yjy

src/yjy/CMakeLists.txt

@@ -0,0 +1,5 @@
+add_library(apsp_yjy SHARED)
+
+set(CMAKE_CXX_FLAGS "${CAMKE_CXX_FLAGS} -std=c++11 -pthread -O0 -msse4.1")
+target_sources(apsp_yjy PRIVATE apsp_yjy.cc)
+# target_compile_options(apsp_baseline PRIVATE -O3 -march=native -fopenmp -Wall -Wextra -Wpedantic -Werror)

src/yjy/apsp_yjy.cc

@@ -0,0 +1,57 @@
+#include "graph.hh"
+#include <thread>
+#include <immintrin.h>    // AVX
+#define MAXTHREADS 512 //这里设置最大线程数，代码会动态设置调用多少线程，所以这里直接设置一个资源最大值——yjy
+
+void funcyjy(int k, int *Distance, int vertex_num_, int n, int total)
+{
+    int ij = 0, ik = k + n * vertex_num_, kj = 0, DistanceNew = 0;
+    for (int i = n; i < vertex_num_; i += total) {
+            kj = k * vertex_num_;
+            ij = i * vertex_num_;
+            int Distance_ik = Distance[ik],j=0;
+            for (; j+4 < vertex_num_; j+=4) {
+                __m128i Disk =  _mm_load_si128((__m128i*)(Distance + kj));
+                __m128i one = _mm_set1_epi32(Distance_ik);
+                Disk=_mm_add_epi32(Disk, one);
+                __m128i Dis =  _mm_load_si128((__m128i*)(Distance + ij));
+                Dis=_mm_min_epi32(Dis,Disk);
+                _mm_storeu_si128((__m128i*)(Distance + ij), Dis);
+                //DistanceNew = Distance_ik + Distance[kj];
+                //Distance[ij] = Distance[ij] < DistanceNew ? Distance[ij] : DistanceNew;
+                //result(i, j) = std::min(Distance[ij], Distance_ik + Distance[kj]);
+                ij+=4;
+                kj+=4;   
+            }
+            for(; j < vertex_num_ ;++j)
+            {
+                DistanceNew = Distance_ik + Distance[kj];
+                Distance[ij] = Distance[ij] < DistanceNew ? Distance[ij] : DistanceNew;
+                //result(i, j) = std::min(Distance[ij], Distance_ik + Distance[kj]);
+                ++ij;
+                ++kj;   
+            }
+            ik += vertex_num_ * total;
+        }
+}
+
+
+Graph Graph::apsp() {
+    Graph result(*this);
+    int *Distance = result.get_raw_ptr();
+    int Numofthread = vertex_num_/64 < MAXTHREADS? vertex_num_/64:MAXTHREADS; //每线程处理64行
+    std::thread n[Numofthread]; //这样用变量定义数组大小居然是可以的
+    for (int k = 0; k < vertex_num_; ++k) {
+        for(int i = 0; i < Numofthread; i++)
+        {
+            n[i]=std::thread(funcyjy,k,Distance,vertex_num_,i,Numofthread);
+        }
+        for(int i = 0; i < Numofthread; i++)
+        {
+            n[i].join();
+        }
+
+
+    }
+    return result;
+}