feat(bridge): add --render offline CLI (Phase 0 of pedalboard replacement)

daw_vst_bridge --render <job.json> --out <clip.wav> renders MIDI notes
through the same instrument engines as realtime (SF2/SFZ/VST3/VST2) to a
16-bit stereo WAV, headless, no SHM. Mirrors plugins.py/vst_engine.py
note semantics (velocity int(vel*127), bank CC0 + program at t=0, beatSec
60/max(30,bpm), totalNeeded max(1024,endSample), no tail). Chunk clamped
to 256 to match VST3 host prepared buffers. Uses vendored sheredom json.h
(gated on HAVE_VST3SDK); rc: 0 ok, 1 arg/job error, 2 load fail, 3 crash.
Partial output removed on failure. Verified: SF2, SFZ, Nexus VST3 render
non-silent at exact duration; 9/9 tests pass.
This commit is contained in:
2026-08-17 08:04:25 +07:00
parent 4111995a5c
commit 0cca84c5f9
6 changed files with 426 additions and 0 deletions
Binary file not shown.
+1
View File
@@ -56,6 +56,7 @@ include_directories(
add_executable(daw_vst_bridge
src/main.cpp
src/RenderJob.cpp
src/NativeInstrumentEngine.cpp
src/SharedMemoryIPC.cpp
src/Vst3Instrument.cpp
+22
View File
@@ -0,0 +1,22 @@
// native_bridge/include/RenderJob.h
#ifndef RENDER_JOB_H
#define RENDER_JOB_H
#include <string>
// Offline render mode: `daw_vst_bridge --render <job.json> --out <clip.wav>`.
// Parses the job, creates the instrument headless, renders the MIDI notes to
// a 16-bit stereo WAV, and exits. Realtime SHM mode is untouched.
//
// job.json: { instrument_type, plugin_path, preset?, sample_rate?, block_size?,
// bpm?, soundfont_bank?, soundfont_program?,
// notes: [{pitch?, velocity?, start_beat?, duration_beats?}] }
// instrument_type: 0=VST3, 1=VST2, 2=SF2/SF3, 3=SFZ (InstrumentType enum).
// pitch/velocity: velocity 0..1 (float) or 0..127; start/duration in beats.
//
// Returns 0 on success. On failure prints a clear error to stderr and returns
// non-zero (1 = job/argument error, 2 = instrument load failure, 3 = plugin
// crashed mid-render). Any partial output file is removed.
int run_render_job(const std::string& jobPath, const std::string& outPath);
#endif // RENDER_JOB_H
+289
View File
@@ -0,0 +1,289 @@
// native_bridge/src/RenderJob.cpp
// Offline render mode (Phase 0 of PLAN_REPLACE_PEDALBOARD_NATIVE_BRIDGE.md):
// `daw_vst_bridge --render <job.json> --out <clip.wav>` renders MIDI notes
// through the same instrument engines as realtime (SF2/SFZ/VST3/VST2) and
// writes a 16-bit stereo WAV. No GUI, no SHM, no Tauri parent — pure CLI.
//
// JSON parsing uses the public-domain sheredom/json.h vendored with the VST3
// SDK (vst3sdk/public.sdk/source/vst/moduleinfo/json.h) — no new dependency.
#include "RenderJob.h"
#include "NativeInstrumentEngine.h"
#ifdef HAVE_VST3SDK
#include "vst3sdk/public.sdk/source/vst/moduleinfo/json.h"
#endif
#ifdef _WIN32
#ifndef NOMINMAX
#define NOMINMAX
#endif
#include <windows.h>
#endif
#include <algorithm>
#include <cmath>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <fstream>
#include <iostream>
#include <iterator>
#include <string>
#include <vector>
namespace {
#ifdef HAVE_VST3SDK
// --- sheredom/json.h helpers -------------------------------------------------
const json_object_element_s* member(const json_object_s* o, const char* key) {
for (const json_object_element_s* e = o ? o->start : nullptr; e; e = e->next)
if (e->name && e->name->string && std::strcmp(e->name->string, key) == 0)
return e;
return nullptr;
}
const json_value_s* memberValue(const json_object_s* o, const char* key) {
const json_object_element_s* m = member(o, key);
return m ? m->value : nullptr;
}
std::string memberString(const json_object_s* o, const char* key, const std::string& def) {
const json_value_s* v = memberValue(o, key);
if (v && v->type == json_type_string) {
const auto* s = static_cast<const json_string_s*>(v->payload);
return std::string(s->string, s->string_size);
}
return def;
}
bool memberNumber(const json_object_s* o, const char* key, double& out) {
const json_value_s* v = memberValue(o, key);
if (v && v->type == json_type_number) {
out = std::atof(static_cast<const json_number_s*>(v->payload)->number);
return true;
}
return false;
}
int64_t memberInt(const json_object_s* o, const char* key, int64_t def) {
double d;
return memberNumber(o, key, d) ? (int64_t)d : def;
}
#endif // HAVE_VST3SDK
// --- WAV writer (stdlib only, 16-bit PCM stereo, little-endian) --------------
void writeU16(std::ofstream& f, uint16_t v) {
char b[2] = { (char)(v & 0xFF), (char)((v >> 8) & 0xFF) };
f.write(b, 2);
}
void writeU32(std::ofstream& f, uint32_t v) {
char b[4] = { (char)(v & 0xFF), (char)((v >> 8) & 0xFF),
(char)((v >> 16) & 0xFF), (char)((v >> 24) & 0xFF) };
f.write(b, 4);
}
void writeWavHeader(std::ofstream& f, uint32_t sampleRate) {
f.write("RIFF", 4);
writeU32(f, 36); // patched in finishWav
f.write("WAVE", 4);
f.write("fmt ", 4);
writeU32(f, 16); // fmt chunk size
writeU16(f, 1); // PCM
writeU16(f, 2); // stereo
writeU32(f, sampleRate);
writeU32(f, sampleRate * 4); // byte rate
writeU16(f, 4); // block align
writeU16(f, 16); // bits per sample
f.write("data", 4);
writeU32(f, 0); // patched in finishWav
}
void finishWav(std::ofstream& f, uint64_t dataBytes) {
f.seekp(4);
writeU32(f, (uint32_t)(36 + dataBytes));
f.seekp(40);
writeU32(f, (uint32_t)dataBytes);
f.flush();
}
void writeFrames(std::ofstream& f, const float* L, const float* R, uint32_t n) {
for (uint32_t i = 0; i < n; ++i) {
auto cl = [](float v) -> int {
if (v > 1.0f) v = 1.0f;
else if (v < -1.0f) v = -1.0f;
return (int)(v * 32767.0f);
};
writeU16(f, (uint16_t)cl(L[i]));
writeU16(f, (uint16_t)cl(R[i]));
}
}
// One scheduled MIDI event at an absolute sample position. kind: 0=noteOn
// (a=pitch,b=velocity 1..127), 1=noteOff (a=pitch), 2=CC (a=cc,b=value),
// 3=program change (a=program).
struct Ev { uint64_t sample; uint8_t kind; uint32_t a; uint32_t b; };
} // namespace
int run_render_job(const std::string& jobPath, const std::string& outPath) {
#ifdef _WIN32
// VST3 hosting needs COM on the loading thread (realtime mode's workers
// call OleInitialize; here the main thread hosts the plugin).
CoInitializeEx(nullptr, COINIT_MULTITHREADED);
struct ComGuard { ~ComGuard() { CoUninitialize(); } } comGuard;
#endif
int rc = 1; // default: job/argument error
std::ofstream wav;
auto fail = [&](const std::string& msg) {
std::cerr << "[RenderJob] " << msg << std::endl;
if (wav.is_open()) wav.close();
std::remove(outPath.c_str()); // never leave a partial clip behind
};
#ifndef HAVE_VST3SDK
fail("built without the VST3 SDK submodule — JSON parser unavailable, --render disabled");
return rc;
#else
struct RootHolder { json_value_s* p = nullptr; ~RootHolder() { if (p) std::free(p); } } rootH;
// 1. Read + parse the job.
std::ifstream in(jobPath, std::ios::binary);
if (!in) { fail("cannot read job file: " + jobPath); return rc; }
std::string data((std::istreambuf_iterator<char>(in)), std::istreambuf_iterator<char>());
if (data.empty()) { fail("job file is empty: " + jobPath); return rc; }
json_parse_result_s pres = {};
json_value_s* root = json_parse_ex(data.data(), data.size(), json_parse_flags_default,
nullptr, nullptr, &pres);
if (!root) {
fail("job JSON parse error (code " + std::to_string(pres.error) + ")");
return rc;
}
rootH.p = root;
if (root->type != json_type_object) {
fail("job root must be a JSON object");
return rc;
}
const json_object_s* job = static_cast<const json_object_s*>(root->payload);
// 2. Instrument.
const int64_t itype = memberInt(job, "instrument_type", -1);
if (itype < 0 || itype > 3) { fail("job missing instrument_type (0=VST3,1=VST2,2=SF2/SF3,3=SFZ)"); return rc; }
const InstrumentType type = (InstrumentType)itype;
const std::string pluginPath = memberString(job, "plugin_path", "");
if (pluginPath.empty()) { fail("job missing plugin_path"); return rc; }
const std::string preset = memberString(job, "preset", "");
if (!preset.empty())
std::cerr << "[RenderJob] preset import is Phase 1 — rendering with the plugin's default state" << std::endl;
// 3. Format.
double srD = 44100.0;
memberNumber(job, "sample_rate", srD);
if (srD < 8000.0 || srD > 192000.0) { fail("bad sample_rate: " + std::to_string(srD)); return rc; }
const uint32_t sampleRate = (uint32_t)srD;
int64_t block = memberInt(job, "block_size", 256);
if (block > 256) {
std::cerr << "[RenderJob] block_size " << block << " > 256 — clamping (VST3 host buffers are prepared at 256)" << std::endl;
block = 256;
}
const uint32_t chunk = (uint32_t)std::max<int64_t>(32, block);
double bpm = 120.0;
memberNumber(job, "bpm", bpm);
bpm = std::max(30.0, bpm); // mirror plugins.py: max(30.0, bpm)
const double beatSec = 60.0 / bpm;
// 4. MIDI events (mirror vst_engine.midi_events_to_messages semantics:
// CC0 bank select + program change at t=0, noteOn at start, noteOff at
// start+duration, all sample-accurate).
std::vector<Ev> evs;
if (memberValue(job, "soundfont_bank")) {
evs.push_back({0, 2, 0, (uint32_t)(memberInt(job, "soundfont_bank", 0) & 0x7F)});
}
if (memberValue(job, "soundfont_program")) {
evs.push_back({0, 3, (uint32_t)(memberInt(job, "soundfont_program", 0) & 0x7F), 0});
}
const json_value_s* nv = memberValue(job, "notes");
if (!nv || nv->type != json_type_array) { fail("job missing notes array"); return rc; }
const json_array_s* notes = static_cast<const json_array_s*>(nv->payload);
if (notes->length == 0) { fail("job notes array is empty"); return rc; }
uint64_t totalNeeded = 1024; // mirror plugins.py: max(total_needed, 1024)
for (const json_array_element_s* el = notes->start; el; el = el->next) {
const json_value_s* v = el->value;
if (!v || v->type != json_type_object) continue;
const json_object_s* n = static_cast<const json_object_s*>(v->payload);
const uint32_t pitch = (uint32_t)(memberInt(n, "pitch", 60) & 0x7F);
double velF = 0.8;
memberNumber(n, "velocity", velF);
int vel = (int)(velF * 127.0); // mirror plugins.py: int(vel*127)
vel = std::max(0, std::min(127, vel));
const double startBeat = [&] { double d = 0.0; memberNumber(n, "start_beat", d); return d; }();
const double durBeats = [&] { double d = 1.0; memberNumber(n, "duration_beats", d); return d; }();
const uint64_t startSample = (uint64_t)std::llround(startBeat * beatSec * sampleRate);
const uint64_t endSample = (uint64_t)std::llround((startBeat + durBeats) * beatSec * sampleRate);
evs.push_back({startSample, 0, pitch, (uint32_t)vel});
evs.push_back({endSample, 1, pitch, 0});
if (endSample > totalNeeded) totalNeeded = endSample;
}
std::stable_sort(evs.begin(), evs.end(),
[](const Ev& a, const Ev& b) { return a.sample < b.sample; });
// 5. Load the instrument headless (same create_instrument path as realtime).
InstrumentEngineManager mgr;
if (!mgr.assign(0, type, pluginPath, sampleRate, chunk)) {
rc = 2; // load failure
fail("instrument load FAILED type=" + std::to_string((int)type) + " path=" + pluginPath);
return rc;
}
std::cout << "[RenderJob] instrument loaded type=" << (int)type
<< " path=" << pluginPath << " sr=" << sampleRate
<< " samples=" << totalNeeded << " bpm=" << bpm << std::endl;
// 6. Offline render loop — same dispatch/render split as the realtime
// loop: events whose sample time has passed are dispatched first, then
// each upcoming event splits the chunk so it lands at sample offset 0 of
// the next rendered sub-block.
wav.open(outPath, std::ios::binary);
if (!wav) { fail("cannot open output: " + outPath); return rc; }
writeWavHeader(wav, sampleRate);
std::vector<float> L(chunk), R(chunk);
uint64_t dataBytes = 0;
uint64_t pos = 0;
size_t ei = 0;
auto dispatch = [&](const Ev& e) {
switch (e.kind) {
case 0: mgr.noteOn(0, e.a, e.b / 127.0f); break;
case 1: mgr.noteOff(0, e.a); break;
case 2: mgr.controlChange(0, e.a, e.b); break;
case 3: mgr.programChange(0, e.a); break;
default: break;
}
};
auto render = [&](uint64_t from, uint64_t to) {
const uint32_t n = (uint32_t)(to - from);
mgr.renderAll(L.data(), R.data(), n);
writeFrames(wav, L.data(), R.data(), n);
dataBytes += (uint64_t)n * 4;
};
while (pos < totalNeeded) {
const uint64_t chunkEnd = std::min(pos + chunk, (uint64_t)totalNeeded);
while (ei < evs.size() && evs[ei].sample <= pos) { dispatch(evs[ei]); ++ei; }
while (ei < evs.size() && evs[ei].sample < chunkEnd) {
const uint64_t off = evs[ei].sample;
if (off > pos) { render(pos, off); pos = off; }
dispatch(evs[ei]);
++ei;
}
if (pos < chunkEnd) { render(pos, chunkEnd); pos = chunkEnd; }
}
finishWav(wav, dataBytes);
wav.close();
if (mgr.isCrashed(0)) {
rc = 3;
fail("plugin crashed inside processAudioBlock — clip invalid, muted channel");
return rc;
}
std::cout << "[RenderJob] wrote " << outPath << " (" << totalNeeded
<< " samples, " << (dataBytes / 4) << " frames)" << std::endl;
rc = 0;
return rc;
#endif // HAVE_VST3SDK
}
+15
View File
@@ -5,6 +5,7 @@
#include "SharedMemoryIPC.h"
#include "NativeInstrumentEngine.h"
#include "StateStore.h"
#include "RenderJob.h"
#ifdef _WIN32
#include <windows.h>
@@ -599,6 +600,20 @@ int main(int argc, char* argv[]) {
#endif
// --render <job.json> --out <clip.wav>: one-shot offline render, no SHM/loop.
std::string renderJob, renderOut;
for (int i = 1; i < argc; ++i) {
if (std::strcmp(argv[i], "--render") == 0 && i + 1 < argc) renderJob = argv[i + 1];
else if (std::strcmp(argv[i], "--out") == 0 && i + 1 < argc) renderOut = argv[i + 1];
}
if (!renderJob.empty()) {
if (renderOut.empty()) {
std::cerr << "[RenderJob] --render requires --out <clip.wav>" << std::endl;
return 1;
}
return run_render_job(renderJob, renderOut);
}
// 1. Shared memory name: argv --shm <name> | env SF_SHM_NAME | default
std::string shmName = "SonicForge_DAW_IPC";
for (int i = 1; i + 1 < argc; ++i) {
@@ -0,0 +1,99 @@
#!/usr/bin/env python3
# Phase 0 offline render test: `daw_vst_bridge --render <job.json> --out <clip.wav>`
# SF2 + SFZ through the same engines as realtime. Asserts:
# - exit 0, WAV exists, valid RIFF, exact duration (totalNeeded frames)
# - audio non-silent (RMS > 0.001)
# - error paths: missing --out -> rc 1; bad plugin_path -> rc 2
# Usage: python tests/test_offline_render.py
import json, math, os, struct, subprocess, sys, tempfile
BRIDGE = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "build", "Release", "daw_vst_bridge.exe"))
SF2 = r"C:\Users\locpham\SonicForgeStudio\app\storage\soundfonts\518e850f-a5d3-4790-b1f9-0c90c203c524.sf2"
SFZ = os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "sfizz-src", "tests", "TestFiles", "g2_test.sfz"))
BPM, SR = 120.0, 44100
BEAT_SEC = 60.0 / BPM
TOTAL = int(max(1024, round((0.0 + 1.0) * BEAT_SEC * SR))) # 22050
def read_wav(path):
b = open(path, "rb").read()
assert b[:4] == b"RIFF" and b[8:12] == b"WAVE", "not a RIFF/WAVE file"
assert b[36:40] == b"data", "no data chunk"
data_size = struct.unpack_from("<I", b, 40)[0]
assert len(b) == 44 + data_size, "trailing bytes"
assert data_size % 4 == 0, "stereo 16-bit: data must be multiple of 4"
frames = data_size // 4
samples = struct.unpack("<%dh" % (frames * 2), b[44:])
rms = math.sqrt(sum(s * s for s in samples) / len(samples)) / 32767.0
return frames, rms
def make_job(itype, path, outdir):
return {
"instrument_type": itype, "plugin_path": path,
"sample_rate": SR, "bpm": BPM,
"notes": [{"pitch": 60, "velocity": 0.8, "start_beat": 0.0, "duration_beats": 1.0}],
}
def run_render(jobdict, out):
job = os.path.join(outdir_of(out), os.path.basename(out) + ".job.json")
with open(job, "w", encoding="utf-8") as f:
json.dump(jobdict, f)
if os.path.exists(out):
os.remove(out)
r = subprocess.run([BRIDGE, "--render", job, "--out", out],
capture_output=True, text=True, timeout=120)
return r.returncode, (r.stdout + r.stderr)
def outdir_of(p):
d = os.path.dirname(p)
return d if d else "."
def main():
results = []
def check(name, ok, detail):
results.append((name, ok, detail))
print(("PASS" if ok else "FAIL"), name, "-", detail)
with tempfile.TemporaryDirectory() as td:
# 1. SF2 render
job = os.path.join(td, "sf2.json"); out = os.path.join(td, "sf2.wav")
rc, log = run_render(make_job(2, SF2, td), out)
if rc == 0 and os.path.exists(out):
frames, rms = read_wav(out)
check("SF2 render exit 0", rc == 0, log.strip().splitlines()[-1:] or "")
check("SF2 exact duration", frames == TOTAL, f"{frames} == {TOTAL}")
check("SF2 non-silent", rms > 0.001, f"rms={rms:.5f}")
else:
check("SF2 render exit 0", False, f"rc={rc} {log[-400:]}")
# 2. SFZ render (g2_test.sfz one-shot decays ~200ms; 0.5s clip still has signal)
job = os.path.join(td, "sfz.json"); out = os.path.join(td, "sfz.wav")
rc, log = run_render(make_job(3, SFZ, td), out)
if rc == 0 and os.path.exists(out):
frames, rms = read_wav(out)
check("SFZ render exit 0", rc == 0, log.strip().splitlines()[-1:] or "")
check("SFZ exact duration", frames == TOTAL, f"{frames} == {TOTAL}")
check("SFZ non-silent", rms > 0.001, f"rms={rms:.5f}")
else:
check("SFZ render exit 0", False, f"rc={rc} {log[-400:]}")
# 3. Missing --out -> rc 1
job = os.path.join(td, "noout.json")
with open(job, "w", encoding="utf-8") as f:
json.dump(make_job(2, SF2, td), f)
r = subprocess.run([BRIDGE, "--render", job], capture_output=True, text=True, timeout=30)
check("missing --out -> rc 1", r.returncode == 1, f"rc={r.returncode}")
# 4. Bad plugin path -> rc 2
job = os.path.join(td, "bad.json"); out = os.path.join(td, "bad.wav")
rc, log = run_render(make_job(2, r"C:\nonexistent\no.sf2", td), out)
check("bad plugin_path -> rc 2", rc == 2, f"rc={rc}")
check("bad render removes partial out", not os.path.exists(out), "no clip left behind")
failed = [n for n, ok, _ in results if not ok]
print()
print(f"{len(results) - len(failed)}/{len(results)} passed")
return 1 if failed else 0
if __name__ == "__main__":
sys.exit(main())