DPDK patches and discussions
 help / color / mirror / Atom feed
From: Tom Rix <trix@redhat.com>
To: Nicolas Chautru <nicolas.chautru@intel.com>,
	dev@dpdk.org, akhil.goyal@nxp.com
Cc: david.marchand@redhat.com
Subject: Re: [dpdk-dev] [PATCH v5 3/7] app/bbdev: include explicit HARQ preloading
Date: Mon, 26 Oct 2020 06:31:56 -0700
Message-ID: <6863b163-cab4-7c60-750b-d96c7d59172c@redhat.com> (raw)
In-Reply-To: <1603496581-35966-4-git-send-email-nicolas.chautru@intel.com>


On 10/23/20 4:42 PM, Nicolas Chautru wrote:
> Run preloading explicitly for unit tests. Load each code block
> by reusing existing input op then restore for the actual test.
>
> Signed-off-by: Nicolas Chautru <nicolas.chautru@intel.com>
> Acked-by: Liu Tianjiao <tianjiao.liu@intel.com>
> ---
>  app/test-bbdev/main.h            |  1 +
>  app/test-bbdev/test_bbdev_perf.c | 51 +++++++++++++++++++++-------------------
>  2 files changed, 28 insertions(+), 24 deletions(-)
>
> diff --git a/app/test-bbdev/main.h b/app/test-bbdev/main.h
> index fb3dec8..dc10a50 100644
> --- a/app/test-bbdev/main.h
> +++ b/app/test-bbdev/main.h
> @@ -17,6 +17,7 @@
>  #define TEST_SKIPPED    1
>  
>  #define MAX_BURST 512U
> +#define MAX_OPS 1024U

This #define is not consistently used.

ex/ see retrieve_harq_ddr, the old 1024 is still being used.

>  #define DEFAULT_BURST 32U
>  #define DEFAULT_OPS 64U
>  #define DEFAULT_ITER 6U
> diff --git a/app/test-bbdev/test_bbdev_perf.c b/app/test-bbdev/test_bbdev_perf.c
> index b62848e..f30cbdb 100644
> --- a/app/test-bbdev/test_bbdev_perf.c
> +++ b/app/test-bbdev/test_bbdev_perf.c
> @@ -2513,20 +2513,20 @@ typedef int (test_case_function)(struct active_device *ad,
>  		bool preload)
>  {
>  	uint16_t j;
> -	int ret;
> -	uint32_t harq_offset = (uint32_t) queue_id * HARQ_INCR * 1024;
> -	struct rte_bbdev_op_data save_hc_in, save_hc_out;
> -	struct rte_bbdev_dec_op *ops_deq[MAX_BURST];
> +	int deq;
> +	uint32_t harq_offset = (uint32_t) queue_id * HARQ_INCR * MAX_OPS;
> +	struct rte_bbdev_op_data save_hc_in[MAX_OPS], save_hc_out[MAX_OPS];
> +	struct rte_bbdev_dec_op *ops_deq[MAX_OPS];
>  	uint32_t flags = ops[0]->ldpc_dec.op_flags;
>  	bool mem_in = flags & RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_IN_ENABLE;
>  	bool hc_in = flags & RTE_BBDEV_LDPC_HQ_COMBINE_IN_ENABLE;
>  	bool mem_out = flags & RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_OUT_ENABLE;
>  	bool hc_out = flags & RTE_BBDEV_LDPC_HQ_COMBINE_OUT_ENABLE;
>  	bool h_comp = flags & RTE_BBDEV_LDPC_HARQ_6BIT_COMPRESSION;
> -	for (j = 0; j < n; ++j) {
> -		if ((mem_in || hc_in) && preload) {
> -			save_hc_in = ops[j]->ldpc_dec.harq_combined_input;
> -			save_hc_out = ops[j]->ldpc_dec.harq_combined_output;
> +	if ((mem_in || hc_in) && preload) {
> +		for (j = 0; j < n; ++j) {
> +			save_hc_in[j] = ops[j]->ldpc_dec.harq_combined_input;
> +			save_hc_out[j] = ops[j]->ldpc_dec.harq_combined_output;
>  			ops[j]->ldpc_dec.op_flags =
>  				RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_LOOPBACK +
>  				RTE_BBDEV_LDPC_INTERNAL_HARQ_MEMORY_OUT_ENABLE;

flags are usually handled with bit operators, not arithmetic.

this seems to be a general issue.

> @@ -2536,16 +2536,23 @@ typedef int (test_case_function)(struct active_device *ad,
>  			ops[j]->ldpc_dec.harq_combined_output.offset =
>  					harq_offset;
>  			ops[j]->ldpc_dec.harq_combined_input.offset = 0;
> -			rte_bbdev_enqueue_ldpc_dec_ops(dev_id, queue_id,
> -					&ops[j], 1);
> -			ret = 0;
> -			while (ret == 0)
> -				ret = rte_bbdev_dequeue_ldpc_dec_ops(
> -					dev_id, queue_id, &ops_deq[j], 1);
> +			harq_offset += HARQ_INCR;
> +		}
> +		rte_bbdev_enqueue_ldpc_dec_ops(dev_id, queue_id, &ops[0], n);
Add check the return is 'n'
> +		deq = 0;
> +		while (deq != n)
> +			deq += rte_bbdev_dequeue_ldpc_dec_ops(
> +					dev_id, queue_id, &ops_deq[deq],
> +					n - deq);

Add check the return >= 0

Tom

> +		/* Restore the operations */
> +		for (j = 0; j < n; ++j) {
>  			ops[j]->ldpc_dec.op_flags = flags;
> -			ops[j]->ldpc_dec.harq_combined_input = save_hc_in;
> -			ops[j]->ldpc_dec.harq_combined_output = save_hc_out;
> +			ops[j]->ldpc_dec.harq_combined_input = save_hc_in[j];
> +			ops[j]->ldpc_dec.harq_combined_output = save_hc_out[j];
>  		}
> +	}
> +	harq_offset = (uint32_t) queue_id * HARQ_INCR * MAX_OPS;
> +	for (j = 0; j < n; ++j) {
>  		/* Adjust HARQ offset when we reach external DDR */
>  		if (mem_in || hc_in)
>  			ops[j]->ldpc_dec.harq_combined_input.offset
> @@ -3231,11 +3238,9 @@ typedef int (test_case_function)(struct active_device *ad,
>  				mbuf_reset(
>  				ops_enq[j]->ldpc_dec.harq_combined_output.data);
>  		}
> -		if (extDdr) {
> -			bool preload = i == (TEST_REPETITIONS - 1);
> +		if (extDdr)
>  			preload_harq_ddr(tp->dev_id, queue_id, ops_enq,
> -					num_ops, preload);
> -		}
> +					num_ops, true);
>  		start_time = rte_rdtsc_precise();
>  
>  		for (enq = 0, deq = 0; enq < num_ops;) {
> @@ -3362,11 +3367,9 @@ typedef int (test_case_function)(struct active_device *ad,
>  				mbuf_reset(
>  				ops_enq[j]->ldpc_dec.harq_combined_output.data);
>  		}
> -		if (extDdr) {
> -			bool preload = i == (TEST_REPETITIONS - 1);
> +		if (extDdr)
>  			preload_harq_ddr(tp->dev_id, queue_id, ops_enq,
> -					num_ops, preload);
> -		}
> +					num_ops, true);
>  		start_time = rte_rdtsc_precise();
>  
>  		for (enq = 0, deq = 0; enq < num_ops;) {


  reply	other threads:[~2020-10-26 13:32 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-23 23:42 [dpdk-dev] [PATCH v5 0/7] BBDEV test updates Nicolas Chautru
2020-10-23 23:42 ` [dpdk-dev] [PATCH v5 1/7] app/bbdev: add explicit ut for latency vs validation Nicolas Chautru
2020-10-26 12:55   ` Tom Rix
2020-10-26 17:30     ` Chautru, Nicolas
2020-10-28 20:37       ` Tom Rix
2020-10-23 23:42 ` [dpdk-dev] [PATCH v5 2/7] app/bbdev: add explicit check for counters Nicolas Chautru
2020-10-26 13:05   ` Tom Rix
2020-10-26 16:29     ` Chautru, Nicolas
2020-10-28 20:31       ` Tom Rix
2020-10-23 23:42 ` [dpdk-dev] [PATCH v5 3/7] app/bbdev: include explicit HARQ preloading Nicolas Chautru
2020-10-26 13:31   ` Tom Rix [this message]
2020-10-26 16:50     ` Chautru, Nicolas
2020-10-28 20:33       ` Tom Rix
2020-10-23 23:42 ` [dpdk-dev] [PATCH v5 4/7] app/bbdev: define wait for offload Nicolas Chautru
2020-10-26 13:33   ` Tom Rix
2020-10-26 16:04     ` Chautru, Nicolas
2020-10-28 20:24       ` Tom Rix
2020-10-23 23:42 ` [dpdk-dev] [PATCH v5 5/7] app/bbdev: skip bler ut when compression is used Nicolas Chautru
2020-10-26 13:35   ` Tom Rix
2020-10-23 23:43 ` [dpdk-dev] [PATCH v5 6/7] app/bbdev: reduce duration of throughput test Nicolas Chautru
2020-10-26 13:39   ` Tom Rix
2020-10-23 23:43 ` [dpdk-dev] [PATCH v5 7/7] app/bbdev: update offload test to dequeue full ring Nicolas Chautru
2020-10-26 13:55   ` Tom Rix
2020-10-26 16:27     ` Chautru, Nicolas
2020-10-28 20:28       ` Tom Rix
2020-10-24  7:47 ` [dpdk-dev] [PATCH v5 0/7] BBDEV test updates David Marchand

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=6863b163-cab4-7c60-750b-d96c7d59172c@redhat.com \
    --to=trix@redhat.com \
    --cc=akhil.goyal@nxp.com \
    --cc=david.marchand@redhat.com \
    --cc=dev@dpdk.org \
    --cc=nicolas.chautru@intel.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

DPDK patches and discussions

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://inbox.dpdk.org/dev/0 dev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 dev dev/ https://inbox.dpdk.org/dev \
		dev@dpdk.org
	public-inbox-index dev

Example config snippet for mirrors.
Newsgroup available over NNTP:
	nntp://inbox.dpdk.org/inbox.dpdk.dev


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git