Diamond75/Panda3DS
1
0
Fork 0
mirror of https://github.com/wheremyfoodat/Panda3DS.git synced 2025-04-20 04:29:13 +12:00
Code Issues Projects Releases Packages Wiki Activity Actions

[Shader JIT] Implement returning

Browse source
This commit is contained in:
wheremyfoodat 2023-06-10 18:20:15 +03:00
parent 3c0ca97b55
commit 7e6f267462
1 changed files with 15 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

17
src/core/PICA/dynapica/shader_rec_emitter_x64.cpp
View file

@ -82,7 +82,17 @@ void ShaderEmitter::compileInstruction(const PICAShader& shaderUnit) {
// See if PC is a possible return PC and emit the proper code if so // See if PC is a possible return PC and emit the proper code if so
if (std::binary_search(returnPCs.begin(), returnPCs.end(), recompilerPC)) { if (std::binary_search(returnPCs.begin(), returnPCs.end(), recompilerPC)) {
int3(); // This is the offset we need to add to rsp to peek the next return address in the callstack
// Ie the PICA PC address which, when reached, will trigger a return
const auto stackOffsetForPC = isWindows() ? (16 + 32) : 16;
Label end;
// Check if return address == recompilerPC, ie if we should return
cmp(dword[rsp + stackOffsetForPC], recompilerPC);
jne(end); // If not, continue with uor lives
ret(); // If yes, return
L(end);
} }
// Fetch instruction and inc PC // Fetch instruction and inc PC
@ -627,11 +637,14 @@ void ShaderEmitter::recCALL(const PICAShader& shader, u32 instruction) {
// Push return PC as stack parameter. This is a decently fast solution and Citra does the same but we should probably switch to a proper PICA-like // Push return PC as stack parameter. This is a decently fast solution and Citra does the same but we should probably switch to a proper PICA-like
// Callstack, because it's not great to have an infinitely expanding call stack where popping from empty stack is undefined as hell // Callstack, because it's not great to have an infinitely expanding call stack where popping from empty stack is undefined as hell
push(qword, dest + num); push(qword, dest + num);
// Realign stack to 64 bits and allocate shadow space on windows // Realign stack to 64 bits and allocate 32 bytes of shadow space on windows
sub(rsp, isWindows() ? (8 + 32) : 8); sub(rsp, isWindows() ? (8 + 32) : 8);
// Call subroutine, Xbyak will update the label if it hasn't been initialized yet // Call subroutine, Xbyak will update the label if it hasn't been initialized yet
call(instructionLabels[dest]); call(instructionLabels[dest]);
// Fix up stack after returning. The 8 from before becomes a 16 because we're also skipping the qword we pushed
add(rsp, isWindows() ? (16 + 32) : 16);
} }
#endif #endif