Skip to content
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

Jump to bottom

Packing use inlined for loops to copy and set bias and weights #7379

Open
wants to merge 1 commit into
base: master
Choose a base branch
from
Open

Packing use inlined for loops to copy and set bias and weights #7379

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
89 changes: 50 additions & 39 deletions src/packing.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,35 +41,46 @@ struct unaligned_int32_t {
XNN_INLINE operator int32_t() const { return value; } // NOLINT: emulating __attribute__((__aligned__(x))) int32_t.
};

template <typename Src, typename Dst>
void copy_n(const Src* src, size_t n, Dst* dst) {
for (size_t i = 0; i < n; ++i) {
dst[i] = src[i];
}
}

template <typename Src, typename Dst>
int32_t copy_n_and_sum(const Src* src, size_t n, Dst* dst) {
int32_t sum = 0;
for (size_t i = 0; i < n; ++i) {
const auto v = src[i];
sum += (int32_t) v;
dst[i] = v;
}
return sum;
}

template <typename Src, typename Dst>
void copy_bias(const Src* b, size_t b_offset, size_t n, Dst* packed_b) {
if (b) {
std::copy_n(b + b_offset, n, packed_b);
copy_n(b + b_offset, n, packed_b);
} else {
std::fill_n(packed_b, n, 0);
for (size_t i = 0; i < n; ++i) {
packed_b[i] = 0;
}
}
}

template <typename Src, typename Dst>
void copy_bias(const Src* b, size_t b_offset, size_t n, Dst* packed_b, Src zero_point) {
if (b) {
for (size_t i = 0; i < n; ++i) {
*packed_b++ = zero_point + b[b_offset + i];
packed_b[i] = zero_point + b[b_offset + i];
}
} else {
std::fill_n(packed_b, n, zero_point);
}
}

template <typename Src, typename Dst>
int32_t copy_n_and_sum(const Src* src, size_t n, Dst* dst) {
int32_t sum = 0;
for (size_t i = 0; i < n; ++i) {
const auto v = *src++;
sum += (int32_t) v;
*dst++ = v;
for (size_t i = 0; i < n; ++i) {
packed_b[i] = zero_point;
}
}
return sum;
}

extern "C" {
Expand Down Expand Up @@ -105,7 +116,7 @@ void xnn_pack_f32_gemm_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -151,7 +162,7 @@ void xnn_pack_f16_gemm_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -197,7 +208,7 @@ void xnn_pack_f32_to_f16_gemm_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -934,7 +945,7 @@ void xnn_pack_f32_qs8w_gemm_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, (int8_t*) packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, (int8_t*) packed_weights);
}
packed_weights = (int8_t*) packed_weights + kr;
}
Expand Down Expand Up @@ -985,7 +996,7 @@ void xnn_pack_f32_qc4w_gemm_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&((const uint8_t*) k)[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, (uint8_t*) packed_weights);
copy_n(&((const uint8_t*) k)[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, (uint8_t*) packed_weights);
}
packed_weights = (uint8_t*) packed_weights + kr;
}
Expand Down Expand Up @@ -1022,7 +1033,7 @@ void xnn_pack_f32_gemm_gio_w(size_t g, size_t nc, size_t kc, size_t nr,
for (size_t kr_block_start = 0; kr_block_start < kc; kr_block_start++) {
const size_t kc_idx = round_down_po2(kr_block_start, skr);
if (kc_idx < kc) {
std::copy_n(&k[kc_idx * k_stride + nr_block_start], nr_block_size,
copy_n(&k[kc_idx * k_stride + nr_block_start], nr_block_size,
packed_weights);
}
packed_weights += nr;
Expand Down Expand Up @@ -1080,7 +1091,7 @@ void xnn_pack_f16_gemm_gio_w(size_t g, size_t nc, size_t kc, size_t nr,
for (size_t kr_block_start = 0; kr_block_start < kc; kr_block_start++) {
const size_t kc_idx = round_down_po2(kr_block_start, skr);
if (kc_idx < kc) {
std::copy_n(&k[kc_idx * k_stride + nr_block_start], nr_block_size,
copy_n(&k[kc_idx * k_stride + nr_block_start], nr_block_size,
packed_weights);
}
packed_weights += nr;
Expand Down Expand Up @@ -1773,7 +1784,7 @@ void xnn_pack_f32_conv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand All @@ -1783,7 +1794,7 @@ void xnn_pack_f32_conv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -1824,7 +1835,7 @@ void xnn_pack_f16_conv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand All @@ -1834,7 +1845,7 @@ void xnn_pack_f16_conv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -1875,7 +1886,7 @@ void xnn_pack_f32_to_f16_conv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand All @@ -1885,7 +1896,7 @@ void xnn_pack_f32_to_f16_conv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[((nr_block_start + nr_block_offset) * ks + ki) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -2081,7 +2092,7 @@ void xnn_pack_f32_conv_kgo_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand Down Expand Up @@ -2125,7 +2136,7 @@ void xnn_pack_f16_conv_kgo_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand Down Expand Up @@ -2169,7 +2180,7 @@ void xnn_pack_f32_to_f16_conv_kgo_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;

Expand Down Expand Up @@ -2354,7 +2365,7 @@ void xnn_pack_f32_deconv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;
for (size_t ky = oy; ky < kh; ky += sh) {
Expand All @@ -2364,7 +2375,7 @@ void xnn_pack_f32_deconv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -2417,7 +2428,7 @@ void xnn_pack_f16_deconv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;
for (size_t ky = oy; ky < kh; ky += sh) {
Expand All @@ -2427,7 +2438,7 @@ void xnn_pack_f16_deconv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -2480,7 +2491,7 @@ void xnn_pack_f32_to_f16_deconv_goki_w(
for (size_t nr_block_start = 0; nr_block_start < nc; nr_block_start += nr) {
const size_t nr_block_size = min(nc - nr_block_start, nr);
if XNN_LIKELY(b != nullptr) {
std::copy_n(&b[nr_block_start], nr_block_size, packed_weights);
copy_n(&b[nr_block_start], nr_block_size, packed_weights);
}
packed_weights += nr;
for (size_t ky = oy; ky < kh; ky += sh) {
Expand All @@ -2490,7 +2501,7 @@ void xnn_pack_f32_to_f16_deconv_goki_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(((nr_block_start + nr_block_offset) * kh + ky) * kw + kx) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
packed_weights += kr;
}
Expand Down Expand Up @@ -4585,7 +4596,7 @@ void xnn_pack_f32_gemminc_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
}
packed_weights += kr;
Expand Down Expand Up @@ -4624,7 +4635,7 @@ void xnn_pack_f16_gemminc_goi_w(
const size_t kc_begin = round_down_po2(kr_block_start, skr) + ((kr_block_start + nr_block_offset * kr) & (skr - 1));
const size_t kc_end = std::min(kc, kc_begin + kr);
if (kc_begin < kc_end) {
std::copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
copy_n(&k[(nr_block_start + nr_block_offset) * kc + kc_begin], kc_end - kc_begin, packed_weights);
}
}
packed_weights += kr;
Expand Down