Add new operator AddMulAdd for Neon™ backend for Float/Quantized types
This is a fused operator that merges Add + Mul + Add [+ Relu-based-Activation] layers and have an intermediate output after the first Add. It's supported for FP16/32/QASYMM8/QASYMM8_SIGNED data types.
The subsequent Add and Mul are intended for scaling and the coefficients only have one dimension (per channel).
The inputs are
- input1 : nD tensor [X, Y, Z, W, ..]
- input2 : nD tensor [X, Y, Z, W, ..]
- add_coef : 1D tensor [X]
- mul_coef : 1D tensor [X]
The outputs are
- out1 : nD tensor (intermediate output) [X, Y, Z, W, ..]
- out2 : nD tensor (final output) [X, Y, Z, W, ..]
The operation can be summarized as follows:
out1 <- input1 + input2
out2 <- Act(out1 * mul_coef + add_coef)
The activation function can be Identity, Relu, Bounded Relu or Lower/Upper Bounded Relu. The intermediate output can be skipped by providing a nullptr.
The reason of providing this operator is to be able to fuse in case of Residual network patterns and save computations by reducing memory back and forward.
Resolves: COMPMID-5463
Signed-off-by: Gunes Bayir <gunes.bayir@arm.com>
Change-Id: I8ef577aa623b036e9a9f655cc088493fd19a6109
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/9055
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Jakub Sujak <jakub.sujak@arm.com>
Reviewed-by: Viet-Hoa Do <viet-hoa.do@arm.com>
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Benchmark: Arm Jenkins <bsgcomp@arm.com>
diff --git a/Android.bp b/Android.bp
index 23b19c2..3af4322 100644
--- a/Android.bp
+++ b/Android.bp
@@ -394,6 +394,7 @@
"src/cpu/CpuTensor.cpp",
"src/cpu/kernels/CpuActivationKernel.cpp",
"src/cpu/kernels/CpuAddKernel.cpp",
+ "src/cpu/kernels/CpuAddMulAddKernel.cpp",
"src/cpu/kernels/CpuCastKernel.cpp",
"src/cpu/kernels/CpuCol2ImKernel.cpp",
"src/cpu/kernels/CpuConcatenateBatchKernel.cpp",
@@ -451,6 +452,10 @@
"src/cpu/kernels/add/generic/neon/qasymm8.cpp",
"src/cpu/kernels/add/generic/neon/qasymm8_signed.cpp",
"src/cpu/kernels/add/generic/neon/qsymm16.cpp",
+ "src/cpu/kernels/addmuladd/generic/neon/fp16.cpp",
+ "src/cpu/kernels/addmuladd/generic/neon/fp32.cpp",
+ "src/cpu/kernels/addmuladd/generic/neon/qasymm8.cpp",
+ "src/cpu/kernels/addmuladd/generic/neon/qasymm8_signed.cpp",
"src/cpu/kernels/boundingboxtransform/generic/neon/fp16.cpp",
"src/cpu/kernels/boundingboxtransform/generic/neon/fp32.cpp",
"src/cpu/kernels/boundingboxtransform/generic/neon/impl.cpp",
@@ -551,6 +556,7 @@
"src/cpu/kernels/sub/neon/qsymm16.cpp",
"src/cpu/operators/CpuActivation.cpp",
"src/cpu/operators/CpuAdd.cpp",
+ "src/cpu/operators/CpuAddMulAdd.cpp",
"src/cpu/operators/CpuCast.cpp",
"src/cpu/operators/CpuConcatenate.cpp",
"src/cpu/operators/CpuConv2d.cpp",
@@ -873,6 +879,7 @@
"src/runtime/NEON/INESimpleFunction.cpp",
"src/runtime/NEON/INESimpleFunctionNoBorder.cpp",
"src/runtime/NEON/functions/NEActivationLayer.cpp",
+ "src/runtime/NEON/functions/NEAddMulAdd.cpp",
"src/runtime/NEON/functions/NEArgMinMaxLayer.cpp",
"src/runtime/NEON/functions/NEArithmeticAddition.cpp",
"src/runtime/NEON/functions/NEArithmeticSubtraction.cpp",