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/arm_compute/runtime/CL/functions/CLMatMul.h b/arm_compute/runtime/CL/functions/CLMatMul.h
index 712bac0..2af9a4a 100644
--- a/arm_compute/runtime/CL/functions/CLMatMul.h
+++ b/arm_compute/runtime/CL/functions/CLMatMul.h
@@ -21,11 +21,12 @@
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
-#ifndef ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
-#define ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
+#ifndef ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
+#define ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL
#include "arm_compute/runtime/IFunction.h"
#include <memory>
+
namespace arm_compute
{
// Forward declarations for used types instead of including their header, that could minimize compile time
@@ -64,10 +65,12 @@
* - 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.
@@ -76,18 +79,18 @@
* @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 (Matrix or Vector A). Data types supported: F16/F32
- * @param[in] rhs RHS input tensor (Matrix B). Data type supported: same as @p lhs.
- * @param[out] output Output tensor. Data type supported: same as @p lhs.
- * @param[in] matmul_info Attributes for MatMul
+ * @param[in] lhs Left-hand side tensor info containing the input activations as Matrix A. Data types supported: F16/F32/QASYMM8_SIGNED/QASYMM8.
+ * @param[in] rhs Right-hand side tensor info containing the input weights as Matrix B. 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.
* @param[in] settings Class containing flags for function level settings
*/
- void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *output, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
+ void configure(const CLCompileContext &compile_context, ICLTensor *rhs, ICLTensor *lhs, ICLTensor *dst, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
/** Initialise the kernel's inputs and output
*
* Similar to @ref CLMatMul::configure()
*/
- void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *output, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
+ void configure(ICLTensor *lhs, ICLTensor *rhs, ICLTensor *dst, const MatMulInfo &matmul_info, const GpuMatMulSettings &settings = GpuMatMulSettings{});
/** Static function to check if given info will lead to a valid configuration of @ref CLMatMul.
*
* Similar to @ref CLMatMul::configure()
@@ -104,4 +107,4 @@
};
} // namespace arm_compute
-#endif /* ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL */
+#endif /* ACL_ARM_COMPUTE_RUNTIME_CL_FUNCTIONS_CLMATMUL */