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

[GPU] Simplify local X offsetting with resolution scaling

Browse Source 
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.
This commit is contained in:
Triang3l
2026-01-13 23:14:15 +03:00
parent 76c531bff2
commit 0f23f05683
169 changed files with 61380 additions and 64785 deletions
Download Patch File Download Diff File Expand all files Collapse all files

74
src/xenia/gpu/shaders/resolve.xesli
View File

@@ -79,8 +79,8 @@ struct XeResolveInfo {
uint dest_format;
float dest_exp_bias_factor;
bool dest_swap;
uint dest_row_pitch_aligned_shr5;
uint dest_slice_pitch_aligned_shr4;
uint dest_row_pitch_macro_tiles;
uint dest_slice_pitch_3d_macro_tiles;
uint2_xe dest_xy_offset_scaled;
uint sample_select;
uint dest_base;
@@ -132,10 +132,11 @@ XeResolveInfo XeResolveGetInfo(param_push_consts_xe) {
(int(dest_info) << (32 - (16 + 6)) >> (32 - 6) << 23) +
float_bits_to_int_xe(1.0f));
resolve_info.dest_swap = (dest_info & (1u << 24u)) != 0u;
resolve_info.dest_row_pitch_aligned_shr5 =
resolve_info.dest_row_pitch_macro_tiles =
dest_coordinate_info & ((1u << 10u) - 1u);
resolve_info.dest_slice_pitch_aligned_shr4 =
((dest_coordinate_info >> 10u) & ((1u << 10u) - 1u)) << 1u;
resolve_info.dest_slice_pitch_3d_macro_tiles =
((dest_coordinate_info >> 10u) & ((1u << 10u) - 1u)) <<
(5 - XENOS_TEXTURE_MACRO_TILE_HEIGHT_3D_LOG2);
resolve_info.dest_xy_offset_scaled =
(((uint_x2_xe(dest_coordinate_info) >> uint2_xe(20u, 24u)) &
((1u << 4u) - 1u)) <<
@@ -158,41 +159,50 @@ uint XeResolveEdramPixelStrideInts(XeResolveInfo resolve_info) {
}
#ifndef XE_RESOLVE_CLEAR
uint XeResolveDestPixelAddress(XeResolveInfo resolve_info, uint2_xe p,
uint bpp_log2) {
p += resolve_info.dest_xy_offset_scaled;
uint XeResolveDestPixelAddress(const XeResolveInfo resolve_info,
uint2_xe host_position,
const uint bytes_per_element_log2) {
host_position += resolve_info.dest_xy_offset_scaled;
uint address;
uint2_xe guest_position = host_position;
#ifdef XE_RESOLVE_RESOLUTION_SCALED
address = XeTextureScaledTiledOffset(
resolve_info.dest_is_array, uint3_xe(p, resolve_info.dest_slice),
resolve_info.dest_row_pitch_aligned_shr5,
resolve_info.dest_slice_pitch_aligned_shr4, bpp_log2,
resolve_info.resolution_scale);
#else
dont_flatten_xe if (resolve_info.dest_is_array) {
address = uint(XenosTextureTiledAddress3D(
int3_xe(uint3_xe(p, resolve_info.dest_slice)),
resolve_info.dest_row_pitch_aligned_shr5,
resolve_info.dest_slice_pitch_aligned_shr4,
bpp_log2));
} else {
address = uint(XenosTextureTiledAddress2D(
int2_xe(p), resolve_info.dest_row_pitch_aligned_shr5, bpp_log2));
}
address += resolve_info.dest_base;
const XeniaTextureResolutionScaledAddressing
resolution_scaled_addressing =
XeniaTextureGetResolutionScaledAddressing(
host_position.xy, resolve_info.resolution_scale,
bytes_per_element_log2);
guest_position = resolution_scaled_addressing.guest_group_origin;
#endif
dont_flatten_xe if (resolve_info.dest_is_array) {
address = uint(XenosTextureTiledAddress3D(
int3_xe(uint3_xe(guest_position, resolve_info.dest_slice)),
resolve_info.dest_row_pitch_macro_tiles,
resolve_info.dest_slice_pitch_3d_macro_tiles,
bytes_per_element_log2));
} else {
address = uint(XenosTextureTiledAddress2D(
int2_xe(guest_position), resolve_info.dest_row_pitch_macro_tiles,
bytes_per_element_log2));
}
#ifdef XE_RESOLVE_RESOLUTION_SCALED
address = address * (resolve_info.resolution_scale.x *
resolve_info.resolution_scale.y) +
resolution_scaled_addressing.host_byte_offset_in_guest_group;
#endif
address += resolve_info.dest_base;
return address;
}
// Offset of the beginning of the odd R32G32/R32G32B32A32 store address from
// the address of the even store.
uint XeResolveDestRightConsecutiveBlocksOffset(uint x, uint bpp_log2,
uint2_xe resolution_scale) {
// XOR to apply to the byte address to flip the bits corresponding to the
// given X coordinate bits the macro tile width, or with resolution scaling,
// within XeniaTextureResolutionScaledGroupElements.x.
uint XeResolveLocalXAddressXor(const uint x,
const uint bytes_per_element_log2) {
#ifdef XE_RESOLVE_RESOLUTION_SCALED
return XeTextureScaledRightSubUnitOffsetInConsecutivePair(
x, bpp_log2, resolution_scale);
return x << bytes_per_element_log2;
#else
return XeTextureTiledOddConsecutiveBlocksOffset(bpp_log2);
return uint(XenosTextureTiledAddressXInMacroXor(int(x),
bytes_per_element_log2));
#endif
}