mirror/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2024-11-16 23:17:29 +00:00
Code Issues Packages Projects Releases Wiki Activity

rsx: Optimize software sampling further for the 6-tap kernel

Browse Source
This commit is contained in:
kd-11 2022-04-01 21:53:25 +03:00 committed by kd-11
parent ca35a75a7d
commit fc05511354
3 changed files with 63 additions and 57 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

114
rpcs3/Emu/RSX/Program/GLSLCommon.cpp
View File

@ -593,10 +593,12 @@ namespace glsl
"#define RENORMALIZE " << rsx::texture_control_bits::RENORMALIZE << "\n"
"#define DEPTH_FLOAT " << rsx::texture_control_bits::DEPTH_FLOAT << "\n"
"#define DEPTH_COMPARE " << rsx::texture_control_bits::DEPTH_COMPARE_OP << "\n"
"#define FILTERED_BIT " << rsx::texture_control_bits::FILTERED << "\n"
"#define FILTERED_MAG_BIT " << rsx::texture_control_bits::FILTERED_MAG << "\n"
"#define FILTERED_MIN_BIT " << rsx::texture_control_bits::FILTERED_MIN << "\n"
"#define INT_COORDS_BIT " << rsx::texture_control_bits::UNNORMALIZED_COORDS << "\n"
"#define GAMMA_CTRL_MASK (GAMMA_R_MASK|GAMMA_G_MASK|GAMMA_B_MASK|GAMMA_A_MASK)\n"
"#define SIGN_EXPAND_MASK (EXPAND_R_MASK|EXPAND_G_MASK|EXPAND_B_MASK|EXPAND_A_MASK)\n\n";
"#define SIGN_EXPAND_MASK (EXPAND_R_MASK|EXPAND_G_MASK|EXPAND_B_MASK|EXPAND_A_MASK)\n"
"#define FILTERED_MASK (FILTERED_MAG_BIT|FILTERED_MIN_BIT)\n\n";
}
}
@ -902,43 +904,37 @@ namespace glsl
"#define TEX2D_Z24X8_RGBA8_MS(index, coord2) process_texel(convert_z24x8_to_rgba8(ZS_READ_MS(index, coord2), texture_parameters[index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))\n\n";
OS <<
R"(
vec3 compute2x2DownsampleWeights(const in float coord, const in float uv_step, const in float actual_step)
{
const float last_sample_point = max(coord - actual_step, 0.);
const float next_sample_point = min(coord + actual_step, 1.);
const float last_coord_step = floor(coord / uv_step) * uv_step;
const float next_coord_step = last_coord_step + uv_step;
const float next_next_coord_step = next_coord_step + uv_step;
const vec3 weights = vec3(next_coord_step - coord,
min(next_next_coord_step, next_sample_point) - next_coord_step,
max(next_next_coord_step, next_sample_point) - next_next_coord_step);
return weights / dot(weights, vec3(1));
}
)";
"vec3 compute2x2DownsampleWeights(const in float coord, const in float uv_step, const in float actual_step)"
"{\n"
" const float next_sample_point = coord + actual_step;\n"
" const float next_coord_step = fma(floor(coord / uv_step), uv_step, uv_step);\n"
" const float next_coord_step_plus_one = next_coord_step + uv_step;\n"
" vec3 weights = vec3(next_coord_step, min(next_coord_step_plus_one, next_sample_point), max(next_coord_step_plus_one, next_sample_point)) - vec3(coord, next_coord_step, next_coord_step_plus_one);\n"
" return weights / actual_step;\n"
"}\n\n";
auto insert_msaa_sample_code = [&OS](const std::string_view& sampler_type)
{
OS <<
"vec4 texelFetch2DMS(in " << sampler_type << " tex, const in ivec2 sample_count, const in ivec2 icoords, const in int index, const in ivec2 offset)\n"
"vec4 texelFetch2DMS(in " << sampler_type << " tex, const in vec2 sample_count, const in ivec2 icoords, const in int index, const in ivec2 offset)\n"
"{\n"
" const ivec2 resolve_coords = (icoords + offset) * ivec2(bvec2(texture_parameters[index].scale_bias.xy));\n"
" const ivec2 aa_coords = resolve_coords / sample_count;\n"
" const ivec2 sample_loc = ivec2(resolve_coords % sample_count);\n"
" const int sample_index = sample_loc.x + (sample_loc.y * sample_count.y);\n"
" return texelFetch(tex, aa_coords, sample_index);\n"
" const vec2 resolve_coords = vec2(icoords + offset);\n"
" const vec2 aa_coords = floor(resolve_coords / sample_count);\n" // AA coords = real_coords / sample_count
" const vec2 sample_loc = fma(aa_coords, -sample_count, resolve_coords);\n" // Sample ID = real_coords % sample_count
" const float sample_index = fma(sample_loc.y, sample_count.y, sample_loc.x);\n"
" return texelFetch(tex, ivec2(aa_coords), int(sample_index));\n"
"}\n\n"
"vec4 sampleTexture2DMS(in " << sampler_type << " tex, const in vec2 coords, const in int index)\n"
"{\n"
" const uint flags = TEX_FLAGS(index);\n"
" const vec2 normalized_coords = COORD_SCALE2(index, coords);\n"
" const ivec2 sample_count = ivec2(2, textureSamples(tex) / 2);\n"
" const ivec2 image_size = textureSize(tex) * sample_count;\n"
" const vec2 sample_count = vec2(2., textureSamples(tex) * 0.5);\n"
" const vec2 image_size = textureSize(tex) * sample_count;\n"
" const ivec2 icoords = ivec2(normalized_coords * image_size);\n"
" const vec4 sample0 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(0));\n"
"\n"
" if (!_test_bit(flags, FILTERED_BIT))\n"
" if (_get_bits(flags, FILTERED_MAG_BIT, 2) == 0)\n"
" {\n"
" return sample0;\n"
" }\n"
@ -947,44 +943,57 @@ namespace glsl
" const vec2 uv_step = 1.0 / vec2(image_size);\n"
" const vec2 actual_step = vec2(dFdx(normalized_coords.x), dFdy(normalized_coords.y));\n"
"\n"
" if (uv_step.x == actual_step.x && uv_step.y == actual_step.y)\n"
" const bvec2 no_filter = lessThan(abs(uv_step - actual_step), vec2(0.000001));\n"
" if (no_filter.x && no_filter.y)\n"
" {\n"
" return sample0;\n"
" }\n"
"\n"
" // Fetch remaining samples\n"
" const vec4 sample1 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(1, 0));\n" // Bottom right
" const vec4 sample2 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(0, 1));\n" // Top left
" const vec4 sample3 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(1, 1));\n" // Top right
"\n"
" vec4 a, b;\n"
" float factor;\n"
" const vec4 sample2 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(0, 1)); // Top left\n"
"\n"
" if (actual_step.x > uv_step.x)\n"
" {\n"
" // Downscale in X, centered\n"
" const vec4 sample4 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(2, 0));\n" // Further bottom right
" const vec4 sample5 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(2, 1));\n" // Further top right
" const vec3 weights = compute2x2DownsampleWeights(normalized_coords.x, uv_step.x, actual_step.x);\n"
"\n"
" a = (sample0 * weights.x + sample1 * weights.y + sample4 * weights.z);\n" // Weighted sum
" b = (sample2 * weights.x + sample3 * weights.y + sample5 * weights.z);\n" // Weighted sum
" }\n"
" else if (actual_step.x < uv_step.x)\n"
" {\n"
" // Upscale in X\n"
" factor = fract(normalized_coords.x * image_size.x);\n"
" a = mix(sample0, sample1, factor);\n"
" b = mix(sample2, sample3, factor);\n"
" }\n"
" else\n"
" if (no_filter.x)\n"
" {\n"
" // No scaling, 1:1\n"
" a = sample0;\n"
" b = sample2;\n"
" }\n"
" else\n"
" {\n"
" // Filter required, sample more data\n"
" const vec4 sample1 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(1, 0)); // Bottom right\n"
" const vec4 sample3 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(1, 1)); // Top right\n"
"\n"
" if (actual_step.y > uv_step.y)\n"
" if (actual_step.x > uv_step.x)\n"
" {\n"
" // Downscale in X, centered\n"
" const vec3 weights = compute2x2DownsampleWeights(normalized_coords.x, uv_step.x, actual_step.x);\n"
"\n"
" const vec4 sample4 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(2, 0)); // Further bottom right\n"
" a = fma(sample0, weights.xxxx, sample1 * weights.y) + (sample4 * weights.z); // Weighted sum\n"
"\n"
" if (!no_filter.y)\n"
" {\n"
" const vec4 sample5 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(2, 1)); // Further top right\n"
" b = fma(sample2, weights.xxxx, sample3 * weights.y) + (sample5 * weights.z); // Weighted sum\n"
" }\n"
" }\n"
" else if (actual_step.x < uv_step.x)\n"
" {\n"
" // Upscale in X\n"
" factor = fract(normalized_coords.x * image_size.x);\n"
" a = mix(sample0, sample1, factor);\n"
" b = mix(sample2, sample3, factor);\n"
" }\n"
" }\n"
"\n"
" if (no_filter.y)\n"
" {\n"
" // 1:1 no scale\n"
" return a;\n"
" }\n"
" else if (actual_step.y > uv_step.y)\n"
" {\n"
" // Downscale in Y\n"
" const vec3 weights = compute2x2DownsampleWeights(normalized_coords.y, uv_step.y, actual_step.y);\n"
@ -997,11 +1006,6 @@ namespace glsl
" factor = fract(normalized_coords.y * image_size.y);\n"
" return mix(a, b, factor);\n"
" }\n"
" else\n"
" {\n"
" // 1:1 no scale\n"
" return a;\n"
" }\n"
"}\n\n";
};

3
rpcs3/Emu/RSX/Program/GLSLCommon.h
View File

@ -23,7 +23,8 @@ namespace rsx
DEPTH_COMPARE_OP,
DEPTH_COMPARE_1,
DEPTH_COMPARE_2,
FILTERED,
FILTERED_MAG,
FILTERED_MIN,
UNNORMALIZED_COORDS,
GAMMA_CTRL_MASK = (1 << GAMMA_R) | (1 << GAMMA_G) | (1 << GAMMA_B) | (1 << GAMMA_A),

3
rpcs3/Emu/RSX/RSXThread.cpp
View File

@ -2002,7 +2002,8 @@ namespace rsx
{
current_fp_texture_state.multisampled_textures |= (1 << i);
texture_control |= (static_cast<u32>(tex.zfunc()) << texture_control_bits::DEPTH_COMPARE_OP);
texture_control |= (static_cast<u32>(tex.mag_filter() != rsx::texture_magnify_filter::nearest) << texture_control_bits::FILTERED);
texture_control |= (static_cast<u32>(tex.mag_filter() != rsx::texture_magnify_filter::nearest) << texture_control_bits::FILTERED_MAG);
texture_control |= (static_cast<u32>(tex.min_filter() != rsx::texture_minify_filter::nearest) << texture_control_bits::FILTERED_MIN);
texture_control |= (((tex.format() & CELL_GCM_TEXTURE_UN) >> 6) << texture_control_bits::UNNORMALIZED_COORDS);
}