Update with clang format (#51)

Summary:
Pull Request resolved: https://github.com/pytorch/FBGEMM/pull/51

Use Clang formatting with "clang-format -i *.cc *.h".

Reviewed By: dskhudia

Differential Revision: D13532121

fbshipit-source-id: 6792d008f3295c128942f4896e8221aebbf2566e

22 files changed, 42 insertions, 94 deletions

diff --git a/bench/PackedFloatInOutBenchmark.cc b/bench/PackedFloatInOutBenchmark.cc
index 79a750e..f2589f7 100644
--- a/bench/PackedFloatInOutBenchmark.cc
+++ b/bench/PackedFloatInOutBenchmark.cc
@@ -187,13 +187,7 @@ void performance_test() {
         row_offset_buf.data());
 
     PackBMatrix<int8_t> packedBN(
-        matrix_op_t::NoTranspose,
-        k,
-        n,
-        Bint8.data(),
-        n,
-        nullptr,
-        1);
+        matrix_op_t::NoTranspose, k, n, Bint8.data(), n, nullptr, 1);
 
     DoNothing<float, float> doNothingObj{};
     ReQuantizeForFloat<false> outputProcObj(
diff --git a/bench/PackedRequantizeAcc16Benchmark.cc b/bench/PackedRequantizeAcc16Benchmark.cc
index f60332f..89cca4f 100644
--- a/bench/PackedRequantizeAcc16Benchmark.cc
+++ b/bench/PackedRequantizeAcc16Benchmark.cc
@@ -296,13 +296,7 @@ void performance_test() {
               PackAWithRowOffset<uint8_t, int16_t>::rowOffsetBufferSize());
 
           PackAMatrix<uint8_t, int16_t> packA(
-              matrix_op_t::NoTranspose,
-              m,
-              k,
-              Aint8.data(),
-              k,
-              nullptr,
-              1);
+              matrix_op_t::NoTranspose, m, k, Aint8.data(), k, nullptr, 1);
           PackAWithRowOffset<uint8_t, int16_t> packAWithRowOffset(
               matrix_op_t::NoTranspose,
               m,
diff --git a/bench/PackedRequantizeAcc32Benchmark.cc b/bench/PackedRequantizeAcc32Benchmark.cc
index b255b8c..5096475 100644
--- a/bench/PackedRequantizeAcc32Benchmark.cc
+++ b/bench/PackedRequantizeAcc32Benchmark.cc
@@ -209,13 +209,7 @@ void performance_test() {
     // offsets before");
 
     PackBMatrix<int8_t> packedBN(
-        matrix_op_t::NoTranspose,
-        k,
-        n,
-        Bint8.data(),
-        n,
-        nullptr,
-        1);
+        matrix_op_t::NoTranspose, k, n, Bint8.data(), n, nullptr, 1);
 
     ttot = 0.0;
     runType = "FBGEMM_i8_acc32";
diff --git a/include/fbgemm/Fbgemm.h b/include/fbgemm/Fbgemm.h
index 3cf832c..b729d56 100644
--- a/include/fbgemm/Fbgemm.h
+++ b/include/fbgemm/Fbgemm.h
@@ -15,11 +15,11 @@
 #include <memory>
 #include <type_traits>
 #include "ConvUtils.h"
+#include "FbgemmBuild.h"
 #include "FbgemmI8Spmdm.h"
+#include "QuantUtilsAvx2.h"
 #include "Types.h"
 #include "Utils.h"
-#include "FbgemmBuild.h"
-#include "QuantUtilsAvx2.h"
 
 // Turning on this option will print out time breakdown of each stage (e.g.,
 // input packing, the main GEMM kernel, each output processing pipeline).
diff --git a/include/fbgemm/FbgemmI8Spmdm.h b/include/fbgemm/FbgemmI8Spmdm.h
index 01416d5..93af6ea 100644
--- a/include/fbgemm/FbgemmI8Spmdm.h
+++ b/include/fbgemm/FbgemmI8Spmdm.h
@@ -9,8 +9,8 @@
 #include <cstdint>
 #include <vector>
 #include "ConvUtils.h"
-#include "Utils.h"
 #include "FbgemmBuild.h"
+#include "Utils.h"
 
 // #define FBGEMM_MEASURE_TIME_BREAKDOWN
 
diff --git a/include/fbgemm/OutputProcessing-inl.h b/include/fbgemm/OutputProcessing-inl.h
index 2ff64f3..c250942 100644
--- a/include/fbgemm/OutputProcessing-inl.h
+++ b/include/fbgemm/OutputProcessing-inl.h
@@ -111,7 +111,7 @@ inline int ReQuantizeOutput<FUSE_RELU, Q_GRAN, outT, inT, nextOPType>::f(
   } else if (instSet == inst_set_t::avx2 || instSet == inst_set_t::avx512) {
     if (std::is_same<outT, uint8_t>::value) {
       bool b_symmetric = (Q_GRAN == QuantizationGranularity::TENSOR &&
-                             Bq_zero_point_[0] == 0) ||
+                          Bq_zero_point_[0] == 0) ||
           q_row_offsets_ == nullptr;
 
       requantizationParams_t r = {Aq_zero_point_,
diff --git a/include/fbgemm/QuantUtils.h b/include/fbgemm/QuantUtils.h
index f658f73..43855d8 100644
--- a/include/fbgemm/QuantUtils.h
+++ b/include/fbgemm/QuantUtils.h
@@ -68,8 +68,7 @@ FBGEMM_API T Quantize(
 
 template <typename T>
 FBGEMM_API T Quantize(float src, const TensorQuantizationParams& qparams) {
-  return Quantize<T>(
-      src, qparams.zero_point, qparams.scale, qparams.precision);
+  return Quantize<T>(src, qparams.zero_point, qparams.scale, qparams.precision);
 }
 
 template <typename T>
diff --git a/include/fbgemm/Utils.h b/include/fbgemm/Utils.h
index a7e91cd..775e18c 100644
--- a/include/fbgemm/Utils.h
+++ b/include/fbgemm/Utils.h
@@ -7,8 +7,8 @@
 #pragma once
 #include <string>
 #include <type_traits>
-#include "UtilsAvx2.h"
 #include "FbgemmBuild.h"
+#include "UtilsAvx2.h"
 
 namespace fbgemm {
 
diff --git a/src/ExecuteKernelU8S8.cc b/src/ExecuteKernelU8S8.cc
index 3b6e059..e7f7c70 100644
--- a/src/ExecuteKernelU8S8.cc
+++ b/src/ExecuteKernelU8S8.cc
@@ -288,7 +288,7 @@ INSTANTIATE_ACC_T(PackAWithRowOffset);
   INSTANTIATE_Q_GRANS(ACC_T, RELU, 2);       \
   INSTANTIATE_Q_GRANS(ACC_T, RELU, 3);
 
-#define INSTANTIATE_RELU(ACC_T)   \
+#define INSTANTIATE_RELU(ACC_T)          \
   INSTANTIATE_SPATIAL_DIM(ACC_T, false); \
   INSTANTIATE_SPATIAL_DIM(ACC_T, true);
 
@@ -410,7 +410,7 @@ template class ExecuteKernel<
   template class ExecuteKernel<         \
       PACK_A<uint8_t, ACC_T>,           \
       PackBMatrix<int8_t, ACC_T>,       \
-      int32_t,                            \
+      int32_t,                          \
       memCopy<>>;
 
 #define INSTANTIATE_ACC_T(PACK_A)   \
@@ -427,7 +427,7 @@ INSTANTIATE_ACC_T(PackAWithRowOffset);
   template class ExecuteKernel<                     \
       PackAWithIm2Col<uint8_t, ACC_T, SPATIAL_DIM>, \
       PackBMatrix<int8_t, ACC_T>,                   \
-      int32_t,                                        \
+      int32_t,                                      \
       memCopy<>>;
 
 #define INSTANTIATE_SPATIAL_DIM(ACC_T) \
diff --git a/src/Fbgemm.cc b/src/Fbgemm.cc
index b7a99c6..45108d0 100644
--- a/src/Fbgemm.cc
+++ b/src/Fbgemm.cc
@@ -254,7 +254,7 @@ INSTANTIATE_ACC_T(PackAWithRowOffset);
   INSTANTIATE_Q_GRANS(ACC_T, RELU, 2);       \
   INSTANTIATE_Q_GRANS(ACC_T, RELU, 3);
 
-#define INSTANTIATE_RELU(ACC_T)   \
+#define INSTANTIATE_RELU(ACC_T)          \
   INSTANTIATE_SPATIAL_DIM(ACC_T, false); \
   INSTANTIATE_SPATIAL_DIM(ACC_T, true);
 
diff --git a/src/FbgemmI8DepthwiseAvx2.cc b/src/FbgemmI8DepthwiseAvx2.cc
index ace3f7e..9dcd7e1 100644
--- a/src/FbgemmI8DepthwiseAvx2.cc
+++ b/src/FbgemmI8DepthwiseAvx2.cc
@@ -112,13 +112,12 @@ PackedDepthWiseConvMatrix<KERNEL_PROD>::PackedDepthWiseConvMatrix(
           reinterpret_cast<const __m256i*>(masks[remainder / 4]));
       for (int i = 0; i < KERNEL_PROD; ++i) {
         b_v[i] = _mm256_maskload_epi32(
-            reinterpret_cast<const int*>(smat_transposed + i * K + k1),
-            mask_v);
+            reinterpret_cast<const int*>(smat_transposed + i * K + k1), mask_v);
       }
     } else {
       for (int i = 0; i < KERNEL_PROD; ++i) {
-        b_v[i] = _mm256_lddqu_si256(reinterpret_cast<const __m256i*>(
-            smat_transposed + i * K + k1));
+        b_v[i] = _mm256_lddqu_si256(
+            reinterpret_cast<const __m256i*>(smat_transposed + i * K + k1));
       }
     }
 
@@ -457,14 +456,7 @@ static inline __attribute__((always_inline)) void inner_prod_packed_(
           a_v[k], Bp + k, &c[0], &c[1], &c[2], &c[3], a_sum_temp);
     } else if (K - k == 2) {
       madd_epi16x2_packed<SUM_A>(
-          a_v[k],
-          a_v[k + 1],
-          Bp + k,
-          &c[0],
-          &c[1],
-          &c[2],
-          &c[3],
-          a_sum_temp);
+          a_v[k], a_v[k + 1], Bp + k, &c[0], &c[1], &c[2], &c[3], a_sum_temp);
     }
 
     c[0] = _mm256_add_epi32(c[0], c_temp[0]);
@@ -860,11 +852,7 @@ static inline __attribute__((always_inline)) void inner_prod_3x3_packed_(
 
   __m256i a_sum[4];
   inner_prod_3x3_packed_<SUM_A, REMAINDER>(
-      a_v,
-      reinterpret_cast<const __m256i*>(Bp),
-      C,
-      remainder,
-      a_sum);
+      a_v, reinterpret_cast<const __m256i*>(Bp), C, remainder, a_sum);
   if (SUM_A) {
     __m256i B_zero_point_v;
     for (int i = 0; i < (REMAINDER ? (remainder / 8) : 4); ++i) {
@@ -974,11 +962,7 @@ static inline __attribute__((always_inline)) void inner_prod_3x3x3_packed_(
 
   __m256i a_sum[4];
   inner_prod_packed_<8, SUM_A, REMAINDER>(
-      a_v,
-      reinterpret_cast<const __m256i*>(Bp),
-      C,
-      remainder,
-      a_sum);
+      a_v, reinterpret_cast<const __m256i*>(Bp), C, remainder, a_sum);
 
   a_v[0] = A_zero_point_v;
   a_v[1] = A_zero_point_v;
@@ -1031,11 +1015,7 @@ static inline __attribute__((always_inline)) void inner_prod_3x3x3_packed_(
 
   __m256i a_sum_temp[4];
   inner_prod_packed_<8, SUM_A, REMAINDER, true /* acc */>(
-      a_v,
-      reinterpret_cast<const __m256i*>(Bp) + 8,
-      C,
-      remainder,
-      a_sum_temp);
+      a_v, reinterpret_cast<const __m256i*>(Bp) + 8, C, remainder, a_sum_temp);
   if (SUM_A) {
     a_sum[0] = _mm256_add_epi32(a_sum[0], a_sum_temp[0]);
     a_sum[1] = _mm256_add_epi32(a_sum[1], a_sum_temp[1]);
@@ -1090,11 +1070,7 @@ static inline __attribute__((always_inline)) void inner_prod_3x3x3_packed_(
   }
 
   inner_prod_packed_<8, SUM_A, REMAINDER, true /* acc */>(
-      a_v,
-      reinterpret_cast<const __m256i*>(Bp) + 16,
-      C,
-      remainder,
-      a_sum_temp);
+      a_v, reinterpret_cast<const __m256i*>(Bp) + 16, C, remainder, a_sum_temp);
   if (SUM_A) {
     a_sum[0] = _mm256_add_epi32(a_sum[0], a_sum_temp[0]);
     a_sum[1] = _mm256_add_epi32(a_sum[1], a_sum_temp[1]);
@@ -1121,11 +1097,7 @@ static inline __attribute__((always_inline)) void inner_prod_3x3x3_packed_(
   }
 
   inner_prod_packed_<3, SUM_A, REMAINDER, true /* acc */>(
-      a_v,
-      reinterpret_cast<const __m256i*>(Bp) + 24,
-      C,
-      remainder,
-      a_sum_temp);
+      a_v, reinterpret_cast<const __m256i*>(Bp) + 24, C, remainder, a_sum_temp);
 
   if (SUM_A) {
     a_sum[0] = _mm256_add_epi32(a_sum[0], a_sum_temp[0]);
diff --git a/src/FbgemmI8Spmdm.cc b/src/FbgemmI8Spmdm.cc
index 8d69c9f..c249871 100644
--- a/src/FbgemmI8Spmdm.cc
+++ b/src/FbgemmI8Spmdm.cc
@@ -236,8 +236,7 @@ void CompressedSparseColumn::SparseConv(
   // TODO: if not hyper sparse, transpose a block of A matrix as in SpMDM.
   if (!accumulation) {
     for (int i = block.row_start; i < block.row_start + block.row_size; ++i) {
-      for (int j = block.col_start; j < block.col_start + block.col_size;
-           ++j) {
+      for (int j = block.col_start; j < block.col_start + block.col_size; ++j) {
         C[(i - block.row_start) * ldc + j - block.col_start] = 0;
       }
     }
@@ -245,8 +244,7 @@ void CompressedSparseColumn::SparseConv(
   for (int j = block.col_start; j < block.col_start + block.col_size; ++j) {
     for (int k = colptr_[j]; k < colptr_[j + 1]; ++k) {
       int v = values_[k];
-      for (int i = block.row_start; i < block.row_start + block.row_size;
-           ++i) {
+      for (int i = block.row_start; i < block.row_start + block.row_size; ++i) {
         int ow = i % conv_p.OUT_DIM[1];
         int oh = i / conv_p.OUT_DIM[1] % conv_p.OUT_DIM[0];
         int n = i / conv_p.OUT_DIM[1] / conv_p.OUT_DIM[0];
diff --git a/src/GenerateKernelU8S8S32ACC16.cc b/src/GenerateKernelU8S8S32ACC16.cc
index f845e44..b7bc676 100644
--- a/src/GenerateKernelU8S8S32ACC16.cc
+++ b/src/GenerateKernelU8S8S32ACC16.cc
@@ -21,7 +21,6 @@ thread_local std::map<
     typename CodeGenBase<TA, TB, TC, accT>::jit_micro_kernel_fp>
     CodeGenBase<TA, TB, TC, accT>::codeCache_;
 
-
 namespace x86 = asmjit::x86;
 
 /**
diff --git a/src/PackAWithIm2Col.cc b/src/PackAWithIm2Col.cc
index 5856942..f096d7a 100644
--- a/src/PackAWithIm2Col.cc
+++ b/src/PackAWithIm2Col.cc
@@ -11,8 +11,8 @@
 #include <iostream>
 #include <numeric>
 
-#include "fbgemm/Fbgemm.h"
 #include "OptimizedKernelsAvx2.h"
+#include "fbgemm/Fbgemm.h"
 
 namespace fbgemm {
 
diff --git a/src/PackAWithQuantRowOffset.cc b/src/PackAWithQuantRowOffset.cc
index 98e862b..02e701e 100644
--- a/src/PackAWithQuantRowOffset.cc
+++ b/src/PackAWithQuantRowOffset.cc
@@ -11,9 +11,9 @@
 #include <iomanip>
 #include <iostream>
 #include <stdexcept>
+#include "OptimizedKernelsAvx2.h"
 #include "fbgemm/Fbgemm.h"
 #include "fbgemm/QuantUtilsAvx2.h"
-#include "OptimizedKernelsAvx2.h"
 
 namespace fbgemm {
 
diff --git a/src/PackAWithRowOffset.cc b/src/PackAWithRowOffset.cc
index 30da4c0..4882bb5 100644
--- a/src/PackAWithRowOffset.cc
+++ b/src/PackAWithRowOffset.cc
@@ -10,8 +10,8 @@
 #include <iomanip>
 #include <iostream>
 #include <stdexcept>
-#include "fbgemm/Fbgemm.h"
 #include "OptimizedKernelsAvx2.h"
+#include "fbgemm/Fbgemm.h"
 
 namespace fbgemm {
 
diff --git a/src/QuantUtils.cc b/src/QuantUtils.cc
index b346952..1d057e2 100644
--- a/src/QuantUtils.cc
+++ b/src/QuantUtils.cc
@@ -153,16 +153,16 @@ void ChooseRequantizationMultiplier(
 ////////////////////////////////////////////////////////////////////////////////
 // Utility functions
 
-#define FBGEMM_SPECIALIZED_QUANTIZE(T)      \
-  template <>                                 \
-  void Quantize<T>(                         \
-      const float* src,                     \
-      T* dst,                                 \
-      const int len,                          \
-      const TensorQuantizationParams& qparams) {   \
-    for (int i = 0; i < len; ++i) {           \
-      dst[i] = Quantize<T>(src[i], qparams); \
-    }                                         \
+#define FBGEMM_SPECIALIZED_QUANTIZE(T)           \
+  template <>                                    \
+  void Quantize<T>(                              \
+      const float* src,                          \
+      T* dst,                                    \
+      const int len,                             \
+      const TensorQuantizationParams& qparams) { \
+    for (int i = 0; i < len; ++i) {              \
+      dst[i] = Quantize<T>(src[i], qparams);     \
+    }                                            \
   }
 FBGEMM_SPECIALIZED_QUANTIZE(int8_t)
 FBGEMM_SPECIALIZED_QUANTIZE(uint16_t)
diff --git a/src/Utils.cc b/src/Utils.cc
index 88d029d..fcd8ade 100644
--- a/src/Utils.cc
+++ b/src/Utils.cc
@@ -5,7 +5,6 @@
  * LICENSE file in the root directory of this source tree.
  */
 #include "fbgemm/Utils.h"
-#include "TransposeUtils.h"
 #include <cpuinfo.h>
 #include <cassert>
 #include <cinttypes>
@@ -14,6 +13,7 @@
 #include <iostream>
 #include <limits>
 #include <stdexcept>
+#include "TransposeUtils.h"
 
 namespace fbgemm {
 
diff --git a/src/UtilsAvx2.cc b/src/UtilsAvx2.cc
index badf70b..5659200 100644
--- a/src/UtilsAvx2.cc
+++ b/src/UtilsAvx2.cc
@@ -4,8 +4,8 @@
  * This source code is licensed under the BSD-style license found in the
  * LICENSE file in the root directory of this source tree.
  */
-#include "TransposeUtils.h"
 #include <immintrin.h>
+#include "TransposeUtils.h"
 
 namespace fbgemm {
 
@@ -145,7 +145,6 @@ inline void transpose_kernel_8x8_avx2(
   _mm256_storeu_ps(&dst[7 * ld_dst], h);
 }
 
-
 void transpose_8x8(
     int M,
     int N,
diff --git a/src/UtilsAvx512.cc b/src/UtilsAvx512.cc
index f49bb6f..44d4f9a 100644
--- a/src/UtilsAvx512.cc
+++ b/src/UtilsAvx512.cc
@@ -5,8 +5,8 @@
  * LICENSE file in the root directory of this source tree.
  */
 
-#include "TransposeUtils.h"
 #include <immintrin.h>
+#include "TransposeUtils.h"
 
 namespace fbgemm {
 
diff --git a/test/Im2ColFusedRequantizeTest.cc b/test/Im2ColFusedRequantizeTest.cc
index 73971b9..feae002 100644
--- a/test/Im2ColFusedRequantizeTest.cc
+++ b/test/Im2ColFusedRequantizeTest.cc
@@ -244,7 +244,7 @@ TEST_P(fbgemmIm2colTest, Acc16Test) {
   }
 }
 
-template<QuantizationGranularity Q_GRAN>
+template <QuantizationGranularity Q_GRAN>
 void SConvTest() {
   for (auto conv_p : shapes) {
     for (int groups : {1, 4}) {
diff --git a/test/PackedRequantizeTest.cc b/test/PackedRequantizeTest.cc
index 9873e3f..ab3e1a7 100644
--- a/test/PackedRequantizeTest.cc
+++ b/test/PackedRequantizeTest.cc
@@ -38,8 +38,7 @@ class fbgemmu8s8acc32WithQuantGranularityTest
     : public testing::TestWithParam<
           tuple<matrix_op_t, matrix_op_t, bool, QuantizationGranularity>> {};
 class fbgemmu8s8acc32Test
-    : public testing::TestWithParam<
-          tuple<matrix_op_t, matrix_op_t, bool>> {};
+    : public testing::TestWithParam<tuple<matrix_op_t, matrix_op_t, bool>> {};
 }; // namespace
 
 INSTANTIATE_TEST_CASE_P(