Fall back to CPU in small FRI.

starkware-libs · Sep 24, 2024 · d14ba6b · d14ba6b
1 parent e1acffa
commit d14ba6b
Show file tree

Hide file tree

Showing 5 changed files with 44 additions and 9 deletions.
diff --git a/crates/prover/src/core/backend/cpu/fri.rs b/crates/prover/src/core/backend/cpu/fri.rs
@@ -17,6 +17,7 @@ impl FriOps for CpuBackend {
     ) -> LineEvaluation<Self> {
         fold_line(eval, alpha)
     }
+
     fn fold_circle_into_line(
         dst: &mut LineEvaluation<Self>,
         src: &SecureEvaluation<Self, BitReversedOrder>,

diff --git a/crates/prover/src/core/backend/simd/fri.rs b/crates/prover/src/core/backend/simd/fri.rs
@@ -12,7 +12,7 @@ use crate::core::backend::Column;
 use crate::core::fields::m31::BaseField;
 use crate::core::fields::qm31::SecureField;
 use crate::core::fields::secure_column::SecureColumnByCoords;
-use crate::core::fri::{self, FriOps};
+use crate::core::fri::{self, fold_circle_into_line, FriOps};
 use crate::core::poly::circle::SecureEvaluation;
 use crate::core::poly::line::LineEvaluation;
 use crate::core::poly::twiddles::TwiddleTree;
@@ -63,7 +63,16 @@ impl FriOps for SimdBackend {
         twiddles: &TwiddleTree<Self>,
     ) {
         let log_size = src.len().ilog2();
-        assert!(log_size > LOG_N_LANES, "Evaluation too small");
+        if log_size <= LOG_N_LANES {
+            // Fall back to CPU implementation.
+            let mut cpu_dst = dst.to_cpu();
+            fold_circle_into_line(&mut cpu_dst, &src.to_cpu(), alpha);
+            *dst = LineEvaluation::new(
+                cpu_dst.domain(),
+                SecureColumnByCoords::from_cpu(cpu_dst.values),
+            );
+            return;
+        }
 
         let domain = src.domain;
         let alpha_sq = alpha * alpha;
@@ -249,7 +258,7 @@ mod tests {
         };
         let avx_eval = SecureEvaluation::new(domain, avx_column.clone());
         let cpu_eval =
-            SecureEvaluation::<CpuBackend, BitReversedOrder>::new(domain, avx_eval.to_cpu());
+            SecureEvaluation::<CpuBackend, BitReversedOrder>::new(domain, avx_eval.values.to_cpu());
         let (cpu_g, cpu_lambda) = CpuBackend::decompose(&cpu_eval);
         let (avx_g, avx_lambda) = SimdBackend::decompose(&avx_eval);
 

diff --git a/crates/prover/src/core/poly/circle/secure_poly.rs b/crates/prover/src/core/poly/circle/secure_poly.rs
@@ -84,6 +84,14 @@ impl<B: FieldOps<BaseField>, EvalOrder> SecureEvaluation<B, EvalOrder> {
         let Self { domain, values, .. } = self;
         values.columns.map(|c| CircleEvaluation::new(domain, c))
     }
+
+    pub fn to_cpu(&self) -> SecureEvaluation<CpuBackend, EvalOrder> {
+        SecureEvaluation {
+            domain: self.domain,
+            values: self.values.to_cpu(),
+            _eval_order: PhantomData,
+        }
+    }
 }
 
 impl<B: FieldOps<BaseField>, EvalOrder> Deref for SecureEvaluation<B, EvalOrder> {

diff --git a/crates/prover/src/examples/wide_fibonacci/mod.rs b/crates/prover/src/examples/wide_fibonacci/mod.rs
@@ -1,7 +1,7 @@
 use itertools::Itertools;
 
 use crate::constraint_framework::{EvalAtRow, FrameworkComponent, FrameworkEval};
-use crate::core::backend::simd::m31::{PackedBaseField, LOG_N_LANES};
+use crate::core::backend::simd::m31::PackedBaseField;
 use crate::core::backend::simd::SimdBackend;
 use crate::core::backend::{Col, Column};
 use crate::core::fields::m31::BaseField;
@@ -47,8 +47,6 @@ pub fn generate_trace<const N: usize>(
     log_size: u32,
     inputs: &[FibInput],
 ) -> ColumnVec<CircleEvaluation<SimdBackend, BaseField, BitReversedOrder>> {
-    assert!(log_size >= LOG_N_LANES);
-    assert_eq!(inputs.len(), 1 << (log_size - LOG_N_LANES));
     let mut trace = (0..N)
         .map(|_| Col::<SimdBackend, BaseField>::zeros(1 << log_size))
         .collect_vec();
@@ -72,7 +70,7 @@ pub fn generate_trace<const N: usize>(
 #[cfg(test)]
 mod tests {
     use itertools::Itertools;
-    use num_traits::One;
+    use num_traits::{One, Zero};
 
     use super::WideFibonacciEval;
     use crate::constraint_framework::{
@@ -101,6 +99,26 @@ mod tests {
     fn generate_test_trace(
         log_n_instances: u32,
     ) -> ColumnVec<CircleEvaluation<SimdBackend, BaseField, BitReversedOrder>> {
+        if log_n_instances < LOG_N_LANES {
+            let n_instances = 1 << log_n_instances;
+            let inputs = vec![FibInput {
+                a: PackedBaseField::from_array(std::array::from_fn(|j| {
+                    if j < n_instances {
+                        BaseField::one()
+                    } else {
+                        BaseField::zero()
+                    }
+                })),
+                b: PackedBaseField::from_array(std::array::from_fn(|j| {
+                    if j < n_instances {
+                        BaseField::from_u32_unchecked((j) as u32)
+                    } else {
+                        BaseField::zero()
+                    }
+                })),
+            }];
+            return generate_trace::<FIB_SEQUENCE_LENGTH>(log_n_instances, &inputs);
+        }
         let inputs = (0..(1 << (log_n_instances - LOG_N_LANES)))
             .map(|i| FibInput {
                 a: PackedBaseField::one(),
@@ -151,7 +169,7 @@ mod tests {
 
     #[test_log::test]
     fn test_wide_fib_prove_with_blake() {
-        for log_n_instances in 4..=6 {
+        for log_n_instances in 2..=6 {
             let config = PcsConfig::default();
             // Precompute twiddles.
             let twiddles = SimdBackend::precompute_twiddles(

diff --git a/crates/prover/src/examples/xor/gkr_lookups/mle_eval.rs b/crates/prover/src/examples/xor/gkr_lookups/mle_eval.rs
@@ -389,7 +389,6 @@ mod tests {
     }
 
     #[test]
-    #[ignore = "SimdBackend `MIN_FFT_LOG_SIZE` is 5"]
     fn eq_constraints_with_4_variables() {
         const N_VARIABLES: usize = 4;
         const EQ_EVAL_TRACE: usize = 0;