[EXT] Re: [PATCH v6 09/12] app/mldev: enable support for inference batches

[Srikanth Yalavarthi]

> -----Original Message-----
> From: Thomas Monjalon <thomas at monjalon.net>
> Sent: 16 March 2023 23:18
> To: Srikanth Yalavarthi <syalavarthi at marvell.com>
> Cc: dev at dpdk.org; Shivah Shankar Shankar Narayan Rao
> <sshankarnara at marvell.com>; Jerin Jacob Kollanukkaran
> <jerinj at marvell.com>; Anup Prabhu <aprabhu at marvell.com>; Prince Takkar
> <ptakkar at marvell.com>; Parijat Shukla <pshukla at marvell.com>
> Subject: [EXT] Re: [PATCH v6 09/12] app/mldev: enable support for inference
> batches
> 
> External Email
> 
> ----------------------------------------------------------------------
> 11/03/2023 16:09, Srikanth Yalavarthi:
> > @@ -528,8 +533,8 @@ ml_request_initialize(struct rte_mempool *mp,
> void *opaque, void *obj, unsigned
> >  	req->niters = 0;
> >
> >  	/* quantize data */
> > -	rte_ml_io_quantize(t->cmn.opt->dev_id, t->model[t->fid].id,
> > -			   t->model[t->fid].info.batch_size, t->model[t-
> >fid].input, req->input);
> > +	rte_ml_io_quantize(t->cmn.opt->dev_id, t->model[t->fid].id, t-
> >model[t->fid].nb_batches,
> > +			   t->model[t->fid].input, req->input);
> >  }
> >
> >  int
> > @@ -547,7 +552,7 @@ ml_inference_iomem_setup(struct ml_test *test,
> struct ml_options *opt, uint16_t
> >  	int ret;
> >
> >  	/* get input buffer size */
> > -	ret = rte_ml_io_input_size_get(opt->dev_id, t->model[fid].id, t-
> >model[fid].info.batch_size,
> > +	ret = rte_ml_io_input_size_get(opt->dev_id, t->model[fid].id,
> > +t->model[fid].nb_batches,
> >  				       &t->model[fid].inp_qsize, &t-
> >model[fid].inp_dsize);
> >  	if (ret != 0) {
> >  		ml_err("Failed to get input size, model : %s\n",
> > opt->filelist[fid].model); @@ -555,9 +560,8 @@
> ml_inference_iomem_setup(struct ml_test *test, struct ml_options *opt,
> uint16_t
> >  	}
> >
> >  	/* get output buffer size */
> > -	ret = rte_ml_io_output_size_get(opt->dev_id, t->model[fid].id,
> > -					t->model[fid].info.batch_size, &t-
> >model[fid].out_qsize,
> > -					&t->model[fid].out_dsize);
> > +	ret = rte_ml_io_output_size_get(opt->dev_id, t->model[fid].id, t-
> >model[fid].nb_batches,
> > +					&t->model[fid].out_qsize, &t-
> >model[fid].out_dsize);
> >  	if (ret != 0) {
> >  		ml_err("Failed to get input size, model : %s\n", opt-
> >filelist[fid].model);
> >  		return ret;
> > @@ -702,7 +706,7 @@ ml_request_finish(struct rte_mempool *mp, void
> *opaque, void *obj, unsigned int
> >  		return;
> >
> >  	t->nb_used++;
> > -	rte_ml_io_dequantize(t->cmn.opt->dev_id, model->id, t->model[req-
> >fid].info.batch_size,
> > +	rte_ml_io_dequantize(t->cmn.opt->dev_id, model->id,
> > +t->model[req->fid].nb_batches,
> >  			     req->output, model->output);
> 
> These changes look unrelated with the topic of the patch.
> You should probably fix it when adding those lines at first.

The changes are related to the patch. Initially the number of batches run per inference is set to the default batch_size value of the mode, which is reported to the user through rte_ml_model_info_get.

This patch adds support to specify the number of batches to be run per inference. Hence, the default batch_size is replace with nb_batches values specified by the user.
>