Port Ocarina of Time (Switch 2) to the recomp host

The NVN2 front end and the SM86 to Maxwell shader lowering (tools/xlate/AmpereLowering.cs, mesh and task stages,
bindless texture pools, BRX jump tables, the NVN2 compute c0 layout) with four translator patches (0017-0020) bring the
opening of the game to 58-60 fps on the Vulkan renderer.

This round fixes the haze: SM86 IPA is lowered as a plain Maxwell IPA.PASS (texture coordinates were corrected for
perspective twice); LDSM, TXD, both F2IP byte forms, HMMA, MOVM and HFMA2.RELU are lowered and spill mode has eight
uniform register groups; conditional rendering keeps the newest counter report per address and treats a report as
pending until its value is in guest memory, so occlusion tests and the depth clears around the depth copy behave; and
the upscaler stand-in decodes the post pass's sRGB picture before writing the linear target. Tensor-core programs keep
the stand-in unless NX2NAVI_NATIVE_TENSOR=1, because their lowering does not match the hardware yet.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
chrg6
2026-11-06 03:03:08 -05:00
parent 0478c13e45
commit 77d0c0b84e
65 changed files with 8136 additions and 554 deletions

View File

@@ -7,6 +7,9 @@
// "ok <key>" or "fail <key> <message>"; a malformed line gets "fail ? <message>" and the loop continues.
// nx2navi_xlate <jobs.json> <outdir> batch mode: a JSON array of jobs, prints a summary, exit 1 if any failed.
// job: { "key", "stage": "vs|ps|cs", "code": path (graphics = 0x50-byte SPH + instructions, compute = instructions), "c1": hex,
// "sm86": bool + "size": bytes (an NVN2 program, lowered by AmpereLowering.cs), "nvn2_table": [start, textures,
// images] (NVN2 compute binding table), "lowered_out": path, "mesh": {"stage": 6|7, "threads", "maxv", "maxp",
// "topo", "payload", "task"} (stage "ts" / "ms": an NVN2 task or mesh program, AmpereLowering.MeshInfo),
// "tex": {slot: type}, "img": {slot: {type, fmt}}, "attr": {loc: float|sint|uint}, "fout": {idx: float|sint|uint},
// "local": [x, y, z], "shared": bytes, "debug": bool, "reduced": bool, "softfloat64": bool, "zero_locals": bool,
// "sbuf_align": int, "subgroup": int }
@@ -37,13 +40,21 @@ static class SwshXlate
// 全部资源放进第 0 组,各占不同绑定号(spirv-cross 出成一个参数缓冲);常量段号 index 的缓冲由翻译器给 binding = index+1 的惯例不依赖
// All resources go into set 0, each with a distinct binding number (spirv-cross emits them as one argument buffer); the translator's convention of giving the buffer of constant segment number index binding = index+1 is not relied on
public SetBindingPair CreateConstantBufferBinding(int index) => new(0, _binding++);
public SetBindingPair CreateImageBinding(int count, bool isBuffer) { int b = _binding; _binding += count; return new(0, b); }
// A pool array (bindless) takes one binding number: the runtime moves it to its own set (vk_pipeline.c
// spirv_pool_rebind), and 32768 skipped numbers pushed the bindings after it past the runtime's 16-bit field.
static bool PoolArray(int count) => count == TexturePoolLength || count == SamplerPoolLength;
public SetBindingPair CreateImageBinding(int count, bool isBuffer) { int b = _binding; _binding += PoolArray(count) ? 1 : count; return new(0, b); }
public SetBindingPair CreateStorageBufferBinding(int index) => new(0, _binding++);
public SetBindingPair CreateTextureBinding(int count, bool isBuffer) { int b = _binding; _binding += count; return new(0, b); }
public SetBindingPair CreateTextureBinding(int count, bool isBuffer) { int b = _binding; _binding += PoolArray(count) ? 1 : count; return new(0, b); }
public ReadOnlySpan<ulong> GetCode(ulong address, int minimumSize) => MemoryMarshal.Cast<byte, ulong>(new ReadOnlySpan<byte>(_code)[(int)address..]);
public int QuerySamplerArrayLengthFromPool() => 32;
// Bindless handles loaded from memory index the whole texture and sampler pools (Ryujinx BindlessElimination
// GenerateBindlessAccess): arrays of TexturePoolLength textures (handle bits 0-19) and SamplerPoolLength samplers
// (bits 20-31; NVN sampler ids are below 4096). res.json marks them "pool": 1 (textures) / 2 (samplers); the
// runtime binds them from its global pool set (vk_exec_draw.c, texture and sampler pools).
public const int TexturePoolLength = 32768, SamplerPoolLength = 4096;
public int QuerySamplerArrayLengthFromPool() => SamplerPoolLength;
public int QueryTextureArrayLengthFromBuffer(int slot) => 32;
public int QueryTextureArrayLengthFromPool() => 32;
public int QueryTextureArrayLengthFromPool() => TexturePoolLength;
string Get(string table, int slot) => _job[table] is JsonObject t && t[slot.ToString()] is JsonNode n ? (n is JsonObject o ? (string)o["type"] : (string)n) : null;
int TexSlot(int handle, int cbufSlot)
@@ -68,9 +79,19 @@ static class SwshXlate
_ => SamplerType.Texture2D,
};
}
// Image handles: NVN1 c[2] 0x120 + 8 * slot (NVN2 compute handles are renumbered there by AmpereLowering.Handle);
// TexSlot knows only the separate texture range, so image formats were never answered (an R32UI view of an
// R11G11B10F target was declared float, and the raw words the shader packs were converted on store).
int ImgSlot(int handle, int cbufSlot)
{
(int off0, _, _) = TextureHandle.UnpackOffsets(handle);
int bytes = off0 * 4;
return (cbufSlot < 0 || (cbufSlot & 0xffff) == 2) && bytes >= 0x120 && bytes < TexBase && (bytes - 0x120) % 8 == 0 ? (bytes - 0x120) / 8 : -1;
}
public TextureFormat QueryTextureFormat(int handle, int cbufSlot = -1)
{
int slot = TexSlot(handle, cbufSlot);
int slot = ImgSlot(handle, cbufSlot);
if (slot < 0) slot = TexSlot(handle, cbufSlot);
if (slot >= 0) ImgQueried.Add(slot);
string f = _job["img"] is JsonObject t && slot >= 0 && t[slot.ToString()] is JsonObject o ? (string)o["fmt"] : null;
if (f == null) { Notes.Add($"图像句柄 {handle:X} 槽 {slot} 无 PC 格式,按 R8G8B8A8Unorm"); return TextureFormat.R8G8B8A8Unorm; }
@@ -80,6 +101,8 @@ static class SwshXlate
public int QueryComputeLocalSizeX() => Local(0);
public int QueryComputeLocalSizeY() => Local(1);
public int QueryComputeLocalSizeZ() => Local(2);
// "local_mem": local memory bytes of a compute program (the lowering's spill area adds to it); else Ryujinx's 0x1000.
public int QueryComputeLocalMemorySize() => _job["local_mem"] is JsonNode n ? (int)n : 0x1000;
public int QueryComputeSharedMemorySize() => _job["shared"] is JsonNode n ? (int)n : 0; // 配对 PC 的 groupshared 声明之和 | sum of the paired PC shader's groupshared declarations
static AttributeType Attr(string s) => s switch { "sint" => AttributeType.Sint, "uint" => AttributeType.Uint, _ => AttributeType.Float };
public GpuGraphicsState QueryGraphicsState()
@@ -134,13 +157,58 @@ static class SwshXlate
try
{
byte[] code = File.ReadAllBytes((string)job["code"]);
AmpereLowering.MeshInfo mesh = null;
// "sm86": true: the code is an NVN2 (Switch 2) program as stored, "size" bytes long (control section +0x38);
// it is lowered to Maxwell first (AmpereLowering.cs). "lowered_out": a path that receives the Maxwell code.
if (job["sm86"] is JsonValue sm && sm.TryGetValue<bool>(out bool isSm86) && isSm86)
{
int size = job["size"] is JsonNode sz ? (int)sz : code.Length;
AmpereLowering.C1Bytes = job["c1"] is JsonNode c1h ? Convert.FromHexString((string)c1h) : null;
int[] table = job["nvn2_table"] is JsonArray t && t.Count == 3 ? new[] { (int)t[0], (int)t[1], (int)t[2] } : null;
if (job["mesh"] is JsonObject mj)
{
// A task or mesh program: lowered and translated as a compute program whose shared memory holds
// the stage buffer, then rewritten into a TaskEXT / MeshEXT module (MeshSpirv.cs).
mesh = new AmpereLowering.MeshInfo
{
Stage = (int)mj["stage"], Threads = (int)mj["threads"], MaxVertices = (int)mj["maxv"],
MaxPrimitives = (int)mj["maxp"], Topology = (int)mj["topo"], PayloadBytes = (int)mj["payload"],
HasTask = mj["task"] is JsonValue tv && tv.TryGetValue<bool>(out bool tb) && tb,
};
code = AmpereLowering.Lower(code, size, false, out int meshLocal, null, mesh);
if (meshLocal != 0) job["local_mem"] = meshLocal;
job["local"] = new JsonArray(mesh.Threads, 1, 1);
job["shared"] = mesh.SharedBytes;
}
else
{
code = AmpereLowering.Lower(code, size, (string)job["stage"] != "cs", out int local, table);
if (local != 0) job["local_mem"] = local;
}
if (job["lowered_out"] is JsonNode lo) File.WriteAllBytes((string)lo, code);
}
var acc = new Accessor(code, job);
// 作业 debug = false:关 DebugMode(DebugMode 会跳过 AST 优化器,产物满是临时变量;不改语义)
// Job debug = false: turn off DebugMode (DebugMode skips the AST optimizer and the output is full of temporaries; semantics unchanged)
var flags = (job["debug"] is JsonNode dbg && !(bool)dbg ? 0 : TranslationFlags.DebugMode) | ((string)job["stage"] == "cs" ? TranslationFlags.Compute : 0);
var flags = (job["debug"] is JsonNode dbg && !(bool)dbg ? 0 : TranslationFlags.DebugMode) | ((string)job["stage"] is "cs" or "ts" or "ms" ? TranslationFlags.Compute : 0);
var ctx = Translator.CreateContext(0, acc, new TranslationOptions(TargetLanguage.Spirv, TargetApi.Vulkan, flags));
ShaderProgram p = ctx.Translate();
File.WriteAllBytes(Path.Combine(outDir, key + ".spv"), p.BinaryCode);
// "glsl_out": a path that receives the same program as GLSL, for reading (diagnosis only).
if (job["glsl_out"] is JsonNode go)
{
var gctx = Translator.CreateContext(0, new Accessor(code, job), new TranslationOptions(TargetLanguage.Glsl, TargetApi.OpenGL, flags));
File.WriteAllText((string)go, gctx.Translate().Code);
}
byte[] spirv = p.BinaryCode;
// A recursive call graph (subroutines the translator could not tell apart) is refused: Vulkan forbids it
// and NVIDIA's compiler overflows its stack on it (MeshSpirv.CallCycle).
if (MeshSpirv.CallCycle(spirv) is long cyc && cyc >= 0) throw new Exception($"recursive call graph (function %{cyc})");
if (mesh != null)
{
if (job["compute_out"] is JsonNode co) File.WriteAllBytes((string)co, spirv);
spirv = MeshSpirv.Convert(spirv, mesh);
}
File.WriteAllBytes(Path.Combine(outDir, key + ".spv"), spirv);
var res = new JsonObject
{
["key"] = key, ["stage"] = (string)job["stage"], ["notes"] = new JsonArray(acc.Notes.ConvertAll(n => (JsonNode)n).ToArray()),
@@ -149,10 +217,14 @@ static class SwshXlate
["tex_queried"] = new JsonArray(acc.TexQueried.Select(x => (JsonNode)x).ToArray()), ["img_queried"] = new JsonArray(acc.ImgQueried.Select(x => (JsonNode)x).ToArray()),
};
if (acc.QueryMotionPrevious()) (res["mvprev"], res["mvprev_banks"]) = (acc.MotionPreviousLocation, acc.MotionPreviousBanks);
if (mesh != null)
res["mesh"] = new JsonObject { ["lines"] = mesh.Lines.Count, ["shared"] = mesh.SharedBytes,
["outputs"] = new JsonArray(MeshSpirv.GenericLocations(mesh).Select(x => (JsonNode)x).ToArray()) };
foreach (var b in p.Info.CBuffers) res["cbufs"].AsArray().Add((JsonNode)new JsonObject { ["binding"] = b.Binding, ["slot"] = b.Slot });
foreach (var b in p.Info.SBuffers) res["sbufs"].AsArray().Add((JsonNode)new JsonObject { ["binding"] = b.Binding, ["slot"] = b.Slot, ["cb_slot"] = b.SbCbSlot, ["cb_offset"] = b.SbCbOffset, ["flags"] = (int)b.Flags });
foreach (var (list, name) in new[] { (p.Info.Textures, "textures"), (p.Info.Images, "images") })
foreach (var t in list) res[name].AsArray().Add((JsonNode)new JsonObject { ["binding"] = t.Binding, ["type"] = (int)t.Type, ["format"] = t.Format.ToString(), ["cbuf_slot"] = t.CbufSlot, ["handle"] = t.HandleIndex, ["array"] = t.ArrayLength, ["separate"] = t.Separate, ["flags"] = (int)t.Flags });
foreach (var t in list) res[name].AsArray().Add((JsonNode)new JsonObject { ["binding"] = t.Binding, ["type"] = (int)t.Type, ["format"] = t.Format.ToString(), ["cbuf_slot"] = t.CbufSlot, ["handle"] = t.HandleIndex, ["array"] = t.ArrayLength, ["separate"] = t.Separate, ["flags"] = (int)t.Flags,
["pool"] = t.ArrayLength == Accessor.TexturePoolLength ? 1 : t.ArrayLength == Accessor.SamplerPoolLength ? 2 : 0 });
File.WriteAllText(Path.Combine(outDir, key + ".res.json"), res.ToJsonString(new JsonSerializerOptions { WriteIndented = true }));
return null;
}