Re: [dpdk-dev] [PATCH v5 1/7] app/bbdev: add explicit ut for latency vs validation

[Tom Rix <trix@redhat.com>]

On 10/26/20 10:30 AM, Chautru, Nicolas wrote:
>> -----Original Message-----
>> From: Tom Rix <trix@redhat.com>
>> Sent: Monday, October 26, 2020 5:56 AM
>> To: Chautru, Nicolas <nicolas.chautru@intel.com>; dev@dpdk.org;
>> akhil.goyal@nxp.com
>> Cc: david.marchand@redhat.com
>> Subject: Re: [PATCH v5 1/7] app/bbdev: add explicit ut for latency vs
>> validation
>>
>>
>> On 10/23/20 4:42 PM, Nicolas Chautru wrote:
>>> Adding explicit different ut when testing for validation or latency
>>> (early termination enabled or not).
>>>
>>> Signed-off-by: Nicolas Chautru <nicolas.chautru@intel.com>
>>> Acked-by: Aidan Goddard <aidan.goddard@accelercomm.com>
>>> Acked-by: Dave Burley <dave.burley@accelercomm.com>
>>> ---
>>>  app/test-bbdev/test_bbdev_perf.c | 92
>>> ++++++++++++++++++++++++++++++++++++++--
>> Should update the copyright.
>>>  1 file changed, 88 insertions(+), 4 deletions(-)
>>>
>>> diff --git a/app/test-bbdev/test_bbdev_perf.c
>>> b/app/test-bbdev/test_bbdev_perf.c
>>> index 6e5535d..3554a77 100644
>>> --- a/app/test-bbdev/test_bbdev_perf.c
>>> +++ b/app/test-bbdev/test_bbdev_perf.c
>>> @@ -3999,12 +3999,14 @@ typedef int (test_case_function)(struct
>> active_device *ad,
>>>  	return i;
>>>  }
>>>
>>> +/* Test case for latency/validation for LDPC Decoder */
>>>  static int
>>>  latency_test_ldpc_dec(struct rte_mempool *mempool,
>>>  		struct test_buffers *bufs, struct rte_bbdev_dec_op *ref_op,
>>>  		int vector_mask, uint16_t dev_id, uint16_t queue_id,
>>>  		const uint16_t num_to_process, uint16_t burst_sz,
>>> -		uint64_t *total_time, uint64_t *min_time, uint64_t
>> *max_time)
>>> +		uint64_t *total_time, uint64_t *min_time, uint64_t
>> *max_time,
>>> +		bool disable_et)
>>>  {
>>>  	int ret = TEST_SUCCESS;
>>>  	uint16_t i, j, dequeued;
>>> @@ -4026,7 +4028,7 @@ typedef int (test_case_function)(struct
>> active_device *ad,
>>>  				"rte_bbdev_dec_op_alloc_bulk() failed");
>>>
>>>  		/* For latency tests we need to disable early termination */
>>> -		if (check_bit(ref_op->ldpc_dec.op_flags,
>>> +		if (disable_et && check_bit(ref_op->ldpc_dec.op_flags,
>>>
>> 	RTE_BBDEV_LDPC_ITERATION_STOP_ENABLE))
>>>  			ref_op->ldpc_dec.op_flags -=
>>>
>> 	RTE_BBDEV_LDPC_ITERATION_STOP_ENABLE;
>> Bit clearing is usually done with &= ~()
> This is the coding style for rest of the file hence sticking to it. 
>
>>> @@ -4248,7 +4250,7 @@ typedef int (test_case_function)(struct
>> active_device *ad,
>>>  	TEST_ASSERT_NOT_NULL(op_type_str, "Invalid op type: %u",
>> op_type);
>>>  	printf("+ ------------------------------------------------------- +\n");
>>> -	printf("== test: validation/latency\ndev: %s, burst size: %u, num ops:
>> %u, op type: %s\n",
>>> +	printf("== test: latency\ndev: %s, burst size: %u, num ops: %u, op
>>> +type: %s\n",
>>>  			info.dev_name, burst_sz, num_to_process,
>> op_type_str);
>>>  	if (op_type == RTE_BBDEV_OP_TURBO_DEC) @@ -4270,7 +4272,83
>> @@
>>> typedef int (test_case_function)(struct active_device *ad,
>>>  		iter = latency_test_ldpc_dec(op_params->mp, bufs,
>>>  				op_params->ref_dec_op, op_params-
>>> vector_mask,
>>>  				ad->dev_id, queue_id, num_to_process,
>>> +				burst_sz, &total_time, &min_time,
>> &max_time,
>>> +				true);
>>> +	else
>>> +		iter = latency_test_enc(op_params->mp, bufs,
>>> +					op_params->ref_enc_op,
>>> +					ad->dev_id, queue_id,
>>> +					num_to_process, burst_sz,
>> &total_time,
>>> +					&min_time, &max_time);
>> This is a repeat of RTE_BBDEV_OP_TURBO_ENC.
>>
>> Do not need both.
> Fair enough. That is part of previous code but can simplify. 
>
>> If the point is to have a else and not fail when the op_type is unknown, then
>>
>> remove the earlier all and comment the else something like
>>
>> else /* RTE_BBDEC_OP_TURBO_ENC */
>>
>>> +
>>> +	if (iter <= 0)
>>> +		return TEST_FAILED;
>>> +
>>> +	printf("Operation latency:\n"
>>> +			"\tavg: %lg cycles, %lg us\n"
>>> +			"\tmin: %lg cycles, %lg us\n"
>>> +			"\tmax: %lg cycles, %lg us\n",
>>> +			(double)total_time / (double)iter,
>>> +			(double)(total_time * 1000000) / (double)iter /
>>> +			(double)rte_get_tsc_hz(), (double)min_time,
>>> +			(double)(min_time * 1000000) /
>> (double)rte_get_tsc_hz(),
>>> +			(double)max_time, (double)(max_time * 1000000) /
>>> +			(double)rte_get_tsc_hz());
>> Could remove a tab from the last 9 lines for better alignment with printf
> I am unsure I follow. The recommended spacing is 2 tabs for continuation and unsure how the alignment would be better.
> I typically only reduce to 1 tab only if I have to (80 chars limit becoming cumbersome with nested statements).

This is just an observation i don't want to get into the weeds with whitespace issues.


>
>>> +
>>> +	return TEST_SUCCESS;
>>> +}
>>> +
>>> +static int
>>> +validation_test(struct active_device *ad,
>>> +		struct test_op_params *op_params)
>>> +{
>>> +	int iter;
>>> +	uint16_t burst_sz = op_params->burst_sz;
>>> +	const uint16_t num_to_process = op_params->num_to_process;
>>> +	const enum rte_bbdev_op_type op_type = test_vector.op_type;
>>> +	const uint16_t queue_id = ad->queue_ids[0];
>>> +	struct test_buffers *bufs = NULL;
>>> +	struct rte_bbdev_info info;
>>> +	uint64_t total_time, min_time, max_time;
>>> +	const char *op_type_str;
>>> +
>>> +	total_time = max_time = 0;
>>> +	min_time = UINT64_MAX;
>>> +
>>> +	TEST_ASSERT_SUCCESS((burst_sz > MAX_BURST),
>>> +			"BURST_SIZE should be <= %u", MAX_BURST);
>>> +
>>> +	rte_bbdev_info_get(ad->dev_id, &info);
>>> +	bufs = &op_params-
>>> q_bufs[GET_SOCKET(info.socket_id)][queue_id];
>>> +
>>> +	op_type_str = rte_bbdev_op_type_str(op_type);
>>> +	TEST_ASSERT_NOT_NULL(op_type_str, "Invalid op type: %u",
>> op_type);
>>> +
>>> +	printf("+ ------------------------------------------------------- +\n");
>>> +	printf("== test: validation\ndev: %s, burst size: %u, num ops: %u, op
>> type: %s\n",
>>> +			info.dev_name, burst_sz, num_to_process,
>> op_type_str);
>>> +
>>> +	if (op_type == RTE_BBDEV_OP_TURBO_DEC)
>>> +		iter = latency_test_dec(op_params->mp, bufs,
>>> +				op_params->ref_dec_op, op_params-
>>> vector_mask,
>>> +				ad->dev_id, queue_id, num_to_process,
>>>  				burst_sz, &total_time, &min_time,
>> &max_time);
>>> +	else if (op_type == RTE_BBDEV_OP_TURBO_ENC)
>>> +		iter = latency_test_enc(op_params->mp, bufs,
>>> +				op_params->ref_enc_op, ad->dev_id,
>> queue_id,
>>> +				num_to_process, burst_sz, &total_time,
>>> +				&min_time, &max_time);
>>> +	else if (op_type == RTE_BBDEV_OP_LDPC_ENC)
>>> +		iter = latency_test_ldpc_enc(op_params->mp, bufs,
>>> +				op_params->ref_enc_op, ad->dev_id,
>> queue_id,
>>> +				num_to_process, burst_sz, &total_time,
>>> +				&min_time, &max_time);
>>> +	else if (op_type == RTE_BBDEV_OP_LDPC_DEC)
>>> +		iter = latency_test_ldpc_dec(op_params->mp, bufs,
>>> +				op_params->ref_dec_op, op_params-
>>> vector_mask,
>>> +				ad->dev_id, queue_id, num_to_process,
>>> +				burst_sz, &total_time, &min_time,
>> &max_time,
>>> +				false);
>> This 'false' is the only change from f latency_test.
>>
>> These should be refactored to a common function. Then use a #define or
>> similar wrapper for calling with/without this flag.
> Fair enough. Thanks. I will push an update later today. 

Thanks.  This was the only serious thing in the patchset.

Tom

>
>> Tom
>>
>>>  	else
>>>  		iter = latency_test_enc(op_params->mp, bufs,
>>>  					op_params->ref_enc_op,
>>> @@ -4930,6 +5008,12 @@ typedef int (test_case_function)(struct
>>> active_device *ad,  }
>>>
>>>  static int
>>> +validation_tc(void)
>>> +{
>>> +	return run_test_case(validation_test); }
>>> +
>>> +static int
>>>  interrupt_tc(void)
>>>  {
>>>  	return run_test_case(throughput_test); @@ -4960,7 +5044,7 @@
>> typedef
>>> int (test_case_function)(struct active_device *ad,
>>>  	.setup = testsuite_setup,
>>>  	.teardown = testsuite_teardown,
>>>  	.unit_test_cases = {
>>> -		TEST_CASE_ST(ut_setup, ut_teardown, latency_tc),
>>> +		TEST_CASE_ST(ut_setup, ut_teardown, validation_tc),
>>>  		TEST_CASES_END() /**< NULL terminate unit test array */
>>>  	}
>>>  };