From: Jan Viktorin <viktorin@rehivetech.com>
To: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Cc: dev@dpdk.org
Subject: Re: [dpdk-dev] [PATCH 1/3] eal: introduce rte_vect_* abstractions
Date: Wed, 2 Dec 2015 14:43:34 +0100 [thread overview]
Message-ID: <20151202144334.1a66676d@pcviktorin.fit.vutbr.cz> (raw)
In-Reply-To: <1448904253-12929-2-git-send-email-jerin.jacob@caviumnetworks.com>
On Mon, 30 Nov 2015 22:54:11 +0530
Jerin Jacob <jerin.jacob@caviumnetworks.com> wrote:
> introduce rte_vect_* abstractions to remove SSE/AVX specific
> code in the common code(i.e the test applications)
>
> The patch does not provide any functional change for IA, the goal is to
Does IA mean Intel Architecture?
> have infrastructure to reuse the common vector-based test code across
> all the architectures.
>
> Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
> ---
> lib/librte_eal/common/include/arch/arm/rte_vect.h | 17 ++++++++++++++++-
> lib/librte_eal/common/include/arch/x86/rte_vect.h | 8 ++++++++
> 2 files changed, 24 insertions(+), 1 deletion(-)
>
> diff --git a/lib/librte_eal/common/include/arch/arm/rte_vect.h b/lib/librte_eal/common/include/arch/arm/rte_vect.h
> index 21cdb4d..d300951 100644
> --- a/lib/librte_eal/common/include/arch/arm/rte_vect.h
> +++ b/lib/librte_eal/common/include/arch/arm/rte_vect.h
> @@ -33,13 +33,14 @@
> #ifndef _RTE_VECT_ARM_H_
> #define _RTE_VECT_ARM_H_
>
> -#include "arm_neon.h"
> +#include <arm_neon.h>
>
> #ifdef __cplusplus
> extern "C" {
> #endif
>
> typedef int32x4_t xmm_t;
> +typedef int32x4_t __m128i;
As Jianbo pointed out recently, the __m128i type should be refactored in
a general rte_vect API too. If we do something like
#if SSE
typedef __m128i rte_128i;
#elif NEON
typedef int32x4_y rte_128i;
#endif
does it make somebody angry? I am afraid that it will influence a lot of
code. However, from the ABI point of view, it is OK, isn't it?
>
> #define XMM_SIZE (sizeof(xmm_t))
> #define XMM_MASK (XMM_SIZE - 1)
> @@ -53,6 +54,20 @@ typedef union rte_xmm {
> double pd[XMM_SIZE / sizeof(double)];
> } __attribute__((aligned(16))) rte_xmm_t;
>
> +/* rte_vect_* abstraction implementation using NEON */
> +
> +/* loads the __m128i value from address p(does not need to be 16-byte aligned)*/
> +#define rte_vect_loadu_sil128(p) vld1q_s32((const int32_t *)p)
> +
> +/* sets the 4 signed 32-bit integer values and returns the __m128i variable */
> +static inline __m128i __attribute__((always_inline))
> +rte_vect_set_epi32(int i3, int i2, int i1, int i0)
> +{
> + int32_t data[4] = {i0, i1, i2, i3};
> +
> + return vld1q_s32(data);
> +}
> +
> #ifdef __cplusplus
> }
> #endif
> diff --git a/lib/librte_eal/common/include/arch/x86/rte_vect.h b/lib/librte_eal/common/include/arch/x86/rte_vect.h
> index b698797..91c6523 100644
> --- a/lib/librte_eal/common/include/arch/x86/rte_vect.h
> +++ b/lib/librte_eal/common/include/arch/x86/rte_vect.h
> @@ -125,6 +125,14 @@ typedef union rte_ymm {
> })
> #endif /* (defined(__ICC) && __ICC < 1210) */
>
> +/* rte_vect_* abstraction implementation using SSE */
> +
> +/* loads the __m128i value from address p(does not need to be 16-byte aligned)*/
> +#define rte_vect_loadu_sil128(p) _mm_loadu_si128(p)
> +
> +/* sets the 4 signed 32-bit integer values and returns the __m128i variable */
> +#define rte_vect_set_epi32(i3, i2, i1, i0) _mm_set_epi32(i3, i2, i1, i0)
> +
> #ifdef __cplusplus
> }
> #endif
I like this approach. It is a question whether to inherit names from
SSE. However, why to reinvent the wheel...
We probably need other people to give their ideas about such
generalization of the API.
I think, there should be an autotest of the rte_vect API. Is it
possible to create one?
Regards
Jan
--
Jan Viktorin E-mail: Viktorin@RehiveTech.com
System Architect Web: www.RehiveTech.com
RehiveTech
Brno, Czech Republic
next prev parent reply other threads:[~2015-12-02 13:46 UTC|newest]
Thread overview: 47+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-11-30 17:24 [dpdk-dev] [PATCH 0/3] add lpm support for NEON Jerin Jacob
2015-11-30 17:24 ` [dpdk-dev] [PATCH 1/3] eal: introduce rte_vect_* abstractions Jerin Jacob
2015-12-02 13:43 ` Jan Viktorin [this message]
2015-12-02 14:51 ` Jerin Jacob
2015-11-30 17:24 ` [dpdk-dev] [PATCH 2/3] lpm: add support for NEON Jerin Jacob
2015-12-02 13:43 ` Jan Viktorin
2015-12-02 14:56 ` Jerin Jacob
2015-12-02 15:00 ` Jan Viktorin
2015-11-30 17:24 ` [dpdk-dev] [PATCH 3/3] maintainers: claim responsibility for arm64 specific files of hash and lpm Jerin Jacob
2015-12-02 13:43 ` Jan Viktorin
2015-12-02 14:57 ` Jerin Jacob
2015-12-02 13:43 ` [dpdk-dev] [PATCH 0/3] add lpm support for NEON Jan Viktorin
2015-12-02 14:41 ` Jerin Jacob
2015-12-04 15:14 ` [dpdk-dev] [PATCH v2 " Jerin Jacob
2015-12-04 15:14 ` [dpdk-dev] [PATCH v2 1/3] lpm: make rte_lpm_lookupx4 API definition architecture agnostic Jerin Jacob
2015-12-07 6:15 ` Jianbo Liu
2015-12-07 6:57 ` Jerin Jacob
2015-12-07 14:06 ` Ananyev, Konstantin
2015-12-04 15:14 ` [dpdk-dev] [PATCH v2 2/3] lpm: add support for NEON Jerin Jacob
2015-12-04 15:14 ` [dpdk-dev] [PATCH v2 3/3] maintainers: claim responsibility for arm64 specific files of hash and lpm Jerin Jacob
2016-01-29 4:10 ` [dpdk-dev] [PATCH v3 0/3] add lpm support for NEON Jerin Jacob
2016-01-29 4:10 ` [dpdk-dev] [PATCH v3 1/3] lpm: make rte_lpm_lookupx4 API definition architecture agnostic Jerin Jacob
2016-01-29 4:10 ` [dpdk-dev] [PATCH v3 2/3] lpm: add support for NEON Jerin Jacob
2016-02-11 11:46 ` Thomas Monjalon
2016-02-12 6:47 ` Jerin Jacob
2016-02-12 8:42 ` Thomas Monjalon
2016-01-29 4:10 ` [dpdk-dev] [PATCH v3 3/3] maintainers: claim responsibility for arm64 specific files of hash and lpm Jerin Jacob
2016-02-08 9:29 ` [dpdk-dev] [PATCH v3 0/3] add lpm support for NEON Jerin Jacob
2016-02-12 12:28 ` [dpdk-dev] [PATCH v4 " Jerin Jacob
2016-02-12 12:28 ` [dpdk-dev] [PATCH v4 1/3] lpm: make rte_lpm_lookupx4 API definition architecture agnostic Jerin Jacob
2016-03-01 17:42 ` Thomas Monjalon
2016-03-02 6:28 ` Jerin Jacob
2016-02-12 12:28 ` [dpdk-dev] [PATCH v4 2/3] lpm: add support for NEON Jerin Jacob
2016-03-01 17:46 ` Thomas Monjalon
2016-03-02 6:45 ` Jerin Jacob
2016-02-12 12:28 ` [dpdk-dev] [PATCH v4 3/3] maintainers: claim responsibility for arm64 specific files of hash and lpm Jerin Jacob
2016-03-01 17:47 ` Thomas Monjalon
2016-03-02 6:46 ` Jerin Jacob
2016-02-16 13:27 ` [dpdk-dev] [PATCH v4 0/3] add lpm support for NEON Kobylinski, MichalX
2016-02-16 16:44 ` Jerin Jacob
2016-02-18 10:26 ` Kobylinski, MichalX
2016-02-19 0:34 ` Jerin Jacob
2016-03-11 3:52 ` [dpdk-dev] [PATCH v5 " Jerin Jacob
2016-03-11 3:52 ` [dpdk-dev] [PATCH v5 1/3] lpm: make rte_lpm_lookupx4 API definition architecture agnostic Jerin Jacob
2016-03-11 3:52 ` [dpdk-dev] [PATCH v5 2/3] lpm: add support for NEON Jerin Jacob
2016-03-11 3:52 ` [dpdk-dev] [PATCH v5 3/3] Maintainers: claim responsibility for arm64 specific files of hash Jerin Jacob
2016-03-11 14:24 ` [dpdk-dev] [PATCH v5 0/3] add lpm support for NEON 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=20151202144334.1a66676d@pcviktorin.fit.vutbr.cz \
--to=viktorin@rehivetech.com \
--cc=dev@dpdk.org \
--cc=jerin.jacob@caviumnetworks.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).