[v4,6/8] net/mlx5: introduce Tx burst routine template

Message ID 1563719100-368-7-git-send-email-viacheslavo@mellanox.com
State Accepted, archived
Delegated to: Raslan Darawsheh
Headers show
Series
  • net/mlx5: consolidate Tx datapath
Related show

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/checkpatch success coding style OK

Commit Message

Viacheslav Ovsiienko July 21, 2019, 2:24 p.m.
Mellanox NICs support the wide set of Tx offloads. The supported
offloads are reported by the mlx5 PMD in rte_eth_dev_info tx_offload_capa
field. An application may choose any combination of supported offloads
and configure the device appropriately. Some of Tx offloads may be
not requested by application, or ever all of them may be omitted.
Most of the Tx offloads require some code branches in tx_burst routine
to support ones. If Tx offload is not requested the tx_burst routine
code may be significantly simplified and consume less CPU cycles.

For example, if application does not engage TSO offload this code
can be omitted, if multi-segment packet is not supposed the tx_burst
may assume single mbuf packets only, etc.

Currently, the mlx5 PMD implements multiple tx_burst subroutines
for most common combinations of requested Tx offloads, each branch
has its own dedicated implementation. It is not very easy to update,
support and develop such kind of code - multiple branches impose
the multiple points to process. Also many of frequently requested
offload combinations are not supported yet. That leads to selecting of
not completely matching tx_burst routine and harms the performance.

This patch introduces the new approach for tx_burst code. It is proposed
to develop the unified template for tx_burst routine, which supports
all the Tx offloads and takes the compile time defined parameter
describing the supposed set of supported offloads. On the base
of this template, the compiler is able to generate multiple tx_burst
routines highly optimized for the statically specified set of Tx offloads.
Next, in runtime, at Tx queue configuration the best matching optimized
implementation of tx_burst is chosen.

This patch intentionally omits the template internal implementation,
but just introduces the template itself to emboss the approach of
the multiple specially tuned tx_burst routines.

Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
---
 drivers/net/mlx5/mlx5_rxtx.c | 520 ++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 516 insertions(+), 4 deletions(-)

Comments

Yongseok Koh July 22, 2019, 5:34 a.m. | #1
> On Jul 21, 2019, at 7:24 AM, Viacheslav Ovsiienko <viacheslavo@mellanox.com> wrote:
> 
> Mellanox NICs support the wide set of Tx offloads. The supported
> offloads are reported by the mlx5 PMD in rte_eth_dev_info tx_offload_capa
> field. An application may choose any combination of supported offloads
> and configure the device appropriately. Some of Tx offloads may be
> not requested by application, or ever all of them may be omitted.
> Most of the Tx offloads require some code branches in tx_burst routine
> to support ones. If Tx offload is not requested the tx_burst routine
> code may be significantly simplified and consume less CPU cycles.
> 
> For example, if application does not engage TSO offload this code
> can be omitted, if multi-segment packet is not supposed the tx_burst
> may assume single mbuf packets only, etc.
> 
> Currently, the mlx5 PMD implements multiple tx_burst subroutines
> for most common combinations of requested Tx offloads, each branch
> has its own dedicated implementation. It is not very easy to update,
> support and develop such kind of code - multiple branches impose
> the multiple points to process. Also many of frequently requested
> offload combinations are not supported yet. That leads to selecting of
> not completely matching tx_burst routine and harms the performance.
> 
> This patch introduces the new approach for tx_burst code. It is proposed
> to develop the unified template for tx_burst routine, which supports
> all the Tx offloads and takes the compile time defined parameter
> describing the supposed set of supported offloads. On the base
> of this template, the compiler is able to generate multiple tx_burst
> routines highly optimized for the statically specified set of Tx offloads.
> Next, in runtime, at Tx queue configuration the best matching optimized
> implementation of tx_burst is chosen.
> 
> This patch intentionally omits the template internal implementation,
> but just introduces the template itself to emboss the approach of
> the multiple specially tuned tx_burst routines.
> 
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> ---

Acked-by: Yongseok Koh <yskoh@mellanox.com>

> drivers/net/mlx5/mlx5_rxtx.c | 520 ++++++++++++++++++++++++++++++++++++++++++-
> 1 file changed, 516 insertions(+), 4 deletions(-)
> 
> diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
> index 13f9431..16be493 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.c
> +++ b/drivers/net/mlx5/mlx5_rxtx.c
> @@ -1,6 +1,6 @@
> /* SPDX-License-Identifier: BSD-3-Clause
>  * Copyright 2015 6WIND S.A.
> - * Copyright 2015 Mellanox Technologies, Ltd
> + * Copyright 2015-2019 Mellanox Technologies, Ltd
>  */
> 
> #include <assert.h>
> @@ -34,6 +34,57 @@
> #include "mlx5_defs.h"
> #include "mlx5_prm.h"
> 
> +/* TX burst subroutines return codes. */
> +enum mlx5_txcmp_code {
> +	MLX5_TXCMP_CODE_EXIT = 0,
> +	MLX5_TXCMP_CODE_ERROR,
> +	MLX5_TXCMP_CODE_SINGLE,
> +	MLX5_TXCMP_CODE_MULTI,
> +	MLX5_TXCMP_CODE_TSO,
> +	MLX5_TXCMP_CODE_EMPW,
> +};
> +
> +/*
> + * These defines are used to configure Tx burst routine option set
> + * supported at compile time. The not specified options are optimized out
> + * out due to if conditions can be explicitly calculated at compile time.
> + * The offloads with bigger runtime check (require more CPU cycles to
> + * skip) overhead should have the bigger index - this is needed to
> + * select the better matching routine function if no exact match and
> + * some offloads are not actually requested.
> + */
> +#define MLX5_TXOFF_CONFIG_MULTI (1u << 0) /* Multi-segment packets.*/
> +#define MLX5_TXOFF_CONFIG_TSO (1u << 1) /* TCP send offload supported.*/
> +#define MLX5_TXOFF_CONFIG_SWP (1u << 2) /* Tunnels/SW Parser offloads.*/
> +#define MLX5_TXOFF_CONFIG_CSUM (1u << 3) /* Check Sums offloaded. */
> +#define MLX5_TXOFF_CONFIG_INLINE (1u << 4) /* Data inlining supported. */
> +#define MLX5_TXOFF_CONFIG_VLAN (1u << 5) /* VLAN insertion supported.*/
> +#define MLX5_TXOFF_CONFIG_METADATA (1u << 6) /* Flow metadata. */
> +#define MLX5_TXOFF_CONFIG_EMPW (1u << 8) /* Enhanced MPW supported.*/
> +
> +/* The most common offloads groups. */
> +#define MLX5_TXOFF_CONFIG_NONE 0
> +#define MLX5_TXOFF_CONFIG_FULL (MLX5_TXOFF_CONFIG_MULTI | \
> +				MLX5_TXOFF_CONFIG_TSO | \
> +				MLX5_TXOFF_CONFIG_SWP | \
> +				MLX5_TXOFF_CONFIG_CSUM | \
> +				MLX5_TXOFF_CONFIG_INLINE | \
> +				MLX5_TXOFF_CONFIG_VLAN | \
> +				MLX5_TXOFF_CONFIG_METADATA)
> +
> +#define MLX5_TXOFF_CONFIG(mask) (olx & MLX5_TXOFF_CONFIG_##mask)
> +
> +#define MLX5_TXOFF_DECL(func, olx) \
> +static uint16_t mlx5_tx_burst_##func(void *txq, \
> +				     struct rte_mbuf **pkts, \
> +				    uint16_t pkts_n) \
> +{ \
> +	return mlx5_tx_burst_tmpl((struct mlx5_txq_data *restrict)txq, \
> +		    pkts, pkts_n, (olx)); \
> +}
> +
> +#define MLX5_TXOFF_INFO(func, olx) {mlx5_tx_burst_##func, olx},
> +
> static __rte_always_inline uint32_t
> rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe);
> 
> @@ -1531,7 +1582,323 @@
> }
> 
> /**
> - * Configure the TX function to use.
> + * DPDK Tx callback template. This is configured template
> + * used to generate routines optimized for specified offload setup.
> + * One of this generated functions is chosen at SQ configuration
> + * time.
> + *
> + * @param txq
> + *   Generic pointer to TX queue structure.
> + * @param[in] pkts
> + *   Packets to transmit.
> + * @param pkts_n
> + *   Number of packets in array.
> + * @param olx
> + *   Configured offloads mask, presents the bits of MLX5_TXOFF_CONFIG_xxx
> + *   values. Should be static to take compile time static configuration
> + *   advantages.
> + *
> + * @return
> + *   Number of packets successfully transmitted (<= pkts_n).
> + */
> +static __rte_always_inline uint16_t
> +mlx5_tx_burst_tmpl(struct mlx5_txq_data *restrict txq,
> +		   struct rte_mbuf **restrict pkts,
> +		   uint16_t pkts_n,
> +		   unsigned int olx)
> +{
> +	(void)txq;
> +	(void)pkts;
> +	(void)pkts_n;
> +	(void)olx;
> +	return 0;
> +}
> +
> +/* Generate routines with Enhanced Multi-Packet Write support. */
> +MLX5_TXOFF_DECL(full_empw,
> +		MLX5_TXOFF_CONFIG_FULL | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(none_empw,
> +		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(md_empw,
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(mt_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(mtsc_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(mti_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(mtv_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(mtiv_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(sc_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(sci_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(scv_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(sciv_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(i_empw,
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(v_empw,
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_DECL(iv_empw,
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +/* Generate routines without Enhanced Multi-Packet Write support. */
> +MLX5_TXOFF_DECL(full,
> +		MLX5_TXOFF_CONFIG_FULL)
> +
> +MLX5_TXOFF_DECL(none,
> +		MLX5_TXOFF_CONFIG_NONE)
> +
> +MLX5_TXOFF_DECL(md,
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(mt,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(mtsc,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(mti,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +
> +MLX5_TXOFF_DECL(mtv,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +
> +MLX5_TXOFF_DECL(mtiv,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(sc,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(sci,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +
> +MLX5_TXOFF_DECL(scv,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +
> +MLX5_TXOFF_DECL(sciv,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(i,
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(v,
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_DECL(iv,
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +/*
> + * Array of declared and compiled Tx burst function and corresponding
> + * supported offloads set. The array is used to select the Tx burst
> + * function for specified offloads set at Tx queue configuration time.
> + */
> +const struct {
> +	eth_tx_burst_t func;
> +	unsigned int olx;
> +} txoff_func[] = {
> +MLX5_TXOFF_INFO(full_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(none_empw,
> +		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(md_empw,
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(mt_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(mtsc_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(mti_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(mtv_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(mtiv_empw,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(sc_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(sci_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(scv_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(sciv_empw,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(i_empw,
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(v_empw,
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(iv_empw,
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
> +
> +MLX5_TXOFF_INFO(full,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(none,
> +		MLX5_TXOFF_CONFIG_NONE)
> +
> +MLX5_TXOFF_INFO(md,
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(mt,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(mtsc,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(mti,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +
> +MLX5_TXOFF_INFO(mtv,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(mtiv,
> +		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(sc,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(sci,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(scv,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(sciv,
> +		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(i,
> +		MLX5_TXOFF_CONFIG_INLINE |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(v,
> +		MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +
> +MLX5_TXOFF_INFO(iv,
> +		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
> +		MLX5_TXOFF_CONFIG_METADATA)
> +};
> +
> +/**
> + * Configure the Tx function to use. The routine checks configured
> + * Tx offloads for the device and selects appropriate Tx burst
> + * routine. There are multiple Tx burst routines compiled from
> + * the same template in the most optimal way for the dedicated
> + * Tx offloads set.
>  *
>  * @param dev
>  *   Pointer to private data structure.
> @@ -1542,8 +1909,153 @@
> eth_tx_burst_t
> mlx5_select_tx_function(struct rte_eth_dev *dev)
> {
> -	(void)dev;
> -	return removed_tx_burst;
> +	struct mlx5_priv *priv = dev->data->dev_private;
> +	struct mlx5_dev_config *config = &priv->config;
> +	uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
> +	unsigned int diff = 0, olx = 0, i, m;
> +
> +	static_assert(MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE <=
> +		      MLX5_DSEG_MAX, "invalid WQE max size");
> +	static_assert(MLX5_WQE_CSEG_SIZE == MLX5_WSEG_SIZE,
> +		      "invalid WQE Control Segment size");
> +	static_assert(MLX5_WQE_ESEG_SIZE == MLX5_WSEG_SIZE,
> +		      "invalid WQE Ethernet Segment size");
> +	static_assert(MLX5_WQE_DSEG_SIZE == MLX5_WSEG_SIZE,
> +		      "invalid WQE Data Segment size");
> +	static_assert(MLX5_WQE_SIZE == 4 * MLX5_WSEG_SIZE,
> +		      "invalid WQE size");
> +	assert(priv);
> +	if (tx_offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
> +		/* We should support Multi-Segment Packets. */
> +		olx |= MLX5_TXOFF_CONFIG_MULTI;
> +	}
> +	if (tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO |
> +			   DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
> +			   DEV_TX_OFFLOAD_GRE_TNL_TSO |
> +			   DEV_TX_OFFLOAD_IP_TNL_TSO |
> +			   DEV_TX_OFFLOAD_UDP_TNL_TSO)) {
> +		/* We should support TCP Send Offload. */
> +		olx |= MLX5_TXOFF_CONFIG_TSO;
> +	}
> +	if (tx_offloads & (DEV_TX_OFFLOAD_IP_TNL_TSO |
> +			   DEV_TX_OFFLOAD_UDP_TNL_TSO |
> +			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
> +		/* We should support Software Parser for Tunnels. */
> +		olx |= MLX5_TXOFF_CONFIG_SWP;
> +	}
> +	if (tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
> +			   DEV_TX_OFFLOAD_UDP_CKSUM |
> +			   DEV_TX_OFFLOAD_TCP_CKSUM |
> +			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
> +		/* We should support IP/TCP/UDP Checksums. */
> +		olx |= MLX5_TXOFF_CONFIG_CSUM;
> +	}
> +	if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT) {
> +		/* We should support VLAN insertion. */
> +		olx |= MLX5_TXOFF_CONFIG_VLAN;
> +	}
> +	if (priv->txqs_n && (*priv->txqs)[0]) {
> +		struct mlx5_txq_data *txd = (*priv->txqs)[0];
> +
> +		if (txd->inlen_send) {
> +			/*
> +			 * Check the data inline requirements. Data inline
> +			 * is enabled on per device basis, we can check
> +			 * the first Tx queue only.
> +			 *
> +			 * If device does not support VLAN insertion in WQE
> +			 * and some queues are requested to perform VLAN
> +			 * insertion offload than inline must be enabled.
> +			 */
> +			olx |= MLX5_TXOFF_CONFIG_INLINE;
> +		}
> +	}
> +	if (config->mps == MLX5_MPW_ENHANCED &&
> +	    config->txq_inline_min <= 0) {
> +		/*
> +		 * The NIC supports Enhanced Multi-Packet Write.
> +		 * We do not support legacy MPW due to its
> +		 * hardware related problems, so we just ignore
> +		 * legacy MLX5_MPW settings. There should be no
> +		 * minimal required inline data.
> +		 */
> +		olx |= MLX5_TXOFF_CONFIG_EMPW;
> +	}
> +	if (tx_offloads & DEV_TX_OFFLOAD_MATCH_METADATA) {
> +		/* We should support Flow metadata. */
> +		olx |= MLX5_TXOFF_CONFIG_METADATA;
> +	}
> +	/*
> +	 * Scan the routines table to find the minimal
> +	 * satisfying routine with requested offloads.
> +	 */
> +	m = RTE_DIM(txoff_func);
> +	for (i = 0; i < RTE_DIM(txoff_func); i++) {
> +		unsigned int tmp;
> +
> +		tmp = txoff_func[i].olx;
> +		if (tmp == olx) {
> +			/* Meets requested offloads exactly.*/
> +			m = i;
> +			break;
> +		}
> +		if ((tmp & olx) != olx) {
> +			/* Does not meet requested offloads at all. */
> +			continue;
> +		}
> +		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_EMPW)
> +			/* Do not enable eMPW if not configured. */
> +			continue;
> +		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_INLINE)
> +			/* Do not enable inlining if not configured. */
> +			continue;
> +		/*
> +		 * Some routine meets the requirements.
> +		 * Check whether it has minimal amount
> +		 * of not requested offloads.
> +		 */
> +		tmp = __builtin_popcountl(tmp & ~olx);
> +		if (m >= RTE_DIM(txoff_func) || tmp < diff) {
> +			/* First or better match, save and continue. */
> +			m = i;
> +			diff = tmp;
> +			continue;
> +		}
> +		if (tmp == diff) {
> +			tmp = txoff_func[i].olx ^ txoff_func[m].olx;
> +			if (__builtin_ffsl(txoff_func[i].olx & ~tmp) <
> +			    __builtin_ffsl(txoff_func[m].olx & ~tmp)) {
> +				/* Lighter not requested offload. */
> +				m = i;
> +			}
> +		}
> +	}
> +	if (m >= RTE_DIM(txoff_func)) {
> +		DRV_LOG(DEBUG, "port %u has no selected Tx function"
> +			       " for requested offloads %04X",
> +				dev->data->port_id, olx);
> +		return NULL;
> +	}
> +	DRV_LOG(DEBUG, "port %u has selected Tx function"
> +		       " supporting offloads %04X/%04X",
> +			dev->data->port_id, olx, txoff_func[m].olx);
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_MULTI)
> +		DRV_LOG(DEBUG, "\tMULTI (multi segment)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_TSO)
> +		DRV_LOG(DEBUG, "\tTSO   (TCP send offload)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_SWP)
> +		DRV_LOG(DEBUG, "\tSWP   (software parser)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_CSUM)
> +		DRV_LOG(DEBUG, "\tCSUM  (checksum offload)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_INLINE)
> +		DRV_LOG(DEBUG, "\tINLIN (inline data)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_VLAN)
> +		DRV_LOG(DEBUG, "\tVLANI (VLAN insertion)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_METADATA)
> +		DRV_LOG(DEBUG, "\tMETAD (tx Flow metadata)");
> +	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_EMPW)
> +		DRV_LOG(DEBUG, "\tEMPW  (Enhanced MPW)");
> +	return txoff_func[m].func;
> }
> 
> 
> -- 
> 1.8.3.1
>
Ferruh Yigit July 22, 2019, 11:53 a.m. | #2
On 7/21/2019 3:24 PM, Viacheslav Ovsiienko wrote:
> Mellanox NICs support the wide set of Tx offloads. The supported
> offloads are reported by the mlx5 PMD in rte_eth_dev_info tx_offload_capa
> field. An application may choose any combination of supported offloads
> and configure the device appropriately. Some of Tx offloads may be
> not requested by application, or ever all of them may be omitted.
> Most of the Tx offloads require some code branches in tx_burst routine
> to support ones. If Tx offload is not requested the tx_burst routine
> code may be significantly simplified and consume less CPU cycles.
> 
> For example, if application does not engage TSO offload this code
> can be omitted, if multi-segment packet is not supposed the tx_burst
> may assume single mbuf packets only, etc.
> 
> Currently, the mlx5 PMD implements multiple tx_burst subroutines
> for most common combinations of requested Tx offloads, each branch
> has its own dedicated implementation. It is not very easy to update,
> support and develop such kind of code - multiple branches impose
> the multiple points to process. Also many of frequently requested
> offload combinations are not supported yet. That leads to selecting of
> not completely matching tx_burst routine and harms the performance.
> 
> This patch introduces the new approach for tx_burst code. It is proposed
> to develop the unified template for tx_burst routine, which supports
> all the Tx offloads and takes the compile time defined parameter
> describing the supposed set of supported offloads. On the base
> of this template, the compiler is able to generate multiple tx_burst
> routines highly optimized for the statically specified set of Tx offloads.
> Next, in runtime, at Tx queue configuration the best matching optimized
> implementation of tx_burst is chosen.
> 
> This patch intentionally omits the template internal implementation,
> but just introduces the template itself to emboss the approach of
> the multiple specially tuned tx_burst routines.
> 
> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>

Getting following build error with icc, can you please check?


.../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4649): error #191: type qualifier is
meaningless on cast type


  MLX5_TXOFF_DECL(full_empw,



  ^







.../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4652): error #191: type qualifier is
meaningless on cast type


  MLX5_TXOFF_DECL(none_empw,



  ^

Many of same error for "MLX5_TXOFF_DECL" usage.
Viacheslav Ovsiienko July 22, 2019, 1:07 p.m. | #3
> -----Original Message-----
> From: Ferruh Yigit <ferruh.yigit@intel.com>
> Sent: Monday, July 22, 2019 14:53
> To: Slava Ovsiienko <viacheslavo@mellanox.com>; dev@dpdk.org
> Cc: Yongseok Koh <yskoh@mellanox.com>
> Subject: Re: [dpdk-dev] [PATCH v4 6/8] net/mlx5: introduce Tx burst routine
> template
> 
> On 7/21/2019 3:24 PM, Viacheslav Ovsiienko wrote:
> > Mellanox NICs support the wide set of Tx offloads. The supported
> > offloads are reported by the mlx5 PMD in rte_eth_dev_info
> > tx_offload_capa field. An application may choose any combination of
> > supported offloads and configure the device appropriately. Some of Tx
> > offloads may be not requested by application, or ever all of them may be
> omitted.
> > Most of the Tx offloads require some code branches in tx_burst routine
> > to support ones. If Tx offload is not requested the tx_burst routine
> > code may be significantly simplified and consume less CPU cycles.
> >
> > For example, if application does not engage TSO offload this code can
> > be omitted, if multi-segment packet is not supposed the tx_burst may
> > assume single mbuf packets only, etc.
> >
> > Currently, the mlx5 PMD implements multiple tx_burst subroutines for
> > most common combinations of requested Tx offloads, each branch has its
> > own dedicated implementation. It is not very easy to update, support
> > and develop such kind of code - multiple branches impose the multiple
> > points to process. Also many of frequently requested offload
> > combinations are not supported yet. That leads to selecting of not
> > completely matching tx_burst routine and harms the performance.
> >
> > This patch introduces the new approach for tx_burst code. It is
> > proposed to develop the unified template for tx_burst routine, which
> > supports all the Tx offloads and takes the compile time defined
> > parameter describing the supposed set of supported offloads. On the
> > base of this template, the compiler is able to generate multiple
> > tx_burst routines highly optimized for the statically specified set of Tx
> offloads.
> > Next, in runtime, at Tx queue configuration the best matching
> > optimized implementation of tx_burst is chosen.
> >
> > This patch intentionally omits the template internal implementation,
> > but just introduces the template itself to emboss the approach of the
> > multiple specially tuned tx_burst routines.
> >
> > Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> 
> Getting following build error with icc, can you please check?
> 
> 
> .../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4649): error #191: type qualifier is
> meaningless on cast type
> 
> 
>   MLX5_TXOFF_DECL(full_empw,
> 
Sorry, I have no icc at my disposal now. What I can say (sorry again, it is obvious) 
- It definitely does not like the MLX5_TXOFF_DECL macro. Most likely - "restrict" qualifier, I think.

With best regards, Slava
Ferruh Yigit July 22, 2019, 4:45 p.m. | #4
On 7/22/2019 2:07 PM, Slava Ovsiienko wrote:
>> -----Original Message-----
>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>> Sent: Monday, July 22, 2019 14:53
>> To: Slava Ovsiienko <viacheslavo@mellanox.com>; dev@dpdk.org
>> Cc: Yongseok Koh <yskoh@mellanox.com>
>> Subject: Re: [dpdk-dev] [PATCH v4 6/8] net/mlx5: introduce Tx burst routine
>> template
>>
>> On 7/21/2019 3:24 PM, Viacheslav Ovsiienko wrote:
>>> Mellanox NICs support the wide set of Tx offloads. The supported
>>> offloads are reported by the mlx5 PMD in rte_eth_dev_info
>>> tx_offload_capa field. An application may choose any combination of
>>> supported offloads and configure the device appropriately. Some of Tx
>>> offloads may be not requested by application, or ever all of them may be
>> omitted.
>>> Most of the Tx offloads require some code branches in tx_burst routine
>>> to support ones. If Tx offload is not requested the tx_burst routine
>>> code may be significantly simplified and consume less CPU cycles.
>>>
>>> For example, if application does not engage TSO offload this code can
>>> be omitted, if multi-segment packet is not supposed the tx_burst may
>>> assume single mbuf packets only, etc.
>>>
>>> Currently, the mlx5 PMD implements multiple tx_burst subroutines for
>>> most common combinations of requested Tx offloads, each branch has its
>>> own dedicated implementation. It is not very easy to update, support
>>> and develop such kind of code - multiple branches impose the multiple
>>> points to process. Also many of frequently requested offload
>>> combinations are not supported yet. That leads to selecting of not
>>> completely matching tx_burst routine and harms the performance.
>>>
>>> This patch introduces the new approach for tx_burst code. It is
>>> proposed to develop the unified template for tx_burst routine, which
>>> supports all the Tx offloads and takes the compile time defined
>>> parameter describing the supposed set of supported offloads. On the
>>> base of this template, the compiler is able to generate multiple
>>> tx_burst routines highly optimized for the statically specified set of Tx
>> offloads.
>>> Next, in runtime, at Tx queue configuration the best matching
>>> optimized implementation of tx_burst is chosen.
>>>
>>> This patch intentionally omits the template internal implementation,
>>> but just introduces the template itself to emboss the approach of the
>>> multiple specially tuned tx_burst routines.
>>>
>>> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
>>
>> Getting following build error with icc, can you please check?
>>
>>
>> .../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4649): error #191: type qualifier is
>> meaningless on cast type
>>
>>
>>   MLX5_TXOFF_DECL(full_empw,
>>
> Sorry, I have no icc at my disposal now. What I can say (sorry again, it is obvious) 
> - It definitely does not like the MLX5_TXOFF_DECL macro. Most likely - "restrict" qualifier, I think.
> 
> With best regards, Slava
> 

Hi Slava,

I was hoping error log can give enough hint, but I tested by removing the
'restrict' keyword and it fixed the icc warning.

Can you please check if the 'icc' is right and 'restrict' is meaningless, if so
can you please send a patch to remove it?
else please let me know and I can make a patch to disable the specific icc warning.
Yongseok Koh July 22, 2019, 4:49 p.m. | #5
> On Jul 22, 2019, at 9:45 AM, Ferruh Yigit <ferruh.yigit@intel.com> wrote:
> 
> On 7/22/2019 2:07 PM, Slava Ovsiienko wrote:
>>> -----Original Message-----
>>> From: Ferruh Yigit <ferruh.yigit@intel.com>
>>> Sent: Monday, July 22, 2019 14:53
>>> To: Slava Ovsiienko <viacheslavo@mellanox.com>; dev@dpdk.org
>>> Cc: Yongseok Koh <yskoh@mellanox.com>
>>> Subject: Re: [dpdk-dev] [PATCH v4 6/8] net/mlx5: introduce Tx burst routine
>>> template
>>> 
>>> On 7/21/2019 3:24 PM, Viacheslav Ovsiienko wrote:
>>>> Mellanox NICs support the wide set of Tx offloads. The supported
>>>> offloads are reported by the mlx5 PMD in rte_eth_dev_info
>>>> tx_offload_capa field. An application may choose any combination of
>>>> supported offloads and configure the device appropriately. Some of Tx
>>>> offloads may be not requested by application, or ever all of them may be
>>> omitted.
>>>> Most of the Tx offloads require some code branches in tx_burst routine
>>>> to support ones. If Tx offload is not requested the tx_burst routine
>>>> code may be significantly simplified and consume less CPU cycles.
>>>> 
>>>> For example, if application does not engage TSO offload this code can
>>>> be omitted, if multi-segment packet is not supposed the tx_burst may
>>>> assume single mbuf packets only, etc.
>>>> 
>>>> Currently, the mlx5 PMD implements multiple tx_burst subroutines for
>>>> most common combinations of requested Tx offloads, each branch has its
>>>> own dedicated implementation. It is not very easy to update, support
>>>> and develop such kind of code - multiple branches impose the multiple
>>>> points to process. Also many of frequently requested offload
>>>> combinations are not supported yet. That leads to selecting of not
>>>> completely matching tx_burst routine and harms the performance.
>>>> 
>>>> This patch introduces the new approach for tx_burst code. It is
>>>> proposed to develop the unified template for tx_burst routine, which
>>>> supports all the Tx offloads and takes the compile time defined
>>>> parameter describing the supposed set of supported offloads. On the
>>>> base of this template, the compiler is able to generate multiple
>>>> tx_burst routines highly optimized for the statically specified set of Tx
>>> offloads.
>>>> Next, in runtime, at Tx queue configuration the best matching
>>>> optimized implementation of tx_burst is chosen.
>>>> 
>>>> This patch intentionally omits the template internal implementation,
>>>> but just introduces the template itself to emboss the approach of the
>>>> multiple specially tuned tx_burst routines.
>>>> 
>>>> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
>>> 
>>> Getting following build error with icc, can you please check?
>>> 
>>> 
>>> .../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4649): error #191: type qualifier is
>>> meaningless on cast type
>>> 
>>> 
>>>  MLX5_TXOFF_DECL(full_empw,
>>> 
>> Sorry, I have no icc at my disposal now. What I can say (sorry again, it is obvious) 
>> - It definitely does not like the MLX5_TXOFF_DECL macro. Most likely - "restrict" qualifier, I think.
>> 
>> With best regards, Slava
>> 
> 
> Hi Slava,
> 
> I was hoping error log can give enough hint, but I tested by removing the
> 'restrict' keyword and it fixed the icc warning.
> 
> Can you please check if the 'icc' is right and 'restrict' is meaningless, if so
> can you please send a patch to remove it?
> else please let me know and I can make a patch to disable the specific icc warning.

Yes, it looks quite obvious from the error message. mlx5_tx_burst_tmpl() has the qualifier,
so compiler thinks it isn't necessary.

Slava is testing it with gcc/clang. He will get back to you soon.


Thanks,
Yongseok
Viacheslav Ovsiienko July 22, 2019, 7 p.m. | #6
> >> On 7/21/2019 3:24 PM, Viacheslav Ovsiienko wrote:
> >>> Mellanox NICs support the wide set of Tx offloads. The supported
> >>> offloads are reported by the mlx5 PMD in rte_eth_dev_info
> >>> tx_offload_capa field. An application may choose any combination of
> >>> supported offloads and configure the device appropriately. Some of
> >>> Tx offloads may be not requested by application, or ever all of them
> >>> may be
> >> omitted.
> >>> Most of the Tx offloads require some code branches in tx_burst
> >>> routine to support ones. If Tx offload is not requested the tx_burst
> >>> routine code may be significantly simplified and consume less CPU
> cycles.
> >>>
> >>> For example, if application does not engage TSO offload this code
> >>> can be omitted, if multi-segment packet is not supposed the tx_burst
> >>> may assume single mbuf packets only, etc.
> >>>
> >>> Currently, the mlx5 PMD implements multiple tx_burst subroutines for
> >>> most common combinations of requested Tx offloads, each branch has
> >>> its own dedicated implementation. It is not very easy to update,
> >>> support and develop such kind of code - multiple branches impose the
> >>> multiple points to process. Also many of frequently requested
> >>> offload combinations are not supported yet. That leads to selecting
> >>> of not completely matching tx_burst routine and harms the
> performance.
> >>>
> >>> This patch introduces the new approach for tx_burst code. It is
> >>> proposed to develop the unified template for tx_burst routine, which
> >>> supports all the Tx offloads and takes the compile time defined
> >>> parameter describing the supposed set of supported offloads. On the
> >>> base of this template, the compiler is able to generate multiple
> >>> tx_burst routines highly optimized for the statically specified set
> >>> of Tx
> >> offloads.
> >>> Next, in runtime, at Tx queue configuration the best matching
> >>> optimized implementation of tx_burst is chosen.
> >>>
> >>> This patch intentionally omits the template internal implementation,
> >>> but just introduces the template itself to emboss the approach of
> >>> the multiple specially tuned tx_burst routines.
> >>>
> >>> Signed-off-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> >>
> >> Getting following build error with icc, can you please check?
> >>
> >>
> >> .../dpdk/drivers/net/mlx5/mlx5_rxtx.c(4649): error #191: type
> >> qualifier is meaningless on cast type
> >>
> >>
> >>   MLX5_TXOFF_DECL(full_empw,
> >>
> > Sorry, I have no icc at my disposal now. What I can say (sorry again,
> > it is obvious)
> > - It definitely does not like the MLX5_TXOFF_DECL macro. Most likely -
> "restrict" qualifier, I think.
> >
> > With best regards, Slava
> >
> 
> Hi Slava,
> 
> I was hoping error log can give enough hint, but I tested by removing the
> 'restrict' keyword and it fixed the icc warning.
> 
> Can you please check if the 'icc' is right and 'restrict' is meaningless, if so can
> you please send a patch to remove it?
> else please let me know and I can make a patch to disable the specific icc
> warning.

Hi,
it is difficult to say whether qualifier is meaningless, at least I was unable to find the
proof in C99 std - neither in type cast nor in qualifiers chapters. From common sense
 the routine expects the qualified type, we should provide. Anyway, I've checked gcc/clang
compilation with dropped restrict qualifier  and it was OK. Optimization was not harmed as well.

With best regards, Slava

Patch

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 13f9431..16be493 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -1,6 +1,6 @@ 
 /* SPDX-License-Identifier: BSD-3-Clause
  * Copyright 2015 6WIND S.A.
- * Copyright 2015 Mellanox Technologies, Ltd
+ * Copyright 2015-2019 Mellanox Technologies, Ltd
  */
 
 #include <assert.h>
@@ -34,6 +34,57 @@ 
 #include "mlx5_defs.h"
 #include "mlx5_prm.h"
 
+/* TX burst subroutines return codes. */
+enum mlx5_txcmp_code {
+	MLX5_TXCMP_CODE_EXIT = 0,
+	MLX5_TXCMP_CODE_ERROR,
+	MLX5_TXCMP_CODE_SINGLE,
+	MLX5_TXCMP_CODE_MULTI,
+	MLX5_TXCMP_CODE_TSO,
+	MLX5_TXCMP_CODE_EMPW,
+};
+
+/*
+ * These defines are used to configure Tx burst routine option set
+ * supported at compile time. The not specified options are optimized out
+ * out due to if conditions can be explicitly calculated at compile time.
+ * The offloads with bigger runtime check (require more CPU cycles to
+ * skip) overhead should have the bigger index - this is needed to
+ * select the better matching routine function if no exact match and
+ * some offloads are not actually requested.
+ */
+#define MLX5_TXOFF_CONFIG_MULTI (1u << 0) /* Multi-segment packets.*/
+#define MLX5_TXOFF_CONFIG_TSO (1u << 1) /* TCP send offload supported.*/
+#define MLX5_TXOFF_CONFIG_SWP (1u << 2) /* Tunnels/SW Parser offloads.*/
+#define MLX5_TXOFF_CONFIG_CSUM (1u << 3) /* Check Sums offloaded. */
+#define MLX5_TXOFF_CONFIG_INLINE (1u << 4) /* Data inlining supported. */
+#define MLX5_TXOFF_CONFIG_VLAN (1u << 5) /* VLAN insertion supported.*/
+#define MLX5_TXOFF_CONFIG_METADATA (1u << 6) /* Flow metadata. */
+#define MLX5_TXOFF_CONFIG_EMPW (1u << 8) /* Enhanced MPW supported.*/
+
+/* The most common offloads groups. */
+#define MLX5_TXOFF_CONFIG_NONE 0
+#define MLX5_TXOFF_CONFIG_FULL (MLX5_TXOFF_CONFIG_MULTI | \
+				MLX5_TXOFF_CONFIG_TSO | \
+				MLX5_TXOFF_CONFIG_SWP | \
+				MLX5_TXOFF_CONFIG_CSUM | \
+				MLX5_TXOFF_CONFIG_INLINE | \
+				MLX5_TXOFF_CONFIG_VLAN | \
+				MLX5_TXOFF_CONFIG_METADATA)
+
+#define MLX5_TXOFF_CONFIG(mask) (olx & MLX5_TXOFF_CONFIG_##mask)
+
+#define MLX5_TXOFF_DECL(func, olx) \
+static uint16_t mlx5_tx_burst_##func(void *txq, \
+				     struct rte_mbuf **pkts, \
+				    uint16_t pkts_n) \
+{ \
+	return mlx5_tx_burst_tmpl((struct mlx5_txq_data *restrict)txq, \
+		    pkts, pkts_n, (olx)); \
+}
+
+#define MLX5_TXOFF_INFO(func, olx) {mlx5_tx_burst_##func, olx},
+
 static __rte_always_inline uint32_t
 rxq_cq_to_pkt_type(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cqe);
 
@@ -1531,7 +1582,323 @@ 
 }
 
 /**
- * Configure the TX function to use.
+ * DPDK Tx callback template. This is configured template
+ * used to generate routines optimized for specified offload setup.
+ * One of this generated functions is chosen at SQ configuration
+ * time.
+ *
+ * @param txq
+ *   Generic pointer to TX queue structure.
+ * @param[in] pkts
+ *   Packets to transmit.
+ * @param pkts_n
+ *   Number of packets in array.
+ * @param olx
+ *   Configured offloads mask, presents the bits of MLX5_TXOFF_CONFIG_xxx
+ *   values. Should be static to take compile time static configuration
+ *   advantages.
+ *
+ * @return
+ *   Number of packets successfully transmitted (<= pkts_n).
+ */
+static __rte_always_inline uint16_t
+mlx5_tx_burst_tmpl(struct mlx5_txq_data *restrict txq,
+		   struct rte_mbuf **restrict pkts,
+		   uint16_t pkts_n,
+		   unsigned int olx)
+{
+	(void)txq;
+	(void)pkts;
+	(void)pkts_n;
+	(void)olx;
+	return 0;
+}
+
+/* Generate routines with Enhanced Multi-Packet Write support. */
+MLX5_TXOFF_DECL(full_empw,
+		MLX5_TXOFF_CONFIG_FULL | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(none_empw,
+		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(md_empw,
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(mt_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(mtsc_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(mti_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(mtv_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(mtiv_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(sc_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(sci_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(scv_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(sciv_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(i_empw,
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(v_empw,
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_DECL(iv_empw,
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+/* Generate routines without Enhanced Multi-Packet Write support. */
+MLX5_TXOFF_DECL(full,
+		MLX5_TXOFF_CONFIG_FULL)
+
+MLX5_TXOFF_DECL(none,
+		MLX5_TXOFF_CONFIG_NONE)
+
+MLX5_TXOFF_DECL(md,
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(mt,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(mtsc,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(mti,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+
+MLX5_TXOFF_DECL(mtv,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+
+MLX5_TXOFF_DECL(mtiv,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(sc,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(sci,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+
+MLX5_TXOFF_DECL(scv,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+
+MLX5_TXOFF_DECL(sciv,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(i,
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(v,
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_DECL(iv,
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+/*
+ * Array of declared and compiled Tx burst function and corresponding
+ * supported offloads set. The array is used to select the Tx burst
+ * function for specified offloads set at Tx queue configuration time.
+ */
+const struct {
+	eth_tx_burst_t func;
+	unsigned int olx;
+} txoff_func[] = {
+MLX5_TXOFF_INFO(full_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(none_empw,
+		MLX5_TXOFF_CONFIG_NONE | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(md_empw,
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(mt_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(mtsc_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(mti_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(mtv_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(mtiv_empw,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(sc_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(sci_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(scv_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(sciv_empw,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(i_empw,
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(v_empw,
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(iv_empw,
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA | MLX5_TXOFF_CONFIG_EMPW)
+
+MLX5_TXOFF_INFO(full,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(none,
+		MLX5_TXOFF_CONFIG_NONE)
+
+MLX5_TXOFF_INFO(md,
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(mt,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(mtsc,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(mti,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+
+MLX5_TXOFF_INFO(mtv,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(mtiv,
+		MLX5_TXOFF_CONFIG_MULTI | MLX5_TXOFF_CONFIG_TSO |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(sc,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(sci,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(scv,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(sciv,
+		MLX5_TXOFF_CONFIG_SWP |	MLX5_TXOFF_CONFIG_CSUM |
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(i,
+		MLX5_TXOFF_CONFIG_INLINE |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(v,
+		MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+
+MLX5_TXOFF_INFO(iv,
+		MLX5_TXOFF_CONFIG_INLINE | MLX5_TXOFF_CONFIG_VLAN |
+		MLX5_TXOFF_CONFIG_METADATA)
+};
+
+/**
+ * Configure the Tx function to use. The routine checks configured
+ * Tx offloads for the device and selects appropriate Tx burst
+ * routine. There are multiple Tx burst routines compiled from
+ * the same template in the most optimal way for the dedicated
+ * Tx offloads set.
  *
  * @param dev
  *   Pointer to private data structure.
@@ -1542,8 +1909,153 @@ 
 eth_tx_burst_t
 mlx5_select_tx_function(struct rte_eth_dev *dev)
 {
-	(void)dev;
-	return removed_tx_burst;
+	struct mlx5_priv *priv = dev->data->dev_private;
+	struct mlx5_dev_config *config = &priv->config;
+	uint64_t tx_offloads = dev->data->dev_conf.txmode.offloads;
+	unsigned int diff = 0, olx = 0, i, m;
+
+	static_assert(MLX5_WQE_SIZE_MAX / MLX5_WSEG_SIZE <=
+		      MLX5_DSEG_MAX, "invalid WQE max size");
+	static_assert(MLX5_WQE_CSEG_SIZE == MLX5_WSEG_SIZE,
+		      "invalid WQE Control Segment size");
+	static_assert(MLX5_WQE_ESEG_SIZE == MLX5_WSEG_SIZE,
+		      "invalid WQE Ethernet Segment size");
+	static_assert(MLX5_WQE_DSEG_SIZE == MLX5_WSEG_SIZE,
+		      "invalid WQE Data Segment size");
+	static_assert(MLX5_WQE_SIZE == 4 * MLX5_WSEG_SIZE,
+		      "invalid WQE size");
+	assert(priv);
+	if (tx_offloads & DEV_TX_OFFLOAD_MULTI_SEGS) {
+		/* We should support Multi-Segment Packets. */
+		olx |= MLX5_TXOFF_CONFIG_MULTI;
+	}
+	if (tx_offloads & (DEV_TX_OFFLOAD_TCP_TSO |
+			   DEV_TX_OFFLOAD_VXLAN_TNL_TSO |
+			   DEV_TX_OFFLOAD_GRE_TNL_TSO |
+			   DEV_TX_OFFLOAD_IP_TNL_TSO |
+			   DEV_TX_OFFLOAD_UDP_TNL_TSO)) {
+		/* We should support TCP Send Offload. */
+		olx |= MLX5_TXOFF_CONFIG_TSO;
+	}
+	if (tx_offloads & (DEV_TX_OFFLOAD_IP_TNL_TSO |
+			   DEV_TX_OFFLOAD_UDP_TNL_TSO |
+			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
+		/* We should support Software Parser for Tunnels. */
+		olx |= MLX5_TXOFF_CONFIG_SWP;
+	}
+	if (tx_offloads & (DEV_TX_OFFLOAD_IPV4_CKSUM |
+			   DEV_TX_OFFLOAD_UDP_CKSUM |
+			   DEV_TX_OFFLOAD_TCP_CKSUM |
+			   DEV_TX_OFFLOAD_OUTER_IPV4_CKSUM)) {
+		/* We should support IP/TCP/UDP Checksums. */
+		olx |= MLX5_TXOFF_CONFIG_CSUM;
+	}
+	if (tx_offloads & DEV_TX_OFFLOAD_VLAN_INSERT) {
+		/* We should support VLAN insertion. */
+		olx |= MLX5_TXOFF_CONFIG_VLAN;
+	}
+	if (priv->txqs_n && (*priv->txqs)[0]) {
+		struct mlx5_txq_data *txd = (*priv->txqs)[0];
+
+		if (txd->inlen_send) {
+			/*
+			 * Check the data inline requirements. Data inline
+			 * is enabled on per device basis, we can check
+			 * the first Tx queue only.
+			 *
+			 * If device does not support VLAN insertion in WQE
+			 * and some queues are requested to perform VLAN
+			 * insertion offload than inline must be enabled.
+			 */
+			olx |= MLX5_TXOFF_CONFIG_INLINE;
+		}
+	}
+	if (config->mps == MLX5_MPW_ENHANCED &&
+	    config->txq_inline_min <= 0) {
+		/*
+		 * The NIC supports Enhanced Multi-Packet Write.
+		 * We do not support legacy MPW due to its
+		 * hardware related problems, so we just ignore
+		 * legacy MLX5_MPW settings. There should be no
+		 * minimal required inline data.
+		 */
+		olx |= MLX5_TXOFF_CONFIG_EMPW;
+	}
+	if (tx_offloads & DEV_TX_OFFLOAD_MATCH_METADATA) {
+		/* We should support Flow metadata. */
+		olx |= MLX5_TXOFF_CONFIG_METADATA;
+	}
+	/*
+	 * Scan the routines table to find the minimal
+	 * satisfying routine with requested offloads.
+	 */
+	m = RTE_DIM(txoff_func);
+	for (i = 0; i < RTE_DIM(txoff_func); i++) {
+		unsigned int tmp;
+
+		tmp = txoff_func[i].olx;
+		if (tmp == olx) {
+			/* Meets requested offloads exactly.*/
+			m = i;
+			break;
+		}
+		if ((tmp & olx) != olx) {
+			/* Does not meet requested offloads at all. */
+			continue;
+		}
+		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_EMPW)
+			/* Do not enable eMPW if not configured. */
+			continue;
+		if ((olx ^ tmp) & MLX5_TXOFF_CONFIG_INLINE)
+			/* Do not enable inlining if not configured. */
+			continue;
+		/*
+		 * Some routine meets the requirements.
+		 * Check whether it has minimal amount
+		 * of not requested offloads.
+		 */
+		tmp = __builtin_popcountl(tmp & ~olx);
+		if (m >= RTE_DIM(txoff_func) || tmp < diff) {
+			/* First or better match, save and continue. */
+			m = i;
+			diff = tmp;
+			continue;
+		}
+		if (tmp == diff) {
+			tmp = txoff_func[i].olx ^ txoff_func[m].olx;
+			if (__builtin_ffsl(txoff_func[i].olx & ~tmp) <
+			    __builtin_ffsl(txoff_func[m].olx & ~tmp)) {
+				/* Lighter not requested offload. */
+				m = i;
+			}
+		}
+	}
+	if (m >= RTE_DIM(txoff_func)) {
+		DRV_LOG(DEBUG, "port %u has no selected Tx function"
+			       " for requested offloads %04X",
+				dev->data->port_id, olx);
+		return NULL;
+	}
+	DRV_LOG(DEBUG, "port %u has selected Tx function"
+		       " supporting offloads %04X/%04X",
+			dev->data->port_id, olx, txoff_func[m].olx);
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_MULTI)
+		DRV_LOG(DEBUG, "\tMULTI (multi segment)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_TSO)
+		DRV_LOG(DEBUG, "\tTSO   (TCP send offload)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_SWP)
+		DRV_LOG(DEBUG, "\tSWP   (software parser)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_CSUM)
+		DRV_LOG(DEBUG, "\tCSUM  (checksum offload)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_INLINE)
+		DRV_LOG(DEBUG, "\tINLIN (inline data)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_VLAN)
+		DRV_LOG(DEBUG, "\tVLANI (VLAN insertion)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_METADATA)
+		DRV_LOG(DEBUG, "\tMETAD (tx Flow metadata)");
+	if (txoff_func[m].olx & MLX5_TXOFF_CONFIG_EMPW)
+		DRV_LOG(DEBUG, "\tEMPW  (Enhanced MPW)");
+	return txoff_func[m].func;
 }