| 1189 | // Helper: time 4-wide scalar s8x8 with a binary op |
| 1190 | template <typename Op> |
| 1191 | inline int64_t benchS8x8_4(int iters, Op op) { |
| 1192 | fl::s8x8 a0(1.5f), a1(2.3f), a2(0.7f), a3(3.1f); |
| 1193 | fl::s8x8 b0(0.5f), b1(1.2f), b2(2.0f), b3(0.9f); |
| 1194 | fl::s8x8 bump = fl::s8x8::from_raw(1); |
| 1195 | uint32_t t0 = micros(); |
| 1196 | for (int i = 0; i < iters; i++) { |
| 1197 | a0 = op(a0, b0); a1 = op(a1, b1); |
| 1198 | a2 = op(a2, b2); a3 = op(a3, b3); |
| 1199 | b0 = a0 + bump; b1 = a1 + bump; |
| 1200 | b2 = a2 + bump; b3 = a3 + bump; |
| 1201 | } |
| 1202 | uint32_t t1 = micros(); |
| 1203 | g_bench_sink = static_cast<uint32_t>(a0.raw()); |
| 1204 | return static_cast<int64_t>(t1 - t0); |
| 1205 | } |
| 1206 | |
| 1207 | // Helper: time 4-wide scalar u16x16 with a binary op |
| 1208 | template <typename Op> |
no test coverage detected