mirror of
https://github.com/PCSX2/pcsx2.git
synced 2026-11-15 18:50:32 +01:00
GSdx: Reworked the sw renderer texture cache a bit, at page level its correctness was questionable if the base address was not page aligned, so now keeping track of blocks (32x more), may be slower or faster by a few percent (number of blocks vs. finer resolution).
git-svn-id: http://pcsx2.googlecode.com/svn/trunk@1142 96395faa-99c1-11dd-bbfe-3dabce05a288
This commit is contained in:
gabest11 committed
1 parent
e6b9275c13
commit
3202a42578
6 files changed
+554
-413
No files matched your search
+119
-154
@@ -22,18 +22,13 @@
|
||||
#include "StdAfx.h"
|
||||
#include "GSTextureCacheSW.h"
|
||||
|
||||
// static FILE* m_log = NULL;
|
||||
|
||||
GSTextureCacheSW::GSTextureCacheSW(GSState* state)
|
||||
: m_state(state)
|
||||
{
|
||||
// m_log = _tfopen(_T("c:\\log.txt"), _T("w"));
|
||||
}
|
||||
|
||||
GSTextureCacheSW::~GSTextureCacheSW()
|
||||
{
|
||||
// fclose(m_log);
|
||||
|
||||
RemoveAll();
|
||||
}
|
||||
|
||||
@@ -43,20 +38,16 @@ const GSTextureCacheSW::GSTexture* GSTextureCacheSW::Lookup(const GIFRegTEX0& TE
|
||||
|
||||
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
|
||||
|
||||
const CAtlList<GSTexturePage*>& t2p = m_p2t[TEX0.TBP0 >> 5];
|
||||
|
||||
// fprintf(m_log, "lu %05x %d %d (%d) ", TEX0.TBP0, TEX0.TBW, TEX0.PSM, t2p.GetCount());
|
||||
|
||||
// if(r) fprintf(m_log, "(%d %d %d %d) ", r->left, r->top, r->right, r->bottom);
|
||||
const CAtlMap<GSTexture*, bool>& map = m_map[TEX0.TBP0 >> 5];
|
||||
|
||||
GSTexture* t = NULL;
|
||||
|
||||
POSITION pos = t2p.GetHeadPosition();
|
||||
POSITION pos = map.GetStartPosition();
|
||||
|
||||
while(pos)
|
||||
{
|
||||
GSTexture* t2 = t2p.GetNext(pos)->t;
|
||||
|
||||
GSTexture* t2 = map.GetNextKey(pos);
|
||||
|
||||
// if(t2->m_TEX0.TBP0 != TEX0.TBP0 || t2->m_TEX0.TBW != TEX0.TBW || t2->m_TEX0.PSM != TEX0.PSM || t2->m_TEX0.TW != TEX0.TW || t2->m_TEX0.TH != TEX0.TH)
|
||||
if(((t2->m_TEX0.ai32[0] ^ TEX0.ai32[0]) | ((t2->m_TEX0.ai32[1] ^ TEX0.ai32[1]) & 3)) != 0)
|
||||
{
|
||||
@@ -68,8 +59,6 @@ const GSTextureCacheSW::GSTexture* GSTextureCacheSW::Lookup(const GIFRegTEX0& TE
|
||||
continue;
|
||||
}
|
||||
|
||||
// fprintf(m_log, "cache hit\n");
|
||||
|
||||
t = t2;
|
||||
|
||||
t->m_age = 0;
|
||||
@@ -79,8 +68,6 @@ const GSTextureCacheSW::GSTexture* GSTextureCacheSW::Lookup(const GIFRegTEX0& TE
|
||||
|
||||
if(t == NULL)
|
||||
{
|
||||
// fprintf(m_log, "cache miss\n");
|
||||
|
||||
t = new GSTexture(m_state);
|
||||
|
||||
t->m_pos = m_textures.AddTail(t);
|
||||
@@ -91,37 +78,38 @@ const GSTextureCacheSW::GSTexture* GSTextureCacheSW::Lookup(const GIFRegTEX0& TE
|
||||
DWORD bp = TEX0.TBP0;
|
||||
DWORD bw = TEX0.TBW;
|
||||
|
||||
for(int j = 0, y = 0; y < th; j++, y += psm.pgs.cy)
|
||||
CSize s = (bp & 31) == 0 ? psm.pgs : psm.bs;
|
||||
|
||||
for(int y = 0; y < th; y += s.cy)
|
||||
{
|
||||
DWORD page = psm.pgn(0, y, bp, bw);
|
||||
DWORD base = psm.bn(0, y, bp, bw);
|
||||
|
||||
for(int i = 0, x = 0; x < tw && page < MAX_PAGES; i++, x += psm.pgs.cx, page++)
|
||||
for(int x = 0; x < tw; x += s.cx)
|
||||
{
|
||||
GSTexturePage* p = new GSTexturePage();
|
||||
|
||||
p->t = t;
|
||||
p->row = j;
|
||||
p->col = i;
|
||||
DWORD page = (base + psm.blockOffset[x >> 3]) >> 5;
|
||||
|
||||
GSTexturePageEntry* p2te = new GSTexturePageEntry();
|
||||
if(page >= MAX_PAGES)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
p2te->p2t = &m_p2t[page];
|
||||
p2te->pos = m_p2t[page].AddHead(p);
|
||||
|
||||
t->m_p2te.AddTail(p2te);
|
||||
|
||||
t->m_maxpages++;
|
||||
m_map[page].SetAt(t, true);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
if(!t->Update(TEX0, TEXA, r))
|
||||
{
|
||||
printf("!@#$%\n"); // memory allocation may fail if the game is too hungry
|
||||
|
||||
m_textures.RemoveAt(t->m_pos);
|
||||
|
||||
delete t;
|
||||
for(int i = 0; i < MAX_PAGES; i++)
|
||||
{
|
||||
m_map[i].RemoveKey(t);
|
||||
}
|
||||
|
||||
printf("!@#$%\n"); // memory allocation may fail if the game is too hungry
|
||||
delete t;
|
||||
|
||||
return NULL;
|
||||
}
|
||||
@@ -142,18 +130,7 @@ void GSTextureCacheSW::RemoveAll()
|
||||
|
||||
for(int i = 0; i < MAX_PAGES; i++)
|
||||
{
|
||||
CAtlList<GSTexturePage*>& t2p = m_p2t[i];
|
||||
|
||||
ASSERT(t2p.IsEmpty());
|
||||
|
||||
POSITION pos = t2p.GetHeadPosition();
|
||||
|
||||
while(pos)
|
||||
{
|
||||
delete t2p.GetNext(pos);
|
||||
}
|
||||
|
||||
t2p.RemoveAll();
|
||||
m_map[i].RemoveAll();
|
||||
}
|
||||
}
|
||||
|
||||
@@ -167,61 +144,60 @@ void GSTextureCacheSW::IncAge()
|
||||
|
||||
GSTexture* t = m_textures.GetNext(pos);
|
||||
|
||||
if(++t->m_age > 3)
|
||||
if(++t->m_age > 30)
|
||||
{
|
||||
m_textures.RemoveAt(cur);
|
||||
|
||||
for(int i = 0; i < MAX_PAGES; i++)
|
||||
{
|
||||
m_map[i].RemoveKey(t);
|
||||
}
|
||||
|
||||
delete t;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
void GSTextureCacheSW::InvalidateVideoMem(const GIFRegBITBLTBUF& BITBLTBUF, const CRect& r)
|
||||
void GSTextureCacheSW::InvalidateVideoMem(const GIFRegBITBLTBUF& BITBLTBUF, const CRect& rect)
|
||||
{
|
||||
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[BITBLTBUF.DPSM];
|
||||
|
||||
CRect r2;
|
||||
|
||||
r2.left = r.left & ~(psm.pgs.cx - 1);
|
||||
r2.top = r.top & ~(psm.pgs.cy - 1);
|
||||
r2.right = (r.right + (psm.pgs.cx - 1)) & ~(psm.pgs.cx - 1);
|
||||
r2.bottom = (r.bottom + (psm.pgs.cy - 1)) & ~(psm.pgs.cy - 1);
|
||||
|
||||
DWORD bp = BITBLTBUF.DBP;
|
||||
DWORD bw = BITBLTBUF.DBW;
|
||||
|
||||
// fprintf(m_log, "ivm %05x %d %d (%d %d %d %d)\n", bp, bw, BITBLTBUF.DPSM, r2.left, r2.top, r2.right, r2.bottom);
|
||||
CSize s = (bp & 31) == 0 ? psm.pgs : psm.bs;
|
||||
|
||||
for(int y = r2.top; y < r2.bottom; y += psm.pgs.cy)
|
||||
CRect r;
|
||||
|
||||
r.left = rect.left & ~(s.cx - 1);
|
||||
r.top = rect.top & ~(s.cy - 1);
|
||||
r.right = (rect.right + (s.cx - 1)) & ~(s.cx - 1);
|
||||
r.bottom = (rect.bottom + (s.cy - 1)) & ~(s.cy - 1);
|
||||
|
||||
for(int y = r.top; y < r.bottom; y += s.cy)
|
||||
{
|
||||
DWORD page = psm.pgn(r2.left, y, bp, bw);
|
||||
DWORD base = psm.bn(0, y, bp, bw);
|
||||
|
||||
for(int x = r2.left; x < r2.right && page < MAX_PAGES; x += psm.pgs.cx, page++)
|
||||
for(int x = r.left; x < r.right; x += s.cx)
|
||||
{
|
||||
const CAtlList<GSTexturePage*>& t2p = m_p2t[page];
|
||||
DWORD page = (base + psm.blockOffset[x >> 3]) >> 5;
|
||||
|
||||
POSITION pos = t2p.GetHeadPosition();
|
||||
if(page >= MAX_PAGES)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
const CAtlMap<GSTexture*, bool>& map = m_map[page];
|
||||
|
||||
POSITION pos = map.GetStartPosition();
|
||||
|
||||
while(pos)
|
||||
{
|
||||
GSTexturePage* p = t2p.GetNext(pos);
|
||||
GSTexture* t = map.GetNextKey(pos);
|
||||
|
||||
DWORD flag = 1 << p->col;
|
||||
t->m_valid[page] = 0;
|
||||
|
||||
if((p->t->m_valid[p->row] & flag) == 0)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
if(GSUtil::HasSharedBits(BITBLTBUF.DPSM, p->t->m_TEX0.PSM))
|
||||
{
|
||||
p->t->m_valid[p->row] &= ~flag;
|
||||
p->t->m_pages--;
|
||||
|
||||
// fprintf(m_log, "ivm hit %05x %d %d (%d %d) (%d)", p->t->m_TEX0.TBP0, p->t->m_TEX0.TBW, p->t->m_TEX0.PSM, p->row, p->col, p->t->m_pages);
|
||||
// if(p->t->m_pages == 0) fprintf(m_log, " *");
|
||||
// fprintf(m_log, "\n");
|
||||
}
|
||||
t->m_complete = false;
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -233,10 +209,9 @@ GSTextureCacheSW::GSTexture::GSTexture(GSState* state)
|
||||
: m_state(state)
|
||||
, m_buff(NULL)
|
||||
, m_tw(0)
|
||||
, m_maxpages(0)
|
||||
, m_pages(0)
|
||||
, m_pos(NULL)
|
||||
, m_age(0)
|
||||
, m_pos(NULL)
|
||||
, m_complete(false)
|
||||
{
|
||||
memset(m_valid, 0, sizeof(m_valid));
|
||||
}
|
||||
@@ -247,30 +222,11 @@ GSTextureCacheSW::GSTexture::~GSTexture()
|
||||
{
|
||||
_aligned_free(m_buff);
|
||||
}
|
||||
|
||||
POSITION pos = m_p2te.GetHeadPosition();
|
||||
|
||||
while(pos)
|
||||
{
|
||||
GSTexturePageEntry* p2te = m_p2te.GetNext(pos);
|
||||
|
||||
GSTexturePage* p = p2te->p2t->GetAt(p2te->pos);
|
||||
|
||||
ASSERT(p->t == this);
|
||||
|
||||
delete p;
|
||||
|
||||
p2te->p2t->RemoveAt(p2te->pos);
|
||||
|
||||
delete p2te;
|
||||
}
|
||||
|
||||
m_p2te.RemoveAll();
|
||||
}
|
||||
|
||||
bool GSTextureCacheSW::GSTexture::Update(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const CRect* r)
|
||||
bool GSTextureCacheSW::GSTexture::Update(const GIFRegTEX0& TEX0, const GIFRegTEXA& TEXA, const CRect* rect)
|
||||
{
|
||||
if(m_pages == m_maxpages)
|
||||
if(m_complete)
|
||||
{
|
||||
return true;
|
||||
}
|
||||
@@ -282,16 +238,16 @@ bool GSTextureCacheSW::GSTexture::Update(const GIFRegTEX0& TEX0, const GIFRegTEX
|
||||
|
||||
const GSLocalMemory::psm_t& psm = GSLocalMemory::m_psm[TEX0.PSM];
|
||||
|
||||
int tw = 1 << TEX0.TW;
|
||||
int th = 1 << TEX0.TH;
|
||||
DWORD bp = TEX0.TBP0;
|
||||
DWORD bw = TEX0.TBW;
|
||||
|
||||
if(tw < psm.bs.cx) tw = psm.bs.cx;
|
||||
if(th < psm.bs.cy) th = psm.bs.cy;
|
||||
CSize s = psm.bs;
|
||||
|
||||
int tw = max(1 << TEX0.TW, s.cx);
|
||||
int th = max(1 << TEX0.TH, s.cy);
|
||||
|
||||
if(m_buff == NULL)
|
||||
{
|
||||
// fprintf(m_log, "up new (%d %d)\n", tw, th);
|
||||
|
||||
m_buff = _aligned_malloc(tw * th * sizeof(DWORD), 16);
|
||||
|
||||
if(m_buff == NULL)
|
||||
@@ -302,75 +258,84 @@ bool GSTextureCacheSW::GSTexture::Update(const GIFRegTEX0& TEX0, const GIFRegTEX
|
||||
m_tw = max(psm.pal > 0 ? 5 : 3, TEX0.TW); // makes one row 32 bytes at least, matches the smallest block size that is allocated above for m_buff
|
||||
}
|
||||
|
||||
CRect r2;
|
||||
CRect r(0, 0, tw, th);
|
||||
|
||||
if(r)
|
||||
if(rect)
|
||||
{
|
||||
r2.left = r->left & ~(psm.pgs.cx - 1);
|
||||
r2.top = r->top & ~(psm.pgs.cy - 1);
|
||||
r2.right = (r->right + (psm.pgs.cx - 1)) & ~(psm.pgs.cx - 1);
|
||||
r2.bottom = (r->bottom + (psm.pgs.cy - 1)) & ~(psm.pgs.cy - 1);
|
||||
r.left = rect->left & ~(s.cx - 1);
|
||||
r.top = rect->top & ~(s.cy - 1);
|
||||
r.right = (rect->right + (s.cx - 1)) & ~(s.cx - 1);
|
||||
r.bottom = (rect->bottom + (s.cy - 1)) & ~(s.cy - 1);
|
||||
}
|
||||
|
||||
// TODO
|
||||
if(r.left == 0 && r.top == 0 && r.right == tw && r.bottom == th)
|
||||
{
|
||||
m_complete = true; // lame, but better than nothing
|
||||
}
|
||||
|
||||
GSLocalMemory::readTexture rt = psm.pal > 0 ? psm.rtxP : psm.rtx;
|
||||
GSLocalMemory::readTextureBlock rtxb = psm.rtxbP;
|
||||
|
||||
int bytes = psm.pal > 0 ? 1 : 4;
|
||||
|
||||
BYTE* dst = (BYTE*)m_buff;
|
||||
|
||||
DWORD pitch = (1 << m_tw) * bytes;
|
||||
DWORD mask = pitch - 1;
|
||||
|
||||
for(int j = 0, y = 0; y < th; j++, y += psm.pgs.cy, dst += pitch * psm.pgs.cy)
|
||||
BYTE* dst = (BYTE*)m_buff + pitch * r.top;
|
||||
|
||||
DWORD blocks = 0;
|
||||
|
||||
for(int y = r.top, o = pitch * s.cy; y < r.bottom; y += s.cy, dst += o)
|
||||
{
|
||||
if(m_valid[j] == mask)
|
||||
DWORD base = psm.bn(0, y, bp, bw);
|
||||
|
||||
for(int x = r.left; x < r.right; x += s.cx)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
DWORD block = base + psm.blockOffset[x >> 3];
|
||||
|
||||
if(r)
|
||||
{
|
||||
if(y < r2.top) continue;
|
||||
if(y >= r2.bottom) break;
|
||||
}
|
||||
|
||||
DWORD page = psm.pgn(0, y, TEX0.TBP0, TEX0.TBW);
|
||||
|
||||
for(int i = 0, x = 0; x < tw && page < MAX_PAGES; i++, x += psm.pgs.cx, page++)
|
||||
{
|
||||
if(r)
|
||||
{
|
||||
if(x < r2.left) continue;
|
||||
if(x >= r2.right) break;
|
||||
}
|
||||
|
||||
DWORD flag = 1 << i;
|
||||
|
||||
if(m_valid[j] & flag)
|
||||
if(block >= MAX_BLOCKS)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
m_valid[j] |= flag;
|
||||
m_pages++;
|
||||
DWORD row = block >> 5;
|
||||
DWORD col = 1 << (block & 31);
|
||||
|
||||
ASSERT(m_pages <= m_maxpages);
|
||||
if(m_valid[row] & col)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
CRect r;
|
||||
|
||||
r.left = x;
|
||||
r.top = y;
|
||||
r.right = min(x + psm.pgs.cx, tw);
|
||||
r.bottom = min(y + psm.pgs.cy, th);
|
||||
// unfortunatelly a block may be part of the same texture multiple times at different places (when (1 << tw) > (tbw << 6), ex. 1024 > 640),
|
||||
// so just can't set the block's bit to valid in one pass, even if 99.9% of the games don't address the repeated part at the right side
|
||||
|
||||
// fprintf(m_log, "up fetch (%d %d) (%d %d %d %d)\n", j, i, r.left, r.top, r.right, r.bottom);
|
||||
// m_valid[row] |= col;
|
||||
|
||||
(mem.*rt)(r, &dst[x * bytes], pitch, TEX0, TEXA);
|
||||
(mem.*rtxb)(block, &dst[x * bytes], pitch, TEXA);
|
||||
|
||||
m_state->m_perfmon.Put(GSPerfMon::Unswizzle, r.Width() * r.Height() * bytes);
|
||||
blocks++;
|
||||
}
|
||||
}
|
||||
|
||||
m_state->m_perfmon.Put(GSPerfMon::Unswizzle, s.cx * s.cy * bytes * blocks);
|
||||
|
||||
for(int y = r.top; y < r.bottom; y += s.cy)
|
||||
{
|
||||
DWORD base = psm.bn(0, y, bp, bw);
|
||||
|
||||
for(int x = r.left; x < r.right; x += s.cx)
|
||||
{
|
||||
DWORD block = base + psm.blockOffset[x >> 3];
|
||||
|
||||
if(block >= MAX_BLOCKS)
|
||||
{
|
||||
continue;
|
||||
}
|
||||
|
||||
DWORD row = block >> 5;
|
||||
DWORD col = 1 << (block & 31);
|
||||
|
||||
m_valid[row] |= col;
|
||||
}
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user