RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-22 02:32:36 +01:00
Code Issues Releases Wiki Activity

rsx: Rework invalidation tagging

Browse Source
This commit is contained in:
kd-11 2022-01-05 20:25:16 +03:00 committed by kd-11
parent 7563655221
commit 92824b6729
2 changed files with 85 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

100
rpcs3/Emu/RSX/Common/surface_store.h
View File

@ -662,35 +662,97 @@ namespace rsx
{
// Generic painter's algorithm to detect obsolete sections
ensure(range.length() < 64 * 0x100000);
std::vector<u8> marker(range.length());
std::memset(marker.data(), 0, range.length());
std::vector<u8> marker(range.length(), 0);
auto compare_and_tag_row = [&](u32 offset, u32 length) -> bool
{
bool valid = false;
for (u32 i = 0; i < (length / 8); ++i, offset += 8, length -= 8)
{
auto dest = reinterpret_cast<u64*>(marker.data() + offset);
valid |= (*dest != umax);
*dest = umax;
}
if (length >= 4)
{
auto dest = reinterpret_cast<u32*>(marker.data() + offset);
valid |= (*dest != umax);
*dest = umax;
offset += 4;
length -= 4;
}
if (length >= 2)
{
auto dest = reinterpret_cast<u16*>(marker.data() + offset);
valid |= (*dest != umax);
*dest = umax;
offset += 2;
length -= 2;
}
if (length)
{
auto dest = (marker.data() + offset);
valid |= (*dest != umax);
*dest = umax;
}
return valid;
};
for (auto it = sections.crbegin(); it != sections.crend(); ++it)
{
if (!it->surface->get_memory_range().inside(range))
{
continue;
}
const auto true_pitch_in_bytes = it->surface->get_surface_width(rsx::surface_metrics::bytes);
const auto true_height_in_rows = it->surface->get_surface_height(rsx::surface_metrics::samples);
auto this_range = it->surface->get_memory_range();
ensure(this_range.overlaps(range));
const auto native_pitch = it->surface->get_surface_width(rsx::surface_metrics::bytes);
const auto rsx_pitch = it->surface->get_rsx_pitch();
auto num_rows = it->surface->get_surface_height(rsx::surface_metrics::samples);
bool valid = false;
auto addr = it->base_address - range.start;
auto data = marker.data();
for (usz row = 0; row < true_height_in_rows; ++row)
if (this_range.start < range.start)
{
for (usz col = 0; col < true_pitch_in_bytes; ++col)
// Starts outside bounds
const auto internal_offset = (range.start - this_range.start);
const auto row_num = internal_offset / rsx_pitch;
const auto row_offset = internal_offset % rsx_pitch;
// This section is unconditionally valid
valid = true;
if (row_offset < native_pitch)
{
if (const auto loc = col + addr; !data[loc])
{
valid = true;
data[loc] = 1;
}
compare_and_tag_row(0, native_pitch - row_offset);
}
addr += true_pitch_in_bytes;
// Jump to next row...
this_range.start = this_range.start + (row_num + 1) * rsx_pitch;
}
if (this_range.end > range.end)
{
// Unconditionally valid
valid = true;
this_range.end = range.end;
}
if (valid)
{
if (this_range.start >= this_range.end)
{
continue;
}
num_rows = utils::aligned_div(this_range.length(), rsx_pitch);
}
for (u32 row = 0, offset = (this_range.start - range.start); row < num_rows; ++row, offset += rsx_pitch)
{
valid |= compare_and_tag_row(offset, std::min<u32>(native_pitch, (this_range.end - offset + 1)));
}
if (!valid)

8
rpcs3/Emu/RSX/Common/texture_cache.h
View File

@ -1879,16 +1879,16 @@ namespace rsx
switch (result.external_subresource_desc.op)
{
case deferred_request_command::atlas_gather:
max_safe_sections = 32 + attr.mipmaps; break;
max_safe_sections = 8 + 2 * attr.mipmaps; break;
case deferred_request_command::cubemap_gather:
max_safe_sections = 16 * attr.mipmaps; break;
max_safe_sections = 6 * 2 * attr.mipmaps; break;
case deferred_request_command::_3d_gather:
max_safe_sections = (attr.depth * attr.mipmaps * 6); break;
max_safe_sections = (attr.depth * attr.mipmaps * 150) / 100; break;
default:
break;
}
if (overlapping_fbos.size() > std::max<usz>(max_safe_sections, 32u))
if (overlapping_fbos.size() > max_safe_sections)
{
rsx_log.error("[Performance warning] Texture gather routine encountered too many objects!");
m_rtts.check_for_duplicates(overlapping_fbos, memory_range);