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
//use crate::common::validate_inputs;
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::indicators::simd_indicators::stddev_simd::SimdState;
use crate::indicators::stddev::{
min_data, output_length, IndicatorState, State, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::types::IndicatorError;
use crate::{common::validate_options, common_simd::assets::validate_inputs};
use std::simd::Simd;
/// SIMD driver that advances the Standard Deviation (STDDEV) across `N` asset lanes per scheduling epoch.
struct StddevDriver {
multiplier: f64,
period: usize,
want_optional_outputs: bool,
}
impl Driver<State> for StddevDriver {
/// Processes one epoch of bars for `N` assets simultaneously using SIMD.
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();
// Optimization 1: Direct array construction instead of collect+try_into
let mut state = SimdState::new(&states);
let multiplier_simd = Simd::splat(self.multiplier);
// Optimization 2: Pre-compute all input and output pointers
let input_ptrs = crate::extract_input_ptrs!(inputs, N, input_ptrs);
let (stddev_line_ptr, sma_line_ptr) =
crate::extract_output_ptrs!(outputs, N, stddev_line_ptr, sma_line_ptr);
// Optimization 3: Simplified main loop with pre-computed offsets
for (j, i) in (self.period..len).enumerate() {
// Get new and old values using pre-computed pointers
let (new_vals, old_vals) = crate::extract_simd_at_indices!(N, input_ptrs,
new_vals @ i,
old_vals @ j
);
let (stddev, sma) = state.calc_simd(new_vals, old_vals, multiplier_simd);
// Store results using pre-computed pointers
crate::write_simd_at_indices!(N, j,
stddev_line_ptr => stddev
);
crate::store_simd_optional_outputs!(j, N,
self.want_optional_outputs, sma_line_ptr => sma
);
}
state.write_states(&mut states);
}
}
/// Calculates the Standard Deviation (STDDEV) for `N` assets simultaneously using SIMD
/// parallelism.
///
/// STDDEV computes the rolling population standard deviation over a sliding window.
/// 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` - `[period]` — the rolling-window period for the standard deviation.
/// * `optional_outputs` - Optional slice of booleans enabling extra outputs:
/// `[0]` → `sma`.
///
/// # Returns
/// `Ok((outputs, states))` where `outputs[i][0]` is the STDDEV line and
/// `outputs[i][1]` is the SMA line (empty unless requested) for asset `i`.
/// `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 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, multiplier) = 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);
let mut want_optional_outputs = false;
for (i, state) in states.into_iter().enumerate() {
let asset_inputs = vec![inputs[i][0]];
let (stddev_line, sma_line) = {
let capacity = output_length(inputs[i][0].len(), options);
(
crate::uninit_vec!(f64, capacity),
crate::init_optional_outputs_eff!(
optional_outputs, &[false],
sma_line: capacity
),
)
};
if i == 0 {
(_, want_optional_outputs) = crate::calc_want_flags!(sma_line);
}
let mut output_buffer = vec![stddev_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,
state,
None,
));
output_buffers.push(output_buffer);
}
let mut driver = StddevDriver {
multiplier,
period,
want_optional_outputs,
};
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,
multiplier,
period,
));
}
Ok((output_buffers, states))
}