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.
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
NSG support MetricType IP #554
Merged
Merged
Changes from all commits
Commits
File filter
Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
247 changes: 247 additions & 0 deletions
247
core/src/index/knowhere/knowhere/index/vector_index/nsg/Distance.cpp
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,247 @@ | ||
// Licensed to the Apache Software Foundation (ASF) under one | ||
// or more contributor license agreements. See the NOTICE file | ||
// distributed with this work for additional information | ||
// regarding copyright ownership. The ASF licenses this file | ||
// to you under the Apache License, Version 2.0 (the | ||
// "License"); you may not use this file except in compliance | ||
// with the License. You may obtain a copy of the License at | ||
// | ||
// http://www.apache.org/licenses/LICENSE-2.0 | ||
// | ||
// Unless required by applicable law or agreed to in writing, | ||
// software distributed under the License is distributed on an | ||
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY | ||
// KIND, either express or implied. See the License for the | ||
// specific language governing permissions and limitations | ||
// under the License. | ||
|
||
#pragma once | ||
|
||
#include <immintrin.h> | ||
|
||
#include "knowhere/index/vector_index/nsg/Distance.h" | ||
|
||
namespace knowhere { | ||
namespace algo { | ||
|
||
float | ||
DistanceL2::Compare(const float* a, const float* b, unsigned size) const { | ||
float result = 0; | ||
|
||
#ifdef __GNUC__ | ||
#ifdef __AVX__ | ||
|
||
#define AVX_L2SQR(addr1, addr2, dest, tmp1, tmp2) \ | ||
tmp1 = _mm256_loadu_ps(addr1); \ | ||
tmp2 = _mm256_loadu_ps(addr2); \ | ||
tmp1 = _mm256_sub_ps(tmp1, tmp2); \ | ||
tmp1 = _mm256_mul_ps(tmp1, tmp1); \ | ||
dest = _mm256_add_ps(dest, tmp1); | ||
|
||
__m256 sum; | ||
__m256 l0, l1; | ||
__m256 r0, r1; | ||
unsigned D = (size + 7) & ~7U; | ||
unsigned DR = D % 16; | ||
unsigned DD = D - DR; | ||
const float* l = a; | ||
const float* r = b; | ||
const float* e_l = l + DD; | ||
const float* e_r = r + DD; | ||
float unpack[8] __attribute__((aligned(32))) = {0, 0, 0, 0, 0, 0, 0, 0}; | ||
|
||
sum = _mm256_loadu_ps(unpack); | ||
if (DR) { | ||
AVX_L2SQR(e_l, e_r, sum, l0, r0); | ||
} | ||
|
||
for (unsigned i = 0; i < DD; i += 16, l += 16, r += 16) { | ||
AVX_L2SQR(l, r, sum, l0, r0); | ||
AVX_L2SQR(l + 8, r + 8, sum, l1, r1); | ||
} | ||
_mm256_storeu_ps(unpack, sum); | ||
result = unpack[0] + unpack[1] + unpack[2] + unpack[3] + unpack[4] + unpack[5] + unpack[6] + unpack[7]; | ||
|
||
#else | ||
#ifdef __SSE2__ | ||
#define SSE_L2SQR(addr1, addr2, dest, tmp1, tmp2) \ | ||
tmp1 = _mm_load_ps(addr1); \ | ||
tmp2 = _mm_load_ps(addr2); \ | ||
tmp1 = _mm_sub_ps(tmp1, tmp2); \ | ||
tmp1 = _mm_mul_ps(tmp1, tmp1); \ | ||
dest = _mm_add_ps(dest, tmp1); | ||
|
||
__m128 sum; | ||
__m128 l0, l1, l2, l3; | ||
__m128 r0, r1, r2, r3; | ||
unsigned D = (size + 3) & ~3U; | ||
unsigned DR = D % 16; | ||
unsigned DD = D - DR; | ||
const float* l = a; | ||
const float* r = b; | ||
const float* e_l = l + DD; | ||
const float* e_r = r + DD; | ||
float unpack[4] __attribute__((aligned(16))) = {0, 0, 0, 0}; | ||
|
||
sum = _mm_load_ps(unpack); | ||
switch (DR) { | ||
case 12: | ||
SSE_L2SQR(e_l + 8, e_r + 8, sum, l2, r2); | ||
case 8: | ||
SSE_L2SQR(e_l + 4, e_r + 4, sum, l1, r1); | ||
case 4: | ||
SSE_L2SQR(e_l, e_r, sum, l0, r0); | ||
default: | ||
break; | ||
} | ||
for (unsigned i = 0; i < DD; i += 16, l += 16, r += 16) { | ||
SSE_L2SQR(l, r, sum, l0, r0); | ||
SSE_L2SQR(l + 4, r + 4, sum, l1, r1); | ||
SSE_L2SQR(l + 8, r + 8, sum, l2, r2); | ||
SSE_L2SQR(l + 12, r + 12, sum, l3, r3); | ||
} | ||
_mm_storeu_ps(unpack, sum); | ||
result += unpack[0] + unpack[1] + unpack[2] + unpack[3]; | ||
|
||
// nomal distance | ||
#else | ||
|
||
float diff0, diff1, diff2, diff3; | ||
const float* last = a + size; | ||
const float* unroll_group = last - 3; | ||
|
||
/* Process 4 items with each loop for efficiency. */ | ||
while (a < unroll_group) { | ||
diff0 = a[0] - b[0]; | ||
diff1 = a[1] - b[1]; | ||
diff2 = a[2] - b[2]; | ||
diff3 = a[3] - b[3]; | ||
result += diff0 * diff0 + diff1 * diff1 + diff2 * diff2 + diff3 * diff3; | ||
a += 4; | ||
b += 4; | ||
} | ||
/* Process last 0-3 pixels. Not needed for standard vector lengths. */ | ||
while (a < last) { | ||
diff0 = *a++ - *b++; | ||
result += diff0 * diff0; | ||
} | ||
#endif | ||
#endif | ||
#endif | ||
|
||
return result; | ||
} | ||
|
||
float | ||
DistanceIP::Compare(const float* a, const float* b, unsigned size) const { | ||
float result = 0; | ||
|
||
#ifdef __GNUC__ | ||
#ifdef __AVX__ | ||
#define AVX_DOT(addr1, addr2, dest, tmp1, tmp2) \ | ||
tmp1 = _mm256_loadu_ps(addr1); \ | ||
tmp2 = _mm256_loadu_ps(addr2); \ | ||
tmp1 = _mm256_mul_ps(tmp1, tmp2); \ | ||
dest = _mm256_add_ps(dest, tmp1); | ||
|
||
__m256 sum; | ||
__m256 l0, l1; | ||
__m256 r0, r1; | ||
unsigned D = (size + 7) & ~7U; | ||
unsigned DR = D % 16; | ||
unsigned DD = D - DR; | ||
const float* l = a; | ||
const float* r = b; | ||
const float* e_l = l + DD; | ||
const float* e_r = r + DD; | ||
float unpack[8] __attribute__((aligned(32))) = {0, 0, 0, 0, 0, 0, 0, 0}; | ||
|
||
sum = _mm256_loadu_ps(unpack); | ||
if (DR) { | ||
AVX_DOT(e_l, e_r, sum, l0, r0); | ||
} | ||
|
||
for (unsigned i = 0; i < DD; i += 16, l += 16, r += 16) { | ||
AVX_DOT(l, r, sum, l0, r0); | ||
AVX_DOT(l + 8, r + 8, sum, l1, r1); | ||
} | ||
_mm256_storeu_ps(unpack, sum); | ||
result = unpack[0] + unpack[1] + unpack[2] + unpack[3] + unpack[4] + unpack[5] + unpack[6] + unpack[7]; | ||
|
||
#else | ||
#ifdef __SSE2__ | ||
#define SSE_DOT(addr1, addr2, dest, tmp1, tmp2) \ | ||
tmp1 = _mm128_loadu_ps(addr1); \ | ||
tmp2 = _mm128_loadu_ps(addr2); \ | ||
tmp1 = _mm128_mul_ps(tmp1, tmp2); \ | ||
dest = _mm128_add_ps(dest, tmp1); | ||
__m128 sum; | ||
__m128 l0, l1, l2, l3; | ||
__m128 r0, r1, r2, r3; | ||
unsigned D = (size + 3) & ~3U; | ||
unsigned DR = D % 16; | ||
unsigned DD = D - DR; | ||
const float* l = a; | ||
const float* r = b; | ||
const float* e_l = l + DD; | ||
const float* e_r = r + DD; | ||
float unpack[4] __attribute__((aligned(16))) = {0, 0, 0, 0}; | ||
|
||
sum = _mm_load_ps(unpack); | ||
switch (DR) { | ||
case 12: | ||
SSE_DOT(e_l + 8, e_r + 8, sum, l2, r2); | ||
case 8: | ||
SSE_DOT(e_l + 4, e_r + 4, sum, l1, r1); | ||
case 4: | ||
SSE_DOT(e_l, e_r, sum, l0, r0); | ||
default: | ||
break; | ||
} | ||
for (unsigned i = 0; i < DD; i += 16, l += 16, r += 16) { | ||
SSE_DOT(l, r, sum, l0, r0); | ||
SSE_DOT(l + 4, r + 4, sum, l1, r1); | ||
SSE_DOT(l + 8, r + 8, sum, l2, r2); | ||
SSE_DOT(l + 12, r + 12, sum, l3, r3); | ||
} | ||
_mm_storeu_ps(unpack, sum); | ||
result += unpack[0] + unpack[1] + unpack[2] + unpack[3]; | ||
#else | ||
|
||
float dot0, dot1, dot2, dot3; | ||
const float* last = a + size; | ||
const float* unroll_group = last - 3; | ||
|
||
/* Process 4 items with each loop for efficiency. */ | ||
while (a < unroll_group) { | ||
dot0 = a[0] * b[0]; | ||
dot1 = a[1] * b[1]; | ||
dot2 = a[2] * b[2]; | ||
dot3 = a[3] * b[3]; | ||
result += dot0 + dot1 + dot2 + dot3; | ||
a += 4; | ||
b += 4; | ||
} | ||
/* Process last 0-3 pixels. Not needed for standard vector lengths. */ | ||
while (a < last) { | ||
result += *a++ * *b++; | ||
} | ||
#endif | ||
#endif | ||
#endif | ||
return result; | ||
} | ||
|
||
//#include <faiss/utils/distances.h> | ||
// float | ||
// DistanceL2::Compare(const float* a, const float* b, unsigned size) const { | ||
// return faiss::fvec_L2sqr(a,b,size); | ||
//} | ||
// | ||
// float | ||
// DistanceIP::Compare(const float* a, const float* b, unsigned size) const { | ||
// return faiss::fvec_inner_product(a,b,size); | ||
//} | ||
|
||
} // namespace algo | ||
} // namespace knowhere |
39 changes: 39 additions & 0 deletions
39
core/src/index/knowhere/knowhere/index/vector_index/nsg/Distance.h
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,39 @@ | ||
// Licensed to the Apache Software Foundation (ASF) under one | ||
// or more contributor license agreements. See the NOTICE file | ||
// distributed with this work for additional information | ||
// regarding copyright ownership. The ASF licenses this file | ||
// to you under the Apache License, Version 2.0 (the | ||
// "License"); you may not use this file except in compliance | ||
// with the License. You may obtain a copy of the License at | ||
// | ||
// http://www.apache.org/licenses/LICENSE-2.0 | ||
// | ||
// Unless required by applicable law or agreed to in writing, | ||
// software distributed under the License is distributed on an | ||
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY | ||
// KIND, either express or implied. See the License for the | ||
// specific language governing permissions and limitations | ||
// under the License. | ||
|
||
#pragma once | ||
|
||
namespace knowhere { | ||
namespace algo { | ||
|
||
struct Distance { | ||
There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. why struct not class ? There was a problem hiding this comment. Choose a reason for hiding this commentThe reason will be displayed to describe this comment to others. Learn more. should be struct |
||
virtual float | ||
Compare(const float* a, const float* b, unsigned size) const = 0; | ||
}; | ||
|
||
struct DistanceL2 : public Distance { | ||
float | ||
Compare(const float* a, const float* b, unsigned size) const override; | ||
}; | ||
|
||
struct DistanceIP : public Distance { | ||
float | ||
Compare(const float* a, const float* b, unsigned size) const override; | ||
}; | ||
|
||
} // namespace algo | ||
} // namespace knowhere |
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
why unsigned ?
can use int32_t or int64_t ?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No