diff options
| author | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-05-21 13:32:43 +0100 |
|---|---|---|
| committer | Georgios Pinitas <georgios.pinitas@arm.com> | 2019-06-03 14:51:29 +0000 |
| commit | 4c5469b192665c94118a8a558787cb9cec2d0765 (patch) | |
| tree | 168aa969de8243bdbb1f25247dd9f54d037ae32c /src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp | |
| parent | 43a129e94df41f9ac8bc78b702da5a387ada0494 (diff) | |
| download | ComputeLibrary-4c5469b192665c94118a8a558787cb9cec2d0765.tar.gz | |
COMPMID-2225: Add interface support for new quantized data types.
Add support for:
-QSYMM8, 8-bit quantized symmetric
-QSYMM8_PER_CHANNEL, 8-bit quantized symmetric with per channel quantization
Change-Id: I00c4ff98e44af37419470af61419ee95d0de2463
Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Reviewed-on: https://review.mlplatform.org/c/1236
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: Gian Marco Iodice <gianmarco.iodice@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>
Diffstat (limited to 'src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp')
| -rw-r--r-- | src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp | 10 |
1 files changed, 7 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp b/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp index 45e1562d8d..8874b52e19 100644 --- a/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp +++ b/src/core/NEON/kernels/NEArithmeticSubtractionKernel.cpp @@ -87,10 +87,14 @@ void sub_saturate_QAYSMM8_QAYSMM8_QAYSMM8(const ITensor *in1, const ITensor *in2 Iterator input2(in2, window.broadcast_if_dimension_le_one(in2->info()->tensor_shape())); Iterator output(out, window); + const UniformQuantizationInfo iq1_info = in1->info()->quantization_info().uniform(); + const UniformQuantizationInfo iq2_info = in2->info()->quantization_info().uniform(); + const UniformQuantizationInfo oq_info = out->info()->quantization_info().uniform(); + execute_window_loop(window, [&](const Coordinates &) { - const float32x4x4_t ta1 = vdequantize(vld1q_u8(reinterpret_cast<const qasymm8_t *>(input1.ptr())), in1->info()->quantization_info()); - const float32x4x4_t ta2 = vdequantize(vld1q_u8(reinterpret_cast<const qasymm8_t *>(input2.ptr())), in2->info()->quantization_info()); + const float32x4x4_t ta1 = vdequantize(vld1q_u8(reinterpret_cast<const qasymm8_t *>(input1.ptr())), iq1_info); + const float32x4x4_t ta2 = vdequantize(vld1q_u8(reinterpret_cast<const qasymm8_t *>(input2.ptr())), iq2_info); const float32x4x4_t ta3 = { @@ -102,7 +106,7 @@ void sub_saturate_QAYSMM8_QAYSMM8_QAYSMM8(const ITensor *in1, const ITensor *in2 } }; - const uint8x16_t result = vquantize(ta3, out->info()->quantization_info()); + const uint8x16_t result = vquantize(ta3, oq_info); vst1q_u8(reinterpret_cast<qasymm8_t *>(output.ptr()), result); }, |