Merge pull request #4 from pkuyym/fix-gpu-numerical

Fix numerical problem for gpu version.
gangliao · Aug 1, 2017 · 430e949 · 430e949
2 parents 0a60ee5 + 10ab30b
commit 430e949
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 1 deletion.
diff --git a/include/detail/ctc_helper.h b/include/detail/ctc_helper.h
@@ -44,6 +44,13 @@ template <typename Arg, typename Res = Arg> struct exponential {
     HOSTDEVICE Res operator()(const Arg& x) const {return std::exp(x);}
 };
 
+template <typename Arg, typename Res = Arg> struct clipped_exponential {
+    HOSTDEVICE Res operator()(const Arg& x) const {
+        Arg min_Arg = std::numeric_limits<Arg>::min();
+        return maximum<Res>()(min_Arg, std::exp(x));
+    }
+};
+
 template<typename Arg1, typename Arg2 = Arg1, typename Res=Arg1>
 struct log_plus {
     typedef Res result_type;

diff --git a/include/detail/gpu_ctc.h b/include/detail/gpu_ctc.h
@@ -392,7 +392,7 @@ GpuCTC<ProbT>::compute_probs(const ProbT* const activations) {
 
     // Kernel launch to calculate probabilities
     compute_probs_kernel<ProbT, VT><<<grid_size, NT, 0, stream_>>>
-        (ctc_helper::exponential<ProbT>(), probs_,
+        (ctc_helper::clipped_exponential<ProbT>(), probs_,
          denoms_, out_dim_, num_elements);
 
     return CTC_STATUS_SUCCESS;