fix: use static array (#222)

* fix: use static array

* fix: clang-format

* fix: clang-format

* fix: clang-format

src/ailego/math_batch/inner_product_distance_batch_impl_fp16_avx2.cc

@@ -28,8 +28,7 @@ compute_one_to_many_inner_product_avx2_fp16(
     const ailego::Float16 *query, const ailego::Float16 **ptrs,
     std::array<const ailego::Float16 *, dp_batch> &prefetch_ptrs,
     size_t dimensionality, float *results) {
-  std::array<__m256, dp_batch> accs;
-
+  __m256 accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm256_setzero_ps();
   }
@@ -42,8 +41,8 @@ compute_one_to_many_inner_product_avx2_fp16(
     __m256 q1 = _mm256_cvtph_ps(_mm256_castsi256_si128(q));
     __m256 q2 = _mm256_cvtph_ps(_mm256_extractf128_si256(q, 1));
 
-    std::array<__m256, dp_batch> data_regs_1;
-    std::array<__m256, dp_batch> data_regs_2;
+    __m256 data_regs_1[dp_batch];
+    __m256 data_regs_2[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       __m256i m =
           _mm256_loadu_si256(reinterpret_cast<const __m256i *>(ptrs[i] + dim));
@@ -68,7 +67,7 @@ compute_one_to_many_inner_product_avx2_fp16(
     __m256 q = _mm256_cvtph_ps(
         _mm_loadu_si128(reinterpret_cast<const __m128i *>(query + dim)));
 
-    std::array<__m256, dp_batch> data_regs;
+    __m256 data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm256_cvtph_ps(
           _mm_loadu_si128(reinterpret_cast<const __m128i *>(ptrs[i] + dim)));

src/ailego/math_batch/inner_product_distance_batch_impl_fp16_avx512.cc

@@ -27,8 +27,7 @@ compute_one_to_many_inner_product_avx512fp16_fp16(
     const ailego::Float16 *query, const ailego::Float16 **ptrs,
     std::array<const ailego::Float16 *, dp_batch> &prefetch_ptrs,
     size_t dimensionality, float *results) {
-  std::array<__m512h, dp_batch> accs;
-
+  __m512h accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm512_setzero_ph();
   }
@@ -37,7 +36,7 @@ compute_one_to_many_inner_product_avx512fp16_fp16(
   for (; dim + 32 <= dimensionality; dim += 32) {
     __m512h q = _mm512_loadu_ph(query + dim);
 
-    std::array<__m512h, dp_batch> data_regs;
+    __m512h data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm512_loadu_ph(ptrs[i] + dim);
     }
@@ -83,8 +82,7 @@ compute_one_to_many_inner_product_avx512f_fp16(
     const ailego::Float16 *query, const ailego::Float16 **ptrs,
     std::array<const ailego::Float16 *, dp_batch> &prefetch_ptrs,
     size_t dimensionality, float *results) {
-  std::array<__m512, dp_batch> accs;
-
+  __m512 accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm512_setzero_ps();
   }
@@ -97,8 +95,8 @@ compute_one_to_many_inner_product_avx512f_fp16(
     __m512 q1 = _mm512_cvtph_ps(_mm512_castsi512_si256(q));
     __m512 q2 = _mm512_cvtph_ps(_mm512_extracti64x4_epi64(q, 1));
 
-    std::array<__m512, dp_batch> data_regs_1;
-    std::array<__m512, dp_batch> data_regs_2;
+    __m512 data_regs_1[dp_batch];
+    __m512 data_regs_2[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       __m512i m =
           _mm512_loadu_si512(reinterpret_cast<const __m512i *>(ptrs[i] + dim));
@@ -123,7 +121,7 @@ compute_one_to_many_inner_product_avx512f_fp16(
     __m512 q = _mm512_cvtph_ps(
         _mm256_loadu_si256(reinterpret_cast<const __m256i *>(query + dim)));
 
-    std::array<__m512, dp_batch> data_regs;
+    __m512 data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm512_cvtph_ps(
           _mm256_loadu_si256(reinterpret_cast<const __m256i *>(ptrs[i] + dim)));
@@ -133,7 +131,7 @@ compute_one_to_many_inner_product_avx512f_fp16(
     dim += 16;
   }
 
-  std::array<__m256, dp_batch> acc_new;
+  __m256 acc_new[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     acc_new[i] = _mm256_add_ps(
         _mm512_castps512_ps256(accs[i]),

src/ailego/math_batch/inner_product_distance_batch_impl_fp32_avx2.cc

@@ -39,14 +39,15 @@ compute_one_to_many_inner_product_avx2_fp32(
     const ValueType *query, const ValueType **ptrs,
     std::array<const ValueType *, dp_batch> &prefetch_ptrs,
     size_t dimensionality, float *results) {
-  std::array<__m256, dp_batch> accs;
+  __m256 accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm256_setzero_ps();
   }
   size_t dim = 0;
   for (; dim + 8 <= dimensionality; dim += 8) {
     __m256 q = _mm256_loadu_ps(query + dim);
-    std::array<__m256, dp_batch> data_regs;
+
+    __m256 data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm256_loadu_ps(ptrs[i] + dim);
     }
@@ -59,13 +60,15 @@ compute_one_to_many_inner_product_avx2_fp32(
       accs[i] = _mm256_fnmadd_ps(q, data_regs[i], accs[i]);
     }
   }
-  std::array<__m128, dp_batch> sum128_regs;
+
+  __m128 sum128_regs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     sum128_regs[i] = sum_top_bottom_avx(accs[i]);
   }
   if (dim + 4 <= dimensionality) {
     __m128 q = _mm_loadu_ps(query + dim);
-    std::array<__m128, dp_batch> data_regs;
+
+    __m128 data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm_loadu_ps(ptrs[i] + dim);
     }
@@ -81,7 +84,8 @@ compute_one_to_many_inner_product_avx2_fp32(
   }
   if (dim + 2 <= dimensionality) {
     __m128 q = _mm_setzero_ps();
-    std::array<__m128, dp_batch> data_regs;
+
+    __m128 data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm_setzero_ps();
     }
@@ -95,7 +99,8 @@ compute_one_to_many_inner_product_avx2_fp32(
     }
     dim += 2;
   }
-  std::array<float, dp_batch> res;
+
+  float res[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     res[i] = sum4(sum128_regs[i]);
   }

src/ailego/math_batch/inner_product_distance_batch_impl_int8_avx2.cc

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#include <vector>
+#include <array>
 #include <ailego/utility/math_helper.h>
 #include <zvec/ailego/internal/platform.h>
 #include <zvec/ailego/utility/type_helper.h>
@@ -27,14 +27,15 @@ compute_one_to_many_inner_product_avx2_int8(
     const int8_t *query, const int8_t **ptrs,
     std::array<const int8_t *, dp_batch> &prefetch_ptrs, size_t dimensionality,
     float *results) {
-  std::vector<__m256i> accs(dp_batch);
+  __m256i accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm256_setzero_si256();
   }
   size_t dim = 0;
   for (; dim + 32 <= dimensionality; dim += 32) {
     __m256i q = _mm256_loadu_si256((const __m256i *)(query + dim));
-    std::vector<__m256i> data_regs(dp_batch);
+
+    __m256i data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] = _mm256_loadu_si256((const __m256i *)(ptrs[i] + dim));
     }
@@ -63,7 +64,8 @@ compute_one_to_many_inner_product_avx2_int8(
           _mm256_add_epi32(accs[i], _mm256_add_epi32(prod_lo[i], prod_hi[i]));
     }
   }
-  std::array<int, dp_batch> temp_results;
+
+  int temp_results[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     __m128i lo = _mm256_castsi256_si128(accs[i]);
     __m128i hi = _mm256_extracti128_si256(accs[i], 1);

src/ailego/math_batch/inner_product_distance_batch_impl_int8_avx512.cc

@@ -52,7 +52,7 @@ static void compute_one_to_many_inner_product_avx512_vnni_int8(
     const int8_t *query, const int8_t **ptrs,
     std::array<const int8_t *, dp_batch> &prefetch_ptrs, size_t dimensionality,
     float *results) {
-  std::array<__m512i, dp_batch> accs;
+  __m512i accs[dp_batch];
   for (size_t i = 0; i < dp_batch; ++i) {
     accs[i] = _mm512_setzero_si512();
   }
@@ -60,7 +60,8 @@ static void compute_one_to_many_inner_product_avx512_vnni_int8(
   for (; dim + 64 <= dimensionality; dim += 64) {
     __m512i q =
         _mm512_loadu_si512(reinterpret_cast<const __m512i *>(query + dim));
-    std::array<__m512i, dp_batch> data_regs;
+
+    __m512i data_regs[dp_batch];
     for (size_t i = 0; i < dp_batch; ++i) {
       data_regs[i] =
           _mm512_loadu_si512(reinterpret_cast<const __m512i *>(ptrs[i] + dim));
@@ -74,7 +75,8 @@ static void compute_one_to_many_inner_product_avx512_vnni_int8(
       accs[i] = _mm512_dpbusd_epi32(accs[i], q, data_regs[i]);
     }
   }
-  std::array<int, dp_batch> temp_results{};
+
+  int temp_results[dp_batch]{};
   for (size_t i = 0; i < dp_batch; ++i) {
     temp_results[i] = _mm512_reduce_add_epi32(accs[i]);
   }
@@ -88,6 +90,7 @@ static void compute_one_to_many_inner_product_avx512_vnni_int8(
     results[i] = static_cast<float>(temp_results[i]);
   }
 }
+
 //
 // #elif defined(__AVX512BW__)
 // // TODO: this version is problematic