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
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
//! The whole source-to-`DeviceFnMut` compilation pipeline

use crate::cache::*;
use crate::device::*;
use crate::error::*;
use crate::pool::*;

use std::borrow::BorrowMut;
use std::collections::hash_map::DefaultHasher;

use std::hash::{Hash, Hasher};
use std::io::{Read, Seek};
use std::sync::Arc;

use wgpu::read_spirv;

// TODO in the future, generalize this to other types, not just struct
/// A trait for structures that can exist in both Rust and GLSL
pub trait GlslStruct {
    /// Provides the GLSL structure definition code to define this structure in GLSL
    fn as_glsl() -> String;
}

/// The trait to implement when adding support for a new source language (e.g. - HLSL, XLA, Swift SIL, etc.).
///
/// This trait is generic over the input language (which must be hash-able so we can do caching) and the target bytecode (which can be a `Vec<u32>` or `&mut [u32]` for example).
/// We want the target bytecode to be mutable for a reason. We want people to be able to mutate the bytecode later on to
/// do - for example - various linting, optimization, compression, etc. before finishing compilation to DeviceFnMut.
///
/// Also, the target bytecode should be a GLSL compute shader with a single entry point of interest.
pub trait CompileToSpirv<I: Hash, P: BorrowMut<[u32]>> {
    /// Compiles the source language into SPIR-V
    fn compile_to_spirv(src: I) -> Result<Spirv<P>, CompileError>;
}

/// A wrapper for SPIR-V bytecode
///
/// The wrapper adds a few important details including parameters and the name of the relevant entry point in the bytecode.
/// You can construct a `Spirv` using a [`SpirvBuilder`](struct.SpirvBuilder.html).
#[derive(Hash)]
pub struct Spirv<P: BorrowMut<[u32]>> {
    pub params: DeviceFnMutParams,
    pub name: String,
    pub code: P,
}

/// A builder for constructing a [`Spirv`](struct.Spirv.html)
///
/// You can use it in the case where you are starting from either `u8` bytes or 4-byte `u32` words.
/// ```
/// # use {emu_core::prelude::*, emu_glsl::*, zerocopy::*, std::io::Cursor};
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// # let mut device = &mut futures::executor::block_on(Device::all())[0];
/// # let data = vec![0.0; 2048];
/// # let mut data_on_gpu: DeviceBox<[f32]> = device.create_from_mut(data.as_slice());
/// let kernel: Vec<u8> = vec![
///     // Magic number.           Version number: 1.0.
///     0x03, 0x02, 0x23, 0x07,    0x00, 0x00, 0x01, 0x00,
///     // Generator number: 0.    Bound: 0.
///     0x00, 0x00, 0x00, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // Reserved word: 0.
///     0x00, 0x00, 0x00, 0x00,
///     // OpMemoryModel.          Logical.
///     0x0e, 0x00, 0x03, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // GLSL450.
///     0x01, 0x00, 0x00, 0x00];
///
/// // if we actually compile to a `DeviceFnMut` later on with `finish`, it will panic
/// // at runtime because our SPIR-V doesn't actually include a main function
//  // nor does it have a mutable and a constant storage buffer
/// let spirv: Spirv<Vec<u32>> = SpirvBuilder::new()
///     .set_entry_point_name("main")
///     .add_param_mut::<[f32]>()
///     .add_param::<[f32]>()
///     .set_code_with_u8(Cursor::new(kernel))?
///     .build();
/// # Ok(())
/// # }
/// ```
pub struct SpirvBuilder<P: BorrowMut<[u32]>> {
    params_builder: ParamsBuilder,
    name: String,
    code: Option<P>,
}

impl<P: BorrowMut<[u32]>> SpirvBuilder<P> {
    /// Creates a new builder
    pub fn new() -> Self {
        Self {
            params_builder: ParamsBuilder::new(),
            name: String::from("main"),
            code: None,
        }
    }

    /// Sets the name of the point in this chunk of SPIR-V where it should be entered
    ///
    /// If you are compiling from GLSL, for example, your entry point name might be "main" if you have a "void main" function.
    pub fn set_entry_point_name(mut self, name: impl Into<String>) -> Self {
        self.name = name.into();
        self
    }

    /// Appends a new constant parameter to the kernel in this SPIR-V
    pub fn add_param<T: ?Sized>(mut self) -> Self {
        self.params_builder = self.params_builder.param::<T>(Mutability::Const);
        self
    }

    /// Appends a new mutable parameter to the kernel in this SPIR-V
    pub fn add_param_mut<T: ?Sized>(mut self) -> Self {
        self.params_builder = self.params_builder.param::<T>(Mutability::Mut);
        self
    }

    /// Set the actual code itself using an owned or borrowed slice of `u32`
    pub fn set_code_with_u32(mut self, code: P) -> Result<Self, std::io::Error> {
        self.code = Some(code);
        Ok(self)
    }

    /// Finish building
    pub fn build(self) -> Spirv<P> {
        Spirv {
            params: self.params_builder.build(),
            name: self.name,
            code: self
                .code
                .expect("no SPIR-V code was given to this SpirvBuilder with either `set_code_with_u8` or `set_code_with_u32`"),
        }
    }
}

impl SpirvBuilder<Vec<u32>> {
    /// Set the actual code itself using an owned `Vec<u32>``
    pub fn set_code_with_u8(mut self, code: impl Read + Seek) -> Result<Self, std::io::Error> {
        self.code = Some(read_spirv(code)?);
        Ok(self)
    }
}

/// Compiles the given source to `SpirvOrFinished`
///
/// There are 4 things this function is generic over.
/// 1. The source language which must be `Hash`able for caching
/// 2. The compiler implementing [`CompileToSpirv`](trait.CompileToSpirv.html)
/// 3. The target bytecode, a mutable borrow of a `u32` slice
/// 4. The cache, implementing [`Cache`](../cache/trait.Cache.html)
///
/// The returned [`SpirvOrFinished`](enum.SpirvOrFinished.html) is a finished `DeviceFnMut` if the source was in the cache or just the compiled SPIR-V if not.
/// You can then call `finish` on the result to finish the compiled SPIR-V to a `DeviceFnMut` in the case that source _wasn't_ in cache.
///
/// Here's how you might use it.
/// ```
/// # use {emu_core::prelude::*, emu_glsl::*, zerocopy::*, std::io::Cursor};
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// # let mut device = &mut futures::executor::block_on(Device::all())[0];
/// # let data = vec![0.0; 2048];
/// # let mut data_on_gpu: DeviceBox<[f32]> = device.create_from(data.as_slice());
/// let kernel: Vec<u32> = convert_to_spirv(Cursor::new(vec![
///     // Magic number.           Version number: 1.0.
///     0x03, 0x02, 0x23, 0x07,    0x00, 0x00, 0x01, 0x00,
///     // Generator number: 0.    Bound: 0.
///     0x00, 0x00, 0x00, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // Reserved word: 0.
///     0x00, 0x00, 0x00, 0x00,
///     // OpMemoryModel.          Logical.
///     0x0e, 0x00, 0x03, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // GLSL450.
///     0x01, 0x00, 0x00, 0x00]))?;
///
/// // later on, if we finish by compiling to a `DeviceFnMut`, it will panic
/// // at runtime just because our SPIR-V doesn't actually include a main function
//  // nor does it have a mutable and a constant storage buffer
/// let spirv: Spirv<Vec<u32>> = SpirvBuilder::new()
///     .set_entry_point_name("main")
///     .add_param_mut::<[f32]>()
///     .add_param::<[f32]>()
///     .set_code_with_u32(kernel)?
///     .build();
///
/// // in this case, `compile` doesn't do much other than caching
/// // but where `Spirv` is replaced with `Glsl` or `GlslKernel`,
/// // actual compilation takes place here
/// let spirv_or_finished = compile::<Spirv<_>, SpirvCompile, Vec<u32>, GlobalCache>(spirv)?;
/// // now at this point you can call `finish` to turn `spirv_or_finished` into
/// // a finished `DeviceFnMut`
/// # Ok(())
/// # }
/// ```
pub fn compile<I: Hash, U: CompileToSpirv<I, P>, P, C: Cache>(
    src: I,
) -> Result<SpirvOrFinished<P, C>, CompileError>
where
    P: BorrowMut<[u32]>,
{
    // get the hash of the source
    let mut hasher = DefaultHasher::new();
    src.hash(&mut hasher);
    let hash = hasher.finish();

    // check if source is in cache
    // if not, compile to SPIR-V before returning
    if C::contains(hash) {
        Ok(SpirvOrFinished::Finished(C::get(hash)))
    } else {
        let spirv = U::compile_to_spirv(src)?;
        Ok(SpirvOrFinished::SpirvAndHash((
            spirv,
            hash,
            std::marker::PhantomData,
        )))
    }
}

/// Either a finished `DeviceFnMut` or compiled SPIR-V
///
/// You can either call `finish` on this to get your final compiled `DeviceFnMut` or you can inspect/mutate the inner SPIR-V before finishing.
/// ```should_panic
/// # use {emu_core::prelude::*, emu_glsl::*, zerocopy::*, std::io::Cursor};
/// # fn main() -> Result<(), Box<dyn std::error::Error>> {
/// # let mut device = &mut futures::executor::block_on(Device::all())[0];
/// # let data = vec![0.0; 2048];
/// # let mut data_on_gpu: DeviceBox<[f32]> = device.create_from(data.as_slice());
/// let kernel: Vec<u8> = vec![
///     // Magic number.           Version number: 1.0.
///     0x03, 0x02, 0x23, 0x07,    0x00, 0x00, 0x01, 0x00,
///     // Generator number: 0.    Bound: 0.
///     0x00, 0x00, 0x00, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // Reserved word: 0.
///     0x00, 0x00, 0x00, 0x00,
///     // OpMemoryModel.          Logical.
///     0x0e, 0x00, 0x03, 0x00,    0x00, 0x00, 0x00, 0x00,
///     // GLSL450.
///     0x01, 0x00, 0x00, 0x00];
/// let spirv: Spirv<Vec<u32>> = SpirvBuilder::new()
///     .set_entry_point_name("main")
///     .add_param_mut::<[f32]>()
///     .add_param::<[f32]>()
///     .set_code_with_u8(Cursor::new(kernel))?
///     .build();
///
/// // in this case, `compile` doesn't do much other than caching
/// // but where `Spirv` is replaced with `Glsl` or `GlslKernel`,
/// // actual compilation takes place here
/// let mut spirv_or_finished = compile::<Spirv<_>, SpirvCompile, Vec<u32>, GlobalCache>(spirv)?;
/// // print out the SPIR-V and finish
/// if let Some(code) = spirv_or_finished.get_code_mut() {
///     println!("{:?}", code);
/// }
///
/// // the returned result is an ARC of a `DeviceFnMut`
/// // ARC just stands for "automatic reference counting"
/// // it means that there is runtime reference counting to
/// // ensure the compiled kernel can be safely used by multiple threads simultaneously
/// let finished_device_fn_mut = spirv_or_finished.finish()?;
/// // the above `finish` completes the compilation and in this case it
/// // will panic because our SPIR-V doesn't actually include a main function
//  // nor does it have a mutable and a constant storage buffer
/// # Ok(())
/// # }
/// ```
pub enum SpirvOrFinished<P: BorrowMut<[u32]>, C: Cache> {
    SpirvAndHash((Spirv<P>, u64, std::marker::PhantomData<C>)), // we need to pass in the hash of the source so that we can store the finished result in the cache
    Finished(Arc<DeviceFnMut>),
}

impl<P: BorrowMut<[u32]>, C: Cache> SpirvOrFinished<P, C> {
    /// Get a mutable reference to the code stored here
    ///
    /// This is useful when inspecting or linting the bytecode somehow before finishing compilation.
    /// If you decide to pass the SPIR-V through some sort of bytecode optimizer or such, this is the place
    pub fn get_code_mut(&mut self) -> Option<&mut [u32]> {
        match self {
            SpirvOrFinished::SpirvAndHash((spirv, _, _)) => Some(spirv.code.borrow_mut()),
            _ => None,
        }
    }

    /// Mutate the entry point name
    pub fn get_name_mut(&mut self) -> Option<&mut String> {
        match self {
            SpirvOrFinished::SpirvAndHash((spirv, _, _)) => Some(&mut spirv.name),
            _ => None,
        }
    }

    /// Mutate the parameters
    pub fn get_params_mut(&mut self) -> Option<&mut DeviceFnMutParams> {
        match self {
            SpirvOrFinished::SpirvAndHash((spirv, _, _)) => Some(&mut spirv.params),
            _ => None,
        }
    }

    /// Finish the compilation and return a `DeviceFnMut`
    pub fn finish(&self) -> Result<Arc<DeviceFnMut>, CompileOrNoDeviceError> {
        match self {
            SpirvOrFinished::SpirvAndHash((spirv, src_hash, _)) => {
                // compile SPIR-V to machine code (DeviceFnMut)
                // then put it in the cache and return it
                C::insert(
                    *src_hash,
                    Arc::new(
                        take()
                            .map_err(|_| CompileOrNoDeviceError::NoDevice)?
                            .lock()
                            .unwrap()
                            .compile::<_, &[u32]>(
                                spirv.params.clone(),
                                spirv.name.clone(),
                                spirv.code.borrow(),
                            )
                            .map_err(|_| CompileOrNoDeviceError::Compile)?,
                    ),
                );
                Ok(C::get(*src_hash))
            }
            SpirvOrFinished::Finished(device_fn_mut) => Ok(device_fn_mut.clone()),
        }
    }
}