GL: Safer vertex loading (should fix the new zelda crash), added vertex format logging to "show some statistics"

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@1295 8ced0084-cf51-0410-be5f-012b33b47a6e
This commit is contained in:
hrydgard 2008-11-25 21:15:31 +00:00
parent f1ab625f5c
commit 30a5c91c99
7 changed files with 85 additions and 37 deletions

View File

@ -39,6 +39,7 @@
#include "rasterfont.h"
#include "VertexShader.h"
#include "PixelShaderManager.h"
#include "VertexLoaderManager.h"
#include "VertexLoader.h"
#include "XFB.h"
#if !defined(OSX64)
@ -791,7 +792,11 @@ void Renderer::SwapBuffers()
p+=sprintf(p,"Num BP loads (DL): %i\n",stats.thisFrame.numBPLoadsInDL);
p+=sprintf(p,"Num vertex loaders: %i\n",stats.numVertexLoaders);
Renderer::RenderText(st, 20, 20, 0xFF00FFFF);
std::string text = st;
VertexLoaderManager::AppendListToString(&text);
Renderer::RenderText(text.c_str(), 20, 20, 0xFF00FFFF);
}
else
{

View File

@ -23,6 +23,7 @@
#include "Config.h"
#include "Profiler.h"
#include "MemoryUtil.h"
#include "StringUtil.h"
#include "x64Emitter.h"
#include "ABI.h"
@ -110,6 +111,7 @@ void LOADERDECL TexMtx_Write_Short3()
VertexLoader::VertexLoader(const TVtxDesc &vtx_desc, const VAT &vtx_attr)
{
m_numLoadedVertices = 0;
m_VertexSize = 0;
m_numPipelineStages = 0;
m_NativeFmt = new NativeVertexFormat();
@ -135,9 +137,18 @@ void VertexLoader::CompileVertexTranslator()
{
m_VertexSize = 0;
#ifdef USE_JIT
u8 *old_code_ptr = GetWritableCodePtr();
SetCodePtr(m_compiledCode);
ABI_EmitPrologue(4);
// Start loop here
MOV(32, M(&tcIndex), Imm32(0));
MOV(32, M(&colIndex), Imm32(0));
MOV(32, M(&s_texmtxwrite), Imm32(0));
MOV(32, M(&s_texmtxread), Imm32(0));
#endif
// Colors
const int col[2] = {m_VtxDesc.Color0, m_VtxDesc.Color1};
// TextureCoord
@ -450,6 +461,9 @@ void VertexLoader::CompileVertexTranslator()
if (vtx_decl.stride != offset)
PanicAlert("offset/stride mismatch, %i %i", vtx_decl.stride, offset);
#ifdef USE_JIT
// End loop here
// SUB(32, M(&vtxCounter), Imm8(1));
// J_CC(CC_NZ, loop);
ABI_EmitEpilogue(4);
#endif
SetCodePtr(old_code_ptr);
@ -565,6 +579,8 @@ void VertexLoader::RunVertices(int vtx_attr_group, int primitive, int count)
{
DVSTARTPROFILE();
m_numLoadedVertices += count;
// Flush if our vertex format is different from the currently set.
if (g_nativeVertexFmt != NULL && g_nativeVertexFmt != m_NativeFmt)
{
@ -618,15 +634,9 @@ void VertexLoader::RunVertices(int vtx_attr_group, int primitive, int count)
if (VertexManager::GetRemainingSize() < 2 * native_stride)
VertexManager::Flush();
break;
case 0: // quads
granularity = 4;
break;
case 2: // tris
granularity = 3;
break;
case 5: // lines
granularity = 2;
break;
case 0: granularity = 4; break; // quads
case 2: granularity = 3; break; // tris
case 5: granularity = 2; break; // lines
}
int startv = 0, extraverts = 0;
@ -634,7 +644,8 @@ void VertexLoader::RunVertices(int vtx_attr_group, int primitive, int count)
while (v < count)
{
if (VertexManager::GetRemainingSize() < granularity*native_stride) {
int remainingVerts = VertexManager::GetRemainingSize() / native_stride;
if (remainingVerts < granularity) {
INCSTAT(stats.thisFrame.numBufferSplits);
// This buffer full - break current primitive and flush, to switch to the next buffer.
u8* plastptr = VertexManager::s_pCurBufferPointer;
@ -675,22 +686,26 @@ void VertexLoader::RunVertices(int vtx_attr_group, int primitive, int count)
}
startv = v;
}
int remainingPrims = remainingVerts / granularity;
remainingVerts = remainingPrims * granularity;
if (count - v < remainingVerts)
remainingVerts = count - v;
for (int s = 0; s < granularity; s++)
// Clean tight loader loop. Todo - build the loop into the JIT code.
for (int s = 0; s < remainingVerts; s++)
{
tcIndex = 0;
colIndex = 0;
s_texmtxwrite = s_texmtxread = 0;
#ifdef USE_JIT
((void (*)())(void*)m_compiledCode)();
#else
tcIndex = 0;
colIndex = 0;
s_texmtxwrite = s_texmtxread = 0;
for (int i = 0; i < m_numPipelineStages; i++)
m_PipelineStages[i]();
#endif
PRIM_LOG("\n");
v++;
}
v += remainingVerts;
}
if (startv < count)
@ -742,3 +757,18 @@ void VertexLoader::SetVAT(u32 _group0, u32 _group1, u32 _group2)
m_VtxAttr.texCoord[7].Format = vat.g2.Tex7CoordFormat;
m_VtxAttr.texCoord[7].Frac = vat.g2.Tex7Frac;
};
void VertexLoader::AppendToString(std::string *dest) {
static const char *posMode[4] = {
"Invalid",
"Direct",
"Idx8",
"Idx16",
};
static const char *posFormats[5] = {
"u8", "s8", "u16", "s16", "flt",
};
dest->append(StringFromFormat("sz: %i skin: %i Pos: %i %s %s Nrm: %i %s %s - %i vtx\n",
m_VertexSize, m_VtxDesc.PosMatIdx, m_VtxAttr.PosElements ? 3 : 2, posMode[m_VtxDesc.Position], posFormats[m_VtxAttr.PosFormat],
m_VtxAttr.NormalElements, posMode[m_VtxDesc.Normal], posFormats[m_VtxAttr.NormalFormat], m_numLoadedVertices));
}

View File

@ -18,6 +18,8 @@
#ifndef _VERTEXLOADER_H
#define _VERTEXLOADER_H
#include <string>
#include "CPMemory.h"
#include "DataReader.h"
@ -59,6 +61,9 @@ public:
int GetVertexSize() const {return m_VertexSize;}
void RunVertices(int vtx_attr_group, int primitive, int count);
// For debugging / profiling
void AppendToString(std::string *dest);
private:
enum
{
@ -83,6 +88,8 @@ private:
u8 *m_compiledCode;
int m_numLoadedVertices;
void SetupColor(int num, int _iMode, int _iFormat, int _iElements);
void SetupTexCoord(int num, int _iMode, int _iFormat, int _iElements, int _iFrac);

View File

@ -50,6 +50,14 @@ void Shutdown()
g_VertexLoaderMap.clear();
}
void AppendListToString(std::string *dest)
{
for (VertexLoaderMap::iterator iter = g_VertexLoaderMap.begin(); iter != g_VertexLoaderMap.end(); ++iter)
{
iter->second->AppendToString(dest);
}
}
void MarkAllDirty()
{
s_attr_dirty = 0xff;

View File

@ -30,6 +30,9 @@ namespace VertexLoaderManager
int GetVertexSize(int vtx_attr_group);
void RunVertices(int vtx_attr_group, int primitive, int count);
// For debugging
void AppendListToString(std::string *dest);
// TODO - don't expose these like this.
// static u8* s_pCurBufferPointer;
};

View File

@ -85,19 +85,6 @@ void ResetBuffer()
s_vStoredPrimitives.resize(0);
}
void ResetComponents()
{
s_prevcomponents = 0;
glDisableVertexAttribArray(SHADER_POSMTX_ATTRIB);
glDisableClientState(GL_NORMAL_ARRAY);
glDisableVertexAttribArray(SHADER_NORM1_ATTRIB);
glDisableVertexAttribArray(SHADER_NORM2_ATTRIB);
glDisableClientState(GL_COLOR_ARRAY);
glDisableClientState(GL_SECONDARY_COLOR_ARRAY);
for (int i = 0; i < 8; i++)
glDisableClientState(GL_TEXTURE_COORD_ARRAY);
}
int GetRemainingSize()
{
return MAX_BUFFER_SIZE - (int)(s_pCurBufferPointer - s_pBaseBufferPointer);
@ -108,12 +95,21 @@ void AddVertices(int primitive, int numvertices)
_assert_( numvertices > 0 );
ADDSTAT(stats.thisFrame.numPrims, numvertices);
/*
if (s_vStoredPrimitives.size() && s_vStoredPrimitives[s_vStoredPrimitives.size() - 1].first == primitive) {
// Actually, just count potential primitive joins.
// Doesn't seem worth it in Metroid Prime games.
INCSTAT(stats.thisFrame.numPrimitiveJoins);
}*/
if (s_vStoredPrimitives.size() && s_vStoredPrimitives[s_vStoredPrimitives.size() - 1].first == c_primitiveType[primitive]) {
// We can join primitives for free here. Not likely to help much, though, but whatever...
if (c_primitiveType[primitive] == GL_TRIANGLES ||
c_primitiveType[primitive] == GL_LINES ||
c_primitiveType[primitive] == GL_POINTS ||
c_primitiveType[primitive] == GL_QUADS) {
INCSTAT(stats.thisFrame.numPrimitiveJoins);
// Easy join
std::pair<int, int> &last_pair = s_vStoredPrimitives[s_vStoredPrimitives.size() - 1];
last_pair.second += numvertices;
return;
}
// Joining strips is a lot more work but would bring more gain. Not sure if it's worth it though.
}
s_vStoredPrimitives.push_back(std::pair<int, int>(c_primitiveType[primitive], numvertices));
#if defined(_DEBUG) || defined(DEBUGFAST)

View File

@ -30,7 +30,6 @@ bool Init();
void Shutdown();
void ResetBuffer();
void ResetComponents();
void AddVertices(int primitive, int numvertices);
void Flush(); // flushes the current buffer