// Copyright 2020 Google LLC // // This source code is licensed under the BSD-style license found in the // LICENSE file in the root directory of this source tree. #include #include #include #include void xnn_math_f32_expm1minus__avx512f_rr1_p6( size_t n, const float* input, float* output) { assert(n % (16 * sizeof(float)) == 0); // The largest x for which expm1f(x) is saturated at -1.0f. const __m512 vsat_cutoff = _mm512_set1_ps(-0x1.154246p+4f); // Large number such that ulp(magic bias) == 1 and magic bias === 127 mod 2**22. const __m512 vmagic_bias = _mm512_set1_ps(0x1.8000FEp23f); const __m512 vlog2e = _mm512_set1_ps(0x1.715476p+0f); const __m512 vminus_ln2 = _mm512_set1_ps(-0x1.62E43p-1f); // Coefficient of polynomial approximation // exp(t) - 1 ~ t * (1 + t * (c2 + t * (c3 + t * (c4 + t * (c5 + t * c6))))) // on [-log(2)/2, log(2)/2] const __m512 vc6 = _mm512_set1_ps(0x1.6b7338p-10f); const __m512 vc5 = _mm512_set1_ps(0x1.12278Ep-7f); const __m512 vc4 = _mm512_set1_ps(0x1.555716p-5f); const __m512 vc3 = _mm512_set1_ps(0x1.5554B0p-3f); const __m512 vc2 = _mm512_set1_ps(0x1.FFFFFEp-2f); const __m512 vone = _mm512_set1_ps(1.0f); for (; n != 0; n -= 16 * sizeof(float)) { __m512 vx = _mm512_loadu_ps(input); // The function saturates at -1 for large negative inputs: expm1f(x) == -1.0f for x <= sat_cutoff ~= -17.328680. // To guarantee this behaviour, we clip input at sat_cutoff, and leverage the fact that for our implementation // expm1f(sat_cutoff) == -1.0f. The order of operands in the [V]MAXPS instruction matters: it ensures that NaN // inputs are passed unchanged. vx = _mm512_max_ps(vsat_cutoff, vx); // Compute reduced argument n := round(x / log(2)). // We do it by adding a large number (magic bias), which cause rounding of the result to integer, then subtracing // the large number back. The first addition is combined with multiplication by log2e into a single FMA // instruction. The trick with adding large number is valid only within certain bounds (|x / log(2)| <= 2**22, // i.e. |x| <= 0x1.62E43p+21 = 2907270.0), but that is acceptable, because inputs x are restricted to // [-17.328680, 0]. // Note that addition-subtraction of the large number doesn't cause overflow for inputs in this range. __m512 vn = _mm512_fmadd_ps(vx, vlog2e, vmagic_bias); // Create a floating-point number s (scale) such that s == 2**n for valid inputs, i.e. // -17.328680 <= x <= 0.0, and -25 <= n <= 0 accordingly. // For NaN inputs, s would have zero mantissa and can have arbitrary sign and exponent, depending on the input // NaN payload. In these cases, n and t are NaNs with the same payload as input while s is non-NaN, and thus // input payload would be propagated in all computations. const __m512 vs = _mm512_castsi512_ps(_mm512_slli_epi32(_mm512_castps_si512(vn), 23)); // Subtract the large number back to get final n := round(x / log(2)). vn = _mm512_sub_ps(vn, vmagic_bias); // Compute reduced argument t := x - n * log(2). __m512 vt = _mm512_fmadd_ps(vn, vminus_ln2, vx); // Compute degree-6 polynomial approximation for exp(t) - 1 on [-log(2)/2, log(2)/2]. // P(t) = t * (1 + t * (c2 + t * (c3 + t * (c4 + t * (c5 + t * c6))))) // = t + t * (t * (c2 + t * (c3 + t * (c4 + t * (c5 + t * c6))))) = t + t * p __m512 vp = _mm512_fmadd_ps(vc6, vt, vc5); vp = _mm512_fmadd_ps(vp, vt, vc4); vp = _mm512_fmadd_ps(vp, vt, vc3); vp = _mm512_fmadd_ps(vp, vt, vc2); vp = _mm512_mul_ps(vp, vt); // Reconstruct the exp(x) - 1 value: // exp(x) - 1 = s * (1 + t * (1 + t * (c2 + t * (c3 + t * (c4 + t * (c5 + t * c6)))))) - 1 // = (s - 1) + s * (t + t * p) // = ((t * s) + (t * s) * p) + (s - 1) vt = _mm512_mul_ps(vt, vs); const __m512 vsm1 = _mm512_sub_ps(vs, vone); vp = _mm512_fmadd_ps(vp, vt, vt); const __m512 vf = _mm512_add_ps(vp, vsm1); _mm512_storeu_ps(output, vf); input += 16; output += 16; } }