.file "atan.s" // Copyright (c) 2000 - 2003, Intel Corporation // All rights reserved. // // Contributed 2000 by the Intel Numerics Group, Intel Corporation // // Redistribution and use in source and binary forms, with or without // modification, are permitted provided that the following conditions are // met: // // * Redistributions of source code must retain the above copyright // notice, this list of conditions and the following disclaimer. // // * Redistributions in binary form must reproduce the above copyright // notice, this list of conditions and the following disclaimer in the // documentation and/or other materials provided with the distribution. // // * The name of Intel Corporation may not be used to endorse or promote // products derived from this software without specific prior written // permission. // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. // // Intel Corporation is the author of this code, and requests that all // problem reports or change requests be submitted to it directly at // http://www.intel.com/software/products/opensource/libraries/num.htm. // // History //============================================================== // 02/02/00 Initial version // 04/13/00 Improved speed // 04/19/00 Removed the qualifying predicate from the fmerge.s that // takes the absolute value. // 06/16/00 Reassigned FP registers to eliminate stalls on loads // 08/30/00 Saved 5 cycles in main path by rearranging large argument logic // and delaying use of result of fcmp in load by 1 group // 05/20/02 Cleaned up namespace and sf0 syntax // 08/20/02 Use atan2 algorithm with x=1 for better accuracy // 02/06/03 Reordered header: .section, .global, .proc, .align // // API //============================================================== // double atan(double Y) // // Overview of operation //============================================================== // // The atan function returns values in the interval [-pi/2,+pi/2]. // // The algorithm used is the atan2(Y,X) algorithm where we fix X=1.0. // // There are two basic paths: swap true and swap false. // atan2(Y,X) ==> atan2(V/U) where U >= V. If Y > X, we must swap. // // p6 swap True |Y| > |X| // p7 swap False |Y| <= |X| // // // Simple trigonometric identities show // Region 1 // |Y|<=1.0, V=Y, U=1.0 atan2(Y,X) = sgnY * (0 + atan(V/U)) // // Region 2 // |Y|>1.0, V=1.0, U=Y atan2(Y,X) = sgnY * (pi/2 - atan(V/U)) // // // We compute atan(V/U) from the identity // atan(z) + atan([(V/U)-z] / [1+(V/U)z]) // where z is a limited precision approximation (16 bits) to V/U // // z is calculated with the assistance of the frcpa instruction. // // atan(z) is calculated by a polynomial z + z^3 * p(w), w=z^2 // where p(w) = P0+P1*w+...+P22*w^22 // // Let d = [(V/U)-z] / [1+(V/U)z]) = (V-U*z)/(U+V*z) // // Approximate atan(d) by d + P0*d^3 // Let F = 1/(U+V*z) * (1-a), where |a|< 2^-8.8. // Compute q(a) = 1 + a + ... + a^5. // Then F*q(a) approximates the reciprocal to more than 50 bits. // Special values //============================================================== // atan(QNAN) = QNAN // atan(SNAN) = quieted SNAN // atan(+-inf) = +- pi/2 // atan(+-0) = +-0 // Registers used //============================================================== // predicate registers used: // p6 -> p15 // floating-point registers used: // f8, input // f32 -> f116 // general registers used // r14 -> r16 // Assembly macros //============================================================== EXP_AD_P1 = r14 EXP_AD_P2 = r15 rsig_near_one = r16 atan2_Y = f8 atan2_X = f1 atan2_u1_X = f32 atan2_u1_Y = f33 atan2_z2_X = f34 atan2_two = f36 atan2_B1sq_Y = f37 atan2_z1_X = f38 atan2_B1X = f40 atan2_B1Y = f41 atan2_wp_X = f42 atan2_B1sq_X = f43 atan2_z = f44 atan2_w = f45 atan2_P0 = f46 atan2_P1 = f47 atan2_P2 = f48 atan2_P3 = f49 atan2_P4 = f50 atan2_P5 = f51 atan2_P6 = f52 atan2_P7 = f53 atan2_P8 = f54 atan2_P9 = f55 atan2_P10 = f56 atan2_P11 = f57 atan2_P12 = f58 atan2_P13 = f59 atan2_P14 = f60 atan2_P15 = f61 atan2_P16 = f62 atan2_P17 = f63 atan2_P18 = f64 atan2_P19 = f65 atan2_P20 = f66 atan2_P21 = f67 atan2_P22 = f68 atan2_pi_by_2 = f69 atan2_sgn_pi_by_2 = f69 atan2_V13 = f70 atan2_W11 = f71 atan2_E = f72 atan2_wp_Y = f73 atan2_V11 = f74 atan2_V12 = f75 atan2_V7 = f76 atan2_V8 = f77 atan2_W7 = f78 atan2_W8 = f79 atan2_W3 = f80 atan2_W4 = f81 atan2_V3 = f82 atan2_V4 = f83 atan2_F = f84 atan2_gV = f85 atan2_V10 = f86 atan2_zcub = f87 atan2_V6 = f88 atan2_V9 = f89 atan2_W10 = f90 atan2_W6 = f91 atan2_W2 = f92 atan2_V2 = f93 atan2_alpha = f94 atan2_alpha_1 = f95 atan2_gVF = f96 atan2_V5 = f97 atan2_W12 = f98 atan2_W5 = f99 atan2_alpha_sq = f100 atan2_Cp = f101 atan2_V1 = f102 atan2_ysq = f103 atan2_W1 = f104 atan2_alpha_cub = f105 atan2_C = f106 atan2_d = f108 atan2_A_hi = f109 atan2_dsq = f110 atan2_pd = f111 atan2_A_lo = f112 atan2_A = f113 atan2_Pp = f114 atan2_sgnY = f115 atan2_sig_near_one = f116 atan2_near_one = f116 ///////////////////////////////////////////////////////////// RODATA .align 16 LOCAL_OBJECT_START(atan2_tb1) data8 0xA21922DC45605EA1 , 0x00003FFA // P11 data8 0xB199DD6D2675C40F , 0x0000BFFA // P10 data8 0xC2F01E5DDD100DBE , 0x00003FFA // P9 data8 0xD78F28FC2A592781 , 0x0000BFFA // P8 data8 0xF0F03ADB3FC930D3 , 0x00003FFA // P7 data8 0x88887EBB209E3543 , 0x0000BFFB // P6 data8 0x9D89D7D55C3287A5 , 0x00003FFB // P5 data8 0xBA2E8B9793955C77 , 0x0000BFFB // P4 data8 0xE38E38E320A8A098 , 0x00003FFB // P3 data8 0x9249249247E37913 , 0x0000BFFC // P2 data8 0xCCCCCCCCCCC906CD , 0x00003FFC // P1 data8 0xAAAAAAAAAAAAA8A9 , 0x0000BFFD // P0 data8 0x0000000000000000 , 0x00000000 // pad to avoid bank conflict LOCAL_OBJECT_END(atan2_tb1) LOCAL_OBJECT_START(atan2_tb2) data8 0xCE585A259BD8374C , 0x00003FF0 // P21 data8 0x9F90FB984D8E39D0 , 0x0000BFF3 // P20 data8 0x9D3436AABE218776 , 0x00003FF5 // P19 data8 0xDEC343E068A6D2A8 , 0x0000BFF6 // P18 data8 0xF396268151CFB11C , 0x00003FF7 // P17 data8 0xD818B4BB43D84BF2 , 0x0000BFF8 // P16 data8 0xA2270D30A90AA220 , 0x00003FF9 // P15 data8 0xD5F4F2182E7A8725 , 0x0000BFF9 // P14 data8 0x80D601879218B53A , 0x00003FFA // P13 data8 0x9297B23CCFFB291F , 0x0000BFFA // P12 data8 0xFE7E52D2A89995B3 , 0x0000BFEC // P22 data8 0xC90FDAA22168C235 , 0x00003FFF // pi/2 LOCAL_OBJECT_END(atan2_tb2) .section .text GLOBAL_LIBM_ENTRY(atan) { .mfi nop.m 999 frcpa.s1 atan2_u1_Y,p7 = f1,atan2_Y nop.i 999 } { .mfi addl EXP_AD_P1 = @ltoff(atan2_tb1), gp fma.s1 atan2_two = f1,f1,f1 nop.i 999 ;; } { .mfi ld8 EXP_AD_P1 = [EXP_AD_P1] frcpa.s1 atan2_u1_X,p6 = f1,atan2_X nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_ysq = atan2_Y,atan2_Y,f0 nop.i 999 } ;; { .mfi add EXP_AD_P2 = 0xd0,EXP_AD_P1 fmerge.s atan2_sgnY = atan2_Y,f1 nop.i 999 } ;; { .mfi ldfe atan2_P11 = [EXP_AD_P1],16 fclass.m p10,p0 = atan2_Y, 0xc3 // Test for y=nan nop.i 999 } { .mfi ldfe atan2_P21 = [EXP_AD_P2],16 nop.f 999 nop.i 999 ;; } { .mfi ldfe atan2_P10 = [EXP_AD_P1],16 fnma.s1 atan2_B1Y = atan2_u1_Y, atan2_Y, atan2_two nop.i 999 } { .mfi ldfe atan2_P20 = [EXP_AD_P2],16 fma.s1 atan2_wp_Y = atan2_u1_Y, atan2_u1_Y, f0 nop.i 999 ;; } { .mfi ldfe atan2_P9 = [EXP_AD_P1],16 fma.s1 atan2_z1_X = atan2_u1_X, atan2_Y, f0 nop.i 999 } { .mfi ldfe atan2_P19 = [EXP_AD_P2],16 fnma.s1 atan2_B1X = atan2_u1_X, atan2_X, atan2_two nop.i 999 } ;; { .mfi ldfe atan2_P8 = [EXP_AD_P1],16 fma.s1 atan2_z2_X = atan2_u1_X, atan2_ysq, f0 nop.i 999 } { .mfb ldfe atan2_P18 = [EXP_AD_P2],16 (p10) fma.d.s0 f8 = atan2_Y,atan2_X,f0 // If y=nan, result quietized y (p10) br.ret.spnt b0 // Exit if y=nan } ;; // p6 true if swap, means |y| > 1.0 or ysq > 1.0 // p7 true if no swap, means 1.0 >= |y| or 1.0 >= ysq { .mfi ldfe atan2_P7 = [EXP_AD_P1],16 fcmp.ge.s1 p7,p6 = f1, atan2_ysq nop.i 999 } { .mmf ldfe atan2_P17 = [EXP_AD_P2],16 nop.m 999 nop.f 999 } ;; { .mfi ldfe atan2_P6 = [EXP_AD_P1],16 fma.s1 atan2_E = atan2_u1_Y, atan2_B1Y, atan2_Y nop.i 999 } { .mfi ldfe atan2_P16 = [EXP_AD_P2],16 fma.s1 atan2_B1sq_Y = atan2_B1Y, atan2_B1Y, f0 nop.i 999 ;; } { .mfi ldfe atan2_P5 = [EXP_AD_P1],16 (p7) fma.s1 atan2_wp_X = atan2_z1_X, atan2_z1_X, f0 nop.i 999 } { .mfi ldfe atan2_P15 = [EXP_AD_P2],16 (p7) fma.s1 atan2_B1sq_X = atan2_B1X, atan2_B1X, f0 nop.i 999 ;; } { .mfi ldfe atan2_P4 = [EXP_AD_P1],16 (p6) fma.s1 atan2_z = atan2_u1_Y, atan2_B1Y, f0 nop.i 999 } { .mfi ldfe atan2_P14 = [EXP_AD_P2],16 (p7) fma.s1 atan2_E = atan2_z2_X, atan2_B1X, atan2_X nop.i 999 ;; } { .mfi ldfe atan2_P3 = [EXP_AD_P1],16 fcmp.eq.s0 p14,p15=atan2_X,atan2_Y // Dummy for denorm and invalid nop.i 999 } { .mmf ldfe atan2_P13 = [EXP_AD_P2],16 nop.m 999 (p7) fma.s1 atan2_z = atan2_z1_X, atan2_B1X, f0 ;; } { .mfi ldfe atan2_P2 = [EXP_AD_P1],16 (p6) fma.s1 atan2_w = atan2_wp_Y, atan2_B1sq_Y,f0 nop.i 999 } { .mlx ldfe atan2_P12 = [EXP_AD_P2],16 movl rsig_near_one = 0x8000000000000001 // signif near 1.0 ;; } { .mfi ldfe atan2_P1 = [EXP_AD_P1],16 fclass.m p9,p0 = atan2_Y, 0x23 // test if y inf nop.i 999 } { .mfi ldfe atan2_P22 = [EXP_AD_P2],16 (p7) fma.s1 atan2_w = atan2_wp_X, atan2_B1sq_X,f0 nop.i 999 ;; } { .mfi ldfe atan2_P0 = [EXP_AD_P1],16 frcpa.s1 atan2_F,p0 = f1, atan2_E nop.i 999 } { .mfi ldfe atan2_pi_by_2 = [EXP_AD_P2],16 (p6) fnma.s1 atan2_gV = atan2_Y, atan2_z, atan2_X nop.i 999 ;; } { .mfi setf.sig atan2_sig_near_one = rsig_near_one (p7) fnma.s1 atan2_gV = atan2_X, atan2_z, atan2_Y nop.i 999 } { .mfb nop.m 999 (p9) fma.d.s0 f8 = atan2_sgnY, atan2_pi_by_2, f0 // +-pi/2 if y inf (p9) br.ret.spnt b0 // exit if y inf, result is +-pi/2 ;; } { .mfi nop.m 999 fma.s1 atan2_V13 = atan2_w, atan2_P11, atan2_P10 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W11 = atan2_w, atan2_P21, atan2_P20 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V11 = atan2_w, atan2_P9, atan2_P8 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_V12 = atan2_w, atan2_w, f0 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V8 = atan2_w, atan2_P7 , atan2_P6 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W8 = atan2_w, atan2_P19, atan2_P18 nop.i 999 ;; } { .mfi nop.m 999 fnma.s1 atan2_alpha = atan2_E, atan2_F, f1 nop.i 999 } { .mfi nop.m 999 fnma.s1 atan2_alpha_1 = atan2_E, atan2_F, atan2_two nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V7 = atan2_w, atan2_P5 , atan2_P4 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W7 = atan2_w, atan2_P17, atan2_P16 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V4 = atan2_w, atan2_P3 , atan2_P2 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W4 = atan2_w, atan2_P15, atan2_P14 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V3 = atan2_w, atan2_P1 , atan2_P0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W3 = atan2_w, atan2_P13, atan2_P12 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V10 = atan2_V12, atan2_V13, atan2_V11 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_gVF = atan2_gV, atan2_F, f0 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_alpha_sq = atan2_alpha, atan2_alpha, f0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_Cp = atan2_alpha, atan2_alpha_1, f1 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V9 = atan2_V12, atan2_V12, f0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W10 = atan2_V12, atan2_P22 , atan2_W11 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V6 = atan2_V12, atan2_V8 , atan2_V7 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W6 = atan2_V12, atan2_W8 , atan2_W7 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V2 = atan2_V12, atan2_V4 , atan2_V3 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W2 = atan2_V12, atan2_W4 , atan2_W3 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_alpha_cub = atan2_alpha, atan2_alpha_sq, f0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_C = atan2_gVF, atan2_Cp, f0 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_W12 = atan2_V9, atan2_V9, f0 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_V5 = atan2_V9, atan2_V10, atan2_V6 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W5 = atan2_V9, atan2_W10, atan2_W6 nop.i 999 ;; } { .mfi nop.m 999 fclass.m p8,p0 = atan2_Y, 0x07 // Test for y=0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_d = atan2_alpha_cub, atan2_C, atan2_C nop.i 999 } ;; { .mfi nop.m 999 fma.s1 atan2_W12 = atan2_V9, atan2_W12, f0 nop.i 999 } ;; { .mfi nop.m 999 fma.s1 atan2_V1 = atan2_V9, atan2_V5, atan2_V2 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_W1 = atan2_V9, atan2_W5, atan2_W2 nop.i 999 ;; } { .mfi nop.m 999 (p8) fmerge.s f8 = atan2_sgnY, f0 // +-0 if y=0 nop.i 999 } { .mfb nop.m 999 fma.s1 atan2_zcub = atan2_z, atan2_w, f0 (p8) br.ret.spnt b0 // Exit if y=0 ;; } { .mfi nop.m 999 fma.s1 atan2_pd = atan2_P0, atan2_d, f0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_dsq = atan2_d, atan2_d, f0 nop.i 999 ;; } { .mfi nop.m 999 fmerge.se atan2_near_one = f1, atan2_sig_near_one // Const ~1.0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_Pp = atan2_W12, atan2_W1, atan2_V1 nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_sgn_pi_by_2 = atan2_pi_by_2, atan2_sgnY, f0 nop.i 999 } { .mfi nop.m 999 fma.s1 atan2_A_lo = atan2_pd, atan2_dsq, atan2_d nop.i 999 ;; } { .mfi nop.m 999 fma.s1 atan2_A_hi = atan2_zcub, atan2_Pp, atan2_z nop.i 999 ;; } { .mfi nop.m 999 (p6) fma.s1 atan2_A = atan2_A_hi, f1, atan2_A_lo nop.i 999 } // For |Y| <= |X| and X > 0, result is A_hi + A_lo { .mfi nop.m 999 (p7) fma.d.s0 f8 = atan2_A_hi, f1, atan2_A_lo nop.i 999 ;; } // For |Y| > |X|, result is +- pi/2 - (A_hi + A_lo) // We perturb A by multiplying by 1.0+1ulp as we produce the result // in order to get symmetrically rounded results in directed rounding modes. // If we don't do this, there are a few cases where the trailing 11 bits of // the significand of the result, before converting to double, are zero. These // cases do not round symmetrically in round to +infinity or round to -infinity. { .mfb nop.m 999 (p6) fnma.d.s0 f8 = atan2_A, atan2_near_one, atan2_sgn_pi_by_2 br.ret.sptk b0 ;; } GLOBAL_LIBM_END(atan)