fix avg pooling, count_include_pad

nits

ggml-cuda.cu

@@ -6056,7 +6056,7 @@ static  __global__ void pool2d_nchw_kernel(
         const int start_w = cur_ow * sw - pw;
         const int bw = max(0, start_w);
         const int ew = min(iw, start_w + kw);
-        const To scale = 1. / ((eh - bh) * (ew - bw));
+        const To scale = 1. / (kh * kw);
         To res = 0;
         switch(op){
             case GGML_OP_POOL_AVG: res = 0; break;

ggml.c

@@ -5569,8 +5569,8 @@ struct ggml_tensor * ggml_pool_2d(
         int                   k1,
         int                   s0,
         int                   s1,
-        float                 p0,
-        float                 p1) {
+        int                   p0,
+        int                   p1) {
 
     bool is_node = false;
 

ggml.h

@@ -1600,8 +1600,8 @@ extern "C" {
             int                   k1,
             int                   s0,
             int                   s1,
-            float                 p0,
-            float                 p1);
+            int                   p0,
+            int                   p1);
 
     // nearest interpolate
     // used in stable-diffusion