diff options
Diffstat (limited to 'src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp')
-rw-r--r-- | src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp | 7 |
1 files changed, 4 insertions, 3 deletions
diff --git a/src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp b/src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp index 9e8907d60f..c725815859 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_qint8.cpp @@ -80,7 +80,7 @@ static const GemmImplementation<int8_t, int8_t, Requantize32> gemm_qint8_methods { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8q_mopa_1VLx4VL", - [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, + [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && args._maxthreads == 1 && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, [](const GemmArgs &args, const Requantize32 &) { const auto VL = sme::get_vector_length<int32_t>(); return args._Msize <= VL || (2*VL < args._Msize && args._Msize <= 3*VL); }, [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedPretransposedNoMergeQuantizedInline<cls_sme2_interleaved_nomerge_s8q_mopa_1VLx4VL, int8_t, int8_t>(args, qp); } @@ -88,7 +88,7 @@ static const GemmImplementation<int8_t, int8_t, Requantize32> gemm_qint8_methods { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8q_mopa_4VLx1VL", - [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, + [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && args._maxthreads == 1 && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, [](const GemmArgs &args, const Requantize32 &) { const auto VL = sme::get_vector_length<int32_t>(); return args._Nsize <= VL || (2*VL < args._Nsize && args._Nsize <= 3*VL); }, [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedPretransposedNoMergeQuantizedInline<cls_sme2_interleaved_nomerge_s8q_mopa_4VLx1VL, int8_t, int8_t>(args, qp); } @@ -96,7 +96,7 @@ static const GemmImplementation<int8_t, int8_t, Requantize32> gemm_qint8_methods { GemmMethod::GEMM_INTERLEAVED, "sme2_interleaved_nomerge_s8q_mopa_2VLx2VL", - [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, + [](const GemmArgs &args, const Requantize32 &qp) { return args._ci->has_sme2() && args._maxthreads == 1 && ((qp.per_channel_requant && (qp.per_channel_left_shifts == nullptr)) || (!qp.per_channel_requant && (qp.per_layer_left_shift == 0)));}, nullptr, [](const GemmArgs &args, const Requantize32 &qp) { return new GemmInterleavedPretransposedNoMergeQuantizedInline<cls_sme2_interleaved_nomerge_s8q_mopa_2VLx2VL, int8_t, int8_t>(args, qp); } }, @@ -265,6 +265,7 @@ const GemmImplementation<int8_t, int8_t, Requantize32> *gemm_implementation_list template UniqueGemmCommon<int8_t, int8_t> gemm<int8_t, int8_t, Requantize32>(const GemmArgs &args, const Requantize32 &os); template bool has_opt_gemm<int8_t, int8_t, Requantize32>(WeightFormat &weight_format, const GemmArgs &args, const Requantize32 &os); +template KernelDescription get_gemm_method<int8_t, int8_t, Requantize32>(const GemmArgs &args, const Requantize32 &os); template std::vector<KernelDescription> get_compatible_kernels<int8_t, int8_t, Requantize32>(const GemmArgs &args, const Requantize32 &os); } // namespace arm_gemm |