diff mbox series

[v1,1/1] ml/cnxk: fix incorrect handling of TVM model I/O

Message ID 20240731054104.7576-1-syalavarthi@marvell.com (mailing list archive)
State New
Delegated to: Jerin Jacob
Headers
Series [v1,1/1] ml/cnxk: fix incorrect handling of TVM model I/O |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/loongarch-compilation success Compilation OK
ci/loongarch-unit-testing success Unit Testing PASS
ci/Intel-compilation success Compilation OK
ci/intel-Testing success Testing PASS
ci/github-robot: build success github build: passed
ci/intel-Functional success Functional PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-marvell-Functional success Functional Testing PASS
ci/iol-abi-testing success Testing PASS
ci/iol-broadcom-Performance success Performance Testing PASS
ci/iol-broadcom-Functional success Functional Testing PASS
ci/iol-compile-arm64-testing success Testing PASS
ci/iol-unit-arm64-testing success Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-unit-amd64-testing success Testing PASS
ci/iol-compile-amd64-testing pending Testing pending
ci/iol-sample-apps-testing success Testing PASS

Commit Message

Srikanth Yalavarthi July 31, 2024, 5:41 a.m. UTC 
Fixed incorrect handling of TVM models with single MRVL
layer. Set the I/O layout to packed and fixed calculation
of quantized and dequantized data buffer addresses.

Fixes: 5cea2c67edfc ("ml/cnxk: update internal TVM model info structure")
Fixes: df2358f3adce ("ml/cnxk: add structures for TVM model type")

Signed-off-by: Srikanth Yalavarthi <syalavarthi@marvell.com>
---
 drivers/ml/cnxk/cnxk_ml_ops.c    | 12 ++++++++----
 drivers/ml/cnxk/mvtvm_ml_model.c |  2 +-
 2 files changed, 9 insertions(+), 5 deletions(-)
diff mbox series

Patch

diff --git a/drivers/ml/cnxk/cnxk_ml_ops.c b/drivers/ml/cnxk/cnxk_ml_ops.c
index 971362b2420..07c7001b175 100644
--- a/drivers/ml/cnxk/cnxk_ml_ops.c
+++ b/drivers/ml/cnxk/cnxk_ml_ops.c
@@ -1462,7 +1462,8 @@  cnxk_ml_io_quantize(struct rte_ml_dev *dev, uint16_t model_id, struct rte_ml_buf
 	d_offset = 0;
 	q_offset = 0;
 	for (i = 0; i < info->nb_inputs; i++) {
-		if (model->type == ML_CNXK_MODEL_TYPE_TVM) {
+		if (model->type == ML_CNXK_MODEL_TYPE_TVM &&
+		    model->subtype != ML_CNXK_MODEL_SUBTYPE_TVM_MRVL) {
 			lcl_dbuffer = dbuffer[i]->addr;
 			lcl_qbuffer = qbuffer[i]->addr;
 		} else {
@@ -1474,7 +1475,8 @@  cnxk_ml_io_quantize(struct rte_ml_dev *dev, uint16_t model_id, struct rte_ml_buf
 		if (ret < 0)
 			return ret;
 
-		if (model->type == ML_CNXK_MODEL_TYPE_GLOW) {
+		if ((model->type == ML_CNXK_MODEL_TYPE_GLOW) ||
+		    (model->subtype == ML_CNXK_MODEL_SUBTYPE_TVM_MRVL)) {
 			d_offset += info->input[i].sz_d;
 			q_offset += info->input[i].sz_q;
 		}
@@ -1516,7 +1518,8 @@  cnxk_ml_io_dequantize(struct rte_ml_dev *dev, uint16_t model_id, struct rte_ml_b
 	q_offset = 0;
 	d_offset = 0;
 	for (i = 0; i < info->nb_outputs; i++) {
-		if (model->type == ML_CNXK_MODEL_TYPE_TVM) {
+		if (model->type == ML_CNXK_MODEL_TYPE_TVM &&
+		    model->subtype != ML_CNXK_MODEL_SUBTYPE_TVM_MRVL) {
 			lcl_qbuffer = qbuffer[i]->addr;
 			lcl_dbuffer = dbuffer[i]->addr;
 		} else {
@@ -1528,7 +1531,8 @@  cnxk_ml_io_dequantize(struct rte_ml_dev *dev, uint16_t model_id, struct rte_ml_b
 		if (ret < 0)
 			return ret;
 
-		if (model->type == ML_CNXK_MODEL_TYPE_GLOW) {
+		if ((model->type == ML_CNXK_MODEL_TYPE_GLOW) ||
+		    (model->subtype == ML_CNXK_MODEL_SUBTYPE_TVM_MRVL)) {
 			q_offset += info->output[i].sz_q;
 			d_offset += info->output[i].sz_d;
 		}
diff --git a/drivers/ml/cnxk/mvtvm_ml_model.c b/drivers/ml/cnxk/mvtvm_ml_model.c
index e3234ae4422..c8c4f61f134 100644
--- a/drivers/ml/cnxk/mvtvm_ml_model.c
+++ b/drivers/ml/cnxk/mvtvm_ml_model.c
@@ -356,7 +356,7 @@  mvtvm_ml_model_info_set(struct cnxk_ml_dev *cnxk_mldev, struct cnxk_ml_model *mo
 	metadata = &model->mvtvm.metadata;
 	strlcpy(info->name, metadata->model.name, TVMDP_NAME_STRLEN);
 
-	info->io_layout = RTE_ML_IO_LAYOUT_SPLIT;
+	info->io_layout = RTE_ML_IO_LAYOUT_PACKED;
 }
 
 void