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
//use crate::common::validate_inputs;
use crate::common::validate_options;
use crate::common_simd::assets::validate_inputs;
use crate::indicators::cmo::{
min_data, output_length, IndicatorState, State, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::indicators::simd_indicators::cmo_simd::{calc_simd, SimdState};
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::types::IndicatorError;
use std::simd::Simd;
/// SIMD driver that advances the Chande Momentum Oscillator (CMO) across `N` asset lanes per
/// scheduling epoch.
struct CmoDriver {
/// The look-back period used to compute up/down momentum sums.
period: usize,
}
impl Driver<State> for CmoDriver {
/// Processes one epoch of bars for `N` assets simultaneously using SIMD.
///
/// Reads from `inputs[asset][0]` (real prices), writes CMO values to
/// `outputs[asset][0]`, 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 State>,
_options: Vec<Option<&()>>,
) {
let len = inputs[0][0].len();
//let output_len = len - self.period;
// Optimization 1: Direct array construction instead of collect+try_into
let mut state = SimdState::new(&states);
// Optimization 2: Pre-compute all input and output pointers
let input_ptrs = crate::extract_input_ptrs!(inputs, N, real_ptrs);
let cmo_line_ptr = crate::extract_output_ptrs!(outputs, N, cmo_line_ptr);
// Optimization 3: Simplified main loop with pre-computed offsets
for (j, i) in (self.period + 1..len).enumerate() {
// Get new and old values using pre-computed pointers
let (current, prev, prev_period, prev_before) = crate::extract_simd_at_indices!(N, input_ptrs,
current @ i,
prev @ i - 1,
prev_period @ j+1,
prev_before @ j
);
let cmo = calc_simd(&mut state, prev_before, prev_period, current, prev);
// Store results using pre-computed pointers
crate::write_simd_at_indices!(N, j,
cmo_line_ptr => cmo
);
}
state.write_states(&mut states);
}
}
/// Calculates the Chande Momentum Oscillator (CMO) for `N` assets simultaneously using SIMD
/// parallelism.
///
/// CMO measures the sum of recent up-moves minus recent down-moves, normalised to a ±100 scale.
/// All assets share the same `options`. 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 the real price series for asset `i`.
/// * `options` - Shared options applied to all `N` assets: `[period]`.
/// * `_optional_outputs` - Unused; CMO has no optional output lines.
///
/// # Returns
/// `Ok((outputs, states))` where `outputs[i][0]` is the CMO series for asset `i`
/// and `states[i]` is the final [`IndicatorState`] for asset `i`.
/// Returns `Err(IndicatorError)` if any input 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 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 simd_state = SimdState::init_state(&real, period);
let states = simd_state.to_states();
let mut road_train = PrimeMover::<N, State>::new();
let mut output_buffers = Vec::with_capacity(N);
for (i, state) in states.into_iter().enumerate() {
let asset_inputs = vec![inputs[i][0]];
let cmo_line = {
let capacity = output_length(inputs[i][0].len(), options);
crate::uninit_vec!(f64, capacity)
};
let mut output_buffer = vec![cmo_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 + 1,
state,
None,
));
output_buffers.push(output_buffer);
}
let mut driver = CmoDriver { period };
let states_vec = road_train.drive(&mut driver);
let mut states = Vec::with_capacity(N);
for (i, state) in states_vec.into_iter().enumerate() {
states.push(IndicatorState::new(
unsafe { inputs.get_unchecked(i).get_unchecked(0) },
state,
period,
));
}
Ok((output_buffers, states))
}