// 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 #include // Table of exp2(k / 64) values decremented (as integer) by (k << 17), k = 0..63 extern XNN_INTERNAL const float xnn_table_exp2minus_k_over_64[64]; void xnn_math_f32_sigmoid__avx_rr2_lut64_p2_div( size_t n, const float* input, float* output) { assert(n % (8 * sizeof(float)) == 0); // Floating-point mask with only the sign bit set const __m256 vsign_mask = _mm256_set1_ps(-0.0f); // Large number such that ulp(magic bias) == exp2(-6) const __m256 vmagic_bias = _mm256_set1_ps(0x1.800000p17f); const __m256 vlog2e = _mm256_set1_ps(0x1.715476p0f); // Mask for the lowest 6 bits const __m256 vindex_mask = _mm256_castsi256_ps(_mm256_set1_epi32(INT32_C(0x3F))); // Last 13 bits are zeroes const __m256 vminus_ln2_hi = _mm256_set1_ps(-0x1.630000p-1f); const __m256 vminus_ln2_lo = _mm256_set1_ps(0x1.BD0106p-13f); // Coefficient of polynomial approximation of exp(t) ~ 1 + t * (1 + t * c2) on [-log(2)/128, log(2)/128] const __m256 vc2 = _mm256_set1_ps(0x1.FFFF0Ap-2f); const __m256 vone = _mm256_set1_ps(1.0f); // The smallest x for which sigmoidf(x) is normalized. // This number is also the smallest x for which expf(x) is normalized. const __m256 vdenorm_cutoff = _mm256_set1_ps(-0x1.5D589Ep+6f); for (; n != 0; n -= 8 * sizeof(float)) { const __m256 vx = _mm256_loadu_ps(input); // General structure of the algorithm: // // / exp(x) / (1 + exp(x)) if x <= 0 // f[x] := // \ 1 - f[-x] if x >= 0 // // First we compute f[z] := exp(z) / (1 + exp(z)) where z = -abs(x), then replace result with 1 - f[z] if x >= 0. const __m256 vz = _mm256_or_ps(vx, vsign_mask); // Compute reduced argument n := round(z / log(2), 6). // We do it by adding a large number (magic bias), which cause rounding of the result to 6 fractional bits, then // subtracing the large number back. The trick with adding large number is valid only within certain bounds // (|z / log(2)| <= 2**16, i.e. |z| <= 0x1.62E43p+15 = 45426.09375), but that is acceptable, because inputs x // outside of [-87.336544, 17.328678] (i.e. z outsize [87.336544, 0]) underflow or saturate sigmoidf(x). We fixup // the result for such inputs at the very end of the algorithm. __m256 vn = _mm256_add_ps(_mm256_mul_ps(vz, vlog2e), vmagic_bias); // Create a floating-point number s (scale) such that s := 2**n for such inputs that sigmoidf(z) is normalized, // i.e. -87.33642 <= z <= 0. As n has 6 fractional bits, we split s == 2**n = 2**int(n) * 2**frac(n). We create s // in two steps: // 1. Fetch 2**frac(n) from the table using the 6 low bits of n, as integer. Note that the fetched values are in // the [1.0, 2.0) range, i.e. their floating-point exponent is 0. // 2. Adjust fecthed value by addition of int(n) to its floating-point exponent. The result is always a normalized // number, because for -87.33642 <= z <= 0 (inputs for which sigmoidf(z) is normalized) we have // -126 <= int(n) <= 0, and thus the adjusted exponent is not lower than -126. // // Shift bits 6:14 into 23:31 (position of floating-point exponent). __m128i ve_lo = _mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vn)), 17); __m128i ve_hi = _mm_slli_epi32(_mm_castps_si128(_mm256_extractf128_ps(vn, 1)), 17); // Use bits 0:6 of n, as integer, as an index for table lookup of l := 2**frac(n). const __m256 vidx = _mm256_and_ps(vn, vindex_mask); const __m128i vidx_lo = _mm_slli_epi32(_mm_castps_si128(_mm256_castps256_ps128(vidx)), 2); const __m128i vidx_hi = _mm_slli_epi32(_mm_castps_si128(_mm256_extractf128_ps(vidx, 1)), 2); #if XNN_ARCH_X86_64 const uint64_t vidx_ll = (uint64_t) _mm_cvtsi128_si64(vidx_lo); const uint64_t vidx_lh = (uint64_t) _mm_extract_epi64(vidx_lo, 1); const uint64_t vidx_hl = (uint64_t) _mm_cvtsi128_si64(vidx_hi); const uint64_t vidx_hh = (uint64_t) _mm_extract_epi64(vidx_hi, 1); __m128i vl_ll = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) vidx_ll))); __m128i vl_lh = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) vidx_lh))); __m128i vl_hl = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) vidx_hl))); __m128i vl_hh = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) vidx_hh))); vl_ll = _mm_insert_epi32(vl_ll, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) (vidx_ll >> 32))), 1); vl_lh = _mm_insert_epi32(vl_lh, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) (vidx_lh >> 32))), 1); vl_hl = _mm_insert_epi32(vl_hl, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) (vidx_hl >> 32))), 1); vl_hh = _mm_insert_epi32(vl_hh, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) (vidx_hh >> 32))), 1); #else __m128i vl_ll = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_cvtsi128_si32(vidx_lo)))); __m128i vl_lh = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_lo, 2)))); __m128i vl_hl = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_cvtsi128_si32(vidx_hi)))); __m128i vl_hh = _mm_cvtsi32_si128(*((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_hi, 2)))); vl_ll = _mm_insert_epi32(vl_ll, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_lo, 1))), 1); vl_lh = _mm_insert_epi32(vl_lh, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_lo, 3))), 1); vl_hl = _mm_insert_epi32(vl_hl, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_hi, 1))), 1); vl_hh = _mm_insert_epi32(vl_hh, *((const int*) ((uintptr_t) xnn_table_exp2minus_k_over_64 + (uint32_t) _mm_extract_epi32(vidx_hi, 3))), 1); #endif const __m128i vl_lo = _mm_unpacklo_epi64(vl_ll, vl_lh); const __m128i vl_hi = _mm_unpacklo_epi64(vl_hl, vl_hh); // Adjust exponent of the value l fetched from the table to get the final s value. const __m128 vs_lo = _mm_castsi128_ps(_mm_add_epi32(vl_lo, ve_lo)); const __m128 vs_hi = _mm_castsi128_ps(_mm_add_epi32(vl_hi, ve_hi)); const __m256 vs = _mm256_insertf128_ps(_mm256_castps128_ps256(vs_lo), vs_hi, 1); // Subtract the large number back to get the final n := round(z / log(2), 6) as a floating-point number. vn = _mm256_sub_ps(vn, vmagic_bias); // Compute reduced argument t := z - n * log(2). // Use Cody-Waite range reduction method (note two constants to represent log(2)) to improve accuracy. __m256 vt = _mm256_add_ps(_mm256_mul_ps(vn, vminus_ln2_hi), vz); vt = _mm256_add_ps(_mm256_mul_ps(vn, vminus_ln2_lo), vt); // Compute degree-2 polynomial approximation for exp(t) on [-log(2)/128, log(2)/128]. // P(t) = 1 + t * (1 + t * c2) = 1 + (t + t * (t * c2)) = 1 + p __m256 vp = _mm256_mul_ps(vt, vc2); vp = _mm256_add_ps(vt, _mm256_mul_ps(vp, vt)); // Reconstruct the exp(z) value: // e = s * (1 + t * (1 + t * c2)) // = s * (1 + p) // = s + s * p const __m256 ve = _mm256_add_ps(vs, _mm256_mul_ps(vs, vp)); // Denominator of the sigmoid fraction: 1.0 + exp(z) const __m256 vd = _mm256_add_ps(ve, vone); // Reconstruct sigmoid(z) = exp(z) / (1.0 + exp(z)) __m256 vf = _mm256_div_ps(ve, vd); // For inputs below denormal cutoff, replace output with +0.0f. // Note that for NaN inputs, comparison result is false, and outputs are left unchanged. vf = _mm256_andnot_ps(_mm256_cmp_ps(vz, vdenorm_cutoff, _CMP_LT_OS), vf); // Reconstruct sigmoid(x) = x < 0 ? sigmoid(z) : 1.0 - sigmoid(z) vf = _mm256_blendv_ps(_mm256_sub_ps(vone, vf), vf, vx); _mm256_storeu_ps(output, vf); input += 8; output += 8; } }