Adding native code related to (#138204)

libs/simdvec/native/publish_vec_binaries.sh

@@ -20,7 +20,7 @@ if [ -z "$ARTIFACTORY_API_KEY" ]; then
   exit 1;
 fi
 
-VERSION="1.0.13"
+VERSION="1.0.14"
 ARTIFACTORY_REPOSITORY="${ARTIFACTORY_REPOSITORY:-https://artifactory.elastic.dev/artifactory/elasticsearch-native/}"
 TEMP=$(mktemp -d)
 

libs/simdvec/native/src/vec/c/aarch64/vec.c

@@ -95,6 +95,31 @@ EXPORT int32_t dot7u(int8_t* a, int8_t* b, size_t dims) {
     return res;
 }
 
+EXPORT void dot7u_bulk(int8_t* a, int8_t* b, size_t dims, size_t count, float_t* results) {
+    int32_t res = 0;
+    if (dims > DOT7U_STRIDE_BYTES_LEN) {
+        int limit = dims & ~(DOT7U_STRIDE_BYTES_LEN - 1);
+        for (size_t c = 0; c < count; c++) {
+            int i = limit;
+            res = dot7u_inner(a, b, i);
+            for (; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    } else {
+        for (size_t c = 0; c < count; c++) {
+            res = 0;
+            for (size_t i = 0; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    }
+}
+
 static inline int32_t sqr7u_inner(int8_t *a, int8_t *b, size_t dims) {
     int32x4_t acc1 = vdupq_n_s32(0);
     int32x4_t acc2 = vdupq_n_s32(0);

libs/simdvec/native/src/vec/c/amd64/vec.c

@@ -153,6 +153,31 @@ EXPORT int32_t dot7u(int8_t* a, int8_t* b, size_t dims) {
     return res;
 }
 
+EXPORT void dot7u_bulk(int8_t* a, int8_t* b, size_t dims, size_t count, float_t* results) {
+    int32_t res = 0;
+    if (dims > STRIDE_BYTES_LEN) {
+        int limit = dims & ~(STRIDE_BYTES_LEN - 1);
+        for (size_t c = 0; c < count; c++) {
+            int i = limit;
+            res = dot7u_inner(a, b, i);
+            for (; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    } else {
+        for (size_t c = 0; c < count; c++) {
+            res = 0;
+            for (size_t i = 0; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    }
+}
+
 static inline int32_t sqr7u_inner(int8_t *a, int8_t *b, size_t dims) {
     // Init accumulator(s) with 0
     __m256i acc1 = _mm256_setzero_si256();

libs/simdvec/native/src/vec/c/amd64/vec_2.cpp

@@ -114,6 +114,32 @@ EXPORT int32_t dot7u_2(int8_t* a, int8_t* b, size_t dims) {
     return res;
 }
 
+extern "C"
+EXPORT void dot7u_bulk_2(int8_t* a, int8_t* b, size_t dims, size_t count, float_t* results) {
+    int32_t res = 0;
+    if (dims > STRIDE_BYTES_LEN) {
+        int limit = dims & ~(STRIDE_BYTES_LEN - 1);
+        for (size_t c = 0; c < count; c++) {
+            int i = limit;
+            res = dot7u_inner_avx512(a, b, i);
+            for (; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    } else {
+        for (size_t c = 0; c < count; c++) {
+            res = 0;
+            for (size_t i = 0; i < dims; i++) {
+                res += a[i] * b[i];
+            }
+            results[c] = (float_t)res;
+            a += dims;
+        }
+    }
+}
+
 template<int offsetRegs>
 inline __m512i sqr8(__m512i acc, const int8_t* p1, const int8_t* p2) {
     constexpr int lanes = offsetRegs * STRIDE_BYTES_LEN;

libs/simdvec/native/src/vec/headers/vec.h

@@ -19,6 +19,8 @@ EXPORT int vec_caps();
 
 EXPORT int32_t dot7u(int8_t* a, int8_t* b, size_t dims);
 
+EXPORT void dot7u_bulk(int8_t* a, int8_t* b, size_t dims, size_t count, float_t* results);
+
 EXPORT int32_t sqr7u(int8_t *a, int8_t *b, size_t length);
 
 EXPORT float cosf32(const float *a, const float *b, size_t elementCount);