remove "m_" from POD struct members in cube_atlas

This commit is contained in:
Jeremie Roy 2013-05-17 23:47:08 +02:00
parent 565d0ca543
commit be925b64c4
2 changed files with 54 additions and 54 deletions

View file

@ -51,16 +51,16 @@ private:
struct Node struct Node
{ {
Node(int16_t _x, int16_t _y, int16_t _width) : m_x(_x), m_y(_y), m_width(_width) Node(int16_t _x, int16_t _y, int16_t _width) : x(_x), y(_y), width(_width)
{ {
} }
/// The starting x-coordinate (leftmost). /// The starting x-coordinate (leftmost).
int16_t m_x; int16_t x;
/// The y-coordinate of the skyline level line. /// The y-coordinate of the skyline level line.
int16_t m_y; int16_t y;
/// The line _width. The ending coordinate (inclusive) will be x+width-1. /// The line _width. The ending coordinate (inclusive) will be x+width-1.
int32_t m_width; // 32bit to avoid padding int32_t width; // 32bit to avoid padding
}; };
/// width (in pixels) of the underlying texture /// width (in pixels) of the underlying texture
@ -120,12 +120,12 @@ bool RectanglePacker::addRectangle(uint16_t _width, uint16_t _height, uint16_t&
node = &m_skyline[ii]; node = &m_skyline[ii];
if ( ( (y + _height) < best_height) if ( ( (y + _height) < best_height)
|| ( ( (y + _height) == best_height) || ( ( (y + _height) == best_height)
&& (node->m_width < best_width) ) ) && (node->width < best_width) ) )
{ {
best_height = y + _height; best_height = y + _height;
best_index = ii; best_index = ii;
best_width = node->m_width; best_width = node->width;
_outX = node->m_x; _outX = node->x;
_outY = y; _outY = y;
} }
} }
@ -143,12 +143,12 @@ bool RectanglePacker::addRectangle(uint16_t _width, uint16_t _height, uint16_t&
{ {
node = &m_skyline[ii]; node = &m_skyline[ii];
prev = &m_skyline[ii - 1]; prev = &m_skyline[ii - 1];
if (node->m_x < (prev->m_x + prev->m_width) ) if (node->x < (prev->x + prev->width) )
{ {
int shrink = prev->m_x + prev->m_width - node->m_x; int shrink = prev->x + prev->width - node->x;
node->m_x += shrink; node->x += shrink;
node->m_width -= shrink; node->width -= shrink;
if (node->m_width <= 0) if (node->width <= 0)
{ {
m_skyline.erase(m_skyline.begin() + ii); m_skyline.erase(m_skyline.begin() + ii);
--ii; --ii;
@ -199,7 +199,7 @@ int32_t RectanglePacker::fit(uint32_t _skylineNodeIndex, uint16_t _width, uint16
const Node& baseNode = m_skyline[_skylineNodeIndex]; const Node& baseNode = m_skyline[_skylineNodeIndex];
int32_t x = baseNode.m_x, y; int32_t x = baseNode.x, y;
int32_t _width_left = width; int32_t _width_left = width;
int32_t i = _skylineNodeIndex; int32_t i = _skylineNodeIndex;
@ -208,13 +208,13 @@ int32_t RectanglePacker::fit(uint32_t _skylineNodeIndex, uint16_t _width, uint16
return -1; return -1;
} }
y = baseNode.m_y; y = baseNode.y;
while (_width_left > 0) while (_width_left > 0)
{ {
const Node& node = m_skyline[i]; const Node& node = m_skyline[i];
if (node.m_y > y) if (node.y > y)
{ {
y = node.m_y; y = node.y;
} }
if ( (y + height) > (int32_t)(m_height - 1) ) if ( (y + height) > (int32_t)(m_height - 1) )
@ -222,7 +222,7 @@ int32_t RectanglePacker::fit(uint32_t _skylineNodeIndex, uint16_t _width, uint16
return -1; return -1;
} }
_width_left -= node.m_width; _width_left -= node.width;
++i; ++i;
} }
@ -239,9 +239,9 @@ void RectanglePacker::merge()
{ {
node = (Node*) &m_skyline[ii]; node = (Node*) &m_skyline[ii];
next = (Node*) &m_skyline[ii + 1]; next = (Node*) &m_skyline[ii + 1];
if (node->m_y == next->m_y) if (node->y == next->y)
{ {
node->m_width += next->m_width; node->width += next->width;
m_skyline.erase(m_skyline.begin() + ii + 1); m_skyline.erase(m_skyline.begin() + ii + 1);
--ii; --ii;
} }
@ -375,10 +375,10 @@ uint16_t Atlas::addRegion(uint16_t _width, uint16_t _height, const uint8_t* _bit
//create new layers //create new layers
for (int ii = 0; ii < _type; ++ii) for (int ii = 0; ii < _type; ++ii)
{ {
m_layers[idx + ii].faceRegion.m_x = 0; m_layers[idx + ii].faceRegion.x = 0;
m_layers[idx + ii].faceRegion.m_y = 0; m_layers[idx + ii].faceRegion.y = 0;
m_layers[idx + ii].faceRegion.m_width = m_textureSize; m_layers[idx + ii].faceRegion.width = m_textureSize;
m_layers[idx + ii].faceRegion.m_height = m_textureSize; m_layers[idx + ii].faceRegion.height = m_textureSize;
m_layers[idx + ii].faceRegion.setMask(_type, m_usedFaces, ii); m_layers[idx + ii].faceRegion.setMask(_type, m_usedFaces, ii);
} }
@ -393,37 +393,37 @@ uint16_t Atlas::addRegion(uint16_t _width, uint16_t _height, const uint8_t* _bit
} }
AtlasRegion& region = m_regions[m_regionCount]; AtlasRegion& region = m_regions[m_regionCount];
region.m_x = x; region.x = x;
region.m_y = y; region.y = y;
region.m_width = _width; region.width = _width;
region.m_height = _height; region.height = _height;
region.m_mask = m_layers[idx].faceRegion.m_mask; region.mask = m_layers[idx].faceRegion.mask;
updateRegion(region, _bitmapBuffer); updateRegion(region, _bitmapBuffer);
region.m_x += outline; region.x += outline;
region.m_y += outline; region.y += outline;
region.m_width -= (outline * 2); region.width -= (outline * 2);
region.m_height -= (outline * 2); region.height -= (outline * 2);
return m_regionCount++; return m_regionCount++;
} }
void Atlas::updateRegion(const AtlasRegion& _region, const uint8_t* _bitmapBuffer) void Atlas::updateRegion(const AtlasRegion& _region, const uint8_t* _bitmapBuffer)
{ {
const bgfx::Memory* mem = bgfx::alloc(_region.m_width * _region.m_height * 4); const bgfx::Memory* mem = bgfx::alloc(_region.width * _region.height * 4);
//BAD! //BAD!
memset(mem->data, 0, mem->size); memset(mem->data, 0, mem->size);
if (_region.getType() == AtlasRegion::TYPE_BGRA8) if (_region.getType() == AtlasRegion::TYPE_BGRA8)
{ {
const uint8_t* inLineBuffer = _bitmapBuffer; const uint8_t* inLineBuffer = _bitmapBuffer;
uint8_t* outLineBuffer = m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.m_y * m_textureSize) + _region.m_x) * 4); uint8_t* outLineBuffer = m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.y * m_textureSize) + _region.x) * 4);
//update the cpu buffer //update the cpu buffer
for (int yy = 0; yy < _region.m_height; ++yy) for (int yy = 0; yy < _region.height; ++yy)
{ {
memcpy(outLineBuffer, inLineBuffer, _region.m_width * 4); memcpy(outLineBuffer, inLineBuffer, _region.width * 4);
inLineBuffer += _region.m_width * 4; inLineBuffer += _region.width * 4;
outLineBuffer += m_textureSize * 4; outLineBuffer += m_textureSize * 4;
} }
@ -435,24 +435,24 @@ void Atlas::updateRegion(const AtlasRegion& _region, const uint8_t* _bitmapBuffe
uint32_t layer = _region.getComponentIndex(); uint32_t layer = _region.getComponentIndex();
//uint32_t face = _region.getFaceIndex(); //uint32_t face = _region.getFaceIndex();
const uint8_t* inLineBuffer = _bitmapBuffer; const uint8_t* inLineBuffer = _bitmapBuffer;
uint8_t* outLineBuffer = (m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.m_y * m_textureSize) + _region.m_x) * 4) ); uint8_t* outLineBuffer = (m_textureBuffer + _region.getFaceIndex() * (m_textureSize * m_textureSize * 4) + ( ( (_region.y * m_textureSize) + _region.x) * 4) );
//update the cpu buffer //update the cpu buffer
for (int yy = 0; yy < _region.m_height; ++yy) for (int yy = 0; yy < _region.height; ++yy)
{ {
for (int xx = 0; xx < _region.m_width; ++xx) for (int xx = 0; xx < _region.width; ++xx)
{ {
outLineBuffer[(xx * 4) + layer] = inLineBuffer[xx]; outLineBuffer[(xx * 4) + layer] = inLineBuffer[xx];
} }
//update the GPU buffer //update the GPU buffer
memcpy(mem->data + yy * _region.m_width * 4, outLineBuffer, _region.m_width * 4); memcpy(mem->data + yy * _region.width * 4, outLineBuffer, _region.width * 4);
inLineBuffer += _region.m_width; inLineBuffer += _region.width;
outLineBuffer += m_textureSize * 4; outLineBuffer += m_textureSize * 4;
} }
} }
bgfx::updateTextureCube(m_textureHandle, (uint8_t)_region.getFaceIndex(), 0, _region.m_x, _region.m_y, _region.m_width, _region.m_height, mem); bgfx::updateTextureCube(m_textureHandle, (uint8_t)_region.getFaceIndex(), 0, _region.x, _region.y, _region.width, _region.height, mem);
} }
void Atlas::packFaceLayerUV(uint32_t _idx, uint8_t* _vertexBuffer, uint32_t _offset, uint32_t _stride) void Atlas::packFaceLayerUV(uint32_t _idx, uint8_t* _vertexBuffer, uint32_t _offset, uint32_t _stride)
@ -472,10 +472,10 @@ void Atlas::packUV(const AtlasRegion& _region, uint8_t* _vertexBuffer, uint32_t
static const int16_t minVal = -32768; static const int16_t minVal = -32768;
static const int16_t maxVal = 32767; static const int16_t maxVal = 32767;
int16_t x0 = (int16_t)(_region.m_x * texMult) - 32767; int16_t x0 = (int16_t)(_region.x * texMult) - 32767;
int16_t y0 = (int16_t)(_region.m_y * texMult) - 32767; int16_t y0 = (int16_t)(_region.y * texMult) - 32767;
int16_t x1 = (int16_t)( (_region.m_x + _region.m_width) * texMult) - 32767; int16_t x1 = (int16_t)( (_region.x + _region.width) * texMult) - 32767;
int16_t y1 = (int16_t)( (_region.m_y + _region.m_height) * texMult) - 32767; int16_t y1 = (int16_t)( (_region.y + _region.height) * texMult) - 32767;
int16_t w = (int16_t) ( (32767.0f / 4.0f) * _region.getComponentIndex() ); int16_t w = (int16_t) ( (32767.0f / 4.0f) * _region.getComponentIndex() );
_vertexBuffer += _offset; _vertexBuffer += _offset;

View file

@ -25,25 +25,25 @@ struct AtlasRegion
TYPE_BGRA8 = 4 // 4 components TYPE_BGRA8 = 4 // 4 components
}; };
uint16_t m_x, m_y; uint16_t x, y;
uint16_t m_width, m_height; uint16_t width, height;
uint32_t m_mask; //encode the region type, the face index and the component index in case of a gray region uint32_t mask; //encode the region type, the face index and the component index in case of a gray region
Type getType() const Type getType() const
{ {
return (Type) ( (m_mask >> 0) & 0x0000000F); return (Type) ( (mask >> 0) & 0x0000000F);
} }
uint32_t getFaceIndex() const uint32_t getFaceIndex() const
{ {
return (m_mask >> 4) & 0x0000000F; return (mask >> 4) & 0x0000000F;
} }
uint32_t getComponentIndex() const uint32_t getComponentIndex() const
{ {
return (m_mask >> 8) & 0x0000000F; return (mask >> 8) & 0x0000000F;
} }
void setMask(Type _type, uint32_t _faceIndex, uint32_t _componentIndex) void setMask(Type _type, uint32_t _faceIndex, uint32_t _componentIndex)
{ {
m_mask = (_componentIndex << 8) + (_faceIndex << 4) + (uint32_t)_type; mask = (_componentIndex << 8) + (_faceIndex << 4) + (uint32_t)_type;
} }
}; };