Re: [dpdk-dev] [PATCH v4 3/5] hash: add fallback to software CRC32 implementation

[Neil Horman <nhorman@tuxdriver.com>]

On Tue, Nov 18, 2014 at 04:04:26PM +0000, Bruce Richardson wrote:
> On Tue, Nov 18, 2014 at 11:00:05AM -0500, Neil Horman wrote:
> > On Tue, Nov 18, 2014 at 09:06:35PM +0600, Yerden Zhumabekov wrote:
> > > 
> > > 18.11.2014 20:41, Neil Horman пишет:
> > > > On Tue, Nov 18, 2014 at 08:03:40PM +0600, Yerden Zhumabekov wrote:
> > > >> Initially, SSE4.2 support is detected via CPUID instruction.
> > > >>
> > > >> Added rte_hash_crc_set_alg() function to detect and set CRC32
> > > >> implementation if necessary. SSE4.2 is allowed by default. If it's
> > > >> not available, fall back to sw implementation.
> > > >>
> > > >> Best available algorithm is detected upon application startup
> > > >> through the constructor function rte_hash_crc_try_sse442().
> > > >>
> > > >> Signed-off-by: Yerden Zhumabekov <e_zhumabekov@sts.kz>
> > > >> ---
> > > >>  lib/librte_hash/rte_hash_crc.h |   53 ++++++++++++++++++++++++++++++++++++++--
> > > >>  1 file changed, 51 insertions(+), 2 deletions(-)
> > > >>
> > > >> diff --git a/lib/librte_hash/rte_hash_crc.h b/lib/librte_hash/rte_hash_crc.h
> > > >> index 15f687a..332ed99 100644
> > > >> --- a/lib/librte_hash/rte_hash_crc.h
> > > >> +++ b/lib/librte_hash/rte_hash_crc.h
> > > >> @@ -45,7 +45,11 @@ extern "C" {
> > > >>  #endif
> > > >>  
> > > >>  #include <stdint.h>
> > > >> +#ifdef RTE_MACHINE_CPUFLAG_SSE4_2
> > > >>  #include <nmmintrin.h>
> > > >> +#endif
> > > >> +#include <rte_cpuflags.h>
> > > >> +#include <rte_branch_prediction.h>
> > > >>  
> > > >>  /* Lookup tables for software implementation of CRC32C */
> > > >>  static uint32_t crc32c_tables[8][256] = {{
> > > >> @@ -363,8 +367,41 @@ crc32c_2words(uint64_t data, uint32_t init_val)
> > > >>  	return crc;
> > > >>  }
> > > >>  
> > > >> +enum crc32_alg_t {
> > > >> +	CRC32_SW = 0,
> > > >> +	CRC32_SSE42
> > > >> +};
> > > >> +
> > > >> +static enum crc32_alg_t crc32_alg;
> > > >> +
> > > >> +/**
> > > >> + * Allow or disallow use of SSE4.2 instrinsics for CRC32 hash
> > > >> + * calculation.
> > > >> + *
> > > >> + * @param flag
> > > >> + *   unsigned integer flag
> > > >> + *   - (CRC32_SW) Don't use SSE4.2 intrinsics
> > > >> + *   - (CRC32_SSE42) Use SSE4.2 intrinsics if available, set by default
> > > >> + */
> > > >> +static inline void
> > > >> +rte_hash_crc_set_alg(enum crc32_alg_t alg)
> > > >> +{
> > > >> +	int sse42_supp = rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_2);
> > > >> +	enum crc32_alg_t alg_supp = sse42_supp ? CRC32_SSE42 : CRC32_SW;
> > > >> +	crc32_alg = (alg == CRC32_SSE42) ? alg_supp : CRC32_SW;
> > > >> +}
> > > >> +
> > > >> +/* Best available algorithm is detected via CPUID instruction */
> > > >> +static inline void __attribute__((constructor))
> > > >> +rte_hash_crc_try_sse42(void)
> > > >> +{
> > > >> +	rte_hash_crc_set_alg(CRC32_SSE42);
> > > >> +}
> > > >> +
> > > >>  /**
> > > >>   * Use single crc32 instruction to perform a hash on a 4 byte value.
> > > >> + * Fall back to software crc32 implementation in case SSE4.2 is
> > > >> + * not supported
> > > >>   *
> > > >>   * @param data
> > > >>   *   Data to perform hash on.
> > > >> @@ -376,11 +413,18 @@ crc32c_2words(uint64_t data, uint32_t init_val)
> > > >>  static inline uint32_t
> > > >>  rte_hash_crc_4byte(uint32_t data, uint32_t init_val)
> > > >>  {
> > > >> -	return _mm_crc32_u32(init_val, data);
> > > >> +#ifdef RTE_MACHINE_CPUFLAG_SSE4_2
> > > >> +	if (likely(crc32_alg == CRC32_SSE42))
> > > >> +		return _mm_crc32_u32(init_val, data);
> > > >> +#endif
> > > > you don't really need these ifdefs here anymore given that you have a
> > > > constructor to do the algorithm selection.  In fact you need to remove them, in
> > > > the event you build on a system that doesn't support SSE42, but run on a system
> > > > that does.
> > > 
> > > Originally, I thought so as well. I wrote the code without these ifdefs,
> > > but it didn't compile on my machine which doesn't support SSE4.2. Error
> > > was triggered by nmmintrin.h which has a check for respective GCC
> > > extension. So I think these ifdefs are indeed required.
> > > 
> > You need to edit the makefile so that the compiler gets passed the option
> > -msse42.  That way it will know to emit sse42 instructions. It will also allow
> > you to remove the ifdef from the include file
> > Neil
> > 
> 
> Question: does that then limit the compiler to emitting sse42 instructions? If,
> for instance, the rest of DPDK is being compiled for a target supporting AVX2,
> does that flag then prevent the compiler from auto-vectorising using SSE?
> 
It should be a last-option-wins model I think (that is to say if the command
line specifies -msse42 and -march=core-avx2, then you will get both instructions,
since core-avx2 supports sse42.  

Neil

> /Bruce
> 
>