diff --git a/src/gallium/frontends/teflon/tfl_device.c b/src/gallium/frontends/teflon/tfl_device.c
index a21f95ba68d..6394bb3285a 100644
--- a/src/gallium/frontends/teflon/tfl_device.c
+++ b/src/gallium/frontends/teflon/tfl_device.c
@@ -166,9 +166,22 @@ fill_operation(struct teflon_delegate *delegate, TfLiteContext *tf_context, TfLi
                                   operation->conv.weight_tensor->dims[2] == 1;
       break;
    }
-   case kTfLiteBuiltinAveragePool2d:
+   case kTfLiteBuiltinMaxPool2d:
+      operation->pooling.type = PIPE_ML_POOLING_TYPE_MAX;
+      __attribute__((fallthrough));
+   case kTfLiteBuiltinAveragePool2d: {
+      TfLitePoolParams *params = node->builtin_data;
+
       operation->type = PIPE_ML_OPERATION_TYPE_POOLING;
+
+      /* Skip setting operation->pooling.type for PIPE_ML_POOLING_TYPE_AVG==0 */
+      operation->pooling.filter_height = params->filter_height;
+      operation->pooling.filter_width = params->filter_width;
+      operation->pooling.stride_x = params->stride_width;
+      operation->pooling.stride_y = params->stride_height;
+      operation->pooling.padding_same = params->padding == kTfLitePaddingSame;
       break;
+   }
    case kTfLiteBuiltinAdd:
       operation->type = PIPE_ML_OPERATION_TYPE_ADD;
       break;
@@ -573,6 +586,8 @@ tflite_builtin_op_name(TfLiteBuiltinOperator op)
       return "ADD";
    case kTfLiteBuiltinAveragePool2d:
       return "AVGPOOL";
+   case kTfLiteBuiltinMaxPool2d:
+      return "MAXPOOL";
    case kTfLiteBuiltinConv2d:
       return "CONV";
    case kTfLiteBuiltinDepthwiseConv2d:
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 208ac317400..d9c887bab9f 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -1056,6 +1056,11 @@ enum pipe_ml_operation_type {
    PIPE_ML_OPERATION_TYPE_TRANSPOSE,
 };
 
+enum pipe_ml_pooling_type {
+   PIPE_ML_POOLING_TYPE_AVG,
+   PIPE_ML_POOLING_TYPE_MAX,
+};
+
 /**
  * Information about a single operation inside a ML subgraph.
  */
@@ -1124,6 +1129,12 @@ struct pipe_ml_operation
          unsigned dilation_height_factor;
       } conv;
       struct {
+
+         /**
+          * Type of pooling operation.
+          */
+         enum pipe_ml_pooling_type type;
+
          /**
           * Stride used to access the input tensor on the x axis.
           */