diff options
Diffstat (limited to 'src/core/NEON')
26 files changed, 64 insertions, 35 deletions
diff --git a/src/core/NEON/kernels/NEBoundingBoxTransformKernel.cpp b/src/core/NEON/kernels/NEBoundingBoxTransformKernel.cpp index 5a40b99609..e195c96722 100644 --- a/src/core/NEON/kernels/NEBoundingBoxTransformKernel.cpp +++ b/src/core/NEON/kernels/NEBoundingBoxTransformKernel.cpp @@ -79,6 +79,7 @@ Status validate_arguments(const ITensorInfo *boxes, const ITensorInfo *pred_boxe std::pair<Status, Window> validate_and_configure_window(ITensorInfo *boxes, ITensorInfo *pred_boxes, ITensorInfo *deltas, const BoundingBoxTransformInfo &bb_info) { + ARM_COMPUTE_UNUSED(bb_info); ARM_COMPUTE_ERROR_ON_NULLPTR(boxes, pred_boxes); auto_init_if_empty(*pred_boxes, deltas->clone()->set_data_type(boxes->data_type()).set_quantization_info(boxes->quantization_info())); @@ -130,7 +131,7 @@ Status NEBoundingBoxTransformKernel::validate(const ITensorInfo *boxes, const IT } template <> -void NEBoundingBoxTransformKernel::internal_run<uint16_t>(const Window &window, const ThreadInfo &info) +void NEBoundingBoxTransformKernel::internal_run<uint16_t>(const Window &window) { const size_t num_classes = _deltas->info()->tensor_shape()[0] >> 2; const size_t deltas_width = _deltas->info()->tensor_shape()[0]; @@ -187,7 +188,7 @@ void NEBoundingBoxTransformKernel::internal_run<uint16_t>(const Window &window, } template <typename T> -void NEBoundingBoxTransformKernel::internal_run(const Window &window, const ThreadInfo &info) +void NEBoundingBoxTransformKernel::internal_run(const Window &window) { const size_t num_classes = _deltas->info()->tensor_shape()[0] >> 2; const size_t deltas_width = _deltas->info()->tensor_shape()[0]; @@ -242,24 +243,25 @@ void NEBoundingBoxTransformKernel::internal_run(const Window &window, const Thre void NEBoundingBoxTransformKernel::run(const Window &window, const ThreadInfo &info) { + ARM_COMPUTE_UNUSED(info); ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); switch(_boxes->info()->data_type()) { case DataType::F32: { - internal_run<float>(window, info); + internal_run<float>(window); break; } case DataType::QASYMM16: { - internal_run<uint16_t>(window, info); + internal_run<uint16_t>(window); break; } #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC case DataType::F16: { - internal_run<float16_t>(window, info); + internal_run<float16_t>(window); break; } #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC diff --git a/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp b/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp index 10bbe8324c..d00c5009d2 100644 --- a/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDepthConvertLayerKernel.cpp @@ -49,7 +49,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, C ARM_COMPUTE_RETURN_ERROR_ON(shift >= 8); ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::QASYMM8 && (output->data_type() != DataType::S16 && output->data_type() != DataType::U16 - && output->data_type() != DataType::S32 && output->data_type() != DataType::F16 && output->data_type() != DataType::F32), + && output->data_type() != DataType::S32 && output->data_type() != DataType::F16 && output->data_type() != DataType::F32), "Only data_types supported [in] QASYMM8 -> [out] U16, S16, S32, F16, F32"); ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::U8 && (output->data_type() != DataType::S16 && output->data_type() != DataType::U16 @@ -62,13 +62,16 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, C ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::S16 && (output->data_type() != DataType::U8 && output->data_type() != DataType::S32), "Only data_types supported [in] S16 -> [out] U8, S32"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::F16 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::U8 && output->data_type() != DataType::F32 && output->data_type() != DataType::S32), + ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::F16 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::U8 && output->data_type() != DataType::F32 + && output->data_type() != DataType::S32), "Only data_types supported [in] F16 -> [out] QASYMM8, F32, S32, U8"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::F32 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::F16 && output->data_type() != DataType::S32 && output->data_type() != DataType::U8), + ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::F32 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::F16 && output->data_type() != DataType::S32 + && output->data_type() != DataType::U8), "Only data_types supported [in] F32 -> [out] QASYMM8, F16, S32, U8"); - ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::S32 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::F16 && output->data_type() != DataType::F32 && output->data_type() != DataType::U8), + ARM_COMPUTE_RETURN_ERROR_ON_MSG(input->data_type() == DataType::S32 && (output->data_type() != DataType::QASYMM8 && output->data_type() != DataType::F16 && output->data_type() != DataType::F32 + && output->data_type() != DataType::U8), "Only data_types supported [in] S32 -> [out] QASYMM8, F16, F32, U8"); // Validate in case of configured output @@ -427,7 +430,7 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info const float16x8_t scale = vdupq_n_f16(1 << _shift); /* Up-conversion F16 -> U8 */ - execute_window_loop(window, [&](const Coordinates & id) + execute_window_loop(window, [&](const Coordinates &) { const float16x8x2_t texels = { @@ -447,7 +450,7 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info const float32x4_t scale = vdupq_n_f32(1 << _shift); /* Up-conversion F16 -> F32 */ - execute_window_loop(window, [&](const Coordinates & id) + execute_window_loop(window, [&](const Coordinates &) { const float16x8x2_t texels = { @@ -470,7 +473,7 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info const float32x4_t scale = vdupq_n_f32(1 << _shift); /* Up-conversion F16 -> S32 */ - execute_window_loop(window, [&](const Coordinates & id) + execute_window_loop(window, [&](const Coordinates &) { const float16x8x2_t texels = { @@ -565,13 +568,12 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info }; vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[0])), vqmovun_s32(vcvtq_s32_f32(texels.val[1]))))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr())+8, vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[2])), vqmovun_s32(vcvtq_s32_f32(texels.val[3]))))); + vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(vcvtq_s32_f32(texels.val[2])), vqmovun_s32(vcvtq_s32_f32(texels.val[3]))))); }, input, output); break; } - default: ARM_COMPUTE_ERROR("Output data type not supported"); } @@ -650,7 +652,7 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info } }; vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[0]), vqmovun_s32(texels.val[1])))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr())+8, vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[2]), vqmovun_s32(texels.val[3])))); + vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vqmovn_u16(vcombine_u16(vqmovun_s32(texels.val[2]), vqmovun_s32(texels.val[3])))); }, input, output); } @@ -668,8 +670,8 @@ void NEDepthConvertLayerKernel::run(const Window &window, const ThreadInfo &info } }; - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[0])),vmovn_u32(vreinterpretq_u32_s32(texels.val[1]))))); - vst1_u8(reinterpret_cast<uint8_t *>(output.ptr())+8, vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[2])),vmovn_u32(vreinterpretq_u32_s32(texels.val[3]))))); + vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()), vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[0])), vmovn_u32(vreinterpretq_u32_s32(texels.val[1]))))); + vst1_u8(reinterpret_cast<uint8_t *>(output.ptr()) + 8, vmovn_u16(vcombine_u16(vmovn_u32(vreinterpretq_u32_s32(texels.val[2])), vmovn_u32(vreinterpretq_u32_s32(texels.val[3]))))); }, input, output); } diff --git a/src/core/NEON/kernels/NEDequantizationLayerKernel.cpp b/src/core/NEON/kernels/NEDequantizationLayerKernel.cpp index 49de3ec8b3..5abd6a122d 100644 --- a/src/core/NEON/kernels/NEDequantizationLayerKernel.cpp +++ b/src/core/NEON/kernels/NEDequantizationLayerKernel.cpp @@ -219,7 +219,7 @@ void run_dequantization_qasymm8_per_channel_nhwc(const ITensor *input, ITensor * Iterator in(input, win); Iterator out(output, win); - execute_window_loop(win, [&](const Coordinates & id) + execute_window_loop(win, [&](const Coordinates &) { const auto in_ptr = reinterpret_cast<const uint8_t *>(in.ptr()); const auto out_ptr = reinterpret_cast<T *>(out.ptr()); diff --git a/src/core/NEON/kernels/NEGenerateProposalsLayerKernel.cpp b/src/core/NEON/kernels/NEGenerateProposalsLayerKernel.cpp index ba5ca78955..940ccabe65 100644 --- a/src/core/NEON/kernels/NEGenerateProposalsLayerKernel.cpp +++ b/src/core/NEON/kernels/NEGenerateProposalsLayerKernel.cpp @@ -100,7 +100,7 @@ Status NEComputeAllAnchorsKernel::validate(const ITensorInfo *anchors, const ITe } template <> -void NEComputeAllAnchorsKernel::internal_run<int16_t>(const Window &window, const ThreadInfo &info) +void NEComputeAllAnchorsKernel::internal_run<int16_t>(const Window &window) { Iterator all_anchors_it(_all_anchors, window); Iterator anchors_it(_all_anchors, window); @@ -136,7 +136,7 @@ void NEComputeAllAnchorsKernel::internal_run<int16_t>(const Window &window, cons } template <typename T> -void NEComputeAllAnchorsKernel::internal_run(const Window &window, const ThreadInfo &info) +void NEComputeAllAnchorsKernel::internal_run(const Window &window) { Iterator all_anchors_it(_all_anchors, window); Iterator anchors_it(_all_anchors, window); @@ -174,19 +174,19 @@ void NEComputeAllAnchorsKernel::run(const Window &window, const ThreadInfo &info { case DataType::QSYMM16: { - internal_run<int16_t>(window, info); + internal_run<int16_t>(window); break; } #ifdef __ARM_FEATURE_FP16_VECTOR_ARITHMETIC case DataType::F16: { - internal_run<float16_t>(window, info); + internal_run<float16_t>(window); break; } #endif // __ARM_FEATURE_FP16_VECTOR_ARITHMETIC case DataType::F32: { - internal_run<float>(window, info); + internal_run<float>(window); break; } default: diff --git a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp index 1dab5d955d..4bd03e959e 100644 --- a/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp +++ b/src/core/NEON/kernels/NEPixelWiseMultiplicationKernel.cpp @@ -191,7 +191,7 @@ inline uint16x8_t scale255_U16_U16(uint16x8_t in) return vreinterpretq_u16_s16(vcombine_s16(vmovn_s32(tmp_s2), vmovn_s32(tmp_s1))); } -inline void mul_saturate_QASYMM8_QASYMM8_QASYMM8_n_opt(const void *__restrict input1_ptr, const void *__restrict input2_ptr, void *__restrict output_ptr, float scale, +inline void mul_saturate_QASYMM8_QASYMM8_QASYMM8_n_opt(const void *__restrict input1_ptr, const void *__restrict input2_ptr, void *__restrict output_ptr, float32x4_t input1_vscale, int32x4_t input1_voffset, float32x4_t input2_vscale, int32x4_t input2_voffset, float32x4_t output_voffset, float32x4_t vinvscale) { const auto input1 = static_cast<const qasymm8_t *__restrict>(input1_ptr); @@ -739,7 +739,7 @@ void NEPixelWiseMultiplicationKernel::run(const Window &window, const ThreadInfo execute_window_loop(collapsed, [&](const Coordinates &) { - mul_saturate_QASYMM8_QASYMM8_QASYMM8_n_opt(input1.ptr(), input2.ptr(), output.ptr(), _scale, + mul_saturate_QASYMM8_QASYMM8_QASYMM8_n_opt(input1.ptr(), input2.ptr(), output.ptr(), input1_vscale, input1_voffset, input2_vscale, input2_voffset, output_voffset, vinvscale); ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input1)); ARM_COMPUTE_UNUSED(collapsed.slide_window_slice_3D(slice_input2)); diff --git a/src/core/NEON/kernels/arm_gemm/buffer_manager.hpp b/src/core/NEON/kernels/arm_gemm/buffer_manager.hpp index 03f099d57b..001cab7f09 100644 --- a/src/core/NEON/kernels/arm_gemm/buffer_manager.hpp +++ b/src/core/NEON/kernels/arm_gemm/buffer_manager.hpp @@ -303,22 +303,32 @@ public: BufferManager(BufferManager &) = delete; BufferManager & operator=(BufferManager &) = delete; - BufferManager(const int maxthreads, const size_t buffersize, void *storage) : _storage(storage) { } + BufferManager(const int maxthreads, const size_t buffersize, void *storage) : _storage(storage) { + UNUSED(maxthreads); + UNUSED(buffersize); + } ~BufferManager() { } // Say how much storage is needed. static inline size_t get_storage_requirement(const int maxthreads, const size_t buffersize) { + UNUSED(maxthreads); return buffersize; } template <typename T> - void try_populate(const int index, T func) { } + void try_populate(const int index, T func) { + UNUSED(index); + UNUSED(func); + } - void release(const int index) { } + void release(const int index) { + UNUSED(index); + } template <typename T> void *get(const int index, T func) { + UNUSED(index); func(_storage); return _storage; } diff --git a/src/core/NEON/kernels/arm_gemm/gemm_hybrid.hpp b/src/core/NEON/kernels/arm_gemm/gemm_hybrid.hpp index a744376393..436f55dee2 100644 --- a/src/core/NEON/kernels/arm_gemm/gemm_hybrid.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemm_hybrid.hpp @@ -146,6 +146,7 @@ public: // Execute void execute(unsigned int start, unsigned int end, int threadid) override { + UNUSED(threadid); #ifdef CYCLE_PROFILING profiler prof; #endif diff --git a/src/core/NEON/kernels/arm_gemm/gemv_batched.hpp b/src/core/NEON/kernels/arm_gemm/gemv_batched.hpp index 8dae7c3098..4453ee8243 100644 --- a/src/core/NEON/kernels/arm_gemm/gemv_batched.hpp +++ b/src/core/NEON/kernels/arm_gemm/gemv_batched.hpp @@ -48,6 +48,8 @@ public: void set_arrays(const To *A, const int lda, const int A_batch_stride, const int A_multi_stride, const To *B, const int ldb, const int B_multi_stride, Tr *C, const int ldc, const int C_batch_stride, const int C_multi_stride) override { + UNUSED(lda); + UNUSED(ldc); /* A and C's batch stride becomes their new row stride. New batch stride is 0 as nbatches for subgemm is always 1. */ _subgemm->set_arrays(A, A_batch_stride, 0, A_multi_stride, B, ldb, B_multi_stride, diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s16_12x8.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s16_12x8.hpp index 2fcb587df1..a19e07d30f 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s16_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s16_12x8.hpp @@ -65,7 +65,7 @@ public: kern_type kernel = a64_gemm_s16_asimd_12x8; - gemm_s16_12x8(const CPUInfo *ci) { } + gemm_s16_12x8(const CPUInfo *ci) { UNUSED(ci); } }; } // namespace arm_gemm diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp index 71c666ad00..7856eb55e0 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_s8_4x4.hpp @@ -59,7 +59,7 @@ public: kern_type kernel=a64_gemm_s8_4x4; - gemm_s8_4x4(const CPUInfo *ci) { } + gemm_s8_4x4(const CPUInfo *ci) { UNUSED(ci); } }; } // namespace arm_gemm diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u16_12x8.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u16_12x8.hpp index 3d5c92c622..a8255629fc 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u16_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u16_12x8.hpp @@ -65,7 +65,7 @@ public: kern_type kernel = a64_gemm_u16_asimd_12x8; - gemm_u16_12x8(const CPUInfo *ci) { } + gemm_u16_12x8(const CPUInfo *ci) { UNUSED(ci); } }; } // namespace arm_gemm diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp index fda7657b2b..56044e9280 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_gemm_u8_4x4.hpp @@ -68,6 +68,7 @@ public: kern_type kernel = a64_gemm_u8_4x4; gemm_u8_4x4(const CPUInfo *ci) { + UNUSED(ci); } }; diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp index 95e3712e84..9b70827125 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_nativeA_pretransposeB_16x4.hpp @@ -69,7 +69,7 @@ public: kern_type kernel=a64_sgemm_nativeA_pretransposeB_16x4; sgemm_nativeA_pretransposeB_16x4(const CPUInfo *ci) { - + UNUSED(ci); } }; diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4.hpp index 3d2b324314..a5511931d8 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemm_native_16x4.hpp @@ -62,7 +62,7 @@ public: kern_type kernel=a64_sgemm_native_16x4; sgemm_native_16x4(const CPUInfo *ci) { - + UNUSED(ci); } }; diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_pretransposed.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_pretransposed.hpp index f5b4f4aa19..c5b410f0b8 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_pretransposed.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_pretransposed.hpp @@ -69,7 +69,7 @@ public: kern_type kernel = a64_sgemv_pretransposed; - sgemv_pretransposed(const CPUInfo *ci) { } + sgemv_pretransposed(const CPUInfo *ci) { UNUSED(ci); } }; } // namespace arm_gemm diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans.hpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans.hpp index cbaa0cfb1b..c904dc8e7a 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans.hpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_sgemv_trans.hpp @@ -49,7 +49,7 @@ public: kern_type kernel=a64_sgemv_trans; - sgemv_trans(const CPUInfo *ci) { } + sgemv_trans(const CPUInfo *ci) { UNUSED(ci); } }; } // namespace arm_gemm diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/a55.cpp index c957d11608..e8ac2d7489 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/a55.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_s8s32_dot_4x6_a55(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = (N / 4) - 1; const long ldab = lda * sizeof(int8_t); const long ldcb = ldc * sizeof(int32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/generic.cpp index 05d8bd5ad1..64ab6f0fdc 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x6/generic.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_s8s32_dot_4x6(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = (N / 4) - 1; const long ldab = lda * sizeof(int8_t); const long ldcb = ldc * sizeof(int32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/a55.cpp index f050fff84a..4ed2bc823c 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/a55.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_s8s32_dot_4x8_a55(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = (N / 4) - 1; const long ldab = lda * sizeof(int8_t); const long ldcb = ldc * sizeof(int32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/generic.cpp index 1881cf57a8..1eb58e1cd6 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_s8s32_dot_4x8/generic.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_s8s32_dot_4x8(const int8_t *A, int lda, const int8_t *B, int32_t *C, int ldc, int32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = (N / 4) - 1; const long ldab = lda * sizeof(int8_t); const long ldcb = ldc * sizeof(int32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/a55.cpp index 88ea940fb7..8bf5debc37 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/a55.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_u8u32_dot_4x6_a55(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, uint32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = iceildiv(N, (int)4) - 1; const long ldab = lda * sizeof(uint8_t); const long ldcb = ldc * sizeof(uint32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/generic.cpp index ec3ba4ad2a..1f279afd43 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x6/generic.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_u8u32_dot_4x6(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, uint32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = iceildiv(N, (int)4) - 1; const long ldab = lda * sizeof(uint8_t); const long ldcb = ldc * sizeof(uint32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/a55.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/a55.cpp index fdd928d567..ec29698f48 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/a55.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/a55.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_u8u32_dot_4x8_a55(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, uint32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = iceildiv(N, (int)4) - 1; const long ldab = lda * sizeof(uint8_t); const long ldcb = ldc * sizeof(uint32_t); diff --git a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp index 7cc8c285ca..46ca013270 100644 --- a/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp +++ b/src/core/NEON/kernels/arm_gemm/kernels/a64_smallK_hybrid_u8u32_dot_4x8/generic.cpp @@ -32,6 +32,7 @@ namespace arm_gemm { void a64_smallK_hybrid_u8u32_dot_4x8(const uint8_t *A, int lda, const uint8_t *B, uint32_t *C, int ldc, uint32_t beta, int M, int N, int K) { + UNUSED(beta); const long loops_count = iceildiv(N, (int)4) - 1; const long ldab = lda * sizeof(uint8_t); const long ldcb = ldc * sizeof(uint32_t); diff --git a/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp b/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp index ce7310d13d..410a0a1dc9 100644 --- a/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp +++ b/src/core/NEON/kernels/arm_gemm/merges/a64_merge_int32_12x8.hpp @@ -27,6 +27,7 @@ template<> inline void MergeResults<12, 8, false>(int32_t *out, const int32_t *in, const int ldout, const int y0, const int ymax, const int x0, const int xmax, const int32_t alpha, const int32_t beta) { + UNUSED(alpha); const int32_t *inptr = in; prefetch_6x(inptr); prefetch_6x(inptr + 96); diff --git a/src/core/NEON/kernels/arm_gemm/utils.hpp b/src/core/NEON/kernels/arm_gemm/utils.hpp index 4271997cdd..7dbbe91ba2 100644 --- a/src/core/NEON/kernels/arm_gemm/utils.hpp +++ b/src/core/NEON/kernels/arm_gemm/utils.hpp @@ -32,6 +32,8 @@ // Paranoid option for the above with assert // #define UNREACHABLE(why) assert(0 && why) +#define UNUSED(x) (void)(x) + template<typename T> inline T iceildiv(const T a, const T b) { return (a + b - 1) / b; |