Replace __builtin_popcountl in armv7, since long is 32-bit here

JDAI-CV · May 28, 2019 · b47929a · b47929a
1 parent 275a855
commit b47929a
Show file tree

Hide file tree

Showing 5 changed files with 37 additions and 4 deletions.
diff --git a/.daq_pm/configs/bgemm_test_v7 b/.daq_pm/configs/bgemm_test_v7
@@ -0,0 +1,7 @@
+# It is a configuration file for [project_manager.vim](https://github.com/daquexian/project_manager.vim)
+name binary-nn
+type cpp
+build_dir build_test_v7
+target bgemm_test
+cmake_options -DCMAKE_TOOLCHAIN_FILE=~/Android/Sdk/ndk-bundle/build/cmake/android.toolchain.cmake -DANDROID_PLATFORM=android-28 -DANDROID_ABI=armeabi-v7a -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DCMAKE_BUILD_TYPE=Release -GNinja
+binary ~/adb_push_and_run.sh tests/bgemm_test
diff --git a/.daq_pm/configs/bgemm_test_x86 b/.daq_pm/configs/bgemm_test_x86
@@ -0,0 +1,7 @@
+# It is a configuration file for [project_manager.vim](https://github.com/daquexian/project_manager.vim)
+name binary-nn
+type cpp
+build_dir build_test_x86
+target bgemm_test
+cmake_options  -DCMAKE_EXPORT_COMPILE_COMMANDS=ON -DBNN_BUILD_MAIN_LIB=ON -GNinja
+binary tests/bgemm_test
diff --git a/common/baseline.h b/common/baseline.h
@@ -10,8 +10,12 @@
 #include <dabnn/mat.h>
 
 inline int bitcount(uint64_t x) {
+#ifdef __aarch64__
+    return __builtin_popcountl(x);
+#else
     std::bitset<64> bs(x);
     return bs.count();
+#endif
 }
 
 inline void baseline_pack_mat(const bnn::Mat &float_mat, bnn::Mat &binary_mat) {

diff --git a/dabnn/bgemm.h b/dabnn/bgemm.h
@@ -6,6 +6,7 @@
 #if __ARM_NEON
 #include <arm_neon.h>
 #endif  // __ARM_NEON
+#include <common/baseline.h>
 #include <common/helper.h>
 
 #define P 8
@@ -86,7 +87,7 @@ inline void inner_kernel(const int m, const int n, const int k,
         FOR(_j, 0, j) {
             FOR(_i, i, m) {
                 FORZ(_k, k) {
-                    C(_i, _j) += __builtin_popcountl(A(_i, _k) ^ B(_k, _j));
+                    C(_i, _j) += bitcount(A(_i, _k) ^ B(_k, _j));
                 }
             }
         }
@@ -95,7 +96,7 @@ inline void inner_kernel(const int m, const int n, const int k,
         FOR(_j, j, n) {
             FOR(_i, 0, i) {
                 FORZ(_k, k) {
-                    C(_i, _j) += __builtin_popcountl(A(_i, _k) ^ B(_k, _j));
+                    C(_i, _j) += bitcount(A(_i, _k) ^ B(_k, _j));
                 }
             }
         }
@@ -104,7 +105,7 @@ inline void inner_kernel(const int m, const int n, const int k,
         FOR(_j, j, n) {
             FOR(_i, i, m) {
                 FORZ(_k, k) {
-                    C(_i, _j) += __builtin_popcountl(A(_i, _k) ^ B(_k, _j));
+                    C(_i, _j) += bitcount(A(_i, _k) ^ B(_k, _j));
                 }
             }
         }
@@ -430,7 +431,7 @@ inline void bgemm_naive(const int m, const int n, const int k,
         FORZ(j, n) {
             FORZ(h, k) {
                 C(i, j) += static_cast<float>(
-                    __builtin_popcountl((A(i, h) ^ B(h, j))));
+                    bitcount((A(i, h) ^ B(h, j))));
             }
         }
     }

diff --git a/tests/bitpack_test.cpp b/tests/bitpack_test.cpp
@@ -74,3 +74,17 @@ TEST(bitpack, pack_mat_fallback) {
 
     ASSERT_EQ(a_binary, expected);
 }
+
+TEST(bitpack, addv_v7) {
+    uint64_t data[2];
+    fill_rand_uint64(data, 2);
+    uint8x16_t v = vld1q_u8(reinterpret_cast<unsigned char *>(data));
+    auto v1 = vcntq_u8(v);
+    auto v2 = vpaddlq_u8(v1);
+    auto v3 = vpaddlq_u16(v2);
+    auto v4 = vpaddlq_u32(v3);
+
+    auto res = vgetq_lane_u64(v4, 0) + vgetq_lane_u64(v4, 1);
+
+    ASSERT_EQ(res, __builtin_popcountl(data[0]) + __builtin_popcountl(data[1]));
+}