embedsom_cuda.cpp

Go to the documentation of this file.

/*
The MIT License
 
Copyright (c) 2021 Martin Krulis
                   Mirek Kratochvil
                   Sona Molnarova
 
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 "embedsom_cuda.h"
 
#include "cuda_runtime.h"
 
void
EmbedSOMCUDAContext::run(size_t n,
                         size_t g,
                         size_t d,
                         float boost,
                         size_t k,
                         float adjust,
                         const float *hidim_points,
                         const float *hidim_landmarks,
                         const float *lodim_landmarks,
                         float *lodim_points)
{
 
    /* first make sure all sizes are acceptable and there's actual sensible
     * work that can be done */
    if (!n)
        return;
    if (d < 2 || d > 256)
        return; // TODO try d=1; d>256 should warn
    if (k > g)
        k = g;
    if (k < 3)
        return;
 
    /* make sure the buffers are sufficiently big. The allocations are kept as
     * such: |data| = d*n |points| = 2*n |lm_hi| = d*g |lm_lo| = 2*g |topk| =
     * adjusted_k()*n
     *
     * Because `n` and `g` is highly variable in this usecase, we _only_
     * _increase_ the buffer sizes. Reallocation in case the sizes grow totally
     * out of limits would help too (feel free to add into the conditions).
     */
 
    CUCH(cudaSetDevice(0));
 
#define size_check(tgt, cur, buffer, type)                                     \
    if (tgt > cur) {                                                           \
        if (buffer)                                                            \
            cudaFree(buffer);                                                  \
        CUCH(cudaMalloc(&buffer, tgt * sizeof(type)));                         \
        cur = tgt;                                                             \
    }
 
    size_check(d * n, ndata, data, float);
    size_check(2 * n, npoints, points, float);
    size_check(d * g, nlm_hi, lm_hi, float);
    size_check(2 * g, nlm_lo, lm_lo, float);
 
    size_t adjusted_k = std::min(g, k + 1);
    size_check(adjusted_k * n, nknns, knns, knn_entry<float>);
    CUCH(cudaGetLastError());
#undef size_check
 
    /* all is safely allocated, transfer the data! */
    CUCH(cudaMemcpy(
      data, hidim_points, d * n * sizeof(float), cudaMemcpyHostToDevice));
    CUCH(cudaMemcpy(
      lm_hi, hidim_landmarks, d * g * sizeof(float), cudaMemcpyHostToDevice));
    CUCH(cudaMemcpy(
      lm_lo, lodim_landmarks, 2 * g * sizeof(float), cudaMemcpyHostToDevice));
    CUCH(cudaGetLastError());
 
    /* run the main embedding */
    runKNNKernel(d, n, g, adjusted_k);
    runProjectionKernel(d, n, g, k, boost, adjust);
 
    /* get the results */
    CUCH(cudaMemcpy(
      lodim_points, points, 2 * n * sizeof(float), cudaMemcpyDeviceToHost));
    CUCH(cudaGetLastError());
}
 
EmbedSOMCUDAContext::~EmbedSOMCUDAContext()
{
    if (data)
        cudaFree(data);
    if (points)
        cudaFree(points);
    if (lm_hi)
        cudaFree(lm_hi);
    if (lm_lo)
        cudaFree(lm_lo);
    if (knns)
        cudaFree(knns);
}