diff --git a/pocketfft_hdronly.h b/pocketfft_hdronly.h
index b74b1c653c748d72c0613506099d5a733ba12e01..ea6538eb194e888ad1cc94e7cfcdfa6893ce0983 100644
--- a/pocketfft_hdronly.h
+++ b/pocketfft_hdronly.h
@@ -2902,25 +2902,16 @@ template<typename T> POCKETFFT_NOINLINE void general_c(
     while (it.remaining()>0)
       {
       it.advance(1);
-      auto tdata = reinterpret_cast<cmplx<T> *>(storage.data());
-      if ((&tin[0]==&out[0]) && (it.stride_out()==sizeof(cmplx<T>))) // fully in-place
-        forward ? plan->forward (&out[it.oofs(0)], fct)
-                : plan->backward(&out[it.oofs(0)], fct);
-      else if (it.stride_out()==sizeof(cmplx<T>)) // compute FFT in output location
-        {
-        for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tin[it.iofs(i)];
-        forward ? plan->forward (&out[it.oofs(0)], fct)
-                : plan->backward(&out[it.oofs(0)], fct);
-        }
-      else
-        {
+      auto buf = it.stride_out() == sizeof(cmplx<T>) ?
+        &out[it.oofs(0)] : reinterpret_cast<cmplx<T> *>(storage.data());
+
+      if (buf != &tin[it.iofs(0)])
         for (size_t i=0; i<len; ++i)
-          tdata[i] = tin[it.iofs(i)];
-        forward ? plan->forward (tdata, fct) : plan->backward(tdata, fct);
+          buf[i] = tin[it.iofs(i)];
+      forward ? plan->forward (buf, fct) : plan->backward(buf, fct);
+      if (buf != &out[it.oofs(0)])
         for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tdata[i];
-        }
+          out[it.oofs(i)] = buf[i];
       }
 } // end of parallel region
     fct = T(1); // factor has been applied, use 1 for remaining axes
@@ -3034,23 +3025,16 @@ template<typename Trafo, typename T> POCKETFFT_NOINLINE void general_dcst(
     while (it.remaining()>0)
       {
       it.advance(1);
-      auto tdata = reinterpret_cast<T *>(storage.data());
-      if ((&tin[0]==&out[0]) && (it.stride_out()==sizeof(T))) // fully in-place
-        plan->exec(&out[it.oofs(0)], fct, ortho, type, cosine);
-      else if (it.stride_out()==sizeof(T)) // compute FFT in output location
-        {
-        for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tin[it.iofs(i)];
-        plan->exec(&out[it.oofs(0)], fct, ortho, type, cosine);
-        }
-      else
-        {
+      auto buf = it.stride_out() == sizeof(T) ? &out[it.oofs(0)]
+        : reinterpret_cast<T *>(storage.data());
+
+      if (buf != &tin[it.iofs(0)])
         for (size_t i=0; i<len; ++i)
-          tdata[i] = tin[it.iofs(i)];
-        plan->exec(tdata, fct, ortho, type, cosine);
+          buf[i] = tin[it.iofs(i)];
+      plan->exec(buf, fct, ortho, type, cosine);
+      if (buf != &out[it.oofs(0)])
         for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tdata[i];
-        }
+          out[it.oofs(i)] = buf[i];
       }
 } // end of parallel region
     fct = T(1); // factor has been applied, use 1 for remaining axes
@@ -3242,45 +3226,22 @@ template<typename T> POCKETFFT_NOINLINE void general_r(
     while (it.remaining()>0)
       {
       it.advance(1);
-      auto tdata = reinterpret_cast<T *>(storage.data());
-      if ((&tin[0]==&out[0]) && (it.stride_out()==sizeof(T))) // fully in-place
-        {
-        if ((!r2c) && forward)
-          for (size_t i=2; i<len; i+=2)
-            out[it.oofs(i)] = -out[it.oofs(i)];
-        forward ? plan->forward (&out[it.oofs(0)], fct)
-                : plan->backward(&out[it.oofs(0)], fct);
-        if (r2c && (!forward))
-          for (size_t i=2; i<len; i+=2)
-            out[it.oofs(i)] = -out[it.oofs(i)];
-        }
-      else if (it.stride_out()==sizeof(T)) // compute FFT in output location
-        {
-        for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tin[it.iofs(i)];
-        if ((!r2c) && forward)
-          for (size_t i=2; i<len; i+=2)
-            out[it.oofs(i)] = -out[it.oofs(i)];
-        forward ? plan->forward (&out[it.oofs(0)], fct)
-                : plan->backward(&out[it.oofs(0)], fct);
-        if (r2c && (!forward))
-          for (size_t i=2; i<len; i+=2)
-            out[it.oofs(i)] = -out[it.oofs(i)];
-        }
-      else
-        {
+      auto buf = it.stride_out() == sizeof(T) ?
+        &out[it.oofs(0)] : reinterpret_cast<T *>(storage.data());
+
+      if (buf != &tin[it.iofs(0)])
         for (size_t i=0; i<len; ++i)
-          tdata[i] = tin[it.iofs(i)];
-        if ((!r2c) && forward)
-          for (size_t i=2; i<len; i+=2)
-            tdata[i] = -tdata[i];
-        forward ? plan->forward (tdata, fct) : plan->backward(tdata, fct);
-        if (r2c && (!forward))
-          for (size_t i=2; i<len; i+=2)
-            tdata[i] = -tdata[i];
+          buf[i] = tin[it.iofs(i)];
+      if ((!r2c) && forward)
+        for (size_t i=2; i<len; i+=2)
+          buf[i] = -buf[i];
+      forward ? plan->forward(buf, fct) : plan->backward(buf, fct);
+      if (r2c && (!forward))
+        for (size_t i=2; i<len; i+=2)
+          buf[i] = -buf[i];
+      if (buf != &out[it.oofs(0)])
         for (size_t i=0; i<len; ++i)
-          out[it.oofs(i)] = tdata[i];
-        }
+          out[it.oofs(i)] = buf[i];
       }
 } // end of parallel region
     fct = T(1); // factor has been applied, use 1 for remaining axes