reorder struct members

Signed-off-by: Pantelis Antoniou <[email protected]>

src/blake3/blake3.c

@@ -144,7 +144,7 @@ static inline void blake3_chunk_state_update(blake3_host_state *hs, blake3_chunk
 // bytes.
 static inline void blake3_output_chaining_value(blake3_host_state *hs, const blake3_output_t *self, uint8_t cv[32])
 {
-	uint32_t cv_words[8];
+	uint32_t cv_words[8] BLAKE3_ALIGN;
 
 	memcpy(cv_words, self->input_cv, 32);
 	hs->compress_in_place(cv_words, self->block, self->block_len, self->counter, self->flags);
@@ -153,9 +153,9 @@ static inline void blake3_output_chaining_value(blake3_host_state *hs, const bla
 
 static inline void blake3_output_root_bytes(blake3_host_state *hs, const blake3_output_t *self, uint64_t seek, uint8_t *out, size_t out_len)
 {
+	uint8_t wide_buf[64] BLAKE3_ALIGN;
 	uint64_t output_block_counter;
 	size_t offset_within_block;
-	uint8_t wide_buf[64];
 	size_t available_bytes;
 	size_t memcpy_len;
 
@@ -651,9 +651,9 @@ void blake3_hasher_reset(blake3_hasher *self)
 
 void blake3_hasher_finalize_seek(const blake3_hasher *self, uint64_t seek, uint8_t *out, size_t out_len)
 {
+	uint8_t parent_block[BLAKE3_BLOCK_LEN] BLAKE3_ALIGN;
 	blake3_output_t output;
 	size_t cvs_remaining;
-	uint8_t parent_block[BLAKE3_BLOCK_LEN];
 
 	// Explicitly checking for zero avoids causing UB by passing a null pointer
 	// to memcpy. This comes up in practice with things like:

src/blake3/blake3_internal.h

@@ -33,32 +33,32 @@ enum blake3_flags {
 
 typedef struct _blake3_output {
 	uint32_t input_cv[8] BLAKE3_ALIGN;
-	uint64_t counter;
 	uint8_t block[BLAKE3_BLOCK_LEN] BLAKE3_ALIGN;
+	uint64_t counter;
 	uint8_t block_len;
 	uint8_t flags;
 } blake3_output_t BLAKE3_ALIGN;
 
 typedef struct _blake3_chunk_state {
 	uint32_t cv[8] BLAKE3_ALIGN;
-	uint64_t chunk_counter;
 	uint8_t buf[BLAKE3_BLOCK_LEN] BLAKE3_ALIGN;
+	uint64_t chunk_counter;
 	uint8_t buf_len;
 	uint8_t blocks_compressed;
 	uint8_t flags;
 } blake3_chunk_state BLAKE3_ALIGN;
 
 typedef struct _blake3_hasher {
-	struct _blake3_host_state *hs;
 	uint32_t key[8] BLAKE3_ALIGN;
 	blake3_chunk_state chunk;
-	uint8_t cv_stack_len;
 	// The stack size is MAX_DEPTH + 1 because we do lazy merging. For example,
 	// with 7 chunks, we have 3 entries in the stack. Adding an 8th chunk
 	// requires a 4th entry, rather than merging everything down to 1, because we
 	// don't know whether more input is coming. This is different from how the
 	// reference implementation does things.
 	uint8_t cv_stack[(BLAKE3_MAX_DEPTH + 1) * BLAKE3_OUT_LEN] BLAKE3_ALIGN;
+	uint8_t cv_stack_len;
+	struct _blake3_host_state *hs;
 } blake3_hasher BLAKE3_ALIGN;
 
 typedef void (*blake3_hash_many_f)(const uint8_t *const *inputs, size_t num_inputs,