diff --git a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_indirect.hpp b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_indirect.hpp
index 5cbdf20..20c8230 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_indirect.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_indirect.hpp
@@ -523,7 +523,7 @@
         return size;
     }
 
-    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
         if (std::is_same<OutputStage, Requantize32>::value) {
             _col_bias = reinterpret_cast<int32_t *>(in_buffer);
 
@@ -534,6 +534,10 @@
                 compute_col_sums(*qp_ptr, _args._Nsize, _args._Ksize * _args._Ksections, B + (i * B_multi_stride), ldb, _col_bias + (i * _args._Nsize), _args._Ksize * _args._Ksections, i, 0);
             }
         }
+    }
+
+    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        requantize_bias(in_buffer, B, ldb, B_multi_stride);
 
         // Put the transposed data after the column sums - in non-transposing cases get_col_sum_size() == 0
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(in_buffer);
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
index c72dca2..efb5bd1 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized.hpp
@@ -269,12 +269,16 @@
         return get_col_sum_size() + (roundup(_Nsize, strategy::out_width()) * roundup(_Ksize, strategy::k_unroll()) * _nmulti * sizeof(Toi));
     }
 
-    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
         col_bias = reinterpret_cast<int32_t *>(in_buffer);
 
         for (unsigned int i=0; i<_nmulti; i++) {
             compute_col_sums(_qp, _Nsize, _Ksize, B + (i * B_multi_stride), ldb, col_bias + (i * _Nsize),  _Ksize, i, 0);
         }
+    }
+
+    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        requantize_bias(in_buffer, B, ldb, B_multi_stride);
 
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(in_buffer);
         Toi *buffer = reinterpret_cast<Toi *>(buffer_int + get_col_sum_size());
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized_inline.hpp b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized_inline.hpp
index 7376b5f..820b542 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized_inline.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_hybrid_quantized_inline.hpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017-2019 Arm Limited.
+ * Copyright (c) 2017-2019,2021 Arm Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -219,12 +219,16 @@
         return get_col_sum_size() + (roundup(_Nsize, strategy::out_width()) * roundup(_Ksize, strategy::k_unroll()) * _nmulti * sizeof(Toi));
     }
 
-    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
         col_bias = reinterpret_cast<int32_t *>(in_buffer);
 
         for (unsigned int i=0; i<_nmulti; i++) {
             compute_col_sums(_qp, _Nsize, _Ksize, B + (i * B_multi_stride), ldb, col_bias + (i * _Nsize),  _Ksize, i, 0);
         }
+    }
+
+    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        requantize_bias(in_buffer, B, ldb, B_multi_stride);
 
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(in_buffer);
         Toi *buffer = reinterpret_cast<Toi *>(buffer_int + get_col_sum_size());
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp
index 5639cb4..c75c320 100644
--- a/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemm_interleaved.hpp
@@ -923,7 +923,7 @@
         return (x_size * _Ktotal * _nmulti * sizeof(Toi)) + get_col_sum_size();
     }
 
-    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
         if (std::is_same<OutputStage, Requantize32>::value) {
             col_bias = reinterpret_cast<int32_t *>(in_buffer);
 
@@ -934,6 +934,10 @@
                 compute_col_sums(*qp_ptr, _Nsize, _Ksize * _Ksections, B + (i * B_multi_stride), ldb, col_bias + (i * _Nsize), _Ksize * _Ksections, i, 0);
             }
         }
+    }
+
+    void pretranspose_B_array(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        requantize_bias(in_buffer, B, ldb, B_multi_stride);
 
         // Put the transposed data after the column sums - in non-transposing cases get_col_sum_size() == 0
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(in_buffer);
diff --git a/src/core/NEON/kernels/arm_gemm/gemv_pretransposed.hpp b/src/core/NEON/kernels/arm_gemm/gemv_pretransposed.hpp
index d4348be..f0b4e5d 100644
--- a/src/core/NEON/kernels/arm_gemm/gemv_pretransposed.hpp
+++ b/src/core/NEON/kernels/arm_gemm/gemv_pretransposed.hpp
@@ -201,11 +201,11 @@
         return _buffer_per_multi * _args._nmulti * sizeof(To) + get_col_sum_size();
     }
 
-    void pretranspose_B_array(void *buffer, const To *B, const int ldb, const int B_multi_stride) override {
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
         // Column sums go on the front of the pretransposed buffer in requantized cases.
         // We could optimize here in case we don't actually need to sum the columns, but this code is only run on setup.
         if (std::is_same<OutputStage, Requantize32>::value) {
-            col_bias = reinterpret_cast<int32_t *>(buffer);
+            col_bias = reinterpret_cast<int32_t *>(in_buffer);
 
             Requantize32 *qp_ptr = reinterpret_cast<Requantize32 *>(&_os);
 
@@ -213,6 +213,10 @@
                 compute_col_sums(*qp_ptr, _args._Nsize, _args._Ksize, B + (i * B_multi_stride), ldb, col_bias + (i * _args._Nsize), _args._Ksize, i, 0);
             }
         }
+    }
+
+    void pretranspose_B_array(void *buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        requantize_bias(buffer, B, ldb, B_multi_stride);
 
         // The actual transposed buffer goes after the column sums (if any)
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(buffer);
diff --git a/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp b/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
index 1e2a9ac..ce72703 100644
--- a/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
+++ b/src/core/NEON/kernels/arm_gemm/quantize_wrapper.hpp
@@ -179,13 +179,16 @@
         return _subgemm->get_B_pretransposed_array_size() + col_sum_size();
     }
 
+    void requantize_bias(void *in_buffer, const To *B, const int ldb, const int B_multi_stride) override {
+        _col_sums = reinterpret_cast<int32_t *>(in_buffer);
+        col_sums_pretransposed(B, ldb, B_multi_stride);
+    }
+
     void pretranspose_B_array(void *buffer, const To *B, const int ldb, const int B_multi_stride) override {
         uintptr_t buffer_int = reinterpret_cast<uintptr_t>(buffer);
         _subgemm->pretranspose_B_array(reinterpret_cast<void *>(buffer_int + col_sum_size()), B, ldb, B_multi_stride);
 
-        _col_sums = reinterpret_cast<int32_t *>(buffer);
-
-        col_sums_pretransposed(B, ldb, B_multi_stride);
+        requantize_bias(buffer, B, ldb, B_multi_stride);
     }
 
     void set_pretransposed_B_data(void *buffer) override {