From d317f37a76d6654d9de500c8afce1f0b37aefa8b Mon Sep 17 00:00:00 2001
From: "-T.K.-" <t_k_233@outlook.com>
Date: Tue, 15 Oct 2024 15:24:13 -0700
Subject: [PATCH] FIX: fix simple example

---
 examples/simple/main.c         |  50 +++++++++++++++++----------------
 examples/simple/model.bin      |   2 +-
 examples/simple/model.pth      | Bin 2316 -> 2316 bytes
 examples/simple/scripts/run.py |  10 +++++--
 4 files changed, 34 insertions(+), 28 deletions(-)

diff --git a/examples/simple/main.c b/examples/simple/main.c
index 1c72f82..6ec230d 100644
--- a/examples/simple/main.c
+++ b/examples/simple/main.c
@@ -12,7 +12,7 @@
 #include "nn.h"
 
 // load the weight data block from the model.bin file
-INCLUDE_FILE(".rodata", "../model.bin", weights);
+INCLUDE_FILE(".rodata", "./model.bin", weights);
 extern uint8_t weights_data[];
 extern size_t weights_start[];
 extern size_t weights_end[];
@@ -21,36 +21,38 @@ extern size_t weights_end[];
 
 
 // Tensors can be defined either globally or locally
-Tensor A;
-Tensor B;
-Tensor C;
-Tensor D;
+Tensor2D_F32 A;
+Tensor2D_F32 B;
+Tensor2D_F32 C;
+Tensor1D_F32 D;
 
 /**
  * Initialize the required tensors for the model
  */
-void init(Tensor *A, Tensor *B, Tensor *C, Tensor *D) {
-  nn_init_tensor(A, 2, (size_t[]){3, 3}, DTYPE_F32, (float *)malloc(9 * sizeof(float)));
-  nn_init_tensor(B, 2, (size_t[]){3, 3}, DTYPE_F32, (float *)(weights_data + 3 * sizeof(float)));
-  nn_init_tensor(C, 2, (size_t[]){3, 3}, DTYPE_F32, (float *)malloc(9 * sizeof(float)));
-  nn_init_tensor(D, 1, (size_t[]){3}, DTYPE_F32, (float *)(weights_data + 0 * sizeof(float)));
+void init(Tensor2D_F32 *A, Tensor2D_F32 *B, Tensor2D_F32 *C, Tensor1D_F32 *D) {
+  A->shape[0] = 3;  A->shape[1] = 3;
+  A->data = (float *)malloc(9 * sizeof(float));
+  B->shape[0] = 3;  B->shape[1] = 3;
+  B->data = (float *)(weights_data + 3 * sizeof(float));
+  C->shape[0] = 3;  C->shape[1] = 3;
+  C->data = (float *)malloc(9 * sizeof(float));
+  D->shape[0] = 3;
+  D->data = (float *)(weights_data + 0 * sizeof(float));
 }
 
 /**
  * Deinitialize the tensors used for the model
  */
-void deinit(Tensor *A, Tensor *B, Tensor *C, Tensor *D) {
-  nn_freeTensor(A);
-  nn_freeTensor(B);
-  nn_freeTensor(C);
-  nn_freeTensor(D);
+void deinit(Tensor2D_F32 *A, Tensor2D_F32 *B, Tensor2D_F32 *C, Tensor1D_F32 *D) {
+  free(A->data);
+  free(C->data);
 }
 
 /**
  * Forward pass of the model
  */
-void forward(Tensor *C, Tensor *A, Tensor *B, Tensor *D) {
-  nn_Linear_F32(C, A, B, D);
+void forward(Tensor2D_F32 *C, Tensor2D_F32 *A, Tensor2D_F32 *B, Tensor1D_F32 *D) {
+  nn_addmm_f32(C, A, B, D);
 }
 
 
@@ -59,25 +61,25 @@ int main() {
 
   // load the input data to the tensor
   float input_data[] = {
-    1., 2., 3.,
-    1., 2., 3.,
-    1., 2., 3.,
+    1.0, 2.0, 3.0,
+    1.0, 2.0, 3.0,
+    1.0, 2.0, 3.0,
   };
   memcpy(A.data, input_data, 9 * sizeof(float));
   
   forward(&C, &A, &B, &D);
 
   printf("A:\n");
-  nn_printf(&A);
+  nn_print_tensor2d_f32(&A);
 
   printf("B:\n");
-  nn_printf(&B);
+  nn_print_tensor2d_f32(&B);
   
   printf("C:\n");
-  nn_printf(&C);
+  nn_print_tensor2d_f32(&C);
   
   printf("D:\n");
-  nn_printf(&D);
+  nn_print_tensor1d_f32(&D);
   
   deinit(&A, &B, &C, &D);
 
diff --git a/examples/simple/model.bin b/examples/simple/model.bin
index 6797f84..223968e 100644
--- a/examples/simple/model.bin
+++ b/examples/simple/model.bin
@@ -1 +1 @@
-裍���>~K�ɋپ��c�S�>@k;�za�>B�Q��p>��2��k�
\ No newline at end of file
+�p>��2��k�裍���>~K�ɋپ��c�S�>@k;�za�>B�Q�
\ No newline at end of file
diff --git a/examples/simple/model.pth b/examples/simple/model.pth
index 35b668828ecab5e03ea0d250a21386b5d1c7c7f6..8f23055024fda254fcfd1b9627fec60e79f2f2ce 100644
GIT binary patch
delta 78
zcmeAX>Jgfd!`QhoH<yXAbMtMc-Ha?w%tl(9XR*{WNtzj&8=4syTbP&{m>V0K85sn4
bvvXM9|95k;0lNf5oe#STBMVrlgkuH(pz#<5

delta 78
zcmeAX>Jgfd!`QJgH<yXAWAkmM-Ha>&mX!&cXR*{WNtzoO8JU=w7+4sYn;V!Ln;Qjq
bvvY8sxDq_sfL#Kj&WBxvkp(PN!Z8B?x@s4p

diff --git a/examples/simple/scripts/run.py b/examples/simple/scripts/run.py
index 927a595..6fd36d4 100644
--- a/examples/simple/scripts/run.py
+++ b/examples/simple/scripts/run.py
@@ -2,6 +2,7 @@
 import torch
 import torch.nn as nn
 
+
 torch.manual_seed(0)
 
 class Simple(nn.Module):
@@ -20,8 +21,10 @@ def forward(self, x: torch.Tensor):
 # Create model
 model = Simple(dim=3)
 
+model.eval()
+
 # Save model
-torch.save(model, "model.pth")
+# torch.save(model, "model.pth")
 
 # Load model
 # model = torch.load("model.pth")
@@ -39,9 +42,10 @@ def forward(self, x: torch.Tensor):
 b1_flat = b1.astype(np.float32).flatten()
 
 with open("model.bin", "wb") as f:
-    f.write(w1_flat.tobytes())
+    # data here is ordered as bias first and then weight,
+    # following the XNNPACK convention
     f.write(b1_flat.tobytes())
-
+    f.write(w1_flat.tobytes())
 
 
 # Test model