Switch between even and odd 16-byte element sequences along X by simply flipping a bit rather than going to a different resolution-scaled group of pixels, by increasing the size of the group within the constraints imposed by tiling.
87 lines
3.8 KiB
Plaintext
87 lines
3.8 KiB
Plaintext
/**
|
|
******************************************************************************
|
|
* Xenia : Xbox 360 Emulator Research Project *
|
|
******************************************************************************
|
|
* Copyright 2022 Ben Vanik. All rights reserved. *
|
|
* Released under the BSD license - see LICENSE in the root for more details. *
|
|
******************************************************************************
|
|
*/
|
|
|
|
#include "pixel_formats.xesli"
|
|
#include "texture_load.xesli"
|
|
|
|
array_buffer_wo_declare_xe(uint4_xe, xe_texture_load_dest, set=0, binding=0, u0,
|
|
space0)
|
|
array_buffer_declare_xe(uint4_xe, xe_texture_load_source, set=1, binding=0, t0,
|
|
space0)
|
|
entry_bindings_begin_compute_xe
|
|
XE_TEXTURE_LOAD_PUSH_CONST_BINDING
|
|
entry_binding_next_xe
|
|
array_buffer_wo_binding_xe(uint4_xe, xe_texture_load_dest, buffer(1))
|
|
entry_binding_next_xe
|
|
array_buffer_binding_xe(uint4_xe, xe_texture_load_source, buffer(2))
|
|
entry_bindings_end_inputs_begin_compute_xe
|
|
entry_in_global_thread_id_xe
|
|
entry_inputs_end_code_begin_compute_xe
|
|
{
|
|
// 1 thread = 2 DXT3 blocks to 8x4 R8G8B8A8 texels.
|
|
XeTextureLoadInfo load_info = XeTextureLoadGetInfo(pass_push_consts_xe);
|
|
uint3_xe block_index = in_global_thread_id_xe << uint3_xe(1u, 0u, 0u);
|
|
dont_flatten_xe
|
|
if (any(greater_than_equal_xe(block_index.xy, load_info.size_blocks.xy))) {
|
|
return;
|
|
}
|
|
uint3_xe texel_index_host = block_index << uint3_xe(2u, 2u, 0u);
|
|
uint block_offset_host = uint(
|
|
(XeTextureHostLinearOffset(int3_xe(texel_index_host),
|
|
load_info.host_pitch, load_info.height_texels,
|
|
4u) +
|
|
load_info.host_offset) >> 4u);
|
|
uint elements_pitch_host = load_info.host_pitch >> 4u;
|
|
uint block_offset_guest =
|
|
XeTextureLoadSourceAddress(load_info, block_index, 4u) >> 4u;
|
|
uint i;
|
|
unroll_xe for (i = 0u; i < 2u; ++i) {
|
|
if (i != 0u) {
|
|
++block_offset_host;
|
|
// Odd block = even block + 32 guest bytes when tiled.
|
|
block_offset_guest += load_info.is_tiled ? 2u : 1u;
|
|
}
|
|
uint4_xe block = XeEndianSwap32(
|
|
array_buffer_load_xe(xe_texture_load_source, block_offset_guest),
|
|
load_info.endian_32);
|
|
uint2_xe bgr_end_8in10 = XeDXTColorEndpointsToBGR8In10(block.z);
|
|
// Sort the color indices so they can be used as weights for the second
|
|
// endpoint.
|
|
uint bgr_weights = XeDXTHighColorWeights(block.w);
|
|
array_buffer_store_xe(
|
|
xe_texture_load_dest, block_offset_host,
|
|
XeDXTOpaqueRowToRGB8(bgr_end_8in10, bgr_weights) +
|
|
((block.xxxx >> uint4_xe(0u, 4u, 8u, 12u)) & 0xFu) * 0x11000000u);
|
|
dont_flatten_xe if (texel_index_host.y + 1u < load_info.height_texels) {
|
|
array_buffer_store_xe(
|
|
xe_texture_load_dest, block_offset_host + elements_pitch_host,
|
|
XeDXTOpaqueRowToRGB8(bgr_end_8in10, bgr_weights >> 8u) +
|
|
((block.xxxx >> uint4_xe(16u, 20u, 24u, 28u)) & 0xFu) *
|
|
0x11000000u);
|
|
dont_flatten_xe if (texel_index_host.y + 2u < load_info.height_texels) {
|
|
array_buffer_store_xe(
|
|
xe_texture_load_dest, block_offset_host + 2u * elements_pitch_host,
|
|
XeDXTOpaqueRowToRGB8(bgr_end_8in10, bgr_weights >> 16u) +
|
|
((block.yyyy >> uint4_xe(0u, 4u, 8u, 12u)) & 0xFu) *
|
|
0x11000000u);
|
|
dont_flatten_xe
|
|
if (texel_index_host.y + 3u < load_info.height_texels) {
|
|
array_buffer_store_xe(
|
|
xe_texture_load_dest,
|
|
block_offset_host + 3u * elements_pitch_host,
|
|
XeDXTOpaqueRowToRGB8(bgr_end_8in10, bgr_weights >> 24u) +
|
|
((block.yyyy >> uint4_xe(16u, 20u, 24u, 28u)) & 0xFu) *
|
|
0x11000000u);
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
entry_code_end_compute_xe
|