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

Fix dx4 calculation in scalar remainder code and add null checks

Browse Source 
- Fix dx4 computation in scalar remainder loops (should be dx2*dx2)
- Add missing null pointer check in benchmark_sinf for consistency

Co-authored-by: wvbbreu <185333235+wvbbreu@users.noreply.github.com>
This commit is contained in:
copilot-swe-agent[bot]
2025-10-29 16:26:27 +00:00
parent 3addf2b05e
commit 807b9d5284
2 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
benchmarks/benchmark_utils.hpp
View File

@@ -26,6 +26,9 @@ static void benchmark_sinf(benchmark::State &state) {
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
float *s =
reinterpret_cast<float *>(backend.allocate_memory(N * sizeof(float)));
if (!x || !s) {
throw std::runtime_error("Buffer allocation failed");
}
auto end = std::chrono::high_resolution_clock::now();
state.counters["init_ms"] =
std::chrono::duration_cast<std::chrono::microseconds>(end - start)

6
src/lookup_xsimd.cpp
View File

@@ -78,7 +78,7 @@ template <std::size_t NR_SAMPLES> struct cosf_dispatcher {
const float dx = a[i] - idx * lookup_table_.PI_FRAC;
const float dx2 = dx * dx;
const float dx3 = dx2 * dx;
const float dx4 = dx3 * dx;
const float dx4 = dx2 * dx2;
const float cosdx =
1.0f - lookup_table_.TERM2 * dx2 + lookup_table_.TERM4 * dx4;
const float sindx = dx - lookup_table_.TERM3 * dx3;
@@ -138,7 +138,7 @@ template <std::size_t NR_SAMPLES> struct sinf_dispatcher {
const float dx = a[i] - idx * lookup_table_.PI_FRAC;
const float dx2 = dx * dx;
const float dx3 = dx2 * dx;
const float dx4 = dx3 * dx;
const float dx4 = dx2 * dx2;
const float cosdx =
1.0f - lookup_table_.TERM2 * dx2 + lookup_table_.TERM4 * dx4;
const float sindx = dx - lookup_table_.TERM3 * dx3;
@@ -202,7 +202,7 @@ template <std::size_t NR_SAMPLES> struct sin_cosf_dispatcher {
const float dx = a[i] - idx * lookup_table_.PI_FRAC;
const float dx2 = dx * dx;
const float dx3 = dx2 * dx;
const float dx4 = dx3 * dx;
const float dx4 = dx2 * dx2;
const float cosdx =
1.0f - lookup_table_.TERM2 * dx2 + lookup_table_.TERM4 * dx4;
const float sindx = dx - lookup_table_.TERM3 * dx3;