RPCS3/rpcs3
1
0
Fork 0
mirror of https://github.com/RPCS3/rpcs3.git synced 2026-01-21 16:10:23 +01:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

vk: Move texture parameter stream to SSBO layout

Browse source
This commit is contained in:
kd-11 2025-07-31 01:51:52 +03:00 committed by kd-11
parent 183acd4196
commit c347435151
8 changed files with 39 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

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

@ -189,7 +189,9 @@ void GLFragmentDecompilerThread::insertConstants(std::stringstream & OS)
"layout(std140, binding = " << GL_RASTERIZER_STATE_BIND_SLOT << ") uniform RasterizerHeap\n"
"{\n"
" uvec4 stipple_pattern[8];\n"
"};\n\n";
"};\n\n"
"#define texture_base_index 0\n\n";
}
void GLFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)

16
rpcs3/Emu/RSX/Program/GLSLInterpreter/FragmentInterpreter.glsl
View file

@ -366,19 +366,19 @@ vec4 _texture(in vec4 coord, float bias)
switch (type)
{
case RSX_SAMPLE_TEXTURE_1D:
coord.x = _texcoord_xform(coord.x, texture_parameters[ur0]);
coord.x = _texcoord_xform(coord.x, texture_parameters[ur0 + texture_base_index]);
vr0 = texture(SAMPLER1D(ur0), coord.x, bias);
break;
case RSX_SAMPLE_TEXTURE_2D:
coord.xy = _texcoord_xform(coord.xy, texture_parameters[ur0]);
coord.xy = _texcoord_xform(coord.xy, texture_parameters[ur0 + texture_base_index]);
vr0 = texture(SAMPLER2D(ur0), coord.xy, bias);
break;
case RSX_SAMPLE_TEXTURE_CUBE:
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0]);
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0 + texture_base_index]);
vr0 = texture(SAMPLERCUBE(ur0), coord.xyz, bias);
break;
case RSX_SAMPLE_TEXTURE_3D:
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0]);
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0 + texture_base_index]);
vr0 = texture(SAMPLER3D(ur0), coord.xyz, bias);
break;
}
@ -405,19 +405,19 @@ vec4 _textureLod(in vec4 coord, float lod)
switch (type)
{
case RSX_SAMPLE_TEXTURE_1D:
coord.x = _texcoord_xform(coord.x, texture_parameters[ur0]);
coord.x = _texcoord_xform(coord.x, texture_parameters[ur0 + texture_base_index]);
vr0 = textureLod(SAMPLER1D(ur0), coord.x, lod);
break;
case RSX_SAMPLE_TEXTURE_2D:
coord.xy = _texcoord_xform(coord.xy, texture_parameters[ur0]);
coord.xy = _texcoord_xform(coord.xy, texture_parameters[ur0 + texture_base_index]);
vr0 = textureLod(SAMPLER2D(ur0), coord.xy, lod);
break;
case RSX_SAMPLE_TEXTURE_CUBE:
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0]);
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0 + texture_base_index]);
vr0 = textureLod(SAMPLERCUBE(ur0), coord.xyz, lod);
break;
case RSX_SAMPLE_TEXTURE_3D:
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0]);
coord.xyz = _texcoord_xform(coord.xyz, texture_parameters[ur0 + texture_base_index]);
vr0 = textureLod(SAMPLER3D(ur0), coord.xyz, lod);
break;
}

6
rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXFragmentTextureDepthConversion.glsl
View file

@ -1,8 +1,8 @@
R"(
#define ZS_READ(index, coord) vec2(texture(TEX_NAME(index), coord).r, float(texture(TEX_NAME_STENCIL(index), coord).x))
#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))
#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))
#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))
#define TEX1D_Z24X8_RGBA8(index, coord1) _process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE1(index, coord1)), texture_parameters[index + texture_base_index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))
#define TEX2D_Z24X8_RGBA8(index, coord2) _process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE2(index, coord2)), texture_parameters[index + texture_base_index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))
#define TEX3D_Z24X8_RGBA8(index, coord3) _process_texel(convert_z24x8_to_rgba8(ZS_READ(index, COORD_SCALE3(index, coord3)), texture_parameters[index + texture_base_index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))
// 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)

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

@ -4,7 +4,7 @@ R"(
#define TEX2D_MS(index, coord2) _process_texel(sampleTexture2DMS(TEX_NAME(index), coord2, index), TEX_FLAGS(index))
#define TEX2D_SHADOW_MS(index, coord3) vec4(comparison_passes(sampleTexture2DMS(TEX_NAME(index), coord3.xy, index).x, coord3.z, ZCOMPARE_FUNC(index)))
#define TEX2D_SHADOWPROJ_MS(index, coord4) TEX2D_SHADOW_MS(index, (coord4.xyz / coord4.w))
#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))
#define TEX2D_Z24X8_RGBA8_MS(index, coord2) _process_texel(convert_z24x8_to_rgba8(ZS_READ_MS(index, coord2), texture_parameters[index + texture_base_index].remap, TEX_FLAGS(index)), TEX_FLAGS(index))
vec3 compute2x2DownsampleWeights(const in float coord, const in float uv_step, const in float actual_step)
{

16
rpcs3/Emu/RSX/Program/GLSLSnippets/RSXProg/RSXFragmentTextureOps.glsl
View file

@ -24,17 +24,17 @@ R"(
uint _texture_flag_override = 0;
#define _enable_texture_expand() _texture_flag_override = SIGN_EXPAND_MASK
#define _disable_texture_expand() _texture_flag_override = 0
#define TEX_FLAGS(index) (texture_parameters[index].flags | _texture_flag_override)
#define TEX_FLAGS(index) (texture_parameters[index + texture_base_index].flags | _texture_flag_override)
#else
#define TEX_FLAGS(index) texture_parameters[index].flags
#define TEX_FLAGS(index) texture_parameters[index + texture_base_index].flags
#endif
#define TEX_NAME(index) tex##index
#define TEX_NAME_STENCIL(index) tex##index##_stencil
#define COORD_SCALE1(index, coord1) _texcoord_xform(coord1, texture_parameters[index])
#define COORD_SCALE2(index, coord2) _texcoord_xform(coord2, texture_parameters[index])
#define COORD_SCALE3(index, coord3) _texcoord_xform(coord3, texture_parameters[index])
#define COORD_SCALE1(index, coord1) _texcoord_xform(coord1, texture_parameters[index + texture_base_index])
#define COORD_SCALE2(index, coord2) _texcoord_xform(coord2, texture_parameters[index + texture_base_index])
#define COORD_SCALE3(index, coord3) _texcoord_xform(coord3, texture_parameters[index + texture_base_index])
#define COORD_PROJ1(index, coord2) COORD_SCALE1(index, coord2.x / coord2.y)
#define COORD_PROJ2(index, coord3) COORD_SCALE2(index, coord3.xy / coord3.z)
#define COORD_PROJ3(index, coord4) COORD_SCALE3(index, coord4.xyz / coord4.w)
@ -57,9 +57,9 @@ R"(
#ifdef _ENABLE_SHADOW
#ifdef _EMULATED_TEXSHADOW
#define SHADOW_COORD(index, coord3) _texcoord_xform_shadow(coord3, texture_parameters[index])
#define SHADOW_COORD4(index, coord4) _texcoord_xform_shadow(coord4, texture_parameters[index])
#define SHADOW_COORD_PROJ(index, coord4) _texcoord_xform_shadow(coord4.xyz / coord4.w, texture_parameters[index])
#define SHADOW_COORD(index, coord3) _texcoord_xform_shadow(coord3, texture_parameters[index + texture_base_index])
#define SHADOW_COORD4(index, coord4) _texcoord_xform_shadow(coord4, texture_parameters[index + texture_base_index])
#define SHADOW_COORD_PROJ(index, coord4) _texcoord_xform_shadow(coord4.xyz / coord4.w, texture_parameters[index + texture_base_index])
#define TEX2D_SHADOW(index, coord3) texture(TEX_NAME(index), SHADOW_COORD(index, coord3))
#define TEX3D_SHADOW(index, coord4) texture(TEX_NAME(index), SHADOW_COORD4(index, coord4))

12
rpcs3/Emu/RSX/VK/VKFragmentProgram.cpp
View file

@ -240,9 +240,9 @@ void VKFragmentDecompilerThread::insertConstants(std::stringstream & OS)
" fragment_context_t fs_contexts[];\n"
"};\n\n";
OS << "layout(std140, set=1, binding=" << vk_prog->binding_table.tex_param_location << ") uniform TextureParametersBuffer\n";
OS << "layout(std430, set=1, binding=" << vk_prog->binding_table.tex_param_location << ") readonly buffer TextureParametersBuffer\n";
OS << "{\n";
OS << " sampler_info texture_parameters[16];\n";
OS << " sampler_info texture_parameters[];\n";
OS << "};\n\n";
OS << "layout(std140, set=1, binding=" << vk_prog->binding_table.polygon_stipple_params_location << ") uniform RasterizerHeap\n";
@ -271,10 +271,11 @@ void VKFragmentDecompilerThread::insertConstants(std::stringstream & OS)
in.location = vk_prog->binding_table.tex_param_location;
in.name = "TextureParametersBuffer";
in.type = vk::glsl::input_type_uniform_buffer;
in.type = vk::glsl::input_type_storage_buffer;
inputs.push_back(in);
in.location = vk_prog->binding_table.polygon_stipple_params_location;
in.type = vk::glsl::input_type_uniform_buffer;
in.name = "RasterizerHeap";
inputs.push_back(in);
@ -283,13 +284,14 @@ void VKFragmentDecompilerThread::insertConstants(std::stringstream & OS)
"{\n"
" layout(offset=12) uint fs_constants_offset;\n"
" uint fs_context_offset;\n"
" uint fs_texture_base_index;\n"
"};\n\n";
const vk::glsl::program_input push_constants
{
.domain = glsl::glsl_fragment_program,
.type = vk::glsl::input_type_push_constant,
.bound_data = vk::glsl::push_constant_ref{.offset = 12, .size = 8 },
.bound_data = vk::glsl::push_constant_ref{.offset = 12, .size = 12 },
.set = vk::glsl::binding_set_index_vertex,
.location = umax,
.name = "fs_push_constants_block"
@ -338,6 +340,8 @@ void VKFragmentDecompilerThread::insertGlobalFunctions(std::stringstream &OS)
"const float wpos_bias fs_contexts[fs_context_offset].wpos_bias;\n\n";
}
OS << "#define texture_base_index fs_texture_base_index\n\n";
glsl::insert_glsl_legacy_function(OS, m_shader_props);
}

12
rpcs3/Emu/RSX/VK/VKGSRender.cpp
View file

@ -512,7 +512,7 @@ VKGSRender::VKGSRender(utils::serial* ar) noexcept : GSRender(ar)
m_attrib_ring_info.create(VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT, VK_ATTRIB_RING_BUFFER_SIZE_M * 0x100000, "attrib buffer", 0x400000, VK_TRUE);
m_fragment_env_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "fragment env buffer");
m_vertex_env_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "vertex env buffer");
m_fragment_texture_params_ring_info.create(VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "fragment texture params buffer");
m_fragment_texture_params_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "fragment texture params buffer");
m_vertex_layout_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "vertex layout buffer", 0x10000, VK_TRUE);
m_fragment_constants_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_UBO_RING_BUFFER_SIZE_M * 0x100000, "fragment constants buffer");
m_transform_constants_ring_info.create(VK_BUFFER_USAGE_STORAGE_BUFFER_BIT, VK_TRANSFORM_CONSTANTS_BUFFER_SIZE_M * 0x100000, "transform constants buffer");
@ -554,7 +554,7 @@ VKGSRender::VKGSRender(utils::serial* ar) noexcept : GSRender(ar)
m_vertex_env_buffer_info = { m_vertex_env_ring_info.heap->value, 0, VK_WHOLE_SIZE };
m_vertex_constants_buffer_info = { m_transform_constants_ring_info.heap->value, 0, VK_WHOLE_SIZE };
m_fragment_env_buffer_info = { m_fragment_env_ring_info.heap->value, 0, VK_WHOLE_SIZE };
m_fragment_texture_params_buffer_info = { m_fragment_texture_params_ring_info.heap->value, 0, 16 };
m_fragment_texture_params_buffer_info = { m_fragment_texture_params_ring_info.heap->value, 0, VK_WHOLE_SIZE };
m_raster_env_buffer_info = { m_raster_env_ring_info.heap->value, 0, 128 };
m_vertex_layout_stream_info = { m_vertex_layout_ring_info.heap->value, 0, VK_WHOLE_SIZE };
m_fragment_constants_buffer_info = { m_fragment_constants_ring_info.heap->value, 0, VK_WHOLE_SIZE };
@ -2029,12 +2029,11 @@ void VKGSRender::load_program_env()
if (update_fragment_texture_env)
{
auto mem = m_fragment_texture_params_ring_info.static_alloc<256, 768>();
auto buf = m_fragment_texture_params_ring_info.map(mem, 768);
m_texture_parameters_dynamic_offset = m_fragment_texture_params_ring_info.static_alloc<16, 768>();
auto buf = m_fragment_texture_params_ring_info.map(m_texture_parameters_dynamic_offset, 768);
current_fragment_program.texture_params.write_to(buf, current_fp_metadata.referenced_textures_mask);
m_fragment_texture_params_ring_info.unmap();
m_fragment_texture_params_buffer_info = { m_fragment_texture_params_ring_info.heap->value, mem, 768 };
}
if (update_raster_env)
@ -2196,6 +2195,7 @@ void VKGSRender::update_vertex_env(u32 id, const vk::vertex_upload_info& vertex_
{
u32 fs_constants_offset;
u32 fs_context_offset;
u32 fs_texture_base_index;
};
struct rsx_prog_vertex_layout_entry_t
@ -2214,6 +2214,7 @@ void VKGSRender::update_vertex_env(u32 id, const vk::vertex_upload_info& vertex_
const u32 vertex_layout_offset = static_cast<u32>(m_vertex_layout_dynamic_offset) / 144u;
const u32 fs_constant_id_offset = static_cast<u32>(m_fragment_constants_dynamic_offset) / 16u;
const u32 fs_context_offset = static_cast<u32>(m_fragment_env_dynamic_offset) / 32u;
const u32 fs_texture_base_index = static_cast<u32>(m_texture_parameters_dynamic_offset) / 48u;
// Pack
rsx_vs_prog_push_constants_block_t vs_push_constants;
@ -2224,6 +2225,7 @@ void VKGSRender::update_vertex_env(u32 id, const vk::vertex_upload_info& vertex_
rsx_fs_prog_push_constants_block_t fs_push_constants;
fs_push_constants.fs_constants_offset = fs_constant_id_offset;
fs_push_constants.fs_context_offset = fs_context_offset;
fs_push_constants.fs_texture_base_index = fs_texture_base_index;
vkCmdPushConstants(
*m_current_command_buffer,

1
rpcs3/Emu/RSX/VK/VKGSRender.h
View file

@ -155,6 +155,7 @@ private:
u64 m_vertex_layout_dynamic_offset = 0;
u64 m_fragment_constants_dynamic_offset = 0;
u64 m_fragment_env_dynamic_offset = 0;
u64 m_texture_parameters_dynamic_offset = 0;
std::array<vk::frame_context_t, VK_MAX_ASYNC_FRAMES> frame_context_storage;
//Temp frame context to use if the real frame queue is overburdened. Only used for storage