Skip to content

GS/DX/GL/VK: Add SW AF #14188

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
TheLastRar wants to merge 1 commit into
base: master
Choose a base branch
from
Open

GS/DX/GL/VK: Add SW AF #14188

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
159 changes: 146 additions & 13 deletions bin/resources/shaders/dx11/tfx.fx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -220,6 +220,148 @@ cbuffer cb1
float RcpScaleFactor;
};

#if (PS_AUTOMATIC_LOD != 1) && (PS_MANUAL_LOD == 1)
float manual_lod(float uv_w)
{
// FIXME add LOD: K - ( LOG2(Q) * (1 << L))
float K = LODParams.x;
float L = LODParams.y;
float bias = LODParams.z;
float max_lod = LODParams.w;

float gs_lod = K - log2(abs(uv_w)) * L;
// FIXME max useful ?
//return max(min(gs_lod, max_lod) - bias, 0.0f);
return min(gs_lod, max_lod) - bias;
}
#endif

#if PS_ANISOTROPIC_FILTERING > 1
float4 sample_c_af(float2 uv, float uv_w)
{
// Below taken from https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec.htm#7.18.11%20LOD%20Calculations
// With guidance from https://pema.dev/2025/05/09/mipmaps-too-much-detail/
float2 sz;
Texture.GetDimensions(sz.x, sz.y);
float2 dX = ddx(uv) * sz;
float2 dY = ddy(uv) * sz;

// Calculate Ellipse Transform
bool d_zero = length(dX) == 0 || length(dY) == 0;
bool d_par = (dX.x * dY.y - dY.x * dX.y) == 0;
bool d_per = dot(dX, dY) == 0;
bool d_inf_nan = any(isinf(dX) | isinf(dY) | isnan(dX) | isnan(dY));

if (!(d_zero || d_par || d_per || d_inf_nan))
{
float A = dX.y * dX.y + dY.y * dY.y;
float B = -2 * (dX.x * dX.y + dY.x * dY.y);
float C = dX.x * dX.x + dY.x * dY.x;
float f = (dX.x * dY.y - dY.x * dX.y);
float F = f * f;

float p = A - C;
float q = A + C;
float t = sqrt(p * p + B * B);

float2 new_dX = float2(
sqrt(F * (t + p) / (t * (q + t))),
sqrt(F * (t - p) / (t * (q + t))) * sign(B)
);

float2 new_dY = float2(
sqrt(F * (t - p) / (t * (q - t))) * -sign(B),
sqrt(F * (t + p) / (t * (q - t)))
);

d_inf_nan = any(isinf(new_dX) | isinf(new_dY) | isnan(new_dX) | isnan(new_dY));
if (!d_inf_nan)
{
dX = new_dX;
dY = new_dY;
}
}

// Compute AF values
float squared_length_x = dX.x * dX.x + dX.y * dX.y;
float squared_length_y = dY.x * dY.x + dY.y * dY.y;
float determinant = abs(dX.x * dY.y - dX.y * dY.x);
bool is_major_x = squared_length_x > squared_length_y;
float squared_length_major = is_major_x ? squared_length_x : squared_length_y;
float length_major = sqrt(squared_length_major);

float aniso_ratio;
float length_lod;
float2 aniso_line;
if (length_major <= 1.0f)
{
// A zero length_major would result in NaN Lod and break sampling.
// A small length_major would result in aniso_ratio getting clamped to 1.
// Perform isotropic filtering instead.
aniso_ratio = 1.0f;
length_lod = length_major;
aniso_line = float2(0, 0);
}
else
{
float norm_major = 1.0f / length_major;

float2 aniso_line_dir = float2(
(is_major_x ? dX.x : dY.x) * norm_major,
(is_major_x ? dX.y : dY.y) * norm_major
);

aniso_ratio = squared_length_major / determinant;

// Calculate the minor length of the ellipse for Lod, while also clamping the ratio of anisotropy.
if (aniso_ratio > PS_ANISOTROPIC_FILTERING)
{
// ratio is clamped - Lod is based on ratio (preserves area)
aniso_ratio = PS_ANISOTROPIC_FILTERING;
length_lod = length_major / PS_ANISOTROPIC_FILTERING;
}
else
{
// ratio not clamped - Lod is based on area
length_lod = determinant / length_major;
}

// clamp to top Lod
if (length_lod < 1.0f)
aniso_ratio = max(1.0f, aniso_ratio * length_lod);

aniso_ratio = round(aniso_ratio);
aniso_line = aniso_line_dir * 0.5f * length_major * (1.0f / sz);
}

#if PS_AUTOMATIC_LOD == 1
float lod = log2(length_lod);
#elif PS_MANUAL_LOD == 1
float lod = manual_lod(uv_w);
#else
float lod = 0; // No Lod
#endif

float4 colour;
if (aniso_ratio == 1.0f)
colour = Texture.SampleLevel(TextureSampler, uv, lod);
else
{
float4 num = float4(0, 0, 0, 0);
for (int i = 0; i < aniso_ratio; i++)
{
float2 d = -aniso_line + (0.5f + i) * (2.0f * aniso_line) / aniso_ratio;
float2 uv_sample = uv + d;
float4 sample_colour = Texture.SampleLevel(TextureSampler, uv_sample, lod);
num += sample_colour;
}

colour = num / aniso_ratio;
}
return colour;
}
#endif

float4 sample_c(float2 uv, float uv_w, int2 xy)
{
#if PS_TEX_IS_FB == 1
Expand Down Expand Up @@ -252,21 +394,12 @@ float4 sample_c(float2 uv, float uv_w, int2 xy)
#endif
#endif

#if PS_AUTOMATIC_LOD == 1
#if PS_ANISOTROPIC_FILTERING > 1
return sample_c_af(uv, uv_w);
#elif PS_AUTOMATIC_LOD == 1
return Texture.Sample(TextureSampler, uv);
#elif PS_MANUAL_LOD == 1
// FIXME add LOD: K - ( LOG2(Q) * (1 << L))
float K = LODParams.x;
float L = LODParams.y;
float bias = LODParams.z;
float max_lod = LODParams.w;

float gs_lod = K - log2(abs(uv_w)) * L;
// FIXME max useful ?
//float lod = max(min(gs_lod, max_lod) - bias, 0.0f);
float lod = min(gs_lod, max_lod) - bias;

return Texture.SampleLevel(TextureSampler, uv, lod);
return Texture.SampleLevel(TextureSampler, uv, manual_lod(uv_w));
#else
return Texture.SampleLevel(TextureSampler, uv, 0); // No lod
#endif
Expand Down
158 changes: 145 additions & 13 deletions bin/resources/shaders/opengl/tfx_fs.glsl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -182,6 +182,147 @@ vec4 sample_from_depth()

#if NEEDS_TEX

#if (PS_AUTOMATIC_LOD != 1) && (PS_MANUAL_LOD == 1)
float manual_lod(float uv_w)
{
// FIXME add LOD: K - ( LOG2(Q) * (1 << L))
float K = LODParams.x;
float L = LODParams.y;
float bias = LODParams.z;
float max_lod = LODParams.w;

float gs_lod = K - log2(abs(uv_w)) * L;
// FIXME max useful ?
//return max(min(gs_lod, max_lod) - bias, 0.0f);
return min(gs_lod, max_lod) - bias;
}
#endif

#if PS_ANISOTROPIC_FILTERING > 1
vec4 sample_c_af(vec2 uv, float uv_w)
{
// Below taken from https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec.htm#7.18.11%20LOD%20Calculations
// With guidance from https://pema.dev/2025/05/09/mipmaps-too-much-detail/
vec2 sz = textureSize(TextureSampler, 0);
vec2 dX = dFdx(uv) * sz;
vec2 dY = dFdy(uv) * sz;

// Calculate Ellipse Transform
bool d_zero = length(dX) == 0 || length(dY) == 0;
bool d_par = (dX.x * dY.y - dY.x * dX.y) == 0;
bool d_per = dot(dX, dY) == 0;
bool d_inf_nan = any(isinf(dX)) || any(isinf(dY)) || any(isnan(dX)) || any(isnan(dY));

if (!(d_zero || d_par || d_per || d_inf_nan))
{
float A = dX.y * dX.y + dY.y * dY.y;
float B = -2 * (dX.x * dX.y + dY.x * dY.y);
float C = dX.x * dX.x + dY.x * dY.x;
float f = (dX.x * dY.y - dY.x * dX.y);
float F = f * f;

float p = A - C;
float q = A + C;
float t = sqrt(p * p + B * B);

vec2 new_dX = vec2(
sqrt(F * (t + p) / (t * (q + t))),
sqrt(F * (t - p) / (t * (q + t))) * sign(B)
);

vec2 new_dY = vec2(
sqrt(F * (t - p) / (t * (q - t))) * -sign(B),
sqrt(F * (t + p) / (t * (q - t)))
);

d_inf_nan = any(isinf(new_dX)) || any(isinf(new_dY)) || any(isnan(new_dX)) || any(isnan(new_dY));
if (!d_inf_nan)
{
dX = new_dX;
dY = new_dY;
}
}

// Compute AF values
float squared_length_x = dX.x * dX.x + dX.y * dX.y;
float squared_length_y = dY.x * dY.x + dY.y * dY.y;
float determinant = abs(dX.x * dY.y - dX.y * dY.x);
bool is_major_x = squared_length_x > squared_length_y;
float squared_length_major = is_major_x ? squared_length_x : squared_length_y;
float length_major = sqrt(squared_length_major);

float aniso_ratio;
float length_lod;
vec2 aniso_line;
if (length_major <= 1.0f)
{
// A zero length_major would result in NaN Lod and break sampling.
// A small length_major would result in aniso_ratio getting clamped to 1.
// Perform isotropic filtering instead.
aniso_ratio = 1.0f;
length_lod = length_major;
aniso_line = vec2(0, 0);
}
else
{
float norm_major = 1.0f / length_major;

vec2 aniso_line_dir = vec2(
(is_major_x ? dX.x : dY.x) * norm_major,
(is_major_x ? dX.y : dY.y) * norm_major
);

aniso_ratio = squared_length_major / determinant;

// Calculate the minor length of the ellipse for Lod, while also clamping the ratio of anisotropy.
if (aniso_ratio > PS_ANISOTROPIC_FILTERING)
{
// ratio is clamped - Lod is based on ratio (preserves area)
aniso_ratio = PS_ANISOTROPIC_FILTERING;
length_lod = length_major / PS_ANISOTROPIC_FILTERING;
}
else
{
// ratio not clamped - Lod is based on area
length_lod = determinant / length_major;
}

// clamp to top Lod
if (length_lod < 1.0f)
aniso_ratio = max(1.0f, aniso_ratio * length_lod);

aniso_ratio = round(aniso_ratio);
aniso_line = aniso_line_dir * 0.5f * length_major * (1.0f / sz);
}

#if PS_AUTOMATIC_LOD == 1
float lod = log2(length_lod);
#elif PS_MANUAL_LOD == 1
float lod = manual_lod(uv_w);
#else
float lod = 0; // No Lod
#endif

vec4 colour;
if (aniso_ratio == 1.0f)
colour = textureLod(TextureSampler, uv, lod);
else
{
vec4 num = vec4(0, 0, 0, 0);
for (int i = 0; i < aniso_ratio; i++)
{
vec2 d = -aniso_line + (0.5f + i) * (2.0f * aniso_line) / aniso_ratio;
vec2 uv_sample = uv + d;
vec4 sample_colour = textureLod(TextureSampler, uv_sample, lod);
num += sample_colour;
}

colour = num / aniso_ratio;
}
return colour;
}
#endif

vec4 sample_c(vec2 uv)
{
#if PS_TEX_IS_FB == 1
Expand All @@ -205,21 +346,12 @@ vec4 sample_c(vec2 uv)
#endif
#endif

#if PS_AUTOMATIC_LOD == 1
#if PS_ANISOTROPIC_FILTERING > 1
return sample_c_af(uv, PSin.t_float.w);
#elif PS_AUTOMATIC_LOD == 1
return texture(TextureSampler, uv);
#elif PS_MANUAL_LOD == 1
// FIXME add LOD: K - ( LOG2(Q) * (1 << L))
float K = LODParams.x;
float L = LODParams.y;
float bias = LODParams.z;
float max_lod = LODParams.w;

float gs_lod = K - log2(abs(PSin.t_float.w)) * L;
// FIXME max useful ?
//float lod = max(min(gs_lod, max_lod) - bias, 0.0f);
float lod = min(gs_lod, max_lod) - bias;

return textureLod(TextureSampler, uv, lod);
return textureLod(TextureSampler, uv, manual_lod(PSin.t_float.w));
#else
return textureLod(TextureSampler, uv, 0.0f); // No lod
#endif
Expand Down
Loading
Loading