diff options
Diffstat (limited to 'sysdeps/ia64/fpu/s_tan.S')
| -rw-r--r-- | sysdeps/ia64/fpu/s_tan.S | 554 |
1 files changed, 331 insertions, 223 deletions
diff --git a/sysdeps/ia64/fpu/s_tan.S b/sysdeps/ia64/fpu/s_tan.S index 3a497fcf4c..a2f80c8858 100644 --- a/sysdeps/ia64/fpu/s_tan.S +++ b/sysdeps/ia64/fpu/s_tan.S @@ -1,10 +1,10 @@ -.file "tan.s" +.file "tancot.s" -// Copyright (C) 2000, 2001, Intel Corporation + +// Copyright (c) 2000 - 2003, Intel Corporation // All rights reserved. -// -// Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story, -// and Ping Tak Peter Tang of the Computational Software Lab, Intel Corporation. +// +// 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 @@ -20,7 +20,7 @@ // * 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 @@ -32,20 +32,24 @@ // 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://developer.intel.com/opensource. +// http://www.intel.com/software/products/opensource/libraries/num.htm. // // History //============================================================== -// 2/02/00: Initial version -// 4/04/00 Unwind support added +// 02/02/00 Initial version +// 04/04/00 Unwind support added // 12/27/00 Improved speed +// 02/21/01 Updated to call tanl +// 05/30/02 Added cot +// 02/10/03 Reordered header: .section, .global, .proc, .align // // API //============================================================== -// double tan( double x); +// double tan(double x); +// double cot(double x); // // Overview of operation //============================================================== @@ -61,11 +65,14 @@ // Nfloat = round_int(tan_W) // // tan_r = x - Nfloat * (pi/2)_hi -// tan_r = tan_r - Nfloat * (pi/2)_lo +// a) tan_r = tan_r - Nfloat * (pi/2)_lo (for tan) +// b) tan_r = Nfloat * (pi/2)_lo - tan_r (for cot) // // We have two paths: p8, when Nfloat is even and p9. when Nfloat is odd. -// p8: tan(X) = tan(r) -// p9: tan(X) = -cot(r) +// a) for tan: p8: tan(X) = tan(r) +// p9: tan(X) = -cot(r) +// b) for cot: p9: cot(X) = cot(r) +// p8: cot(X) = -tan(r) // // Each is evaluated as a series. The p9 path requires 1/r. // @@ -75,19 +82,16 @@ // Registers used //============================================================== // -// predicate registers used: -// p6-10 +// predicate registers used: +// p6-12 // -// floating-point registers used: -// f10-15, f32-105 +// floating-point registers used: +// f10-15, f32-106 // f8, input // // general registers used -// r14-18, r32-43 +// r14-26, r32-39 // - -#include "libm_support.h" - // Assembly macros //============================================================== TAN_INV_PI_BY_2_2TO64 = f10 @@ -105,28 +109,28 @@ tan_Pi_by_2_lo = f34 tan_P0 = f35 tan_P1 = f36 tan_P2 = f37 -tan_P3 = f38 -tan_P4 = f39 -tan_P5 = f40 +tan_P3 = f38 +tan_P4 = f39 +tan_P5 = f40 tan_P6 = f41 tan_P7 = f42 -tan_P8 = f43 -tan_P9 = f44 -tan_P10 = f45 +tan_P8 = f43 +tan_P9 = f44 +tan_P10 = f45 tan_P11 = f46 -tan_P12 = f47 +tan_P12 = f47 tan_P13 = f48 tan_P14 = f49 tan_P15 = f50 -tan_Q0 = f51 -tan_Q1 = f52 -tan_Q2 = f53 -tan_Q3 = f54 -tan_Q4 = f55 -tan_Q5 = f56 -tan_Q6 = f57 -tan_Q7 = f58 +tan_Q0 = f51 +tan_Q1 = f52 +tan_Q2 = f53 +tan_Q3 = f54 +tan_Q4 = f55 +tan_Q5 = f56 +tan_Q6 = f57 +tan_Q7 = f58 tan_Q8 = f59 tan_Q9 = f60 tan_Q10 = f61 @@ -153,19 +157,19 @@ tan_v10 = f79 tan_v2 = f80 tan_v9 = f81 tan_v1 = f82 -tan_int_Nfloat = f83 -tan_Nfloat = f84 +tan_int_Nfloat = f83 +tan_Nfloat = f84 -tan_NORM_f8 = f85 +tan_NORM_f8 = f85 tan_W = f86 tan_y0 = f87 -tan_d = f88 -tan_y1 = f89 -tan_dsq = f90 -tan_y2 = f91 -tan_d4 = f92 -tan_inv_r = f93 +tan_d = f88 +tan_y1 = f89 +tan_dsq = f90 +tan_y2 = f91 +tan_d4 = f92 +tan_inv_r = f93 tan_z1 = f94 tan_z2 = f95 @@ -180,6 +184,7 @@ tan_z10 = f103 tan_z11 = f104 tan_z12 = f105 +arg_copy = f106 ///////////////////////////////////////////////////////////// @@ -188,37 +193,33 @@ tan_GR_rshf_2to64 = r15 tan_GR_exp_2tom64 = r16 tan_GR_n = r17 tan_GR_rshf = r18 - -tan_AD = r33 -tan_GR_10009 = r34 -tan_GR_17_ones = r35 -tan_GR_N_odd_even = r36 -tan_GR_N = r37 -tan_signexp = r38 -tan_exp = r39 -tan_ADQ = r40 - -GR_SAVE_PFS = r41 -GR_SAVE_B0 = r42 -GR_SAVE_GP = r43 - - -#ifdef _LIBC -.rodata -#else -.data -#endif +tan_AD = r19 +tan_GR_10009 = r20 +tan_GR_17_ones = r21 +tan_GR_N_odd_even = r22 +tan_GR_N = r23 +tan_signexp = r24 +tan_exp = r25 +tan_ADQ = r26 + +GR_SAVE_B0 = r33 +GR_SAVE_PFS = r34 +GR_SAVE_GP = r35 +GR_Parameter_X = r36 +GR_Parameter_Y = r37 +GR_Parameter_RESULT = r38 +GR_Parameter_Tag = r39 + + +RODATA .align 16 -double_tan_constants: -ASM_TYPE_DIRECTIVE(double_tan_constants,@object) -// data8 0xA2F9836E4E44152A, 0x00003FFE // 2/pi +LOCAL_OBJECT_START(double_tan_constants) data8 0xC90FDAA22168C234, 0x00003FFF // pi/2 hi - - data8 0xBEEA54580DDEA0E1 // P14 + data8 0xBEEA54580DDEA0E1 // P14 data8 0x3ED3021ACE749A59 // P15 - data8 0xBEF312BD91DC8DA1 // P12 + data8 0xBEF312BD91DC8DA1 // P12 data8 0x3EFAE9AFC14C5119 // P13 data8 0x3F2F342BF411E769 // P8 data8 0x3F1A60FC9F3B0227 // P9 @@ -232,10 +233,9 @@ ASM_TYPE_DIRECTIVE(double_tan_constants,@object) data8 0x3FC11111111111C2 // P1 data8 0x3FABA1BA1BA0E850 // P2 data8 0x3F9664F4886725A7 // P3 -ASM_SIZE_DIRECTIVE(double_tan_constants) +LOCAL_OBJECT_END(double_tan_constants) -double_Q_tan_constants: -ASM_TYPE_DIRECTIVE(double_Q_tan_constants,@object) +LOCAL_OBJECT_START(double_Q_tan_constants) data8 0xC4C6628B80DC1CD1, 0x00003FBF // pi/2 lo data8 0x3E223A73BA576E48 // Q8 data8 0x3DF54AD8D1F2CA43 // Q9 @@ -248,35 +248,19 @@ ASM_TYPE_DIRECTIVE(double_Q_tan_constants,@object) data8 0x3F61566ABBFFB489 // Q2 data8 0x3F2BBD77945C1733 // Q3 data8 0x3D927FB33E2B0E04 // Q10 -ASM_SIZE_DIRECTIVE(double_Q_tan_constants) +LOCAL_OBJECT_END(double_Q_tan_constants) - -.align 32 -.global tan# -#ifdef _LIBC -.global __tan# -#endif +.section .text //////////////////////////////////////////////////////// - - -.section .text -.proc tan# -#ifdef _LIBC -.proc __tan# -#endif -.align 32 -tan: -#ifdef _LIBC -__tan: -#endif +LOCAL_LIBM_ENTRY(cot) // The initial fnorm will take any unmasked faults and // normalize any single/double unorms { .mlx - alloc r32=ar.pfs,1,11,0,0 + cmp.eq p12, p11 = r0, r0 // set p12=1, p11=0 for cot movl tan_GR_sig_inv_pi_by_2 = 0xA2F9836E4E44152A // significand of 2/pi } { .mlx @@ -285,18 +269,48 @@ __tan: } ;; -{ .mfi - ld8 tan_AD = [tan_AD] - fnorm tan_NORM_f8 = f8 +{ .mlx mov tan_GR_exp_2tom64 = 0xffff-64 // exponent of scaling factor 2^-64 + movl tan_GR_rshf = 0x43e8000000000000 // 1.1000 2^63 for right shift } +{ .mfb + ld8 tan_AD = [tan_AD] + fnorm.s0 tan_NORM_f8 = f8 + br.cond.sptk COMMON_PATH +} +;; + +LOCAL_LIBM_END(cot) + + +GLOBAL_IEEE754_ENTRY(tan) +// The initial fnorm will take any unmasked faults and +// normalize any single/double unorms + { .mlx - nop.m 999 + cmp.eq p11, p12 = r0, r0 // set p11=1, p12=0 for tan + movl tan_GR_sig_inv_pi_by_2 = 0xA2F9836E4E44152A // significand of 2/pi +} +{ .mlx + addl tan_AD = @ltoff(double_tan_constants), gp + movl tan_GR_rshf_2to64 = 0x47e8000000000000 // 1.1000 2^(63+63+1) +} +;; + +{ .mlx + mov tan_GR_exp_2tom64 = 0xffff-64 // exponent of scaling factor 2^-64 movl tan_GR_rshf = 0x43e8000000000000 // 1.1000 2^63 for right shift } +{ .mfi + ld8 tan_AD = [tan_AD] + fnorm.s0 tan_NORM_f8 = f8 + nop.i 0 +} ;; +// Common path for both tan and cot +COMMON_PATH: // Form two constants we need // 2/pi * 2^1 * 2^63, scaled by 2^64 since we just loaded the significand // 1.1000...000 * 2^(63+63+1) to right shift int(W) into the significand @@ -313,7 +327,7 @@ __tan: { .mmf setf.exp TAN_2TOM64 = tan_GR_exp_2tom64 adds tan_ADQ = double_Q_tan_constants - double_tan_constants, tan_AD - fclass.m.unc p6,p0 = f8, 0x07 // Test for x=0 +(p11) fclass.m.unc p6,p0 = f8, 0x07 // Test for x=0 (tan) } ;; @@ -323,79 +337,79 @@ __tan: // 1.1000...000 * 2^63, the right shift constant { .mmf setf.d TAN_RSHF = tan_GR_rshf - ldfe tan_Pi_by_2_hi = [tan_AD],16 + ldfe tan_Pi_by_2_hi = [tan_AD],16 fclass.m.unc p7,p0 = f8, 0x23 // Test for x=inf } ;; { .mfb - ldfe tan_Pi_by_2_lo = [tan_ADQ],16 + ldfe tan_Pi_by_2_lo = [tan_ADQ],16 fclass.m.unc p8,p0 = f8, 0xc3 // Test for x=nan -(p6) br.ret.spnt b0 ;; // Exit for x=0 +(p6) br.ret.spnt b0 ;; // Exit for x=0 (tan only) } { .mfi - ldfpd tan_P14,tan_P15 = [tan_AD],16 + ldfpd tan_P14,tan_P15 = [tan_AD],16 (p7) frcpa.s0 f8,p9=f0,f0 // Set qnan indef if x=inf mov tan_GR_10009 = 0x10009 } { .mib - ldfpd tan_Q8,tan_Q9 = [tan_ADQ],16 + ldfpd tan_Q8,tan_Q9 = [tan_ADQ],16 nop.i 999 (p7) br.ret.spnt b0 ;; // Exit for x=inf } { .mfi - ldfpd tan_P12,tan_P13 = [tan_AD],16 -(p8) fma.d f8=f8,f1,f8 // Set qnan if x=nan + ldfpd tan_P12,tan_P13 = [tan_AD],16 +(p12) fclass.m.unc p6,p0 = f8, 0x07 // Test for x=0 (cot) nop.i 999 } -{ .mib - ldfpd tan_Q4,tan_Q5 = [tan_ADQ],16 - nop.i 999 +{ .mfb + ldfpd tan_Q4,tan_Q5 = [tan_ADQ],16 +(p8) fma.d.s0 f8=f8,f1,f8 // Set qnan if x=nan (p8) br.ret.spnt b0 ;; // Exit for x=nan } -{ .mmi - getf.exp tan_signexp = tan_NORM_f8 - ldfpd tan_P8,tan_P9 = [tan_AD],16 - nop.i 999 ;; +{ .mmf + getf.exp tan_signexp = tan_NORM_f8 + ldfpd tan_P8,tan_P9 = [tan_AD],16 + fmerge.s arg_copy = f8, f8 ;; // Save input for error call } -// Multiply x by scaled 2/pi and add large const to shift integer part of W to +// Multiply x by scaled 2/pi and add large const to shift integer part of W to // rightmost bits of significand -{ .mfi +{ .mmf + alloc r32=ar.pfs,0,4,4,0 ldfpd tan_Q6,tan_Q7 = [tan_ADQ],16 fma.s1 TAN_W_2TO64_RSH = tan_NORM_f8,TAN_INV_PI_BY_2_2TO64,TAN_RSHF_2TO64 - nop.i 999 ;; -} +};; -{ .mmi - ldfpd tan_P10,tan_P11 = [tan_AD],16 - nop.m 999 - and tan_exp = tan_GR_17_ones, tan_signexp ;; +{ .mmf + ldfpd tan_P10,tan_P11 = [tan_AD],16 + and tan_exp = tan_GR_17_ones, tan_signexp +(p6) frcpa.s0 f8, p0 = f1, f8 ;; // cot(+-0) = +-Inf } // p7 is true if we must call DBX TAN // p7 is true if f8 exp is > 0x10009 (which includes all ones // NAN or inf) -{ .mmi - ldfpd tan_Q0,tan_Q1 = [tan_ADQ],16 - cmp.ge.unc p7,p0 = tan_exp,tan_GR_10009 - nop.i 999 ;; +{ .mmb + ldfpd tan_Q0,tan_Q1 = [tan_ADQ],16 + cmp.ge.unc p7,p0 = tan_exp,tan_GR_10009 +(p7) br.cond.spnt TAN_DBX ;; } { .mmb - ldfpd tan_P4,tan_P5 = [tan_AD],16 - nop.m 999 -(p7) br.cond.spnt L(TAN_DBX) ;; + ldfpd tan_P4,tan_P5 = [tan_AD],16 +(p6) mov GR_Parameter_Tag = 226 // (cot) +(p6) br.cond.spnt __libm_error_region ;; // call error support if cot(+-0) } { .mmi - ldfpd tan_Q2,tan_Q3 = [tan_ADQ],16 + ldfpd tan_Q2,tan_Q3 = [tan_ADQ],16 nop.m 999 nop.i 999 ;; } @@ -404,8 +418,8 @@ __tan: // TAN_NFLOAT = Round_Int_Nearest(tan_W) { .mfi - ldfpd tan_P6,tan_P7 = [tan_AD],16 - fms.s1 TAN_NFLOAT = TAN_W_2TO64_RSH,TAN_2TOM64,TAN_RSHF + ldfpd tan_P6,tan_P7 = [tan_AD],16 + fms.s1 TAN_NFLOAT = TAN_W_2TO64_RSH,TAN_2TOM64,TAN_RSHF nop.i 999 ;; } @@ -418,22 +432,22 @@ __tan: { .mfi - ldfpd tan_P0,tan_P1 = [tan_AD],16 + ldfpd tan_P0,tan_P1 = [tan_AD],16 nop.f 999 nop.i 999 ;; } -{ .mfi +{ .mmi getf.sig tan_GR_n = TAN_W_2TO64_RSH - nop.f 999 + ldfpd tan_P2,tan_P3 = [tan_AD] nop.i 999 ;; } // tan_r = -tan_Nfloat * tan_Pi_by_2_hi + x { .mfi - ldfpd tan_P2,tan_P3 = [tan_AD] - fnma.s1 tan_r = TAN_NFLOAT, tan_Pi_by_2_hi, tan_NORM_f8 +(p12) add tan_GR_n = 0x1, tan_GR_n // N = N + 1 (for cot) + fnma.s1 tan_r = TAN_NFLOAT, tan_Pi_by_2_hi, tan_NORM_f8 nop.i 999 ;; } @@ -441,42 +455,49 @@ __tan: // p8 ==> even // p9 ==> odd { .mmi - and tan_GR_N_odd_even = 0x1, tan_GR_n ;; + and tan_GR_N_odd_even = 0x1, tan_GR_n ;; nop.m 999 cmp.eq.unc p8,p9 = tan_GR_N_odd_even, r0 ;; } -// tan_r = tan_r -tan_Nfloat * tan_Pi_by_2_lo +.pred.rel "mutex", p11, p12 +// tan_r = tan_r -tan_Nfloat * tan_Pi_by_2_lo (tan) { .mfi nop.m 999 - fnma.s1 tan_r = TAN_NFLOAT, tan_Pi_by_2_lo, tan_r +(p11) fnma.s1 tan_r = TAN_NFLOAT, tan_Pi_by_2_lo, tan_r + nop.i 999 +} +// tan_r = -(tan_r -tan_Nfloat * tan_Pi_by_2_lo) (cot) +{ .mfi + nop.m 999 +(p12) fms.s1 tan_r = TAN_NFLOAT, tan_Pi_by_2_lo, tan_r nop.i 999 ;; } { .mfi nop.m 999 - fma.s1 tan_rsq = tan_r, tan_r, f0 + fma.s1 tan_rsq = tan_r, tan_r, f0 nop.i 999 ;; } { .mfi nop.m 999 -(p9) frcpa.s1 tan_y0, p10 = f1,tan_r +(p9) frcpa.s1 tan_y0, p0 = f1,tan_r nop.i 999 ;; } { .mfi nop.m 999 -(p8) fma.s1 tan_v18 = tan_rsq, tan_P15, tan_P14 +(p8) fma.s1 tan_v18 = tan_rsq, tan_P15, tan_P14 nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v4 = tan_rsq, tan_P1, tan_P0 +(p8) fma.s1 tan_v4 = tan_rsq, tan_P1, tan_P0 nop.i 999 ;; } @@ -484,12 +505,12 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v16 = tan_rsq, tan_P13, tan_P12 - nop.i 999 +(p8) fma.s1 tan_v16 = tan_rsq, tan_P13, tan_P12 + nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v17 = tan_rsq, tan_rsq, f0 +(p8) fma.s1 tan_v17 = tan_rsq, tan_rsq, f0 nop.i 999 ;; } @@ -497,12 +518,12 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v12 = tan_rsq, tan_P9, tan_P8 - nop.i 999 +(p8) fma.s1 tan_v12 = tan_rsq, tan_P9, tan_P8 + nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v13 = tan_rsq, tan_P11, tan_P10 +(p8) fma.s1 tan_v13 = tan_rsq, tan_P11, tan_P10 nop.i 999 ;; } @@ -510,12 +531,12 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v7 = tan_rsq, tan_P5, tan_P4 - nop.i 999 +(p8) fma.s1 tan_v7 = tan_rsq, tan_P5, tan_P4 + nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v8 = tan_rsq, tan_P7, tan_P6 +(p8) fma.s1 tan_v8 = tan_rsq, tan_P7, tan_P6 nop.i 999 ;; } @@ -523,12 +544,12 @@ __tan: { .mfi nop.m 999 -(p9) fnma.s1 tan_d = tan_r, tan_y0, f1 - nop.i 999 +(p9) fnma.s1 tan_d = tan_r, tan_y0, f1 + nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v5 = tan_rsq, tan_P3, tan_P2 +(p8) fma.s1 tan_v5 = tan_rsq, tan_P3, tan_P2 nop.i 999 ;; } @@ -536,36 +557,36 @@ __tan: { .mfi nop.m 999 -(p9) fma.s1 tan_z11 = tan_rsq, tan_Q9, tan_Q8 +(p9) fma.s1 tan_z11 = tan_rsq, tan_Q9, tan_Q8 nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z12 = tan_rsq, tan_rsq, f0 +(p9) fma.s1 tan_z12 = tan_rsq, tan_rsq, f0 nop.i 999 ;; } { .mfi nop.m 999 -(p8) fma.s1 tan_v15 = tan_v17, tan_v18, tan_v16 - nop.i 999 +(p8) fma.s1 tan_v15 = tan_v17, tan_v18, tan_v16 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z7 = tan_rsq, tan_Q5, tan_Q4 +(p9) fma.s1 tan_z7 = tan_rsq, tan_Q5, tan_Q4 nop.i 999 ;; } { .mfi nop.m 999 -(p8) fma.s1 tan_v11 = tan_v17, tan_v13, tan_v12 +(p8) fma.s1 tan_v11 = tan_v17, tan_v13, tan_v12 nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z8 = tan_rsq, tan_Q7, tan_Q6 +(p9) fma.s1 tan_z8 = tan_rsq, tan_Q7, tan_Q6 nop.i 999 ;; } @@ -573,13 +594,13 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v14 = tan_v17, tan_v17, f0 - nop.i 999 +(p8) fma.s1 tan_v14 = tan_v17, tan_v17, f0 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z3 = tan_rsq, tan_Q1, tan_Q0 - nop.i 999 ;; +(p9) fma.s1 tan_z3 = tan_rsq, tan_Q1, tan_Q0 + nop.i 999 ;; } @@ -587,12 +608,12 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v3 = tan_v17, tan_v5, tan_v4 +(p8) fma.s1 tan_v3 = tan_v17, tan_v5, tan_v4 nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v6 = tan_v17, tan_v8, tan_v7 +(p8) fma.s1 tan_v6 = tan_v17, tan_v8, tan_v7 nop.i 999 ;; } @@ -600,89 +621,89 @@ __tan: { .mfi nop.m 999 -(p9) fma.s1 tan_y1 = tan_y0, tan_d, tan_y0 - nop.i 999 +(p9) fma.s1 tan_y1 = tan_y0, tan_d, tan_y0 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_dsq = tan_d, tan_d, f0 - nop.i 999 ;; +(p9) fma.s1 tan_dsq = tan_d, tan_d, f0 + nop.i 999 ;; } { .mfi nop.m 999 -(p9) fma.s1 tan_z10 = tan_z12, tan_Q10, tan_z11 - nop.i 999 +(p9) fma.s1 tan_z10 = tan_z12, tan_Q10, tan_z11 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z9 = tan_z12, tan_z12,f0 +(p9) fma.s1 tan_z9 = tan_z12, tan_z12,f0 nop.i 999 ;; } { .mfi nop.m 999 -(p9) fma.s1 tan_z4 = tan_rsq, tan_Q3, tan_Q2 - nop.i 999 +(p9) fma.s1 tan_z4 = tan_rsq, tan_Q3, tan_Q2 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z6 = tan_z12, tan_z8, tan_z7 - nop.i 999 ;; +(p9) fma.s1 tan_z6 = tan_z12, tan_z8, tan_z7 + nop.i 999 ;; } { .mfi nop.m 999 -(p8) fma.s1 tan_v10 = tan_v14, tan_v15, tan_v11 - nop.i 999 ;; +(p8) fma.s1 tan_v10 = tan_v14, tan_v15, tan_v11 + nop.i 999 ;; } { .mfi nop.m 999 -(p9) fma.s1 tan_y2 = tan_y1, tan_d, tan_y0 - nop.i 999 +(p9) fma.s1 tan_y2 = tan_y1, tan_d, tan_y0 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_d4 = tan_dsq, tan_dsq, tan_d +(p9) fma.s1 tan_d4 = tan_dsq, tan_dsq, tan_d nop.i 999 ;; } { .mfi nop.m 999 -(p8) fma.s1 tan_v2 = tan_v14, tan_v6, tan_v3 +(p8) fma.s1 tan_v2 = tan_v14, tan_v6, tan_v3 nop.i 999 } { .mfi nop.m 999 -(p8) fma.s1 tan_v9 = tan_v14, tan_v14, f0 +(p8) fma.s1 tan_v9 = tan_v14, tan_v14, f0 nop.i 999 ;; } { .mfi nop.m 999 -(p9) fma.s1 tan_z2 = tan_z12, tan_z4, tan_z3 - nop.i 999 +(p9) fma.s1 tan_z2 = tan_z12, tan_z4, tan_z3 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z5 = tan_z9, tan_z10, tan_z6 +(p9) fma.s1 tan_z5 = tan_z9, tan_z10, tan_z6 nop.i 999 ;; } { .mfi nop.m 999 -(p9) fma.s1 tan_inv_r = tan_d4, tan_y2, tan_y0 - nop.i 999 +(p9) fma.s1 tan_inv_r = tan_d4, tan_y2, tan_y0 + nop.i 999 } { .mfi nop.m 999 @@ -694,12 +715,12 @@ __tan: { .mfi nop.m 999 -(p8) fma.s1 tan_v1 = tan_v9, tan_v10, tan_v2 - nop.i 999 +(p8) fma.s1 tan_v1 = tan_v9, tan_v10, tan_v2 + nop.i 999 } { .mfi nop.m 999 -(p9) fma.s1 tan_z1 = tan_z9, tan_z5, tan_z2 +(p9) fma.s1 tan_z1 = tan_z9, tan_z5, tan_z2 nop.i 999 ;; } @@ -707,64 +728,151 @@ __tan: { .mfi nop.m 999 -(p8) fma.d.s0 f8 = tan_v1, tan_rcube, tan_r - nop.i 999 +(p8) fma.d.s0 f8 = tan_v1, tan_rcube, tan_r + nop.i 999 } { .mfb nop.m 999 -(p9) fms.d.s0 f8 = tan_r, tan_z1, tan_inv_r - br.ret.sptk b0 ;; +(p9) fms.d.s0 f8 = tan_r, tan_z1, tan_inv_r + br.ret.sptk b0 ;; } -.endp tan# -ASM_SIZE_DIRECTIVE(tan) +GLOBAL_IEEE754_END(tan) -.proc __libm_callout -__libm_callout: -L(TAN_DBX): +LOCAL_LIBM_ENTRY(__libm_callout) +TAN_DBX: .prologue { .mfi - nop.m 0 - fmerge.s f9 = f0,f0 -.save ar.pfs,GR_SAVE_PFS - mov GR_SAVE_PFS=ar.pfs + nop.m 0 + fmerge.s f9 = f0,f0 +.save ar.pfs,GR_SAVE_PFS + mov GR_SAVE_PFS=ar.pfs } ;; { .mfi - mov GR_SAVE_GP=gp - nop.f 0 -.save b0, GR_SAVE_B0 - mov GR_SAVE_B0=b0 + mov GR_SAVE_GP=gp + nop.f 0 +.save b0, GR_SAVE_B0 + mov GR_SAVE_B0=b0 } .body -{ .mfb +{ .mmb nop.m 999 - nop.f 999 - br.call.sptk.many b0=__libm_tan# ;; + nop.m 999 +(p11) br.cond.sptk.many call_tanl ;; } +// Here if we should call cotl +{ .mmb + nop.m 999 + nop.m 999 + br.call.sptk.many b0=__libm_cotl# ;; +} { .mfi - mov gp = GR_SAVE_GP - fnorm.d f8 = f8 - mov b0 = GR_SAVE_B0 + mov gp = GR_SAVE_GP + fnorm.d.s0 f8 = f8 + mov b0 = GR_SAVE_B0 } ;; +{ .mib + nop.m 999 + mov ar.pfs = GR_SAVE_PFS + br.ret.sptk b0 +;; +} + +// Here if we should call tanl +call_tanl: +{ .mmb + nop.m 999 + nop.m 999 + br.call.sptk.many b0=__libm_tanl# ;; +} + +{ .mfi + mov gp = GR_SAVE_GP + fnorm.d.s0 f8 = f8 + mov b0 = GR_SAVE_B0 +} +;; { .mib - nop.m 999 + nop.m 999 mov ar.pfs = GR_SAVE_PFS br.ret.sptk b0 ;; } +LOCAL_LIBM_END(__libm_callout) + +.type __libm_tanl#,@function +.global __libm_tanl# +.type __libm_cotl#,@function +.global __libm_cotl# + +LOCAL_LIBM_ENTRY(__libm_error_region) +.prologue + +// (1) +{ .mfi + add GR_Parameter_Y=-32,sp // Parameter 2 value + nop.f 0 +.save ar.pfs,GR_SAVE_PFS + mov GR_SAVE_PFS=ar.pfs // Save ar.pfs +} +{ .mfi +.fframe 64 + add sp=-64,sp // Create new stack + nop.f 0 + mov GR_SAVE_GP=gp // Save gp +};; + +// (2) +{ .mmi + stfd [GR_Parameter_Y] = f1,16 // STORE Parameter 2 on stack + add GR_Parameter_X = 16,sp // Parameter 1 address +.save b0, GR_SAVE_B0 + mov GR_SAVE_B0=b0 // Save b0 +};; + +.body +// (3) +{ .mib + stfd [GR_Parameter_X] = arg_copy // STORE Parameter 1 on stack + add GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address + nop.b 0 +} +{ .mib + stfd [GR_Parameter_Y] = f8 // STORE Parameter 3 on stack + add GR_Parameter_Y = -16,GR_Parameter_Y + br.call.sptk b0=__libm_error_support# // Call error handling function +};; +{ .mmi + nop.m 0 + nop.m 0 + add GR_Parameter_RESULT = 48,sp +};; + +// (4) +{ .mmi + ldfd f8 = [GR_Parameter_RESULT] // Get return result off stack +.restore sp + add sp = 64,sp // Restore stack pointer + mov b0 = GR_SAVE_B0 // Restore return address +};; +{ .mib + mov gp = GR_SAVE_GP // Restore gp + mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs + br.ret.sptk b0 // Return +};; + +LOCAL_LIBM_END(__libm_error_region) -.endp __libm_callout -ASM_SIZE_DIRECTIVE(__libm_callout) +.type __libm_error_support#,@function +.global __libm_error_support# -.type __libm_tan#,@function -.global __libm_tan# |