GSdx: still working on the rasterizer, would be nice to add some avx code there, but it's just so unfitting for anything.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4407 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/GPUDrawScanlineCodeGenerator.cpp

@@ -23,6 +23,7 @@
 
 #include "stdafx.h"
 #include "GPUDrawScanlineCodeGenerator.h"
+#include "GSVertexSW.h"
 
 static const int _args = 8;
 static const int _top = _args + 4;
@@ -152,7 +153,7 @@ void GPUDrawScanlineCodeGenerator::Init()
 
 		// GSVector4i vt = GSVector4i(v.t).xxzzl();
 
-		cvttps2dq(xmm4, ptr[edx + 32]);
+		cvttps2dq(xmm4, ptr[edx + offsetof(GSVertexSW, t)]);
 		pshuflw(xmm4, xmm4, _MM_SHUFFLE(2, 2, 0, 0));
 
 		// s = vt.xxxx().add16(m_local.d.s);
@@ -186,7 +187,7 @@ void GPUDrawScanlineCodeGenerator::Init()
 	{
 		// GSVector4i vc = GSVector4i(v.c).xxzzlh();
 
-		cvttps2dq(xmm6, ptr[edx]);
+		cvttps2dq(xmm6, ptr[edx + offsetof(GSVertexSW, c)]);
 		pshuflw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
 		pshufhw(xmm6, xmm6, _MM_SHUFFLE(2, 2, 0, 0));
 

plugins/GSdx/GPUSetupPrimCodeGenerator.cpp

@@ -22,8 +22,8 @@
 // TODO: x64
 
 #include "stdafx.h"
-#include "GSVertexSW.h"
 #include "GPUSetupPrimCodeGenerator.h"
+#include "GSVertexSW.h"
 
 using namespace Xbyak;
 
@@ -50,7 +50,7 @@ void GPUSetupPrimCodeGenerator::Generate()
 		{
 			// t = (GSVector4i(vertices[1].t) >> 8) - GSVector4i::x00000001();
 
-			cvttps2dq(xmm1, ptr[ecx + sizeof(GSVertexSW) * 1 + 32]);
+			cvttps2dq(xmm1, ptr[ecx + sizeof(GSVertexSW) * 1 + offsetof(GSVertexSW, t)]);
 			psrld(xmm1, 8);
 			psrld(xmm0, 31);
 			psubd(xmm1, xmm0);
@@ -94,8 +94,8 @@ void GPUSetupPrimCodeGenerator::Generate()
 		// GSVector4 dt = dscan.t;
 		// GSVector4 dc = dscan.c;
 
-		movaps(xmm4, ptr[edx]);
+		movaps(xmm4, ptr[edx + offsetof(GSVertexSW, c)]);
-		movaps(xmm3, ptr[edx + 32]);
+		movaps(xmm3, ptr[edx + offsetof(GSVertexSW, t)]);
 
 		// GSVector4i dtc8 = GSVector4i(dt * 8.0f).ps32(GSVector4i(dc * 8.0f));
 

plugins/GSdx/GSDrawScanlineCodeGenerator.x64.avx.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSDrawScanlineCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE >= 0x500 && (defined(_M_AMD64) || defined(_WIN64))
 
@@ -264,7 +265,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.fwrite && m_sel.fge || m_sel.zb)
 		{
-			vmovaps(xmm0, ptr[r9 + 16]); // v.p
+			vmovaps(xmm0, ptr[r9 + offsetof(GSVertexSW, p)]); // v.p
 
 			if(m_sel.fwrite && m_sel.fge)
 			{
@@ -297,7 +298,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.edge || m_sel.tfx != TFX_NONE)
 		{
-			vmovaps(xmm0, ptr[r9 + 32]); // v.t
+			vmovaps(xmm0, ptr[r9 + offsetof(GSVertexSW, t)]); // v.t
 		}
 
 		if(m_sel.edge)
@@ -361,7 +362,7 @@ void GSDrawScanlineCodeGenerator::Init()
 			{
 				// GSVector4i vc = GSVector4i(v.c);
 
-				vcvttps2dq(xmm0, ptr[r9]); // v.c
+				vcvttps2dq(xmm0, ptr[r9 + offsetof(GSVertexSW, c)]); // v.c
 
 				// vc = vc.upl16(vc.zwxy());
 

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.avx.cpp

@@ -24,6 +24,7 @@
 
 #include "stdafx.h"
 #include "GSDrawScanlineCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE >= 0x500 && !(defined(_M_AMD64) || defined(_WIN64))
 
@@ -296,7 +297,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.fwrite && m_sel.fge || m_sel.zb)
 		{
-			vmovaps(xmm0, ptr[ebx + 16]); // v.p
+			vmovaps(xmm0, ptr[ebx + offsetof(GSVertexSW, p)]); // v.p
 
 			if(m_sel.fwrite && m_sel.fge)
 			{
@@ -333,7 +334,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.edge || m_sel.tfx != TFX_NONE)
 		{
-			vmovaps(xmm4, ptr[ebx + 32]); // v.t
+			vmovaps(xmm4, ptr[ebx + offsetof(GSVertexSW, t)]); // v.t
 		}
 
 		if(m_sel.edge)
@@ -410,7 +411,7 @@ void GSDrawScanlineCodeGenerator::Init()
 			{
 				// GSVector4i vc = GSVector4i(v.c);
 
-				vcvttps2dq(xmm6, ptr[ebx]); // v.c
+				vcvttps2dq(xmm6, ptr[ebx + offsetof(GSVertexSW, c)]); // v.c
 
 				// vc = vc.upl16(vc.zwxy());
 

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSDrawScanlineCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE < 0x500 && !(defined(_M_AMD64) || defined(_WIN64))
 
@@ -293,7 +294,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.fwrite && m_sel.fge || m_sel.zb)
 		{
-			movaps(xmm0, ptr[ebx + 16]); // v.p
+			movaps(xmm0, ptr[ebx + offsetof(GSVertexSW, p)]); // v.p
 
 			if(m_sel.fwrite && m_sel.fge)
 			{
@@ -330,7 +331,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		if(m_sel.edge || m_sel.tfx != TFX_NONE)
 		{
-			movaps(xmm4, ptr[ebx + 32]); // v.t
+			movaps(xmm4, ptr[ebx + offsetof(GSVertexSW, t)]); // v.t
 		}
 
 		if(m_sel.edge)
@@ -410,7 +411,7 @@ void GSDrawScanlineCodeGenerator::Init()
 			{
 				// GSVector4i vc = GSVector4i(v.c);
 
-				cvttps2dq(xmm6, ptr[ebx]); // v.c
+				cvttps2dq(xmm6, ptr[ebx + offsetof(GSVertexSW, c)]); // v.c
 
 				// vc = vc.upl16(vc.zwxy());
 

plugins/GSdx/GSRasterizer.cpp

@@ -44,13 +44,13 @@ GSRasterizer::GSRasterizer(IDrawScanline* ds)
 	, m_id(0)
 	, m_threads(1)
 {
-	m_edge.buff = (GSScanline*)vmalloc(sizeof(GSScanline) * 2048, false);
+	m_edge.buff = (GSVertexSW*)vmalloc(sizeof(GSVertexSW) * 2048, false);
 	m_edge.count = 0;
 }
 
 GSRasterizer::~GSRasterizer()
 {
-	if(m_edge.buff != NULL) vmfree(m_edge.buff, sizeof(GSScanline) * 2048);
+	if(m_edge.buff != NULL) vmfree(m_edge.buff, sizeof(GSVertexSW) * 2048);
 
 	delete m_ds;
 }
@@ -119,8 +119,6 @@ void GSRasterizer::GetStats(GSRasterizerStats& stats)
 
 void GSRasterizer::DrawPoint(const GSVertexSW* v)
 {
-	// TODO: round to closest for point, prestep for line
-
 	GSVector4i p(v->p);
 
 	if(m_scissor.left <= p.x && p.x < m_scissor.right && m_scissor.top <= p.y && p.y < m_scissor.bottom)
@@ -142,22 +140,20 @@ void GSRasterizer::DrawLine(const GSVertexSW* v)
 
 	GSVector4 dp = dv.p.abs();
 
+	int i = (dp < dp.yxwz()).mask() & 1; // |dx| <= |dy|
+
 	if(m_ds->IsEdge())
 	{
-		int i = (dp < dp.yxwz()).mask() & 1; // |x| <= |y|
-
 		GSVertexSW dscan;
 
 		dscan.p = GSVector4::zero();
 		dscan.t = GSVector4::zero();
 		dscan.c = GSVector4::zero();
 
-		m_ds->SetupPrim(v, dscan);
-
 		DrawEdge(v[0], v[1], dv, i, 0);
 		DrawEdge(v[0], v[1], dv, i, 1);
 
-		FlushEdge();
+		Flush(v, dscan, true);
 
 		return;
 	}
@@ -188,34 +184,60 @@ void GSRasterizer::DrawLine(const GSVertexSW* v)
 			{
 				GSVertexSW dscan = dv / dv.p.xxxx();
 
-				m_ds->SetupPrim(v, dscan);
-
 				l.p = l.p.upl(r).xyzw(l.p); // r.x => l.y
 
 				DrawTriangleSection(p.y, p.y + 1, l, dl, dscan);
 
-				Flush();
+				Flush(v, dscan);
 			}
 		}
 
 		return;
 	}
 
-	int i = dpi.x > dpi.y ? 0 : 1;
-
-	GSVertexSW edge = v[0];
-	GSVertexSW dedge = dv / dp.v[i];
-
-	// TODO: prestep + clip with the scissor
-	// TODO: inline drawpoint + Flush()
-
 	int steps = dpi.v[i];
 
-	while(steps-- > 0)
+	if(steps > 0)
 	{
-		DrawPoint(&edge);
+		GSVertexSW edge = v[0];
+		GSVertexSW dedge = dv / GSVector4(dp.v[i]);
 
-		edge += dedge;
+		GSVertexSW* RESTRICT e = m_edge.buff;
+
+		while(1)
+		{
+			GSVector4i p(edge.p);
+
+			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))
+				{
+					*e = edge;
+
+					e->p.i16[0] = (int16)p.x;
+					e->p.i16[1] = (int16)p.y;
+					e->p.i16[2] = (int16)(p.x + 1);
+
+					e++;
+				}
+			}
+
+			if(--steps == 0) break;
+
+			edge += dedge;
+		}
+
+		m_edge.count = e - m_edge.buff;
+
+		m_stats.pixels += m_edge.count;
+
+		GSVertexSW dscan;
+
+		dscan.p = GSVector4::zero();
+		dscan.t = GSVector4::zero();
+		dscan.c = GSVector4::zero();
+
+		Flush(v, dscan);
 	}
 }
 
@@ -233,7 +255,7 @@ static const int s_abc[8][4] =
 
 void GSRasterizer::DrawTriangle(const GSVertexSW* vertices)
 {
-	// edge buffer is used here to avoid xmm save-restores (except when we do aa1 in the middle)
+	// TODO: GSVertexSW::c/t could be merged into a GSVector8
 
 	GSVertexSW v[4];
 	GSVertexSW dv[3];
@@ -276,7 +298,7 @@ void GSRasterizer::DrawTriangle(const GSVertexSW* vertices)
 	case 1: // a == b < c
 		ddv[1] = dv[1] / dv[1].p.yyyy();
 		ddv[2] = dv[2] / dv[2].p.yyyy();
-		longest = dv[0];
+		longest = dv[0]; // should be negated to be equal to "ddv[1] * dv[0].p.yyyy() - dv[0]", but it's easier to change the index of v/ddv later
 		break;
 	case 4: // a < b == c
 		ddv[0] = dv[0] / dv[0].p.yyyy();
@@ -319,9 +341,7 @@ void GSRasterizer::DrawTriangle(const GSVertexSW* vertices)
 		dscan.t = GSVector4::zero();
 		dscan.c = GSVector4::zero();
 
-		m_ds->SetupPrim(v, dscan);
+		Flush(v, dscan, true);
-
-		FlushEdge();
 	}
 
 	switch(i)
@@ -402,50 +422,42 @@ void GSRasterizer::DrawTriangle(const GSVertexSW* vertices)
 		__assume(0);
 	}
 
-	m_ds->SetupPrim(v, dscan);
+	Flush(v, dscan);
-
-	Flush();
 }
 
 void GSRasterizer::DrawTriangleSection(int top, int bottom, GSVertexSW& l, const GSVertexSW& dl, const GSVertexSW& dscan)
 {
 	ASSERT(top < bottom);
 
-	GSScanline* RESTRICT e = &m_edge.buff[m_edge.count];
+	GSVertexSW* RESTRICT e = &m_edge.buff[m_edge.count];
 
 	while(1)
 	{
-		do
+		if(IsOneOfMyScanlines(top))
 		{
-			if(IsOneOfMyScanlines(top))
+			GSVector4 lrf = l.p.ceil();
+			GSVector4 lrmax = lrf.max(m_fscissor.xxxx());
+			GSVector4 lrmin = lrf.min(m_fscissor.zzzz());
+			GSVector4i lr = GSVector4i(lrmax.xxyy(lrmin));
+
+			int left = lr.extract32<0>();
+			int right = lr.extract32<2>();
+
+			int pixels = right - left;
+
+			if(pixels > 0)
 			{
-				GSVector4 lr = l.p.ceil();
+				m_stats.pixels += pixels;
 
-				GSVector4 lrmax = lr.max(m_fscissor.xxxx());
+				*e = l + dscan * (lrmax - l.p).xxxx();
-				GSVector4 lrmin = lr.min(m_fscissor.zzzz());
 
-				GSVector4i lri = GSVector4i(lrmax.xxyy(lrmin));
+				e->p.i16[0] = (int16)left;
+				e->p.i16[1] = (int16)top;
+				e->p.i16[2] = (int16)right;
 
-				int left = lri.extract32<0>();
+				e++;
-				int right = lri.extract32<2>();
-
-				int pixels = right - left;
-
-				if(pixels > 0)
-				{
-					m_stats.pixels += pixels;
-
-					e->scan = l + dscan * (lrmax - l.p).xxxx();
-
-					e->p.left = left;
-					e->p.top = top;
-					e->p.right = right;
-
-					e++;
-				}
 			}
 		}
-		while(0);
 
 		if(++top >= bottom) break;
 
@@ -508,7 +520,7 @@ void GSRasterizer::DrawSprite(const GSVertexSW* vertices)
 
 	m_ds->SetupPrim(v, dscan);
 
-	for(; r.top < r.bottom; r.top++, scan.t += dedge.t)
+	while(1)
 	{
 		if(IsOneOfMyScanlines(r.top))
 		{
@@ -516,6 +528,10 @@ void GSRasterizer::DrawSprite(const GSVertexSW* vertices)
 
 			m_ds->DrawScanline(r.right, r.left, r.top, scan);
 		}
+
+		if(++r.top >= r.bottom) break;
+
+		scan.t += dedge.t;
 	}
 }
 
@@ -531,7 +547,7 @@ 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)
 
-	GSScanline* RESTRICT dst = &m_edge.buff[m_edge.count];
+	GSVertexSW* RESTRICT e = &m_edge.buff[m_edge.count];
 			
 	GSVector4 lrtb = v0.p.upl(v1.p).ceil();
 
@@ -540,7 +556,7 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		GSVector4 tbmax = lrtb.max(m_fscissor.yyyy());
 		GSVector4 tbmin = lrtb.min(m_fscissor.wwww());
 
-		GSVector4i tbi = GSVector4i(tbmax.zwzw(tbmin));
+		GSVector4i tb = GSVector4i(tbmax.zwzw(tbmin));
 
 		int top, bottom;
 
@@ -548,8 +564,8 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 
 		if((dv.p >= GSVector4::zero()).mask() & 2)
 		{
-			top = tbi.extract32<0>();
+			top = tb.extract32<0>();
-			bottom = tbi.extract32<3>();
+			bottom = tb.extract32<3>();
 
 			if(top >= bottom) return;
 
@@ -560,8 +576,8 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		}
 		else
 		{
-			top = tbi.extract32<1>();
+			top = tb.extract32<1>();
-			bottom = tbi.extract32<2>();
+			bottom = tb.extract32<2>();
 
 			if(top >= bottom) return;
 
@@ -580,26 +596,23 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		{
 			while(1)
 			{
-				do
+				int xi = x >> 16;
+				int xf = x & 0xffff;
+
+				if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
 				{
-					int xi = x >> 16;
+					m_stats.pixels++;
-					int xf = x & 0xffff;
 
-					if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
+					*e = edge;
-					{
-						m_stats.pixels++;
 
-						dst->scan = edge;
+					e->t.u32[3] = (0x10000 - xf) & 0xffff;
-						dst->scan.t.u32[3] = (0x10000 - xf) & 0xffff;
 
-						dst->p.left = xi;
+					e->p.i16[0] = (int16)xi;
-						dst->p.top = top;
+					e->p.i16[1] = (int16)top;
-						dst->p.right = xi + 1;
+					e->p.i16[2] = (int16)(xi + 1);
 
-						dst++;
+					e++;
-					}
 				}
-				while(0);
 
 				if(++top >= bottom) break;
 
@@ -611,26 +624,23 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		{
 			while(1)
 			{
-				do
+				int xi = (x >> 16) + 1;
+				int xf = x & 0xffff;
+
+				if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
 				{
-					int xi = (x >> 16) + 1;
+					m_stats.pixels++;
-					int xf = x & 0xffff;
 
-					if(m_scissor.left <= xi && xi < m_scissor.right && IsOneOfMyScanlines(xi))
+					*e = edge;
-					{
-						m_stats.pixels++;
 
-						dst->scan = edge;
+					e->t.u32[3] = xf;
-						dst->scan.t.u32[3] = xf;
 
-						dst->p.left = xi;
+					e->p.i16[0] = (int16)xi;
-						dst->p.top = top;
+					e->p.i16[1] = (int16)top;
-						dst->p.right = xi + 1;
+					e->p.i16[2] = (int16)(xi + 1);
 
-						dst++;
+					e++;
-					}
 				}
-				while(0);
 
 				if(++top >= bottom) break;
 
@@ -644,7 +654,7 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		GSVector4 lrmax = lrtb.max(m_fscissor.xxxx());
 		GSVector4 lrmin = lrtb.min(m_fscissor.zzzz());
 
-		GSVector4i lri = GSVector4i(lrmax.xyxy(lrmin));
+		GSVector4i lr = GSVector4i(lrmax.xyxy(lrmin));
 
 		int left, right;
 
@@ -652,8 +662,8 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 
 		if((dv.p >= GSVector4::zero()).mask() & 1)
 		{
-			left = lri.extract32<0>();
+			left = lr.extract32<0>();
-			right = lri.extract32<3>();
+			right = lr.extract32<3>();
 
 			if(left >= right) return;
 
@@ -664,8 +674,8 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		}
 		else
 		{
-			left = lri.extract32<1>();
+			left = lr.extract32<1>();
-			right = lri.extract32<2>();
+			right = lr.extract32<2>();
 
 			if(left >= right) return;
 
@@ -684,26 +694,23 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		{
 			while(1)
 			{
-				do
+				int yi = y >> 16;
+				int yf = y & 0xffff;
+
+				if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
 				{
-					int yi = y >> 16;
+					m_stats.pixels++;
-					int yf = y & 0xffff;
 
-					if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
+					*e = edge;
-					{
+					
-						m_stats.pixels++;
+					e->t.u32[3] = (0x10000 - yf) & 0xffff;
 
-						dst->scan = edge;
+					e->p.i16[0] = (int16)left;
-						dst->scan.t.u32[3] = (0x10000 - yf) & 0xffff;
+					e->p.i16[1] = (int16)yi;
+					e->p.i16[2] = (int16)(left + 1);
 
-						dst->p.left = left;
+					e++;
-						dst->p.top = yi;
-						dst->p.right = left + 1;
-
-						dst++;
-					}
 				}
-				while(0);
 
 				if(++left >= right) break;
 
@@ -715,26 +722,23 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		{
 			while(1)
 			{
-				do
+				int yi = (y >> 16) + 1;
+				int yf = y & 0xffff;
+
+				if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
 				{
-					int yi = (y >> 16) + 1;
+					m_stats.pixels++;
-					int yf = y & 0xffff;
 
-					if(m_scissor.top <= yi && yi < m_scissor.bottom && IsOneOfMyScanlines(yi))
+					*e = edge;
-					{
+					
-						m_stats.pixels++;
+					e->t.u32[3] = yf;
 
-						dst->scan = edge;
+					e->p.i16[0] = (int16)left;
-						dst->scan.t.u32[3] = yf;
+					e->p.i16[1] = (int16)yi;
+					e->p.i16[2] = (int16)(left + 1);
 
-						dst->p.left = left;
+					e++;
-						dst->p.top = yi;
-						dst->p.right = left + 1;
-
-						dst++;
-					}
 				}
-				while(0);
 
 				if(++left >= right) break;
 
@@ -744,33 +748,36 @@ void GSRasterizer::DrawEdge(const GSVertexSW& v0, const GSVertexSW& v1, const GS
 		}
 	}
 
-	m_edge.count += dst - &m_edge.buff[m_edge.count];
+	m_edge.count += e - &m_edge.buff[m_edge.count];
 }
 
-void GSRasterizer::Flush()
+void GSRasterizer::Flush(const GSVertexSW* vertices, const GSVertexSW& dscan, bool edge)
 {
 	// TODO: on win64 this could be the place where xmm6-15 are preserved (not by each DrawScanline)
 
-	const GSScanline* s = m_edge.buff;
+	int count = m_edge.count;
 
-	for(int count = m_edge.count; count > 0; count--, s++)
+	if(count > 0)
 	{
-		m_ds->DrawScanline(s->p.right, s->p.left, s->p.top, s->scan);
+		m_ds->SetupPrim(vertices, dscan);
+
+		const GSVertexSW* RESTRICT e = m_edge.buff;
+
+		int i = 0;
+
+		if(!edge)
+		{
+			do {m_ds->DrawScanline(e[i].p.i16[2], e[i].p.i16[0], e[i].p.i16[1], e[i]);}
+			while(++i < count);
+		}
+		else
+		{
+			do {m_ds->DrawEdge(e[i].p.i16[2], e[i].p.i16[0], e[i].p.i16[1], e[i]);}
+			while(++i < count);
+		}
+
+		m_edge.count = 0;
 	}
-
-	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

@@ -81,8 +81,6 @@ public:
 
 __aligned(class, 32) GSRasterizer : public GSAlignedClass<32>, public IRasterizer
 {
-	struct GSScanline {GSVertexSW scan; GSVector4i p;};
-
 protected:
 	IDrawScanline* m_ds;
 	int m_id;
@@ -90,7 +88,7 @@ protected:
 	GSRasterizerStats m_stats;
 	GSVector4i m_scissor;
 	GSVector4 m_fscissor;
-	struct {GSScanline* buff; int count;} m_edge;
+	struct {GSVertexSW* buff; int count;} m_edge;
 
 	void DrawPoint(const GSVertexSW* v);
 	void DrawLine(const GSVertexSW* v);
@@ -104,8 +102,7 @@ protected:
 
 	__forceinline bool IsOneOfMyScanlines(int scanline) const;
 
-	void Flush();
+	__forceinline void Flush(const GSVertexSW* vertices, const GSVertexSW& dscan, bool edge = false);
-	void FlushEdge();
 
 public:
 	GSRasterizer(IDrawScanline* ds);

plugins/GSdx/GSSetupPrimCodeGenerator.x64.avx.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSSetupPrimCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE >= 0x500 && (defined(_M_AMD64) || defined(_WIN64))
 
@@ -70,7 +71,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = dscan.p;
 
-		vmovaps(xmm0, ptr[rdx + 16]);
+		vmovaps(xmm0, ptr[rdx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -122,7 +123,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = vertices[0].p;
 
-		vmovaps(xmm0, ptr[rcx + 16]);
+		vmovaps(xmm0, ptr[rcx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -179,7 +180,7 @@ void GSSetupPrimCodeGenerator::Texture()
 
 	// GSVector4 t = dscan.t;
 
-	vmovaps(xmm0, ptr[rdx + 32]);
+	vmovaps(xmm0, ptr[rdx + offsetof(GSVertexSW, t)]);
 
 	vmulps(xmm1, xmm0, xmm3);
 
@@ -249,7 +250,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4 c = dscan.c;
 
-		vmovaps(xmm0, ptr[rdx]);
+		vmovaps(xmm0, ptr[rdx + offsetof(GSVertexSW, c)]);
 
 		// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
 
@@ -289,7 +290,7 @@ void GSSetupPrimCodeGenerator::Color()
 
 		// GSVector4 c = dscan.c;
 
-		vmovaps(xmm0, ptr[rdx]); // not enough regs, have to reload it
+		vmovaps(xmm0, ptr[rdx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
 
 		// GSVector4 dg = c.yyyy();
 		// GSVector4 da = c.wwww();
@@ -321,7 +322,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4i c = GSVector4i(vertices[0].c);
 
-		vcvttps2dq(xmm0, ptr[rcx]);
+		vcvttps2dq(xmm0, ptr[rcx + offsetof(GSVertexSW, c)]);
 
 		// c = c.upl16(c.zwxy());
 

plugins/GSdx/GSSetupPrimCodeGenerator.x64.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSSetupPrimCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE < 0x500 && (defined(_M_AMD64) || defined(_WIN64))
 
@@ -68,7 +69,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = dscan.p;
 
-		movaps(xmm0, ptr[rdx + 16]);
+		movaps(xmm0, ptr[rdx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -125,7 +126,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = vertices[0].p;
 
-		movaps(xmm0, ptr[rcx + 16]);
+		movaps(xmm0, ptr[rcx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -183,7 +184,7 @@ void GSSetupPrimCodeGenerator::Texture()
 
 	// GSVector4 t = dscan.t;
 
-	movaps(xmm0, ptr[rdx + 32]);
+	movaps(xmm0, ptr[rdx + offsetof(GSVertexSW, t)]);
 
 	movaps(xmm1, xmm0);
 	mulps(xmm1, xmm3);
@@ -256,7 +257,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4 c = dscan.c;
 
-		movaps(xmm0, ptr[rdx]);
+		movaps(xmm0, ptr[rdx + offsetof(GSVertexSW, c)]);
 		movaps(xmm1, xmm0);
 
 		// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
@@ -300,7 +301,7 @@ void GSSetupPrimCodeGenerator::Color()
 
 		// GSVector4 c = dscan.c;
 
-		movaps(xmm0, ptr[rdx]); // not enough regs, have to reload it
+		movaps(xmm0, ptr[rdx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
 		movaps(xmm1, xmm0);
 
 		// GSVector4 dg = c.yyyy();
@@ -335,7 +336,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4i c = GSVector4i(vertices[0].c);
 
-		cvttps2dq(xmm0, ptr[rcx]);
+		cvttps2dq(xmm0, ptr[rcx + offsetof(GSVertexSW, c)]);
 
 		// c = c.upl16(c.zwxy());
 

plugins/GSdx/GSSetupPrimCodeGenerator.x86.avx.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSSetupPrimCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE >= 0x500 && !(defined(_M_AMD64) || defined(_WIN64))
 
@@ -56,7 +57,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = dscan.p;
 
-		vmovaps(xmm0, ptr[edx + 16]);
+		vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -108,7 +109,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = vertices[0].p;
 
-		vmovaps(xmm0, ptr[ecx + 16]);
+		vmovaps(xmm0, ptr[ecx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -163,7 +164,7 @@ void GSSetupPrimCodeGenerator::Texture()
 
 	// GSVector4 t = dscan.t;
 
-	vmovaps(xmm0, ptr[edx + 32]);
+	vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, t)]);
 
 	vmulps(xmm1, xmm0, xmm3);
 
@@ -233,7 +234,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4 c = dscan.c;
 
-		vmovaps(xmm0, ptr[edx]);
+		vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]);
 
 		// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
 
@@ -273,7 +274,7 @@ void GSSetupPrimCodeGenerator::Color()
 
 		// GSVector4 c = dscan.c;
 
-		vmovaps(xmm0, ptr[edx]); // not enough regs, have to reload it
+		vmovaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
 
 		// GSVector4 dg = c.yyyy();
 		// GSVector4 da = c.wwww();
@@ -305,7 +306,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4i c = GSVector4i(vertices[0].c);
 
-		vcvttps2dq(xmm0, ptr[ecx]);
+		vcvttps2dq(xmm0, ptr[ecx + offsetof(GSVertexSW, c)]);
 
 		// c = c.upl16(c.zwxy());
 

plugins/GSdx/GSSetupPrimCodeGenerator.x86.cpp

@@ -21,6 +21,7 @@
 
 #include "stdafx.h"
 #include "GSSetupPrimCodeGenerator.h"
+#include "GSVertexSW.h"
 
 #if _M_SSE < 0x500 && !(defined(_M_AMD64) || defined(_WIN64))
 
@@ -56,7 +57,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = dscan.p;
 
-		movaps(xmm0, ptr[edx + 16]);
+		movaps(xmm0, ptr[edx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -113,7 +114,7 @@ void GSSetupPrimCodeGenerator::Depth()
 	{
 		// GSVector4 p = vertices[0].p;
 
-		movaps(xmm0, ptr[ecx + 16]);
+		movaps(xmm0, ptr[ecx + offsetof(GSVertexSW, p)]);
 
 		if(m_en.f)
 		{
@@ -168,7 +169,7 @@ void GSSetupPrimCodeGenerator::Texture()
 
 	// GSVector4 t = dscan.t;
 
-	movaps(xmm0, ptr[edx + 32]);
+	movaps(xmm0, ptr[edx + offsetof(GSVertexSW, t)]);
 
 	movaps(xmm1, xmm0);
 	mulps(xmm1, xmm3);
@@ -241,7 +242,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4 c = dscan.c;
 
-		movaps(xmm0, ptr[edx]);
+		movaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]);
 		movaps(xmm1, xmm0);
 
 		// m_local.d4.c = GSVector4i(c * 4.0f).xzyw().ps32();
@@ -285,7 +286,7 @@ void GSSetupPrimCodeGenerator::Color()
 
 		// GSVector4 c = dscan.c;
 
-		movaps(xmm0, ptr[edx]); // not enough regs, have to reload it
+		movaps(xmm0, ptr[edx + offsetof(GSVertexSW, c)]); // not enough regs, have to reload it
 		movaps(xmm1, xmm0);
 
 		// GSVector4 dg = c.yyyy();
@@ -320,7 +321,7 @@ void GSSetupPrimCodeGenerator::Color()
 	{
 		// GSVector4i c = GSVector4i(vertices[0].c);
 
-		movaps(xmm0, ptr[ecx]);
+		movaps(xmm0, ptr[ecx + offsetof(GSVertexSW, c)]);
 		cvttps2dq(xmm0, xmm0);
 
 		// c = c.upl16(c.zwxy());

plugins/GSdx/GSVector.h

@@ -3030,6 +3030,8 @@ public:
 		uint32 u32[8];
 		uint64 u64[4];
 		__m256 m;
+
+		// TODO: _M_SSE < 0x500 => union {__m128 m0, m1;}; and replace each function with a pair of 128 bit intructions
 	};
 
 	__forceinline GSVector8()
@@ -3050,7 +3052,7 @@ public:
 
 	__forceinline GSVector8(__m128 m0, __m128 m1)
 	{
-		m = zero().insert<0>(m0).insert<1>(m1);
+		m = _mm256_permute2f128_ps(_mm256_castps128_ps256(m0), _mm256_castps128_ps256(m1), 0x20);
 	}
 
 	__forceinline GSVector8(const GSVector8& v)
@@ -3065,7 +3067,8 @@ public:
 
 	__forceinline explicit GSVector8(__m128 m)
 	{
-		this->m = zero().insert<0>(m).insert<1>(m);
+		this->m = _mm256_castps128_ps256(m);
+		this->m = _mm256_permute2f128_ps(this->m, this->m, 0);
 	}
 
 	__forceinline explicit GSVector8(__m256 m)
@@ -3087,7 +3090,8 @@ public:
 
 	__forceinline void operator = (__m128 m)
 	{
-		this->m = zero().insert<0>(m).insert<1>(m);
+		this->m = _mm256_castps128_ps256(m);
+		this->m = _mm256_permute2f128_ps(this->m, this->m, 0);
 	}
 
 	__forceinline void operator = (__m256 m)
@@ -3104,7 +3108,7 @@ public:
 
 	__forceinline GSVector8 abs() const
 	{
-		return *this & cast(GSVector8i(GSVector4i::x7fffffff()));
+		return *this & cast(GSVector8i(GSVector4i::x7fffffff())); // TODO: add GSVector8 consts
 	}
 
 	__forceinline GSVector8 neg() const
@@ -3143,17 +3147,27 @@ public:
 
 	// TODO
 
+	__forceinline GSVector8 min(const GSVector8& a) const
+	{
+		return GSVector8(_mm256_min_ps(m, a));
+	}
+
+	__forceinline GSVector8 max(const GSVector8& a) const
+	{
+		return GSVector8(_mm256_max_ps(m, a));
+	}
+
 	__forceinline GSVector8 blend8(const GSVector8& a, const GSVector8& mask)  const
 	{
 		return GSVector8(_mm256_blendv_ps(m, a, mask));
 	}
 
-	__forceinline GSVector8 upl32(const GSVector8& a) const
+	__forceinline GSVector8 upl(const GSVector8& a) const
 	{
 		return GSVector8(_mm256_unpacklo_ps(m, a));
 	}
 
-	__forceinline GSVector8 uph32(const GSVector8& a) const
+	__forceinline GSVector8 uph(const GSVector8& a) const
 	{
 		return GSVector8(_mm256_unpackhi_ps(m, a));
 	}
@@ -3392,6 +3406,23 @@ public:
 		return GSVector8(_mm256_cmp_ps(v1, v2, _CMP_LE_OQ));
 	}
 
+	#define VECTOR8_PERMUTE_2(xs, xn, ys, yn) \
+		__forceinline GSVector8 xs##ys() const {return GSVector8(_mm256_permute2f128_ps(m, m, xn | (yn << 4)));} \
+		__forceinline GSVector8 xs##ys(const GSVector8& v) const {return GSVector8(_mm256_permute2f128_ps(m, v.m, xn | (yn << 4)));} \
+
+	#define VECTOR8_PERMUTE_1(xs, xn) \
+		VECTOR8_PERMUTE_2(xs, xn, x, 0) \
+		VECTOR8_PERMUTE_2(xs, xn, y, 1) \
+		VECTOR8_PERMUTE_2(xs, xn, z, 2) \
+		VECTOR8_PERMUTE_2(xs, xn, w, 3) \
+		VECTOR8_PERMUTE_2(xs, xn, _, 8) \
+
+	VECTOR8_PERMUTE_1(x, 0)
+	VECTOR8_PERMUTE_1(y, 1)
+	VECTOR8_PERMUTE_1(z, 2)
+	VECTOR8_PERMUTE_1(w, 3)
+	VECTOR8_PERMUTE_1(_, 8)
+
 	#define VECTOR8_SHUFFLE_4(xs, xn, ys, yn, zs, zn, ws, wn) \
 		__forceinline GSVector8 xs##ys##zs##ws() const {return GSVector8(_mm256_permute_ps(m, _MM_SHUFFLE(wn, zn, yn, xn)));} \
 		__forceinline GSVector8 xs##ys##zs##ws(const GSVector8& v) const {return GSVector8(_mm256_shuffle_ps(m, v.m, _MM_SHUFFLE(wn, zn, yn, xn)));} \

plugins/GSdx/GSVertexSW.h

@@ -25,10 +25,10 @@
 
 __aligned(struct, 16) GSVertexSW
 {
-	GSVector4 c, p, t;
+	GSVector4 p, t, c;
 
-	GSVertexSW() {}
+	__forceinline GSVertexSW() {}
-	GSVertexSW(const GSVertexSW& v) {*this = v;}
+	__forceinline GSVertexSW(const GSVertexSW& v) {*this = v;}
 
 	__forceinline 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;}
@@ -37,8 +37,6 @@ __aligned(struct, 16) GSVertexSW
 	friend GSVertexSW operator - (const GSVertexSW& v1, const GSVertexSW& v2);
 	friend GSVertexSW operator * (const GSVertexSW& v, const GSVector4& vv);
 	friend GSVertexSW operator / (const GSVertexSW& v, const GSVector4& vv);
-	friend GSVertexSW operator * (const GSVertexSW& v, float f);
-	friend GSVertexSW operator / (const GSVertexSW& v, float f);
 
 	static bool IsQuad(const GSVertexSW* v, int& tl, int& br)
 	{
@@ -192,22 +190,3 @@ __forceinline GSVertexSW operator / (const GSVertexSW& v, const GSVector4& vv)
 	return v0;
 }
 
-__forceinline GSVertexSW operator * (const GSVertexSW& v, float f)
-{
-	GSVertexSW v0;
-	GSVector4 vf(f);
-	v0.c = v.c * vf;
-	v0.p = v.p * vf;
-	v0.t = v.t * vf;
-	return v0;
-}
-
-__forceinline GSVertexSW operator / (const GSVertexSW& v, float f)
-{
-	GSVertexSW v0;
-	GSVector4 vf(f);
-	v0.c = v.c / vf;
-	v0.p = v.p / vf;
-	v0.t = v.t / vf;
-	return v0;
-}

plugins/GSdx/GSVertexTrace.h

@@ -31,9 +31,11 @@ class GSState;
 
 __aligned(class, 32) GSVertexTrace
 {
+public:
 	struct Vertex {GSVector4i c; GSVector4 p, t;};
 	struct VertexAlpha {int min, max; bool valid;};
 
+private:
 	typedef void (*VertexTracePtr)(int count, const void* v, Vertex& min, Vertex& max);
 
 	class CGSW : public GSCodeGenerator

plugins/GSdx/GSVertexTrace.x64.avx.cpp

@@ -90,7 +90,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		vmovaps(xmm1, ptr[rdx + 1 * sizeof(GSVertexSW) + 32]);
+		vmovaps(xmm1, ptr[rdx + 1 * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 		vshufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
 	}
 
@@ -101,7 +101,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.c = min.c.minv(v[i + j].c);
 			// max.c = max.c.maxv(v[i + j].c);
 
-			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW)]);
+			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, c)]);
 
 			vminps(xmm2, xmm0);
 			vmaxps(xmm3, xmm0);
@@ -110,7 +110,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + 16]);
+		vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, p)]);
 
 		vminps(xmm4, xmm0);
 		vmaxps(xmm5, xmm0);
@@ -120,7 +120,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.t = min.t.minv(v[i + j].t);
 			// max.t = max.t.maxv(v[i + j].t);
 
-			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + 32]);
+			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 
 			if(!fst)
 			{
@@ -149,20 +149,20 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 	{
 		vcvttps2dq(xmm2, xmm2);
 		vpsrld(xmm2, 7);
-		vmovaps(ptr[r8], xmm2);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
 
 		vcvttps2dq(xmm3, xmm3);
 		vpsrld(xmm3, 7);
-		vmovaps(ptr[r9], xmm3);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
-	vmovaps(ptr[r8 + 16], xmm4);
+	vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[r9 + 16], xmm5);
+	vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
-		vmovaps(ptr[r8 + 32], xmm6);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[r9 + 32], xmm7);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	vmovdqa(xmm6, ptr[rsp + 0]);
@@ -239,7 +239,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		vmovaps(xmm1, ptr[rdx + 5 * sizeof(GSVertexHW9) + 16]);
+		vmovaps(xmm1, ptr[rdx + 5 * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 		vshufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
 	}
 
@@ -248,7 +248,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + 16]);
+		vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 
 		vminps(xmm4, xmm0);
 		vmaxps(xmm5, xmm0);
@@ -260,7 +260,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 		if(color && (iip || j == n - 1) || tme)
 		{
-			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9)]);
+			vmovaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, t)]);
 		}
 
 		if(color && (iip || j == n - 1))
@@ -309,15 +309,15 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		vpshufd(xmm3, xmm3, _MM_SHUFFLE(2, 2, 2, 2));
 		vpmovzxbd(xmm3, xmm3);
 
-		vmovaps(ptr[r8], xmm2);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		vmovaps(ptr[r9], xmm3);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin;
 	// m_max.p = pmax;
 
-	vmovaps(ptr[r8 + 16], xmm4);
+	vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[r9 + 16], xmm5);
+	vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
@@ -327,8 +327,8 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		vshufps(xmm6, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 		vshufps(xmm7, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-		vmovaps(ptr[r8 + 32], xmm6);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[r9 + 32], xmm7);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	vmovdqa(xmm6, ptr[rsp + 0]);
@@ -463,8 +463,8 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 		vpshufd(xmm3, xmm3, _MM_SHUFFLE(2, 2, 2, 2));
 		vpmovzxbd(xmm3, xmm3);
 
-		vmovaps(ptr[r8], xmm2);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		vmovaps(ptr[r9], xmm3);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin.xyww();
@@ -473,16 +473,16 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 	vshufps(xmm4, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 	vshufps(xmm5, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-	vmovaps(ptr[r8 + 16], xmm4);
+	vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[r9 + 16], xmm5);
+	vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
 		// m_min.t = tmin;
 		// m_max.t = tmax;
 
-		vmovaps(ptr[r8 + 32], xmm6);
+		vmovaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[r9 + 32], xmm7);
+		vmovaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	vmovdqa(xmm6, ptr[rsp + 0]);

plugins/GSdx/GSVertexTrace.x64.cpp

@@ -93,7 +93,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		movaps(xmm1, ptr[rdx + 1 * sizeof(GSVertexSW) + 32]);
+		movaps(xmm1, ptr[rdx + 1 * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 		shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
 	}
 
@@ -104,7 +104,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.c = min.c.minv(v[i + j].c);
 			// max.c = max.c.maxv(v[i + j].c);
 
-			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW)]);
+			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, c)]);
 
 			minps(xmm2, xmm0);
 			maxps(xmm3, xmm0);
@@ -113,7 +113,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + 16]);
+		movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, p)]);
 
 		minps(xmm4, xmm0);
 		maxps(xmm5, xmm0);
@@ -123,7 +123,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.t = min.t.minv(v[i + j].t);
 			// max.t = max.t.maxv(v[i + j].t);
 
-			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + 32]);
+			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 
 			if(!fst)
 			{
@@ -153,20 +153,20 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 	{
 		cvttps2dq(xmm2, xmm2);
 		psrld(xmm2, 7);
-		movaps(ptr[r8], xmm2);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
 
 		cvttps2dq(xmm3, xmm3);
 		psrld(xmm3, 7);
-		movaps(ptr[r9], xmm3);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
-	movaps(ptr[r8 + 16], xmm4);
+	movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[r9 + 16], xmm5);
+	movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
-		movaps(ptr[r8 + 32], xmm6);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[r9 + 32], xmm7);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	movdqa(xmm6, ptr[rsp + 0]);
@@ -246,7 +246,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		movaps(xmm1, ptr[rdx + 5 * sizeof(GSVertexHW9) + 16]);
+		movaps(xmm1, ptr[rdx + 5 * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 		shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
 	}
 
@@ -255,7 +255,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		movaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + 16]);
+		movaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 
 		minps(xmm4, xmm0);
 		maxps(xmm5, xmm0);
@@ -268,7 +268,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 		if(color && (iip || j == n - 1) || tme)
 		{
-			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9)]);
+			movaps(xmm0, ptr[rdx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, t)]);
 		}
 
 		if(color && (iip || j == n - 1))
@@ -330,15 +330,15 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 			punpcklwd(xmm3, xmm0);
 		}
 
-		movaps(ptr[r8], xmm2);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		movaps(ptr[r9], xmm3);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin;
 	// m_max.p = pmax;
 
-	movaps(ptr[r8 + 16], xmm4);
+	movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[r9 + 16], xmm5);
+	movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
@@ -348,8 +348,8 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		shufps(xmm6, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 		shufps(xmm7, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-		movaps(ptr[r8 + 32], xmm6);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[r9 + 32], xmm7);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	movdqa(xmm6, ptr[rsp + 0]);
@@ -510,8 +510,8 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 			punpcklwd(xmm3, xmm0);
 		}
 
-		movaps(ptr[r8], xmm2);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		movaps(ptr[r9], xmm3);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin.xyww();
@@ -520,16 +520,16 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 	shufps(xmm4, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 	shufps(xmm5, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-	movaps(ptr[r8 + 16], xmm4);
+	movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[r9 + 16], xmm5);
+	movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
 		// m_min.t = tmin;
 		// m_max.t = tmax;
 
-		movaps(ptr[r8 + 32], xmm6);
+		movaps(ptr[r8 + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[r9 + 32], xmm7);
+		movaps(ptr[r9 + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	movdqa(xmm6, ptr[rsp + 0]);

plugins/GSdx/GSVertexTrace.x86.avx.cpp

@@ -92,7 +92,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		vmovaps(xmm1, ptr[edx + 1 * sizeof(GSVertexSW) + 32]);
+		vmovaps(xmm1, ptr[edx + 1 * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 		vshufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
 	}
 
@@ -103,7 +103,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.c = min.c.minv(v[i + j].c);
 			// max.c = max.c.maxv(v[i + j].c);
 
-			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW)]);
+			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, c)]);
 
 			vminps(xmm2, xmm0);
 			vmaxps(xmm3, xmm0);
@@ -112,7 +112,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + 16]);
+		vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, p)]);
 
 		vminps(xmm4, xmm0);
 		vmaxps(xmm5, xmm0);
@@ -122,7 +122,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.t = min.t.minv(v[i + j].t);
 			// max.t = max.t.maxv(v[i + j].t);
 
-			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + 32]);
+			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 
 			if(!fst)
 			{
@@ -154,20 +154,20 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 	{
 		vcvttps2dq(xmm2, xmm2);
 		vpsrld(xmm2, 7);
-		vmovaps(ptr[eax], xmm2);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
 
 		vcvttps2dq(xmm3, xmm3);
 		vpsrld(xmm3, 7);
-		vmovaps(ptr[edx], xmm3);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
-	vmovaps(ptr[eax + 16], xmm4);
+	vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[edx + 16], xmm5);
+	vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
-		vmovaps(ptr[eax + 32], xmm6);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[edx + 32], xmm7);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();
@@ -235,7 +235,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		vmovaps(xmm1, ptr[edx + 5 * sizeof(GSVertexHW9) + 16]);
+		vmovaps(xmm1, ptr[edx + 5 * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 		vshufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
 	}
 
@@ -244,7 +244,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + 16]);
+		vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 
 		vminps(xmm4, xmm0);
 		vmaxps(xmm5, xmm0);
@@ -256,7 +256,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 		if(color && (iip || j == n - 1) || tme)
 		{
-			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9)]);
+			vmovaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, t)]);
 		}
 
 		if(color && (iip || j == n - 1))
@@ -308,15 +308,15 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		vpshufd(xmm3, xmm3, _MM_SHUFFLE(2, 2, 2, 2));
 		vpmovzxbd(xmm3, xmm3);
 
-		vmovaps(ptr[eax], xmm2);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		vmovaps(ptr[edx], xmm3);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin;
 	// m_max.p = pmax;
 
-	vmovaps(ptr[eax + 16], xmm4);
+	vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[edx + 16], xmm5);
+	vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
@@ -326,8 +326,8 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		vshufps(xmm6, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 		vshufps(xmm7, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-		vmovaps(ptr[eax + 32], xmm6);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[edx + 32], xmm7);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();
@@ -456,8 +456,8 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 		vpshufd(xmm3, xmm3, _MM_SHUFFLE(2, 2, 2, 2));
 		vpmovzxbd(xmm3, xmm3);
 
-		vmovaps(ptr[eax], xmm2);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		vmovaps(ptr[edx], xmm3);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin.xyww();
@@ -466,16 +466,16 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 	vshufps(xmm4, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 	vshufps(xmm5, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-	vmovaps(ptr[eax + 16], xmm4);
+	vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	vmovaps(ptr[edx + 16], xmm5);
+	vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
 		// m_min.t = tmin;
 		// m_max.t = tmax;
 
-		vmovaps(ptr[eax + 32], xmm6);
+		vmovaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		vmovaps(ptr[edx + 32], xmm7);
+		vmovaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();

plugins/GSdx/GSVertexTrace.x86.cpp

@@ -95,7 +95,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		movaps(xmm1, ptr[edx + 1 * sizeof(GSVertexSW) + 32]);
+		movaps(xmm1, ptr[edx + 1 * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 		shufps(xmm1, xmm1, _MM_SHUFFLE(2, 2, 2, 2));
 	}
 
@@ -106,7 +106,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.c = min.c.minv(v[i + j].c);
 			// max.c = max.c.maxv(v[i + j].c);
 
-			movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW)]);
+			movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, c)]);
 
 			minps(xmm2, xmm0);
 			maxps(xmm3, xmm0);
@@ -115,7 +115,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + 16]);
+		movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, p)]);
 
 		minps(xmm4, xmm0);
 		maxps(xmm5, xmm0);
@@ -125,7 +125,7 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 			// min.t = min.t.minv(v[i + j].t);
 			// max.t = max.t.maxv(v[i + j].t);
 
-			movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + 32]);
+			movaps(xmm0, ptr[edx + j * sizeof(GSVertexSW) + offsetof(GSVertexSW, t)]);
 
 			if(!fst)
 			{
@@ -158,20 +158,20 @@ GSVertexTrace::CGSW::CGSW(const void* param, uint32 key, void* code, size_t maxs
 	{
 		cvttps2dq(xmm2, xmm2);
 		psrld(xmm2, 7);
-		movaps(ptr[eax], xmm2);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
 
 		cvttps2dq(xmm3, xmm3);
 		psrld(xmm3, 7);
-		movaps(ptr[edx], xmm3);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
-	movaps(ptr[eax + 16], xmm4);
+	movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[edx + 16], xmm5);
+	movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
-		movaps(ptr[eax + 32], xmm6);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[edx + 32], xmm7);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();
@@ -242,7 +242,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 	if(tme && !fst && primclass == GS_SPRITE_CLASS)
 	{
-		movaps(xmm1, ptr[edx + 5 * sizeof(GSVertexHW9) + 16]);
+		movaps(xmm1, ptr[edx + 5 * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 		shufps(xmm1, xmm1, _MM_SHUFFLE(3, 3, 3, 3));
 	}
 
@@ -251,7 +251,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		// min.p = min.p.minv(v[i + j].p);
 		// max.p = max.p.maxv(v[i + j].p);
 
-		movaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + 16]);
+		movaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, p)]);
 
 		minps(xmm4, xmm0);
 		maxps(xmm5, xmm0);
@@ -264,7 +264,7 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 
 		if(color && (iip || j == n - 1) || tme)
 		{
-			movaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9)]);
+			movaps(xmm0, ptr[edx + j * sizeof(GSVertexHW9) + offsetof(GSVertexHW9, t)]);
 		}
 
 		if(color && (iip || j == n - 1))
@@ -329,15 +329,15 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 			punpcklwd(xmm3, xmm0);
 		}
 
-		movaps(ptr[eax], xmm2);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		movaps(ptr[edx], xmm3);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin;
 	// m_max.p = pmax;
 
-	movaps(ptr[eax + 16], xmm4);
+	movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[edx + 16], xmm5);
+	movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
@@ -347,8 +347,8 @@ GSVertexTrace::CGHW9::CGHW9(const void* param, uint32 key, void* code, size_t ma
 		shufps(xmm6, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 		shufps(xmm7, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-		movaps(ptr[eax + 32], xmm6);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[edx + 32], xmm7);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();
@@ -503,8 +503,8 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 			punpcklwd(xmm3, xmm0);
 		}
 
-		movaps(ptr[eax], xmm2);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, c)], xmm2);
-		movaps(ptr[edx], xmm3);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, c)], xmm3);
 	}
 
 	// m_min.p = pmin.xyww();
@@ -513,16 +513,16 @@ GSVertexTrace::CGHW11::CGHW11(const void* param, uint32 key, void* code, size_t
 	shufps(xmm4, xmm4, _MM_SHUFFLE(3, 3, 1, 0));
 	shufps(xmm5, xmm5, _MM_SHUFFLE(3, 3, 1, 0));
 
-	movaps(ptr[eax + 16], xmm4);
+	movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, p)], xmm4);
-	movaps(ptr[edx + 16], xmm5);
+	movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, p)], xmm5);
 
 	if(tme)
 	{
 		// m_min.t = tmin;
 		// m_max.t = tmax;
 
-		movaps(ptr[eax + 32], xmm6);
+		movaps(ptr[eax + offsetof(GSVertexTrace::Vertex, t)], xmm6);
-		movaps(ptr[edx + 32], xmm7);
+		movaps(ptr[edx + offsetof(GSVertexTrace::Vertex, t)], xmm7);
 	}
 
 	ret();