COMPMID-1836: Remove CLGEMMTranspose1xWKernel and replace with CLGEMMReshapeRHSMatrixKernel
Change-Id: Ic5a4f32657a155380684dcd4b44fbb608ef40cb4
Reviewed-on: https://review.mlplatform.org/418
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/core/CL/CLHelpers.cpp b/src/core/CL/CLHelpers.cpp
index 924fb1d..18ef185 100644
--- a/src/core/CL/CLHelpers.cpp
+++ b/src/core/CL/CLHelpers.cpp
@@ -148,7 +148,7 @@
const GPUTarget gpu_target = get_target_from_name(device_name);
// SW_WORKAROUND: Workaround for DDK revision r14p0.to enable cl_arm_integer_dot_product_int8
- std::set<GPUTarget> sw_workaround_issue = {GPUTarget::G76};
+ std::set<GPUTarget> sw_workaround_issue = { GPUTarget::G76 };
return (device_supports_extension(device, "cl_arm_integer_dot_product_int8") || sw_workaround_issue.count(gpu_target) != 0);
}
@@ -255,5 +255,4 @@
return 1;
}
}
-
} // namespace arm_compute
diff --git a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
index b2fb3e0..66fafe4 100644
--- a/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMLowpMatrixMultiplyKernel.cpp
@@ -71,11 +71,17 @@
}
else
{
- const int m = reshape_info.m();
- const int n = reshape_info.n();
- const int k = reshape_info.k();
- const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width();
- const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height();
+ GEMMRHSMatrixInfo rhs_info;
+ const int m = reshape_info.m();
+ const int n = reshape_info.n();
+ const int k = reshape_info.k();
+ const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width();
+ const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height();
+ rhs_info.n0 = 16 / input1->element_size();
+ rhs_info.k0 = 1;
+ rhs_info.h0 = mult_transpose1xW_width;
+ rhs_info.interleave = false;
+ rhs_info.transpose = false;
TensorShape tensor_shape0{ input0->tensor_shape() };
tensor_shape0.set(0, k);
@@ -89,7 +95,7 @@
const TensorInfo tensor_info1 = input1->clone()->set_tensor_shape(tensor_shape1);
const TensorInfo tensor_info_reshaped0 = input0->clone()->set_tensor_shape(compute_interleaved_shape(tensor_info0, mult_interleave4x4_height));
- const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_transpose1xW_with_element_size_shape(tensor_info1, mult_transpose1xW_width));
+ const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_rhs_reshaped_shape(tensor_info1, rhs_info));
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input0, &tensor_info_reshaped0);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, &tensor_info_reshaped1);
diff --git a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
index c9ed776..69455cf 100644
--- a/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
+++ b/src/core/CL/kernels/CLGEMMMatrixMultiplyKernel.cpp
@@ -66,11 +66,17 @@
}
else
{
- const int m = reshape_info.m();
- const int n = reshape_info.n();
- const int k = reshape_info.k();
- const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width();
- const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height();
+ GEMMRHSMatrixInfo rhs_info;
+ const int m = reshape_info.m();
+ const int n = reshape_info.n();
+ const int k = reshape_info.k();
+ const int mult_transpose1xW_width = reshape_info.mult_transpose1xW_width();
+ const int mult_interleave4x4_height = reshape_info.mult_interleave4x4_height();
+ rhs_info.n0 = 16 / input1->element_size();
+ rhs_info.k0 = 1;
+ rhs_info.h0 = mult_transpose1xW_width;
+ rhs_info.interleave = false;
+ rhs_info.transpose = false;
TensorShape tensor_shape0{ input0->tensor_shape() };
tensor_shape0.set(0, k);
@@ -84,7 +90,7 @@
const TensorInfo tensor_info1 = input1->clone()->set_tensor_shape(tensor_shape1);
const TensorInfo tensor_info_reshaped0 = input0->clone()->set_tensor_shape(compute_interleaved_shape(tensor_info0, mult_interleave4x4_height));
- const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_transpose1xW_with_element_size_shape(tensor_info1, mult_transpose1xW_width));
+ const TensorInfo tensor_info_reshaped1 = input1->clone()->set_tensor_shape(compute_rhs_reshaped_shape(tensor_info1, rhs_info));
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input0, &tensor_info_reshaped0);
ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input1, &tensor_info_reshaped1);