sp1-core-machine 6.0.0-beta.1

SP1 is a performant, 100% open-source, contributor-friendly zkVM.
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
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
use super::MemoryChipType;
use crate::{
    air::{SP1CoreAirBuilder, SP1Operation, WordAirBuilder},
    operations::{
        IsZeroOperation, IsZeroOperationInput, LtOperationUnsigned, LtOperationUnsignedInput,
    },
    utils::next_multiple_of_32,
};
use core::{
    borrow::{Borrow, BorrowMut},
    mem::{size_of, MaybeUninit},
};
use slop_air::{Air, AirBuilder, BaseAir};
use slop_algebra::{AbstractField, PrimeField32};
use slop_matrix::Matrix;
use slop_maybe_rayon::prelude::{
    IndexedParallelIterator, IntoParallelRefIterator, ParallelIterator, ParallelSlice,
    ParallelSliceMut,
};
use sp1_core_executor::{
    events::{ByteRecord, GlobalInteractionEvent, MemoryInitializeFinalizeEvent},
    ExecutionRecord, Program,
};
use sp1_derive::AlignedBorrow;
use sp1_hypercube::{
    air::{AirInteraction, InteractionScope, MachineAir},
    InteractionKind, Word,
};
use sp1_primitives::consts::u64_to_u16_limbs;
use std::iter::once;

/// A memory chip that can initialize or finalize values in memory.
pub struct MemoryGlobalChip {
    pub kind: MemoryChipType,
}

impl MemoryGlobalChip {
    /// Creates a new memory chip with a certain type.
    pub const fn new(kind: MemoryChipType) -> Self {
        Self { kind }
    }
}

impl<F> BaseAir<F> for MemoryGlobalChip {
    fn width(&self) -> usize {
        NUM_MEMORY_INIT_COLS
    }
}

impl<F: PrimeField32> MachineAir<F> for MemoryGlobalChip {
    type Record = ExecutionRecord;

    type Program = Program;

    fn name(&self) -> &'static str {
        match self.kind {
            MemoryChipType::Initialize => "MemoryGlobalInit",
            MemoryChipType::Finalize => "MemoryGlobalFinalize",
        }
    }

    fn generate_dependencies(&self, input: &ExecutionRecord, output: &mut ExecutionRecord) {
        let mut memory_events = match self.kind {
            MemoryChipType::Initialize => input.global_memory_initialize_events.clone(),
            MemoryChipType::Finalize => input.global_memory_finalize_events.clone(),
        };

        let is_receive = match self.kind {
            MemoryChipType::Initialize => false,
            MemoryChipType::Finalize => true,
        };

        match self.kind {
            MemoryChipType::Initialize => {
                output.public_values.global_init_count += memory_events.len() as u32;
            }
            MemoryChipType::Finalize => {
                output.public_values.global_finalize_count += memory_events.len() as u32;
            }
        };

        let previous_addr = match self.kind {
            MemoryChipType::Initialize => input.public_values.previous_init_addr,
            MemoryChipType::Finalize => input.public_values.previous_finalize_addr,
        };

        memory_events.sort_by_key(|event| event.addr);

        let chunk_size = std::cmp::max(memory_events.len() / num_cpus::get(), 1);
        let indices = (0..memory_events.len()).collect::<Vec<_>>();
        let blu_batches = indices
            .par_chunks(chunk_size)
            .map(|chunk| {
                let mut blu = Vec::new();
                let mut row = [F::zero(); NUM_MEMORY_INIT_COLS];
                let cols: &mut MemoryInitCols<F> = row.as_mut_slice().borrow_mut();
                chunk.iter().for_each(|&i| {
                    let addr = memory_events[i].addr;
                    let value = memory_events[i].value;
                    let prev_addr = if i == 0 { previous_addr } else { memory_events[i - 1].addr };
                    blu.add_u16_range_checks(&u64_to_u16_limbs(value));
                    blu.add_u16_range_checks(&u64_to_u16_limbs(prev_addr)[0..3]);
                    blu.add_u16_range_checks(&u64_to_u16_limbs(addr)[0..3]);
                    let value_lower = (value >> 32 & 0xFF) as u8;
                    let value_upper = (value >> 40 & 0xFF) as u8;
                    blu.add_u8_range_check(value_lower, value_upper);
                    if i != 0 || prev_addr != 0 {
                        cols.lt_cols.populate_unsigned(&mut blu, 1, prev_addr, addr);
                    }
                });
                blu
            })
            .collect::<Vec<_>>();
        output.add_byte_lookup_events(blu_batches.into_iter().flatten().collect());

        let events = memory_events.into_iter().map(|event| {
            let interaction_clk_high = if is_receive { (event.timestamp >> 24) as u32 } else { 0 };
            let interaction_clk_low =
                if is_receive { (event.timestamp & 0xFFFFFF) as u32 } else { 0 };
            let limb_1 =
                (event.value & 0xFFFF) as u32 + (1 << 16) * (event.value >> 32 & 0xFF) as u32;
            let limb_2 =
                (event.value >> 16 & 0xFFFF) as u32 + (1 << 16) * (event.value >> 40 & 0xFF) as u32;

            GlobalInteractionEvent {
                message: [
                    interaction_clk_high,
                    interaction_clk_low,
                    (event.addr & 0xFFFF) as u32,
                    ((event.addr >> 16) & 0xFFFF) as u32,
                    ((event.addr >> 32) & 0xFFFF) as u32,
                    limb_1,
                    limb_2,
                    ((event.value >> 48) & 0xFFFF) as u32,
                ],
                is_receive,
                kind: InteractionKind::Memory as u8,
            }
        });
        output.global_interaction_events.extend(events);
    }

    fn num_rows(&self, input: &Self::Record) -> Option<usize> {
        let events = match self.kind {
            MemoryChipType::Initialize => &input.global_memory_initialize_events,
            MemoryChipType::Finalize => &input.global_memory_finalize_events,
        };
        let nb_rows = events.len();
        let size_log2 = input.fixed_log2_rows::<F, Self>(self);
        let padded_nb_rows = next_multiple_of_32(nb_rows, size_log2);
        Some(padded_nb_rows)
    }

    fn generate_trace_into(
        &self,
        input: &ExecutionRecord,
        _output: &mut ExecutionRecord,
        buffer: &mut [MaybeUninit<F>],
    ) {
        let mut memory_events = match self.kind {
            MemoryChipType::Initialize => input.global_memory_initialize_events.clone(),
            MemoryChipType::Finalize => input.global_memory_finalize_events.clone(),
        };

        let previous_addr = match self.kind {
            MemoryChipType::Initialize => input.public_values.previous_init_addr,
            MemoryChipType::Finalize => input.public_values.previous_finalize_addr,
        };

        memory_events.sort_by_key(|event| event.addr);

        let padded_nb_rows = <MemoryGlobalChip as MachineAir<F>>::num_rows(self, input).unwrap();
        let num_event_rows = memory_events.len();

        unsafe {
            let padding_start = num_event_rows * NUM_MEMORY_INIT_COLS;
            let padding_size = (padded_nb_rows - num_event_rows) * NUM_MEMORY_INIT_COLS;
            if padding_size > 0 {
                core::ptr::write_bytes(buffer[padding_start..].as_mut_ptr(), 0, padding_size);
            }
        }

        let buffer_ptr = buffer.as_mut_ptr() as *mut F;
        let values = unsafe {
            core::slice::from_raw_parts_mut(buffer_ptr, num_event_rows * NUM_MEMORY_INIT_COLS)
        };

        values.par_chunks_exact_mut(NUM_MEMORY_INIT_COLS).zip(memory_events.par_iter()).for_each(
            |(row, event)| {
                let cols: &mut MemoryInitCols<F> = row.borrow_mut();
                let MemoryInitializeFinalizeEvent { addr, value, timestamp } = event.to_owned();

                cols.addr[0] = F::from_canonical_u16((addr & 0xFFFF) as u16);
                cols.addr[1] = F::from_canonical_u16(((addr >> 16) & 0xFFFF) as u16);
                cols.addr[2] = F::from_canonical_u16(((addr >> 32) & 0xFFFF) as u16);
                cols.clk_high = F::from_canonical_u32((timestamp >> 24) as u32);
                cols.clk_low = F::from_canonical_u32((timestamp & 0xFFFFFF) as u32);
                cols.value = Word::from(value);
                cols.is_real = F::one();
                cols.value_lower = F::from_canonical_u32((value >> 32 & 0xFF) as u32);
                cols.value_upper = F::from_canonical_u32((value >> 40 & 0xFF) as u32);
            },
        );

        let mut blu = vec![];
        for i in 0..memory_events.len() {
            let row_start = i * NUM_MEMORY_INIT_COLS;
            let row = &mut values[row_start..row_start + NUM_MEMORY_INIT_COLS];
            let cols: &mut MemoryInitCols<F> = row.borrow_mut();

            let addr = memory_events[i].addr;
            let prev_addr = if i == 0 { previous_addr } else { memory_events[i - 1].addr };

            if prev_addr == 0 && i != 0 {
                cols.prev_valid = F::zero();
            } else {
                cols.prev_valid = F::one();
            }
            cols.index = F::from_canonical_u32(i as u32);
            cols.prev_addr[0] = F::from_canonical_u16((prev_addr & 0xFFFF) as u16);
            cols.prev_addr[1] = F::from_canonical_u16(((prev_addr >> 16) & 0xFFFF) as u16);
            cols.prev_addr[2] = F::from_canonical_u16(((prev_addr >> 32) & 0xFFFF) as u16);
            cols.is_prev_addr_zero.populate_from_field_element(
                cols.prev_addr[0] + cols.prev_addr[1] + cols.prev_addr[2],
            );
            cols.is_index_zero.populate(i as u64);
            if prev_addr != 0 || i != 0 {
                cols.is_comp = F::one();
                cols.lt_cols.populate_unsigned(&mut blu, 1, prev_addr, addr);
            } else {
                cols.is_comp = F::zero();
                cols.lt_cols = LtOperationUnsigned::<F>::default();
            }
        }
    }

    fn included(&self, shard: &Self::Record) -> bool {
        if let Some(shape) = shard.shape.as_ref() {
            shape.included::<F, _>(self)
        } else {
            match self.kind {
                MemoryChipType::Initialize => !shard.global_memory_initialize_events.is_empty(),
                MemoryChipType::Finalize => !shard.global_memory_finalize_events.is_empty(),
            }
        }
    }
}

#[derive(AlignedBorrow, Clone, Copy)]
#[repr(C)]
pub struct MemoryInitCols<T: Copy> {
    /// The top bits of the timestamp of the memory access.
    pub clk_high: T,

    /// The low bits of the timestamp of the memory access.
    pub clk_low: T,

    /// The index of the memory access.
    pub index: T,

    /// The address of the previous memory access.
    pub prev_addr: [T; 3],

    /// The address of the memory access.
    pub addr: [T; 3],

    /// Comparison assertions for address to be strictly increasing.
    pub lt_cols: LtOperationUnsigned<T>,

    /// The value of the memory access.
    pub value: Word<T>,

    /// Lower half of third limb of the value
    pub value_lower: T,

    /// Upper half of third limb of the value
    pub value_upper: T,

    /// Whether the memory access is a real access.
    pub is_real: T,

    /// Whether or not we are making the assertion `prev_addr < addr`.
    pub is_comp: T,

    /// The validity of previous state.
    /// The unique invalid state is when the chip only initializes address 0 once.
    pub prev_valid: T,

    /// A witness to assert whether or not `prev_addr` is zero.
    pub is_prev_addr_zero: IsZeroOperation<T>,

    /// A witness to assert whether or not the index is zero.
    pub is_index_zero: IsZeroOperation<T>,
}

pub(crate) const NUM_MEMORY_INIT_COLS: usize = size_of::<MemoryInitCols<u8>>();

impl<AB> Air<AB> for MemoryGlobalChip
where
    AB: SP1CoreAirBuilder,
{
    fn eval(&self, builder: &mut AB) {
        let main = builder.main();
        let local = main.row_slice(0);
        let local: &MemoryInitCols<AB::Var> = (*local).borrow();

        // Constrain that `local.is_real` is boolean.
        builder.assert_bool(local.is_real);
        // Constrain that the value is a valid `Word`.
        builder.slice_range_check_u16(&local.value.0, local.is_real);
        // Constrain that the previous address is a valid `Word`.
        builder.slice_range_check_u16(&local.prev_addr, local.is_real);
        // Constrain that the address is a valid `Word`.
        builder.slice_range_check_u16(&local.addr, local.is_real);

        // Assert that value_lower and value_upper are the lower and upper halves of the third limb.
        builder.assert_eq(
            local.value.0[2],
            local.value_lower + local.value_upper * AB::F::from_canonical_u32(1 << 8),
        );
        builder.slice_range_check_u8(&[local.value_lower, local.value_upper], local.is_real);

        let interaction_kind = match self.kind {
            MemoryChipType::Initialize => InteractionKind::MemoryGlobalInitControl,
            MemoryChipType::Finalize => InteractionKind::MemoryGlobalFinalizeControl,
        };

        // Receive the previous index, address, and validity state.
        builder.receive(
            AirInteraction::new(
                vec![local.index]
                    .into_iter()
                    .chain(local.prev_addr)
                    .chain(once(local.prev_valid))
                    .map(Into::into)
                    .collect(),
                local.is_real.into(),
                interaction_kind,
            ),
            InteractionScope::Local,
        );

        // Send the next index, address, and validity state.
        builder.send(
            AirInteraction::new(
                vec![local.index + AB::Expr::one()]
                    .into_iter()
                    .chain(local.addr.map(Into::into))
                    .chain(once(local.is_comp.into()))
                    .collect(),
                local.is_real.into(),
                interaction_kind,
            ),
            InteractionScope::Local,
        );

        if self.kind == MemoryChipType::Initialize {
            // Send the "send interaction" to the global table.
            builder.send(
                AirInteraction::new(
                    vec![
                        AB::Expr::zero(),
                        AB::Expr::zero(),
                        local.addr[0].into(),
                        local.addr[1].into(),
                        local.addr[2].into(),
                        local.value.0[0] + local.value_lower * AB::F::from_canonical_u32(1 << 16),
                        local.value.0[1] + local.value_upper * AB::F::from_canonical_u32(1 << 16),
                        local.value.0[3].into(),
                        AB::Expr::one(),
                        AB::Expr::zero(),
                        AB::Expr::from_canonical_u8(InteractionKind::Memory as u8),
                    ],
                    local.is_real.into(),
                    InteractionKind::Global,
                ),
                InteractionScope::Local,
            );
        } else {
            // Send the "receive interaction" to the global table.
            builder.send(
                AirInteraction::new(
                    vec![
                        local.clk_high.into(),
                        local.clk_low.into(),
                        local.addr[0].into(),
                        local.addr[1].into(),
                        local.addr[2].into(),
                        local.value.0[0] + local.value_lower * AB::F::from_canonical_u32(1 << 16),
                        local.value.0[1] + local.value_upper * AB::F::from_canonical_u32(1 << 16),
                        local.value.0[3].into(),
                        AB::Expr::zero(),
                        AB::Expr::one(),
                        AB::Expr::from_canonical_u8(InteractionKind::Memory as u8),
                    ],
                    local.is_real.into(),
                    InteractionKind::Global,
                ),
                InteractionScope::Local,
            );
        }

        // Assert that `prev_addr < addr` when `prev_addr != 0` or `index != 0`.
        // First, check if `prev_addr != 0`, and check if `index != 0`.
        // SAFETY: Since `prev_addr` are composed of valid u16 limbs, adding them to check if
        // all three limbs are zero is safe, as overflows are impossible.
        IsZeroOperation::<AB::F>::eval(
            builder,
            IsZeroOperationInput::new(
                local.prev_addr[0] + local.prev_addr[1] + local.prev_addr[2],
                local.is_prev_addr_zero,
                local.is_real.into(),
            ),
        );
        IsZeroOperation::<AB::F>::eval(
            builder,
            IsZeroOperationInput::new(
                local.index.into(),
                local.is_index_zero,
                local.is_real.into(),
            ),
        );

        // Comparison will be done unless both `prev_addr == 0` and `index == 0`.
        // If `is_real = 0`, then `is_comp` will be zero.
        // If `is_real = 1`, then `is_comp` will be zero when `prev_addr == 0` and `index == 0`.
        // If `is_real = 1`, then `is_comp` will be one when `prev_addr != 0` or `index != 0`.
        builder.assert_eq(
            local.is_comp,
            local.is_real
                * (AB::Expr::one() - local.is_prev_addr_zero.result * local.is_index_zero.result),
        );
        builder.assert_bool(local.is_comp);

        // If `is_comp = 1`, then `prev_addr < addr` should hold.
        <LtOperationUnsigned<AB::F> as SP1Operation<AB>>::eval(
            builder,
            LtOperationUnsignedInput::<AB>::new(
                Word([
                    local.prev_addr[0].into(),
                    local.prev_addr[1].into(),
                    local.prev_addr[2].into(),
                    AB::Expr::zero(),
                ]),
                Word([
                    local.addr[0].into(),
                    local.addr[1].into(),
                    local.addr[2].into(),
                    AB::Expr::zero(),
                ]),
                local.lt_cols,
                local.is_comp.into(),
            ),
        );
        builder.when(local.is_comp).assert_one(local.lt_cols.u16_compare_operation.bit);

        // If `prev_addr == 0` and `index == 0`, then `addr == 0`, and the `value` should be zero.
        // SAFETY: Since `local.addr` is valid u16 limbs, one can constrain that the sum of the
        // limbs is zero in order to constrain that `addr == 0`, as no overflow is possible.
        // This forces the initialization of address 0 with value 0.
        // Constraints related to register %x0: Register %x0 should always be 0.
        // See 2.6 Load and Store Instruction on P.18 of the RISC-V spec.
        let is_not_comp = local.is_real - local.is_comp;
        builder
            .when(is_not_comp.clone())
            .assert_zero(local.addr[0] + local.addr[1] + local.addr[2]);
        builder.when(is_not_comp.clone()).assert_word_zero(local.value);
    }
}

// #[cfg(test)]
// mod tests {
//     #![allow(clippy::print_stdout)]

//     use super::*;
//     use crate::programs::tests::*;
//     use crate::{
//         riscv::RiscvAir, syscall::precompiles::sha256::extend_tests::sha_extend_program,
//         utils::setup_logger,
//     };
//     use sp1_primitives::SP1Field;
//     use sp1_core_executor::{Executor, Trace};
//     use sp1_hypercube::InteractionKind;
//     use sp1_hypercube::{
//         koala_bear_poseidon2::SP1InnerPcs, debug_interactions_with_all_chips,
// SP1CoreOpts,         StarkMachine,
//     };

//     #[test]
//     fn test_memory_generate_trace() {
//         let program = simple_program();
//         let mut runtime = Executor::new(program, SP1CoreOpts::default());
//         runtime.run::<Trace>().unwrap();
//         let shard = runtime.record.clone();

//         let chip: MemoryGlobalChip = MemoryGlobalChip::new(MemoryChipType::Initialize);

//         let trace: RowMajorMatrix<SP1Field> =
//             chip.generate_trace(&shard, &mut ExecutionRecord::default());
//         println!("{:?}", trace.values);

//         let chip: MemoryGlobalChip = MemoryGlobalChip::new(MemoryChipType::Finalize);
//         let trace: RowMajorMatrix<SP1Field> =
//             chip.generate_trace(&shard, &mut ExecutionRecord::default());
//         println!("{:?}", trace.values);

//         for mem_event in shard.global_memory_finalize_events {
//             println!("{:?}", mem_event);
//         }
//     }

//     #[test]
//     fn test_memory_lookup_interactions() {
//         setup_logger();
//         let program = sha_extend_program();
//         let program_clone = program.clone();
//         let mut runtime = Executor::new(program, SP1CoreOpts::default());
//         runtime.run::<Trace>().unwrap();
//         let machine: StarkMachine<SP1InnerPcs, RiscvAir<SP1Field>> =
//             RiscvAir::machine(SP1InnerPcs::new());
//         let (pkey, _) = machine.setup(&program_clone);
//         let opts = SP1CoreOpts::default();
//         machine.generate_dependencies(
//             &mut runtime.records.clone().into_iter().map(|r| *r).collect::<Vec<_>>(),
//             &opts,
//             None,
//         );

//         let shards = runtime.records;
//         for shard in shards.clone() {
//             debug_interactions_with_all_chips::<SP1InnerPcs, RiscvAir<SP1Field>>(
//                 &machine,
//                 &pkey,
//                 &[*shard],
//                 vec![InteractionKind::Memory],
//                 InteractionScope::Local,
//             );
//         }
//         debug_interactions_with_all_chips::<SP1InnerPcs, RiscvAir<SP1Field>>(
//             &machine,
//             &pkey,
//             &shards.into_iter().map(|r| *r).collect::<Vec<_>>(),
//             vec![InteractionKind::Memory],
//             InteractionScope::Global,
//         );
//     }

//     #[test]
//     fn test_byte_lookup_interactions() {
//         setup_logger();
//         let program = sha_extend_program();
//         let program_clone = program.clone();
//         let mut runtime = Executor::new(program, SP1CoreOpts::default());
//         runtime.run::<Trace>().unwrap();
//         let machine = RiscvAir::machine(SP1InnerPcs::new());
//         let (pkey, _) = machine.setup(&program_clone);
//         let opts = SP1CoreOpts::default();
//         machine.generate_dependencies(
//             &mut runtime.records.clone().into_iter().map(|r| *r).collect::<Vec<_>>(),
//             &opts,
//             None,
//         );

//         let shards = runtime.records;
//         debug_interactions_with_all_chips::<SP1InnerPcs, RiscvAir<SP1Field>>(
//             &machine,
//             &pkey,
//             &shards.into_iter().map(|r| *r).collect::<Vec<_>>(),
//             vec![InteractionKind::Byte],
//             InteractionScope::Global,
//         );
//     }
// }