N-archive/citra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

GPU: Bitwise texture swizzling

Browse source 
Replace the loop-based texture address swizzling code by a bit-twiddling
implementation, providing a very small speed up. Also simplify
addressing code.
This commit is contained in:
Yuri Kunde Schlesner 2014-12-23 12:27:56 -02:00
parent 2012e1420f
commit fe186d3a59
1 changed files with 23 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

49
src/video_core/debug_utils/debug_utils.cpp
View file

@ -304,7 +304,6 @@ std::unique_ptr<PicaTrace> FinishPicaTracing()
} }
const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info, bool disable_alpha) { const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const TextureInfo& info, bool disable_alpha) {
// Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each // Images are split into 8x8 tiles. Each tile is composed of four 4x4 subtiles each
// of which is composed of four 2x2 subtiles each of which is composed of four texels. // of which is composed of four 2x2 subtiles each of which is composed of four texels.
// Each structure is embedded into the next-bigger one in a diagonal pattern, e.g. // Each structure is embedded into the next-bigger one in a diagonal pattern, e.g.
@ -323,41 +322,39 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
// 02 03 06 07 18 19 22 23 // 02 03 06 07 18 19 22 23
// 00 01 04 05 16 17 20 21 // 00 01 04 05 16 17 20 21
// TODO(neobrain): Not sure if this swizzling pattern is used for all textures. const unsigned int block_width = 8;
// To be flexible in case different but similar patterns are used, we keep this const unsigned int block_height = 8;
// somewhat inefficient code around for now.
int texel_index_within_tile = 0;
for (int block_size_index = 0; block_size_index < 3; ++block_size_index) {
int sub_tile_width = 1 << block_size_index;
int sub_tile_height = 1 << block_size_index;
int sub_tile_index = (x & sub_tile_width) << block_size_index; const unsigned int coarse_x = x & ~7;
sub_tile_index += 2 * ((y & sub_tile_height) << block_size_index); const unsigned int coarse_y = y & ~7;
texel_index_within_tile += sub_tile_index;
}
const int block_width = 8; // Interleave the lower 3 bits of each coordinate to get the intra-block offsets, which are
const int block_height = 8; // arranged in a Z-order curve. More details on the bit manipulation at:
// https://fgiesen.wordpress.com/2009/12/13/decoding-morton-codes/
unsigned int i = (x | (y << 8)) & 0x0707; // ---- -210
i = (i ^ (i << 2)) & 0x1313; // ---2 --10
i = (i ^ (i << 1)) & 0x1515; // ---2 -1-0
i = (i | (i >> 7)) & 0x3F;
int coarse_x = (x / block_width) * block_width; source += coarse_y * info.stride;
int coarse_y = (y / block_height) * block_height; const unsigned int offset = coarse_x * block_height + i;
switch (info.format) { switch (info.format) {
case Regs::TextureFormat::RGBA8: case Regs::TextureFormat::RGBA8:
{ {
const u8* source_ptr = source + coarse_x * block_height * 4 + coarse_y * info.stride + texel_index_within_tile * 4; const u8* source_ptr = source + offset * 4;
return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? (u8)255 : source_ptr[0] }; return { source_ptr[3], source_ptr[2], source_ptr[1], disable_alpha ? (u8)255 : source_ptr[0] };
} }
case Regs::TextureFormat::RGB8: case Regs::TextureFormat::RGB8:
{ {
const u8* source_ptr = source + coarse_x * block_height * 3 + coarse_y * info.stride + texel_index_within_tile * 3; const u8* source_ptr = source + offset * 3;
return { source_ptr[2], source_ptr[1], source_ptr[0], 255 }; return { source_ptr[2], source_ptr[1], source_ptr[0], 255 };
} }
case Regs::TextureFormat::RGBA5551: case Regs::TextureFormat::RGBA5551:
{ {
const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2); const u16 source_ptr = *(const u16*)(source + offset * 2);
u8 r = (source_ptr >> 11) & 0x1F; u8 r = (source_ptr >> 11) & 0x1F;
u8 g = ((source_ptr) >> 6) & 0x1F; u8 g = ((source_ptr) >> 6) & 0x1F;
u8 b = (source_ptr >> 1) & 0x1F; u8 b = (source_ptr >> 1) & 0x1F;
@ -367,7 +364,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::RGB565: case Regs::TextureFormat::RGB565:
{ {
const u16 source_ptr = *(const u16*)(source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2); const u16 source_ptr = *(const u16*)(source + offset * 2);
u8 r = (source_ptr >> 11) & 0x1F; u8 r = (source_ptr >> 11) & 0x1F;
u8 g = ((source_ptr) >> 5) & 0x3F; u8 g = ((source_ptr) >> 5) & 0x3F;
u8 b = (source_ptr) & 0x1F; u8 b = (source_ptr) & 0x1F;
@ -376,7 +373,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::RGBA4: case Regs::TextureFormat::RGBA4:
{ {
const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2; const u8* source_ptr = source + offset * 2;
u8 r = source_ptr[1] >> 4; u8 r = source_ptr[1] >> 4;
u8 g = source_ptr[1] & 0xFF; u8 g = source_ptr[1] & 0xFF;
u8 b = source_ptr[0] >> 4; u8 b = source_ptr[0] >> 4;
@ -390,7 +387,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::IA8: case Regs::TextureFormat::IA8:
{ {
const u8* source_ptr = source + coarse_x * block_height * 2 + coarse_y * info.stride + texel_index_within_tile * 2; const u8* source_ptr = source + offset * 2;
// TODO: component order not verified // TODO: component order not verified
@ -404,13 +401,13 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::I8: case Regs::TextureFormat::I8:
{ {
const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile; const u8* source_ptr = source + offset;
return { *source_ptr, *source_ptr, *source_ptr, 255 }; return { *source_ptr, *source_ptr, *source_ptr, 255 };
} }
case Regs::TextureFormat::A8: case Regs::TextureFormat::A8:
{ {
const u8* source_ptr = source + coarse_x * block_height + coarse_y * info.stride + texel_index_within_tile; const u8* source_ptr = source + offset;
if (disable_alpha) { if (disable_alpha) {
return { *source_ptr, *source_ptr, *source_ptr, 255 }; return { *source_ptr, *source_ptr, *source_ptr, 255 };
@ -421,7 +418,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::IA4: case Regs::TextureFormat::IA4:
{ {
const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2; const u8* source_ptr = source + offset / 2;
// TODO: component order not verified // TODO: component order not verified
@ -440,7 +437,7 @@ const Math::Vec4<u8> LookupTexture(const u8* source, int x, int y, const Texture
case Regs::TextureFormat::A4: case Regs::TextureFormat::A4:
{ {
const u8* source_ptr = source + coarse_x * block_height / 2 + coarse_y * info.stride + texel_index_within_tile / 2; const u8* source_ptr = source + offset / 2;
// TODO: component order not verified // TODO: component order not verified