Skip to content

hexagon: add Q4_1 support to MUL_MAT #55

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 1 commit into
base: master
Choose a base branch
from
Open

hexagon: add Q4_1 support to MUL_MAT #55

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
248 changes: 248 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 @@ -330,6 +330,242 @@ static void dump_packed_block_q4x4x2(const uint8_t * v, unsigned int i, size_t k
GGML_FP16_TO_FP32(d[4]), GGML_FP16_TO_FP32(d[5]), GGML_FP16_TO_FP32(d[6]), GGML_FP16_TO_FP32(d[7]));
}






// initialize q4_1 buffer (padding with zeros)
static void init_row_q4_1x4x2(block_q4_1 * x, int64_t k) {
static const int qk = QK_Q4_1x4x2;
const int nb = (k + qk - 1) / qk;

for (int i = 0; i < nb; i++) {
x[i * 8 + 0].d = 0;
x[i * 8 + 1].d = 0;
x[i * 8 + 2].d = 0;
x[i * 8 + 3].d = 0;
x[i * 8 + 4].d = 0;
x[i * 8 + 5].d = 0;
x[i * 8 + 6].d = 0;
x[i * 8 + 7].d = 0;
x[i * 8 + 0].m = 0;
x[i * 8 + 1].m = 0;
x[i * 8 + 2].m = 0;
x[i * 8 + 3].m = 0;
x[i * 8 + 4].m = 0;
x[i * 8 + 5].m = 0;
x[i * 8 + 6].m = 0;
x[i * 8 + 7].m = 0;
}
}

static void unpack_q4_1_quants(uint8_t * x, const block_q4_1 * b, unsigned int bi) {
static const int qk = QK4_1;

for (unsigned int i = 0; i < qk / 2; ++i) {
const uint8_t q0 = b->qs[i] & 0x0F;
const uint8_t q1 = b->qs[i] >> 4;
x[bi * qk + i + 0] = q0;
x[bi * qk + i + qk / 2] = q1;
}
}

static void pack_q4_1_quants(block_q4_1 * x, const uint8_t * qs, unsigned int bi) {
static const int qk = QK4_1;

for (unsigned int i = 0; i < qk / 2; ++i) {
const uint8_t x0 = qs[bi * qk + i + 0];
const uint8_t x1 = qs[bi * qk + i + qk / 2];
x->qs[i] = x0 | (x1 << 4);
}
}


static void repack_row_q4_1x4x2(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;
const int nloe = k % qk;

const int dblk_size = 8 * 2;
const int qblk_size = qk / 2;
const int qrow_size = k / 2;
const int drow_size = k / 32 * 2; // k/32 blocks, each has 2 bytes d

uint8_t * y_q = y + 0;
uint8_t * y_d = y + qrow_size;
uint8_t * y_m = y_d + drow_size;

for (int i = 0; i < nb; i++) {
uint8_t qs[QK_Q4_1x4x2];
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++) {
if (partial && j >= nloe/2) break;
const uint8_t x0 = qs[j + 0];
const uint8_t x1 = qs[j + qk / 2];
q[j] = x0 | (x1 << 4);
}

uint8_t * d = y_d + (i * dblk_size);
uint8_t * m = y_m + (i * dblk_size);

int max_blks = partial ? nloe / 32 : 8;
for (int j = 0; j < 8; j++) {
if (j < max_blks) {
((ggml_fp16_t*)d)[j] = x[i * 8 + j].d;
((ggml_fp16_t*)m)[j] = x[i * 8 + j].m;
} else {
((ggml_fp16_t*)d)[j] = 0;
((ggml_fp16_t*)m)[j] = 0;
}
}
}
}

static void unpack_row_q4_1x4x2(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;
const int nloe = k % qk;

const int dblk_size = 8 * 2;
const int qblk_size = qk / 2;
const int qrow_size = k / 2;
const int drow_size = k / 32 * 2;

const uint8_t * y_q = (const uint8_t *)x;
const uint8_t * y_d = y_q + qrow_size;
const uint8_t * y_m = y_d + drow_size;

block_q4_1 * out = (block_q4_1 *)y;

for (int i = 0; i < nb; i++) {
uint8_t qs[QK_Q4_1x4x2];
const uint8_t * q = y_q + (i * qblk_size);

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

for (int j = 0; j < qk / 2; j++) {
if (partial && j >= nloe/2) {
qs[j + 0] = 0;
qs[j + qk / 2] = 0;
} else {
qs[j + 0] = q[j] & 0x0f;
qs[j + qk / 2] = q[j] >> 4;
}
}

int max_blks = partial ? nloe / 32 : 8;
for (int j = 0; j < max_blks; j++) {
pack_q4_1_quants(&out[i * 8 + j], qs, j);
}

const uint8_t * d = y_d + (i * dblk_size);
const uint8_t * m = y_m + (i * dblk_size);

for (int j = 0; j < max_blks; j++) {
out[i * 8 + j].d = ((ggml_fp16_t*)d)[j];
out[i * 8 + j].m = ((ggml_fp16_t*)m)[j];
}
}
}

static void repack_q4_1_q4_1x4x2(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-q4_1x4x2 %s : data %p size %zu dims %ldx%ld row-size %zu
", 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_q4_1x4x2((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);

memcpy(buf_pd, src, n_rem_bytes);
repack_row_q4_1x4x2((uint8_t *) buf_rp, (const block_q4_1 *) buf_pd, t->ne[0]);
memcpy(dst, buf_rp, row_size);
}

ggml_aligned_free(buf_rp);
ggml_aligned_free(buf_pd);
}

static void repack_q4_1x4x2_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));

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);

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);

unpack_row_q4_1x4x2((uint8_t *) buf_pd, (const block_q4_1 *) src, t->ne[0]);
memcpy(dst, buf_pd, 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);

unpack_row_q4_1x4x2((uint8_t *) buf_pd, (const block_q4_1 *) src, t->ne[0]);
memcpy(dst, buf_pd, n_rem_bytes);
}

ggml_aligned_free(buf_pd);
}




static void unpack_q4_0_quants(uint8_t * qs, const block_q4_0 * x, unsigned int bi) {
static const int qk = QK4_0;

Expand Down Expand Up @@ -1349,6 +1585,11 @@ static void ggml_backend_hexagon_buffer_set_tensor(ggml_backend_buffer_t buffer,
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
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_q4_1x4x2(tensor, data, size);
break;

case GGML_TYPE_Q8_0:
GGML_ASSERT(offset == 0);
Expand Down Expand Up @@ -1391,6 +1632,11 @@ static void ggml_backend_hexagon_buffer_get_tensor(ggml_backend_buffer_t buffer,
GGML_ASSERT(offset + size <= ggml_nbytes(tensor));
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_q4_1x4x2_q4_1(data, tensor, size);
break;

case GGML_TYPE_Q8_0:
GGML_ASSERT(offset == 0);
Expand Down Expand Up @@ -2163,6 +2409,7 @@ 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 @@ -2213,6 +2460,7 @@ 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
24 changes: 24 additions & 0 deletions ggml/src/ggml-hexagon/htp/hmx-matmul-ops.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,8 @@ static inline size_t get_x4x2_row_stride(int weight_type, int k) {
case HTP_TYPE_Q4_0:
case HTP_TYPE_IQ4_NL:
return (size_t) nb * (QK_Q4_0x4x2 / 2 + HMX_X4X2_DBLK_SIZE); // 144 * nb
case HTP_TYPE_Q4_1:
return (size_t) nb * (QK_Q4_0x4x2 / 2 + HMX_X4X2_DBLK_SIZE * 2); // 160 * nb
case HTP_TYPE_Q8_0:
return (size_t) nb * (QK_Q8_0x4x2 + HMX_X4X2_DBLK_SIZE); // 272 * nb
case HTP_TYPE_MXFP4:
Expand Down Expand Up @@ -430,6 +432,18 @@ static void dequantize_x4x2_weight_to_fp16_tiles_task(
v1[0] = v1[1] = v1[2] = v1[3] = Q6_V_vzero();
}

if (weight_type == HTP_TYPE_Q4_1) {
int m_off = (k_block / 32) * 2;
for (int g = 0; g < 4; g++) {
HVX_Vector m0 = Q6_V_vsplat_R(*(uint32_t *)(r0 + scale_off + g * sizeof(__fp16) + m_off));
v0[g] = Q6_Vhf_vadd_VhfVhf(v0[g], m0);
if (row1 < n_cols) {
HVX_Vector m1 = Q6_V_vsplat_R(*(uint32_t *)(r1 + scale_off + g * sizeof(__fp16) + m_off));
v1[g] = Q6_Vhf_vadd_VhfVhf(v1[g], m1);
}
}
}

for (int g = 0; g < 4; g++) { Q6_vscatter_QRMVwV(q_mask64, (size_t)tile_bases[g], HMX_FP16_TILE_SIZE - 1, v_off, v0[g]); }
v_off = Q6_Vw_vadd_VwVw(v_off, v_scat_step);
for (int g = 0; g < 4; g++) { Q6_vscatter_QRMVwV(q_mask64, (size_t)tile_bases[g], HMX_FP16_TILE_SIZE - 1, v_off, v1[g]); }
Expand Down Expand Up @@ -517,6 +531,16 @@ static void dequantize_x4x2_weight_to_fp16_tiles_task(
r1 + byte_off, upper, (const __fp16 *)(r1 + scale_off), vlut_cvt)
: Q6_V_vzero();

if (weight_type == HTP_TYPE_Q4_1) {
int m_off = (k_block / 32) * 2;
HVX_Vector m0 = Q6_V_vsplat_R(*(uint32_t *)(r0 + scale_off + m_off));
v0 = Q6_Vhf_vadd_VhfVhf(v0, m0);
if (row1 < n_cols) {
HVX_Vector m1 = Q6_V_vsplat_R(*(uint32_t *)(r1 + scale_off + m_off));
v1 = Q6_Vhf_vadd_VhfVhf(v1, m1);
}
}

Q6_vscatter_QRMVwV(q_mask64, (size_t)tile_base, HMX_FP16_TILE_SIZE - 1, v_off, v0);
v_off = Q6_Vw_vadd_VwVw(v_off, v_scat_step);
Q6_vscatter_QRMVwV(q_mask64, (size_t)tile_base, HMX_FP16_TILE_SIZE - 1, v_off, v1);
Expand Down
6 changes: 5 additions & 1 deletion ggml/src/ggml-hexagon/htp/htp-ops.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,7 @@ enum htp_data_type {
HTP_TYPE_F32 = 0,
HTP_TYPE_F16 = 1,
HTP_TYPE_Q4_0 = 2,
HTP_TYPE_Q4_1 = 3,
HTP_TYPE_Q8_0 = 8,
HTP_TYPE_IQ4_NL = 20,
HTP_TYPE_I32 = 26,
Expand All @@ -28,14 +29,17 @@ enum htp_data_type {

// types used internally for repack, dyn.quant, etc
HTP_TYPE_Q4_0x4x2 = 200,
HTP_TYPE_Q4_1x4x2,
HTP_TYPE_Q8_1x4x2,
HTP_TYPE_Q8_0x4x2,
HTP_TYPE_MXFP4x4x2,

HTP_TYPE_INVALID
};

// Constats for internal types
#define QK_Q4_0x4x2 256 // 4x Q4_0 blocks packed with next 4x Q4_0 blocks (size in bytes 128)
#define QK_Q4_0x4x2 256
#define QK_Q4_1x4x2 256 // 4x Q4_0 blocks packed with next 4x Q4_0 blocks (size in bytes 128)
#define QK_Q8_0x4x2 256 // 4x Q8_0 blocks concat with next 4x Q8_0 blocks
#define QK_MXFP4x4x2 256 // 4x MXFP4 blocks concat with next 4x MXFP4 blocks

Expand Down
Loading
Loading