RE: [RFC PATCH 06/14] net/ice: use the new common vector capability function

["Loftus, Ciara" <ciara.loftus@intel.com>]

> 
> On Fri, Jul 25, 2025 at 12:49:11PM +0000, Ciara Loftus wrote:
> > Use the new function for determining the maximum simd bitwidth in
> > the ice driver.
> >
> > Signed-off-by: Ciara Loftus <ciara.loftus@intel.com>
> 
> Few comments inline below.
> > ---
> >  drivers/net/intel/ice/ice_ethdev.h       |  5 +--
> >  drivers/net/intel/ice/ice_rxtx.c         | 52 ++++++------------------
> >  drivers/net/intel/ice/ice_rxtx.h         |  1 +
> >  drivers/net/intel/ice/ice_rxtx_vec_sse.c |  6 +++
> >  4 files changed, 21 insertions(+), 43 deletions(-)
> >
> > diff --git a/drivers/net/intel/ice/ice_ethdev.h
> b/drivers/net/intel/ice/ice_ethdev.h
> > index 5fda814f06..992fcc9175 100644
> > --- a/drivers/net/intel/ice/ice_ethdev.h
> > +++ b/drivers/net/intel/ice/ice_ethdev.h
> > @@ -11,6 +11,7 @@
> >
> >  #include <ethdev_driver.h>
> >  #include <rte_tm_driver.h>
> > +#include <rte_vect.h>
> >
> >  #include "base/ice_common.h"
> >  #include "base/ice_adminq_cmd.h"
> > @@ -674,9 +675,7 @@ struct ice_adapter {
> >  	/* Set bit if the engine is disabled */
> >  	unsigned long disabled_engine_mask;
> >  	struct ice_parser *psr;
> > -	/* used only on X86, zero on other Archs */
> > -	bool tx_use_avx2;
> > -	bool tx_use_avx512;
> > +	enum rte_vect_max_simd tx_simd_width;
> >  	bool rx_vec_offload_support;
> >  };
> >
> > diff --git a/drivers/net/intel/ice/ice_rxtx.c b/drivers/net/intel/ice/ice_rxtx.c
> > index 85832d95a3..79217249b9 100644
> > --- a/drivers/net/intel/ice/ice_rxtx.c
> > +++ b/drivers/net/intel/ice/ice_rxtx.c
> > @@ -3703,7 +3703,7 @@ ice_set_rx_function(struct rte_eth_dev *dev)
> >  	struct ci_rx_queue *rxq;
> >  	int i;
> >  	int rx_check_ret = -1;
> > -	bool rx_use_avx512 = false, rx_use_avx2 = false;
> > +	enum rte_vect_max_simd rx_simd_width =
> RTE_VECT_SIMD_DISABLED;
> >
> >  	rx_check_ret = ice_rx_vec_dev_check(dev);
> >  	if (ad->ptp_ena)
> > @@ -3720,35 +3720,22 @@ ice_set_rx_function(struct rte_eth_dev *dev)
> >  				break;
> >  			}
> >  		}
> > +		rx_simd_width = ice_get_max_simd_bitwidth();
> >
> 
> Since this whole block is in #ifdef X86_64, do we need a generic ice
> function here? Is it worth just just calling the x86 function directly?

We'd then need to include the rx_vec_x86.h file in the common rxtx.c code which I think is probably not desired.
We could move the function to rx.h but then that file would have some arch specific stuff which again is probably not ideal either.

> 
> > -		if (rte_vect_get_max_simd_bitwidth() >=
> RTE_VECT_SIMD_512 &&
> > -
> 	rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) == 1 &&
> > -
> 	rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512BW) == 1)
> > -#ifdef CC_AVX512_SUPPORT
> > -			rx_use_avx512 = true;
> > -#else
> > -		PMD_DRV_LOG(NOTICE,
> > -			"AVX512 is not supported in build env");
> > -#endif
> > -		if (!rx_use_avx512 &&
> > -
> 	(rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1 ||
> > -
> 	rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) == 1) &&
> > -				rte_vect_get_max_simd_bitwidth() >=
> RTE_VECT_SIMD_256)
> > -			rx_use_avx2 = true;
> >  	} else {
> >  		ad->rx_vec_allowed = false;
> >  	}
> >
> >  	if (ad->rx_vec_allowed) {
> >  		if (dev->data->scattered_rx) {
> > -			if (rx_use_avx512) {
> > +			if (rx_simd_width == RTE_VECT_SIMD_512) {
> >  #ifdef CC_AVX512_SUPPORT
> >  				if (ad->rx_vec_offload_support)
> >  					ad->rx_func_type =
> ICE_RX_AVX512_SCATTERED_OFFLOAD;
> >  				else
> >  					ad->rx_func_type =
> ICE_RX_AVX512_SCATTERED;
> >  #endif
> > -			} else if (rx_use_avx2) {
> > +			} else if (rx_simd_width == RTE_VECT_SIMD_256) {
> >  				if (ad->rx_vec_offload_support)
> >  					ad->rx_func_type =
> ICE_RX_AVX2_SCATTERED_OFFLOAD;
> >  				else
> > @@ -3757,14 +3744,14 @@ ice_set_rx_function(struct rte_eth_dev *dev)
> >  				ad->rx_func_type = ICE_RX_SSE_SCATTERED;
> >  			}
> >  		} else {
> > -			if (rx_use_avx512) {
> > +			if (rx_simd_width == RTE_VECT_SIMD_512) {
> >  #ifdef CC_AVX512_SUPPORT
> >  				if (ad->rx_vec_offload_support)
> >  					ad->rx_func_type =
> ICE_RX_AVX512_OFFLOAD;
> >  				else
> >  					ad->rx_func_type = ICE_RX_AVX512;
> >  #endif
> > -			} else if (rx_use_avx2) {
> > +			} else if (rx_simd_width == RTE_VECT_SIMD_256) {
> >  				if (ad->rx_vec_offload_support)
> >  					ad->rx_func_type =
> ICE_RX_AVX2_OFFLOAD;
> >  				else
> > @@ -4032,29 +4019,14 @@ ice_set_tx_function(struct rte_eth_dev *dev)
> >  	int tx_check_ret = -1;
> >
> >  	if (rte_eal_process_type() == RTE_PROC_PRIMARY) {
> > -		ad->tx_use_avx2 = false;
> > -		ad->tx_use_avx512 = false;
> > +		ad->tx_simd_width = RTE_VECT_SIMD_DISABLED;
> >  		tx_check_ret = ice_tx_vec_dev_check(dev);
> > +		ad->tx_simd_width = ice_get_max_simd_bitwidth();
> >  		if (tx_check_ret >= 0 &&
> >  		    rte_vect_get_max_simd_bitwidth() >=
> RTE_VECT_SIMD_128) {
> >  			ad->tx_vec_allowed = true;
> >
> > -			if (rte_vect_get_max_simd_bitwidth() >=
> RTE_VECT_SIMD_512 &&
> > -			rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F)
> == 1 &&
> > -
> 	rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512BW) == 1)
> > -#ifdef CC_AVX512_SUPPORT
> > -				ad->tx_use_avx512 = true;
> > -#else
> > -			PMD_DRV_LOG(NOTICE,
> > -				"AVX512 is not supported in build env");
> > -#endif
> > -			if (!ad->tx_use_avx512 &&
> > -
> 	(rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2) == 1 ||
> > -
> 	rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F) == 1) &&
> > -				rte_vect_get_max_simd_bitwidth() >=
> RTE_VECT_SIMD_256)
> > -				ad->tx_use_avx2 = true;
> > -
> > -			if (!ad->tx_use_avx2 && !ad->tx_use_avx512 &&
> > +			if (ad->tx_simd_width < RTE_VECT_SIMD_256 &&
> >  				tx_check_ret ==
> ICE_VECTOR_OFFLOAD_PATH)
> >  				ad->tx_vec_allowed = false;
> >
> > @@ -4074,7 +4046,7 @@ ice_set_tx_function(struct rte_eth_dev *dev)
> >
> >  	if (ad->tx_vec_allowed) {
> >  		dev->tx_pkt_prepare = NULL;
> > -		if (ad->tx_use_avx512) {
> > +		if (ad->tx_simd_width == RTE_VECT_SIMD_512) {
> >  #ifdef CC_AVX512_SUPPORT
> >  			if (tx_check_ret == ICE_VECTOR_OFFLOAD_PATH) {
> >  				PMD_DRV_LOG(NOTICE,
> > @@ -4100,9 +4072,9 @@ ice_set_tx_function(struct rte_eth_dev *dev)
> >  				dev->tx_pkt_prepare = ice_prep_pkts;
> >  			} else {
> >  				PMD_DRV_LOG(DEBUG, "Using %sVector Tx
> (port %d).",
> > -					    ad->tx_use_avx2 ? "avx2 " : "",
> > +					    ad->tx_simd_width ==
> RTE_VECT_SIMD_256 ? "avx2 " : "",
> >  					    dev->data->port_id);
> > -				dev->tx_pkt_burst = ad->tx_use_avx2 ?
> > +				dev->tx_pkt_burst = ad->tx_simd_width ==
> RTE_VECT_SIMD_256 ?
> >  						    ice_xmit_pkts_vec_avx2 :
> >  						    ice_xmit_pkts_vec;
> >  			}
> > diff --git a/drivers/net/intel/ice/ice_rxtx.h b/drivers/net/intel/ice/ice_rxtx.h
> > index 0301d05888..8c3d6c413a 100644
> > --- a/drivers/net/intel/ice/ice_rxtx.h
> > +++ b/drivers/net/intel/ice/ice_rxtx.h
> > @@ -261,6 +261,7 @@ uint16_t ice_xmit_pkts_vec_avx512_offload(void
> *tx_queue,
> >  int ice_fdir_programming(struct ice_pf *pf, struct ice_fltr_desc *fdir_desc);
> >  int ice_tx_done_cleanup(void *txq, uint32_t free_cnt);
> >  int ice_get_monitor_addr(void *rx_queue, struct rte_power_monitor_cond
> *pmc);
> > +enum rte_vect_max_simd ice_get_max_simd_bitwidth(void);
> >
> >  #define FDIR_PARSING_ENABLE_PER_QUEUE(ad, on) do { \
> >  	int i; \
> > diff --git a/drivers/net/intel/ice/ice_rxtx_vec_sse.c
> b/drivers/net/intel/ice/ice_rxtx_vec_sse.c
> > index d818b3b728..1545bc3b6e 100644
> > --- a/drivers/net/intel/ice/ice_rxtx_vec_sse.c
> > +++ b/drivers/net/intel/ice/ice_rxtx_vec_sse.c
> > @@ -735,3 +735,9 @@ ice_tx_vec_dev_check(struct rte_eth_dev *dev)
> >  {
> >  	return ice_tx_vec_dev_check_default(dev);
> >  }
> > +
> > +enum rte_vect_max_simd
> > +ice_get_max_simd_bitwidth(void)
> > +{
> > +	return ci_get_x86_max_simd_bitwidth();
> > +}
> 
> If we do wrap the x86 bitwidth function in an ice-specific one, we probably
> need to provide one for other architectures. However, as I comment above, I
> don't think we need to wrap this - though perhaps I'm missing something or
> its needed in later patches...
> 
> > --
> > 2.34.1
> >