contrek 1.2.7 → 1.2.8

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: ae46fa52ddce3383ef534142814ed28ddc5cae45042e677c8f445140bee67e82
-  data.tar.gz: 57cbaf53a0d6e8c7d31bdfcef0718c94654ceaf0fd7b2be84325751dc29f6224
+  metadata.gz: d6702682d1ddd30acbdc45a4344b1d761473afc336b7c07734f169739228f855
+  data.tar.gz: 8fde7e85a08f0e8037b7711b303fefb0d4edcff1f7c9725a767ff09f815445a6
 SHA512:
-  metadata.gz: ee922b6b787ac0f91b98a2e7d608c351499ea5405e325a0eb6d23f751a29075b390c08d757824132c7fbef7d64cef9f3af7dbb11709506e84777ff652f9f15fc
-  data.tar.gz: 848102a6980857b1f4d3914e60de04408d0f7a4cc4c950a20b3fa8ef7231e31db029220eb6e3fb71847b6986a09c0d7b1fedae70f8e30e8ef758a3adeabeddcd
+  metadata.gz: f498831ee81c1ff533ce412b39efdfe9c9dc42ff40679749600a2e2cf833cd8193b1f3ca8bf00e8ff98ae3a3798f9670380274d3cf63e485bad4428200256af4
+  data.tar.gz: 4fd128cf5cd14d929f2baaf9c611708cd9bd8ce01550e267f1d93252f5f23a9012b6f66bb59a4ab4390cb02c38b31a37f3fc6fd56d22478cf046359834a3bf0d

data/ext/cpp_polygon_finder/PolygonFinder/CMakeLists.txt

@@ -11,8 +11,10 @@ if(CMAKE_BUILD_TYPE STREQUAL "Debug")
     list(FILTER CMAKE_CXX_FLAGS EXCLUDE REGEX "-DNDEBUG")
     list(FILTER CMAKE_C_FLAGS EXCLUDE REGEX "-DNDEBUG")
 else()
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra -pthread -march=native -DNDEBUG -Ofast -flto")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3 -march=native -fPIC -DNDEBUG")
+    #set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra -pthread -march=native -DNDEBUG -Ofast -flto")
+    #set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O3 -march=native -fPIC -DNDEBUG")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall -Wextra -pthread -march=native -DNDEBUG -Ofast -flto -fopenmp-simd -ftree-vectorize")
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall -Wextra -pthread -march=native -DNDEBUG -Ofast -flto -fopenmp-simd -ftree-vectorize")
     set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} -pthread -flto=auto -Wl,--no-as-needed")
     find_library(TCMALLOC_LIB tcmalloc)
     if(TCMALLOC_LIB)

data/ext/cpp_polygon_finder/PolygonFinder/src/polygon/finder/PolygonFinder.cpp

@@ -46,6 +46,7 @@ PolygonFinder::PolygonFinder(Bitmap *bitmap,
   cpu_timer.start();
   scan();
   reports["scan"] = cpu_timer.stop();
+  //std::cout << "scan " << reports["scan"] << std::endl;
   //=====================//
 
   //= BUILD_TANGS_SEQUENCE ===//

data/ext/cpp_polygon_finder/PolygonFinder/src/polygon/finder/PolygonFinder.h

@@ -20,6 +20,8 @@
 #include <sstream>
 #include <stdexcept>
 #include <limits>
+#include <cstdint>
+
 #include "../bitmaps/Bitmap.h"
 #include "NodeCluster.h"
 #include "Node.h"
@@ -205,7 +207,7 @@ class PolygonFinder {
   std::map<std::string, double> reports;
   void scan();
   CpuTimer cpu_timer;
-
+/*
   template <typename M, typename F>
   void run_loop(M* specific_matcher, F&& fetch_color, int offset) {
     int img_h = this->source_bitmap->h();
@@ -236,7 +238,299 @@ class PolygonFinder {
         }
       }
     }
-  }
+  }*/
+
+template <typename M, typename F>
+void run_loop(M* specific_matcher, F&& fetch_color, int offset) {
+    int img_h = this->source_bitmap->h();
+    int bpp = this->source_bitmap->get_bytes_per_pixel();
+
+    for (int y = 0; y < img_h; y++) {
+        const unsigned char* row_ptr = this->source_bitmap->get_row_ptr(y);
+        const unsigned char* p = row_ptr + (this->start_x * bpp);
+        
+        int min_x = 0;
+        bool matching = false;
+        unsigned char last_red_value = 0;
+        
+        int x = this->start_x;
+
+        if (bpp == 4) {
+            for (; x <= this->end_x - 4; x += 4) {
+                // 1. Unico accesso alla RAM per pixel: leggiamo i 4 byte completi
+                unsigned int c0 = fetch_color(p);
+                unsigned int c1 = fetch_color(p + 4);
+                unsigned int c2 = fetch_color(p + 8);
+                unsigned int c3 = fetch_color(p + 12);
+
+                // 2. Estrazione a costo zero dai registri della CPU (Operazione bitwise, NO RAM)
+                // Se fetch_color ritorna il mascheramento standard, il primo byte si prende con il cast o lo shift
+                unsigned char v0 = static_cast<unsigned char>(c0); 
+                unsigned char v1 = static_cast<unsigned char>(c1);
+                unsigned char v2 = static_cast<unsigned char>(c2);
+                unsigned char v3 = static_cast<unsigned char>(c3);
+
+                p += 16; 
+
+                bool m0 = specific_matcher->match(c0);
+                bool m1 = specific_matcher->match(c1);
+                bool m2 = specific_matcher->match(c2);
+                bool m3 = specific_matcher->match(c3);
+
+                // [Resto della logica dei pixel 0, 1, 2, 3 invariata...]
+                if (m0) {
+                    if (!matching) { min_x = x; last_red_value = v0; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                if (m1) {
+                    if (!matching) { min_x = x + 1; last_red_value = v1; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                if (m2) {
+                    if (!matching) { min_x = x + 2; last_red_value = v2; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                if (m3) {
+                    if (!matching) { min_x = x + 3; last_red_value = v3; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 2, y, last_red_value, offset);
+                    matching = false;
+                }
+            }
+        }
+
+        // --- CLEANUP SCALARE ---
+        for (; x < this->end_x; x++) {
+            unsigned int color = fetch_color(p);
+            // Applichiamo la stessa ottimizzazione anche qui
+            unsigned char current_val = static_cast<unsigned char>(color);
+            p += bpp;
+            
+            if (specific_matcher->match(color)) {
+                if (!matching) {
+                    min_x = x;
+                    last_red_value = current_val;
+                    matching = true;
+                }
+            } else if (matching) {
+                this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                matching = false;
+            }
+        }
+
+        if (matching) {
+            this->node_cluster->add_node(min_x, this->end_x - 1, y, last_red_value, offset);
+        }
+    }
+}
+
+ /*template <typename M, typename F>
+void run_loop(M* specific_matcher, F&& fetch_color, int offset) {
+    int img_h = this->source_bitmap->h();
+    int bpp = this->source_bitmap->get_bytes_per_pixel();
+
+    for (int y = 0; y < img_h; y++) {
+        const unsigned char* row_ptr = this->source_bitmap->get_row_ptr(y);
+        const unsigned char* p = row_ptr + (this->start_x * bpp);
+        
+        int min_x = 0;
+        bool matching = false;
+        unsigned char last_red_value = 0;
+        
+        int x = this->start_x;
+
+        // --- IL TUO SWEET SPOT: UNROLLING A 4 VIE RIFINITO ---
+        if (bpp == 4) {
+            for (; x <= this->end_x - 4; x += 4) {
+                // Leggiamo in sequenza stretta per ottimizzare i register e la cache L1
+                unsigned int c0 = fetch_color(p);
+                unsigned char v0 = p[0];
+                
+                unsigned int c1 = fetch_color(p + 4);
+                unsigned char v1 = p[4];
+                
+                unsigned int c2 = fetch_color(p + 8);
+                unsigned char v2 = p[8];
+                
+                unsigned int c3 = fetch_color(p + 12);
+                unsigned char v3 = p[12];
+
+                p += 16; // Spostato qui: avanza di 4 pixel RGBA precisi
+
+                bool m0 = specific_matcher->match(c0);
+                bool m1 = specific_matcher->match(c1);
+                bool m2 = specific_matcher->match(c2);
+                bool m3 = specific_matcher->match(c3);
+
+                // Pixel 0
+                if (m0) {
+                    if (!matching) { min_x = x; last_red_value = v0; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 1
+                if (m1) {
+                    if (!matching) { min_x = x + 1; last_red_value = v1; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 2
+                if (m2) {
+                    if (!matching) { min_x = x + 2; last_red_value = v2; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 3
+                if (m3) {
+                    if (!matching) { min_x = x + 3; last_red_value = v3; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 2, y, last_red_value, offset);
+                    matching = false;
+                }
+            }
+        }
+
+        // --- CLEANUP SCALARE PER I RIMANENTI ---
+        for (; x < this->end_x; x++) {
+            unsigned int color = fetch_color(p);
+            unsigned char current_val = p[0];
+            p += bpp;
+            
+            if (specific_matcher->match(color)) {
+                if (!matching) {
+                    min_x = x;
+                    last_red_value = current_val;
+                    matching = true;
+                }
+            } else if (matching) {
+                this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                matching = false;
+            }
+        }
+
+        // --- CHIUSURA DI SICUREZZA FINE RIGA ---
+        if (matching) {
+            this->node_cluster->add_node(min_x, this->end_x - 1, y, last_red_value, offset);
+        }
+    }
+}*/
+
+
+
+/*
+
+  #include <cstdint>
+
+template <typename M, typename F>
+void run_loop(M* specific_matcher, F&& fetch_color, int offset) {
+    int img_h = this->source_bitmap->h();
+    int bpp = this->source_bitmap->get_bytes_per_pixel();
+
+    for (int y = 0; y < img_h; y++) {
+        const unsigned char* row_ptr = this->source_bitmap->get_row_ptr(y);
+        const unsigned char* p = row_ptr + (this->start_x * bpp);
+        
+        int min_x = 0;
+        bool matching = false;
+        unsigned char last_red_value = 0;
+        
+        int x = this->start_x;
+
+        // --- LINEA 1: UNROLLING A 4 VIE (SIMD FRIENDLY, NO DEPENDENCY) ---
+        if (bpp == 4) {
+            for (; x <= this->end_x - 4; x += 4) {
+                // Carichiamo 4 pixel in variabili separate (zero dipendenze tra loro)
+                unsigned int c0 = fetch_color(p);
+                unsigned int c1 = fetch_color(p + 4);
+                unsigned int c2 = fetch_color(p + 8);
+                unsigned int c3 = fetch_color(p + 12);
+
+                unsigned char v0 = p[0];
+                unsigned char v1 = p[4];
+                unsigned char v2 = p[8];
+                unsigned char v3 = p[12];
+
+                p += 16; // Avanza di 4 pixel (4 * 4 byte)
+
+                // Eseguiamo il match in parallelo. Il compilatore trasforma questo in istruzioni vettoriali
+                bool m0 = specific_matcher->match(c0);
+                bool m1 = specific_matcher->match(c1);
+                bool m2 = specific_matcher->match(c2);
+                bool m3 = specific_matcher->match(c3);
+
+                // Pixel 0
+                if (m0) {
+                    if (!matching) { min_x = x; last_red_value = v0; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 1
+                if (m1) {
+                    if (!matching) { min_x = x + 1; last_red_value = v1; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 2
+                if (m2) {
+                    if (!matching) { min_x = x + 2; last_red_value = v2; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 1, y, last_red_value, offset);
+                    matching = false;
+                }
+
+                // Pixel 3
+                if (m3) {
+                    if (!matching) { min_x = x + 3; last_red_value = v3; matching = true; }
+                } else if (matching) {
+                    this->node_cluster->add_node(min_x, x + 2, y, last_red_value, offset);
+                    matching = false;
+                }
+            }
+        }
+
+        // --- LINEA 2: SCALAR CLEANUP ---
+        for (; x < this->end_x; x++) {
+            unsigned int color = fetch_color(p);
+            unsigned char current_val = p[0];
+            p += bpp;
+            
+            if (specific_matcher->match(color)) {
+                if (!matching) {
+                    min_x = x;
+                    last_red_value = current_val;
+                    matching = true;
+                }
+            } else if (matching) {
+                this->node_cluster->add_node(min_x, x - 1, y, last_red_value, offset);
+                matching = false;
+            }
+        }
+
+        // --- CHIUSURA DI SICUREZZA ---
+        if (matching) {
+            this->node_cluster->add_node(min_x, this->end_x - 1, y, last_red_value, offset);
+        }
+    }
+}*/
 
  public:
   PolygonFinder(Bitmap *bitmap,

data/lib/contrek/version.rb

@@ -1,3 +1,3 @@
 module Contrek
-  VERSION = "1.2.7"
+  VERSION = "1.2.8"
 end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: contrek
 version: !ruby/object:Gem::Version
-  version: 1.2.7
+  version: 1.2.8
 platform: ruby
 authors:
 - Emanuele Cesaroni
-autorequire: 
+autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-06-02 00:00:00.000000000 Z
+date: 2026-06-06 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: rspec
@@ -301,7 +301,7 @@ metadata:
   homepage_uri: https://github.com/runout77/contrek
   documentation_uri: https://github.com/runout77/contrek#readme
   changelog_uri: https://github.com/runout77/contrek/blob/main/CHANGELOG.md
-post_install_message: 
+post_install_message:
 rdoc_options: []
 require_paths:
 - lib
@@ -317,7 +317,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubygems_version: 3.5.22
-signing_key: 
+signing_key:
 specification_version: 4
 summary: Fast PNG contour tracing and shape detection for Ruby
 test_files: []