Skip to content

batched-bench : fix llama_synchronize usage during prompt processing #15835

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

Merged
merged 1 commit into from
Sep 8, 2025
+12 −8
Merged

batched-bench : fix llama_synchronize usage during prompt processing #15835

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
20 changes: 12 additions & 8 deletions tools/batched-bench/batched-bench.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,7 +71,7 @@ int main(int argc, char ** argv) {
llama_batch batch = llama_batch_init(n_kv_max, 0, 1);

// decode in batches of ctx_params.n_batch tokens
auto decode_helper = [](llama_context * ctx, llama_batch & batch, int32_t n_batch) {
auto decode_helper = [](llama_context * ctx, llama_batch & batch, int32_t n_batch, bool synchronize) {
for (int32_t i = 0; i < (int32_t) batch.n_tokens; i += n_batch) {
const int32_t n_tokens = std::min(n_batch, (int32_t) (batch.n_tokens - i));

Expand All @@ -91,7 +91,9 @@ int main(int argc, char ** argv) {
return false;
}

llama_synchronize(ctx);
if (synchronize) {
llama_synchronize(ctx);
}
}

return true;
Expand All @@ -103,7 +105,7 @@ int main(int argc, char ** argv) {
common_batch_add(batch, get_token_rand(), i, { 0 }, false);
}

if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
if (!decode_helper(ctx, batch, ctx_params.n_batch, true)) {
LOG_ERR("%s: llama_decode() failed\n", __func__);
return 1;
}
Expand Down Expand Up @@ -138,15 +140,17 @@ int main(int argc, char ** argv) {
}
}

const auto t_pp_start = ggml_time_us();

llama_memory_clear(mem, false);

if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
const auto t_pp_start = ggml_time_us();

if (!decode_helper(ctx, batch, ctx_params.n_batch, false)) {
LOG_ERR("%s: llama_decode() failed\n", __func__);
return 1;
}

llama_synchronize(ctx);

const auto t_pp_end = ggml_time_us();

if (is_pp_shared) {
Expand All @@ -158,7 +162,7 @@ int main(int argc, char ** argv) {
// run one dummy token to apply the memory copy
common_batch_clear(batch);
common_batch_add(batch, get_token_rand(), pp + 0, { 0 }, true);
if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
if (!decode_helper(ctx, batch, ctx_params.n_batch, true)) {
LOG_ERR("%s: llama_decode() failed\n", __func__);
return 1;
}
Expand All @@ -175,7 +179,7 @@ int main(int argc, char ** argv) {
common_batch_add(batch, get_token_rand(), pp + i, { j }, true);
}

if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
if (!decode_helper(ctx, batch, ctx_params.n_batch, true)) {
LOG_ERR("%s: llama_decode() failed\n", __func__);
return 1;
}
Expand Down
Loading