Fix up some pragmas to align unrolling

dgburnette · Apr 30, 2024 · e753456 · e753456
1 parent b25f41a
commit e753456
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 8 deletions.
diff --git a/hls4ml/templates/catapult/nnet_utils/nnet_conv2d_stream.h b/hls4ml/templates/catapult/nnet_utils/nnet_conv2d_stream.h
@@ -52,7 +52,7 @@ void conv_2d_encoded_cl(
     constexpr int ce_reuse_factor =
         CONFIG_T::reuse_factor * (CONFIG_T::strategy == nnet::latency && data_T::size / CONFIG_T::n_chan == 1);
     (void)ce_reuse_factor;
-    #pragma hls_pipeline_init_interval ce_reuse_factor
+    #pragma hls_pipeline_init_interval 1
 ReadInputHeight:
     for (unsigned i_ih = 0; i_ih < CONFIG_T::in_height; i_ih++) {
     ReadInputWidth:
@@ -82,7 +82,7 @@ void conv_2d_buffer_cl(
 
     constexpr int ce_reuse_factor = CONFIG_T::reuse_factor * (CONFIG_T::strategy == nnet::latency);
     (void)ce_reuse_factor;
-    #pragma hls_pipeline_init_interval ce_reuse_factor
+    #pragma hls_pipeline_init_interval 1
 ReadInputHeight:
     for (unsigned i_ih = 0; i_ih < CONFIG_T::in_height; i_ih++) {
     ReadInputWidth:

diff --git a/hls4ml/templates/catapult/nnet_utils/nnet_dense_latency.h b/hls4ml/templates/catapult/nnet_utils/nnet_dense_latency.h
@@ -15,14 +15,15 @@ void dense_latency(data_T data[CONFIG_T::n_in], res_T res[CONFIG_T::n_out],
                    typename CONFIG_T::weight_t weights[CONFIG_T::n_in * CONFIG_T::n_out],
                    typename CONFIG_T::bias_t biases[CONFIG_T::n_out]) {
     constexpr int ce_reuse_factor = CONFIG_T::reuse_factor;
+
     // Partial unroll config
     constexpr int prod1_unroll =
         (ce_reuse_factor < CONFIG_T::n_in) ? CONFIG_T::n_in : (int)(CONFIG_T::n_in * CONFIG_T::n_out) / ce_reuse_factor;
     constexpr int prod2_unroll = (int)CONFIG_T::n_out / ce_reuse_factor;
 
     (void)ce_reuse_factor; // to silence compiler warnings
-    (void)prod1_unroll;
     (void)prod2_unroll;
+    (void)prod1_unroll;
 
     // For Catapult, add an extra scope so that we can apply the pipeline pragma as if it applied to the function
     #pragma hls_pipeline_init_interval ce_reuse_factor
@@ -68,17 +69,17 @@ void dense_latency(data_T data[CONFIG_T::n_in], res_T res[CONFIG_T::n_out],
         }
 
     // Initialize accumulator with input biases
-    #pragma hls_unroll
+    #pragma hls_unroll prod2_unroll
     ResetAccum:
         for (unsigned int iacc = 0; iacc < CONFIG_T::n_out; iacc++) {
             acc[iacc] = (typename CONFIG_T::accum_t)biases[iacc];
         }
 
     // Accumulate multiplication result
-    #pragma hls_unroll
+    #pragma hls_unroll prod1_unroll
     Accum1:
         for (unsigned int ii = 0; ii < CONFIG_T::n_in; ii++) {
-        #pragma hls_unroll
+        #pragma hls_unroll prod2_unroll
         Accum2:
             for (unsigned int jj = 0; jj < CONFIG_T::n_out; jj++) {
                 int index = ii * CONFIG_T::n_out + jj;
@@ -87,15 +88,14 @@ void dense_latency(data_T data[CONFIG_T::n_in], res_T res[CONFIG_T::n_out],
         }
 
     // Cast to "res_t" type
-    #pragma hls_unroll
+    #pragma hls_unroll prod2_unroll
     Result:
         for (unsigned int ires = 0; ires < CONFIG_T::n_out; ires++) {
             // res[ires] = (res_T) (acc[ires]);
             res[ires] = cast<data_T, res_T, CONFIG_T>(acc[ires]);
         }
     } while (false); // one iteration loop
 }
-
 } // namespace nnet
 
 #endif