diff options
Diffstat (limited to 'sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S')
| -rw-r--r-- | sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S | 265 |
1 files changed, 265 insertions, 0 deletions
diff --git a/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S new file mode 100644 index 0000000000..a3f6d21ce1 --- /dev/null +++ b/sysdeps/x86_64/fpu/multiarch/svml_s_hypotf4_core_sse4.S @@ -0,0 +1,265 @@ +/* Function hypotf vectorized with SSE4. + Copyright (C) 2021 Free Software Foundation, Inc. + This file is part of the GNU C Library. + + The GNU C Library is free software; you can redistribute it and/or + modify it under the terms of the GNU Lesser General Public + License as published by the Free Software Foundation; either + version 2.1 of the License, or (at your option) any later version. + + The GNU C Library is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + Lesser General Public License for more details. + + You should have received a copy of the GNU Lesser General Public + License along with the GNU C Library; if not, see + https://www.gnu.org/licenses/. */ + +/* + * ALGORITHM DESCRIPTION: + * + * HIGH LEVEL OVERVIEW + * + * Calculate z = (x*x+y*y) + * Calculate reciplicle sqrt (z) + * Calculate make two NR iterations + * + * ALGORITHM DETAILS + * + * Multiprecision branch for _HA_ only + * Remove sigm from both arguments + * Find maximum (_x) and minimum (_y) (by abs value) between arguments + * Split _x int _a and _b for multiprecision + * If _x >> _y we will we will not split _y for multiprecision + * all _y will be put into lower part (_d) and higher part (_c = 0) + * Fixing _hilo_mask for the case _x >> _y + * Split _y into _c and _d for multiprecision with fixed mask + * + * compute Hi and Lo parts of _z = _x*_x + _y*_y + * + * _zHi = _a*_a + _c*_c + * _zLo = (_x + _a)*_b + _d*_y + _d*_c + * _z = _zHi + _zLo + * + * No multiprecision branch for _LA_ and _EP_ + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 + * + * Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout + * + * Compute resciplicle sqrt s0 ~ 1.0/sqrt(_z), + * that multiplied by _z, is final result for _EP_ version. + * + * First iteration (or zero iteration): + * s = z * s0 + * h = .5 * s0 + * d = s * h - .5 + * + * Second iteration: + * h = d * h + h + * s = s * d + s + * d = s * s - z (in multiprecision for _HA_) + * + * result = s - h * d + * + * EP version of the function can be implemented as y[i]=sqrt(a[i]^2+b[i]^2) + * with all intermediate operations done in target precision for i=1,..,n. + * It can return result y[i]=0 in case a[i]^2 and b[i]^2 underflow in target + * precision (for some i). It can return result y[i]=NAN in case + * a[i]^2+b[i]^2 overflow in target precision, for some i. It can return + * result y[i]=NAN in case a[i] or b[i] is infinite, for some i. + * + * + */ + +/* Offsets for data table __svml_shypot_data_internal + */ +#define _sHiLoMask 0 +#define _sAbsMask 16 +#define _sHalf 32 +#define _LowBoundary 48 +#define _HighBoundary 64 + +#include <sysdep.h> + + .text + .section .text.sse4,"ax",@progbits +ENTRY(_ZGVbN4vv_hypotf_sse4) + subq $88, %rsp + cfi_def_cfa_offset(96) + +/* + * Implementation + * Multiprecision branch for _HA_ only + * No multiprecision branch for _LA_ + * _z = _VARG1 * _VARG1 + _VARG2 * _VARG2 + */ + movaps %xmm0, %xmm8 + movaps %xmm1, %xmm2 + mulps %xmm0, %xmm8 + mulps %xmm1, %xmm2 + +/* + * Variables + * Defines + * Constants loading + */ + movups _sHalf+__svml_shypot_data_internal(%rip), %xmm5 + addps %xmm2, %xmm8 + +/* _s0 ~ 1.0/sqrt(_z) */ + rsqrtps %xmm8, %xmm10 + +/* First iteration */ + movaps %xmm10, %xmm2 + movaps %xmm8, %xmm3 + mulps %xmm8, %xmm2 + mulps %xmm5, %xmm10 + movaps %xmm2, %xmm6 + mulps %xmm10, %xmm6 + +/* Check _z exponent to be withing borders [1E3 ; 60A] else goto Callout */ + movdqu _LowBoundary+__svml_shypot_data_internal(%rip), %xmm4 + subps %xmm6, %xmm5 + +/* Second iteration */ + movaps %xmm5, %xmm7 + pcmpgtd %xmm8, %xmm4 + mulps %xmm2, %xmm5 + mulps %xmm10, %xmm7 + addps %xmm5, %xmm2 + addps %xmm7, %xmm10 + +/* Finish second iteration in native precision for _LA_ */ + movaps %xmm2, %xmm9 + mulps %xmm2, %xmm9 + pcmpgtd _HighBoundary+__svml_shypot_data_internal(%rip), %xmm3 + subps %xmm8, %xmm9 + mulps %xmm9, %xmm10 + por %xmm3, %xmm4 + movmskps %xmm4, %edx + subps %xmm10, %xmm2 + +/* The end of implementation */ + testl %edx, %edx + +/* Go to special inputs processing branch */ + jne L(SPECIAL_VALUES_BRANCH) + # LOE rbx rbp r12 r13 r14 r15 edx xmm0 xmm1 xmm2 + +/* Restore registers + * and exit the function + */ + +L(EXIT): + movaps %xmm2, %xmm0 + addq $88, %rsp + cfi_def_cfa_offset(8) + ret + cfi_def_cfa_offset(96) + +/* Branch to process + * special inputs + */ + +L(SPECIAL_VALUES_BRANCH): + movups %xmm0, 32(%rsp) + movups %xmm1, 48(%rsp) + movups %xmm2, 64(%rsp) + # LOE rbx rbp r12 r13 r14 r15 edx + + xorl %eax, %eax + movq %r12, 16(%rsp) + cfi_offset(12, -80) + movl %eax, %r12d + movq %r13, 8(%rsp) + cfi_offset(13, -88) + movl %edx, %r13d + movq %r14, (%rsp) + cfi_offset(14, -96) + # LOE rbx rbp r15 r12d r13d + +/* Range mask + * bits check + */ + +L(RANGEMASK_CHECK): + btl %r12d, %r13d + +/* Call scalar math function */ + jc L(SCALAR_MATH_CALL) + # LOE rbx rbp r15 r12d r13d + +/* Special inputs + * processing loop + */ + +L(SPECIAL_VALUES_LOOP): + incl %r12d + cmpl $4, %r12d + +/* Check bits in range mask */ + jl L(RANGEMASK_CHECK) + # LOE rbx rbp r15 r12d r13d + + movq 16(%rsp), %r12 + cfi_restore(12) + movq 8(%rsp), %r13 + cfi_restore(13) + movq (%rsp), %r14 + cfi_restore(14) + movups 64(%rsp), %xmm2 + +/* Go to exit */ + jmp L(EXIT) + cfi_offset(12, -80) + cfi_offset(13, -88) + cfi_offset(14, -96) + # LOE rbx rbp r12 r13 r14 r15 xmm2 + +/* Scalar math fucntion call + * to process special input + */ + +L(SCALAR_MATH_CALL): + movl %r12d, %r14d + movss 32(%rsp,%r14,4), %xmm0 + movss 48(%rsp,%r14,4), %xmm1 + call hypotf@PLT + # LOE rbx rbp r14 r15 r12d r13d xmm0 + + movss %xmm0, 64(%rsp,%r14,4) + +/* Process special inputs in loop */ + jmp L(SPECIAL_VALUES_LOOP) + # LOE rbx rbp r15 r12d r13d +END(_ZGVbN4vv_hypotf_sse4) + + .section .rodata, "a" + .align 16 + +#ifdef __svml_shypot_data_internal_typedef +typedef unsigned int VUINT32; +typedef struct +{ + __declspec(align(16)) VUINT32 _sHiLoMask[4][1]; + __declspec(align(16)) VUINT32 _sAbsMask[4][1]; + __declspec(align(16)) VUINT32 _sHalf[4][1]; + __declspec(align(16)) VUINT32 _LowBoundary[4][1]; + __declspec(align(16)) VUINT32 _HighBoundary[4][1]; +} __svml_shypot_data_internal; +#endif +__svml_shypot_data_internal: + /* legacy algorithm */ + .long 0xFFF80000, 0xFFF80000, 0xFFF80000, 0xFFF80000 /* _sHiLoMask */ + .align 16 + .long 0x7fffffff, 0x7fffffff, 0x7fffffff, 0x7fffffff /* _sAbsMask */ + .align 16 + .long 0x3f000000, 0x3f000000, 0x3f000000, 0x3f000000 /* _sHalf */ + .align 16 + .long 0x1E300000, 0x1E300000, 0x1E300000, 0x1E300000 /* _LowBoundary */ + .align 16 + .long 0x60A00000, 0x60A00000, 0x60A00000, 0x60A00000 /* _HighBoundary */ + .align 16 + .type __svml_shypot_data_internal,@object + .size __svml_shypot_data_internal,.-__svml_shypot_data_internal |