First step - duplicate the convolution shader code in two child class of a new base class.

One will become the new 2 pass implementation and the base class allows easy runtime switch between the 1 pass and 2 pass implementations.
The bicubic and Lanczos2 activate 1 pass, Lanczos3 optimized and Lanczos3 activate 2 pass, making it very easy to compare performance and to check for visual bugs...

1 files changed, 176 insertions, 0 deletions

diff --git a/system/shaders/convolutionsep-6x6_d3d.fx b/system/shaders/convolutionsep-6x6_d3d.fx
new file mode 100644
index 0000000000..dbb67926b4
--- /dev/null
+++ b/system/shaders/convolutionsep-6x6_d3d.fx
@@ -0,0 +1,176 @@
+/*
+ *      Copyright (C) 2005-2010 Team XBMC
+ *      http://www.xbmc.org
+ *
+ *  This Program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2, or (at your option)
+ *  any later version.
+ *
+ *  This Program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with XBMC; see the file COPYING.  If not, write to
+ *  the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139, USA.
+ *  http://www.gnu.org/copyleft/gpl.html
+ *
+ */
+
+texture g_Texture;
+texture g_KernelTexture;
+float2  g_StepXY;
+
+sampler RGBSampler =
+  sampler_state {
+    Texture = <g_Texture>;
+    AddressU = CLAMP;
+    AddressV = CLAMP;
+    MipFilter = LINEAR;
+    MinFilter = POINT;
+    MagFilter = POINT;
+  };
+
+sampler KernelSampler =
+  sampler_state
+  {
+    Texture = <g_KernelTexture>;
+    AddressU = CLAMP;
+    AddressV = CLAMP;
+    MipFilter = LINEAR;
+    MinFilter = LINEAR;
+    MagFilter = LINEAR;
+  };
+
+struct VS_OUTPUT
+{
+  float4 Position   : POSITION;
+  float2 TextureUV  : TEXCOORD0;
+};
+
+struct PS_OUTPUT
+{
+  float4 RGBColor : COLOR0;
+};
+
+half3 weight(float pos)
+{
+  half3 w;
+#ifdef HAS_RGBA
+  w = tex1D(KernelSampler, pos).rgb;
+#else
+  w = tex1D(KernelSampler, pos).bgr;
+#endif
+
+#ifdef HAS_FLOAT_TEXTURE
+  return w;
+#else
+  return w * 2.0 - 1.0;
+#endif
+}
+
+half3 pixel(float xpos, float ypos)
+{
+  return tex2D(RGBSampler, float2(xpos, ypos)).rgb;
+}
+
+half3 getLine(float ypos, float3 xpos1, float3 xpos2, half3 linetaps1, half3 linetaps2)
+{
+  return
+    pixel(xpos1.r, ypos) * linetaps1.r +
+    pixel(xpos1.g, ypos) * linetaps2.r +
+    pixel(xpos1.b, ypos) * linetaps1.g +
+    pixel(xpos2.r, ypos) * linetaps2.g +
+    pixel(xpos2.g, ypos) * linetaps1.b +
+    pixel(xpos2.b, ypos) * linetaps2.b;
+}
+
+PS_OUTPUT CONVOLUTION6x6Horiz(VS_OUTPUT In)
+{
+  PS_OUTPUT OUT;
+
+  float2 pos = In.TextureUV + g_StepXY * 0.5;
+  float2 f = frac(pos / g_StepXY);
+
+  half3 linetaps1   = weight((1.0 - f.x) / 2.0);
+  half3 linetaps2   = weight((1.0 - f.x) / 2.0 + 0.5);
+  half3 columntaps1 = weight((1.0 - f.y) / 2.0);
+  half3 columntaps2 = weight((1.0 - f.y) / 2.0 + 0.5);
+
+  // kernel generation code made sure taps add up to 1, no need to adjust here.
+
+  float2 xystart = (-2.0 - f) * g_StepXY + In.TextureUV;
+  float3 xpos1 = float3(
+      xystart.x,
+      xystart.x + g_StepXY.x,
+      xystart.x + g_StepXY.x * 2.0);
+  float3 xpos2 = half3(
+      xystart.x + g_StepXY.x * 3.0,
+      xystart.x + g_StepXY.x * 4.0,
+      xystart.x + g_StepXY.x * 5.0);
+
+  OUT.RGBColor.rgb = getLine(xystart.y                   , xpos1, xpos2, linetaps1, linetaps2) * columntaps1.r +
+					 getLine(xystart.y + g_StepXY.y      , xpos1, xpos2, linetaps1, linetaps2) * columntaps2.r +
+					 getLine(xystart.y + g_StepXY.y * 2.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps1.g +
+					 getLine(xystart.y + g_StepXY.y * 3.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps2.g +
+					 getLine(xystart.y + g_StepXY.y * 4.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps1.b +
+					 getLine(xystart.y + g_StepXY.y * 5.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps2.b;
+
+  OUT.RGBColor.a = 1.0;
+  return OUT;
+}
+
+PS_OUTPUT CONVOLUTION6x6Vert(VS_OUTPUT In)
+{
+  PS_OUTPUT OUT;
+
+  float2 pos = In.TextureUV + g_StepXY * 0.5;
+  float2 f = frac(pos / g_StepXY);
+
+  half3 linetaps1   = weight((1.0 - f.x) / 2.0);
+  half3 linetaps2   = weight((1.0 - f.x) / 2.0 + 0.5);
+  half3 columntaps1 = weight((1.0 - f.y) / 2.0);
+  half3 columntaps2 = weight((1.0 - f.y) / 2.0 + 0.5);
+
+  // kernel generation code made sure taps add up to 1, no need to adjust here.
+
+  float2 xystart = (-2.0 - f) * g_StepXY + In.TextureUV;
+  float3 xpos1 = float3(
+      xystart.x,
+      xystart.x + g_StepXY.x,
+      xystart.x + g_StepXY.x * 2.0);
+  float3 xpos2 = half3(
+      xystart.x + g_StepXY.x * 3.0,
+      xystart.x + g_StepXY.x * 4.0,
+      xystart.x + g_StepXY.x * 5.0);
+
+  OUT.RGBColor.rgb = getLine(xystart.y                   , xpos1, xpos2, linetaps1, linetaps2) * columntaps1.r +
+					 getLine(xystart.y + g_StepXY.y      , xpos1, xpos2, linetaps1, linetaps2) * columntaps2.r +
+					 getLine(xystart.y + g_StepXY.y * 2.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps1.g +
+					 getLine(xystart.y + g_StepXY.y * 3.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps2.g +
+					 getLine(xystart.y + g_StepXY.y * 4.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps1.b +
+					 getLine(xystart.y + g_StepXY.y * 5.0, xpos1, xpos2, linetaps1, linetaps2) * columntaps2.b;
+
+  OUT.RGBColor.a = 1.0;
+  return OUT;
+}
+
+technique SCALER_T
+{
+  pass P0
+  {
+    PixelShader  = compile ps_3_0 CONVOLUTION6x6Horiz();
+    ZEnable = False;
+    FillMode = Solid;
+    FogEnable = False;
+  }
+  pass P1
+  {
+    PixelShader  = compile ps_3_0 CONVOLUTION6x6Vert();
+    ZEnable = False;
+    FillMode = Solid;
+    FogEnable = False;
+  }
+};