commit 11c562c94fa6a0399aff798bfd970ed8c1942730
author Michele Di Giorgio <michele.digiorgio@arm.com> Wed Jun 10 16:34:50 2020 +0100
committer Michele Di Giorgio <michele.digiorgio@arm.com> Fri Jun 12 09:26:10 2020 +0000
tree 6eb2ca628659b1b579a2f99a6dd598d5e6ee5ea2
parent d004a7a707feab36e51f51cfc9eb2cb70729d5ad

COMPMID-3527: QLSTM should allow nullptr for optional projection bias

Change-Id: I5568ca1e9383f3960886f211d8320fc4525a8804
Signed-off-by: Michele Di Giorgio <michele.digiorgio@arm.com>
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/3326
Reviewed-by: James Conroy <james.conroy@arm.com>
Reviewed-by: Georgios Pinitas <georgios.pinitas@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>

src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp

diff --git a/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp b/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
index f8ee578..3878c76 100644
--- a/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
+++ b/src/core/NEON/kernels/NEArithmeticAdditionKernel.cpp
@@ -815,8 +815,12 @@
     ARM_COMPUTE_UNUSED(policy);
 
     ARM_COMPUTE_RETURN_ERROR_ON_CPU_F16_UNSUPPORTED(&input1);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S16, DataType::QSYMM16, DataType::F16, DataType::F32);
-    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input2, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED, DataType::S16, DataType::QSYMM16, DataType::F16, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input1, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
+                                                         DataType::S32, DataType::F32);
+    ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(&input2, 1, DataType::U8, DataType::QASYMM8, DataType::QASYMM8_SIGNED,
+                                                         DataType::S16, DataType::QSYMM16, DataType::F16,
+                                                         DataType::S32, DataType::F32);
 
     const TensorShape out_shape = TensorShape::broadcast_shape(input1.tensor_shape(), input2.tensor_shape());
 
@@ -834,6 +838,7 @@
             && !(input1.data_type() == DataType::U8 && input2.data_type() == DataType::S16 && output.data_type() == DataType::S16)
             && !(input1.data_type() == DataType::S16 && input2.data_type() == DataType::U8 && output.data_type() == DataType::S16)
             && !(input1.data_type() == DataType::S16 && input2.data_type() == DataType::S16 && output.data_type() == DataType::S16)
+            && !(input1.data_type() == DataType::S32 && input2.data_type() == DataType::S32 && output.data_type() == DataType::S32)
             && !(input1.data_type() == DataType::F32 && input2.data_type() == DataType::F32 && output.data_type() == DataType::F32)
             && !(input1.data_type() == DataType::F16 && input2.data_type() == DataType::F16 && output.data_type() == DataType::F16)
             && !(input1.data_type() == DataType::QASYMM8 && input2.data_type() == DataType::QASYMM8 && output.data_type() == DataType::QASYMM8)
@@ -862,6 +867,10 @@
         {
             set_format_if_unknown(output, Format::S16);
         }
+        if(input1.data_type() == DataType::S32 || input2.data_type() == DataType::S32)
+        {
+            set_format_if_unknown(output, Format::S32);
+        }
         else if(input1.data_type() == DataType::F16 || input2.data_type() == DataType::F16)
         {
             set_format_if_unknown(output, Format::F16);
@@ -926,6 +935,8 @@
         { "add_saturate_U8_U8_S16", &add_U8_U8_S16 },
         { "add_wrap_S16_S16_S16", &add_same<int16_t> },
         { "add_saturate_S16_S16_S16", &add_same<int16_t> },
+        { "add_wrap_S32_S32_S32", &add_same<int32_t> },
+        { "add_saturate_S32_S32_S32", &add_same<int32_t> },
         { "add_wrap_F32_F32_F32", &add_same<float> },
         { "add_saturate_F32_F32_F32", &add_same<float> },
 #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC