flystem-usls/src/core/engine.rs

379 lines
14 KiB
Rust

use anyhow::Result;
use half::f16;
use ndarray::{Array, IxDyn};
use ort::{
ExecutionProvider, ExecutionProviderDispatch, Session, SessionBuilder, TensorElementType,
TensorRTExecutionProvider, ValueType,
};
use crate::{home_dir, Device, MinOptMax, Options, CHECK_MARK, CROSS_MARK, SAFE_CROSS_MARK};
/// ONNXRuntime Backend
#[derive(Debug)]
pub struct OrtEngine {
session: Session,
device: Device,
inputs_minoptmax: Vec<Vec<MinOptMax>>,
inames: Vec<String>,
ishapes: Vec<Vec<isize>>,
idtypes: Vec<TensorElementType>,
onames: Vec<String>,
oshapes: Vec<Vec<isize>>,
odtypes: Vec<TensorElementType>,
profile: bool,
num_dry_run: usize,
}
impl OrtEngine {
pub fn dry_run(&self) -> Result<()> {
if self.num_dry_run == 0 {
println!("{SAFE_CROSS_MARK} No dry run count specified, skipping the dry run.");
return Ok(());
}
let mut xs: Vec<Array<f32, IxDyn>> = Vec::new();
for i in self.inputs_minoptmax.iter() {
let mut x: Vec<usize> = Vec::new();
for i_ in i.iter() {
x.push(i_.opt as usize);
}
let x: Array<f32, IxDyn> = Array::ones(x).into_dyn();
xs.push(x);
}
for _ in 0..self.num_dry_run {
self.run(xs.as_ref())?;
}
println!("{CHECK_MARK} Dry run x{}", self.num_dry_run);
Ok(())
}
pub fn new(config: &Options) -> Result<Self> {
ort::init().commit()?;
let session = Session::builder()?.with_model_from_file(&config.onnx_path)?;
// inputs
let mut ishapes = Vec::new();
let mut idtypes = Vec::new();
let mut inames = Vec::new();
for x in session.inputs.iter() {
inames.push(x.name.to_owned());
if let ValueType::Tensor { ty, dimensions } = &x.input_type {
ishapes.push(dimensions.iter().map(|x| *x as isize).collect::<Vec<_>>());
idtypes.push(*ty);
} else {
ishapes.push(vec![-1_isize]);
idtypes.push(ort::TensorElementType::Float32);
}
}
// outputs
let mut oshapes = Vec::new();
let mut odtypes = Vec::new();
let mut onames = Vec::new();
for x in session.outputs.iter() {
onames.push(x.name.to_owned());
if let ValueType::Tensor { ty, dimensions } = &x.output_type {
oshapes.push(dimensions.iter().map(|x| *x as isize).collect::<Vec<_>>());
odtypes.push(*ty);
} else {
oshapes.push(vec![-1_isize]);
odtypes.push(ort::TensorElementType::Float32);
}
}
let mut inputs_minoptmax: Vec<Vec<MinOptMax>> = Vec::new();
for (i, dims) in ishapes.iter().enumerate() {
let mut v_: Vec<MinOptMax> = Vec::new();
for (ii, &x) in dims.iter().enumerate() {
let x_default: MinOptMax = (ishapes[i][ii], ishapes[i][ii], ishapes[i][ii]).into();
let x: MinOptMax = match (i, ii) {
(0, 0) => Self::_set_ixx(x, &config.i00, i, ii).unwrap_or(x_default),
(0, 1) => Self::_set_ixx(x, &config.i01, i, ii).unwrap_or(x_default),
(0, 2) => Self::_set_ixx(x, &config.i02, i, ii).unwrap_or(x_default),
(0, 3) => Self::_set_ixx(x, &config.i03, i, ii).unwrap_or(x_default),
(0, 4) => Self::_set_ixx(x, &config.i04, i, ii).unwrap_or(x_default),
(0, 5) => Self::_set_ixx(x, &config.i05, i, ii).unwrap_or(x_default),
(1, 0) => Self::_set_ixx(x, &config.i10, i, ii).unwrap_or(x_default),
(1, 1) => Self::_set_ixx(x, &config.i11, i, ii).unwrap_or(x_default),
(1, 2) => Self::_set_ixx(x, &config.i12, i, ii).unwrap_or(x_default),
(1, 3) => Self::_set_ixx(x, &config.i13, i, ii).unwrap_or(x_default),
(1, 4) => Self::_set_ixx(x, &config.i14, i, ii).unwrap_or(x_default),
(1, 5) => Self::_set_ixx(x, &config.i15, i, ii).unwrap_or(x_default),
(2, 0) => Self::_set_ixx(x, &config.i20, i, ii).unwrap_or(x_default),
(2, 1) => Self::_set_ixx(x, &config.i21, i, ii).unwrap_or(x_default),
(2, 2) => Self::_set_ixx(x, &config.i22, i, ii).unwrap_or(x_default),
(2, 3) => Self::_set_ixx(x, &config.i23, i, ii).unwrap_or(x_default),
(2, 4) => Self::_set_ixx(x, &config.i24, i, ii).unwrap_or(x_default),
(2, 5) => Self::_set_ixx(x, &config.i25, i, ii).unwrap_or(x_default),
(3, 0) => Self::_set_ixx(x, &config.i30, i, ii).unwrap_or(x_default),
(3, 1) => Self::_set_ixx(x, &config.i31, i, ii).unwrap_or(x_default),
(3, 2) => Self::_set_ixx(x, &config.i32_, i, ii).unwrap_or(x_default),
(3, 3) => Self::_set_ixx(x, &config.i33, i, ii).unwrap_or(x_default),
(3, 4) => Self::_set_ixx(x, &config.i34, i, ii).unwrap_or(x_default),
(3, 5) => Self::_set_ixx(x, &config.i35, i, ii).unwrap_or(x_default),
_ => todo!(),
};
v_.push(x);
}
inputs_minoptmax.push(v_);
}
// build again
let builder = Session::builder()?;
let device = config.device.to_owned();
let _ep = match device {
Device::Trt(device_id) => Self::build_trt(
&inames,
&inputs_minoptmax,
&builder,
device_id,
config.trt_int8_enable,
config.trt_fp16_enable,
config.trt_engine_cache_enable,
)?,
Device::Cuda(device_id) => Self::build_cuda(&builder, device_id)?,
Device::CoreML(_) => {
let coreml = ort::CoreMLExecutionProvider::default()
.with_subgraphs()
// .with_ane_only()
.build();
if coreml.is_available()? && coreml.register(&builder).is_ok() {
println!("{CHECK_MARK} Using CoreML");
coreml
} else {
println!("{CROSS_MARK} CoreML initialization failed");
println!("{CHECK_MARK} Using CPU");
ort::CPUExecutionProvider::default().build()
}
}
Device::Cpu(_) => {
println!("{CHECK_MARK} Using CPU");
ort::CPUExecutionProvider::default().build()
} // _ => todo!(),
};
let session = builder
.with_optimization_level(ort::GraphOptimizationLevel::Level3)?
.with_model_from_file(&config.onnx_path)?;
Ok(Self {
session,
device,
inputs_minoptmax,
inames,
ishapes,
idtypes,
onames,
oshapes,
odtypes,
profile: config.profile,
num_dry_run: config.num_dry_run,
})
}
fn build_trt(
inames: &[String],
inputs_minoptmax: &[Vec<MinOptMax>],
builder: &SessionBuilder,
device_id: usize,
int8_enable: bool,
fp16_enable: bool,
engine_cache_enable: bool,
) -> Result<ExecutionProviderDispatch> {
// auto generate shapes
let mut spec_min = String::new();
let mut spec_opt = String::new();
let mut spec_max = String::new();
for (i, name) in inames.iter().enumerate() {
if i != 0 {
spec_min.push(',');
spec_opt.push(',');
spec_max.push(',');
}
let mut s_min = format!("{}:", name);
let mut s_opt = format!("{}:", name);
let mut s_max = format!("{}:", name);
for d in inputs_minoptmax[i].iter() {
let min_ = &format!("{}x", d.min);
let opt_ = &format!("{}x", d.opt);
let max_ = &format!("{}x", d.max);
s_min += min_;
s_opt += opt_;
s_max += max_;
}
s_min.pop();
s_opt.pop();
s_max.pop();
spec_min += &s_min;
spec_opt += &s_opt;
spec_max += &s_max;
}
let trt = TensorRTExecutionProvider::default()
.with_device_id(device_id as i32)
.with_int8(int8_enable)
.with_fp16(fp16_enable)
.with_engine_cache(engine_cache_enable)
.with_engine_cache_path(format!(
"{}/{}",
home_dir(None).to_str().unwrap(),
"trt-cache"
))
.with_timing_cache(false)
.with_profile_min_shapes(spec_min)
.with_profile_opt_shapes(spec_opt)
.with_profile_max_shapes(spec_max)
.build();
if trt.is_available()? && trt.register(builder).is_ok() {
println!(
"{CHECK_MARK} Using TensorRT (Initial model serialization may require a wait)"
);
Ok(trt)
} else {
println!("{CROSS_MARK} TensorRT initialization failed. Try CUDA...");
Self::build_cuda(builder, device_id)
}
}
fn build_cuda(builder: &SessionBuilder, device_id: usize) -> Result<ExecutionProviderDispatch> {
let cuda = ort::CUDAExecutionProvider::default()
.with_device_id(device_id as i32)
.build();
if cuda.is_available()? && cuda.register(builder).is_ok() {
println!("{CHECK_MARK} Using CUDA");
Ok(cuda)
} else {
println!("{CROSS_MARK} CUDA initialization failed");
println!("{CHECK_MARK} Using CPU");
Ok(ort::CPUExecutionProvider::default().build())
}
}
pub fn run(&self, xs: &[Array<f32, IxDyn>]) -> Result<Vec<Array<f32, IxDyn>>> {
// input
let mut xs_ = Vec::new();
let t_pre = std::time::Instant::now();
for (idtype, x) in self.idtypes.iter().zip(xs.iter()) {
let x_ = match idtype {
TensorElementType::Float32 => ort::Value::from_array(x.view())?,
TensorElementType::Float16 => ort::Value::from_array(x.mapv(f16::from_f32).view())?,
TensorElementType::Int32 => ort::Value::from_array(x.mapv(|x_| x_ as i32).view())?,
TensorElementType::Int64 => ort::Value::from_array(x.mapv(|x_| x_ as i64).view())?,
_ => todo!(),
};
xs_.push(x_);
}
let t_pre = t_pre.elapsed();
// inference
let t_run = std::time::Instant::now();
let ys = self.session.run(xs_.as_ref())?;
let t_run = t_run.elapsed();
// oputput
let mut ys_ = Vec::new();
let t_post = std::time::Instant::now();
for (dtype, name) in self.odtypes.iter().zip(self.onames.iter()) {
let y = &ys[name.as_str()];
let y_ = match &dtype {
TensorElementType::Float32 => y.extract_tensor::<f32>()?.view().to_owned(),
TensorElementType::Float16 => y.extract_tensor::<f16>()?.view().mapv(f16::to_f32),
TensorElementType::Int64 => y
.extract_tensor::<i64>()?
.view()
.to_owned()
.mapv(|x| x as f32),
_ => todo!(),
};
ys_.push(y_);
}
let t_post = t_post.elapsed();
if self.profile {
println!(
"[Profile] batch: {:?} => {:.4?} (i: {t_pre:.4?}, run: {t_run:.4?}, o: {t_post:.4?})",
self.batch().opt,
t_pre + t_run + t_post
);
}
Ok(ys_)
}
pub fn _set_ixx(x: isize, ixx: &Option<MinOptMax>, i: usize, ii: usize) -> Option<MinOptMax> {
match x {
-1 => {
match ixx {
None => panic!(
"{CROSS_MARK} Using dynamic shapes in inputs without specifying it: the {}-th input, the {}-th dimension.",
i + 1,
ii + 1
),
Some(ixx) => Some(ixx.to_owned()), // customized
}
}
_ => Some((x, x, x).into()), // customized, but not dynamic
}
}
pub fn oshapes(&self) -> &Vec<Vec<isize>> {
&self.oshapes
}
pub fn onames(&self) -> &Vec<String> {
&self.onames
}
pub fn odtypes(&self) -> &Vec<ort::TensorElementType> {
&self.odtypes
}
pub fn ishapes(&self) -> &Vec<Vec<isize>> {
&self.ishapes
}
pub fn inames(&self) -> &Vec<String> {
&self.inames
}
pub fn idtypes(&self) -> &Vec<ort::TensorElementType> {
&self.idtypes
}
pub fn device(&self) -> &Device {
&self.device
}
pub fn inputs_minoptmax(&self) -> &Vec<Vec<MinOptMax>> {
&self.inputs_minoptmax
}
pub fn batch(&self) -> &MinOptMax {
&self.inputs_minoptmax[0][0]
}
pub fn height(&self) -> &MinOptMax {
&self.inputs_minoptmax[0][2]
}
pub fn width(&self) -> &MinOptMax {
&self.inputs_minoptmax[0][3]
}
pub fn is_batch_dyn(&self) -> bool {
self.ishapes[0][0] == -1
}
pub fn try_fetch(&self, key: &str) -> Option<String> {
match self.session.metadata() {
Err(_) => None,
Ok(metadata) => match metadata.custom(key) {
Err(_) => None,
Ok(value) => value,
},
}
}
pub fn session(&self) -> &Session {
&self.session
}
pub fn version(&self) -> Option<String> {
self.try_fetch("version")
}
}