Skip to content

GS/TC: Improve target selection on source lookup #15089

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
refractionpcsx2 wants to merge 1 commit into
base: master
Choose a base branch
from
Open

GS/TC: Improve target selection on source lookup #15089

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 6 additions & 3 deletions pcsx2/GS/GSState.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6236,11 +6236,14 @@ GSState::TextureMinMaxResult GSState::GetTextureMinMax(GIFRegTEX0 TEX0, GIFRegCL
// since this was making it take the full texture even though it wasn't needed.
if (!clamp_to_tsize)
{
const u32 mask = (m_vt.m_min.t.floor() == GSVector4::zero()).mask();
GSVector4 min_t = m_vt.m_min.t;
min_t.x = min_t.x - (static_cast<int>(min_t.x) & ~(GSLocalMemory::m_psm[TEX0.PSM].pgs.x - 1));
min_t.y = min_t.y - (static_cast<int>(min_t.y) & ~(GSLocalMemory::m_psm[TEX0.PSM].pgs.y - 1));
const u32 mask = (min_t.floor() == GSVector4::zero()).mask();
if (mask & 1) // X == 0
st.x = st.max(GSVector4::zero()).x;
st.x = st.max(m_vt.m_min.t.floor()).x;
if (mask & 2) // Y == 0
st.y = st.max(GSVector4::zero()).y;
st.y = st.max(m_vt.m_min.t.floor()).y;
}
}

Expand Down
71 changes: 47 additions & 24 deletions pcsx2/GS/Renderers/HW/GSTextureCache.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -369,14 +369,14 @@ GSVector4i GSTextureCache::TranslateAlignedRectByPage(u32 tbp, u32 tebp, u32 tbw

if (src_pgw != dst_pgw)
{
const int horizontal_dst_page_offset = page_offset % clamped_tbw;
const bool single_row = ((src_pgw + horizontal_dst_page_offset) <= clamped_tbw) && (in_rect.height() <= dst_page_size.y);
const bool single_page = (in_rect.width() <= t_psm.pgs.x) && (in_rect.height() <= t_psm.pgs.y);
const int vertical_offset = in_rect.y / t_psm.pgs.y;
const int horizontal_offset = in_rect.x / t_psm.pgs.x;
const int rect_offset = horizontal_offset + (vertical_offset * src_pgw);
const int rect_pages = std::max(((in_rect.width() / t_psm.pgs.x) % src_pgw) + ((in_rect.height() / t_psm.pgs.y) * src_pgw), 1);
page_offset += rect_offset;
const int horizontal_dst_page_offset = page_offset % clamped_tbw;
const bool single_row = ((src_pgw + horizontal_dst_page_offset) <= clamped_tbw) && (in_rect.height() <= dst_page_size.y);
in_rect -= GSVector4i(horizontal_offset * t_psm.pgs.x, vertical_offset * t_psm.pgs.y).xyxy();

if (sbw == 0) // Intentionally check this separately
Expand Down Expand Up @@ -1379,6 +1379,18 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
// we want to create a temporary source and load the data from memory.
bool target_bp_hit_outside_valid_area = false;

GSVector4i req_rect = r;

// The read area might be offset but the start of the texture is at the beginning of the space.
req_rect.x = region.HasX() ? region.GetMinX() : req_rect.x;
req_rect.y = region.HasY() ? region.GetMinY() : req_rect.y;

GSVector4i block_boundary_rect = req_rect;
block_boundary_rect.x = block_boundary_rect.x & ~(psm_s.bs.x - 1);
block_boundary_rect.y = block_boundary_rect.y & ~(psm_s.bs.y - 1);
// Round up to the nearst block boundary for lookup to avoid problems due to bilinear and inclusive rects.
block_boundary_rect.z = std::max(req_rect.x + 1, (block_boundary_rect.z + (psm_s.bs.x / 2)) & ~(psm_s.bs.x - 1));
block_boundary_rect.w = std::max(req_rect.y + 1, (block_boundary_rect.w + (psm_s.bs.y / 2)) & ~(psm_s.bs.y - 1));
#ifdef DISABLE_HW_TEXTURE_CACHE
if (0)
#else
Expand All @@ -1389,18 +1401,6 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
const u32 psm = TEX0.PSM;
const u32 bw = TEX0.TBW;

GSVector4i req_rect = r;

// The read area might be offset but the start of the texture is at the beginning of the space.
req_rect.x = region.HasX() ? region.GetMinX() : 0;
req_rect.y = region.HasY() ? region.GetMinY() : 0;

GSVector4i block_boundary_rect = req_rect;
block_boundary_rect.x = block_boundary_rect.x & ~(psm_s.bs.x - 1);
block_boundary_rect.y = block_boundary_rect.y & ~(psm_s.bs.y - 1);
// Round up to the nearst block boundary for lookup to avoid problems due to bilinear and inclusive rects.
block_boundary_rect.z = std::max(req_rect.x + 1, (block_boundary_rect.z + (psm_s.bs.x / 2)) & ~(psm_s.bs.x - 1));
block_boundary_rect.w = std::max(req_rect.y + 1, (block_boundary_rect.w + (psm_s.bs.y / 2)) & ~(psm_s.bs.y - 1));

// Arc the Lad finds the wrong surface here when looking for a depth stencil.
// Since we're currently not caching depth stencils (check ToDo in CreateSource) we should not look for it here.
Expand All @@ -1424,7 +1424,12 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
// it's probable that's more correct than being inside (Tomb Raider Legends + Project Snowblind)
// Vakyrie Profile 2 also has some in draws which get done on a different target due to a slight offset, so we need to make sure we have the newer one.
if (!overlaps || (found_t && (GSState::s_n - dst->m_last_draw) < (GSState::s_n - t->m_last_draw)))
{
if (!overlaps && bp == t->m_TEX0.TBP0)
target_bp_hit_outside_valid_area |= t->HasValidBitsForFormat(psm, req_color, req_alpha, t->m_TEX0.TBW == TEX0.TBW);

continue;
}

// If the BP is offset in to a page and the format does not match, trying to match up the correct position is very difficult since we don't swizzle.
// Tomb Raider Legends does a block level BP in PSMT8 over a C16 target, which is just a nightmare to get right.
Expand Down Expand Up @@ -1820,6 +1825,7 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const

const u32 end_block = GSLocalMemory::GetEndBlockAddress(bp, TEX0.TBW, TEX0.PSM, r);
const u32 adj_bp = (end_block < t->m_TEX0.TBP0 && t->UnwrappedEndBlock() > GS_MAX_BLOCKS) ? (bp + GS_MAX_BLOCKS) : bp;

u32 horz_page_offset = ((adj_bp - t->m_TEX0.TBP0) >> 5) % std::max(1U, t->m_TEX0.TBW);
if (GSConfig.UserHacks_TargetPageRearranging && GSLocalMemory::m_psm[color_psm].bpp != 8 && (((horz_page_offset * 64) + r.width()) > (t->m_TEX0.TBW * 64) || (r.w / psm_s.pgs.y) > 1))
{
Expand All @@ -1829,8 +1835,11 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const

u32 rt_tbw = std::max(1U, t->m_TEX0.TBW);
horz_page_offset = ((adj_bp - t->m_TEX0.TBP0) >> 5) % rt_tbw;
const GSVector2i& page_size = GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs;
// when checking if there's page boundary crossing, we need to check from any offset in to the page, not just the raw height, which could be less than a page.
const u32 height_from_offset = block_boundary_rect.w - (block_boundary_rect.y & ~(page_size.y - 1));

if (GSLocalMemory::m_psm[psm].bpp == GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp && bw != rt_tbw && block_boundary_rect.height() > GSLocalMemory::m_psm[psm].pgs.y)
if (GSLocalMemory::m_psm[psm].bpp == GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp && bw != rt_tbw && height_from_offset > GSLocalMemory::m_psm[psm].pgs.y)
continue;

// Reading 16bit as 32bit, or vice versa (when there isn't a shuffle) isn't really possible and no conversion is done.
Expand Down Expand Up @@ -1909,15 +1918,22 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
if (can_translate)
{
const bool swizzle_match = GSLocalMemory::m_psm[src_psm].depth == GSLocalMemory::m_psm[t->m_TEX0.PSM].depth;
const GSVector2i& page_size = GSLocalMemory::m_psm[t->m_TEX0.PSM].pgs;
const GSVector4i page_mask(GSVector4i((page_size.x - 1), (page_size.y - 1)).xyxy());
rect = new_rect & ~page_mask;
// Round it up so we don't end up with a completely empty rect.
new_rect.z = new_rect.z + (page_size.x - 1);
new_rect.w = new_rect.w + (page_size.y - 1);

rect = new_rect & ~page_mask;
/*if (region.HasEither())
{
rect.x -= region.GetMinX() & ~(page_size.x - 1);
rect.y -= region.GetMinY() & ~(page_size.y - 1);
}*/
if (swizzle_match)
{
rect = TranslateAlignedRectByPage(t->m_TEX0.TBP0, t->m_end_block, rt_tbw, t->m_TEX0.PSM, t->m_valid, adj_bp, src_psm, bw, new_rect);
rect.x -= new_rect.x;
rect.y -= new_rect.y;
rect = TranslateAlignedRectByPage(t->m_TEX0.TBP0, t->m_end_block, rt_tbw, t->m_TEX0.PSM, t->m_valid, adj_bp, src_psm, bw, rect);
rect.x -= new_rect.x & ~(page_size.x - 1);
rect.y -= new_rect.y & ~(page_size.y - 1);
}
else
{
Expand Down Expand Up @@ -1950,6 +1966,8 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
const GSVector4i dirty_rect = t->m_dirty.GetTotalRect(t->m_TEX0, GSVector2i(rect.z, rect.w)).rintersect(rect);
if (dirty_rect.eq(rect))
continue;
if (!dirty_rect.rempty())
t->Update();
}

if (!t->HasValidBitsForFormat(psm, req_color, req_alpha, rt_tbw == TEX0.TBW) && !(possible_shuffle && GSLocalMemory::m_psm[psm].bpp == 16 && GSLocalMemory::m_psm[t->m_TEX0.PSM].bpp == 32))
Expand Down Expand Up @@ -2082,8 +2100,8 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
region = SourceRegion::Create(TEX0, fake_CLAMP);

const GSVector4i custom_offset_rect = TranslateAlignedRectByPage(t, bp, psm, bw, block_boundary_rect);
x_offset = custom_offset_rect.x;
y_offset = custom_offset_rect.y;
x_offset = custom_offset_rect.x - block_boundary_rect.x;
y_offset = custom_offset_rect.y - block_boundary_rect.y;
dst = t;
tex_merge_rt = false;
found_t = true;
Expand Down Expand Up @@ -2306,7 +2324,12 @@ GSTextureCache::Source* GSTextureCache::LookupSource(const bool is_color, const
}
else if (src->m_TEX0.TBP0 > src->m_from_target->m_TEX0.TBP0)
{
GSVector4i dst_offset = TranslateAlignedRectByPage(src->m_from_target, src->m_TEX0.TBP0, src->m_TEX0.PSM, src->m_TEX0.TBW, GSVector4i(0, 0, 1, 1), false);
GSVector4i dst_offset = TranslateAlignedRectByPage(src->m_from_target, src->m_TEX0.TBP0, src->m_TEX0.PSM, src->m_TEX0.TBW, block_boundary_rect, false);
if (dst_offset.x >= block_boundary_rect.x && dst_offset.y >= block_boundary_rect.y)
{
dst_offset.x -= block_boundary_rect.x;
dst_offset.y -= block_boundary_rect.y;
}
src->m_region.bits = 0;
src->m_region.SetX(dst_offset.x, dst_offset.x + (region.HasX() ? std::min(region.GetMaxX(), (1 << TEX0.TW)) : (1 << TEX0.TW)));
src->m_region.SetY(dst_offset.y, dst_offset.y + (region.HasY() ? std::min(region.GetMaxY(), (1 << TEX0.TH)) : (1 << TEX0.TH)));
Expand Down Expand Up @@ -6319,7 +6342,7 @@ GSTextureCache::Source* GSTextureCache::CreateSource(const GIFRegTEX0& TEX0, con
src->m_unscaled_size = dst->m_unscaled_size;
src->m_shared_texture = true;

if (channel_shuffle)
if (channel_shuffle || y_offset < 0 || x_offset < 0)
m_temporary_source = src;
}

Expand Down
Loading