Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Reverse stride align restriction on FastSwizzle due to lost performance

Browse source
This commit is contained in:
FernandoS27 2018-09-21 12:09:59 -04:00
parent d2dd1289bd
commit 57b44200a2
1 changed files with 2 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/video_core/textures/decoders.cpp
View file

@ -85,7 +85,6 @@ static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_
const std::size_t pixel_base{y * width * out_bytes_per_pixel}; const std::size_t pixel_base{y * width * out_bytes_per_pixel};
const auto& table = fast_swizzle_table[y % gobs_in_y]; const auto& table = fast_swizzle_table[y % gobs_in_y];
for (std::size_t xb = 0; xb < stride; xb += copy_size) { for (std::size_t xb = 0; xb < stride; xb += copy_size) {
const std::size_t truncated_copy = std::min(copy_size, stride - xb);
const std::size_t gob_address{initial_gob + const std::size_t gob_address{initial_gob +
(xb / gobs_in_x) * gobs_size * block_height}; (xb / gobs_in_x) * gobs_size * block_height};
const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]}; const std::size_t swizzle_offset{gob_address + table[(xb / 16) % 4]};
@ -93,14 +92,14 @@ static void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_
const std::size_t pixel_index{out_x + pixel_base}; const std::size_t pixel_index{out_x + pixel_base};
data_ptrs[unswizzle] = swizzled_data + swizzle_offset; data_ptrs[unswizzle] = swizzled_data + swizzle_offset;
data_ptrs[!unswizzle] = unswizzled_data + pixel_index; data_ptrs[!unswizzle] = unswizzled_data + pixel_index;
std::memcpy(data_ptrs[0], data_ptrs[1], truncated_copy); std::memcpy(data_ptrs[0], data_ptrs[1], copy_size);
} }
} }
} }
void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel, void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) { u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) {
if (bytes_per_pixel % 3 != 0) { if (bytes_per_pixel % 3 != 0 && (width * bytes_per_pixel) % 16 == 0) {
FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data, FastSwizzleData(width, height, bytes_per_pixel, out_bytes_per_pixel, swizzled_data,
unswizzled_data, unswizzle, block_height); unswizzled_data, unswizzle, block_height);
} else { } else {