gaming/rpcs3
gaming/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3 synced 2024-07-20 17:45:48 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

rsx: Switch common codegen to use the glsl scripts

Browse source
This commit is contained in:
kd-11 2023-06-20 00:20:51 +03:00 committed by kd-11
parent cffcfad42a
commit 89c81d9f22
8 changed files with 193 additions and 545 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

4
rpcs3/Emu/RSX/GL/GLFragmentProgram.cpp
View file

@ -191,6 +191,7 @@ void GLFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
m_shader_props.require_texture_expand = properties.has_exp_tex_op;
m_shader_props.require_srgb_to_linear = properties.has_upg;
m_shader_props.require_linear_to_srgb = properties.has_pkg;
m_shader_props.require_fog_read = properties.in_register_mask & in_fogc;
m_shader_props.emulate_coverage_tests = true; // g_cfg.video.antialiasing_level == msaa_level::none;
m_shader_props.emulate_shadow_compare = device_props.emulate_depth_compare;
m_shader_props.low_precision_tests = ::gl::get_driver_caps().vendor_NVIDIA && !(m_prog.ctrl & RSX_SHADER_CONTROL_ATTRIBUTE_INTERPOLATION);
@ -203,9 +204,6 @@ void GLFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
void GLFragmentDecompilerThread::insertMainStart(std::stringstream & OS)
{
if (properties.in_register_mask & in_fogc)
program_common::insert_fog_declaration(OS);
std::set<std::string> output_registers;
if (m_ctrl & CELL_GCM_SHADER_CONTROL_32_BITS_EXPORTS)
{

645
rpcs3/Emu/RSX/Program/GLSLCommon.cpp
View file

@ -20,104 +20,14 @@ namespace program_common
OS << "\n";
}
void insert_compare_op(std::ostream& OS)
void define_glsl_switches(std::ostream& OS, std::vector<std::string_view>& enums)
{
OS <<
"bool comparison_passes(const in float a, const in float b, const in uint func)\n"
"{\n"
" switch (func)\n"
" {\n"
" default:\n"
" case 0: return false; //never\n"
" case 1: return (CMP_FIXUP(a) < CMP_FIXUP(b)); //less\n"
" case 2: return (CMP_FIXUP(a) == CMP_FIXUP(b)); //equal\n"
" case 3: return (CMP_FIXUP(a) <= CMP_FIXUP(b)); //lequal\n"
" case 4: return (CMP_FIXUP(a) > CMP_FIXUP(b)); //greater\n"
" case 5: return (CMP_FIXUP(a) != CMP_FIXUP(b)); //nequal\n"
" case 6: return (CMP_FIXUP(a) >= CMP_FIXUP(b)); //gequal\n"
" case 7: return true; //always\n"
" }\n"
"}\n\n";
}
void insert_compare_op_vector(std::ostream& OS)
{
OS <<
"bvec4 comparison_passes(const in vec4 a, const in vec4 b, const in uint func)\n"
"{\n"
" switch (func)\n"
" {\n"
" default:\n"
" case 0: return bvec4(false); //never\n"
" case 1: return lessThan(CMP_FIXUP(a), CMP_FIXUP(b)); //less\n"
" case 2: return equal(CMP_FIXUP(a), CMP_FIXUP(b)); //equal\n"
" case 3: return lessThanEqual(CMP_FIXUP(a), CMP_FIXUP(b)); //lequal\n"
" case 4: return greaterThan(CMP_FIXUP(a), CMP_FIXUP(b)); //greater\n"
" case 5: return notEqual(CMP_FIXUP(a), CMP_FIXUP(b)); //nequal\n"
" case 6: return greaterThanEqual(CMP_FIXUP(a), CMP_FIXUP(b)); //gequal\n"
" case 7: return bvec4(true); //always\n"
" }\n"
"}\n\n";
}
void insert_fog_declaration(std::ostream& OS, std::string_view wide_vector_type, std::string_view input_coord)
{
define_glsl_constants<rsx::fog_mode>(OS,
for (const auto& e : enums)
{
{ "FOG_LINEAR", rsx::fog_mode::linear },
{ "FOG_EXP", rsx::fog_mode::exponential },
{ "FOG_EXP2", rsx::fog_mode::exponential2 },
{ "FOG_LINEAR_ABS", rsx::fog_mode::linear_abs },
{ "FOG_EXP_ABS", rsx::fog_mode::exponential_abs },
{ "FOG_EXP2_ABS", rsx::fog_mode::exponential2_abs }
});
OS << "#define " << e << "\n";
}
std::string template_body = "$T fetch_fog_value(const in uint mode)\n";
template_body +=
"{\n"
" $T result = $T($I.x, 0., 0., 0.);\n"
" switch(mode)\n"
" {\n"
" default:\n"
" return result;\n"
" case FOG_LINEAR:\n"
" //linear\n"
" result.y = fog_param1 * $I.x + (fog_param0 - 1.);\n"
" break;\n"
" case FOG_EXP:\n"
" //exponential\n"
" result.y = exp(11.084 * (fog_param1 * $I.x + fog_param0 - 1.5));\n"
" break;\n"
" case FOG_EXP2:\n"
" //exponential2\n"
" result.y = exp(-pow(4.709 * (fog_param1 * $I.x + fog_param0 - 1.5), 2.));\n"
" break;\n"
" case FOG_EXP_ABS:\n"
" //exponential_abs\n"
" result.y = exp(11.084 * (fog_param1 * abs($I.x) + fog_param0 - 1.5));\n"
" break;\n"
" case FOG_EXP2_ABS:\n"
" //exponential2_abs\n"
" result.y = exp(-pow(4.709 * (fog_param1 * abs($I.x) + fog_param0 - 1.5), 2.));\n"
" break;\n"
" case FOG_LINEAR_ABS:\n"
" //linear_abs\n"
" result.y = fog_param1 * abs($I.x) + (fog_param0 - 1.);\n"
" break;\n"
" }\n"
"\n"
" result.y = clamp(result.y, 0., 1.);\n"
" return result;\n"
"}\n\n";
std::pair<std::string_view, std::string> replacements[] =
{
std::make_pair("$T", std::string(wide_vector_type)),
std::make_pair("$I", std::string(input_coord))
};
OS << fmt::replace_all(template_body, replacements);
OS << "\n";
}
}
@ -522,6 +432,17 @@ namespace glsl
void insert_glsl_legacy_function(std::ostream& OS, const shader_properties& props)
{
std::vector<std::string_view> enabled_options;
if (props.low_precision_tests)
{
enabled_options.push_back("_GPU_LOW_PRECISION_COMPARE");
}
if (props.require_lit_emulation)
{
enabled_options.push_back("_ENABLE_LIT_EMULATION");
}
OS << "#define _select mix\n";
OS << "#define _saturate(x) clamp(x, 0., 1.)\n";
OS << "#define _get_bits(x, off, count) bitfieldExtract(x, off, count)\n";
@ -529,508 +450,182 @@ namespace glsl
OS << "#define _test_bit(x, y) (_get_bits(x, y, 1) != 0)\n";
OS << "#define _rand(seed) fract(sin(dot(seed.xy, vec2(12.9898f, 78.233f))) * 43758.5453f)\n\n";
if (props.low_precision_tests)
{
OS << "#define CMP_FIXUP(a) (sign(a) * 16. + a)\n\n";
}
else
{
OS << "#define CMP_FIXUP(a) (a)\n\n";
}
if (props.domain == glsl::program_domain::glsl_fragment_program)
{
OS << "// ROP control\n";
OS << "#define ALPHA_TEST_ENABLE_BIT " << rsx::ROP_control_bits::ALPHA_TEST_ENABLE_BIT << "\n";
OS << "#define SRGB_FRAMEBUFFER_BIT " << rsx::ROP_control_bits::SRGB_FRAMEBUFFER_BIT << "\n";
OS << "#define ALPHA_TO_COVERAGE_ENABLE_BIT " << rsx::ROP_control_bits::ALPHA_TO_COVERAGE_ENABLE_BIT << "\n";
OS << "#define MSAA_WRITE_ENABLE_BIT " << rsx::ROP_control_bits::MSAA_WRITE_ENABLE_BIT << "\n";
OS << "#define INT_FRAMEBUFFER_BIT " << rsx::ROP_control_bits::INT_FRAMEBUFFER_BIT << "\n";
OS << "#define POLYGON_STIPPLE_ENABLE_BIT " << rsx::ROP_control_bits::POLYGON_STIPPLE_ENABLE_BIT << "\n";
OS << "#define ALPHA_TEST_FUNC_OFFSET " << rsx::ROP_control_bits::ALPHA_FUNC_OFFSET << "\n";
OS << "#define ALPHA_TEST_FUNC_LENGTH " << rsx::ROP_control_bits::ALPHA_FUNC_NUM_BITS << "\n";
OS << "#define MSAA_SAMPLE_CTRL_OFFSET " << rsx::ROP_control_bits::MSAA_SAMPLE_CTRL_OFFSET << "\n";
OS << "#define MSAA_SAMPLE_CTRL_LENGTH " << rsx::ROP_control_bits::MSAA_SAMPLE_CTRL_NUM_BITS << "\n";
OS << "#define ROP_CMD_MASK " << rsx::ROP_control_bits::ROP_CMD_MASK << "\n\n";
// 8-bit rounding/quantization
program_common::define_glsl_constants<rsx::ROP_control_bits>(OS,
{
const auto _16bit_outputs = (!props.fp32_outputs && props.supports_native_fp16);
const auto _255 = _16bit_outputs ? "f16vec4(255.)" : "vec4(255.)";
const auto _1_over_2 = _16bit_outputs ? "f16vec4(0.5)" : "vec4(0.5)";
OS << "#define round_to_8bit(v4) (floor(fma(v4, " << _255 << ", " << _1_over_2 << ")) / " << _255 << ")\n\n";
{ "ALPHA_TEST_ENABLE_BIT ", rsx::ROP_control_bits::ALPHA_TEST_ENABLE_BIT },
{ "SRGB_FRAMEBUFFER_BIT ", rsx::ROP_control_bits::SRGB_FRAMEBUFFER_BIT },
{ "ALPHA_TO_COVERAGE_ENABLE_BIT ", rsx::ROP_control_bits::ALPHA_TO_COVERAGE_ENABLE_BIT },
{ "MSAA_WRITE_ENABLE_BIT ", rsx::ROP_control_bits::MSAA_WRITE_ENABLE_BIT },
{ "INT_FRAMEBUFFER_BIT ", rsx::ROP_control_bits::INT_FRAMEBUFFER_BIT },
{ "POLYGON_STIPPLE_ENABLE_BIT ", rsx::ROP_control_bits::POLYGON_STIPPLE_ENABLE_BIT },
{ "ALPHA_TEST_FUNC_OFFSET ", rsx::ROP_control_bits::ALPHA_FUNC_OFFSET },
{ "ALPHA_TEST_FUNC_LENGTH ", rsx::ROP_control_bits::ALPHA_FUNC_NUM_BITS },
{ "MSAA_SAMPLE_CTRL_OFFSET ", rsx::ROP_control_bits::MSAA_SAMPLE_CTRL_OFFSET },
{ "MSAA_SAMPLE_CTRL_LENGTH ", rsx::ROP_control_bits::MSAA_SAMPLE_CTRL_NUM_BITS },
{ "ROP_CMD_MASK ", rsx::ROP_control_bits::ROP_CMD_MASK }
});
if (props.fp32_outputs || !props.supports_native_fp16)
{
enabled_options.push_back("_32_BIT_OUTPUT");
}
OS << "// Workaround for broken early discard in some drivers\n";
if (props.disable_early_discard)
{
OS << "bool _fragment_discard = false;\n";
OS << "#define _kill() _fragment_discard = true\n\n";
}
else
{
OS << "#define _kill() discard\n\n";
}
if (props.require_texture_ops)
{
// Declare special texture control flags
OS << "#define GAMMA_R_MASK (1 << " << rsx::texture_control_bits::GAMMA_R << ")\n";
OS << "#define GAMMA_G_MASK (1 << " << rsx::texture_control_bits::GAMMA_G << ")\n";
OS << "#define GAMMA_B_MASK (1 << " << rsx::texture_control_bits::GAMMA_B << ")\n";
OS << "#define GAMMA_A_MASK (1 << " << rsx::texture_control_bits::GAMMA_A << ")\n";
OS << "#define EXPAND_R_MASK (1 << " << rsx::texture_control_bits::EXPAND_R << ")\n";
OS << "#define EXPAND_G_MASK (1 << " << rsx::texture_control_bits::EXPAND_G << ")\n";
OS << "#define EXPAND_B_MASK (1 << " << rsx::texture_control_bits::EXPAND_B << ")\n";
OS << "#define EXPAND_A_MASK (1 << " << rsx::texture_control_bits::EXPAND_A << ")\n\n";
OS << "#define ALPHAKILL " << rsx::texture_control_bits::ALPHAKILL << "\n";
OS << "#define RENORMALIZE " << rsx::texture_control_bits::RENORMALIZE << "\n";
OS << "#define DEPTH_FLOAT " << rsx::texture_control_bits::DEPTH_FLOAT << "\n";
OS << "#define DEPTH_COMPARE " << rsx::texture_control_bits::DEPTH_COMPARE_OP << "\n";
OS << "#define FILTERED_MAG_BIT " << rsx::texture_control_bits::FILTERED_MAG << "\n";
OS << "#define FILTERED_MIN_BIT " << rsx::texture_control_bits::FILTERED_MIN << "\n";
OS << "#define INT_COORDS_BIT " << rsx::texture_control_bits::UNNORMALIZED_COORDS << "\n";
OS << "#define GAMMA_CTRL_MASK (GAMMA_R_MASK|GAMMA_G_MASK|GAMMA_B_MASK|GAMMA_A_MASK)\n";
OS << "#define SIGN_EXPAND_MASK (EXPAND_R_MASK|EXPAND_G_MASK|EXPAND_B_MASK|EXPAND_A_MASK)\n";
OS << "#define FILTERED_MASK (FILTERED_MAG_BIT|FILTERED_MIN_BIT)\n\n";
enabled_options.push_back("_DISABLE_EARLY_DISCARD");
}
}
if (props.require_lit_emulation)
{
OS <<
"vec4 lit_legacy(const in vec4 val)"
"{\n"
" vec4 clamped_val = val;\n"
" clamped_val.x = max(val.x, 0.);\n"
" clamped_val.y = max(val.y, 0.);\n"
" vec4 result;\n"
" result.x = 1.;\n"
" result.w = 1.;\n"
" result.y = clamped_val.x;\n"
" result.z = clamped_val.x > 0. ? exp(clamped_val.w * log(max(clamped_val.y, 0.0000000001))) : 0.;\n"
" return result;\n"
"}\n\n";
}
// Import common header
program_common::define_glsl_switches(OS, enabled_options);
enabled_options.clear();
OS <<
#include "GLSLSnippets/RSXProg/RSXProgramCommon.glsl"
;
if (props.domain == glsl::program_domain::glsl_vertex_program)
{
if (props.require_explicit_invariance)
{
// PS3 has shader invariance, but we don't really care about most attributes outside ATTR0
OS << "invariant gl_Position;\n\n";
enabled_options.push_back("_FORCE_POSITION_INVARIANCE");
}
if (props.emulate_zclip_transform)
{
if (props.emulate_depth_clip_only)
{
// Technically the depth value here is the 'final' depth that should be stored in the Z buffer.
// Forward mapping eqn is d' = d * (f - n) + n, where d' is the stored Z value (this) and d is the normalized API value.
OS <<
"vec4 apply_zclip_xform(const in vec4 pos, const in float near_plane, const in float far_plane)\n"
"{\n"
" if (pos.w != 0.0)\n"
" {\n"
" const float real_n = min(far_plane, near_plane);\n"
" const float real_f = max(far_plane, near_plane);\n"
" const double depth_range = double(real_f - real_n);\n"
" const double inv_range = (depth_range > 0.000001) ? (1.0 / (depth_range * pos.w)) : 0.0;\n"
" const double actual_d = (double(pos.z) - double(real_n * pos.w)) * inv_range;\n"
" const double nearest_d = floor(actual_d + 0.5);\n"
" const double epsilon = (inv_range * pos.w) / 16777215.;\n" // Epsilon value is the minimum discernable change in Z that should affect the stored Z
" const double d = _select(actual_d, nearest_d, abs(actual_d - nearest_d) < epsilon);\n"
" return vec4(pos.xy, float(d * pos.w), pos.w);\n"
" }\n"
" else\n"
" {\n"
" return pos;\n" // Only values where Z=0 can ever pass this clip
" }\n"
"}\n\n";
enabled_options.push_back("_EMULATE_ZCLIP_XFORM_STANDARD");
}
else
{
OS <<
"vec4 apply_zclip_xform(const in vec4 pos, const in float near_plane, const in float far_plane)\n"
"{\n"
" float d = float(pos.z / pos.w);\n"
" if (d < 0.f && d >= near_plane)\n"
" {\n"
" // Clamp\n"
" d = 0.f;\n"
" }\n"
" else if (d > 1.f && d <= far_plane)\n"
" {\n"
" // Compress Z and store towards highest end of the range\n"
" d = min(1., 0.99 + (0.01 * (pos.z - near_plane) / (far_plane - near_plane)));\n"
" }\n"
" else\n" // This catch-call also handles w=0 since d=inf
" {\n"
" return pos;\n"
" }\n"
"\n"
" return vec4(pos.x, pos.y, d * pos.w, pos.w);\n"
"}\n\n";
enabled_options.push_back("_EMULATE_ZCLIP_XFORM_FALLBACK");
}
}
// Import vertex header
program_common::define_glsl_switches(OS, enabled_options);
OS <<
#include "GLSLSnippets/RSXProg/RSXVertexPrologue.glsl"
;
return;
}
program_common::insert_compare_op(OS);
if (props.emulate_coverage_tests)
{
// Purely stochastic
OS <<
"bool coverage_test_passes(const in vec4 _sample)\n"
"{\n"
" float random = _rand(gl_FragCoord);\n"
" return (_sample.a > random);\n"
"}\n\n";
enabled_options.push_back("_EMULATE_COVERAGE_TEST");
}
if (!props.fp32_outputs || props.require_linear_to_srgb)
{
OS <<
"vec4 linear_to_srgb(const in vec4 cl)\n"
"{\n"
" vec4 low = cl * 12.92;\n"
" vec4 high = 1.055 * pow(cl, vec4(1. / 2.4)) - 0.055;\n"
" bvec4 selection = lessThan(cl, vec4(0.0031308));\n"
" return clamp(mix(high, low, selection), 0., 1.);\n"
"}\n\n";
enabled_options.push_back("_ENABLE_LINEAR_TO_SRGB");
}
if (props.require_texture_ops || props.require_srgb_to_linear)
{
OS <<
"vec4 srgb_to_linear(const in vec4 cs)\n"
"{\n"
" vec4 a = cs / 12.92;\n"
" vec4 b = pow((cs + 0.055) / 1.055, vec4(2.4));\n"
" return _select(a, b, greaterThan(cs, vec4(0.04045)));\n"
"}\n\n";
enabled_options.push_back("_ENABLE_SRGB_TO_LINEAR");
}
if (props.require_depth_conversion)
if (props.require_wpos)
{
ensure(props.require_texture_ops);
//NOTE: Memory layout is fetched as byteswapped BGRA [GBAR] (GOW collection, DS2, DeS)
//The A component (Z) is useless (should contain stencil8 or just 1)
OS <<
"vec4 decode_depth24(const in float depth_value, const in bool depth_float)\n"
"{\n"
" uint value;\n"
" if (!depth_float)\n"
" value = uint(depth_value * 16777215.);\n"
" else\n"
" value = _get_bits(floatBitsToUint(depth_value), 7, 24);\n"
"\n"
" uint b = _get_bits(value, 0, 8);\n"
" uint g = _get_bits(value, 8, 8);\n"
" uint r = _get_bits(value, 16, 8);\n"
" return vec4(float(g)/255., float(b)/255., 1., float(r)/255.);\n"
"}\n\n"
"vec4 remap_vector(const in vec4 color, const in uint remap)\n"
"{\n"
" vec4 result;\n"
" if (_get_bits(remap, 0, 8) == 0xE4)\n"
" {\n"
" result = color;\n"
" }\n"
" else\n"
" {\n"
" uvec4 remap_channel = uvec4(remap) >> uvec4(2, 4, 6, 0);\n"
" remap_channel &= 3;\n"
" remap_channel = (remap_channel + 3) % 4; // Map A-R-G-B to R-G-B-A\n\n"
" // Generate remapped result\n"
" result.a = color[remap_channel.a];\n"
" result.r = color[remap_channel.r];\n"
" result.g = color[remap_channel.g];\n"
" result.b = color[remap_channel.b];\n"
" }\n\n"
" if (_get_bits(remap, 8, 8) == 0xAA)\n"
" return result;\n\n"
" uvec4 remap_select = uvec4(remap) >> uvec4(10, 12, 14, 8);\n"
" remap_select &= 3;\n"
" bvec4 choice = lessThan(remap_select, uvec4(2));\n"
" return _select(result, vec4(remap_select), choice);\n"
"}\n\n"
"vec4 convert_z24x8_to_rgba8(const in vec2 depth_stencil, const in uint remap, const in uint flags)\n"
"{\n"
" vec4 result = decode_depth24(depth_stencil.x, _test_bit(flags, DEPTH_FLOAT));\n"
" result.z = depth_stencil.y / 255.;\n\n"
" if (remap == 0xAAE4)\n"
" return result;\n\n"
" return remap_vector(result, remap);\n"
"}\n\n";
enabled_options.push_back("_ENABLE_WPOS");
}
if (props.require_fog_read)
{
program_common::define_glsl_constants<rsx::fog_mode>(OS,
{
{ "FOG_LINEAR", rsx::fog_mode::linear },
{ "FOG_EXP", rsx::fog_mode::exponential },
{ "FOG_EXP2", rsx::fog_mode::exponential2 },
{ "FOG_LINEAR_ABS", rsx::fog_mode::linear_abs },
{ "FOG_EXP_ABS", rsx::fog_mode::exponential_abs },
{ "FOG_EXP2_ABS", rsx::fog_mode::exponential2_abs },
});
enabled_options.push_back("_ENABLE_FOG_READ");
}
// Import fragment header
program_common::define_glsl_switches(OS, enabled_options);
enabled_options.clear();
OS <<
#include "GLSLSnippets/RSXProg/RSXFragmentPrologue.glsl"
;
if (props.require_texture_ops)
{
OS <<
// Declare special texture control flags
OS << "#define GAMMA_R_MASK (1 << " << rsx::texture_control_bits::GAMMA_R << ")\n";
OS << "#define GAMMA_G_MASK (1 << " << rsx::texture_control_bits::GAMMA_G << ")\n";
OS << "#define GAMMA_B_MASK (1 << " << rsx::texture_control_bits::GAMMA_B << ")\n";
OS << "#define GAMMA_A_MASK (1 << " << rsx::texture_control_bits::GAMMA_A << ")\n";
OS << "#define EXPAND_R_MASK (1 << " << rsx::texture_control_bits::EXPAND_R << ")\n";
OS << "#define EXPAND_G_MASK (1 << " << rsx::texture_control_bits::EXPAND_G << ")\n";
OS << "#define EXPAND_B_MASK (1 << " << rsx::texture_control_bits::EXPAND_B << ")\n";
OS << "#define EXPAND_A_MASK (1 << " << rsx::texture_control_bits::EXPAND_A << ")\n\n";
//TODO: Move all the texture read control operations here
"vec4 process_texel(in vec4 rgba, const in uint control_bits)\n"
"{\n"
" if (control_bits == 0)\n"
" {\n"
" return rgba;\n"
" }\n"
"\n"
" if (_test_bit(control_bits, ALPHAKILL))\n"
" {\n"
" // Alphakill\n"
" if (rgba.a < 0.000001)\n"
" {\n"
" _kill();\n"
" return rgba;\n"
" }\n"
" }\n"
"\n"
" if (_test_bit(control_bits, RENORMALIZE))\n"
" {\n"
" // Renormalize to 8-bit (PS3) accuracy\n"
" rgba = floor(rgba * 255.);\n"
" rgba /= 255.;\n"
" }\n"
"\n"
" uvec4 mask;\n"
" vec4 convert;\n"
" uint op_mask = control_bits & uint(SIGN_EXPAND_MASK);\n"
"\n"
" if (op_mask != 0)\n"
" {\n"
" // Expand to signed normalized\n"
" mask = uvec4(op_mask) & uvec4(EXPAND_R_MASK, EXPAND_G_MASK, EXPAND_B_MASK, EXPAND_A_MASK);\n"
" convert = (rgba * 2.f - 1.f);\n"
" rgba = _select(rgba, convert, notEqual(mask, uvec4(0)));\n"
" }\n"
"\n"
" op_mask = control_bits & uint(GAMMA_CTRL_MASK);\n"
" if (op_mask != 0u)\n"
" {\n"
" // Gamma correction\n"
" mask = uvec4(op_mask) & uvec4(GAMMA_R_MASK, GAMMA_G_MASK, GAMMA_B_MASK, GAMMA_A_MASK);\n"
" convert = srgb_to_linear(rgba);\n"
" return _select(rgba, convert, notEqual(mask, uvec4(0)));\n"
" }\n"
"\n"
" return rgba;\n"
"}\n\n";
OS << "#define ALPHAKILL " << rsx::texture_control_bits::ALPHAKILL << "\n";
OS << "#define RENORMALIZE " << rsx::texture_control_bits::RENORMALIZE << "\n";
OS << "#define DEPTH_FLOAT " << rsx::texture_control_bits::DEPTH_FLOAT << "\n";
OS << "#define DEPTH_COMPARE " << rsx::texture_control_bits::DEPTH_COMPARE_OP << "\n";
OS << "#define FILTERED_MAG_BIT " << rsx::texture_control_bits::FILTERED_MAG << "\n";
OS << "#define FILTERED_MIN_BIT " << rsx::texture_control_bits::FILTERED_MIN << "\n";
OS << "#define INT_COORDS_BIT " << rsx::texture_control_bits::UNNORMALIZED_COORDS << "\n";
OS << "#define GAMMA_CTRL_MASK (GAMMA_R_MASK|GAMMA_G_MASK|GAMMA_B_MASK|GAMMA_A_MASK)\n";
OS << "#define SIGN_EXPAND_MASK (EXPAND_R_MASK|EXPAND_G_MASK|EXPAND_B_MASK|EXPAND_A_MASK)\n";
OS << "#define FILTERED_MASK (FILTERED_MAG_BIT|FILTERED_MIN_BIT)\n\n";
if (props.require_texture_expand)
{
OS <<
"uint _texture_flag_override = 0;\n"
"#define _enable_texture_expand() _texture_flag_override = SIGN_EXPAND_MASK\n"
"#define _disable_texture_expand() _texture_flag_override = 0\n"
"#define TEX_FLAGS(index) (texture_parameters[index].flags | _texture_flag_override)\n";
enabled_options.push_back("_ENABLE_TEXTURE_EXPAND");
}
else
{
OS <<
"#define TEX_FLAGS(index) texture_parameters[index].flags\n";
}
OS <<
"#define TEX_NAME(index) tex##index\n"
"#define TEX_NAME_STENCIL(index) tex##index##_stencil\n\n"
"#define COORD_SCALE1(index, coord1) ((coord1 + texture_parameters[index].scale_bias.w) * texture_parameters[index].scale_bias.x)\n"
"#define COORD_SCALE2(index, coord2) ((coord2 + texture_parameters[index].scale_bias.w) * texture_parameters[index].scale_bias.xy)\n"
"#define COORD_SCALE3(index, coord3) ((coord3 + texture_parameters[index].scale_bias.w) * texture_parameters[index].scale_bias.xyz)\n\n"
"#define TEX1D(index, coord1) process_texel(texture(TEX_NAME(index), COORD_SCALE1(index, coord1)), TEX_FLAGS(index))\n"
"#define TEX1D_BIAS(index, coord1, bias) process_texel(texture(TEX_NAME(index), COORD_SCALE1(index, coord1), bias), TEX_FLAGS(index))\n"
"#define TEX1D_LOD(index, coord1, lod) process_texel(textureLod(TEX_NAME(index), COORD_SCALE1(index, coord1), lod), TEX_FLAGS(index))\n"
"#define TEX1D_GRAD(index, coord1, dpdx, dpdy) process_texel(textureGrad(TEX_NAME(index), COORD_SCALE1(index, coord1), dpdx, dpdy), TEX_FLAGS(index))\n"
"#define TEX1D_PROJ(index, coord4) process_texel(textureProj(TEX_NAME(index), vec2(COORD_SCALE1(index, coord4.x), coord4.w)), TEX_FLAGS(index))\n"
"#define TEX2D(index, coord2) process_texel(texture(TEX_NAME(index), COORD_SCALE2(index, coord2)), TEX_FLAGS(index))\n"
"#define TEX2D_BIAS(index, coord2, bias) process_texel(texture(TEX_NAME(index), COORD_SCALE2(index, coord2), bias), TEX_FLAGS(index))\n"
"#define TEX2D_LOD(index, coord2, lod) process_texel(textureLod(TEX_NAME(index), COORD_SCALE2(index, coord2), lod), TEX_FLAGS(index))\n"
"#define TEX2D_GRAD(index, coord2, dpdx, dpdy) process_texel(textureGrad(TEX_NAME(index), COORD_SCALE2(index, coord2), dpdx, dpdy), TEX_FLAGS(index))\n"
"#define TEX2D_PROJ(index, coord4) process_texel(textureProj(TEX_NAME(index), vec4(COORD_SCALE2(index, coord4.xy), coord4.z, coord4.w)), TEX_FLAGS(index))\n\n";
if (props.emulate_shadow_compare)
{
OS <<
"#define SHADOW_COORD(index, coord3) vec3(COORD_SCALE2(index, coord3.xy), _test_bit(TEX_FLAGS(index), DEPTH_FLOAT)? coord3.z : min(float(coord3.z), 1.0))\n"
"#define SHADOW_COORD4(index, coord4) vec4(SHADOW_COORD(index, coord4.xyz), coord4.w)\n"
"#define SHADOW_COORD_PROJ(index, coord4) vec4(COORD_SCALE2(index, coord4.xy), _test_bit(TEX_FLAGS(index), DEPTH_FLOAT)? coord4.z : min(coord4.z, coord4.w), coord4.w)\n\n"
enabled_options.push_back("_EMULATED_TEXSHADOW");
}
"#define TEX2D_SHADOW(index, coord3) texture(TEX_NAME(index), SHADOW_COORD(index, coord3))\n"
"#define TEX3D_SHADOW(index, coord4) texture(TEX_NAME(index), SHADOW_COORD4(index, coord4))\n"
"#define TEX2D_SHADOWPROJ(index, coord4) textureProj(TEX_NAME(index), SHADOW_COORD_PROJ(index, coord4))\n";
}
else
{
OS <<
"#define TEX2D_SHADOW(index, coord3) texture(TEX_NAME(index), vec3(COORD_SCALE2(index, coord3.xy), coord3.z))\n"
"#define TEX3D_SHADOW(index, coord4) texture(TEX_NAME(index), vec4(COORD_SCALE3(index, coord4.xyz), coord4.w))\n"
"#define TEX2D_SHADOWPROJ(index, coord4) textureProj(TEX_NAME(index), vec4(COORD_SCALE2(index, coord4.xy), coord4.zw))\n";
}
program_common::define_glsl_switches(OS, enabled_options);
enabled_options.clear();
OS <<
"#define TEX3D(index, coord3) process_texel(texture(TEX_NAME(index), COORD_SCALE3(index, coord3)), TEX_FLAGS(index))\n"
"#define TEX3D_BIAS(index, coord3, bias) process_texel(texture(TEX_NAME(index), COORD_SCALE3(index, coord3), bias), TEX_FLAGS(index))\n"
"#define TEX3D_LOD(index, coord3, lod) process_texel(textureLod(TEX_NAME(index), COORD_SCALE3(index, coord3), lod), TEX_FLAGS(index))\n"
"#define TEX3D_GRAD(index, coord3, dpdx, dpdy) process_texel(textureGrad(TEX_NAME(index), COORD_SCALE3(index, coord3), dpdx, dpdy), TEX_FLAGS(index))\n"
"#define TEX3D_PROJ(index, coord4) process_texel(texture(TEX_NAME(index), COORD_SCALE3(index, coord4.xyz) / coord4.w), TEX_FLAGS(index))\n\n";
#include "GLSLSnippets/RSXProg/RSXFragmentTextureOps.glsl"
;
if (props.require_depth_conversion)
{
OS <<
"#define ZS_READ(index, coord) vec2(texture(TEX_NAME(index), coord).r, float(texture(TEX_NAME_STENCIL(index), coord).x))\n"
"#define TEX1D_Z24X8_RGBA8(index, coord1) process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE1(index, coord1)), texture_parameters[index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))\n"
"#define TEX2D_Z24X8_RGBA8(index, coord2) process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE2(index, coord2)), texture_parameters[index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))\n"
"#define TEX3D_Z24X8_RGBA8(index, coord3) process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE3(index, coord3)), texture_parameters[index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))\n\n";
#include "GLSLSnippets/RSXProg/RSXFragmentTextureDepthConversion.glsl"
;
}
if (props.require_msaa_ops)
{
OS <<
"#define ZCOMPARE_FUNC(index) _get_bits(TEX_FLAGS(index), DEPTH_COMPARE, 3)\n"
"#define ZS_READ_MS(index, coord) vec2(sampleTexture2DMS(TEX_NAME(index), coord, index).r, float(sampleTexture2DMS(TEX_NAME_STENCIL(index), coord, index).x))\n"
"#define TEX2D_MS(index, coord2) process_texel(sampleTexture2DMS(TEX_NAME(index), coord2, index), TEX_FLAGS(index))\n"
"#define TEX2D_SHADOW_MS(index, coord3) vec4(comparison_passes(sampleTexture2DMS(TEX_NAME(index), coord3.xy, index).x, coord3.z, ZCOMPARE_FUNC(index)))\n"
"#define TEX2D_SHADOWPROJ_MS(index, coord4) TEX2D_SHADOW_MS(index, (coord4.xyz / coord4.w))\n"
"#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";
#include "GLSLSnippets/RSXProg/RSXFragmentTextureMSAAOps.glsl"
;
OS <<
"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";
// Generate multiple versions of the actual sampler code.
// We could use defines to generate these, but I don't trust some OpenGL compilers to do the right thing.
const std::string_view msaa_sampling_impl =
#include "GLSLSnippets/RSXProg/RSXFragmentTextureMSAAOpsInternal.glsl"
;
auto insert_msaa_sample_code = [&OS](const std::string_view& sampler_type)
{
OS <<
"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 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 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 (_get_bits(flags, FILTERED_MAG_BIT, 2) == 0)\n"
" {\n"
" return sample0;\n"
" }\n"
"\n"
" // Bilinear scaling, with upto 2x2 downscaling with simple weights\n"
" 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"
" 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"
" vec4 a, b;\n"
" float factor;\n"
" const vec4 sample2 = texelFetch2DMS(tex, sample_count, icoords, index, ivec2(0, 1)); // Top left\n"
"\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.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"
" // We only have 2 rows computed for performance reasons, so combine rows 1 and 2\n"
" return a * weights.x + b * (weights.y + weights.z);\n"
" }\n"
" else if (actual_step.y < uv_step.y)\n"
" {\n"
" // Upscale in Y\n"
" factor = fract(normalized_coords.y * image_size.y);\n"
" return mix(a, b, factor);\n"
" }\n"
"}\n\n";
};
insert_msaa_sample_code("sampler2DMS");
OS << fmt::replace_all(msaa_sampling_impl, "_MSAA_SAMPLER_TYPE_", "sampler2DMS");
if (props.require_depth_conversion)
{
insert_msaa_sample_code("usampler2DMS");
OS << fmt::replace_all(msaa_sampling_impl, "_MSAA_SAMPLER_TYPE_", "usampler2DMS");
}
}
}
if (props.require_wpos)
{
OS <<
"vec4 get_wpos()\n"
"{\n"
" float abs_scale = abs(wpos_scale);\n"
" return (gl_FragCoord * vec4(abs_scale, wpos_scale, 1., 1.)) + vec4(0., wpos_bias, 0., 0.);\n"
"}\n\n";
}
}
std::string getFunctionImpl(FUNCTION f)
@ -1152,12 +747,8 @@ namespace glsl
// Global types and stuff
// Must be compatible with std140 packing rules
OS <<
"struct sampler_info\n"
"{\n"
" vec4 scale_bias;\n"
" uint remap;\n"
" uint flags;\n"
"};\n\n";
#include "GLSLSnippets/RSXProg/RSXDefines2.glsl"
;
}
void insert_fragment_shader_inputs_block(

7
rpcs3/Emu/RSX/Program/GLSLCommon.h
View file

@ -76,13 +76,6 @@ namespace rsx
};
}
namespace program_common
{
void insert_compare_op(std::ostream& OS, bool low_precision);
void insert_compare_op_vector(std::ostream& OS);
void insert_fog_declaration(std::ostream& OS, std::string_view vector_type = "vec4", std::string_view input_coord = "fog_c");
}
namespace glsl
{
struct two_sided_lighting_config

73
rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXFragmentPrologue.glsl
View file

@ -1,5 +1,4 @@
R"(
#ifdef _32_BIT_OUTPUT
// Default. Used when we're not utilizing native fp16
#define round_to_8bit(v4) (floor(fma(v4, vec4(255.), vec4(0.5))) / vec4(255.))
@ -9,9 +8,10 @@ R"(
#endif
#ifdef _DISABLE_EARLY_DISCARD
#define kill() _fragment_discard = true
bool _fragment_discard = false;
#define _kill() _fragment_discard = true
#else
#define kill() discard
#define _kill() discard
#endif
#ifdef _ENABLE_WPOS
@ -22,6 +22,73 @@ vec4 get_wpos()
}
#endif
#ifdef _ENABLE_FOG_READ
vec4 fetch_fog_value(const in uint mode)
{
vec4 result = vec4(fog_c.x, 0., 0., 0.);
switch(mode)
{
default:
return result;
case FOG_LINEAR:
// linear
result.y = fog_param1 * fog_c.x + (fog_param0 - 1.);
break;
case FOG_EXP:
// exponential
result.y = exp(11.084 * (fog_param1 * fog_c.x + fog_param0 - 1.5));
break;
case FOG_EXP2:
// exponential2
result.y = exp(-pow(4.709 * (fog_param1 * fog_c.x + fog_param0 - 1.5), 2.));
break;
case FOG_EXP_ABS:
// exponential_abs
result.y = exp(11.084 * (fog_param1 * abs(fog_c.x) + fog_param0 - 1.5));
break;
case FOG_EXP2_ABS:
// exponential2_abs
result.y = exp(-pow(4.709 * (fog_param1 * abs(fog_c.x) + fog_param0 - 1.5), 2.));
break;
case FOG_LINEAR_ABS:
// linear_abs
result.y = fog_param1 * abs(fog_c.x) + (fog_param0 - 1.);
break;
}
result.y = clamp(result.y, 0., 1.);
return result;
}
#endif
#ifdef _EMULATE_COVERAGE_TEST
// Purely stochastic
bool coverage_test_passes(const in vec4 _sample)
{
float random = _rand(gl_FragCoord);
return (_sample.a > random);
}
#endif
#ifdef _ENABLE_LINEAR_TO_SRGB
vec4 linear_to_srgb(const in vec4 cl)
{
vec4 low = cl * 12.92;
vec4 high = 1.055 * pow(cl, vec4(1. / 2.4)) - 0.055;
bvec4 selection = lessThan(cl, vec4(0.0031308));
return clamp(mix(high, low, selection), 0., 1.);
}
#endif
#ifdef _ENABLE_SRGB_TO_LINEAR
vec4 srgb_to_linear(const in vec4 cs)
{
vec4 a = cs / 12.92;
vec4 b = pow((cs + 0.055) / 1.055, vec4(2.4));
return _select(a, b, greaterThan(cs, vec4(0.04045)));
}
#endif
// Required by all fragment shaders for alpha test
bool comparison_passes(const in float a, const in float b, const in uint func)
{

2
rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXProgramCommon.glsl
View file

@ -24,4 +24,4 @@ vec4 lit_legacy(const in vec4 val)
}
#endif
)"
)"

2
rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXVertexPrologue.glsl
View file

@ -1,5 +1,6 @@
R"(
#ifdef _FORCE_POSITION_INVARIANCE
// PS3 has shader invariance, but we don't really care about most attributes outside ATTR0
invariant gl_Position;
#endif
@ -54,5 +55,4 @@ vec4 apply_zclip_xform(
}\n
#endif
)"

1
rpcs3/Emu/RSX/Program/GLSLTypes.h
View file

@ -32,6 +32,7 @@ namespace glsl
bool require_srgb_to_linear : 1;
bool require_linear_to_srgb : 1;
bool require_explicit_invariance: 1;
bool require_fog_read : 1;
bool emulate_coverage_tests : 1;
bool emulate_shadow_compare : 1;
bool emulate_zclip_transform : 1;

4
rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp
View file

@ -243,6 +243,7 @@ void VKFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
m_shader_props.require_texture_expand = properties.has_exp_tex_op;
m_shader_props.require_srgb_to_linear = properties.has_upg;
m_shader_props.require_linear_to_srgb = properties.has_pkg;
m_shader_props.require_fog_read = properties.in_register_mask & in_fogc;
m_shader_props.emulate_coverage_tests = g_cfg.video.antialiasing_level == msaa_level::none;
m_shader_props.emulate_shadow_compare = device_props.emulate_depth_compare;
m_shader_props.low_precision_tests = device_props.has_low_precision_rounding && !(m_prog.ctrl & RSX_SHADER_CONTROL_ATTRIBUTE_INTERPOLATION);
@ -255,9 +256,6 @@ void VKFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
void VKFragmentDecompilerThread::insertMainStart(std::stringstream & OS)
{
if (properties.in_register_mask & in_fogc)
program_common::insert_fog_declaration(OS);
std::set<std::string> output_registers;
if (m_ctrl & CELL_GCM_SHADER_CONTROL_32_BITS_EXPORTS)
{