dolphin/Source/Core/VideoCommon/VertexShaderManager.cpp

// Copyright 2013 Dolphin Emulator Project
// Licensed under GPLv2
// Refer to the license.txt file included.

#include <cmath>
#include <sstream>

#include "Common/Common.h"
#include "Common/MathUtil.h"
#include "VideoCommon/BPMemory.h"
#include "VideoCommon/CPMemory.h"
#include "VideoCommon/RenderBase.h"
#include "VideoCommon/Statistics.h"
#include "VideoCommon/VertexManagerBase.h"
#include "VideoCommon/VertexShaderGen.h"
#include "VideoCommon/VertexShaderManager.h"
#include "VideoCommon/VideoCommon.h"
#include "VideoCommon/VideoConfig.h"
#include "VideoCommon/XFMemory.h"

float GC_ALIGNED16(g_fProjectionMatrix[16]);

// track changes
static bool bTexMatricesChanged[2], bPosNormalMatrixChanged, bProjectionChanged, bViewportChanged;
static int nMaterialsChanged;
static int nTransformMatricesChanged[2]; // min,max
static int nNormalMatricesChanged[2]; // min,max
static int nPostTransformMatricesChanged[2]; // min,max
static int nLightsChanged[2]; // min,max

static Matrix44 s_viewportCorrection;
static Matrix33 s_viewRotationMatrix;
static Matrix33 s_viewInvRotationMatrix;
static float s_fViewTranslationVector[3];
static float s_fViewRotation[2];

VertexShaderConstants VertexShaderManager::constants;
bool VertexShaderManager::dirty;

struct ProjectionHack
{
	float sign;
	float value;
	ProjectionHack() { }
	ProjectionHack(float new_sign, float new_value)
		: sign(new_sign), value(new_value) {}
};

namespace
{
// Control Variables
static ProjectionHack g_ProjHack1;
static ProjectionHack g_ProjHack2;
} // Namespace

float PHackValue(std::string sValue)
{
	float f = 0;
	bool fp = false;
	const char *cStr = sValue.c_str();
	char *c = new char[strlen(cStr)+1];
	std::istringstream sTof("");

	for (unsigned int i=0; i<=strlen(cStr); ++i)
	{
		if (i == 20)
		{
			c[i] = '\0';
			break;
		}

		c[i] = (cStr[i] == ',') ? '.' : *(cStr+i);
		if (c[i] == '.')
			fp = true;
	}

	cStr = c;
	sTof.str(cStr);
	sTof >> f;

	if (!fp)
		f /= 0xF4240;

	delete [] c;
	return f;
}

void UpdateProjectionHack(int iPhackvalue[], std::string sPhackvalue[])
{
	float fhackvalue1 = 0, fhackvalue2 = 0;
	float fhacksign1 = 1.0, fhacksign2 = 1.0;
	const char *sTemp[2];

	if (iPhackvalue[0] == 1)
	{
		NOTICE_LOG(VIDEO, "\t\t--- Orthographic Projection Hack ON ---");

		fhacksign1 *= (iPhackvalue[1] == 1) ? -1.0f : fhacksign1;
		sTemp[0] = (iPhackvalue[1] == 1) ? " * (-1)" : "";
		fhacksign2 *= (iPhackvalue[2] == 1) ? -1.0f : fhacksign2;
		sTemp[1] = (iPhackvalue[2] == 1) ? " * (-1)" : "";

		fhackvalue1 = PHackValue(sPhackvalue[0]);
		NOTICE_LOG(VIDEO, "- zNear Correction = (%f + zNear)%s", fhackvalue1, sTemp[0]);

		fhackvalue2 = PHackValue(sPhackvalue[1]);
		NOTICE_LOG(VIDEO, "- zFar Correction =  (%f + zFar)%s", fhackvalue2, sTemp[1]);

	}

	// Set the projections hacks
	g_ProjHack1 = ProjectionHack(fhacksign1, fhackvalue1);
	g_ProjHack2 = ProjectionHack(fhacksign2, fhackvalue2);
}


// Viewport correction:
// In D3D, the viewport rectangle must fit within the render target.
// Say you want a viewport at (ix, iy) with size (iw, ih),
// but your viewport must be clamped at (ax, ay) with size (aw, ah).
// Just multiply the projection matrix with the following to get the same
// effect:
// [   (iw/aw)         0     0    ((iw - 2*(ax-ix)) / aw - 1)   ]
// [         0   (ih/ah)     0   ((-ih + 2*(ay-iy)) / ah + 1)   ]
// [         0         0     1                              0   ]
// [         0         0     0                              1   ]
static void ViewportCorrectionMatrix(Matrix44& result)
{
	int scissorXOff = bpmem.scissorOffset.x * 2;
	int scissorYOff = bpmem.scissorOffset.y * 2;

	// TODO: ceil, floor or just cast to int?
	// TODO: Directly use the floats instead of rounding them?
	float intendedX = xfregs.viewport.xOrig - xfregs.viewport.wd - scissorXOff;
	float intendedY = xfregs.viewport.yOrig + xfregs.viewport.ht - scissorYOff;
	float intendedWd = 2.0f * xfregs.viewport.wd;
	float intendedHt = -2.0f * xfregs.viewport.ht;
	
	if (intendedWd < 0.f)
	{
		intendedX += intendedWd;
		intendedWd = -intendedWd;
	}
	if (intendedHt < 0.f)
	{
		intendedY += intendedHt;
		intendedHt = -intendedHt;
	}

	// fit to EFB size
	float X = (intendedX >= 0.f) ? intendedX : 0.f;
	float Y = (intendedY >= 0.f) ? intendedY : 0.f;
	float Wd = (X + intendedWd <= EFB_WIDTH) ? intendedWd : (EFB_WIDTH - X);
	float Ht = (Y + intendedHt <= EFB_HEIGHT) ? intendedHt : (EFB_HEIGHT - Y);
	
	Matrix44::LoadIdentity(result);
	if (Wd == 0 || Ht == 0)
		return;
	
	result.data[4*0+0] = intendedWd / Wd;
	result.data[4*0+3] = (intendedWd - 2.f * (X - intendedX)) / Wd - 1.f;
	result.data[4*1+1] = intendedHt / Ht;
	result.data[4*1+3] = (-intendedHt + 2.f * (Y - intendedY)) / Ht + 1.f;
}

void VertexShaderManager::Init()
{
	Dirty();

	memset(&xfregs, 0, sizeof(xfregs));
	memset(xfmem, 0, sizeof(xfmem));
	memset(&constants, 0 , sizeof(constants));
	ResetView();

	// TODO: should these go inside ResetView()?
	Matrix44::LoadIdentity(s_viewportCorrection);
	memset(g_fProjectionMatrix, 0, sizeof(g_fProjectionMatrix));
	for (int i = 0; i < 4; ++i)
		g_fProjectionMatrix[i*5] = 1.0f;
}

void VertexShaderManager::Shutdown()
{
}

void VertexShaderManager::Dirty()
{
	nTransformMatricesChanged[0] = 0;
	nTransformMatricesChanged[1] = 256;

	nNormalMatricesChanged[0] = 0;
	nNormalMatricesChanged[1] = 96;

	nPostTransformMatricesChanged[0] = 0;
	nPostTransformMatricesChanged[1] = 256;

	nLightsChanged[0] = 0;
	nLightsChanged[1] = 0x80;

	bPosNormalMatrixChanged = true;
	bTexMatricesChanged[0] = true;
	bTexMatricesChanged[1] = true;

	bProjectionChanged = true;

	nMaterialsChanged = 15;

	dirty = true;
}

// Syncs the shader constant buffers with xfmem
// TODO: A cleaner way to control the matrices without making a mess in the parameters field
void VertexShaderManager::SetConstants()
{
	if (nTransformMatricesChanged[0] >= 0)
	{
		int startn = nTransformMatricesChanged[0] / 4;
		int endn = (nTransformMatricesChanged[1] + 3) / 4;
		memcpy(constants.transformmatrices[startn], &xfmem[startn * 4], (endn - startn) * 16);
		dirty = true;
		nTransformMatricesChanged[0] = nTransformMatricesChanged[1] = -1;
	}

	if (nNormalMatricesChanged[0] >= 0)
	{
		int startn = nNormalMatricesChanged[0] / 3;
		int endn = (nNormalMatricesChanged[1] + 2) / 3;
		for (int i=startn; i<endn; i++)
		{
			memcpy(constants.normalmatrices[i], &xfmem[XFMEM_NORMALMATRICES + 3*i], 12);
		}
		dirty = true;
		nNormalMatricesChanged[0] = nNormalMatricesChanged[1] = -1;
	}

	if (nPostTransformMatricesChanged[0] >= 0)
	{
		int startn = nPostTransformMatricesChanged[0] / 4;
		int endn = (nPostTransformMatricesChanged[1] + 3 ) / 4;
		memcpy(constants.posttransformmatrices[startn], &xfmem[XFMEM_POSTMATRICES + startn * 4], (endn - startn) * 16);
		dirty = true;
		nPostTransformMatricesChanged[0] = nPostTransformMatricesChanged[1] = -1;
	}

	if (nLightsChanged[0] >= 0)
	{
		// TODO: Outdated comment
		// lights don't have a 1 to 1 mapping, the color component needs to be converted to 4 floats
		int istart = nLightsChanged[0] / 0x10;
		int iend = (nLightsChanged[1] + 15) / 0x10;
		const float* xfmemptr = (const float*)&xfmem[0x10 * istart + XFMEM_LIGHTS];

		for (int i = istart; i < iend; ++i)
		{
			u32 color = *(const u32*)(xfmemptr + 3);
			constants.light_colors[i][0] = (color >> 24) & 0xFF;
			constants.light_colors[i][1] = (color >> 16) & 0xFF;
			constants.light_colors[i][2] = (color >> 8)  & 0xFF;
			constants.light_colors[i][3] = (color)       & 0xFF;
			xfmemptr += 4;

			for (int j = 0; j < 4; ++j, xfmemptr += 3)
			{
				if (j == 1 &&
					fabs(xfmemptr[0]) < 0.00001f &&
					fabs(xfmemptr[1]) < 0.00001f &&
					fabs(xfmemptr[2]) < 0.00001f)
				{
					// dist attenuation, make sure not equal to 0!!!
					constants.lights[4*i+j][0] = 0.00001f;
				}
				else
					constants.lights[4*i+j][0] = xfmemptr[0];
				constants.lights[4*i+j][1] = xfmemptr[1];
				constants.lights[4*i+j][2] = xfmemptr[2];
			}
		}
		dirty = true;

		nLightsChanged[0] = nLightsChanged[1] = -1;
	}

	if (nMaterialsChanged)
	{
		for (int i = 0; i < 2; ++i)
		{
			if (nMaterialsChanged & (1 << i))
			{
				u32 data = *(xfregs.ambColor + i);
				constants.materials[i][0] = (data >> 24) & 0xFF;
				constants.materials[i][1] = (data >> 16) & 0xFF;
				constants.materials[i][2] = (data >>  8) & 0xFF;
				constants.materials[i][3] =  data        & 0xFF;
			}
		}

		for (int i = 0; i < 2; ++i)
		{
			if (nMaterialsChanged & (1 << (i + 2)))
			{
				u32 data = *(xfregs.matColor + i);
				constants.materials[i+2][0] = (data >> 24) & 0xFF;
				constants.materials[i+2][1] = (data >> 16) & 0xFF;
				constants.materials[i+2][2] = (data >>  8) & 0xFF;
				constants.materials[i+2][3] =  data        & 0xFF;
			}
		}
		dirty = true;

		nMaterialsChanged = 0;
	}

	if (bPosNormalMatrixChanged)
	{
		bPosNormalMatrixChanged = false;

		const float *pos = (const float *)xfmem + MatrixIndexA.PosNormalMtxIdx * 4;
		const float *norm = (const float *)xfmem + XFMEM_NORMALMATRICES + 3 * (MatrixIndexA.PosNormalMtxIdx & 31);

		memcpy(constants.posnormalmatrix, pos, 3*16);
		memcpy(constants.posnormalmatrix[3], norm, 12);
		memcpy(constants.posnormalmatrix[4], norm+3, 12);
		memcpy(constants.posnormalmatrix[5], norm+6, 12);
		dirty = true;
	}

	if (bTexMatricesChanged[0])
	{
		bTexMatricesChanged[0] = false;
		const float *fptrs[] =
		{
			(const float *)xfmem + MatrixIndexA.Tex0MtxIdx * 4, (const float *)xfmem + MatrixIndexA.Tex1MtxIdx * 4,
			(const float *)xfmem + MatrixIndexA.Tex2MtxIdx * 4, (const float *)xfmem + MatrixIndexA.Tex3MtxIdx * 4
		};

		for (int i = 0; i < 4; ++i)
		{
			memcpy(constants.texmatrices[3*i], fptrs[i], 3*16);
		}
		dirty = true;
	}

	if (bTexMatricesChanged[1])
	{
		bTexMatricesChanged[1] = false;
		const float *fptrs[] = {
			(const float *)xfmem + MatrixIndexB.Tex4MtxIdx * 4, (const float *)xfmem + MatrixIndexB.Tex5MtxIdx * 4,
			(const float *)xfmem + MatrixIndexB.Tex6MtxIdx * 4, (const float *)xfmem + MatrixIndexB.Tex7MtxIdx * 4
		};

		for (int i = 0; i < 4; ++i)
		{
			memcpy(constants.texmatrices[3*i+12], fptrs[i], 3*16);
		}
		dirty = true;
	}

	if (bViewportChanged)
	{
		bViewportChanged = false;
		constants.depthparams[0] = xfregs.viewport.farZ / 16777216.0f;
		constants.depthparams[1] = xfregs.viewport.zRange / 16777216.0f;
		dirty = true;
		// This is so implementation-dependent that we can't have it here.
		g_renderer->SetViewport();
		
		// Update projection if the viewport isn't 1:1 useable
		if (!g_ActiveConfig.backend_info.bSupportsOversizedViewports)
		{
			ViewportCorrectionMatrix(s_viewportCorrection);
			bProjectionChanged = true;
		}
	}

	if (bProjectionChanged)
	{
		bProjectionChanged = false;

		float *rawProjection = xfregs.projection.rawProjection;

		switch (xfregs.projection.type)
		{
		case GX_PERSPECTIVE:

			g_fProjectionMatrix[0] = rawProjection[0] * g_ActiveConfig.fAspectRatioHackW;
			g_fProjectionMatrix[1] = 0.0f;
			g_fProjectionMatrix[2] = rawProjection[1];
			g_fProjectionMatrix[3] = 0.0f;

			g_fProjectionMatrix[4] = 0.0f;
			g_fProjectionMatrix[5] = rawProjection[2] * g_ActiveConfig.fAspectRatioHackH;
			g_fProjectionMatrix[6] = rawProjection[3];
			g_fProjectionMatrix[7] = 0.0f;

			g_fProjectionMatrix[8] = 0.0f;
			g_fProjectionMatrix[9] = 0.0f;
			g_fProjectionMatrix[10] = rawProjection[4];

			g_fProjectionMatrix[11] = rawProjection[5];

			g_fProjectionMatrix[12] = 0.0f;
			g_fProjectionMatrix[13] = 0.0f;
			// donkopunchstania suggested the GC GPU might round differently
			// He had thus changed this to -(1 + epsilon) to fix clipping issues.
			// I (neobrain) don't think his conjecture is true and thus reverted his change.
			g_fProjectionMatrix[14] = -1.0f;
			g_fProjectionMatrix[15] = 0.0f;

			SETSTAT_FT(stats.gproj_0, g_fProjectionMatrix[0]);
			SETSTAT_FT(stats.gproj_1, g_fProjectionMatrix[1]);
			SETSTAT_FT(stats.gproj_2, g_fProjectionMatrix[2]);
			SETSTAT_FT(stats.gproj_3, g_fProjectionMatrix[3]);
			SETSTAT_FT(stats.gproj_4, g_fProjectionMatrix[4]);
			SETSTAT_FT(stats.gproj_5, g_fProjectionMatrix[5]);
			SETSTAT_FT(stats.gproj_6, g_fProjectionMatrix[6]);
			SETSTAT_FT(stats.gproj_7, g_fProjectionMatrix[7]);
			SETSTAT_FT(stats.gproj_8, g_fProjectionMatrix[8]);
			SETSTAT_FT(stats.gproj_9, g_fProjectionMatrix[9]);
			SETSTAT_FT(stats.gproj_10, g_fProjectionMatrix[10]);
			SETSTAT_FT(stats.gproj_11, g_fProjectionMatrix[11]);
			SETSTAT_FT(stats.gproj_12, g_fProjectionMatrix[12]);
			SETSTAT_FT(stats.gproj_13, g_fProjectionMatrix[13]);
			SETSTAT_FT(stats.gproj_14, g_fProjectionMatrix[14]);
			SETSTAT_FT(stats.gproj_15, g_fProjectionMatrix[15]);
			break;

		case GX_ORTHOGRAPHIC:

			g_fProjectionMatrix[0] = rawProjection[0];
			g_fProjectionMatrix[1] = 0.0f;
			g_fProjectionMatrix[2] = 0.0f;
			g_fProjectionMatrix[3] = rawProjection[1];

			g_fProjectionMatrix[4] = 0.0f;
			g_fProjectionMatrix[5] = rawProjection[2];
			g_fProjectionMatrix[6] = 0.0f;
			g_fProjectionMatrix[7] = rawProjection[3];

			g_fProjectionMatrix[8] = 0.0f;
			g_fProjectionMatrix[9] = 0.0f;
			g_fProjectionMatrix[10] = (g_ProjHack1.value + rawProjection[4]) * ((g_ProjHack1.sign == 0) ? 1.0f : g_ProjHack1.sign);
			g_fProjectionMatrix[11] = (g_ProjHack2.value + rawProjection[5]) * ((g_ProjHack2.sign == 0) ? 1.0f : g_ProjHack2.sign);

			g_fProjectionMatrix[12] = 0.0f;
			g_fProjectionMatrix[13] = 0.0f;

			g_fProjectionMatrix[14] = 0.0f;
			g_fProjectionMatrix[15] = 1.0f;

			SETSTAT_FT(stats.g2proj_0, g_fProjectionMatrix[0]);
			SETSTAT_FT(stats.g2proj_1, g_fProjectionMatrix[1]);
			SETSTAT_FT(stats.g2proj_2, g_fProjectionMatrix[2]);
			SETSTAT_FT(stats.g2proj_3, g_fProjectionMatrix[3]);
			SETSTAT_FT(stats.g2proj_4, g_fProjectionMatrix[4]);
			SETSTAT_FT(stats.g2proj_5, g_fProjectionMatrix[5]);
			SETSTAT_FT(stats.g2proj_6, g_fProjectionMatrix[6]);
			SETSTAT_FT(stats.g2proj_7, g_fProjectionMatrix[7]);
			SETSTAT_FT(stats.g2proj_8, g_fProjectionMatrix[8]);
			SETSTAT_FT(stats.g2proj_9, g_fProjectionMatrix[9]);
			SETSTAT_FT(stats.g2proj_10, g_fProjectionMatrix[10]);
			SETSTAT_FT(stats.g2proj_11, g_fProjectionMatrix[11]);
			SETSTAT_FT(stats.g2proj_12, g_fProjectionMatrix[12]);
			SETSTAT_FT(stats.g2proj_13, g_fProjectionMatrix[13]);
			SETSTAT_FT(stats.g2proj_14, g_fProjectionMatrix[14]);
			SETSTAT_FT(stats.g2proj_15, g_fProjectionMatrix[15]);
			SETSTAT_FT(stats.proj_0, rawProjection[0]);
			SETSTAT_FT(stats.proj_1, rawProjection[1]);
			SETSTAT_FT(stats.proj_2, rawProjection[2]);
			SETSTAT_FT(stats.proj_3, rawProjection[3]);
			SETSTAT_FT(stats.proj_4, rawProjection[4]);
			SETSTAT_FT(stats.proj_5, rawProjection[5]);
			break;

		default:
			ERROR_LOG(VIDEO, "Unknown projection type: %d", xfregs.projection.type);
		}

		PRIM_LOG("Projection: %f %f %f %f %f %f\n", rawProjection[0], rawProjection[1], rawProjection[2], rawProjection[3], rawProjection[4], rawProjection[5]);

		if ((g_ActiveConfig.bFreeLook || g_ActiveConfig.bAnaglyphStereo ) && xfregs.projection.type == GX_PERSPECTIVE)
		{
			Matrix44 mtxA;
			Matrix44 mtxB;
			Matrix44 viewMtx;

			Matrix44::Translate(mtxA, s_fViewTranslationVector);
			Matrix44::LoadMatrix33(mtxB, s_viewRotationMatrix);
			Matrix44::Multiply(mtxB, mtxA, viewMtx); // view = rotation x translation
			Matrix44::Set(mtxB, g_fProjectionMatrix);
			Matrix44::Multiply(mtxB, viewMtx, mtxA); // mtxA = projection x view
			Matrix44::Multiply(s_viewportCorrection, mtxA, mtxB); // mtxB = viewportCorrection x mtxA
			memcpy(constants.projection, mtxB.data, 4*16);
		}
		else
		{
			Matrix44 projMtx;
			Matrix44::Set(projMtx, g_fProjectionMatrix);

			Matrix44 correctedMtx;
			Matrix44::Multiply(s_viewportCorrection, projMtx, correctedMtx);
			memcpy(constants.projection, correctedMtx.data, 4*16);
		}
		dirty = true;
	}
}

void VertexShaderManager::InvalidateXFRange(int start, int end)
{
	if (((u32)start >= (u32)MatrixIndexA.PosNormalMtxIdx * 4 &&
		 (u32)start <  (u32)MatrixIndexA.PosNormalMtxIdx * 4 + 12) ||
		((u32)start >= XFMEM_NORMALMATRICES + ((u32)MatrixIndexA.PosNormalMtxIdx & 31) * 3 &&
		 (u32)start <  XFMEM_NORMALMATRICES + ((u32)MatrixIndexA.PosNormalMtxIdx & 31) * 3 + 9))
	{
		bPosNormalMatrixChanged = true;
	}

	if (((u32)start >= (u32)MatrixIndexA.Tex0MtxIdx*4 && (u32)start < (u32)MatrixIndexA.Tex0MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexA.Tex1MtxIdx*4 && (u32)start < (u32)MatrixIndexA.Tex1MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexA.Tex2MtxIdx*4 && (u32)start < (u32)MatrixIndexA.Tex2MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexA.Tex3MtxIdx*4 && (u32)start < (u32)MatrixIndexA.Tex3MtxIdx*4+12))
	{
		bTexMatricesChanged[0] = true;
	}

	if (((u32)start >= (u32)MatrixIndexB.Tex4MtxIdx*4 && (u32)start < (u32)MatrixIndexB.Tex4MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexB.Tex5MtxIdx*4 && (u32)start < (u32)MatrixIndexB.Tex5MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexB.Tex6MtxIdx*4 && (u32)start < (u32)MatrixIndexB.Tex6MtxIdx*4+12) ||
		((u32)start >= (u32)MatrixIndexB.Tex7MtxIdx*4 && (u32)start < (u32)MatrixIndexB.Tex7MtxIdx*4+12))
	{
		bTexMatricesChanged[1] = true;
	}

	if (start < XFMEM_POSMATRICES_END)
	{
		if (nTransformMatricesChanged[0] == -1)
		{
			nTransformMatricesChanged[0] = start;
			nTransformMatricesChanged[1] = end>XFMEM_POSMATRICES_END?XFMEM_POSMATRICES_END:end;
		}
		else
		{
			if (nTransformMatricesChanged[0] > start) nTransformMatricesChanged[0] = start;
			if (nTransformMatricesChanged[1] < end) nTransformMatricesChanged[1] = end>XFMEM_POSMATRICES_END?XFMEM_POSMATRICES_END:end;
		}
	}

	if (start < XFMEM_NORMALMATRICES_END && end > XFMEM_NORMALMATRICES)
	{
		int _start = start < XFMEM_NORMALMATRICES ? 0 : start-XFMEM_NORMALMATRICES;
		int _end = end < XFMEM_NORMALMATRICES_END ? end-XFMEM_NORMALMATRICES : XFMEM_NORMALMATRICES_END-XFMEM_NORMALMATRICES;

		if (nNormalMatricesChanged[0] == -1)
		{
			nNormalMatricesChanged[0] = _start;
			nNormalMatricesChanged[1] = _end;
		}
		else
		{
			if (nNormalMatricesChanged[0] > _start) nNormalMatricesChanged[0] = _start;
			if (nNormalMatricesChanged[1] < _end) nNormalMatricesChanged[1] = _end;
		}
	}

	if (start < XFMEM_POSTMATRICES_END && end > XFMEM_POSTMATRICES)
	{
		int _start = start < XFMEM_POSTMATRICES ? XFMEM_POSTMATRICES : start-XFMEM_POSTMATRICES;
		int _end = end < XFMEM_POSTMATRICES_END ? end-XFMEM_POSTMATRICES : XFMEM_POSTMATRICES_END-XFMEM_POSTMATRICES;

		if (nPostTransformMatricesChanged[0] == -1)
		{
			nPostTransformMatricesChanged[0] = _start;
			nPostTransformMatricesChanged[1] = _end;
		}
		else
		{
			if (nPostTransformMatricesChanged[0] > _start) nPostTransformMatricesChanged[0] = _start;
			if (nPostTransformMatricesChanged[1] < _end) nPostTransformMatricesChanged[1] = _end;
		}
	}

	if (start < XFMEM_LIGHTS_END && end > XFMEM_LIGHTS)
	{
		int _start = start < XFMEM_LIGHTS ? XFMEM_LIGHTS : start-XFMEM_LIGHTS;
		int _end = end < XFMEM_LIGHTS_END ? end-XFMEM_LIGHTS : XFMEM_LIGHTS_END-XFMEM_LIGHTS;

		if (nLightsChanged[0] == -1 )
		{
			nLightsChanged[0] = _start;
			nLightsChanged[1] = _end;
		}
		else
		{
			if (nLightsChanged[0] > _start) nLightsChanged[0] = _start;
			if (nLightsChanged[1] < _end)   nLightsChanged[1] = _end;
		}
	}
}

void VertexShaderManager::SetTexMatrixChangedA(u32 Value)
{
	if (MatrixIndexA.Hex != Value)
	{
		VertexManager::Flush();
		if (MatrixIndexA.PosNormalMtxIdx != (Value&0x3f))
			bPosNormalMatrixChanged = true;
		bTexMatricesChanged[0] = true;
		MatrixIndexA.Hex = Value;
	}
}

void VertexShaderManager::SetTexMatrixChangedB(u32 Value)
{
	if (MatrixIndexB.Hex != Value)
	{
		VertexManager::Flush();
		bTexMatricesChanged[1] = true;
		MatrixIndexB.Hex = Value;
	}
}

void VertexShaderManager::SetViewportChanged()
{
	bViewportChanged = true;
}

void VertexShaderManager::SetProjectionChanged()
{
	bProjectionChanged = true;
}

void VertexShaderManager::SetMaterialColorChanged(int index, u32 color)
{
	nMaterialsChanged  |= (1 << index);
}

void VertexShaderManager::TranslateView(float x, float y, float z)
{
	float result[3];
	float vector[3] = { x,z,y };

	Matrix33::Multiply(s_viewInvRotationMatrix, vector, result);

	for (int i = 0; i < 3; i++)
		s_fViewTranslationVector[i] += result[i];

	bProjectionChanged = true;
}

void VertexShaderManager::RotateView(float x, float y)
{
	s_fViewRotation[0] += x;
	s_fViewRotation[1] += y;

	Matrix33 mx;
	Matrix33 my;
	Matrix33::RotateX(mx, s_fViewRotation[1]);
	Matrix33::RotateY(my, s_fViewRotation[0]);
	Matrix33::Multiply(mx, my, s_viewRotationMatrix);

	// reverse rotation
	Matrix33::RotateX(mx, -s_fViewRotation[1]);
	Matrix33::RotateY(my, -s_fViewRotation[0]);
	Matrix33::Multiply(my, mx, s_viewInvRotationMatrix);

	bProjectionChanged = true;
}

void VertexShaderManager::ResetView()
{
	memset(s_fViewTranslationVector, 0, sizeof(s_fViewTranslationVector));
	Matrix33::LoadIdentity(s_viewRotationMatrix);
	Matrix33::LoadIdentity(s_viewInvRotationMatrix);
	s_fViewRotation[0] = s_fViewRotation[1] = 0.0f;

	bProjectionChanged = true;
}

void VertexShaderManager::DoState(PointerWrap &p)
{
	p.Do(g_fProjectionMatrix);
	p.Do(s_viewportCorrection);
	p.Do(s_viewRotationMatrix);
	p.Do(s_viewInvRotationMatrix);
	p.Do(s_fViewTranslationVector);
	p.Do(s_fViewRotation);
	p.Do(constants);
	p.Do(dirty);

	if (p.GetMode() == PointerWrap::MODE_READ)
	{
		Dirty();
	}
}