Improve texture clamping and mirroring

src/rdp.cpp

@@ -45,12 +45,12 @@ enum CycleType
 
 struct Tile
 {
- uint16_t sBase;
+ uint16_t s1, s2;
  uint16_t sMask;
  bool sMirror;
  bool sClamp;
 
- uint16_t tBase;
+ uint16_t t1, t2;
  uint16_t tMask;
  bool tMirror;
  bool tClamp;
@@ -365,8 +365,8 @@ inline uint32_t RDP::colorToAlpha(uint32_t color)
 uint32_t RDP::getTexel(Tile &tile, int s, int t, bool rect)
 {
  // Offset the texture coordinates relative to the tile
- s -= tile.sBase;
- t -= tile.tBase;
+ s -= tile.s1;
+ t -= tile.t1;
 
  // Fall back to nearest sampling, or filter based on https://www.shadertoy.com/view/Ws2fWV
  if (!Settings::texFilter || !texFilter || cycleType >= COPY_MODE)
@@ -407,20 +407,14 @@ uint32_t RDP::getTexel(Tile &tile, int s, int t, bool rect)
 uint32_t RDP::getRawTexel(Tile &tile, int s, int t)
 {
  // Clamp, mirror, or mask the S-coordinate based on tile settings
- if (tile.sClamp)
- s = std::max<int>(std::min<int>(s, tile.sMask), 0);
- else if (tile.sMirror)
- s = ((s & (tile.sMask + 1)) ? ~s : s) & tile.sMask;
- else
- s &= tile.sMask;
+ if (tile.sClamp) s = std::max<int>(std::min<int>(s, (tile.s2 - tile.s1) >> 5), 0);
+ if (tile.sMirror && (s & (tile.sMask + 1))) s = ~s;
+ s &= tile.sMask;
 
  // Clamp, mirror, or mask the T-coordinate based on tile settings
- if (tile.tClamp)
- t = std::max<int>(std::min<int>(t, tile.tMask), 0);
- else if (tile.tMirror)
- t = ((t & (tile.tMask + 1)) ? ~t : t) & tile.tMask;
- else
- t &= tile.tMask;
+ if (tile.tClamp) t = std::max<int>(std::min<int>(t, (tile.t2 - tile.t1) >> 5), 0);
+ if (tile.tMirror && (t & (tile.tMask + 1))) t = ~t;
+ t &= tile.tMask;
 
  // Get an RGBA32 texel from a tile at the given coordinates
  switch (tile.format)
@@ -1438,40 +1432,48 @@ void RDP::setOtherModes()
 
 void RDP::loadTlut()
 {
- // Decode the operands
+ // Decode the operands and set texture coordinate bounds
  Tile &tile = tiles[(opcode[0] >> 24) & 0x7];
- uint16_t indexL = ((opcode[0] >> 44) & 0xFFF) >> 1;
- uint16_t indexH = ((opcode[0] >> 12) & 0xFFF) >> 1;
-
- // Copy 16-bit texture lookup values into TMEM
- for (int i = indexL; i <= indexH; i += 2)
+ uint16_t s1 = (tile.s1 = ((opcode[0] >> 44) & 0xFFF) << 3) >> 4;
+ uint16_t t1 = (tile.t1 = ((opcode[0] >> 32) & 0xFFF) << 3) >> 4;
+ uint16_t s2 = (tile.s2 = ((opcode[0] >> 12) & 0xFFF) << 3) >> 4;
+ uint16_t t2 = (tile.t2 = ((opcode[0] >> 0) & 0xFFF) << 3) >> 4;
+
+ // Copy 16-bit texture lookup values into TMEM, duplicated 4 times
+ // TODO: actually use T-coordinates?
+ for (int s = s1; s <= s2; s += 2)
  {
- uint16_t src = Memory::read<uint16_t>(texAddress + i);
- uint8_t *dst = &tmem[(tile.address + i * 4) & 0xFF8];
- dst[0] = src >> 8;
- dst[1] = src >> 0;
+ uint16_t src = Memory::read<uint16_t>(texAddress + s);
+ uint8_t *dst = &tmem[(tile.address + s * 4) & 0xFF8];
+ for (int i = 0; i < 8; i += 2)
+ {
+ dst[i + 0] = src >> 8;
+ dst[i + 1] = src >> 0;
+ }
  }
 }
 
 void RDP::setTileSize()
 {
- // Set the low texture coordinates for reference
- // TODO: use the high coordinates for something?
+ // Set the texture coordinate bounds
  Tile &tile = tiles[(opcode[0] >> 24) & 0x7];
- tile.sBase = ((opcode[0] >> 44) & 0xFFF) << 3;
- tile.tBase = ((opcode[0] >> 32) & 0xFFF) << 3;
+ tile.s1 = ((opcode[0] >> 44) & 0xFFF) << 3;
+ tile.t1 = ((opcode[0] >> 32) & 0xFFF) << 3;
+ tile.s2 = ((opcode[0] >> 12) & 0xFFF) << 3;
+ tile.t2 = ((opcode[0] >> 0) & 0xFFF) << 3;
 }
 
 void RDP::loadBlock()
 {
- // Decode the operands
+ // Decode the operands and set texture coordinate bounds
  Tile &tile = tiles[(opcode[0] >> 24) & 0x7];
- uint16_t count = ((opcode[0] >> 12) & 0xFFF);
- uint16_t dxt = (opcode[0] & 0xFFF);
-
- // Adjust the byte count based on the texel size
- count = (count << 2) >> (~texFormat & 0x3);
+ uint16_t s1 = (tile.s1 = ((opcode[0] >> 44) & 0xFFF) << 3) >> 1;
+ uint16_t t1 = (tile.t1 = ((opcode[0] >> 32) & 0xFFF) << 3) >> 1;
+ uint16_t s2 = (tile.s2 = ((opcode[0] >> 12) & 0xFFF) << 3) >> 1;
+ uint16_t dxt = (tile.t2 = ((opcode[0] >> 0) & 0xFFF) << 3) >> 3;
 
+ // Adjust the byte count based on texel size
+ uint16_t count = (s2 - s1) >> (~texFormat & 0x3);
  uint16_t d = 0;
  bool odd = false;
 
@@ -1523,16 +1525,12 @@ void RDP::loadBlock()
 
 void RDP::loadTile()
 {
- // Decode the operands
+ // Decode the operands and set texture coordinate bounds
  Tile &tile = tiles[(opcode[0] >> 24) & 0x7];
- uint16_t t2 = ((opcode[0] >> 0) & 0xFFF) >> 2;
- uint16_t s2 = ((opcode[0] >> 12) & 0xFFF) >> 2;
- uint16_t t1 = ((opcode[0] >> 32) & 0xFFF) >> 2;
- uint16_t s1 = ((opcode[0] >> 44) & 0xFFF) >> 2;
-
- // Save the low texture coordinates for reference
- tile.sBase = ((opcode[0] >> 44) & 0xFFF) << 3;
- tile.tBase = ((opcode[0] >> 32) & 0xFFF) << 3;
+ uint16_t s1 = (tile.s1 = ((opcode[0] >> 44) & 0xFFF) << 3) >> 5;
+ uint16_t t1 = (tile.t1 = ((opcode[0] >> 32) & 0xFFF) << 3) >> 5;
+ uint16_t s2 = (tile.s2 = ((opcode[0] >> 12) & 0xFFF) << 3) >> 5;
+ uint16_t t2 = (tile.t2 = ((opcode[0] >> 0) & 0xFFF) << 3) >> 5;
 
  // Only support loading textures without conversion for now
  if (texFormat != tile.format)