[RUNTIME] Lower-level interface for executing functions

python/examples/tutorials/conv2d.py

@@ -8,7 +8,9 @@ class _conv(torch.autograd.Function):
                          TYPE *C __noalias __aligned(16), 
                          float alpha,
                          // equivalent matmul
-                         int M, int N, int K,
+                         int M __retune,
+                         int N __retune,
+                         int K __retune,
                          // convolution properties
                          int pad_h, int pad_w, int stride_h, int stride_w,
                          // pointer increment
@@ -197,4 +199,4 @@ c = conv(a, b, pad, stride, time)
 print((cc - c).abs().max() / max(cc.max(), c.max()))
 print(time[0], 2*Z*H*W*CI*CO*R*S/(time[0]*1e-9)*1e-12)
 #zc  = torch.matmul(a,b)
-#zc_ = dot(a,b)
+#zc_ = dot(a,b)