Add tile declaration capability in KernelWriter
Resolves: COMPMID-5816
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Change-Id: Ibd885707a842550a058252f9d01e072129896055
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9901
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/compute_kernel_writer/src/cl/CLHelpers.cpp b/compute_kernel_writer/src/cl/CLHelpers.cpp
index d940a5a..5a3d0fa 100644
--- a/compute_kernel_writer/src/cl/CLHelpers.cpp
+++ b/compute_kernel_writer/src/cl/CLHelpers.cpp
@@ -88,4 +88,36 @@
return res;
}
-} // namespace ckw
\ No newline at end of file
+
+int32_t width_to_cl_vector_size(int32_t width)
+{
+ switch(width)
+ {
+ case 1:
+ return 1;
+ case 2:
+ return 2;
+ case 3:
+ return 3;
+ case 4:
+ return 4;
+ case 5:
+ case 6:
+ case 7:
+ case 8:
+ return 8;
+ case 9:
+ case 10:
+ case 11:
+ case 12:
+ case 13:
+ case 14:
+ case 15:
+ case 16:
+ return 16;
+ default:
+ CKW_THROW_MSG("Unsupported width to convert to OpenCL vector");
+ return 0;
+ }
+}
+} // namespace ckw
diff --git a/compute_kernel_writer/src/cl/CLHelpers.h b/compute_kernel_writer/src/cl/CLHelpers.h
index 915d59f..a9a84e2 100644
--- a/compute_kernel_writer/src/cl/CLHelpers.h
+++ b/compute_kernel_writer/src/cl/CLHelpers.h
@@ -21,10 +21,11 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef COMPUTE_KERNEL_WRITER_SRC_CL_CLHELPERS_H
-#define COMPUTE_KERNEL_WRITER_SRC_CL_CLHELPERS_H
+#ifndef CKW_SRC_CL_CLHELPERS_H
+#define CKW_SRC_CL_CLHELPERS_H
#include <string>
+#include <cstdint>
/** OpenCL specific helper functions */
namespace ckw
@@ -48,6 +49,15 @@
* @return the OpenCL datatype as a string
*/
std::string cl_get_variable_datatype_as_string(DataType dt, int32_t len);
+
+/** Helper function to return the OpenCL vector size that accommodate the the desired width
+ *
+ * @param[in] width The desired width
+ *
+ * @return the OpenCL vector size
+*/
+int32_t width_to_cl_vector_size(int32_t width);
+
} // namespace ckw
#endif /* COMPUTE_KERNEL_WRITER_SRC_CL_CLHELPERS_H */
diff --git a/compute_kernel_writer/src/cl/CLKernelWriter.cpp b/compute_kernel_writer/src/cl/CLKernelWriter.cpp
index 231d321..7faf2e6 100644
--- a/compute_kernel_writer/src/cl/CLKernelWriter.cpp
+++ b/compute_kernel_writer/src/cl/CLKernelWriter.cpp
@@ -24,14 +24,14 @@
#include "src/cl/CLKernelWriter.h"
#include "ckw/Error.h"
+#include "src/cl/CLTile.h"
+#include "src/cl/CLHelpers.h"
+#include <cstdint>
namespace ckw
{
-CLKernelWriter::CLKernelWriter()
-{
-}
-
+CLKernelWriter::CLKernelWriter() = default;
CLKernelWriter::~CLKernelWriter() = default;
std::unique_ptr<Kernel> CLKernelWriter::emit_kernel(const std::string &name)
@@ -61,4 +61,27 @@
return _body_source_code;
}
+ITileOperand &CLKernelWriter::declare_tile(const std::string &name, const TileInfo &tile_info)
+{
+ const std::string fullname = generate_full_name(name);
+
+ const int32_t height = tile_info.height();
+ const int32_t width = tile_info.width();
+ const DataType data_type = tile_info.data_type();
+
+ for(int32_t row = 0; row < height; ++row)
+ {
+ const std::string cl_type = cl_get_variable_datatype_as_string(data_type, width);
+ append_code(cl_type, " ", fullname, std::to_string(row), ";\n");
+ }
+
+ return add_operand(fullname, tile_info);
+}
+
+ITileOperand &CLKernelWriter::add_operand(const std::string &name, const TileInfo &tile_info)
+{
+ std::unique_ptr<ITileOperand> operand = std::make_unique<CLTile>(name, tile_info);
+ return KernelWriter::add_operand(operand);
+}
+
} // namespace ckw
diff --git a/compute_kernel_writer/src/cl/CLKernelWriter.h b/compute_kernel_writer/src/cl/CLKernelWriter.h
index e6f0641..d0c4b7c 100644
--- a/compute_kernel_writer/src/cl/CLKernelWriter.h
+++ b/compute_kernel_writer/src/cl/CLKernelWriter.h
@@ -57,6 +57,13 @@
std::unique_ptr<Kernel> emit_kernel(const std::string &name) override;
+ /** Declare a tile given name and tile information
+ *
+ * Similar to @ref KernelWriter::declare_tile()
+ */
+ ITileOperand &declare_tile(const ::std::string &name, const TileInfo &tile_info) override;
+
+
protected:
/** Append the specified code to the kernel body source code. */
template <typename T, typename... TArgs>
@@ -76,6 +83,9 @@
/** Get the current kernel body source code. */
const std::string &body_source_code() const;
+ /** Add a tile operand to the kernel and return it */
+ ITileOperand &add_operand(const std::string &code, const TileInfo &tile_info) override;
+
private:
/** This string contains the kernel body source code, not the full CL source code.
* The full source code will only be generated when the user calls @ref KernelWriter::emit_kernel.