COMPMID-1979: Fuse Activation Function in CLGEMM - part 1

Implementing a new struct to contains the information for the
OpenCL GEMM kernels

Change-Id: I6c641c312f9c3b025a7c69dd0df3b730d2d2c2cb
Signed-off-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1434
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Giuseppe Rossini <giuseppe.rossini@arm.com>

4 files changed, 78 insertions, 15 deletions

diff --git a/tests/validation/CL/GEMMMatrixMultiplyNative.cpp b/tests/validation/CL/GEMMMatrixMultiplyNative.cpp
index b0d1fd2ad1..031b807ef8 100644
--- a/tests/validation/CL/GEMMMatrixMultiplyNative.cpp
+++ b/tests/validation/CL/GEMMMatrixMultiplyNative.cpp
@@ -22,6 +22,7 @@
  * SOFTWARE.
  */
 #include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyNativeKernel.h"
+#include "arm_compute/core/KernelDescriptors.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
@@ -128,7 +129,11 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
     rhs_info.n0         = n0_value;
     rhs_info.k0         = k0_value;
 
-    GEMMReshapeInfo gemm_info(M, N, K, false, false, 0, false, broadcast_bias);
+    GEMMKernelInfo kernel_info;
+    kernel_info.m              = M;
+    kernel_info.n              = N;
+    kernel_info.k              = K;
+    kernel_info.broadcast_bias = broadcast_bias;
 
     const TensorShape lhs_shape(K, M, b_value);
     const TensorShape rhs_shape(N, K, b_value);
@@ -137,7 +142,8 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
                                  broadcast_bias? 1 : b_value);
     const TensorShape dst_shape = compute_mm_shape(TensorInfo(lhs_shape, 1, data_type),
                                                    TensorInfo(rhs_shape, 1, data_type),
-                                                   gemm_info);
+                                                   kernel_info);
+
     // Create tensors
     CLTensor lhs  = create_tensor<CLTensor>(lhs_shape, data_type);
     CLTensor rhs  = create_tensor<CLTensor>(rhs_shape, data_type);
@@ -151,7 +157,7 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
 
     // Create and configure function
     CLGEMMMatrixMultiplyNative gemm;
-    gemm.configure(&lhs, &rhs, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, gemm_info);
+    gemm.configure(&lhs, &rhs, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, kernel_info);
 }
 } // namespace
 
diff --git a/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp b/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp
index 69e58303f3..bd70ddbe8c 100644
--- a/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp
+++ b/tests/validation/CL/GEMMMatrixMultiplyReshaped.cpp
@@ -24,6 +24,7 @@
 #include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyReshapedKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMReshapeLHSMatrixKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
+#include "arm_compute/core/KernelDescriptors.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
@@ -157,7 +158,13 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
     rhs_info.interleave = i_value_rhs;
     rhs_info.transpose  = true;
 
-    GEMMReshapeInfo gemm_info(M, N, K, false, false, 0, false, broadcast_bias);
+    GEMMKernelInfo kernel_info;
+    kernel_info.m                       = M;
+    kernel_info.n                       = N;
+    kernel_info.k                       = K;
+    kernel_info.depth_output_gemm3d     = 0;
+    kernel_info.reinterpret_input_as_3d = false;
+    kernel_info.broadcast_bias          = broadcast_bias;
 
     const TensorShape lhs_shape(K, M, b_value);
     const TensorShape lhs_shape_reshaped = compute_lhs_reshaped_shape(TensorInfo(lhs_shape, 1, data_type),
@@ -170,7 +177,7 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
 
     const TensorShape dst_shape = compute_mm_shape(TensorInfo(lhs_shape_reshaped, 1, data_type),
                                                    TensorInfo(rhs_shape_reshaped, 1, data_type),
-                                                   gemm_info);
+                                                   kernel_info);
 
     const TensorShape bias_shape(N,
                                  broadcast_bias? 1 : M,
@@ -189,7 +196,7 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
 
     // Create and configure function
     CLGEMMMatrixMultiplyReshaped gemm;
-    gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, gemm_info);
+    gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, kernel_info);
 }
 } // namespace
 
diff --git a/tests/validation/CL/GEMMMatrixMultiplyReshapedOnlyRHS.cpp b/tests/validation/CL/GEMMMatrixMultiplyReshapedOnlyRHS.cpp
index 133170e2d3..6c34fc870a 100644
--- a/tests/validation/CL/GEMMMatrixMultiplyReshapedOnlyRHS.cpp
+++ b/tests/validation/CL/GEMMMatrixMultiplyReshapedOnlyRHS.cpp
@@ -23,6 +23,7 @@
  */
 #include "arm_compute/core/CL/kernels/CLGEMMMatrixMultiplyReshapedOnlyRHSKernel.h"
 #include "arm_compute/core/CL/kernels/CLGEMMReshapeRHSMatrixKernel.h"
+#include "arm_compute/core/KernelDescriptors.h"
 #include "arm_compute/core/Types.h"
 #include "arm_compute/core/utils/misc/ShapeCalculator.h"
 #include "arm_compute/runtime/CL/CLTensor.h"
@@ -144,7 +145,13 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
     rhs_info.interleave = i_value_rhs;
     rhs_info.transpose  = t_value_rhs;
 
-    GEMMReshapeInfo gemm_info(M, N, K, false, false, 0, false, broadcast_bias);
+    GEMMKernelInfo kernel_info;
+    kernel_info.m                       = M;
+    kernel_info.n                       = N;
+    kernel_info.k                       = K;
+    kernel_info.depth_output_gemm3d     = 0;
+    kernel_info.reinterpret_input_as_3d = false;
+    kernel_info.broadcast_bias          = broadcast_bias;
 
     const TensorShape lhs_shape(K, M, b_value);
     const TensorShape rhs_shape(N, K, b_value);
@@ -153,7 +160,7 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
 
     const TensorShape dst_shape = compute_mm_shape(TensorInfo(lhs_shape, 1, data_type),
                                                    TensorInfo(rhs_shape_reshaped, 1, data_type),
-                                                   gemm_info);
+                                                   kernel_info);
 
     const TensorShape bias_shape(N,
                                  broadcast_bias? 1 : M,
@@ -172,7 +179,7 @@ void validate_configuration(unsigned int m_value, unsigned int n_value, unsigned
 
     // Create and configure function
     CLGEMMMatrixMultiplyReshapedOnlyRHS gemm;
-    gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, gemm_info);
+    gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, 1.0f, 1.0f, lhs_info, rhs_info, kernel_info);
 }
 } // namespace
 
diff --git a/tests/validation/fixtures/GEMMFixture.h b/tests/validation/fixtures/GEMMFixture.h
index b721d841f7..a225c71ab7 100644
--- a/tests/validation/fixtures/GEMMFixture.h
+++ b/tests/validation/fixtures/GEMMFixture.h
@@ -24,6 +24,7 @@
 #ifndef ARM_COMPUTE_TEST_GEMM_FIXTURE
 #define ARM_COMPUTE_TEST_GEMM_FIXTURE
 
+#include "arm_compute/core/KernelDescriptors.h"
 #include "arm_compute/core/TensorShape.h"
 #include "arm_compute/core/Types.h"
 #include "tests/AssetsLibrary.h"
@@ -210,6 +211,13 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = 0;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = broadcast_bias;
 
         // The output tensor will be auto-initialized within the function
 
@@ -219,7 +227,7 @@ protected:
         GEMMFunctionType       gemm;
         reshape_lhs.configure(&lhs, &lhs_reshaped, lhs_info);
         reshape_rhs.configure(&rhs, &rhs_reshaped, rhs_info);
-        gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, 0, false, broadcast_bias));
+        gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);
@@ -346,6 +354,13 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = m_h;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = true;
 
         // The output tensor will be auto-initialized within the function
 
@@ -355,7 +370,7 @@ protected:
         GEMMFunctionType       gemm;
         reshape_lhs.configure(&lhs, &lhs_reshaped, lhs_info);
         reshape_rhs.configure(&rhs, &rhs_reshaped, rhs_info);
-        gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, m_h, false, true));
+        gemm.configure(&lhs_reshaped, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);
@@ -479,6 +494,13 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = 0;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = broadcast_bias;
 
         // The output tensor will be auto-initialized within the function
 
@@ -486,7 +508,7 @@ protected:
         ReshapeRHSFunctionType reshape_rhs;
         GEMMFunctionType       gemm;
         reshape_rhs.configure(&rhs, &rhs_reshaped, rhs_info);
-        gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, 0, false, broadcast_bias));
+        gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);
@@ -606,6 +628,13 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = m_h;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = true;
 
         // The output tensor will be auto-initialized within the function
 
@@ -613,7 +642,7 @@ protected:
         ReshapeRHSFunctionType reshape_rhs;
         GEMMFunctionType       gemm;
         reshape_rhs.configure(&rhs, &rhs_reshaped, rhs_info);
-        gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, m_h, false, true));
+        gemm.configure(&lhs, &rhs_reshaped, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);
@@ -729,10 +758,17 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = 0;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = broadcast_bias;
 
         // Create and configure function
         GEMMFunctionType gemm;
-        gemm.configure(&lhs, &rhs, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, 0, false, broadcast_bias));
+        gemm.configure(&lhs, &rhs, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);
@@ -843,12 +879,19 @@ protected:
         const unsigned int M = lhs_shape[1];
         const unsigned int N = rhs_shape[0];
         const unsigned int K = lhs_shape[0];
+        GEMMKernelInfo     kernel_info;
+        kernel_info.m                       = M;
+        kernel_info.n                       = N;
+        kernel_info.k                       = K;
+        kernel_info.depth_output_gemm3d     = m_h;
+        kernel_info.reinterpret_input_as_3d = false;
+        kernel_info.broadcast_bias          = true;
 
         // The output tensor will be auto-initialized within the function
 
         // Create and configure function
         GEMMFunctionType gemm;
-        gemm.configure(&lhs, &rhs, &bias, &dst, alpha, beta, lhs_info, rhs_info, GEMMReshapeInfo(M, N, K, 1, 1, m_h, false, true));
+        gemm.configure(&lhs, &rhs, &bias, &dst, alpha, beta, lhs_info, rhs_info, kernel_info);
 
         ARM_COMPUTE_EXPECT(lhs.info()->is_resizable(), framework::LogLevel::ERRORS);
         ARM_COMPUTE_EXPECT(rhs.info()->is_resizable(), framework::LogLevel::ERRORS);