/* * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. * See https://llvm.org/LICENSE.txt for license information. * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception * */ #include #include #include #if ! defined(VL) #error VL must be specified #endif #define CONFIG 1 #if VL == 2 #include "helperavx2_128.h" #elif VL == 4 #include "helperavx2.h" #elif VL == 8 #include "helperavx512f.h" #else #error VL must be 2, 4, or 8 #endif #define SINCOS_COMMA #if defined(SINE) && !(defined(COSINE) || defined(SINCOS)) #define S(...) __VA_ARGS__ #define C(...) #define FNAME sin #elif defined(COSINE) && !(defined(SINE) || defined(SINCOS)) #define S(...) #define C(...) __VA_ARGS__ #define FNAME cos #elif defined(SINCOS) && !(defined(SINE) || defined(COSINE)) #define S(...) __VA_ARGS__ #define C(...) __VA_ARGS__ #define FNAME sincos #include #undef SINCOS_COMMA #define SINCOS_COMMA , #else #error One of SINE, COSINE, or SINCOS must be defined. #endif #define _CONCAT(l,r) l##r #define CONCAT(l,r) _CONCAT(l,r) #define _CONCAT4(ll,l,r,rr) ll##l##r##rr #define CONCAT4(ll,l,r,rr) _CONCAT4(ll,l,r,rr) #if VL != 8 #define FCN_NAME CONCAT(CONCAT4(__fd_,FNAME,_,VL),_avx2) #else #define FCN_NAME CONCAT(CONCAT4(__fd_,FNAME,_,VL),FCN_AVX512()) #endif #include "sincos.h" #if defined(SINCOS) extern "C" vdouble __mth_return2vectors(const vdouble, const vdouble); #endif extern "C" vdouble FCN_NAME(const vdouble x); static void INLINE __reduction_slowpath(vdouble const a, S(vdouble *vrs, vmask *vhs) SINCOS_COMMA C(vdouble *vrc, vmask *vhc)) { int ivl = sizeof a / sizeof(a[0]); union { vdouble vd; vmask vm; double sd[sizeof(vdouble) / sizeof(double)]; uint64_t sm[sizeof(vmask) / sizeof(uint64_t)]; } S(rs, hs) SINCOS_COMMA C(rc, hc); for (int i = 0; i < ivl; ++i) { reduction_slowpath(a[i], S(&rs.sd[i], &hs.sm[i]) SINCOS_COMMA C(&rc.sd[i], &hc.sm[i])); } S(*vrs = rs.vd;) S(*vhs = hs.vm;) C(*vrc = rc.vd;) C(*vhc = hc.vm;) } vdouble __attribute__((noinline)) FCN_NAME(const vdouble x) { S(vdouble as, ks, rs;) S(vint2 hs;) S(vdouble ss, fs, ts;) C(vdouble ac, kc, rc;) C(vint2 hc;) C(vdouble sc, fc, tc;) vint2 p; p = vand_vi2_vi2_vi2((vint2)x, (vint2)vcast_vm_i_i(0x7fffffff, 0xffffffff)); S(ks = vfma_vd_vd_vd_vd(x, vcast_vd_d(_1_OVER_PI), vcast_vd_d(6755399441055744.0));) S(hs = vsll64_vi2_vi2_i((vint2)ks, 63);) S(ks = vsub_vd_vd_vd(ks, vcast_vd_d(6755399441055744.0));) C(kc = vfma_vd_vd_vd_vd(x, vcast_vd_d(_1_OVER_PI), vcast_vd_d(-0.5));) C(kc = vadd_vd_vd_vd(kc, vcast_vd_d(6755399441055744.0));) C(hc = vsll64_vi2_vi2_i((vint2)kc, 63);) C(kc = vsub_vd_vd_vd(kc, vcast_vd_d(6755399441055744.0));) C(kc = vadd_vd_vd_vd(kc, vcast_vd_d(0.5));) S(as = vfma_vd_vd_vd_vd(ks, vcast_vd_d(-PI_HI), x);) C(ac = vfma_vd_vd_vd_vd(kc, vcast_vd_d(-PI_HI), x);) S(as = vfma_vd_vd_vd_vd(ks, vcast_vd_d(-PI_MI), as);) C(ac = vfma_vd_vd_vd_vd(kc, vcast_vd_d(-PI_MI), ac);) S(as = vfma_vd_vd_vd_vd(ks, vcast_vd_d(-PI_LO), as);) C(ac = vfma_vd_vd_vd_vd(kc, vcast_vd_d(-PI_LO), ac);) vopmask mask_lrg_args = vgt64_vo_vi2_vi2(p, (vint2)vcast_vd_d(THRESHOLD)); if (__builtin_expect(!vtestz_i_vo(mask_lrg_args), 0)) { S(vdouble spas;) S(vmask sphs;) C(vdouble spac;) C(vmask sphc;) __reduction_slowpath(x, S(&spas, &sphs) SINCOS_COMMA C(&spac, &sphc)); S(as = vsel_vd_vo_vd_vd(mask_lrg_args, spas, as);) C(ac = vsel_vd_vo_vd_vd(mask_lrg_args, spac, ac);) S(hs = (vmask)vsel_vi2_vo_vi2_vi2(mask_lrg_args, (vint2)sphs, (vint2)hs);) C(hc = (vmask)vsel_vi2_vo_vi2_vi2(mask_lrg_args, (vint2)sphc, (vint2)hc);) } S(ss = vmul_vd_vd_vd(as, as);) C(sc = vmul_vd_vd_vd(ac, ac);) S(rs = vfma_vd_vd_vd_vd(vcast_vd_d(A_D), ss, vcast_vd_d(B_D));) C(rc = vfmapn_vd_vd_vd_vd(vcast_vd_d(-A_D), sc, vcast_vd_d(B_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(C_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(C_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(D_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(D_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(E_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(E_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(F_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(F_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(G_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(G_D));) S(rs = vfma_vd_vd_vd_vd(rs, ss, vcast_vd_d(H_D));) C(rc = vfmapn_vd_vd_vd_vd(rc, sc, vcast_vd_d(H_D));) S(fs = (vdouble)vxor_vi2_vi2_vi2((vint2)as, hs);) C(fc = (vdouble)vxor_vi2_vi2_vi2((vint2)ac, hc);) S(ts = vfma_vd_vd_vd_vd(ss, fs, vcast_vd_d(0.0));) C(tc = vfma_vd_vd_vd_vd(sc, fc, vcast_vd_d(0.0));) S(rs = vfma_vd_vd_vd_vd(rs, ts, fs);) C(rc = vfmapn_vd_vd_vd_vd(rc, tc, fc);) C(vopmask m0 = vgt64_vo_vi2_vi2(p, (vint2)vcast_vm_i_i(0x3e46a09e, 0x667f3bcc));) C(rc = vsel_vd_vo_vd_vd(m0, rc, vcast_vd_d(1.0));) vopmask ninf = vgt64_vo_vi2_vi2((vint2)vcast_vm_i_i(0x7ff00000, 0), p); S(rs = vsel_vd_vo_vd_vd(ninf, rs, vmul_vd_vd_vd(x, vcast_vd_d(0.0)));) C(rc = vsel_vd_vo_vd_vd(ninf, rc, vmul_vd_vd_vd(x, vcast_vd_d(0.0)));) #if defined (SINCOS) return __mth_return2vectors(rs, rc); #else S(return rs;) C(return rc;) #endif } #ifdef UNIT_TEST int main() { #if VL == 2 //vdouble a = {-M_PI+(-M_PI/6), -M_PI/6, M_PI/6, M_PI-(M_PI/6)}; //vdouble a = {-M_PI+(-M_PI/6), -M_PI/6}; //vdouble a = {10*(-M_PI+(-M_PI/6)), 10*(-M_PI/6), 10*(M_PI/6), 10*(M_PI-(M_PI/6))}; //vdouble a = {11*(-M_PI+(-M_PI/6)), 11*(-M_PI/6)}; //vdouble a = {11*(M_PI/6), 11*(M_PI-(M_PI/6))}; vdouble a = {0.1250000, 0.1250000*2.5}; //vdouble a = {-M_PI+(M_PI/3), -M_PI/3, M_PI/3, M_PI-(M_PI/3)}; //vdouble a = {-M_PI+(M_PI/3), -M_PI/4}; //vdouble a = {-40000-M_PI+(M_PI/3), -40000-M_PI/3, 40000+M_PI/3, 40000+M_PI-(M_PI/3)}; //vdouble a = {-40000-M_PI+(M_PI/3), -INFINITY, 40000+M_PI/3, 40000+M_PI-(M_PI/3)}; //vdouble a = {-40000-M_PI+(M_PI/3), -INFINITY}; //vdouble a = {.1, .2}; //vdouble a = {THRESHOLD*2+M_PI/6, -THRESHOLD*2-M_PI/4}; //vdouble a = {THRESHOLD*100000+M_PI/6, -THRESHOLD*2-M_PI/4}; //vdouble a = {THRESHOLD*100000+M_PI/6, INFINITY}; //vdouble a = {M_PI/6, 0.00000001053671213}; //vdouble a = {M_PI/6, 0.00000000000671213}; vdouble r; // uint32_t inf32 = 0x7f800000; // a[1] = *(float *)&inf32; uint64_t someval = 0x3e46a09e667f3bcc; printf("someval=%20.17f\n", *(double *)&someval); #ifdef SINCOS vdouble rs, rc; rs = FCN_NAME(a); asm( //"vmovups %%xmm0,%0\n" "vmovups %%xmm1,%1\n" : "=m"(rs), "=m"(rc) : :); #else S(r = FCN_NAME(a);) C(r = FCN_NAME(a);) #endif printf("%f %f\n", a[0], a[1]); S(printf("ref(sin)\t%f %f\n", sin(a[0]), sin(a[1]));) C(printf("ref(cos)\t%f %f\n", cos(a[0]), cos(a[1]));) #ifdef SINCOS printf("new(sin)\t%f %f %f %f\n", rs[0], rs[1], rs[0]-sin(a[0]), rs[1]-sin(a[1])); printf("new(cos)\t%f %f %f %f\n", rc[0], rc[1], rc[0]-cos(a[0]), rc[1]-cos(a[1])); #else printf("%f %f\n", r[0], r[1]); #endif #endif return 0; } asm ("\n\ .text\n\ .globl ret2vrs\n\ .type ret2vrs,@function\n\ ret2vrs:\n\ ret\n\ .size ret2vrs,.-ret2vrs\n\ "); #endif // #ifdef UNIT_TEST // vim: ts=4 expandtab