flodavid/suyu
1
0
Fork 0
forked from suyu/suyu
Code Pull requests Activity

Merge pull request #1311 from FernandoS27/fast-swizzle

Browse source 
Optimized Texture Swizzling
This commit is contained in:
bunnei 2018-09-17 12:39:34 -04:00 committed by GitHub
commit a94b623dfb
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 49 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

51
src/video_core/textures/decoders.cpp
View file

@ -46,6 +46,48 @@ void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_
} }
} }
template <std::size_t N, std::size_t M>
struct alignas(64) SwizzleTable {
constexpr SwizzleTable() {
for (u32 y = 0; y < N; ++y) {
for (u32 x = 0; x < M; ++x) {
const u32 x2 = x * 16;
values[y][x] = static_cast<u16>(((x2 % 64) / 32) * 256 + ((y % 8) / 2) * 64 +
((x2 % 32) / 16) * 32 + (y % 2) * 16);
}
}
}
const std::array<u16, M>& operator[](std::size_t index) const {
return values[index];
}
std::array<std::array<u16, M>, N> values{};
};
constexpr auto swizzle_table = SwizzleTable<8, 4>();
void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u8* swizzled_data,
u8* unswizzled_data, bool unswizzle, u32 block_height) {
std::array<u8*, 2> data_ptrs;
const std::size_t stride{width * bytes_per_pixel};
const std::size_t image_width_in_gobs{(stride + 63) / 64};
const std::size_t copy_size{16};
for (std::size_t y = 0; y < height; ++y) {
const std::size_t initial_gob =
(y / (8 * block_height)) * 512 * block_height * image_width_in_gobs +
(y % (8 * block_height) / 8) * 512;
const std::size_t pixel_base{y * width * bytes_per_pixel};
const auto& table = swizzle_table[y % 8];
for (std::size_t xb = 0; xb < stride; xb += copy_size) {
const std::size_t gob_address{initial_gob + (xb / 64) * 512 * block_height};
const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]};
const std::size_t pixel_index{xb + pixel_base};
data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
std::memcpy(data_ptrs[0], data_ptrs[1], copy_size);
}
}
}
u32 BytesPerPixel(TextureFormat format) { u32 BytesPerPixel(TextureFormat format) {
switch (format) { switch (format) {
case TextureFormat::DXT1: case TextureFormat::DXT1:
@ -92,8 +134,13 @@ u32 BytesPerPixel(TextureFormat format) {
std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width, std::vector<u8> UnswizzleTexture(VAddr address, u32 tile_size, u32 bytes_per_pixel, u32 width,
u32 height, u32 block_height) { u32 height, u32 block_height) {
std::vector<u8> unswizzled_data(width * height * bytes_per_pixel); std::vector<u8> unswizzled_data(width * height * bytes_per_pixel);
CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel, if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) {
Memory::GetPointer(address), unswizzled_data.data(), true, block_height); FastSwizzleData(width / tile_size, height / tile_size, bytes_per_pixel,
Memory::GetPointer(address), unswizzled_data.data(), true, block_height);
} else {
CopySwizzledData(width / tile_size, height / tile_size, bytes_per_pixel, bytes_per_pixel,
Memory::GetPointer(address), unswizzled_data.data(), true, block_height);
}
return unswizzled_data; return unswizzled_data;
} }