RGB to YUV converison was previously baked into every scale shader, but this work has been moved to the YUV packing shaders. The scale shaders now write RGBA instead. In the case where base and output resolutions are identical, the render texture is forwarded directly to the YUV pack step, skipping an entire fullscreen pass. Intel GPA, SetStablePowerState, Intel HD Graphics 530, NV12 1920x1080, Before: RGBA -> UYVX: ~321 us UYVX -> Y: ~480 us UYVX -> UV: ~127 us 1920x1080, After: [forward render texture] RGBA -> Y: ~487 us RGBA -> UV: ~131 us 1920x1080 -> 1280x720, Before: RGBA -> UYVX: ~268 us UYVX -> Y: ~209 us UYVX -> UV: ~57 us 1920x1080 -> 1280x720, After: RGBA -> RGBA (rescale): ~268 us RGBA -> Y: ~210 us RGBA -> UV: ~58 us
174 lines
4.5 KiB
Plaintext
174 lines
4.5 KiB
Plaintext
/*
|
|
* lanczos sharper
|
|
* note - this shader is adapted from the GPL bsnes shader, very good stuff
|
|
* there.
|
|
*/
|
|
|
|
uniform float4x4 ViewProj;
|
|
uniform texture2d image;
|
|
uniform float2 base_dimension_i;
|
|
uniform float undistort_factor = 1.0;
|
|
|
|
sampler_state textureSampler
|
|
{
|
|
AddressU = Clamp;
|
|
AddressV = Clamp;
|
|
Filter = Linear;
|
|
};
|
|
|
|
struct VertData {
|
|
float4 pos : POSITION;
|
|
float2 uv : TEXCOORD0;
|
|
};
|
|
|
|
struct FragData {
|
|
float4 pos : POSITION;
|
|
float2 uv : TEXCOORD0;
|
|
float2 scale : TEXCOORD1;
|
|
};
|
|
|
|
FragData VSDefault(VertData v_in)
|
|
{
|
|
FragData vert_out;
|
|
vert_out.pos = mul(float4(v_in.pos.xyz, 1.0), ViewProj);
|
|
vert_out.uv = v_in.uv;
|
|
vert_out.scale = min(0.25 + abs(0.75 / mul(float4(1.0 / base_dimension_i.xy, 1.0, 1.0), ViewProj).xy), 1.0);
|
|
|
|
return vert_out;
|
|
}
|
|
|
|
float sinc(float x)
|
|
{
|
|
const float PIval = 3.1415926535897932384626433832795;
|
|
return sin(x * PIval) / (x * PIval);
|
|
}
|
|
|
|
float weight(float x, float radius)
|
|
{
|
|
float ax = abs(x);
|
|
if (x == 0.0)
|
|
return 1.0;
|
|
else if (ax < radius)
|
|
return sinc(x) * sinc(x / radius);
|
|
else
|
|
return 0.0;
|
|
}
|
|
|
|
float3 weight3(float x, float scale)
|
|
{
|
|
return float3(
|
|
weight((x * 2.0 + 0.0 * 2.0 - 3.0) * scale, 3.0),
|
|
weight((x * 2.0 + 1.0 * 2.0 - 3.0) * scale, 3.0),
|
|
weight((x * 2.0 + 2.0 * 2.0 - 3.0) * scale, 3.0));
|
|
}
|
|
|
|
float AspectUndistortX(float x, float a)
|
|
{
|
|
// The higher the power, the longer the linear part will be.
|
|
return (1.0 - a) * (x * x * x * x * x) + a * x;
|
|
}
|
|
|
|
float AspectUndistortU(float u)
|
|
{
|
|
// Normalize texture coord to -1.0 to 1.0 range, and back.
|
|
return AspectUndistortX((u - 0.5) * 2.0, undistort_factor) * 0.5 + 0.5;
|
|
}
|
|
|
|
float2 pixel_coord(float xpos, float ypos)
|
|
{
|
|
return float2(AspectUndistortU(xpos), ypos);
|
|
}
|
|
|
|
float4 pixel(float xpos, float ypos, bool undistort)
|
|
{
|
|
if (undistort)
|
|
return image.Sample(textureSampler, pixel_coord(xpos, ypos));
|
|
else
|
|
return image.Sample(textureSampler, float2(xpos, ypos));
|
|
}
|
|
|
|
float4 get_line(float ypos, float3 xpos1, float3 xpos2, float3 rowtap1,
|
|
float3 rowtap2, bool undistort)
|
|
{
|
|
return
|
|
pixel(xpos1.r, ypos, undistort) * rowtap1.r +
|
|
pixel(xpos1.g, ypos, undistort) * rowtap2.r +
|
|
pixel(xpos1.b, ypos, undistort) * rowtap1.g +
|
|
pixel(xpos2.r, ypos, undistort) * rowtap2.g +
|
|
pixel(xpos2.g, ypos, undistort) * rowtap1.b +
|
|
pixel(xpos2.b, ypos, undistort) * rowtap2.b;
|
|
}
|
|
|
|
float4 DrawLanczos(FragData v_in, bool undistort)
|
|
{
|
|
float2 stepxy = base_dimension_i;
|
|
float2 pos = v_in.uv + stepxy * 0.5;
|
|
float2 f = frac(pos / stepxy);
|
|
|
|
float3 rowtap1 = weight3((1.0 - f.x) / 2.0, v_in.scale.x);
|
|
float3 rowtap2 = weight3((1.0 - f.x) / 2.0 + 0.5, v_in.scale.x);
|
|
float3 coltap1 = weight3((1.0 - f.y) / 2.0, v_in.scale.y);
|
|
float3 coltap2 = weight3((1.0 - f.y) / 2.0 + 0.5, v_in.scale.y);
|
|
|
|
/* make sure all taps added together is exactly 1.0, otherwise some
|
|
* (very small) distortion can occur */
|
|
float suml = rowtap1.r + rowtap1.g + rowtap1.b + rowtap2.r + rowtap2.g + rowtap2.b;
|
|
float sumc = coltap1.r + coltap1.g + coltap1.b + coltap2.r + coltap2.g + coltap2.b;
|
|
rowtap1 /= suml;
|
|
rowtap2 /= suml;
|
|
coltap1 /= sumc;
|
|
coltap2 /= sumc;
|
|
|
|
float2 xystart = (-2.5 - f) * stepxy + pos;
|
|
float3 xpos1 = float3(xystart.x , xystart.x + stepxy.x , xystart.x + stepxy.x * 2.0);
|
|
float3 xpos2 = float3(xystart.x + stepxy.x * 3.0, xystart.x + stepxy.x * 4.0, xystart.x + stepxy.x * 5.0);
|
|
|
|
return
|
|
get_line(xystart.y , xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap1.r +
|
|
get_line(xystart.y + stepxy.y , xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap2.r +
|
|
get_line(xystart.y + stepxy.y * 2.0, xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap1.g +
|
|
get_line(xystart.y + stepxy.y * 3.0, xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap2.g +
|
|
get_line(xystart.y + stepxy.y * 4.0, xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap1.b +
|
|
get_line(xystart.y + stepxy.y * 5.0, xpos1, xpos2, rowtap1, rowtap2, undistort) * coltap2.b;
|
|
}
|
|
|
|
float4 PSDrawLanczosRGBA(FragData v_in, bool undistort) : TARGET
|
|
{
|
|
return DrawLanczos(v_in, undistort);
|
|
}
|
|
|
|
float4 PSDrawLanczosRGBADivide(FragData v_in) : TARGET
|
|
{
|
|
float4 rgba = DrawLanczos(v_in, false);
|
|
float alpha = rgba.a;
|
|
float multiplier = (alpha > 0.0) ? (1.0 / alpha) : 0.0;
|
|
return float4(rgba.rgb * multiplier, alpha);
|
|
}
|
|
|
|
technique Draw
|
|
{
|
|
pass
|
|
{
|
|
vertex_shader = VSDefault(v_in);
|
|
pixel_shader = PSDrawLanczosRGBA(v_in, false);
|
|
}
|
|
}
|
|
|
|
technique DrawAlphaDivide
|
|
{
|
|
pass
|
|
{
|
|
vertex_shader = VSDefault(v_in);
|
|
pixel_shader = PSDrawLanczosRGBADivide(v_in);
|
|
}
|
|
}
|
|
|
|
technique DrawUndistort
|
|
{
|
|
pass
|
|
{
|
|
vertex_shader = VSDefault(v_in);
|
|
pixel_shader = PSDrawLanczosRGBA(v_in, true);
|
|
}
|
|
}
|