GSdx: just a small mipmapping optimization when lod is constant, it was already a TODO.

git-svn-id: http://pcsx2.googlecode.com/svn/trunk@4518 96395faa-99c1-11dd-bbfe-3dabce05a288

plugins/GSdx/GSDrawScanline.cpp

@@ -40,6 +40,16 @@ void GSDrawScanline::BeginDraw(const void* param)
 {
 	memcpy(&m_global, param, sizeof(m_global));
 
+	if(m_global.sel.mmin && m_global.sel.lcm)
+	{
+		GSVector4i v = m_global.t.minmax.srl16(m_global.lod.i.x);
+
+		v = v.upl16(v);
+
+		m_local.temp.uv_minmax[0] = v.upl32(v);
+		m_local.temp.uv_minmax[1] = v.uph32(v);
+	}
+
 	m_ds = m_ds_map[m_global.sel];
 
 	if(m_global.sel.aa1)

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.avx.cpp

@@ -293,8 +293,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		// edx = &m_local.d[skip]
 
-		shl(edx, 3);
+		lea(edx, ptr[edx * 8 + (size_t)m_local.d]);
-		lea(edx, ptr[edx + (size_t)m_local.d]);
 
 		// ebx = &v
 
@@ -1243,18 +1242,8 @@ return;
 		vmovdqa(ptr[&m_local.temp.uv[0]], xmm2);
 		vmovdqa(ptr[&m_local.temp.uv[1]], xmm3);
 
-		// TODO: precalc
+		vmovdqa(xmm5, ptr[&m_local.temp.uv_minmax[0]]);
-
+		vmovdqa(xmm6, ptr[&m_local.temp.uv_minmax[1]]);
-		vmovq(xmm4, ptr[&m_local.gd->t.minmax]);
-
-		vpsrlw(xmm4, xmm0);
-
-		vpunpcklwd(xmm4, xmm4);
-		vpunpckldq(xmm5, xmm4, xmm4);
-		vpunpckhdq(xmm6, xmm4, xmm4);
-
-		vmovdqa(ptr[&m_local.temp.uv_minmax[0]], xmm5);
-		vmovdqa(ptr[&m_local.temp.uv_minmax[1]], xmm6);
 	}
 
 	// xmm2 = m_local.temp.uv[0] = u (level m)
@@ -2762,7 +2751,7 @@ void GSDrawScanlineCodeGenerator::ReadTexel(int pixels, int mip_offset)
 
 	const GSVector4i* lod_i = m_sel.lcm ? &m_local.gd->lod.i : &m_local.temp.lod.i;
 
-	if(m_sel.mmin)
+	if(m_sel.mmin && !m_sel.lcm)
 	{
 		const int r[] = {5, 6, 2, 4, 0, 1, 3, 7};
 
@@ -2790,6 +2779,12 @@ void GSDrawScanlineCodeGenerator::ReadTexel(int pixels, int mip_offset)
 	}
 	else
 	{
+		if(m_sel.mmin && m_sel.lcm)
+		{
+			mov(ebx, ptr[&lod_i->u32[0]]);
+			mov(ebx, ptr[edx + ebx * sizeof(void*) + mip_offset]);
+		}
+
 		const int r[] = {5, 6, 2, 4, 0, 1, 3, 5};
 
 		for(int i = 0; i < pixels; i++)

plugins/GSdx/GSDrawScanlineCodeGenerator.x86.cpp

@@ -290,8 +290,7 @@ void GSDrawScanlineCodeGenerator::Init()
 	{
 		// edx = &m_local.d[skip]
 
-		shl(edx, 3);
+		lea(edx, ptr[edx * 8 + (size_t)m_local.d]);
-		lea(edx, ptr[edx + (size_t)m_local.d]);
 
 		// ebx = &v
 
@@ -1295,20 +1294,8 @@ void GSDrawScanlineCodeGenerator::SampleTextureLOD()
 		movdqa(ptr[&m_local.temp.uv[0]], xmm2);
 		movdqa(ptr[&m_local.temp.uv[1]], xmm3);
 
-		// TODO: precalc
+		movdqa(xmm5, ptr[&m_local.temp.uv_minmax[0]]);
-
+		movdqa(xmm6, ptr[&m_local.temp.uv_minmax[1]]);
-		movq(xmm4, ptr[&m_local.gd->t.minmax]);
-
-		psrlw(xmm4, xmm0);
-
-		punpcklwd(xmm4, xmm4);
-		movdqa(xmm5, xmm4);
-		movdqa(xmm6, xmm4);
-		punpckldq(xmm5, xmm4);
-		punpckhdq(xmm6, xmm4);
-
-		movdqa(ptr[&m_local.temp.uv_minmax[0]], xmm5);
-		movdqa(ptr[&m_local.temp.uv_minmax[1]], xmm6);
 	}
 
 	// xmm2 = m_local.temp.uv[0] = u (level m)
@@ -2924,7 +2911,7 @@ void GSDrawScanlineCodeGenerator::ReadTexel(int pixels, int mip_offset)
 
 	const GSVector4i* lod_i = m_sel.lcm ? &m_local.gd->lod.i : &m_local.temp.lod.i;
 
-	if(m_sel.mmin)
+	if(m_sel.mmin && !m_sel.lcm)
 	{
 		#if _M_SSE >= 0x401
 
@@ -3075,6 +3062,12 @@ void GSDrawScanlineCodeGenerator::ReadTexel(int pixels, int mip_offset)
 	}
 	else
 	{
+		if(m_sel.mmin && m_sel.lcm)
+		{
+			mov(ebx, ptr[&lod_i->u32[0]]);
+			mov(ebx, ptr[edx + ebx * sizeof(void*) + mip_offset]);
+		}
+
 		const int r[] = {5, 6, 2, 4, 0, 1, 3, 5};
 
 		#if _M_SSE >= 0x401

plugins/GSdx/GSRasterizer.cpp

@@ -785,7 +785,7 @@ GSRasterizerMT::GSRasterizerMT(IDrawScanline* ds, volatile long& sync)
 
 GSRasterizerMT::~GSRasterizerMT()
 {
-    Draw(NULL);
+	Draw(NULL);
 
 	CloseThread();
 }

plugins/GSdx/GSScanlineEnvironment.h

@@ -137,8 +137,6 @@ __aligned(struct, 32) GSScanlineGlobalData // per batch variables, this is like
 
 __aligned(struct, 32) GSScanlineLocalData // per prim variables, each thread has its own
 {
-	const GSScanlineGlobalData* gd;
-
 	struct skip {GSVector4 z, s, t, q; GSVector4i rb, ga, f, _pad;} d[4];
 	struct step {GSVector4 z, stq; GSVector4i c, f;} d4;
 	struct {GSVector4i rb, ga;} c;
@@ -156,10 +154,15 @@ __aligned(struct, 32) GSScanlineLocalData // per prim variables, each thread has
 		GSVector4i cov;
 
 		// mipmapping
+
 		struct {GSVector4i i, f;} lod;
 		GSVector4i uv[2];
 		GSVector4i uv_minmax[2];
 		GSVector4i trb, tga;
 		GSVector4i test;
 	} temp; 
+
+	//
+
+	const GSScanlineGlobalData* gd;
 };