moonpower/P3DS-test
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #745 from wheremyfoodat/ios

Browse Source 
iOS driver & Metal renderer improvements
This commit is contained in:
wheremyfoodat
2025-03-16 17:12:51 +02:00
committed by GitHub
parent da9dd8522a 449c14093d
commit 761f9264ba
16 changed files with 587 additions and 384 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/config.cpp
View File

@@ -72,14 +72,14 @@ void EmulatorConfig::load() {
auto gpu = gpuResult.unwrap();
// Get renderer
auto rendererName = toml::find_or<std::string>(gpu, "Renderer", "OpenGL");
auto rendererName = toml::find_or<std::string>(gpu, "Renderer", Renderer::typeToString(rendererDefault));
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;
rendererType = rendererDefault;
}
shaderJitEnabled = toml::find_or<toml::boolean>(gpu, "EnableShaderJIT", shaderJitDefault);

116
src/core/renderer_mtl/mtl_etc1.cpp
View File

@@ -1,116 +0,0 @@
#include <algorithm>
#include "colour.hpp"
#include "renderer_mtl/mtl_texture.hpp"
#include "renderer_mtl/renderer_mtl.hpp"
using namespace Helpers;
namespace Metal {
static constexpr u32 signExtend3To32(u32 val) {
return (u32)(s32(val) << 29 >> 29);
}
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) {
offs >>= 1;
}
// In-tile offsets for u/v
u &= 7;
v &= 7;
// ETC1(A4) also subdivide the 8x8 tile to 4 4x4 tiles
// Each tile is 8 bytes for ETC1, but since ETC1A4 has 4 alpha bits per pixel, that becomes 16 bytes
const u32 subTileSize = hasAlpha ? 16 : 8;
const u32 subTileIndex = (u / 4) + 2 * (v / 4); // Which of the 4 subtiles is this texel in?
// In-subtile offsets for u/v
u &= 3;
v &= 3;
offs += subTileSize * subTileIndex;
u32 alpha;
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
const u64 alphaData = *ptr++;
alpha = Colour::convert4To8Bit((alphaData >> (4 * (u * 4 + v))) & 0xf);
} else {
alpha = 0xff; // ETC1 without alpha uses ff for every pixel
}
// Next 64 bits of the subtile are colour data
u64 colourData = *ptr;
return decodeETC(alpha, u, v, colourData);
}
u32 Texture::decodeETC(u32 alpha, u32 u, u32 v, u64 colourData) {
static constexpr u32 modifiers[8][2] = {
{2, 8}, {5, 17}, {9, 29}, {13, 42}, {18, 60}, {24, 80}, {33, 106}, {47, 183},
};
// Parse colour data for 4x4 block
const u32 subindices = getBits<0, 16, u32>(colourData);
const u32 negationFlags = getBits<16, 16, u32>(colourData);
const bool flip = getBit<32>(colourData);
const bool diffMode = getBit<33>(colourData);
// Note: index1 is indeed stored on the higher bits, with index2 in the lower bits
const u32 tableIndex1 = getBits<37, 3, u32>(colourData);
const u32 tableIndex2 = getBits<34, 3, u32>(colourData);
const u32 texelIndex = u * 4 + v; // Index of the texel in the block
if (flip) std::swap(u, v);
s32 r, g, b;
if (diffMode) {
r = getBits<59, 5, s32>(colourData);
g = getBits<51, 5, s32>(colourData);
b = getBits<43, 5, s32>(colourData);
if (u >= 2) {
r += signExtend3To32(getBits<56, 3, u32>(colourData));
g += signExtend3To32(getBits<48, 3, u32>(colourData));
b += signExtend3To32(getBits<40, 3, u32>(colourData));
}
// Expand from 5 to 8 bits per channel
r = Colour::convert5To8Bit(r);
g = Colour::convert5To8Bit(g);
b = Colour::convert5To8Bit(b);
} else {
if (u < 2) {
r = getBits<60, 4, s32>(colourData);
g = getBits<52, 4, s32>(colourData);
b = getBits<44, 4, s32>(colourData);
} else {
r = getBits<56, 4, s32>(colourData);
g = getBits<48, 4, s32>(colourData);
b = getBits<40, 4, s32>(colourData);
}
// Expand from 4 to 8 bits per channel
r = Colour::convert4To8Bit(r);
g = Colour::convert4To8Bit(g);
b = Colour::convert4To8Bit(b);
}
const u32 index = (u < 2) ? tableIndex1 : tableIndex2;
s32 modifier = modifiers[index][(subindices >> texelIndex) & 1];
if (((negationFlags >> texelIndex) & 1) != 0) {
modifier = -modifier;
}
r = std::clamp(r + modifier, 0, 255);
g = std::clamp(g + modifier, 0, 255);
b = std::clamp(b + modifier, 0, 255);
return (alpha << 24) | (u32(b) << 16) | (u32(g) << 8) | u32(r);
}
} // namespace Metal

223
src/core/renderer_mtl/mtl_texture.cpp
View File

@@ -1,16 +1,18 @@
#include "renderer_mtl/mtl_texture.hpp"
#include <fmt/format.h>
#include <array>
#include <memory>
#include "colour.hpp"
#include "renderer_mtl/objc_helper.hpp"
using namespace Helpers;
namespace Metal {
void Texture::allocate() {
formatInfo = PICA::getPixelFormatInfo(format);
formatInfo = PICA::getMTLPixelFormatInfo(format);
MTL::TextureDescriptor* descriptor = MTL::TextureDescriptor::alloc()->init();
descriptor->setTextureType(MTL::TextureType2D);
@@ -20,11 +22,14 @@ namespace Metal {
descriptor->setUsage(MTL::TextureUsageShaderRead);
descriptor->setStorageMode(MTL::StorageModeShared); // TODO: use private + staging buffers?
texture = device->newTexture(descriptor);
texture->setLabel(toNSString(
"Texture " + std::string(PICA::textureFormatToString(format)) + " " + std::to_string(size.u()) + "x" + std::to_string(size.v())
));
texture->setLabel(toNSString(fmt::format("Base texture {} {}x{}", std::string(PICA::textureFormatToString(format)), size.u(), size.v())));
descriptor->release();
if (formatInfo.needsSwizzle) {
base = texture;
texture = base->newTextureView(formatInfo.pixelFormat, MTL::TextureType2D, NS::Range(0, 1), NS::Range(0, 1), formatInfo.swizzle);
}
setNewConfig(config);
}
@@ -58,6 +63,11 @@ namespace Metal {
if (texture) {
texture->release();
}
if (base) {
base->release();
}
if (sampler) {
sampler->release();
}
@@ -99,210 +109,19 @@ namespace Metal {
}
}
// u and v are the UVs of the relevant texel
// Texture data is stored interleaved in Morton order, ie in a Z - order curve as shown here
// https://en.wikipedia.org/wiki/Z-order_curve
// Textures are split into 8x8 tiles.This function returns the in - tile offset depending on the u & v of the texel
// The in - tile offset is the sum of 2 offsets, one depending on the value of u % 8 and the other on the value of y % 8
// As documented in this picture https ://en.wikipedia.org/wiki/File:Moser%E2%80%93de_Bruijn_addition.svg
u32 Texture::mortonInterleave(u32 u, u32 v) {
static constexpr u32 xOffsets[] = {0, 1, 4, 5, 16, 17, 20, 21};
static constexpr u32 yOffsets[] = {0, 2, 8, 10, 32, 34, 40, 42};
return xOffsets[u & 7] + yOffsets[v & 7];
}
// Get the byte offset of texel (u, v) in the texture
u32 Texture::getSwizzledOffset(u32 u, u32 v, u32 width, u32 bytesPerPixel) {
u32 offset = ((u & ~7) * 8) + ((v & ~7) * width); // Offset of the 8x8 tile the texel belongs to
offset += mortonInterleave(u, v); // Add the in-tile offset of the texel
return offset * bytesPerPixel;
}
// Same as the above code except we need to divide by 2 because 4 bits is smaller than a byte
u32 Texture::getSwizzledOffset_4bpp(u32 u, u32 v, u32 width) {
u32 offset = ((u & ~7) * 8) + ((v & ~7) * width); // Offset of the 8x8 tile the texel belongs to
offset += mortonInterleave(u, v); // Add the in-tile offset of the texel
return offset / 2;
}
u8 Texture::decodeTexelU8(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data) {
switch (fmt) {
case PICA::TextureFmt::A4: {
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 = data[offset] >> ((u % 2) ? 4 : 0);
alpha = Colour::convert4To8Bit(getBits<0, 4>(alpha));
// A8
return alpha;
}
case PICA::TextureFmt::A8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 alpha = data[offset];
// A8
return alpha;
}
default: Helpers::panic("[Texture::DecodeTexel] Unimplemented format = %d", static_cast<int>(fmt));
}
}
u16 Texture::decodeTexelU16(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data) {
switch (fmt) {
case PICA::TextureFmt::RG8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
constexpr u8 b = 0;
const u8 g = data[offset];
const u8 r = data[offset + 1];
// RG8
return (g << 8) | r;
}
case PICA::TextureFmt::RGBA4: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
u16 texel = u16(data[offset]) | (u16(data[offset + 1]) << 8);
u8 alpha = getBits<0, 4, u8>(texel);
u8 b = getBits<4, 4, u8>(texel);
u8 g = getBits<8, 4, u8>(texel);
u8 r = getBits<12, 4, u8>(texel);
// ABGR4
return (r << 12) | (g << 8) | (b << 4) | alpha;
}
case PICA::TextureFmt::RGBA5551: {
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 = getBits<1, 5, u8>(texel);
u8 g = getBits<6, 5, u8>(texel);
u8 r = getBits<11, 5, u8>(texel);
// BGR5A1
return (alpha << 15) | (r << 10) | (g << 5) | b;
}
case PICA::TextureFmt::RGB565: {
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(data[offset]) | (u16(data[offset + 1]) << 8);
const u8 b = getBits<0, 5, u8>(texel);
const u8 g = getBits<5, 6, u8>(texel);
const u8 r = getBits<11, 5, u8>(texel);
// B5G6R5
return (r << 11) | (g << 5) | b;
}
case PICA::TextureFmt::IA4: {
const u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 texel = data[offset];
const u8 alpha = texel & 0xf;
const u8 intensity = texel >> 4;
// ABGR4
return (intensity << 12) | (intensity << 8) | (intensity << 4) | alpha;
}
case PICA::TextureFmt::I4: {
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 intensity = data[offset] >> ((u % 2) ? 4 : 0);
intensity = getBits<0, 4>(intensity);
// ABGR4
return (intensity << 12) | (intensity << 8) | (intensity << 4) | 0xff;
}
default: Helpers::panic("[Texture::DecodeTexel] Unimplemented format = %d", static_cast<int>(fmt));
}
}
u32 Texture::decodeTexelU32(u32 u, u32 v, PICA::TextureFmt fmt, std::span<const u8> data) {
switch (fmt) {
case PICA::TextureFmt::RGB8: {
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];
// RGBA8
return (0xff << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::RGBA8: {
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];
// RGBA8
return (alpha << 24) | (b << 16) | (g << 8) | r;
}
case PICA::TextureFmt::I8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 intensity = data[offset];
// RGBA8
return (0xff << 24) | (intensity << 16) | (intensity << 8) | intensity;
}
case PICA::TextureFmt::IA8: {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
// Same as I8 except each pixel gets its own alpha value too
const u8 alpha = data[offset];
const u8 intensity = data[offset + 1];
// RGBA8
return (alpha << 24) | (intensity << 16) | (intensity << 8) | intensity;
}
case PICA::TextureFmt::ETC1: return getTexelETC(false, u, v, size.u(), data);
case PICA::TextureFmt::ETC1A4: return getTexelETC(true, u, v, size.u(), data);
default: Helpers::panic("[Texture::DecodeTexel] Unimplemented format = %d", static_cast<int>(fmt));
}
}
void Texture::decodeTexture(std::span<const u8> data) {
std::vector<u8> decoded;
decoded.reserve(u64(size.u()) * u64(size.v()) * formatInfo.bytesPerTexel);
std::unique_ptr<u8[]> decodedData(new u8[u64(size.u()) * u64(size.v()) * formatInfo.bytesPerTexel]);
// This pointer will be incremented by our texture decoders
u8* decodePtr = decodedData.get();
// Decode texels line by line
for (u32 v = 0; v < size.v(); v++) {
for (u32 u = 0; u < size.u(); u++) {
if (formatInfo.bytesPerTexel == 1) {
u8 texel = decodeTexelU8(u, v, format, data);
decoded.push_back(texel);
} else if (formatInfo.bytesPerTexel == 2) {
u16 texel = decodeTexelU16(u, v, format, data);
decoded.push_back((texel & 0x00ff) >> 0);
decoded.push_back((texel & 0xff00) >> 8);
} else if (formatInfo.bytesPerTexel == 4) {
u32 texel = decodeTexelU32(u, v, format, data);
decoded.push_back((texel & 0x000000ff) >> 0);
decoded.push_back((texel & 0x0000ff00) >> 8);
decoded.push_back((texel & 0x00ff0000) >> 16);
decoded.push_back((texel & 0xff000000) >> 24);
} else {
Helpers::panic("[Texture::decodeTexture] Unimplemented bytesPerTexel (%u)", formatInfo.bytesPerTexel);
}
formatInfo.decoder(size, u, v, data, decodePtr);
decodePtr += formatInfo.bytesPerTexel;
}
}
texture->replaceRegion(MTL::Region(0, 0, size.u(), size.v()), 0, 0, decoded.data(), formatInfo.bytesPerTexel * size.u(), 0);
texture->replaceRegion(MTL::Region(0, 0, size.u(), size.v()), 0, 0, decodedData.get(), formatInfo.bytesPerTexel * size.u(), 0);
}
} // namespace Metal

62
src/core/renderer_mtl/pica_to_mtl.cpp Normal file
View File

@@ -0,0 +1,62 @@
#include "renderer_mtl/pica_to_mtl.hpp"
#include "renderer_mtl/texture_decoder.hpp"
using namespace Helpers;
namespace PICA {
MTLPixelFormatInfo mtlPixelFormatInfos[14] = {
{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelABGR8ToRGBA8}, // RGBA8
{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelBGR8ToRGBA8}, // RGB8
{MTL::PixelFormatBGR5A1Unorm, 2, decodeTexelA1BGR5ToBGR5A1}, // RGBA5551
{MTL::PixelFormatB5G6R5Unorm, 2, decodeTexelB5G6R5ToB5G6R5}, // RGB565
{MTL::PixelFormatABGR4Unorm, 2, decodeTexelABGR4ToABGR4}, // RGBA4
{MTL::PixelFormatRG8Unorm,
2,
decodeTexelAI8ToRG8,
true,
{
.red = MTL::TextureSwizzleRed,
.green = MTL::TextureSwizzleRed,
.blue = MTL::TextureSwizzleRed,
.alpha = MTL::TextureSwizzleGreen,
}}, // IA8
{MTL::PixelFormatRG8Unorm, 2, decodeTexelGR8ToRG8}, // RG8
{MTL::PixelFormatR8Unorm,
1,
decodeTexelI8ToR8,
true,
{.red = MTL::TextureSwizzleRed, .green = MTL::TextureSwizzleRed, .blue = MTL::TextureSwizzleRed, .alpha = MTL::TextureSwizzleOne}}, // I8
{MTL::PixelFormatA8Unorm, 1, decodeTexelA8ToA8}, // A8
{MTL::PixelFormatABGR4Unorm, 2, decodeTexelAI4ToABGR4}, // IA4
{MTL::PixelFormatR8Unorm,
1,
decodeTexelI4ToR8,
true,
{.red = MTL::TextureSwizzleRed, .green = MTL::TextureSwizzleRed, .blue = MTL::TextureSwizzleRed, .alpha = MTL::TextureSwizzleOne}}, // I4
{MTL::PixelFormatA8Unorm, 1, decodeTexelA4ToA8}, // A4
{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1ToRGBA8}, // ETC1
{MTL::PixelFormatRGBA8Unorm, 4, decodeTexelETC1A4ToRGBA8}, // ETC1A4
};
void checkForMTLPixelFormatSupport(MTL::Device* device) {
if (!device->supportsFamily(MTL::GPUFamilyApple1)) {
mtlPixelFormatInfos[2] = {MTL::PixelFormatRGBA8Unorm, 4, decodeTexelA1BGR5ToRGBA8};
mtlPixelFormatInfos[3] = {MTL::PixelFormatRGBA8Unorm, 4, decodeTexelB5G6R5ToRGBA8};
mtlPixelFormatInfos[4] = {MTL::PixelFormatRGBA8Unorm, 4, decodeTexelABGR4ToRGBA8};
mtlPixelFormatInfos[9] = {
MTL::PixelFormatRG8Unorm,
2,
decodeTexelAI4ToRG8,
true,
{
.red = MTL::TextureSwizzleRed,
.green = MTL::TextureSwizzleRed,
.blue = MTL::TextureSwizzleRed,
.alpha = MTL::TextureSwizzleGreen,
}
};
}
}
} // namespace PICA

20
src/core/renderer_mtl/renderer_mtl.cpp
View File

@@ -30,7 +30,6 @@ PICA::ColorFmt ToColorFormat(u32 format) {
}
MTL::Library* loadLibrary(MTL::Device* device, const cmrc::file& shaderSource) {
// MTL::CompileOptions* compileOptions = MTL::CompileOptions::alloc()->init();
NS::Error* error = nullptr;
MTL::Library* library = device->newLibrary(Metal::createDispatchData(shaderSource.begin(), shaderSource.size()), &error);
// MTL::Library* library = device->newLibrary(NS::String::string(source.c_str(), NS::ASCIIStringEncoding), compileOptions, &error);
@@ -56,12 +55,18 @@ void RendererMTL::reset() {
colorRenderTargetCache.reset();
}
void RendererMTL::setMTKLayer(void* layer) {
metalLayer = (CA::MetalLayer*)layer;
}
void RendererMTL::display() {
CA::MetalDrawable* drawable = metalLayer->nextDrawable();
if (!drawable) {
return;
}
MTL::Texture* texture = drawable->texture();
using namespace PICA::ExternalRegs;
// Top screen
@@ -87,13 +92,13 @@ void RendererMTL::display() {
MTL::RenderPassDescriptor* renderPassDescriptor = MTL::RenderPassDescriptor::alloc()->init();
MTL::RenderPassColorAttachmentDescriptor* colorAttachment = renderPassDescriptor->colorAttachments()->object(0);
colorAttachment->setTexture(drawable->texture());
colorAttachment->setTexture(texture);
colorAttachment->setLoadAction(MTL::LoadActionClear);
colorAttachment->setClearColor(MTL::ClearColor{0.0f, 0.0f, 0.0f, 1.0f});
colorAttachment->setStoreAction(MTL::StoreActionStore);
nextRenderPassName = "Display";
beginRenderPassIfNeeded(renderPassDescriptor, false, drawable->texture());
beginRenderPassIfNeeded(renderPassDescriptor, false, texture);
renderCommandEncoder->setRenderPipelineState(displayPipeline);
renderCommandEncoder->setFragmentSamplerState(nearestSampler, 0);
@@ -119,17 +124,22 @@ void RendererMTL::display() {
// Inform the vertex buffer cache that the frame ended
vertexBufferCache.endFrame();
// Release
drawable->release();
}
void RendererMTL::initGraphicsContext(SDL_Window* window) {
// On iOS, the SwiftUI side handles the MetalLayer
#ifdef PANDA3DS_IOS
device = MTL::CreateSystemDefaultDevice();
#else
// TODO: what should be the type of the view?
void* view = SDL_Metal_CreateView(window);
metalLayer = (CA::MetalLayer*)SDL_Metal_GetLayer(view);
device = MTL::CreateSystemDefaultDevice();
metalLayer->setDevice(device);
#endif
checkForMTLPixelFormatSupport(device);
commandQueue = device->newCommandQueue();
// Textures

334
src/core/renderer_mtl/texture_decoder.cpp Normal file
View File

@@ -0,0 +1,334 @@
#include "renderer_mtl/texture_decoder.hpp"
#include <array>
#include <string>
#include "colour.hpp"
#include "math_util.hpp"
using namespace Helpers;
// u and v are the UVs of the relevant texel
// Texture data is stored interleaved in Morton order, ie in a Z - order curve as shown here
// https://en.wikipedia.org/wiki/Z-order_curve
// Textures are split into 8x8 tiles.This function returns the in - tile offset depending on the u & v of the texel
// The in - tile offset is the sum of 2 offsets, one depending on the value of u % 8 and the other on the value of y % 8
// As documented in this picture https ://en.wikipedia.org/wiki/File:Moser%E2%80%93de_Bruijn_addition.svg
u32 mortonInterleave(u32 u, u32 v) {
static constexpr u32 xOffsets[] = {0, 1, 4, 5, 16, 17, 20, 21};
static constexpr u32 yOffsets[] = {0, 2, 8, 10, 32, 34, 40, 42};
return xOffsets[u & 7] + yOffsets[v & 7];
}
// Get the byte offset of texel (u, v) in the texture
u32 getSwizzledOffset(u32 u, u32 v, u32 width, u32 bytesPerPixel) {
u32 offset = ((u & ~7) * 8) + ((v & ~7) * width); // Offset of the 8x8 tile the texel belongs to
offset += mortonInterleave(u, v); // Add the in-tile offset of the texel
return offset * bytesPerPixel;
}
// Same as the above code except we need to divide by 2 because 4 bits is smaller than a byte
u32 getSwizzledOffset_4bpp(u32 u, u32 v, u32 width) {
u32 offset = ((u & ~7) * 8) + ((v & ~7) * width); // Offset of the 8x8 tile the texel belongs to
offset += mortonInterleave(u, v); // Add the in-tile offset of the texel
return offset / 2;
}
void decodeTexelABGR8ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 4);
const u8 alpha = inData[offset];
const u8 b = inData[offset + 1];
const u8 g = inData[offset + 2];
const u8 r = inData[offset + 3];
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = alpha;
}
void decodeTexelBGR8ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 3);
const u8 b = inData[offset];
const u8 g = inData[offset + 1];
const u8 r = inData[offset + 2];
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = 0xff;
}
void decodeTexelA1BGR5ToBGR5A1(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
u8 alpha = getBit<0>(texel);
u8 b = getBits<1, 5, u8>(texel);
u8 g = getBits<6, 5, u8>(texel);
u8 r = getBits<11, 5, u8>(texel);
u16 outTexel = (alpha << 15) | (r << 10) | (g << 5) | b;
*outData++ = outTexel & 0xff;
*outData++ = (outTexel >> 8) & 0xff;
}
void decodeTexelA1BGR5ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
u8 alpha = getBit<0>(texel) ? 0xff : 0;
u8 b = Colour::convert5To8Bit(getBits<1, 5, u8>(texel));
u8 g = Colour::convert5To8Bit(getBits<6, 5, u8>(texel));
u8 r = Colour::convert5To8Bit(getBits<11, 5, u8>(texel));
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = alpha;
}
void decodeTexelB5G6R5ToB5G6R5(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
*outData++ = texel & 0xff;
*outData++ = (texel >> 8) & 0xff;
}
void decodeTexelB5G6R5ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 2);
const u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
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));
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = 0xff;
}
void decodeTexelABGR4ToABGR4(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
u8 alpha = getBits<0, 4, u8>(texel);
u8 b = getBits<4, 4, u8>(texel);
u8 g = getBits<8, 4, u8>(texel);
u8 r = getBits<12, 4, u8>(texel);
*outData++ = (b << 4) | alpha;
*outData++ = (r << 4) | g;
}
void decodeTexelABGR4ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
u16 texel = u16(inData[offset]) | (u16(inData[offset + 1]) << 8);
u8 alpha = Colour::convert4To8Bit(getBits<0, 4, u8>(texel));
u8 b = Colour::convert4To8Bit(getBits<4, 4, u8>(texel));
u8 g = Colour::convert4To8Bit(getBits<8, 4, u8>(texel));
u8 r = Colour::convert4To8Bit(getBits<12, 4, u8>(texel));
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = alpha;
}
void decodeTexelAI8ToRG8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
// Same as I8 except each pixel gets its own alpha value too
const u8 alpha = inData[offset];
const u8 intensity = inData[offset + 1];
*outData++ = intensity;
*outData++ = alpha;
}
void decodeTexelGR8ToRG8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 2);
constexpr u8 b = 0;
const u8 g = inData[offset];
const u8 r = inData[offset + 1];
*outData++ = r;
*outData++ = g;
}
void decodeTexelI8ToR8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 intensity = inData[offset];
*outData++ = intensity;
}
void decodeTexelA8ToA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 alpha = inData[offset];
*outData++ = alpha;
}
void decodeTexelAI4ToABGR4(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 texel = inData[offset];
const u8 alpha = texel & 0xf;
const u8 intensity = texel >> 4;
*outData++ = (intensity << 4) | intensity;
*outData++ = (alpha << 4) | intensity;
}
void decodeTexelAI4ToRG8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
const u32 offset = getSwizzledOffset(u, v, size.u(), 1);
const u8 texel = inData[offset];
const u8 alpha = Colour::convert4To8Bit(texel & 0xf);
const u8 intensity = Colour::convert4To8Bit(texel >> 4);
*outData++ = intensity;
*outData++ = alpha;
}
void decodeTexelI4ToR8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
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 intensity = inData[offset] >> ((u % 2) ? 4 : 0);
intensity = Colour::convert4To8Bit(getBits<0, 4>(intensity));
*outData++ = intensity;
}
void decodeTexelA4ToA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
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 = inData[offset] >> ((u % 2) ? 4 : 0);
alpha = Colour::convert4To8Bit(getBits<0, 4>(alpha));
*outData++ = alpha;
}
static constexpr u32 signExtend3To32(u32 val) { return (u32)(s32(val) << 29 >> 29); }
void decodeETC(u32 u, u32 v, u64 colourData, u32 alpha, u8* outData) {
static constexpr u32 modifiers[8][2] = {
{2, 8}, {5, 17}, {9, 29}, {13, 42}, {18, 60}, {24, 80}, {33, 106}, {47, 183},
};
// Parse colour data for 4x4 block
const u32 subindices = getBits<0, 16, u32>(colourData);
const u32 negationFlags = getBits<16, 16, u32>(colourData);
const bool flip = getBit<32>(colourData);
const bool diffMode = getBit<33>(colourData);
// Note: index1 is indeed stored on the higher bits, with index2 in the lower bits
const u32 tableIndex1 = getBits<37, 3, u32>(colourData);
const u32 tableIndex2 = getBits<34, 3, u32>(colourData);
const u32 texelIndex = u * 4 + v; // Index of the texel in the block
if (flip) std::swap(u, v);
s32 r, g, b;
if (diffMode) {
r = getBits<59, 5, s32>(colourData);
g = getBits<51, 5, s32>(colourData);
b = getBits<43, 5, s32>(colourData);
if (u >= 2) {
r += signExtend3To32(getBits<56, 3, u32>(colourData));
g += signExtend3To32(getBits<48, 3, u32>(colourData));
b += signExtend3To32(getBits<40, 3, u32>(colourData));
}
// Expand from 5 to 8 bits per channel
r = Colour::convert5To8Bit(r);
g = Colour::convert5To8Bit(g);
b = Colour::convert5To8Bit(b);
} else {
if (u < 2) {
r = getBits<60, 4, s32>(colourData);
g = getBits<52, 4, s32>(colourData);
b = getBits<44, 4, s32>(colourData);
} else {
r = getBits<56, 4, s32>(colourData);
g = getBits<48, 4, s32>(colourData);
b = getBits<40, 4, s32>(colourData);
}
// Expand from 4 to 8 bits per channel
r = Colour::convert4To8Bit(r);
g = Colour::convert4To8Bit(g);
b = Colour::convert4To8Bit(b);
}
const u32 index = (u < 2) ? tableIndex1 : tableIndex2;
s32 modifier = modifiers[index][(subindices >> texelIndex) & 1];
if (((negationFlags >> texelIndex) & 1) != 0) {
modifier = -modifier;
}
r = std::clamp(r + modifier, 0, 255);
g = std::clamp(g + modifier, 0, 255);
b = std::clamp(b + modifier, 0, 255);
*outData++ = r;
*outData++ = g;
*outData++ = b;
*outData++ = alpha;
}
template <bool hasAlpha>
void getTexelETC(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
// Pixel offset of the 8x8 tile based on u, v and the width of the texture
u32 offs = ((u & ~7) * 8) + ((v & ~7) * size.u());
if (!hasAlpha) {
offs >>= 1;
}
// In-tile offsets for u/v
u &= 7;
v &= 7;
// ETC1(A4) also subdivide the 8x8 tile to 4 4x4 tiles
// Each tile is 8 bytes for ETC1, but since ETC1A4 has 4 alpha bits per pixel, that becomes 16 bytes
const u32 subTileSize = hasAlpha ? 16 : 8;
const u32 subTileIndex = (u / 4) + 2 * (v / 4); // Which of the 4 subtiles is this texel in?
// In-subtile offsets for u/v
u &= 3;
v &= 3;
offs += subTileSize * subTileIndex;
u32 alpha;
const u64* ptr = reinterpret_cast<const u64*>(inData.data() + offs); // Cast to u64*
if (hasAlpha) {
// First 64 bits of the 4x4 subtile are alpha data
const u64 alphaData = *ptr++;
alpha = Colour::convert4To8Bit((alphaData >> (4 * (u * 4 + v))) & 0xf);
} else {
alpha = 0xff; // ETC1 without alpha uses ff for every pixel
}
// Next 64 bits of the subtile are colour data
u64 colourData = *ptr;
decodeETC(u, v, colourData, alpha, outData);
}
void decodeTexelETC1ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
getTexelETC<false>(size, u, v, inData, outData);
}
void decodeTexelETC1A4ToRGBA8(OpenGL::uvec2 size, u32 u, u32 v, std::span<const u8> inData, u8* outData) {
getTexelETC<true>(size, u, v, inData, outData);
}

29
src/host_shaders/metal_shaders.metal
View File

@@ -1,4 +1,6 @@
#include <metal_stdlib>
#include <TargetConditionals.h>
using namespace metal;
struct BasicVertexOut {
@@ -219,12 +221,6 @@ struct Globals {
uint GPUREG_LIGHTING_LUTINPUT_SELECT;
uint GPUREG_LIGHTi_CONFIG;
// HACK
//bool lightingEnabled;
//uint8_t lightingNumLights;
//uint32_t lightingConfig1;
//uint16_t alphaControl;
float3 normal;
};
@@ -655,14 +651,15 @@ float4 performLogicOp(LogicOp logicOp, float4 s, float4 d) {
return as_type<float4>(performLogicOpU(logicOp, as_type<uint4>(s), as_type<uint4>(d)));
}
fragment float4 fragmentDraw(DrawVertexOut in [[stage_in]], float4 prevColor [[color(0)]], constant PicaRegs& picaRegs [[buffer(0)]], constant FragTEV& tev [[buffer(1)]], constant LogicOp& logicOp [[buffer(2)]], constant uint2& lutSlices [[buffer(3)]], texture2d<float> tex0 [[texture(0)]], texture2d<float> tex1 [[texture(1)]], texture2d<float> tex2 [[texture(2)]], texture2d_array<float> texLightingLut [[texture(3)]], texture1d_array<float> texFogLut [[texture(4)]], sampler samplr0 [[sampler(0)]], sampler samplr1 [[sampler(1)]], sampler samplr2 [[sampler(2)]], sampler linearSampler [[sampler(3)]]) {
Globals globals;
// iOS simulator doesn't support fb fetch, so don't enable it
#ifndef TARGET_OS_SIMULATOR
#define PREVIOUS_COLOR_DECL float4 prevColor [[color(0)]],
#else
#define PREVIOUS_COLOR_DECL
#endif
// HACK
//globals.lightingEnabled = picaRegs.read(0x008Fu) != 0u;
//globals.lightingNumLights = picaRegs.read(0x01C2u);
//globals.lightingConfig1 = picaRegs.read(0x01C4u);
//globals.alphaControl = picaRegs.read(0x104);
fragment float4 fragmentDraw(DrawVertexOut in [[stage_in]], PREVIOUS_COLOR_DECL constant PicaRegs& picaRegs [[buffer(0)]], constant FragTEV& tev [[buffer(1)]], constant LogicOp& logicOp [[buffer(2)]], constant uint2& lutSlices [[buffer(3)]], texture2d<float> tex0 [[texture(0)]], texture2d<float> tex1 [[texture(1)]], texture2d<float> tex2 [[texture(2)]], texture2d_array<float> texLightingLut [[texture(3)]], texture1d_array<float> texFogLut [[texture(4)]], sampler samplr0 [[sampler(0)]], sampler samplr1 [[sampler(1)]], sampler samplr2 [[sampler(2)]], sampler linearSampler [[sampler(3)]]) {
Globals globals;
globals.tevSources[0] = in.color;
if (lightingEnabled) {
@@ -755,5 +752,9 @@ fragment float4 fragmentDraw(DrawVertexOut in [[stage_in]], float4 prevColor [[c
}
}
#ifndef TARGET_OS_SIMULATOR
return performLogicOp(logicOp, color, prevColor);
}
#else
return performLogicOp(logicOp, color, float4(0.0));
#endif
}

37
src/ios_driver.mm Normal file
View File

@@ -0,0 +1,37 @@
#import <Foundation/Foundation.h>
extern "C" {
#include "ios_driver.h"
}
// Apple's Foundation headers define some macros globablly that create issues with our own code, so remove the definitions
#undef ABS
#undef NO
#include <memory>
#include "emulator.hpp"
// The Objective-C++ bridge functions must be exported without name mangling in order for the SwiftUI frontend to be able to call them
#define IOS_EXPORT extern "C" __attribute__((visibility("default")))
std::unique_ptr<Emulator> emulator = nullptr;
HIDService* hidService = nullptr;
IOS_EXPORT void iosCreateEmulator() {
printf("Creating emulator\n");
emulator = std::make_unique<Emulator>();
hidService = &emulator->getServiceManager().getHID();
emulator->initGraphicsContext(nullptr);
// TODO: Add game selection on iOS frontend
auto path = emulator->getAppDataRoot() / "toon_shading.elf";
emulator->loadROM(path);
}
IOS_EXPORT void iosRunFrame(CAMetalLayer* layer) {
void* layerBridged = (__bridge void*)layer;
emulator->getRenderer()->setMTKLayer(layerBridged);
emulator->runFrame();
}