GSdx: optimized the triangle setup of the rasterizer a bit, while it isn't the bottle-neck of drawing, it can still add a few percent to the fps.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4404 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/GSDrawScanlineCodeGenerator.x64.avx.cpp

@@ -379,8 +379,8 @@ void GSDrawScanlineCodeGenerator::Init()
 			}
 			else
 			{
-				vmovdqa(xmm13, ptr[&m_local.c.rb]);
+				vmovdqa(xmm13, ptr[r11 + offsetof(GSScanlineLocalData, c.rb)]);
-				vmovdqa(xmm14, ptr[&m_local.c.ga]);
+				vmovdqa(xmm14, ptr[r11 + offsetof(GSScanlineLocalData, c.ga)]);
 			}
 		}
 	}

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.avx.cpp

@@ -273,12 +273,12 @@ void GSDrawScanlineCodeGenerator::Init()
 
 	mov(esi, dword[esp + _top]);
 	lea(esi, ptr[esi * 8]);
-	add(esi, dword[&m_local.gd->fzbr]);
+	add(esi, ptr[&m_local.gd->fzbr]);
 
 	// GSVector2i* fza_offset = &m_local.gd->fzbc[left >> 2];
 
 	lea(edi, ptr[ebx * 2]);
-	add(edi, dword[&m_local.gd->fzbc]);
+	add(edi, ptr[&m_local.gd->fzbc]);
 
 	if(!m_sel.sprite && (m_sel.fwrite && m_sel.fge || m_sel.zb) || m_sel.fb && (m_sel.edge || m_sel.tfx != TFX_NONE || m_sel.iip))
 	{
@@ -585,8 +585,8 @@ void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
 
 	// int za = fza_base.y + fza_offset->y;
 
-	mov(ebp, dword[esi + 4]);
+	mov(ebp, ptr[esi + 4]);
-	add(ebp, dword[edi + 4]);
+	add(ebp, ptr[edi + 4]);
 
 	// GSVector4i zs = zi;
 
@@ -682,7 +682,7 @@ void GSDrawScanlineCodeGenerator::SampleTexture()
 		return;
 	}
 
-	mov(ebx, dword[&m_local.gd->tex]);
+	mov(ebx, ptr[&m_local.gd->tex]);
 
 	// ebx = tex
 
@@ -1446,8 +1446,8 @@ void GSDrawScanlineCodeGenerator::ReadFrame()
 
 	// int fa = fza_base.x + fza_offset->x;
 
-	mov(ebx, dword[esi]);
+	mov(ebx, ptr[esi]);
-	add(ebx, dword[edi]);
+	add(ebx, ptr[edi]);
 
 	if(!m_sel.rfb)
 	{
@@ -1805,7 +1805,7 @@ void GSDrawScanlineCodeGenerator::WriteFrame()
 
 	if(m_sel.fpsm == 2 && m_sel.dthe)
 	{
-		mov(eax, dword[esp + _top]);
+		mov(eax, ptr[esp + _top]);
 		and(eax, 3);
 		shl(eax, 5);
 		vpaddw(xmm5, ptr[eax + (size_t)&m_local.gd->dimx[0]]);

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.cpp

@@ -268,14 +268,14 @@ void GSDrawScanlineCodeGenerator::Init()
 
 	// GSVector2i* fza_base = &m_local.gd->fzbr[top];
 
-	mov(esi, dword[esp + _top]);
+	mov(esi, ptr[esp + _top]);
 	lea(esi, ptr[esi * 8]);
-	add(esi, dword[&m_local.gd->fzbr]);
+	add(esi, ptr[&m_local.gd->fzbr]);
 
 	// GSVector2i* fza_offset = &m_local.gd->fzbc[left >> 2];
 
 	lea(edi, ptr[ebx * 2]);
-	add(edi, dword[&m_local.gd->fzbc]);
+	add(edi, ptr[&m_local.gd->fzbc]);
 
 	if(!m_sel.sprite && (m_sel.fwrite && m_sel.fge || m_sel.zb) || m_sel.fb && (m_sel.edge || m_sel.tfx != TFX_NONE || m_sel.iip))
 	{
@@ -286,7 +286,7 @@ void GSDrawScanlineCodeGenerator::Init()
 
 		// ebx = &v
 
-		mov(ebx, dword[esp + _v]);
+		mov(ebx, ptr[esp + _v]);
 	}
 
 	if(!m_sel.sprite)
@@ -587,8 +587,8 @@ void GSDrawScanlineCodeGenerator::TestZ(const Xmm& temp1, const Xmm& temp2)
 
 	// int za = fza_base.y + fza_offset->y;
 
-	mov(ebp, dword[esi + 4]);
+	mov(ebp, ptr[esi + 4]);
-	add(ebp, dword[edi + 4]);
+	add(ebp, ptr[edi + 4]);
 
 	// GSVector4i zs = zi;
 
@@ -684,7 +684,7 @@ void GSDrawScanlineCodeGenerator::SampleTexture()
 		return;
 	}
 
-	mov(ebx, dword[&m_local.gd->tex]);
+	mov(ebx, ptr[&m_local.gd->tex]);
 
 	// ebx = tex
 
@@ -1495,8 +1495,8 @@ void GSDrawScanlineCodeGenerator::ReadFrame()
 
 	// int fa = fza_base.x + fza_offset->x;
 
-	mov(ebx, dword[esi]);
+	mov(ebx, ptr[esi]);
-	add(ebx, dword[edi]);
+	add(ebx, ptr[edi]);
 
 	if(!m_sel.rfb)
 	{
@@ -1875,7 +1875,7 @@ void GSDrawScanlineCodeGenerator::WriteFrame()
 
 	if(m_sel.fpsm == 2 && m_sel.dthe)
 	{
-		mov(eax, dword[esp + _top]);
+		mov(eax, ptr[esp + _top]);
 		and(eax, 3);
 		shl(eax, 5);
 		paddw(xmm5, ptr[eax + (size_t)&m_local.gd->dimx[0]]);

plugins/GSdx/GSFunctionMap.h

@@ -234,6 +234,24 @@ public:
 				ml.method_size = (unsigned int)cg->getSize();
 
 				iJIT_NotifyEvent(iJVM_EVENT_TYPE_METHOD_LOAD_FINISHED, &ml);
+/*
+				name = format("c:/temp/%s_%016llx.bin", m_name.c_str(), (uint64)key);
+
+				if(FILE* fp = fopen(name.c_str(), "wb"))
+				{
+					fputc(0x0F, fp); fputc(0x0B, fp);
+					fputc(0xBB, fp); fputc(0x6F, fp); fputc(0x00, fp); fputc(0x00, fp); fputc(0x00, fp);
+					fputc(0x64, fp); fputc(0x67, fp); fputc(0x90, fp);
+
+					fwrite(cg->getCode(), cg->getSize(), 1, fp);
+					
+					fputc(0xBB, fp); fputc(0xDE, fp); fputc(0x00, fp); fputc(0x00, fp); fputc(0x00, fp);
+					fputc(0x64, fp); fputc(0x67, fp); fputc(0x90, fp);
+					fputc(0x0F, fp); fputc(0x0B, fp);
+
+					fclose(fp);
+				}
+*/
 			}
 
             #endif

plugins/GSdx/GSRasterizer.cpp

@@ -44,10 +44,14 @@ GSRasterizer::GSRasterizer(IDrawScanline* ds)
 	, m_id(0)
 	, m_threads(1)
 {
+	m_edge.buff = (GSScanline*)vmalloc(sizeof(GSScanline) * 2048, false);
+	m_edge.count = 0;
 }
 
 GSRasterizer::~GSRasterizer()
 {
+	if(m_edge.buff != NULL) vmfree(m_edge.buff, sizeof(GSScanline) * 2048);
+
 	delete m_ds;
 }
 
@@ -68,10 +72,12 @@ void GSRasterizer::Draw(const GSRasterizerData* data)
 {
 	m_ds->BeginDraw(data->param);
 
-	const GSVector4i scissor = data->scissor;
 	const GSVertexSW* vertices = data->vertices;
 	const int count = data->count;
 
+	m_scissor = data->scissor;
+	m_fscissor = GSVector4(data->scissor);
+
 	m_stats.Reset();
 
 	int64 start = __rdtsc();
@@ -80,22 +86,22 @@ void GSRasterizer::Draw(const GSRasterizerData* data)
 	{
 	case GS_POINT_CLASS:
 		m_stats.prims = count;
-		for(int i = 0; i < count; i++) DrawPoint(&vertices[i], scissor);
+		for(int i = 0; i < count; i++) DrawPoint(&vertices[i]);
 		break;
 	case GS_LINE_CLASS:
 		ASSERT(!(count & 1));
 		m_stats.prims = count / 2;
-		for(int i = 0; i < count; i += 2) DrawLine(&vertices[i], scissor);
+		for(int i = 0; i < count; i += 2) DrawLine(&vertices[i]);
 		break;
 	case GS_TRIANGLE_CLASS:
 		ASSERT(!(count % 3));
 		m_stats.prims = count / 3;
-		for(int i = 0; i < count; i += 3) DrawTriangle(&vertices[i], scissor);
+		for(int i = 0; i < count; i += 3) DrawTriangle(&vertices[i]);
 		break;
 	case GS_SPRITE_CLASS:
 		ASSERT(!(count & 1));
 		m_stats.prims = count / 2;
-		for(int i = 0; i < count; i += 2) DrawSprite(&vertices[i], scissor);
+		for(int i = 0; i < count; i += 2) DrawSprite(&vertices[i]);
 		break;
 	default:
 		__assume(0);
@@ -111,26 +117,26 @@ void GSRasterizer::GetStats(GSRasterizerStats& stats)
 	stats = m_stats;
 }
 
-void GSRasterizer::DrawPoint(const GSVertexSW* v, const GSVector4i& scissor)
+void GSRasterizer::DrawPoint(const GSVertexSW* v)
 {
 	// TODO: round to closest for point, prestep for line
 
 	GSVector4i p(v->p);
 
-	if(scissor.left <= p.x && p.x < scissor.right && scissor.top <= p.y && p.y < scissor.bottom)
+	if(m_scissor.left <= p.x && p.x < m_scissor.right && m_scissor.top <= p.y && p.y < m_scissor.bottom)
 	{
 		if(IsOneOfMyScanlines(p.y))
 		{
+			m_stats.pixels++;
+
 			m_ds->SetupPrim(v, *v);
 
 			m_ds->DrawScanline(p.x + 1, p.x, p.y, *v);
-
-			m_stats.pixels++;
 		}
 	}
 }
 
-void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
+void GSRasterizer::DrawLine(const GSVertexSW* v)
 {
 	GSVertexSW dv = v[1] - v[0];
 
@@ -148,8 +154,10 @@ void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
 
 		m_ds->SetupPrim(v, dscan);
 
-		DrawEdge(v[0], v[1], dv, scissor, i, 0);
+		DrawEdge(v[0], v[1], dv, i, 0);
-		DrawEdge(v[0], v[1], dv, scissor, i, 1);
+		DrawEdge(v[0], v[1], dv, i, 1);
+
+		FlushEdge();
 
 		return;
 	}
@@ -176,7 +184,7 @@ void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
 
 			GSVector4i p(l.p);
 
-			if(scissor.top <= p.y && p.y < scissor.bottom)
+			if(m_scissor.top <= p.y && p.y < m_scissor.bottom)
 			{
 				GSVertexSW dscan = dv / dv.p.xxxx();
 
@@ -184,9 +192,9 @@ void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
 
 				l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
 
-				GSVector4 fscissor(scissor);
+				DrawTriangleSection(p.y, p.y + 1, l, dl, dscan);
 
-				DrawTriangleSection(p.y, p.y + 1, l, dl, dscan, fscissor);
+				Flush();
 			}
 		}
 
@@ -199,12 +207,13 @@ void GSRasterizer::DrawLine(const GSVertexSW* v, const GSVector4i& scissor)
 	GSVertexSW dedge = dv / dp.v[i];
 
 	// TODO: prestep + clip with the scissor
+	// TODO: inline drawpoint + Flush()
 
 	int steps = dpi.v[i];
 
 	while(steps-- > 0)
 	{
-		DrawPoint(&edge, scissor);
+		DrawPoint(&edge);
 
 		edge += dedge;
 	}
@@ -222,301 +231,187 @@ static const int s_abc[8][4] =
 	{2, 1, 0, 0}, // a > b > c
 };
 
-void GSRasterizer::DrawTriangle(const GSVertexSW* vertices, const GSVector4i& scissor)
+void GSRasterizer::DrawTriangle(const GSVertexSW* vertices)
 {
-	GSVertexSW v[3];
+	// edge buffer is used here to avoid xmm save-restores (except when we do aa1 in the middle)
+
+	GSVertexSW v[4];
+	GSVertexSW dv[3];
+	GSVertexSW ddv[3];
+	GSVertexSW longest;
+	GSVertexSW dscan;
 
 	GSVector4 aabb = vertices[0].p.yyyy(vertices[1].p);
 	GSVector4 bccb = vertices[1].p.yyyy(vertices[2].p).xzzx();
 
-	int i = (aabb > bccb).mask() & 7;
+	int abc = (aabb > bccb).mask() & 7;
 
-	v[0] = vertices[s_abc[i][0]];
+	v[0] = vertices[s_abc[abc][0]];
-	v[1] = vertices[s_abc[i][1]];
+	v[1] = vertices[s_abc[abc][1]];
-	v[2] = vertices[s_abc[i][2]];
+	v[2] = vertices[s_abc[abc][2]];
 
 	aabb = v[0].p.yyyy(v[1].p);
 	bccb = v[1].p.yyyy(v[2].p).xzzx();
 
-	i = (aabb == bccb).mask() & 7;
+	int i = (aabb == bccb).mask() & 7;
 
-	if(m_ds->IsEdge())
+	GSVector4 tbf = aabb.xzxz(bccb).ceil();
-	{
+	GSVector4 tbmax = tbf.max(m_fscissor.yyyy());
-		DrawEdge(v, scissor);
+	GSVector4 tbmin = tbf.min(m_fscissor.wwww());
-	}
+	GSVector4i tb = GSVector4i(tbmax.xzyw(tbmin));
-
-	switch(i)
-	{
-	case 0: // a < b < c
-		DrawTriangleTopBottom(v, scissor);
-		break;
-	case 1: // a == b < c
-		DrawTriangleBottom(v, scissor);
-		break;
-	case 4: // a < b == c
-		DrawTriangleTop(v, scissor);
-		break;
-	case 7: // a == b == c
-		break;
-	default:
-		__assume(0);
-	}
-}
-
-void GSRasterizer::DrawEdge(const GSVertexSW* v, const GSVector4i& scissor)
-{
-	GSVertexSW dv[3];
 
 	dv[0] = v[1] - v[0];
 	dv[1] = v[2] - v[0];
 	dv[2] = v[2] - v[1];
 
-	GSVector4 dx = dv[0].p.upl(dv[1].p).xyxy(dv[2].p);
+	switch(i)
-	GSVector4 dy = dv[0].p.upl(dv[1].p).zwyx(dv[2].p);
-
-	GSVector4 a = dx.abs() < dy.abs(); // |x| <= |y|
-	GSVector4 b = dx < GSVector4::zero(); // x < 0
-	GSVector4 c = dv[1].p * (dv[0].p / dv[1].p).yyyy() < dv[0].p; // longest.p.x < 0
-
-	int i = a.mask();
-	int j = ((a | b) ^ c.xxxx()).mask() ^ 2; // evil
-
-	GSVertexSW dscan;
-
-	dscan.p = GSVector4::zero();
-	dscan.t = GSVector4::zero();
-	dscan.c = GSVector4::zero();
-
-	m_ds->SetupPrim(v, dscan); // TODO: don't call it twice (can't be sure about the second call if the triangle is too small)
-
-	DrawEdge(v[0], v[1], dv[0], scissor, i & 1, j & 1);
-	DrawEdge(v[0], v[2], dv[1], scissor, i & 2, j & 2);
-	DrawEdge(v[1], v[2], dv[2], scissor, i & 4, j & 4);
-}
-
-void GSRasterizer::DrawTriangleTop(GSVertexSW* v, const GSVector4i& scissor)
-{
-	GSVertexSW longest;
-
-	longest.p = v[2].p - v[1].p;
-
-	int i = longest.p.upl(longest.p == GSVector4::zero()).mask();
-
-	if(i & 2) return;
-
-	i &= 1;
-
-	GSVertexSW& l = v[0];
-	GSVector4& r = v[0].p;
-
-	GSVector4 fscissor(scissor);
-
-	GSVector4 tb = l.p.upl(v[2].p).ceil();
-
-	GSVector4 tbmax = tb.max(fscissor.yyyy());
-	GSVector4 tbmin = tb.min(fscissor.wwww());
-
-	GSVector4i tbi = GSVector4i(tbmax.zzww(tbmin));
-
-	int top = tbi.extract32<0>();
-	int bottom = tbi.extract32<2>();
-
-	if(top >= bottom) return;
-
-	longest.t = v[2].t - v[1].t;
-	longest.c = v[2].c - v[1].c;
-
-	GSVertexSW dscan = longest * longest.p.xxxx().rcp();
-
-	GSVertexSW vl = v[1 + i] - l;
-	GSVector4 vr = v[2 - i].p - r;
-
-	GSVertexSW dl = vl / vl.p.yyyy();
-	GSVector4 dr = vr / vr.yyyy();
-
-	GSVector4 dy = tbmax.zzzz() - l.p.yyyy();
-
-	l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
-	dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
-
-	l += dl * dy;
-
-	m_ds->SetupPrim(v, dscan);
-
-	DrawTriangleSection(top, bottom, l, dl, dscan, fscissor);
-}
-
-void GSRasterizer::DrawTriangleBottom(GSVertexSW* v, const GSVector4i& scissor)
-{
-	GSVertexSW longest;
-
-	longest.p = v[1].p - v[0].p;
-
-	int i = longest.p.upl(longest.p == GSVector4::zero()).mask();
-
-	if(i & 2) return;
-
-	i &= 1;
-
-	GSVertexSW& l = v[i];
-	GSVector4& r = v[1 - i].p;
-
-	GSVector4 fscissor(scissor);
-
-	GSVector4 tb = l.p.upl(v[2].p).ceil();
-
-	GSVector4 tbmax = tb.max(fscissor.yyyy());
-	GSVector4 tbmin = tb.min(fscissor.wwww());
-
-	GSVector4i tbi = GSVector4i(tbmax.zzww(tbmin));
-
-	int top = tbi.extract32<0>();
-	int bottom = tbi.extract32<2>();
-
-	if(top >= bottom) return;
-
-	longest.t = v[1].t - v[0].t;
-	longest.c = v[1].c - v[0].c;
-
-	GSVertexSW dscan = longest * longest.p.xxxx().rcp();
-
-	GSVertexSW vl = v[2] - l;
-	GSVector4 vr = v[2].p - r;
-
-	GSVertexSW dl = vl / vl.p.yyyy();
-	GSVector4 dr = vr / vr.yyyy();
-
-	GSVector4 dy = tbmax.zzzz() - l.p.yyyy();
-
-	l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
-	dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
-
-	l += dl * dy;
-
-	m_ds->SetupPrim(v, dscan);
-
-	DrawTriangleSection(top, bottom, l, dl, dscan, fscissor);
-}
-
-void GSRasterizer::DrawTriangleTopBottom(GSVertexSW* v, const GSVector4i& scissor)
-{
-	GSVertexSW dv[3];
-
-	dv[0] = v[1] - v[0];
-	dv[1] = v[2] - v[0];
-
-	GSVertexSW longest = dv[1] * (dv[0].p / dv[1].p).yyyy() - dv[0];
-
-	int i = longest.p.upl(longest.p == GSVector4::zero()).mask();
-
-	if(i & 2) return;
-
-	i &= 1;
-
-	GSVertexSW dscan = longest * longest.p.xxxx().rcp();
-
-	m_ds->SetupPrim(v, dscan);
-
-	GSVector4 fscissor(scissor);
-
-	GSVector4 tb = v[0].p.upl(v[1].p).zwzw(v[1].p.upl(v[2].p)).ceil();
-
-	GSVector4 tbmax = tb.max(fscissor.yyyy());
-	GSVector4 tbmin = tb.min(fscissor.wwww());
-
-	GSVector4i tbi = GSVector4i(tbmax.xzyw(tbmin));
-
-	int top = tbi.extract32<0>();
-	int bottom = tbi.extract32<2>();
-
-	GSVertexSW& l = v[0];
-	GSVector4 r = v[0].p;
-
-	GSVertexSW dl = dv[i] / dv[i].p.yyyy();
-	GSVector4 dr = dv[1 - i].p / dv[1 - i].p.yyyy();
-
-	GSVector4 dy = tbmax.xxxx() - l.p.yyyy();
-
-	l += dl * dy;
-	r += dr * dy;
-
-	if(top < bottom)
 	{
-		DrawTriangleSection(top, bottom, l, dl, r, dr, dscan, fscissor);
+	case 0: // a < b < c
+		ddv[0] = dv[0] / dv[0].p.yyyy();
+		ddv[1] = dv[1] / dv[1].p.yyyy();
+		ddv[2] = dv[2] / dv[2].p.yyyy();
+		longest = ddv[1] * dv[0].p.yyyy() - dv[0];
+		v[3] = v[1] + longest; // point between v[0] and v[2] where y == v[1].y
+		break;
+	case 1: // a == b < c
+		ddv[1] = dv[1] / dv[1].p.yyyy();
+		ddv[2] = dv[2] / dv[2].p.yyyy();
+		longest = dv[0];
+		break;
+	case 4: // a < b == c
+		ddv[0] = dv[0] / dv[0].p.yyyy();
+		ddv[1] = dv[1] / dv[1].p.yyyy();
+		longest = dv[2];
+		break;
+	case 7: // a == b == c
+		return;
+	default:
+		__assume(0);
 	}
 
-	top = tbi.y;
+	int j = longest.p.upl(longest.p == GSVector4::zero()).mask();
-	bottom = tbi.w;
 
-	if(top < bottom)
+	if(j & 2) return;
+
+	j &= 1;
+
+	dscan = longest * longest.p.xxxx().rcp();
+	
+	if(m_ds->IsEdge())
 	{
-		if(i == 0)
+		GSVector4 dx = dv[0].p.upl(dv[1].p).xyxy(dv[2].p);
-		{
+		GSVector4 dy = dv[0].p.upl(dv[1].p).zwyx(dv[2].p);
-			l = v[1];
-			dv[2] = v[2] - v[1];
-			dl = dv[2] / dv[2].p.yyyy();
-		}
-		else
-		{
-			r = v[1].p;
-			dv[2].p = v[2].p - v[1].p;
-			dr = dv[2].p / dv[2].p.yyyy();
-		}
 
-		l += dl * (tbmax.zzzz() - l.p.yyyy());
+		GSVector4 a = dx.abs() < dy.abs(); // |dx| <= |dy|
-		r += dr * (tbmax.zzzz() - r.yyyy());
+		GSVector4 b = dx < GSVector4::zero(); // dx < 0
+		GSVector4 c = longest.p.xxxx() < GSVector4::zero(); // longest.p.x < 0
 
-		l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
+		int i = a.mask();
-		dl.p = dl.p.upl(dr).xyzw(dl.p); // dr.x => dl.y
+		int j = ((a | b) ^ c).mask() ^ 2; // evil
 
-		DrawTriangleSection(top, bottom, l, dl, dscan, fscissor);
+		DrawEdge(v[0], v[1], dv[0], i & 1, j & 1);
+		DrawEdge(v[0], v[2], dv[1], i & 2, j & 2);
+		DrawEdge(v[1], v[2], dv[2], i & 4, j & 4);
+
+		GSVertexSW dscan;
+
+		dscan.p = GSVector4::zero();
+		dscan.t = GSVector4::zero();
+		dscan.c = GSVector4::zero();
+
+		m_ds->SetupPrim(v, dscan);
+
+		FlushEdge();
 	}
+
+	switch(i)
+	{
+	case 0: // a < b < c
+		
+		if(tb.x < tb.z)
+		{
+			GSVertexSW l = v[0];
+			GSVertexSW dl = ddv[j];
+
+			GSVector4 dy = tbmax.xxxx() - l.p.yyyy();
+
+			l.p = l.p.xxzw(); // r.x => l.y
+			dl.p = dl.p.upl(ddv[1 - j].p).xyzw(dl.p); // dr.x => dl.y
+
+			l += dl * dy;
+
+			DrawTriangleSection(tb.x, tb.z, l, dl, dscan);
+		}
+
+		if(tb.y < tb.w)
+		{
+			GSVertexSW l = v[1 + (1 << j)];
+			GSVertexSW dl = ddv[2 - j];
+
+			GSVector4 dy = tbmax.zzzz() - l.p.yyyy();
+
+			l.p = l.p.upl(v[3 - (1 << j)].p).xyzw(l.p); // r.x => l.y
+			dl.p = dl.p.upl(ddv[1 + j].p).xyzw(dl.p); // dr.x => dl.y
+
+			l += dl * dy;
+
+			DrawTriangleSection(tb.y, tb.w, l, dl, dscan);
+		}
+
+		break;
+
+	case 1: // a == b < c
+
+		if(tb.x < tb.w)
+		{
+			GSVertexSW l = v[j];
+			GSVertexSW dl = ddv[1 + j];
+
+			GSVector4 dy = tbmax.xxxx() - l.p.yyyy();
+
+			l.p = l.p.upl(v[1 - j].p).xyzw(l.p); // r.x => l.y
+			dl.p = dl.p.upl(ddv[2 - j].p).xyzw(dl.p); // dr.x => dl.y
+
+			l += dl * dy;
+
+			DrawTriangleSection(tb.x, tb.w, l, dl, dscan);
+		}
+
+		break;
+
+	case 4: // a < b == c
+
+		if(tb.x < tb.w)
+		{
+			GSVertexSW l = v[0];
+			GSVertexSW dl = ddv[j];
+
+			GSVector4 dy = tbmax.xxxx() - l.p.yyyy();
+
+			l.p = l.p.xxzw(); // r.x => l.y
+			dl.p = dl.p.upl(ddv[1 - j].p).xyzw(dl.p); // dr.x => dl.y
+
+			l += dl * dy;
+
+			DrawTriangleSection(tb.x, tb.w, l, dl, dscan);
+		}
+
+		break;
+
+	default:
+		__assume(0);
+	}
+
+	m_ds->SetupPrim(v, dscan);
+
+	Flush();
 }
 
-void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, GSVector4& r, const GSVector4& dr, const GSVertexSW& dscan, const GSVector4& fscissor)
+void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, const GSVertexSW& dscan)
 {
 	ASSERT(top < bottom);
 
-	while(1)
+	GSScanline* RESTRICT e = &m_edge.buff[m_edge.count];
-	{
-		do
-		{
-			if(IsOneOfMyScanlines(top))
-			{
-				GSVector4 lr = l.p.xyxy(r).ceil();
-
-				GSVector4 lrmax = lr.max(fscissor.xxxx());
-				GSVector4 lrmin = lr.min(fscissor.zzzz());
-
-				GSVector4i lri = GSVector4i(lrmax.xxzz(lrmin));
-
-				int left = lri.extract32<0>();
-				int right = lri.extract32<2>();
-
-				int pixels = right - left;
-
-				if(pixels > 0)
-				{
-					m_stats.pixels += pixels;
-
-					GSVertexSW scan = l + dscan * (lrmax - l.p).xxxx();
-
-					m_ds->DrawScanline(right, left, top, scan);
-				}
-			}
-		}
-		while(0);
-
-		if(++top >= bottom) break;
-
-		l += dl;
-		r += dr;
-	}
-}
-
-void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, const GSVertexSW& dscan, const GSVector4& fscissor)
-{
-	ASSERT(top < bottom);
 
 	while(1)
 	{
@@ -526,8 +421,8 @@ void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const
 			{
 				GSVector4 lr = l.p.ceil();
 
-				GSVector4 lrmax = lr.max(fscissor.xxxx());
+				GSVector4 lrmax = lr.max(m_fscissor.xxxx());
-				GSVector4 lrmin = lr.min(fscissor.zzzz());
+				GSVector4 lrmin = lr.min(m_fscissor.zzzz());
 
 				GSVector4i lri = GSVector4i(lrmax.xxyy(lrmin));
 
@@ -540,9 +435,13 @@ void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const
 				{
 					m_stats.pixels += pixels;
 
-					GSVertexSW scan = l + dscan * (lrmax - l.p).xxxx();
+					e->scan = l + dscan * (lrmax - l.p).xxxx();
 
-					m_ds->DrawScanline(right, left, top, scan);
+					e->p.left = left;
+					e->p.top = top;
+					e->p.right = right;
+
+					e++;
 				}
 			}
 		}
@@ -552,9 +451,11 @@ void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const
 
 		l += dl;
 	}
+
+	m_edge.count += e - &m_edge.buff[m_edge.count];
 }
 
-void GSRasterizer::DrawSprite(const GSVertexSW* vertices, const GSVector4i& scissor)
+void GSRasterizer::DrawSprite(const GSVertexSW* vertices)
 {
 	GSVertexSW v[2];
 
@@ -569,7 +470,7 @@ void GSRasterizer::DrawSprite(const GSVertexSW* vertices, const GSVector4i& scis
 
 	GSVector4i r(v[0].p.xyxy(v[1].p).ceil());
 
-	r = r.rintersect(scissor);
+	r = r.rintersect(m_scissor);
 
 	if(r.rempty()) return;
 
@@ -611,14 +512,14 @@ void GSRasterizer::DrawSprite(const GSVertexSW* vertices, const GSVector4i& scis
 	{
 		if(IsOneOfMyScanlines(r.top))
 		{
-			m_ds->DrawScanline(r.right, r.left, r.top, scan);
-
 			m_stats.pixels += r.width();
+
+			m_ds->DrawScanline(r.right, r.left, r.top, scan);
 		}
 	}
 }
 
-void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GSVertexSW& dv, const GSVector4i& scissor, int orientation, int side)
+void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GSVertexSW& dv, int orientation, int side)
 {
 	// orientation:
 	// - true: |dv.p.y| > |dv.p.x|
@@ -630,14 +531,14 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 	// TODO: bit slow and too much duplicated code
 	// TODO: inner pre-step is still missing (hardly noticable)
 
-	GSVector4 fscissor(scissor);
+	GSScanline* RESTRICT dst = &m_edge.buff[m_edge.count];
-
+			
 	GSVector4 lrtb = v0.p.upl(v1.p).ceil();
 
 	if(orientation)
 	{
-		GSVector4 tbmax = lrtb.max(fscissor.yyyy());
+		GSVector4 tbmax = lrtb.max(m_fscissor.yyyy());
-		GSVector4 tbmin = lrtb.min(fscissor.wwww());
+		GSVector4 tbmin = lrtb.min(m_fscissor.wwww());
 
 		GSVector4i tbi = GSVector4i(tbmax.zwzw(tbmin));
 
@@ -684,15 +585,18 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 					int xi = x >> 16;
 					int xf = x & 0xffff;
 
-					if(scissor.left <= xi && xi < scissor.right && IsOneOfMyScanlines(xi))
+					if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
 					{
 						m_stats.pixels++;
 
-						edge.t.u32[3] = (0x10000 - xf) & 0xffff;
+						dst->scan = edge;
+						dst->scan.t.u32[3] = (0x10000 - xf) & 0xffff;
 
-						m_ds->DrawEdge(xi + 1, xi, top, edge);
+						dst->p.left = xi;
+						dst->p.top = top;
+						dst->p.right = xi + 1;
 
-						edge.t.u32[3] = 0;
+						dst++;
 					}
 				}
 				while(0);
@@ -712,15 +616,18 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 					int xi = (x >> 16) + 1;
 					int xf = x & 0xffff;
 
-					if(scissor.left <= xi && xi < scissor.right && IsOneOfMyScanlines(xi))
+					if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
 					{
 						m_stats.pixels++;
 
-						edge.t.u32[3] = xf;
+						dst->scan = edge;
+						dst->scan.t.u32[3] = xf;
 
-						m_ds->DrawEdge(xi + 1, xi, top, edge);
+						dst->p.left = xi;
+						dst->p.top = top;
+						dst->p.right = xi + 1;
 
-						edge.t.u32[3] = 0;
+						dst++;
 					}
 				}
 				while(0);
@@ -734,8 +641,8 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 	}
 	else
 	{
-		GSVector4 lrmax = lrtb.max(fscissor.xxxx());
+		GSVector4 lrmax = lrtb.max(m_fscissor.xxxx());
-		GSVector4 lrmin = lrtb.min(fscissor.zzzz());
+		GSVector4 lrmin = lrtb.min(m_fscissor.zzzz());
 
 		GSVector4i lri = GSVector4i(lrmax.xyxy(lrmin));
 
@@ -782,15 +689,18 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 					int yi = y >> 16;
 					int yf = y & 0xffff;
 
-					if(scissor.top <= yi && yi < scissor.bottom && IsOneOfMyScanlines(yi))
+					if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
 					{
 						m_stats.pixels++;
 
-						edge.t.u32[3] = (0x10000 - yf) & 0xffff;
+						dst->scan = edge;
+						dst->scan.t.u32[3] = (0x10000 - yf) & 0xffff;
 
-						m_ds->DrawEdge(left + 1, left, yi, edge);
+						dst->p.left = left;
+						dst->p.top = yi;
+						dst->p.right = left + 1;
 
-						edge.t.u32[3] = 0;
+						dst++;
 					}
 				}
 				while(0);
@@ -810,15 +720,18 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 					int yi = (y >> 16) + 1;
 					int yf = y & 0xffff;
 
-					if(scissor.top <= yi && yi < scissor.bottom && IsOneOfMyScanlines(yi))
+					if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
 					{
 						m_stats.pixels++;
 
-						edge.t.u32[3] = yf;
+						dst->scan = edge;
+						dst->scan.t.u32[3] = yf;
 
-						m_ds->DrawEdge(left + 1, left, yi, edge);
+						dst->p.left = left;
+						dst->p.top = yi;
+						dst->p.right = left + 1;
 
-						edge.t.u32[3] = 0;
+						dst++;
 					}
 				}
 				while(0);
@@ -830,6 +743,34 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 			}
 		}
 	}
+
+	m_edge.count += dst - &m_edge.buff[m_edge.count];
+}
+
+void GSRasterizer::Flush()
+{
+	// TODO: on win64 this could be the place where xmm6-15 are preserved (not by each DrawScanline)
+
+	const GSScanline* s = m_edge.buff;
+
+	for(int count = m_edge.count; count > 0; count--, s++)
+	{
+		m_ds->DrawScanline(s->p.right, s->p.left, s->p.top, s->scan);
+	}
+
+	m_edge.count = 0;
+}
+
+void GSRasterizer::FlushEdge()
+{
+	const GSScanline* s = m_edge.buff;
+
+	for(int count = m_edge.count; count > 0; count--, s++)
+	{
+		m_ds->DrawEdge(s->p.right, s->p.left, s->p.top, s->scan);
+	}
+
+	m_edge.count = 0;
 }
 
 //

plugins/GSdx/GSRasterizer.h

@@ -59,7 +59,7 @@ public:
 	virtual void EndDraw(const GSRasterizerStats& stats, uint64 frame) = 0;
 	virtual void PrintStats() = 0;
 
-	__forceinline void SetupPrim(const GSVertexSW* v, const GSVertexSW& dscan) {m_sp(v, dscan);}
+	__forceinline void SetupPrim(const GSVertexSW* vertices, const GSVertexSW& dscan) {m_sp(vertices, dscan);}
 	__forceinline void DrawScanline(int right, int left, int top, const GSVertexSW& scan) {m_ds(right, left, top, scan);}
 	__forceinline void DrawEdge(int right, int left, int top, const GSVertexSW& scan) {m_de(right, left, top, scan);}
 	__forceinline void DrawRect(const GSVector4i& r, const GSVertexSW& v) {(this->*m_dr)(r, v);}
@@ -79,30 +79,33 @@ public:
 	virtual void SetThreadId(int id, int threads) = 0;
 };
 
-class GSRasterizer : public IRasterizer
+__aligned(class, 32) GSRasterizer : public GSAlignedClass<32>, public IRasterizer
 {
+	struct GSScanline {GSVertexSW scan; GSVector4i p;};
+
 protected:
 	IDrawScanline* m_ds;
 	int m_id;
 	int m_threads;
 	GSRasterizerStats m_stats;
+	GSVector4i m_scissor;
+	GSVector4 m_fscissor;
+	struct {GSScanline* buff; int count;} m_edge;
 
-	void DrawPoint(const GSVertexSW* v, const GSVector4i& scissor);
+	void DrawPoint(const GSVertexSW* v);
-	void DrawLine(const GSVertexSW* v, const GSVector4i& scissor);
+	void DrawLine(const GSVertexSW* v);
-	void DrawTriangle(const GSVertexSW* v, const GSVector4i& scissor);
+	void DrawTriangle(const GSVertexSW* v);
-	void DrawEdge(const GSVertexSW* v, const GSVector4i& scissor);
+	void DrawSprite(const GSVertexSW* v);
-	void DrawSprite(const GSVertexSW* v, const GSVector4i& scissor);
+	void DrawEdge(const GSVertexSW* v);
 
-	void DrawTriangleTop(GSVertexSW* v, const GSVector4i& scissor);
+	__forceinline void DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, const GSVertexSW& dscan);
-	void DrawTriangleBottom(GSVertexSW* v, const GSVector4i& scissor);
-	void DrawTriangleTopBottom(GSVertexSW* v, const GSVector4i& scissor);
 
-	__forceinline void DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, GSVector4& r, const GSVector4& dr, const GSVertexSW& dscan, const GSVector4& scissor);
+	void DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GSVertexSW& dv, int orientation, int side);
-	__forceinline void DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, const GSVertexSW& dscan, const GSVector4& scissor);
 
-	void DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GSVertexSW& dv, const GSVector4i& scissor, int orientation, int side);
+	__forceinline bool IsOneOfMyScanlines(int scanline) const;
 
-	inline bool IsOneOfMyScanlines(int scanline) const;
+	void Flush();
+	void FlushEdge();
 
 public:
 	GSRasterizer(IDrawScanline* ds);

plugins/GSdx/GSRenderer.cpp

@@ -76,8 +76,6 @@ bool GSRenderer::CreateDevice(GSDevice* dev)
 
 void GSRenderer::ResetDevice()
 {
-    InvalidateTextureCache();
-
     ResetPrim();
 
     if(m_dev) m_dev->Reset(1, 1);

plugins/GSdx/GSRenderer.h

@@ -139,7 +139,7 @@ protected:
 	{
 		if(m_vertices != NULL) _aligned_free(m_vertices);
 
-		m_maxcount = max(10000, m_maxcount * 3/2);
+		m_maxcount = std::max<int>(10000, m_maxcount * 3 / 2);
 		m_vertices = (Vertex*)_aligned_malloc(sizeof(Vertex) * m_maxcount, 32);
 		m_maxcount -= 100;
 	}

plugins/GSdx/GSRendererHW.h

@@ -484,13 +484,10 @@ protected:
 		}
 	}
 
-	void InvalidateTextureCache()
-	{
-		m_tc->RemoveAll();
-	}
-
 	void ResetDevice()
 	{
+		m_tc->RemoveAll();
+
 		__super::ResetDevice();
 	}
 

plugins/GSdx/GSRendererSW.cpp

@@ -372,12 +372,12 @@ void GSRendererSW::GetScanlineGlobalData(GSScanlineGlobalData& gd)
 
 			if(gd.sel.ltf)
 			{
-				GSVector4 half(0x8000, 0x8000);
-
 				if(gd.sel.fst)
 				{
 					// if q is constant we can do the half pel shift for bilinear sampling on the vertices
 
+					GSVector4 half(0x8000, 0x8000);
+
 					GSVertexSW* v = m_vertices;
 
 					for(int i = 0, j = m_count; i < j; i++)

plugins/GSdx/GSState.cpp

@@ -196,8 +196,6 @@ void GSState::Reset()
 	m_env.Reset();
 
 	m_context = &m_env.CTXT[0];
-
-	InvalidateTextureCache();
 }
 
 void GSState::ResetHandlers()
@@ -770,8 +768,6 @@ void GSState::GIFRegHandlerFOGCOL(const GIFReg* r)
 void GSState::GIFRegHandlerTEXFLUSH(const GIFReg* r)
 {
 	// TRACE(_T("TEXFLUSH\n"));
-
-	// InvalidateTextureCache();
 }
 
 template<int i> void GSState::GIFRegHandlerSCISSOR(const GIFReg* r)
@@ -903,6 +899,7 @@ template<int i> void GSState::GIFRegHandlerFRAME(const GIFReg* r)
 template<int i> void GSState::GIFRegHandlerZBUF(const GIFReg* r)
 {
 	GIFRegZBUF ZBUF = r->ZBUF;
+
 	if(ZBUF.u32[0] == 0)
 	{
 		// during startup all regs are cleared to 0 (by the bios or something), so we mask z until this register becomes valid
@@ -1396,9 +1393,6 @@ template void GSState::Transfer<3>(const uint8* mem, uint32 size);
 
 template<int index> void GSState::Transfer(const uint8* mem, uint32 size)
 {
-	// [TODO] make me into a template parameter... I think.  --air
-	static const bool FrameSkipIt = false;
-
 	GSPerfMonAutoTimer pmat(m_perfmon);
 
 	const uint8* start = mem;
@@ -1420,7 +1414,7 @@ template<int index> void GSState::Transfer(const uint8* mem, uint32 size)
 
 				// ASSERT(!(path.tag.PRE && path.tag.FLG == GIF_FLG_REGLIST)); // kingdom hearts
 
-				if(path.tag.PRE && (path.tag.FLG == GIF_FLG_PACKED) && !FrameSkipIt)
+				if(path.tag.PRE && path.tag.FLG == GIF_FLG_PACKED)
 				{
 					GIFRegPRIM r;
 					r.u64 = path.tag.PRIM;
@@ -1551,7 +1545,7 @@ template<int index> void GSState::Transfer(const uint8* mem, uint32 size)
 			if(m_mt)
 			{
 				// Hackfix for BIOS, which sends an incomplete packet when it does an XGKICK without
-				// having an EOP specified anywhere in VU1 memory.  Needed until PCSX2 is fixed t
+				// having an EOP specified anywhere in VU1 memory.  Needed until PCSX2 is fixed to
 				// handle it more properly (ie, without looping infinitely).
 
 				path.nloop = 0;
@@ -1802,7 +1796,7 @@ bool GSState::GSTransferBuffer::Update(int tw, int th, int bpp, int& len)
 	if(total == 0)
 	{
 		start = end = 0;
-		total = min((tw * bpp >> 3) * th, 1024 * 1024 * 4);
+		total = std::min<int>((tw * bpp >> 3) * th, 1024 * 1024 * 4);
 		overflow = false;
 	}
 

plugins/GSdx/GSState.h

@@ -28,7 +28,6 @@
 #include "GSVertex.h"
 #include "GSVertexList.h"
 #include "GSUtil.h"
-#include "GSDirtyRect.h"
 #include "GSPerfMon.h"
 #include "GSVector.h"
 #include "GSDevice.h"
@@ -208,7 +207,6 @@ public:
 	virtual void ResetPrim() = 0;
 	virtual void InvalidateVideoMem(const GIFRegBITBLTBUF& BITBLTBUF, const GSVector4i& r) {}
 	virtual void InvalidateLocalMem(const GIFRegBITBLTBUF& BITBLTBUF, const GSVector4i& r) {}
-	virtual void InvalidateTextureCache() {}
 
 	void Move();
 	void Write(const uint8* mem, int len);

plugins/GSdx/GSTextureCache.h

@@ -22,6 +22,7 @@
 #pragma once
 
 #include "GSRenderer.h"
+#include "GSDirtyRect.h"
 
 class GSTextureCache
 {

plugins/GSdx/GSVertexSW.h

@@ -23,15 +23,15 @@
 
 #include "GSVector.h"
 
-__aligned(struct, 32) GSVertexSW
+__aligned(struct, 16) GSVertexSW
 {
 	GSVector4 c, p, t;
 
 	GSVertexSW() {}
 	GSVertexSW(const GSVertexSW& v) {*this = v;}
 
-	void operator = (const GSVertexSW& v) {c = v.c; p = v.p; t = v.t;}
+	__forceinline void operator = (const GSVertexSW& v) {c = v.c; p = v.p; t = v.t;}
-	void operator += (const GSVertexSW& v) {c += v.c; p += v.p; t += v.t;}
+	__forceinline void operator += (const GSVertexSW& v) {c += v.c; p += v.p; t += v.t;}
 
 	friend GSVertexSW operator + (const GSVertexSW& v1, const GSVertexSW& v2);
 	friend GSVertexSW operator - (const GSVertexSW& v1, const GSVertexSW& v2);

plugins/GSdx/GSdx.vcxproj

@@ -528,7 +528,7 @@
     <ClCompile Include="GSLocalMemory.cpp" />
     <ClCompile Include="GSPerfMon.cpp" />
     <ClCompile Include="GSRasterizer.cpp">
-      <AssemblerOutput Condition="'$(Configuration)|$(Platform)'=='Release AVX|x64'">All</AssemblerOutput>
+      <AssemblerOutput Condition="'$(Configuration)|$(Platform)'=='Release AVX|x64'">AssemblyAndSourceCode</AssemblerOutput>
       <AssemblerOutput Condition="'$(Configuration)|$(Platform)'=='Release AVX|Win32'">AssemblyAndSourceCode</AssemblerOutput>
     </ClCompile>
     <ClCompile Include="GSRenderer.cpp" />

plugins/GSdx/vtune/iacaMarks.h

@@ -0,0 +1,75 @@
+/*
+* INTEL CONFIDENTIAL
+* Copyright (2008-2009) Intel Corporation All Rights Reserved. 
+* The source code contained or described herein and all documents 
+* related to the source code ("Material") are owned by Intel Corporation 
+* or its suppliers or licensors. Title to the Material remains with 
+* Intel Corporation or its suppliers and licensors. The Material 
+* contains trade secrets and proprietary and confidential information 
+* of Intel or its suppliers and licensors. The Material is protected 
+* by worldwide copyright and trade secret laws and treaty provisions. 
+* No part of the Material may be used, copied, reproduced, modified, 
+* published, uploaded, posted, transmitted, distributed, or disclosed 
+* in any way without Intel’s prior express written permission.
+*
+* No license under any patent, copyright, trade secret or other 
+* intellectual property right is granted to or conferred upon you by 
+* disclosure or delivery of the Materials, either expressly, by implication,
+* inducement, estoppel or otherwise. Any license under such intellectual 
+* property rights must be express and approved by Intel in writing.
+*/
+
+/********************************************************/
+/* Binaries that contain IACA_MARKS will not run.       */
+/* Define IACA_MARKS_OFF when you compile your sources, */
+/* to disable IACA_START, IACA_END, IACA_MSC64_START	*/
+/* and IACA_MSC64_END									*/
+/********************************************************/
+#ifdef IACA_MARKS_OFF
+
+#define IACA_START
+#define IACA_END
+#define IACA_MSC64_START
+#define IACA_MSC64_END
+
+#else
+#if defined (__GNUC__) 
+#define IACA_SSC_MARK( MARK_ID )						\
+__asm__ __volatile__ (									\
+					  "\n\t  movl $"#MARK_ID", %%ebx"	\
+					  "\n\t  .byte 0x64, 0x67, 0x90"	\
+					  : : : "memory" );
+
+#define IACA_UD_BYTES __asm__ __volatile__ ("\n\t .byte 0x0F, 0x0B");
+
+#else
+#define IACA_UD_BYTES {__asm _emit 0x0F \
+	__asm _emit 0x0B}
+
+#define IACA_SSC_MARK(x) {__asm  mov ebx, x\
+	__asm  _emit 0x64 \
+	__asm  _emit 0x67 \
+	__asm  _emit 0x90 }
+
+#define IACA_VC64_START __writegsbyte(111, 111);
+#define IACA_VC64_END   __writegsbyte(222, 222);
+
+#endif
+
+#define IACA_START {IACA_UD_BYTES \
+					IACA_SSC_MARK(111)}
+#define IACA_END {IACA_SSC_MARK(222) \
+					IACA_UD_BYTES}
+
+#endif
+
+/**************** asm *****************
+;START_MARKER
+mov ebx, 111
+db 0x64, 0x67, 0x90
+
+;END_MARKER
+mov ebx, 222
+db 0x64, 0x67, 0x90
+
+**************************************/