| Kyle Swenson | 8d8f654 | 2021-03-15 11:02:55 -0600 | [diff] [blame^] | 1 | /* |
| 2 | * Copyright (C) 1999-2000 Hewlett-Packard Co |
| 3 | * Copyright (C) 1999-2000 David Mosberger-Tang <davidm@hpl.hp.com> |
| 4 | * |
| 5 | * 64-bit integer division. |
| 6 | * |
| 7 | * This code is based on the application note entitled "Divide, Square Root |
| 8 | * and Remainder Algorithms for the IA-64 Architecture". This document |
| 9 | * is available as Intel document number 248725-002 or via the web at |
| 10 | * http://developer.intel.com/software/opensource/numerics/ |
| 11 | * |
| 12 | * For more details on the theory behind these algorithms, see "IA-64 |
| 13 | * and Elementary Functions" by Peter Markstein; HP Professional Books |
| 14 | * (http://www.hp.com/go/retailbooks/) |
| 15 | */ |
| 16 | |
| 17 | #include <asm/asmmacro.h> |
| 18 | |
| 19 | #ifdef MODULO |
| 20 | # define OP mod |
| 21 | #else |
| 22 | # define OP div |
| 23 | #endif |
| 24 | |
| 25 | #ifdef UNSIGNED |
| 26 | # define SGN u |
| 27 | # define INT_TO_FP(a,b) fcvt.xuf.s1 a=b |
| 28 | # define FP_TO_INT(a,b) fcvt.fxu.trunc.s1 a=b |
| 29 | #else |
| 30 | # define SGN |
| 31 | # define INT_TO_FP(a,b) fcvt.xf a=b |
| 32 | # define FP_TO_INT(a,b) fcvt.fx.trunc.s1 a=b |
| 33 | #endif |
| 34 | |
| 35 | #define PASTE1(a,b) a##b |
| 36 | #define PASTE(a,b) PASTE1(a,b) |
| 37 | #define NAME PASTE(PASTE(__,SGN),PASTE(OP,di3)) |
| 38 | |
| 39 | GLOBAL_ENTRY(NAME) |
| 40 | .regstk 2,0,0,0 |
| 41 | // Transfer inputs to FP registers. |
| 42 | setf.sig f8 = in0 |
| 43 | setf.sig f9 = in1 |
| 44 | ;; |
| 45 | // Convert the inputs to FP, to avoid FP software-assist faults. |
| 46 | INT_TO_FP(f8, f8) |
| 47 | INT_TO_FP(f9, f9) |
| 48 | ;; |
| 49 | frcpa.s1 f11, p6 = f8, f9 // y0 = frcpa(b) |
| 50 | ;; |
| 51 | (p6) fmpy.s1 f7 = f8, f11 // q0 = a*y0 |
| 52 | (p6) fnma.s1 f6 = f9, f11, f1 // e0 = -b*y0 + 1 |
| 53 | ;; |
| 54 | (p6) fma.s1 f10 = f7, f6, f7 // q1 = q0*e0 + q0 |
| 55 | (p6) fmpy.s1 f7 = f6, f6 // e1 = e0*e0 |
| 56 | ;; |
| 57 | #ifdef MODULO |
| 58 | sub in1 = r0, in1 // in1 = -b |
| 59 | #endif |
| 60 | (p6) fma.s1 f10 = f10, f7, f10 // q2 = q1*e1 + q1 |
| 61 | (p6) fma.s1 f6 = f11, f6, f11 // y1 = y0*e0 + y0 |
| 62 | ;; |
| 63 | (p6) fma.s1 f6 = f6, f7, f6 // y2 = y1*e1 + y1 |
| 64 | (p6) fnma.s1 f7 = f9, f10, f8 // r = -b*q2 + a |
| 65 | ;; |
| 66 | #ifdef MODULO |
| 67 | setf.sig f8 = in0 // f8 = a |
| 68 | setf.sig f9 = in1 // f9 = -b |
| 69 | #endif |
| 70 | (p6) fma.s1 f11 = f7, f6, f10 // q3 = r*y2 + q2 |
| 71 | ;; |
| 72 | FP_TO_INT(f11, f11) // q = trunc(q3) |
| 73 | ;; |
| 74 | #ifdef MODULO |
| 75 | xma.l f11 = f11, f9, f8 // r = q*(-b) + a |
| 76 | ;; |
| 77 | #endif |
| 78 | getf.sig r8 = f11 // transfer result to result register |
| 79 | br.ret.sptk.many rp |
| 80 | END(NAME) |