Implement RadixSort and sort by modification time and size.

lib/Base/Arena.cpp

@@ -254,3 +254,59 @@ struct Auto_Release {
     }
   }
 };
+
+// #FixedArena procedures:
+FixedArena* bootstrap_fixed_arena (s64 size, Allocator backing_allocator) {
+  push_allocator(backing_allocator);
+  Assert(size >= sizeof(FixedArena));
+  ArrayView<u8> memory = ArrayView<u8>(size);
+  FixedArena* result = (FixedArena*)memory.data;
+  result->memory = memory;
+  result->cursor = sizeof(FixedArena);
+  result->allocator = backing_allocator;
+  return result;
+}
+
+force_inline void destroy_arena (FixedArena* arena) {
+  Delete(arena->allocator, arena);
+}
+
+Allocator allocator (FixedArena* arena) {
+  return { fixed_arena_allocator_proc, arena };
+}
+
+void* fixed_arena_allocator_proc (Allocator_Mode mode, s64 requested_size, s64 old_size, void* old_memory, void* allocator_data) {
+  constexpr s64 DEFAULT_ALIGNMENT = 16; // maybe make this modifiable as part of FixedArena struct?
+  FixedArena* arena = (FixedArena*)allocator_data;
+  Assert(arena != nullptr);
+  
+  switch (mode) {
+    case Allocator_Mode::ALLOCATE: {
+      arena->cursor = Align<s64>(arena->cursor, DEFAULT_ALIGNMENT);
+      void* result = &arena->memory[arena->cursor];
+      arena->cursor += requested_size;
+      Assert(arena->cursor <= arena->memory.count);
+      return result;
+    } break;
+    case Allocator_Mode::RESIZE: {
+      arena->cursor = Align<s64>(arena->cursor, DEFAULT_ALIGNMENT);
+      void* result = &arena->memory[arena->cursor];
+      arena->cursor += requested_size;
+      Assert(arena->cursor <= arena->memory.count);
+      s64 size_to_copy = old_size < requested_size ? old_size : requested_size;
+      if (result && size_to_copy) { memcpy(result, old_memory, size_to_copy); }
+      return result;
+    } break;
+    case Allocator_Mode::DEALLOCATE: {
+      return nullptr; // unused
+    } break;
+    case Allocator_Mode::DETAILS: {
+      if (allocator_data == nullptr) {
+        return "fixed_arena_allocator_proc: data pointer is null!";
+      }
+      return "fixed_arena_allocator_proc: with valid data";
+    } break;
+  }
+  
+  return nullptr;
+}

lib/Base/Arena.h

@@ -139,3 +139,19 @@ struct Push_Alignment { // #rename to Arena_Push_Alignment?
 // Do this later:
 // arena_lock_pages :: (using arena: *Arena, start_address: *u8, byte_count: s64)
 // arena_unlock_pages :: (using arena: *Arena, start_address: *u8, byte_count: s64)
+
+// #FixedArena is a super simple arena where you allocate a fixed block up front (fully committed), 
+// and use it as-is. 
+// #NOTE: we can save space be always backing with a known allocator (e.g. GPAllocator()).
+struct FixedArena {
+  ArrayView<u8> memory;
+  s64 cursor;
+  Allocator allocator;
+};
+
+void* fixed_arena_allocator_proc (Allocator_Mode mode, s64 requested_size, s64 old_size, void* old_memory, void* allocator_data);
+
+// #FixedArena API
+FixedArena* bootstrap_fixed_arena (s64 size, Allocator backing_allocator);
+force_inline void destroy_arena (FixedArena* arena);
+Allocator allocator (FixedArena* arena);

lib/Base/Expandable_Arena.cpp

@@ -107,6 +107,7 @@ Allocator allocator (ExpandableArena* arena_ex) {
 }
 
 // #TODO: currently this keeps the final arena's memory. Fix this!
+// This is not implemented correctly! 
 void arena_reset_to (ExpandableArena* arena_ex, Arena* last_arena, u8* starting_point) {
   // going backwards from end of arena list
   

lib/Base/RadixSort.cpp

@@ -0,0 +1,128 @@
+struct RadixSort {
+  ArrayView<u32> ranks;
+  ArrayView<u32> ranks2;
+  Allocator allocator;
+  bool valid_ranks;
+};
+
+void radix_sort_init (RadixSort* r, u32 items_to_allocate) {
+  if (r->allocator.proc == nullptr) {
+    r->allocator = context_allocator();
+  }
+  push_allocator(r->allocator);
+  r->ranks = ArrayView<u32>(items_to_allocate);
+  r->ranks2 = ArrayView<u32>(items_to_allocate);
+  r->valid_ranks = false;
+}
+
+void radix_sort_free (RadixSort* r) {
+  Assert(r->allocator.proc != nullptr);
+  push_allocator(r->allocator);
+  array_free(r->ranks);
+  array_free(r->ranks2);
+}
+
+// RadixSort provides an array of indices in sorted order.
+u32 rank (RadixSort* r, s64 i) {
+  Assert(r != nullptr);
+#if ARRAY_ENABLE_BOUNDS_CHECKING
+  if (i < 0 || i >= r->ranks.count) { debug_break(); /*INDEX OOB*/ }
+#endif
+  return r->ranks[i];
+}
+
+template <typename T> void create_histograms (RadixSort* r, T* buffer, u32 count, u32* histogram) {
+  constexpr u32 bucket_count = sizeof(T);
+  // Init bucket pointers:
+  u32* h[bucket_count] = {};
+  for (u32 i = 0; i < bucket_count; i += 1) {
+    h[i] = histogram + (256 * i);
+  }
+  
+  // Build histogram:
+  u8* p  = (u8*)buffer;
+  u8* pe = (p + count * sizeof(T));
+  
+  while (p != pe) {
+    h[0][*p] += 1; p += 1;
+    if (bucket_count > 1) { // how to make compile time if?
+      h[1][*p] += 1; p += 1;
+      
+      if (bucket_count > 2) {
+        h[2][*p] += 1; p += 1;
+        h[3][*p] += 1; p += 1;
+                
+        if (bucket_count == 8) {
+          h[4][*p] += 1; p += 1;
+          h[5][*p] += 1; p += 1;
+          h[6][*p] += 1; p += 1;
+          h[7][*p] += 1; p += 1;  
+        }
+      }
+    }
+  }
+}
+
+template <typename T> void radix_sort (RadixSort* r, T* input, u32 count) {
+  constexpr u32 T_SIZE = sizeof(T);
+  // Allocate histograms & offsets on the stack:
+  u32 histogram [256 * T_SIZE] = {};
+  u32* link [256];
+  
+  create_histograms(r, input, count, histogram);
+  
+  // Radix sort, j is the pass number, (0 = LSB, P = MSB)
+  for (u32 j = 0; j < T_SIZE; j += 1) {
+    u32* h = &histogram[j * 256];
+    
+    u8* input_bytes = (u8*)input;
+    input_bytes += j; // Assumes little endian!
+    
+    if (h[input_bytes[0]] == count) {
+      continue;
+    }
+    
+    // Create offsets
+    link[0] = r->ranks2.data;
+    for (u32 i = 1; i < 256; i += 1) { // 1..255
+      link[i] = link[i-1] + h[i-1];
+    }
+    
+    // Perform Radix Sort
+    if (!r->valid_ranks) {
+      for (u32 i = 0; i < count; i += 1) {
+        *link[input_bytes[i*T_SIZE]] =  i;
+         link[input_bytes[i*T_SIZE]] += 1;
+      }
+      r->valid_ranks = true;
+    } else {
+      for (u32 i = 0; i < count; i += 1) {
+        u32 idx = r->ranks[i];
+        *link[input_bytes[idx*T_SIZE]] =  idx;
+         link[input_bytes[idx*T_SIZE]] += 1;
+      }
+    }
+    
+    // Swap pointers for next pass. Valid indices - the most recent ones - are in ranks after the swap.
+    ArrayView<u32> ranks2_temp = r->ranks2;
+    r->ranks2 = r->ranks;
+    r->ranks  = ranks2_temp;
+  }
+  
+  // All values were equal; generate linear ranks
+  if (!r->valid_ranks) {
+    for (u32 i = 0; i < count; i += 1) {
+      r->ranks[i] = i;
+      r->valid_ranks = true;
+    }
+  }
+}
+
+// NOTE: For a small number of elements it's more efficient to use insertion sort
+void radix_sort_u64 (RadixSort* r, u64* input, u32 count) {
+  if (input == nullptr || count == 0) return;
+  if (r->ranks.count == 0) {
+    radix_sort_init(r, count);
+  }
+  radix_sort(r, input, count);
+}

lib/OS/OS_Win32.cpp

@@ -1214,6 +1214,16 @@ string get_file_copy (ST_File_Enumeration* stfe, s64 index) {
   string file = {strlength, string_ptr};
   return copy_string(file);
 }
+
+string get_file_string_view (ST_File_Enumeration* stfe, s64 index) {
+  STFE_Results* r = &stfe->files;
+  s64 strlength = (*r->lengths)[index];
+  u32 offset    = (*r->offsets)[index];
+  u8* string_ptr = &r->strings->data[offset];
+  string file = {strlength, string_ptr};
+  return file;
+}
+
 string get_path_copy (ST_File_Enumeration* stfe, s64 index) {
   STFE_Results* r = &stfe->dirs;
   Assert(index >= 0 && index < count_paths(stfe));
@@ -1261,6 +1271,7 @@ s64 win32_file_enum_thread_proc (Thread* thread) {
       push_allocator(temp());
       auto_release_temp();
       // This needs to be null-terminated:
+      // #TODO: Replace this #LIFO array with an arena-backed FIFO stack (singly linked-list).
       string next_directory = copy_string(pop(paths_to_enumerate)); // LIFO. maybe not the best way?
       wstring wildcard_name = utf8_to_wide(format_string("%s\\*", next_directory.data));
       

lib_main.cpp

@@ -36,6 +36,7 @@
 
 #include "lib/Base/Arena.cpp"
 #include "lib/Base/String.cpp"
+#include "lib/Base/RadixSort.cpp"
 
 #include "lib/Base/Base_Thread_Context.cpp"
 #include "lib/Base/ErrorType.cpp"

src/Ex1.cpp

@@ -124,15 +124,37 @@ void Ex1_show_ntfs_workspace () { using namespace ImGui;
   }
 }
 
+// #Workspaces are FOR DEVELOPMENT ONLY.
 struct Ex1_Workspace {
   s32 path_select;
   s32 file_select;
+  
+  RadixSort file_size_radix;
+  RadixSort file_modtime_radix;
+  RadixSort dir_modtime_radix;
+  bool sort_completed;
+  
+  // Reordered strings:
+  ArrayView<string> files_sorted_by_size;
+  ArrayView<string> files_sorted_by_modtime;
 };
+
+void reorder_files_by_radix (RadixSort* r, ArrayView<string>* files, bool reverse_order=false) {
+  Timed_Block_Print("reorder_files_by_radix");
+  // Where are my source files!?
+  (*files) = ArrayView<string>(r->ranks.count);
+  for_each(f, (*files)) {
+    // (*files)[f] = get_file_copy(stfe, r->ranks[f]);
+    (*files)[f] = get_file_string_view(stfe, r->ranks[f]);
+  }
+}
+
 global Ex1_Workspace ex1w;
 
 void Ex1_show_enumeration_workspace () { using namespace ImGui;
   push_imgui_window("Enumerated Data Workspace");
-  // #TODO SliderInt for each 
+
+  /*
   SliderInt("Select path index", &ex1w.path_select, 0, count_paths(stfe)-1);
   Text("%s", get_path_copy(stfe, ex1w.path_select).data);
   Text("time modified: %s", format_time_datetime(get_path_modtime(stfe, ex1w.path_select)).data);
@@ -141,7 +163,44 @@ void Ex1_show_enumeration_workspace () { using namespace ImGui;
   Text("%s", get_file_copy(stfe, ex1w.file_select).data);
   Text("size: %s", format_bytes(get_file_size_bytes(stfe, ex1w.file_select)).data);
   Text("time modified: %s", format_time_datetime(get_file_modtime(stfe, ex1w.file_select)).data);
+  */
   // #TODO: size, modtime
+  
+  if (!ex1w.sort_completed || Button("sort file sizes")) {
+    push_allocator(GPAllocator());
+    Timed_Block_Print("radix_sort_u64: file sizes, file modtimes, directory modtimes");
+    ArrayView<u64> sizes = to_view(*stfe->files.sizes);
+    radix_sort_u64(&ex1w.file_size_radix, sizes.data, (u32)sizes.count);
+    ArrayView<u64> file_modtimes = to_view(*stfe->files.modtimes);
+    radix_sort_u64(&ex1w.file_modtime_radix, file_modtimes.data, (u32)file_modtimes.count);
+    ArrayView<u64> dirs_modtimes = to_view(*stfe->dirs.modtimes);
+    radix_sort_u64(&ex1w.dir_modtime_radix, dirs_modtimes.data, (u32)dirs_modtimes.count);
+    // Create ArrayView<string>, ArrayView<u64> sizes, and ArrayView<u64> modtimes
+    reorder_files_by_radix(&ex1w.file_size_radix, &ex1w.files_sorted_by_size);
+    reorder_files_by_radix(&ex1w.file_modtime_radix, &ex1w.files_sorted_by_modtime);
+    // reordering by the rank permutations generated by RadixSort.
+    ex1w.sort_completed = true;
+  }
+  
+  if (ex1w.sort_completed) {
+    SeparatorText("Files ordered by modtime");
+    s32 file_count = (s32)ex1w.files_sorted_by_modtime.count;
+    SliderInt("Select file index", &ex1w.file_select, 0, file_count-1);
+    string file_name = copy_string(ex1w.files_sorted_by_modtime[ex1w.file_select]);
+    Text("%s", file_name.data);
+    u32 radix_index = rank(&ex1w.file_modtime_radix, ex1w.file_select);
+    Text("date modified: %s", format_time_datetime(get_file_modtime(stfe, radix_index)).data);
+  }
+  if (ex1w.sort_completed) {
+    SeparatorText("Files ordered by size");
+    s32 file_count = (s32)ex1w.files_sorted_by_size.count;
+    // SliderInt("Select file index", &ex1w.file_select, 0, file_count-1);
+    string file_name = copy_string(ex1w.files_sorted_by_size[ex1w.file_select]);
+    Text("%s", file_name.data);
+    u32 radix_index = rank(&ex1w.file_size_radix, ex1w.file_select);
+    Text("size: %s", format_bytes(get_file_size_bytes(stfe, radix_index)).data);
+    // Text("date modified: %s", format_time_datetime(get_file_modtime(stfe, radix_index)).data);
+  }
 }
 
 void Ex1_Control_Panel () { using namespace ImGui;
@@ -195,9 +254,9 @@ void Ex1_Control_Panel () { using namespace ImGui;
   //   && (drives_enumerated == drives.count);
   
   string file_path = format_string_temp("%s_DriveData.bin", os_get_machine_name().data);
-  if (!all_drives_enumerated && file_exists(file_path)) {
-    Deserialize_ST_File_Enumeration(file_path);
-  }
+  // if (!all_drives_enumerated && file_exists(file_path)) {
+  //   Deserialize_ST_File_Enumeration(file_path);
+  // }
   if (drives.count > 0 && !all_drives_enumerated && file_exists(file_path) && Button("Load from file (this machine)")) {
     Deserialize_ST_File_Enumeration(file_path);
   //   Deserialize_Win32_Drives(file_path);
@@ -368,12 +427,12 @@ void ImGui_Debug_Panel () { using namespace ImGui;
   push_allocator(temp());
   
   Begin("Debug Panel");
-  SeparatorText("ex1_ntfs");
+  /*SeparatorText("ex1_ntfs");
   Text("Threads in flight count: %d", ex1_ntfs.threads_in_flight.count);
   for_each(i, ex1_ntfs.threads) {
     Text(" [%d] initialized: %d, has_context: %d, has_data: %d", 
       i, ex1_ntfs.threads[i].proc != nullptr, ex1_ntfs.threads[i].context != nullptr, ex1_ntfs.threads[i].data != nullptr);
-  }
+  }*/
   // #cpuid
   // Text("[cpus] physical: %d, logical: %d, primary: %d, secondary: %d", os_cpu_physical_core_count(), os_cpu_logical_core_count(), os_cpu_primary_core_count(), os_cpu_secondary_core_count());
   { SeparatorText("Arena In-Use List");