From d12dd8b1deab03526cc44abd3f229a6a40db43fe Mon Sep 17 00:00:00 2001
From: daquexian <daquexian566@gmail.com>
Date: Thu, 22 Aug 2019 18:06:51 +0800
Subject: [PATCH] Revert "use bgemm_naive instead of bconv_naive as fallback
 due to the 128-align weight"

This reverts commit be606fa6f87db0a4e9166b79f5d9cb86804525ba.
---
 dabnn/layers/BinConv.cpp | 57 +++++++++++++++++++---------------------
 1 file changed, 27 insertions(+), 30 deletions(-)

diff --git a/dabnn/layers/BinConv.cpp b/dabnn/layers/BinConv.cpp
index 7481337..27ad5ce 100644
--- a/dabnn/layers/BinConv.cpp
+++ b/dabnn/layers/BinConv.cpp
@@ -25,16 +25,14 @@ BinConv::BinConv(NetCP net, const std::string &name, css input, css weight,
       stride_h(stride_h),
       stride_w(stride_w) {
     auto &mat_map = net.lock()->mat_map_;
-    if (direct_conv_compatible()) {
-        const auto binaized_name = "binaized_for_" + output + "_cal";
-        if (mat_map.find(binaized_name) == mat_map.end()) {
-            auto &input_mat = *mat_map[input];
-            mat_map[binaized_name] =
-                std::make_shared<Mat>(input_mat.h, input_mat.w, input_mat.elem_c,
-                                      DataType::Bit, binaized_name);
-        }
-        binarized_mat = mat(binaized_name);
+    const auto binaized_name = "binaized_for_" + output + "_cal";
+    if (mat_map.find(binaized_name) == mat_map.end()) {
+        auto &input_mat = *mat_map[input];
+        mat_map[binaized_name] =
+            std::make_shared<Mat>(input_mat.h, input_mat.w, input_mat.elem_c,
+                                  DataType::Bit, binaized_name);
     }
+    binarized_mat = mat(binaized_name);
 
     const auto pad_name = "pad_for_" + output + "_cal";
     if (mat_map.find(pad_name) == mat_map.end()) {
@@ -45,18 +43,18 @@ BinConv::BinConv(NetCP net, const std::string &name, css input, css weight,
     }
     padded_mat = mat(pad_name);
 
-    if (net.lock()->optimize && !direct_conv_compatible()) {
-
-        const auto col_mat_name = "col_for_" + output + "_cal";
-        if (mat_map.find(col_mat_name) == mat_map.end()) {
-            const auto len =
-                output_mat->h * output_mat->w *
-                align_to(weight_mat->h * weight_mat->w * input_mat->elem_c, 128);
-            mat_map[col_mat_name] =
-                std::make_shared<Mat>(1, 1, len, bnn::DataType::Bit);
-        }
-        col_mat = mat(col_mat_name);
+    const auto col_mat_name = "col_for_" + output + "_cal";
+    if (mat_map.find(col_mat_name) == mat_map.end()) {
+        const auto len =
+            output_mat->h * output_mat->w *
+            align_to(weight_mat->h * weight_mat->w * input_mat->elem_c, 128);
+        mat_map[col_mat_name] =
+            std::make_shared<Mat>(1, 1, len, bnn::DataType::Bit);
+    }
+    col_mat = mat(col_mat_name);
 
+    if (net.lock()->optimize && !direct_conv_compatible() &&
+        gemm_compatible()) {
         const auto trans_weight_mat_name = "trans_" + weight;
         // transpose the weight for bgemm
         const int m = weight_mat->n;
@@ -128,7 +126,7 @@ void BinConv::forward_impl() const {
             pack_mat(*input_mat, *binarized_mat);
             pad(*binarized_mat, pad_h, pad_w, *padded_mat);
             bconv_3x3(*padded_mat, *weight_mat, *output_mat, stride_h);
-        } else {
+        } else if (gemm_compatible()) {
             output_mat->fill<float>(0.f);
 
             bnn::fused_binarize_im2col(*input_mat, weight_mat->h, weight_mat->w,
@@ -138,15 +136,14 @@ void BinConv::forward_impl() const {
             const int m = weight_mat->n;
             const int n = output_mat->h * output_mat->w;
             const int k = weight_mat->total() / weight_mat->n;
-            if (gemm_compatible()) {
-                bgemm(m, n, k, static_cast<uint64_t *>(transposed_weight_mat->data),
-                      m, static_cast<uint64_t *>(col_mat->data), k,
-                      static_cast<float *>(output_mat->data), m);
-            } else {
-                bgemm_naive(m, n, k, static_cast<uint64_t *>(transposed_weight_mat->data),
-                      m, static_cast<uint64_t *>(col_mat->data), k,
-                      static_cast<float *>(output_mat->data), m);
-            }
+            bgemm(m, n, k, static_cast<uint64_t *>(transposed_weight_mat->data),
+                  m, static_cast<uint64_t *>(col_mat->data), k,
+                  static_cast<float *>(output_mat->data), m);
+        } else {
+            pack_mat(*input_mat, *binarized_mat);
+            baseline_bconv(*binarized_mat, *weight_mat, weight_mat->h,
+                           weight_mat->w, pad_h, pad_w, stride_h, stride_w, 1,
+                           1, output_mat->c, *output_mat);
         }
     } else {
         pack_mat(*input_mat, *binarized_mat);