Connect CLMatMul function to quantized kernels and resolve NE BatchMatMul int_8 failures
* Adapt the CLMatMul function and ClMatMul operator to use quantized kernels.
* Add function-level tests.
Resolves: COMPMID-5929 and COMPMID-5811
Change-Id: I5348cdcf07b8074c138e04dfef0a73399377accd
Signed-off-by: Jakub Sujak <jakub.sujak@arm.com>
Signed-off-by: Omar Al Khatib <omar.alkhatib@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9575
Reviewed-by: Mohmun02 <MohammedSuhail.Munshi@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
diff --git a/src/gpu/cl/operators/ClMatMul.h b/src/gpu/cl/operators/ClMatMul.h
index 20beda9..3d98632 100644
--- a/src/gpu/cl/operators/ClMatMul.h
+++ b/src/gpu/cl/operators/ClMatMul.h
@@ -21,11 +21,13 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef ARM_COMPUTE_SRC_GPU_CL_OPERATORS_ClMatMul
-#define ARM_COMPUTE_SRC_GPU_CL_OPERATORS_ClMatMul
+#ifndef ACL_ARM_COMPUTE_SRC_GPU_CL_OPERATORS_CLMATMUL
+#define ACL_ARM_COMPUTE_SRC_GPU_CL_OPERATORS_CLMATMUL
#include "src/gpu/cl/IClOperator.h"
#include "src/gpu/cl/kernels/ClMatMulNativeKernel.h"
+#include "src/gpu/cl/kernels/ClMatMulLowpNativeKernel.h"
+
#include <memory>
namespace arm_compute
@@ -41,17 +43,20 @@
public:
/** Constructor */
ClMatMul();
- ~ClMatMul();
+ /** Default destructor */
+ ~ClMatMul() = default;
/** Initialise the kernel's inputs and output
*
* Valid data layouts:
* - All
*
* Valid data type configurations:
- * |lhs |rhs |output |
- * |:------------|:------------|:------------|
- * |F32 |F32 |F32 |
- * |F16 |F16 |F16 |
+ * |lhs |rhs |dst |
+ * |:--------------|:--------------|:--------------|
+ * |F32 |F32 |F32 |
+ * |F16 |F16 |F16 |
+ * |QASYMM8_SIGNED |QASYMM8_SIGNED |QASYMM8_SIGNED |
+ * |QASYMM8 |QASYMM8 |QASYMM8 |
*
* @note BatchMatMul: Batched Matrix Multiply - [A * B], Multiplies all slices (slice is an element of a batch) of Tensors A and B
* and stores the result in the dst tensor of the same batch size.
@@ -60,25 +65,28 @@
* @note All tensors must have the same data type.
*
* @param[in] compile_context The compile context to be used.
- * @param[in] lhs LHS input tensor info (Matrix A). Data types supported: F16/F32
- * @param[in] rhs RHS input tensor info (Matrix B). Data types supported: same as @p lhs.
- * @param[out] output Output tensor info. Data types supported: same as @p lhs
- * @param[in] matmul_info Attributes for MatMul
+ * @param[in] lhs Left-hand side tensor info. Data types supported: F16/F32/QASYMM8_SIGNED/QASYMM8.
+ * @param[in] rhs Right-hand side tensor info. Data types supported: same as @p lhs.
+ * @param[out] dst Output tensor to store the result of the batched matrix multiplication. Data types supported: same as @p lhs.
+ * @param[in] matmul_info Contains MatMul operation information described in @ref MatMulInfo.
*/
- void configure(const CLCompileContext &compile_context, ITensorInfo *lhs, ITensorInfo *rhs, ITensorInfo *output, const MatMulInfo &matmul_info);
+ void configure(const CLCompileContext &compile_context, ITensorInfo *lhs, ITensorInfo *rhs, ITensorInfo *dst, const MatMulInfo &matmul_info);
/** Static function to check if given info will lead to a valid configuration
*
* Similar to @ref ClMatMul::configure()
*
* @return a status
*/
- static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *output, const MatMulInfo &matmul_info);
+ static Status validate(const ITensorInfo *lhs, const ITensorInfo *rhs, const ITensorInfo *dst, const MatMulInfo &matmul_info);
// Inherited methods overridden:
void run(ITensorPack &tensors) override;
private:
- std::unique_ptr<kernels::ClMatMulNativeKernel> _native_matmul_kernel;
+ std::unique_ptr<kernels::ClMatMulNativeKernel> _matmul_native_kernel{nullptr};
+ std::unique_ptr<kernels::ClMatMulLowpNativeKernel> _matmul_lowp_native_kernel{nullptr};
+
+ bool _is_quantized{ false };
};
} // namespace opencl
} // namespace arm_compute
-#endif // ARM_COMPUTE_SRC_GPU_CL_OPERATORS_ClMatMul
+#endif /* ACL_ARM_COMPUTE_SRC_GPU_CL_OPERATORS_CLMATMUL */