`i)QdZddlmZddlZgdZdZejedgdZd Zd Z d Z d Z d Z Gdde Z ddZdS)a upfirdn implementation. Functions defined here were ported directly from cuSignal under terms of the MIT license, under the following notice: Copyright (c) 2019-2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved. 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. )ceilN) constantwrapedgesmooth symmetricreflect antisymmetric antireflectlinea 0 #include /////////////////////////////////////////////////////////////////////////////// // UPFIRDN1D // /////////////////////////////////////////////////////////////////////////////// template __device__ void _cupy_upfirdn1D( const T *__restrict__ inp, const T *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, T *__restrict__ out, const int outW ) { const int t { static_cast( blockIdx.x * blockDim.x + threadIdx.x ) }; const int stride { static_cast( blockDim.x * gridDim.x ) }; for ( size_t tid = t; tid < outW; tid += stride ) { __builtin_assume( padded_len > 0 ); __builtin_assume( up > 0 ); __builtin_assume( down > 0 ); __builtin_assume( tid > 0 ); const int x_idx { static_cast( ( tid * down ) / up ) % padded_len }; int h_idx { static_cast( ( tid * down ) % up * h_per_phase ) }; int x_conv_idx { x_idx - h_per_phase + 1 }; if ( x_conv_idx < 0 ) { h_idx -= x_conv_idx; x_conv_idx = 0; } T temp {}; int stop = ( x_shape_a < ( x_idx + 1 ) ) ? x_shape_a : ( x_idx + 1 ); for ( int x_c = x_conv_idx; x_c < stop; x_c++ ) { temp += inp[x_c] * h_trans_flip[h_idx]; h_idx += 1; } out[tid] = temp; } } extern "C" __global__ void __launch_bounds__( 512 ) _cupy_upfirdn1D_float32( const float *__restrict__ inp, const float *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, float *__restrict__ out, const int outW ) { _cupy_upfirdn1D( inp, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW ); } extern "C" __global__ void __launch_bounds__( 512 ) _cupy_upfirdn1D_float64( const double *__restrict__ inp, const double *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, double *__restrict__ out, const int outW ) { _cupy_upfirdn1D( inp, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW ); } extern "C" __global__ void __launch_bounds__( 512 ) _cupy_upfirdn1D_complex64( const thrust::complex *__restrict__ inp, const thrust::complex *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, thrust::complex *__restrict__ out, const int outW ) { _cupy_upfirdn1D>( inp, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW ); } extern "C" __global__ void __launch_bounds__( 512 ) _cupy_upfirdn1D_complex128( const thrust::complex *__restrict__ inp, const thrust::complex *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, thrust::complex *__restrict__ out, const int outW ) { _cupy_upfirdn1D>( inp, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW ); } /////////////////////////////////////////////////////////////////////////////// // UPFIRDN2D // /////////////////////////////////////////////////////////////////////////////// template __device__ void _cupy_upfirdn2D( const T *__restrict__ inp, const int inpH, const T *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, T *__restrict__ out, const int outW, const int outH ) { const int ty { static_cast( blockIdx.x * blockDim.x + threadIdx.x ) }; const int tx { static_cast( blockIdx.y * blockDim.y + threadIdx.y ) }; const int stride_y { static_cast( blockDim.x * gridDim.x ) }; const int stride_x { static_cast( blockDim.y * gridDim.y ) }; for ( int x = tx; x < outH; x += stride_x ) { for ( int y = ty; y < outW; y += stride_y ) { int x_idx {}; int h_idx {}; __builtin_assume( padded_len > 0 ); __builtin_assume( up > 0 ); __builtin_assume( down > 0 ); if ( axis == 1 ) { __builtin_assume( x > 0 ); x_idx = ( static_cast( x * down ) / up ) % padded_len; h_idx = ( x * down ) % up * h_per_phase; } else { __builtin_assume( y > 0 ); x_idx = ( static_cast( y * down ) / up ) % padded_len; h_idx = ( y * down ) % up * h_per_phase; } int x_conv_idx { x_idx - h_per_phase + 1 }; if ( x_conv_idx < 0 ) { h_idx -= x_conv_idx; x_conv_idx = 0; } T temp {}; int stop = ( x_shape_a < ( x_idx + 1 ) ) ? x_shape_a : ( x_idx + 1 ); for ( int x_c = x_conv_idx; x_c < stop; x_c++ ) { if ( axis == 1 ) { temp += inp[y * inpH + x_c] * h_trans_flip[h_idx]; } else { temp += inp[x_c * inpH + x] * h_trans_flip[h_idx]; } h_idx += 1; } out[y * outH + x] = temp; } } } extern "C" __global__ void __launch_bounds__( 64 ) _cupy_upfirdn2D_float32( const float *__restrict__ inp, const int inpH, const float *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, float *__restrict__ out, const int outW, const int outH ) { _cupy_upfirdn2D( inp, inpH, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW, outH ); } extern "C" __global__ void _cupy_upfirdn2D_float64( const double *__restrict__ inp, const int inpH, const double *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, double *__restrict__ out, const int outW, const int outH ) { _cupy_upfirdn2D( inp, inpH, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW, outH ); } extern "C" __global__ void __launch_bounds__( 64 ) _cupy_upfirdn2D_complex64( const thrust::complex *__restrict__ inp, const int inpH, const thrust::complex *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, thrust::complex *__restrict__ out, const int outW, const int outH ) { _cupy_upfirdn2D>( inp, inpH, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW, outH ); } extern "C" __global__ void __launch_bounds__( 64 ) _cupy_upfirdn2D_complex128( const thrust::complex *__restrict__ inp, const int inpH, const thrust::complex *__restrict__ h_trans_flip, const int up, const int down, const int axis, const int x_shape_a, const int h_per_phase, const int padded_len, thrust::complex *__restrict__ out, const int outW, const int outH ) { _cupy_upfirdn2D>( inp, inpH, h_trans_flip, up, down, axis, x_shape_a, h_per_phase, padded_len, out, outW, outH ); } )z -std=c++11)_cupy_upfirdn1D_float32_cupy_upfirdn1D_float64_cupy_upfirdn1D_complex64_cupy_upfirdn1D_complex128_cupy_upfirdn2D_float32_cupy_upfirdn2D_float64_cupy_upfirdn2D_complex64_cupy_upfirdn2D_complex128)codeoptionsname_expressionsct|t| |zz}tj||j}||dt|<|d|jdddddf}|S)aStore coefficients in a transposed, flipped arrangement. For example, suppose upRate is 3, and the input number of coefficients is 10, represented as h[0], ..., h[9]. Then the internal buffer will look like this:: h[9], h[6], h[3], h[0], // flipped phase 0 coefs 0, h[7], h[4], h[1], // flipped phase 1 coefs (zero-padded) 0, h[8], h[5], h[2], // flipped phase 2 coefs (zero-padded) N)lencupyzerosdtypereshapeTravel)huph_padlenh_fulls o/home/jaya/work/projects/VOICE-AGENT/VIET/agent-env/lib/python3.11/site-packages/cupyx/scipy/signal/_upfirdn.py_pad_hr&$s1vv#a&&2&H Z!' * *FF8SVV8 ^^B # # %aaa2g . 4 4 6 6F Mc*|dz |z|zdz |zdzS)N)len_hin_lenr"downs r% _output_lenr.4s%qjB  &! + 4q 88r'cXtj}|jdS)N MaxGridDimXrcudaDevice attributes device_ids r% _get_max_gdxr7:#   ""I   ..r'cXtj}|jdS)N MaxGridDimYr1r5s r% _get_max_gdyr;?r8r'cntj}|jd}d}|dz}||fS)NMultiProcessorCountir1)r6numSMthreadsperblock blockspergrids r% _get_tpb_bpgrBDs<   ""I  !6 7EOBJM M ))r'ceZdZdZdZdS)_UpFIRDncntj|}|jdks |jdkrt dtj|j|tj|_tj||j}t||_ t||_ |j dks |j dkrt dt||j |_ tj|j |_ tj|j |_ t||_dS)zHelper for resamplingr)rz!h must be 1D with non-zero lengthzBoth up and down must be >= 1N)rasarrayndimsize ValueError result_typerfloat32 _output_typeint_up_downr& _h_trans_flipascontiguousarrayr _h_len_orig)selfr!x_dtyper"r-s r%__init__z_UpFIRDn.__init__Ns LOO 6Q;;!&A++@AA A ,QWgt|LL LD- . .r77YY 8a<<4:>><== =#Atx00!\$*<==!3D4FGGq66r'c tj||j}t|j|j||j|j}t|j}|||<tj ||jd}||j z}|j|}t|j |jz}|j|t|j |jzzdz }|j dkrt\} } td|jj} | |jddzdz dzfd||j |j|j||||||jdf n|j dkrd } t%|jd| dz } | t'kr| n t'} t%|jd| dz } | t)kr| n t)}| |f} td |jj} | | | ||jd|j |j|j||||||jd|jdf nt+d |S) z@Apply the prepared filter to the specified axis of a nD signal xC)rorderr)_cupy_upfirdn1D_r)rZ)r\_cupy_upfirdn2D_zupfirdn() requires ndim <= 2)rrFrLr.rRshaperNrOlistemptyrGrrPrBUPFIRDN_MODULE get_functionrnamerr7r;NotImplementedError)rSxaxis output_len output_shapeout x_shape_a h_per_phase padded_lenr@rAkernelblocksblockspergrid_xblockspergrid_ys r% apply_filterz_UpFIRDn.apply_filter`s LD- . .  agdmTXtzCC AG}} ' TjT->cJJJaf}GDM $,--9 WT]c$*<&=&=&IJQN 8q==-9^^ *O]#00339>3355F FQWQZ#%)c13V&HJIaL     X]]$O#)A,);;<>11|~~ #)A,);;<>11|~~ -o>M$00339>3355F F?MGAJ&HJIaLIaL      &&DEE E r'N)__name__ __module__ __qualname__rUrqr*r'r%rDrDMs7"""$KKKKKr'rDr)rrc|d}|dks|dkrtd|d|dt||jt|t|}|||S)a Upsample, FIR filter, and downsample. Parameters ---------- h : array_like 1-dimensional FIR (finite-impulse response) filter coefficients. x : array_like Input signal array. up : int, optional Upsampling rate. Default is 1. down : int, optional Downsampling rate. Default is 1. axis : int, optional The axis of the input data array along which to apply the linear filter. The filter is applied to each subarray along this axis. Default is -1. mode : str, optional This parameter is not implemented for values other than ``"constant"``. cval : float, optional This parameter is not implemented for values other than 0. Returns ------- y : ndarray The output signal array. Dimensions will be the same as `x` except for along `axis`, which will change size according to the `h`, `up`, and `down` parameters. Notes ----- The algorithm is an implementation of the block diagram shown on page 129 of the Vaidyanathan text [1]_ (Figure 4.3-8d). The direct approach of upsampling by factor of P with zero insertion, FIR filtering of length ``N``, and downsampling by factor of Q is O(N*Q) per output sample. The polyphase implementation used here is O(N/P). See Also -------- scipy.signal.upfirdn References ---------- .. [1] P. P. Vaidyanathan, Multirate Systems and Filter Banks, Prentice Hall, 1993. Nrrzmode=z and cval=z not implemented.)rdrDrrMrq)r!rer"r-rfmodecvalufds r%upfirdnrys{r | zTQYY!"IT"I"I"I"I"IJJJ 1ags2wwD 2 2C   At $ $$r')r)r)rrr)__doc__mathrr_upfirdn_modesUPFIRDN_KERNEL RawModulerar&r.r7r;rBobjectrDryr*r'r%rs'8  m`            999 /// /// ***^^^^^v^^^H  @%@%@%@%@%@%r'