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

Merge branch 'master' into ir

Browse source
This commit is contained in:
wheremyfoodat 2023-07-21 15:19:57 +03:00
commit d470a8c8d3
46 changed files with 2206 additions and 1206 deletions
Download patch file Download diff file Expand all files Collapse all files

17
src/config.cpp
View file

@ -1,6 +1,7 @@
#include "config.hpp"
#include <fstream>
#include <string>
#include "helpers.hpp"
#include "toml.hpp"
@ -9,6 +10,8 @@
// We are legally allowed, as per the author's wish, to use the above code without any licensing restrictions
// However we still want to follow the license as closely as possible and offer the proper attributions.
EmulatorConfig::EmulatorConfig(const std::filesystem::path& path) { load(path); }
void EmulatorConfig::load(const std::filesystem::path& path) {
// If the configuration file does not exist, create it and return
std::error_code error;
@ -31,6 +34,17 @@ void EmulatorConfig::load(const std::filesystem::path& path) {
if (gpuResult.is_ok()) {
auto gpu = gpuResult.unwrap();
// Get renderer
auto rendererName = toml::find_or<std::string>(gpu, "Renderer", "OpenGL");
auto configRendererType = Renderer::typeFromString(rendererName);
if (configRendererType.has_value()) {
rendererType = configRendererType.value();
} else {
Helpers::warn("Invalid renderer specified: %s\n", rendererName.c_str());
rendererType = RendererType::OpenGL;
}
shaderJitEnabled = toml::find_or<toml::boolean>(gpu, "EnableShaderJIT", false);
}
}
@ -43,7 +57,7 @@ void EmulatorConfig::save(const std::filesystem::path& path) {
if (std::filesystem::exists(path, error)) {
try {
data = toml::parse<toml::preserve_comments>(path);
} catch (std::exception& ex) {
} catch (const std::exception& ex) {
Helpers::warn("Exception trying to parse config file. Exception: %s\n", ex.what());
return;
}
@ -55,6 +69,7 @@ void EmulatorConfig::save(const std::filesystem::path& path) {
}
data["GPU"]["EnableShaderJIT"] = shaderJitEnabled;
data["GPU"]["Renderer"] = std::string(Renderer::typeToString(rendererType));
std::ofstream file(path, std::ios::out);
file << data;

285
src/core/PICA/dynapica/shader_rec_emitter_x64.cpp
View file

@ -61,11 +61,14 @@ void ShaderEmitter::compile(const PICAShader& shaderUnit) {
// Tail call to shader code entrypoint
jmp(arg2);
align(16);
// Scan the shader code for call instructions and add them to the list of possible return PCs. We need to do this because the PICA callstack works
// Pretty weirdly
scanForCalls(shaderUnit);
// Scan the code for call, exp2, log2, etc instructions which need some special care
// After that, emit exp2 and log2 functions if the corresponding instructions are present
scanCode(shaderUnit);
if (codeHasExp2) exp2Func = emitExp2Func();
if (codeHasLog2) log2Func = emitLog2Func();
align(16);
// Compile every instruction in the shader
// This sounds horrible but the PICA instruction memory is tiny, and most of the time it's padded wtih nops that compile to nothing
recompilerPC = 0;
@ -73,17 +76,23 @@ void ShaderEmitter::compile(const PICAShader& shaderUnit) {
compileUntil(shaderUnit, PICAShader::maxInstructionCount);
}
void ShaderEmitter::scanForCalls(const PICAShader& shaderUnit) {
void ShaderEmitter::scanCode(const PICAShader& shaderUnit) {
returnPCs.clear();
for (u32 i = 0; i < PICAShader::maxInstructionCount; i++) {
const u32 instruction = shaderUnit.loadedShader[i];
const u32 opcode = instruction >> 26;
if (isCall(instruction)) {
const u32 num = instruction & 0xff;
const u32 dest = getBits<10, 12>(instruction);
const u32 returnPC = num + dest; // Add them to get the return PC
returnPCs.push_back(returnPC);
} else if (opcode == ShaderOpcodes::EX2) {
codeHasExp2 = true;
} else if (opcode == ShaderOpcodes::LG2) {
codeHasLog2 = true;
}
}
@ -877,7 +886,6 @@ void ShaderEmitter::recLOOP(const PICAShader& shader, u32 instruction) {
loopLevel--;
}
// SSE does not have a log2 instruction so we temporarily emulate this using x87 FPU
void ShaderEmitter::recLG2(const PICAShader& shader, u32 instruction) {
const u32 operandDescriptor = shader.operandDescriptors[instruction & 0x7f];
const u32 src = getBits<12, 7>(instruction);
@ -885,30 +893,16 @@ void ShaderEmitter::recLG2(const PICAShader& shader, u32 instruction) {
const u32 dest = getBits<21, 5>(instruction);
const u32 writeMask = getBits<0, 4>(operandDescriptor);
// Load swizzled source, push 1.0 to the x87 stack
loadRegister<1>(src1_xmm, shader, src, idx, operandDescriptor);
fld1();
// Push source to the x87 stack
movd(eax, src1_xmm);
push(rax);
fld(dword[rsp]);
// Perform log2, load result to src1_xmm, write it back and undo the previous push rax
fyl2x();
fstp(dword[rsp]);
movss(src1_xmm, dword[rsp]);
add(rsp, 8);
// If we only write back the x component to the result, we needn't perform a shuffle to do res = res.xxxx
// Otherwise we do
call(log2Func); // Result is output in src1_xmm
if (writeMask != 0x8) { // Copy bottom lane to all lanes if we're not simply writing back x
shufps(src1_xmm, src1_xmm, 0); // src1_xmm = src1_xmm.xxxx
}
storeRegister(src1_xmm, shader, dest, operandDescriptor);
}
// SSE does not have an exp2 instruction so we temporarily emulate this using x87 FPU
void ShaderEmitter::recEX2(const PICAShader& shader, u32 instruction) {
const u32 operandDescriptor = shader.operandDescriptors[instruction & 0x7f];
const u32 src = getBits<12, 7>(instruction);
@ -917,31 +911,12 @@ void ShaderEmitter::recEX2(const PICAShader& shader, u32 instruction) {
const u32 writeMask = getBits<0, 4>(operandDescriptor);
loadRegister<1>(src1_xmm, shader, src, idx, operandDescriptor);
call(exp2Func); // Result is output in src1_xmm
// Push source to the x87 stack, then do some insane compiler-generated x87 math
movd(eax, src1_xmm);
push(rax);
fld(dword[rsp]);
fld(st0);
frndint();
fsub(st1, st0);
fxch(st1);
f2xm1();
fadd(dword[rip + onesVector]);
fscale();
// Load result to src1_xmm, write it back and undo the previous push rax
fstp(st1);
fstp(dword[rsp]);
movss(src1_xmm, dword[rsp]);
add(rsp, 8);
// If we only write back the x component to the result, we needn't perform a shuffle to do res = res.xxxx
// Otherwise we do
if (writeMask != 0x8) { // Copy bottom lane to all lanes if we're not simply writing back x
shufps(src1_xmm, src1_xmm, 0); // src1_xmm = src1_xmm.xxxx
}
storeRegister(src1_xmm, shader, dest, operandDescriptor);
}
@ -962,6 +937,228 @@ void ShaderEmitter::printLog(const PICAShader& shaderUnit) {
printf("cmp: (%d, %d)\n", shaderUnit.cmpRegister[0], shaderUnit.cmpRegister[1]);
}
// For EXP2/LOG2, we have permission to adjust and relicense the SSE implementation from Citra for this project from the original authors
// So we do it since EXP2/LOG2 are pretty terrible to implement.
// ABI: Input is in the bottom bits of src1_xmm, same for output. If the result needs swizzling, the caller must handle it
// Assume src1, src2, scratch1, scratch2, eax, edx all thrashed
Xbyak::Label ShaderEmitter::emitLog2Func() {
Xbyak::Label subroutine;
// This code uses the fact that log2(float) = log2(2^exponent * mantissa)
// = log2(2^exponent) + log2(mantissa) = exponent + log2(mantissa) where mantissa has a limited range of values
// https://stackoverflow.com/a/45787548
// SSE does not have a log instruction, thus we must approximate.
// We perform this approximation first performing a range reduction into the range [1.0, 2.0).
// A minimax polynomial which was fit for the function log2(x) / (x - 1) is then evaluated.
// We multiply the result by (x - 1) then restore the result into the appropriate range.
// Coefficients for the minimax polynomial.
// f(x) computes approximately log2(x) / (x - 1).
// f(x) = c4 + x * (c3 + x * (c2 + x * (c1 + x * c0)).
// We align the table of coefficients to 64 bytes, so that the whole thing will fit in 1 cache line
align(64);
const void* c0 = getCurr();
dd(0x3d74552f);
const void* c1 = getCurr();
dd(0xbeee7397);
const void* c2 = getCurr();
dd(0x3fbd96dd);
const void* c3 = getCurr();
dd(0xc02153f6);
const void* c4 = getCurr();
dd(0x4038d96c);
align(16);
const void* negative_infinity_vector = getCurr();
dd(0xff800000);
dd(0xff800000);
dd(0xff800000);
dd(0xff800000);
const void* default_qnan_vector = getCurr();
dd(0x7fc00000);
dd(0x7fc00000);
dd(0x7fc00000);
dd(0x7fc00000);
Xbyak::Label inputIsNan, inputIsZero, inputOutOfRange;
align(16);
L(inputOutOfRange);
je(inputIsZero);
movaps(src1_xmm, xword[rip + default_qnan_vector]);
ret();
L(inputIsZero);
movaps(src1_xmm, xword[rip + negative_infinity_vector]);
ret();
align(16);
L(subroutine);
// Here we handle edge cases: input in {NaN, 0, -Inf, Negative}.
xorps(scratch1, scratch1);
ucomiss(scratch1, src1_xmm);
jp(inputIsNan);
jae(inputOutOfRange);
// Split input: SRC1=MANT[1,2) SCRATCH2=Exponent
if (cpuCaps.has(Cpu::tAVX512F | Cpu::tAVX512VL)) {
vgetexpss(scratch2, src1_xmm, src1_xmm);
vgetmantss(src1_xmm, src1_xmm, src1_xmm, 0);
} else {
movd(eax, src1_xmm);
mov(edx, eax);
and_(eax, 0x7f800000);
and_(edx, 0x007fffff);
or_(edx, 0x3f800000);
movd(src1_xmm, edx);
// SRC1 now contains the mantissa of the input.
shr(eax, 23);
sub(eax, 0x7f);
cvtsi2ss(scratch2, eax);
// scratch2 now contains the exponent of the input.
}
movss(scratch1, xword[rip + c0]);
// Complete computation of polynomial
if (haveFMA3) {
vfmadd213ss(scratch1, src1_xmm, xword[rip + c1]);
vfmadd213ss(scratch1, src1_xmm, xword[rip + c2]);
vfmadd213ss(scratch1, src1_xmm, xword[rip + c3]);
vfmadd213ss(scratch1, src1_xmm, xword[rip + c4]);
subss(src1_xmm, dword[rip + onesVector]);
vfmadd231ss(scratch2, scratch1, src1_xmm);
} else {
mulss(scratch1, src1_xmm);
addss(scratch1, xword[rip + c1]);
mulss(scratch1, src1_xmm);
addss(scratch1, xword[rip + c2]);
mulss(scratch1, src1_xmm);
addss(scratch1, xword[rip + c3]);
mulss(scratch1, src1_xmm);
subss(src1_xmm, dword[rip + onesVector]);
addss(scratch1, xword[rip + c4]);
mulss(scratch1, src1_xmm);
addss(scratch2, scratch1);
}
xorps(src1_xmm, src1_xmm); // break dependency chain
movss(src1_xmm, scratch2);
L(inputIsNan);
ret();
return subroutine;
}
Xbyak::Label ShaderEmitter::emitExp2Func() {
Xbyak::Label subroutine;
// SSE does not have a exp instruction, thus we must approximate.
// We perform this approximation first performaing a range reduction into the range [-0.5, 0.5).
// A minimax polynomial which was fit for the function exp2(x) is then evaluated.
// We then restore the result into the appropriate range.
// Similarly to log2, we align our literal pool to 64 bytes to make sure the whole thing fits in 1 cache line
align(64);
const void* input_max = getCurr();
dd(0x43010000);
const void* input_min = getCurr();
dd(0xc2fdffff);
const void* c0 = getCurr();
dd(0x3c5dbe69);
const void* half = getCurr();
dd(0x3f000000);
const void* c1 = getCurr();
dd(0x3d5509f9);
const void* c2 = getCurr();
dd(0x3e773cc5);
const void* c3 = getCurr();
dd(0x3f3168b3);
const void* c4 = getCurr();
dd(0x3f800016);
Xbyak::Label retLabel;
align(16);
L(subroutine);
// Handle edge cases
ucomiss(src1_xmm, src1_xmm);
jp(retLabel);
// Decompose input:
// SCRATCH=2^round(input)
// SRC1=input-round(input) [-0.5, 0.5)
if (cpuCaps.has(Cpu::tAVX512F | Cpu::tAVX512VL)) {
// Cheat a bit and store ones in src2 since the register is unused
vmovaps(src2_xmm, xword[rip + onesVector]);
// input - 0.5
vsubss(scratch1, src1_xmm, xword[rip + half]);
// trunc(input - 0.5)
vrndscaless(scratch2, scratch1, scratch1, _MM_FROUND_TRUNC);
// SCRATCH = 1 * 2^(trunc(input - 0.5))
vscalefss(scratch1, src2_xmm, scratch2);
// SRC1 = input-trunc(input - 0.5)
vsubss(src1_xmm, src1_xmm, scratch2);
} else {
// Clamp to maximum range since we shift the value directly into the exponent.
minss(src1_xmm, xword[rip + input_max]);
maxss(src1_xmm, xword[rip + input_min]);
if (cpuCaps.has(Cpu::tAVX)) {
vsubss(scratch1, src1_xmm, xword[rip + half]);
} else {
movss(scratch1, src1_xmm);
subss(scratch1, xword[rip + half]);
}
if (cpuCaps.has(Cpu::tSSE41)) {
roundss(scratch1, scratch1, _MM_FROUND_TRUNC);
cvtss2si(eax, scratch1);
} else {
cvtss2si(eax, scratch1);
cvtsi2ss(scratch1, eax);
}
// SCRATCH now contains input rounded to the nearest integer.
add(eax, 0x7f);
subss(src1_xmm, scratch1);
// SRC1 contains input - round(input), which is in [-0.5, 0.5).
shl(eax, 23);
movd(scratch1, eax);
// SCRATCH contains 2^(round(input)).
}
// Complete computation of polynomial.
movss(scratch2, xword[rip + c0]);
if (haveFMA3) {
vfmadd213ss(scratch2, src1_xmm, xword[rip + c1]);
vfmadd213ss(scratch2, src1_xmm, xword[rip + c2]);
vfmadd213ss(scratch2, src1_xmm, xword[rip + c3]);
vfmadd213ss(src1_xmm, scratch2, xword[rip + c4]);
} else {
mulss(scratch2, src1_xmm);
addss(scratch2, xword[rip + c1]);
mulss(scratch2, src1_xmm);
addss(scratch2, xword[rip + c2]);
mulss(scratch2, src1_xmm);
addss(scratch2, xword[rip + c3]);
mulss(src1_xmm, scratch2);
addss(src1_xmm, xword[rip + c4]);
}
mulss(src1_xmm, scratch1);
L(retLabel);
ret();
return subroutine;
}
// As we mentioned above, this function is uber slow because we don't expect the shader JIT to call HLL functions in real scenarios
// Aside from debugging code. So we don't care for this function to be performant or anything of the like. It is quick and dirty
// And mostly meant to be used for generating logs to diff the JIT and interpreter

101
src/core/PICA/gpu.cpp
View file

@ -2,19 +2,45 @@
#include <array>
#include <bitset>
#include <cstdio>
#include <cstddef>
#include <cstdio>
#include "PICA/float_types.hpp"
#include "PICA/regs.hpp"
#include "renderer_null/renderer_null.hpp"
#ifdef PANDA3DS_ENABLE_OPENGL
#include "renderer_gl/renderer_gl.hpp"
#endif
using namespace Floats;
// Note: For when we have multiple backends, the GL state manager can stay here and have the constructor for the Vulkan-or-whatever renderer ignore it
// Thus, our GLStateManager being here does not negatively impact renderer-agnosticness
GPU::GPU(Memory& mem, GLStateManager& gl, EmulatorConfig& config) : mem(mem), renderer(*this, gl, regs), config(config) {
GPU::GPU(Memory& mem, EmulatorConfig& config) : mem(mem), config(config) {
vram = new u8[vramSize];
mem.setVRAM(vram); // Give the bus a pointer to our VRAM
mem.setVRAM(vram); // Give the bus a pointer to our VRAM
switch (config.rendererType) {
case RendererType::Null: {
renderer.reset(new RendererNull(*this, regs));
break;
}
#ifdef PANDA3DS_ENABLE_OPENGL
case RendererType::OpenGL: {
renderer.reset(new RendererGL(*this, regs));
break;
}
#endif
case RendererType::Vulkan: {
Helpers::panic("Vulkan is not supported yet, please pick another renderer");
}
default: {
Helpers::panic("Rendering backend not supported: %s", Renderer::typeToString(config.rendererType));
break;
}
}
}
void GPU::reset() {
@ -41,7 +67,7 @@ void GPU::reset() {
e.config2 = 0;
}
renderer.reset();
renderer->reset();
}
// Call the correct version of drawArrays based on whether this is an indexed draw (first template parameter)
@ -73,15 +99,14 @@ void GPU::drawArrays() {
// Base address for vertex attributes
// The vertex base is always on a quadword boundary because the PICA does weird alignment shit any time possible
const u32 vertexBase = ((regs[PICA::InternalRegs::VertexAttribLoc] >> 1) & 0xfffffff) * 16;
const u32 vertexCount = regs[PICA::InternalRegs::VertexCountReg]; // Total # of vertices to transfer
const u32 vertexCount = regs[PICA::InternalRegs::VertexCountReg]; // Total # of vertices to transfer
// Configures the type of primitive and the number of vertex shader outputs
const u32 primConfig = regs[PICA::InternalRegs::PrimitiveConfig];
const PICA::PrimType primType = static_cast<PICA::PrimType>(Helpers::getBits<8, 2>(primConfig));
if (vertexCount > Renderer::vertexBufferSize) Helpers::panic("[PICA] vertexCount > vertexBufferSize");
if ((primType == PICA::PrimType::TriangleList && vertexCount % 3) ||
(primType == PICA::PrimType::TriangleStrip && vertexCount < 3) ||
if ((primType == PICA::PrimType::TriangleList && vertexCount % 3) || (primType == PICA::PrimType::TriangleStrip && vertexCount < 3) ||
(primType == PICA::PrimType::TriangleFan && vertexCount < 3)) {
Helpers::panic("Invalid vertex count for primitive. Type: %d, vert count: %d\n", primType, vertexCount);
}
@ -89,10 +114,10 @@ void GPU::drawArrays() {
// Get the configuration for the index buffer, used only for indexed drawing
u32 indexBufferConfig = regs[PICA::InternalRegs::IndexBufferConfig];
u32 indexBufferPointer = vertexBase + (indexBufferConfig & 0xfffffff);
bool shortIndex = Helpers::getBit<31>(indexBufferConfig); // Indicates whether vert indices are 16-bit or 8-bit
bool shortIndex = Helpers::getBit<31>(indexBufferConfig); // Indicates whether vert indices are 16-bit or 8-bit
// Stuff the global attribute config registers in one u64 to make attr parsing easier
// TODO: Cache this when the vertex attribute format registers are written to
// TODO: Cache this when the vertex attribute format registers are written to
u64 vertexCfg = u64(regs[PICA::InternalRegs::AttribFormatLow]) | (u64(regs[PICA::InternalRegs::AttribFormatHigh]) << 32);
if constexpr (!indexed) {
@ -111,24 +136,24 @@ void GPU::drawArrays() {
constexpr size_t vertexCacheSize = 64;
struct {
std::bitset<vertexCacheSize> validBits{0}; // Shows which tags are valid. If the corresponding bit is 1, then there's an entry
std::array<u32, vertexCacheSize> ids; // IDs (ie indices of the cached vertices in the 3DS vertex buffer)
std::array<u32, vertexCacheSize> bufferPositions; // Positions of the cached vertices in our own vertex buffer
std::bitset<vertexCacheSize> validBits{0}; // Shows which tags are valid. If the corresponding bit is 1, then there's an entry
std::array<u32, vertexCacheSize> ids; // IDs (ie indices of the cached vertices in the 3DS vertex buffer)
std::array<u32, vertexCacheSize> bufferPositions; // Positions of the cached vertices in our own vertex buffer
} vertexCache;
for (u32 i = 0; i < vertexCount; i++) {
u32 vertexIndex; // Index of the vertex in the VBO for indexed rendering
u32 vertexIndex; // Index of the vertex in the VBO for indexed rendering
if constexpr (!indexed) {
vertexIndex = i + regs[PICA::InternalRegs::VertexOffsetReg];
} else {
if (shortIndex) {
auto ptr = getPointerPhys<u16>(indexBufferPointer);
vertexIndex = *ptr; // TODO: This is very unsafe
vertexIndex = *ptr; // TODO: This is very unsafe
indexBufferPointer += 2;
} else {
auto ptr = getPointerPhys<u8>(indexBufferPointer);
vertexIndex = *ptr; // TODO: This is also very unsafe
vertexIndex = *ptr; // TODO: This is also very unsafe
indexBufferPointer += 1;
}
}
@ -152,22 +177,22 @@ void GPU::drawArrays() {
}
int attrCount = 0;
int buffer = 0; // Vertex buffer index for non-fixed attributes
int buffer = 0; // Vertex buffer index for non-fixed attributes
while (attrCount < totalAttribCount) {
// Check if attribute is fixed or not
if (fixedAttribMask & (1 << attrCount)) { // Fixed attribute
vec4f& fixedAttr = shaderUnit.vs.fixedAttributes[attrCount]; // TODO: Is this how it works?
if (fixedAttribMask & (1 << attrCount)) { // Fixed attribute
vec4f& fixedAttr = shaderUnit.vs.fixedAttributes[attrCount]; // TODO: Is this how it works?
vec4f& inputAttr = currentAttributes[attrCount];
std::memcpy(&inputAttr, &fixedAttr, sizeof(vec4f)); // Copy fixed attr to input attr
std::memcpy(&inputAttr, &fixedAttr, sizeof(vec4f)); // Copy fixed attr to input attr
attrCount++;
} else { // Non-fixed attribute
auto& attr = attributeInfo[buffer]; // Get information for this attribute
u64 attrCfg = attr.getConfigFull(); // Get config1 | (config2 << 32)
} else { // Non-fixed attribute
auto& attr = attributeInfo[buffer]; // Get information for this attribute
u64 attrCfg = attr.getConfigFull(); // Get config1 | (config2 << 32)
u32 attrAddress = vertexBase + attr.offset + (vertexIndex * attr.size);
for (int j = 0; j < attr.componentCount; j++) {
uint index = (attrCfg >> (j * 4)) & 0xf; // Get index of attribute in vertexCfg
uint index = (attrCfg >> (j * 4)) & 0xf; // Get index of attribute in vertexCfg
// Vertex attributes used as padding
// 12, 13, 14 and 15 are equivalent to 4, 8, 12 and 16 bytes of padding respectively
@ -179,15 +204,15 @@ void GPU::drawArrays() {
}
u32 attribInfo = (vertexCfg >> (index * 4)) & 0xf;
u32 attribType = attribInfo & 0x3; // Type of attribute(sbyte/ubyte/short/float)
u32 size = (attribInfo >> 2) + 1; // Total number of components
u32 attribType = attribInfo & 0x3; // Type of attribute(sbyte/ubyte/short/float)
u32 size = (attribInfo >> 2) + 1; // Total number of components
//printf("vertex_attribute_strides[%d] = %d\n", attrCount, attr.size);
// printf("vertex_attribute_strides[%d] = %d\n", attrCount, attr.size);
vec4f& attribute = currentAttributes[attrCount];
uint component; // Current component
uint component; // Current component
switch (attribType) {
case 0: { // Signed byte
case 0: { // Signed byte
s8* ptr = getPointerPhys<s8>(attrAddress);
for (component = 0; component < size; component++) {
float val = static_cast<float>(*ptr++);
@ -197,7 +222,7 @@ void GPU::drawArrays() {
break;
}
case 1: { // Unsigned byte
case 1: { // Unsigned byte
u8* ptr = getPointerPhys<u8>(attrAddress);
for (component = 0; component < size; component++) {
float val = static_cast<float>(*ptr++);
@ -207,7 +232,7 @@ void GPU::drawArrays() {
break;
}
case 2: { // Short
case 2: { // Short
s16* ptr = getPointerPhys<s16>(attrAddress);
for (component = 0; component < size; component++) {
float val = static_cast<float>(*ptr++);
@ -217,7 +242,7 @@ void GPU::drawArrays() {
break;
}
case 3: { // Float
case 3: { // Float
float* ptr = getPointerPhys<float>(attrAddress);
for (component = 0; component < size; component++) {
float val = *ptr++;
@ -251,8 +276,8 @@ void GPU::drawArrays() {
const u32 mapping = (inputAttrCfg >> (j * 4)) & 0xf;
std::memcpy(&shaderUnit.vs.inputs[mapping], &currentAttributes[j], sizeof(vec4f));
}
if constexpr (useShaderJIT) {
if constexpr (useShaderJIT) {
shaderJIT.run(shaderUnit.vs);
} else {
shaderUnit.vs.run();
@ -264,14 +289,14 @@ void GPU::drawArrays() {
for (int i = 0; i < totalShaderOutputs; i++) {
const u32 config = regs[PICA::InternalRegs::ShaderOutmap0 + i];
for (int j = 0; j < 4; j++) { // pls unroll
for (int j = 0; j < 4; j++) { // pls unroll
const u32 mapping = (config >> (j * 8)) & 0x1F;
out.raw[mapping] = shaderUnit.vs.outputs[i][j];
}
}
}
renderer.drawVertices(primType, std::span(vertices).first(vertexCount));
renderer->drawVertices(primType, std::span(vertices).first(vertexCount));
}
PICA::Vertex GPU::getImmediateModeVertex() {
@ -289,7 +314,9 @@ PICA::Vertex GPU::getImmediateModeVertex() {
std::memcpy(&v.s.colour, &shaderUnit.vs.outputs[1], sizeof(vec4f));
std::memcpy(&v.s.texcoord0, &shaderUnit.vs.outputs[2], 2 * sizeof(f24));
printf("(x, y, z, w) = (%f, %f, %f, %f)\n", (double)v.s.positions[0], (double)v.s.positions[1], (double)v.s.positions[2], (double)v.s.positions[3]);
printf(
"(x, y, z, w) = (%f, %f, %f, %f)\n", (double)v.s.positions[0], (double)v.s.positions[1], (double)v.s.positions[2], (double)v.s.positions[3]
);
printf("(r, g, b, a) = (%f, %f, %f, %f)\n", (double)v.s.colour[0], (double)v.s.colour[1], (double)v.s.colour[2], (double)v.s.colour[3]);
printf("(u, v ) = (%f, %f)\n", (double)v.s.texcoord0[0], (double)v.s.texcoord0[1]);

122
src/core/PICA/regs.cpp
View file

@ -1,11 +1,12 @@
#include "PICA/gpu.hpp"
#include "PICA/regs.hpp"
#include "PICA/gpu.hpp"
using namespace Floats;
using namespace Helpers;
u32 GPU::readReg(u32 address) {
if (address >= 0x1EF01000 && address < 0x1EF01C00) { // Internal registers
if (address >= 0x1EF01000 && address < 0x1EF01C00) { // Internal registers
const u32 index = (address - 0x1EF01000) / sizeof(u32);
return readInternalReg(index);
} else {
@ -15,7 +16,7 @@ u32 GPU::readReg(u32 address) {
}
void GPU::writeReg(u32 address, u32 value) {
if (address >= 0x1EF01000 && address < 0x1EF01C00) { // Internal registers
if (address >= 0x1EF01000 && address < 0x1EF01C00) { // Internal registers
const u32 index = (address - 0x1EF01000) / sizeof(u32);
writeInternalReg(index, value, 0xffffffff);
} else {
@ -59,7 +60,7 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
}
u32 currentValue = regs[index];
u32 newValue = (currentValue & ~mask) | (value & mask); // Only overwrite the bits specified by "mask"
u32 newValue = (currentValue & ~mask) | (value & mask); // Only overwrite the bits specified by "mask"
regs[index] = newValue;
// TODO: Figure out if things like the shader index use the unmasked value or the masked one
@ -74,38 +75,38 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
break;
case AttribFormatHigh:
totalAttribCount = (value >> 28) + 1; // Total number of vertex attributes
fixedAttribMask = getBits<16, 12>(value); // Determines which vertex attributes are fixed for all vertices
totalAttribCount = (value >> 28) + 1; // Total number of vertex attributes
fixedAttribMask = getBits<16, 12>(value); // Determines which vertex attributes are fixed for all vertices
break;
case ColourBufferLoc: {
u32 loc = (value & 0x0fffffff) << 3;
renderer.setColourBufferLoc(loc);
renderer->setColourBufferLoc(loc);
break;
};
case ColourBufferFormat: {
u32 format = getBits<16, 3>(value);
renderer.setColourFormat(static_cast<PICA::ColorFmt>(format));
renderer->setColourFormat(static_cast<PICA::ColorFmt>(format));
break;
}
case DepthBufferLoc: {
u32 loc = (value & 0x0fffffff) << 3;
renderer.setDepthBufferLoc(loc);
renderer->setDepthBufferLoc(loc);
break;
}
case DepthBufferFormat: {
u32 format = value & 0x3;
renderer.setDepthFormat(static_cast<PICA::DepthFmt>(format));
renderer->setDepthFormat(static_cast<PICA::DepthFmt>(format));
break;
}
case FramebufferSize: {
const u32 width = value & 0x7ff;
const u32 height = getBits<12, 10>(value) + 1;
renderer.setFBSize(width, height);
renderer->setFBSize(width, height);
break;
}
@ -116,7 +117,7 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
case LightingLUTData4:
case LightingLUTData5:
case LightingLUTData6:
case LightingLUTData7:{
case LightingLUTData7: {
const uint32_t index = regs[LightingLUTIndex]; // Get full LUT index register
const uint32_t lutID = getBits<8, 5>(index); // Get which LUT we're actually writing to
uint32_t lutIndex = getBits<0, 8>(index); // And get the index inside the LUT we're writing to
@ -133,15 +134,22 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
break;
}
case VertexFloatUniformIndex:
case VertexFloatUniformIndex: {
shaderUnit.vs.setFloatUniformIndex(value);
break;
}
case VertexFloatUniformData0: case VertexFloatUniformData1: case VertexFloatUniformData2:
case VertexFloatUniformData3: case VertexFloatUniformData4: case VertexFloatUniformData5:
case VertexFloatUniformData6: case VertexFloatUniformData7:
case VertexFloatUniformData0:
case VertexFloatUniformData1:
case VertexFloatUniformData2:
case VertexFloatUniformData3:
case VertexFloatUniformData4:
case VertexFloatUniformData5:
case VertexFloatUniformData6:
case VertexFloatUniformData7: {
shaderUnit.vs.uploadFloatUniform(value);
break;
}
case FixedAttribIndex:
fixedAttribCount = 0;
@ -162,7 +170,9 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
}
break;
case FixedAttribData0: case FixedAttribData1: case FixedAttribData2:
case FixedAttribData0:
case FixedAttribData1:
case FixedAttribData2:
fixedAttrBuff[fixedAttribCount++] = value;
if (fixedAttribCount == 3) {
@ -170,15 +180,15 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
vec4f attr;
// These are stored in the reverse order anyone would expect them to be in
attr.x() = f24::fromRaw(fixedAttrBuff[2] & 0xffffff);
attr.y() = f24::fromRaw(((fixedAttrBuff[1] & 0xffff) << 8) | (fixedAttrBuff[2] >> 24));
attr.z() = f24::fromRaw(((fixedAttrBuff[0] & 0xff) << 16) | (fixedAttrBuff[1] >> 16));
attr.w() = f24::fromRaw(fixedAttrBuff[0] >> 8);
attr[0] = f24::fromRaw(fixedAttrBuff[2] & 0xffffff);
attr[1] = f24::fromRaw(((fixedAttrBuff[1] & 0xffff) << 8) | (fixedAttrBuff[2] >> 24));
attr[2] = f24::fromRaw(((fixedAttrBuff[0] & 0xff) << 16) | (fixedAttrBuff[1] >> 16));
attr[3] = f24::fromRaw(fixedAttrBuff[0] >> 8);
// If the fixed attribute index is < 12, we're just writing to one of the fixed attributes
if (fixedAttribIndex < 12) [[likely]] {
shaderUnit.vs.fixedAttributes[fixedAttribIndex++] = attr;
} else if (fixedAttribIndex == 15) { // Otherwise if it's 15, we're submitting an immediate mode vertex
} else if (fixedAttribIndex == 15) { // Otherwise if it's 15, we're submitting an immediate mode vertex
const uint totalAttrCount = (regs[PICA::InternalRegs::VertexShaderAttrNum] & 0xf) + 1;
if (totalAttrCount <= immediateModeAttrIndex) {
printf("Broken state in the immediate mode vertex submission pipeline. Failing silently\n");
@ -199,13 +209,15 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
// If we've reached 3 verts, issue a draw call
// Handle rendering depending on the primitive type
if (immediateModeVertIndex == 3) {
renderer.drawVertices(PICA::PrimType::TriangleList, immediateModeVertices);
renderer->drawVertices(PICA::PrimType::TriangleList, immediateModeVertices);
switch (primType) {
// Triangle or geometry primitive. Draw a triangle and discard all vertices
case 0: case 3:
case 0:
case 3: {
immediateModeVertIndex = 0;
break;
}
// Triangle strip. Draw triangle, discard first vertex and keep the last 2
case 1:
@ -223,54 +235,72 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
}
}
}
} else { // Writing to fixed attributes 13 and 14 probably does nothing, but we'll see
} else { // Writing to fixed attributes 13 and 14 probably does nothing, but we'll see
log("Wrote to invalid fixed vertex attribute %d\n", fixedAttribIndex);
}
}
break;
case VertexShaderOpDescriptorIndex:
case VertexShaderOpDescriptorIndex: {
shaderUnit.vs.setOpDescriptorIndex(value);
break;
}
case VertexShaderOpDescriptorData0: case VertexShaderOpDescriptorData1: case VertexShaderOpDescriptorData2:
case VertexShaderOpDescriptorData3: case VertexShaderOpDescriptorData4: case VertexShaderOpDescriptorData5:
case VertexShaderOpDescriptorData6: case VertexShaderOpDescriptorData7:
case VertexShaderOpDescriptorData0:
case VertexShaderOpDescriptorData1:
case VertexShaderOpDescriptorData2:
case VertexShaderOpDescriptorData3:
case VertexShaderOpDescriptorData4:
case VertexShaderOpDescriptorData5:
case VertexShaderOpDescriptorData6:
case VertexShaderOpDescriptorData7: {
shaderUnit.vs.uploadDescriptor(value);
break;
}
case VertexBoolUniform:
case VertexBoolUniform: {
shaderUnit.vs.boolUniform = value & 0xffff;
break;
}
case VertexIntUniform0: case VertexIntUniform1: case VertexIntUniform2: case VertexIntUniform3:
case VertexIntUniform0:
case VertexIntUniform1:
case VertexIntUniform2:
case VertexIntUniform3: {
shaderUnit.vs.uploadIntUniform(index - VertexIntUniform0, value);
break;
}
case VertexShaderData0: case VertexShaderData1: case VertexShaderData2: case VertexShaderData3:
case VertexShaderData4: case VertexShaderData5: case VertexShaderData6: case VertexShaderData7:
case VertexShaderData0:
case VertexShaderData1:
case VertexShaderData2:
case VertexShaderData3:
case VertexShaderData4:
case VertexShaderData5:
case VertexShaderData6:
case VertexShaderData7: {
shaderUnit.vs.uploadWord(value);
break;
}
case VertexShaderEntrypoint:
case VertexShaderEntrypoint: {
shaderUnit.vs.entrypoint = value & 0xffff;
break;
}
case VertexShaderTransferEnd:
if (value != 0) shaderUnit.vs.finalize();
break;
case VertexShaderTransferIndex:
shaderUnit.vs.setBufferIndex(value);
break;
case VertexShaderTransferIndex: shaderUnit.vs.setBufferIndex(value); break;
// Command lists can write to the command processor registers and change the command list stream
// Several games are known to do this, including New Super Mario Bros 2 and Super Mario 3D Land
case CmdBufTrigger0:
case CmdBufTrigger1: {
if (value != 0) { // A non-zero value triggers command list processing
int bufferIndex = index - CmdBufTrigger0; // Index of the command buffer to execute (0 or 1)
if (value != 0) { // A non-zero value triggers command list processing
int bufferIndex = index - CmdBufTrigger0; // Index of the command buffer to execute (0 or 1)
u32 addr = (regs[CmdBufAddr0 + bufferIndex] & 0xfffffff) << 3;
u32 size = (regs[CmdBufSize0 + bufferIndex] & 0xfffff) << 3;
@ -285,15 +315,13 @@ void GPU::writeInternalReg(u32 index, u32 value, u32 mask) {
default:
// Vertex attribute registers
if (index >= AttribInfoStart && index <= AttribInfoEnd) {
uint attributeIndex = (index - AttribInfoStart) / 3; // Which attribute are we writing to
uint reg = (index - AttribInfoStart) % 3; // Which of this attribute's registers are we writing to?
uint attributeIndex = (index - AttribInfoStart) / 3; // Which attribute are we writing to
uint reg = (index - AttribInfoStart) % 3; // Which of this attribute's registers are we writing to?
auto& attr = attributeInfo[attributeIndex];
switch (reg) {
case 0: attr.offset = value & 0xfffffff; break; // Attribute offset
case 1:
attr.config1 = value;
break;
case 0: attr.offset = value & 0xfffffff; break; // Attribute offset
case 1: attr.config1 = value; break;
case 2:
attr.config2 = value;
attr.size = getBits<16, 8>(value);
@ -339,13 +367,13 @@ void GPU::startCommandList(u32 addr, u32 size) {
u32 id = header & 0xffff;
u32 paramMaskIndex = getBits<16, 4>(header);
u32 paramCount = getBits<20, 8>(header); // Number of additional parameters
u32 paramCount = getBits<20, 8>(header); // Number of additional parameters
// Bit 31 tells us whether this command is going to write to multiple sequential registers (if the bit is 1)
// Or if all written values will go to the same register (If the bit is 0). It's essentially the value that
// gets added to the "id" field after each register write
bool consecutiveWritingMode = (header >> 31) != 0;
u32 mask = maskLUT[paramMaskIndex]; // Actual parameter mask
u32 mask = maskLUT[paramMaskIndex]; // Actual parameter mask
// Increment the ID by 1 after each write if we're in consecutive mode, or 0 otherwise
u32 idIncrement = (consecutiveWritingMode) ? 1 : 0;

133
src/core/PICA/shader_interpreter.cpp
View file

@ -1,6 +1,7 @@
#include "PICA/shader.hpp"
#include <cmath>
#include "PICA/shader.hpp"
using namespace Helpers;
void PICAShader::run() {
@ -11,20 +12,23 @@ void PICAShader::run() {
while (true) {
const u32 instruction = loadedShader[pc++];
const u32 opcode = instruction >> 26; // Top 6 bits are the opcode
const u32 opcode = instruction >> 26; // Top 6 bits are the opcode
switch (opcode) {
case ShaderOpcodes::ADD: add(instruction); break;
case ShaderOpcodes::CALL: call(instruction); break;
case ShaderOpcodes::CALLC: callc(instruction); break;
case ShaderOpcodes::CALLU: callu(instruction); break;
case ShaderOpcodes::CMP1: case ShaderOpcodes::CMP2:
case ShaderOpcodes::CMP1:
case ShaderOpcodes::CMP2: {
cmp(instruction);
break;
}
case ShaderOpcodes::DP3: dp3(instruction); break;
case ShaderOpcodes::DP4: dp4(instruction); break;
case ShaderOpcodes::DPHI: dphi(instruction); break;
case ShaderOpcodes::END: return; // Stop running shader
case ShaderOpcodes::END: return; // Stop running shader
case ShaderOpcodes::EX2: ex2(instruction); break;
case ShaderOpcodes::FLR: flr(instruction); break;
case ShaderOpcodes::IFC: ifc(instruction); break;
@ -38,31 +42,47 @@ void PICAShader::run() {
case ShaderOpcodes::MOV: mov(instruction); break;
case ShaderOpcodes::MOVA: mova(instruction); break;
case ShaderOpcodes::MUL: mul(instruction); break;
case ShaderOpcodes::NOP: break; // Do nothing
case ShaderOpcodes::NOP: break; // Do nothing
case ShaderOpcodes::RCP: rcp(instruction); break;
case ShaderOpcodes::RSQ: rsq(instruction); break;
case ShaderOpcodes::SGEI: sgei(instruction); break;
case ShaderOpcodes::SLT: slt(instruction); break;
case ShaderOpcodes::SLTI: slti(instruction); break;
case 0x30: case 0x31: case 0x32: case 0x33: case 0x34: case 0x35: case 0x36: case 0x37:
case 0x30:
case 0x31:
case 0x32:
case 0x33:
case 0x34:
case 0x35:
case 0x36:
case 0x37: {
madi(instruction);
break;
}
case 0x38: case 0x39: case 0x3A: case 0x3B: case 0x3C: case 0x3D: case 0x3E: case 0x3F:
case 0x38:
case 0x39:
case 0x3A:
case 0x3B:
case 0x3C:
case 0x3D:
case 0x3E:
case 0x3F: {
mad(instruction);
break;
}
default:Helpers::panic("Unimplemented PICA instruction %08X (Opcode = %02X)", instruction, opcode);
default: Helpers::panic("Unimplemented PICA instruction %08X (Opcode = %02X)", instruction, opcode);
}
// Handle control flow statements. The ordering is important as the priority goes: LOOP > IF > CALL
// Handle loop
if (loopIndex != 0) {
auto& loop = loopInfo[loopIndex - 1];
if (pc == loop.endingPC) { // Check if the loop needs to start over
if (pc == loop.endingPC) { // Check if the loop needs to start over
loop.iterations -= 1;
if (loop.iterations == 0) // If the loop ended, go one level down on the loop stack
if (loop.iterations == 0) // If the loop ended, go one level down on the loop stack
loopIndex -= 1;
loopCounter += loop.increment;
@ -73,7 +93,7 @@ void PICAShader::run() {
// Handle ifs
if (ifIndex != 0) {
auto& info = conditionalInfo[ifIndex - 1];
if (pc == info.endingPC) { // Check if the IF block ended
if (pc == info.endingPC) { // Check if the IF block ended
pc = info.newPC;
ifIndex -= 1;
}
@ -82,7 +102,7 @@ void PICAShader::run() {
// Handle calls
if (callIndex != 0) {
auto& info = callInfo[callIndex - 1];
if (pc == info.endingPC) { // Check if the CALL block ended
if (pc == info.endingPC) { // Check if the CALL block ended
pc = info.returnPC;
callIndex -= 1;
}
@ -92,15 +112,15 @@ void PICAShader::run() {
// Calculate the actual source value using an instruction's source field and it's respective index value
// The index value is used to apply relative addressing when index != 0 by adding one of the 3 addr registers to the
// source field, but only with the original source field is pointing at a vector uniform register
// source field, but only with the original source field is pointing at a vector uniform register
u8 PICAShader::getIndexedSource(u32 source, u32 index) {
if (source < 0x20) // No offset is applied if the source isn't pointing to a vector uniform reg
if (source < 0x20) // No offset is applied if the source isn't pointing to a vector uniform reg
return source;
switch (index) {
case 0: [[likely]] return u8(source); // No offset applied
case 1: return u8(source + addrRegister.x());
case 2: return u8(source + addrRegister.y());
case 0: [[likely]] return u8(source); // No offset applied
case 1: return u8(source + addrRegister[0]);
case 2: return u8(source + addrRegister[1]);
case 3: return u8(source + loopCounter);
}
@ -117,7 +137,7 @@ PICAShader::vec4f PICAShader::getSource(u32 source) {
return floatUniforms[source - 0x20];
else {
Helpers::warn("[PICA] Unimplemented source value: %X\n", source);
return vec4f({ f24::zero(), f24::zero(), f24::zero(), f24::zero() });
return vec4f({f24::zero(), f24::zero(), f24::zero(), f24::zero()});
}
}
@ -136,13 +156,13 @@ bool PICAShader::isCondTrue(u32 instruction) {
bool refX = (getBit<25>(instruction)) != 0;
switch (condition) {
case 0: // Either cmp register matches
case 0: // Either cmp register matches
return cmpRegister[0] == refX || cmpRegister[1] == refY;
case 1: // Both cmp registers match
case 1: // Both cmp registers match
return cmpRegister[0] == refX && cmpRegister[1] == refY;
case 2: // At least cmp.x matches
case 2: // At least cmp.x matches
return cmpRegister[0] == refX;
default: // At least cmp.y matches
default: // At least cmp.y matches
return cmpRegister[1] == refY;
}
}
@ -150,7 +170,7 @@ bool PICAShader::isCondTrue(u32 instruction) {
void PICAShader::add(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -171,7 +191,7 @@ void PICAShader::add(u32 instruction) {
void PICAShader::mul(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -210,7 +230,7 @@ void PICAShader::flr(u32 instruction) {
void PICAShader::max(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
const u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -232,7 +252,7 @@ void PICAShader::max(u32 instruction) {
void PICAShader::min(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
const u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -278,16 +298,16 @@ void PICAShader::mova(u32 instruction) {
vec4f srcVector = getSourceSwizzled<1>(src, operandDescriptor);
u32 componentMask = operandDescriptor & 0xf;
if (componentMask & 0b1000) // x component
addrRegister.x() = static_cast<s32>(srcVector.x().toFloat32());
if (componentMask & 0b0100) // y component
addrRegister.y() = static_cast<s32>(srcVector.y().toFloat32());
if (componentMask & 0b1000) // x component
addrRegister[0] = static_cast<s32>(srcVector[0].toFloat32());
if (componentMask & 0b0100) // y component
addrRegister[1] = static_cast<s32>(srcVector[1].toFloat32());
}
void PICAShader::dp3(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -309,7 +329,7 @@ void PICAShader::dp3(u32 instruction) {
void PICAShader::dp4(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -480,7 +500,7 @@ void PICAShader::madi(u32 instruction) {
void PICAShader::slt(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 dest = getBits<21, 5>(instruction);
@ -542,11 +562,11 @@ void PICAShader::slti(u32 instruction) {
void PICAShader::cmp(u32 instruction) {
const u32 operandDescriptor = operandDescriptors[instruction & 0x7f];
const u32 src1 = getBits<12, 7>(instruction);
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 src2 = getBits<7, 5>(instruction); // src2 coming first because PICA moment
const u32 idx = getBits<19, 2>(instruction);
const u32 cmpY = getBits<21, 3>(instruction);
const u32 cmpX = getBits<24, 3>(instruction);
const u32 cmpOperations[2] = { cmpX, cmpY };
const u32 cmpOperations[2] = {cmpX, cmpY};
if (idx) Helpers::panic("[PICA] CMP: idx != 0");
vec4f srcVec1 = getSourceSwizzled<1>(src1, operandDescriptor);
@ -554,33 +574,34 @@ void PICAShader::cmp(u32 instruction) {
for (int i = 0; i < 2; i++) {
switch (cmpOperations[i]) {
case 0: // Equal
case 0: // Equal
cmpRegister[i] = srcVec1[i] == srcVec2[i];
break;
case 1: // Not equal
case 1: // Not equal
cmpRegister[i] = srcVec1[i] != srcVec2[i];
break;
case 2: // Less than
case 2: // Less than
cmpRegister[i] = srcVec1[i] < srcVec2[i];
break;
case 3: // Less than or equal
case 3: // Less than or equal
cmpRegister[i] = srcVec1[i] <= srcVec2[i];
break;
case 4: // Greater than
case 4: // Greater than
cmpRegister[i] = srcVec1[i] > srcVec2[i];
break;
case 5: // Greater than or equal
case 5: // Greater than or equal
cmpRegister[i] = srcVec1[i] >= srcVec2[i];
break;
default:
default: {
cmpRegister[i] = true;
break;
}
}
}
}
@ -604,7 +625,7 @@ void PICAShader::ifc(u32 instruction) {
void PICAShader::ifu(u32 instruction) {
const u32 dest = getBits<10, 12>(instruction);
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
if (boolUniform & (1 << bit)) {
if (ifIndex >= 8) [[unlikely]]
@ -615,8 +636,7 @@ void PICAShader::ifu(u32 instruction) {
auto& block = conditionalInfo[ifIndex++];
block.endingPC = dest;
block.newPC = dest + num;
}
else {
} else {
pc = dest;
}
}
@ -637,12 +657,12 @@ void PICAShader::call(u32 instruction) {
void PICAShader::callc(u32 instruction) {
if (isCondTrue(instruction)) {
call(instruction); // Pls inline
call(instruction); // Pls inline
}
}
void PICAShader::callu(u32 instruction) {
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
if (boolUniform & (1 << bit)) {
if (callIndex >= 4) [[unlikely]]
@ -664,26 +684,27 @@ void PICAShader::loop(u32 instruction) {
Helpers::panic("[PICA] Overflowed loop stack");
u32 dest = getBits<10, 12>(instruction);
auto& uniform = intUniforms[getBits<22, 2>(instruction)]; // The uniform we'll get loop info from
loopCounter = uniform.y();
auto& uniform = intUniforms[getBits<22, 2>(instruction)]; // The uniform we'll get loop info from
loopCounter = uniform[1];
auto& loop = loopInfo[loopIndex++];
loop.startingPC = pc;
loop.endingPC = dest + 1; // Loop is inclusive so we need + 1 here
loop.iterations = uniform.x() + 1;
loop.increment = uniform.z();
loop.endingPC = dest + 1; // Loop is inclusive so we need + 1 here
loop.iterations = uniform[0] + 1;
loop.increment = uniform[2];
}
void PICAShader::jmpc(u32 instruction) {
if (isCondTrue(instruction))
if (isCondTrue(instruction)) {
pc = getBits<10, 12>(instruction);
}
}
void PICAShader::jmpu(u32 instruction) {
const u32 test = (instruction & 1) ^ 1; // If the LSB is 0 we want to compare to true, otherwise compare to false
const u32 test = (instruction & 1) ^ 1; // If the LSB is 0 we want to compare to true, otherwise compare to false
const u32 dest = getBits<10, 12>(instruction);
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
const u32 bit = getBits<22, 4>(instruction); // Bit of the bool uniform to check
if (((boolUniform >> bit) & 1) == test) // Jump if the bool uniform is the value we want
if (((boolUniform >> bit) & 1) == test) // Jump if the bool uniform is the value we want
pc = dest;
}

9
src/core/PICA/shader_unit.cpp
View file

@ -1,4 +1,5 @@
#include "PICA/shader_unit.hpp"
#include "cityhash.hpp"
void ShaderUnit::reset() {
@ -18,18 +19,18 @@ void PICAShader::reset() {
opDescriptorIndex = 0;
f32UniformTransfer = false;
const vec4f zero = vec4f({ f24::zero(), f24::zero(), f24::zero(), f24::zero() });
const vec4f zero = vec4f({f24::zero(), f24::zero(), f24::zero(), f24::zero()});
inputs.fill(zero);
floatUniforms.fill(zero);
outputs.fill(zero);
tempRegisters.fill(zero);
for (auto& e : intUniforms) {
e.x() = e.y() = e.z() = e.w() = 0;
e[0] = e[1] = e[2] = e[3] = 0;
}
addrRegister.x() = 0;
addrRegister.y() = 0;
addrRegister[0] = 0;
addrRegister[1] = 0;
loopCounter = 0;
codeHashDirty = true;

210
src/core/action_replay.cpp Normal file
View file

@ -0,0 +1,210 @@
#include "action_replay.hpp"
ActionReplay::ActionReplay(Memory& mem, HIDService& hid) : mem(mem), hid(hid) { reset(); }
void ActionReplay::reset() {
// Default value of storage regs is 0
storage1 = 0;
storage2 = 0;
// TODO: Is the active storage persistent or not?
activeStorage = &storage1;
}
void ActionReplay::runCheat(const Cheat& cheat) {
// Set offset and data registers to 0 at the start of a cheat
data1 = data2 = offset1 = offset2 = 0;
pc = 0;
ifStackIndex = 0;
loopStackIndex = 0;
running = true;
activeOffset = &offset1;
activeData = &data1;
while (running) {
// See if we can fetch 1 64-bit opcode, otherwise we're out of bounds. Cheats seem to end when going out of bounds?
if (pc + 1 >= cheat.size()) {
return;
}
// Fetch instruction
const u32 instruction = cheat[pc++];
// Instructions D0000000 00000000 and D2000000 00000000 are unconditional
bool isUnconditional = cheat[pc] == 0 && (instruction == 0xD0000000 || instruction == 0xD2000000);
if (ifStackIndex > 0 && !isUnconditional && !ifStack[ifStackIndex - 1]) {
pc++; // Eat up dummy word
continue; // Skip conditional instructions where the condition is false
}
runInstruction(cheat, instruction);
}
}
u8 ActionReplay::read8(u32 addr) { return mem.read8(addr); }
u16 ActionReplay::read16(u32 addr) { return mem.read16(addr); }
u32 ActionReplay::read32(u32 addr) { return mem.read32(addr); }
// Some AR cheats seem to want to write to unmapped memory or memory that straight up does not exist
#define MAKE_WRITE_HANDLER(size) \
void ActionReplay::write##size(u32 addr, u##size value) { \
auto pointerWrite = mem.getWritePointer(addr); \
if (pointerWrite) { \
*(u##size*)pointerWrite = value; \
} else { \
auto pointerRead = mem.getReadPointer(addr); \
if (pointerRead) { \
*(u##size*)pointerRead = value; \
} else { \
Helpers::warn("AR code tried to write to invalid address: %08X\n", addr); \
} \
} \
}
MAKE_WRITE_HANDLER(8)
MAKE_WRITE_HANDLER(16)
MAKE_WRITE_HANDLER(32)
#undef MAKE_WRITE_HANDLER
void ActionReplay::runInstruction(const Cheat& cheat, u32 instruction) {
// Top nibble determines the instruction type
const u32 type = instruction >> 28;
switch (type) {
// 32-bit write to [XXXXXXX + offset]
case 0x0: {
const u32 baseAddr = Helpers::getBits<0, 28>(instruction);
const u32 value = cheat[pc++];
write32(baseAddr + *activeOffset, value);
break;
}
// 16-bit write to [XXXXXXX + offset]
case 0x1: {
const u32 baseAddr = Helpers::getBits<0, 28>(instruction);
const u16 value = u16(cheat[pc++]);
write16(baseAddr + *activeOffset, value);
break;
}
// 8-bit write to [XXXXXXX + offset]
case 0x2: {
const u32 baseAddr = Helpers::getBits<0, 28>(instruction);
const u8 value = u8(cheat[pc++]);
write8(baseAddr + *activeOffset, value);
break;
}
// Less Than (YYYYYYYY < [XXXXXXX + offset])
case 0x4: {
const u32 baseAddr = Helpers::getBits<0, 28>(instruction);
const u32 imm = cheat[pc++];
const u32 value = read32(baseAddr + *activeOffset);
Helpers::panic("TODO: How do ActionReplay conditional blocks work?");
break;
}
case 0xD: executeDType(cheat, instruction); break;
default: Helpers::panic("Unimplemented ActionReplay instruction type %X", type); break;
}
}
void ActionReplay::executeDType(const Cheat& cheat, u32 instruction) {
switch (instruction) {
case 0xD3000000: offset1 = cheat[pc++]; break;
case 0xD3000001: offset2 = cheat[pc++]; break;
case 0xDC000000: *activeOffset += cheat[pc++]; break;
// DD000000 XXXXXXXX - if KEYPAD has value XXXXXXXX execute next block
case 0xDD000000: {
const u32 mask = cheat[pc++];
const u32 buttons = hid.getOldButtons();
pushConditionBlock((buttons & mask) == mask);
break;
}
// Offset register ops
case 0xDF000000: {
const u32 subopcode = cheat[pc++];
switch (subopcode) {
case 0x00000000: activeOffset = &offset1; break;
case 0x00000001: activeOffset = &offset2; break;
case 0x00010000: offset2 = offset1; break;
case 0x00010001: offset1 = offset2; break;
case 0x00020000: data1 = offset1; break;
case 0x00020001: data2 = offset2; break;
default:
Helpers::warn("Unknown ActionReplay offset operation");
running = false;
break;
}
break;
}
// Data register operations
case 0xDF000001: {
const u32 subopcode = cheat[pc++];
switch (subopcode) {
case 0x00000000: activeData = &data1; break;
case 0x00000001: activeData = &data2; break;
case 0x00010000: data2 = data1; break;
case 0x00010001: data1 = data2; break;
case 0x00020000: offset1 = data1; break;
case 0x00020001: offset2 = data2; break;
default:
Helpers::warn("Unknown ActionReplay data operation");
running = false;
break;
}
break;
}
// Storage register operations
case 0xDF000002: {
const u32 subopcode = cheat[pc++];
switch (subopcode) {
case 0x00000000: activeStorage = &storage1; break;
case 0x00000001: activeStorage = &storage2; break;
case 0x00010000: data1 = storage1; break;
case 0x00010001: data2 = storage2; break;
case 0x00020000: storage1 = data1; break;
case 0x00020001: storage2 = data2; break;
default:
Helpers::warn("Unknown ActionReplay data operation: %08X", subopcode);
running = false;
break;
}
break;
}
// Control flow block operations
case 0xD2000000: {
const u32 subopcode = cheat[pc++];
switch (subopcode) {
// Ends all loop/execute blocks
case 0:
loopStackIndex = 0;
ifStackIndex = 0;
break;
default: Helpers::panic("Unknown ActionReplay control flow operation: %08X", subopcode); break;
}
break;
}
default: Helpers::panic("ActionReplay: Unimplemented d-type opcode: %08X", instruction); break;
}
}
void ActionReplay::pushConditionBlock(bool condition) {
if (ifStackIndex >= 32) {
Helpers::warn("ActionReplay if stack overflowed");
running = false;
return;
}
ifStack[ifStackIndex++] = condition;
}

28
src/core/cheats.cpp Normal file
View file

@ -0,0 +1,28 @@
#include "cheats.hpp"
Cheats::Cheats(Memory& mem, HIDService& hid) : ar(mem, hid) { reset(); }
void Cheats::reset() {
cheats.clear(); // Unload loaded cheats
ar.reset(); // Reset ActionReplay
}
void Cheats::addCheat(const Cheat& cheat) { cheats.push_back(cheat); }
void Cheats::run() {
for (const Cheat& cheat : cheats) {
switch (cheat.type) {
case CheatType::ActionReplay: {
ar.runCheat(cheat.instructions);
break;
}
case CheatType::Gateway: {
Helpers::panic("Gateway cheats not supported yet! Only Action Replay is supported!");
break;
}
default: Helpers::panic("Unknown cheat type");
}
}
}

147
src/core/kernel/directory_operations.cpp
View file

@ -1,3 +1,10 @@
#include <array>
#include <cctype>
#include <filesystem>
#include <string>
#include <utility>
#include "ipc.hpp"
#include "kernel.hpp"
namespace DirectoryOps {
@ -7,6 +14,79 @@ namespace DirectoryOps {
};
}
// Helper to convert std::string to an 8.3 filename to mimic how Directory::Read works
using ShortFilename = std::array<char, 9>;
using ShortExtension = std::array<char, 4>;
using Filename83 = std::pair<ShortFilename, ShortExtension>;
// The input string should be the stem and extension together, not separately
// Eg something like "boop.png", "panda.txt", etc
Filename83 convertTo83(const std::string& path) {
ShortFilename filename;
ShortExtension extension;
// Convert a character to add it to the 8.3 name
// "Characters such as + are changed to the underscore _, and letters are put in uppercase"
// For now we put letters in uppercase until we find out what is supposed to be converted to _ and so on
auto convertCharacter = [](char c) { return (char) std::toupper(c); };
// List of forbidden character for 8.3 filenames, from Citra
// TODO: Use constexpr when C++20 support is solid
const std::string forbiddenChars = ".\"/\\[]:;=, ";
// By default space-initialize the whole name, append null terminator in the end for both the filename and extension
filename.fill(' ');
extension.fill(' ');
filename[filename.size() - 1] = '\0';
extension[extension.size() - 1] = '\0';
// Find the position of the dot in the string
auto dotPos = path.rfind('.');
// Wikipedia: If a file name has no extension, a trailing . has no effect
// Thus check if the last character is a dot and ignore it, prefering the previous dot if it exists
if (dotPos == path.size() - 1) {
dotPos = path.rfind('.', dotPos); // Get previous dot
}
// If pointPos is not npos we have a valid dot character, and as such an extension
bool haveExtension = dotPos != std::string::npos;
int validCharacterCount = 0;
bool filenameTooBig = false;
// Parse characters until we're done OR until we reach 9 characters, in which case according to Wikipedia we must truncate to 6 letters
// And append ~1 in the end
for (auto c : path.substr(0, dotPos)) {
// Character is forbidden, we must ignore it
if (forbiddenChars.find(c) != std::string::npos) {
continue;
}
// We already have capped the amount of characters, thus our filename is too big
if (validCharacterCount == 8) {
filenameTooBig = true;
break;
}
filename[validCharacterCount++] = convertCharacter(c); // Append character to filename
}
// Truncate name to 6 characters and denote that it is too big
// TODO: Wikipedia says we should also do this if the filename contains an invalid character, including spaces. Must test
if (filenameTooBig) {
filename[6] = '~';
filename[7] = '1';
}
if (haveExtension) {
int extensionLen = 0;
// Copy up to 3 characters from the dot onwards to the extension
for (auto c : path.substr(dotPos + 1, 3)) {
extension[extensionLen++] = convertCharacter(c);
}
}
return {filename, extension};
}
void Kernel::handleDirectoryOperation(u32 messagePointer, Handle directory) {
const u32 cmd = mem.read32(messagePointer);
switch (cmd) {
@ -25,16 +105,77 @@ void Kernel::closeDirectory(u32 messagePointer, Handle directory) {
}
p->getData<DirectorySession>()->isOpen = false;
mem.write32(messagePointer, IPC::responseHeader(0x802, 1, 0));
mem.write32(messagePointer + 4, Result::Success);
}
void Kernel::readDirectory(u32 messagePointer, Handle directory) {
const u32 entryCount = mem.read32(messagePointer + 4);
const u32 outPointer = mem.read32(messagePointer + 12);
logFileIO("Directory::Read (handle = %X, entry count = %d, out pointer = %08X)\n", directory, entryCount, outPointer);
Helpers::panicDev("Unimplemented FsDir::Read");
const auto p = getObject(directory, KernelObjectType::Directory);
if (p == nullptr) [[unlikely]] {
Helpers::panic("Called ReadDirectory on non-existent directory");
}
DirectorySession* session = p->getData<DirectorySession>();
if (!session->pathOnDisk.has_value()) [[unlikely]] {
Helpers::panic("Called ReadDirectory on directory that doesn't have a path on disk");
}
std::filesystem::path dirPath = session->pathOnDisk.value();
int count = 0;
while (count < entryCount && session->currentEntry < session->entries.size()) {
const auto& entry = session->entries[session->currentEntry];
std::filesystem::path path = entry.path;
std::filesystem::path filename = path.filename();
std::filesystem::path relative = path.lexically_relative(dirPath);
bool isDirectory = std::filesystem::is_directory(relative);
std::u16string nameU16 = relative.u16string();
bool isHidden = nameU16[0] == u'.'; // If the first character is a dot then this is a hidden file/folder
const u32 entryPointer = outPointer + (count * 0x228); // 0x228 is the size of a single entry
u32 utfPointer = entryPointer;
u32 namePointer = entryPointer + 0x20C;
u32 extensionPointer = entryPointer + 0x216;
u32 attributePointer = entryPointer + 0x21C;
u32 sizePointer = entryPointer + 0x220;
std::string filenameString = filename.string();
auto [shortFilename, shortExtension] = convertTo83(filenameString);
for (auto c : nameU16) {
mem.write16(utfPointer, u16(c));
utfPointer += sizeof(u16);
}
mem.write16(utfPointer, 0); // Null terminate the UTF16 name
// Write 8.3 filename-extension
for (auto c : shortFilename) {
mem.write8(namePointer, u8(c));
namePointer += sizeof(u8);
}
for (auto c : shortExtension) {
mem.write8(extensionPointer, u8(c));
extensionPointer += sizeof(u8);
}
mem.write8(outPointer + 0x21A, 1); // Always 1 according to 3DBrew
mem.write8(attributePointer, entry.isDirectory ? 1 : 0); // "Is directory" attribute
mem.write8(attributePointer + 1, isHidden ? 1 : 0); // "Is hidden" attribute
mem.write8(attributePointer + 2, entry.isDirectory ? 0 : 1); // "Is archive" attribute
mem.write8(attributePointer + 3, 0); // "Is read-only" attribute
count++; // Increment number of read directories
session->currentEntry++; // Increment index of the entry currently being read
}
mem.write32(messagePointer, IPC::responseHeader(0x801, 2, 2));
mem.write32(messagePointer + 4, Result::Success);
mem.write32(messagePointer + 8, 0);
mem.write32(messagePointer + 8, count);
}

27
src/core/kernel/kernel.cpp
View file

@ -95,14 +95,29 @@ KernelObject* Kernel::getProcessFromPID(Handle handle) {
}
void Kernel::deleteObjectData(KernelObject& object) {
using enum KernelObjectType;
// Resource limit and thread objects do not allocate heap data, so we don't delete anything
if (object.data == nullptr || object.type == ResourceLimit || object.type == Thread) {
if (object.data == nullptr) {
return;
}
delete object.data;
// Resource limit and thread objects do not allocate heap data, so we don't delete anything
switch (object.type) {
case KernelObjectType::AddressArbiter: delete object.getData<AddressArbiter>(); return;
case KernelObjectType::Archive: delete object.getData<ArchiveSession>(); return;
case KernelObjectType::Directory: delete object.getData<DirectorySession>(); return;
case KernelObjectType::Event: delete object.getData<Event>(); return;
case KernelObjectType::File: delete object.getData<FileSession>(); return;
case KernelObjectType::MemoryBlock: delete object.getData<MemoryBlock>(); return;
case KernelObjectType::Port: delete object.getData<Port>(); return;
case KernelObjectType::Process: delete object.getData<Process>(); return;
case KernelObjectType::ResourceLimit: return;
case KernelObjectType::Session: delete object.getData<Session>(); return;
case KernelObjectType::Mutex: delete object.getData<Mutex>(); return;
case KernelObjectType::Semaphore: delete object.getData<Semaphore>(); return;
case KernelObjectType::Thread: return;
case KernelObjectType::Dummy: return;
default: [[unlikely]] Helpers::warn("unknown object type"); return;
}
}
void Kernel::reset() {
@ -240,4 +255,4 @@ std::string Kernel::getProcessName(u32 pid) {
} else {
Helpers::panic("Attempted to name non-current process");
}
}
}

7
src/core/renderer_gl/etc1.cpp
View file

@ -9,7 +9,7 @@ static constexpr u32 signExtend3To32(u32 val) {
return (u32)(s32(val) << 29 >> 29);
}
u32 Texture::getTexelETC(bool hasAlpha, u32 u, u32 v, u32 width, const void* data) {
u32 Texture::getTexelETC(bool hasAlpha, u32 u, u32 v, u32 width, std::span<const u8> data) {
// Pixel offset of the 8x8 tile based on u, v and the width of the texture
u32 offs = ((u & ~7) * 8) + ((v & ~7) * width);
if (!hasAlpha)
@ -30,8 +30,7 @@ u32 Texture::getTexelETC(bool hasAlpha, u32 u, u32 v, u32 width, const void* dat
offs += subTileSize * subTileIndex;
u32 alpha;
const u8* tmp = static_cast<const u8*>(data) + offs; // Pointer to colour and alpha data as u8*
const u64* ptr = reinterpret_cast<const u64*>(tmp); // Cast to u64*
const u64* ptr = reinterpret_cast<const u64*>(data.data() + offs); // Cast to u64*
if (hasAlpha) {
// First 64 bits of the 4x4 subtile are alpha data
@ -118,4 +117,4 @@ u32 Texture::decodeETC(u32 alpha, u32 u, u32 v, u64 colourData) {
b = std::clamp(b + modifier, 0, 255);
return (alpha << 24) | (u32(b) << 16) | (u32(g) << 8) | u32(r);
}
}

2
src/gl_state.cpp → src/core/renderer_gl/gl_state.cpp
View file

@ -1,4 +1,4 @@
#include "gl_state.hpp"
#include "renderer_gl/gl_state.hpp"
void GLStateManager::resetBlend() {
blendEnabled = false;

726
src/core/renderer_gl/renderer_gl.cpp
View file

@ -1,582 +1,22 @@
#include "renderer_gl/renderer_gl.hpp"
#include <stb_image_write.h>
#include <cmrc/cmrc.hpp>
#include "PICA/float_types.hpp"
#include "PICA/gpu.hpp"
#include "PICA/regs.hpp"
CMRC_DECLARE(RendererGL);
using namespace Floats;
using namespace Helpers;
using namespace PICA;
const char* vertexShader = R"(
#version 410 core
layout (location = 0) in vec4 a_coords;
layout (location = 1) in vec4 a_quaternion;
layout (location = 2) in vec4 a_vertexColour;
layout (location = 3) in vec2 a_texcoord0;
layout (location = 4) in vec2 a_texcoord1;
layout (location = 5) in float a_texcoord0_w;
layout (location = 6) in vec3 a_view;
layout (location = 7) in vec2 a_texcoord2;
RendererGL::~RendererGL() {}
out vec3 v_normal;
out vec3 v_tangent;
out vec3 v_bitangent;
out vec4 v_colour;
out vec3 v_texcoord0;
out vec2 v_texcoord1;
out vec3 v_view;
out vec2 v_texcoord2;
flat out vec4 v_textureEnvColor[6];
flat out vec4 v_textureEnvBufferColor;
out float gl_ClipDistance[2];
// TEV uniforms
uniform uint u_textureEnvColor[6];
uniform uint u_picaRegs[0x200 - 0x48];
// Helper so that the implementation of u_pica_regs can be changed later
uint readPicaReg(uint reg_addr){
return u_picaRegs[reg_addr - 0x48];
}
vec4 abgr8888ToVec4(uint abgr) {
const float scale = 1.0 / 255.0;
return scale * vec4(
float(abgr & 0xffu),
float((abgr >> 8) & 0xffu),
float((abgr >> 16) & 0xffu),
float(abgr >> 24)
);
}
vec3 rotateVec3ByQuaternion(vec3 v, vec4 q){
vec3 u = q.xyz;
float s = q.w;
return 2.0 * dot(u, v) * u + (s * s - dot(u, u))* v + 2.0 * s * cross(u, v);
}
// Convert an arbitrary-width floating point literal to an f32
float decodeFP(uint hex, uint E, uint M){
uint width = M + E + 1u;
uint bias = 128u - (1u << (E - 1u));
uint exponent = (hex >> M) & ((1u << E) - 1u);
uint mantissa = hex & ((1u << M) - 1u);
uint sign = (hex >> (E + M)) << 31u;
if ((hex & ((1u << (width - 1u)) - 1u)) != 0) {
if (exponent == (1u << E) - 1u) exponent = 255u;
else exponent += bias;
hex = sign | (mantissa << (23u - M)) | (exponent << 23u);
} else {
hex = sign;
}
return uintBitsToFloat(hex);
}
void main() {
gl_Position = a_coords;
v_colour = a_vertexColour;
// Flip y axis of UVs because OpenGL uses an inverted y for texture sampling compared to the PICA
v_texcoord0 = vec3(a_texcoord0.x, 1.0 - a_texcoord0.y, a_texcoord0_w);
v_texcoord1 = vec2(a_texcoord1.x, 1.0 - a_texcoord1.y);
v_texcoord2 = vec2(a_texcoord2.x, 1.0 - a_texcoord2.y);
v_view = a_view;
v_normal = normalize(rotateVec3ByQuaternion(vec3(0.0, 0.0, 1.0), a_quaternion));
v_tangent = normalize(rotateVec3ByQuaternion(vec3(1.0, 0.0, 0.0), a_quaternion));
v_bitangent = normalize(rotateVec3ByQuaternion(vec3(0.0, 1.0, 0.0), a_quaternion));
for (int i = 0; i < 6; i++) {
v_textureEnvColor[i] = abgr8888ToVec4(u_textureEnvColor[i]);
}
v_textureEnvBufferColor = abgr8888ToVec4(readPicaReg(0xFD));
// Parse clipping plane registers
// The plane registers describe a clipping plane in the form of Ax + By + Cz + D = 0
// With n = (A, B, C) being the normal vector and D being the origin point distance
// Therefore, for the second clipping plane, we can just pass the dot product of the clip vector and the input coordinates to gl_ClipDistance[1]
vec4 clipData = vec4(
decodeFP(readPicaReg(0x48) & 0xffffffu, 7, 16),
decodeFP(readPicaReg(0x49) & 0xffffffu, 7, 16),
decodeFP(readPicaReg(0x4A) & 0xffffffu, 7, 16),
decodeFP(readPicaReg(0x4B) & 0xffffffu, 7, 16)
);
// There's also another, always-on clipping plane based on vertex z
gl_ClipDistance[0] = -a_coords.z;
gl_ClipDistance[1] = dot(clipData, a_coords);
}
)";
const char* fragmentShader = R"(
#version 410 core
in vec3 v_tangent;
in vec3 v_normal;
in vec3 v_bitangent;
in vec4 v_colour;
in vec3 v_texcoord0;
in vec2 v_texcoord1;
in vec3 v_view;
in vec2 v_texcoord2;
flat in vec4 v_textureEnvColor[6];
flat in vec4 v_textureEnvBufferColor;
out vec4 fragColour;
// TEV uniforms
uniform uint u_textureEnvSource[6];
uniform uint u_textureEnvOperand[6];
uniform uint u_textureEnvCombiner[6];
uniform uint u_textureEnvScale[6];
// Depth control uniforms
uniform float u_depthScale;
uniform float u_depthOffset;
uniform bool u_depthmapEnable;
uniform sampler2D u_tex0;
uniform sampler2D u_tex1;
uniform sampler2D u_tex2;
uniform sampler1DArray u_tex_lighting_lut;
uniform uint u_picaRegs[0x200 - 0x48];
// Helper so that the implementation of u_pica_regs can be changed later
uint readPicaReg(uint reg_addr){
return u_picaRegs[reg_addr - 0x48];
}
vec4 tevSources[16];
vec4 tevNextPreviousBuffer;
bool tevUnimplementedSourceFlag = false;
// OpenGL ES 1.1 reference pages for TEVs (this is what the PICA200 implements):
// https://registry.khronos.org/OpenGL-Refpages/es1.1/xhtml/glTexEnv.xml
vec4 tevFetchSource(uint src_id) {
if (src_id >= 6u && src_id < 13u) {
tevUnimplementedSourceFlag = true;
}
return tevSources[src_id];
}
vec4 tevGetColorAndAlphaSource(int tev_id, int src_id) {
vec4 result;
vec4 colorSource = tevFetchSource((u_textureEnvSource[tev_id] >> (src_id * 4)) & 15u);
vec4 alphaSource = tevFetchSource((u_textureEnvSource[tev_id] >> (src_id * 4 + 16)) & 15u);
uint colorOperand = (u_textureEnvOperand[tev_id] >> (src_id * 4)) & 15u;
uint alphaOperand = (u_textureEnvOperand[tev_id] >> (12 + src_id * 4)) & 7u;
// TODO: figure out what the undocumented values do
switch (colorOperand) {
case 0u: result.rgb = colorSource.rgb; break; // Source color
case 1u: result.rgb = 1.0 - colorSource.rgb; break; // One minus source color
case 2u: result.rgb = vec3(colorSource.a); break; // Source alpha
case 3u: result.rgb = vec3(1.0 - colorSource.a); break; // One minus source alpha
case 4u: result.rgb = vec3(colorSource.r); break; // Source red
case 5u: result.rgb = vec3(1.0 - colorSource.r); break; // One minus source red
case 8u: result.rgb = vec3(colorSource.g); break; // Source green
case 9u: result.rgb = vec3(1.0 - colorSource.g); break; // One minus source green
case 12u: result.rgb = vec3(colorSource.b); break; // Source blue
case 13u: result.rgb = vec3(1.0 - colorSource.b); break; // One minus source blue
default: break;
}
// TODO: figure out what the undocumented values do
switch (alphaOperand) {
case 0u: result.a = alphaSource.a; break; // Source alpha
case 1u: result.a = 1.0 - alphaSource.a; break; // One minus source alpha
case 2u: result.a = alphaSource.r; break; // Source red
case 3u: result.a = 1.0 - alphaSource.r; break; // One minus source red
case 4u: result.a = alphaSource.g; break; // Source green
case 5u: result.a = 1.0 - alphaSource.g; break; // One minus source green
case 6u: result.a = alphaSource.b; break; // Source blue
case 7u: result.a = 1.0 - alphaSource.b; break; // One minus source blue
default: break;
}
return result;
}
vec4 tevCalculateCombiner(int tev_id) {
vec4 source0 = tevGetColorAndAlphaSource(tev_id, 0);
vec4 source1 = tevGetColorAndAlphaSource(tev_id, 1);
vec4 source2 = tevGetColorAndAlphaSource(tev_id, 2);
uint colorCombine = u_textureEnvCombiner[tev_id] & 15u;
uint alphaCombine = (u_textureEnvCombiner[tev_id] >> 16) & 15u;
vec4 result = vec4(1.0);
// TODO: figure out what the undocumented values do
switch (colorCombine) {
case 0u: result.rgb = source0.rgb; break; // Replace
case 1u: result.rgb = source0.rgb * source1.rgb; break; // Modulate
case 2u: result.rgb = min(vec3(1.0), source0.rgb + source1.rgb); break; // Add
case 3u: result.rgb = clamp(source0.rgb + source1.rgb - 0.5, 0.0, 1.0); break; // Add signed
case 4u: result.rgb = mix(source1.rgb, source0.rgb, source2.rgb); break; // Interpolate
case 5u: result.rgb = max(source0.rgb - source1.rgb, 0.0); break; // Subtract
case 6u: result.rgb = vec3(4.0 * dot(source0.rgb - 0.5 , source1.rgb - 0.5)); break; // Dot3 RGB
case 7u: result = vec4(4.0 * dot(source0.rgb - 0.5 , source1.rgb - 0.5)); break; // Dot3 RGBA
case 8u: result.rgb = min(source0.rgb * source1.rgb + source2.rgb, 1.0); break; // Multiply then add
case 9u: result.rgb = min((source0.rgb + source1.rgb) * source2.rgb, 1.0); break; // Add then multiply
default: break;
}
if (colorCombine != 7u) { // The color combiner also writes the alpha channel in the "Dot3 RGBA" mode.
// TODO: figure out what the undocumented values do
// TODO: test if the alpha combiner supports all the same modes as the color combiner.
switch (alphaCombine) {
case 0u: result.a = source0.a; break; // Replace
case 1u: result.a = source0.a * source1.a; break; // Modulate
case 2u: result.a = min(1.0, source0.a + source1.a); break; // Add
case 3u: result.a = clamp(source0.a + source1.a - 0.5, 0.0, 1.0); break; // Add signed
case 4u: result.a = mix(source1.a, source0.a, source2.a); break; // Interpolate
case 5u: result.a = max(0.0, source0.a - source1.a); break; // Subtract
case 8u: result.a = min(1.0, source0.a * source1.a + source2.a); break; // Multiply then add
case 9u: result.a = min(1.0, (source0.a + source1.a) * source2.a); break; // Add then multiply
default: break;
}
}
result.rgb *= float(1 << (u_textureEnvScale[tev_id] & 3u));
result.a *= float(1 << ((u_textureEnvScale[tev_id] >> 16) & 3u));
return result;
}
#define D0_LUT 0u
#define D1_LUT 1u
#define SP_LUT 2u
#define FR_LUT 3u
#define RB_LUT 4u
#define RG_LUT 5u
#define RR_LUT 6u
float lutLookup(uint lut, uint light, float value){
if (lut >= FR_LUT && lut <= RR_LUT)
lut -= 1;
if (lut==SP_LUT)
lut = light + 8;
return texture(u_tex_lighting_lut, vec2(value, lut)).r;
}
vec3 regToColor(uint reg) {
// Normalization scale to convert from [0...255] to [0.0...1.0]
const float scale = 1.0 / 255.0;
return scale * vec3(
float(bitfieldExtract(reg, 20, 8)),
float(bitfieldExtract(reg, 10, 8)),
float(bitfieldExtract(reg, 00, 8))
);
}
// Convert an arbitrary-width floating point literal to an f32
float decodeFP(uint hex, uint E, uint M){
uint width = M + E + 1u;
uint bias = 128u - (1u << (E - 1u));
uint exponent = (hex >> M) & ((1u << E) - 1u);
uint mantissa = hex & ((1u << M) - 1u);
uint sign = (hex >> (E + M)) << 31u;
if ((hex & ((1u << (width - 1u)) - 1u)) != 0) {
if (exponent == (1u << E) - 1u) exponent = 255u;
else exponent += bias;
hex = sign | (mantissa << (23u - M)) | (exponent << 23u);
} else {
hex = sign;
}
return uintBitsToFloat(hex);
}
// Implements the following algorthm: https://mathb.in/26766
void calcLighting(out vec4 primary_color, out vec4 secondary_color){
// Quaternions describe a transformation from surface-local space to eye space.
// In surface-local space, by definition (and up to permutation) the normal vector is (0,0,1),
// the tangent vector is (1,0,0), and the bitangent vector is (0,1,0).
vec3 normal = normalize(v_normal );
vec3 tangent = normalize(v_tangent );
vec3 bitangent = normalize(v_bitangent);
vec3 view = normalize(v_view);
uint GPUREG_LIGHTING_ENABLE = readPicaReg(0x008F);
if (bitfieldExtract(GPUREG_LIGHTING_ENABLE, 0, 1) == 0){
primary_color = secondary_color = vec4(1.0);
return;
}
uint GPUREG_LIGHTING_AMBIENT = readPicaReg(0x01C0);
uint GPUREG_LIGHTING_NUM_LIGHTS = (readPicaReg(0x01C2) & 0x7u) +1;
uint GPUREG_LIGHTING_LIGHT_PERMUTATION = readPicaReg(0x01D9);
primary_color = vec4(vec3(0.0),1.0);
secondary_color = vec4(vec3(0.0),1.0);
primary_color.rgb += regToColor(GPUREG_LIGHTING_AMBIENT);
uint GPUREG_LIGHTING_LUTINPUT_ABS = readPicaReg(0x01D0);
uint GPUREG_LIGHTING_LUTINPUT_SELECT = readPicaReg(0x01D1);
uint GPUREG_LIGHTING_CONFIG0 = readPicaReg(0x01C3);
uint GPUREG_LIGHTING_CONFIG1 = readPicaReg(0x01C4);
uint GPUREG_LIGHTING_LUTINPUT_SCALE = readPicaReg(0x01D2);
float d[7];
bool error_unimpl = false;
for (uint i = 0; i < GPUREG_LIGHTING_NUM_LIGHTS; i++) {
uint light_id = bitfieldExtract(GPUREG_LIGHTING_LIGHT_PERMUTATION,int(i*3),3);
uint GPUREG_LIGHTi_SPECULAR0 = readPicaReg(0x0140 + 0x10 * light_id);
uint GPUREG_LIGHTi_SPECULAR1 = readPicaReg(0x0141 + 0x10 * light_id);
uint GPUREG_LIGHTi_DIFFUSE = readPicaReg(0x0142 + 0x10 * light_id);
uint GPUREG_LIGHTi_AMBIENT = readPicaReg(0x0143 + 0x10 * light_id);
uint GPUREG_LIGHTi_VECTOR_LOW = readPicaReg(0x0144 + 0x10 * light_id);
uint GPUREG_LIGHTi_VECTOR_HIGH= readPicaReg(0x0145 + 0x10 * light_id);
uint GPUREG_LIGHTi_CONFIG = readPicaReg(0x0149 + 0x10 * light_id);
vec3 light_vector = normalize(vec3(
decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_LOW, 0, 16), 5, 10),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_LOW, 16, 16), 5, 10),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_HIGH, 0, 16), 5, 10)
));
// Positional Light
if (bitfieldExtract(GPUREG_LIGHTi_CONFIG, 0, 1) == 0)
error_unimpl = true;
vec3 half_vector = normalize(normalize(light_vector) + view);
for (int c = 0; c < 7; c++) {
if (bitfieldExtract(GPUREG_LIGHTING_CONFIG1, 16 + c, 1) == 0){
uint scale_id = bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_SCALE, c * 4, 3);
float scale = float(1u << scale_id);
if (scale_id >= 6u)
scale/=256.0;
uint input_id = bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_SELECT, c * 4, 3);
if (input_id == 0u) d[c] = dot(normal,half_vector);
else if (input_id == 1u) d[c] = dot(view,half_vector);
else if (input_id == 2u) d[c] = dot(normal,view);
else if (input_id == 3u) d[c] = dot(light_vector,normal);
else if (input_id == 4u){
uint GPUREG_LIGHTi_SPOTDIR_LOW = readPicaReg(0x0146 + 0x10 * light_id);
uint GPUREG_LIGHTi_SPOTDIR_HIGH= readPicaReg(0x0147 + 0x10 * light_id);
vec3 spot_light_vector = normalize(vec3(
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_LOW, 0, 16), 1, 11),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_LOW, 16, 16), 1, 11),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_HIGH, 0, 16), 1, 11)
));
d[c] = dot(-light_vector, spot_light_vector); // -L dot P (aka Spotlight aka SP);
} else if (input_id == 5u) {
d[c] = 1.0; // TODO: cos <greek symbol> (aka CP);
error_unimpl = true;
} else {
d[c] = 1.0;
}
d[c] = lutLookup(c, light_id, d[c] * 0.5 + 0.5) * scale;
if (bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_ABS, 2 * c, 1) != 0u)
d[c] = abs(d[c]);
} else {
d[c] = 1.0;
}
}
uint lookup_config = bitfieldExtract(GPUREG_LIGHTi_CONFIG,4,4);
if (lookup_config == 0) {
d[D1_LUT] = 0.0;
d[FR_LUT] = 0.0;
d[RG_LUT]= d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 1) {
d[D0_LUT] = 0.0;
d[D1_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 2) {
d[FR_LUT] = 0.0;
d[SP_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 3) {
d[SP_LUT] = 0.0;
d[RG_LUT]= d[RB_LUT] = d[RR_LUT] = 1.0;
} else if (lookup_config == 4) {
d[FR_LUT] = 0.0;
} else if (lookup_config == 5) {
d[D1_LUT] = 0.0;
} else if (lookup_config == 6) {
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
}
float distance_factor = 1.0; // a
float indirect_factor = 1.0; // fi
float shadow_factor = 1.0; // o
float NdotL = dot(normal, light_vector); //Li dot N
// Two sided diffuse
if (bitfieldExtract(GPUREG_LIGHTi_CONFIG, 1, 1) == 0) NdotL = max(0.0, NdotL);
else NdotL = abs(NdotL);
float light_factor = distance_factor*d[SP_LUT]*indirect_factor*shadow_factor;
primary_color.rgb += light_factor * (regToColor(GPUREG_LIGHTi_AMBIENT) + regToColor(GPUREG_LIGHTi_DIFFUSE)*NdotL);
secondary_color.rgb += light_factor * (
regToColor(GPUREG_LIGHTi_SPECULAR0) * d[D0_LUT] +
regToColor(GPUREG_LIGHTi_SPECULAR1) * d[D1_LUT] * vec3(d[RR_LUT], d[RG_LUT], d[RB_LUT])
);
}
uint fresnel_output1 = bitfieldExtract(GPUREG_LIGHTING_CONFIG0, 2, 1);
uint fresnel_output2 = bitfieldExtract(GPUREG_LIGHTING_CONFIG0, 3, 1);
if (fresnel_output1 == 1u) primary_color.a = d[FR_LUT];
if (fresnel_output2 == 1u) secondary_color.a = d[FR_LUT];
if (error_unimpl) {
secondary_color = primary_color = vec4(1.0,0.,1.0,1.0);
}
}
void main() {
// TODO: what do invalid sources and disabled textures read as?
// And what does the "previous combiner" source read initially?
tevSources[0] = v_colour; // Primary/vertex color
calcLighting(tevSources[1],tevSources[2]);
uint textureConfig = readPicaReg(0x80);
vec2 tex2UV = (textureConfig & (1u << 13)) != 0u ? v_texcoord1 : v_texcoord2;
if ((textureConfig & 1u) != 0u) tevSources[3] = texture(u_tex0, v_texcoord0.xy);
if ((textureConfig & 2u) != 0u) tevSources[4] = texture(u_tex1, v_texcoord1);
if ((textureConfig & 4u) != 0u) tevSources[5] = texture(u_tex2, tex2UV);
tevSources[13] = vec4(0.0); // Previous buffer
tevSources[15] = vec4(0.0); // Previous combiner
tevNextPreviousBuffer = v_textureEnvBufferColor;
uint textureEnvUpdateBuffer = readPicaReg(0xE0);
for (int i = 0; i < 6; i++) {
tevSources[14] = v_textureEnvColor[i]; // Constant color
tevSources[15] = tevCalculateCombiner(i);
tevSources[13] = tevNextPreviousBuffer;
if (i < 4) {
if ((textureEnvUpdateBuffer & (0x100u << i)) != 0u) {
tevNextPreviousBuffer.rgb = tevSources[15].rgb;
}
if ((textureEnvUpdateBuffer & (0x1000u << i)) != 0u) {
tevNextPreviousBuffer.a = tevSources[15].a;
}
}
}
fragColour = tevSources[15];
if (tevUnimplementedSourceFlag) {
// fragColour = vec4(1.0, 0.0, 1.0, 1.0);
}
// fragColour.rg = texture(u_tex_lighting_lut,vec2(gl_FragCoord.x/200.,float(int(gl_FragCoord.y/2)%24))).rr;
// Get original depth value by converting from [near, far] = [0, 1] to [-1, 1]
// We do this by converting to [0, 2] first and subtracting 1 to go to [-1, 1]
float z_over_w = gl_FragCoord.z * 2.0f - 1.0f;
float depth = z_over_w * u_depthScale + u_depthOffset;
if (!u_depthmapEnable) // Divide z by w if depthmap enable == 0 (ie using W-buffering)
depth /= gl_FragCoord.w;
// Write final fragment depth
gl_FragDepth = depth;
// Perform alpha test
uint alphaControl = readPicaReg(0x104);
if ((alphaControl & 1u) != 0u) { // Check if alpha test is on
uint func = (alphaControl >> 4u) & 7u;
float reference = float((alphaControl >> 8u) & 0xffu) / 255.0;
float alpha = fragColour.a;
switch (func) {
case 0: discard; // Never pass alpha test
case 1: break; // Always pass alpha test
case 2: // Pass if equal
if (alpha != reference)
discard;
break;
case 3: // Pass if not equal
if (alpha == reference)
discard;
break;
case 4: // Pass if less than
if (alpha >= reference)
discard;
break;
case 5: // Pass if less than or equal
if (alpha > reference)
discard;
break;
case 6: // Pass if greater than
if (alpha <= reference)
discard;
break;
case 7: // Pass if greater than or equal
if (alpha < reference)
discard;
break;
}
}
}
)";
const char* displayVertexShader = R"(
#version 410 core
out vec2 UV;
void main() {
const vec4 positions[4] = vec4[](
vec4(-1.0, 1.0, 1.0, 1.0), // Top-left
vec4(1.0, 1.0, 1.0, 1.0), // Top-right
vec4(-1.0, -1.0, 1.0, 1.0), // Bottom-left
vec4(1.0, -1.0, 1.0, 1.0) // Bottom-right
);
// The 3DS displays both screens' framebuffer rotated 90 deg counter clockwise
// So we adjust our texcoords accordingly
const vec2 texcoords[4] = vec2[](
vec2(1.0, 1.0), // Top-right
vec2(1.0, 0.0), // Bottom-right
vec2(0.0, 1.0), // Top-left
vec2(0.0, 0.0) // Bottom-left
);
gl_Position = positions[gl_VertexID];
UV = texcoords[gl_VertexID];
}
)";
const char* displayFragmentShader = R"(
#version 410 core
in vec2 UV;
out vec4 FragColor;
uniform sampler2D u_texture;
void main() {
FragColor = texture(u_texture, UV);
}
)";
void Renderer::reset() {
void RendererGL::reset() {
depthBufferCache.reset();
colourBufferCache.reset();
textureCache.reset();
@ -592,10 +32,10 @@ void Renderer::reset() {
const auto oldProgram = OpenGL::getProgram();
gl.useProgram(triangleProgram);
oldDepthScale = -1.0; // Default depth scale to -1.0, which is what games typically use
oldDepthOffset = 0.0; // Default depth offset to 0
oldDepthmapEnable = false; // Enable w buffering
oldDepthScale = -1.0; // Default depth scale to -1.0, which is what games typically use
oldDepthOffset = 0.0; // Default depth offset to 0
oldDepthmapEnable = false; // Enable w buffering
glUniform1f(depthScaleLoc, oldDepthScale);
glUniform1f(depthOffsetLoc, oldDepthOffset);
@ -605,10 +45,17 @@ void Renderer::reset() {
}
}
void Renderer::initGraphicsContext() {
OpenGL::Shader vert(vertexShader, OpenGL::Vertex);
OpenGL::Shader frag(fragmentShader, OpenGL::Fragment);
triangleProgram.create({ vert, frag });
void RendererGL::initGraphicsContext() {
gl.reset();
auto gl_resources = cmrc::RendererGL::get_filesystem();
auto vertexShaderSource = gl_resources.open("opengl_vertex_shader.vert");
auto fragmentShaderSource = gl_resources.open("opengl_fragment_shader.frag");
OpenGL::Shader vert({vertexShaderSource.begin(), vertexShaderSource.size()}, OpenGL::Vertex);
OpenGL::Shader frag({fragmentShaderSource.begin(), fragmentShaderSource.size()}, OpenGL::Fragment);
triangleProgram.create({vert, frag});
gl.useProgram(triangleProgram);
textureEnvSourceLoc = OpenGL::uniformLocation(triangleProgram, "u_textureEnvSource");
@ -628,12 +75,15 @@ void Renderer::initGraphicsContext() {
glUniform1i(OpenGL::uniformLocation(triangleProgram, "u_tex2"), 2);
glUniform1i(OpenGL::uniformLocation(triangleProgram, "u_tex_lighting_lut"), 3);
OpenGL::Shader vertDisplay(displayVertexShader, OpenGL::Vertex);
OpenGL::Shader fragDisplay(displayFragmentShader, OpenGL::Fragment);
displayProgram.create({ vertDisplay, fragDisplay });
auto displayVertexShaderSource = gl_resources.open("opengl_display.vert");
auto displayFragmentShaderSource = gl_resources.open("opengl_display.frag");
OpenGL::Shader vertDisplay({displayVertexShaderSource.begin(), displayVertexShaderSource.size()}, OpenGL::Vertex);
OpenGL::Shader fragDisplay({displayFragmentShaderSource.begin(), displayFragmentShaderSource.size()}, OpenGL::Fragment);
displayProgram.create({vertDisplay, fragDisplay});
gl.useProgram(displayProgram);
glUniform1i(OpenGL::uniformLocation(displayProgram, "u_texture"), 0); // Init sampler object
glUniform1i(OpenGL::uniformLocation(displayProgram, "u_texture"), 0); // Init sampler object
vbo.createFixedSize(sizeof(Vertex) * vertexBufferSize, GL_STREAM_DRAW);
gl.bindVBO(vbo);
@ -669,10 +119,10 @@ void Renderer::initGraphicsContext() {
dummyVAO.create();
// Create texture and framebuffer for the 3DS screen
const u32 screenTextureWidth = 400; // Top screen is 400 pixels wide, bottom is 320
const u32 screenTextureHeight = 2 * 240; // Both screens are 240 pixels tall
glGenTextures(1,&lightLUTTextureArray);
const u32 screenTextureWidth = 400; // Top screen is 400 pixels wide, bottom is 320
const u32 screenTextureHeight = 2 * 240; // Both screens are 240 pixels tall
glGenTextures(1, &lightLUTTextureArray);
auto prevTexture = OpenGL::getTex2D();
screenTexture.create(screenTextureWidth, screenTextureHeight, GL_RGBA8);
@ -684,8 +134,7 @@ void Renderer::initGraphicsContext() {
screenFramebuffer.createWithDrawTexture(screenTexture);
screenFramebuffer.bind(OpenGL::DrawAndReadFramebuffer);
if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE)
Helpers::panic("Incomplete framebuffer");
if (glCheckFramebufferStatus(GL_FRAMEBUFFER) != GL_FRAMEBUFFER_COMPLETE) Helpers::panic("Incomplete framebuffer");
// TODO: This should not clear the framebuffer contents. It should load them from VRAM.
GLint oldViewport[4];
@ -699,19 +148,32 @@ void Renderer::initGraphicsContext() {
}
// Set up the OpenGL blending context to match the emulated PICA
void Renderer::setupBlending() {
void RendererGL::setupBlending() {
const bool blendingEnabled = (regs[PICA::InternalRegs::ColourOperation] & (1 << 8)) != 0;
// Map of PICA blending equations to OpenGL blending equations. The unused blending equations are equivalent to equation 0 (add)
static constexpr std::array<GLenum, 8> blendingEquations = {
GL_FUNC_ADD, GL_FUNC_SUBTRACT, GL_FUNC_REVERSE_SUBTRACT, GL_MIN, GL_MAX, GL_FUNC_ADD, GL_FUNC_ADD, GL_FUNC_ADD
GL_FUNC_ADD, GL_FUNC_SUBTRACT, GL_FUNC_REVERSE_SUBTRACT, GL_MIN, GL_MAX, GL_FUNC_ADD, GL_FUNC_ADD, GL_FUNC_ADD,
};
// Map of PICA blending funcs to OpenGL blending funcs. Func = 15 is undocumented and stubbed to GL_ONE for now
static constexpr std::array<GLenum, 16> blendingFuncs = {
GL_ZERO, GL_ONE, GL_SRC_COLOR, GL_ONE_MINUS_SRC_COLOR, GL_DST_COLOR, GL_ONE_MINUS_DST_COLOR, GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA,
GL_DST_ALPHA, GL_ONE_MINUS_DST_ALPHA, GL_CONSTANT_COLOR, GL_ONE_MINUS_CONSTANT_COLOR, GL_CONSTANT_ALPHA, GL_ONE_MINUS_CONSTANT_ALPHA,
GL_SRC_ALPHA_SATURATE, GL_ONE
GL_ZERO,
GL_ONE,
GL_SRC_COLOR,
GL_ONE_MINUS_SRC_COLOR,
GL_DST_COLOR,
GL_ONE_MINUS_DST_COLOR,
GL_SRC_ALPHA,
GL_ONE_MINUS_SRC_ALPHA,
GL_DST_ALPHA,
GL_ONE_MINUS_DST_ALPHA,
GL_CONSTANT_COLOR,
GL_ONE_MINUS_CONSTANT_COLOR,
GL_CONSTANT_ALPHA,
GL_ONE_MINUS_CONSTANT_ALPHA,
GL_SRC_ALPHA_SATURATE,
GL_ONE,
};
if (!blendingEnabled) {
@ -743,13 +205,12 @@ void Renderer::setupBlending() {
}
}
void Renderer::setupTextureEnvState() {
void RendererGL::setupTextureEnvState() {
// TODO: Only update uniforms when the TEV config changed. Use an UBO potentially.
static constexpr std::array<u32, 6> ioBases = {
PICA::InternalRegs::TexEnv0Source, PICA::InternalRegs::TexEnv1Source,
PICA::InternalRegs::TexEnv2Source, PICA::InternalRegs::TexEnv3Source,
PICA::InternalRegs::TexEnv4Source, PICA::InternalRegs::TexEnv5Source
PICA::InternalRegs::TexEnv0Source, PICA::InternalRegs::TexEnv1Source, PICA::InternalRegs::TexEnv2Source,
PICA::InternalRegs::TexEnv3Source, PICA::InternalRegs::TexEnv4Source, PICA::InternalRegs::TexEnv5Source,
};
u32 textureEnvSourceRegs[6];
@ -775,9 +236,11 @@ void Renderer::setupTextureEnvState() {
glUniform1uiv(textureEnvScaleLoc, 6, textureEnvScaleRegs);
}
void Renderer::bindTexturesToSlots() {
void RendererGL::bindTexturesToSlots() {
static constexpr std::array<u32, 3> ioBases = {
PICA::InternalRegs::Tex0BorderColor, PICA::InternalRegs::Tex1BorderColor, PICA::InternalRegs::Tex2BorderColor
PICA::InternalRegs::Tex0BorderColor,
PICA::InternalRegs::Tex1BorderColor,
PICA::InternalRegs::Tex2BorderColor,
};
for (int i = 0; i < 3; i++) {
@ -805,13 +268,13 @@ void Renderer::bindTexturesToSlots() {
glActiveTexture(GL_TEXTURE0);
}
void Renderer::updateLightingLUT() {
void RendererGL::updateLightingLUT() {
gpu.lightingLUTDirty = false;
std::array<u16, GPU::LightingLutSize> u16_lightinglut;
std::array<u16, GPU::LightingLutSize> u16_lightinglut;
for (int i = 0; i < gpu.lightingLUT.size(); i++) {
uint64_t value = gpu.lightingLUT[i] & ((1 << 12) - 1);
u16_lightinglut[i] = value * 65535 / 4095;
uint64_t value = gpu.lightingLUT[i] & ((1 << 12) - 1);
u16_lightinglut[i] = value * 65535 / 4095;
}
glActiveTexture(GL_TEXTURE0 + 3);
@ -824,19 +287,22 @@ void Renderer::updateLightingLUT() {
glActiveTexture(GL_TEXTURE0);
}
void Renderer::drawVertices(PICA::PrimType primType, std::span<const Vertex> vertices) {
void RendererGL::drawVertices(PICA::PrimType primType, std::span<const Vertex> vertices) {
// The fourth type is meant to be "Geometry primitive". TODO: Find out what that is
static constexpr std::array<OpenGL::Primitives, 4> primTypes = {
OpenGL::Triangle, OpenGL::TriangleStrip, OpenGL::TriangleFan, OpenGL::Triangle
OpenGL::Triangle,
OpenGL::TriangleStrip,
OpenGL::TriangleFan,
OpenGL::Triangle,
};
const auto primitiveTopology = primTypes[static_cast<usize>(primType)];
const auto primitiveTopology = primTypes[static_cast<usize>(primType)];
gl.disableScissor();
gl.bindVBO(vbo);
gl.bindVAO(vao);
gl.useProgram(triangleProgram);
OpenGL::enableClipPlane(0); // Clipping plane 0 is always enabled
OpenGL::enableClipPlane(0); // Clipping plane 0 is always enabled
if (regs[PICA::InternalRegs::ClipEnable] & 1) {
OpenGL::enableClipPlane(1);
}
@ -852,9 +318,7 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const Vertex> ver
const int colourMask = getBits<8, 4>(depthControl);
gl.setColourMask(colourMask & 1, colourMask & 2, colourMask & 4, colourMask & 8);
static constexpr std::array<GLenum, 8> depthModes = {
GL_NEVER, GL_ALWAYS, GL_EQUAL, GL_NOTEQUAL, GL_LESS, GL_LEQUAL, GL_GREATER, GL_GEQUAL
};
static constexpr std::array<GLenum, 8> depthModes = {GL_NEVER, GL_ALWAYS, GL_EQUAL, GL_NOTEQUAL, GL_LESS, GL_LEQUAL, GL_GREATER, GL_GEQUAL};
const float depthScale = f24::fromRaw(regs[PICA::InternalRegs::DepthScale] & 0xffffff).toFloat32();
const float depthOffset = f24::fromRaw(regs[PICA::InternalRegs::DepthOffset] & 0xffffff).toFloat32();
@ -865,7 +329,7 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const Vertex> ver
oldDepthScale = depthScale;
glUniform1f(depthScaleLoc, depthScale);
}
if (oldDepthOffset != depthOffset) {
oldDepthOffset = depthOffset;
glUniform1f(depthOffsetLoc, depthOffset);
@ -917,7 +381,7 @@ void Renderer::drawVertices(PICA::PrimType primType, std::span<const Vertex> ver
constexpr u32 topScreenBuffer = 0x1f000000;
constexpr u32 bottomScreenBuffer = 0x1f05dc00;
void Renderer::display() {
void RendererGL::display() {
gl.disableScissor();
glBindFramebuffer(GL_DRAW_FRAMEBUFFER, 0);
@ -925,7 +389,7 @@ void Renderer::display() {
glBlitFramebuffer(0, 0, 400, 480, 0, 0, 400, 480, GL_COLOR_BUFFER_BIT, GL_LINEAR);
}
void Renderer::clearBuffer(u32 startAddress, u32 endAddress, u32 value, u32 control) {
void RendererGL::clearBuffer(u32 startAddress, u32 endAddress, u32 value, u32 control) {
return;
log("GPU: Clear buffer\nStart: %08X End: %08X\nValue: %08X Control: %08X\n", startAddress, endAddress, value, control);
@ -947,10 +411,10 @@ void Renderer::clearBuffer(u32 startAddress, u32 endAddress, u32 value, u32 cont
OpenGL::clearColor();
}
OpenGL::Framebuffer Renderer::getColourFBO() {
//We construct a colour buffer object and see if our cache has any matching colour buffers in it
// If not, we allocate a texture & FBO for our framebuffer and store it in the cache
ColourBuffer sampleBuffer(colourBufferLoc, colourBufferFormat, fbSize.x(), fbSize.y());
OpenGL::Framebuffer RendererGL::getColourFBO() {
// We construct a colour buffer object and see if our cache has any matching colour buffers in it
// If not, we allocate a texture & FBO for our framebuffer and store it in the cache
ColourBuffer sampleBuffer(colourBufferLoc, colourBufferFormat, fbSize[0], fbSize[1]);
auto buffer = colourBufferCache.find(sampleBuffer);
if (buffer.has_value()) {
@ -960,9 +424,9 @@ OpenGL::Framebuffer Renderer::getColourFBO() {
}
}
void Renderer::bindDepthBuffer() {
void RendererGL::bindDepthBuffer() {
// Similar logic as the getColourFBO function
DepthBuffer sampleBuffer(depthBufferLoc, depthBufferFormat, fbSize.x(), fbSize.y());
DepthBuffer sampleBuffer(depthBufferLoc, depthBufferFormat, fbSize[0], fbSize[1]);
auto buffer = depthBufferCache.find(sampleBuffer);
GLuint tex;
@ -979,14 +443,14 @@ void Renderer::bindDepthBuffer() {
glFramebufferTexture2D(GL_FRAMEBUFFER, attachment, GL_TEXTURE_2D, tex, 0);
}
OpenGL::Texture Renderer::getTexture(Texture& tex) {
OpenGL::Texture RendererGL::getTexture(Texture& tex) {
// Similar logic as the getColourFBO/bindDepthBuffer functions
auto buffer = textureCache.find(tex);
if (buffer.has_value()) {
return buffer.value().get().texture;
} else {
const void* textureData = gpu.getPointerPhys<void*>(tex.location); // Get pointer to the texture data in 3DS memory
const auto textureData = std::span{gpu.getPointerPhys<u8>(tex.location), tex.sizeInBytes()}; // Get pointer to the texture data in 3DS memory
Texture& newTex = textureCache.add(tex);
newTex.decodeTexture(textureData);
@ -994,7 +458,7 @@ OpenGL::Texture Renderer::getTexture(Texture& tex) {
}
}
void Renderer::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags) {
void RendererGL::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags) {
const u32 inputWidth = inputSize & 0xffff;
const u32 inputGap = inputSize >> 16;
@ -1022,21 +486,21 @@ void Renderer::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32
// Hack: Detect whether we are writing to the top or bottom screen by checking output gap and drawing to the proper part of the output texture
// We consider output gap == 320 to mean bottom, and anything else to mean top
if (outputGap == 320) {
OpenGL::setViewport(40, 0, 320, 240); // Bottom screen viewport
OpenGL::setViewport(40, 0, 320, 240); // Bottom screen viewport
} else {
OpenGL::setViewport(0, 240, 400, 240); // Top screen viewport
OpenGL::setViewport(0, 240, 400, 240); // Top screen viewport
}
OpenGL::draw(OpenGL::TriangleStrip, 4); // Actually draw our 3DS screen
OpenGL::draw(OpenGL::TriangleStrip, 4); // Actually draw our 3DS screen
}
void Renderer::screenshot(const std::string& name) {
void RendererGL::screenshot(const std::string& name) {
constexpr uint width = 400;
constexpr uint height = 2 * 240;
std::vector<uint8_t> pixels, flippedPixels;
pixels.resize(width * height * 4);
flippedPixels.resize(pixels.size());;
pixels.resize(width * height * 4);
flippedPixels.resize(pixels.size());
OpenGL::bindScreenFramebuffer();
glReadPixels(0, 0, width, height, GL_BGRA, GL_UNSIGNED_BYTE, pixels.data());
@ -1053,4 +517,4 @@ void Renderer::screenshot(const std::string& name) {
}
stbi_write_png(name.c_str(), width, height, 4, flippedPixels.data(), 0);
}
}

78
src/core/renderer_gl/textures.cpp
View file

@ -112,12 +112,11 @@ u32 Texture::getSwizzledOffset_4bpp(u32 u, u32 v, u32 width) {
// Get the texel at position (u, v)
// fmt: format of the texture
// data: texture data of the texture
u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data) {
switch (fmt) {
case PICA::TextureFmt::RGBA4: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
auto ptr = static_cast<const u8*>(data);
u16 texel = u16(ptr[offset]) | (u16(ptr[offset + 1]) << 8);
u16 texel = u16(data[offset]) | (u16(data[offset + 1]) << 8);
u8 alpha = Colour::convert4To8Bit(getBits<0, 4, u8>(texel));
u8 b = Colour::convert4To8Bit(getBits<4, 4, u8>(texel));
@ -128,9 +127,8 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
}
case PICA::TextureFmt::RGBA5551: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
auto ptr = static_cast<const u8*>(data);
u16 texel = u16(ptr[offset]) | (u16(ptr[offset + 1]) << 8);
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(data[offset]) | (u16(data[offset + 1]) << 8);
u8 alpha = getBit<0>(texel) ? 0xff : 0;
u8 b = Colour::convert5To8Bit(getBits<1, 5, u8>(texel));
@ -141,56 +139,47 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
}
case PICA::TextureFmt::RGB565: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
auto ptr = static_cast<const u8*>(data);
u16 texel = u16(ptr[offset]) | (u16(ptr[offset + 1]) << 8);
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(data[offset]) | (u16(data[offset + 1]) << 8);
u8 b = Colour::convert5To8Bit(getBits<0, 5, u8>(texel));
u8 g = Colour::convert6To8Bit(getBits<5, 6, u8>(texel));
u8 r = Colour::convert5To8Bit(getBits<11, 5, u8>(texel));
const u8 b = Colour::convert5To8Bit(getBits<0, 5, u8>(texel));
const u8 g = Colour::convert6To8Bit(getBits<5, 6, u8>(texel));
const u8 r = Colour::convert5To8Bit(getBits<11, 5, u8>(texel));
return (0xff << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::RG8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
auto ptr = static_cast<const u8*>(data);
constexpr u8 b = 0;
u8 g = ptr[offset];
u8 r = ptr[offset + 1];
const u8 g = data[offset];
const u8 r = data[offset + 1];
return (0xff << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::RGB8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 3);
auto ptr = static_cast<const u8*>(data);
u8 b = ptr[offset];
u8 g = ptr[offset + 1];
u8 r = ptr[offset + 2];
const u32 offset = getSwizzledOffset(u, v, size.u(), 3);
const u8 b = data[offset];
const u8 g = data[offset + 1];
const u8 r = data[offset + 2];
return (0xff << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::RGBA8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 4);
auto ptr = static_cast<const u8*>(data);
u8 alpha = ptr[offset];
u8 b = ptr[offset + 1];
u8 g = ptr[offset + 2];
u8 r = ptr[offset + 3];
const u32 offset = getSwizzledOffset(u, v, size.u(), 4);
const u8 alpha = data[offset];
const u8 b = data[offset + 1];
const u8 g = data[offset + 2];
const u8 r = data[offset + 3];
return (alpha << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::IA4: {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
auto ptr = static_cast<const u8*>(data);
const u8 texel = ptr[offset];
const u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 texel = data[offset];
const u8 alpha = Colour::convert4To8Bit(texel & 0xf);
const u8 intensity = Colour::convert4To8Bit(texel >> 4);
@ -199,11 +188,10 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
}
case PICA::TextureFmt::A4: {
u32 offset = getSwizzledOffset_4bpp(u, v, size.u());
auto ptr = static_cast<const u8*>(data);
const u32 offset = getSwizzledOffset_4bpp(u, v, size.u());
// For odd U coordinates, grab the top 4 bits, and the low 4 bits for even coordinates
u8 alpha = ptr[offset] >> ((u % 2) ? 4 : 0);
u8 alpha = data[offset] >> ((u % 2) ? 4 : 0);
alpha = Colour::convert4To8Bit(getBits<0, 4>(alpha));
// A8 sets RGB to 0
@ -212,8 +200,7 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
case PICA::TextureFmt::A8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
auto ptr = static_cast<const u8*>(data);
const u8 alpha = ptr[offset];
const u8 alpha = data[offset];
// A8 sets RGB to 0
return (alpha << 24) | (0 << 16) | (0 << 8) | 0;
@ -221,10 +208,9 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
case PICA::TextureFmt::I4: {
u32 offset = getSwizzledOffset_4bpp(u, v, size.u());
auto ptr = static_cast<const u8*>(data);
// For odd U coordinates, grab the top 4 bits, and the low 4 bits for even coordinates
u8 intensity = ptr[offset] >> ((u % 2) ? 4 : 0);
u8 intensity = data[offset] >> ((u % 2) ? 4 : 0);
intensity = Colour::convert4To8Bit(getBits<0, 4>(intensity));
// Intensity formats just copy the intensity value to every colour channel
@ -233,8 +219,7 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
case PICA::TextureFmt::I8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
auto ptr = static_cast<const u8*>(data);
const u8 intensity = ptr[offset];
const u8 intensity = data[offset];
// Intensity formats just copy the intensity value to every colour channel
return (0xff << 24) | (intensity << 16) | (intensity << 8) | intensity;
@ -242,11 +227,10 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
case PICA::TextureFmt::IA8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
auto ptr = static_cast<const u8*>(data);
// Same as I8 except each pixel gets its own alpha value too
const u8 alpha = ptr[offset];
const u8 intensity = ptr[offset + 1];
const u8 alpha = data[offset];
const u8 intensity = data[offset + 1];
return (alpha << 24) | (intensity << 16) | (intensity << 8) | intensity;
}
@ -258,7 +242,7 @@ u32 Texture::decodeTexel(u32 u, u32 v, PICA::TextureFmt fmt, const void* data) {
}
}
void Texture::decodeTexture(const void* data) {
void Texture::decodeTexture(std::span<const u8> data) {
std::vector<u32> decoded;
decoded.reserve(u64(size.u()) * u64(size.v()));
@ -272,4 +256,4 @@ void Texture::decodeTexture(const void* data) {
texture.bind();
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, size.u(), size.v(), GL_RGBA, GL_UNSIGNED_BYTE, decoded.data());
}
}

12
src/core/renderer_null/renderer_null.cpp Normal file
View file

@ -0,0 +1,12 @@
#include "renderer_null/renderer_null.hpp"
RendererNull::RendererNull(GPU& gpu, const std::array<u32, regNum>& internalRegs) : Renderer(gpu, internalRegs) {}
RendererNull::~RendererNull() {}
void RendererNull::reset() {}
void RendererNull::display() {}
void RendererNull::initGraphicsContext() {}
void RendererNull::clearBuffer(u32 startAddress, u32 endAddress, u32 value, u32 control) {}
void RendererNull::displayTransfer(u32 inputAddr, u32 outputAddr, u32 inputSize, u32 outputSize, u32 flags) {}
void RendererNull::drawVertices(PICA::PrimType primType, std::span<const PICA::Vertex> vertices) {}
void RendererNull::screenshot(const std::string& name) {}

179
src/emulator.cpp
View file

@ -1,6 +1,8 @@
#include "emulator.hpp"
#include <stb_image_write.h>
#ifdef PANDA3DS_ENABLE_OPENGL
#include <glad/gl.h>
#endif
#ifdef _WIN32
#include <windows.h>
@ -12,7 +14,9 @@ __declspec(dllexport) DWORD AmdPowerXpressRequestHighPerformance = 1;
}
#endif
Emulator::Emulator() : kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory, gl, config), memory(cpu.getTicksRef()) {
Emulator::Emulator()
: config(std::filesystem::current_path() / "config.toml"), kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory, config),
memory(cpu.getTicksRef()), cheats(memory, kernel.getServiceManager().getHID()) {
if (SDL_Init(SDL_INIT_VIDEO | SDL_INIT_EVENTS) < 0) {
Helpers::panic("Failed to initialize SDL2");
}
@ -23,25 +27,29 @@ Emulator::Emulator() : kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory
Helpers::warn("Failed to initialize SDL2 GameController: %s", SDL_GetError());
}
// Request OpenGL 4.1 Core (Max available on MacOS)
// MacOS gets mad if we don't explicitly demand a core profile
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 4);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 1);
window = SDL_CreateWindow("Alber", SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED, width, height, SDL_WINDOW_OPENGL);
#ifdef PANDA3DS_ENABLE_OPENGL
if (config.rendererType == RendererType::OpenGL) {
// Request OpenGL 4.1 Core (Max available on MacOS)
// MacOS gets mad if we don't explicitly demand a core profile
SDL_GL_SetAttribute(SDL_GL_CONTEXT_PROFILE_MASK, SDL_GL_CONTEXT_PROFILE_CORE);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MAJOR_VERSION, 4);
SDL_GL_SetAttribute(SDL_GL_CONTEXT_MINOR_VERSION, 1);
window = SDL_CreateWindow("Alber", SDL_WINDOWPOS_CENTERED, SDL_WINDOWPOS_CENTERED, width, height, SDL_WINDOW_OPENGL);
if (window == nullptr) {
Helpers::panic("Window creation failed: %s", SDL_GetError());
}
if (window == nullptr) {
Helpers::panic("Window creation failed: %s", SDL_GetError());
}
glContext = SDL_GL_CreateContext(window);
if (glContext == nullptr) {
Helpers::panic("OpenGL context creation failed: %s", SDL_GetError());
}
glContext = SDL_GL_CreateContext(window);
if (glContext == nullptr) {
Helpers::panic("OpenGL context creation failed: %s", SDL_GetError());
}
if (!gladLoadGL(reinterpret_cast<GLADloadfunc>(SDL_GL_GetProcAddress))) {
Helpers::panic("OpenGL init failed: %s", SDL_GetError());
if (!gladLoadGL(reinterpret_cast<GLADloadfunc>(SDL_GL_GetProcAddress))) {
Helpers::panic("OpenGL init failed: %s", SDL_GetError());
}
}
#endif
if (SDL_WasInit(SDL_INIT_GAMECONTROLLER)) {
gameController = SDL_GameControllerOpen(0);
@ -52,7 +60,6 @@ Emulator::Emulator() : kernel(cpu, memory, gpu), cpu(memory, kernel), gpu(memory
}
}
config.load(std::filesystem::current_path() / "config.toml");
reset(ReloadOption::NoReload);
}
@ -69,6 +76,12 @@ void Emulator::reset(ReloadOption reload) {
// Otherwise resetting the kernel or cpu might nuke them
cpu.setReg(13, VirtualAddrs::StackTop); // Set initial SP
// We're resetting without reloading the ROM, so yeet cheats
if (reload == ReloadOption::NoReload) {
haveCheats = false;
cheats.reset();
}
// If a ROM is active and we reset, with the reload option enabled then reload it.
// This is necessary to set up stack, executable memory, .data/.rodata/.bss all over again
if (reload == ReloadOption::Reload && romType != ROMType::None && romPath.has_value()) {
@ -91,19 +104,8 @@ void Emulator::run() {
#endif
while (running) {
ServiceManager& srv = kernel.getServiceManager();
if (romType != ROMType::None) {
#ifdef PANDA3DS_ENABLE_HTTP_SERVER
pollHttpServer();
#endif
runFrame(); // Run 1 frame of instructions
gpu.display(); // Display graphics
// Send VBlank interrupts
srv.sendGPUInterrupt(GPUInterrupt::VBlank0);
srv.sendGPUInterrupt(GPUInterrupt::VBlank1);
}
runFrame();
HIDService& hid = kernel.getServiceManager().getHID();
SDL_Event event;
while (SDL_PollEvent(&event)) {
@ -119,41 +121,41 @@ void Emulator::run() {
if (romType == ROMType::None) break;
switch (event.key.keysym.sym) {
case SDLK_l: srv.pressKey(Keys::A); break;
case SDLK_k: srv.pressKey(Keys::B); break;
case SDLK_o: srv.pressKey(Keys::X); break;
case SDLK_i: srv.pressKey(Keys::Y); break;
case SDLK_l: hid.pressKey(Keys::A); break;
case SDLK_k: hid.pressKey(Keys::B); break;
case SDLK_o: hid.pressKey(Keys::X); break;
case SDLK_i: hid.pressKey(Keys::Y); break;
case SDLK_q: srv.pressKey(Keys::L); break;
case SDLK_p: srv.pressKey(Keys::R); break;
case SDLK_q: hid.pressKey(Keys::L); break;
case SDLK_p: hid.pressKey(Keys::R); break;
case SDLK_RIGHT: srv.pressKey(Keys::Right); break;
case SDLK_LEFT: srv.pressKey(Keys::Left); break;
case SDLK_UP: srv.pressKey(Keys::Up); break;
case SDLK_DOWN: srv.pressKey(Keys::Down); break;
case SDLK_RIGHT: hid.pressKey(Keys::Right); break;
case SDLK_LEFT: hid.pressKey(Keys::Left); break;
case SDLK_UP: hid.pressKey(Keys::Up); break;
case SDLK_DOWN: hid.pressKey(Keys::Down); break;
case SDLK_w:
srv.setCirclepadY(0x9C);
hid.setCirclepadY(0x9C);
keyboardAnalogY = true;
break;
case SDLK_a:
srv.setCirclepadX(-0x9C);
hid.setCirclepadX(-0x9C);
keyboardAnalogX = true;
break;
case SDLK_s:
srv.setCirclepadY(-0x9C);
hid.setCirclepadY(-0x9C);
keyboardAnalogY = true;
break;
case SDLK_d:
srv.setCirclepadX(0x9C);
hid.setCirclepadX(0x9C);
keyboardAnalogX = true;
break;
case SDLK_RETURN: srv.pressKey(Keys::Start); break;
case SDLK_BACKSPACE: srv.pressKey(Keys::Select); break;
case SDLK_RETURN: hid.pressKey(Keys::Start); break;
case SDLK_BACKSPACE: hid.pressKey(Keys::Select); break;
}
break;
@ -161,34 +163,34 @@ void Emulator::run() {
if (romType == ROMType::None) break;
switch (event.key.keysym.sym) {
case SDLK_l: srv.releaseKey(Keys::A); break;
case SDLK_k: srv.releaseKey(Keys::B); break;
case SDLK_o: srv.releaseKey(Keys::X); break;
case SDLK_i: srv.releaseKey(Keys::Y); break;
case SDLK_l: hid.releaseKey(Keys::A); break;
case SDLK_k: hid.releaseKey(Keys::B); break;
case SDLK_o: hid.releaseKey(Keys::X); break;
case SDLK_i: hid.releaseKey(Keys::Y); break;
case SDLK_q: srv.releaseKey(Keys::L); break;
case SDLK_p: srv.releaseKey(Keys::R); break;
case SDLK_q: hid.releaseKey(Keys::L); break;
case SDLK_p: hid.releaseKey(Keys::R); break;
case SDLK_RIGHT: srv.releaseKey(Keys::Right); break;
case SDLK_LEFT: srv.releaseKey(Keys::Left); break;
case SDLK_UP: srv.releaseKey(Keys::Up); break;
case SDLK_DOWN: srv.releaseKey(Keys::Down); break;
case SDLK_RIGHT: hid.releaseKey(Keys::Right); break;
case SDLK_LEFT: hid.releaseKey(Keys::Left); break;
case SDLK_UP: hid.releaseKey(Keys::Up); break;
case SDLK_DOWN: hid.releaseKey(Keys::Down); break;
// Err this is probably not ideal
case SDLK_w:
case SDLK_s:
srv.setCirclepadY(0);
hid.setCirclepadY(0);
keyboardAnalogY = false;
break;
case SDLK_a:
case SDLK_d:
srv.setCirclepadX(0);
hid.setCirclepadX(0);
keyboardAnalogX = false;
break;
case SDLK_RETURN: srv.releaseKey(Keys::Start); break;
case SDLK_BACKSPACE: srv.releaseKey(Keys::Select); break;
case SDLK_RETURN: hid.releaseKey(Keys::Start); break;
case SDLK_BACKSPACE: hid.releaseKey(Keys::Select); break;
}
break;
@ -205,9 +207,9 @@ void Emulator::run() {
u16 x_converted = static_cast<u16>(x) - 40;
u16 y_converted = static_cast<u16>(y) - 240;
srv.setTouchScreenPress(x_converted, y_converted);
hid.setTouchScreenPress(x_converted, y_converted);
} else {
srv.releaseTouchScreen();
hid.releaseTouchScreen();
}
} else if (event.button.button == SDL_BUTTON_RIGHT) {
holdingRightClick = true;
@ -219,7 +221,7 @@ void Emulator::run() {
if (romType == ROMType::None) break;
if (event.button.button == SDL_BUTTON_LEFT) {
srv.releaseTouchScreen();
hid.releaseTouchScreen();
} else if (event.button.button == SDL_BUTTON_RIGHT) {
holdingRightClick = false;
}
@ -262,9 +264,9 @@ void Emulator::run() {
if (key != 0) {
if (event.cbutton.state == SDL_PRESSED) {
srv.pressKey(key);
hid.pressKey(key);
} else {
srv.releaseKey(key);
hid.releaseKey(key);
}
}
break;
@ -283,8 +285,8 @@ void Emulator::run() {
// So up until then, we will set the gyroscope euler angles to fixed values based on the direction of the relative motion
const s32 roll = motionX > 0 ? 0x7f : -0x7f;
const s32 pitch = motionY > 0 ? 0x7f : -0x7f;
srv.setRoll(roll);
srv.setPitch(pitch);
hid.setRoll(roll);
hid.setPitch(pitch);
break;
}
@ -311,19 +313,19 @@ void Emulator::run() {
// Avoid overriding the keyboard's circlepad input
if (abs(stickX) < deadzone && !keyboardAnalogX) {
srv.setCirclepadX(0);
hid.setCirclepadX(0);
} else {
srv.setCirclepadX(stickX / div);
hid.setCirclepadX(stickX / div);
}
if (abs(stickY) < deadzone && !keyboardAnalogY) {
srv.setCirclepadY(0);
hid.setCirclepadY(0);
} else {
srv.setCirclepadY(-(stickY / div));
hid.setCirclepadY(-(stickY / div));
}
}
srv.updateInputs(cpu.getTicks());
hid.updateInputs(cpu.getTicks());
}
// Update inputs in the HID module
@ -331,7 +333,24 @@ void Emulator::run() {
}
}
void Emulator::runFrame() { cpu.runFrame(); }
void Emulator::runFrame() {
if (romType != ROMType::None) {
#ifdef PANDA3DS_ENABLE_HTTP_SERVER
pollHttpServer();
#endif
cpu.runFrame(); // Run 1 frame of instructions
gpu.display(); // Display graphics
// Send VBlank interrupts
ServiceManager& srv = kernel.getServiceManager();
srv.sendGPUInterrupt(GPUInterrupt::VBlank0);
srv.sendGPUInterrupt(GPUInterrupt::VBlank1);
if (haveCheats) [[unlikely]] {
cheats.run();
}
}
}
bool Emulator::loadROM(const std::filesystem::path& path) {
// Reset the emulator if we've already loaded a ROM
@ -427,15 +446,13 @@ bool Emulator::loadELF(std::ifstream& file) {
}
// Reset our graphics context and initialize the GPU's graphics context
void Emulator::initGraphicsContext() {
gl.reset(); // TODO (For when we have multiple backends): Only do this if we are using OpenGL
gpu.initGraphicsContext();
}
void Emulator::initGraphicsContext() { gpu.initGraphicsContext(); }
#ifdef PANDA3DS_ENABLE_HTTP_SERVER
void Emulator::pollHttpServer() {
std::scoped_lock lock(httpServer.actionMutex);
ServiceManager& srv = kernel.getServiceManager();
HIDService& hid = kernel.getServiceManager().getHID();
if (httpServer.pendingAction) {
switch (httpServer.action) {
@ -443,14 +460,14 @@ void Emulator::pollHttpServer() {
case HttpAction::PressKey:
if (httpServer.pendingKey != 0) {
srv.pressKey(httpServer.pendingKey);
hid.pressKey(httpServer.pendingKey);
httpServer.pendingKey = 0;
}
break;
case HttpAction::ReleaseKey:
if (httpServer.pendingKey != 0) {
srv.releaseKey(httpServer.pendingKey);
hid.releaseKey(httpServer.pendingKey);
httpServer.pendingKey = 0;
}
break;

8
src/host_shaders/opengl_display.frag Normal file
View file

@ -0,0 +1,8 @@
#version 410 core
in vec2 UV;
out vec4 FragColor;
uniform sampler2D u_texture;
void main() {
FragColor = texture(u_texture, UV);
}

23
src/host_shaders/opengl_display.vert Normal file
View file

@ -0,0 +1,23 @@
#version 410 core
out vec2 UV;
void main() {
const vec4 positions[4] = vec4[](
vec4(-1.0, 1.0, 1.0, 1.0), // Top-left
vec4(1.0, 1.0, 1.0, 1.0), // Top-right
vec4(-1.0, -1.0, 1.0, 1.0), // Bottom-left
vec4(1.0, -1.0, 1.0, 1.0) // Bottom-right
);
// The 3DS displays both screens' framebuffer rotated 90 deg counter clockwise
// So we adjust our texcoords accordingly
const vec2 texcoords[4] = vec2[](
vec2(1.0, 1.0), // Top-right
vec2(1.0, 0.0), // Bottom-right
vec2(0.0, 1.0), // Top-left
vec2(0.0, 0.0) // Bottom-left
);
gl_Position = positions[gl_VertexID];
UV = texcoords[gl_VertexID];
}

409
src/host_shaders/opengl_fragment_shader.frag Normal file
View file

@ -0,0 +1,409 @@
#version 410 core
in vec3 v_tangent;
in vec3 v_normal;
in vec3 v_bitangent;
in vec4 v_colour;
in vec3 v_texcoord0;
in vec2 v_texcoord1;
in vec3 v_view;
in vec2 v_texcoord2;
flat in vec4 v_textureEnvColor[6];
flat in vec4 v_textureEnvBufferColor;
out vec4 fragColour;
// TEV uniforms
uniform uint u_textureEnvSource[6];
uniform uint u_textureEnvOperand[6];
uniform uint u_textureEnvCombiner[6];
uniform uint u_textureEnvScale[6];
// Depth control uniforms
uniform float u_depthScale;
uniform float u_depthOffset;
uniform bool u_depthmapEnable;
uniform sampler2D u_tex0;
uniform sampler2D u_tex1;
uniform sampler2D u_tex2;
uniform sampler1DArray u_tex_lighting_lut;
uniform uint u_picaRegs[0x200 - 0x48];
// Helper so that the implementation of u_pica_regs can be changed later
uint readPicaReg(uint reg_addr) { return u_picaRegs[reg_addr - 0x48]; }
vec4 tevSources[16];
vec4 tevNextPreviousBuffer;
bool tevUnimplementedSourceFlag = false;
// OpenGL ES 1.1 reference pages for TEVs (this is what the PICA200 implements):
// https://registry.khronos.org/OpenGL-Refpages/es1.1/xhtml/glTexEnv.xml
vec4 tevFetchSource(uint src_id) {
if (src_id >= 6u && src_id < 13u) {
tevUnimplementedSourceFlag = true;
}
return tevSources[src_id];
}
vec4 tevGetColorAndAlphaSource(int tev_id, int src_id) {
vec4 result;
vec4 colorSource = tevFetchSource((u_textureEnvSource[tev_id] >> (src_id * 4)) & 15u);
vec4 alphaSource = tevFetchSource((u_textureEnvSource[tev_id] >> (src_id * 4 + 16)) & 15u);
uint colorOperand = (u_textureEnvOperand[tev_id] >> (src_id * 4)) & 15u;
uint alphaOperand = (u_textureEnvOperand[tev_id] >> (12 + src_id * 4)) & 7u;
// TODO: figure out what the undocumented values do
switch (colorOperand) {
case 0u: result.rgb = colorSource.rgb; break; // Source color
case 1u: result.rgb = 1.0 - colorSource.rgb; break; // One minus source color
case 2u: result.rgb = vec3(colorSource.a); break; // Source alpha
case 3u: result.rgb = vec3(1.0 - colorSource.a); break; // One minus source alpha
case 4u: result.rgb = vec3(colorSource.r); break; // Source red
case 5u: result.rgb = vec3(1.0 - colorSource.r); break; // One minus source red
case 8u: result.rgb = vec3(colorSource.g); break; // Source green
case 9u: result.rgb = vec3(1.0 - colorSource.g); break; // One minus source green
case 12u: result.rgb = vec3(colorSource.b); break; // Source blue
case 13u: result.rgb = vec3(1.0 - colorSource.b); break; // One minus source blue
default: break;
}
// TODO: figure out what the undocumented values do
switch (alphaOperand) {
case 0u: result.a = alphaSource.a; break; // Source alpha
case 1u: result.a = 1.0 - alphaSource.a; break; // One minus source alpha
case 2u: result.a = alphaSource.r; break; // Source red
case 3u: result.a = 1.0 - alphaSource.r; break; // One minus source red
case 4u: result.a = alphaSource.g; break; // Source green
case 5u: result.a = 1.0 - alphaSource.g; break; // One minus source green
case 6u: result.a = alphaSource.b; break; // Source blue
case 7u: result.a = 1.0 - alphaSource.b; break; // One minus source blue
default: break;
}
return result;
}
vec4 tevCalculateCombiner(int tev_id) {
vec4 source0 = tevGetColorAndAlphaSource(tev_id, 0);
vec4 source1 = tevGetColorAndAlphaSource(tev_id, 1);
vec4 source2 = tevGetColorAndAlphaSource(tev_id, 2);
uint colorCombine = u_textureEnvCombiner[tev_id] & 15u;
uint alphaCombine = (u_textureEnvCombiner[tev_id] >> 16) & 15u;
vec4 result = vec4(1.0);
// TODO: figure out what the undocumented values do
switch (colorCombine) {
case 0u: result.rgb = source0.rgb; break; // Replace
case 1u: result.rgb = source0.rgb * source1.rgb; break; // Modulate
case 2u: result.rgb = min(vec3(1.0), source0.rgb + source1.rgb); break; // Add
case 3u: result.rgb = clamp(source0.rgb + source1.rgb - 0.5, 0.0, 1.0); break; // Add signed
case 4u: result.rgb = mix(source1.rgb, source0.rgb, source2.rgb); break; // Interpolate
case 5u: result.rgb = max(source0.rgb - source1.rgb, 0.0); break; // Subtract
case 6u: result.rgb = vec3(4.0 * dot(source0.rgb - 0.5, source1.rgb - 0.5)); break; // Dot3 RGB
case 7u: result = vec4(4.0 * dot(source0.rgb - 0.5, source1.rgb - 0.5)); break; // Dot3 RGBA
case 8u: result.rgb = min(source0.rgb * source1.rgb + source2.rgb, 1.0); break; // Multiply then add
case 9u: result.rgb = min((source0.rgb + source1.rgb) * source2.rgb, 1.0); break; // Add then multiply
default: break;
}
if (colorCombine != 7u) { // The color combiner also writes the alpha channel in the "Dot3 RGBA" mode.
// TODO: figure out what the undocumented values do
// TODO: test if the alpha combiner supports all the same modes as the color combiner.
switch (alphaCombine) {
case 0u: result.a = source0.a; break; // Replace
case 1u: result.a = source0.a * source1.a; break; // Modulate
case 2u: result.a = min(1.0, source0.a + source1.a); break; // Add
case 3u: result.a = clamp(source0.a + source1.a - 0.5, 0.0, 1.0); break; // Add signed
case 4u: result.a = mix(source1.a, source0.a, source2.a); break; // Interpolate
case 5u: result.a = max(0.0, source0.a - source1.a); break; // Subtract
case 8u: result.a = min(1.0, source0.a * source1.a + source2.a); break; // Multiply then add
case 9u: result.a = min(1.0, (source0.a + source1.a) * source2.a); break; // Add then multiply
default: break;
}
}
result.rgb *= float(1 << (u_textureEnvScale[tev_id] & 3u));
result.a *= float(1 << ((u_textureEnvScale[tev_id] >> 16) & 3u));
return result;
}
#define D0_LUT 0u
#define D1_LUT 1u
#define SP_LUT 2u
#define FR_LUT 3u
#define RB_LUT 4u
#define RG_LUT 5u
#define RR_LUT 6u
float lutLookup(uint lut, uint light, float value) {
if (lut >= FR_LUT && lut <= RR_LUT) lut -= 1;
if (lut == SP_LUT) lut = light + 8;
return texture(u_tex_lighting_lut, vec2(value, lut)).r;
}
vec3 regToColor(uint reg) {
// Normalization scale to convert from [0...255] to [0.0...1.0]
const float scale = 1.0 / 255.0;
return scale * vec3(float(bitfieldExtract(reg, 20, 8)), float(bitfieldExtract(reg, 10, 8)), float(bitfieldExtract(reg, 00, 8)));
}
// Convert an arbitrary-width floating point literal to an f32
float decodeFP(uint hex, uint E, uint M) {
uint width = M + E + 1u;
uint bias = 128u - (1u << (E - 1u));
uint exponent = (hex >> M) & ((1u << E) - 1u);
uint mantissa = hex & ((1u << M) - 1u);
uint sign = (hex >> (E + M)) << 31u;
if ((hex & ((1u << (width - 1u)) - 1u)) != 0) {
if (exponent == (1u << E) - 1u)
exponent = 255u;
else
exponent += bias;
hex = sign | (mantissa << (23u - M)) | (exponent << 23u);
} else {
hex = sign;
}
return uintBitsToFloat(hex);
}
// Implements the following algorthm: https://mathb.in/26766
void calcLighting(out vec4 primary_color, out vec4 secondary_color) {
// Quaternions describe a transformation from surface-local space to eye space.
// In surface-local space, by definition (and up to permutation) the normal vector is (0,0,1),
// the tangent vector is (1,0,0), and the bitangent vector is (0,1,0).
vec3 normal = normalize(v_normal);
vec3 tangent = normalize(v_tangent);
vec3 bitangent = normalize(v_bitangent);
vec3 view = normalize(v_view);
uint GPUREG_LIGHTING_ENABLE = readPicaReg(0x008F);
if (bitfieldExtract(GPUREG_LIGHTING_ENABLE, 0, 1) == 0) {
primary_color = secondary_color = vec4(1.0);
return;
}
uint GPUREG_LIGHTING_AMBIENT = readPicaReg(0x01C0);
uint GPUREG_LIGHTING_NUM_LIGHTS = (readPicaReg(0x01C2) & 0x7u) + 1;
uint GPUREG_LIGHTING_LIGHT_PERMUTATION = readPicaReg(0x01D9);
primary_color = vec4(vec3(0.0), 1.0);
secondary_color = vec4(vec3(0.0), 1.0);
primary_color.rgb += regToColor(GPUREG_LIGHTING_AMBIENT);
uint GPUREG_LIGHTING_LUTINPUT_ABS = readPicaReg(0x01D0);
uint GPUREG_LIGHTING_LUTINPUT_SELECT = readPicaReg(0x01D1);
uint GPUREG_LIGHTING_CONFIG0 = readPicaReg(0x01C3);
uint GPUREG_LIGHTING_CONFIG1 = readPicaReg(0x01C4);
uint GPUREG_LIGHTING_LUTINPUT_SCALE = readPicaReg(0x01D2);
float d[7];
bool error_unimpl = false;
for (uint i = 0; i < GPUREG_LIGHTING_NUM_LIGHTS; i++) {
uint light_id = bitfieldExtract(GPUREG_LIGHTING_LIGHT_PERMUTATION, int(i * 3), 3);
uint GPUREG_LIGHTi_SPECULAR0 = readPicaReg(0x0140 + 0x10 * light_id);
uint GPUREG_LIGHTi_SPECULAR1 = readPicaReg(0x0141 + 0x10 * light_id);
uint GPUREG_LIGHTi_DIFFUSE = readPicaReg(0x0142 + 0x10 * light_id);
uint GPUREG_LIGHTi_AMBIENT = readPicaReg(0x0143 + 0x10 * light_id);
uint GPUREG_LIGHTi_VECTOR_LOW = readPicaReg(0x0144 + 0x10 * light_id);
uint GPUREG_LIGHTi_VECTOR_HIGH = readPicaReg(0x0145 + 0x10 * light_id);
uint GPUREG_LIGHTi_CONFIG = readPicaReg(0x0149 + 0x10 * light_id);
vec3 light_vector = normalize(vec3(
decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_LOW, 0, 16), 5, 10), decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_LOW, 16, 16), 5, 10),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_VECTOR_HIGH, 0, 16), 5, 10)
));
// Positional Light
if (bitfieldExtract(GPUREG_LIGHTi_CONFIG, 0, 1) == 0) error_unimpl = true;
vec3 half_vector = normalize(normalize(light_vector) + view);
for (int c = 0; c < 7; c++) {
if (bitfieldExtract(GPUREG_LIGHTING_CONFIG1, 16 + c, 1) == 0) {
uint scale_id = bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_SCALE, c * 4, 3);
float scale = float(1u << scale_id);
if (scale_id >= 6u) scale /= 256.0;
uint input_id = bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_SELECT, c * 4, 3);
if (input_id == 0u)
d[c] = dot(normal, half_vector);
else if (input_id == 1u)
d[c] = dot(view, half_vector);
else if (input_id == 2u)
d[c] = dot(normal, view);
else if (input_id == 3u)
d[c] = dot(light_vector, normal);
else if (input_id == 4u) {
uint GPUREG_LIGHTi_SPOTDIR_LOW = readPicaReg(0x0146 + 0x10 * light_id);
uint GPUREG_LIGHTi_SPOTDIR_HIGH = readPicaReg(0x0147 + 0x10 * light_id);
vec3 spot_light_vector = normalize(vec3(
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_LOW, 0, 16), 1, 11),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_LOW, 16, 16), 1, 11),
decodeFP(bitfieldExtract(GPUREG_LIGHTi_SPOTDIR_HIGH, 0, 16), 1, 11)
));
d[c] = dot(-light_vector, spot_light_vector); // -L dot P (aka Spotlight aka SP);
} else if (input_id == 5u) {
d[c] = 1.0; // TODO: cos <greek symbol> (aka CP);
error_unimpl = true;
} else {
d[c] = 1.0;
}
d[c] = lutLookup(c, light_id, d[c] * 0.5 + 0.5) * scale;
if (bitfieldExtract(GPUREG_LIGHTING_LUTINPUT_ABS, 2 * c, 1) != 0u) d[c] = abs(d[c]);
} else {
d[c] = 1.0;
}
}
uint lookup_config = bitfieldExtract(GPUREG_LIGHTi_CONFIG, 4, 4);
if (lookup_config == 0) {
d[D1_LUT] = 0.0;
d[FR_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 1) {
d[D0_LUT] = 0.0;
d[D1_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 2) {
d[FR_LUT] = 0.0;
d[SP_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
} else if (lookup_config == 3) {
d[SP_LUT] = 0.0;
d[RG_LUT] = d[RB_LUT] = d[RR_LUT] = 1.0;
} else if (lookup_config == 4) {
d[FR_LUT] = 0.0;
} else if (lookup_config == 5) {
d[D1_LUT] = 0.0;
} else if (lookup_config == 6) {
d[RG_LUT] = d[RB_LUT] = d[RR_LUT];
}
float distance_factor = 1.0; // a
float indirect_factor = 1.0; // fi
float shadow_factor = 1.0; // o
float NdotL = dot(normal, light_vector); // Li dot N
// Two sided diffuse
if (bitfieldExtract(GPUREG_LIGHTi_CONFIG, 1, 1) == 0)
NdotL = max(0.0, NdotL);
else
NdotL = abs(NdotL);
float light_factor = distance_factor * d[SP_LUT] * indirect_factor * shadow_factor;
primary_color.rgb += light_factor * (regToColor(GPUREG_LIGHTi_AMBIENT) + regToColor(GPUREG_LIGHTi_DIFFUSE) * NdotL);
secondary_color.rgb += light_factor * (regToColor(GPUREG_LIGHTi_SPECULAR0) * d[D0_LUT] +
regToColor(GPUREG_LIGHTi_SPECULAR1) * d[D1_LUT] * vec3(d[RR_LUT], d[RG_LUT], d[RB_LUT]));
}
uint fresnel_output1 = bitfieldExtract(GPUREG_LIGHTING_CONFIG0, 2, 1);
uint fresnel_output2 = bitfieldExtract(GPUREG_LIGHTING_CONFIG0, 3, 1);
if (fresnel_output1 == 1u) primary_color.a = d[FR_LUT];
if (fresnel_output2 == 1u) secondary_color.a = d[FR_LUT];
if (error_unimpl) {
secondary_color = primary_color = vec4(1.0, 0., 1.0, 1.0);
}
}
void main() {
// TODO: what do invalid sources and disabled textures read as?
// And what does the "previous combiner" source read initially?
tevSources[0] = v_colour; // Primary/vertex color
calcLighting(tevSources[1], tevSources[2]);
uint textureConfig = readPicaReg(0x80);
vec2 tex2UV = (textureConfig & (1u << 13)) != 0u ? v_texcoord1 : v_texcoord2;
if ((textureConfig & 1u) != 0u) tevSources[3] = texture(u_tex0, v_texcoord0.xy);
if ((textureConfig & 2u) != 0u) tevSources[4] = texture(u_tex1, v_texcoord1);
if ((textureConfig & 4u) != 0u) tevSources[5] = texture(u_tex2, tex2UV);
tevSources[13] = vec4(0.0); // Previous buffer
tevSources[15] = vec4(0.0); // Previous combiner
tevNextPreviousBuffer = v_textureEnvBufferColor;
uint textureEnvUpdateBuffer = readPicaReg(0xE0);
for (int i = 0; i < 6; i++) {
tevSources[14] = v_textureEnvColor[i]; // Constant color
tevSources[15] = tevCalculateCombiner(i);
tevSources[13] = tevNextPreviousBuffer;
if (i < 4) {
if ((textureEnvUpdateBuffer & (0x100u << i)) != 0u) {
tevNextPreviousBuffer.rgb = tevSources[15].rgb;
}
if ((textureEnvUpdateBuffer & (0x1000u << i)) != 0u) {
tevNextPreviousBuffer.a = tevSources[15].a;
}
}
}
fragColour = tevSources[15];
if (tevUnimplementedSourceFlag) {
// fragColour = vec4(1.0, 0.0, 1.0, 1.0);
}
// fragColour.rg = texture(u_tex_lighting_lut,vec2(gl_FragCoord.x/200.,float(int(gl_FragCoord.y/2)%24))).rr;
// Get original depth value by converting from [near, far] = [0, 1] to [-1, 1]
// We do this by converting to [0, 2] first and subtracting 1 to go to [-1, 1]
float z_over_w = gl_FragCoord.z * 2.0f - 1.0f;
float depth = z_over_w * u_depthScale + u_depthOffset;
if (!u_depthmapEnable) // Divide z by w if depthmap enable == 0 (ie using W-buffering)
depth /= gl_FragCoord.w;
// Write final fragment depth
gl_FragDepth = depth;
// Perform alpha test
uint alphaControl = readPicaReg(0x104);
if ((alphaControl & 1u) != 0u) { // Check if alpha test is on
uint func = (alphaControl >> 4u) & 7u;
float reference = float((alphaControl >> 8u) & 0xffu) / 255.0;
float alpha = fragColour.a;
switch (func) {
case 0: discard; // Never pass alpha test
case 1: break; // Always pass alpha test
case 2: // Pass if equal
if (alpha != reference) discard;
break;
case 3: // Pass if not equal
if (alpha == reference) discard;
break;
case 4: // Pass if less than
if (alpha >= reference) discard;
break;
case 5: // Pass if less than or equal
if (alpha > reference) discard;
break;
case 6: // Pass if greater than
if (alpha <= reference) discard;
break;
case 7: // Pass if greater than or equal
if (alpha < reference) discard;
break;
}
}
}

97
src/host_shaders/opengl_vertex_shader.vert Normal file
View file

@ -0,0 +1,97 @@
#version 410 core
layout(location = 0) in vec4 a_coords;
layout(location = 1) in vec4 a_quaternion;
layout(location = 2) in vec4 a_vertexColour;
layout(location = 3) in vec2 a_texcoord0;
layout(location = 4) in vec2 a_texcoord1;
layout(location = 5) in float a_texcoord0_w;
layout(location = 6) in vec3 a_view;
layout(location = 7) in vec2 a_texcoord2;
out vec3 v_normal;
out vec3 v_tangent;
out vec3 v_bitangent;
out vec4 v_colour;
out vec3 v_texcoord0;
out vec2 v_texcoord1;
out vec3 v_view;
out vec2 v_texcoord2;
flat out vec4 v_textureEnvColor[6];
flat out vec4 v_textureEnvBufferColor;
out float gl_ClipDistance[2];
// TEV uniforms
uniform uint u_textureEnvColor[6];
uniform uint u_picaRegs[0x200 - 0x48];
// Helper so that the implementation of u_pica_regs can be changed later
uint readPicaReg(uint reg_addr) { return u_picaRegs[reg_addr - 0x48]; }
vec4 abgr8888ToVec4(uint abgr) {
const float scale = 1.0 / 255.0;
return scale * vec4(float(abgr & 0xffu), float((abgr >> 8) & 0xffu), float((abgr >> 16) & 0xffu), float(abgr >> 24));
}
vec3 rotateVec3ByQuaternion(vec3 v, vec4 q) {
vec3 u = q.xyz;
float s = q.w;
return 2.0 * dot(u, v) * u + (s * s - dot(u, u)) * v + 2.0 * s * cross(u, v);
}
// Convert an arbitrary-width floating point literal to an f32
float decodeFP(uint hex, uint E, uint M) {
uint width = M + E + 1u;
uint bias = 128u - (1u << (E - 1u));
uint exponent = (hex >> M) & ((1u << E) - 1u);
uint mantissa = hex & ((1u << M) - 1u);
uint sign = (hex >> (E + M)) << 31u;
if ((hex & ((1u << (width - 1u)) - 1u)) != 0) {
if (exponent == (1u << E) - 1u)
exponent = 255u;
else
exponent += bias;
hex = sign | (mantissa << (23u - M)) | (exponent << 23u);
} else {
hex = sign;
}
return uintBitsToFloat(hex);
}
void main() {
gl_Position = a_coords;
v_colour = a_vertexColour;
// Flip y axis of UVs because OpenGL uses an inverted y for texture sampling compared to the PICA
v_texcoord0 = vec3(a_texcoord0.x, 1.0 - a_texcoord0.y, a_texcoord0_w);
v_texcoord1 = vec2(a_texcoord1.x, 1.0 - a_texcoord1.y);
v_texcoord2 = vec2(a_texcoord2.x, 1.0 - a_texcoord2.y);
v_view = a_view;
v_normal = normalize(rotateVec3ByQuaternion(vec3(0.0, 0.0, 1.0), a_quaternion));
v_tangent = normalize(rotateVec3ByQuaternion(vec3(1.0, 0.0, 0.0), a_quaternion));
v_bitangent = normalize(rotateVec3ByQuaternion(vec3(0.0, 1.0, 0.0), a_quaternion));
for (int i = 0; i < 6; i++) {
v_textureEnvColor[i] = abgr8888ToVec4(u_textureEnvColor[i]);
}
v_textureEnvBufferColor = abgr8888ToVec4(readPicaReg(0xFD));
// Parse clipping plane registers
// The plane registers describe a clipping plane in the form of Ax + By + Cz + D = 0
// With n = (A, B, C) being the normal vector and D being the origin point distance
// Therefore, for the second clipping plane, we can just pass the dot product of the clip vector and the input coordinates to gl_ClipDistance[1]
vec4 clipData = vec4(
decodeFP(readPicaReg(0x48) & 0xffffffu, 7, 16), decodeFP(readPicaReg(0x49) & 0xffffffu, 7, 16),
decodeFP(readPicaReg(0x4A) & 0xffffffu, 7, 16), decodeFP(readPicaReg(0x4B) & 0xffffffu, 7, 16)
);
// There's also another, always-on clipping plane based on vertex z
gl_ClipDistance[0] = -a_coords.z;
gl_ClipDistance[1] = dot(clipData, a_coords);
}

6
src/main.cpp
View file

@ -1,9 +1,9 @@
#include "emulator.hpp"
int main (int argc, char *argv[]) {
Emulator emu;
int main(int argc, char *argv[]) {
Emulator emu;
emu.initGraphicsContext();
emu.initGraphicsContext();
if (argc > 1) {
auto romPath = std::filesystem::current_path() / argv[1];

35
src/renderer.cpp Normal file
View file

@ -0,0 +1,35 @@
#include "renderer.hpp"
#include <algorithm>
#include <unordered_map>
Renderer::Renderer(GPU& gpu, const std::array<u32, regNum>& internalRegs) : gpu(gpu), regs(internalRegs) {}
Renderer::~Renderer() {}
std::optional<RendererType> Renderer::typeFromString(std::string inString) {
// Transform to lower-case to make the setting case-insensitive
std::transform(inString.begin(), inString.end(), inString.begin(), [](unsigned char c) { return std::tolower(c); });
// Huge table of possible names and misspellings
// Please stop misspelling Vulkan as Vulcan
static const std::unordered_map<std::string, RendererType> map = {
{"null", RendererType::Null}, {"nil", RendererType::Null}, {"none", RendererType::Null},
{"gl", RendererType::OpenGL}, {"ogl", RendererType::OpenGL}, {"opengl", RendererType::OpenGL},
{"vk", RendererType::Vulkan}, {"vulkan", RendererType::Vulkan}, {"vulcan", RendererType::Vulkan},
};
if (auto search = map.find(inString); search != map.end()) {
return search->second;
}
return std::nullopt;
}
const char* Renderer::typeToString(RendererType rendererType) {
switch (rendererType) {
case RendererType::Null: return "null";
case RendererType::OpenGL: return "opengl";
case RendererType::Vulkan: return "vulkan";
default: return "Invalid";
}
}