Skip to content

Commit

Permalink
ggml : optimize Q4_0 into Q4_0_X_Y repack (#10324)
Browse files Browse the repository at this point in the history
  • Loading branch information
@eddnjjn
eddnjjn authored Nov 16, 2024
1 parent 89e4caa commit 1e58ee1
Show file tree
Hide file tree
Showing 2 changed files with 86 additions and 28 deletions.
57 changes: 43 additions & 14 deletions ggml/src/ggml-aarch64.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,19 +8,42 @@

#define UNUSED GGML_UNUSED

static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_interleave, unsigned int xor_mask) {
static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_interleave) {
block_q4_0x4 out;

for (int i = 0; i < 4; i++) {
out.d[i] = in[i].d;
}

for (int i = 0; i < QK4_0 * 2; i++) {
int src_offset = (i / (4 * blck_size_interleave)) * blck_size_interleave;
int src_id = (i % (4 * blck_size_interleave)) / blck_size_interleave;
src_offset += (i % blck_size_interleave);
const int end = QK4_0 * 2 / blck_size_interleave;

out.qs[i] = in[src_id].qs[src_offset] ^ xor_mask;
if (blck_size_interleave == 8) {
const uint64_t xor_mask = 0x8888888888888888ULL;
for (int i = 0; i < end; ++i) {
int src_id = i % 4;
int src_offset = (i / 4) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

uint64_t elems;
// Using memcpy to avoid unaligned memory accesses
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint64_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint64_t));
}
} else if (blck_size_interleave == 4) {
const uint32_t xor_mask = 0x88888888;
for (int i = 0; i < end; ++i) {
int src_id = i % 4;
int src_offset = (i / 4) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

uint32_t elems;
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint32_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint32_t));
}
} else {
GGML_ASSERT(false);
}

return out;
Expand All @@ -30,19 +53,25 @@ static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_in
// returns an interleaved block_q4_0x8
// in the interleaved block_q4_0x8, place deltas for 8 block_q4_0 blocks
// first, then interleave quants from 8 block_q4_0s in blocks of blck_size_interleave
static block_q4_0x8 make_block_q4_0x8(block_q4_0 * in, unsigned int blck_size_interleave, unsigned int xor_mask) {
static block_q4_0x8 make_block_q4_0x8(block_q4_0 * in, unsigned int blck_size_interleave) {
block_q4_0x8 out;

for (int i = 0; i < 8; i++) {
out.d[i] = in[i].d;
}

for (int i = 0; i < QK4_0 * 4; i++) {
int src_offset = (i / (8 * blck_size_interleave)) * blck_size_interleave;
int src_id = (i % (8 * blck_size_interleave)) / blck_size_interleave;
src_offset += (i % blck_size_interleave);
const int end = QK4_0 * 4 / blck_size_interleave;
const uint64_t xor_mask = 0x8888888888888888ULL;

for (int i = 0; i < end; ++i) {
int src_id = i % 8;
int src_offset = (i / 8) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

out.qs[i] = in[src_id].qs[src_offset] ^ xor_mask;
uint64_t elems;
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint64_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint64_t));
}

return out;
Expand Down Expand Up @@ -71,11 +100,11 @@ static size_t quantize_q4_0_nr_bl(const float * restrict src, void * restrict ds
}

if (nrows_interleaved == 8) {
*(block_q4_0x8 *) out_ptr = make_block_q4_0x8(dst_tmp, blck_size_interleave, 0x88);
*(block_q4_0x8 *) out_ptr = make_block_q4_0x8(dst_tmp, blck_size_interleave);
out_ptr = (block_q4_0x8 *) out_ptr + 1;
}
else if (nrows_interleaved == 4) {
*(block_q4_0x4 *) out_ptr = make_block_q4_0x4(dst_tmp, blck_size_interleave, 0x88);
*(block_q4_0x4 *) out_ptr = make_block_q4_0x4(dst_tmp, blck_size_interleave);
out_ptr = (block_q4_0x4 *) out_ptr + 1;
}
}
Expand Down
57 changes: 43 additions & 14 deletions ggml/src/ggml-cpu/ggml-cpu-aarch64.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -3387,19 +3387,42 @@ void ggml_gemm_q4_0_8x8_q8_0(int n, float * restrict s, size_t bs, const void *
}

// FIXME: this code is duplicated from ggml-aarch64.c
static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_interleave, unsigned int xor_mask) {
static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_interleave) {
block_q4_0x4 out;

for (int i = 0; i < 4; i++) {
out.d[i] = in[i].d;
}

for (int i = 0; i < QK4_0 * 2; i++) {
int src_offset = (i / (4 * blck_size_interleave)) * blck_size_interleave;
int src_id = (i % (4 * blck_size_interleave)) / blck_size_interleave;
src_offset += (i % blck_size_interleave);
const int end = QK4_0 * 2 / blck_size_interleave;

out.qs[i] = in[src_id].qs[src_offset] ^ xor_mask;
if (blck_size_interleave == 8) {
const uint64_t xor_mask = 0x8888888888888888ULL;
for (int i = 0; i < end; ++i) {
int src_id = i % 4;
int src_offset = (i / 4) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

uint64_t elems;
// Using memcpy to avoid unaligned memory accesses
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint64_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint64_t));
}
} else if (blck_size_interleave == 4) {
const uint32_t xor_mask = 0x88888888;
for (int i = 0; i < end; ++i) {
int src_id = i % 4;
int src_offset = (i / 4) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

uint32_t elems;
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint32_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint32_t));
}
} else {
GGML_ASSERT(false);
}

return out;
Expand All @@ -3409,19 +3432,25 @@ static block_q4_0x4 make_block_q4_0x4(block_q4_0 * in, unsigned int blck_size_in
// returns an interleaved block_q4_0x8
// in the interleaved block_q4_0x8, place deltas for 8 block_q4_0 blocks
// first, then interleave quants from 8 block_q4_0s in blocks of blck_size_interleave
static block_q4_0x8 make_block_q4_0x8(block_q4_0 * in, unsigned int blck_size_interleave, unsigned int xor_mask) {
static block_q4_0x8 make_block_q4_0x8(block_q4_0 * in, unsigned int blck_size_interleave) {
block_q4_0x8 out;

for (int i = 0; i < 8; i++) {
out.d[i] = in[i].d;
}

for (int i = 0; i < QK4_0 * 4; i++) {
int src_offset = (i / (8 * blck_size_interleave)) * blck_size_interleave;
int src_id = (i % (8 * blck_size_interleave)) / blck_size_interleave;
src_offset += (i % blck_size_interleave);
const int end = QK4_0 * 4 / blck_size_interleave;
const uint64_t xor_mask = 0x8888888888888888ULL;

for (int i = 0; i < end; ++i) {
int src_id = i % 8;
int src_offset = (i / 8) * blck_size_interleave;
int dst_offset = i * blck_size_interleave;

out.qs[i] = in[src_id].qs[src_offset] ^ xor_mask;
uint64_t elems;
memcpy(&elems, &in[src_id].qs[src_offset], sizeof(uint64_t));
elems ^= xor_mask;
memcpy(&out.qs[dst_offset], &elems, sizeof(uint64_t));
}

return out;
Expand Down Expand Up @@ -3449,7 +3478,7 @@ static int repack_q4_0_to_q4_0_4_bl(struct ggml_tensor * t, int interleave_block
for (int i = 0; i < nrows_interleaved; i++) {
dst_tmp[i] = src[x + i * nblocks];
}
*dst++ = make_block_q4_0x4(dst_tmp, interleave_block, 0x88);
*dst++ = make_block_q4_0x4(dst_tmp, interleave_block);
}
src += nrows_interleaved * nblocks;
}
Expand Down Expand Up @@ -3480,7 +3509,7 @@ static int repack_q4_0_to_q4_0_8_bl(struct ggml_tensor *t, int interleave_block,
for (int i = 0; i < nrows_interleaved; i++ ) {
dst_tmp[i] = src[x + i * nblocks];
}
*dst++ = make_block_q4_0x8(dst_tmp, interleave_block, 0x88);
*dst++ = make_block_q4_0x8(dst_tmp, interleave_block);
}
src += nrows_interleaved * nblocks;
}
Expand Down

0 comments on commit 1e58ee1

Please sign in to comment.