PS2 games in the browser: static recompilation to WebAssembly with a native runtime
C++
GS: internal resolution scale (PS2_SCALE 1-3)
Samir Nuri37c8d755ff11
diff --git a/runtime/src/gs/gpu/gpu.cpp b/runtime/src/gs/gpu/gpu.cppindex 457c1fb..9178010 100644--- a/runtime/src/gs/gpu/gpu.cpp+++ b/runtime/src/gs/gpu/gpu.cpp@@ -26,13 +26,12 @@ namespace ps2::gs::gpu {namespace {-enum Pipeline { kUpload, kCopyRead, kCopyWrite, kLoadClut, kDisplay, kRaster, kPipelines };-const char *const kEntryPoints[kPipelines] = {"upload", "copy_read", "copy_write", "load_clut", "display", "raster"};+enum Pipeline { kUpload, kCopyRead, kCopyWrite, kLoadClut, kDisplay, kRaster, kResolve, kPipelines };+const char *const kEntryPoints[kPipelines] = {"upload", "copy_read", "copy_write", "load_clut", "display", "raster", "resolve"};constexpr std::uint32_t kParamStride = 256;constexpr std::uint32_t kClutSlots = 2048;constexpr std::uint32_t kClutEntries = 512;-constexpr std::uint64_t kScratchBytes = 16ull << 20;constexpr std::uint32_t kPrimWords = 40;constexpr std::uint32_t kSetupWords = 24;constexpr int kTileShift = 4;@@ -52,7 +51,9 @@ WGPUBindGroup g_blit_group = nullptr;WGPUBuffer g_blit_params = nullptr;std::uint32_t g_field = 0, g_last_width = 0, g_last_height = 0;bool g_have_previous = false;-constexpr std::uint32_t kFieldWords = 2u << 20;+std::uint32_t g_scale = 1, g_copy_words = 1u << 20, g_field_words = 1u << 19;++std::uint32_t field_base(std::uint32_t field) { return g_copy_words + (field & 1u) * g_field_words; }bool g_timestamps = false;WGPUQuerySet g_queries = nullptr;@@ -685,14 +686,17 @@ bool init(Mode m) {if (!g_device) return false;g_queue = wgpuDeviceGetQueue(g_device);- g_vram = make_buffer(kMemorySize, WGPUBufferUsage_Storage | WGPUBufferUsage_CopySrc | WGPUBufferUsage_CopyDst);+ if (const char *scale = std::getenv("PS2_SCALE")) g_scale = std::clamp<std::uint32_t>(static_cast<std::uint32_t>(std::atoi(scale)), 1u, 3u);+ g_copy_words = (1u << 20) * g_scale * g_scale;+ g_field_words = (1u << 19) * g_scale * g_scale;+ g_vram = make_buffer(static_cast<std::uint64_t>(kMemorySize) * g_scale * g_scale, WGPUBufferUsage_Storage | WGPUBufferUsage_CopySrc | WGPUBufferUsage_CopyDst);std::vector<std::uint32_t> tables;swizzle_tables(tables);g_tables = make_buffer(tables.size() * 4, WGPUBufferUsage_Storage | WGPUBufferUsage_CopyDst);wgpuQueueWriteBuffer(g_queue, g_tables, 0, tables.data(), tables.size() * 4);g_clut = make_buffer(static_cast<std::uint64_t>(kClutSlots) * kClutEntries * 4, WGPUBufferUsage_Storage);- g_scratch = make_buffer(kScratchBytes, WGPUBufferUsage_Storage | WGPUBufferUsage_CopySrc);- g_readback = make_buffer(kScratchBytes, WGPUBufferUsage_MapRead | WGPUBufferUsage_CopyDst);+ g_scratch = make_buffer((static_cast<std::uint64_t>(g_copy_words) + 2ull * g_field_words) * 4, WGPUBufferUsage_Storage | WGPUBufferUsage_CopySrc);+ g_readback = make_buffer(static_cast<std::uint64_t>(g_copy_words) * 4, WGPUBufferUsage_MapRead | WGPUBufferUsage_CopyDst);if (g_timestamps) {WGPUQuerySetDescriptor qd = WGPU_QUERY_SET_DESCRIPTOR_INIT;qd.type = WGPUQueryType_Timestamp;@@ -732,6 +736,11 @@ bool init(Mode m) {cd.layout = g_pipeline_layout;cd.compute.module = module;cd.compute.entryPoint = view(kEntryPoints[i]);+ WGPUConstantEntry constant = WGPU_CONSTANT_ENTRY_INIT;+ constant.key = view("SCALE");+ constant.value = static_cast<double>(g_scale);+ cd.compute.constantCount = 1;+ cd.compute.constants = &constant;g_pipelines[i] = wgpuDeviceCreateComputePipeline(g_device, &cd);if (!g_pipelines[i]) return false;}@@ -828,6 +837,10 @@ void draw(const State &st, const Vertex *v, int count) {default:return;}+ if (g_scale > 1) {+ --minx;+ --miny;+ }minx = std::max(minx, 0);miny = std::max(miny, 0);maxx = std::min(maxx, 2047);@@ -933,8 +946,8 @@ void local_copy(const State &s) {const std::uint32_t write = push_params({static_cast<std::uint32_t>(buf >> 32) & 0x3FFFu, static_cast<std::uint32_t>(buf >> 48) & 0x3Fu, static_cast<std::uint32_t>(buf >> 56) & 0x3Fu,static_cast<std::uint32_t>(pos >> 32) & 0x7FFu, static_cast<std::uint32_t>(pos >> 48) & 0x7FFu, w, h,static_cast<std::uint32_t>(pos >> 59) & 3u});- dispatch_linear(kCopyRead, w * h, read);- dispatch_linear(kCopyWrite, w * h, write);+ dispatch_linear(kCopyRead, w * h * g_scale * g_scale, read);+ dispatch_linear(kCopyWrite, w * h * g_scale * g_scale, write);}void load_clut(std::uint64_t tex0, std::uint64_t texclut) {@@ -968,8 +981,14 @@ void flush() {void download(std::uint8_t *memory) {flush_all();+ if (g_scale == 1) {+ submit();+ read_back(g_vram, kMemorySize, memory);+ return;+ }+ dispatch_linear(kResolve, kMemorySize / 4, push_params({}));submit();- read_back(g_vram, kMemorySize, memory);+ read_back(g_scratch, kMemorySize, memory);}void stats() {@@ -1100,14 +1119,14 @@ bool has_surface() { return g_surface != nullptr; }bool present(const DisplaySource &src) {if (!g_surface) return false;flush_all();- const std::uint32_t current = (g_field & 1u) * kFieldWords, previous = ((g_field + 1) & 1u) * kFieldWords;+ const std::uint32_t current = field_base(g_field), previous = field_base(g_field + 1);static const bool weave_off = std::getenv("PS2_NO_DEINTERLACE") != nullptr;const bool blend = src.halved && g_have_previous && g_last_width == src.width && g_last_height == src.height && !weave_off;const std::uint32_t param = push_params({src.fbp, src.fbw, src.psm, src.dbx, src.dby, src.width, src.height, src.halved ? 1u : 0u, current});- g_commands.push_back({kDisplay, (src.width + 7) / 8, (src.height + 7) / 8, param});+ g_commands.push_back({kDisplay, (src.width * g_scale + 7) / 8, (src.height * g_scale + 7) / 8, param});const float sw = static_cast<float>(g_surface_width), sh = static_cast<float>(g_surface_height);const float vw = std::min(sw, sh * 4.0f / 3.0f), vh = vw * 3.0f / 4.0f;- const std::uint32_t blit[8] = {src.width, src.height, current, previous, blend ? 1u : 0u, static_cast<std::uint32_t>(vw), static_cast<std::uint32_t>(vh), 0};+ const std::uint32_t blit[8] = {src.width * g_scale, src.height * g_scale, current, previous, blend ? 1u : 0u, static_cast<std::uint32_t>(vw), static_cast<std::uint32_t>(vh), 0};wgpuQueueWriteBuffer(g_queue, g_blit_params, 0, blit, sizeof blit);++g_field;g_have_previous = true;@@ -1174,7 +1193,7 @@ bool display_async(const DisplaySource &src, std::vector<std::uint32_t> &rgba, sfor (Copy &c : g_copies)if (!c.busy) slot = &c;if (slot) {- const std::uint64_t bytes = static_cast<std::uint64_t>(src.width) * src.height * 4;+ const std::uint64_t bytes = static_cast<std::uint64_t>(src.width) * src.height * g_scale * g_scale * 4;if (slot->size < bytes) {if (slot->buffer) wgpuBufferRelease(slot->buffer);slot->size = bytes;@@ -1182,7 +1201,7 @@ bool display_async(const DisplaySource &src, std::vector<std::uint32_t> &rgba, s}flush_all();const std::uint32_t param = push_params({src.fbp, src.fbw, src.psm, src.dbx, src.dby, src.width, src.height, src.halved ? 1u : 0u});- g_commands.push_back({kDisplay, (src.width + 7) / 8, (src.height + 7) / 8, param});+ g_commands.push_back({kDisplay, (src.width * g_scale + 7) / 8, (src.height * g_scale + 7) / 8, param});submit();WGPUCommandEncoder enc = wgpuDeviceCreateCommandEncoder(g_device, nullptr);wgpuCommandEncoderCopyBufferToBuffer(enc, g_scratch, 0, slot->buffer, 0, bytes);@@ -1192,8 +1211,8 @@ bool display_async(const DisplaySource &src, std::vector<std::uint32_t> &rgba, swgpuCommandEncoderRelease(enc);slot->busy = true;slot->order = ++g_copy_order;- slot->width = src.width;- slot->height = src.height;+ slot->width = src.width * g_scale;+ slot->height = src.height * g_scale;WGPUBufferMapCallbackInfo cb = WGPU_BUFFER_MAP_CALLBACK_INFO_INIT;cb.mode = WGPUCallbackMode_AllowProcessEvents;cb.callback = [](WGPUMapAsyncStatus status, WGPUStringView, void *copy, void *) {@@ -1207,12 +1226,14 @@ bool display_async(const DisplaySource &src, std::vector<std::uint32_t> &rgba, sreturn newest != nullptr;}-bool display(const DisplaySource &src, std::vector<std::uint32_t> &rgba) {+bool display(const DisplaySource &src, std::vector<std::uint32_t> &rgba, std::uint32_t &width, std::uint32_t &height) {flush_all();const std::uint32_t param = push_params({src.fbp, src.fbw, src.psm, src.dbx, src.dby, src.width, src.height, src.halved ? 1u : 0u});- g_commands.push_back({kDisplay, (src.width + 7) / 8, (src.height + 7) / 8, param});+ g_commands.push_back({kDisplay, (src.width * g_scale + 7) / 8, (src.height * g_scale + 7) / 8, param});submit();- rgba.resize(static_cast<std::size_t>(src.width) * src.height);+ width = src.width * g_scale;+ height = src.height * g_scale;+ rgba.resize(static_cast<std::size_t>(width) * height);read_back(g_scratch, rgba.size() * 4, rgba.data());return true;}diff --git a/runtime/src/gs/gpu/gpu.hpp b/runtime/src/gs/gpu/gpu.hppindex 4e4b9c9..a606cc9 100644--- a/runtime/src/gs/gpu/gpu.hpp+++ b/runtime/src/gs/gpu/gpu.hpp@@ -32,7 +32,7 @@ void local_copy(const State &s);void load_clut(std::uint64_t tex0, std::uint64_t texclut);void flush();void download(std::uint8_t *memory);-bool display(const DisplaySource &src, std::vector<std::uint32_t> &rgba);+bool display(const DisplaySource &src, std::vector<std::uint32_t> &rgba, std::uint32_t &width, std::uint32_t &height);bool display_async(const DisplaySource &src, std::vector<std::uint32_t> &rgba, std::uint32_t &width, std::uint32_t &height);bool attach_surface(const SurfaceTarget &target);void resize_surface(std::uint32_t width, std::uint32_t height);diff --git a/runtime/src/gs/gpu/gs.wgsl b/runtime/src/gs/gpu/gs.wgslindex 0e1620a..5280f20 100644--- a/runtime/src/gs/gpu/gs.wgsl+++ b/runtime/src/gs/gpu/gs.wgsl@@ -9,6 +9,8 @@ struct Params {@group(0) @binding(4) var<storage, read_write> clut: array<u32>;@group(0) @binding(5) var<storage, read_write> scratch: array<u32>;+override SCALE: u32 = 1u;+const PSMCT32: u32 = 0x00u;const PSMCT24: u32 = 0x01u;const PSMCT16: u32 = 0x02u;@@ -118,19 +120,34 @@ fn locate(psm: u32, bp: u32, bw: u32, x: u32, y: u32) -> Loc {return Loc((a & 0x3FFFFFu) >> 2u, shift, mask);}-fn read_pixel(psm: u32, bp: u32, bw: u32, x: u32, y: u32) -> u32 {+fn read_sample(psm: u32, bp: u32, bw: u32, x: u32, y: u32, k: u32) -> u32 {let l = locate(psm, bp, bw, x, y);- return (atomicLoad(&vram[l.word]) >> l.shift) & l.mask;+ return (atomicLoad(&vram[l.word * SCALE * SCALE + k]) >> l.shift) & l.mask;}-fn write_pixel(psm: u32, bp: u32, bw: u32, x: u32, y: u32, value: u32) {- let l = locate(psm, bp, bw, x, y);+fn read_pixel(psm: u32, bp: u32, bw: u32, x: u32, y: u32) -> u32 {+ return read_sample(psm, bp, bw, x, y, 0u);+}++fn store(l: Loc, k: u32, value: u32) {+ let w = l.word * SCALE * SCALE + k;if (l.mask == 0xFFFFFFFFu) {- atomicStore(&vram[l.word], value);+ atomicStore(&vram[w], value);return;}- atomicAnd(&vram[l.word], ~(l.mask << l.shift));- atomicOr(&vram[l.word], (value & l.mask) << l.shift);+ atomicAnd(&vram[w], ~(l.mask << l.shift));+ atomicOr(&vram[w], (value & l.mask) << l.shift);+}++fn write_sample(psm: u32, bp: u32, bw: u32, x: u32, y: u32, k: u32, value: u32) {+ store(locate(psm, bp, bw, x, y), k, value);+}++fn write_pixel(psm: u32, bp: u32, bw: u32, x: u32, y: u32, value: u32) {+ let l = locate(psm, bp, bw, x, y);+ for (var k = 0u; k < SCALE * SCALE; k = k + 1u) {+ store(l, k, value);+ }}fn bits_at(offset: u32, bit: u32, count: u32) -> u32 {@@ -178,12 +195,14 @@ fn upload(@builtin(global_invocation_id) gid: vec3<u32>) {fn copy_read(@builtin(global_invocation_id) gid: vec3<u32>) {let i = gid.x + gid.y * 64u * 1024u;let w = prm(5u);- if (i >= w * prm(6u)) {+ let n = SCALE * SCALE;+ if (i >= w * prm(6u) * n) {return;}- let x = (prm(3u) + i % w) & 0x7FFu;- let y = (prm(4u) + i / w) & 0x7FFu;- scratch[i] = read_pixel(prm(2u), prm(0u), prm(1u), x, y);+ let pixel = i / n;+ let x = (prm(3u) + pixel % w) & 0x7FFu;+ let y = (prm(4u) + pixel / w) & 0x7FFu;+ scratch[i] = read_sample(prm(2u), prm(0u), prm(1u), x, y, i % n);}@compute @workgroup_size(64)@@ -191,12 +210,14 @@ fn copy_write(@builtin(global_invocation_id) gid: vec3<u32>) {let i = gid.x + gid.y * 64u * 1024u;let w = prm(5u);let h = prm(6u);- if (i >= w * h) {+ let n = SCALE * SCALE;+ if (i >= w * h * n) {return;}let dir = prm(7u);- let x = i % w;- let y = i / w;+ let k = i % n;+ let x = (i / n) % w;+ let y = (i / n) / w;var sx = x;var sy = y;if ((dir & 1u) != 0u) {@@ -205,7 +226,7 @@ fn copy_write(@builtin(global_invocation_id) gid: vec3<u32>) {if ((dir & 2u) != 0u) {sy = h - 1u - y;}- write_pixel(prm(2u), prm(0u), prm(1u), (prm(3u) + x) & 0x7FFu, (prm(4u) + y) & 0x7FFu, scratch[sy * w + sx]);+ write_sample(prm(2u), prm(0u), prm(1u), (prm(3u) + x) & 0x7FFu, (prm(4u) + y) & 0x7FFu, k, scratch[(sy * w + sx) * n + k]);}@compute @workgroup_size(256)@@ -249,8 +270,8 @@ fn load_clut(@builtin(local_invocation_index) k: u32) {@compute @workgroup_size(8, 8)fn display(@builtin(global_invocation_id) gid: vec3<u32>) {- let w = prm(5u);- let h = prm(6u);+ let w = prm(5u) * SCALE;+ let h = prm(6u) * SCALE;if (gid.x >= w || gid.y >= h) {return;}@@ -259,7 +280,7 @@ fn display(@builtin(global_invocation_id) gid: vec3<u32>) {if (prm(7u) != 0u) {sy = gid.y / 2u;}- var c = read_pixel(psm, prm(0u), prm(1u), prm(3u) + gid.x, prm(4u) + sy);+ var c = read_sample(psm, prm(0u), prm(1u), prm(3u) + gid.x / SCALE, prm(4u) + sy / SCALE, (sy % SCALE) * SCALE + gid.x % SCALE);if (psm == PSMCT16 || psm == PSMCT16S) {c = ((c & 0x1Fu) << 3u) | ((c & 0x3E0u) << 6u) | ((c & 0x7C00u) << 9u);}@@ -393,10 +414,19 @@ fn wrap_coord(v: i32, size: i32, m: u32, mn: u32, mx: u32) -> i32 {}}-fn fetch(s: Setup, xi: i32, yi: i32) -> u32 {- let x = wrap_coord(xi, 1 << s.tw, mode(s, 24u, 2u), s.minu, s.maxu);- let y = wrap_coord(yi, 1 << s.th, mode(s, 26u, 2u), s.minv, s.maxv);- let raw = read_pixel(s.tpsm, s.tbp, s.tbw, u32(x), u32(y));+fn texel_x(s: Setup, xi: i32) -> u32 {+ return u32(wrap_coord(xi, 1 << s.tw, mode(s, 24u, 2u), s.minu, s.maxu));+}++fn texel_y(s: Setup, yi: i32) -> u32 {+ return u32(wrap_coord(yi, 1 << s.th, mode(s, 26u, 2u), s.minv, s.maxv));+}++fn fetch(s: Setup, xi: i32, yi: i32, k: u32) -> u32 {+ return convert(s, read_sample(s.tpsm, s.tbp, s.tbw, texel_x(s, xi), texel_y(s, yi), k));+}++fn convert(s: Setup, raw: u32) -> u32 {switch s.tpsm {case 0x00u, 0x30u {return raw;@@ -431,20 +461,18 @@ fn fetch(s: Setup, xi: i32, yi: i32) -> u32 {}}-fn sample(s: Setup, uin: f32, vin: f32) -> u32 {+fn sample(s: Setup, uin: f32, vin: f32, k: u32) -> u32 {if (!has(s, F_BILINEAR)) {- return fetch(s, i32(floor(uin)), i32(floor(vin)));+ return fetch(s, i32(floor(uin)), i32(floor(vin)), k);}let u = uin - 0.5;let v = vin - 0.5;let x0 = i32(floor(u));let y0 = i32(floor(v));- let fx = u - f32(x0);- let fy = v - f32(y0);- let c00 = fetch(s, x0, y0);- let c10 = fetch(s, x0 + 1, y0);- let c01 = fetch(s, x0, y0 + 1);- let c11 = fetch(s, x0 + 1, y0 + 1);+ return blend4(fetch(s, x0, y0, k), fetch(s, x0 + 1, y0, k), fetch(s, x0, y0 + 1, k), fetch(s, x0 + 1, y0 + 1, k), u - f32(x0), v - f32(y0));+}++fn blend4(c00: u32, c10: u32, c01: u32, c11: u32, fx: f32, fy: f32) -> u32 {var out = 0u;for (var ch = 0u; ch < 32u; ch = ch + 8u) {let a = f32((c00 >> ch) & 0xFFu);@@ -458,8 +486,8 @@ fn sample(s: Setup, uin: f32, vin: f32) -> u32 {return out;}-fn read_frame(s: Setup, x: u32, y: u32) -> u32 {- let raw = read_pixel(s.fpsm, s.fbp, s.fbw, x, y);+fn read_frame(s: Setup, x: u32, y: u32, k: u32) -> u32 {+ let raw = read_sample(s.fpsm, s.fbp, s.fbw, x, y, k);switch s.fpsm {case 0x01u {return raw | 0x80000000u;@@ -485,28 +513,28 @@ fn pack16(v: u32) -> u32 {return ((v >> 3u) & 0x1Fu) | (((v >> 11u) & 0x1Fu) << 5u) | (((v >> 19u) & 0x1Fu) << 10u) | a;}-fn write_frame(s: Setup, x: u32, y: u32, c: u32, mask: u32) {+fn write_frame(s: Setup, x: u32, y: u32, k: u32, c: u32, mask: u32) {switch s.fpsm {case 0x02u, 0x0Au {- let old = read_pixel(s.fpsm, s.fbp, s.fbw, x, y);+ let old = read_sample(s.fpsm, s.fbp, s.fbw, x, y, k);var m16 = pack16(mask);if ((mask & 0x80000000u) != 0u) {m16 = m16 | 0x8000u;}- write_pixel(s.fpsm, s.fbp, s.fbw, x, y, (pack16(c) & ~m16) | (old & m16));+ write_sample(s.fpsm, s.fbp, s.fbw, x, y, k, (pack16(c) & ~m16) | (old & m16));}case 0x01u {- let old = read_pixel(PSMCT32, s.fbp, s.fbw, x, y);+ let old = read_sample(PSMCT32, s.fbp, s.fbw, x, y, k);let m = mask | 0xFF000000u;- write_pixel(PSMCT32, s.fbp, s.fbw, x, y, (c & ~m) | (old & m));+ write_sample(PSMCT32, s.fbp, s.fbw, x, y, k, (c & ~m) | (old & m));}default {var v = c;if (mask != 0u) {- let old = read_pixel(s.fpsm, s.fbp, s.fbw, x, y);+ let old = read_sample(s.fpsm, s.fbp, s.fbw, x, y, k);v = (c & ~mask) | (old & mask);}- write_pixel(s.fpsm, s.fbp, s.fbw, x, y, v);+ write_sample(s.fpsm, s.fbp, s.fbw, x, y, k, v);}}}@@ -523,7 +551,7 @@ struct Attr {z: u32,}-fn plot(s: Setup, x: i32, y: i32, at: Attr) {+fn plot(s: Setup, x: i32, y: i32, k: u32, at: Attr) {if (x < s.scax0 || x > s.scax1 || y < s.scay0 || y > s.scay1) {return;}@@ -556,7 +584,7 @@ fn plot(s: Setup, x: i32, y: i32, at: Attr) {u = at.s / q * f32(1u << s.tw);v = at.t / q * f32(1u << s.th);}- let t = sample(s, u, v);+ let t = sample(s, u, v, k);let tr = i32(t & 0xFFu);let tg = i32((t >> 8u) & 0xFFu);let tb = i32((t >> 16u) & 0xFFu);@@ -661,7 +689,7 @@ fn plot(s: Setup, x: i32, y: i32, at: Attr) {var dst = 0u;var have_dst = false;if (has(s, F_DATE)) {- dst = read_frame(s, ux, uy);+ dst = read_frame(s, ux, uy, k);have_dst = true;if (((dst & 0x80000000u) != 0u) != has(s, F_DATM)) {return;@@ -670,7 +698,7 @@ fn plot(s: Setup, x: i32, y: i32, at: Attr) {let zte = has(s, F_ZTE);if (zte) {- let old = read_pixel(s.zpsm, s.zbp, s.fbw, ux, uy) & s.zmax;+ let old = read_sample(s.zpsm, s.zbp, s.fbw, ux, uy, k) & s.zmax;var ok = false;switch mode(s, 8u, 2u) {case 0u {@@ -695,7 +723,7 @@ fn plot(s: Setup, x: i32, y: i32, at: Attr) {let colclamp = has(s, F_COLCLAMP);if (has(s, F_ABE) && (!has(s, F_PABE) || (a & 0x80) != 0)) {if (!have_dst) {- dst = read_frame(s, ux, uy);+ dst = read_frame(s, ux, uy, k);}let cd = vec3<i32>(i32(dst & 0xFFu), i32((dst >> 8u) & 0xFFu), i32((dst >> 16u) & 0xFFu));let cs = vec3<i32>(r, g, b);@@ -750,10 +778,10 @@ fn plot(s: Setup, x: i32, y: i32, at: Attr) {if (!write_alpha) {mask = mask | 0xFF000000u;}- write_frame(s, ux, uy, u32(r) | (u32(g) << 8u) | (u32(b) << 16u) | (out_a << 24u), mask);+ write_frame(s, ux, uy, k, u32(r) | (u32(g) << 8u) | (u32(b) << 16u) | (out_a << 24u), mask);}if (write_z && zte) {- write_pixel(s.zpsm, s.zbp, s.fbw, ux, uy, z);+ write_sample(s.zpsm, s.zbp, s.fbw, ux, uy, k, z);}}@@ -782,9 +810,15 @@ fn z_at(p: u32, d: f32) -> u32 {}fn shade(p: u32, s: Setup, x: i32, y: i32) {+ for (var k = 0u; k < SCALE * SCALE; k = k + 1u) {+ let px = f32(x) + f32(k % SCALE) / f32(SCALE);+ let py = f32(y) + f32(k / SCALE) / f32(SCALE);+ shade_at(p, s, x, y, k, px, py);+ }+}++fn shade_at(p: u32, s: Setup, x: i32, y: i32, k: u32, px: f32, py: f32) {let kind = data[p + 1u];- let px = f32(x);- let py = f32(y);var at: Attr;switch kind {case 0u {@@ -810,6 +844,9 @@ fn shade(p: u32, s: Setup, x: i32, y: i32) {let y0 = fat(p + 5u);let dx = fat(p + 6u);let dy = fat(p + 7u);+ if (SCALE > 1u && (px < min(x0, x0 + dx) || px >= max(x0, x0 + dx) || py < min(y0, y0 + dy) || py >= max(y0, y0 + dy))) {+ return;+ }let a = vertex_tex(p, 0u);let b = vertex_tex(p, 1u);var tx = 0.0;@@ -871,7 +908,16 @@ fn shade(p: u32, s: Setup, x: i32, y: i32) {at = Attr(c.x, c.y, c.z, c.w, t.x, t.y, t.z, t.w, data[p + 36u]);}}- plot(s, x, y, at);+ plot(s, x, y, k, at);+}++@compute @workgroup_size(64)+fn resolve(@builtin(global_invocation_id) gid: vec3<u32>) {+ let i = gid.x + gid.y * 64u * 1024u;+ if (i >= 1048576u) {+ return;+ }+ scratch[i] = atomicLoad(&vram[i * SCALE * SCALE]);}@compute @workgroup_size(16, 16)diff --git a/runtime/src/gs/gs.cpp b/runtime/src/gs/gs.cppindex 22262fa..9a53c52 100644--- a/runtime/src/gs/gs.cpp+++ b/runtime/src/gs/gs.cpp@@ -756,9 +756,10 @@ bool display(const DisplayRegs &r, std::vector<std::uint32_t> &rgba, std::uint32#ifdef PS2_GPUif (gpu_on()) {const gpu::DisplaySource src{fbp * 32, fbw, psm, dbx, dby, width, height, halved};- if (!cpu_raster()) return gpu::display(src, rgba);+ if (!cpu_raster()) return gpu::display(src, rgba, width, height);static const bool compare = std::getenv("PS2_GS_COMPARE") != nullptr;- if (compare) g_gpu_frame.clear(), gpu::display(src, g_gpu_frame);+ std::uint32_t gw = 0, gh = 0;+ if (compare) g_gpu_frame.clear(), gpu::display(src, g_gpu_frame, gw, gh);}#endifrgba.resize(static_cast<std::size_t>(width) * height);diff --git a/web/shell.html b/web/shell.htmlindex ea79a8a..fc1d6bb 100644--- a/web/shell.html+++ b/web/shell.html@@ -13,6 +13,7 @@#bar { display: flex; gap: 16px; align-items: center; padding: 6px 12px; background: #141418; border-top: 1px solid #222; }#status { flex: 1; white-space: nowrap; overflow: hidden; text-overflow: ellipsis; }#stats { font-variant-numeric: tabular-nums; color: #9ad; }+ select { background: #2a3a55; color: #fff; border: 0; padding: 5px 8px; border-radius: 4px; }button { background: #2a3a55; color: #fff; border: 0; padding: 6px 14px; border-radius: 4px; cursor: pointer; }#overlay { position: fixed; inset: 0; display: flex; align-items: center; justify-content: center; background: rgba(0,0,0,.6); padding: 16px; }#panel { display: flex; flex-direction: column; align-items: center; gap: 14px; max-width: 520px; text-align: center; }@@ -94,7 +95,13 @@</div></div></div>-<div id="bar"><span id="status">Loading…</span><span id="stats"></span><button id="controls-toggle" type="button" hidden>Hide controls</button></div>+<div id="bar"><span id="status">Loading…</span><span id="stats"></span>+ <select id="scale" aria-label="Resolution" title="Internal resolution, the page reloads to apply it">+ <option value="1">640×512</option>+ <option value="2">1280×1024</option>+ <option value="3">1920×1536</option>+ </select>+ <button id="controls-toggle" type="button" hidden>Hide controls</button></div><div id="overlay"><div id="panel"><p id="help" hidden></p><button id="start" disabled>Loading…</button></div></div><script>const statusEl = document.getElementById('status');@@ -172,6 +179,20 @@ function onTouch(e) {}for (const type of ['touchstart', 'touchmove', 'touchend', 'touchcancel']) touchRoot.addEventListener(type, onTouch, { passive: false });+const scaleSelect = document.getElementById('scale');+let scale = params.get('PS2_SCALE');+if (!scale) {+ try { scale = localStorage.getItem('ps2web-scale'); } catch (e) {}+}+scale = ['1', '2', '3'].includes(scale) ? scale : '1';+scaleSelect.value = scale;+scaleSelect.addEventListener('change', () => {+ try { localStorage.setItem('ps2web-scale', scaleSelect.value); } catch (e) {}+ const url = new URL(location.href);+ url.searchParams.delete('PS2_SCALE');+ location.href = url.href;+});+const controlsButton = document.getElementById('controls-toggle');let hideControls = false;try { hideControls = localStorage.getItem('ps2web-hide-controls') === '1'; } catch (e) {}@@ -385,6 +406,7 @@ var Module = {},preRun: [() => {for (const [key, value] of params) if (key.startsWith('PS2_')) Module.ENV[key] = value;+ Module.ENV.PS2_SCALE = scale;Module.addRunDependency('disc');Promise.all([fetch(Module.discUrl, { method: 'HEAD' }), chooseGpu().catch((e) => 'GPU check failed: ' + e)]).then(([r, gpu]) => {Module.discSize = Number(r.headers.get('content-length')) || 0;