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
//use crate::common::validate_inputs;
use crate::common_simd::assets::validate_inputs;
use crate::indicators::simd_indicators::road_train::{Asset, Driver, PrimeMover};
use crate::indicators::simd_indicators::wad_simd::SimdState;
use crate::indicators::wad::{
min_data, output_length, IndicatorState as State, INPUTS_WIDTH, OPTIONS_WIDTH,
};
use crate::types::IndicatorError;
use std::simd::Simd;
/// SIMD driver that advances the WAD Indicator across `N` asset lanes per scheduling epoch.
struct WadDriver;
impl Driver<State> for WadDriver {
/// 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();
let mut state = SimdState::new(&states);
// Optimization 2: Pre-compute all input and output pointers
let (high_ptrs, low_ptrs, close_ptrs) =
crate::extract_input_ptrs!(inputs, N, high_ptrs, low_ptrs, close_ptrs);
let output_ptrs = crate::extract_output_ptrs!(outputs, N, output_ptr);
// Optimization 3: Simplified main loop with pre-computed offsets
for i in 0..len {
let (high, low, close) = crate::extract_simd_inputs_at_index!(i, N,
high @ high_ptrs,
low @ low_ptrs,
close @ close_ptrs
);
let wad = state.calc_simd(high, low, close);
// Store results using pre-computed pointers
crate::write_simd_at_indices!(N, i,
output_ptrs => wad
);
}
state.write_states(&mut states);
}
}
/// Calculates the WAD Indicator for `N` assets simultaneously using SIMD parallelism.
///
/// WAD requires no configurable options and produces no optional outputs. 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 `[high, low, close]` for asset `i`.
/// * `_options` - Unused; WAD has no configurable options.
/// * `_optional_outputs` - Unused; WAD has no optional outputs.
///
/// # Returns
/// `Ok((outputs, states))` where `outputs[i][0]` is the WAD line for asset `i` and
/// `states[i]` is the final state for asset `i`.
/// Returns `Err(IndicatorError)` if any input slice is too short.
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<State>), IndicatorError> {
validate_inputs::<INPUTS_WIDTH>(inputs, min_data(_options))?;
let mut road_train = PrimeMover::<N, State>::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], inputs[i][1], inputs[i][2]];
let state = State::new(inputs[i][2][0], 0.0);
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,
1,
0,
state,
None,
));
}
}
let mut driver = WadDriver;
let states = road_train.drive(&mut driver);
Ok((output_buffers, states))
}
/*pub fn indicator_by_assets_from_state<const N: usize>(
inputs: &[ &[ &[f64]; INPUTS_WIDTH]; N],
states: &mut [IndicatorState; N],
_optional_outputs: Option<&[bool]>,
) -> Result<[Vec<Vec<f64>>; N], IndicatorError>
{
let len = inputs[0][0].len();
// Validate all inputs have same length
for i in 0..N {
if inputs[i][0].len() != len {
return Err(IndicatorError::InvalidInputs);
}
}
// Extract EMAs and multipliers from states
let mut emas = Simd::from_array(std::array::from_fn(|i| states[i].get_ema()));
let multipliers = states[0].get_multipliers();
let multipliers_simd = (Simd::splat(multipliers.0), Simd::splat(multipliers.1));
// Create output arrays and process directly
let mut ema_lines: [Vec<Vec<f64>>; N] = std::array::from_fn(|_| {
vec![crate::uninit_vec!(f64, len)]
});
for i in 0..len {
//let values: [f64; N] = (0..N).map(|j| inputs[j][0][i]).collect::<Vec<_>>().try_into().unwrap();
let values: [f64; N] = std::array::from_fn(|j| inputs[j][0][i]);
let vals = Simd::from_array(values);
emas = calc_simd(vals, emas, multipliers_simd);
let outputs = emas.to_array();
for j in 0..N {
unsafe { *ema_lines[j].get_unchecked_mut(0).get_unchecked_mut(i) = outputs[j] }
}
}
// Update states with final EMA values
let final_emas = emas.to_array();
for i in 0..N {
states[i].set_ema(final_emas[i]);
}
Ok(ema_lines)
}*/