mirror/trigdx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Update GPU memory management

Browse Source 
- Move device memory allocation for output out of init
- Copy directly from device memory to host pointers
This commit is contained in:
Bram Veenboer
2025-09-02 09:33:36 +02:00
parent 9d3af8c202
commit 716f323b26
1 changed files with 15 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
src/gpu.cpp
View File

@@ -14,68 +14,55 @@ struct GPUBackend::Impl {
if (h_x) { if (h_x) {
cudaFreeHost(h_x); cudaFreeHost(h_x);
} }
if (h_s) {
cudaFreeHost(h_s);
}
if (h_c) {
cudaFreeHost(h_c);
}
if (d_x) { if (d_x) {
cudaFree(d_x); cudaFree(d_x);
} }
if (d_s) {
cudaFree(d_s);
}
if (d_c) {
cudaFree(d_c);
}
} }
void init(size_t n) { void init(size_t n) {
const size_t bytes = n * sizeof(float); const size_t bytes = n * sizeof(float);
cudaMallocHost(&h_x, bytes); cudaMallocHost(&h_x, bytes);
cudaMallocHost(&h_s, bytes);
cudaMallocHost(&h_c, bytes);
cudaMalloc(&d_x, bytes); cudaMalloc(&d_x, bytes);
cudaMalloc(&d_s, bytes);
cudaMalloc(&d_c, bytes);
} }
void compute_sinf(size_t n, const float *x, float *s) const { void compute_sinf(size_t n, const float *x, float *s) const {
const size_t bytes = n * sizeof(float); const size_t bytes = n * sizeof(float);
std::memcpy(h_x, x, bytes); std::memcpy(h_x, x, bytes);
float *d_s;
cudaMalloc(&d_s, bytes);
cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice); cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice);
launch_sinf_kernel(d_x, d_s, n); launch_sinf_kernel(d_x, d_s, n);
cudaMemcpy(h_s, d_s, bytes, cudaMemcpyDeviceToHost); cudaMemcpy(s, d_s, bytes, cudaMemcpyDeviceToHost);
std::memcpy(s, h_s, bytes); cudaFree(d_s);
} }
void compute_cosf(size_t n, const float *x, float *c) const { void compute_cosf(size_t n, const float *x, float *c) const {
const size_t bytes = n * sizeof(float); const size_t bytes = n * sizeof(float);
std::memcpy(h_x, x, bytes); std::memcpy(h_x, x, bytes);
float *d_c;
cudaMalloc(&d_c, bytes);
cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice); cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice);
launch_cosf_kernel(d_x, d_c, n); launch_cosf_kernel(d_x, d_c, n);
cudaMemcpy(h_c, d_c, bytes, cudaMemcpyDeviceToHost); cudaMemcpy(c, d_c, bytes, cudaMemcpyDeviceToHost);
std::memcpy(c, h_c, bytes); cudaFree(d_c);
} }
void compute_sincosf(size_t n, const float *x, float *s, float *c) const { void compute_sincosf(size_t n, const float *x, float *s, float *c) const {
const size_t bytes = n * sizeof(float); const size_t bytes = n * sizeof(float);
std::memcpy(h_x, x, bytes); std::memcpy(h_x, x, bytes);
float *d_s, *d_c;
cudaMalloc(&d_s, bytes);
cudaMalloc(&d_c, bytes);
cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice); cudaMemcpy(d_x, h_x, bytes, cudaMemcpyHostToDevice);
launch_sincosf_kernel(d_x, d_s, d_c, n); launch_sincosf_kernel(d_x, d_s, d_c, n);
cudaMemcpy(h_s, d_s, bytes, cudaMemcpyDeviceToHost); cudaMemcpy(s, d_s, bytes, cudaMemcpyDeviceToHost);
cudaMemcpy(h_c, d_c, bytes, cudaMemcpyDeviceToHost); cudaMemcpy(c, d_c, bytes, cudaMemcpyDeviceToHost);
std::memcpy(s, h_s, bytes); cudaFree(d_s);
std::memcpy(c, h_c, bytes); cudaFree(d_c);
} }
float *h_x = nullptr; float *h_x = nullptr;
float *h_s = nullptr;
float *h_c = nullptr;
float *d_x = nullptr; float *d_x = nullptr;
float *d_s = nullptr;
float *d_c = nullptr;
}; };
GPUBackend::GPUBackend() : impl(std::make_unique<Impl>()) {} GPUBackend::GPUBackend() : impl(std::make_unique<Impl>()) {}