Skip to content

SW: Fix fog calculation accuracy across all rendering paths by removing rounding bias + Add optional OpenCL fog processing pipeline #13212

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

Closed
wants to merge 9 commits into from
Closed

SW: Fix fog calculation accuracy across all rendering paths by removing rounding bias + Add optional OpenCL fog processing pipeline #13212

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
b49d87b
Initial plan
Copilot Aug 31, 2025
92f8f89
Implement fog accuracy fix by removing rounding from SW fog calculations
Copilot Aug 31, 2025
e1149a9
Fix fog calculation in C implementation (non-JIT path)
Copilot Aug 31, 2025
991f66b
Fix fog calculation in hardware renderer shaders (OpenGL, Vulkan, DX1…
Copilot Aug 31, 2025
6b41ccc
Add OpenCL fog processing pipeline infrastructure
Copilot Aug 31, 2025
2b50523
Complete OpenCL fog pipeline with shader integration and demo
Copilot Aug 31, 2025
29e6c20
Make C implementation consistent with ASM fog interpolation - use ler…
Copilot Aug 31, 2025
e783fdf
Fix build failures by making OpenCL dependency optional with proper c…
Copilot Aug 31, 2025
d6d55c6
Enable OpenCL in GitHub Actions Linux builds for fog pipeline testing
Copilot Sep 1, 2025
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
17 changes: 15 additions & 2 deletions pcsx2/GS/Renderers/SW/GSDrawScanlineCodeGenerator.all.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -185,6 +185,13 @@ void GSDrawScanlineCodeGenerator::modulate16(const XYm& a, const Operand& f, u8
}
}

void GSDrawScanlineCodeGenerator::modulate16_noround(const XYm& a, const Operand& f)
{
// Non-rounding equivalent of pmulhrsw: ((a << 1) * f) >> 16 = (a * f) >> 15
psllw(a, 1);
pmulhw(a, f);
}

void GSDrawScanlineCodeGenerator::lerp16(const XYm& a, const XYm& b, const XYm& f, u8 shift)
{
psubw(a, b);
Expand Down Expand Up @@ -2395,10 +2402,16 @@ void GSDrawScanlineCodeGenerator::Fog()
movdqa(xym1, _ga);

pbroadcastdLocal(tmp, _rip_global(frb));
lerp16(_rb, tmp, f, 0);
// Use non-rounding interpolation for fog (PS2 hardware doesn't round)
psubw(_rb, tmp);
modulate16_noround(_rb, f);
paddw(_rb, tmp);

pbroadcastdLocal(tmp, _rip_global(fga));
lerp16(_ga, tmp, f, 0);
// Use non-rounding interpolation for fog (PS2 hardware doesn't round)
psubw(_ga, tmp);
modulate16_noround(_ga, f);
paddw(_ga, tmp);

mix16(_ga, xym1, xym0);
}
Expand Down
1 change: 1 addition & 0 deletions pcsx2/GS/Renderers/SW/GSDrawScanlineCodeGenerator.all.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -99,6 +99,7 @@ class GSDrawScanlineCodeGenerator : public GSNewCodeGenerator
/// Broadcast a 32-bit GPR to a vector register
void broadcastGPRToVec(const XYm& vec, const Xbyak::Reg32& gpr);
void modulate16(const XYm& a, const Xbyak::Operand& f, u8 shift);
void modulate16_noround(const XYm& a, const Xbyak::Operand& f);
void lerp16(const XYm& a, const XYm& b, const XYm& f, u8 shift);
void lerp16_4(const XYm& a, const XYm& b, const XYm& f);
void mix16(const XYm& a, const XYm& b, const XYm& temp);
Expand Down
20 changes: 17 additions & 3 deletions pcsx2/GS/Renderers/SW/GSDrawScanlineCodeGenerator.arm64.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1659,9 +1659,16 @@ void GSDrawScanlineCodeGenerator::Fog()
armAsm->Dup(_vscratch2.V4S(), _global_fga);
armAsm->Mov(v1, v6);

lerp16(v5, _vscratch, _temp_f, 0);

lerp16(v6, _vscratch2, _temp_f, 0);
// Use non-rounding interpolation for fog (PS2 hardware doesn't round)
armAsm->Sub(v5.V8H(), v5.V8H(), _vscratch.V8H());
modulate16_noround(v5, _temp_f);
armAsm->Add(v5.V8H(), v5.V8H(), _vscratch.V8H());

// Use non-rounding interpolation for fog (PS2 hardware doesn't round)
armAsm->Sub(v6.V8H(), v6.V8H(), _vscratch2.V8H());
modulate16_noround(v6, _temp_f);
armAsm->Add(v6.V8H(), v6.V8H(), _vscratch2.V8H());

mix16(v6, v1, v0);
}

Expand Down Expand Up @@ -2362,6 +2369,13 @@ void GSDrawScanlineCodeGenerator::modulate16(const VRegister& d, const VRegister
armAsm->Sshr(a.V8H(), a.V8H(), 1);
}

void GSDrawScanlineCodeGenerator::modulate16_noround(const VRegister& a, const VRegister& f)
{
// Non-rounding equivalent of sqrdmulh: use sqdmulh directly after left shift
armAsm->Shl(a.V8H(), a.V8H(), 1);
armAsm->Sqdmulh(a.V8H(), a.V8H(), f.V8H());
}

void GSDrawScanlineCodeGenerator::lerp16(const VRegister& a, const VRegister& b, const VRegister& f, u8 shift)
{
armAsm->Sub(a.V8H(), a.V8H(), b.V8H());
Expand Down
1 change: 1 addition & 0 deletions pcsx2/GS/Renderers/SW/GSDrawScanlineCodeGenerator.arm64.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,7 @@ class GSDrawScanlineCodeGenerator

void modulate16(const vixl::aarch64::VRegister& d, const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& f, u8 shift);
void modulate16(const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& f, u8 shift);
void modulate16_noround(const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& f);
void lerp16(const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& b, const vixl::aarch64::VRegister& f, u8 shift);
void lerp16_4(const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& b, const vixl::aarch64::VRegister& f);
void mix16(const vixl::aarch64::VRegister& a, const vixl::aarch64::VRegister& b, const vixl::aarch64::VRegister& temp);
Expand Down