eitan.c/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Shorten the implementation of 3D swizzle to only 3 functions

Browse source
This commit is contained in:
FernandoS27 2018-10-13 20:42:38 -04:00
parent 1ff20d8538
commit 331ce2942c
1 changed files with 27 additions and 70 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

75
src/video_core/textures/decoders.cpp
View file

@ -45,7 +45,7 @@ constexpr auto fast_swizzle_table = SwizzleTable<8, 4, 16>();
* Instead of going gob by gob, we map the coordinates inside a block and manage from
* those. Block_Width is assumed to be 1.
*/
void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
void PreciseProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end,
const u32 y_end, const u32 z_end, const u32 tile_offset,
const u32 xy_block_size, const u32 layer_z, const u32 stride_x,
@ -76,61 +76,12 @@ void Precise3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool un
}
}
/**
* This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue.
* The body of this function takes care of splitting the swizzled texture into blocks,
* and managing the extents of it. Once all the parameters of a single block are obtained,
* the function calls '3DProcessBlock' to process that particular Block.
*
* Documentation for the memory layout and decoding can be found at:
* https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces
*/
void Precise3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
const u32 width, const u32 height, const u32 depth,
const u32 bytes_per_pixel, const u32 out_bytes_per_pixel,
const u32 block_height, const u32 block_depth) {
auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
const u32 stride_x = width * out_bytes_per_pixel;
const u32 layer_z = height * stride_x;
const u32 gob_x_bytes = 64;
const u32 gob_elements_x = gob_x_bytes / bytes_per_pixel;
const u32 gob_elements_y = 8;
const u32 gob_elements_z = 1;
const u32 block_x_elements = gob_elements_x;
const u32 block_y_elements = gob_elements_y * block_height;
const u32 block_z_elements = gob_elements_z * block_depth;
const u32 blocks_on_x = div_ceil(width, block_x_elements);
const u32 blocks_on_y = div_ceil(height, block_y_elements);
const u32 blocks_on_z = div_ceil(depth, block_z_elements);
const u32 blocks = blocks_on_x * blocks_on_y * blocks_on_z;
const u32 gob_size = gob_x_bytes * gob_elements_y * gob_elements_z;
const u32 xy_block_size = gob_size * block_height;
const u32 block_size = xy_block_size * block_depth;
u32 tile_offset = 0;
for (u32 zb = 0; zb < blocks_on_z; zb++) {
const u32 z_start = zb * block_z_elements;
const u32 z_end = std::min(depth, z_start + block_z_elements);
for (u32 yb = 0; yb < blocks_on_y; yb++) {
const u32 y_start = yb * block_y_elements;
const u32 y_end = std::min(height, y_start + block_y_elements);
for (u32 xb = 0; xb < blocks_on_x; xb++) {
const u32 x_start = xb * block_x_elements;
const u32 x_end = std::min(width, x_start + block_x_elements);
Precise3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);
tile_offset += block_size;
}
}
}
}
/**
* This function manages ALL the GOBs(Group of Bytes) Inside a single block.
* Instead of going gob by gob, we map the coordinates inside a block and manage from
* those. Block_Width is assumed to be 1.
*/
void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
void FastProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
const u32 x_start, const u32 y_start, const u32 z_start, const u32 x_end,
const u32 y_end, const u32 z_end, const u32 tile_offset,
const u32 xy_block_size, const u32 layer_z, const u32 stride_x,
@ -169,15 +120,15 @@ void Fast3DProcessBlock(u8* swizzled_data, u8* unswizzled_data, const bool unswi
* This function unswizzles or swizzles a texture by mapping Linear to BlockLinear Textue.
* The body of this function takes care of splitting the swizzled texture into blocks,
* and managing the extents of it. Once all the parameters of a single block are obtained,
* the function calls '3DProcessBlock' to process that particular Block.
* the function calls 'ProcessBlock' to process that particular Block.
*
* Documentation for the memory layout and decoding can be found at:
* https://envytools.readthedocs.io/en/latest/hw/memory/g80-surface.html#blocklinear-surfaces
*/
void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle,
const u32 width, const u32 height, const u32 depth,
const u32 bytes_per_pixel, const u32 out_bytes_per_pixel,
const u32 block_height, const u32 block_depth) {
template <bool fast>
void SwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswizzle, const u32 width,
const u32 height, const u32 depth, const u32 bytes_per_pixel,
const u32 out_bytes_per_pixel, const u32 block_height, const u32 block_depth) {
auto div_ceil = [](const u32 x, const u32 y) { return ((x + y - 1) / y); };
const u32 stride_x = width * out_bytes_per_pixel;
const u32 layer_z = height * stride_x;
@ -205,9 +156,15 @@ void Fast3DSwizzledData(u8* swizzled_data, u8* unswizzled_data, const bool unswi
for (u32 xb = 0; xb < blocks_on_x; xb++) {
const u32 x_start = xb * block_x_elements;
const u32 x_end = std::min(width, x_start + block_x_elements);
Fast3DProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
if (fast) {
FastProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);
} else {
PreciseProcessBlock(swizzled_data, unswizzled_data, unswizzle, x_start, y_start,
z_start, x_end, y_end, z_end, tile_offset, xy_block_size,
layer_z, stride_x, bytes_per_pixel, out_bytes_per_pixel);
}
tile_offset += block_size;
}
}
@ -218,10 +175,10 @@ void CopySwizzledData(u32 width, u32 height, u32 depth, u32 bytes_per_pixel,
u32 out_bytes_per_pixel, u8* swizzled_data, u8* unswizzled_data,
bool unswizzle, u32 block_height, u32 block_depth) {
if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) {
Fast3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
SwizzledData<true>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
} else {
Precise3DSwizzledData(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
SwizzledData<false>(swizzled_data, unswizzled_data, unswizzle, width, height, depth,
bytes_per_pixel, out_bytes_per_pixel, block_height, block_depth);
}
}