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

Improved Legacy Swizzler to be better documented and work better

Browse source
This commit is contained in:
FernandoS27 2018-09-20 23:48:24 -04:00
parent bf2f2a715f
commit 68aaa83836
1 changed files with 21 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

32
src/video_core/textures/decoders.cpp
View file

@ -14,17 +14,12 @@ namespace Tegra::Texture {
/** /**
* Calculates the offset of an (x, y) position within a swizzled texture. * Calculates the offset of an (x, y) position within a swizzled texture.
* Taken from the Tegra X1 TRM. * Taken from the Tegra X1 Technical Reference Manual. pages 1187-1188
*/ */
static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel, u32 block_height) { static u32 GetSwizzleOffset(u32 x, u32 y, u32 bytes_per_pixel, u32 gob_address) {
// Round up to the next gob // Round up to the next gob
const u32 image_width_in_gobs{(image_width * bytes_per_pixel + 63) / 64};
u32 GOB_address = 0 + (y / (8 * block_height)) * 512 * block_height * image_width_in_gobs +
(x * bytes_per_pixel / 64) * 512 * block_height +
(y % (8 * block_height) / 8) * 512;
x *= bytes_per_pixel; x *= bytes_per_pixel;
u32 address = GOB_address + ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 + u32 address = gob_address + ((x % 64) / 32) * 256 + ((y % 8) / 2) * 64 + ((x % 32) / 16) * 32 +
(y % 2) * 16 + (x % 16); (y % 2) * 16 + (x % 16);
return address; return address;
@ -32,14 +27,24 @@ static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel,
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) {
u8* data_ptrs[2]; std::array<u8*, 2> data_ptrs;
const u32 stride = width * bytes_per_pixel;
const u32 gobs_in_x = 64;
const u32 gobs_in_y = 8;
const u32 gobs_size = gobs_in_x * gobs_in_y;
const u32 image_width_in_gobs{(stride + gobs_in_x - 1) / gobs_in_x};
for (unsigned y = 0; y < height; ++y) { for (unsigned y = 0; y < height; ++y) {
const u32 gob_y_address =
(y / (gobs_in_y * block_height)) * gobs_size * block_height * image_width_in_gobs +
(y % (gobs_in_y * block_height) / gobs_in_y) * gobs_size;
for (unsigned x = 0; x < width; ++x) { for (unsigned x = 0; x < width; ++x) {
u32 swizzle_offset = GetSwizzleOffset(x, y, width, bytes_per_pixel, block_height); const u32 gob_address =
u32 pixel_index = (x + y * width) * out_bytes_per_pixel; gob_y_address + (x * bytes_per_pixel / gobs_in_x) * gobs_size * block_height;
const u32 swizzle_offset = GetSwizzleOffset(x, y, bytes_per_pixel, gob_address);
const u32 pixel_index = (x + y * width) * out_bytes_per_pixel;
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], bytes_per_pixel); std::memcpy(data_ptrs[0], data_ptrs[1], bytes_per_pixel);
} }
@ -83,7 +88,8 @@ void FastSwizzleData(u32 width, u32 height, u32 bytes_per_pixel, u8* swizzled_da
const auto& table = swizzle_table[y % gobs_in_y]; const auto& table = 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 truncated_copy = std::min(copy_size, stride - xb);
const std::size_t gob_address{initial_gob + (xb / gobs_in_x) * gobs_size * block_height}; const std::size_t gob_address{initial_gob +
(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]};
const std::size_t pixel_index{xb + pixel_base}; const std::size_t pixel_index{xb + pixel_base};
data_ptrs[unswizzle] = swizzled_data + swizzle_offset; data_ptrs[unswizzle] = swizzled_data + swizzle_offset;