Tables: Fixed unaligned accesses when using TableSetBgColor(ImGuiTableBgTarget_CellBg). (#3872)

ImSpanAllocator: Support for alignment.

docs/CHANGELOG.txt

@@ -51,6 +51,7 @@ Other Changes:
 - Window: Shrink close button hit-testing region when it covers an abnormally high portion of the window visible
   area (e.g. when window is collapsed + moved in a corner) to facilitate moving the window away. (#3825)
 - Window, Nav: Fixed crash when calling SetWindowFocus(NULL) as the time a new window appears. (#3865) [@nem0]
+- Tables: Fixed unaligned accesses when using TableSetBgColor(ImGuiTableBgTarget_CellBg). (#3872)
 - Added GetAllocatorFunctions() to facilitate sharing allocators accross DLL boundaries. (#3836)
 - ImFontAtlas: Added 'bool TexPixelsUseColors' output to help backend decide of underlying texture format. (#3369)
   This can currently only ever be set by the Freetype renderer.

imgui_internal.h

@@ -214,6 +214,7 @@ namespace ImStb
 #define IM_NEWLINE                      "\n"
 #endif
 #define IM_TABSIZE                      (4)
+#define IM_MEMALIGN(_OFF,_ALIGN)        (((_OFF) + (_ALIGN - 1)) & ~(_ALIGN - 1))               // Memory align e.g. IM_ALIGN(0,4)=0, IM_ALIGN(1,4)=4, IM_ALIGN(4,4)=4, IM_ALIGN(5,4)=8
 #define IM_F32_TO_INT8_UNBOUND(_VAL)    ((int)((_VAL) * 255.0f + ((_VAL)>=0 ? 0.5f : -0.5f)))   // Unsaturated, for display purpose
 #define IM_F32_TO_INT8_SAT(_VAL)        ((int)(ImSaturate(_VAL) * 255.0f + 0.5f))               // Saturated, always output 0..255
 #define IM_FLOOR(_VAL)                  ((float)(int)(_VAL))                                    // ImFloor() is not inlined in MSVC debug builds
@@ -545,20 +546,22 @@ struct ImSpan
 
 // Helper: ImSpanAllocator<>
 // Facilitate storing multiple chunks into a single large block (the "arena")
+// - Usage: call Reserve() N times, allocate GetArenaSizeInBytes() worth, pass it to SetArenaBasePtr(), call GetSpan() N times to retrieve the aligned ranges.
 template<int CHUNKS>
 struct ImSpanAllocator
 {
     char*   BasePtr;
-    int     TotalSize;
+    int     CurrOff;
-    int     CurrSpan;
+    int     CurrIdx;
     int     Offsets[CHUNKS];
+    int     Sizes[CHUNKS];
 
     ImSpanAllocator()                               { memset(this, 0, sizeof(*this)); }
-    inline void  ReserveBytes(int n, size_t sz)     { IM_ASSERT(n == CurrSpan && n < CHUNKS); IM_UNUSED(n); Offsets[CurrSpan++] = TotalSize; TotalSize += (int)sz; }
+    inline void  Reserve(int n, size_t sz, int a=4) { IM_ASSERT(n == CurrIdx && n < CHUNKS); CurrOff = IM_MEMALIGN(CurrOff, a); Offsets[n] = CurrOff; Sizes[n] = (int)sz; CurrIdx++; CurrOff += (int)sz; }
-    inline int   GetArenaSizeInBytes()              { return TotalSize; }
+    inline int   GetArenaSizeInBytes()              { return CurrOff; }
     inline void  SetArenaBasePtr(void* base_ptr)    { BasePtr = (char*)base_ptr; }
-    inline void* GetSpanPtrBegin(int n)             { IM_ASSERT(n >= 0 && n < CHUNKS && CurrSpan == CHUNKS); return (void*)(BasePtr + Offsets[n]); }
+    inline void* GetSpanPtrBegin(int n)             { IM_ASSERT(n >= 0 && n < CHUNKS && CurrIdx == CHUNKS); return (void*)(BasePtr + Offsets[n]); }
-    inline void* GetSpanPtrEnd(int n)               { IM_ASSERT(n >= 0 && n < CHUNKS && CurrSpan == CHUNKS); return (n + 1 < CHUNKS) ? BasePtr + Offsets[n + 1] : (void*)(BasePtr + TotalSize); }
+    inline void* GetSpanPtrEnd(int n)               { IM_ASSERT(n >= 0 && n < CHUNKS && CurrIdx == CHUNKS); return (void*)(BasePtr + Offsets[n] + Sizes[n]); }
     template<typename T>
     inline void  GetSpan(int n, ImSpan<T>* span)    { span->set((T*)GetSpanPtrBegin(n), (T*)GetSpanPtrEnd(n)); }
 };
@@ -592,7 +595,7 @@ struct IMGUI_API ImPool
 // Helper: ImChunkStream<>
 // Build and iterate a contiguous stream of variable-sized structures.
 // This is used by Settings to store persistent data while reducing allocation count.
-// We store the chunk size first, and align the final size on 4 bytes boundaries (this what the '(X + 3) & ~3' statement is for)
+// We store the chunk size first, and align the final size on 4 bytes boundaries.
 // The tedious/zealous amount of casting is to avoid -Wcast-align warnings.
 template<typename T>
 struct IMGUI_API ImChunkStream
@@ -602,7 +605,7 @@ struct IMGUI_API ImChunkStream
     void    clear()                     { Buf.clear(); }
     bool    empty() const               { return Buf.Size == 0; }
     int     size() const                { return Buf.Size; }
-    T*      alloc_chunk(size_t sz)      { size_t HDR_SZ = 4; sz = ((HDR_SZ + sz) + 3u) & ~3u; int off = Buf.Size; Buf.resize(off + (int)sz); ((int*)(void*)(Buf.Data + off))[0] = (int)sz; return (T*)(void*)(Buf.Data + off + (int)HDR_SZ); }
+    T*      alloc_chunk(size_t sz)      { size_t HDR_SZ = 4; sz = IM_MEMALIGN(HDR_SZ + sz, 4u); int off = Buf.Size; Buf.resize(off + (int)sz); ((int*)(void*)(Buf.Data + off))[0] = (int)sz; return (T*)(void*)(Buf.Data + off + (int)HDR_SZ); }
     T*      begin()                     { size_t HDR_SZ = 4; if (!Buf.Data) return NULL; return (T*)(void*)(Buf.Data + HDR_SZ); }
     T*      next_chunk(T* p)            { size_t HDR_SZ = 4; IM_ASSERT(p >= begin() && p < end()); p = (T*)(void*)((char*)(void*)p + chunk_size(p)); if (p == (T*)(void*)((char*)end() + HDR_SZ)) return (T*)0; IM_ASSERT(p < end()); return p; }
     int     chunk_size(const T* p)      { return ((const int*)p)[-1]; }

imgui_tables.cpp

@@ -538,9 +538,9 @@ void ImGui::TableBeginInitMemory(ImGuiTable* table, int columns_count)
 {
     // Allocate single buffer for our arrays
     ImSpanAllocator<3> span_allocator;
-    span_allocator.ReserveBytes(0, columns_count * sizeof(ImGuiTableColumn));
+    span_allocator.Reserve(0, columns_count * sizeof(ImGuiTableColumn));
-    span_allocator.ReserveBytes(1, columns_count * sizeof(ImGuiTableColumnIdx));
+    span_allocator.Reserve(1, columns_count * sizeof(ImGuiTableColumnIdx));
-    span_allocator.ReserveBytes(2, columns_count * sizeof(ImGuiTableCellData));
+    span_allocator.Reserve(2, columns_count * sizeof(ImGuiTableCellData), 4);
     table->RawData = IM_ALLOC(span_allocator.GetArenaSizeInBytes());
     memset(table->RawData, 0, span_allocator.GetArenaSizeInBytes());
     span_allocator.SetArenaBasePtr(table->RawData);