Skip to content
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

Test: use software renderer blending logic in fbfetch blend #11387

Draft
wants to merge 10 commits into
base: master
Choose a base branch
from
Draft

Test: use software renderer blending logic in fbfetch blend #11387

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
18f4b4b
PixelShaderGen: Typo: SRCALHA -> SRCALPHA
Pokechu22 Dec 28, 2022
5ab93e8
Test: use software renderer blending logic in fbfetch blend
Pokechu22 Dec 28, 2022
2a3135d
Force fbfetch
Pokechu22 Dec 28, 2022
eb1fae1
Make the change for ubershaders
Pokechu22 Dec 28, 2022
d56ab51
More temp
Pokechu22 Dec 28, 2022
a1b466d
Revert
Pokechu22 Dec 28, 2022
726183f
Partial unrevert
Pokechu22 Dec 28, 2022
c1cb276
Full unrevert
Pokechu22 Dec 28, 2022
3dd3526
Fix order of operations
Pokechu22 Dec 28, 2022
51f393f
Undo forcing pretending fbfetch is supported
Pokechu22 Dec 28, 2022
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
18 changes: 18 additions & 0 deletions Source/Core/VideoBackends/OGL/OGLRender.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -590,6 +590,24 @@ Renderer::Renderer(std::unique_ptr<GLContext> main_gl_context, float backbuffer_
}
else
{
if (GLExtensions::Supports("GL_EXT_shader_framebuffer_fetch"))
{
INFO_LOG_FMT(VIDEO, "GL_EXT_shader_framebuffer_fetch");
g_ogl_config.SupportedFramebufferFetch = EsFbFetchType::FbFetchExt;
}
else if (GLExtensions::Supports("GL_ARM_shader_framebuffer_fetch"))
{
INFO_LOG_FMT(VIDEO, "GL_ARM_shader_framebuffer_fetch");
g_ogl_config.SupportedFramebufferFetch = EsFbFetchType::FbFetchArm;
}
else
{
INFO_LOG_FMT(VIDEO, "Not supported");
g_ogl_config.SupportedFramebufferFetch = EsFbFetchType::FbFetchNone;
}
g_Config.backend_info.bSupportsFramebufferFetch =
g_ogl_config.SupportedFramebufferFetch != EsFbFetchType::FbFetchNone;

if (GLExtensions::Version() < 300)
{
PanicAlertFmtT("GPU: OGL ERROR: Need at least GLSL 1.30\n"
Expand Down
112 changes: 64 additions & 48 deletions Source/Core/VideoCommon/PixelShaderGen.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1844,75 +1844,91 @@ static void WriteBlend(ShaderCode& out, const pixel_shader_uid_data* uid_data)
{
using Common::EnumMap;
static constexpr EnumMap<const char*, SrcBlendFactor::InvDstAlpha> blend_src_factor{
"float3(0,0,0);", // ZERO
"float3(1,1,1);", // ONE
"initial_ocol0.rgb;", // DSTCLR
"float3(1,1,1) - initial_ocol0.rgb;", // INVDSTCLR
"src_color.aaa;", // SRCALPHA
"float3(1,1,1) - src_color.aaa;", // INVSRCALPHA
"initial_ocol0.aaa;", // DSTALPHA
"float3(1,1,1) - initial_ocol0.aaa;", // INVDSTALPHA
"float3(0,0,0)", // ZERO
"float3(1,1,1)", // ONE
"initial_ocol0.rgb", // DSTCLR
"float3(1,1,1) - initial_ocol0.rgb", // INVDSTCLR
"src_color.aaa", // SRCALPHA
"float3(1,1,1) - src_color.aaa", // INVSRCALPHA
"initial_ocol0.aaa", // DSTALPHA
"float3(1,1,1) - initial_ocol0.aaa", // INVDSTALPHA
};
static constexpr EnumMap<const char*, SrcBlendFactor::InvDstAlpha> blend_src_factor_alpha{
"0.0;", // ZERO
"1.0;", // ONE
"initial_ocol0.a;", // DSTCLR
"1.0 - initial_ocol0.a;", // INVDSTCLR
"src_color.a;", // SRCALPHA
"1.0 - src_color.a;", // INVSRCALPHA
"initial_ocol0.a;", // DSTALPHA
"1.0 - initial_ocol0.a;", // INVDSTALPHA
"0.0", // ZERO
"1.0", // ONE
"initial_ocol0.a", // DSTCLR
"1.0 - initial_ocol0.a", // INVDSTCLR
"src_color.a", // SRCALPHA
"1.0 - src_color.a", // INVSRCALPHA
"initial_ocol0.a", // DSTALPHA
"1.0 - initial_ocol0.a", // INVDSTALPHA
};
static constexpr EnumMap<const char*, DstBlendFactor::InvDstAlpha> blend_dst_factor{
"float3(0,0,0);", // ZERO
"float3(1,1,1);", // ONE
"ocol0.rgb;", // SRCCLR
"float3(1,1,1) - ocol0.rgb;", // INVSRCCLR
"src_color.aaa;", // SRCALHA
"float3(1,1,1) - src_color.aaa;", // INVSRCALPHA
"initial_ocol0.aaa;", // DSTALPHA
"float3(1,1,1) - initial_ocol0.aaa;", // INVDSTALPHA
"float3(0,0,0)", // ZERO
"float3(1,1,1)", // ONE
"ocol0.rgb", // SRCCLR
"float3(1,1,1) - ocol0.rgb", // INVSRCCLR
"src_color.aaa", // SRCALPHA
"float3(1,1,1) - src_color.aaa", // INVSRCALPHA
"initial_ocol0.aaa", // DSTALPHA
"float3(1,1,1) - initial_ocol0.aaa", // INVDSTALPHA
};
static constexpr EnumMap<const char*, DstBlendFactor::InvDstAlpha> blend_dst_factor_alpha{
"0.0;", // ZERO
"1.0;", // ONE
"ocol0.a;", // SRCCLR
"1.0 - ocol0.a;", // INVSRCCLR
"src_color.a;", // SRCALPHA
"1.0 - src_color.a;", // INVSRCALPHA
"initial_ocol0.a;", // DSTALPHA
"1.0 - initial_ocol0.a;", // INVDSTALPHA
"0.0", // ZERO
"1.0", // ONE
"ocol0.a", // SRCCLR
"1.0 - ocol0.a", // INVSRCCLR
"src_color.a", // SRCALPHA
"1.0 - src_color.a", // INVSRCALPHA
"initial_ocol0.a", // DSTALPHA
"1.0 - initial_ocol0.a", // INVDSTALPHA
};
out.Write("\tfloat4 src_color = {};\n"
"\tfloat4 blend_src;",
"\tint4 blend_src;\n",
uid_data->useDstAlpha ? "ocol1" : "ocol0");
out.Write("\tblend_src.rgb = {}\n", blend_src_factor[uid_data->blend_src_factor]);
out.Write("\tblend_src.a = {}\n", blend_src_factor_alpha[uid_data->blend_src_factor_alpha]);
out.Write("\tfloat4 blend_dst;\n");
out.Write("\tblend_dst.rgb = {}\n", blend_dst_factor[uid_data->blend_dst_factor]);
out.Write("\tblend_dst.a = {}\n", blend_dst_factor_alpha[uid_data->blend_dst_factor_alpha]);

out.Write("\tfloat4 blend_result;\n");
out.Write("\tblend_src.rgb = int3(({}) * 255.0);\n",
blend_src_factor[uid_data->blend_src_factor]);
out.Write("\tblend_src.a = int(({}) * 255.0);\n",
blend_src_factor_alpha[uid_data->blend_src_factor_alpha]);
out.Write("\tint4 blend_dst;\n");
out.Write("\tblend_dst.rgb = int3(({}) * 255.0);\n",
blend_dst_factor[uid_data->blend_dst_factor]);
out.Write("\tblend_dst.a = int(({}) * 255.0);\n",
blend_dst_factor_alpha[uid_data->blend_dst_factor_alpha]);

// add MSB of factors to make their range 0 -> 256
out.Write("\tblend_src += blend_src >> 7;\n"
"\tblend_dst += blend_dst >> 7;\n");

out.Write("\tint4 blend_result;\n");
if (uid_data->blend_subtract)
{
out.Write("\tblend_result.rgb = initial_ocol0.rgb * blend_dst.rgb - ocol0.rgb * "
"blend_src.rgb;\n");
// TODO: libogc docs say src/dest factor are ignored in subtract mode
// (we may be handling this elsewhere?)
out.Write("\tblend_result.rgb = int3(initial_ocol0.rgb * 255.0) * blend_dst.rgb "
"- int3(ocol0.rgb * 255.0) * blend_src.rgb;\n");
}
else
{
out.Write(
"\tblend_result.rgb = initial_ocol0.rgb * blend_dst.rgb + ocol0.rgb * blend_src.rgb;\n");
out.Write("\tblend_result.rgb = int3(initial_ocol0.rgb * 255.0) * blend_dst.rgb "
"+ int3(ocol0.rgb * 255.0) * blend_src.rgb;\n");
}

if (uid_data->blend_subtract_alpha)
out.Write("\tblend_result.a = initial_ocol0.a * blend_dst.a - ocol0.a * blend_src.a;\n");
{
out.Write("\tblend_result.a = int(initial_ocol0.a * 255.0) * blend_dst.a "
"- int(ocol0.a * 255.0) * blend_src.a;\n");
}
else
out.Write("\tblend_result.a = initial_ocol0.a * blend_dst.a + ocol0.a * blend_src.a;\n");
{
out.Write("\tblend_result.a = int(initial_ocol0.a * 255.0) * blend_dst.a "
"+ int(ocol0.a * 255.0) * blend_src.a;\n");
}
}
else
{
out.Write("\tfloat4 blend_result = ocol0;\n");
out.Write("\tint4 blend_result = int4(ocol0 * 255.0) * 256;\n");
}

out.Write("\treal_ocol0 = blend_result;\n");
out.Write("\treal_ocol0 = float4(clamp(blend_result >> 8, 0, 255)) / 255.0;\n");
}
9 changes: 6 additions & 3 deletions Source/Core/VideoCommon/ShaderCache.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -641,13 +641,13 @@ static GXPipelineUid ApplyDriverBugs(const GXPipelineUid& in)

if (g_ActiveConfig.backend_info.bSupportsFramebufferFetch)
{
bool fbfetch_blend = false;
bool fbfetch_blend = blend.blendenable;
if ((DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DISCARD_WITH_EARLY_Z) ||
!g_ActiveConfig.backend_info.bSupportsEarlyZ) &&
ps->ztest == EmulatedZ::ForcedEarly)
{
ps->ztest = EmulatedZ::EarlyWithFBFetch;
fbfetch_blend |= static_cast<bool>(out.blending_state.blendenable);
fbfetch_blend |= static_cast<bool>(blend.blendenable);
ps->no_dual_src = true;
}
fbfetch_blend |= blend.logicopenable && !g_ActiveConfig.backend_info.bSupportsLogicOp;
Expand All @@ -670,7 +670,10 @@ static GXPipelineUid ApplyDriverBugs(const GXPipelineUid& in)
ps->blend_dst_factor_alpha = blend.dstfactoralpha;
ps->blend_subtract = blend.subtract;
ps->blend_subtract_alpha = blend.subtractAlpha;
blend.blendenable = false;
blend.hex = 0;
blend.colorupdate = in.blending_state.colorupdate.Value();
blend.alphaupdate = in.blending_state.alphaupdate.Value();
blend.usedualsrc = false;
}
}
}
Expand Down
28 changes: 18 additions & 10 deletions Source/Core/VideoCommon/UberShaderPixel.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1193,20 +1193,28 @@ ShaderCode GenPixelShader(APIType api_type, const ShaderHostConfig& host_config,
WriteSwitch(out, api_type, "blend_dst_factor", blendDstFactor, 4, true);
WriteSwitch(out, api_type, "blend_dst_factor_alpha", blendDstFactorAlpha, 4, true);

out.Write(
" float4 blend_result;\n"
" if (blend_subtract)\n"
" blend_result.rgb = initial_ocol0.rgb * blend_dst.rgb - ocol0.rgb * blend_src.rgb;\n"
" else\n"
" blend_result.rgb = initial_ocol0.rgb * blend_dst.rgb + ocol0.rgb * "
"blend_src.rgb;\n");
out.Write(" int4 blend_src_real = int4(blend_src * 255.0);\n"
" blend_src_real += blend_src_real >> 7;\n"
" int4 blend_src_col = int4(ocol0 * 255.0);\n"
" int4 blend_dst_real = int4(blend_dst * 255.0);\n"
" blend_dst_real += blend_dst_real >> 7;\n"
" int4 blend_dst_col = int4(initial_ocol0 * 255.0);\n"
" int4 blend_result;\n"
" if (blend_subtract)\n"
" blend_result.rgb = blend_dst_col.rgb * blend_dst_real.rgb "
"- blend_src_col.rgb * blend_src_real.rgb;\n"
" else\n"
" blend_result.rgb = blend_dst_col.rgb * blend_dst_real.rgb "
"+ blend_src_col.rgb * blend_src_real.rgb;\n");

out.Write(" if (blend_subtract_alpha)\n"
" blend_result.a = initial_ocol0.a * blend_dst.a - ocol0.a * blend_src.a;\n"
" blend_result.a = blend_dst_col.a * blend_dst_real.a "
"- blend_src_col.a * blend_src_real.a;\n"
" else\n"
" blend_result.a = initial_ocol0.a * blend_dst.a + ocol0.a * blend_src.a;\n");
" blend_result.a = blend_dst_col.a * blend_dst_real.a "
"+ blend_src_col.a * blend_src_real.a;\n");

out.Write(" real_ocol0 = blend_result;\n");
out.Write(" real_ocol0 = float4(clamp(blend_result >> 8, 0, 255)) / 255.0;\n");

out.Write(" }} else {{\n"
" real_ocol0 = ocol0;\n"
Expand Down