COMPMID-1086: Optimizing GCGEMM - Add helpers to get target GPU information

This patch moves some of the helper functions in CLHelpers.[h,cpp] to
GPUTarget.[h,cpp] in order to make them avaialable to both CL and GLES backends.

Change-Id: Ib89b05a2084b73eb643b32fac13b8367cc80b6a5
Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/128988
Tested-by: Jenkins <bsgcomp@arm.com>
Reviewed-by: Anthony Barbier <anthony.barbier@arm.com>

14 files changed, 228 insertions, 140 deletions

diff --git a/src/core/CL/CLHelpers.cpp b/src/core/CL/CLHelpers.cpp
index 9842d4d0ab..cda29d69d1 100644
--- a/src/core/CL/CLHelpers.cpp
+++ b/src/core/CL/CLHelpers.cpp
@@ -27,74 +27,8 @@
 #include "arm_compute/core/Log.h"
 #include "arm_compute/core/Types.h"
 
-#include <map>
-#include <regex>
 #include <vector>
 
-namespace
-{
-arm_compute::GPUTarget get_bifrost_target(const std::string &version)
-{
-    if(version == "G71")
-    {
-        return arm_compute::GPUTarget::G71;
-    }
-    else if(version == "G72")
-    {
-        return arm_compute::GPUTarget::G72;
-    }
-    else if(version == "G51")
-    {
-        return arm_compute::GPUTarget::G51;
-    }
-    else if(version == "G51BIG")
-    {
-        return arm_compute::GPUTarget::G51BIG;
-    }
-    else if(version == "G51LIT")
-    {
-        return arm_compute::GPUTarget::G51LIT;
-    }
-    else if(version == "TNOX")
-    {
-        return arm_compute::GPUTarget::TNOX;
-    }
-    else if(version == "TTRX")
-    {
-        return arm_compute::GPUTarget::TTRX;
-    }
-    else if(version == "TBOX")
-    {
-        return arm_compute::GPUTarget::TBOX;
-    }
-    else
-    {
-        return arm_compute::GPUTarget::BIFROST;
-    }
-}
-
-arm_compute::GPUTarget get_midgard_target(const std::string &version)
-{
-    if(version == "T600")
-    {
-        return arm_compute::GPUTarget::T600;
-    }
-    else if(version == "T700")
-    {
-        return arm_compute::GPUTarget::T700;
-    }
-    else if(version == "T800")
-    {
-        return arm_compute::GPUTarget::T800;
-    }
-    else
-    {
-        return arm_compute::GPUTarget::MIDGARD;
-    }
-}
-
-} // namespace
-
 namespace arm_compute
 {
 std::string get_cl_type_from_data_type(const DataType &dt)
@@ -177,61 +111,6 @@ std::string get_underlying_cl_type_from_data_type(const DataType &dt)
     }
 }
 
-const std::string &string_from_target(GPUTarget target)
-{
-    static std::map<GPUTarget, const std::string> gpu_target_map =
-    {
-        { GPUTarget::MIDGARD, "midgard" },
-        { GPUTarget::BIFROST, "bifrost" },
-        { GPUTarget::T600, "t600" },
-        { GPUTarget::T700, "t700" },
-        { GPUTarget::T800, "t800" },
-        { GPUTarget::G71, "g71" },
-        { GPUTarget::G72, "g72" },
-        { GPUTarget::G51, "g51" },
-        { GPUTarget::G51BIG, "g51big" },
-        { GPUTarget::G51LIT, "g51lit" },
-        { GPUTarget::TNOX, "tnox" },
-        { GPUTarget::TTRX, "ttrx" },
-        { GPUTarget::TBOX, "tbox" }
-    };
-
-    return gpu_target_map[target];
-}
-
-GPUTarget get_target_from_name(const std::string &device_name)
-{
-    std::regex  mali_regex(R"(Mali-(.*))");
-    std::smatch name_parts;
-    const bool  found_mali = std::regex_search(device_name, name_parts, mali_regex);
-
-    if(!found_mali)
-    {
-        ARM_COMPUTE_LOG_INFO_MSG_CORE("Can't find valid Mali GPU. Target is set to UNKNOWN.");
-        return GPUTarget::UNKNOWN;
-    }
-
-    const char         target  = name_parts.str(1)[0];
-    const std::string &version = name_parts.str(1);
-
-    std::regex future_regex(R"(.*X)");
-    const bool is_future_bifrost = std::regex_search(version, future_regex);
-
-    if(target == 'G' || is_future_bifrost)
-    {
-        return get_bifrost_target(version);
-    }
-    else if(target == 'T')
-    {
-        return get_midgard_target(version);
-    }
-    else
-    {
-        ARM_COMPUTE_LOG_INFO_MSG_CORE("Mali GPU unknown. Target is set to the default one. (BIFROST)");
-        return GPUTarget::BIFROST;
-    }
-}
-
 GPUTarget get_target_from_device(cl::Device &device)
 {
     // Query device name size
@@ -240,11 +119,6 @@ GPUTarget get_target_from_device(cl::Device &device)
     return get_target_from_name(device_name);
 }
 
-GPUTarget get_arch_from_target(GPUTarget target)
-{
-    return (target & GPUTarget::GPU_ARCH_MASK);
-}
-
 bool arm_non_uniform_workgroup_supported(const cl::Device &device)
 {
     return device_supports_extension(device, "cl_arm_non_uniform_work_group_size");
diff --git a/src/core/GLES_COMPUTE/GCHelpers.cpp b/src/core/GLES_COMPUTE/GCHelpers.cpp
new file mode 100644
index 0000000000..8970688ca7
--- /dev/null
+++ b/src/core/GLES_COMPUTE/GCHelpers.cpp
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/GLES_COMPUTE/GCHelpers.h"
+
+namespace arm_compute
+{
+GPUTarget get_target_from_device()
+{
+    const std::string device_name = reinterpret_cast<const char *>(glGetString(GL_RENDERER));
+
+    return get_target_from_name(device_name);
+}
+} // namespace arm_compute
diff --git a/src/core/GLES_COMPUTE/IGCKernel.cpp b/src/core/GLES_COMPUTE/IGCKernel.cpp
index 55b7f0da4a..ecd63b54a4 100644
--- a/src/core/GLES_COMPUTE/IGCKernel.cpp
+++ b/src/core/GLES_COMPUTE/IGCKernel.cpp
@@ -62,7 +62,7 @@ void arm_compute::enqueue(IGCKernel &kernel, const Window &window, const gles::N
 }
 
 IGCKernel::IGCKernel()
-    : _kernel(), _lws_hint(gles::NDRange(1U, 1U, 1U))
+    : _kernel(), _lws_hint(gles::NDRange(1U, 1U, 1U)), _target(GPUTarget::MIDGARD)
 {
 }
 
diff --git a/src/core/GLES_COMPUTE/OpenGLES.cpp b/src/core/GLES_COMPUTE/OpenGLES.cpp
index d2539d0775..e93b360125 100644
--- a/src/core/GLES_COMPUTE/OpenGLES.cpp
+++ b/src/core/GLES_COMPUTE/OpenGLES.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -66,7 +66,8 @@ using glMapBufferRange_func          = void *GL_APIENTRY (*)(GLenum target, GLin
 using glMemoryBarrier_func           = void GL_APIENTRY (*)(GLbitfield barriers);
 using glUniform1ui_func              = void GL_APIENTRY (*)(GLint location, GLuint v0);
 using glUnmapBuffer_func             = GLboolean GL_APIENTRY (*)(GLenum target);
-using glGetError_func                = GLenum              GL_APIENTRY (*)();
+using glGetError_func                = GLenum          GL_APIENTRY (*)();
+using glGetString_func               = const GLubyte * GL_APIENTRY (*)(GLenum name);
 using glGetActiveUniformBlockiv_func = void GL_APIENTRY (*)(GLuint program, GLuint uniformBlockIndex, GLenum pname, GLint *params);
 using glUniformBlockBinding_func     = void GL_APIENTRY (*)(GLuint program, GLuint uniformBlockIndex, GLuint uniformBlockBinding);
 using glGetUniformBlockIndex_func    = GLuint GL_APIENTRY (*)(GLuint program, const GLchar *uniformBlockName);
@@ -668,6 +669,19 @@ GLenum GL_APIENTRY glGetError(void)
     }
 }
 
+const GLubyte *GL_APIENTRY glGetString(GLenum name)
+{
+    auto func = GLESSymbols::get().glGetString;
+    if(func != nullptr)
+    {
+        return func(name);
+    }
+    else
+    {
+        return nullptr;
+    }
+}
+
 void GL_APIENTRY glGetActiveUniformBlockiv(GLuint program, GLuint uniformBlockIndex, GLenum pname, GLint *params)
 {
     auto func = GLESSymbols::get().glGetActiveUniformBlockiv;
diff --git a/src/core/GLES_COMPUTE/gl_entries.in b/src/core/GLES_COMPUTE/gl_entries.in
index 15ce8ee819..17e3aeead7 100644
--- a/src/core/GLES_COMPUTE/gl_entries.in
+++ b/src/core/GLES_COMPUTE/gl_entries.in
@@ -61,3 +61,4 @@ GL_ENTRY(glGenFramebuffers)
 GL_ENTRY(glDeleteFramebuffers)
 GL_ENTRY(glBindFramebuffer)
 GL_ENTRY(glFramebufferTexture2D)
+GL_ENTRY(glGetString)
diff --git a/src/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.cpp b/src/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.cpp
index dc86bfb2cc..171fbad702 100644
--- a/src/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCGEMMInterleave4x4Kernel.cpp
@@ -35,7 +35,6 @@
 #include "arm_compute/core/Window.h"
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 
 GCGEMMInterleave4x4Kernel::GCGEMMInterleave4x4Kernel()
     : _input(nullptr), _output(nullptr)
diff --git a/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.cpp
index 43846dcf32..1a68a62ea5 100644
--- a/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixAdditionKernel.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -34,7 +34,6 @@
 #include "arm_compute/core/Window.h"
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 
 GCGEMMMatrixAdditionKernel::GCGEMMMatrixAdditionKernel()
     : _input(nullptr), _output(nullptr)
diff --git a/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.cpp
index b4bb5470ad..2bd769cac4 100644
--- a/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCGEMMMatrixMultiplyKernel.cpp
@@ -42,7 +42,6 @@
 #include <string>
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 using namespace arm_compute::misc::shape_calculator;
 
 namespace
@@ -195,10 +194,13 @@ void GCGEMMMatrixMultiplyKernel::configure(const IGCTensor *input0, const IGCTen
     _input1 = input1;
     _output = output;
 
+    // Get target architecture
+    GPUTarget gpu_target = get_target();
+
     ElementsProcessed num_elements_processed{};
 
     // Configure kernel window
-    auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info(), is_interleaved_transposed, reshape_info, GPUTarget::UNKNOWN, num_elements_processed);
+    auto win_config = validate_and_configure_window(input0->info(), input1->info(), output->info(), is_interleaved_transposed, reshape_info, gpu_target, num_elements_processed);
     ARM_COMPUTE_ERROR_THROW_ON(win_config.first);
     IGCKernel::configure(win_config.second);
 
diff --git a/src/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.cpp
index 21946b7f8d..f0057df769 100644
--- a/src/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCTensorShiftKernel.cpp
@@ -36,7 +36,6 @@
 #include "support/ToolchainSupport.h"
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 
 GCTensorShiftKernel::GCTensorShiftKernel()
     : _input(nullptr), _lws(gles::NDRange(1U, 1U, 1U)), _left_padding(0)
diff --git a/src/core/GLES_COMPUTE/kernels/GCWeightsReshapeKernel.cpp b/src/core/GLES_COMPUTE/kernels/GCWeightsReshapeKernel.cpp
index 55bf9b754b..ccbfaf8410 100644
--- a/src/core/GLES_COMPUTE/kernels/GCWeightsReshapeKernel.cpp
+++ b/src/core/GLES_COMPUTE/kernels/GCWeightsReshapeKernel.cpp
@@ -36,7 +36,6 @@
 #include "arm_compute/core/GLES_COMPUTE/GCHelpers.h"
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 using namespace arm_compute::misc::shape_calculator;
 
 GCWeightsReshapeKernel::GCWeightsReshapeKernel()
diff --git a/src/core/GPUTarget.cpp b/src/core/GPUTarget.cpp
new file mode 100644
index 0000000000..575d858589
--- /dev/null
+++ b/src/core/GPUTarget.cpp
@@ -0,0 +1,154 @@
+/*
+ * Copyright (c) 2018 ARM Limited.
+ *
+ * SPDX-License-Identifier: MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to
+ * deal in the Software without restriction, including without limitation the
+ * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
+ * sell copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in all
+ * copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include "arm_compute/core/GPUTarget.h"
+#include "arm_compute/core/Log.h"
+
+#include <map>
+#include <regex>
+
+namespace
+{
+arm_compute::GPUTarget get_bifrost_target(const std::string &version)
+{
+    if(version == "G71")
+    {
+        return arm_compute::GPUTarget::G71;
+    }
+    else if(version == "G72")
+    {
+        return arm_compute::GPUTarget::G72;
+    }
+    else if(version == "G51")
+    {
+        return arm_compute::GPUTarget::G51;
+    }
+    else if(version == "G51BIG")
+    {
+        return arm_compute::GPUTarget::G51BIG;
+    }
+    else if(version == "G51LIT")
+    {
+        return arm_compute::GPUTarget::G51LIT;
+    }
+    else if(version == "TNOX")
+    {
+        return arm_compute::GPUTarget::TNOX;
+    }
+    else if(version == "TTRX")
+    {
+        return arm_compute::GPUTarget::TTRX;
+    }
+    else if(version == "TBOX")
+    {
+        return arm_compute::GPUTarget::TBOX;
+    }
+    else
+    {
+        return arm_compute::GPUTarget::BIFROST;
+    }
+}
+
+arm_compute::GPUTarget get_midgard_target(const std::string &version)
+{
+    if(version == "T600")
+    {
+        return arm_compute::GPUTarget::T600;
+    }
+    else if(version == "T700")
+    {
+        return arm_compute::GPUTarget::T700;
+    }
+    else if(version == "T800")
+    {
+        return arm_compute::GPUTarget::T800;
+    }
+    else
+    {
+        return arm_compute::GPUTarget::MIDGARD;
+    }
+}
+} // namespace
+
+namespace arm_compute
+{
+const std::string &string_from_target(GPUTarget target)
+{
+    static std::map<GPUTarget, const std::string> gpu_target_map =
+    {
+        { GPUTarget::MIDGARD, "midgard" },
+        { GPUTarget::BIFROST, "bifrost" },
+        { GPUTarget::T600, "t600" },
+        { GPUTarget::T700, "t700" },
+        { GPUTarget::T800, "t800" },
+        { GPUTarget::G71, "g71" },
+        { GPUTarget::G72, "g72" },
+        { GPUTarget::G51, "g51" },
+        { GPUTarget::G51BIG, "g51big" },
+        { GPUTarget::G51LIT, "g51lit" },
+        { GPUTarget::TNOX, "tnox" },
+        { GPUTarget::TTRX, "ttrx" },
+        { GPUTarget::TBOX, "tbox" }
+    };
+
+    return gpu_target_map[target];
+}
+
+GPUTarget get_target_from_name(const std::string &device_name)
+{
+    std::regex  mali_regex(R"(Mali-(.*))");
+    std::smatch name_parts;
+    const bool  found_mali = std::regex_search(device_name, name_parts, mali_regex);
+
+    if(!found_mali)
+    {
+        ARM_COMPUTE_LOG_INFO_MSG_CORE("Can't find valid Mali GPU. Target is set to UNKNOWN.");
+        return GPUTarget::UNKNOWN;
+    }
+
+    const char         target  = name_parts.str(1)[0];
+    const std::string &version = name_parts.str(1);
+
+    std::regex future_regex(R"(.*X)");
+    const bool is_future_bifrost = std::regex_search(version, future_regex);
+
+    if(target == 'G' || is_future_bifrost)
+    {
+        return get_bifrost_target(version);
+    }
+    else if(target == 'T')
+    {
+        return get_midgard_target(version);
+    }
+    else
+    {
+        ARM_COMPUTE_LOG_INFO_MSG_CORE("Mali GPU unknown. Target is set to the default one. (BIFROST)");
+        return GPUTarget::BIFROST;
+    }
+}
+
+GPUTarget get_arch_from_target(GPUTarget target)
+{
+    return (target & GPUTarget::GPU_ARCH_MASK);
+}
+} // namespace arm_compute
diff --git a/src/runtime/GLES_COMPUTE/GCScheduler.cpp b/src/runtime/GLES_COMPUTE/GCScheduler.cpp
index fcc855957f..f7812730fc 100644
--- a/src/runtime/GLES_COMPUTE/GCScheduler.cpp
+++ b/src/runtime/GLES_COMPUTE/GCScheduler.cpp
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2017 ARM Limited.
+ * Copyright (c) 2017-2018 ARM Limited.
  *
  * SPDX-License-Identifier: MIT
  *
@@ -24,6 +24,7 @@
 
 #include "arm_compute/runtime/GLES_COMPUTE/GCScheduler.h"
 
+#include "arm_compute/core/GLES_COMPUTE/GCHelpers.h"
 #include "arm_compute/core/GLES_COMPUTE/GCKernelLibrary.h"
 
 using namespace arm_compute;
@@ -31,7 +32,7 @@ using namespace arm_compute;
 std::once_flag GCScheduler::_initialize_symbols;
 
 GCScheduler::GCScheduler()
-    : _display(EGL_NO_DISPLAY), _context(EGL_NO_CONTEXT)
+    : _display(EGL_NO_DISPLAY), _context(EGL_NO_CONTEXT), _target(GPUTarget::MIDGARD)
 {
 }
 
@@ -48,11 +49,13 @@ void GCScheduler::default_init()
 {
     setup_context();
 
-    GCKernelLibrary::get().init("./cs_shaders/", _display, _context);
+    init(_display, _context);
 }
 
 void GCScheduler::init(EGLDisplay dpy, EGLContext ctx)
 {
+    _target = get_target_from_device();
+
     GCKernelLibrary::get().init("./cs_shaders/", dpy, ctx);
 }
 
diff --git a/src/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.cpp b/src/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.cpp
index dc73eb85e6..d1ef87d32c 100644
--- a/src/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.cpp
+++ b/src/runtime/GLES_COMPUTE/functions/GCConvolutionLayer.cpp
@@ -109,6 +109,10 @@ void GCConvolutionLayer::configure(const IGCTensor *input, const IGCTensor *weig
 
     const DataType dt = input->info()->data_type();
 
+    // Set the GPU target for im2col and col2im
+    _input_im2col_kernel.set_target(GCScheduler::get().get_target());
+    _output_col2im_kernel.set_target(GCScheduler::get().get_target());
+
     const bool       append_bias   = (biases != nullptr);
     const unsigned   bias_element  = (append_bias) ? 1 : 0;
     const IGCTensor *biases_to_use = (append_bias) ? biases : nullptr;
diff --git a/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp b/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
index 0a75a38c50..79f8f71713 100644
--- a/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
+++ b/src/runtime/GLES_COMPUTE/functions/GCGEMM.cpp
@@ -38,7 +38,6 @@
 #include "arm_compute/runtime/ITensorAllocator.h"
 
 using namespace arm_compute;
-using namespace arm_compute::gles_compute;
 
 namespace
 {
@@ -92,6 +91,13 @@ void GCGEMM::configure(const IGCTensor *a, const IGCTensor *b, const IGCTensor *
     const IGCTensor *matrix_a = a;
     const IGCTensor *matrix_b = b;
 
+    // Get the GPU target
+    const GPUTarget gpu_target = GCScheduler::get().get_target();
+
+    // Set the target for the kernels
+    _interleave_kernel.set_target(gpu_target);
+    _mm_kernel.set_target(gpu_target);
+
     // Arguments used by GEMMReshapeInfo
     // If we pass the matrix A and matrix B reshaped to GCGEMMMatrixMultiplyKernel, we need to pass m, n, k, mult_transpose1xW_width and mult_interleave4x4_height to GCGEMMReshapeInfo
     // in order to know how the matrices have been reshaped