Integrate DS4 execution parity in Rust

src/engine/gguf.rs

@@ -1,4 +1,4 @@
-use memmap2::{Mmap, MmapOptions};
+use memmap2::{Advice, Mmap, MmapOptions};
 use sha2::{Digest, Sha256};
 use std::collections::HashMap;
 use std::fs::File;
@@ -215,6 +215,23 @@ impl Gguf {
         self.max_tensor_bytes
     }
 
+    pub(super) fn warm(&self) -> Result<(), String> {
+        let start = self.data_offset as usize;
+        if start >= self.map.len() {
+            return Ok(());
+        }
+        self.map
+            .advise_range(Advice::WillNeed, start, self.map.len() - start)
+            .map_err(|error| format!("Cannot warm {}: {error}", self.path.display()))?;
+        let mut checksum = 0_u64;
+        for offset in (start..self.map.len()).step_by(16 * 1024) {
+            checksum = checksum.wrapping_add(u64::from(self.map[offset]));
+        }
+        checksum = checksum.wrapping_add(u64::from(self.map[self.map.len() - 1]));
+        std::hint::black_box(checksum);
+        Ok(())
+    }
+
     pub(super) fn tensor(&self, name: &str) -> Result<&Tensor, String> {
         self.tensors
             .get(name)
@@ -544,6 +561,7 @@ mod tests {
         assert_eq!(model.bytes("general.architecture").unwrap(), b"deepseek4");
         assert_eq!(model.tensor("weight").unwrap().dims, [1]);
         assert_eq!(model.tensor_data("weight").unwrap(), 1_f32.to_le_bytes());
+        model.warm().unwrap();
         fs::remove_file(path).unwrap();
     }