Skip to content

imatrix: fix oob writes if src1 is not contiguous #13286

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
May 3, 2025
Merged

imatrix: fix oob writes if src1 is not contiguous #13286

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
14 changes: 8 additions & 6 deletions tools/imatrix/imatrix.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ class IMatrixCollector {
common_params m_params;
std::mutex m_mutex;
int m_last_call = 0;
std::vector<float> m_src1_data;
std::vector<char> m_src1_data;
std::vector<char> m_ids; // the expert ids from ggml_mul_mat_id
};

Expand Down Expand Up @@ -93,11 +93,13 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
const bool is_host = ggml_backend_buffer_is_host(src1->buffer);

if (!is_host) {
m_src1_data.resize(ggml_nelements(src1));
ggml_backend_tensor_get(src1, m_src1_data.data(), 0, ggml_nbytes(src1));
const size_t src1_nbytes = ggml_nbytes(src1);
m_src1_data.resize(src1_nbytes);
ggml_backend_tensor_get(src1, m_src1_data.data(), 0, src1_nbytes);
}

const float * data = is_host ? (const float *) src1->data : m_src1_data.data();
const char * data = is_host ? (const char *) src1->data : m_src1_data.data();
GGML_ASSERT(src1->nb[0] == ggml_element_size(src1));

// this has been adapted to the new format of storing merged experts in a single 3d tensor
// ref: https://github.com/ggml-org/llama.cpp/pull/6387
Expand Down Expand Up @@ -144,7 +146,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *

const int64_t i11 = idx % src1->ne[1];
const int64_t i12 = row;
const float * x = (const float *)((const char *)data + i11*src1->nb[1] + i12*src1->nb[2]);
const float * x = (const float *)(data + i11*src1->nb[1] + i12*src1->nb[2]);

for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[e_start + j] += x[j]*x[j];
Expand Down Expand Up @@ -180,7 +182,7 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
++e.ncall;
LOG_DBGV(2, "%s[%d]: %32s, %s, %5d x %5d, %d\n", __func__, m_last_call, wname.c_str(), ggml_op_name(t->op), (int)src1->ne[0], (int)src1->ne[1], (int)src1->type);
for (int row = 0; row < (int)src1->ne[1]; ++row) {
const float * x = data + row * src1->ne[0];
const float * x = (const float *) (data + row * src1->nb[1]);
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Excuse my ignorance if I'm asking silly questions, but why nb[1] instead of nb[0]?

Copy link
Collaborator Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nb[0] is the byte offset when changing dimension 0 by 1, nb[1] is the byte offset when changing dimension 1 by 1.

for (int j = 0; j < (int)src1->ne[0]; ++j) {
e.values[j] += x[j]*x[j];
e.counts[j]++;
Expand Down
Loading