Use fabsf instead of fabs for single floats

fabs() returns a double and under some circumstances the compiler emits
a warning when implicitly casting this to a float.

BUG=aomedia:2256

Change-Id: I7d4a2b867b530c3d2be9549bd76129a6d7551c1c

test/av1_nn_predict_test.cc

diff --git a/test/av1_nn_predict_test.cc b/test/av1_nn_predict_test.cc
index 6c4eed4..0574a15 100644
--- a/test/av1_nn_predict_test.cc
+++ b/test/av1_nn_predict_test.cc
@@ -126,7 +126,7 @@
             << shape_str << ")";
       } else {
         const float error = outputs_ref[node] - outputs_test[node];
-        const float relative_error = fabs(error / outputs_ref[node]);
+        const float relative_error = fabsf(error / outputs_ref[node]);
         ASSERT_LE(relative_error, epsilon)
             << "Excessive relative error between reference and test ("
             << shape_str << ")";