DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Ananyev, Konstantin" <konstantin.ananyev@intel.com>
To: Neil Horman <nhorman@tuxdriver.com>, "dev@dpdk.org" <dev@dpdk.org>
Subject: Re: [dpdk-dev] [PATCHv3] librte_acl make it build/work for 'default' target
Date: Mon, 25 Aug 2014 16:30:05 +0000	[thread overview]
Message-ID: <2601191342CEEE43887BDE71AB9772582135D369@IRSMSX105.ger.corp.intel.com> (raw)
In-Reply-To: <1408652100-29217-1-git-send-email-nhorman@tuxdriver.com>

Hi Neil,

> -----Original Message-----
> From: Neil Horman [mailto:nhorman@tuxdriver.com]
> Sent: Thursday, August 21, 2014 9:15 PM
> To: dev@dpdk.org
> Cc: Ananyev, Konstantin; thomas.monjalon@6wind.com; Neil Horman
> Subject: [PATCHv3] librte_acl make it build/work for 'default' target
> 
> Make ACL library to build/work on 'default' architecture:
> - make rte_acl_classify_scalar really scalar
>  (make sure it wouldn't use sse4 instrincts through resolve_priority()).
> - Provide two versions of rte_acl_classify code path:
>   rte_acl_classify_sse() - could be build and used only on systems with sse4.2
>   and upper, return -ENOTSUP on lower arch.
>   rte_acl_classify_scalar() - a slower version, but could be build and used
>   on all systems.
> - keep common code shared between these two codepaths.
> 
> v2 chages:
>  run-time selection of most appropriate code-path for given ISA.
>  By default the highest supprted one is selected.
>  User can still override that selection by manually assigning new value to
>  the global function pointer rte_acl_default_classify.
>  rte_acl_classify() becomes a macro calling whatever rte_acl_default_classify
>  points to.
> 

I see you decided not to wait for me and fix everything by yourself :)

> V3 Changes
>  Updated classify pointer to be a function so as to better preserve ABI

As I said in my previous mail it generates extra jump...
Though from numbers I got the performance impact is negligible: < 1%.
So I suppose, I don't have a good enough reason to object :)

Though I still think we better keep  rte_acl_classify_scalar() publically available (same as we do for rte acl_classify_sse()):
First of all keep  rte_acl_classify_scalar() is already part of our public API.
Also, as I remember, one of the customers explicitly asked for scalar version and they planned to call it directly.
Plus using rte_acl_select_classify() to always switch between implementations is not always handy:
-  it is global, which means that we can't simultaneously use classify_scalar() and classify_sse() for 2 different ACL contexts.  
- to properly support such switching we then will need to support something like (see app/test/test_acl.c below):
  old_alg = rte_acl_get_classify();
  rte_acl_select_classify(new_alg);
  ...
  rte_acl_select_classify(old_alg); 
  
>  REmoved macro definitions for match check functions to make them static inline

More comments inlined below.

Thanks
Konstantin

> 
> Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
> ---
>  app/test-acl/main.c              |  13 +-
>  app/test/test_acl.c              |  12 +-
>  lib/librte_acl/Makefile          |   5 +-
>  lib/librte_acl/acl_bld.c         |   5 +-
>  lib/librte_acl/acl_match_check.h |  83 ++++
>  lib/librte_acl/acl_run.c         | 944 ---------------------------------------
>  lib/librte_acl/acl_run.h         | 220 +++++++++
>  lib/librte_acl/acl_run_scalar.c  | 198 ++++++++
>  lib/librte_acl/acl_run_sse.c     | 627 ++++++++++++++++++++++++++
>  lib/librte_acl/rte_acl.c         |  46 ++
>  lib/librte_acl/rte_acl.h         |  26 +-
>  11 files changed, 1216 insertions(+), 963 deletions(-)
>  create mode 100644 lib/librte_acl/acl_match_check.h
>  delete mode 100644 lib/librte_acl/acl_run.c
>  create mode 100644 lib/librte_acl/acl_run.h
>  create mode 100644 lib/librte_acl/acl_run_scalar.c
>  create mode 100644 lib/librte_acl/acl_run_sse.c
> 
> diff --git a/app/test-acl/main.c b/app/test-acl/main.c
> index d654409..a77f47d 100644
> --- a/app/test-acl/main.c
> +++ b/app/test-acl/main.c
> @@ -787,6 +787,10 @@ acx_init(void)
>  	/* perform build. */
>  	ret = rte_acl_build(config.acx, &cfg);
> 
> +	/* setup default rte_acl_classify */
> +	if (config.scalar)
> +		rte_acl_select_classify(ACL_CLASSIFY_SCALAR);
> +
>  	dump_verbose(DUMP_NONE, stdout,
>  		"rte_acl_build(%u) finished with %d\n",
>  		config.bld_categories, ret);
> @@ -815,13 +819,8 @@ search_ip5tuples_once(uint32_t categories, uint32_t step, int scalar)
>  			v += config.trace_sz;
>  		}
> 
> -		if (scalar != 0)
> -			ret = rte_acl_classify_scalar(config.acx, data,
> -				results, n, categories);
> -
> -		else
> -			ret = rte_acl_classify(config.acx, data,
> -				results, n, categories);
> +		ret = rte_acl_classify(config.acx, data, results,
> +			n, categories);
> 
>  		if (ret != 0)
>  			rte_exit(ret, "classify for ipv%c_5tuples returns %d\n",
> diff --git a/app/test/test_acl.c b/app/test/test_acl.c
> index 869f6d3..2fcef6e 100644
> --- a/app/test/test_acl.c
> +++ b/app/test/test_acl.c
> @@ -148,7 +148,8 @@ test_classify_run(struct rte_acl_ctx *acx)
>  	}
> 
>  	/* make a quick check for scalar */
> -	ret = rte_acl_classify_scalar(acx, data, results,
> +	rte_acl_select_classify(ACL_CLASSIFY_SCALAR);
> +	ret = rte_acl_classify(acx, data, results,
>  			RTE_DIM(acl_test_data), RTE_ACL_MAX_CATEGORIES);


As I said above, that doesn't seem correct: we set rte_acl_default_classify = rte_acl_classify_scalar and never restore it back to the original value.
To support it properly, we need to:
old_alg = rte_acl_get_classify();
 rte_acl_select_classify(new_alg);
 ...
 rte_acl_select_classify(old_alg);

Make all this just to keep UT valid seems like a big hassle to me.
So I said above - probably better just leave it to call rte_acl_classify_scalar() directly.

>  	if (ret != 0) {
>  		printf("Line %i: SSE classify failed!\n", __LINE__);
> @@ -362,7 +363,8 @@ test_invalid_layout(void)
>  	}
> 
>  	/* classify tuples (scalar) */
> -	ret = rte_acl_classify_scalar(acx, data, results,
> +	rte_acl_select_classify(ACL_CLASSIFY_SCALAR);
> +	ret = rte_acl_classify(acx, data, results,
>  			RTE_DIM(results), 1);
>  	if (ret != 0) {
>  		printf("Line %i: Scalar classify failed!\n", __LINE__);
> @@ -850,7 +852,8 @@ test_invalid_parameters(void)
>  	/* scalar classify test */
> 
>  	/* cover zero categories in classify (should not fail) */
> -	result = rte_acl_classify_scalar(acx, NULL, NULL, 0, 0);
> +	rte_acl_select_classify(ACL_CLASSIFY_SCALAR);
> +	result = rte_acl_classify(acx, NULL, NULL, 0, 0);
>  	if (result != 0) {
>  		printf("Line %i: Scalar classify with zero categories "
>  				"failed!\n", __LINE__);
> @@ -859,7 +862,8 @@ test_invalid_parameters(void)
>  	}
> 
>  	/* cover invalid but positive categories in classify */
> -	result = rte_acl_classify_scalar(acx, NULL, NULL, 0, 3);
> +	rte_acl_select_classify(ACL_CLASSIFY_SCALAR);
> +	result = rte_acl_classify(acx, NULL, NULL, 0, 3);
>  	if (result == 0) {
>  		printf("Line %i: Scalar classify with 3 categories "
>  				"should have failed!\n", __LINE__);
> diff --git a/lib/librte_acl/Makefile b/lib/librte_acl/Makefile
> index 4fe4593..65e566d 100644
> --- a/lib/librte_acl/Makefile
> +++ b/lib/librte_acl/Makefile
> @@ -43,7 +43,10 @@ SRCS-$(CONFIG_RTE_LIBRTE_ACL) += tb_mem.c
>  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += rte_acl.c
>  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_bld.c
>  SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_gen.c
> -SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run.c
> +SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_scalar.c
> +SRCS-$(CONFIG_RTE_LIBRTE_ACL) += acl_run_sse.c
> +
> +CFLAGS_acl_run_sse.o += -msse4.1
> 
>  # install this header file
>  SYMLINK-$(CONFIG_RTE_LIBRTE_ACL)-include := rte_acl_osdep.h
> diff --git a/lib/librte_acl/acl_bld.c b/lib/librte_acl/acl_bld.c
> index 873447b..09d58ea 100644
> --- a/lib/librte_acl/acl_bld.c
> +++ b/lib/librte_acl/acl_bld.c
> @@ -31,7 +31,6 @@
>   *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
>   */
> 
> -#include <nmmintrin.h>
>  #include <rte_acl.h>
>  #include "tb_mem.h"
>  #include "acl.h"
> @@ -1480,8 +1479,8 @@ acl_calc_wildness(struct rte_acl_build_rule *head,
> 
>  			switch (rule->config->defs[n].type) {
>  			case RTE_ACL_FIELD_TYPE_BITMASK:
> -				wild = (size -
> -					_mm_popcnt_u32(fld->mask_range.u8)) /
> +				wild = (size - __builtin_popcount(
> +					fld->mask_range.u8)) /
>  					size;
>  				break;
> 
> diff --git a/lib/librte_acl/acl_match_check.h b/lib/librte_acl/acl_match_check.h
> new file mode 100644
> index 0000000..4dc1982
> --- /dev/null
> +++ b/lib/librte_acl/acl_match_check.h

As a nit: we probably don't need a special header just for one function and can place it inside acl_run.h.

> @@ -0,0 +1,83 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef _ACL_MATCH_CHECK_H_
> +#define _ACL_MATCH_CHECK_H_
> +
> +/*
> + * Detect matches. If a match node transition is found, then this trie
> + * traversal is complete and fill the slot with the next trie
> + * to be processed.
> + */
> +static inline uint64_t
> +acl_match_check(uint64_t transition, int slot,
> +	const struct rte_acl_ctx *ctx, struct parms *parms,
> +	struct acl_flow_data *flows, void (*resolve_priority)(
> +	uint64_t transition, int n, const struct rte_acl_ctx *ctx,
> +	struct parms *parms, const struct rte_acl_match_results *p,
> +	uint32_t categories))

Ugh, that's really hard to read.
Can we create a typedef for resolve_priority function type:
typedef void (*resolve_priority_t)(uint64_t, int,
        const struct rte_acl_ctx *ctx, struct parms *,
        const struct rte_acl_match_results *, uint32_t);
And use it here?

> +{
> +	const struct rte_acl_match_results *p;
> +
> +	p = (const struct rte_acl_match_results *)
> +		(flows->trans + ctx->match_index);
> +
> +	if (transition & RTE_ACL_NODE_MATCH) {
> +
> +		/* Remove flags from index and decrement active traversals */
> +		transition &= RTE_ACL_NODE_INDEX;
> +		flows->started--;
> +
> +		/* Resolve priorities for this trie and running results */
> +		if (flows->categories == 1)
> +			resolve_single_priority(transition, slot, ctx,
> +				parms, p);
> +		else
> +			resolve_priority(transition, slot, ctx, parms,
> +				p, flows->categories);
> +
> +		/* Count down completed tries for this search request */
> +		parms[slot].cmplt->count--;
> +
> +		/* Fill the slot with the next trie or idle trie */
> +		transition = acl_start_next_trie(flows, parms, slot, ctx);
> +
> +	} else if (transition == ctx->idle) {
> +		/* reset indirection table for idle slots */
> +		parms[slot].data_index = idle;
> +	}
> +
> +	return transition;
> +}
> +
> +#endif
> diff --git a/lib/librte_acl/acl_run.c b/lib/librte_acl/acl_run.c
> deleted file mode 100644
> index e3d9fc1..0000000
> --- a/lib/librte_acl/acl_run.c
> +++ /dev/null
> @@ -1,944 +0,0 @@
> -/*-
> - *   BSD LICENSE
> - *
> - *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> - *   All rights reserved.
> - *
> - *   Redistribution and use in source and binary forms, with or without
> - *   modification, are permitted provided that the following conditions
> - *   are met:
> - *
> - *     * Redistributions of source code must retain the above copyright
> - *       notice, this list of conditions and the following disclaimer.
> - *     * Redistributions in binary form must reproduce the above copyright
> - *       notice, this list of conditions and the following disclaimer in
> - *       the documentation and/or other materials provided with the
> - *       distribution.
> - *     * Neither the name of Intel Corporation nor the names of its
> - *       contributors may be used to endorse or promote products derived
> - *       from this software without specific prior written permission.
> - *
> - *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> - *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> - *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> - *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> - *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> - *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> - *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> - *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> - *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> - *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> - *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> - */
> -
> -#include <rte_acl.h>
> -#include "acl_vect.h"
> -#include "acl.h"
> -
> -#define MAX_SEARCHES_SSE8	8
> -#define MAX_SEARCHES_SSE4	4
> -#define MAX_SEARCHES_SSE2	2
> -#define MAX_SEARCHES_SCALAR	2
> -
> -#define GET_NEXT_4BYTES(prm, idx)	\
> -	(*((const int32_t *)((prm)[(idx)].data + *(prm)[idx].data_index++)))
> -
> -
> -#define RTE_ACL_NODE_INDEX	((uint32_t)~RTE_ACL_NODE_TYPE)
> -
> -#define	SCALAR_QRANGE_MULT	0x01010101
> -#define	SCALAR_QRANGE_MASK	0x7f7f7f7f
> -#define	SCALAR_QRANGE_MIN	0x80808080
> -
> -enum {
> -	SHUFFLE32_SLOT1 = 0xe5,
> -	SHUFFLE32_SLOT2 = 0xe6,
> -	SHUFFLE32_SLOT3 = 0xe7,
> -	SHUFFLE32_SWAP64 = 0x4e,
> -};
> -
> -/*
> - * Structure to manage N parallel trie traversals.
> - * The runtime trie traversal routines can process 8, 4, or 2 tries
> - * in parallel. Each packet may require multiple trie traversals (up to 4).
> - * This structure is used to fill the slots (0 to n-1) for parallel processing
> - * with the trie traversals needed for each packet.
> - */
> -struct acl_flow_data {
> -	uint32_t            num_packets;
> -	/* number of packets processed */
> -	uint32_t            started;
> -	/* number of trie traversals in progress */
> -	uint32_t            trie;
> -	/* current trie index (0 to N-1) */
> -	uint32_t            cmplt_size;
> -	uint32_t            total_packets;
> -	uint32_t            categories;
> -	/* number of result categories per packet. */
> -	/* maximum number of packets to process */
> -	const uint64_t     *trans;
> -	const uint8_t     **data;
> -	uint32_t           *results;
> -	struct completion  *last_cmplt;
> -	struct completion  *cmplt_array;
> -};
> -
> -/*
> - * Structure to maintain running results for
> - * a single packet (up to 4 tries).
> - */
> -struct completion {
> -	uint32_t *results;                          /* running results. */
> -	int32_t   priority[RTE_ACL_MAX_CATEGORIES]; /* running priorities. */
> -	uint32_t  count;                            /* num of remaining tries */
> -	/* true for allocated struct */
> -} __attribute__((aligned(XMM_SIZE)));
> -
> -/*
> - * One parms structure for each slot in the search engine.
> - */
> -struct parms {
> -	const uint8_t              *data;
> -	/* input data for this packet */
> -	const uint32_t             *data_index;
> -	/* data indirection for this trie */
> -	struct completion          *cmplt;
> -	/* completion data for this packet */
> -};
> -
> -/*
> - * Define an global idle node for unused engine slots
> - */
> -static const uint32_t idle[UINT8_MAX + 1];
> -
> -static const rte_xmm_t mm_type_quad_range = {
> -	.u32 = {
> -		RTE_ACL_NODE_QRANGE,
> -		RTE_ACL_NODE_QRANGE,
> -		RTE_ACL_NODE_QRANGE,
> -		RTE_ACL_NODE_QRANGE,
> -	},
> -};
> -
> -static const rte_xmm_t mm_type_quad_range64 = {
> -	.u32 = {
> -		RTE_ACL_NODE_QRANGE,
> -		RTE_ACL_NODE_QRANGE,
> -		0,
> -		0,
> -	},
> -};
> -
> -static const rte_xmm_t mm_shuffle_input = {
> -	.u32 = {0x00000000, 0x04040404, 0x08080808, 0x0c0c0c0c},
> -};
> -
> -static const rte_xmm_t mm_shuffle_input64 = {
> -	.u32 = {0x00000000, 0x04040404, 0x80808080, 0x80808080},
> -};
> -
> -static const rte_xmm_t mm_ones_16 = {
> -	.u16 = {1, 1, 1, 1, 1, 1, 1, 1},
> -};
> -
> -static const rte_xmm_t mm_bytes = {
> -	.u32 = {UINT8_MAX, UINT8_MAX, UINT8_MAX, UINT8_MAX},
> -};
> -
> -static const rte_xmm_t mm_bytes64 = {
> -	.u32 = {UINT8_MAX, UINT8_MAX, 0, 0},
> -};
> -
> -static const rte_xmm_t mm_match_mask = {
> -	.u32 = {
> -		RTE_ACL_NODE_MATCH,
> -		RTE_ACL_NODE_MATCH,
> -		RTE_ACL_NODE_MATCH,
> -		RTE_ACL_NODE_MATCH,
> -	},
> -};
> -
> -static const rte_xmm_t mm_match_mask64 = {
> -	.u32 = {
> -		RTE_ACL_NODE_MATCH,
> -		0,
> -		RTE_ACL_NODE_MATCH,
> -		0,
> -	},
> -};
> -
> -static const rte_xmm_t mm_index_mask = {
> -	.u32 = {
> -		RTE_ACL_NODE_INDEX,
> -		RTE_ACL_NODE_INDEX,
> -		RTE_ACL_NODE_INDEX,
> -		RTE_ACL_NODE_INDEX,
> -	},
> -};
> -
> -static const rte_xmm_t mm_index_mask64 = {
> -	.u32 = {
> -		RTE_ACL_NODE_INDEX,
> -		RTE_ACL_NODE_INDEX,
> -		0,
> -		0,
> -	},
> -};
> -
> -/*
> - * Allocate a completion structure to manage the tries for a packet.
> - */
> -static inline struct completion *
> -alloc_completion(struct completion *p, uint32_t size, uint32_t tries,
> -	uint32_t *results)
> -{
> -	uint32_t n;
> -
> -	for (n = 0; n < size; n++) {
> -
> -		if (p[n].count == 0) {
> -
> -			/* mark as allocated and set number of tries. */
> -			p[n].count = tries;
> -			p[n].results = results;
> -			return &(p[n]);
> -		}
> -	}
> -
> -	/* should never get here */
> -	return NULL;
> -}
> -
> -/*
> - * Resolve priority for a single result trie.
> - */
> -static inline void
> -resolve_single_priority(uint64_t transition, int n,
> -	const struct rte_acl_ctx *ctx, struct parms *parms,
> -	const struct rte_acl_match_results *p)
> -{
> -	if (parms[n].cmplt->count == ctx->num_tries ||
> -			parms[n].cmplt->priority[0] <=
> -			p[transition].priority[0]) {
> -
> -		parms[n].cmplt->priority[0] = p[transition].priority[0];
> -		parms[n].cmplt->results[0] = p[transition].results[0];
> -	}
> -
> -	parms[n].cmplt->count--;
> -}
> -
> -/*
> - * Resolve priority for multiple results. This consists comparing
> - * the priority of the current traversal with the running set of
> - * results for the packet. For each result, keep a running array of
> - * the result (rule number) and its priority for each category.
> - */
> -static inline void
> -resolve_priority(uint64_t transition, int n, const struct rte_acl_ctx *ctx,
> -	struct parms *parms, const struct rte_acl_match_results *p,
> -	uint32_t categories)
> -{
> -	uint32_t x;
> -	xmm_t results, priority, results1, priority1, selector;
> -	xmm_t *saved_results, *saved_priority;
> -
> -	for (x = 0; x < categories; x += RTE_ACL_RESULTS_MULTIPLIER) {
> -
> -		saved_results = (xmm_t *)(&parms[n].cmplt->results[x]);
> -		saved_priority =
> -			(xmm_t *)(&parms[n].cmplt->priority[x]);
> -
> -		/* get results and priorities for completed trie */
> -		results = MM_LOADU((const xmm_t *)&p[transition].results[x]);
> -		priority = MM_LOADU((const xmm_t *)&p[transition].priority[x]);
> -
> -		/* if this is not the first completed trie */
> -		if (parms[n].cmplt->count != ctx->num_tries) {
> -
> -			/* get running best results and their priorities */
> -			results1 = MM_LOADU(saved_results);
> -			priority1 = MM_LOADU(saved_priority);
> -
> -			/* select results that are highest priority */
> -			selector = MM_CMPGT32(priority1, priority);
> -			results = MM_BLENDV8(results, results1, selector);
> -			priority = MM_BLENDV8(priority, priority1, selector);
> -		}
> -
> -		/* save running best results and their priorities */
> -		MM_STOREU(saved_results, results);
> -		MM_STOREU(saved_priority, priority);
> -	}
> -
> -	/* Count down completed tries for this search request */
> -	parms[n].cmplt->count--;
> -}
> -
> -/*
> - * Routine to fill a slot in the parallel trie traversal array (parms) from
> - * the list of packets (flows).
> - */
> -static inline uint64_t
> -acl_start_next_trie(struct acl_flow_data *flows, struct parms *parms, int n,
> -	const struct rte_acl_ctx *ctx)
> -{
> -	uint64_t transition;
> -
> -	/* if there are any more packets to process */
> -	if (flows->num_packets < flows->total_packets) {
> -		parms[n].data = flows->data[flows->num_packets];
> -		parms[n].data_index = ctx->trie[flows->trie].data_index;
> -
> -		/* if this is the first trie for this packet */
> -		if (flows->trie == 0) {
> -			flows->last_cmplt = alloc_completion(flows->cmplt_array,
> -				flows->cmplt_size, ctx->num_tries,
> -				flows->results +
> -				flows->num_packets * flows->categories);
> -		}
> -
> -		/* set completion parameters and starting index for this slot */
> -		parms[n].cmplt = flows->last_cmplt;
> -		transition =
> -			flows->trans[parms[n].data[*parms[n].data_index++] +
> -			ctx->trie[flows->trie].root_index];
> -
> -		/*
> -		 * if this is the last trie for this packet,
> -		 * then setup next packet.
> -		 */
> -		flows->trie++;
> -		if (flows->trie >= ctx->num_tries) {
> -			flows->trie = 0;
> -			flows->num_packets++;
> -		}
> -
> -		/* keep track of number of active trie traversals */
> -		flows->started++;
> -
> -	/* no more tries to process, set slot to an idle position */
> -	} else {
> -		transition = ctx->idle;
> -		parms[n].data = (const uint8_t *)idle;
> -		parms[n].data_index = idle;
> -	}
> -	return transition;
> -}
> -
> -/*
> - * Detect matches. If a match node transition is found, then this trie
> - * traversal is complete and fill the slot with the next trie
> - * to be processed.
> - */
> -static inline uint64_t
> -acl_match_check_transition(uint64_t transition, int slot,
> -	const struct rte_acl_ctx *ctx, struct parms *parms,
> -	struct acl_flow_data *flows)
> -{
> -	const struct rte_acl_match_results *p;
> -
> -	p = (const struct rte_acl_match_results *)
> -		(flows->trans + ctx->match_index);
> -
> -	if (transition & RTE_ACL_NODE_MATCH) {
> -
> -		/* Remove flags from index and decrement active traversals */
> -		transition &= RTE_ACL_NODE_INDEX;
> -		flows->started--;
> -
> -		/* Resolve priorities for this trie and running results */
> -		if (flows->categories == 1)
> -			resolve_single_priority(transition, slot, ctx,
> -				parms, p);
> -		else
> -			resolve_priority(transition, slot, ctx, parms, p,
> -				flows->categories);
> -
> -		/* Fill the slot with the next trie or idle trie */
> -		transition = acl_start_next_trie(flows, parms, slot, ctx);
> -
> -	} else if (transition == ctx->idle) {
> -		/* reset indirection table for idle slots */
> -		parms[slot].data_index = idle;
> -	}
> -
> -	return transition;
> -}
> -
> -/*
> - * Extract transitions from an XMM register and check for any matches
> - */
> -static void
> -acl_process_matches(xmm_t *indicies, int slot, const struct rte_acl_ctx *ctx,
> -	struct parms *parms, struct acl_flow_data *flows)
> -{
> -	uint64_t transition1, transition2;
> -
> -	/* extract transition from low 64 bits. */
> -	transition1 = MM_CVT64(*indicies);
> -
> -	/* extract transition from high 64 bits. */
> -	*indicies = MM_SHUFFLE32(*indicies, SHUFFLE32_SWAP64);
> -	transition2 = MM_CVT64(*indicies);
> -
> -	transition1 = acl_match_check_transition(transition1, slot, ctx,
> -		parms, flows);
> -	transition2 = acl_match_check_transition(transition2, slot + 1, ctx,
> -		parms, flows);
> -
> -	/* update indicies with new transitions. */
> -	*indicies = MM_SET64(transition2, transition1);
> -}
> -
> -/*
> - * Check for a match in 2 transitions (contained in SSE register)
> - */
> -static inline void
> -acl_match_check_x2(int slot, const struct rte_acl_ctx *ctx, struct parms *parms,
> -	struct acl_flow_data *flows, xmm_t *indicies, xmm_t match_mask)
> -{
> -	xmm_t temp;
> -
> -	temp = MM_AND(match_mask, *indicies);
> -	while (!MM_TESTZ(temp, temp)) {
> -		acl_process_matches(indicies, slot, ctx, parms, flows);
> -		temp = MM_AND(match_mask, *indicies);
> -	}
> -}
> -
> -/*
> - * Check for any match in 4 transitions (contained in 2 SSE registers)
> - */
> -static inline void
> -acl_match_check_x4(int slot, const struct rte_acl_ctx *ctx, struct parms *parms,
> -	struct acl_flow_data *flows, xmm_t *indicies1, xmm_t *indicies2,
> -	xmm_t match_mask)
> -{
> -	xmm_t temp;
> -
> -	/* put low 32 bits of each transition into one register */
> -	temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1, (__m128)*indicies2,
> -		0x88);
> -	/* test for match node */
> -	temp = MM_AND(match_mask, temp);
> -
> -	while (!MM_TESTZ(temp, temp)) {
> -		acl_process_matches(indicies1, slot, ctx, parms, flows);
> -		acl_process_matches(indicies2, slot + 2, ctx, parms, flows);
> -
> -		temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1,
> -					(__m128)*indicies2,
> -					0x88);
> -		temp = MM_AND(match_mask, temp);
> -	}
> -}
> -
> -/*
> - * Calculate the address of the next transition for
> - * all types of nodes. Note that only DFA nodes and range
> - * nodes actually transition to another node. Match
> - * nodes don't move.
> - */
> -static inline xmm_t
> -acl_calc_addr(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> -	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> -	xmm_t *indicies1, xmm_t *indicies2)
> -{
> -	xmm_t addr, node_types, temp;
> -
> -	/*
> -	 * Note that no transition is done for a match
> -	 * node and therefore a stream freezes when
> -	 * it reaches a match.
> -	 */
> -
> -	/* Shuffle low 32 into temp and high 32 into indicies2 */
> -	temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1, (__m128)*indicies2,
> -		0x88);
> -	*indicies2 = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1,
> -		(__m128)*indicies2, 0xdd);
> -
> -	/* Calc node type and node addr */
> -	node_types = MM_ANDNOT(index_mask, temp);
> -	addr = MM_AND(index_mask, temp);
> -
> -	/*
> -	 * Calc addr for DFAs - addr = dfa_index + input_byte
> -	 */
> -
> -	/* mask for DFA type (0) nodes */
> -	temp = MM_CMPEQ32(node_types, MM_XOR(node_types, node_types));
> -
> -	/* add input byte to DFA position */
> -	temp = MM_AND(temp, bytes);
> -	temp = MM_AND(temp, next_input);
> -	addr = MM_ADD32(addr, temp);
> -
> -	/*
> -	 * Calc addr for Range nodes -> range_index + range(input)
> -	 */
> -	node_types = MM_CMPEQ32(node_types, type_quad_range);
> -
> -	/*
> -	 * Calculate number of range boundaries that are less than the
> -	 * input value. Range boundaries for each node are in signed 8 bit,
> -	 * ordered from -128 to 127 in the indicies2 register.
> -	 * This is effectively a popcnt of bytes that are greater than the
> -	 * input byte.
> -	 */
> -
> -	/* shuffle input byte to all 4 positions of 32 bit value */
> -	temp = MM_SHUFFLE8(next_input, shuffle_input);
> -
> -	/* check ranges */
> -	temp = MM_CMPGT8(temp, *indicies2);
> -
> -	/* convert -1 to 1 (bytes greater than input byte */
> -	temp = MM_SIGN8(temp, temp);
> -
> -	/* horizontal add pairs of bytes into words */
> -	temp = MM_MADD8(temp, temp);
> -
> -	/* horizontal add pairs of words into dwords */
> -	temp = MM_MADD16(temp, ones_16);
> -
> -	/* mask to range type nodes */
> -	temp = MM_AND(temp, node_types);
> -
> -	/* add index into node position */
> -	return MM_ADD32(addr, temp);
> -}
> -
> -/*
> - * Process 4 transitions (in 2 SIMD registers) in parallel
> - */
> -static inline xmm_t
> -transition4(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> -	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> -	const uint64_t *trans, xmm_t *indicies1, xmm_t *indicies2)
> -{
> -	xmm_t addr;
> -	uint64_t trans0, trans2;
> -
> -	 /* Calculate the address (array index) for all 4 transitions. */
> -
> -	addr = acl_calc_addr(index_mask, next_input, shuffle_input, ones_16,
> -		bytes, type_quad_range, indicies1, indicies2);
> -
> -	 /* Gather 64 bit transitions and pack back into 2 registers. */
> -
> -	trans0 = trans[MM_CVT32(addr)];
> -
> -	/* get slot 2 */
> -
> -	/* {x0, x1, x2, x3} -> {x2, x1, x2, x3} */
> -	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT2);
> -	trans2 = trans[MM_CVT32(addr)];
> -
> -	/* get slot 1 */
> -
> -	/* {x2, x1, x2, x3} -> {x1, x1, x2, x3} */
> -	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT1);
> -	*indicies1 = MM_SET64(trans[MM_CVT32(addr)], trans0);
> -
> -	/* get slot 3 */
> -
> -	/* {x1, x1, x2, x3} -> {x3, x1, x2, x3} */
> -	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT3);
> -	*indicies2 = MM_SET64(trans[MM_CVT32(addr)], trans2);
> -
> -	return MM_SRL32(next_input, 8);
> -}
> -
> -static inline void
> -acl_set_flow(struct acl_flow_data *flows, struct completion *cmplt,
> -	uint32_t cmplt_size, const uint8_t **data, uint32_t *results,
> -	uint32_t data_num, uint32_t categories, const uint64_t *trans)
> -{
> -	flows->num_packets = 0;
> -	flows->started = 0;
> -	flows->trie = 0;
> -	flows->last_cmplt = NULL;
> -	flows->cmplt_array = cmplt;
> -	flows->total_packets = data_num;
> -	flows->categories = categories;
> -	flows->cmplt_size = cmplt_size;
> -	flows->data = data;
> -	flows->results = results;
> -	flows->trans = trans;
> -}
> -
> -/*
> - * Execute trie traversal with 8 traversals in parallel
> - */
> -static inline void
> -search_sse_8(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	uint32_t *results, uint32_t total_packets, uint32_t categories)
> -{
> -	int n;
> -	struct acl_flow_data flows;
> -	uint64_t index_array[MAX_SEARCHES_SSE8];
> -	struct completion cmplt[MAX_SEARCHES_SSE8];
> -	struct parms parms[MAX_SEARCHES_SSE8];
> -	xmm_t input0, input1;
> -	xmm_t indicies1, indicies2, indicies3, indicies4;
> -
> -	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> -		total_packets, categories, ctx->trans_table);
> -
> -	for (n = 0; n < MAX_SEARCHES_SSE8; n++) {
> -		cmplt[n].count = 0;
> -		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> -	}
> -
> -	/*
> -	 * indicies1 contains index_array[0,1]
> -	 * indicies2 contains index_array[2,3]
> -	 * indicies3 contains index_array[4,5]
> -	 * indicies4 contains index_array[6,7]
> -	 */
> -
> -	indicies1 = MM_LOADU((xmm_t *) &index_array[0]);
> -	indicies2 = MM_LOADU((xmm_t *) &index_array[2]);
> -
> -	indicies3 = MM_LOADU((xmm_t *) &index_array[4]);
> -	indicies4 = MM_LOADU((xmm_t *) &index_array[6]);
> -
> -	 /* Check for any matches. */
> -	acl_match_check_x4(0, ctx, parms, &flows,
> -		&indicies1, &indicies2, mm_match_mask.m);
> -	acl_match_check_x4(4, ctx, parms, &flows,
> -		&indicies3, &indicies4, mm_match_mask.m);
> -
> -	while (flows.started > 0) {
> -
> -		/* Gather 4 bytes of input data for each stream. */
> -		input0 = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0),
> -			0);
> -		input1 = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 4),
> -			0);
> -
> -		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 1), 1);
> -		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 5), 1);
> -
> -		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 2), 2);
> -		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 6), 2);
> -
> -		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 3), 3);
> -		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 7), 3);
> -
> -		 /* Process the 4 bytes of input on each stream. */
> -
> -		input0 = transition4(mm_index_mask.m, input0,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		input1 = transition4(mm_index_mask.m, input1,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies3, &indicies4);
> -
> -		input0 = transition4(mm_index_mask.m, input0,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		input1 = transition4(mm_index_mask.m, input1,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies3, &indicies4);
> -
> -		input0 = transition4(mm_index_mask.m, input0,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		input1 = transition4(mm_index_mask.m, input1,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies3, &indicies4);
> -
> -		input0 = transition4(mm_index_mask.m, input0,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		input1 = transition4(mm_index_mask.m, input1,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies3, &indicies4);
> -
> -		 /* Check for any matches. */
> -		acl_match_check_x4(0, ctx, parms, &flows,
> -			&indicies1, &indicies2, mm_match_mask.m);
> -		acl_match_check_x4(4, ctx, parms, &flows,
> -			&indicies3, &indicies4, mm_match_mask.m);
> -	}
> -}
> -
> -/*
> - * Execute trie traversal with 4 traversals in parallel
> - */
> -static inline void
> -search_sse_4(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	 uint32_t *results, int total_packets, uint32_t categories)
> -{
> -	int n;
> -	struct acl_flow_data flows;
> -	uint64_t index_array[MAX_SEARCHES_SSE4];
> -	struct completion cmplt[MAX_SEARCHES_SSE4];
> -	struct parms parms[MAX_SEARCHES_SSE4];
> -	xmm_t input, indicies1, indicies2;
> -
> -	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> -		total_packets, categories, ctx->trans_table);
> -
> -	for (n = 0; n < MAX_SEARCHES_SSE4; n++) {
> -		cmplt[n].count = 0;
> -		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> -	}
> -
> -	indicies1 = MM_LOADU((xmm_t *) &index_array[0]);
> -	indicies2 = MM_LOADU((xmm_t *) &index_array[2]);
> -
> -	/* Check for any matches. */
> -	acl_match_check_x4(0, ctx, parms, &flows,
> -		&indicies1, &indicies2, mm_match_mask.m);
> -
> -	while (flows.started > 0) {
> -
> -		/* Gather 4 bytes of input data for each stream. */
> -		input = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0), 0);
> -		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 1), 1);
> -		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 2), 2);
> -		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 3), 3);
> -
> -		/* Process the 4 bytes of input on each stream. */
> -		input = transition4(mm_index_mask.m, input,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		 input = transition4(mm_index_mask.m, input,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		 input = transition4(mm_index_mask.m, input,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		 input = transition4(mm_index_mask.m, input,
> -			mm_shuffle_input.m, mm_ones_16.m,
> -			mm_bytes.m, mm_type_quad_range.m,
> -			flows.trans, &indicies1, &indicies2);
> -
> -		/* Check for any matches. */
> -		acl_match_check_x4(0, ctx, parms, &flows,
> -			&indicies1, &indicies2, mm_match_mask.m);
> -	}
> -}
> -
> -static inline xmm_t
> -transition2(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> -	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> -	const uint64_t *trans, xmm_t *indicies1)
> -{
> -	uint64_t t;
> -	xmm_t addr, indicies2;
> -
> -	indicies2 = MM_XOR(ones_16, ones_16);
> -
> -	addr = acl_calc_addr(index_mask, next_input, shuffle_input, ones_16,
> -		bytes, type_quad_range, indicies1, &indicies2);
> -
> -	/* Gather 64 bit transitions and pack 2 per register. */
> -
> -	t = trans[MM_CVT32(addr)];
> -
> -	/* get slot 1 */
> -	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT1);
> -	*indicies1 = MM_SET64(trans[MM_CVT32(addr)], t);
> -
> -	return MM_SRL32(next_input, 8);
> -}
> -
> -/*
> - * Execute trie traversal with 2 traversals in parallel.
> - */
> -static inline void
> -search_sse_2(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	uint32_t *results, uint32_t total_packets, uint32_t categories)
> -{
> -	int n;
> -	struct acl_flow_data flows;
> -	uint64_t index_array[MAX_SEARCHES_SSE2];
> -	struct completion cmplt[MAX_SEARCHES_SSE2];
> -	struct parms parms[MAX_SEARCHES_SSE2];
> -	xmm_t input, indicies;
> -
> -	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> -		total_packets, categories, ctx->trans_table);
> -
> -	for (n = 0; n < MAX_SEARCHES_SSE2; n++) {
> -		cmplt[n].count = 0;
> -		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> -	}
> -
> -	indicies = MM_LOADU((xmm_t *) &index_array[0]);
> -
> -	/* Check for any matches. */
> -	acl_match_check_x2(0, ctx, parms, &flows, &indicies, mm_match_mask64.m);
> -
> -	while (flows.started > 0) {
> -
> -		/* Gather 4 bytes of input data for each stream. */
> -		input = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0), 0);
> -		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 1), 1);
> -
> -		/* Process the 4 bytes of input on each stream. */
> -
> -		input = transition2(mm_index_mask64.m, input,
> -			mm_shuffle_input64.m, mm_ones_16.m,
> -			mm_bytes64.m, mm_type_quad_range64.m,
> -			flows.trans, &indicies);
> -
> -		input = transition2(mm_index_mask64.m, input,
> -			mm_shuffle_input64.m, mm_ones_16.m,
> -			mm_bytes64.m, mm_type_quad_range64.m,
> -			flows.trans, &indicies);
> -
> -		input = transition2(mm_index_mask64.m, input,
> -			mm_shuffle_input64.m, mm_ones_16.m,
> -			mm_bytes64.m, mm_type_quad_range64.m,
> -			flows.trans, &indicies);
> -
> -		input = transition2(mm_index_mask64.m, input,
> -			mm_shuffle_input64.m, mm_ones_16.m,
> -			mm_bytes64.m, mm_type_quad_range64.m,
> -			flows.trans, &indicies);
> -
> -		/* Check for any matches. */
> -		acl_match_check_x2(0, ctx, parms, &flows, &indicies,
> -			mm_match_mask64.m);
> -	}
> -}
> -
> -/*
> - * When processing the transition, rather than using if/else
> - * construct, the offset is calculated for DFA and QRANGE and
> - * then conditionally added to the address based on node type.
> - * This is done to avoid branch mis-predictions. Since the
> - * offset is rather simple calculation it is more efficient
> - * to do the calculation and do a condition move rather than
> - * a conditional branch to determine which calculation to do.
> - */
> -static inline uint32_t
> -scan_forward(uint32_t input, uint32_t max)
> -{
> -	return (input == 0) ? max : rte_bsf32(input);
> -}
> -
> -static inline uint64_t
> -scalar_transition(const uint64_t *trans_table, uint64_t transition,
> -	uint8_t input)
> -{
> -	uint32_t addr, index, ranges, x, a, b, c;
> -
> -	/* break transition into component parts */
> -	ranges = transition >> (sizeof(index) * CHAR_BIT);
> -
> -	/* calc address for a QRANGE node */
> -	c = input * SCALAR_QRANGE_MULT;
> -	a = ranges | SCALAR_QRANGE_MIN;
> -	index = transition & ~RTE_ACL_NODE_INDEX;
> -	a -= (c & SCALAR_QRANGE_MASK);
> -	b = c & SCALAR_QRANGE_MIN;
> -	addr = transition ^ index;
> -	a &= SCALAR_QRANGE_MIN;
> -	a ^= (ranges ^ b) & (a ^ b);
> -	x = scan_forward(a, 32) >> 3;
> -	addr += (index == RTE_ACL_NODE_DFA) ? input : x;
> -
> -	/* pickup next transition */
> -	transition = *(trans_table + addr);
> -	return transition;
> -}
> -
> -int
> -rte_acl_classify_scalar(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	uint32_t *results, uint32_t num, uint32_t categories)
> -{
> -	int n;
> -	uint64_t transition0, transition1;
> -	uint32_t input0, input1;
> -	struct acl_flow_data flows;
> -	uint64_t index_array[MAX_SEARCHES_SCALAR];
> -	struct completion cmplt[MAX_SEARCHES_SCALAR];
> -	struct parms parms[MAX_SEARCHES_SCALAR];
> -
> -	if (categories != 1 &&
> -		((RTE_ACL_RESULTS_MULTIPLIER - 1) & categories) != 0)
> -		return -EINVAL;
> -
> -	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results, num,
> -		categories, ctx->trans_table);
> -
> -	for (n = 0; n < MAX_SEARCHES_SCALAR; n++) {
> -		cmplt[n].count = 0;
> -		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> -	}
> -
> -	transition0 = index_array[0];
> -	transition1 = index_array[1];
> -
> -	while (flows.started > 0) {
> -
> -		input0 = GET_NEXT_4BYTES(parms, 0);
> -		input1 = GET_NEXT_4BYTES(parms, 1);
> -
> -		for (n = 0; n < 4; n++) {
> -			if (likely((transition0 & RTE_ACL_NODE_MATCH) == 0))
> -				transition0 = scalar_transition(flows.trans,
> -					transition0, (uint8_t)input0);
> -
> -			input0 >>= CHAR_BIT;
> -
> -			if (likely((transition1 & RTE_ACL_NODE_MATCH) == 0))
> -				transition1 = scalar_transition(flows.trans,
> -					transition1, (uint8_t)input1);
> -
> -			input1 >>= CHAR_BIT;
> -
> -		}
> -		if ((transition0 | transition1) & RTE_ACL_NODE_MATCH) {
> -			transition0 = acl_match_check_transition(transition0,
> -				0, ctx, parms, &flows);
> -			transition1 = acl_match_check_transition(transition1,
> -				1, ctx, parms, &flows);
> -
> -		}
> -	}
> -	return 0;
> -}
> -
> -int
> -rte_acl_classify(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	uint32_t *results, uint32_t num, uint32_t categories)
> -{
> -	if (categories != 1 &&
> -		((RTE_ACL_RESULTS_MULTIPLIER - 1) & categories) != 0)
> -		return -EINVAL;
> -
> -	if (likely(num >= MAX_SEARCHES_SSE8))
> -		search_sse_8(ctx, data, results, num, categories);
> -	else if (num >= MAX_SEARCHES_SSE4)
> -		search_sse_4(ctx, data, results, num, categories);
> -	else
> -		search_sse_2(ctx, data, results, num, categories);
> -
> -	return 0;
> -}
> diff --git a/lib/librte_acl/acl_run.h b/lib/librte_acl/acl_run.h
> new file mode 100644
> index 0000000..c39650e
> --- /dev/null
> +++ b/lib/librte_acl/acl_run.h
> @@ -0,0 +1,220 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#ifndef	_ACL_RUN_H_
> +#define	_ACL_RUN_H_
> +
> +#include <rte_acl.h>
> +#include "acl_vect.h"
> +#include "acl.h"
> +
> +#define MAX_SEARCHES_SSE8	8
> +#define MAX_SEARCHES_SSE4	4
> +#define MAX_SEARCHES_SSE2	2
> +#define MAX_SEARCHES_SCALAR	2
> +
> +#define GET_NEXT_4BYTES(prm, idx)	\
> +	(*((const int32_t *)((prm)[(idx)].data + *(prm)[idx].data_index++)))
> +
> +
> +#define RTE_ACL_NODE_INDEX	((uint32_t)~RTE_ACL_NODE_TYPE)
> +
> +#define	SCALAR_QRANGE_MULT	0x01010101
> +#define	SCALAR_QRANGE_MASK	0x7f7f7f7f
> +#define	SCALAR_QRANGE_MIN	0x80808080
> +
> +/*
> + * Structure to manage N parallel trie traversals.
> + * The runtime trie traversal routines can process 8, 4, or 2 tries
> + * in parallel. Each packet may require multiple trie traversals (up to 4).
> + * This structure is used to fill the slots (0 to n-1) for parallel processing
> + * with the trie traversals needed for each packet.
> + */
> +struct acl_flow_data {
> +	uint32_t            num_packets;
> +	/* number of packets processed */
> +	uint32_t            started;
> +	/* number of trie traversals in progress */
> +	uint32_t            trie;
> +	/* current trie index (0 to N-1) */
> +	uint32_t            cmplt_size;
> +	uint32_t            total_packets;
> +	uint32_t            categories;
> +	/* number of result categories per packet. */
> +	/* maximum number of packets to process */
> +	const uint64_t     *trans;
> +	const uint8_t     **data;
> +	uint32_t           *results;
> +	struct completion  *last_cmplt;
> +	struct completion  *cmplt_array;
> +};
> +
> +/*
> + * Structure to maintain running results for
> + * a single packet (up to 4 tries).
> + */
> +struct completion {
> +	uint32_t *results;                          /* running results. */
> +	int32_t   priority[RTE_ACL_MAX_CATEGORIES]; /* running priorities. */
> +	uint32_t  count;                            /* num of remaining tries */
> +	/* true for allocated struct */
> +} __attribute__((aligned(XMM_SIZE)));
> +
> +/*
> + * One parms structure for each slot in the search engine.
> + */
> +struct parms {
> +	const uint8_t              *data;
> +	/* input data for this packet */
> +	const uint32_t             *data_index;
> +	/* data indirection for this trie */
> +	struct completion          *cmplt;
> +	/* completion data for this packet */
> +};
> +
> +/*
> + * Define an global idle node for unused engine slots
> + */
> +static const uint32_t idle[UINT8_MAX + 1];
> +
> +/*
> + * Allocate a completion structure to manage the tries for a packet.
> + */
> +static inline struct completion *
> +alloc_completion(struct completion *p, uint32_t size, uint32_t tries,
> +	uint32_t *results)
> +{
> +	uint32_t n;
> +
> +	for (n = 0; n < size; n++) {
> +
> +		if (p[n].count == 0) {
> +
> +			/* mark as allocated and set number of tries. */
> +			p[n].count = tries;
> +			p[n].results = results;
> +			return &(p[n]);
> +		}
> +	}
> +
> +	/* should never get here */
> +	return NULL;
> +}
> +
> +/*
> + * Resolve priority for a single result trie.
> + */
> +static inline void
> +resolve_single_priority(uint64_t transition, int n,
> +	const struct rte_acl_ctx *ctx, struct parms *parms,
> +	const struct rte_acl_match_results *p)
> +{
> +	if (parms[n].cmplt->count == ctx->num_tries ||
> +			parms[n].cmplt->priority[0] <=
> +			p[transition].priority[0]) {
> +
> +		parms[n].cmplt->priority[0] = p[transition].priority[0];
> +		parms[n].cmplt->results[0] = p[transition].results[0];
> +	}
> +}
> +
> +/*
> + * Routine to fill a slot in the parallel trie traversal array (parms) from
> + * the list of packets (flows).
> + */
> +static inline uint64_t
> +acl_start_next_trie(struct acl_flow_data *flows, struct parms *parms, int n,
> +	const struct rte_acl_ctx *ctx)
> +{
> +	uint64_t transition;
> +
> +	/* if there are any more packets to process */
> +	if (flows->num_packets < flows->total_packets) {
> +		parms[n].data = flows->data[flows->num_packets];
> +		parms[n].data_index = ctx->trie[flows->trie].data_index;
> +
> +		/* if this is the first trie for this packet */
> +		if (flows->trie == 0) {
> +			flows->last_cmplt = alloc_completion(flows->cmplt_array,
> +				flows->cmplt_size, ctx->num_tries,
> +				flows->results +
> +				flows->num_packets * flows->categories);
> +		}
> +
> +		/* set completion parameters and starting index for this slot */
> +		parms[n].cmplt = flows->last_cmplt;
> +		transition =
> +			flows->trans[parms[n].data[*parms[n].data_index++] +
> +			ctx->trie[flows->trie].root_index];
> +
> +		/*
> +		 * if this is the last trie for this packet,
> +		 * then setup next packet.
> +		 */
> +		flows->trie++;
> +		if (flows->trie >= ctx->num_tries) {
> +			flows->trie = 0;
> +			flows->num_packets++;
> +		}
> +
> +		/* keep track of number of active trie traversals */
> +		flows->started++;
> +
> +	/* no more tries to process, set slot to an idle position */
> +	} else {
> +		transition = ctx->idle;
> +		parms[n].data = (const uint8_t *)idle;
> +		parms[n].data_index = idle;
> +	}
> +	return transition;
> +}
> +
> +static inline void
> +acl_set_flow(struct acl_flow_data *flows, struct completion *cmplt,
> +	uint32_t cmplt_size, const uint8_t **data, uint32_t *results,
> +	uint32_t data_num, uint32_t categories, const uint64_t *trans)
> +{
> +	flows->num_packets = 0;
> +	flows->started = 0;
> +	flows->trie = 0;
> +	flows->last_cmplt = NULL;
> +	flows->cmplt_array = cmplt;
> +	flows->total_packets = data_num;
> +	flows->categories = categories;
> +	flows->cmplt_size = cmplt_size;
> +	flows->data = data;
> +	flows->results = results;
> +	flows->trans = trans;
> +}
> +
> +#endif /* _ACL_RUN_H_ */
> diff --git a/lib/librte_acl/acl_run_scalar.c b/lib/librte_acl/acl_run_scalar.c
> new file mode 100644
> index 0000000..a59ff17
> --- /dev/null
> +++ b/lib/librte_acl/acl_run_scalar.c
> @@ -0,0 +1,198 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include "acl_run.h"
> +#include "acl_match_check.h"
> +
> +int
> +rte_acl_classify_scalar(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +        uint32_t *results, uint32_t num, uint32_t categories);
> +
> +/*
> + * Resolve priority for multiple results (scalar version).
> + * This consists comparing the priority of the current traversal with the
> + * running set of results for the packet.
> + * For each result, keep a running array of the result (rule number) and
> + * its priority for each category.
> + */
> +static inline void
> +resolve_priority_scalar(uint64_t transition, int n,
> +	const struct rte_acl_ctx *ctx, struct parms *parms,
> +	const struct rte_acl_match_results *p, uint32_t categories)
> +{
> +	uint32_t i;
> +	int32_t *saved_priority;
> +	uint32_t *saved_results;
> +	const int32_t *priority;
> +	const uint32_t *results;
> +
> +	saved_results = parms[n].cmplt->results;
> +	saved_priority = parms[n].cmplt->priority;
> +
> +	/* results and priorities for completed trie */
> +	results = p[transition].results;
> +	priority = p[transition].priority;
> +
> +	/* if this is not the first completed trie */
> +	if (parms[n].cmplt->count != ctx->num_tries) {
> +		for (i = 0; i < categories; i += RTE_ACL_RESULTS_MULTIPLIER) {
> +
> +			if (saved_priority[i] <= priority[i]) {
> +				saved_priority[i] = priority[i];
> +				saved_results[i] = results[i];
> +			}
> +			if (saved_priority[i + 1] <= priority[i + 1]) {
> +				saved_priority[i + 1] = priority[i + 1];
> +				saved_results[i + 1] = results[i + 1];
> +			}
> +			if (saved_priority[i + 2] <= priority[i + 2]) {
> +				saved_priority[i + 2] = priority[i + 2];
> +				saved_results[i + 2] = results[i + 2];
> +			}
> +			if (saved_priority[i + 3] <= priority[i + 3]) {
> +				saved_priority[i + 3] = priority[i + 3];
> +				saved_results[i + 3] = results[i + 3];
> +			}
> +		}
> +	} else {
> +		for (i = 0; i < categories; i += RTE_ACL_RESULTS_MULTIPLIER) {
> +			saved_priority[i] = priority[i];
> +			saved_priority[i + 1] = priority[i + 1];
> +			saved_priority[i + 2] = priority[i + 2];
> +			saved_priority[i + 3] = priority[i + 3];
> +
> +			saved_results[i] = results[i];
> +			saved_results[i + 1] = results[i + 1];
> +			saved_results[i + 2] = results[i + 2];
> +			saved_results[i + 3] = results[i + 3];
> +		}
> +	}
> +}
> +
> +/*
> + * When processing the transition, rather than using if/else
> + * construct, the offset is calculated for DFA and QRANGE and
> + * then conditionally added to the address based on node type.
> + * This is done to avoid branch mis-predictions. Since the
> + * offset is rather simple calculation it is more efficient
> + * to do the calculation and do a condition move rather than
> + * a conditional branch to determine which calculation to do.
> + */
> +static inline uint32_t
> +scan_forward(uint32_t input, uint32_t max)
> +{
> +	return (input == 0) ? max : rte_bsf32(input);
> +}
> +
> +static inline uint64_t
> +scalar_transition(const uint64_t *trans_table, uint64_t transition,
> +	uint8_t input)
> +{
> +	uint32_t addr, index, ranges, x, a, b, c;
> +
> +	/* break transition into component parts */
> +	ranges = transition >> (sizeof(index) * CHAR_BIT);
> +
> +	/* calc address for a QRANGE node */
> +	c = input * SCALAR_QRANGE_MULT;
> +	a = ranges | SCALAR_QRANGE_MIN;
> +	index = transition & ~RTE_ACL_NODE_INDEX;
> +	a -= (c & SCALAR_QRANGE_MASK);
> +	b = c & SCALAR_QRANGE_MIN;
> +	addr = transition ^ index;
> +	a &= SCALAR_QRANGE_MIN;
> +	a ^= (ranges ^ b) & (a ^ b);
> +	x = scan_forward(a, 32) >> 3;
> +	addr += (index == RTE_ACL_NODE_DFA) ? input : x;
> +
> +	/* pickup next transition */
> +	transition = *(trans_table + addr);
> +	return transition;
> +}
> +
> +int
> +rte_acl_classify_scalar(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +	uint32_t *results, uint32_t num, uint32_t categories)
> +{
> +	int n;
> +	uint64_t transition0, transition1;
> +	uint32_t input0, input1;
> +	struct acl_flow_data flows;
> +	uint64_t index_array[MAX_SEARCHES_SCALAR];
> +	struct completion cmplt[MAX_SEARCHES_SCALAR];
> +	struct parms parms[MAX_SEARCHES_SCALAR];
> +
> +	if (categories != 1 &&
> +		((RTE_ACL_RESULTS_MULTIPLIER - 1) & categories) != 0)
> +		return -EINVAL;
> +
> +	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results, num,
> +		categories, ctx->trans_table);
> +
> +	for (n = 0; n < MAX_SEARCHES_SCALAR; n++) {
> +		cmplt[n].count = 0;
> +		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> +	}
> +
> +	transition0 = index_array[0];
> +	transition1 = index_array[1];
> +
> +	while (flows.started > 0) {
> +
> +		input0 = GET_NEXT_4BYTES(parms, 0);
> +		input1 = GET_NEXT_4BYTES(parms, 1);
> +
> +		for (n = 0; n < 4; n++) {
> +			if (likely((transition0 & RTE_ACL_NODE_MATCH) == 0))
> +				transition0 = scalar_transition(flows.trans,
> +					transition0, (uint8_t)input0);
> +
> +			input0 >>= CHAR_BIT;
> +
> +			if (likely((transition1 & RTE_ACL_NODE_MATCH) == 0))
> +				transition1 = scalar_transition(flows.trans,
> +					transition1, (uint8_t)input1);
> +
> +			input1 >>= CHAR_BIT;
> +
> +		}
> +		if ((transition0 | transition1) & RTE_ACL_NODE_MATCH) {
> +			transition0 = acl_match_check(transition0,
> +				0, ctx, parms, &flows, resolve_priority_scalar);
> +			transition1 = acl_match_check(transition1,
> +				1, ctx, parms, &flows, resolve_priority_scalar);
> +
> +		}
> +	}
> +	return 0;
> +}
> diff --git a/lib/librte_acl/acl_run_sse.c b/lib/librte_acl/acl_run_sse.c
> new file mode 100644
> index 0000000..3f5c721
> --- /dev/null
> +++ b/lib/librte_acl/acl_run_sse.c
> @@ -0,0 +1,627 @@
> +/*-
> + *   BSD LICENSE
> + *
> + *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
> + *   All rights reserved.
> + *
> + *   Redistribution and use in source and binary forms, with or without
> + *   modification, are permitted provided that the following conditions
> + *   are met:
> + *
> + *     * Redistributions of source code must retain the above copyright
> + *       notice, this list of conditions and the following disclaimer.
> + *     * Redistributions in binary form must reproduce the above copyright
> + *       notice, this list of conditions and the following disclaimer in
> + *       the documentation and/or other materials provided with the
> + *       distribution.
> + *     * Neither the name of Intel Corporation nor the names of its
> + *       contributors may be used to endorse or promote products derived
> + *       from this software without specific prior written permission.
> + *
> + *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
> + *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
> + *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
> + *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
> + *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
> + *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
> + *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
> + *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
> + *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
> + *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
> + *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
> + */
> +
> +#include "acl_run.h"
> +#include "acl_match_check.h"
> +
> +enum {
> +	SHUFFLE32_SLOT1 = 0xe5,
> +	SHUFFLE32_SLOT2 = 0xe6,
> +	SHUFFLE32_SLOT3 = 0xe7,
> +	SHUFFLE32_SWAP64 = 0x4e,
> +};
> +
> +static const rte_xmm_t mm_type_quad_range = {
> +	.u32 = {
> +		RTE_ACL_NODE_QRANGE,
> +		RTE_ACL_NODE_QRANGE,
> +		RTE_ACL_NODE_QRANGE,
> +		RTE_ACL_NODE_QRANGE,
> +	},
> +};
> +
> +static const rte_xmm_t mm_type_quad_range64 = {
> +	.u32 = {
> +		RTE_ACL_NODE_QRANGE,
> +		RTE_ACL_NODE_QRANGE,
> +		0,
> +		0,
> +	},
> +};
> +
> +static const rte_xmm_t mm_shuffle_input = {
> +	.u32 = {0x00000000, 0x04040404, 0x08080808, 0x0c0c0c0c},
> +};
> +
> +static const rte_xmm_t mm_shuffle_input64 = {
> +	.u32 = {0x00000000, 0x04040404, 0x80808080, 0x80808080},
> +};
> +
> +static const rte_xmm_t mm_ones_16 = {
> +	.u16 = {1, 1, 1, 1, 1, 1, 1, 1},
> +};
> +
> +static const rte_xmm_t mm_bytes = {
> +	.u32 = {UINT8_MAX, UINT8_MAX, UINT8_MAX, UINT8_MAX},
> +};
> +
> +static const rte_xmm_t mm_bytes64 = {
> +	.u32 = {UINT8_MAX, UINT8_MAX, 0, 0},
> +};
> +
> +static const rte_xmm_t mm_match_mask = {
> +	.u32 = {
> +		RTE_ACL_NODE_MATCH,
> +		RTE_ACL_NODE_MATCH,
> +		RTE_ACL_NODE_MATCH,
> +		RTE_ACL_NODE_MATCH,
> +	},
> +};
> +
> +static const rte_xmm_t mm_match_mask64 = {
> +	.u32 = {
> +		RTE_ACL_NODE_MATCH,
> +		0,
> +		RTE_ACL_NODE_MATCH,
> +		0,
> +	},
> +};
> +
> +static const rte_xmm_t mm_index_mask = {
> +	.u32 = {
> +		RTE_ACL_NODE_INDEX,
> +		RTE_ACL_NODE_INDEX,
> +		RTE_ACL_NODE_INDEX,
> +		RTE_ACL_NODE_INDEX,
> +	},
> +};
> +
> +static const rte_xmm_t mm_index_mask64 = {
> +	.u32 = {
> +		RTE_ACL_NODE_INDEX,
> +		RTE_ACL_NODE_INDEX,
> +		0,
> +		0,
> +	},
> +};
> +
> +
> +/*
> + * Resolve priority for multiple results (sse version).
> + * This consists comparing the priority of the current traversal with the
> + * running set of results for the packet.
> + * For each result, keep a running array of the result (rule number) and
> + * its priority for each category.
> + */
> +static inline void
> +resolve_priority_sse(uint64_t transition, int n, const struct rte_acl_ctx *ctx,
> +	struct parms *parms, const struct rte_acl_match_results *p,
> +	uint32_t categories)
> +{
> +	uint32_t x;
> +	xmm_t results, priority, results1, priority1, selector;
> +	xmm_t *saved_results, *saved_priority;
> +
> +	for (x = 0; x < categories; x += RTE_ACL_RESULTS_MULTIPLIER) {
> +
> +		saved_results = (xmm_t *)(&parms[n].cmplt->results[x]);
> +		saved_priority =
> +			(xmm_t *)(&parms[n].cmplt->priority[x]);
> +
> +		/* get results and priorities for completed trie */
> +		results = MM_LOADU((const xmm_t *)&p[transition].results[x]);
> +		priority = MM_LOADU((const xmm_t *)&p[transition].priority[x]);
> +
> +		/* if this is not the first completed trie */
> +		if (parms[n].cmplt->count != ctx->num_tries) {
> +
> +			/* get running best results and their priorities */
> +			results1 = MM_LOADU(saved_results);
> +			priority1 = MM_LOADU(saved_priority);
> +
> +			/* select results that are highest priority */
> +			selector = MM_CMPGT32(priority1, priority);
> +			results = MM_BLENDV8(results, results1, selector);
> +			priority = MM_BLENDV8(priority, priority1, selector);
> +		}
> +
> +		/* save running best results and their priorities */
> +		MM_STOREU(saved_results, results);
> +		MM_STOREU(saved_priority, priority);
> +	}
> +}
> +
> +/*
> + * Extract transitions from an XMM register and check for any matches
> + */
> +static void
> +acl_process_matches(xmm_t *indicies, int slot, const struct rte_acl_ctx *ctx,
> +	struct parms *parms, struct acl_flow_data *flows)
> +{
> +	uint64_t transition1, transition2;
> +
> +	/* extract transition from low 64 bits. */
> +	transition1 = MM_CVT64(*indicies);
> +
> +	/* extract transition from high 64 bits. */
> +	*indicies = MM_SHUFFLE32(*indicies, SHUFFLE32_SWAP64);
> +	transition2 = MM_CVT64(*indicies);
> +
> +	transition1 = acl_match_check(transition1, slot, ctx,
> +		parms, flows, resolve_priority_sse);
> +	transition2 = acl_match_check(transition2, slot + 1, ctx,
> +		parms, flows, resolve_priority_sse);
> +
> +	/* update indicies with new transitions. */
> +	*indicies = MM_SET64(transition2, transition1);
> +}
> +
> +/*
> + * Check for a match in 2 transitions (contained in SSE register)
> + */
> +static inline void
> +acl_match_check_x2(int slot, const struct rte_acl_ctx *ctx, struct parms *parms,
> +	struct acl_flow_data *flows, xmm_t *indicies, xmm_t match_mask)
> +{
> +	xmm_t temp;
> +
> +	temp = MM_AND(match_mask, *indicies);
> +	while (!MM_TESTZ(temp, temp)) {
> +		acl_process_matches(indicies, slot, ctx, parms, flows);
> +		temp = MM_AND(match_mask, *indicies);
> +	}
> +}
> +
> +/*
> + * Check for any match in 4 transitions (contained in 2 SSE registers)
> + */
> +static inline void
> +acl_match_check_x4(int slot, const struct rte_acl_ctx *ctx, struct parms *parms,
> +	struct acl_flow_data *flows, xmm_t *indicies1, xmm_t *indicies2,
> +	xmm_t match_mask)
> +{
> +	xmm_t temp;
> +
> +	/* put low 32 bits of each transition into one register */
> +	temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1, (__m128)*indicies2,
> +		0x88);
> +	/* test for match node */
> +	temp = MM_AND(match_mask, temp);
> +
> +	while (!MM_TESTZ(temp, temp)) {
> +		acl_process_matches(indicies1, slot, ctx, parms, flows);
> +		acl_process_matches(indicies2, slot + 2, ctx, parms, flows);
> +
> +		temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1,
> +					(__m128)*indicies2,
> +					0x88);
> +		temp = MM_AND(match_mask, temp);
> +	}
> +}
> +
> +/*
> + * Calculate the address of the next transition for
> + * all types of nodes. Note that only DFA nodes and range
> + * nodes actually transition to another node. Match
> + * nodes don't move.
> + */
> +static inline xmm_t
> +acl_calc_addr(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> +	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> +	xmm_t *indicies1, xmm_t *indicies2)
> +{
> +	xmm_t addr, node_types, temp;
> +
> +	/*
> +	 * Note that no transition is done for a match
> +	 * node and therefore a stream freezes when
> +	 * it reaches a match.
> +	 */
> +
> +	/* Shuffle low 32 into temp and high 32 into indicies2 */
> +	temp = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1, (__m128)*indicies2,
> +		0x88);
> +	*indicies2 = (xmm_t)MM_SHUFFLEPS((__m128)*indicies1,
> +		(__m128)*indicies2, 0xdd);
> +
> +	/* Calc node type and node addr */
> +	node_types = MM_ANDNOT(index_mask, temp);
> +	addr = MM_AND(index_mask, temp);
> +
> +	/*
> +	 * Calc addr for DFAs - addr = dfa_index + input_byte
> +	 */
> +
> +	/* mask for DFA type (0) nodes */
> +	temp = MM_CMPEQ32(node_types, MM_XOR(node_types, node_types));
> +
> +	/* add input byte to DFA position */
> +	temp = MM_AND(temp, bytes);
> +	temp = MM_AND(temp, next_input);
> +	addr = MM_ADD32(addr, temp);
> +
> +	/*
> +	 * Calc addr for Range nodes -> range_index + range(input)
> +	 */
> +	node_types = MM_CMPEQ32(node_types, type_quad_range);
> +
> +	/*
> +	 * Calculate number of range boundaries that are less than the
> +	 * input value. Range boundaries for each node are in signed 8 bit,
> +	 * ordered from -128 to 127 in the indicies2 register.
> +	 * This is effectively a popcnt of bytes that are greater than the
> +	 * input byte.
> +	 */
> +
> +	/* shuffle input byte to all 4 positions of 32 bit value */
> +	temp = MM_SHUFFLE8(next_input, shuffle_input);
> +
> +	/* check ranges */
> +	temp = MM_CMPGT8(temp, *indicies2);
> +
> +	/* convert -1 to 1 (bytes greater than input byte */
> +	temp = MM_SIGN8(temp, temp);
> +
> +	/* horizontal add pairs of bytes into words */
> +	temp = MM_MADD8(temp, temp);
> +
> +	/* horizontal add pairs of words into dwords */
> +	temp = MM_MADD16(temp, ones_16);
> +
> +	/* mask to range type nodes */
> +	temp = MM_AND(temp, node_types);
> +
> +	/* add index into node position */
> +	return MM_ADD32(addr, temp);
> +}
> +
> +/*
> + * Process 4 transitions (in 2 SIMD registers) in parallel
> + */
> +static inline xmm_t
> +transition4(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> +	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> +	const uint64_t *trans, xmm_t *indicies1, xmm_t *indicies2)
> +{
> +	xmm_t addr;
> +	uint64_t trans0, trans2;
> +
> +	 /* Calculate the address (array index) for all 4 transitions. */
> +
> +	addr = acl_calc_addr(index_mask, next_input, shuffle_input, ones_16,
> +		bytes, type_quad_range, indicies1, indicies2);
> +
> +	 /* Gather 64 bit transitions and pack back into 2 registers. */
> +
> +	trans0 = trans[MM_CVT32(addr)];
> +
> +	/* get slot 2 */
> +
> +	/* {x0, x1, x2, x3} -> {x2, x1, x2, x3} */
> +	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT2);
> +	trans2 = trans[MM_CVT32(addr)];
> +
> +	/* get slot 1 */
> +
> +	/* {x2, x1, x2, x3} -> {x1, x1, x2, x3} */
> +	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT1);
> +	*indicies1 = MM_SET64(trans[MM_CVT32(addr)], trans0);
> +
> +	/* get slot 3 */
> +
> +	/* {x1, x1, x2, x3} -> {x3, x1, x2, x3} */
> +	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT3);
> +	*indicies2 = MM_SET64(trans[MM_CVT32(addr)], trans2);
> +
> +	return MM_SRL32(next_input, 8);
> +}
> +
> +/*
> + * Execute trie traversal with 8 traversals in parallel
> + */
> +static inline int
> +search_sse_8(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +	uint32_t *results, uint32_t total_packets, uint32_t categories)
> +{
> +	int n;
> +	struct acl_flow_data flows;
> +	uint64_t index_array[MAX_SEARCHES_SSE8];
> +	struct completion cmplt[MAX_SEARCHES_SSE8];
> +	struct parms parms[MAX_SEARCHES_SSE8];
> +	xmm_t input0, input1;
> +	xmm_t indicies1, indicies2, indicies3, indicies4;
> +
> +	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> +		total_packets, categories, ctx->trans_table);
> +
> +	for (n = 0; n < MAX_SEARCHES_SSE8; n++) {
> +		cmplt[n].count = 0;
> +		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> +	}
> +
> +	/*
> +	 * indicies1 contains index_array[0,1]
> +	 * indicies2 contains index_array[2,3]
> +	 * indicies3 contains index_array[4,5]
> +	 * indicies4 contains index_array[6,7]
> +	 */
> +
> +	indicies1 = MM_LOADU((xmm_t *) &index_array[0]);
> +	indicies2 = MM_LOADU((xmm_t *) &index_array[2]);
> +
> +	indicies3 = MM_LOADU((xmm_t *) &index_array[4]);
> +	indicies4 = MM_LOADU((xmm_t *) &index_array[6]);
> +
> +	 /* Check for any matches. */
> +	acl_match_check_x4(0, ctx, parms, &flows,
> +		&indicies1, &indicies2, mm_match_mask.m);
> +	acl_match_check_x4(4, ctx, parms, &flows,
> +		&indicies3, &indicies4, mm_match_mask.m);
> +
> +	while (flows.started > 0) {
> +
> +		/* Gather 4 bytes of input data for each stream. */
> +		input0 = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0),
> +			0);
> +		input1 = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 4),
> +			0);
> +
> +		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 1), 1);
> +		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 5), 1);
> +
> +		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 2), 2);
> +		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 6), 2);
> +
> +		input0 = MM_INSERT32(input0, GET_NEXT_4BYTES(parms, 3), 3);
> +		input1 = MM_INSERT32(input1, GET_NEXT_4BYTES(parms, 7), 3);
> +
> +		 /* Process the 4 bytes of input on each stream. */
> +
> +		input0 = transition4(mm_index_mask.m, input0,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		input1 = transition4(mm_index_mask.m, input1,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies3, &indicies4);
> +
> +		input0 = transition4(mm_index_mask.m, input0,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		input1 = transition4(mm_index_mask.m, input1,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies3, &indicies4);
> +
> +		input0 = transition4(mm_index_mask.m, input0,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		input1 = transition4(mm_index_mask.m, input1,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies3, &indicies4);
> +
> +		input0 = transition4(mm_index_mask.m, input0,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		input1 = transition4(mm_index_mask.m, input1,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies3, &indicies4);
> +
> +		 /* Check for any matches. */
> +		acl_match_check_x4(0, ctx, parms, &flows,
> +			&indicies1, &indicies2, mm_match_mask.m);
> +		acl_match_check_x4(4, ctx, parms, &flows,
> +			&indicies3, &indicies4, mm_match_mask.m);
> +	}
> +
> +	return 0;
> +}
> +
> +/*
> + * Execute trie traversal with 4 traversals in parallel
> + */
> +static inline int
> +search_sse_4(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +	 uint32_t *results, int total_packets, uint32_t categories)
> +{
> +	int n;
> +	struct acl_flow_data flows;
> +	uint64_t index_array[MAX_SEARCHES_SSE4];
> +	struct completion cmplt[MAX_SEARCHES_SSE4];
> +	struct parms parms[MAX_SEARCHES_SSE4];
> +	xmm_t input, indicies1, indicies2;
> +
> +	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> +		total_packets, categories, ctx->trans_table);
> +
> +	for (n = 0; n < MAX_SEARCHES_SSE4; n++) {
> +		cmplt[n].count = 0;
> +		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> +	}
> +
> +	indicies1 = MM_LOADU((xmm_t *) &index_array[0]);
> +	indicies2 = MM_LOADU((xmm_t *) &index_array[2]);
> +
> +	/* Check for any matches. */
> +	acl_match_check_x4(0, ctx, parms, &flows,
> +		&indicies1, &indicies2, mm_match_mask.m);
> +
> +	while (flows.started > 0) {
> +
> +		/* Gather 4 bytes of input data for each stream. */
> +		input = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0), 0);
> +		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 1), 1);
> +		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 2), 2);
> +		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 3), 3);
> +
> +		/* Process the 4 bytes of input on each stream. */
> +		input = transition4(mm_index_mask.m, input,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		 input = transition4(mm_index_mask.m, input,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		 input = transition4(mm_index_mask.m, input,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		 input = transition4(mm_index_mask.m, input,
> +			mm_shuffle_input.m, mm_ones_16.m,
> +			mm_bytes.m, mm_type_quad_range.m,
> +			flows.trans, &indicies1, &indicies2);
> +
> +		/* Check for any matches. */
> +		acl_match_check_x4(0, ctx, parms, &flows,
> +			&indicies1, &indicies2, mm_match_mask.m);
> +	}
> +
> +	return 0;
> +}
> +
> +static inline xmm_t
> +transition2(xmm_t index_mask, xmm_t next_input, xmm_t shuffle_input,
> +	xmm_t ones_16, xmm_t bytes, xmm_t type_quad_range,
> +	const uint64_t *trans, xmm_t *indicies1)
> +{
> +	uint64_t t;
> +	xmm_t addr, indicies2;
> +
> +	indicies2 = MM_XOR(ones_16, ones_16);
> +
> +	addr = acl_calc_addr(index_mask, next_input, shuffle_input, ones_16,
> +		bytes, type_quad_range, indicies1, &indicies2);
> +
> +	/* Gather 64 bit transitions and pack 2 per register. */
> +
> +	t = trans[MM_CVT32(addr)];
> +
> +	/* get slot 1 */
> +	addr = MM_SHUFFLE32(addr, SHUFFLE32_SLOT1);
> +	*indicies1 = MM_SET64(trans[MM_CVT32(addr)], t);
> +
> +	return MM_SRL32(next_input, 8);
> +}
> +
> +/*
> + * Execute trie traversal with 2 traversals in parallel.
> + */
> +static inline int
> +search_sse_2(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +	uint32_t *results, uint32_t total_packets, uint32_t categories)
> +{
> +	int n;
> +	struct acl_flow_data flows;
> +	uint64_t index_array[MAX_SEARCHES_SSE2];
> +	struct completion cmplt[MAX_SEARCHES_SSE2];
> +	struct parms parms[MAX_SEARCHES_SSE2];
> +	xmm_t input, indicies;
> +
> +	acl_set_flow(&flows, cmplt, RTE_DIM(cmplt), data, results,
> +		total_packets, categories, ctx->trans_table);
> +
> +	for (n = 0; n < MAX_SEARCHES_SSE2; n++) {
> +		cmplt[n].count = 0;
> +		index_array[n] = acl_start_next_trie(&flows, parms, n, ctx);
> +	}
> +
> +	indicies = MM_LOADU((xmm_t *) &index_array[0]);
> +
> +	/* Check for any matches. */
> +	acl_match_check_x2(0, ctx, parms, &flows, &indicies, mm_match_mask64.m);
> +
> +	while (flows.started > 0) {
> +
> +		/* Gather 4 bytes of input data for each stream. */
> +		input = MM_INSERT32(mm_ones_16.m, GET_NEXT_4BYTES(parms, 0), 0);
> +		input = MM_INSERT32(input, GET_NEXT_4BYTES(parms, 1), 1);
> +
> +		/* Process the 4 bytes of input on each stream. */
> +
> +		input = transition2(mm_index_mask64.m, input,
> +			mm_shuffle_input64.m, mm_ones_16.m,
> +			mm_bytes64.m, mm_type_quad_range64.m,
> +			flows.trans, &indicies);
> +
> +		input = transition2(mm_index_mask64.m, input,
> +			mm_shuffle_input64.m, mm_ones_16.m,
> +			mm_bytes64.m, mm_type_quad_range64.m,
> +			flows.trans, &indicies);
> +
> +		input = transition2(mm_index_mask64.m, input,
> +			mm_shuffle_input64.m, mm_ones_16.m,
> +			mm_bytes64.m, mm_type_quad_range64.m,
> +			flows.trans, &indicies);
> +
> +		input = transition2(mm_index_mask64.m, input,
> +			mm_shuffle_input64.m, mm_ones_16.m,
> +			mm_bytes64.m, mm_type_quad_range64.m,
> +			flows.trans, &indicies);
> +
> +		/* Check for any matches. */
> +		acl_match_check_x2(0, ctx, parms, &flows, &indicies,
> +			mm_match_mask64.m);
> +	}
> +
> +	return 0;
> +}
> +
> +int
> +rte_acl_classify_sse(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +	uint32_t *results, uint32_t num, uint32_t categories)
> +{
> +	if (categories != 1 &&
> +		((RTE_ACL_RESULTS_MULTIPLIER - 1) & categories) != 0)
> +		return -EINVAL;
> +
> +	if (likely(num >= MAX_SEARCHES_SSE8))
> +		return search_sse_8(ctx, data, results, num, categories);
> +	else if (num >= MAX_SEARCHES_SSE4)
> +		return search_sse_4(ctx, data, results, num, categories);
> +	else
> +		return search_sse_2(ctx, data, results, num, categories);
> +}
> diff --git a/lib/librte_acl/rte_acl.c b/lib/librte_acl/rte_acl.c
> index 7c288bd..b9173c1 100644
> --- a/lib/librte_acl/rte_acl.c
> +++ b/lib/librte_acl/rte_acl.c
> @@ -38,6 +38,52 @@
> 
>  TAILQ_HEAD(rte_acl_list, rte_tailq_entry);
> 
> +typedef int (*rte_acl_classify_t)
> +(const struct rte_acl_ctx *, const uint8_t **, uint32_t *, uint32_t, uint32_t);
> +
> +extern int
> +rte_acl_classify_scalar(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +        uint32_t *results, uint32_t num, uint32_t categories);
> +
> +/* by default, use always avaialbe scalar code path. */
> +rte_acl_classify_t rte_acl_default_classify = rte_acl_classify_scalar;

Why not 'static'?
I thought you'd like to hide it  from external world.

> +
> +void rte_acl_select_classify(enum acl_classify_alg alg)
> +{
> +
> +	switch(alg)
> +	{
> +		case ACL_CLASSIFY_DEFAULT:
> +		case ACL_CLASSIFY_SCALAR:
> +			rte_acl_default_classify = rte_acl_classify_scalar;
> +			break;
> +		case ACL_CLASSIFY_SSE:
> +			rte_acl_default_classify = rte_acl_classify_sse;
> +			break;
> +	}
> +
> +}

As this is init phase function, I suppose we can add check that alg has a valid(supported) value, and return some error as return value, if not.  

> +
> +static void __attribute__((constructor))
> +rte_acl_init(void)
> +{
> +	enum acl_classify_alg alg = ACL_CLASSIFY_DEFAULT;
> +
> +	if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_SSE4_1))
> +		alg = ACL_CLASSIFY_SSE;
> +
> +	rte_acl_select_classify(alg);
> +}
> +
> +inline int rte_acl_classify(const struct rte_acl_ctx *ctx,
> +                            const uint8_t **data,
> +                            uint32_t *results, uint32_t num,
> +                            uint32_t categories)
> +{
> +	return rte_acl_default_classify(ctx, data, results, num, categories);
> +}
> +
> +
>  struct rte_acl_ctx *
>  rte_acl_find_existing(const char *name)
>  {
> diff --git a/lib/librte_acl/rte_acl.h b/lib/librte_acl/rte_acl.h
> index afc0f69..650b306 100644
> --- a/lib/librte_acl/rte_acl.h
> +++ b/lib/librte_acl/rte_acl.h
> @@ -267,6 +267,9 @@ rte_acl_reset(struct rte_acl_ctx *ctx);
>   * RTE_ACL_RESULTS_MULTIPLIER and can't be bigger than RTE_ACL_MAX_CATEGORIES.
>   * If more than one rule is applicable for given input buffer and
>   * given category, then rule with highest priority will be returned as a match.
> + * Note, that this function could be run only on CPUs with SSE4.1 support.
> + * It is up to the caller to make sure that this function is only invoked on
> + * a machine that supports SSE4.1 ISA.
>   * Note, that it is a caller responsibility to ensure that input parameters
>   * are valid and point to correct memory locations.
>   *
> @@ -286,9 +289,10 @@ rte_acl_reset(struct rte_acl_ctx *ctx);
>   * @return
>   *   zero on successful completion.
>   *   -EINVAL for incorrect arguments.
> + *   -ENOTSUP for unsupported platforms.

Please remove the line above: current implementation doesn't return ENOTSUP
(I think that was left from v1).

>   */
>  int
> -rte_acl_classify(const struct rte_acl_ctx *ctx, const uint8_t **data,
> +rte_acl_classify_sse(const struct rte_acl_ctx *ctx, const uint8_t **data,
>  	uint32_t *results, uint32_t num, uint32_t categories);
> 
>  /**
> @@ -323,9 +327,23 @@ rte_acl_classify(const struct rte_acl_ctx *ctx, const uint8_t **data,
>   *   zero on successful completion.
>   *   -EINVAL for incorrect arguments.
>   */
> -int
> -rte_acl_classify_scalar(const struct rte_acl_ctx *ctx, const uint8_t **data,
> -	uint32_t *results, uint32_t num, uint32_t categories);


As I said above we'd better keep it.  

> +
> +enum acl_classify_alg {
> +	ACL_CLASSIFY_DEFAULT = 0,
> +	ACL_CLASSIFY_SCALAR = 1,
> +	ACL_CLASSIFY_SSE = 2,
> +};

As a nit: as this emum is part of public API, I think it is better to add rte_ prefix: enum rte_acl_classify_alg

> +
> +extern inline int rte_acl_classify(const struct rte_acl_ctx *ctx,
> +				   const uint8_t **data,
> +				   uint32_t *results, uint32_t num,
> +				   uint32_t categories);

Again as a nit: here and everywhere can we keep same style through the whole DPDK - function name from the new line:
extern nt
rte_acl_classify(...);

> +/**
> + * Analyze ISA of the current CPU and points rte_acl_default_classify
> + * to the highest applicable version of classify function.
> + */
> +extern void
> +rte_acl_select_classify(enum acl_classify_alg alg);
> 
>  /**
>   * Dump an ACL context structure to the console.
> --
> 1.9.3

  reply	other threads:[~2014-08-25 16:27 UTC|newest]

Thread overview: 21+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-08-07 18:31 [dpdk-dev] [PATCHv2] " Konstantin Ananyev
2014-08-07 20:11 ` Neil Horman
2014-08-07 20:58   ` Vincent JARDIN
2014-08-07 21:28     ` Chris Wright
2014-08-08  2:07       ` Neil Horman
2014-08-08 11:49   ` Ananyev, Konstantin
2014-08-08 12:25     ` Neil Horman
2014-08-08 13:09       ` Ananyev, Konstantin
2014-08-08 14:30         ` Neil Horman
2014-08-11 22:23           ` Thomas Monjalon
2014-08-21 20:15 ` [dpdk-dev] [PATCHv3] " Neil Horman
2014-08-25 16:30   ` Ananyev, Konstantin [this message]
2014-08-26 17:44     ` Neil Horman
2014-08-27 11:25       ` Ananyev, Konstantin
2014-08-27 18:56         ` Neil Horman
2014-08-27 19:18           ` Ananyev, Konstantin
2014-08-28  9:02             ` Richardson, Bruce
2014-08-28 15:55             ` Neil Horman
2014-08-28 20:38 ` [dpdk-dev] [PATCHv4] " Neil Horman
2014-08-29 17:58   ` Ananyev, Konstantin
2014-09-01 11:05     ` Thomas Monjalon

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=2601191342CEEE43887BDE71AB9772582135D369@IRSMSX105.ger.corp.intel.com \
    --to=konstantin.ananyev@intel.com \
    --cc=dev@dpdk.org \
    --cc=nhorman@tuxdriver.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).