ruPRIM 0.1.8

Parallel primitives, reductions, scans, and indexing for Ruda.
Documentation 
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 ruda_core::device::Device;
use ruda_kernel::dsl as kernel_dsl;
use ruda_kernel::dsl::{prelude::*, calculate_ruda_count_elemwise};
use ruda_kernel::library::tensor::layout::linear::LinearView;
use ruda_kernel::tensor::{RudaTensor, element::TensorElement, allocation::empty_device_dtype, layout::address_type};
use ruda_core::tensor::{Shape, element::Element};
use crate::collective::{RudaBinaryOp, RudaCompare, RudaKeyEqual};
use super::{RudaPrimitiveError, check_type, transform::{self, RudaCast, RudaCastLaunch}};

/// Convert integral offset/count descriptors entirely on the device. Input
/// offsets are nonnegative and results must fit the chosen output type.
pub fn indices<R: Runtime, I: TensorElement + Int, O: TensorElement + Int>(input: &RudaTensor<R>)
    -> Result<RudaTensor<R>, RudaPrimitiveError>
{
    transform::unary::<R, I, O, RudaCast>(input, RudaCastLaunch::new())
}

#[ruda(launch_unchecked, explicit_define, address_type = "dynamic")]
fn convert_prefix<I: Int, O: Int>(input: &LinearView<I>, count: &LinearView<u64>, output: &mut LinearView<O, ReadWrite>) {
    let index = ABSOLUTE_POS;
    if index < input.shape() {
        if (index as u64) < count[0] { output[index] = O::cast_from(input[index]); }
    }
}

/// Convert only the device-counted valid prefix, never loading an uninitialized
/// capacity tail. Count must not exceed input length; converted values must fit O.
pub fn indices_prefix<R: Runtime, I: TensorElement + Int, O: TensorElement + Int>(input: &RudaTensor<R>, count: &RudaTensor<R>)
    -> Result<RudaTensor<R>, RudaPrimitiveError>
{
    check_type::<R, I>(input)?;
    check_type::<R, u64>(count)?;
    if count.meta.num_elements() != 1 { return Err(RudaPrimitiveError::Length); }
    if input.device.to_id() != count.device.to_id() { return Err(RudaPrimitiveError::Device); }
    let length = input.meta.num_elements();
    let output = empty_device_dtype(input.client.clone(), input.device.clone(), Shape::new([length]), <O as Element>::dtype());
    if length > 0 {
        let dim = RudaDim::new(input.client.properties(), length);
        let grid = calculate_ruda_count_elemwise(&input.client, length, dim);
        unsafe { convert_prefix::launch_unchecked::<I, O, R>(&input.client, grid, dim, address_type!(input, count, output),
            input.clone().into_linear_view(), count.clone().into_linear_view(), output.clone().into_linear_view()); }
    }
    Ok(output)
}

pub fn reduce<R, T, U, O>(input: &RudaTensor<R>, initial: U, op: O::RuntimeArg<R>, threads: u32)
    -> Result<RudaTensor<R>, RudaPrimitiveError>
where R: Runtime, T: TensorElement, U: TensorElement, O: RudaBinaryOp<U> + LaunchArg, O::RuntimeArg<R>: Clone,
{
    super::reduce::transform_reduce::<R, T, U, RudaCast, O>(input, RudaCastLaunch::new(), initial, op, threads)
}

pub fn reduce_by_key<R, K, T, U, O, E>(keys: &RudaTensor<R>, values: &RudaTensor<R>,
    op: O::RuntimeArg<R>, equal: E::RuntimeArg<R>, threads: u32) -> Result<super::reduce::RudaKeyReduction<R>, RudaPrimitiveError>
where R: Runtime, K: TensorElement, T: TensorElement, U: TensorElement,
    O: RudaBinaryOp<U> + LaunchArg, O::RuntimeArg<R>: Clone, E: RudaKeyEqual<K> + LaunchArg,
{
    check_type::<R, K>(keys)?;
    if keys.meta.num_elements() != values.meta.num_elements() { return Err(RudaPrimitiveError::Length); }
    if keys.device.to_id() != values.device.to_id() { return Err(RudaPrimitiveError::Device); }
    let values = transform::unary::<R, T, U, RudaCast>(values, RudaCastLaunch::new())?;
    super::reduce::by_key::<R, K, U, O, E>(keys, &values, op, equal, threads)
}

/// initial=None is inclusive; Some(initial) is exclusive within each equal-key run.
pub fn scan_by_key<R, K, T, U, O, E>(keys: &RudaTensor<R>, values: &RudaTensor<R>,
    op: O::RuntimeArg<R>, equal: E::RuntimeArg<R>, initial: Option<U>) -> Result<RudaTensor<R>, RudaPrimitiveError>
where R: Runtime, K: TensorElement, T: TensorElement, U: TensorElement,
    O: RudaBinaryOp<U> + LaunchArg, O::RuntimeArg<R>: Clone, E: RudaKeyEqual<K> + LaunchArg,
{
    check_type::<R, K>(keys)?;
    if keys.meta.num_elements() != values.meta.num_elements() { return Err(RudaPrimitiveError::Length); }
    if keys.device.to_id() != values.device.to_id() { return Err(RudaPrimitiveError::Device); }
    let values = transform::unary::<R, T, U, RudaCast>(values, RudaCastLaunch::new())?;
    super::segmented::scan_by_key::<R, K, U, O, E>(keys, &values, op, equal, initial)
}

/// Scan in accumulator type U, converting each input value before the first
/// operation. This does not reduce in T and convert only the final result.
pub fn scan<R, T, U, O>(input: &RudaTensor<R>, initial: Option<U>, exclusive: bool,
    op: O::RuntimeArg<R>, threads: u32) -> Result<RudaTensor<R>, RudaPrimitiveError>
where R: Runtime, T: TensorElement, U: TensorElement, O: RudaBinaryOp<U> + LaunchArg, O::RuntimeArg<R>: Clone,
{
    if exclusive && initial.is_none() { return Err(RudaPrimitiveError::Configuration("exclusive scan requires an initial value")); }
    let values = transform::unary::<R, T, U, RudaCast>(input, RudaCastLaunch::new())?;
    match (exclusive, initial) {
        (true, Some(initial)) => super::scan::exclusive_scan::<R, U, O>(&values, initial, op, threads),
        (false, Some(initial)) => super::scan::inclusive_scan_init::<R, U, O>(&values, initial, op, threads),
        _ => super::scan::inclusive_scan::<R, U, O>(&values, op, threads),
    }
}

pub fn segmented_reduce<R, T, U, I, O>(input: &RudaTensor<R>, begins: &RudaTensor<R>, ends: &RudaTensor<R>,
    initial: U, op: O::RuntimeArg<R>, threads: u32) -> Result<RudaTensor<R>, RudaPrimitiveError>
where R: Runtime, T: TensorElement, U: TensorElement, I: TensorElement + Int, O: RudaBinaryOp<U> + LaunchArg,
{
    check_type::<R, I>(begins)?;
    check_type::<R, I>(ends)?;
    if begins.meta.num_elements() != ends.meta.num_elements() { return Err(RudaPrimitiveError::Length); }
    if input.device.to_id() != begins.device.to_id() || input.device.to_id() != ends.device.to_id() { return Err(RudaPrimitiveError::Device); }
    let values = transform::unary::<R, T, U, RudaCast>(input, RudaCastLaunch::new())?;
    let begins = indices::<R, I, u64>(begins)?;
    let ends = indices::<R, I, u64>(ends)?;
    super::segments::reduce::<R, U, O>(&values, &begins, &ends, initial, op, threads)
}

pub fn segmented_scan_into<R, T, U, I, J, O>(input: &RudaTensor<R>, begins: &RudaTensor<R>, ends: &RudaTensor<R>,
    output_begins: &RudaTensor<R>, output: &RudaTensor<R>, initial: Option<U>, exclusive: bool,
    op: O::RuntimeArg<R>, threads: u32) -> Result<(), RudaPrimitiveError>
where R: Runtime, T: TensorElement, U: TensorElement, I: TensorElement + Int, J: TensorElement + Int, O: RudaBinaryOp<U> + LaunchArg,
{
    check_type::<R, I>(begins)?;
    check_type::<R, I>(ends)?;
    check_type::<R, J>(output_begins)?;
    check_type::<R, U>(output)?;
    let count = begins.meta.num_elements();
    if count != ends.meta.num_elements() || count != output_begins.meta.num_elements() { return Err(RudaPrimitiveError::Length); }
    for tensor in [begins, ends, output_begins, output] {
        if tensor.device.to_id() != input.device.to_id() { return Err(RudaPrimitiveError::Device); }
    }
    if exclusive && initial.is_none() { return Err(RudaPrimitiveError::Configuration("exclusive scan requires an initial value")); }
    let values = transform::unary::<R, T, U, RudaCast>(input, RudaCastLaunch::new())?;
    let begins = indices::<R, I, u64>(begins)?;
    let ends = indices::<R, I, u64>(ends)?;
    let output_begins = indices::<R, J, u64>(output_begins)?;
    super::segments::scan_into::<R, U, O>(&values, &begins, &ends, &output_begins, output, initial, exclusive, op, threads)
}

pub fn segmented_sort_keys<R, K, I, C>(input: &RudaTensor<R>, begins: &RudaTensor<R>, ends: &RudaTensor<R>,
    compare: C::RuntimeArg<R>, threads: u32) -> Result<RudaTensor<R>, RudaPrimitiveError>
where R: Runtime, K: TensorElement, I: TensorElement + Int, C: RudaCompare<K> + LaunchArg, C::RuntimeArg<R>: Clone,
{
    check_type::<R, K>(input)?;
    if begins.meta.num_elements() != ends.meta.num_elements() { return Err(RudaPrimitiveError::Length); }
    if input.device.to_id() != begins.device.to_id() || input.device.to_id() != ends.device.to_id() { return Err(RudaPrimitiveError::Device); }
    let begins = indices::<R, I, u64>(begins)?;
    let ends = indices::<R, I, u64>(ends)?;
    super::segmented_sort::sort_keys::<R, K, C>(input, &begins, &ends, compare, threads)
}

pub fn segmented_sort_pairs<R, K, V, I, C>(keys: &RudaTensor<R>, values: &RudaTensor<R>, begins: &RudaTensor<R>, ends: &RudaTensor<R>,
    compare: C::RuntimeArg<R>, threads: u32) -> Result<(RudaTensor<R>, RudaTensor<R>), RudaPrimitiveError>
where R: Runtime, K: TensorElement, V: TensorElement, I: TensorElement + Int,
    C: RudaCompare<K> + LaunchArg, C::RuntimeArg<R>: Clone,
{
    check_type::<R, K>(keys)?;
    check_type::<R, V>(values)?;
    if begins.meta.num_elements() != ends.meta.num_elements() || keys.meta.num_elements() != values.meta.num_elements() {
        return Err(RudaPrimitiveError::Length);
    }
    for tensor in [values, begins, ends] {
        if tensor.device.to_id() != keys.device.to_id() { return Err(RudaPrimitiveError::Device); }
    }
    let begins = indices::<R, I, u64>(begins)?;
    let ends = indices::<R, I, u64>(ends)?;
    super::segmented_sort::sort_pairs::<R, K, V, C>(keys, values, &begins, &ends, compare, threads)
}