COMPMID-1413 - Improve the performance of GEMMLowp with 8 bit dot product on OpenCL

COMPMID-1424 - Add dot product support for CLDepthwise QASYMM8 3x3 NHWC non-unit stride

With this patch we are able to improve the performance of MobileNet v1-qasymm8 by 37 %
Tried to use the dot product instruction in CLDepthwise QASYMM8 3x3 NHWC non-unit stride
but I have not seen any benefit (maybe because we have few arithemtic operation and we
do not have more load instructions). However Depthwise convolution has been improved by
30%

Change-Id: Id768a99c2e53a04276707e427af5d0ec93419ada
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/155082
Tested-by: bsgcomp <bsgcomp@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>

1 files changed, 40 insertions, 9 deletions

diff --git a/arm_compute/core/Types.h b/arm_compute/core/Types.h
index 5e04bcd0f4..134b8e2905 100644
--- a/arm_compute/core/Types.h
+++ b/arm_compute/core/Types.h
@@ -1205,6 +1205,26 @@ private:
     const bool _reinterpret_input_as_3d;
 };
 
+/** GEMMLowp output stage type */
+enum class GEMMLowpOutputStageType
+{
+    NONE,                     /**< No quantization to uint8 */
+    QUANTIZE_DOWN,            /**< Quantize to uint8 using an integer multiplication */
+    QUANTIZE_DOWN_FIXEDPOINT, /**< Quantize to uint8 using a fixed point multiplication */
+    QUANTIZE_DOWN_FLOAT       /**< Quantize to uint8 using a floating point multiplication */
+};
+
+/** GEMMLowp output stage info */
+struct GEMMLowpOutputStageInfo
+{
+    GEMMLowpOutputStageType type{ GEMMLowpOutputStageType::NONE }; /**< GEMMLowp output stage type */
+    int                     gemmlowp_offset{ 0 };                  /**< GEMMLowp output stage offset used for quantizing to QASYMM8 */
+    int                     gemmlowp_multiplier{ 0 };              /**< GEMMLowp output stage multiplier used for quantizing to QASYMM8 */
+    int                     gemmlowp_shift{ 0 };                   /**< GEMMLowp output stage shift used for quantizing to uint8 */
+    int                     gemmlowp_min_bound{ 0 };               /**< GEMMLowp min value used to saturate down the output result before converting back to QASYMM8 */
+    int                     gemmlowp_max_bound{ 0 };               /**< GEMMLowp max value used to saturate down the output result before converting back to QASYMM8 */
+};
+
 /** GEMM information class. This class stores the necessary information to compute GEMM functions
  *
  * This object also contains the information about how matrix A and matrix B have been reshaped
@@ -1215,7 +1235,7 @@ class GEMMInfo
 public:
     /** Default constructor */
     GEMMInfo()
-        : _is_a_reshaped(false), _is_b_reshaped(false), _reshape_b_only_on_first_run(false), _depth_output_gemm3d(1), _reinterpret_input_as_3d(false), _retain_internal_weights(false)
+        : _is_a_reshaped(false), _is_b_reshaped(false), _reshape_b_only_on_first_run(false), _depth_output_gemm3d(1), _reinterpret_input_as_3d(false), _retain_internal_weights(false), _gemmlowp_output_stage()
     {
     }
     /** Constructor
@@ -1227,11 +1247,13 @@ public:
      * @param[in] reinterpret_input_as_3d     (Optional) Reinterpret the input as 3D tensor. (i.e. this flag should be set to true when GEMM is used
      *                                        to perform 1x1 convolutions with the NHWC data layout)
      * @param[in] retain_internal_weights     (Optional) Retain the weights tensor from previous run
+     * @param[in] gemmlowp_output_stage       (Optional) GEMMLowp Output stage info
      *
      */
-    GEMMInfo(bool is_a_reshaped, bool is_b_reshaped, bool reshape_b_only_on_first_run, int depth_output_gemm3d = 1, bool reinterpret_input_as_3d = false, bool retain_internal_weights = false)
+    GEMMInfo(bool is_a_reshaped, bool is_b_reshaped, bool reshape_b_only_on_first_run, int depth_output_gemm3d = 1, bool reinterpret_input_as_3d = false, bool retain_internal_weights = false,
+             GEMMLowpOutputStageInfo gemmlowp_output_stage = GEMMLowpOutputStageInfo())
         : _is_a_reshaped(is_a_reshaped), _is_b_reshaped(is_b_reshaped), _reshape_b_only_on_first_run(reshape_b_only_on_first_run), _depth_output_gemm3d(depth_output_gemm3d),
-          _reinterpret_input_as_3d(reinterpret_input_as_3d), _retain_internal_weights(retain_internal_weights)
+          _reinterpret_input_as_3d(reinterpret_input_as_3d), _retain_internal_weights(retain_internal_weights), _gemmlowp_output_stage(gemmlowp_output_stage)
     {
     }
     /** Flag which specifies if the matrix A has been reshaped
@@ -1284,14 +1306,23 @@ public:
     {
         return _retain_internal_weights;
     };
+    /** GEMMLowp output stage
+     *
+     * @return the GEMMLowp output stage info
+     */
+    GEMMLowpOutputStageInfo gemmlowp_output_stage() const
+    {
+        return _gemmlowp_output_stage;
+    };
 
 private:
-    const bool _is_a_reshaped;
-    const bool _is_b_reshaped;
-    const bool _reshape_b_only_on_first_run;
-    const int  _depth_output_gemm3d;
-    const bool _reinterpret_input_as_3d;
-    const bool _retain_internal_weights;
+    const bool                    _is_a_reshaped;
+    const bool                    _is_b_reshaped;
+    const bool                    _reshape_b_only_on_first_run;
+    const int                     _depth_output_gemm3d;
+    const bool                    _reinterpret_input_as_3d;
+    const bool                    _retain_internal_weights;
+    const GEMMLowpOutputStageInfo _gemmlowp_output_stage;
 };
 
 /** Winograd information */