sp1-jit 6.2.1

JIT compilation for SP1 trace generation
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
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
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
use crate::shm::ConsumerGuard;

use std::{marker::PhantomData, ops::Deref, sync::Arc};

use memmap2::Mmap;
use serde::{Deserialize, Serialize};

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
#[repr(u8)]
pub enum RiscRegister {
    X0 = 0,
    X1 = 1,
    X2 = 2,
    X3 = 3,
    X4 = 4,
    X5 = 5,
    X6 = 6,
    X7 = 7,
    X8 = 8,
    X9 = 9,
    X10 = 10,
    X11 = 11,
    X12 = 12,
    X13 = 13,
    X14 = 14,
    X15 = 15,
    X16 = 16,
    X17 = 17,
    X18 = 18,
    X19 = 19,
    X20 = 20,
    X21 = 21,
    X22 = 22,
    X23 = 23,
    X24 = 24,
    X25 = 25,
    X26 = 26,
    X27 = 27,
    X28 = 28,
    X29 = 29,
    X30 = 30,
    X31 = 31,
}

impl RiscRegister {
    pub fn all_registers() -> &'static [RiscRegister] {
        &[
            RiscRegister::X0,
            RiscRegister::X1,
            RiscRegister::X2,
            RiscRegister::X3,
            RiscRegister::X4,
            RiscRegister::X5,
            RiscRegister::X6,
            RiscRegister::X7,
            RiscRegister::X8,
            RiscRegister::X9,
            RiscRegister::X10,
            RiscRegister::X11,
            RiscRegister::X12,
            RiscRegister::X13,
            RiscRegister::X14,
            RiscRegister::X15,
            RiscRegister::X16,
            RiscRegister::X17,
            RiscRegister::X18,
            RiscRegister::X19,
            RiscRegister::X20,
            RiscRegister::X21,
            RiscRegister::X22,
            RiscRegister::X23,
            RiscRegister::X24,
            RiscRegister::X25,
            RiscRegister::X26,
            RiscRegister::X27,
            RiscRegister::X28,
            RiscRegister::X29,
            RiscRegister::X30,
            RiscRegister::X31,
        ]
    }
}

/// ALU operations can either have register or immediate operands.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
pub enum RiscOperand {
    Register(RiscRegister),
    Immediate(i32),
}

impl From<RiscRegister> for RiscOperand {
    fn from(reg: RiscRegister) -> Self {
        RiscOperand::Register(reg)
    }
}

impl From<u32> for RiscOperand {
    fn from(imm: u32) -> Self {
        RiscOperand::Immediate(imm as i32)
    }
}

impl From<i32> for RiscOperand {
    fn from(imm: i32) -> Self {
        RiscOperand::Immediate(imm)
    }
}

impl From<u64> for RiscOperand {
    fn from(imm: u64) -> Self {
        RiscOperand::Immediate(imm as i32)
    }
}

#[repr(C)]
#[derive(Default, Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub struct MemValue {
    pub clk: u64,
    pub value: u64,
}

/// Basic elf information. It extracts information from Program, but
/// does not introduce dependency on Program.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub struct ElfInfo {
    pub pc_base: u64,
    pub instruction_count: usize,
    pub untrusted_memory: Option<(u64, u64)>,
}

impl ElfInfo {
    #[inline]
    pub fn enable_untrusted_program(&self) -> bool {
        self.untrusted_memory.is_some()
    }
}

#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub struct PageProtValue {
    pub timestamp: u64,
    pub value: u8,
}

impl Default for PageProtValue {
    fn default() -> Self {
        Self { timestamp: 0, value: sp1_primitives::consts::DEFAULT_PAGE_PROT }
    }
}

impl From<PageProtValue> for MemValue {
    fn from(v: PageProtValue) -> MemValue {
        MemValue { clk: v.timestamp, value: v.value as u64 }
    }
}

impl From<MemValue> for PageProtValue {
    fn from(v: MemValue) -> PageProtValue {
        assert!(v.value & !0xff == 0, "value = {:x}", v.value);
        PageProtValue { timestamp: v.clk, value: (v.value & 0xff).try_into().unwrap() }
    }
}

/// A RISC-V interrupt, right now we are only doing trap with this
/// structure but it might be expanded later.
#[derive(Clone, Debug, PartialEq, Eq)]
pub struct Interrupt {
    /// Trap code
    pub code: u64,
}

/// A convience structure for getting offsets of fields in the actual [TraceChunk].
#[repr(C)]
pub struct TraceChunkHeader {
    pub start_registers: [u64; 32],
    pub pc_start: u64,
    pub clk_start: u64,
    pub clk_end: u64,
    pub num_mem_reads: u64,
    pub global_clk_end: u64,
    // This ensures TraceChunkHeader is aligned to 16 bytes.
    _padding: u64,
}

#[derive(Clone)]
pub enum TraceChunkRaw {
    Mmap(Arc<Mmap>),
    Shm(Arc<ConsumerGuard>),
}

impl TraceChunkRaw {
    /// # Safety
    ///
    /// - The mmap must be a valid [`TraceChunkHeader`].
    /// - The mmap must contain valid [`MemValue`]s in after the header.
    /// - The `num_mem_reads` must be the number of [`MemValue`]s in the mmap after the header.
    pub unsafe fn new(inner: Mmap) -> Self {
        Self::Mmap(Arc::new(inner))
    }

    /// # Safety
    ///
    /// See TraceChunkRaw::new for similar safety requirements
    pub unsafe fn from_shm(inner: ConsumerGuard) -> Self {
        Self::Shm(Arc::new(inner))
    }

    fn as_ref(&self) -> &[u8] {
        match self {
            Self::Mmap(mmap) => mmap.as_ref(),
            Self::Shm(shm) => shm.deref(),
        }
    }

    fn as_ptr(&self) -> *const u8 {
        match self {
            Self::Mmap(mmap) => mmap.as_ptr(),
            Self::Shm(shm) => shm.deref().as_ptr(),
        }
    }

    fn len(&self) -> usize {
        match self {
            Self::Mmap(mmap) => mmap.len(),
            Self::Shm(shm) => shm.deref().len(),
        }
    }

    /// Fetching global_clk when trace ends.
    /// For now, only native executor requires this data. So we implemented
    /// it as a method of TraceChunkRaw, not as a trait method of MinimalTrace.
    /// Adding it to MinimalTrace would complicate SplicingVM, while SplicingVM
    /// does not really need global_clk now.
    pub fn global_clk_end(&self) -> u64 {
        let offset = std::mem::offset_of!(TraceChunkHeader, global_clk_end);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const u64) }
    }
}

impl MinimalTrace for TraceChunkRaw {
    fn start_registers(&self) -> [u64; 32] {
        let offset = std::mem::offset_of!(TraceChunkHeader, start_registers);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const [u64; 32]) }
    }

    fn pc_start(&self) -> u64 {
        let offset = std::mem::offset_of!(TraceChunkHeader, pc_start);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const u64) }
    }

    fn clk_start(&self) -> u64 {
        let offset = std::mem::offset_of!(TraceChunkHeader, clk_start);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const u64) }
    }

    fn clk_end(&self) -> u64 {
        let offset = std::mem::offset_of!(TraceChunkHeader, clk_end);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const u64) }
    }

    fn num_mem_reads(&self) -> u64 {
        let offset = std::mem::offset_of!(TraceChunkHeader, num_mem_reads);

        unsafe { std::ptr::read_unaligned(self.as_ptr().add(offset) as *const u64) }
    }

    fn mem_reads(&self) -> MemReads<'_> {
        let header_end = std::mem::size_of::<TraceChunkHeader>();
        let len = self.num_mem_reads() as usize;

        debug_assert!(self.len() - header_end >= len);

        // SAFETY:
        // - The memory is valid assuming num_mem_reads is correct.
        // - The memory is technically always valid for reads since all bitpatterns are valid for
        //   `MemValue`.
        unsafe { MemReads::new(self.as_ptr().add(header_end) as *const MemValue, len) }
    }
}

pub struct MemReads<'a> {
    inner: *const MemValue,
    end: *const MemValue,
    /// Capture the lifetime of the buffer for saftey reasons.
    _phantom: PhantomData<&'a ()>,
}

impl<'a> MemReads<'a> {
    /// # Safety
    ///
    /// - The underlying memory is valid and contains valid `MemValue`s.
    /// - The length is the number of `MemValue`s in the underlying memory.
    pub(crate) unsafe fn new(inner: *const MemValue, len: usize) -> Self {
        debug_assert!(inner.is_aligned(), "MemReads ptr is not aligned");

        Self { inner, end: inner.add(len), _phantom: PhantomData }
    }

    /// Advance the pointer by `n` elements.
    ///
    /// # Panics
    ///
    /// Panics if `n` is greater than the purported length of the underlying buffer.
    pub fn advance(&mut self, n: usize) {
        unsafe {
            let advanced = self.inner.add(n);

            if advanced > self.end {
                panic!("Cannot advance by more than the length of the slice");
            }

            self.inner = advanced;
        }
    }

    /// Get the raw pointer to the head of the slice.
    pub fn head_raw(&self) -> *const MemValue {
        self.inner
    }

    /// The remaining length of the slice from our current position.
    #[must_use]
    pub fn len(&self) -> usize {
        unsafe { self.end.offset_from_unsigned(self.inner) }
    }

    /// Check if the iterator is empty.
    #[must_use]
    pub fn is_empty(&self) -> bool {
        self.inner == self.end
    }
}

impl<'a> Iterator for MemReads<'a> {
    type Item = MemValue;

    fn next(&mut self) -> Option<Self::Item> {
        if self.inner == self.end {
            None
        } else {
            let value = unsafe { std::ptr::read(self.inner) };
            self.inner = unsafe { self.inner.add(1) };

            Some(value)
        }
    }
}

/// A trace chunk is all the data needed to continue the execution of a program at
/// pc_start/clk_start.
///
/// We transmute this type directly from bytes, and the buffer should be of [TraceChunkRaw] form,
/// plus, a slice of the memory reads.
///
/// When we read this type from the buffer, we will copy the registers, the pc/clk start and end,
/// and take a pointer to the memory reads, by reading the num_mem_vals field.
///
/// The fields should be placed in the buffer according to the layout of [TraceChunkRaw].
#[repr(C)]
#[derive(Default, Debug, Clone, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub struct TraceChunk {
    pub start_registers: [u64; 32],
    pub pc_start: u64,
    pub clk_start: u64,
    pub clk_end: u64,
    #[serde(serialize_with = "ser::serialize_mem_reads")]
    #[serde(deserialize_with = "ser::deserialize_mem_reads")]
    pub mem_reads: Arc<[MemValue]>,
}

impl From<TraceChunkRaw> for TraceChunk {
    fn from(raw: TraceChunkRaw) -> Self {
        TraceChunk::copy_from_bytes(raw.as_ref())
    }
}

impl TraceChunk {
    /// Copy the bytes into a [TraceChunk]. We dont just back it with the original bytes,
    /// since this type is likely to be sent off to worker for proving.
    ///
    /// # Note:
    /// This method will panic if the buffer is not large enough,
    /// or the number of reads causes an overflow.
    pub fn copy_from_bytes(src: &[u8]) -> Self {
        const HDR: usize = size_of::<TraceChunkHeader>();

        /* ---------- 1. header must fit ---------- */
        if src.len() < HDR {
            panic!("TraceChunk header too small");
        }

        /* ---------- 2. copy-out the header ---------- */
        // SAFETY:
        // we just checked that `src` contains at least `HDR` bytes,
        // and `read_unaligne
        //
        // Note: All bit patterns are valid for `TraceChunkRaw`.
        let raw: TraceChunkHeader =
            unsafe { core::ptr::read_unaligned(src.as_ptr() as *const TraceChunkHeader) };

        /* ---------- 3. tail must fit ---------- */
        let n_words = raw.num_mem_reads as usize;
        let n_bytes = n_words.checked_mul(size_of::<MemValue>()).expect("Num mem reads too large");
        let total = HDR.checked_add(n_bytes).expect("Num mem reads too large");
        if src.len() < total {
            panic!("TraceChunk tail too small");
        }

        /* ---------- 4. extract tail ---------- */
        let tail = &src[HDR..total]; // only after the length check

        let mem_reads = Arc::new_uninit_slice(n_words);

        // SAFETY:
        // - The tail contains valid u64s, so doing a bitwise copy preserves the validity and
        //   endianness.
        // - tail is likely unaligned, so casting to a u8 pointer gives the alignmnt guarantee the
        //   compiler needs to do a copy.
        // - `mem_reads` was just allocated to have enough space.
        // - u8 has minimum alignment, so casting the pointer allocated by the vec is valid.
        // - The cast from const -> mut is valid since there are no other references to the memory.
        //
        // This trick is mostly taken from [`std::ptr::read_unaligned`]
        // see: <https://doc.rust-lang.org/src/core/ptr/mod.rs.html#1811>.
        unsafe {
            std::ptr::copy_nonoverlapping(tail.as_ptr(), mem_reads.as_ptr() as *mut u8, n_bytes)
        };

        Self {
            start_registers: raw.start_registers,
            pc_start: raw.pc_start,
            clk_start: raw.clk_start,
            clk_end: raw.clk_end,
            // SAFETY: We know the memory is initialized, so we can assume it.
            mem_reads: unsafe { mem_reads.assume_init() },
        }
    }
}

/// A trait that represents a minimal trace.
///
/// A minimal trace is the minimum required information to rexecute from
/// `pc_start` and `clk_start` -> `clk_end`.
///
/// It effectively acts as an oracle for the results of memory read operations.
pub trait MinimalTrace: Clone + Send + Sync + 'static {
    fn start_registers(&self) -> [u64; 32];

    fn pc_start(&self) -> u64;

    fn clk_start(&self) -> u64;

    fn clk_end(&self) -> u64;

    fn num_mem_reads(&self) -> u64;

    fn mem_reads(&self) -> MemReads<'_>;
}

impl MinimalTrace for TraceChunk {
    fn start_registers(&self) -> [u64; 32] {
        self.start_registers
    }

    fn pc_start(&self) -> u64 {
        self.pc_start
    }

    fn clk_start(&self) -> u64 {
        self.clk_start
    }

    fn clk_end(&self) -> u64 {
        self.clk_end
    }

    fn num_mem_reads(&self) -> u64 {
        self.mem_reads.len() as u64
    }

    fn mem_reads(&self) -> MemReads<'_> {
        // SAFETY:
        // - The memory is technically always valid for reads since all bitpatterns are valid for
        //   `MemValue`.
        // - the length comes directly from the Vec, which we know to be valid.
        unsafe { MemReads::new(self.mem_reads.as_ptr(), self.mem_reads.len()) }
    }
}

mod ser {
    use super::*;
    use serde::{Deserializer, Serializer};

    pub fn serialize_mem_reads<S: Serializer>(
        mem_reads: &Arc<[MemValue]>,
        serializer: S,
    ) -> Result<S::Ok, S::Error> {
        let as_vec: Vec<MemValue> = Vec::from(&mem_reads[..]);

        Vec::serialize(&as_vec, serializer)
    }

    pub fn deserialize_mem_reads<'a, D: Deserializer<'a>>(
        deserializer: D,
    ) -> Result<Arc<[MemValue]>, D::Error> {
        let as_vec = Vec::deserialize(deserializer)?;

        Ok(as_vec.into())
    }

    #[test]
    #[cfg(test)]
    fn test_mem_reads() {
        let mem_reads = Arc::new([MemValue { clk: 0, value: 0 }, MemValue { clk: 1, value: 1 }]);
        let trace = TraceChunk {
            start_registers: [5; 32],
            pc_start: 6,
            clk_start: 7,
            clk_end: 8,
            mem_reads,
        };

        let serialized = bincode::serialize(&trace).unwrap();
        let deserialized = bincode::deserialize(&serialized).unwrap();

        assert_eq!(trace, deserialized);
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    // TraceChunkHeader must be aligned to 16 bytes
    #[test]
    fn test_trace_chunk_header_alignment() {
        assert_eq!(std::mem::size_of::<TraceChunkHeader>() % 16, 0);
    }
}