feat: update cuda tracegen for boundary chip

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: 876pol

crates/vm/src/system/cuda/memory.rs

@@ -1,7 +1,7 @@
 use std::sync::Arc;
 
 use openvm_circuit::{
-    arch::{AddressSpaceHostLayout, DenseRecordArena, MemoryConfig, ADDR_SPACE_OFFSET},
+    arch::{AddressSpaceHostLayout, CONST_BLOCK_SIZE, DenseRecordArena, MemoryConfig, ADDR_SPACE_OFFSET},
     system::{
         memory::{online::LinearMemory, AddressMap, TimestampedValues},
         TouchedMemory,
@@ -151,15 +151,15 @@ impl MemoryInventoryGPU {
                 mem.tracing_info("boundary finalize");
                 let (touched_memory, empty) = if partition.is_empty() {
                     let leftmost_values = 'left: {
-                        let mut res = [F::ZERO; DIGEST_WIDTH];
+                        let mut res = [F::ZERO; CONST_BLOCK_SIZE];
                         if persistent.initial_memory[ADDR_SPACE_OFFSET as usize].is_empty() {
                             break 'left res;
                         }
                         let layout = &persistent.merkle_tree.mem_config().addr_spaces
                             [ADDR_SPACE_OFFSET as usize]
                             .layout;
                         let one_cell_size = layout.size();
-                        let values = vec![0u8; one_cell_size * DIGEST_WIDTH];
+                        let values = vec![0u8; one_cell_size * CONST_BLOCK_SIZE];
                         unsafe {
                             cuda_memcpy::<true, false>(
                                 values.as_ptr() as *mut std::ffi::c_void,
@@ -168,7 +168,7 @@ impl MemoryInventoryGPU {
                                 values.len(),
                             )
                             .unwrap();
-                            for i in 0..DIGEST_WIDTH {
+                            for i in 0..CONST_BLOCK_SIZE {
                                 res[i] = layout.to_field::<F>(&values[i * one_cell_size..]);
                             }
                         }

crates/vm/src/system/cuda/merkle_tree/mod.rs

@@ -1,7 +1,7 @@
 use std::{ffi::c_void, sync::Arc};
 
 use openvm_circuit::{
-    arch::{MemoryConfig, ADDR_SPACE_OFFSET},
+    arch::{MemoryConfig, ADDR_SPACE_OFFSET, CONST_BLOCK_SIZE},
     system::memory::{merkle::MemoryMerkleCols, TimestampedEquipartition},
     utils::next_power_of_two_or_zero,
 };
@@ -26,7 +26,9 @@ pub mod cuda;
 use cuda::merkle_tree::*;
 
 type H = [F; DIGEST_WIDTH];
-pub const TIMESTAMPED_BLOCK_WIDTH: usize = 11;
+/// Width of `((u32, u32), TimestampedValues<F, CONST_BLOCK_SIZE>)` in u32 units.
+/// = 2 (key) + 1 (timestamp) + CONST_BLOCK_SIZE (values)
+pub const TIMESTAMPED_BLOCK_WIDTH: usize = 3 + CONST_BLOCK_SIZE;
 
 /// A Merkle subtree stored in a single flat buffer, combining a vertical path and a heap-ordered
 /// binary tree.
@@ -393,7 +395,7 @@ impl MemoryMerkleTree {
     /// An auxiliary function to calculate the required number of rows for the merkle trace.
     pub fn calculate_unpadded_height(
         &self,
-        touched_memory: &TimestampedEquipartition<F, DIGEST_WIDTH>,
+        touched_memory: &TimestampedEquipartition<F, CONST_BLOCK_SIZE>,
     ) -> usize {
         let md = self.mem_config.memory_dimensions();
         let tree_height = md.overall_height();