Mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-06-05 02:03:16 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 5

texture_cache: detiler: added support for block coded 128bit images

Browse source
This commit is contained in:
psucien 2024-06-09 12:54:19 +02:00
parent 1a66fa098f
commit fa73812f32
8 changed files with 115 additions and 35 deletions
Download patch file Download diff file Expand all files Collapse all files

3
src/video_core/host_shaders/CMakeLists.txt
View file

@ -3,7 +3,8 @@
set(SHADER_FILES
detile_m8x1.comp
detile_m8x4.comp
detile_m32x1.comp
detile_m32x4.comp
)
set(SHADER_INCLUDE ${CMAKE_CURRENT_BINARY_DIR}/include)

23
src/video_core/host_shaders/detile_m8x4.comp → src/video_core/host_shaders/detile_m32x1.comp
View file

@ -8,14 +8,14 @@ layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout(std430, binding = 0) buffer input_buf {
uint in_data[];
};
layout(rgba8ui, binding = 1) uniform writeonly uimage2D output_img;
layout(r32ui, binding = 1) uniform writeonly uimage2D output_img;
layout(push_constant) uniform image_info {
uint pitch;
} info;
// Inverse morton LUT, small enough to fit into K$
uint lut_8x4[16] = {
uint rmort[16] = {
0x11011000, 0x31213020,
0x13031202, 0x33233222,
0x51415040, 0x71617060,
@ -27,17 +27,17 @@ uint lut_8x4[16] = {
0x57475646, 0x77677666,
};
#define MICRO_TILE_DIM 8
#define TEXELS_PER_ELEMENT 1
#define MICRO_TILE_DIM (8)
#define TEXELS_PER_ELEMENT (1)
void main() {
uint src_tx = in_data[gl_GlobalInvocationID.x];
uint bit_ofs = 8 * (gl_LocalInvocationID.x % 4);
uint packed_pos = lut_8x4[gl_LocalInvocationID.x >> 2] >> bit_ofs;
uint packed_pos = rmort[gl_LocalInvocationID.x >> 2] >> bit_ofs;
uint col = bitfieldExtract(packed_pos, 4, 4);
uint row = bitfieldExtract(packed_pos, 0, 4);
uint p0 = in_data[gl_GlobalInvocationID.x];
uint tiles_per_pitch = info.pitch >> 3; // log2(MICRO_TILE_DIM)
uint target_tile_x = gl_WorkGroupID.x % tiles_per_pitch;
uint target_tile_y = gl_WorkGroupID.x / tiles_per_pitch;
@ -46,12 +46,5 @@ void main() {
uint dw_ofs_y = target_tile_y * MICRO_TILE_DIM + row;
ivec2 img_pos = ivec2(dw_ofs_x, dw_ofs_y);
uvec4 dst_tx = uvec4(
bitfieldExtract(src_tx, 0, 8),
bitfieldExtract(src_tx, 8, 8),
bitfieldExtract(src_tx, 16, 8),
bitfieldExtract(src_tx, 24, 8)
);
imageStore(output_img, img_pos, dst_tx);
imageStore(output_img, img_pos, uvec4(p0, 0, 0, 0));
}

50
src/video_core/host_shaders/detile_m32x4.comp Normal file
View file

@ -0,0 +1,50 @@
// SPDX-FileCopyrightText: Copyright 2024 shadPS4 Emulator Project
// SPDX-License-Identifier: GPL-2.0-or-later
#version 450
layout (local_size_x = 64, local_size_y = 1, local_size_z = 1) in;
layout(std430, binding = 0) buffer input_buf {
uint in_data[];
};
layout(rgba32ui, binding = 1) uniform writeonly uimage2D output_img;
layout(push_constant) uniform image_info {
uint pitch;
} info;
// Inverse morton LUT, small enough to fit into K$
uint rmort[16] = {
0x11011000, 0x31213020,
0x13031202, 0x33233222,
0x51415040, 0x71617060,
0x53435242, 0x73637262,
0x15051404, 0x35253424,
0x17071606, 0x37273626,
0x55455444, 0x75657464,
0x57475646, 0x77677666,
};
#define MICRO_TILE_DIM (8)
void main() {
uint bit_ofs = 8 * (gl_LocalInvocationID.x % 4);
uint packed_pos = rmort[gl_LocalInvocationID.x >> 2] >> bit_ofs;
uint col = bitfieldExtract(packed_pos, 4, 4);
uint row = bitfieldExtract(packed_pos, 0, 4);
uint block_ofs = 4 * gl_GlobalInvocationID.x;
uint p0 = in_data[block_ofs + 0];
uint p1 = in_data[block_ofs + 1];
uint p2 = in_data[block_ofs + 2];
uint p3 = in_data[block_ofs + 3];
uint tiles_per_pitch = (info.pitch >> 3) >> 2; // log2(MICRO_TILE_DIM) / 4
ivec2 img_pos = MICRO_TILE_DIM * ivec2(
gl_WorkGroupID.x % tiles_per_pitch,
gl_WorkGroupID.x / tiles_per_pitch
);
imageStore(output_img, img_pos + ivec2(col, row), uvec4(p0, p1, p2, p3));
}