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

[GPU/WGF] Fix resolve_fast_32bpp_4xmsaa sample addressing

Browse Source 
For 4xMSAA 32bpp, samples 2/3 add a +4 byte horizontal offset to the
EDRAM address. The old typed buffer code (Buffer<uint4>) implicitly
truncated this offset via integer division (>> 2), and the .yw swizzle
compensated by selecting odd elements within the aligned uint4 load.

After the ByteAddressBuffer conversion, Load4 uses the exact byte
address including the +4 offset, shifting the load window by one dword.
The .yw swizzle then picks samples from the wrong pixels, causing
visual artifacts like vertical streaks in Resident Evil 5.

Fix by aligning the load address down to 16 bytes and selecting the
swizzle based on whether the address was offset.
This commit is contained in:
Herman S.
2026-03-15 17:21:57 +09:00
parent 7f189f21fe
commit fdd583eced
1 changed files with 18 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

28
src/xenia/gpu/shaders/resolve_fast_32bpp_4xmsaa.xesli
View File

@@ -42,27 +42,35 @@ entry_inputs_end_code_begin_compute_xe
kXenosMsaaSamples_4X, resolve_info.edram_is_depth, 0u,
XeResolveFirstSampleIndex(resolve_info.sample_select),
resolve_info.resolution_scale);
// For 32bpp 4xMSAA, samples 0 and 1 are at consecutive int positions
// (horizontally adjacent in the sample grid). Each Load4 reads 4 consecutive
// samples spanning 2 pixels. Samples 0/1 (sample_select 0/1) produce a
// 16-byte-aligned address, but samples 2/3 (sample_select 2/3) add a +4 byte
// offset (the horizontal sample offset in the EDRAM tile). Align the load
// address down to 16 bytes and select the correct pair via swizzle: .xz for
// aligned (samples 0/1 at positions 0,2 within the uint4), .yw for
// offset-by-4 (samples 2/3 at positions 1,3 within the uint4).
uint source_address_aligned = source_address & ~15u;
uint4_xe pixels_0123, pixels_4567;
dont_flatten_xe
if (resolve_info.sample_select != kXenosCopySampleSelect_2 &&
resolve_info.sample_select != kXenosCopySampleSelect_3) {
if ((source_address & 15u) == 0u) {
pixels_0123.xy = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address).xz;
source_address_aligned).xz;
pixels_0123.zw = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x10u).xz;
source_address_aligned + 0x10u).xz;
pixels_4567.xy = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x20u).xz;
source_address_aligned + 0x20u).xz;
pixels_4567.zw = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x30u).xz;
source_address_aligned + 0x30u).xz;
} else {
pixels_0123.xy = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address).yw;
source_address_aligned).yw;
pixels_0123.zw = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x10u).yw;
source_address_aligned + 0x10u).yw;
pixels_4567.xy = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x20u).yw;
source_address_aligned + 0x20u).yw;
pixels_4567.zw = byte_buffer_align16_load16_xe(xe_resolve_edram,
source_address + 0x30u).yw;
source_address_aligned + 0x30u).yw;
}
dont_flatten_xe
if (pixel_index.x == 0u &&