c | 12 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/net/meson.build | 4 + lib/net/net_crc.h | 11 ++ lib/net/net_crc_zbc.c | 202 +++++++++++++++++++++++++++++++++ lib/net/rte_net_crc.c | 35 ++++++ lib/net/rte_net_crc.h | 2 + 15 files changed, 389 insertions(+), 6 deletions(-) create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Signature · 1 line
2.39.2
PATCH 1/5 config/riscv: add flag for using Zbc extension
The RISC-V Zbc extension adds carry-less multiply instructions we can use to implement more efficient CRC hashing algorithms.
--- config/riscv/meson.build | 7 +++++++ 1 file changed, 7 insertions(+)
config/riscv/meson.build +7 -0
@@ -26,6 +26,13 @@ flags_common = [ # read from /proc/device-tree/cpus/timebase-frequency. This property is # guaranteed on Linux, as riscv time_init() requires it. ['RTE_RISCV_TIME_FREQ', 0], + + # Use RISC-V Carry-less multiplication extension (Zbc) for hardware + # implementations of CRC-32C (lib/hash/rte_crc_riscv64.h), CRC-32 and CRC-16 + # (lib/net/net_crc_zbc.c). Requires intrinsics available in GCC 14.1.0+ and + # Clang 18.1.0+ + # Make sure to add '_zbc' to your target's -march below + ['RTE_RISCV_ZBC', false], ] ## SoC-specific options.
Signature · 1 line
2.39.2
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 6 lines · On Tue, 18 Jun 2024 18:41:29 +0100 Daniel Gregory…
config/riscv/meson.build +0
-0@@ -26,6 +26,13 @@ flags_common = [ # read from /proc/device-tree/cpus/timebase-frequency. This property is
config/riscv/meson.build flags_common = [ · line 36
Quoted 10 lines
config/riscv/meson.build +7
-0# guaranteed on Linux, as riscv time_init() requires it. ['RTE_RISCV_TIME_FREQ', 0], + + # Use RISC-V Carry-less multiplication extension (Zbc) for hardware + # implementations of CRC-32C (lib/hash/rte_crc_riscv64.h), CRC-32 and CRC-16 + # (lib/net/net_crc_zbc.c). Requires intrinsics available in GCC 14.1.0+ and + # Clang 18.1.0+ + # Make sure to add '_zbc' to your target's -march below + ['RTE_RISCV_ZBC', false], ]
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
Instead, detect at compile or runtime
RE: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 1 line
From: Stephen Hemminger [mailto:[email protected]]
1/5] config/riscv: add flag for using Zbc extension
Quoted 23 lines
config/riscv/meson.build flags_common = [ · line 31
Quoted 11 lines · On Tue, 18 Jun 2024 18:41:29 +0100 Daniel Gregory…
config/riscv/meson.build +3
-0@@ -26,6 +26,13 @@ flags_common = [ # read from /proc/device-tree/cpus/timebase-frequency. This property is # guaranteed on Linux, as riscv time_init() requires it. ['RTE_RISCV_TIME_FREQ', 0], + + # Use RISC-V Carry-less multiplication extension (Zbc) for hardware + # implementations of CRC-32C (lib/hash/rte_crc_riscv64.h), CRC-32 andCRC-16
config/riscv/meson.build flags_common = [ · line 32
Quoted 1 line
config/riscv/meson.build +1
-0+ # (lib/net/net_crc_zbc.c). Requires intrinsics available in GCC 14.1.0+and
config/riscv/meson.build flags_common = [ · line 36
Quoted 4 lines
config/riscv/meson.build +3
-0+ # Clang 18.1.0+ + # Make sure to add '_zbc' to your target's -march below + ['RTE_RISCV_ZBC', false], ]
Quoted 4 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
Instead, detect at compile or runtime
Build time detection is not possible for cross builds.
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 7 lines · On Wed, 19 Jun 2024 09:08:14 +0200 Morten Brørup <[email protected]>…
Quoted 5 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
Instead, detect at compile or runtime
Build time detection is not possible for cross builds.
I was thinking some mechanism like the ARM configs.
RE: [PATCH 1/5] config/riscv: add flag for using Zbc extension
In reply to Morten BrørupQuoted 25 lines · Morten Brørup
Quoted 1 line
From: Stephen Hemminger [mailto:[email protected]]
1/5] config/riscv: add flag for using Zbc extension
Quoted 23 lines
config/riscv/meson.build flags_common = [ · line 31
Quoted 11 lines · On Tue, 18 Jun 2024 18:41:29 +0100 Daniel Gregory…
config/riscv/meson.build +3
-0@@ -26,6 +26,13 @@ flags_common = [ # read from /proc/device-tree/cpus/timebase-frequency. This property is # guaranteed on Linux, as riscv time_init() requires it. ['RTE_RISCV_TIME_FREQ', 0], + + # Use RISC-V Carry-less multiplication extension (Zbc) for hardware + # implementations of CRC-32C (lib/hash/rte_crc_riscv64.h), CRC-32 andCRC-16
config/riscv/meson.build flags_common = [ · line 32
Quoted 1 line
config/riscv/meson.build +1
-0+ # (lib/net/net_crc_zbc.c). Requires intrinsics available in GCC 14.1.0+and
Quoted 4 lines
config/riscv/meson.build +3
-0+ # Clang 18.1.0+ + # Make sure to add '_zbc' to your target's -march below + ['RTE_RISCV_ZBC', false], ]
Quoted 7 lines
Quoted 4 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
Instead, detect at compile or runtime
Build time detection is not possible for cross builds.
How about build time detection based on the target's configured instruction set (either specified by cross-file or passed in through -Dinstruction_set)? We could have a map from extensions present in the ISA string to compile flags that should be enabled.
I suggested this whilst discussing a previous patch adding support for the Zawrs extension, but haven't heard back from Stanisław yet:
https://lore.kernel.org/dpdk-dev/20240520094854.GA3672529@ste-uk-lab-gw/
As for runtime detection, newer kernel versions have a hardware probing interface for detecting the presence of extensions, support could be added to rte_cpuflags.c?
https://docs.kernel.org/arch/riscv/hwprobe.html
In combination, distros on newer kernels could ship a version that has these optimisations baked in that falls back to a generic implementation when the extension is detected to not be present, and systems without the latest GCC/Clang can still compile by specifying a target ISA that doesn't include "_zbc".
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 21 lines · Daniel Gregory
Quoted 19 lines · Morten Brørup
Quoted 1 line
From: Stephen Hemminger [mailto:[email protected]]
1/5] config/riscv: add flag for using Zbc extension
Quoted 17 lines
config/riscv/meson.build flags_common = [ · line 31
Quoted 11 lines · On Tue, 18 Jun 2024 18:41:29 +0100 Daniel Gregory…
config/riscv/meson.build +3
-0@@ -26,6 +26,13 @@ flags_common = [ # read from /proc/device-tree/cpus/timebase-frequency. This property is # guaranteed on Linux, as riscv time_init() requires it. ['RTE_RISCV_TIME_FREQ', 0], + + # Use RISC-V Carry-less multiplication extension (Zbc) for hardware + # implementations of CRC-32C (lib/hash/rte_crc_riscv64.h), CRC-32 andCRC-16
Quoted 1 line
config/riscv/meson.build +1
-0+ # (lib/net/net_crc_zbc.c). Requires intrinsics available in GCC 14.1.0+
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
and
config/riscv/meson.build flags_common = [ · line 36
Quoted 4 lines
config/riscv/meson.build +3
-0+ # Clang 18.1.0+ + # Make sure to add '_zbc' to your target's -march below + ['RTE_RISCV_ZBC', false], ]Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
That is a problem with RISC-V in general. Since all features are "extensions" and there is no limit (up to a point) on the permutation of those, we cannot statically build the code for all extensions. Fortunately instructions tend to resolve to nops if an instruction is not present but that still increases the code size for no benefit on platforms without a given extension.
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines
Instead, detect at compile or runtime
Build time detection is not possible for cross builds.
Quoted 8 lines
How about build time detection based on the target's configured instruction set (either specified by cross-file or passed in through -Dinstruction_set)? We could have a map from extensions present in the ISA string to compile flags that should be enabled.
I suggested this whilst discussing a previous patch adding support for the Zawrs extension, but haven't heard back from Stanisław yet:
https://lore.kernel.org/dpdk-dev/20240520094854.GA3672529@ste-uk-lab-gw/
I think we already have 1 case of a cross compile config:
https://git.dpdk.org/dpdk/tree/config/riscv/riscv64_sifive_u740_linux_gcc.
This could serve as a stop gap before runtime detection is sorted out. I would prefer the static option to rather list all the hardware platforms explicitly. This way we will support existing platforms, not some RISC-V vendor plans. Maybe at some point the extension mess gets fixed in the arch.
Quoted 11 lines
As for runtime detection, newer kernel versions have a hardware probing interface for detecting the presence of extensions, support could be added to
rte_cpuflags.c?
https://docs.kernel.org/arch/riscv/hwprobe.htmlIn combination, distros on newer kernels could ship a version that has these optimisations baked in that falls back to a generic implementation when the extension is detected to not be present, and systems without the latest GCC/Clang can still compile by specifying a target ISA that doesn't include "
_zbc".
hwprobe sounds like a good idea, although the key name for extensions (RISCV_HWPROBE_KEY_IMA_EXT_0) suggests that there will be more (it's 64bit and we already have 46 bits taken). That I wonder what options we have to keep the performance characteristics of the code. We either need to live-patch the code (which is problematic for userspace processes) or resort to some .so or a driver-like model. Neither option sounds very appealing.
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 9 lines · On Mon, 7 Oct 2024 10:14:22 +0200 Stanisław Kardach…
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
That is a problem with RISC-V in general. Since all features are "extensions" and there is no limit (up to a point) on the permutation of those, we cannot statically build the code for all extensions. Fortunately instructions tend to resolve to nops if an instruction is not present but that still increases the code size for no benefit on platforms without a given extension.
X86 already has the cpu feature flag infrastructure, why not use similar mechanism on RiscV?
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 14 lines · Stephen Hemminger
Quoted 9 lines · On Mon, 7 Oct 2024 10:14:22 +0200 Stanisław Kardach…
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
That is a problem with RISC-V in general. Since all features are "extensions" and there is no limit (up to a point) on the permutation of those, we cannot statically build the code for all extensions. Fortunately instructions tend to resolve to nops if an instruction is not present but that still increases the code size for no benefit on platforms without a given extension.
Quoted 2 lines
X86 already has the cpu feature flag infrastructure, why not use similar mechanism on RiscV?
We can and some further patches I've seen on the list implemented that. However if that has to be applied in basic intrinsics like rte_prefetch() which means we'd have to pay a conditional or an indirect call. It'd be best to test it on a real dataplane platform to which I don't have access.
Re: [PATCH 1/5] config/riscv: add flag for using Zbc extension
Quoted 16 lines · On Tue, 8 Oct 2024 07:52:42 +0200 Stanisław Kardach…
Quoted 14 lines · Stephen Hemminger
Quoted 9 lines · On Mon, 7 Oct 2024 10:14:22 +0200 Stanisław Kardach…
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Please do not add more config options via compile flags. It makes it impossible for distros to ship one version.
That is a problem with RISC-V in general. Since all features are "extensions" and there is no limit (up to a point) on the permutation of those, we cannot statically build the code for all extensions. Fortunately instructions tend to resolve to nops if an instruction is not present but that still increases the code size for no benefit on platforms without a given extension.
Quoted 7 lines
Quoted 2 lines
X86 already has the cpu feature flag infrastructure, why not use similar mechanism on RiscV?
We can and some further patches I've seen on the list implemented that. However if that has to be applied in basic intrinsics like
rte_prefetch()which means we'd have to pay a conditional or an indirect call. It'd be best to test it on a real dataplane platform to which I don't have access.
That makes sense, is there some config file per cpu type like Arm?
PATCH 2/5 hash: implement crc using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions from RISC-V's Zbc extension, implement a Barrett reduction that calculates CRC-32C checksums.
Based on the approach described by Intel's whitepaper on "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction", which is also described here
(https://web.archive.org/web/20240111232520/https://mary.rs/lab/crc32/)
--- MAINTAINERS | 1 + app/test/test_hash.c | 7 +++ lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++++++++++++++++++++++++++++++++ lib/hash/rte_hash_crc.c | 12 ++++- lib/hash/rte_hash_crc.h | 6 ++- 6 files changed, 114 insertions(+), 2 deletions(-) create mode 100644 lib/hash/rte_crc_riscv64.h
MAINTAINERS +1 -0
@@ -318,6 +318,7 @@ M: Stanislaw Kardach <[email protected]> F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ +F: lib/hash/rte_crc_riscv64.h Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_hash.c +7 -0
@@ -205,6 +205,13 @@ test_crc32_hash_alg_equiv(void)printf ("Failed checking CRC32_SW against CRC32_ARM64\n" );break ; } + +/* Check against 8-byte-operand RISCV64 CRC32 if available */ +rte_hash_crc_set_alg (CRC32_RISCV64 ); +if (hash_val !=rte_hash_crc (data64, data_len, init_val)) { +printf ("Failed checking CRC32_SW against CRC32_RISC64\n" ); +break ; + } }/* Resetting to best available algorithm */
lib/hash/meson.build +1 -0
@@ -12,6 +12,7 @@ headers = files( indirect_headers += files( 'rte_crc_arm64.h', 'rte_crc_generic.h', + 'rte_crc_riscv64.h', 'rte_crc_sw.h', 'rte_crc_x86.h', 'rte_thash_x86_gfni.h',
lib/hash/rte_crc_riscv64.h new file +89 -0
89 lines not on this page · raw message
lib/hash/rte_hash_crc.c +11 -1
@@ -15,7 +15,7 @@ RTE_LOG_REGISTER_SUFFIX(hash_crc_logtype, crc, INFO);uint8_t rte_hash_crc32_alg =CRC32_SW ;/** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -24,6 +24,7 @@ uint8_t rte_hash_crc32_alg = CRC32_SW;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISCV64 Zbc extension if available * */ void
@@ -52,6 +53,13 @@ rte_hash_crc_set_alg(uint8_t alg) rte_hash_crc32_alg =CRC32_ARM64 ;#endif +#if defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_ZBC ) +if (!(alg &CRC32_RISCV64 )) +HASH_CRC_LOG (WARNING , +"Unsupported CRC32 algorithm requested using CRC32_RISCV64" ); + rte_hash_crc32_alg =CRC32_RISCV64 ; +#endif +if (rte_hash_crc32_alg ==CRC32_SW )HASH_CRC_LOG (WARNING ,"Unsupported CRC32 algorithm requested using CRC32_SW" );
@@ -64,6 +72,8 @@ RTE_INIT(rte_hash_crc_init_alg)rte_hash_crc_set_alg (CRC32_SSE42_x64);#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)rte_hash_crc_set_alg (CRC32_ARM64 ); +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_ZBC ) +rte_hash_crc_set_alg (CRC32_RISCV64 );#else rte_hash_crc_set_alg (CRC32_SW );#endif
lib/hash/rte_hash_crc.h +5 -1
@@ -28,6 +28,7 @@ extern "C" {#define CRC32_x64 (1U << 2)#define CRC32_SSE42_x64 (CRC32_x64|CRC32_SSE42)#define CRC32_ARM64 (1U << 3) +#define CRC32_RISCV64 (1U << 4)extern uint8_t rte_hash_crc32_alg;
@@ -35,12 +36,14 @@ extern uint8_t rte_hash_crc32_alg;#include "rte_crc_arm64.h" #elif defined(RTE_ARCH_X86 )#include "rte_crc_x86.h" +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_ZBC ) +#include "rte_crc_riscv64.h" #else #include "rte_crc_generic.h" #endif /** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -49,6 +52,7 @@ extern uint8_t rte_hash_crc32_alg;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISC-V Carry-less multiply if available (default rv64gc_zbc) */ void rte_hash_crc_set_alg (uint8_t alg);
Signature · 1 line
2.39.2
PATCH 3/5 net: implement crc using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions (clmul) from RISC-V's Zbc extension, implement CRC-32 and CRC-16 calculations on buffers.
Based on the approach described in Intel's whitepaper on "Fast CRC Computation for Generic Polynomails Using PCLMULQDQ Instructions", we perfom repeated folds-by-1 whilst the buffer is still big enough, then perform Barrett's reductions on the rest.
Add a case to the crc_autotest suite that tests this implementation.
This implementation is enabled by setting the RTE_RISCV_ZBC flag (see config/riscv/meson.build).
--- MAINTAINERS | 1 + app/test/test_crc.c | 9 ++ lib/net/meson.build | 4 + lib/net/net_crc.h | 11 +++ lib/net/net_crc_zbc.c | 202 ++++++++++++++++++++++++++++++++++++++++++ lib/net/rte_net_crc.c | 35 ++++++++ lib/net/rte_net_crc.h | 2 + 7 files changed, 264 insertions(+) create mode 100644 lib/net/net_crc_zbc.c
MAINTAINERS +1 -0
@@ -319,6 +319,7 @@ F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ F: lib/hash/rte_crc_riscv64.h +F: lib/net/net_crc_zbc.c Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_crc.c +9 -0
@@ -168,6 +168,15 @@ test_crc(void)return ret; } +/* set CRC riscv mode */ +rte_net_crc_set_alg (RTE_NET_CRC_ZBC ); + + ret =test_crc_calc (); +if (ret <0 ) { +printf ("test crc (riscv64 zbc clmul): failed (%d)\n" , ret); +return ret; + } +return 0 ; }
lib/net/meson.build +4 -0
@@ -125,4 +125,8 @@ elif (dpdk_conf.has('RTE_ARCH_ARM64') and cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '') sources += files('net_crc_neon.c') cflags += ['-DCC_ARM64_NEON_PMULL_SUPPORT'] +elif (dpdk_conf.has('RTE_ARCH_RISCV') and dpdk_conf.has('RTE_RISCV_ZBC') and + dpdk_conf.get('RTE_RISCV_ZBC')) + sources += files('net_crc_zbc.c') + cflags += ['-DCC_RISCV64_ZBC_CLMUL_SUPPORT'] endif
lib/net/net_crc.h +11 -0
@@ -42,4 +42,15 @@ rte_crc16_ccitt_neon_handler(const uint8_t *data, uint32_t data_len);uint32_t rte_crc32_eth_neon_handler (const uint8_t *data,uint32_t data_len); +/* RISCV64 Zbc */ +void +rte_net_crc_zbc_init (void ); + +uint32_t +rte_crc16_ccitt_zbc_handler (const uint8_t *data,uint32_t data_len); + +uint32_t +rte_crc32_eth_zbc_handler (const uint8_t *data,uint32_t data_len); + +#endif /* _NET_CRC_H_ */
lib/net/net_crc_zbc.c new file +202 -0
202 lines not on this page · raw message
lib/net/rte_net_crc.c +35 -0
@@ -67,6 +67,12 @@ static const rte_net_crc_handler handlers_neon[] = { [RTE_NET_CRC32_ETH] = rte_crc32_eth_neon_handler, };#endif +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +static const rte_net_crc_handler handlers_zbc[] = { + [RTE_NET_CRC16_CCITT ] = rte_crc16_ccitt_zbc_handler, + [RTE_NET_CRC32_ETH ] = rte_crc32_eth_zbc_handler, +}; +#endif static uint16_t max_simd_bitwidth;
@@ -244,6 +250,26 @@ neon_pmull_init(void)#endif } +/* ZBC/CLMUL handling */ + +static const rte_net_crc_handler * +zbc_clmul_get_handlers (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +return handlers_zbc; +#endif +NET_LOG (INFO ,"Requirements not met, can't use Zbc" ); +return NULL ; +} + +static void +zbc_clmul_init (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +rte_net_crc_zbc_init (); +#endif +} +/* Default handling */ static uint32_t
@@ -260,6 +286,9 @@ rte_crc16_ccitt_default_handler(const uint8_t *data, uint32_t data_len)if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers =neon_pmull_get_handlers (); +if (handlers !=NULL ) +return handlers[RTE_NET_CRC16_CCITT ](data, data_len); + handlers =zbc_clmul_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers = handlers_scalar;
@@ -282,6 +311,8 @@ rte_crc32_eth_default_handler(const uint8_t *data, uint32_t data_len) handlers =neon_pmull_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC32_ETH ](data, data_len); + handlers =zbc_clmul_get_handlers (); +return handlers[RTE_NET_CRC32_ETH ](data, data_len); handlers = handlers_scalar;return handlers[RTE_NET_CRC32_ETH ](data, data_len); }
@@ -306,6 +337,9 @@ rte_net_crc_set_alg(enum rte_net_crc_alg alg)break ;/* for x86, always break here */ case RTE_NET_CRC_NEON : handlers =neon_pmull_get_handlers (); +break ; +case RTE_NET_CRC_ZBC : + handlers =zbc_clmul_get_handlers ();/* fall-through */ case RTE_NET_CRC_SCALAR :/* fall-through */
@@ -338,4 +372,5 @@ RTE_INIT(rte_net_crc_init)sse42_pclmulqdq_init ();avx512_vpclmulqdq_init ();neon_pmull_init (); +zbc_clmul_init (); }
lib/net/rte_net_crc.h +2 -0
@@ -24,6 +24,7 @@ enum rte_net_crc_alg {RTE_NET_CRC_SSE42 ,RTE_NET_CRC_NEON ,RTE_NET_CRC_AVX512 , + RTE_NET_CRC_ZBC, };/**
@@ -37,6 +38,7 @@ enum rte_net_crc_alg {* - RTE_NET_CRC_SSE42 (Use 64-bit SSE4.2 intrinsic) * - RTE_NET_CRC_NEON (Use ARM Neon intrinsic) * - RTE_NET_CRC_AVX512 (Use 512-bit AVX intrinsic) +* - RTE_NET_CRC_ZBC (Use RISC-V Zbc extension) */ void rte_net_crc_set_alg (enum rte_net_crc_alg alg);
Signature · 1 line
2.39.2
PATCH 4/5 examples/l3fwd: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd/l3fwd_em.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd/l3fwd_em.c +1 -1
@@ -29,7 +29,7 @@#include "l3fwd_event.h" #include "em_route_parse.c" -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_ZBC )#define EM_HASH_CRC 1#endif
Signature · 1 line
2.39.2
PATCH 5/5 ipfrag: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/ip_frag/ip_frag_internal.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
lib/ip_frag/ip_frag_internal.c +3 -3
@@ -45,14 +45,14 @@ ipv4_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *)&key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (key->id, v);#else v =rte_jhash_3words (p[0 ], p[1 ], key->id,PRIME_VALUE ); -#endif /* RTE_ARCH_X86 */ +#endif /* RTE_ARCH_X86 || RTE_ARCH_ARM64 || RTE_RISCV_ZBC */ *v1 = v; *v2 = (v <<7 ) + (v >>14 );
@@ -66,7 +66,7 @@ ipv6_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *) &key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (p[2 ], v);
Signature · 1 line
2.39.2
v2 cover riscv: implement accelerated crc using zbc
In reply to Daniel GregoryThe RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in lib/hash/rte_crc_riscv64.h that uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one in lib/net/net_crc_zbc.c that uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implement rte_crc16_ccitt_zbc_handler and rte_crc32_eth_zbc_handler
My approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/
Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:
|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |
These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the hash_functions_autotest benchmark in dpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:
| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in the net/ implementation that would cause segfaults on small unaligned buffers
- refactor net/ implemementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hash
Daniel Gregory (9):
config/riscv: detect presence of Zbc extension
hash: implement crc using riscv carryless multiply
net: implement crc using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated crc on riscv
ipfrag: use accelerated crc on riscv examples/l3fwd-power: use accelerated crc on riscv hash/cuckoo: use accelerated crc on riscv
member: use accelerated crc on riscv
MAINTAINERS | 2 + app/test/test_crc.c | 9 + app/test/test_hash.c | 7 + config/riscv/meson.build | 44 +++- config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/rte_member.h | 2 +- lib/net/meson.build | 4 + lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 191 ++++++++++++++++++ lib/net/rte_net_crc.c | 40 ++++ lib/net/rte_net_crc.h | 2 + 22 files changed, 529 insertions(+), 41 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Signature · 1 line
2.39.2
PATCH v2 1/9 config/riscv: detect presence of Zbc extension
The RISC-V Zbc extension adds carry-less multiply instructions we can use to implement more efficient CRC hashing algorithms.
The RISC-V C api defines architecture extension test macros
https://github.com/riscv-non-isa/riscv-c-api-doc/blob/main/riscv-c-api.md#architecture-extension-test-macros
These let us detect whether the Zbc extension is supported on the compiler and -march we're building with. The C api also defines Zbc intrinsics we can use rather than inline assembly on newer versions of GCC (14.1.0+) and Clang (18.1.0+).
The Linux kernel exposes a RISC-V hardware probing syscall for getting information about the system at run-time including which extensions are available. We detect whether this interface is present by looking for the <asm/hwprobe.h> header, as it's only present in newer kernels (v6.4+). Furthermore, support for detecting certain extensions, including Zbc, wasn't present until versions after this, so we need to check the constants this header exports.
The kernel exposes bitmasks for each extension supported by the probing interface, rather than the bit index that is set if that extensions is present, so modify the existing cpu flag HWCAP table entries to line up with this. The values returned by the interface are 64-bits long, so grow the hwcap registers array to be able to hold them.
If the Zbc extension and intrinsics are both present and we can detect the Zbc extension at runtime, we define a flag, RTE_RISCV_FEATURE_ZBC.
--- config/riscv/meson.build | 41 ++++++++++ lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++++++++++++++-------- 3 files changed, 123 insertions(+), 32 deletions(-)
config/riscv/meson.build +41 -0
@@ -119,6 +119,47 @@ foreach flag: arch_config['machine_args'] endif endforeach +# check if we can do buildtime detection of extensions supported by the target +riscv_extension_macros = false +if (cc.get_define('__riscv_arch_test', args: machine_args) == '1') + message('Detected architecture extension test macros') + riscv_extension_macros = true +else + warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible') +endif + +# check if we can use hwprobe interface for runtime extension detection +riscv_hwprobe = false +if (cc.check_header('asm/hwprobe.h', args: machine_args)) + message('Detected hwprobe interface, enabling runtime detection of supported extensions') + machine_args += ['-DRTE_RISCV_FEATURE_HWPROBE'] + riscv_hwprobe = true +else + warning('Hwprobe interface not available (present in Linux v6.4+), instruction extensions won\'t be enabled') +endif + +# detect extensions +# RISC-V Carry-less multiplication extension (Zbc) for hardware implementations +# of CRC-32C (lib/hash/rte_crc_riscv64.h) and CRC-32/16 (lib/net/net_crc_zbc.c). +# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+ +if (riscv_extension_macros and riscv_hwprobe and + (cc.get_define('__riscv_zbc', args: machine_args) != '')) + if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0')) + or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0'))) + # determine whether we can detect Zbc extension (this wasn't possible until + # Linux kernel v6.8) + if (cc.compiles('''#include <asm/hwprobe.h> + int a = RISCV_HWPROBE_EXT_ZBC;''', args: machine_args)) + message('Compiling with the Zbc extension') + machine_args += ['-DRTE_RISCV_FEATURE_ZBC'] + else + warning('Detected Zbc extension but cannot use because runtime detection doesn\'t support it (support present in Linux kernel v6.8+)') + endif + else + warning('Detected Zbc extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)') + endif +endif + # apply flags foreach flag: dpdk_flags if flag.length() > 0
lib/eal/riscv/include/rte_cpuflags.h +2 -0
@@ -42,6 +42,8 @@ enum rte_cpu_flag_t {RTE_CPUFLAG_RISCV_ISA_X ,/* Non-standard extension present */ RTE_CPUFLAG_RISCV_ISA_Y ,/* Reserved */ RTE_CPUFLAG_RISCV_ISA_Z ,/* Reserved */ + + RTE_CPUFLAG_RISCV_EXT_ZBC, /* Carry-less multiplication */ };#include "generic/rte_cpuflags.h"
lib/eal/riscv/rte_cpuflags.c +80 -32
153 lines not on this page · raw message
Signature · 1 line
2.39.2
PATCH v2 2/9 hash: implement crc using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions from RISC-V's Zbc extension, implement a Barrett reduction that calculates CRC-32C checksums.
Based on the approach described by Intel's whitepaper on "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction", which is also described here
(https://web.archive.org/web/20240111232520/https://mary.rs/lab/crc32/)
Add a case to the autotest_hash unit test.
--- MAINTAINERS | 1 + app/test/test_hash.c | 7 +++ lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++++++++++++++++++++++++++++++++ lib/hash/rte_hash_crc.c | 13 +++++- lib/hash/rte_hash_crc.h | 6 ++- 6 files changed, 115 insertions(+), 2 deletions(-) create mode 100644 lib/hash/rte_crc_riscv64.h
MAINTAINERS +1 -0
@@ -318,6 +318,7 @@ M: Stanislaw Kardach <[email protected]> F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ +F: lib/hash/rte_crc_riscv64.h Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_hash.c +7 -0
@@ -205,6 +205,13 @@ test_crc32_hash_alg_equiv(void)printf ("Failed checking CRC32_SW against CRC32_ARM64\n" );break ; } + +/* Check against 8-byte-operand RISCV64 CRC32 if available */ +rte_hash_crc_set_alg (CRC32_RISCV64 ); +if (hash_val !=rte_hash_crc (data64, data_len, init_val)) { +printf ("Failed checking CRC32_SW against CRC32_RISC64\n" ); +break ; + } }/* Resetting to best available algorithm */
lib/hash/meson.build +1 -0
@@ -12,6 +12,7 @@ headers = files( indirect_headers += files( 'rte_crc_arm64.h', 'rte_crc_generic.h', + 'rte_crc_riscv64.h', 'rte_crc_sw.h', 'rte_crc_x86.h', 'rte_thash_x86_gfni.h',
lib/hash/rte_crc_riscv64.h new file +89 -0
89 lines not on this page · raw message
lib/hash/rte_hash_crc.c +12 -1
@@ -15,7 +15,7 @@ RTE_LOG_REGISTER_SUFFIX(hash_crc_logtype, crc, INFO);uint8_t rte_hash_crc32_alg =CRC32_SW ;/** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -24,6 +24,7 @@ uint8_t rte_hash_crc32_alg = CRC32_SW;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISCV64 Zbc extension if available * */ void
@@ -52,6 +53,14 @@ rte_hash_crc_set_alg(uint8_t alg) rte_hash_crc32_alg =CRC32_ARM64 ;#endif +#if defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +if (!(alg &CRC32_RISCV64 )) +HASH_CRC_LOG (WARNING , +"Unsupported CRC32 algorithm requested using CRC32_RISCV64" ); +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + rte_hash_crc32_alg =CRC32_RISCV64 ; +#endif +if (rte_hash_crc32_alg ==CRC32_SW )HASH_CRC_LOG (WARNING ,"Unsupported CRC32 algorithm requested using CRC32_SW" );
@@ -64,6 +73,8 @@ RTE_INIT(rte_hash_crc_init_alg)rte_hash_crc_set_alg (CRC32_SSE42_x64);#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)rte_hash_crc_set_alg (CRC32_ARM64 ); +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +rte_hash_crc_set_alg (CRC32_RISCV64 );#else rte_hash_crc_set_alg (CRC32_SW );#endif
lib/hash/rte_hash_crc.h +5 -1
@@ -28,6 +28,7 @@ extern "C" {#define CRC32_x64 (1U << 2)#define CRC32_SSE42_x64 (CRC32_x64|CRC32_SSE42)#define CRC32_ARM64 (1U << 3) +#define CRC32_RISCV64 (1U << 4)extern uint8_t rte_hash_crc32_alg;
@@ -35,12 +36,14 @@ extern uint8_t rte_hash_crc32_alg;#include "rte_crc_arm64.h" #elif defined(RTE_ARCH_X86 )#include "rte_crc_x86.h" +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +#include "rte_crc_riscv64.h" #else #include "rte_crc_generic.h" #endif /** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -49,6 +52,7 @@ extern uint8_t rte_hash_crc32_alg;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISC-V Carry-less multiply if available (default rv64gc_zbc) */ void rte_hash_crc_set_alg (uint8_t alg);
Signature · 1 line
2.39.2
PATCH v2 3/9 net: implement crc using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions (clmul) from RISC-V's Zbc extension, implement CRC-32 and CRC-16 calculations on buffers.
Based on the approach described in Intel's whitepaper on "Fast CRC Computation for Generic Polynomails Using PCLMULQDQ Instructions", we perform repeated folds-by-1 whilst the buffer is still big enough, then perform Barrett's reductions on the rest.
Add a case to the crc_autotest suite that tests this implementation.
--- MAINTAINERS | 1 + app/test/test_crc.c | 9 ++ lib/net/meson.build | 4 + lib/net/net_crc.h | 11 +++ lib/net/net_crc_zbc.c | 191 ++++++++++++++++++++++++++++++++++++++++++ lib/net/rte_net_crc.c | 40 +++++++++ lib/net/rte_net_crc.h | 2 + 7 files changed, 258 insertions(+) create mode 100644 lib/net/net_crc_zbc.c
MAINTAINERS +1 -0
@@ -319,6 +319,7 @@ F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ F: lib/hash/rte_crc_riscv64.h +F: lib/net/net_crc_zbc.c Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_crc.c +9 -0
@@ -168,6 +168,15 @@ test_crc(void)return ret; } +/* set CRC riscv mode */ +rte_net_crc_set_alg (RTE_NET_CRC_ZBC ); + + ret =test_crc_calc (); +if (ret <0 ) { +printf ("test crc (riscv64 zbc clmul): failed (%d)\n" , ret); +return ret; + } +return 0 ; }
lib/net/meson.build +4 -0
@@ -125,4 +125,8 @@ elif (dpdk_conf.has('RTE_ARCH_ARM64') and cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '') sources += files('net_crc_neon.c') cflags += ['-DCC_ARM64_NEON_PMULL_SUPPORT'] +elif (dpdk_conf.has('RTE_ARCH_RISCV') and + cc.get_define('RTE_RISCV_FEATURE_ZBC', args: machine_args) != '') + sources += files('net_crc_zbc.c') + cflags += ['-DCC_RISCV64_ZBC_CLMUL_SUPPORT'] endif
lib/net/net_crc.h +11 -0
@@ -42,4 +42,15 @@ rte_crc16_ccitt_neon_handler(const uint8_t *data, uint32_t data_len);uint32_t rte_crc32_eth_neon_handler (const uint8_t *data,uint32_t data_len); +/* RISCV64 Zbc */ +void +rte_net_crc_zbc_init (void ); + +uint32_t +rte_crc16_ccitt_zbc_handler (const uint8_t *data,uint32_t data_len); + +uint32_t +rte_crc32_eth_zbc_handler (const uint8_t *data,uint32_t data_len); + +#endif /* _NET_CRC_H_ */
lib/net/net_crc_zbc.c new file +191 -0
191 lines not on this page · raw message
lib/net/rte_net_crc.c +40 -0
@@ -67,6 +67,12 @@ static const rte_net_crc_handler handlers_neon[] = { [RTE_NET_CRC32_ETH] = rte_crc32_eth_neon_handler, };#endif +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +static const rte_net_crc_handler handlers_zbc[] = { + [RTE_NET_CRC16_CCITT ] = rte_crc16_ccitt_zbc_handler, + [RTE_NET_CRC32_ETH ] = rte_crc32_eth_zbc_handler, +}; +#endif static uint16_t max_simd_bitwidth;
@@ -244,6 +250,31 @@ neon_pmull_init(void)#endif } +/* ZBC/CLMUL handling */ + +#define ZBC_CLMUL_CPU_SUPPORTED \ + rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_EXT_ZBC) + +static const rte_net_crc_handler * +zbc_clmul_get_handlers (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) +return handlers_zbc; +#endif +NET_LOG (INFO ,"Requirements not met, can't use Zbc" ); +return NULL ; +} + +static void +zbc_clmul_init (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) +rte_net_crc_zbc_init (); +#endif +} +/* Default handling */ static uint32_t
@@ -260,6 +291,9 @@ rte_crc16_ccitt_default_handler(const uint8_t *data, uint32_t data_len)if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers =neon_pmull_get_handlers (); +if (handlers !=NULL ) +return handlers[RTE_NET_CRC16_CCITT ](data, data_len); + handlers =zbc_clmul_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers = handlers_scalar;
@@ -282,6 +316,8 @@ rte_crc32_eth_default_handler(const uint8_t *data, uint32_t data_len) handlers =neon_pmull_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC32_ETH ](data, data_len); + handlers =zbc_clmul_get_handlers (); +return handlers[RTE_NET_CRC32_ETH ](data, data_len); handlers = handlers_scalar;return handlers[RTE_NET_CRC32_ETH ](data, data_len); }
@@ -306,6 +342,9 @@ rte_net_crc_set_alg(enum rte_net_crc_alg alg)break ;/* for x86, always break here */ case RTE_NET_CRC_NEON : handlers =neon_pmull_get_handlers (); +break ; +case RTE_NET_CRC_ZBC : + handlers =zbc_clmul_get_handlers ();/* fall-through */ case RTE_NET_CRC_SCALAR :/* fall-through */
@@ -338,4 +377,5 @@ RTE_INIT(rte_net_crc_init)sse42_pclmulqdq_init ();avx512_vpclmulqdq_init ();neon_pmull_init (); +zbc_clmul_init (); }
lib/net/rte_net_crc.h +2 -0
@@ -24,6 +24,7 @@ enum rte_net_crc_alg {RTE_NET_CRC_SSE42 ,RTE_NET_CRC_NEON ,RTE_NET_CRC_AVX512 , + RTE_NET_CRC_ZBC, };/**
@@ -37,6 +38,7 @@ enum rte_net_crc_alg {* - RTE_NET_CRC_SSE42 (Use 64-bit SSE4.2 intrinsic) * - RTE_NET_CRC_NEON (Use ARM Neon intrinsic) * - RTE_NET_CRC_AVX512 (Use 512-bit AVX intrinsic) +* - RTE_NET_CRC_ZBC (Use RISC-V Zbc extension) */ void rte_net_crc_set_alg (enum rte_net_crc_alg alg);
Signature · 1 line
2.39.2
PATCH v2 4/9 config/riscv: add qemu crossbuild target
In reply to Daniel GregoryA new cross-compilation target that has extensions that DPDK uses and QEMU supports. Initially, this is just the Zbc extension for hardware crc support.
--- config/riscv/meson.build | 3 ++- config/riscv/riscv64_qemu_linux_gcc | 17 +++++++++++++++++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 +++++ 3 files changed, 24 insertions(+), 1 deletion(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc
config/riscv/meson.build +2 -1
@@ -43,7 +43,8 @@ vendor_generic = { ['RTE_MAX_NUMA_NODES', 2] ], 'arch_config': { - 'generic': {'machine_args': ['-march=rv64gc']} + 'generic': {'machine_args': ['-march=rv64gc']}, + 'qemu': {'machine_args': ['-march=rv64gc_zbc']}, } }
config/riscv/riscv64_qemu_linux_gcc new file +17 -0
@@ -0,0 +1,17 @@ +[binaries] +c = ['ccache', 'riscv64-linux-gnu-gcc'] +cpp = ['ccache', 'riscv64-linux-gnu-g++'] +ar = 'riscv64-linux-gnu-ar' +strip = 'riscv64-linux-gnu-strip' +pcap-config = '' + +[host_machine] +system = 'linux' +cpu_family = 'riscv64' +cpu = 'rv64gc_zbc' +endian = 'little' + +[properties] +vendor_id = 'generic' +arch_id = 'qemu' +pkg_config_libdir = '/usr/lib/riscv64-linux-gnu/pkgconfig'
doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst +5 -0
@@ -110,6 +110,11 @@ Currently the following targets are supported: * SiFive U740 SoC: ``config/riscv/riscv64_sifive_u740_linux_gcc`` +* QEMU: ``config/riscv/riscv64_qemu_linux_gcc`` + + * A target with all the extensions that QEMU supports that DPDK has a use for + (currently ``rv64gc_zbc``). Requires QEMU version 7.0.0 or newer. + To add a new target support, ``config/riscv/meson.build`` has to be modified by adding a new vendor/architecture id and a corresponding cross-file has to be added to ``config/riscv`` directory.
Signature · 1 line
2.39.2
PATCH v2 5/9 examples/l3fwd: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd/l3fwd_em.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd/l3fwd_em.c +1 -1
@@ -29,7 +29,7 @@#include "l3fwd_event.h" #include "em_route_parse.c" -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#define EM_HASH_CRC 1#endif
Signature · 1 line
2.39.2
PATCH v2 6/9 ipfrag: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/ip_frag/ip_frag_internal.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
lib/ip_frag/ip_frag_internal.c +3 -3
@@ -45,14 +45,14 @@ ipv4_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *)&key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (key->id, v);#else v =rte_jhash_3words (p[0 ], p[1 ], key->id,PRIME_VALUE ); -#endif /* RTE_ARCH_X86 */ +#endif /* RTE_ARCH_X86 || RTE_ARCH_ARM64 || RTE_RISCV_FEATURE_ZBC */ *v1 = v; *v2 = (v <<7 ) + (v >>14 );
@@ -66,7 +66,7 @@ ipv6_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *) &key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (p[2 ], v);
Signature · 1 line
2.39.2
PATCH v2 7/9 examples/l3fwd-power: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd-power/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd-power/main.c +1 -1
@@ -270,7 +270,7 @@ static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];#if (APP_LOOKUP_METHOD ==APP_LOOKUP_EXACT_MATCH ) -#ifdef RTE_ARCH_X86 +#if defined(RTE_ARCH_X86 ) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define DEFAULT_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.39.2
PATCH v2 8/9 hash/cuckoo: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/hash/rte_cuckoo_hash.c | 3 +++ 1 file changed, 3 insertions(+)
lib/hash/rte_cuckoo_hash.c +3 -0
@@ -409,6 +409,9 @@ rte_hash_create(const struct rte_hash_parameters *params)#elif defined(RTE_ARCH_ARM64 )if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_CRC32 )) default_hash_func = (rte_hash_function )rte_hash_crc; +#elif defined(RTE_ARCH_RISCV ) +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + default_hash_func = (rte_hash_function )rte_hash_crc;#endif /* Setup hash context */ strlcpy (h->name, params->name,sizeof (h->name));
Signature · 1 line
2.39.2
PATCH v2 9/9 member: use accelerated crc on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/member/rte_member.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
lib/member/rte_member.h +1 -1
@@ -92,7 +92,7 @@ typedef uint16_t member_set_t;#define RTE_MEMBER_SKETCH_COUNT_BYTE 0x02/** @internal Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.39.2
Quoted 46 lines · Daniel Gregory
The RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in
lib/hash/rte_crc_riscv64.hthat uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one inlib/net/net_crc_zbc.cthat uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implementrte_crc16_ccitt_zbc_handlerandrte_crc32_eth_zbc_handlerMy approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by
rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the
hash_functions_autotestbenchmark indpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:
Quoted 15 lines
| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |
Thanks for the submission.
This series comes late for v24.07 and there was no review, it is deferred to v24.11.
Signature · 1 line
David Marchand
v3 cover riscv: implement accelerated crc using zbc
In reply to Daniel GregoryThe RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in lib/hash/rte_crc_riscv64.h that uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one in lib/net/net_crc_zbc.c that uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implement rte_crc16_ccitt_zbc_handler and rte_crc32_eth_zbc_handler
My approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/
Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:
|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |
These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the hash_functions_autotest benchmark in dpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:
| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |
v3:
- rebase on 24.07
- replace crc with CRC in commits (check-git-log.sh)
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in the net/ implementation that would cause segfaults on small unaligned buffers
- refactor net/ implementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hash
Daniel Gregory (9):
config/riscv: detect presence of Zbc extension
hash: implement CRC using riscv carryless multiply
net: implement CRC using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated CRC on riscv
ipfrag: use accelerated CRC on riscv examples/l3fwd-power: use accelerated CRC on riscv hash/cuckoo: use accelerated CRC on riscv
member: use accelerated CRC on riscv
MAINTAINERS | 2 + app/test/test_crc.c | 9 + app/test/test_hash.c | 7 + config/riscv/meson.build | 44 +++- config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/rte_member.h | 2 +- lib/net/meson.build | 4 + lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 191 ++++++++++++++++++ lib/net/rte_net_crc.c | 40 ++++ lib/net/rte_net_crc.h | 2 + 22 files changed, 529 insertions(+), 41 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Signature · 1 line
2.39.2
PATCH v3 1/9 config/riscv: detect presence of Zbc extension
The RISC-V Zbc extension adds carry-less multiply instructions we can use to implement more efficient CRC hashing algorithms.
The RISC-V C api defines architecture extension test macros
https://github.com/riscv-non-isa/riscv-c-api-doc/blob/main/riscv-c-api.md#architecture-extension-test-macros
These let us detect whether the Zbc extension is supported on the compiler and -march we're building with. The C api also defines Zbc intrinsics we can use rather than inline assembly on newer versions of GCC (14.1.0+) and Clang (18.1.0+).
The Linux kernel exposes a RISC-V hardware probing syscall for getting information about the system at run-time including which extensions are available. We detect whether this interface is present by looking for the <asm/hwprobe.h> header, as it's only present in newer kernels (v6.4+). Furthermore, support for detecting certain extensions, including Zbc, wasn't present until versions after this, so we need to check the constants this header exports.
The kernel exposes bitmasks for each extension supported by the probing interface, rather than the bit index that is set if that extensions is present, so modify the existing cpu flag HWCAP table entries to line up with this. The values returned by the interface are 64-bits long, so grow the hwcap registers array to be able to hold them.
If the Zbc extension and intrinsics are both present and we can detect the Zbc extension at runtime, we define a flag, RTE_RISCV_FEATURE_ZBC.
--- config/riscv/meson.build | 41 ++++++++++ lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++++++++++++++-------- 3 files changed, 123 insertions(+), 32 deletions(-)
config/riscv/meson.build +41 -0
@@ -119,6 +119,47 @@ foreach flag: arch_config['machine_args'] endif endforeach +# check if we can do buildtime detection of extensions supported by the target +riscv_extension_macros = false +if (cc.get_define('__riscv_arch_test', args: machine_args) == '1') + message('Detected architecture extension test macros') + riscv_extension_macros = true +else + warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible') +endif + +# check if we can use hwprobe interface for runtime extension detection +riscv_hwprobe = false +if (cc.check_header('asm/hwprobe.h', args: machine_args)) + message('Detected hwprobe interface, enabling runtime detection of supported extensions') + machine_args += ['-DRTE_RISCV_FEATURE_HWPROBE'] + riscv_hwprobe = true +else + warning('Hwprobe interface not available (present in Linux v6.4+), instruction extensions won\'t be enabled') +endif + +# detect extensions +# RISC-V Carry-less multiplication extension (Zbc) for hardware implementations +# of CRC-32C (lib/hash/rte_crc_riscv64.h) and CRC-32/16 (lib/net/net_crc_zbc.c). +# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+ +if (riscv_extension_macros and riscv_hwprobe and + (cc.get_define('__riscv_zbc', args: machine_args) != '')) + if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0')) + or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0'))) + # determine whether we can detect Zbc extension (this wasn't possible until + # Linux kernel v6.8) + if (cc.compiles('''#include <asm/hwprobe.h> + int a = RISCV_HWPROBE_EXT_ZBC;''', args: machine_args)) + message('Compiling with the Zbc extension') + machine_args += ['-DRTE_RISCV_FEATURE_ZBC'] + else + warning('Detected Zbc extension but cannot use because runtime detection doesn\'t support it (support present in Linux kernel v6.8+)') + endif + else + warning('Detected Zbc extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)') + endif +endif + # apply flags foreach flag: dpdk_flags if flag.length() > 0
lib/eal/riscv/include/rte_cpuflags.h +2 -0
@@ -42,6 +42,8 @@ enum rte_cpu_flag_t {RTE_CPUFLAG_RISCV_ISA_X ,/* Non-standard extension present */ RTE_CPUFLAG_RISCV_ISA_Y ,/* Reserved */ RTE_CPUFLAG_RISCV_ISA_Z ,/* Reserved */ + + RTE_CPUFLAG_RISCV_EXT_ZBC, /* Carry-less multiplication */ };#include "generic/rte_cpuflags.h"
lib/eal/riscv/rte_cpuflags.c +80 -32
153 lines not on this page · raw message
Signature · 1 line
2.39.2
PATCH v3 2/9 hash: implement CRC using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions from RISC-V's Zbc extension, implement a Barrett reduction that calculates CRC-32C checksums.
Based on the approach described by Intel's whitepaper on "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction", which is also described here
(https://web.archive.org/web/20240111232520/https://mary.rs/lab/crc32/)
Add a case to the autotest_hash unit test.
--- MAINTAINERS | 1 + app/test/test_hash.c | 7 +++ lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++++++++++++++++++++++++++++++++ lib/hash/rte_hash_crc.c | 13 +++++- lib/hash/rte_hash_crc.h | 6 ++- 6 files changed, 115 insertions(+), 2 deletions(-) create mode 100644 lib/hash/rte_crc_riscv64.h
MAINTAINERS +1 -0
@@ -322,6 +322,7 @@ M: Stanislaw Kardach <[email protected]> F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ +F: lib/hash/rte_crc_riscv64.h Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_hash.c +7 -0
@@ -231,6 +231,13 @@ test_crc32_hash_alg_equiv(void)printf ("Failed checking CRC32_SW against CRC32_ARM64\n" );break ; } + +/* Check against 8-byte-operand RISCV64 CRC32 if available */ +rte_hash_crc_set_alg (CRC32_RISCV64 ); +if (hash_val !=rte_hash_crc (data64, data_len, init_val)) { +printf ("Failed checking CRC32_SW against CRC32_RISC64\n" ); +break ; + } }/* Resetting to best available algorithm */
lib/hash/meson.build +1 -0
@@ -12,6 +12,7 @@ headers = files( indirect_headers += files( 'rte_crc_arm64.h', 'rte_crc_generic.h', + 'rte_crc_riscv64.h', 'rte_crc_sw.h', 'rte_crc_x86.h', 'rte_thash_x86_gfni.h',
lib/hash/rte_crc_riscv64.h new file +89 -0
89 lines not on this page · raw message
lib/hash/rte_hash_crc.c +12 -1
@@ -15,7 +15,7 @@ RTE_LOG_REGISTER_SUFFIX(hash_crc_logtype, crc, INFO);uint8_t rte_hash_crc32_alg =CRC32_SW ;/** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -24,6 +24,7 @@ uint8_t rte_hash_crc32_alg = CRC32_SW;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISCV64 Zbc extension if available * */ void
@@ -52,6 +53,14 @@ rte_hash_crc_set_alg(uint8_t alg) rte_hash_crc32_alg =CRC32_ARM64 ;#endif +#if defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +if (!(alg &CRC32_RISCV64 )) +HASH_CRC_LOG (WARNING , +"Unsupported CRC32 algorithm requested using CRC32_RISCV64" ); +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + rte_hash_crc32_alg =CRC32_RISCV64 ; +#endif +if (rte_hash_crc32_alg ==CRC32_SW )HASH_CRC_LOG (WARNING ,"Unsupported CRC32 algorithm requested using CRC32_SW" );
@@ -64,6 +73,8 @@ RTE_INIT(rte_hash_crc_init_alg)rte_hash_crc_set_alg (CRC32_SSE42_x64);#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)rte_hash_crc_set_alg (CRC32_ARM64 ); +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +rte_hash_crc_set_alg (CRC32_RISCV64 );#else rte_hash_crc_set_alg (CRC32_SW );#endif
lib/hash/rte_hash_crc.h +5 -1
@@ -28,6 +28,7 @@ extern "C" {#define CRC32_x64 (1U << 2)#define CRC32_SSE42_x64 (CRC32_x64|CRC32_SSE42)#define CRC32_ARM64 (1U << 3) +#define CRC32_RISCV64 (1U << 4)extern uint8_t rte_hash_crc32_alg;
@@ -35,12 +36,14 @@ extern uint8_t rte_hash_crc32_alg;#include "rte_crc_arm64.h" #elif defined(RTE_ARCH_X86 )#include "rte_crc_x86.h" +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +#include "rte_crc_riscv64.h" #else #include "rte_crc_generic.h" #endif /** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -49,6 +52,7 @@ extern uint8_t rte_hash_crc32_alg;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISC-V Carry-less multiply if available (default rv64gc_zbc) */ void rte_hash_crc_set_alg (uint8_t alg);
Signature · 1 line
2.39.2
PATCH v3 3/9 net: implement CRC using riscv carryless multiply
In reply to Daniel GregoryUsing carryless multiply instructions (clmul) from RISC-V's Zbc extension, implement CRC-32 and CRC-16 calculations on buffers.
Based on the approach described in Intel's whitepaper on "Fast CRC Computation for Generic Polynomails Using PCLMULQDQ Instructions", we perform repeated folds-by-1 whilst the buffer is still big enough, then perform Barrett's reductions on the rest.
Add a case to the crc_autotest suite that tests this implementation.
--- MAINTAINERS | 1 + app/test/test_crc.c | 9 ++ lib/net/meson.build | 4 + lib/net/net_crc.h | 11 +++ lib/net/net_crc_zbc.c | 191 ++++++++++++++++++++++++++++++++++++++++++ lib/net/rte_net_crc.c | 40 +++++++++ lib/net/rte_net_crc.h | 2 + 7 files changed, 258 insertions(+) create mode 100644 lib/net/net_crc_zbc.c
MAINTAINERS +1 -0
@@ -323,6 +323,7 @@ F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ F: lib/hash/rte_crc_riscv64.h +F: lib/net/net_crc_zbc.c Intel x86 M: Bruce Richardson <[email protected]>
app/test/test_crc.c +9 -0
@@ -168,6 +168,15 @@ test_crc(void)return ret; } +/* set CRC riscv mode */ +rte_net_crc_set_alg (RTE_NET_CRC_ZBC ); + + ret =test_crc_calc (); +if (ret <0 ) { +printf ("test crc (riscv64 zbc clmul): failed (%d)\n" , ret); +return ret; + } +return 0 ; }
lib/net/meson.build +4 -0
@@ -125,4 +125,8 @@ elif (dpdk_conf.has('RTE_ARCH_ARM64') and cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '') sources += files('net_crc_neon.c') cflags += ['-DCC_ARM64_NEON_PMULL_SUPPORT'] +elif (dpdk_conf.has('RTE_ARCH_RISCV') and + cc.get_define('RTE_RISCV_FEATURE_ZBC', args: machine_args) != '') + sources += files('net_crc_zbc.c') + cflags += ['-DCC_RISCV64_ZBC_CLMUL_SUPPORT'] endif
lib/net/net_crc.h +11 -0
@@ -42,4 +42,15 @@ rte_crc16_ccitt_neon_handler(const uint8_t *data, uint32_t data_len);uint32_t rte_crc32_eth_neon_handler (const uint8_t *data,uint32_t data_len); +/* RISCV64 Zbc */ +void +rte_net_crc_zbc_init (void ); + +uint32_t +rte_crc16_ccitt_zbc_handler (const uint8_t *data,uint32_t data_len); + +uint32_t +rte_crc32_eth_zbc_handler (const uint8_t *data,uint32_t data_len); + +#endif /* _NET_CRC_H_ */
lib/net/net_crc_zbc.c new file +191 -0
191 lines not on this page · raw message
lib/net/rte_net_crc.c +40 -0
@@ -67,6 +67,12 @@ static const rte_net_crc_handler handlers_neon[] = { [RTE_NET_CRC32_ETH] = rte_crc32_eth_neon_handler, };#endif +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +static const rte_net_crc_handler handlers_zbc[] = { + [RTE_NET_CRC16_CCITT ] = rte_crc16_ccitt_zbc_handler, + [RTE_NET_CRC32_ETH ] = rte_crc32_eth_zbc_handler, +}; +#endif static uint16_t max_simd_bitwidth;
@@ -244,6 +250,31 @@ neon_pmull_init(void)#endif } +/* ZBC/CLMUL handling */ + +#define ZBC_CLMUL_CPU_SUPPORTED \ + rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_EXT_ZBC) + +static const rte_net_crc_handler * +zbc_clmul_get_handlers (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) +return handlers_zbc; +#endif +NET_LOG (INFO ,"Requirements not met, can't use Zbc" ); +return NULL ; +} + +static void +zbc_clmul_init (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) +rte_net_crc_zbc_init (); +#endif +} +/* Default handling */ static uint32_t
@@ -260,6 +291,9 @@ rte_crc16_ccitt_default_handler(const uint8_t *data, uint32_t data_len)if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers =neon_pmull_get_handlers (); +if (handlers !=NULL ) +return handlers[RTE_NET_CRC16_CCITT ](data, data_len); + handlers =zbc_clmul_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC16_CCITT ](data, data_len); handlers = handlers_scalar;
@@ -282,6 +316,8 @@ rte_crc32_eth_default_handler(const uint8_t *data, uint32_t data_len) handlers =neon_pmull_get_handlers ();if (handlers !=NULL )return handlers[RTE_NET_CRC32_ETH ](data, data_len); + handlers =zbc_clmul_get_handlers (); +return handlers[RTE_NET_CRC32_ETH ](data, data_len); handlers = handlers_scalar;return handlers[RTE_NET_CRC32_ETH ](data, data_len); }
@@ -306,6 +342,9 @@ rte_net_crc_set_alg(enum rte_net_crc_alg alg)break ;/* for x86, always break here */ case RTE_NET_CRC_NEON : handlers =neon_pmull_get_handlers (); +break ; +case RTE_NET_CRC_ZBC : + handlers =zbc_clmul_get_handlers ();/* fall-through */ case RTE_NET_CRC_SCALAR :/* fall-through */
@@ -338,4 +377,5 @@ RTE_INIT(rte_net_crc_init)sse42_pclmulqdq_init ();avx512_vpclmulqdq_init ();neon_pmull_init (); +zbc_clmul_init (); }
lib/net/rte_net_crc.h +2 -0
@@ -24,6 +24,7 @@ enum rte_net_crc_alg {RTE_NET_CRC_SSE42 ,RTE_NET_CRC_NEON ,RTE_NET_CRC_AVX512 , + RTE_NET_CRC_ZBC, };/**
@@ -37,6 +38,7 @@ enum rte_net_crc_alg {* - RTE_NET_CRC_SSE42 (Use 64-bit SSE4.2 intrinsic) * - RTE_NET_CRC_NEON (Use ARM Neon intrinsic) * - RTE_NET_CRC_AVX512 (Use 512-bit AVX intrinsic) +* - RTE_NET_CRC_ZBC (Use RISC-V Zbc extension) */ void rte_net_crc_set_alg (enum rte_net_crc_alg alg);
Signature · 1 line
2.39.2
PATCH v3 4/9 config/riscv: add qemu crossbuild target
In reply to Daniel GregoryA new cross-compilation target that has extensions that DPDK uses and QEMU supports. Initially, this is just the Zbc extension for hardware CRC support.
--- config/riscv/meson.build | 3 ++- config/riscv/riscv64_qemu_linux_gcc | 17 +++++++++++++++++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 +++++ 3 files changed, 24 insertions(+), 1 deletion(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc
config/riscv/meson.build +2 -1
@@ -43,7 +43,8 @@ vendor_generic = { ['RTE_MAX_NUMA_NODES', 2] ], 'arch_config': { - 'generic': {'machine_args': ['-march=rv64gc']} + 'generic': {'machine_args': ['-march=rv64gc']}, + 'qemu': {'machine_args': ['-march=rv64gc_zbc']}, } }
config/riscv/riscv64_qemu_linux_gcc new file +17 -0
@@ -0,0 +1,17 @@ +[binaries] +c = ['ccache', 'riscv64-linux-gnu-gcc'] +cpp = ['ccache', 'riscv64-linux-gnu-g++'] +ar = 'riscv64-linux-gnu-ar' +strip = 'riscv64-linux-gnu-strip' +pcap-config = '' + +[host_machine] +system = 'linux' +cpu_family = 'riscv64' +cpu = 'rv64gc_zbc' +endian = 'little' + +[properties] +vendor_id = 'generic' +arch_id = 'qemu' +pkg_config_libdir = '/usr/lib/riscv64-linux-gnu/pkgconfig'
doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst +5 -0
@@ -110,6 +110,11 @@ Currently the following targets are supported: * SiFive U740 SoC: ``config/riscv/riscv64_sifive_u740_linux_gcc`` +* QEMU: ``config/riscv/riscv64_qemu_linux_gcc`` + + * A target with all the extensions that QEMU supports that DPDK has a use for + (currently ``rv64gc_zbc``). Requires QEMU version 7.0.0 or newer. + To add a new target support, ``config/riscv/meson.build`` has to be modified by adding a new vendor/architecture id and a corresponding cross-file has to be added to ``config/riscv`` directory.
Signature · 1 line
2.39.2
PATCH v3 5/9 examples/l3fwd: use accelerated CRC on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd/l3fwd_em.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd/l3fwd_em.c +1 -1
@@ -29,7 +29,7 @@#include "l3fwd_event.h" #include "em_route_parse.c" -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#define EM_HASH_CRC 1#endif
Signature · 1 line
2.39.2
PATCH v3 6/9 ipfrag: use accelerated CRC on riscv
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/ip_frag/ip_frag_internal.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
lib/ip_frag/ip_frag_internal.c +3 -3
@@ -45,14 +45,14 @@ ipv4_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *)&key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (key->id, v);#else v =rte_jhash_3words (p[0 ], p[1 ], key->id,PRIME_VALUE ); -#endif /* RTE_ARCH_X86 */ +#endif /* RTE_ARCH_X86 || RTE_ARCH_ARM64 || RTE_RISCV_FEATURE_ZBC */ *v1 = v; *v2 = (v <<7 ) + (v >>14 );
@@ -66,7 +66,7 @@ ipv6_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *) &key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (p[2 ], v);
Signature · 1 line
2.39.2
PATCH v3 7/9 examples/l3fwd-power: use accelerated CRC on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd-power/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd-power/main.c +1 -1
@@ -270,7 +270,7 @@ static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];#if (APP_LOOKUP_METHOD ==APP_LOOKUP_EXACT_MATCH ) -#ifdef RTE_ARCH_X86 +#if defined(RTE_ARCH_X86 ) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define DEFAULT_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.39.2
PATCH v3 8/9 hash/cuckoo: use accelerated CRC on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/hash/rte_cuckoo_hash.c | 3 +++ 1 file changed, 3 insertions(+)
lib/hash/rte_cuckoo_hash.c +3 -0
@@ -427,6 +427,9 @@ rte_hash_create(const struct rte_hash_parameters *params)#elif defined(RTE_ARCH_ARM64 )if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_CRC32 )) default_hash_func = (rte_hash_function )rte_hash_crc; +#elif defined(RTE_ARCH_RISCV ) +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + default_hash_func = (rte_hash_function )rte_hash_crc;#endif /* Setup hash context */ strlcpy (h->name, params->name,sizeof (h->name));
Signature · 1 line
2.39.2
PATCH v3 9/9 member: use accelerated CRC on riscv
In reply to Daniel GregoryWhen the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/member/rte_member.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
lib/member/rte_member.h +1 -1
@@ -92,7 +92,7 @@ typedef uint16_t member_set_t;#define RTE_MEMBER_SKETCH_COUNT_BYTE 0x02/** @internal Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.39.2
Would it be possible to get a review on this patchset? I would be happy to hear any feedback on the approach to RISC-V extension detection or how I have implemented the hardware-optimised CRCs.
Kind regards,
Daniel
Quoted 4 lines · Daniel Gregory
Would it be possible to get a review on this patchset? I would be happy to hear any feedback on the approach to RISC-V extension detection or how I have implemented the hardware-optimised CRCs.
Hi,
Just wanted to follow up on this patch. Are you still interested in moving it forward?
If so, could you please address the remaining comments([Patch 1/5] ...) and resolve the merge conflicts?
Quoted 107 lines · On Tue, 27 Aug 2024 16:32:20 +0100 Daniel Gregory…
The RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in
lib/hash/rte_crc_riscv64.hthat uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one inlib/net/net_crc_zbc.cthat uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implementrte_crc16_ccitt_zbc_handlerandrte_crc32_eth_zbc_handlerMy approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by
rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the
hash_functions_autotestbenchmark indpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |v3:
- rebase on 24.07
- replace crc with CRC in commits (check-git-log.sh)
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in thenet/implementation that would cause segfaults on small unaligned buffers
- refactornet/implementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hashDaniel Gregory (9):
config/riscv: detect presence of Zbc extension
hash: implement CRC using riscv carryless multiply
net: implement CRC using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated CRC on riscv
ipfrag: use accelerated CRC on riscv examples/l3fwd-power: use accelerated CRC on riscv hash/cuckoo: use accelerated CRC on riscv
member: use accelerated CRC on riscvMAINTAINERS | 2 + app/test/test_crc.c | 9 + app/test/test_hash.c | 7 + config/riscv/meson.build | 44 +++- config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 89 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/rte_member.h | 2 +- lib/net/meson.build | 4 +
Quoted 9 lines
lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 191 ++++++++++++++++++ lib/net/rte_net_crc.c | 40 ++++ lib/net/rte_net_crc.h | 2 + 22 files changed, 529 insertions(+), 41 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Not sure all the reasons this never got merged but it won't apply to current DPDK version and AI review has lots of feedback. If still interested, please fix up and resubmit.
# DPDK Patch Review: RISC-V Zbc Extension Series
**Patch Series:** [PATCH v3 1-9/9] RISC-V Zbc extension for hardware CRC **Submitter:** Daniel Gregory <[email protected]> **Date:** August 27, 2024
**Patchwork IDs:** 143405-143413
---
## Overview
This 9-patch series adds support for RISC-V's Zbc (carry-less multiplication) extension to implement hardware-accelerated CRC-32C, CRC-32, and CRC-16 calculations in DPDK. The series includes build-time and runtime detection, library implementations, and integration into various DPDK components.
---
## ERROR Level Issues (Must Fix)
### 1. **SPDX License Identifier Typo** — lib/hash/rte_crc_riscv64.h
```c /* SPDX-License_Identifier: BSD-3-Clause ← WRONG: underscore * Copyright(c) ByteDance 2024 */ ```
**Required:**
```c /* SPDX-License-Identifier: BSD-3-Clause ← hyphen required * Copyright(c) 2024 ByteDance */ ```
**AGENTS.md Reference:** "Every source file must begin with an SPDX license identifier"
---
### 2. **Header Guard After Includes** — lib/hash/rte_crc_riscv64.h
The header guard is placed AFTER the #include statements, which defeats its purpose:
```c /* SPDX-License_Identifier: BSD-3-Clause * Copyright(c) ByteDance 2024 */ #include <assert.h> ← includes BEFORE guard #include <stdint.h> #include <riscv_bitmanip.h> #ifndef _RTE_CRC_RISCV64_H_ ← guard comes too late #define _RTE_CRC_RISCV64_H_ ```
**Required:** Header guard immediately after blank line following copyright:
```c /* SPDX-License-Identifier: BSD-3-Clause * Copyright(c) 2024 ByteDance */ #ifndef _RTE_CRC_RISCV64_H_ #define _RTE_CRC_RISCV64_H_ #include <assert.h> #include <stdint.h> #include <riscv_bitmanip.h> ```
---
### 3. **Missing NULL Check / Logic Bug** — lib/net/rte_net_crc.c
In rte_crc32_eth_default_handler(), after adding zbc support:
```c
handlers = neon_pmull_get_handlers();
if (handlers != NULL)
return handlers[RTE_NET_CRC32_ETH](data, data_len);
handlers = zbc_clmul_get_handlers();
return handlers[RTE_NET_CRC32_ETH](data, data_len); ← Missing NULL check!
handlers = handlers_scalar;
```This will crash if zbc_clmul_get_handlers() returns NULL. **Required:**
```c
handlers = zbc_clmul_get_handlers();
if (handlers != NULL)
return handlers[RTE_NET_CRC32_ETH](data, data_len);
handlers = handlers_scalar;
```---
### 4. **Test Output Typo** — app/test/test_hash.c
```c
printf("Failed checking CRC32_SW against CRC32_RISC64\n");
^^^^ Missing 'V'
```**Should be:** CRC32_RISCV64
---
## WARNING Level Issues (Should Fix)
### 5. **Use of Standard assert() Instead of RTE_ASSERT**
Multiple files use standard library assert():
- lib/hash/rte_crc_riscv64.h line 539
- lib/net/net_crc_zbc.c line 896
**AGENTS.md Guidance:** DPDK prefers RTE_ASSERT() or RTE_VERIFY() for consistency. Standard assert() may be compiled out in release builds unexpectedly.
---
### 6. **Implicit Pointer Comparison in likely() Macro**
In lib/hash/rte_crc_riscv64.h:
```c if (likely(rte_hash_crc32_alg & CRC32_RISCV64)) ```
This is a bitmask check (integer), which is acceptable. However, for clarity and consistency with DPDK style, explicit comparison is preferred:
```c if (likely((rte_hash_crc32_alg & CRC32_RISCV64) != 0)) ```
---
### 7. **Fall-through Comment Missing** — lib/net/rte_net_crc.c
```c
case RTE_NET_CRC_ZBC:
handlers = zbc_clmul_get_handlers();
/* fall-through */ ← MISSING
case RTE_NET_CRC_SCALAR:
```DPDK requires explicit /* fall-through */ or /* FALLTHROUGH */ comments (per AGENTS.md Section on switch statements).
---
### 8. **Commit Body Line Length** — Patch 1/9
Some lines in the commit message body may exceed 75 characters:
``` These let us detect whether the Zbc extension is supported on the compiler and -march we're building with. The C api also defines Zbc ```
**AGENTS.md:** "Body wrapped at 75 characters"
---
### 9. **Meson Build File Comment at End of Conditionals**
In lib/net/meson.build, the endif cascade doesn't clearly indicate which condition is being closed. Consider adding comments for maintainability.
---
### 10. **Copyright Format Inconsistency**
The patches use:
```c Copyright(c) ByteDance 2024 ```
DPDK convention (per AGENTS.md examples) typically uses:
```c Copyright(c) 2024 ByteDance ```
With the year first.
---
## INFO Level Issues (Consider)
### 11. **Static Variables Without static at File Scope** — lib/net/net_crc_zbc.c
```c struct crc_clmul_ctx crc32_eth_clmul; struct crc_clmul_ctx crc16_ccitt_clmul; ```
These should be static to limit scope to the translation unit:
```c static struct crc_clmul_ctx crc32_eth_clmul; static struct crc_clmul_ctx crc16_ccitt_clmul; ```
---
### 12. **Magic Numbers in Constants**
In net_crc_zbc.c, the polynomial constants are well-commented, which is good. Consider also documenting the mathematical derivation source more explicitly (the Intel whitepaper reference is good).
---
### 13. **Documentation Gap**
The series doesn't include updates to release notes (doc/guides/rel_notes/). For a feature addition like hardware CRC acceleration on a new architecture, this should be documented.
**AGENTS.md:** "Update release notes in doc/guides/rel_notes/ for important changes"
---
## Commit Message Analysis
| Patch | Subject Length | Format Check | |-------|---------------|--------------| | 1/9 | `config/riscv: detect presence of Zbc extension` | 47 chars ✓ | | 2/9 | `hash: implement CRC using riscv carryless multiply` | 50 chars ✓ | | 3/9 | `net: implement CRC using riscv carryless multiply` | 49 chars ✓ | | 4/9 | `config/riscv: add qemu crossbuild target` | 41 chars ✓ | | 5/9 | `examples/l3fwd: use accelerated CRC on riscv` | 45 chars ✓ | | 6/9 | `ipfrag: use accelerated CRC on riscv` | 38 chars ✓ | | 7/9 | `examples/l3fwd-power: use accelerated CRC on riscv` | 51 chars ✓ | | 8/9 | `hash/cuckoo: use accelerated CRC on riscv` | 42 chars ✓ | | 9/9 | `member: use accelerated CRC on riscv` | 37 chars ✓ |
**All subject lines:**
- ✓ Under 60 character limit
- ✓ Lowercase after colon
- ✓ Imperative mood
- ✓ No trailing period
- ✓ Correct prefixes (config/riscv, hash, net, examples/*, lib names)
- ✓ Signed-off-by present on all patches
---
## Positive Observations
1. **Well-structured series** — The patches are logically ordered: detection first, then core implementations, then consumers.
2. **Good documentation** — Comments explain the Barrett reduction algorithm and reference the Intel whitepaper.
3. **Proper runtime detection** — Uses hwprobe syscall for runtime CPU feature detection, with appropriate fallbacks.
4. **Test coverage** — Adds test cases to existing test suites (test_hash.c, test_crc.c).
5. **MAINTAINERS updates** — Properly adds new files to the RISC-V maintainer's scope.
6. **Cross-compilation support** — Includes QEMU target for development/testing.
---
## Summary
| Severity | Count | Action Required | |----------|-------|-----------------| | **ERROR** | 4 | Must fix before merge | | **WARNING** | 6 | Should fix | | **INFO** | 3 | Consider fixing |
### Required Actions Before Merge
1. Fix SPDX identifier typo (License_Identifier → License-Identifier)
2. Move header guard before includes in rte_crc_riscv64.h
3. Add NULL check after zbc_clmul_get_handlers() in rte_crc32_eth_default_handler()
4. Fix typo CRC32_RISC64 → CRC32_RISCV64 in test output
### Recommended Actions
5. Add fall-through comment in switch statement
6. Make file-scope variables static
7. Add release notes entry
8. Consider using RTE_ASSERT() instead of assert()
v4 cover riscv: implement accelerated crc using zbc
In reply to Daniel GregoryThe RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in lib/hash/rte_crc_riscv64.h that uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one in lib/net/net_crc_zbc.c that uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implement rte_crc16_ccitt_zbc_handler and rte_crc32_eth_zbc_handler
My approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/
Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:
|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |
These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the hash_functions_autotest benchmark in dpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:
| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |
v4:
- rebase on 26.03-rc1
- RISC64 -> RISCV64 in test_hash.c (Stephen Hemminger)
- Added section to release notes (Stephen Hemminger)
- SPDX-License_Identifier -> SPDX-License-Identifier in rte_crc_riscv64.h (Stephen Hemminger)
- Fix header guard in rte_crc_riscv64.h (Stephen Hemminger)
- assert -> RTE_ASSERT in rte_crc_riscv64.h (Stephen Hemminger)
- Fix copyright statement in net_crc_zbc.c (Stephen Hemminger)
- Make crc context structs static in net_crc_zbc.c (Stephen Hemminger)
- prefer the optimised crc when zbc present over jhash in rte_fbk_hash.c v3:
- rebase on 24.07
- replace crc with CRC in commits (check-git-log.sh)
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in the net/ implementation that would cause segfaults on small unaligned buffers
- refactor net/ implementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hash
Daniel Gregory (10):
config/riscv: detect presence of Zbc extension
hash: implement CRC using riscv carryless multiply
net: implement CRC using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated CRC on riscv
ipfrag: use accelerated CRC on riscv examples/l3fwd-power: use accelerated CRC on riscv
hash: use accelerated CRC on riscv
member: use accelerated CRC on riscv
doc: implement CRC using riscv carryless multiply
.mailmap | 2 +- MAINTAINERS | 2 + app/test/test_crc.c | 10 + app/test/test_hash.c | 7 + config/riscv/meson.build | 33 +++ config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + doc/guides/rel_notes/release_26_03.rst | 8 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 90 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_fbk_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/member.h | 2 +- lib/net/meson.build | 4 + lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 194 ++++++++++++++++++ lib/net/rte_net_crc.c | 30 ++- lib/net/rte_net_crc.h | 3 + 25 files changed, 526 insertions(+), 42 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Signature · 1 line
2.53.0
PATCH v4 1/10 config/riscv: detect presence of Zbc extension
From: Daniel Gregory <[email protected]>
The RISC-V Zbc extension adds carry-less multiply instructions we can use to implement more efficient CRC hashing algorithms.
The RISC-V C api defines Zbc intrinsics we can use rather than inline assembly on newer versions of GCC (14.1.0+) and Clang (18.1.0+). We can detect whether these are supported on the compiler and -march we're building with by using the architecture extension macros in combination with checking compiler version.
The Linux kernel exposes a RISC-V hardware probing syscall for getting information about the system at run-time including which extensions are available. We detect whether this interface is present by looking for the <asm/hwprobe.h> header, as it's only present in newer kernels (v6.4+). Furthermore, support for detecting certain extensions, including Zbc, wasn't present until versions after this, so we need to check the constants this header exports.
The kernel exposes bitmasks for each extension supported by the probing interface, rather than the bit index that is set if that extensions is present, so modify the existing cpu flag HWCAP table entries to line up with this. The values returned by the interface are 64-bits long, so grow the hwcap registers array to be able to hold them.
If the Zbc extension and intrinsics are both present and we can detect the Zbc extension at runtime, we define a flag, RTE_RISCV_FEATURE_ZBC.
--- config/riscv/meson.build | 32 ++++++++ lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++++++++++++++-------- 3 files changed, 114 insertions(+), 32 deletions(-)
config/riscv/meson.build +32 -0
@@ -131,6 +131,16 @@ else warning('RISC-V architecture extension test macros not available. Build-time detection of extensions not possible') endif +# check if we can use hwprobe interface for runtime extension detection +riscv_hwprobe = false +if (cc.check_header('asm/hwprobe.h', args: machine_args)) + message('Detected hwprobe interface, enabling runtime detection of supported extensions') + machine_args += ['-DRTE_RISCV_FEATURE_HWPROBE'] + riscv_hwprobe = true +else + warning('Hwprobe interface not available (present in Linux v6.4+), instruction extensions won\'t be enabled') +endif + # detect extensions # Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+ if (riscv_extension_macros and
@@ -147,6 +157,28 @@ if (riscv_extension_macros and endif endif +# detect extensions +# RISC-V Carry-less multiplication extension (Zbc) for hardware implementations +# of CRC-32C (lib/hash/rte_crc_riscv64.h) and CRC-32/16 (lib/net/net_crc_zbc.c). +# Requires intrinsics available in GCC 14.1.0+ and Clang 18.1.0+ +if (riscv_extension_macros and riscv_hwprobe and + (cc.get_define('__riscv_zbc', args: machine_args) != '')) + if ((cc.get_id() == 'gcc' and cc.version().version_compare('>=14.1.0')) + or (cc.get_id() == 'clang' and cc.version().version_compare('>=18.1.0'))) + # determine whether we can detect Zbc extension (this wasn't possible until + # Linux kernel v6.8) + if (cc.compiles('''#include <asm/hwprobe.h> + int a = RISCV_HWPROBE_EXT_ZBC;''', args: machine_args)) + message('Compiling with the Zbc extension') + machine_args += ['-DRTE_RISCV_FEATURE_ZBC'] + else + warning('Detected Zbc extension but cannot use because runtime detection doesn\'t support it (support present in Linux kernel v6.8+)') + endif + else + warning('Detected Zbc extension but cannot use because intrinsics are not available (present in GCC 14.1.0+ and Clang 18.1.0+)') + endif +endif + # apply flags foreach flag: dpdk_flags if flag.length() > 0
lib/eal/riscv/include/rte_cpuflags.h +2 -0
@@ -38,6 +38,8 @@ enum rte_cpu_flag_t {RTE_CPUFLAG_RISCV_ISA_X ,/* Non-standard extension present */ RTE_CPUFLAG_RISCV_ISA_Y ,/* Reserved */ RTE_CPUFLAG_RISCV_ISA_Z ,/* Reserved */ + + RTE_CPUFLAG_RISCV_EXT_ZBC, /* Carry-less multiplication */ };#include "generic/rte_cpuflags.h"
lib/eal/riscv/rte_cpuflags.c +80 -32
153 lines not on this page · raw message
Signature · 1 line
2.53.0
PATCH v4 2/10 hash: implement CRC using riscv carryless multiply
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
Using carryless multiply instructions from RISC-V's Zbc extension, implement a Barrett reduction that calculates CRC-32C checksums.
Based on the approach described by Intel's whitepaper on "Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction", which is also described here
(https://web.archive.org/web/20240111232520/https://mary.rs/lab/crc32/)
Add a case to the autotest_hash unit test.
--- MAINTAINERS | 1 + app/test/test_hash.c | 7 +++ lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 90 ++++++++++++++++++++++++++++++++++++++ lib/hash/rte_hash_crc.c | 13 +++++- lib/hash/rte_hash_crc.h | 6 ++- 6 files changed, 116 insertions(+), 2 deletions(-) create mode 100644 lib/hash/rte_crc_riscv64.h
MAINTAINERS +1 -0
@@ -338,6 +338,7 @@ M: Stanisław Kardach <[email protected]> F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ +F: lib/hash/rte_crc_riscv64.h M: Sun Yuechi <[email protected]> F: lib/**/*rvv*
app/test/test_hash.c +7 -0
@@ -231,6 +231,13 @@ test_crc32_hash_alg_equiv(void)printf ("Failed checking CRC32_SW against CRC32_ARM64\n" );break ; } + +/* Check against 8-byte-operand RISCV64 CRC32 if available */ +rte_hash_crc_set_alg (CRC32_RISCV64 ); +if (hash_val !=rte_hash_crc (data64, data_len, init_val)) { +printf ("Failed checking CRC32_SW against CRC32_RISCV64\n" ); +break ; + } }/* Resetting to best available algorithm */
lib/hash/meson.build +1 -0
@@ -12,6 +12,7 @@ headers = files( indirect_headers += files( 'rte_crc_arm64.h', 'rte_crc_generic.h', + 'rte_crc_riscv64.h', 'rte_crc_sw.h', 'rte_crc_x86.h', 'rte_thash_x86_gfni.h',
lib/hash/rte_crc_riscv64.h new file +90 -0
90 lines not on this page · raw message
lib/hash/rte_hash_crc.c +12 -1
@@ -17,7 +17,7 @@ RTE_EXPORT_SYMBOL(rte_hash_crc32_alg)uint8_t rte_hash_crc32_alg =CRC32_SW ;/** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -26,6 +26,7 @@ uint8_t rte_hash_crc32_alg = CRC32_SW;* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISCV64 Zbc extension if available */ RTE_EXPORT_SYMBOL (rte_hash_crc_set_alg) void
@@ -54,6 +55,14 @@ rte_hash_crc_set_alg(uint8_t alg) rte_hash_crc32_alg =CRC32_ARM64 ;#endif +#if defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +if (!(alg &CRC32_RISCV64 )) +HASH_CRC_LOG (WARNING , +"Unsupported CRC32 algorithm requested using CRC32_RISCV64" ); +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + rte_hash_crc32_alg =CRC32_RISCV64 ; +#endif +if (rte_hash_crc32_alg ==CRC32_SW )HASH_CRC_LOG (WARNING ,"Unsupported CRC32 algorithm requested using CRC32_SW" );
@@ -66,6 +75,8 @@ RTE_INIT(rte_hash_crc_init_alg)rte_hash_crc_set_alg (CRC32_SSE42_x64);#elif defined(RTE_ARCH_ARM64 ) && defined(__ARM_FEATURE_CRC32)rte_hash_crc_set_alg (CRC32_ARM64 ); +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +rte_hash_crc_set_alg (CRC32_RISCV64 );#else rte_hash_crc_set_alg (CRC32_SW );#endif
lib/hash/rte_hash_crc.h +5 -1
@@ -24,6 +24,7 @@#define CRC32_x64 (1U << 2)#define CRC32_SSE42_x64 (CRC32_x64|CRC32_SSE42)#define CRC32_ARM64 (1U << 3) +#define CRC32_RISCV64 (1U << 4)extern uint8_t rte_hash_crc32_alg;
@@ -31,6 +32,8 @@ extern uint8_t rte_hash_crc32_alg;#include "rte_crc_arm64.h" #elif defined(RTE_ARCH_X86 )#include "rte_crc_x86.h" +#elif defined(RTE_ARCH_RISCV ) && defined(RTE_RISCV_FEATURE_ZBC ) +#include "rte_crc_riscv64.h" #else #include "rte_crc_generic.h" #endif
@@ -40,7 +43,7 @@ extern "C" {#endif /** -* Allow or disallow use of SSE4.2/ARMv8 intrinsics for CRC32 hash +* Allow or disallow use of SSE4.2/ARMv8/RISC-V intrinsics for CRC32 hash * calculation. * * @param alg
@@ -49,6 +52,7 @@ extern "C" {* - (CRC32_SSE42) Use SSE4.2 intrinsics if available * - (CRC32_SSE42_x64) Use 64-bit SSE4.2 intrinsic if available (default x86) * - (CRC32_ARM64) Use ARMv8 CRC intrinsic if available (default ARMv8) +* - (CRC32_RISCV64) Use RISC-V Carry-less multiply if available (default rv64gc_zbc) */ void rte_hash_crc_set_alg (uint8_t alg);
Signature · 1 line
2.53.0
PATCH v4 3/10 net: implement CRC using riscv carryless multiply
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
Using carryless multiply instructions (clmul) from RISC-V's Zbc extension, implement CRC-32 and CRC-16 calculations on buffers.
Based on the approach described in Intel's whitepaper on "Fast CRC Computation for Generic Polynomails Using PCLMULQDQ Instructions", we perform repeated folds-by-1 whilst the buffer is still big enough, then perform Barrett's reductions on the rest.
Add a case to the crc_autotest suite that tests this implementation.
--- MAINTAINERS | 1 + app/test/test_crc.c | 10 +++ lib/net/meson.build | 4 + lib/net/net_crc.h | 11 +++ lib/net/net_crc_zbc.c | 194 ++++++++++++++++++++++++++++++++++++++++++ lib/net/rte_net_crc.c | 30 ++++++- lib/net/rte_net_crc.h | 3 + 7 files changed, 252 insertions(+), 1 deletion(-) create mode 100644 lib/net/net_crc_zbc.c
MAINTAINERS +1 -0
@@ -339,6 +339,7 @@ F: config/riscv/ F: doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst F: lib/eal/riscv/ F: lib/hash/rte_crc_riscv64.h +F: lib/net/net_crc_zbc.c M: Sun Yuechi <[email protected]> F: lib/**/*rvv*
app/test/test_crc.c +10 -0
@@ -96,6 +96,16 @@ crc_all_algs(const char *desc, enum rte_net_crc_type type, }rte_net_crc_free (ctx); + ctx =rte_net_crc_set_alg (RTE_NET_CRC_ZBC , type); +TEST_ASSERT_NOT_NULL (ctx,"cannot allocate the CRC context" ); + crc =rte_net_crc_calc (ctx, data, data_len); +if (crc != res) { +RTE_LOG (ERR ,USER1 ,"TEST FAILED: %s ZBC\n" , desc); +debug_hexdump (stdout,"ZBC" , &crc,4 ); + ret =TEST_FAILED ; + } +rte_net_crc_free (ctx); +return ret; }
lib/net/meson.build +4 -0
@@ -57,4 +57,8 @@ elif (dpdk_conf.has('RTE_ARCH_ARM64') and cc.get_define('__ARM_FEATURE_CRYPTO', args: machine_args) != '') sources += files('net_crc_neon.c') cflags += ['-DCC_ARM64_NEON_PMULL_SUPPORT'] +elif (dpdk_conf.has('RTE_ARCH_RISCV') and + cc.get_define('RTE_RISCV_FEATURE_ZBC', args: machine_args) != '') + sources += files('net_crc_zbc.c') + cflags += ['-DCC_RISCV64_ZBC_CLMUL_SUPPORT'] endif
lib/net/net_crc.h +11 -0
@@ -44,4 +44,15 @@ rte_crc16_ccitt_neon_handler(const uint8_t *data, uint32_t data_len);uint32_t rte_crc32_eth_neon_handler (const uint8_t *data,uint32_t data_len); +/* RISCV64 Zbc */ +void +rte_net_crc_zbc_init (void ); + +uint32_t +rte_crc16_ccitt_zbc_handler (const uint8_t *data,uint32_t data_len); + +uint32_t +rte_crc32_eth_zbc_handler (const uint8_t *data,uint32_t data_len); + +#endif /* _NET_CRC_H_ */
lib/net/net_crc_zbc.c new file +194 -0
194 lines not on this page · raw message
lib/net/rte_net_crc.c +29 -1
@@ -40,7 +40,7 @@ struct rte_net_crc {static struct {rte_net_crc_handler f[RTE_NET_CRC_REQS ]; -} handlers[RTE_NET_CRC_AVX512 +1 ]; +} handlers[RTE_NET_CRC_ZBC +1 ];/* Scalar handling */
@@ -174,6 +174,20 @@ neon_pmull_init(void)#endif } +/* ZBC/CLMUL handling */ + +#define ZBC_CLMUL_CPU_SUPPORTED \ + rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_EXT_ZBC) + +static void +zbc_clmul_init (void ) +{ +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) +rte_net_crc_zbc_init (); +#endif +} +static void handlers_init (enum rte_net_crc_alg alg) {
@@ -205,6 +219,15 @@ handlers_init(enum rte_net_crc_alg alg) handlers[alg].f[RTE_NET_CRC32_ETH ] = rte_crc32_eth_neon_handler; break; } +#endif +break ; +case RTE_NET_CRC_ZBC : +#ifdef CC_RISCV64_ZBC_CLMUL_SUPPORT +if (ZBC_CLMUL_CPU_SUPPORTED ) { + handlers[alg].f[RTE_NET_CRC16_CCITT ] = rte_crc16_ccitt_zbc_handler; + handlers[alg].f[RTE_NET_CRC32_ETH ] = rte_crc32_eth_zbc_handler; +break ; + }#endif /* fall-through */ case RTE_NET_CRC_SCALAR :
@@ -248,6 +271,9 @@ struct rte_net_crc *rte_net_crc_set_alg(enum rte_net_crc_alg alg, enum rte_net_creturn crc; }break ; +case RTE_NET_CRC_ZBC : + crc->alg =RTE_NET_CRC_ZBC ; +return crc;case RTE_NET_CRC_SCALAR :/* fall-through */ default :
@@ -275,8 +301,10 @@ RTE_INIT(rte_net_crc_init)sse42_pclmulqdq_init ();avx512_vpclmulqdq_init ();neon_pmull_init (); +zbc_clmul_init ();handlers_init (RTE_NET_CRC_SCALAR );handlers_init (RTE_NET_CRC_NEON );handlers_init (RTE_NET_CRC_SSE42 );handlers_init (RTE_NET_CRC_AVX512 ); +handlers_init (RTE_NET_CRC_ZBC ); }
lib/net/rte_net_crc.h +3 -0
@@ -25,6 +25,7 @@ enum rte_net_crc_alg {RTE_NET_CRC_SSE42 ,RTE_NET_CRC_NEON ,RTE_NET_CRC_AVX512 , + RTE_NET_CRC_ZBC, };/** CRC context (algorithm, type) */
@@ -51,6 +52,8 @@ rte_net_crc_free(struct rte_net_crc *crc);* - RTE_NET_CRC_SSE42 (Use 64-bit SSE4.2 intrinsic) * - RTE_NET_CRC_NEON (Use ARM Neon intrinsic) * - RTE_NET_CRC_AVX512 (Use 512-bit AVX intrinsic) +* - RTE_NET_CRC_ZBC (Use RISC-V Zbc extension) +* * @param type * CRC type (enum rte_net_crc_type) *
Signature · 1 line
2.53.0
PATCH v4 4/10 config/riscv: add qemu crossbuild target
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
A new cross-compilation target that has extensions that DPDK uses and QEMU supports. This includes the vector extension V for ... and the Zbc extension for hardware CRC support.
--- config/riscv/meson.build | 1 + config/riscv/riscv64_qemu_linux_gcc | 17 +++++++++++++++++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 +++++ 3 files changed, 23 insertions(+) create mode 100644 config/riscv/riscv64_qemu_linux_gcc
config/riscv/meson.build +1 -0
@@ -45,6 +45,7 @@ vendor_generic = { 'arch_config': { 'generic': {'machine_args': ['-march=rv64gc']}, 'rv64gcv': {'machine_args': ['-march=rv64gcv']}, + 'qemu': {'machine_args': ['-march=rv64gcv_zbc']}, } }
config/riscv/riscv64_qemu_linux_gcc new file +17 -0
@@ -0,0 +1,17 @@ +[binaries] +c = ['ccache', 'riscv64-linux-gnu-gcc'] +cpp = ['ccache', 'riscv64-linux-gnu-g++'] +ar = 'riscv64-linux-gnu-ar' +strip = 'riscv64-linux-gnu-strip' +pcap-config = '' + +[host_machine] +system = 'linux' +cpu_family = 'riscv64' +cpu = 'rv64gcv_zbc' +endian = 'little' + +[properties] +vendor_id = 'generic' +arch_id = 'qemu' +pkg_config_libdir = '/usr/lib/riscv64-linux-gnu/pkgconfig'
doc/guides/linux_gsg/cross_build_dpdk_for_riscv.rst +5 -0
@@ -112,6 +112,11 @@ Currently the following targets are supported: * SiFive U740 SoC: ``config/riscv/riscv64_sifive_u740_linux_gcc`` +* QEMU: ``config/riscv/riscv64_qemu_linux_gcc`` + + * A target with all the extensions that QEMU supports that DPDK has a use for + (currently ``rv64gcv_zbc``). Requires QEMU version 7.0.0 or newer. + To add a new target support, ``config/riscv/meson.build`` has to be modified by adding a new vendor/architecture id and a corresponding cross-file has to be added to ``config/riscv`` directory.
Signature · 1 line
2.53.0
PATCH v4 5/10 examples/l3fwd: use accelerated CRC on riscv
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd/l3fwd_em.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd/l3fwd_em.c +1 -1
@@ -29,7 +29,7 @@#include "l3fwd_event.h" #include "em_route_parse.c" -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#define EM_HASH_CRC 1#endif
Signature · 1 line
2.53.0
PATCH v4 6/10 ipfrag: use accelerated CRC on riscv
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/ip_frag/ip_frag_internal.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
lib/ip_frag/ip_frag_internal.c +3 -3
@@ -45,14 +45,14 @@ ipv4_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *)&key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (key->id, v);#else v =rte_jhash_3words (p[0 ], p[1 ], key->id,PRIME_VALUE ); -#endif /* RTE_ARCH_X86 */ +#endif /* RTE_ARCH_X86 || RTE_ARCH_ARM64 || RTE_RISCV_FEATURE_ZBC */ *v1 = v; *v2 = (v <<7 ) + (v >>14 );
@@ -66,7 +66,7 @@ ipv6_frag_hash(const struct ip_frag_key *key, uint32_t *v1, uint32_t *v2) p = (const uint32_t *) &key->src_dst; -#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) +#if defined(RTE_ARCH_X86 ) || defined(RTE_ARCH_ARM64 ) || defined(RTE_RISCV_FEATURE_ZBC ) v =rte_hash_crc_4byte (p[0 ],PRIME_VALUE ); v =rte_hash_crc_4byte (p[1 ], v); v =rte_hash_crc_4byte (p[2 ], v);
Signature · 1 line
2.53.0
PATCH v4 7/10 examples/l3fwd-power: use accelerated CRC on riscv
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- examples/l3fwd-power/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
examples/l3fwd-power/main.c +1 -1
@@ -273,7 +273,7 @@ static struct rte_mempool * pktmbuf_pool[NB_SOCKETS];#if (APP_LOOKUP_METHOD ==APP_LOOKUP_EXACT_MATCH ) -#ifdef RTE_ARCH_X86 +#if defined(RTE_ARCH_X86 ) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define DEFAULT_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.53.0
PATCH v4 8/10 hash: use accelerated CRC on riscv
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/hash/rte_cuckoo_hash.c | 3 +++ lib/hash/rte_fbk_hash.c | 3 +++ 2 files changed, 6 insertions(+)
lib/hash/rte_cuckoo_hash.c +3 -0
@@ -460,6 +460,9 @@ rte_hash_create(const struct rte_hash_parameters *params)#elif defined(RTE_ARCH_ARM64 )if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_CRC32 )) default_hash_func = (rte_hash_function )rte_hash_crc; +#elif defined(RTE_ARCH_RISCV ) +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + default_hash_func = (rte_hash_function )rte_hash_crc;#endif /* Setup hash context */ strlcpy (h->name, params->name,sizeof (h->name));
lib/hash/rte_fbk_hash.c +3 -0
@@ -148,6 +148,9 @@ rte_fbk_hash_create(const struct rte_fbk_hash_params *params)#elif defined(RTE_ARCH_ARM64 )if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_CRC32 )) default_hash_func = (rte_fbk_hash_fn )rte_hash_crc_4byte; +#elif defined(RTE_ARCH_RISCV ) +if (rte_cpu_get_flag_enabled (RTE_CPUFLAG_RISCV_EXT_ZBC )) + default_hash_func = (rte_fbk_hash_fn )rte_hash_crc_4byte;#endif /* Set up hash table context. */
Signature · 1 line
2.53.0
PATCH v4 9/10 member: use accelerated CRC on riscv
In reply to Daniel GregoryFrom: Daniel Gregory <[email protected]>
When the RISC-V Zbc (carryless multiplication) extension is present, an implementation of CRC hashing using hardware instructions is available. Use it rather than jhash.
--- lib/member/member.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
lib/member/member.h +1 -1
@@ -12,7 +12,7 @@ extern int librte_member_logtype; "%s(): ", __func__, __VA_ARGS__)/* Hash function used by membership library. */ -#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) +#if defined(RTE_ARCH_X86 ) || defined(__ARM_FEATURE_CRC32) || defined(RTE_RISCV_FEATURE_ZBC )#include <rte_hash_crc.h> #define MEMBER_HASH_FUNC rte_hash_crc#else
Signature · 1 line
2.53.0
PATCH v4 10/10 doc: implement CRC using riscv carryless multiply
In reply to Daniel GregoryAdd to v26.03 release notes support for RISC-V Zbc extension to accelerate CRC using the Linux RISC-V hardware probing interface to detect its presence at runtime.
--- .mailmap | 2 +- doc/guides/rel_notes/release_26_03.rst | 8 ++++++++ 2 files changed, 9 insertions(+), 1 deletion(-)
.mailmap +1 -1
@@ -314,7 +314,7 @@ Dan Gora <[email protected]> Dan Nowlin <[email protected]> Dan Wei <[email protected]> Dana Vardi <[email protected]> -Daniel Gregory <[email protected]> +Daniel Gregory <[email protected]> Daniel Kaminsky <[email protected]> Daniel Kan <[email protected]> Daniel Martin Buckley <[email protected]>
doc/guides/rel_notes/release_26_03.rst +8 -0
@@ -106,6 +106,14 @@ New Features Added handling of the key combination Control+L to clear the screen before redisplaying the prompt. +* **Added RISC-V carryless multiplication support in CRC computation.** + + The RISC-V Zbc extension adds instructions for carryless multiplication. + Support for using this extension to accelerate CRC computation was added to + hash and net. This acceleration is enabled when the Zbc extension is present + on the build target and detected at runtime using the Linux RISC-V hardware + probing interface. + Removed Items -------------
Signature · 1 line
2.53.0
Quoted 122 lines · On Sun, 22 Feb 2026 16:29:54 +0100 Daniel Gregory <[email protected]> wrote:
The RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in
lib/hash/rte_crc_riscv64.hthat uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one inlib/net/net_crc_zbc.cthat uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implementrte_crc16_ccitt_zbc_handlerandrte_crc32_eth_zbc_handlerMy approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by
rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the
hash_functions_autotestbenchmark indpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |v4:
- rebase on 26.03-rc1
- RISC64 -> RISCV64 intest_hash.c(Stephen Hemminger)
- Added section to release notes (Stephen Hemminger)
- SPDX-License_Identifier -> SPDX-License-Identifier inrte_crc_riscv64.h(Stephen Hemminger)
- Fix header guard inrte_crc_riscv64.h(Stephen Hemminger)
- assert ->RTE_ASSERTinrte_crc_riscv64.h(Stephen Hemminger)
- Fix copyright statement innet_crc_zbc.c(Stephen Hemminger)
- Make crc context structs static innet_crc_zbc.c(Stephen Hemminger)
- prefer the optimised crc when zbc present over jhash inrte_fbk_hash.cv3:
- rebase on 24.07
- replace crc with CRC in commits (check-git-log.sh)
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in thenet/implementation that would cause segfaults on small unaligned buffers
- refactornet/implementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hashDaniel Gregory (10):
config/riscv: detect presence of Zbc extension
hash: implement CRC using riscv carryless multiply
net: implement CRC using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated CRC on riscv
ipfrag: use accelerated CRC on riscv examples/l3fwd-power: use accelerated CRC on riscv
hash: use accelerated CRC on riscv
member: use accelerated CRC on riscv
doc: implement CRC using riscv carryless multiply.mailmap | 2 +- MAINTAINERS | 2 + app/test/test_crc.c | 10 + app/test/test_hash.c | 7 + config/riscv/meson.build | 33 +++ config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + doc/guides/rel_notes/release_26_03.rst | 8 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 90 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_fbk_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/member.h | 2 +- lib/net/meson.build | 4 +
Quoted 9 lines
lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 194 ++++++++++++++++++ lib/net/rte_net_crc.c | 30 ++- lib/net/rte_net_crc.h | 3 + 25 files changed, 526 insertions(+), 42 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
AI patch review summary: the overall approach looks good — the hwprobe integration is clean and the Barrett reduction math appears correct. A few issues need addressing before this can be merged:
1. [ERROR, patch 01] 1 << n used for all 26 HWCAP mask entries
The feature table entries now store masks in a uint64_t field, but all 26 existing RISCV_ISA_* entries still use plain '1 << n' (signed int). This produces 32-bit values stored in a 64-bit field and causes undefined behaviour for n >= 31. All entries must use UINT64_C(1) << n:
FEAT_DEF(RISCV_ISA_A, REG_HWCAP, UINT64_C(1) << 0)
...
FEAT_DEF(RISCV_ISA_Z, REG_HWCAP, UINT64_C(1) << 25)2. [WARNING, patches 03-07, 09] Missing Signed-off-by from submitter address
These patches carry only:
but are submitted from [email protected]. The DCO requires a Signed-off-by from the address used to submit the patch. Please add:
to each of these patches (as done correctly in 01, 08, and 10).
3. [WARNING, patch 02] Inverted condition in rte_hash_crc_set_alg
The new warning log fires when the caller does *not* request CRC32_RISCV64, but the message says the opposite:
if (!(alg & CRC32_RISCV64))
HASH_CRC_LOG(WARNING, "Unsupported CRC32 algorithm requested
using CRC32_RISCV64"); Either flip the condition or reword the message to match the intent (e.g. "Falling back to CRC32_RISCV64 despite a different algorithm being requested").
4. [WARNING, patch 03] Unaligned access in crc32_repeated_barrett_zbc
The function casts 'data' directly to uint64_t*/uint32_t*/uint16_t* and dereferences it. It is called for tail data (after the main fold loop and for buffers < 16 bytes) where alignment is not guaranteed. RISC-V hardware unaligned access support is optional. Use memcpy into a local variable or equivalent to be safe on all implementations.
5. [WARNING, patch 10] .mailmap replaces rather than aliases old address
The change removes the bytedance entry entirely. The correct form maps the old address to the new canonical one:
Daniel Gregory <[email protected]> <[email protected]>
Without this, existing commits with the bytedance address will no longer be attributed to the canonical identity in git shortlog.
Minor: the Barrett reduction in rte_crc_riscv64.h and net_crc_zbc.c both truncate a uint64_t result to uint32_t implicitly on return. A brief comment explaining this is intentional (only the lower 32 bits are the CRC remainder) would help future readers.
Quoted 33 lines
From: Stephen Hemminger [mailto:[email protected]]
Sent: Sunday, 22 February 2026 19.03Quoted 1 line · On Sun, 22 Feb 2026 16:29:54 +0100 Daniel Gregory <[email protected]> wrote:
The RISC-V Zbc extension adds instructions for carry-less
multiplication
Quoted 1 line
we can use to implement CRC in hardware. This patch set contains two
new
Quoted 5 lines
implementations:
- one in
lib/hash/rte_crc_riscv64.hthat uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one inlib/net/net_crc_zbc.cthat uses repeated single-folds toreduce
Quoted 3 lines
the buffer until it is small enough for a Barrett reduction to implement
rte_crc16_ccitt_zbc_handlerandrte_crc32_eth_zbc_handlerAI patch review summary: the overall approach looks good — the hwprobe integration is clean and the Barrett reduction math appears correct. A few issues need addressing before this can be merged:
1. [ERROR, patch 01] 1 << n used for all 26 HWCAP mask entries
The feature table entries now store masks in a
uint64_tfield, but all 26 existing RISCV_ISA_* entries still use plain '1 << n' (signed int). This produces 32-bit values stored in a 64-bit field and
causes
undefined behaviour for n >= 31. All entries must useUINT64_C(1)<<
n:
Quoted 3 lines
FEAT_DEF(RISCV_ISA_A, REG_HWCAP, UINT64_C(1) << 0)
...
FEAT_DEF(RISCV_ISA_Z, REG_HWCAP, UINT64_C(1) << 25)Furthermore, RTE_BIT64(25) [1] is preferred over UINT64_C(1) << 25.
[1]: https://elixir.bootlin.com/dpdk/v25.11/source/lib/eal/include/rte_bitops.h#L36
It seems the AI missed this preference.
Hi Daniel,
I noticed that my recent patch [1] overlaps with your patch 01/10 -- both modify rte_cpuflags.c and rte_cpuflags.h to add hwprobe-based Z sub-extension detection.
The key differences are:
* Scope: my patch adds infrastructure for all Z sub-extensions currently defined in Linux hwprobe.h; your patch 01/10 adds only ZBC as required by the CRC series.
* Caching: my patch calls hwprobe once at EAL init via RTE_INIT and caches the result; your patch currently appears to call it on every invocation of rte_cpu_get_features().
* FEAT_DEF semantics: my patch keeps the existing convention where the third argument is a bit index; your patch changes it to a mask value and updates rte_cpu_get_flag_enabled() accordingly. Mixing the two would produce silent incorrect results, so this needs to be resolved regardless of merge order.
* Enum naming: I use RTE_CPUFLAG_RISCV_ZBC; you use RTE_CPUFLAG_RISCV_EXT_ZBC.
Regardless of which patch lands first, the other can be simplified:
* If mine lands first, your patch 01/10 can drop the hwprobe infrastructure and keep only the meson build-time detection logic (RTE_RISCV_FEATURE_ZBC). Your CRC patches (02/10 onwards) would be unaffected in substance -- they would just reference RTE_CPUFLAG_RISCV_ZBC instead of RTE_CPUFLAG_RISCV_EXT_ZBC.
* If yours lands first, I will rebase on top of it and align with your FEAT_DEF semantics and enum naming.
On naming: I prefer the flat style (RTE_CPUFLAG_RISCV_ZBC) since it is consistent with the existing RTE_CPUFLAG_RISCV_ISA_* entries, but I will follow whatever the maintainer decides.
I would be happy to coordinate on any of the above. Please let me know your thoughts.
[1] https://patchwork.dpdk.org/patch/162581/
Best regards,
dangshiwei
Quoted 122 lines · On Sun, 22 Feb 2026 16:29:54 +0100 Daniel Gregory <[email protected]> wrote:
The RISC-V Zbc extension adds instructions for carry-less multiplication we can use to implement CRC in hardware. This patch set contains two new implementations:
- one in
lib/hash/rte_crc_riscv64.hthat uses a Barrett reduction to implement the four rte_hash_crc_* functions
- one inlib/net/net_crc_zbc.cthat uses repeated single-folds to reduce the buffer until it is small enough for a Barrett reduction to implementrte_crc16_ccitt_zbc_handlerandrte_crc32_eth_zbc_handlerMy approach is largely based on the Intel's "Fast CRC Computation Using PCLMULQDQ Instruction" white paper
https://www.researchgate.net/publication/263424619_Fast_CRC_computation
and a post about "Optimizing CRC32 for small payload sizes on x86"
https://mary.rs/lab/crc32/Whether these new implementations are enabled is controlled by new build-time and run-time detection of the RISC-V extensions present in the compiler and on the target system.
I have carried out some performance comparisons between the generic table implementations and the new hardware implementations. Listed below is the number of cycles it takes to compute the CRC hash for buffers of various sizes (as reported by
rte_get_timer_cycles()). These results were collected on a Kendryte K230 and averaged over 20 samples:|Buffer | CRC32-ETH (lib/net) | CRC32C (lib/hash) | |Size (MB) | Table | Hardware | Table | Hardware | |----------|----------|----------|----------|----------| | 1 | 155168 | 11610 | 73026 | 18385 | | 2 | 311203 | 22998 | 145586 | 35886 | | 3 | 466744 | 34370 | 218536 | 53939 | | 4 | 621843 | 45536 | 291574 | 71944 | | 5 | 777908 | 56989 | 364152 | 89706 | | 6 | 932736 | 68023 | 437016 | 107726 | | 7 | 1088756 | 79236 | 510197 | 125426 | | 8 | 1243794 | 90467 | 583231 | 143614 |These results suggest a speed-up of lib/net by thirteen times, and of lib/hash by four times.
I have also run the
hash_functions_autotestbenchmark indpdk_test, which measures the performance of the lib/hash implementation on small buffers, getting the following times:| Key Length | Time (ticks/op) | | (bytes) | Table | Hardware | |------------|----------|----------| | 1 | 0.47 | 0.85 | | 2 | 0.57 | 0.87 | | 4 | 0.99 | 0.88 | | 8 | 1.35 | 0.88 | | 9 | 1.20 | 1.09 | | 13 | 1.76 | 1.35 | | 16 | 1.87 | 1.02 | | 32 | 2.96 | 0.98 | | 37 | 3.35 | 1.45 | | 40 | 3.49 | 1.12 | | 48 | 4.02 | 1.25 | | 64 | 5.08 | 1.54 |v4:
- rebase on 26.03-rc1
- RISC64 -> RISCV64 intest_hash.c(Stephen Hemminger)
- Added section to release notes (Stephen Hemminger)
- SPDX-License_Identifier -> SPDX-License-Identifier inrte_crc_riscv64.h(Stephen Hemminger)
- Fix header guard inrte_crc_riscv64.h(Stephen Hemminger)
- assert ->RTE_ASSERTinrte_crc_riscv64.h(Stephen Hemminger)
- Fix copyright statement innet_crc_zbc.c(Stephen Hemminger)
- Make crc context structs static innet_crc_zbc.c(Stephen Hemminger)
- prefer the optimised crc when zbc present over jhash inrte_fbk_hash.cv3:
- rebase on 24.07
- replace crc with CRC in commits (check-git-log.sh)
v2:
- replace compile flag with build-time (riscv extension macros) and run-time detection (linux hwprobe syscall) (Stephen Hemminger)
- add qemu target that supports zbc (Stanislaw Kardach)
- fix spelling error in commit message
- fix a bug in thenet/implementation that would cause segfaults on small unaligned buffers
- refactornet/implementation to move variable declarations to top of functions
- enable the optimisation in a couple other places optimised crc is preferred to jhash
- l3fwd-power
- cuckoo-hashDaniel Gregory (10):
config/riscv: detect presence of Zbc extension
hash: implement CRC using riscv carryless multiply
net: implement CRC using riscv carryless multiply config/riscv: add qemu crossbuild target examples/l3fwd: use accelerated CRC on riscv
ipfrag: use accelerated CRC on riscv examples/l3fwd-power: use accelerated CRC on riscv
hash: use accelerated CRC on riscv
member: use accelerated CRC on riscv
doc: implement CRC using riscv carryless multiply.mailmap | 2 +- MAINTAINERS | 2 + app/test/test_crc.c | 10 + app/test/test_hash.c | 7 + config/riscv/meson.build | 33 +++ config/riscv/riscv64_qemu_linux_gcc | 17 ++ .../linux_gsg/cross_build_dpdk_for_riscv.rst | 5 + doc/guides/rel_notes/release_26_03.rst | 8 + examples/l3fwd-power/main.c | 2 +- examples/l3fwd/l3fwd_em.c | 2 +- lib/eal/riscv/include/rte_cpuflags.h | 2 + lib/eal/riscv/rte_cpuflags.c | 112 +++++++--- lib/hash/meson.build | 1 + lib/hash/rte_crc_riscv64.h | 90 ++++++++ lib/hash/rte_cuckoo_hash.c | 3 + lib/hash/rte_fbk_hash.c | 3 + lib/hash/rte_hash_crc.c | 13 +- lib/hash/rte_hash_crc.h | 6 +- lib/ip_frag/ip_frag_internal.c | 6 +- lib/member/member.h | 2 +- lib/net/meson.build | 4 +
Quoted 9 lines
lib/net/net_crc.h | 11 + lib/net/net_crc_zbc.c | 194 ++++++++++++++++++ lib/net/rte_net_crc.c | 30 ++- lib/net/rte_net_crc.h | 3 + 25 files changed, 526 insertions(+), 42 deletions(-) create mode 100644 config/riscv/riscv64_qemu_linux_gcc create mode 100644 lib/hash/rte_crc_riscv64.h create mode 100644 lib/net/net_crc_zbc.c
Since don't have riscv hardware or detailed CPU knowledge, turned to AI review for help. Looks like more work is needed. Not all AI comments are correct; please look at in detail.
Summary of Findings
The patch series correctly introduces RISC-V Zbc (carryless multiplication) support for CRC acceleration in both hash and net libraries, including runtime detection via the Linux hwprobe interface. However, I identified a significant alignment bug in the buffer processing logic and a missing header that will likely break the build on some systems.
---
Correctness Bugs (High Priority)
1. Alignment Logic Error in crc32_eth_calc_zbc (lib/net/net_crc_zbc.c)
The code attempts to align the input buffer to an 8-byte boundary before performing 64-bit wide operations, but the calculation is incorrect:
1 + /* Barrett reduce until buffer aligned to 8-byte word */
2 + uint32_t misalign = (size_t)data & 7;
3 + if (misalign != 0 && misalign <= data_len) {
4 + crc = crc32_repeated_barrett_zbc(data, misalign, crc, params);
5 + data += misalign;
6 + data_len -= misalign;7 + }
Issue: If data is 0x...001 (1-byte misaligned), misalign is 1. After processing 1 byte and incrementing data, the new address is 0x...002, which is still not 8-byte aligned. This leads to misaligned 64-bit loads (*(const uint64_t *)data) later in the function, which can cause performance degradation or traps depending on the RISC-V implementation.
Recommendation: The number of bytes to process to reach alignment should be (8 - ((uintptr_t)data & 7)) & 7.
2. Missing Header for cpu_set_t (lib/eal/riscv/rte_cpuflags.c)
In Patch 01/10, the function rte_cpu_hwprobe_ima_ext introduces the use of cpu_set_t:
1 + /* empty set of cpus returns extensions present on all cpus */
2 + cpu_set_t *cpus = NULL;
Issue: lib/eal/riscv/rte_cpuflags.c does not include <sched.h>, which is required for the cpu_set_t definition. This will cause a compilation error.
Recommendation: Add #include <sched.h> to the EAL CPU flags implementation.
3. Algorithm Overriding in rte_hash_crc_set_alg (lib/hash/rte_hash_crc.c)
The implementation overrides the user's requested algorithm if the Zbc extension is detected:
1 + if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_RISCV_EXT_ZBC)) 2 + rte_hash_crc32_alg = CRC32_RISCV64;
Issue: If a user explicitly calls rte_hash_crc_set_alg(CRC32_SW), the code will still force the use of CRC32_RISCV64. While this matches the existing (and arguably flawed) pattern for ARM64, it violates the expectation that set_alg respects the user's choice.
---
Style and Process Compliance
4. Inconsistent Assertion Macros
- Patch 02/10 uses RTE_ASSERT in lib/hash/rte_crc_riscv64.h.
- Patch 03/10 uses standard C assert() in lib/net/net_crc_zbc.c.
Recommendation: Use RTE_ASSERT consistently across the codebase per DPDK standards.
5. Implicit Pointer Comparison Nits
While most comparisons are explicit, the project style generally favors if (ptr == NULL) over if (!ptr). There are a few instances in Patch 01 (Meson logic) and Patch 02 (intrinsics checks) that could be more explicit, though they mostly follow the local conventions of those specific files.
6. Barrett Reduction Constants
The constants mu and p in lib/hash/rte_crc_riscv64.h are lowercase. While these are static const variables, global/static constants in DPDK are often written in ALL_UPPERCASE.
---
Additional Observations
- Toolchain Requirements: The Meson logic correctly identifies that GCC 14.1+ or Clang 18.1+ is required for the Zbc intrinsics.
- Cross-compilation: The addition of a QEMU cross-build target is a helpful addition for testing these extensions in virtualized environments.
- Hardware Probing: The use of the Linux hwprobe syscall is the correct modern approach for RISC-V extension discovery.