Skip to content

Fix: null-check malloc/realloc in attention, sampling, and tokenizer #47

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
FHult wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix: null-check malloc/realloc in attention, sampling, and tokenizer #47

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 9 additions & 0 deletions iris.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -737,6 +737,11 @@ static iris_image *iris_generate_zimage_with_embeddings(iris_ctx *ctx,
* where latent_ch = in_ch * ps * ps = 64 */
int latent_ch = in_ch * ps * ps;
float *latent = (float *)malloc(latent_ch * post_h * post_w * sizeof(float));
if (!latent) {
free(denoised);
set_error("Out of memory allocating latent");
return NULL;
}
iris_patchify(latent, denoised, 1, in_ch, pre_h, pre_w, ps);
free(denoised);

Expand Down Expand Up @@ -1058,6 +1063,10 @@ iris_image *iris_generate_with_embeddings_and_noise(iris_ctx *ctx,

/* Copy external noise */
float *z = (float *)malloc(expected_noise_size * sizeof(float));
if (!z) {
set_error("Out of memory allocating noise buffer");
return NULL;
}
memcpy(z, noise, expected_noise_size * sizeof(float));

/* Get schedule */
Expand Down
15 changes: 15 additions & 0 deletions iris_kernels.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -658,6 +658,7 @@ void iris_attention(float *out, const float *Q, const float *K, const float *V,
float scale) {
/* Allocate attention scores */
float *scores = (float *)malloc(seq_q * seq_k * sizeof(float));
if (!scores) return;

for (int b = 0; b < batch; b++) {
for (int h = 0; h < heads; h++) {
Expand Down Expand Up @@ -791,6 +792,11 @@ static void flash_attention_head_tiled(float *out,
/* Per-query running statistics: max_score[seq_q], sum_exp[seq_q] */
float *max_scores = (float *)malloc(seq_q * sizeof(float));
float *sum_exps = (float *)malloc(seq_q * sizeof(float));
if (!max_scores || !sum_exps) {
free(max_scores);
free(sum_exps);
return;
}

/* Initialize */
for (int i = 0; i < seq_q; i++) {
Expand Down Expand Up @@ -904,6 +910,7 @@ void iris_flash_attention(float *out, const float *Q, const float *K, const floa

/* Allocate tile scratch buffer */
float *tile_scores = (float *)malloc(q_tile_size * k_tile_size * sizeof(float));
if (!tile_scores) return;

/* Process each head */
for (int h = 0; h < heads; h++) {
Expand All @@ -922,6 +929,10 @@ void iris_flash_attention(float *out, const float *Q, const float *K, const floa
float *K_contig = (float *)malloc(seq_k * head_dim * sizeof(float));
float *V_contig = (float *)malloc(seq_k * head_dim * sizeof(float));
float *out_contig = (float *)malloc(seq_q * head_dim * sizeof(float));
if (!Q_contig || !K_contig || !V_contig || !out_contig) {
free(Q_contig); free(K_contig); free(V_contig); free(out_contig);
break;
}

for (int i = 0; i < seq_q; i++) {
for (int d = 0; d < head_dim; d++) {
Expand Down Expand Up @@ -956,6 +967,10 @@ void iris_flash_attention(float *out, const float *Q, const float *K, const floa
float *K_contig = (float *)malloc(seq_k * head_dim * sizeof(float));
float *V_contig = (float *)malloc(seq_k * head_dim * sizeof(float));
float *out_contig = (float *)malloc(seq_q * head_dim * sizeof(float));
if (!Q_contig || !K_contig || !V_contig || !out_contig) {
free(Q_contig); free(K_contig); free(V_contig); free(out_contig);
break;
}

for (int i = 0; i < seq_q; i++) {
for (int d = 0; d < head_dim; d++) {
Expand Down
17 changes: 15 additions & 2 deletions iris_qwen3_tokenizer.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -753,6 +753,7 @@ static token_node_t *bpe_encode_word(qwen3_tokenizer_t *tok, const char *word) {
int len1 = strlen(best_node->text);
int len2 = strlen(best_node->next->text);
char *merged = malloc(len1 + len2 + 1);
if (!merged) break;
memcpy(merged, best_node->text, len1);
memcpy(merged + len1, best_node->next->text, len2);
merged[len1 + len2] = '\0';
Expand Down Expand Up @@ -876,12 +877,15 @@ static char **pretokenize(const char *text, int *num_chunks) {
if (p > start) {
int len = p - start;
char *chunk = malloc(len + 1);
if (!chunk) break;
memcpy(chunk, start, len);
chunk[len] = '\0';

if (count >= capacity) {
capacity *= 2;
chunks = realloc(chunks, capacity * sizeof(char *));
char **new_chunks = realloc(chunks, capacity * sizeof(char *));
if (!new_chunks) { free(chunk); break; }
chunks = new_chunks;
}
chunks[count++] = chunk;
}
Expand Down Expand Up @@ -926,7 +930,16 @@ int *qwen3_tokenize(qwen3_tokenizer_t *tok, const char *text,
if (id >= 0) {
if (total >= capacity) {
capacity *= 2;
tokens = realloc(tokens, capacity * sizeof(int));
int *new_tokens = realloc(tokens, capacity * sizeof(int));
if (!new_tokens) {
free_token_list(bpe_tokens);
free(byte_text);
for (int i = c + 1; i < num_chunks; i++) free(chunks[i]);
free(chunks);
*num_tokens = total;
return tokens;
}
tokens = new_tokens;
}
tokens[total++] = id;
}
Expand Down