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
//use crate::common::validate_inputs;
use crate::indicators::hma::{
min_data, multiplier, output_length, IndicatorState, State, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::indicators::simd_indicators::hma_simd::assets::{calc_unchecked_simd, SimdState};
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::types::IndicatorError;
use crate::{common::validate_options, common_simd::assets::validate_inputs};
use std::simd::Simd;
/// SIMD driver that advances the Hull Moving Average (HMA) across `N` asset lanes
/// per scheduling epoch.
struct HmaDriver {
multipliers: (f64, f64, (f64, f64, f64), (f64, f64, f64)),
period: usize,
period2: usize,
}
impl Driver<State> for HmaDriver {
/// Processes one epoch of bars for `N` assets simultaneously using SIMD.
///
/// Reads from `inputs[asset][0]` (real), writes the HMA 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 mut state = SimdState::<N>::new(&mut states);
let len = inputs[0][0].len();
//multipliers: (f64, f64, (f64, f64, f64), (f64, f64, f64)),
let multipliers = (
Simd::splat(self.multipliers.0),
Simd::splat(self.multipliers.1),
(
Simd::splat(self.multipliers.2 .0),
Simd::splat(self.multipliers.2 .1),
Simd::splat(self.multipliers.2 .2),
),
(
Simd::splat(self.multipliers.3 .0),
Simd::splat(self.multipliers.3 .1),
Simd::splat(self.multipliers.3 .2),
),
);
//collect outputs
let hma_line_ptr = crate::extract_output_ptrs!(outputs, N, hma_line_ptr);
// Optimization 2: Pre-compute all input and output pointers
let real_ptrs = crate::extract_input_ptrs!(inputs, N, real_ptrs);
// Optimization 3: Simplified main loop with pre-computed offsets
for (j, i) in (self.period..len).enumerate() {
// Get inputs arrays for stocks
let (real, prev_real, prev_real2) = crate::extract_simd_at_indices!(N, real_ptrs,
real @ i,
prev_real @ j,// i - self.period,
prev_real2 @ i- self.period2
);
let hma = unsafe {
calc_unchecked_simd(&mut state, real, prev_real, prev_real2, multipliers)
};
//unsafe { calc_simd(&mut state, high, low, close, multiplier) };
// Store results using pre-computed pointers
crate::write_simd_at_indices!(N, j,
hma_line_ptr => hma
);
}
// Update states efficiently
state.write_states(&mut states);
}
}
/// Calculates the Hull Moving Average (HMA) 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` - Unused; HMA has no optional outputs.
///
/// # Returns
/// `Ok((outputs, states))` where `outputs[i][0]` is the HMA line for asset `i`
/// 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 period2 = period / 2;
let multipliers = multiplier(period);
let mut road_train = PrimeMover::<N, State>::new();
let mut output_buffers = Vec::with_capacity(N);
for i in 0..N {
let asset_inputs = vec![
inputs[i][0], // real
];
let hma_line = {
let capacity = output_length(inputs[i][0].len(), options);
crate::uninit_vec!(f64, capacity)
};
let (start, state) = State::init_state(inputs[i][0], period);
let mut output_buffer = vec![hma_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,
start,
period,
state,
None,
));
output_buffers.push(output_buffer);
}
let mut driver = HmaDriver {
multipliers,
period,
period2,
};
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(
inputs[i][0],
state,
period,
period2,
multipliers,
));
}
Ok((output_buffers, states))
}