fix output of 1d transpose parallel to be faster

Jonathan-Shoemaker · Jonathan-Shoemaker · commit 31af0ccb0907 · 2022-07-26T16:20:09.000-05:00
diff --git a/hls4ml/templates/vivado/nnet_utils/nnet_conv1dtranspose_resource.h b/hls4ml/templates/vivado/nnet_utils/nnet_conv1dtranspose_resource.h
@@ -150,8 +150,8 @@ void conv_1d_transpose_resource_cl(
                     for (unsigned i_res = 0; i_res < mult_n_out; i_res++) {
                         #pragma HLS UNROLL
 
-                        // res[(output_index-CONFIG_T::pad_left)*CONFIG_T::n_filt + i_res]
-                        *(res++) = cast<data_T, res_T, typename CONFIG_T::mult_config>(acc[i_pxl][i_res][i_sw]);
+                        res[(output_index-CONFIG_T::pad_left)*mult_n_out + i_res] = 
+                            cast<data_T, res_T, typename CONFIG_T::mult_config>(acc[i_pxl][i_res][i_sw]);
                     }
                 }
             }

Original file line number	Diff line number	Diff line change
`@@ -150,8 +150,8 @@ void conv_1d_transpose_resource_cl(`
`150`	`150`	`for (unsigned i_res = 0; i_res < mult_n_out; i_res++) {`
`151`	`151`	`#pragma HLS UNROLL`
`152`	`152`
`153`		`- // res[(output_index-CONFIG_T::pad_left)*CONFIG_T::n_filt + i_res]`
`154`		`- *(res++) = cast<data_T, res_T, typename CONFIG_T::mult_config>(acc[i_pxl][i_res][i_sw]);`
	`153`	`+ res[(output_index-CONFIG_T::pad_left)*mult_n_out + i_res] =`
	`154`	`+ cast<data_T, res_T, typename CONFIG_T::mult_config>(acc[i_pxl][i_res][i_sw]);`
`155`	`155`	`}`
`156`	`156`	`}`
`157`	`157`	`}`