Revert "Update CPU kernels and add mixed sign GEMM support"

This reverts commit fc94f4d23abd4bc427b701f54ad85282e9ec7872 and 5d6fff041ade7eb44af0945867212f3979be3d3e (because the latter fixes a build failure caused by the former)

Change-Id: I7d07fea8307e9a7033b30874bbb14ba9202b23d8
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/11815
Benchmark: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Adnan AlSinan <adnan.alsinan@arm.com>

6 files changed, 20 insertions, 26 deletions

diff --git a/src/cpu/kernels/CpuGemmLowpMatrixMultiplyKernel.cpp b/src/cpu/kernels/CpuGemmLowpMatrixMultiplyKernel.cpp
index 87340e566e..5b88735e7a 100644
--- a/src/cpu/kernels/CpuGemmLowpMatrixMultiplyKernel.cpp
+++ b/src/cpu/kernels/CpuGemmLowpMatrixMultiplyKernel.cpp
@@ -684,10 +684,6 @@ Status validate_arguments(const ITensorInfo *src0, const ITensorInfo *src1, cons
                                                          DataType::U8);
     ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(dst, 1, DataType::S32);
 
-    ARM_COMPUTE_RETURN_ERROR_ON_MSG(src0->data_type() == DataType::QASYMM8_SIGNED &&
-                                        src1->data_type() == DataType::QASYMM8,
-                                    "QASYMM8_SIGNED input with QASYMM8 weights not supported");
-
     TensorShape in0_shape = src0->tensor_shape();
     TensorShape in1_shape = src1->tensor_shape();
     TensorShape out_shape = dst->tensor_shape();
diff --git a/src/cpu/kernels/CpuKernelSelectionTypes.h b/src/cpu/kernels/CpuKernelSelectionTypes.h
index 03a474de53..7c1e4772a6 100644
--- a/src/cpu/kernels/CpuKernelSelectionTypes.h
+++ b/src/cpu/kernels/CpuKernelSelectionTypes.h
@@ -105,7 +105,7 @@ struct SoftmaxKernelDataTypeISASelectorData
     cpuinfo::CpuIsaInfo isa;
     bool                is_log;
     int                 axis;
-    uint64_t            sme2_vector_length;
+    unsigned long       sme2_vector_length;
 };
 
 // Selector pointer types
diff --git a/src/cpu/kernels/assembly/CpuGemmAssemblyWrapperKernel.h b/src/cpu/kernels/assembly/CpuGemmAssemblyWrapperKernel.h
index 72fafca1bb..e2a27675b3 100644
--- a/src/cpu/kernels/assembly/CpuGemmAssemblyWrapperKernel.h
+++ b/src/cpu/kernels/assembly/CpuGemmAssemblyWrapperKernel.h
@@ -52,7 +52,7 @@ namespace kernel
   *
   *
   */
-template <typename TypeInput, typename TypeWeight, typename TypeOutput>
+template <typename TypeInput, typename TypeOutput>
 class CpuGemmAssemblyWrapperKernel final : public INEKernel
 {
 public:
@@ -101,7 +101,7 @@ public:
      * @param[in] kernel          Pointer to an assembly kernel implementation.
      * @param[in] kernel_name_tag Tag to be attacehd to the kernel's name.
      */
-    void configure(arm_gemm::GemmCommon<TypeInput, TypeWeight, TypeOutput> *kernel, std::string kernel_name_tag)
+    void configure(arm_gemm::GemmCommon<TypeInput, TypeOutput> *kernel, std::string kernel_name_tag)
     {
         ARM_COMPUTE_ERROR_ON_NULLPTR((reinterpret_cast<void *>(kernel)));
         _kernel = kernel;
@@ -131,8 +131,8 @@ public:
     }
 
 private:
-    arm_gemm::GemmCommon<TypeInput, TypeWeight, TypeOutput> *_kernel;
-    std::string                                              _name;
+    arm_gemm::GemmCommon<TypeInput, TypeOutput> *_kernel;
+    std::string                                  _name;
 };
 } // namespace kernel
 } // namespace cpu
diff --git a/src/cpu/kernels/assembly/arm_gemm.hpp b/src/cpu/kernels/assembly/arm_gemm.hpp
index cbc8be416e..941fed0ba8 100644
--- a/src/cpu/kernels/assembly/arm_gemm.hpp
+++ b/src/cpu/kernels/assembly/arm_gemm.hpp
@@ -277,8 +277,8 @@ struct Nothing
 {
 };
 
-template <typename Tlop, typename Trop, typename Tret>
-using UniqueGemmCommon = std::unique_ptr<GemmCommon<Tlop, Trop, Tret>>;
+template <typename Top, typename Tret>
+using UniqueGemmCommon = std::unique_ptr<GemmCommon<Top, Tret>>;
 
 /* Low level API calls.
  * These are implemented as 'GemmArgs' versions, or with the arguments explicitly listed. */
@@ -288,13 +288,13 @@ using UniqueGemmCommon = std::unique_ptr<GemmCommon<Tlop, Trop, Tret>>;
 template <typename Top, typename Tret, class OutputStage = Nothing>
 KernelDescription get_gemm_method(const GemmArgs &args, const OutputStage & = {});
 
-template <typename Tlop, typename Trop, typename Tret, class OutputStage = Nothing>
-UniqueGemmCommon<Tlop, Trop, Tret> gemm(const GemmArgs &args, const OutputStage & = {});
+template <typename Top, typename Tret, class OutputStage = Nothing>
+UniqueGemmCommon<Top, Tret> gemm(const GemmArgs &args, const OutputStage & = {});
 
-template <typename Tlop, typename Trop, typename Tret, class OutputStage = Nothing>
+template <typename Top, typename Tret, class OutputStage = Nothing>
 std::vector<KernelDescription> get_compatible_kernels(const GemmArgs &args, const OutputStage & = {});
 
-template <typename Tlop, typename Trop, typename Tret, class OutputStage = Nothing>
+template <typename Top, typename Tret, class OutputStage = Nothing>
 bool has_opt_gemm(WeightFormat &weight_format, const GemmArgs &args, const OutputStage & = {});
 
 } // namespace arm_gemm
diff --git a/src/cpu/kernels/assembly/convolution_parameters.hpp b/src/cpu/kernels/assembly/convolution_parameters.hpp
index 09b73ca409..a6cf96344c 100644
--- a/src/cpu/kernels/assembly/convolution_parameters.hpp
+++ b/src/cpu/kernels/assembly/convolution_parameters.hpp
@@ -61,8 +61,6 @@ struct ConvolutionParameters
     int64_t output_stride_w;
     int64_t output_stride_h;
     //          output_channels not included as they do not affect the input.
-    int64_t dilation_w;
-    int64_t dilation_h;
     int64_t padding_top;
     int64_t padding_left;
     float   padding_value;
diff --git a/src/cpu/kernels/assembly/gemm_common.hpp b/src/cpu/kernels/assembly/gemm_common.hpp
index f693021fcb..45d1e43274 100644
--- a/src/cpu/kernels/assembly/gemm_common.hpp
+++ b/src/cpu/kernels/assembly/gemm_common.hpp
@@ -189,7 +189,7 @@ public:
  * 'set_arrays' to capture the provided arguments in protected class
  * members, as essentially any implementation will need these.
  */
-template <typename To, typename Tw, typename Tr>
+template <typename To, typename Tr>
 class GemmCommon : public IGemmCommon
 {
 protected:
@@ -197,7 +197,7 @@ protected:
     int       _lda               = 0;
     int       _A_batch_stride    = 0;
     int       _A_multi_stride    = 0;
-    const Tw *_Bptr              = nullptr;
+    const To *_Bptr              = nullptr;
     int       _ldb               = 0;
     int       _B_multi_stride    = 0;
     Tr       *_Cptr              = nullptr;
@@ -214,7 +214,7 @@ public:
                             const int                                     lda,
                             const int                                     A_batch_stride,
                             const int                                     A_multi_stride,
-                            const Tw                                     *B,
+                            const To                                     *B,
                             const int                                     ldb,
                             /* batches share B */ const int               B_multi_stride,
                             Tr                                           *C,
@@ -254,7 +254,7 @@ public:
                             const void                                   *bias,
                             /* no row or batch stride needed */ const int bias_multi_stride) override
     {
-        set_arrays(static_cast<const To *>(A), lda, A_batch_stride, A_multi_stride, static_cast<const Tw *>(B), ldb,
+        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,
                    static_cast<const Tr *>(bias), bias_multi_stride);
     }
@@ -262,17 +262,17 @@ public:
     /*** "Pretransposed" interface ***/
 
     /* Compute col sums over all columns */
-    virtual void requantize_bias(void *, const Tw *, const int, const int){};
+    virtual void requantize_bias(void *, const To *, const int, const int){};
 
     /* Perform pretranspose - the void * passed in must remain allocated for the duration of any execute calls. */
     /* Arguments are: output buffer pointer, source pointer, source row stride, source multi stride */
-    virtual void pretranspose_B_array(void *, const Tw *, const int, const int, bool){};
+    virtual void pretranspose_B_array(void *, const To *, const int, const int, bool){};
 
     /* Implementation of the void * overload which casts its arguments to the appropriate type. */
     void pretranspose_B_array_generic(
         void *out, const void *in, const int row_stride, const int multi_stride, bool transposed) override
     {
-        pretranspose_B_array(out, static_cast<const Tw *>(in), row_stride, multi_stride, transposed);
+        pretranspose_B_array(out, static_cast<const To *>(in), row_stride, multi_stride, transposed);
     }
 
     /* Threaded versions of the above.
@@ -280,7 +280,7 @@ public:
      * just calls the non-threaded functions to do the work.  This is valid as with window size of 1 the only
      * legal values for start and end are 0 and 1 respectively. */
     virtual void pretranspose_B_array_part(
-        void *out, const Tw *in, const int row_stride, const int multi_stride, bool transposed, size_t, size_t)
+        void *out, const To *in, const int row_stride, const int multi_stride, bool transposed, size_t, size_t)
     {
         pretranspose_B_array(out, in, row_stride, multi_stride, transposed);
     };
@@ -293,7 +293,7 @@ public:
                                            size_t      start,
                                            size_t      end) override
     {
-        pretranspose_B_array_part(out, static_cast<const Tw *>(in), row_stride, multi_stride, transposed, start, end);
+        pretranspose_B_array_part(out, static_cast<const To *>(in), row_stride, multi_stride, transposed, start, end);
     }
 
     /*** Indirect interface ***/