* [PATCH v1 1/1] mldev: add scale and zero point to I/O info struct
@ 2024-07-31 6:25 Srikanth Yalavarthi
2024-10-17 21:37 ` Thomas Monjalon
0 siblings, 1 reply; 2+ messages in thread
From: Srikanth Yalavarthi @ 2024-07-31 6:25 UTC (permalink / raw)
To: Srikanth Yalavarthi; +Cc: dev, sshankarnara, aprabhu, ptakkar
Added scale and zero point to I/O information structure.
This would provision sharing the recommended sclae factor
and zero point to the user for quantization process.
Signed-off-by: Srikanth Yalavarthi <syalavarthi@marvell.com>
---
drivers/ml/cnxk/cn10k_ml_model.c | 4 ++++
drivers/ml/cnxk/cnxk_ml_io.h | 3 +++
drivers/ml/cnxk/mvtvm_ml_model.c | 4 ++++
lib/mldev/rte_mldev.h | 4 ++++
4 files changed, 15 insertions(+)
diff --git a/drivers/ml/cnxk/cn10k_ml_model.c b/drivers/ml/cnxk/cn10k_ml_model.c
index 0325cd54f1f..12a2dda800e 100644
--- a/drivers/ml/cnxk/cn10k_ml_model.c
+++ b/drivers/ml/cnxk/cn10k_ml_model.c
@@ -586,6 +586,8 @@ cn10k_ml_model_info_set(struct cnxk_ml_dev *cnxk_mldev, struct cnxk_ml_model *mo
input[i].nb_elements = io_info->input[i].nb_elements;
input[i].size = io_info->input[i].nb_elements *
rte_ml_io_type_size_get(io_info->input[i].qtype);
+ input[i].scale = 1.0 / io_info->input[i].scale;
+ input[i].zero_point = 0;
}
/* Set output info */
@@ -597,6 +599,8 @@ cn10k_ml_model_info_set(struct cnxk_ml_dev *cnxk_mldev, struct cnxk_ml_model *mo
output[i].nb_elements = io_info->output[i].nb_elements;
output[i].size = io_info->output[i].nb_elements *
rte_ml_io_type_size_get(io_info->output[i].qtype);
+ output[i].scale = io_info->output[i].scale;
+ output[i].zero_point = 0;
}
}
diff --git a/drivers/ml/cnxk/cnxk_ml_io.h b/drivers/ml/cnxk/cnxk_ml_io.h
index c33a9c23a11..83329c237aa 100644
--- a/drivers/ml/cnxk/cnxk_ml_io.h
+++ b/drivers/ml/cnxk/cnxk_ml_io.h
@@ -55,6 +55,9 @@ struct cnxk_ml_io {
/* Scale */
float scale;
+
+ /* Zero point */
+ int64_t zero_point;
};
/* Model / Layer IO structure */
diff --git a/drivers/ml/cnxk/mvtvm_ml_model.c b/drivers/ml/cnxk/mvtvm_ml_model.c
index e3234ae4422..88b0089295d 100644
--- a/drivers/ml/cnxk/mvtvm_ml_model.c
+++ b/drivers/ml/cnxk/mvtvm_ml_model.c
@@ -334,6 +334,8 @@ mvtvm_ml_model_info_set(struct cnxk_ml_dev *cnxk_mldev, struct cnxk_ml_model *mo
input[i].nb_elements = model->mvtvm.info.input[i].nb_elements;
input[i].size = model->mvtvm.info.input[i].nb_elements *
rte_ml_io_type_size_get(model->mvtvm.info.input[i].qtype);
+ input[i].scale = model->mvtvm.info.input[i].scale;
+ input[i].zero_point = 0;
}
/* Set output info */
@@ -345,6 +347,8 @@ mvtvm_ml_model_info_set(struct cnxk_ml_dev *cnxk_mldev, struct cnxk_ml_model *mo
output[i].nb_elements = model->mvtvm.info.output[i].nb_elements;
output[i].size = model->mvtvm.info.output[i].nb_elements *
rte_ml_io_type_size_get(model->mvtvm.info.output[i].qtype);
+ input[i].scale = model->mvtvm.info.output[i].scale;
+ input[i].zero_point = 0;
}
return;
diff --git a/lib/mldev/rte_mldev.h b/lib/mldev/rte_mldev.h
index 634af3d5e1a..e2f715f140d 100644
--- a/lib/mldev/rte_mldev.h
+++ b/lib/mldev/rte_mldev.h
@@ -937,6 +937,10 @@ struct rte_ml_io_info {
/** Number of elements in tensor */
uint64_t size;
/** Size of tensor in bytes */
+ float scale;
+ /** Scale factor */
+ int64_t zero_point;
+ /* Zero point */
};
/** Model information structure */
--
2.45.1
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: [PATCH v1 1/1] mldev: add scale and zero point to I/O info struct
2024-07-31 6:25 [PATCH v1 1/1] mldev: add scale and zero point to I/O info struct Srikanth Yalavarthi
@ 2024-10-17 21:37 ` Thomas Monjalon
0 siblings, 0 replies; 2+ messages in thread
From: Thomas Monjalon @ 2024-10-17 21:37 UTC (permalink / raw)
To: Srikanth Yalavarthi; +Cc: dev, sshankarnara, aprabhu, ptakkar
31/07/2024 08:25, Srikanth Yalavarthi:
> Added scale and zero point to I/O information structure.
> This would provision sharing the recommended sclae factor
> and zero point to the user for quantization process.
>
> Signed-off-by: Srikanth Yalavarthi <syalavarthi@marvell.com>
Applied with a Doxygen comment fixed.
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2024-10-17 21:37 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2024-07-31 6:25 [PATCH v1 1/1] mldev: add scale and zero point to I/O info struct Srikanth Yalavarthi
2024-10-17 21:37 ` Thomas Monjalon
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).