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
//use crate::common::validate_inputs;
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::indicators::sma::{
min_data, multiplier, output_length, IndicatorState, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::types::IndicatorError;
use crate::{common::validate_options, common_simd::assets::validate_inputs};
use std::simd::Simd;
//use crate::indicators::ad::output_length;
use crate::indicators::simd_indicators::sma_simd::calc_simd;
/// SIMD driver for Simple Moving Average (SMA) across `N` asset lanes per epoch.
/// Holds shared parameters used by [`Driver::next_run`] for each scheduled run.
struct SmaDriver {
/// The `1.0 / period` multiplier broadcast to all SIMD lanes.
multiplier: f64,
/// The look-back window length (number of bars summed per average).
period: usize,
}
impl Driver<f64> for SmaDriver {
/// Processes `bar_count` bars for `N` assets simultaneously using SIMD.
/// Reads `inputs[asset][0]` (real prices), writes `outputs[asset][0]` (SMA line),
/// and updates `states[asset]` with the rolling window sum for subsequent epochs.
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();
// 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 = crate::extract_input_ptrs!(inputs, N, input_ptrs);
let output_ptrs = crate::extract_output_ptrs!(outputs, N, output_ptrs);
// Optimization 3: Simplified main loop with pre-computed offsets
for (j, i) in (self.period..len).enumerate() {
let (old_vals, new_vals) = crate::extract_simd_at_indices!(N, input_ptrs,
old_vals @ j,
new_vals @ i
);
let sma = calc_simd(&mut sums, new_vals, old_vals, multiplier_simd);
crate::write_simd_at_indices!(N, j,
output_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];
}
}
}
/// Warms up the SMA state by accumulating the first `period` bars of each of the `N`
/// input slices into a running sum. Returns the per-asset initial sums and the shared multiplier.
///
/// # Returns
///
/// `(sums, multiplier)` — a `Vec<f64>` of `N` window sums and the `1.0 / period` constant.
pub fn init_state<'a, const N: usize>(inputs: &[&'a [f64]; N], period: usize) -> (Vec<f64>, f64) {
let multiplier = multiplier(period);
let mut sums = Simd::<f64, N>::splat(0.0);
// Optimization: Pre-compute input pointers for the initialization loop
let input_ptrs: [*const f64; N] = std::array::from_fn(|i| inputs[i].as_ptr());
for i in 0..period {
let values = Simd::from_array(std::array::from_fn(|j| unsafe { *input_ptrs[j].add(i) }));
sums += values;
}
(sums.to_array().to_vec(), multiplier)
}
/// Calculates the Simple Moving Average (SMA) for `N` assets simultaneously using SIMD.
///
/// All assets share the same `options` (period). Warms up each asset's rolling sum via
/// [`init_state`], then dispatches to [`SmaDriver::next_run`] through the `PrimeMover` scheduler.
///
/// # Arguments
///
/// * `inputs` — `N` asset input sets; `inputs[i][0]` is the real-price slice for asset `i`.
/// * `options` — Shared parameter array: `options[0]` = period.
/// * `_optional_outputs`— Unused; SMA has no optional output lines.
///
/// # Returns
///
/// `Ok((outputs, states))` where:
/// * `outputs[i][0]` — the SMA line for asset `i`.
/// * `states[i]` — the [`IndicatorState`] (rolling sum + multiplier) for resuming computation.
///
/// # Errors
///
/// Returns [`IndicatorError`] if inputs are 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 (sums, multiplier) = init_state(&real, period);
let mut road_train = PrimeMover::<N, f64>::new();
let mut output_buffers: Vec<Vec<Vec<f64>>> = (0..N)
.map(|i| {
vec![{
let capacity = output_length(inputs[i][0].len(), options);
crate::uninit_vec!(f64, capacity)
}]
})
.collect();
for i in 0..N {
let asset_inputs = vec![inputs[i][0]];
unsafe {
// Get a mutable reference to the output buffer for this asset
let output_buffer = &mut output_buffers[i][0];
let asset_outputs = vec![std::slice::from_raw_parts_mut(
output_buffer.as_mut_ptr(),
output_buffer.len(),
)];
road_train.add_asset(Asset::new(
asset_inputs,
asset_outputs,
i,
period,
period,
sums[i],
None,
));
}
}
let mut driver = SmaDriver { multiplier, period };
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,
));
}
Ok((output_buffers, states))
}