diff options
Diffstat (limited to 'src/BUILD.bazel')
| -rw-r--r-- | src/BUILD.bazel | 26 |
1 files changed, 25 insertions, 1 deletions
diff --git a/src/BUILD.bazel b/src/BUILD.bazel index 8b8842dfce..22521d1744 100644 --- a/src/BUILD.bazel +++ b/src/BUILD.bazel @@ -248,6 +248,7 @@ filegroup( "core/NEON/kernels/arm_conv/pooling/kernels/sve_u8q_nhwc_max_generic_depthfirst/generic.cpp", "core/NEON/kernels/arm_gemm/interleave_indirect-sve.cpp", "core/NEON/kernels/arm_gemm/kernels/sme2_gemv_bf16fp32_dot_16VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sme2_gemv_fp16_mla_16VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sme2_gemv_fp16fp32fp16_dot_16VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sme2_gemv_fp32_mla_16VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sme2_gemv_fp32bf16fp32_dot_16VL/generic.cpp", @@ -280,6 +281,7 @@ filegroup( "core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp32_mla_6x4VL/a64fx.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp32_mla_6x4VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_ffhybrid_fp32bf16fp32_mmla_4x6VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sve_ffinterleaved_bf16fp32_dot_8x3VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_ffinterleaved_bf16fp32_mmla_8x3VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_ffinterleaved_fp16_mla_8x3VL/a64fx.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_ffinterleaved_fp16_mla_8x3VL/generic.cpp", @@ -304,6 +306,9 @@ filegroup( "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_s8s32_mmla_6x4VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_dot_4x4VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8qa_mmla_4x4VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8s8qa_dot_4x4VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8s8qa_mmla_4x4VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8s8s32_mmla_6x4VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8u32_dot_6x4VL/a64fx.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8u32_dot_6x4VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_hybrid_u8u32_mmla_6x4VL/generic.cpp", @@ -317,6 +322,7 @@ filegroup( "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_s8s32_dot_8x3VL/a64fx.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_s8s32_dot_8x3VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_s8s32_mmla_8x3VL/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8s8s32_mmla_8x3VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/a64fx.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_dot_8x3VL/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/sve_interleaved_u8u32_mmla_8x3VL/generic.cpp", @@ -521,9 +527,11 @@ filegroup( "core/NEON/kernels/arm_gemm/gemm_fp32.cpp", "core/NEON/kernels/arm_gemm/gemm_int16.cpp", "core/NEON/kernels/arm_gemm/gemm_int8.cpp", + "core/NEON/kernels/arm_gemm/gemm_q8_mixed.cpp", "core/NEON/kernels/arm_gemm/gemm_qint8.cpp", "core/NEON/kernels/arm_gemm/gemm_quint8.cpp", "core/NEON/kernels/arm_gemm/gemm_s8fp32.cpp", + "core/NEON/kernels/arm_gemm/gemm_u8s8fp32.cpp", "core/NEON/kernels/arm_gemm/gemm_uint16.cpp", "core/NEON/kernels/arm_gemm/gemm_uint8.cpp", "core/NEON/kernels/arm_gemm/interleave-8way.cpp", @@ -574,6 +582,10 @@ filegroup( "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/a55.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_dot_4x16/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8qa_mmla_4x16/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8s8qa_dot_4x16/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8s8qa_mmla_4x16/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8s8s32_dot_6x16/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8s8s32_mmla_6x16/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/a55.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_dot_6x16/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_hybrid_u8u32_mmla_6x16/generic.cpp", @@ -583,6 +595,7 @@ filegroup( "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_bf16fp32_mmla_8x12/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12/a510.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_s8s32_mmla_8x12/generic.cpp", + "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_u8s8s32_mmla_8x12/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_u8u32_mmla_8x12/a510.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_interleaved_u8u32_mmla_8x12/generic.cpp", "core/NEON/kernels/arm_gemm/kernels/a64_sgemm_8x12/a53.cpp", @@ -753,6 +766,8 @@ filegroup( "cpu/kernels/depthwiseconv2d/generic/neon/impl.cpp", "cpu/kernels/depthwiseconv2d/generic/neon/qasymm8.cpp", "cpu/kernels/depthwiseconv2d/generic/neon/qasymm8_signed.cpp", + "cpu/kernels/dequantize/generic/neon/fp16.cpp", + "cpu/kernels/dequantize/generic/neon/fp32.cpp", "cpu/kernels/directconv2d/nchw/all.cpp", "cpu/kernels/directconv2d/nchw/fp16.cpp", "cpu/kernels/directconv2d/nhwc/neon/fp16.cpp", @@ -818,9 +833,17 @@ filegroup( "cpu/kernels/pool3d/neon/fp32.cpp", "cpu/kernels/pool3d/neon/qasymm8.cpp", "cpu/kernels/pool3d/neon/qasymm8_signed.cpp", + "cpu/kernels/quantize/generic/neon/fp16.cpp", + "cpu/kernels/quantize/generic/neon/fp32.cpp", + "cpu/kernels/quantize/generic/neon/integer.cpp", "cpu/kernels/range/generic/neon/fp16.cpp", "cpu/kernels/range/generic/neon/fp32.cpp", "cpu/kernels/range/generic/neon/integer.cpp", + "cpu/kernels/reduction_layer/generic/neon/fp16.cpp", + "cpu/kernels/reduction_layer/generic/neon/fp32.cpp", + "cpu/kernels/reduction_layer/generic/neon/integer.cpp", + "cpu/kernels/reduction_layer/generic/neon/qasymm8.cpp", + "cpu/kernels/reduction_layer/generic/neon/qasymm8_signed.cpp", "cpu/kernels/roialign/generic/neon/fp16.cpp", "cpu/kernels/roialign/generic/neon/fp32.cpp", "cpu/kernels/roialign/generic/neon/qasymm8.cpp", @@ -998,7 +1021,8 @@ filegroup( "runtime/SubTensor.cpp", "runtime/Tensor.cpp", "runtime/TensorAllocator.cpp", - "runtime/Utils.cpp"] + + "runtime/Utils.cpp", + "runtime/experimental/operators/CpuGemm.cpp"] + glob(["**/*.h", "**/*.hpp", "**/*.inl"]), |