1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
//use crate::common::validate_inputs;
use crate::common::validate_options;
use crate::common_simd::assets::validate_inputs;
use crate::indicators::dpo::{
min_data, output_length, IndicatorState, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::indicators::simd_indicators::{by_asset::sma::init_state, dpo_simd::calc_simd};
use crate::types::IndicatorError;
use std::simd::Simd;
/// SIMD driver that advances the Detrended Price Oscillator (DPO) across `N` asset lanes
/// per scheduling epoch.
struct DpoDriver {
multiplier: f64,
period: usize,
dpo_period: usize,
want_sma: bool,
}
impl Driver<f64> for DpoDriver {
/// Processes one epoch of bars for `N` assets simultaneously using SIMD.
///
/// Reads from `inputs[asset][0]` (real), writes the DPO to `outputs[asset][0]`,
/// optional SMA to `outputs[asset][1]`, and updates `states[asset]` in place.
fn next_run<const N: usize>(
&mut self,
inputs: Vec<Vec<&[f64]>>,
mut outputs: Vec<Vec<&mut [f64]>>,
mut states: Vec<&mut f64>,
_options: Vec<Option<&()>>,
) {
let len = inputs[0][0].len();
let want_sma = self.want_sma;
// Optimization 1: Direct array construction instead of collect+try_into
let mut sums = Simd::<f64, N>::from_array(std::array::from_fn(|i| unsafe {
**states.get_unchecked(i)
}));
let multiplier_simd = Simd::splat(self.multiplier);
// Optimization 2: Pre-compute all input and output pointers
let input_ptrs: [*const f64; N] =
std::array::from_fn(|j| unsafe { inputs.get_unchecked(j).get_unchecked(0).as_ptr() });
let (dpo_line_ptrs, sma_line_ptrs) =
crate::extract_output_ptrs!(outputs, N, dpo_ptrs, sma_ptrs);
// Optimization 3: Simplified main loop with pre-computed offsets
for (j, i) in (self.period..len).enumerate() {
let (old_vals, dpo_vals, new_vals) = crate::extract_simd_at_indices!(N, input_ptrs,
old_vals @ j,
dpo_vals @ i - self.dpo_period,
new_vals @ i
);
let (dpo, sma) = calc_simd(new_vals, &mut sums, (old_vals, dpo_vals), multiplier_simd);
// Store results using pre-computed pointers
crate::write_simd_at_indices!(N, j,
dpo_line_ptrs => dpo
);
crate::store_simd_optional_outputs!(j, N,
want_sma, sma_line_ptrs => sma
);
}
// Update states efficiently
let final_sums = sums.to_array();
for (i, state) in states.iter_mut().enumerate().take(N) {
**state = final_sums[i];
}
}
}
/// Calculates the Detrended Price Oscillator (DPO) for `N` assets simultaneously using SIMD
/// parallelism.
///
/// Uses the [`PrimeMover`] scheduler to batch assets into SIMD-width groups.
///
/// # Arguments
/// * `inputs` - An array of `N` asset input sets; `inputs[i]` is `[&[f64]; INPUTS_WIDTH]`
/// containing `[real]` for asset `i`.
/// * `options` - Shared options slice; `options[0]` is the period.
/// * `optional_outputs` - Optional slice selecting extra outputs: index `0` = `sma`.
///
/// # Returns
/// `Ok((outputs, states))` where `outputs[i][0]` is the DPO line for asset `i`,
/// `outputs[i][1]` is the optional SMA, and `states[i]` is the final [`IndicatorState`]
/// for asset `i`.
/// Returns `Err(IndicatorError)` if any input slice is too short or options are invalid.
pub fn indicator_by_assets<const N: usize>(
inputs: &[&[&[f64]; INPUTS_WIDTH]; N], //stock[ fields [ field [f64] ] ]
options: &[f64; OPTIONS_WIDTH],
optional_outputs: Option<&[bool]>,
) -> Result<(Vec<Vec<Vec<f64>>>, Vec<IndicatorState>), IndicatorError> {
validate_inputs::<INPUTS_WIDTH>(inputs, min_data(options))?;
validate_options(options)?;
let period = options[0] as usize;
let dpo_period = period / 2 + 1;
//let real: Vec<&[f64]> = (0..N).map(|i| inputs[i][0]).collect();
let real: [&[f64]; N] = std::array::from_fn(|i| inputs[i][0]);
//init ema, sliced inputs and multipliers
let (sums, multiplier) = init_state(&real, period);
let mut road_train = PrimeMover::<N, f64>::new();
let mut want_sma = false;
let mut output_buffers = Vec::with_capacity(N);
for i in 0..N {
let asset_inputs = vec![inputs[i][0]];
let (dpo_line, sma_line) = {
let len = inputs[i][0].len();
let capacity = output_length(len, options);
(
crate::uninit_vec!(f64, capacity),
crate::init_optional_outputs_eff!(
optional_outputs, &[false],
sma_line: capacity
),
)
};
if i == 0 {
(_, want_sma) = crate::calc_want_flags!(sma_line);
}
let mut output_buffer = vec![dpo_line, sma_line];
//let adosc_len = output_buffer[0].len();
let mut asset_outputs = Vec::with_capacity(output_buffer.len());
for j in 0..output_buffer.len() {
unsafe {
//let slice_len = output_buffer.len() - starts[j];
// Get a mutable reference to the output buffer for this asset
let output_buffer = &mut output_buffer[j];
asset_outputs.push(std::slice::from_raw_parts_mut(
output_buffer.as_mut_ptr(), //slice from
output_buffer.len(), // slice to
));
}
}
road_train.add_asset(Asset::new(
asset_inputs,
asset_outputs,
i,
period,
period,
sums[i],
None,
));
output_buffers.push(output_buffer);
}
let mut driver = DpoDriver {
multiplier,
period,
dpo_period,
want_sma,
};
let sums = road_train.drive(&mut driver);
let mut states = Vec::with_capacity(N);
for (i, &sum) in sums.iter().enumerate() {
states.push(IndicatorState::new(
unsafe { inputs.get_unchecked(i).get_unchecked(0) },
sum,
multiplier,
period,
dpo_period,
));
}
Ok((output_buffers, states))
}