handle 64 filter SE networks (#624)

* minor performance fixes

~5-10% improvement in CPU limited cases (tested with 32x4 network on GTX 970)

* handle 64 filter SE networks

 - need numFc1Out of 16

* Update params.cc

* fix diff

* fix whitespace

src/neural/cuda/fp16_kernels.cu

@@ -134,7 +134,15 @@ void Se_Fp16_NHWC(int N, int C, int numFc1Out, half* output, const half* skip,
                   const half* input, const half* w1, const half* b1,
                   const half* w2, const half* b2, const half* bPrev) {
   // TODO: Think of more elegant way to avoid this hardcoding :-/
-  if (numFc1Out == 32) {
+  if (numFc1Out == 16) {
+    if (C == 64) {
+      SE_Layer_NHWC<64, 16>
+          <<<N, C>>>(output, skip, input, w1, b1, w2, b2, bPrev);
+    } else {
+      // TODO: support other channel counts.
+      throw Exception("channel count unsupported by SE layer");
+    }
+  } else if (numFc1Out == 32) {
     if (C == 64) {
       SE_Layer_NHWC<64, 32>
           <<<N, C>>>(output, skip, input, w1, b1, w2, b2, bPrev);