fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[GPU] XeSL texture load shaders + minor XeSL cleanup

Browse Source
This commit is contained in:
Triang3l
2022-04-04 21:48:27 +03:00
parent c4eae232f1
commit 3d48fde5ca
197 changed files with 56046 additions and 7249 deletions
Download Patch File Download Diff File Expand all files Collapse all files

83
src/xenia/gpu/shaders/texture_load_dxt3aas1111.xesli Normal file
View File

@@ -0,0 +1,83 @@
/**
******************************************************************************
* 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 "texture_load.xesli"
xesl_entry
xesl_writeTypedStorageBuffer(xesl_uint4, xe_texture_load_dest, set=0,
binding=0, u0, space0)
xesl_entry_binding_next
xesl_typedStorageBuffer(xesl_uint4, xe_texture_load_source, set=1, binding=0,
t0, space0)
xesl_entry_bindings_end_local_size(8, 32, 1)
xesl_input_global_invocation_id
xesl_entry_signature_end
// 1 thread = 4 DXT3A-as-1111 blocks to 16x4 16bpp texels passed through an
// externally provided
// `uint4 XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(uint2 halfblocks)`
// conversion function.
xesl_uint3 block_index = xesl_GlobalInvocationID << xesl_uint3(2u, 0u, 0u);
xesl_dont_flatten
if (any(xesl_greaterThanEqual(block_index.xy,
XeTextureLoadSizeBlocks().xy))) {
return;
}
xesl_uint3 texel_index_host = block_index << xesl_uint3(2u, 2u, 0u);
uint blocks_pitch_host = XeTextureLoadHostPitch();
uint height_texels = XeTextureLoadHeightTexels();
uint block_offset_host = uint(
(XeTextureHostLinearOffset(xesl_int3(texel_index_host), blocks_pitch_host,
height_texels, 2u) +
XeTextureLoadHostOffset()) >> 4u);
uint elements_pitch_host = blocks_pitch_host >> 4u;
uint block_offset_guest =
XeTextureLoadGuestBlockOffset(block_index, 8u, 3u) >> 4u;
uint endian = XeTextureLoadEndian32();
xesl_uint4 blocks_01 = XeEndianSwap32(
xesl_typedStorageBufferLoad(xe_texture_load_source, block_offset_guest),
endian);
// Odd 2 blocks = even 2 blocks + 32 bytes when tiled.
block_offset_guest += XeTextureLoadIsTiled() ? 2u : 1u;
xesl_uint4 blocks_23 = XeEndianSwap32(
xesl_typedStorageBufferLoad(xe_texture_load_source, block_offset_guest),
endian);
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_01.xz));
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host + 1u,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_23.xz));
xesl_dont_flatten if (++texel_index_host.y < height_texels) {
block_offset_host += elements_pitch_host;
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_01.xz >> 16u));
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host + 1u,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_23.xz >> 16u));
xesl_dont_flatten if (++texel_index_host.y < height_texels) {
block_offset_host += elements_pitch_host;
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_01.yw));
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host + 1u,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_23.yw));
xesl_dont_flatten if (++texel_index_host.y < height_texels) {
block_offset_host += elements_pitch_host;
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_01.yw >> 16u));
xesl_writeTypedStorageBufferStore(
xe_texture_load_dest, block_offset_host + 1u,
XE_TEXTURE_LOAD_DXT3A_AS_1_1_1_1_TO_16BPP(blocks_23.yw >> 16u));
}
}
}
xesl_entry_end