COMPMID-2178: Update GEMM assembly code.

Perform offset reduction and requantization within the assembly wrapper.

Change-Id: I5d5b3e1f6f9ef4c71805362c57f88ff199c027a3
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1541
Comments-Addressed: Pablo Marquez <pablo.tello@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 7 insertions, 3 deletions

diff --git a/arm_compute/core/NEON/kernels/assembly/gemm_common.hpp b/arm_compute/core/NEON/kernels/assembly/gemm_common.hpp
index bb32fea9da..f59a61703f 100644
--- a/arm_compute/core/NEON/kernels/assembly/gemm_common.hpp
+++ b/arm_compute/core/NEON/kernels/assembly/gemm_common.hpp
@@ -46,7 +46,7 @@ public:
      */
     virtual void set_arrays_generic(const void *A, const int lda, const int A_batch_stride, const int A_multi_stride,
                                     const void *B, const int ldb, /* batches share B */     const int B_multi_stride,
-                                    void *C, const int ldc, const int C_batch_stride, const int C_multi_stride) = 0;
+                                          void *C, const int ldc, const int C_batch_stride, const int C_multi_stride) = 0;
 
     /* For threading, we divide the work into some number of units and work
      * out internally what unit corresponds to what work.  This returns the
@@ -93,6 +93,10 @@ public:
     /* Set pretransposed data - the void * passed in must previously have been passed to pretranspose_B_array() for the same or a similar GEMM. */
     virtual void set_pretransposed_B_data(void *) { }
 
+    /*** "Quantized bias" interface (optional) ***/
+    /* Set the bias vector for quantized GEMMs */
+    virtual void set_quantized_bias(const int32_t *bias) { }
+
     // Destructor
     virtual ~IGemmCommon() { }
 };
@@ -125,7 +129,7 @@ public:
      * strides (templated version with appropriate types). */
     virtual void set_arrays(const To *A, const int lda, const int A_batch_stride, const int A_multi_stride,
                             const To *B, const int ldb, /* batches share B */     const int B_multi_stride,
-                            Tr *C, const int ldc, const int C_batch_stride, const int C_multi_stride) {
+                                  Tr *C, const int ldc, const int C_batch_stride, const int C_multi_stride) {
         _Aptr = A;
         _lda = lda;
         _A_batch_stride = A_batch_stride;
@@ -142,7 +146,7 @@ public:
     /* Implementation of the void * overload which casts its arguments to the appropriate type. */
     void set_arrays_generic(const void *A, const int lda, const int A_batch_stride, const int A_multi_stride,
                             const void *B, const int ldb, /* batches share B */     const int B_multi_stride,
-                            void *C, const int ldc, const int C_batch_stride, const int C_multi_stride) override {
+                                  void *C, const int ldc, const int C_batch_stride, const int C_multi_stride) override {
         set_arrays(static_cast<const To *>(A), lda, A_batch_stride, A_multi_stride,
                    static_cast<const To *>(B), ldb, B_multi_stride,
                    static_cast<Tr *>(C), ldc, C_batch_stride, C_multi_stride);