Gitea Mirror Bot
50 lines
1.6 KiB
C++
50 lines
1.6 KiB
C++
#include <wasm_simd128.h>
|
|
|
|
void multiply_arrays(int* out, int* in_a, int* in_b, int size) {
|
|
for (int i = 0; i < size; i += 4) {
|
|
v128_t a = wasm_v128_load(&in_a[i]);
|
|
v128_t b = wasm_v128_load(&in_b[i]);
|
|
v128_t prod = wasm_i32x4_mul(a, b);
|
|
wasm_v128_store(&out[i], prod);
|
|
}
|
|
}
|
|
|
|
#ifdef __wasm_relaxed_simd__
|
|
|
|
float dot_product_accumulate(const float* A, const float* B, int len, float initial_sum) {
|
|
int i = 0;
|
|
|
|
// Initialize accumulator vector with the initial sum spread across all lanes
|
|
// We will sum these lanes at the end.
|
|
v128_t acc = wasm_f32x4_splat(initial_sum);
|
|
|
|
// Main loop: process 4 elements at a time
|
|
// Requires len to be at least 4. If len < 4, the loop is skipped.
|
|
for (; i + 4 <= len; i += 4) {
|
|
// Load 4 floats from A and B
|
|
v128_t va = wasm_v128_load(&A[i]);
|
|
v128_t vb = wasm_v128_load(&B[i]);
|
|
|
|
// Fused Multiply-Add: acc = (va * vb) + acc
|
|
// Uses relaxed rounding for performance
|
|
acc = __builtin_wasm_relaxed_madd_f32x4(va, vb, acc);
|
|
}
|
|
|
|
// Horizontal add: sum the 4 lanes of the accumulator
|
|
// acc = {s0, s1, s2, s3}
|
|
// sum = s0 + s1 + s2 + s3
|
|
v128_t sum1 = wasm_f32x4_add(acc, wasm_i32x4_shuffle(acc, acc, 1, 0, 3, 2)); // {s0+s1, s0+s1, s2+s3, s2+s3}
|
|
v128_t sum2 = wasm_f32x4_add(sum1, wasm_i32x4_shuffle(sum1, sum1, 2, 3, 0, 1)); // {total, total, total, total}
|
|
|
|
float result = wasm_f32x4_extract_lane(sum2, 0);
|
|
|
|
// Handle remaining elements (tail loop)
|
|
for (; i < len; i++) {
|
|
result += A[i] * B[i];
|
|
}
|
|
|
|
return result;
|
|
}
|
|
|
|
#endif
|