Merge 680bdb58f7 into 4ec3be8bfa

MemorySanitizer does not support assembly, and therefore produces a
false positive on `blake3_hasher_finalize` and related functions.

c/blake3.c

@@ -5,6 +5,13 @@
 #include "blake3.h"
 #include "blake3_impl.h"
 
+#ifdef __has_feature
+#if __has_feature(memory_sanitizer)
+#include <sanitizer/msan_interface.h>
+#define BLAKE3_MEMORY_SANITIZER_BUILD
+#endif
+#endif
+
 const char *blake3_version(void) { return BLAKE3_VERSION_STRING; }
 
 INLINE void chunk_state_init(blake3_chunk_state *self, const uint32_t key[8],
@@ -583,6 +590,10 @@ void blake3_hasher_finalize_seek(const blake3_hasher *self, uint64_t seek,
   if (self->cv_stack_len == 0) {
     output_t output = chunk_state_output(&self->chunk);
     output_root_bytes(&output, seek, out, out_len);
+#ifdef BLAKE3_MEMORY_SANITIZER_BUILD
+    // MemorySanitizer gives a false positive due to use of assembly.
+    __msan_unpoison(out, out_len);
+#endif
     return;
   }
   // If there are any bytes in the chunk state, finalize that chunk and do a
@@ -611,6 +622,10 @@ void blake3_hasher_finalize_seek(const blake3_hasher *self, uint64_t seek,
     output = parent_output(parent_block, self->key, self->chunk.flags);
   }
   output_root_bytes(&output, seek, out, out_len);
+#ifdef BLAKE3_MEMORY_SANITIZER_BUILD
+  // MemorySanitizer gives a false positive due to use of assembly.
+  __msan_unpoison(out, out_len);
+#endif
 }
 
 void blake3_hasher_reset(blake3_hasher *self) {

reference_impl/reference_impl.rs

@@ -78,23 +78,14 @@ fn compress(
     block_len: u32,
     flags: u32,
 ) -> [u32; 16] {
+    let counter_low = counter as u32;
+    let counter_high = (counter >> 32) as u32;
+    #[rustfmt::skip]
     let mut state = [
-        chaining_value[0],
-        chaining_value[1],
-        chaining_value[2],
-        chaining_value[3],
-        chaining_value[4],
-        chaining_value[5],
-        chaining_value[6],
-        chaining_value[7],
-        IV[0],
-        IV[1],
-        IV[2],
-        IV[3],
-        counter as u32,
-        (counter >> 32) as u32,
-        block_len,
-        flags,
+        chaining_value[0], chaining_value[1], chaining_value[2], chaining_value[3],
+        chaining_value[4], chaining_value[5], chaining_value[6], chaining_value[7],
+        IV[0],             IV[1],             IV[2],             IV[3],
+        counter_low,       counter_high,      block_len,         flags,
     ];
     let mut block = *block_words;