From 05fb448bf48e31d723dfd9f4bbf3899ff65f0fba Mon Sep 17 00:00:00 2001 From: giuros01 Date: Tue, 26 Mar 2019 17:44:40 +0000 Subject: COMPMID-1963: Implement FFT (2D) on NEON Change-Id: I3b564be8d7949e00c6544071ef62dd51de838c96 Signed-off-by: giuros01 Reviewed-on: https://review.mlplatform.org/c/1048 Comments-Addressed: Arm Jenkins Tested-by: Arm Jenkins Reviewed-by: Georgios Pinitas --- src/core/NEON/kernels/NEFFTDigitReverseKernel.cpp | 21 +- src/core/NEON/kernels/NEFFTRadixStageKernel.cpp | 579 +++++++++++++++------- src/core/NEON/kernels/NEFFTScaleKernel.cpp | 136 +++++ 3 files changed, 548 insertions(+), 188 deletions(-) create mode 100644 src/core/NEON/kernels/NEFFTScaleKernel.cpp (limited to 'src/core') diff --git a/src/core/NEON/kernels/NEFFTDigitReverseKernel.cpp b/src/core/NEON/kernels/NEFFTDigitReverseKernel.cpp index 845fcef4f3..b2ffb01e99 100644 --- a/src/core/NEON/kernels/NEFFTDigitReverseKernel.cpp +++ b/src/core/NEON/kernels/NEFFTDigitReverseKernel.cpp @@ -37,7 +37,7 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c { ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(idx, 1, DataType::U32); - ARM_COMPUTE_RETURN_ERROR_ON(axis != 0); + ARM_COMPUTE_RETURN_ERROR_ON(axis > 1); // Checks performed when output is configured if((output != nullptr) && (output->total_size() != 0)) @@ -96,15 +96,24 @@ void NEFFTDigitReverseKernel::run(const Window &window, const ThreadInfo &info) Iterator out(_output, window); const size_t element_size = _input->info()->element_size(); + // Pointers to the buffers + const size_t offset = _input->info()->offset_first_element_in_bytes(); + auto *idx_ptr = reinterpret_cast(_idx->buffer()); + uint8_t *input_ptr = offset + _input->buffer(); + + // Strides + const size_t stride_x = _input->info()->strides_in_bytes()[0]; + const size_t stride_y = _input->info()->strides_in_bytes()[1]; + const size_t stride_z = _input->info()->strides_in_bytes()[2]; + const size_t stride_w = _input->info()->strides_in_bytes()[3]; + execute_window_loop(window, [&](const Coordinates & id) { - unsigned int in_index_1d = *reinterpret_cast(_idx->ptr_to_element(Coordinates(id.x()))); - - auto reverse_id = id; + unsigned int in_index_1d = idx_ptr[id[_axis]]; + auto reverse_id = id; reverse_id.set(_axis, in_index_1d); - memcpy(out.ptr(), _input->ptr_to_element(reverse_id), 2 * element_size); - + memcpy(out.ptr(), input_ptr + reverse_id.x() * stride_x + reverse_id.y() * stride_y + reverse_id.z() * stride_z + reverse_id[3] * stride_w, element_size); }, out); diff --git a/src/core/NEON/kernels/NEFFTRadixStageKernel.cpp b/src/core/NEON/kernels/NEFFTRadixStageKernel.cpp index b264590791..148bbe915a 100644 --- a/src/core/NEON/kernels/NEFFTRadixStageKernel.cpp +++ b/src/core/NEON/kernels/NEFFTRadixStageKernel.cpp @@ -24,8 +24,6 @@ #include "arm_compute/core/NEON/kernels/NEFFTRadixStageKernel.h" #include "arm_compute/core/ITensor.h" -#include "arm_compute/core/NEON/wrapper/traits.h" -#include "arm_compute/core/NEON/wrapper/wrapper.h" #include "arm_compute/core/TensorInfo.h" #include "arm_compute/core/Types.h" #include "arm_compute/core/Utils.h" @@ -34,28 +32,53 @@ #include #include #include +#include + +#include "arm_compute/core/NEON/wrapper/traits.h" +#include "arm_compute/core/NEON/wrapper/wrapper.h" namespace arm_compute { namespace { -constexpr float PI = 3.141592653589793f; +// PI constant (from cmath) +constexpr float kPi = float(M_PI); + +// Constant used in the fft_3 kernel +constexpr float kSqrt3Div2 = 0.866025403784438; + +// Constants used in the fft_5 kernel +constexpr float kW5_0 = 0.30901699437494f; +constexpr float kW5_1 = 0.95105651629515f; +constexpr float kW5_2 = 0.80901699437494f; +constexpr float kW5_3 = 0.58778525229247f; + +// Constants used in the fft_7 kernel +constexpr float kW7_0 = 0.62348980185873f; +constexpr float kW7_1 = 0.78183148246802f; +constexpr float kW7_2 = 0.22252093395631f; +constexpr float kW7_3 = 0.97492791218182f; +constexpr float kW7_4 = 0.90096886790241f; +constexpr float kW7_5 = 0.43388373911755f; + +// Constant used in the fft_8 kernel +constexpr float kSqrt2Div2 = 0.707106781186548; float32x2_t c_mul_neon(float32x2_t a, float32x2_t b) { - float32x2_t tmp = wrapper::vmul(a, b); + using ExactTagType = typename wrapper::traits::neon_vector::tag_type; - const float P1 = wrapper::vgetlane(tmp, 0); - const float P2 = wrapper::vgetlane(tmp, 1); + const float32x2_t mask = { -1.0, 1.0 }; + const float32x2_t tmp0 = wrapper::vdup_n(wrapper::vgetlane(a, 0), ExactTagType{}); + const float32x2_t tmp1 = wrapper::vdup_n(wrapper::vgetlane(a, 1), ExactTagType{}); - const float a_r = wrapper::vgetlane(a, 0); - const float a_i = wrapper::vgetlane(a, 1); - const float b_r = wrapper::vgetlane(b, 0); - const float b_i = wrapper::vgetlane(b, 1); + float32x2_t res = wrapper::vmul(tmp0, b); - const float P3 = (a_r + a_i) * (b_r + b_i); - float32x2_t out = { P1 - P2, P3 - P2 - P1 }; - return out; + b = wrapper::vrev64(b); + b = wrapper::vmul(b, mask); + res = wrapper::vmla(res, tmp1, b); + + return res; } float32x2_t c_mul_neon_img(float32x2_t a, float img_constant) @@ -107,7 +130,6 @@ void fft_2(float32x2_t &x, float32x2_t &y, float32x2_t &w) y = wrapper::vsub(a, b); } -constexpr float sqrt3div2 = 0.866025403784438; void fft_3(float32x2_t &x, float32x2_t &y, float32x2_t &z, const float32x2_t &w, const float32x2_t &w2) { float32x2_t a = x; @@ -118,7 +140,7 @@ void fft_3(float32x2_t &x, float32x2_t &y, float32x2_t &z, const float32x2_t &w, x = wrapper::vadd(x, c); const auto v1 = wrapper::vmul(float32x2_t{ 0.5f, 0.5 }, wrapper::vadd(b, c)); - const auto v2 = c_mul_neon(float32x2_t{ 0.f, -sqrt3div2 }, wrapper::vsub(b, c)); + const auto v2 = c_mul_neon(float32x2_t{ 0.f, -kSqrt3Div2 }, wrapper::vsub(b, c)); y = z = wrapper::vsub(a, v1); y = wrapper::vadd(y, v2); @@ -149,10 +171,6 @@ void fft_4(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, c x4 = wrapper::vadd(x41, x42); } -constexpr float W5_0 = 0.30901699437494f; -constexpr float W5_1 = 0.95105651629515f; -constexpr float W5_2 = 0.80901699437494f; -constexpr float W5_3 = 0.58778525229247f; void fft_5(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, float32x2_t &x5, const float32x2_t &w, const float32x2_t &w2, const float32x2_t &w3, const float32x2_t &w4) { const auto a = x1; @@ -161,25 +179,25 @@ void fft_5(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto d = c_mul_neon(w3, x4); const auto e = c_mul_neon(w4, x5); - const auto b0 = c_mul_neon(float32x2_t{ W5_0, -W5_1 }, b); - const auto b1 = c_mul_neon(float32x2_t{ -W5_2, -W5_3 }, b); - const auto b2 = c_mul_neon(float32x2_t{ -W5_2, W5_3 }, b); - const auto b3 = c_mul_neon(float32x2_t{ W5_0, W5_1 }, b); + const auto b0 = c_mul_neon(float32x2_t{ kW5_0, -kW5_1 }, b); + const auto b1 = c_mul_neon(float32x2_t{ -kW5_2, -kW5_3 }, b); + const auto b2 = c_mul_neon(float32x2_t{ -kW5_2, kW5_3 }, b); + const auto b3 = c_mul_neon(float32x2_t{ kW5_0, kW5_1 }, b); - const auto c0 = c_mul_neon(float32x2_t{ -W5_2, -W5_3 }, c); - const auto c1 = c_mul_neon(float32x2_t{ W5_0, W5_1 }, c); - const auto c2 = c_mul_neon(float32x2_t{ W5_0, -W5_1 }, c); - const auto c3 = c_mul_neon(float32x2_t{ -W5_2, W5_3 }, c); + const auto c0 = c_mul_neon(float32x2_t{ -kW5_2, -kW5_3 }, c); + const auto c1 = c_mul_neon(float32x2_t{ kW5_0, kW5_1 }, c); + const auto c2 = c_mul_neon(float32x2_t{ kW5_0, -kW5_1 }, c); + const auto c3 = c_mul_neon(float32x2_t{ -kW5_2, kW5_3 }, c); - const auto d0 = c_mul_neon(float32x2_t{ -W5_2, W5_3 }, d); - const auto d1 = c_mul_neon(float32x2_t{ W5_0, -W5_1 }, d); - const auto d2 = c_mul_neon(float32x2_t{ W5_0, W5_1 }, d); - const auto d3 = c_mul_neon(float32x2_t{ -W5_2, -W5_3 }, d); + const auto d0 = c_mul_neon(float32x2_t{ -kW5_2, kW5_3 }, d); + const auto d1 = c_mul_neon(float32x2_t{ kW5_0, -kW5_1 }, d); + const auto d2 = c_mul_neon(float32x2_t{ kW5_0, kW5_1 }, d); + const auto d3 = c_mul_neon(float32x2_t{ -kW5_2, -kW5_3 }, d); - const auto e0 = c_mul_neon(float32x2_t{ W5_0, W5_1 }, e); - const auto e1 = c_mul_neon(float32x2_t{ -W5_2, W5_3 }, e); - const auto e2 = c_mul_neon(float32x2_t{ -W5_2, -W5_3 }, e); - const auto e3 = c_mul_neon(float32x2_t{ W5_0, -W5_1 }, e); + const auto e0 = c_mul_neon(float32x2_t{ kW5_0, kW5_1 }, e); + const auto e1 = c_mul_neon(float32x2_t{ -kW5_2, kW5_3 }, e); + const auto e2 = c_mul_neon(float32x2_t{ -kW5_2, -kW5_3 }, e); + const auto e3 = c_mul_neon(float32x2_t{ kW5_0, -kW5_1 }, e); x1 = reduce_sum_5(a, b, c, d, e); x2 = reduce_sum_5(a, b0, c0, d0, e0); @@ -188,12 +206,6 @@ void fft_5(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f x5 = reduce_sum_5(a, b3, c3, d3, e3); } -constexpr float W7_0 = 0.62348980185873f; -constexpr float W7_1 = 0.78183148246802f; -constexpr float W7_2 = 0.22252093395631f; -constexpr float W7_3 = 0.97492791218182f; -constexpr float W7_4 = 0.90096886790241f; -constexpr float W7_5 = 0.43388373911755f; void fft_7(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, float32x2_t &x5, float32x2_t &x6, float32x2_t &x7, const float32x2_t &w, const float32x2_t &w2, const float32x2_t &w3, const float32x2_t &w4, const float32x2_t &w5, const float32x2_t &w6) @@ -206,47 +218,47 @@ void fft_7(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto f = c_mul_neon(w5, x6); const auto g = c_mul_neon(w6, x7); - const auto b0 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, b); - const auto b1 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, b); - const auto b2 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, b); - const auto b3 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, b); - const auto b4 = c_mul_neon(float32x2_t{ -W7_2, W7_3 }, b); - const auto b5 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, b); - - const auto c0 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, c); - const auto c1 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, c); - const auto c2 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, c); - const auto c3 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, c); - const auto c4 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, c); - const auto c5 = c_mul_neon(float32x2_t{ -W7_2, W7_3 }, c); - - const auto d0 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, d); - const auto d1 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, d); - const auto d2 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, d); - const auto d3 = c_mul_neon(float32x2_t{ -W7_2, +W7_3 }, d); - const auto d4 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, d); - const auto d5 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, d); - - const auto e0 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, e); - const auto e1 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, e); - const auto e2 = c_mul_neon(float32x2_t{ -W7_2, W7_3 }, e); - const auto e3 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, e); - const auto e4 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, e); - const auto e5 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, e); - - const auto f0 = c_mul_neon(float32x2_t{ -W7_2, W7_3 }, f); - const auto f1 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, f); - const auto f2 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, f); - const auto f3 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, f); - const auto f4 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, f); - const auto f5 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, f); - - const auto g0 = c_mul_neon(float32x2_t{ W7_0, W7_1 }, g); - const auto g1 = c_mul_neon(float32x2_t{ -W7_2, W7_3 }, g); - const auto g2 = c_mul_neon(float32x2_t{ -W7_4, W7_5 }, g); - const auto g3 = c_mul_neon(float32x2_t{ -W7_4, -W7_5 }, g); - const auto g4 = c_mul_neon(float32x2_t{ -W7_2, -W7_3 }, g); - const auto g5 = c_mul_neon(float32x2_t{ W7_0, -W7_1 }, g); + const auto b0 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, b); + const auto b1 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, b); + const auto b2 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, b); + const auto b3 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, b); + const auto b4 = c_mul_neon(float32x2_t{ -kW7_2, kW7_3 }, b); + const auto b5 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, b); + + const auto c0 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, c); + const auto c1 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, c); + const auto c2 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, c); + const auto c3 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, c); + const auto c4 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, c); + const auto c5 = c_mul_neon(float32x2_t{ -kW7_2, kW7_3 }, c); + + const auto d0 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, d); + const auto d1 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, d); + const auto d2 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, d); + const auto d3 = c_mul_neon(float32x2_t{ -kW7_2, +kW7_3 }, d); + const auto d4 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, d); + const auto d5 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, d); + + const auto e0 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, e); + const auto e1 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, e); + const auto e2 = c_mul_neon(float32x2_t{ -kW7_2, kW7_3 }, e); + const auto e3 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, e); + const auto e4 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, e); + const auto e5 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, e); + + const auto f0 = c_mul_neon(float32x2_t{ -kW7_2, kW7_3 }, f); + const auto f1 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, f); + const auto f2 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, f); + const auto f3 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, f); + const auto f4 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, f); + const auto f5 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, f); + + const auto g0 = c_mul_neon(float32x2_t{ kW7_0, kW7_1 }, g); + const auto g1 = c_mul_neon(float32x2_t{ -kW7_2, kW7_3 }, g); + const auto g2 = c_mul_neon(float32x2_t{ -kW7_4, kW7_5 }, g); + const auto g3 = c_mul_neon(float32x2_t{ -kW7_4, -kW7_5 }, g); + const auto g4 = c_mul_neon(float32x2_t{ -kW7_2, -kW7_3 }, g); + const auto g5 = c_mul_neon(float32x2_t{ kW7_0, -kW7_1 }, g); x1 = reduce_sum_7(a, b, c, d, e, f, g); x2 = reduce_sum_7(a, b0, c0, d0, e0, f0, g0); @@ -257,7 +269,6 @@ void fft_7(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f x7 = reduce_sum_7(a, b5, c5, d5, e5, f5, g5); } -constexpr float sqrt2div2 = 0.707106781186548; void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, float32x2_t &x5, float32x2_t &x6, float32x2_t &x7, float32x2_t &x8, const float32x2_t &w, const float32x2_t &w2, const float32x2_t &w3, const float32x2_t &w4, const float32x2_t &w5, const float32x2_t &w6, @@ -272,13 +283,13 @@ void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto g = c_mul_neon(w6, x7); const auto h = c_mul_neon(w7, x8); - const auto b0 = c_mul_neon(float32x2_t{ sqrt2div2, -sqrt2div2 }, b); + const auto b0 = c_mul_neon(float32x2_t{ kSqrt2Div2, -kSqrt2Div2 }, b); const auto b1 = c_mul_neon(float32x2_t{ 0, -1 }, b); - const auto b2 = c_mul_neon(float32x2_t{ -sqrt2div2, -sqrt2div2 }, b); + const auto b2 = c_mul_neon(float32x2_t{ -kSqrt2Div2, -kSqrt2Div2 }, b); const auto b3 = c_mul_neon(float32x2_t{ -1, 0 }, b); - const auto b4 = c_mul_neon(float32x2_t{ -sqrt2div2, sqrt2div2 }, b); + const auto b4 = c_mul_neon(float32x2_t{ -kSqrt2Div2, kSqrt2Div2 }, b); const auto b5 = c_mul_neon(float32x2_t{ 0, 1 }, b); - const auto b6 = c_mul_neon(float32x2_t{ sqrt2div2, sqrt2div2 }, b); + const auto b6 = c_mul_neon(float32x2_t{ kSqrt2Div2, kSqrt2Div2 }, b); const auto c0 = c_mul_neon(float32x2_t{ 0, -1 }, c); const auto c1 = c_mul_neon(float32x2_t{ -1, 0 }, c); @@ -288,13 +299,13 @@ void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto c5 = c_mul_neon(float32x2_t{ -1, 0 }, c); const auto c6 = c_mul_neon(float32x2_t{ 0, 1 }, c); - const auto d0 = c_mul_neon(float32x2_t{ -sqrt2div2, -sqrt2div2 }, d); + const auto d0 = c_mul_neon(float32x2_t{ -kSqrt2Div2, -kSqrt2Div2 }, d); const auto d1 = c_mul_neon(float32x2_t{ 0, 1 }, d); - const auto d2 = c_mul_neon(float32x2_t{ sqrt2div2, -sqrt2div2 }, d); + const auto d2 = c_mul_neon(float32x2_t{ kSqrt2Div2, -kSqrt2Div2 }, d); const auto d3 = c_mul_neon(float32x2_t{ -1, 0 }, d); - const auto d4 = c_mul_neon(float32x2_t{ sqrt2div2, sqrt2div2 }, d); + const auto d4 = c_mul_neon(float32x2_t{ kSqrt2Div2, kSqrt2Div2 }, d); const auto d5 = c_mul_neon(float32x2_t{ 0, -1 }, d); - const auto d6 = c_mul_neon(float32x2_t{ -sqrt2div2, sqrt2div2 }, d); + const auto d6 = c_mul_neon(float32x2_t{ -kSqrt2Div2, kSqrt2Div2 }, d); const auto e0 = c_mul_neon(float32x2_t{ -1, 0 }, e); const auto e1 = c_mul_neon(float32x2_t{ 1, 0 }, e); @@ -304,13 +315,13 @@ void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto e5 = c_mul_neon(float32x2_t{ 1, 0 }, e); const auto e6 = c_mul_neon(float32x2_t{ -1, 0 }, e); - const auto f0 = c_mul_neon(float32x2_t{ -sqrt2div2, sqrt2div2 }, f); + const auto f0 = c_mul_neon(float32x2_t{ -kSqrt2Div2, kSqrt2Div2 }, f); const auto f1 = c_mul_neon(float32x2_t{ 0, -1 }, f); - const auto f2 = c_mul_neon(float32x2_t{ sqrt2div2, sqrt2div2 }, f); + const auto f2 = c_mul_neon(float32x2_t{ kSqrt2Div2, kSqrt2Div2 }, f); const auto f3 = c_mul_neon(float32x2_t{ -1, 0 }, f); - const auto f4 = c_mul_neon(float32x2_t{ sqrt2div2, -sqrt2div2 }, f); + const auto f4 = c_mul_neon(float32x2_t{ kSqrt2Div2, -kSqrt2Div2 }, f); const auto f5 = c_mul_neon(float32x2_t{ 0, 1 }, f); - const auto f6 = c_mul_neon(float32x2_t{ -sqrt2div2, -sqrt2div2 }, f); + const auto f6 = c_mul_neon(float32x2_t{ -kSqrt2Div2, -kSqrt2Div2 }, f); const auto g0 = c_mul_neon(float32x2_t{ 0, 1 }, g); const auto g1 = c_mul_neon(float32x2_t{ -1, 0 }, g); @@ -320,13 +331,13 @@ void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f const auto g5 = c_mul_neon(float32x2_t{ -1, 0 }, g); const auto g6 = c_mul_neon(float32x2_t{ 0, -1 }, g); - const auto h0 = c_mul_neon(float32x2_t{ sqrt2div2, sqrt2div2 }, h); + const auto h0 = c_mul_neon(float32x2_t{ kSqrt2Div2, kSqrt2Div2 }, h); const auto h1 = c_mul_neon(float32x2_t{ 0, 1 }, h); - const auto h2 = c_mul_neon(float32x2_t{ -sqrt2div2, sqrt2div2 }, h); + const auto h2 = c_mul_neon(float32x2_t{ -kSqrt2Div2, kSqrt2Div2 }, h); const auto h3 = c_mul_neon(float32x2_t{ -1, 0 }, h); - const auto h4 = c_mul_neon(float32x2_t{ -sqrt2div2, -sqrt2div2 }, h); + const auto h4 = c_mul_neon(float32x2_t{ -kSqrt2Div2, -kSqrt2Div2 }, h); const auto h5 = c_mul_neon(float32x2_t{ 0, -1 }, h); - const auto h6 = c_mul_neon(float32x2_t{ sqrt2div2, -sqrt2div2 }, h); + const auto h6 = c_mul_neon(float32x2_t{ kSqrt2Div2, -kSqrt2Div2 }, h); x1 = reduce_sum_8(a, b, c, d, e, f, g, h); x2 = reduce_sum_8(a, b0, c0, d0, e0, f0, g0, h0); @@ -339,17 +350,12 @@ void fft_8(float32x2_t &x1, float32x2_t &x2, float32x2_t &x3, float32x2_t &x4, f } template -void fft_radix_2_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_2_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) { - unsigned int Nx2 = 2 * Nx; - float alpha = 2 * PI / Nx2; - - float32x2_t w{ 1, 0 }; - const float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; - + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx2) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { auto a = float32x2_t{ 0, 0 }; auto b = float32x2_t{ 0, 0 }; @@ -386,19 +392,38 @@ void fft_radix_2_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } -template -void fft_radix_3_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_2_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) { - const unsigned int Nx3 = 3 * Nx; - const float alpha = 2 * PI / float(Nx3); - float32x2_t w{ 1, 0 }; - const float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + // Base-case prime transform + fft_2(a, b, w); + + // Write outputs + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + } + + w = c_mul_neon(w, w_m); + } +} + +template +void fft_radix_3_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { const auto w2 = c_mul_neon(w, w); - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx3) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { // Load inputs float32x2_t a = { 0, 0 }; @@ -435,21 +460,42 @@ void fft_radix_3_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } -template -void fft_radix_4_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_3_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) { - unsigned int Nx4 = 4 * Nx; - const float alpha = 2 * PI / float(Nx4); + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + const auto w2 = c_mul_neon(w, w); + + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + float32x2_t c = wrapper::vload(x + M * (k + 4 * Nx)); + + // Base-case prime transform + fft_3(a, b, c, w, w2); - float32x2_t w{ 1, 0 }; - float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + // Store the output + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + wrapper::vstore(X + M * (k + 4 * Nx), c); + } + w = c_mul_neon(w, w_m); + } +} +template +void fft_radix_4_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { const auto w2 = c_mul_neon(w, w); const auto w3 = c_mul_neon(w2, w); - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx4) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { float32x2_t a = { 0, 0 }; float32x2_t b = { 0, 0 }; @@ -494,22 +540,46 @@ void fft_radix_4_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } -template -void fft_radix_5_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_4_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) { - unsigned int Nx5 = 5 * Nx; - const float alpha = 2 * PI / float(Nx5); + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + const auto w2 = c_mul_neon(w, w); + const auto w3 = c_mul_neon(w2, w); + + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + float32x2_t c = wrapper::vload(x + M * (k + 4 * Nx)); + float32x2_t d = wrapper::vload(x + M * (k + 6 * Nx)); + + // Base-case prime transform + fft_4(a, b, c, d, w, w2, w3); - float32x2_t w{ 1, 0 }; - float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + wrapper::vstore(X + M * (k + 4 * Nx), c); + wrapper::vstore(X + M * (k + 6 * Nx), d); + } + + w = c_mul_neon(w, w_m); + } +} +template +void fft_radix_5_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { const float32x2_t w2 = c_mul_neon(w, w); const float32x2_t w3 = c_mul_neon(w2, w); const float32x2_t w4 = c_mul_neon(w3, w); - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx5) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { float32x2_t a = { 0, 0 }; float32x2_t b = { 0, 0 }; @@ -560,15 +630,43 @@ void fft_radix_5_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } -template -void fft_radix_7_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_5_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) { - unsigned int Nx7 = 7 * Nx; - const float alpha = 2 * PI / float(Nx7); + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + const float32x2_t w2 = c_mul_neon(w, w); + const float32x2_t w3 = c_mul_neon(w2, w); + const float32x2_t w4 = c_mul_neon(w3, w); - float32x2_t w{ 1, 0 }; - float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + float32x2_t c = wrapper::vload(x + M * (k + 4 * Nx)); + float32x2_t d = wrapper::vload(x + M * (k + 6 * Nx)); + float32x2_t e = wrapper::vload(x + M * (k + 8 * Nx)); + + // Base-case prime transform + fft_5(a, b, c, d, e, w, w2, w3, w4); + + // Store outputs + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + wrapper::vstore(X + M * (k + 4 * Nx), c); + wrapper::vstore(X + M * (k + 6 * Nx), d); + wrapper::vstore(X + M * (k + 8 * Nx), e); + } + + w = c_mul_neon(w, w_m); + } +} +template +void fft_radix_7_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { const float32x2_t w2 = c_mul_neon(w, w); @@ -577,7 +675,7 @@ void fft_radix_7_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) const float32x2_t w5 = c_mul_neon(w4, w); const float32x2_t w6 = c_mul_neon(w5, w); - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx7) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { float32x2_t a = { 0, 0 }; float32x2_t b = { 0, 0 }; @@ -637,15 +735,49 @@ void fft_radix_7_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } -template -void fft_radix_8_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) +void fft_radix_7_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) { - unsigned int Nx8 = 8 * Nx; - const float alpha = 2 * PI / float(Nx8); + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + const float32x2_t w2 = c_mul_neon(w, w); + const float32x2_t w3 = c_mul_neon(w2, w); + const float32x2_t w4 = c_mul_neon(w3, w); + const float32x2_t w5 = c_mul_neon(w4, w); + const float32x2_t w6 = c_mul_neon(w5, w); - float32x2_t w{ 1, 0 }; - const float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + float32x2_t c = wrapper::vload(x + M * (k + 4 * Nx)); + float32x2_t d = wrapper::vload(x + M * (k + 6 * Nx)); + float32x2_t e = wrapper::vload(x + M * (k + 8 * Nx)); + float32x2_t f = wrapper::vload(x + M * (k + 10 * Nx)); + float32x2_t g = wrapper::vload(x + M * (k + 12 * Nx)); + + // Base-case prime transform + fft_7(a, b, c, d, e, f, g, w, w2, w3, w4, w5, w6); + + // Store outputs + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + wrapper::vstore(X + M * (k + 4 * Nx), c); + wrapper::vstore(X + M * (k + 6 * Nx), d); + wrapper::vstore(X + M * (k + 8 * Nx), e); + wrapper::vstore(X + M * (k + 10 * Nx), f); + wrapper::vstore(X + M * (k + 12 * Nx), g); + } + + w = c_mul_neon(w, w_m); + } +} +template +void fft_radix_8_axes_0(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; for(unsigned int j = 0; j < Nx; j++) { const float32x2_t w2 = c_mul_neon(w, w); @@ -655,7 +787,7 @@ void fft_radix_8_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) const float32x2_t w6 = c_mul_neon(w5, w); const float32x2_t w7 = c_mul_neon(w6, w); - for(unsigned int k = 2 * j; k < 2 * N; k += 2 * Nx8) + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) { // Load inputs float32x2_t a = { 0, 0 }; @@ -724,11 +856,54 @@ void fft_radix_8_axes_0(float *X, float *x, unsigned int Nx, unsigned int N) } } +void fft_radix_8_axes_1(float *X, float *x, unsigned int Nx, unsigned int NxRadix, const float32x2_t &w_m, unsigned int M, unsigned int N) +{ + float32x2_t w{ 1.0f, 0.0f }; + for(unsigned int j = 0; j < Nx; j++) + { + const float32x2_t w2 = c_mul_neon(w, w); + const float32x2_t w3 = c_mul_neon(w2, w); + const float32x2_t w4 = c_mul_neon(w3, w); + const float32x2_t w5 = c_mul_neon(w4, w); + const float32x2_t w6 = c_mul_neon(w5, w); + const float32x2_t w7 = c_mul_neon(w6, w); + + for(unsigned int k = 2 * j; k < 2 * N; k += 2 * NxRadix) + { + // Load inputs + float32x2_t a = wrapper::vload(x + M * k); + float32x2_t b = wrapper::vload(x + M * (k + 2 * Nx)); + float32x2_t c = wrapper::vload(x + M * (k + 4 * Nx)); + float32x2_t d = wrapper::vload(x + M * (k + 6 * Nx)); + float32x2_t e = wrapper::vload(x + M * (k + 8 * Nx)); + float32x2_t f = wrapper::vload(x + M * (k + 10 * Nx)); + float32x2_t g = wrapper::vload(x + M * (k + 12 * Nx)); + float32x2_t h = wrapper::vload(x + M * (k + 14 * Nx)); + + // Base-case prime transform + fft_8(a, b, c, d, e, f, g, h, w, w2, w3, w4, w5, w6, w7); + + // Store outputs + wrapper::vstore(X + M * k, a); + wrapper::vstore(X + M * (k + 2 * Nx), b); + wrapper::vstore(X + M * (k + 4 * Nx), c); + wrapper::vstore(X + M * (k + 6 * Nx), d); + wrapper::vstore(X + M * (k + 8 * Nx), e); + wrapper::vstore(X + M * (k + 10 * Nx), f); + wrapper::vstore(X + M * (k + 12 * Nx), g); + wrapper::vstore(X + M * (k + 14 * Nx), h); + } + + w = c_mul_neon(w, w_m); + } +} + Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, const FFTRadixStageKernelInfo &config) { ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); - ARM_COMPUTE_RETURN_ERROR_ON(config.axis != 0); + ARM_COMPUTE_RETURN_ERROR_ON(config.axis > 1); ARM_COMPUTE_RETURN_ERROR_ON(NEFFTRadixStageKernel::supported_radix().count(config.radix) == 0); + ARM_COMPUTE_UNUSED(config); // Checks performed when output is configured if((output != nullptr) && (output->total_size() != 0)) @@ -742,12 +917,14 @@ Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output, c std::pair validate_and_configure_window(ITensorInfo *input, ITensorInfo *output, const FFTRadixStageKernelInfo &config) { + ARM_COMPUTE_UNUSED(config); + if(output != nullptr) { auto_init_if_empty(*output, *input); } - Window win = calculate_max_window(*input, Steps(config.radix)); + Window win = calculate_max_window(*input, Steps()); if(output != nullptr) { output->set_valid_region(ValidRegion(Coordinates(), output->tensor_shape())); @@ -758,36 +935,51 @@ std::pair validate_and_configure_window(ITensorInfo *input, ITen } // namespace NEFFTRadixStageKernel::NEFFTRadixStageKernel() - : _input(nullptr), _output(nullptr), _run_in_place(false), _Nx(0), _func() + : _input(nullptr), _output(nullptr), _run_in_place(false), _Nx(0), _axis(0), _radix(0), _func_0(), _func_1() { } -template -void NEFFTRadixStageKernel::set_radix_stage_fun(unsigned int radix) +void NEFFTRadixStageKernel::set_radix_stage_axis0(const FFTRadixStageKernelInfo &config) { - switch(radix) + // FFT table axis 0: [radix, first_stage] + static std::map> fft_table_axis0; + + if(fft_table_axis0.empty()) { - case 2: - _func = &fft_radix_2_axes_0; - break; - case 3: - _func = &fft_radix_3_axes_0; - break; - case 4: - _func = &fft_radix_4_axes_0; - break; - case 5: - _func = &fft_radix_5_axes_0; - break; - case 7: - _func = &fft_radix_7_axes_0; - break; - case 8: - _func = &fft_radix_8_axes_0; - break; - default: - ARM_COMPUTE_ERROR("Radix not supported"); + fft_table_axis0[2][false] = &fft_radix_2_axes_0; + fft_table_axis0[3][false] = &fft_radix_3_axes_0; + fft_table_axis0[4][false] = &fft_radix_4_axes_0; + fft_table_axis0[5][false] = &fft_radix_5_axes_0; + fft_table_axis0[7][false] = &fft_radix_7_axes_0; + fft_table_axis0[8][false] = &fft_radix_8_axes_0; + + fft_table_axis0[2][true] = &fft_radix_2_axes_0; + fft_table_axis0[3][true] = &fft_radix_3_axes_0; + fft_table_axis0[4][true] = &fft_radix_4_axes_0; + fft_table_axis0[5][true] = &fft_radix_5_axes_0; + fft_table_axis0[7][true] = &fft_radix_7_axes_0; + fft_table_axis0[8][true] = &fft_radix_8_axes_0; + } + + _func_0 = fft_table_axis0[config.radix][config.is_first_stage]; +} + +void NEFFTRadixStageKernel::set_radix_stage_axis1(const FFTRadixStageKernelInfo &config) +{ + // FFT table axis 1: [radix, first_stage] + static std::map fft_table_axis1; + + if(fft_table_axis1.empty()) + { + fft_table_axis1[2] = &fft_radix_2_axes_1; + fft_table_axis1[3] = &fft_radix_3_axes_1; + fft_table_axis1[4] = &fft_radix_4_axes_1; + fft_table_axis1[5] = &fft_radix_5_axes_1; + fft_table_axis1[7] = &fft_radix_7_axes_1; + fft_table_axis1[8] = &fft_radix_8_axes_1; } + + _func_1 = fft_table_axis1[config.radix]; } void NEFFTRadixStageKernel::configure(ITensor *input, ITensor *output, const FFTRadixStageKernelInfo &config) @@ -806,14 +998,20 @@ void NEFFTRadixStageKernel::configure(ITensor *input, ITensor *output, const FFT _output = output; _run_in_place = (output == nullptr) || (output == input); _Nx = config.Nx; + _axis = config.axis; + _radix = config.radix; - if(config.is_first_stage) - { - set_radix_stage_fun(config.radix); - } - else + switch(config.axis) { - set_radix_stage_fun(config.radix); + case 0: + set_radix_stage_axis0(config); + break; + case 1: + set_radix_stage_axis1(config); + break; + default: + ARM_COMPUTE_ERROR("Axis not supported"); + break; } // Configure kernel window @@ -841,23 +1039,40 @@ std::set NEFFTRadixStageKernel::supported_radix() void NEFFTRadixStageKernel::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); + ARM_COMPUTE_UNUSED(info); Window input_window = window; - input_window.set(Window::DimX, 0); - - unsigned int N = _input->info()->dimension(0); + input_window.set(_axis, 0); Iterator in(_input, input_window); Iterator out(_run_in_place ? _input : _output, input_window); - execute_window_loop(input_window, [&](const Coordinates &) + // Precompute FFT constants + const unsigned int NxRadix = _radix * _Nx; + const float alpha = 2.0f * kPi / float(NxRadix); + const float32x2_t w_m{ cosf(alpha), -sinf(alpha) }; + + if(_axis == 0) + { + const unsigned int N = _input->info()->dimension(0); + execute_window_loop(input_window, [&](const Coordinates &) + { + _func_0(reinterpret_cast(out.ptr()), reinterpret_cast(in.ptr()), _Nx, NxRadix, w_m, N); + }, + in, out); + } + else { - _func(reinterpret_cast(out.ptr()), reinterpret_cast(in.ptr()), _Nx, N); - }, - in, out); + const unsigned int N = _input->info()->dimension(0); + const unsigned int M = _input->info()->dimension(1); + execute_window_loop(input_window, [&](const Coordinates &) + { + _func_1(reinterpret_cast(out.ptr()), reinterpret_cast(in.ptr()), _Nx, NxRadix, w_m, N, M); + }, + in, out); + } ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); diff --git a/src/core/NEON/kernels/NEFFTScaleKernel.cpp b/src/core/NEON/kernels/NEFFTScaleKernel.cpp new file mode 100644 index 0000000000..6568755e5d --- /dev/null +++ b/src/core/NEON/kernels/NEFFTScaleKernel.cpp @@ -0,0 +1,136 @@ +/* + * Copyright (c) 2019 ARM Limited. + * + * SPDX-License-Identifier: MIT + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to + * deal in the Software without restriction, including without limitation the + * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or + * sell copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in all + * copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE + * SOFTWARE. + */ +#include "arm_compute/core/NEON/kernels/NEFFTScaleKernel.h" + +#include "arm_compute/core/ITensor.h" +#include "arm_compute/core/NEON/wrapper/wrapper.h" +#include "arm_compute/core/TensorInfo.h" +#include "arm_compute/core/Types.h" +#include "arm_compute/core/Validate.h" +#include "arm_compute/core/Window.h" + +#include + +namespace arm_compute +{ +namespace +{ +void scale_complex(float *c_in, float *c_out, bool is_conjugate, float scale) +{ + const auto a = wrapper::vload(c_in); + auto b = wrapper::vdiv(a, float32x2_t{ scale, scale }); + if(is_conjugate) + { + const float img_part = wrapper::vgetlane(b, 1); + b = wrapper::vsetlane(-img_part, b, 1); + } + + wrapper::vstore(c_out, b); +} + +Status validate_arguments(const ITensorInfo *input, const ITensorInfo *output) +{ + ARM_COMPUTE_RETURN_ERROR_ON_DATA_TYPE_CHANNEL_NOT_IN(input, 2, DataType::F32); + + // Checks performed when output is configured + if((output != nullptr) && (output->total_size() != 0)) + { + ARM_COMPUTE_RETURN_ERROR_ON(output->num_channels() != 1 && output->num_channels() != 2); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_SHAPES(input, output); + ARM_COMPUTE_RETURN_ERROR_ON_MISMATCHING_DATA_TYPES(input, output); + } + + return Status{}; +} + +std::pair validate_and_configure_window(ITensorInfo *input, ITensorInfo *output) +{ + // Configure kernel window + Window win = calculate_max_window(*input, Steps()); + + if(output != nullptr) + { + // Output auto inizialitation if not yet initialized + auto_init_if_empty(*output, *input->clone()); + + // NEFFTScaleKernel doesn't need padding so update_window_and_padding() can be skipped + Coordinates coord; + coord.set_num_dimensions(output->num_dimensions()); + output->set_valid_region(ValidRegion(coord, output->tensor_shape())); + } + + return std::make_pair(Status{}, win); +} +} // namespace + +NEFFTScaleKernel::NEFFTScaleKernel() + : _input(nullptr), _output(nullptr), _scale(), _run_in_place(false), _is_conj(false) +{ +} + +void NEFFTScaleKernel::configure(ITensor *input, ITensor *output, const FFTScaleKernelInfo &config) +{ + ARM_COMPUTE_ERROR_ON_NULLPTR(input); + ARM_COMPUTE_ERROR_THROW_ON(validate_arguments(input->info(), (output != nullptr) ? output->info() : nullptr)); + + _input = input; + _output = output; + _run_in_place = (output == nullptr) || (output == input); + _is_conj = config.conjugate; + _scale = config.scale; + + // Configure kernel window + auto win_config = validate_and_configure_window(input->info(), _run_in_place ? nullptr : output->info()); + ARM_COMPUTE_ERROR_THROW_ON(win_config.first); + INEKernel::configure(win_config.second); +} + +Status NEFFTScaleKernel::validate(const ITensorInfo *input, const ITensorInfo *output, const FFTScaleKernelInfo &config) +{ + ARM_COMPUTE_UNUSED(config); + ARM_COMPUTE_RETURN_ON_ERROR(validate_arguments(input, output)); + ARM_COMPUTE_RETURN_ON_ERROR(validate_and_configure_window(input->clone().get(), output->clone().get()).first); + + return Status{}; +} + +void NEFFTScaleKernel::run(const Window &window, const ThreadInfo &info) +{ + ARM_COMPUTE_ERROR_ON_UNCONFIGURED_KERNEL(this); + ARM_COMPUTE_ERROR_ON_INVALID_SUBWINDOW(INEKernel::window(), window); + ARM_COMPUTE_UNUSED(info); + + Window input_window = window; + input_window.set(Window::DimX, 0); + + Iterator in(_input, input_window); + Iterator out(_run_in_place ? _input : _output, input_window); + + execute_window_loop(window, [&](const Coordinates &) + { + scale_complex(reinterpret_cast(out.ptr()), reinterpret_cast(in.ptr()), _is_conj, _scale); + }, + in, out); +} +} // namespace arm_compute -- cgit v1.2.1