Skip to content

feat(hexagon): implement Q4_1 support for MUL_MAT ops #56

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
max-krasnyansky wants to merge 4 commits into
base: master
Choose a base branch
from
Open

feat(hexagon): implement Q4_1 support for MUL_MAT ops #56

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
386302e
feat(hexagon): implement Q4_1 support for MUL_MAT ops
google-labs-jules[bot] May 10, 2026
7ee209c
feat(hexagon): implement Q4_1 support for MUL_MAT ops
google-labs-jules[bot] May 10, 2026
ea6feeb
feat(hexagon): implement Q4_1 support for MUL_MAT ops
google-labs-jules[bot] May 10, 2026
8cf393a
feat(hexagon): implement Q4_1 support for MUL_MAT ops
google-labs-jules[bot] May 10, 2026
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
260 changes: 260 additions & 0 deletions ggml/src/ggml-hexagon/ggml-hexagon.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -544,6 +544,19 @@ static void init_row_q4x4x2(block_q4_0 * x, int64_t k) {
}
}


static inline void unpack_q4_1_quants(uint8_t * y, const block_q4_1 * x, int b) {
for (int i = 0; i < QK4_1 / 2; i++) {
y[b * QK4_1 / 2 + i + 000] = x->qs[i] & 0x0F;
y[b * QK4_1 / 2 + i + 128] = x->qs[i] >> 4;
}
}
static inline void pack_q4_1_quants(block_q4_1 * y, const uint8_t * x, int b) {
for (int i = 0; i < QK4_1 / 2; i++) {
y->qs[i] = (x[b * QK4_1 / 2 + i + 000] & 0x0F) | (x[b * QK4_1 / 2 + i + 128] << 4);
}
}

// repack q4_0 data into q4x4x2 tensor
static void repack_q4_0_q4x4x2(ggml_tensor * t, const void * data, size_t size) {
int64_t nrows = ggml_nrows(t);
Expand Down Expand Up @@ -605,6 +618,237 @@ static void repack_q4_0_q4x4x2(ggml_tensor * t, const void * data, size_t size)
ggml_aligned_free(buf_rp, row_size_rp);
}


static void repack_row_q4x4x2_q4_1(uint8_t * y, const block_q4_1 * x, int64_t k) {
static const int qk = QK_Q4_1x4x2;
const int nb = (k + qk - 1) / qk; // number of blocks (padded)
const int nloe = k % qk; // leftovers

const int dblk_size = 8 * 2; // 8x __fp16 for d
const int mblk_size = 8 * 2; // 8x __fp16 for m
const int qblk_size = qk / 2; // int4
const int qrow_size = k / 2; // int4 (not padded to blocks)
const int drow_size = (k + qk - 1) / qk * (qk / 32 * 2); // padded drow_size

uint8_t * y_q = y + 0; // quants first
uint8_t * y_d = y + qrow_size; // then scales
uint8_t * y_m = y + qrow_size + drow_size; // then mins

// Repack the quants
for (int i = 0; i < nb; i++) {
uint8_t qs[QK_Q4_1x4x2]; // unpacked quants
unpack_q4_1_quants(qs, &x[i * 8 + 0], 0);
unpack_q4_1_quants(qs, &x[i * 8 + 1], 1);
unpack_q4_1_quants(qs, &x[i * 8 + 2], 2);
unpack_q4_1_quants(qs, &x[i * 8 + 3], 3);
unpack_q4_1_quants(qs, &x[i * 8 + 4], 4);
unpack_q4_1_quants(qs, &x[i * 8 + 5], 5);
unpack_q4_1_quants(qs, &x[i * 8 + 6], 6);
unpack_q4_1_quants(qs, &x[i * 8 + 7], 7);

bool partial = (nloe && i == nb-1);

uint8_t * q = y_q + (i * qblk_size);
for (int j = 0; j < qk / 2; j++) {
q[j] = partial ? (qs[j*2+1] << 4) | qs[j*2+0] : (qs[j+128] << 4) | qs[j+000];
}
}

// Repack the scales and minimums
for (int i = 0; i < nb; i++) {
// Repack the scales
ggml_half * d = (ggml_half *) (y_d + i * dblk_size);
d[0] = x[i * 8 + 0].d;
d[1] = x[i * 8 + 1].d;
d[2] = x[i * 8 + 2].d;
d[3] = x[i * 8 + 3].d;
d[4] = x[i * 8 + 4].d;
d[5] = x[i * 8 + 5].d;
d[6] = x[i * 8 + 6].d;
d[7] = x[i * 8 + 7].d;

// Repack the minimums
ggml_half * m = (ggml_half *) (y_m + i * mblk_size);
m[0] = x[i * 8 + 0].m;
m[1] = x[i * 8 + 1].m;
m[2] = x[i * 8 + 2].m;
m[3] = x[i * 8 + 3].m;
m[4] = x[i * 8 + 4].m;
m[5] = x[i * 8 + 5].m;
m[6] = x[i * 8 + 6].m;
m[7] = x[i * 8 + 7].m;
}
}

static void unpack_row_q4x4x2_q4_1(block_q4_1 * y, const uint8_t * x, int64_t k) {
static const int qk = QK_Q4_1x4x2;
const int nb = (k + qk - 1) / qk; // number of blocks (padded)
const int nloe = k % qk; // leftovers

const int dblk_size = 8 * 2; // 8x __fp16 for d
const int mblk_size = 8 * 2; // 8x __fp16 for m
const int qblk_size = qk / 2; // int4
const int qrow_size = k / 2; // int4 (not padded to blocks)
const int drow_size = (k + qk - 1) / qk * (qk / 32 * 2); // padded drow_size

const uint8_t * x_q = x + 0; // quants first
const uint8_t * x_d = x + qrow_size; // then scales
const uint8_t * x_m = x + qrow_size + drow_size; // then mins

// Unpack the quants
for (int i = 0; i < nb; i++) {
uint8_t qs[QK_Q4_1x4x2]; // unpacked quants

bool partial = (nloe && i == nb-1);

const uint8_t * q = x_q + (i * qblk_size);
for (int j = 0; j < qk / 2; j++) {
if (partial) {
qs[j*2+0] = q[j] & 0x0F;
qs[j*2+1] = q[j] >> 4;
} else {
qs[j+000] = q[j] & 0x0F;
qs[j+128] = q[j] >> 4;
}
}

pack_q4_1_quants(&y[i * 8 + 0], qs, 0);
pack_q4_1_quants(&y[i * 8 + 1], qs, 1);
pack_q4_1_quants(&y[i * 8 + 2], qs, 2);
pack_q4_1_quants(&y[i * 8 + 3], qs, 3);
pack_q4_1_quants(&y[i * 8 + 4], qs, 4);
pack_q4_1_quants(&y[i * 8 + 5], qs, 5);
pack_q4_1_quants(&y[i * 8 + 6], qs, 6);
pack_q4_1_quants(&y[i * 8 + 7], qs, 7);
}

// Unpack the scales and minimums
for (int i = 0; i < nb; i++) {
// Unpack the scales
const ggml_half * d = (const ggml_half *) (x_d + i * dblk_size);
y[i * 8 + 0].d = d[0];
y[i * 8 + 1].d = d[1];
y[i * 8 + 2].d = d[2];
y[i * 8 + 3].d = d[3];
y[i * 8 + 4].d = d[4];
y[i * 8 + 5].d = d[5];
y[i * 8 + 6].d = d[6];
y[i * 8 + 7].d = d[7];

// Unpack the minimums
const ggml_half * m = (const ggml_half *) (x_m + i * mblk_size);
y[i * 8 + 0].m = m[0];
y[i * 8 + 1].m = m[1];
y[i * 8 + 2].m = m[2];
y[i * 8 + 3].m = m[3];
y[i * 8 + 4].m = m[4];
y[i * 8 + 5].m = m[5];
y[i * 8 + 6].m = m[6];
y[i * 8 + 7].m = m[7];
}
}

static inline void init_row_q4_1x4x2(block_q4_1 * x, int64_t k) {
const int nb = (k + QK_Q4_1x4x2 - 1) / QK_Q4_1x4x2;
memset(x, 0, nb * QK_Q4_1x4x2 / 2 + nb * 8 * 2 * 2);
}

static void repack_q4_1_q4x4x2(ggml_tensor * t, const void * data, size_t size) {
int64_t nrows = ggml_nrows(t);

size_t row_size = ggml_row_size(t->type, t->ne[0]);
size_t row_size_pd = ggml_row_size(t->type, hex_round_up(t->ne[0], QK_Q4_1x4x2)); // extra elements for the pad
size_t row_size_rp = row_size * 2; // extra space for tmp pad (if any)

const size_t total_tensor_size = (size_t)nrows * row_size;
const size_t n_bytes_to_copy = size < total_tensor_size ? size : total_tensor_size;

const int64_t n_full_rows = n_bytes_to_copy / row_size;
const size_t n_rem_bytes = n_bytes_to_copy % row_size;

void * buf_pd = ggml_aligned_malloc(row_size_pd);
GGML_ASSERT(buf_pd != NULL);

void * buf_rp = ggml_aligned_malloc(row_size_rp);
GGML_ASSERT(buf_rp != NULL);

HEX_VERBOSE("ggml-hex: repack-q4_1-q4x4x2 %s : data %p size %zu dims %ldx%ld row-size %zu\n", t->name, data, size,
t->ne[0], nrows, row_size);

init_row_q4_1x4x2((block_q4_1 *) buf_pd, t->ne[0]);

for (int64_t i = 0; i < n_full_rows; i++) {
const uint8_t * src = (const uint8_t *) data + (i * row_size);
uint8_t * dst = (uint8_t *) t->data + (i * row_size);

memcpy(buf_pd, src, row_size);
repack_row_q4x4x2_q4_1((uint8_t *) buf_rp, (const block_q4_1 *) buf_pd, t->ne[0]);
memcpy(dst, buf_rp, row_size);
}

if (n_rem_bytes > 0) {
const int64_t i = n_full_rows;
const uint8_t * src = (const uint8_t *) data + (i * row_size);
uint8_t * dst = (uint8_t *) t->data + (i * row_size);

init_row_q4_1x4x2((block_q4_1 *) buf_pd, t->ne[0]);
memcpy(buf_pd, src, n_rem_bytes);
repack_row_q4x4x2_q4_1((uint8_t *) buf_rp, (const block_q4_1 *) buf_pd, t->ne[0]);
memcpy(dst, buf_rp, n_rem_bytes);
}

ggml_aligned_free(buf_pd, row_size_pd);
ggml_aligned_free(buf_rp, row_size_rp);
}

static void repack_q4x4x2_q4_1(void * data, const ggml_tensor * t, size_t size) {
int64_t nrows = ggml_nrows(t);

size_t row_size = ggml_row_size(t->type, t->ne[0]);
size_t row_size_pd = ggml_row_size(t->type, hex_round_up(t->ne[0], QK_Q4_1x4x2)); // extra elements for the pad
size_t row_size_rp = row_size * 2; // extra space for tmp pad (if any)

const size_t total_tensor_size = (size_t)nrows * row_size;
const size_t n_bytes_to_copy = size < total_tensor_size ? size : total_tensor_size;

const int64_t n_full_rows = n_bytes_to_copy / row_size;
const size_t n_rem_bytes = n_bytes_to_copy % row_size;

void * buf_pd = ggml_aligned_malloc(row_size_pd);
GGML_ASSERT(buf_pd != NULL);

void * buf_rp = ggml_aligned_malloc(row_size_rp);
GGML_ASSERT(buf_rp != NULL);

HEX_VERBOSE("ggml-hex: repack-q4x4x2-q4_1 %s : data %p size %zu dims %ldx%ld row-size %zu\n", t->name, data, size,
t->ne[0], nrows, row_size);

memset(buf_pd, 0, row_size_pd);

for (int64_t i = 0; i < n_full_rows; i++) {
const uint8_t * src = (const uint8_t *) t->data + (i * row_size);
uint8_t * dst = (uint8_t *) data + (i * row_size);

memcpy(buf_pd, src, row_size);
unpack_row_q4x4x2_q4_1((block_q4_1 *) buf_rp, (const uint8_t *) buf_pd, t->ne[0]);
memcpy(dst, buf_rp, row_size);
}

if (n_rem_bytes > 0) {
const int64_t i = n_full_rows;
const uint8_t * src = (const uint8_t *) t->data + (i * row_size);
uint8_t * dst = (uint8_t *) data + (i * row_size);

memcpy(buf_pd, src, n_rem_bytes);
unpack_row_q4x4x2_q4_1((block_q4_1 *) buf_rp, (const uint8_t *) buf_pd, t->ne[0]);
memcpy(dst, buf_rp, n_rem_bytes);
}

ggml_aligned_free(buf_pd, row_size_pd);
ggml_aligned_free(buf_rp, row_size_rp);
}


// repack q4x4x2 tensor into q4_0 data
static void repack_q4x4x2_q4_0(void * data, const ggml_tensor * t, size_t size) {
int64_t nrows = ggml_nrows(t);
Expand Down Expand Up @@ -1365,6 +1609,12 @@ static void ggml_backend_hexagon_buffer_set_tensor(ggml_backend_buffer_t buffer,
repack_q4_0_q4x4x2(tensor, data, size);
break;

case GGML_TYPE_Q4_1:
GGML_ASSERT(offset == 0);
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
repack_q4_1_q4x4x2(tensor, data, size);
break;

case GGML_TYPE_Q8_0:
GGML_ASSERT(offset == 0);
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
Expand Down Expand Up @@ -1407,6 +1657,12 @@ static void ggml_backend_hexagon_buffer_get_tensor(ggml_backend_buffer_t buffer,
repack_q4x4x2_q4_0(data, tensor, size);
break;

case GGML_TYPE_Q4_1:
GGML_ASSERT(offset == 0);
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
repack_q4x4x2_q4_1(data, tensor, size);
break;

case GGML_TYPE_Q8_0:
GGML_ASSERT(offset == 0);
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
Expand Down Expand Up @@ -2327,6 +2583,7 @@ static bool ggml_hexagon_supported_mul_mat(const struct ggml_hexagon_session * s

switch (src0->type) {
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q8_0:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_MXFP4:
Expand Down Expand Up @@ -2377,6 +2634,7 @@ static bool ggml_hexagon_supported_mul_mat_id(const struct ggml_hexagon_session

switch (src0->type) {
case GGML_TYPE_Q4_0:
case GGML_TYPE_Q4_1:
case GGML_TYPE_Q8_0:
case GGML_TYPE_IQ4_NL:
case GGML_TYPE_MXFP4:
Expand Down Expand Up @@ -3558,6 +3816,8 @@ static void ggml_hexagon_init(ggml_backend_reg * reg) {
// Basic sanity checks to make sure definitions match
static_assert((unsigned int) HTP_TYPE_Q4_0 == (unsigned int) GGML_TYPE_Q4_0,
"please update hexagon_type to match ggml_type");
static_assert((unsigned int) HTP_TYPE_Q4_1 == (unsigned int) GGML_TYPE_Q4_1,
"please update hexagon_type to match ggml_type");
static_assert((unsigned int) HTP_TYPE_Q8_0 == (unsigned int) GGML_TYPE_Q8_0,
"please update hexagon_type to match ggml_type");
static_assert((unsigned int) HTP_TYPE_MXFP4 == (unsigned int) GGML_TYPE_MXFP4,
Expand Down
Loading
Loading