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

[XeSL] Metal Shading Language definitions

Browse Source
This commit is contained in:
Triang3l
2022-06-16 21:39:16 +03:00
parent 820b7ba217
commit 166be463be
176 changed files with 83740 additions and 83682 deletions
Download Patch File Download Diff File Expand all files Collapse all files

146
src/xenia/gpu/shaders/texture_load.xesli
View File

@@ -22,10 +22,12 @@
// in 128x16 blocks per group for 8bpb / 16bpb, and for a 32x32 mip tail, there
// would be two groups rather than one, for a total of 128x32 blocks - 75% of
// the work will be wasted rather than 50% with one 64x32-block group.
#define kXeTextureLoadGroupSizeX 4
#define kXeTextureLoadGroupSizeY 32
#define xesl_localSize_x 4
#define xesl_localSize_y 32
#define xesl_localSize_z 1
xesl_cbuffer_begin(xe_texture_load_constants, set=2, binding=0, b0, space0)
xesl_constantBuffer_begin(xe_texture_load_constants, set=2, binding=0, b0,
space0)
uint xe_texture_load_is_tiled_3d_endian_scale;
// Base offset in bytes, resolution-scaled.
uint xe_texture_load_guest_offset;
@@ -47,115 +49,109 @@ xesl_cbuffer_begin(xe_texture_load_constants, set=2, binding=0, b0, space0)
uint xe_texture_load_host_pitch;
uint xe_texture_load_height_texels;
xesl_cbuffer_end(xe_texture_load_constants)
xesl_constantBuffer_end(xe_texture_load_constants)
bool XeTextureLoadIsTiled() {
#define XE_TEXTURE_LOAD_CONSTANT_BUFFER_BINDING \
xesl_constantBuffer_binding(xe_texture_load_constants, buffer(0))
struct XeTextureLoadInfo {
bool is_tiled;
bool is_3d;
uint endian_32;
xesl_uint2 resolution_scale;
uint guest_offset;
uint guest_pitch_aligned;
uint guest_z_stride_block_rows_aligned;
xesl_uint3 size_blocks;
uint host_offset;
uint host_pitch;
uint height_texels;
};
XeTextureLoadInfo XeTextureLoadGetInfo(
xesl_function_param_constantBuffer(xe_texture_load_constants)) {
XeTextureLoadInfo load_info;
uint is_tiled_3d_endian_scale = xesl_constant(
xe_texture_load_constants, xe_texture_load_is_tiled_3d_endian_scale);
#ifdef XE_TEXTURE_LOAD_RESOLUTION_SCALED
// Only resolved textures can be resolution-scaled, and resolving is only
// possible to a tiled destination.
return true;
load_info.is_tiled = true;
#else
return (xesl_constant(xe_texture_load_constants,
xe_texture_load_is_tiled_3d_endian_scale) & 1u) != 0u;
load_info.is_tiled = (is_tiled_3d_endian_scale & 1u) != 0u;
#endif
}
bool XeTextureLoadIs3D() {
return (xesl_constant(xe_texture_load_constants,
xe_texture_load_is_tiled_3d_endian_scale) & (1u << 1u))
!= 0u;
}
uint XeTextureLoadEndian32() {
return (xesl_constant(xe_texture_load_constants,
xe_texture_load_is_tiled_3d_endian_scale) >> 2u) & 3u;
}
xesl_uint2 XeTextureLoadResolutionScale() {
load_info.is_3d = (is_tiled_3d_endian_scale & (1u << 1u)) != 0u;
load_info.endian_32 = (is_tiled_3d_endian_scale >> 2u) & 3u;
#ifdef XE_TEXTURE_LOAD_RESOLUTION_SCALED
return (xesl_constant(xe_texture_load_constants,
xe_texture_load_is_tiled_3d_endian_scale.xx) >>
xesl_uint2(4u, 6u)) & 3u;
load_info.resolution_scale =
((xesl_uint_x2(is_tiled_3d_endian_scale)) >> xesl_uint2(4u, 6u)) & 3u;
#else
return xesl_uint2(1u, 1u);
load_info.resolution_scale = xesl_uint2(1u, 1u);
#endif
}
xesl_uint3 XeTextureLoadSizeBlocks() {
return xesl_constant(xe_texture_load_constants, xe_texture_load_size_blocks);
}
uint XeTextureLoadHostOffset() {
return xesl_constant(xe_texture_load_constants, xe_texture_load_host_offset);
}
uint XeTextureLoadHostPitch() {
return xesl_constant(xe_texture_load_constants, xe_texture_load_host_pitch);
}
uint XeTextureLoadHeightTexels() {
return xesl_constant(xe_texture_load_constants,
xe_texture_load_height_texels);
load_info.guest_offset =
xesl_constant(xe_texture_load_constants, xe_texture_load_guest_offset);
load_info.guest_pitch_aligned = xesl_constant(
xe_texture_load_constants, xe_texture_load_guest_pitch_aligned);
load_info.guest_z_stride_block_rows_aligned =
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_z_stride_block_rows_aligned);
load_info.size_blocks =
xesl_constant(xe_texture_load_constants, xe_texture_load_size_blocks);
load_info.host_offset =
xesl_constant(xe_texture_load_constants, xe_texture_load_host_offset);
load_info.host_pitch =
xesl_constant(xe_texture_load_constants, xe_texture_load_host_pitch);
load_info.height_texels =
xesl_constant(xe_texture_load_constants, xe_texture_load_height_texels);
return load_info;
}
// bpb and bpb_log2 are separate because bpb may be not a power of 2 (like 96).
uint XeTextureLoadGuestBlockOffset(xesl_uint3 block_index, uint bpb,
uint XeTextureLoadGuestBlockOffset(XeTextureLoadInfo load_info,
xesl_uint3 block_index, uint bpb,
uint bpb_log2) {
#ifdef XE_TEXTURE_LOAD_RESOLUTION_SCALED
// Only resolved textures can be resolution-scaled, and resolving is only
// possible to a tiled destination.
return
xesl_constant(xe_texture_load_constants, xe_texture_load_guest_offset) +
XeTextureScaledTiledOffset(
XeTextureLoadIs3D(), block_index,
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_pitch_aligned),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_z_stride_block_rows_aligned),
bpb_log2, XeTextureLoadResolutionScale());
load_info.guest_offset +
XeTextureScaledTiledOffset(load_info.is_3d, block_index,
load_info.guest_pitch_aligned,
load_info.guest_z_stride_block_rows_aligned,
bpb_log2, load_info.resolution_scale);
#else
int block_offset_guest;
xesl_dont_flatten if (XeTextureLoadIsTiled()) {
xesl_dont_flatten if (XeTextureLoadIs3D()) {
xesl_dont_flatten if (load_info.is_tiled) {
xesl_dont_flatten if (load_info.is_3d) {
block_offset_guest = XeTextureTiledOffset3D(
xesl_int3(block_index),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_pitch_aligned),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_z_stride_block_rows_aligned),
bpb_log2);
xesl_int3(block_index), load_info.guest_pitch_aligned,
load_info.guest_z_stride_block_rows_aligned, bpb_log2);
} else {
block_offset_guest = XeTextureTiledOffset2D(
xesl_int2(block_index.xy),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_pitch_aligned), bpb_log2);
xesl_int2(block_index.xy), load_info.guest_pitch_aligned,
bpb_log2);
}
} else {
block_offset_guest = XeTextureGuestLinearOffset(
xesl_int3(block_index),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_pitch_aligned),
xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_z_stride_block_rows_aligned),
bpb);
xesl_int3(block_index), load_info.guest_pitch_aligned,
load_info.guest_z_stride_block_rows_aligned, bpb);
}
return uint(int(xesl_constant(xe_texture_load_constants,
xe_texture_load_guest_offset)) +
block_offset_guest);
return uint(int(load_info.guest_offset) + block_offset_guest);
#endif
}
// Offset of the beginning of the odd R32G32/R32G32B32A32 load address from the
// address of the even load, for power-of-two-sized textures.
uint XeTextureLoadRightConsecutiveBlocksOffset(uint block_x, uint bpb_log2) {
uint XeTextureLoadRightConsecutiveBlocksOffset(XeTextureLoadInfo load_info,
uint block_x, uint bpb_log2) {
#ifdef XE_TEXTURE_LOAD_RESOLUTION_SCALED
return XeTextureScaledRightSubUnitOffsetInConsecutivePair(
block_x, bpb_log2, XeTextureLoadResolutionScale());
block_x, bpb_log2, load_info.resolution_scale);
#else
uint offset;
uint consecutive_blocks_log2 =
XeTextureTiledConsecutiveBlocksLog2(bpb_log2);
xesl_dont_flatten if (XeTextureLoadIsTiled()) {
xesl_dont_flatten if (load_info.is_tiled) {
offset = XeTextureTiledOddConsecutiveBlocksOffset(bpb_log2);
} else {
offset = 1u << (consecutive_blocks_log2 + bpb_log2);