/* * ppc64_helpers.h * Copyright (C) 2016-2017 Will Schmidt * * This file contains helper functions for the ISA 3.0 test suite. */ /* * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License V2 * as published by the Free Software Foundation * * This program 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 General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program; if not, see . */ #include "tests/malloc.h" // memalign32 typedef uint64_t HWord_t; #if defined (DEBUG_ARGS_BUILD) #define AB_DPRINTF(fmt, args...) do { fprintf(stderr, fmt , ##args); } while (0) #else #define AB_DPRINTF(fmt, args...) do { } while (0) #endif /* Exhaustive tests? * Due to the excessive size of the test results, allow a #ifdef to * enable/disable most of the input values. * Off by default. */ // #define EXHAUSTIVE_TESTS 1 #define ALLCR "cr0","cr1","cr2","cr3","cr4","cr5","cr6","cr7" #define SET_CR(_arg) \ __asm__ __volatile__ ("mtcr %0" : : "b"(_arg) : ALLCR ); #define SET_CR0_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x80,%0 " : : "b" (_arg):"cr0"); #define SET_CR1_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x40,%0 " : : "b" (_arg):"cr1"); #define SET_CR2_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x20,%0 " : : "b" (_arg):"cr2"); #define SET_CR3_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x10,%0 " : : "b" (_arg):"cr3"); #define SET_CR4_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x08,%0 " : : "r" (_arg):"cr4"); #define SET_CR5_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x04,%0 " : : "r" (_arg):"cr5"); #define SET_CR6_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x02,%0 " : : "r" (_arg):"cr6"); #define SET_CR7_FIELD(_arg) __asm__ __volatile__ ("mtocrf 0x01,%0 " : : "r" (_arg):"cr7"); #define SET_XER(_arg) \ __asm__ __volatile__ ("mtxer %0" : : "b"(_arg) : "xer" ); #define GET_CR(_lval) \ __asm__ __volatile__ ("mfcr %0" : "=b"(_lval) ) #define GET_XER(_lval) \ __asm__ __volatile__ ("mfxer %0" : "=b"(_lval) ) #define SET_CR_ZERO \ SET_CR(0) #define SET_FPSCR_ZERO \ do { \ double _d = 0.0; \ __asm__ __volatile__ ("mtfsf 0xFF, %0" : : "f"(_d) ); \ } while (0); #define GET_FPSCR(_arg) \ __asm__ __volatile__ ("mffs %0" : "=f"(_arg) ); /* The bit definitions for the FPSCR are as follows. Bit(s) Description 0:31 Reserved 32 Floating-Point Exception Summary (FX) 33 Floating-Point Enabled Exception Summary (FEX) 34 Floating-Point Invalid Operation Exception Summary (VX) 35 Floating-Point Overflow Exception (OX) 36 Floating-Point Underflow Exception (UX) 37 Floating-Point Zero Divide Exception (ZX) 38 Floating-Point Inexact Exception (XX) 39 Floating-Point Invalid Operation Exception (SNaN) (VXSNAN) 40 Floating-Point Invalid Operation Exception (∞ - ∞) (VXISI) 41 Floating-Point Invalid Operation Exception (∞ ÷ ∞) (VXIDI) 42 Floating-Point Invalid Operation Exception (0 ÷ 0) (VXZDZ) 43 Floating-Point Invalid Operation Exception (∞ × 0) (VXIMZ) 44 Floating-Point Invalid Operation Exception (Invalid Compare) (VXVC) 45 Floating-Point Fraction Rounded (FR) 46 Floating-Point Fraction Inexact (FI) 47:51 Floating-Point Result Flags (FPRF) 47 Floating-Point Result Class Descriptor (C) 48:51 Floating-Point Condition Code (FPCC) 48 Floating-Point Less Than or Negative (FL or <) 49 Floating-Point Greater Than or Positive (FG or >) 50 Floating-Point Equal or Zero (FE or =) 51 Floating-Point Unordered or NaN (FU or ?) 52 Reserved 53 Floating-Point Invalid Operation Exception (Software-Defined Condition) (VXSOFT) 54 Floating-Point Invalid Operation Exception (Invalid Square Root) (VXSQRT) 55 Floating-Point Invalid Operation Exception (Invalid Integer Convert) (VXCVI) 56 Floating-Point Invalid Operation Exception Enable (VE) 57 Floating-Point Overflow Exception Enable (OE) 58 Floating-Point Underflow Exception Enable (UE) 59 Floating-Point Zero Divide Exception Enable (ZE) 60 Floating-Point Inexact Exception Enable (XE) 61 Floating-Point Non-IEEE Mode (NI) 62:63 Floating-Point Rounding Control (RN) 00 Round to Nearest 01 Round toward Zero 10 Round toward +Infinity 11 Round toward -Infinity */ /* NOTE, currently Valgrind only tracks the rounding mode, C and FPCC fields in the * FPSCR register. */ static char * fpscr_strings[] = { " 0-RSVD", " 1-RSVD", " 2-RSVD", " 3-RSVD", " 4-RSVD", " 5-RSVD", " 6-RSVD", " 7-RSVD", " 8-RSVD", " 9-RSVD", "10-RSVD", "11-RSVD", "12-RSVD", "13-RSVD", "14-RSVD", "15-RSVD", "16-RSVD", "17-RSVD", "18-RSVD", "19-RSVD", "20-RSVD", "21-RSVD", "22-RSVD", "23-RSVD", "24-RSVD", "25-RSVD", "26-RSVD", "27-RSVD", "28-RSVD", "29-DRN0", "30-DRN1", "31-DRN2", /* 32 */ "FX", "FEX", "VX", /* 35 */ "OX", "UX", "ZX", "XX", "VXSNAN", /* 40 */ "VXISI (inf-inf)", "VXIDI (inf/inf)", "VXZDZ (0/0)", /* 43 */ "VXIMZ (inf*0)", "VXVC", /* 45 */ "FR", "FI", /* 47 */ "FPRF-C", "FPCC-FL", "FPCC-FG", /* 50 */ "FPCC-FE", "FPCC-FU", /* 52 */ "52-RSVD", "FXSOFT", "VXSQRT", /* 55 */ "VXCVI", "VE", "OE", "UE", "ZE", /* 60 */ "XE", "NI", "RN-bit62", "RN-bit63" }; #define FPCC_C_BIT (0x1 << (63-47)) #define FPCC_FL_BIT (0x1 << (63-48)) #define FPCC_FG_BIT (0x1 << (63-49)) #define FPCC_FE_BIT (0x1 << (63-50)) #define FPCC_FU_BIT (0x1 << (63-51)) #define FPCC_FPRF_MASK FPCC_C_BIT|FPCC_FL_BIT|FPCC_FG_BIT|FPCC_FE_BIT|FPCC_FU_BIT #define FPSCR_RN_BIT62 (0x1 << (63-62)) #define FPSCR_RN_BIT63 (0x1 << (63-63)) #define CRFIELD_BIT0 0x8 #define CRFIELD_BIT1 0x4 #define CRFIELD_BIT2 0x2 #define CRFIELD_BIT3 0x1 /* dissect_cr*: * display the condition register bits in a * human readable format. */ static inline int cr_overflow_set(unsigned this_cr) { return (this_cr & CRFIELD_BIT3); } static inline int cr_zero_set(unsigned this_cr) { return (this_cr & CRFIELD_BIT2); } static inline int cr_positive_set(unsigned this_cr) { return (this_cr & CRFIELD_BIT1); } static inline int cr_negative_set(unsigned this_cr) { return (this_cr & CRFIELD_BIT0); } /* __dissect_cr takes a bitfield directly, not the full condition register. * This is a helper for dissect_cr_rn. */ inline static void __dissect_cr(unsigned this_cr) { if (cr_negative_set(this_cr)) printf("%s(LT)", verbose ? " 0x1=Negative" : ""); if (cr_positive_set(this_cr)) printf("%s(GT)", verbose ? " 0x2=Positive" : ""); if (cr_zero_set(this_cr)) printf("%s(EQ)", verbose ? " 0x4=Zero" : ""); if (cr_overflow_set(this_cr)) printf("%s(SO)", verbose ? " 0x8=Overflow" : ""); } /* Extract one CR field */ static int extract_cr_rn(unsigned long local_cr,unsigned long rn) { unsigned int masked_cr; unsigned long shifted_value; shifted_value = local_cr >> ( ( (7 - rn) * 4 ) ); masked_cr = shifted_value & 0xf; return masked_cr; } /* Display one CR field */ static void dissect_cr_rn(unsigned long local_cr, unsigned long rn) { unsigned int masked_cr; masked_cr = extract_cr_rn(local_cr, rn); __dissect_cr(masked_cr); } /* Display all of the CR fields... */ static void dissect_cr(unsigned long local_cr) { unsigned int crn; for (crn = 0; crn < 8; crn++) { dissect_cr_rn(local_cr, crn); } } /* dissect the fpscr bits that are valid under valgrind. * Valgrind tracks the C (FPSCR[47]), FPCC (FPSCR[48:51) * DRN (FPSCR[29:31]) and RN (FPSCR[62:63]). */ static void dissect_fpscr_valgrind(unsigned long local_fpscr) { int i; long mybit; /* Print DRN fields */ for (i = 29; i < 32; i++) { mybit = 1LL << (63 - i); if (mybit & local_fpscr) { printf(" %s",fpscr_strings[i]); } } /* Print C and FPCC fields */ for (i = 47; i < 52; i++) { mybit = 1LL << (63 - i); if (mybit & local_fpscr) { printf(" %s",fpscr_strings[i]); } } /* Print RN field */ for (i = 62; i < 64; i++) { mybit = 1LL << (63 - i); if (mybit & local_fpscr) { printf(" %s",fpscr_strings[i]); } } } /* dissect the fpscr bits. * This prints the entire FPSCR field. This is only called under higher * verbosities, as valgrind does not track most of these bits. */ static void dissect_fpscr_raw(unsigned long local_fpscr) { /* Due to the additional involved logic, the rounding mode (RN) bits 61-62 * are handled within dissect_fpscr_rounding_mode(). */ int i; long mybit; for (i = 0; i < 61; i++) { /* also note that the bit numbering is backwards. */ mybit = 1LL << (63 - i); if (mybit & local_fpscr) { printf(" %s", fpscr_strings[i]); } } } static void dissect_fpscr(unsigned long local_fpscr) { if (verbose > 1) { printf(" [[ fpscr:%lx ]] ", local_fpscr); dissect_fpscr_raw(local_fpscr); } else { dissect_fpscr_valgrind(local_fpscr); } } /* Display the rounding mode */ static void dissect_fpscr_rounding_mode(unsigned long local_fpscr) { /* special case handing for the rounding mode round-nearest (RN) bits. 62:63 */ printf("Rounding Mode: "); if (local_fpscr & FPSCR_RN_BIT62) if (local_fpscr & FPSCR_RN_BIT63) /* 0b11 */ printf("RN-to--INF"); else /* 0b10 */ printf("RN-to-+INF"); else if (local_fpscr & FPSCR_RN_BIT63) /* 0b01 */ printf("RN-to-Nearest"); else /* 0b00 */ printf("RN-to-Zero"); } /* * Arithmetic, rounding, and Convert From Integer instructions will set * bits in the FPCC field to indicate the class of the result. * The table is described as follows; flags / Result value class C < > = ? 1 0 0 0 1 Quiet NaN 0 1 0 0 1 -Infinity 0 1 0 0 0 -Normalized Number 1 1 0 0 0 -Denormalized Number 1 0 0 1 0 -Zero 0 0 0 1 0 +Zero 1 0 1 0 0 +Denormalized Number 0 0 1 0 0 +Normalized Number 0 0 1 0 1 +Infinity */ static void dissect_fpscr_result_value_class(unsigned long local_fpscr) { if (local_fpscr & FPCC_C_BIT) { if (local_fpscr & FPCC_FL_BIT) printf("-Denormalized"); else if (local_fpscr & FPCC_FG_BIT) printf("+Denormalized"); else if (local_fpscr & FPCC_FE_BIT) printf("-Zero "); else if (local_fpscr & FPCC_FU_BIT) printf("Quiet NaN "); } else { if (local_fpscr & FPCC_FL_BIT) { if (local_fpscr & FPCC_FU_BIT) printf("-Infinity "); else printf("-Normalized "); } else if (local_fpscr & FPCC_FG_BIT) { if (local_fpscr & FPCC_FU_BIT) printf("+Infinity "); else printf("+Normalized "); if (local_fpscr & FPCC_FE_BIT) printf("+Zero "); } } } /* Interpret the fields in the FPCC as they apply to the DCMX checks. * The 'Match' indicator will typically be evaluated by the caller. * * DMCX: * DCMX bit / 0x value / Data Class * 0 0x01 NaN * 1 0x02 +Infinity * 2 0x04 -Infinity * 3 0x08 +Zero * 4 0x10 -Zero * 5 0x20 +Denormal * 6 0x40 -Denormal * 7 0x7f ALL bits set. */ static void dissect_fpscr_dcmx_indicator(unsigned long local_fpscr) { if (verbose > 2) printf("fpscr_cc:%lx ", local_fpscr & (FPCC_FPRF_MASK) ); // See if the data class of the src value matches the set DCMX bits. if (verbose > 1) printf("%s ", (local_fpscr&FPCC_FE_BIT) ? "Match":""); // Display the sign bit of the src value. if (verbose > 1) printf("SRC sign:%s ", (local_fpscr&FPCC_FL_BIT) ? "-" : "+"); // The src value can be either a SP or DP value, this indicates // if it is a valid SP value. if (verbose > 1) printf("%s ", (local_fpscr&FPCC_FE_BIT) ? "SP" : ""); } /* dissect_xer helpers*/ static char * xer_strings[] = { " 0-RSVD", " 1-RSVD", " 2-RSVD", " 3-RSVD", " 4-RSVD", " 5-RSVD", " 6-RSVD", " 7-RSVD", " 8-RSVD", " 9-RSVD", "10-RSVD", "11-RSVD", "12-RSVD", "13-RSVD", "14-RSVD", "15-RSVD", "16-RSVD", "17-RSVD", "18-RSVD", "19-RSVD", "20-RSVD", "21-RSVD", "22-RSVD", "23-RSVD", "24-RSVD", "25-RSVD", "26-RSVD", "27-RSVD", "28-RSVD", "29-RSVD", "30-RSVD", "31-RSVD", /* 32 */ "SO", "OV", "CA", /* 35 */ "35-RSVD", "36-RSVD", "37-RSVD", "38-RSVD", "39-RSVD", /* 40 */ "40-RSVD", "41-RSVD", "42-RSVD", "43-RSVD", /* 44 */ "OV32", "CA32", /* 46 */ "46-RSVD", "47-RSVD", "48-RSVD", "49-RSVD", "50-RSVD", "51-RSVD", "52-RSVD", "53-RSVD", "54-RSVD", "55-RSVD", "56-RSVD", /* 57:63 # bytes transferred by a Load/Store String Indexed instruction. */ "LSI/SSI-0", "LSI/SSI-1", "LSI/SSI-2", "LSI/SSI-3", "LSI/SSI-4", "LSI/SSI-5", "LSI/SSI-6", }; /* Dissect the XER register contents. */ static void dissect_xer_raw(unsigned long local_xer) { int i; long mybit; for (i = 0; i <= 63; i++) { mybit = 1ULL << (63 - i); /* compensate for reversed bit numbering. */ if (mybit & local_xer) printf(" %s", xer_strings[i]); } } /* Display only the XER contents that are relevant for our tests. * this is currently the OV and OV32 bits. */ static void dissect_xer_valgrind(unsigned long local_xer) { int i; long mybit; i = 33; // OV mybit = 1ULL << (63 - i); if (mybit & local_xer) printf(" %s", xer_strings[i]); i = 44; // OV32 mybit = 1ULL << (63 - i); if (mybit & local_xer) printf(" %s", xer_strings[i]); } /* */ static void dissect_xer(unsigned long local_xer) { if (verbose > 1) printf(" [[ xer:%lx ]]", local_xer); if (verbose > 2 ) dissect_xer_raw(local_xer); else dissect_xer_valgrind(local_xer); } /* DFP helpers for bcd-to-dpd, dpd-to-bcd, misc. * pulled from vex/.../host_generic_simd64.c */ /*------------------------------------------------------------------*/ /* Decimal Floating Point (DFP) helper functions */ /*------------------------------------------------------------------*/ #define NOT( x ) ( ( ( x ) == 0) ? 1 : 0) #define GET( x, y ) ( ( ( x ) & ( 0x1UL << ( y ) ) ) >> ( y ) ) #define PUT( x, y ) ( ( x )<< ( y ) ) static unsigned long dpb_to_bcd( unsigned long chunk ) { int a, b, c, d, e, f, g, h, i, j, k, m; int p, q, r, s, t, u, v, w, x, y; unsigned long value; /* convert 10 bit densely packed BCD to BCD */ p = GET( chunk, 9 ); q = GET( chunk, 8 ); r = GET( chunk, 7 ); s = GET( chunk, 6 ); t = GET( chunk, 5 ); u = GET( chunk, 4 ); v = GET( chunk, 3 ); w = GET( chunk, 2 ); x = GET( chunk, 1 ); y = GET( chunk, 0 ); /* The BCD bit values are given by the following boolean equations.*/ a = ( NOT(s) & v & w ) | ( t & v & w & s ) | ( v & w & NOT(x) ); b = ( p & s & x & NOT(t) ) | ( p & NOT(w) ) | ( p & NOT(v) ); c = ( q & s & x & NOT(t) ) | ( q & NOT(w) ) | ( q & NOT(v) ); d = r; e = ( v & NOT(w) & x ) | ( s & v & w & x ) | ( NOT(t) & v & x & w ); f = ( p & t & v & w & x & NOT(s) ) | ( s & NOT(x) & v ) | ( s & NOT(v) ); g = ( q & t & w & v & x & NOT(s) ) | ( t & NOT(x) & v ) | ( t & NOT(v) ); h = u; i = ( t & v & w & x ) | ( s & v & w & x ) | ( v & NOT(w) & NOT(x) ); j = ( p & NOT(s) & NOT(t) & w & v ) | ( s & v & NOT(w) & x ) | ( p & w & NOT(x) & v ) | ( w & NOT(v) ); k = ( q & NOT(s) & NOT(t) & v & w ) | ( t & v & NOT(w) & x ) | ( q & v & w & NOT(x) ) | ( x & NOT(v) ); m = y; value = PUT(a, 11) | PUT(b, 10) | PUT(c, 9) | PUT(d, 8) | PUT(e, 7) | PUT(f, 6) | PUT(g, 5) | PUT(h, 4) | PUT(i, 3) | PUT(j, 2) | PUT(k, 1) | PUT(m, 0); return value; } #undef NOT #undef GET #undef PUT typedef union dfp_union { _Decimal128 dec_val128; struct { #if defined(VGP_ppc64le_linux) unsigned long vall; unsigned long valu; #else unsigned long valu; unsigned long vall; #endif } u128; } dfp_val_t; /* Based on and enhanced from the dfp128_vals table in test_dfp5.c. * Todo: Refine/refactor and turn into a build_table function. */ static unsigned long dfp128_vals[] = { #ifdef EXHAUSTIVE_TESTS // Some finite numbers 0x2208000000000000ULL, 0x0000000000000001ULL, // 1 *10^0 0xa208800000000000ULL, 0x0000000000000001ULL, // -1 *10^1 0x0000000000000000ULL, 0x0000000000000001ULL, // 1 *10^-6176. (smallest exp) 0x43ffc00000000000ULL, 0x0000000000000001ULL, // 1 *10^6111 0x6fffc00000000000ULL, 0x0000000000000001ULL, // foo *10^2015. 0x67ffc00000000000ULL, 0x0000000000000001ULL, // foo *10^-2081. 0x77ffc00000000000ULL, 0x0000000000000001ULL, // 1 *10^6111 (largest exp) 0x77ffffffffffffffULL, 0xffffffffffffffffULL, // max possible value *10^6111 (largest exp) 0x0000000000000000ULL, 0x0000000000000001ULL, // min possible value 1 *10^-6176. (smallest exp) 0x8000000000000000ULL, 0x0000000000000001ULL, // -1 *10^-6176. (smallest exp) /* data bits sprinkled across the significand field. */ 0xa208800001000000ULL, 0x0000000000010000ULL, //-foo *10^1 0xa208800000000100ULL, 0x0000000000000100ULL, //-foo *10^1 0xa208800000000000ULL, 0x0000100000000000ULL, //-foo *10^1 0xa208800000000000ULL, 0x0000000001000000ULL, //-foo *10^1 0xa208800000000000ULL, 0x0000000000000001ULL, //-foo *10^1 // pre-existing dfp128 values: 0x2207c00000000000ULL, 0x0000000000000e50ULL, // foo * 10^-1 0x2207c00000000000ULL, 0x000000000014c000ULL, // foo * 10^-1 0xa207c00000000000ULL, 0x00000000000000e0ULL, // foo * 10^-1 0x2206c00000000000ULL, 0x00000000000000cfULL, // foo * 10^-5 0xa205c00000000000ULL, 0x000000010a395bcfULL, // foo * 10^-9 0x6209400000fd0000ULL, 0x00253f1f534acdd4ULL, // foo * 10^-4091 0x000400000089b000ULL, 0x0a6000d000000049ULL, // very small number // foo * 10^-6160 // flavors of zero 0x2208000000000000ULL, 0x0000000000000000ULL, // 0*10^256 0xa208000000000000ULL, 0x0000000000000000ULL, // -0*10^0 0xa248000000000000ULL, 0x0000000000000000ULL, // 0*10^256 // flavors of NAN 0x7c00000000000000ULL, 0x0000000000000000ULL, // quiet 0xfc00000000000000ULL, 0xc00100035b007700ULL, // NAN 0x7e00000000000000ULL, 0xfe000000d0e0a0d0ULL, // signaling NAN // flavors of Infinity 0x7800000000000000ULL, 0x0000000000000000ULL, // +inf 0xf800000000000000ULL, 0x0000000000000000ULL, // -inf 0xf900000000000000ULL, 0x0000000000000000ULL // -inf #else 0x2208000000000000ULL, 0x0000000000000001ULL, // 1 *10^0 0x77ffffffffffffffULL, 0xffffffffffffffffULL, // max possible value *10^6111 (largest exp) 0xa208000000000000ULL, 0x0000000000000000ULL, // -0*10^0 0xfc00000000000000ULL, 0xc00100035b007700ULL, // NAN 0x7e00000000000000ULL, 0xfe000000d0e0a0d0ULL, // signaling NAN 0xf800000000000000ULL, 0x0000000000000000ULL, // -inf #endif }; #define NUM_DFP128_VALS (sizeof(dfp128_vals) / sizeof(unsigned long)) unsigned long nb_dfp128_vals = NUM_DFP128_VALS; /* Todo: update dfp64_vals to match dfp128_vals content. */ static unsigned long dfp64_vals[] = { #ifdef EXHAUSTIVE_TESTS 0x77fcffffffffffffULL, // max possible value 9..9 *10^369 (largest exp) 0x0000000000000001ULL, // min possible nonzero value 1 *10^-398. (smallest exp) 0x4248000000000001ULL, // 1*10^260 0x2234000000000e50ULL, // foo*10^-1 0x223400000014c000ULL, // 0xa2340000000000e0ULL, // 0x22240000000000cfULL, // foo*10^-5 0xa21400010a395bcfULL, // negative -foo*10^-9 0x6e4d3f1f534acdd4ULL, // huge number foo*10^5 0x000400000089b000ULL, // very small number foo*10^-397 // flavors of zero 0x2238000000000000ULL, 0xa238000000000000ULL, // 0 * 10 ^0 0x4248000000000000ULL, // 0 * 10 ^260 // flavors of NAN 0x7e34000000000111ULL, //signaling NaN 0xfe000000d0e0a0d0ULL, //signaling NaN 0xfc00000000000000ULL, //quiet NaN // flavors of Infinity 0x7800000000000000ULL, //+Inf 0xf800000000000000ULL, //-Inf 0x7a34000000000000ULL, //+Inf #else 0x77fcffffffffffffULL, // max possible value 9..9 *10^369 (largest exp) 0x4248000000000000ULL, // 0 * 10 ^260 0xfe000000d0e0a0d0ULL, //signaling NaN 0xf800000000000000ULL, //-Inf #endif }; #define NUM_DFP64_VALS (sizeof(dfp64_vals) / sizeof(unsigned long)) unsigned long nb_dfp64_vals = NUM_DFP64_VALS; /* shift helpers */ #define SH_0 0 #define SH_1 1 #define SH_2 15 #define SH_3 63 static uint64_t shift_amounts[] = { SH_0, SH_1, SH_2, SH_3, #define SHIFT_ARRAY_SIZE 4 }; /* vector splat helpers */ #define SPLAT0 0 #define SPLAT1 1 #define SPLAT2 0xaa #define SPLAT3 0x55 #define SPLAT4 0xff static uint64_t splat_values[] = { SPLAT0, SPLAT1, SPLAT2, SPLAT3, SPLAT4, #define SPLAT_ARRAY_SIZE 5 }; /* a small memory range used to test load-from and store-to vsx */ #define BUFFER_SIZE 4 #define MAX_BUFFER_PATTERNS 6 unsigned long buffer[BUFFER_SIZE]; static void initialize_buffer(int t) { int x; for (x = 0; x < BUFFER_SIZE; x++) /* Don't want each of the 32-bit chunks to be identical. Loads of a * byte from the wrong 32-bit chuck are not detectable if the chunks * are identical. */ switch((t+x)%BUFFER_SIZE) { case 0: buffer[x] = 0xffffffffffffffff; break; case 1: buffer[x] = 0x0001020304050607; break; case 2: buffer[x] = 0x5555555555555555; break; case 3: buffer[x] = 0x0000000000000000; break; case 4: buffer[x] = 0x5a05a05a05a05a05; break; case 5: buffer[x] = 0x0102030405060708; break; default: buffer[x] = 0x1010101010101010; break; } } #define PATTERN_SIZE 5 unsigned long pattern[PATTERN_SIZE] = { 0xffffffffffffffff, 0xaaaaaaaaaaaaaaaa, 0x5152535455565758, 0x0000000000000000, 0xffaa5599113377cc, }; static void dump_small_buffer(void) { int x; printf("[ "); for (x = 0; x < BUFFER_SIZE; x++) printf("%016lx ", buffer[x] ); printf("]"); } /* value to be shifted */ static uint64_t values_to_shift[] = { 0x0, 0x1, 0x10, 0x100, 0x1000, 0x10000, 0x100000, 0x1000000, 0x10000000, 0x100000000, 0x1000000000, 0x10000000000, 0x100000000000, 0x1000000000000, 0x10000000000000, 0x100000000000000, 0x1000000000000000, 0xf, 0x1f, 0x10f, 0x100f, 0x1000f, 0x10000f, 0x100000f, 0x1000000f, 0x10000000f, 0x100000000f, 0x1000000000f, 0x10000000000f, 0x100000000000f, 0x1000000000000f, 0x10000000000000f, 0x100000000000000f, 0x7, 0x70, 0x700, 0x7000, 0x70000, 0x700000, 0x7000000, 0x70000000, 0x700000000, 0x7000000000, 0x70000000000, 0x700000000000, 0x7000000000000, 0x70000000000000, 0x700000000000000, 0x7000000000000000, 0x8, 0x80, 0x800, 0x8000, 0x80000, 0x800000, 0x8000000, 0x80000000, 0x800000000, 0x8000000000, 0x80000000000, 0x800000000000, 0x8000000000000, 0x80000000000000, 0x800000000000000, 0x8000000000000000, 0xffffffffffffffff, 0 #define SHIFT_VALUES_SIZE 66 }; /* DFP related helper functions: */ /* For DFP finite numbers, the combination field (G field) is a * combination of the exponent and the LMD (Left Most Digit) of the * significand. The fields are encoded/decoded as described in the * table here. * 00 01 10 -< Exponent bits. * 0: 00000 01000 10000 * ... * 7: 00111 01111 10111 * 8: 11000 11010 11100 * 9: 11001 11011 11101 (encoded special field). * | * ^ LMD value. */ #define DFP_GFIELD_MASK 0x7c00000000000000UL #define DFP_GFIELD_SHIFT 58 static unsigned int special_field_LMD(uint64_t dword1) { unsigned long g_field_specials; int left_two_bits; int right_three_bits; g_field_specials = (dword1 & DFP_GFIELD_MASK) >> DFP_GFIELD_SHIFT; left_two_bits = (g_field_specials & 0x18) >> 3; right_three_bits = g_field_specials & 0x07; /* The LMD result maps directly to the right_three_bits value as * long as the left two bits are 0b00,0b01,0b10. So a compare * against 3 is sufficient to determine if we can return the right * three bits directly. (LMD values 0..7). */ if (left_two_bits < 3) { return (right_three_bits); } /* LMD values of 8 or 9 require a bit of swizzle, but a check of * the right-most bit is sufficient to determine whether LMD value * is 8 or 9. */ if (right_three_bits & 0x1) return 9; else return 8; } /* Returns the exponent bits, as decoded from the G field. */ static inline int special_field_exponent_bits(unsigned long dword1) { unsigned long g_field_specials; int left_two_bits; int right_three_bits; g_field_specials = (dword1 & DFP_GFIELD_MASK) >> DFP_GFIELD_SHIFT; left_two_bits = (g_field_specials & 0x18) >> 3; right_three_bits = g_field_specials & 0x07; /* The special field exponent bits maps directly to the left_two_bits * value as long as the left two bits are 0b00,0b01,0b10. So a compare * against 3 is sufficient for those values. */ if (left_two_bits < 3) { return (left_two_bits); } switch(right_three_bits) { case 0: case 1: return 0x0; case 2: case 3: return 0x1; case 4: case 5: return 0x2; case 6: /* Infinity */ return 0x0; case 7: /* NaN */ return 0x0; } return -1; /* should never hit this */ } /* get_declet(). Return a 10-bit declet, beginning at the 'start' * offset. * * | dword1 | dword0 | * | 0 63|64 127| */ #define TEN_BITS 0x03ffULL static inline int get_declet(int start, uint64_t dword1, uint64_t dword0) { unsigned long local_declet; unsigned int dword0_shift; unsigned int dword1_shift; dword1_shift = 63 - (start + 9); dword0_shift = 127 - (start + 9); if (verbose>5) printf("\n%s (%d) %016lx %016lx", __FUNCTION__, start, dword1, dword0); if ((start + 9) < 63) { /* fully within dword1 */ local_declet = (dword1 >> dword1_shift) & TEN_BITS; } else if (start >= 65) {/* fully within dword0 */ local_declet = (dword0 >> dword0_shift) & TEN_BITS; } else { /* straddling the two dwords*/ unsigned long mask_dword0; unsigned long mask_dword1; mask_dword1 = TEN_BITS >> (64 - dword0_shift); mask_dword0 = TEN_BITS << (dword0_shift); local_declet = ((dword1 & mask_dword1) << (64-dword0_shift)) + ((dword0 & mask_dword0) >> dword0_shift); } return local_declet; } static int get_bcd_digit_from_dpd(int start, uint64_t dword1, uint64_t dword0) { long bcd_digit; long declet; declet = get_declet(start, dword1, dword0); bcd_digit = dpb_to_bcd(declet); return bcd_digit; } /* The 'exponent left' shift is for moving the leftmost two bits * of the exponent down to where they can be easily merged with the * rest of the exponent. */ #define DFP128_EXPONENT_RIGHT_MASK 0x03ffc00000000000 #define DFP64_EXPONENT_RIGHT_MASK 0x03fc000000000000 #define DFP128_EXPONENT_RIGHT_MASK_SHIFT 46 #define DFP64_EXPONENT_RIGHT_MASK_SHIFT 50 #define DFP128_EXPONENT_LEFT_SHIFT 12 #define DFP64_EXPONENT_LEFT_SHIFT 8 #define DFP_NAN 0x1f #define DFP_INF 0x1e #define DFP_SIGNALING_NAN_BIT 0x0200000000000000 /* Start of the Trailing Significand field is at bit # .. */ #define DFP128_T_START 18 #define DFP64_T_START 14 //The exponent bias value is 101 for DFP Short, 398 //for DFP Long, and 6176 for DFP Extended. #define DFP128_EXPONENT_BIAS 6176 #define DFP64_EXPONENT_BIAS 398 /* return the dfp exponent from the leading dword. */ static inline signed long dfp128_exponent(unsigned long dword1) { unsigned long exponent_left; unsigned long exponent_right; unsigned long biased_exponent; signed long exponent; exponent_left = special_field_exponent_bits(dword1); exponent_right = (dword1 & DFP128_EXPONENT_RIGHT_MASK); biased_exponent = (exponent_left << DFP128_EXPONENT_LEFT_SHIFT) + (exponent_right >> DFP128_EXPONENT_RIGHT_MASK_SHIFT); /* Unbias the exponent. */ exponent = biased_exponent - DFP128_EXPONENT_BIAS; return exponent; } /* Interpret the paired 64-bit values as a extended (quad) 128 bit DFP. * * | Significand | Combination Field/ | | * | sign bit | Encoded Exponent | remainder of significand | * |0 |1 17|18 127| * ^ (bit0) Significand sign bit. * ^ (bit 1:17) Combination field. Contains high bits of * exponent (encoded), LMD of significand (encoded), * and the remainder of the exponent. First five bits * will indicate special cases NAN or INF. * ^ (bit 18:127) Remainder of the * significand. */ #define DFP128_COMBINATION_MASK 0x7fffc #define DFP64_COMBINATION_MASK 0x7ffc #define DFP128_COMBINATION_SHIFT 46 #define DFP64_COMBINATION_SHIFT 50 #define DFP_SPECIAL_SYMBOLS_MASK 0x1f #define DFP_SPECIAL_SYMBOLS_SHIFT 58 static inline void dissect_dfp128_float(uint64_t dword1, uint64_t dword0) { long signbit; signed long exponent; unsigned long gfield_special_symbols; unsigned long lmd_digit; unsigned long bcd_digits[13]; int i; int silent=0; // suppress leading zeros from the output. if (verbose > 5) printf("RAW128: %016lx %016lx ", dword1, dword0); signbit = (dword1 >> 63); if (signbit) printf("-"); else printf("+"); gfield_special_symbols = ((dword1 >> DFP_SPECIAL_SYMBOLS_SHIFT) & DFP_SPECIAL_SYMBOLS_MASK); switch (gfield_special_symbols) { case DFP_INF: printf( "inf "); break; case DFP_NAN: if (dword1 & DFP_SIGNALING_NAN_BIT) printf("SNaN "); else printf("QNaN "); break; default: printf( "Finite "); exponent = dfp128_exponent(dword1); lmd_digit = special_field_LMD(dword1); for (i = 0; i < 11; i++) { bcd_digits[i] = get_bcd_digit_from_dpd((DFP128_T_START + 10 * i), dword1, dword0); } if (lmd_digit) { silent++; printf("%01lx", lmd_digit); } else { printf(" "); } for (i = 0; i < 11; i++) { if (bcd_digits[i] || silent ) { silent++; printf("%03lx", bcd_digits[i]); } else { /* always print at least the last zero */ if (i == 10) printf(" 0"); else printf(" "); } } printf(" * 10 ^ "); printf("%ld ", exponent); } } /* Interpret the 64-bit values as a 64 bit DFP. * * | Significand | Combination Field/ | | * | sign bit | Encoded Exponent | remainder of significand | * |0 |1 13|14 63| * ^ (bit0) Significand sign bit. * ^ (bit 1:13) Combination field. Contains high bits of * exponent (encoded), LMD of significand (encoded), * and the remainder of the exponent. First five bits * will indicate special cases NAN or INF. * ^ (bit 14:63) Remainder of the * significand. */ /* return the dfp exponent from the leading dword. */ static inline signed long dfp64_exponent(unsigned long dword1) { unsigned long exponent_left; unsigned long exponent_right; unsigned long biased_exponent; signed long exponent; exponent_left = special_field_exponent_bits(dword1); exponent_right = (dword1 & DFP64_EXPONENT_RIGHT_MASK); biased_exponent = (exponent_left << DFP64_EXPONENT_LEFT_SHIFT) + (exponent_right >> DFP64_EXPONENT_RIGHT_MASK_SHIFT); /* Unbias the exponent. */ exponent = biased_exponent - DFP64_EXPONENT_BIAS; return exponent; } static inline void dissect_dfp64_float(uint64_t dword1) { long signbit; signed long exponent; unsigned long gfield_special_symbols; unsigned long lmd_digit; unsigned long bcd_digits[13]; int i; int silent=0; // suppress leading zeros from the output. if (verbose > 5) printf("RAW64: %016lx ", dword1); signbit = (dword1 >> 63); if (signbit) printf("-"); else printf("+"); gfield_special_symbols = ((dword1 >> DFP_SPECIAL_SYMBOLS_SHIFT) & DFP_SPECIAL_SYMBOLS_MASK); switch (gfield_special_symbols) { case DFP_INF: printf( "inf "); break; case DFP_NAN: if (dword1 & DFP_SIGNALING_NAN_BIT) printf("SNaN "); else printf("QNaN "); break; default: printf( "Finite "); exponent = dfp64_exponent(dword1); lmd_digit = special_field_LMD(dword1); for (i = 0; i < 5; i++) bcd_digits[i] = get_bcd_digit_from_dpd((DFP64_T_START + 10 * i), dword1, 0); if (lmd_digit) { silent++; printf("%01lx", lmd_digit); } else { printf(" "); } for (i = 0; i < 5; i++) { if (bcd_digits[i] || silent) { silent++; printf("%03lx", bcd_digits[i]); } else { // suppress leading zeros. /* always print at least the last zero */ if (i == 4) printf(" 0"); else printf(" "); } } printf(" * 10 ^ "); printf("%ld ", exponent); } } static void dump_dfp128_table(void) { int i; printf("DFP 128 table:\n"); for (i = 0; i < nb_dfp128_vals; i += 2) { printf("i=:%2d ", i); dissect_dfp128_float(dfp128_vals[i], dfp128_vals[i+1]); printf("\n"); } } static void dump_dfp64_table(void) { int i; printf("DFP 64 table:\n"); for (i = 0; i 3) printf("%s %lx \n", __FUNCTION__, (unsigned long)thevalue); return thevalue; } /* double (64bit): * | Sign | EXPonent+Bias | FRACTION/Mantissa | * 0 1 11 12 63 * exponent is 11 bits. ranging from: 0x000 .. 0x7ff * 0 = (zero if fraction==0, DeNormal if fraction !=0 ) * 1...0x7fe = normalized * 7ff = (infinite if fraction==0, NaN if fraction !=0) */ #define DOUBLE_EXP_MASK 0x7ff #define DOUBLE_EXP_SHIFT 52 #define DOUBLE_MANTISSA_MASK 0x000fffffffffffff static inline unsigned long build_binary64_float(unsigned long signbit, unsigned long exponent, unsigned long mantissa) { unsigned long thevalue; thevalue = (unsigned long ) (signbit << 63) | ((exponent & DOUBLE_EXP_MASK) << DOUBLE_EXP_SHIFT) | (mantissa & DOUBLE_MANTISSA_MASK ); if (verbose > 3) printf("%s %lx \n", __FUNCTION__, (unsigned long)thevalue); return thevalue; } /* floating point single (32bit): * | Sign | EXPonent+Bias | FRACTION/Mantissa | * 0 1 8 9 31 * exponent is 8 bits. ranging from: 0x00 .. 0xff * 0 = (zero if fraction==0, DeNormal if fraction !=0 ) * 1...0x7e = normalized * 7f = (infinite if fraction==0, NaN if fraction !=0) */ #define SINGLE_EXP_MASK 0xff #define SINGLE_EXP_SHIFT 23 #define SINGLE_MANTISSA_MASK 0x007fffff /* This is building the 32-bit float. */ static inline unsigned long build_binary32_float(unsigned long signbit, unsigned long exponent, unsigned long mantissa) { unsigned long thevalue; unsigned long local_signbit; unsigned long local_exponent; unsigned long local_mantissa; local_signbit = (signbit != 0) << 31; local_exponent = ((exponent & SINGLE_EXP_MASK) << SINGLE_EXP_SHIFT); local_mantissa = (mantissa & SINGLE_MANTISSA_MASK); thevalue = (unsigned long) (local_signbit) | (local_exponent) | (local_mantissa); if (verbose > 3) printf("%s %lx \n", __FUNCTION__, (unsigned long)thevalue); return thevalue; } /* floating point half (16bit): * | Sign | EXPonent+Bias | FRACTION/Mantissa | * 0 1 6 7 15 * exponent is 6 bits. 0x00 .. 0x7e masked with EXP_MASK * 0 = (zero if fraction==0, DeNormal if fraction !=0 ) * 1...0x7d = normalized * 7e = (infinite if fraction==0, NaN if fraction !=0) */ /* when extracting the exponent from the 16-bit half-word, use this mask. */ #define HALF_EXP_MASK 0x7e00 /* when building the 16-bit half-word, mask against this, * then shift into place */ #define HALF_EXP_MASK_NORMALIZED 0x3f #define HALF_EXP_SHIFT 9 #define HALF_MANTISSA_MASK 0x01ff /* This is building the 16-bit float. */ static inline unsigned long build_binary16_float(unsigned long in_signbit, unsigned long exponent, unsigned mantissa) { unsigned long thevalue; unsigned long local_signbit; unsigned long local_exponent; unsigned long local_mantissa; local_signbit = (in_signbit != 0) << 15; local_exponent= ((exponent & HALF_EXP_MASK_NORMALIZED) << HALF_EXP_SHIFT); local_mantissa = (mantissa & HALF_MANTISSA_MASK); thevalue = (unsigned long) (local_signbit) | (local_exponent) | (local_mantissa); if (verbose > 3) printf("%s %lx \n", __FUNCTION__, (unsigned long)thevalue); return thevalue; } /* dissect_binary128_float: * Interpret the (high half) 64-bit value as normal/denormal/inf/NaN. * This is as it would be interpreted as the MSB portion of * a 128-bit wide QUAD. */ static inline void dissect_binary128_float(uint64_t value) { unsigned long signbit; unsigned long exponent; unsigned long mantissa; signbit = (value >> 63); exponent = ( QUAD_EXP_MASK & (value >> QUAD_EXP_SHIFT)); mantissa = ( QUAD_MANTISSA_MASK & value); if (verbose > 4) printf("128 bit:"); if (signbit) printf("-"); else printf("+"); switch (exponent) { case 0x0: if (mantissa == 0) printf("zero "); else printf("denormal "); break; case QUAD_EXP_MASK: if (mantissa == 0) printf("inf "); else printf("NaN "); break; default: printf("Normal "); } if (verbose > 4) printf("%lx %4lx %16lx %16lx \n", signbit, exponent, mantissa, value); } /* Interpret the 64-bit value as normal/denormal/inf/NaN * this is as interpreted as the 64-bit float */ static inline void dissect_binary64_float(uint64_t value) { unsigned long signbit; unsigned long exponent; unsigned long mantissa; signbit = (value >> 63); // bit0 exponent = ( DOUBLE_EXP_MASK & (value >> DOUBLE_EXP_SHIFT)); mantissa = ( DOUBLE_MANTISSA_MASK & value); if (verbose > 4) printf(" 64 bit:"); if (signbit) printf("-"); else printf("+"); switch (exponent) { case 0x0: if (mantissa == 0) printf("zero "); else printf("denormal "); break; case DOUBLE_EXP_MASK: if (mantissa == 0) printf("inf "); else printf("NaN "); break; default: printf("Normal "); } if (verbose>4) printf("%lx %4lx %16lx %16lx\n", signbit, exponent, mantissa, value); } /* interpret the 32-bit value as normal/denormal/inf/NaN. * Note that the value is stored in the upper half of a * 64-bit, which is itself in the upper half of a quad. */ static inline void dissect_binary32_float(uint64_t value) { unsigned long signbit; unsigned long exponent; unsigned long mantissa; unsigned long adj_value; /* shift down to where the offsets make more sense.*/ adj_value = value; //>>32; signbit = (adj_value >> 31); exponent = ( SINGLE_EXP_MASK & (adj_value >> SINGLE_EXP_SHIFT)); mantissa = ( SINGLE_MANTISSA_MASK & adj_value); if (verbose > 4) printf(" 32 bit:"); if (signbit) printf("-"); else printf("+"); switch (exponent) { case 0x0: if (mantissa == 0) printf("zero "); else printf("denormal "); break; case SINGLE_EXP_MASK: if (mantissa == 0) printf("inf "); else printf("NaN "); break; default: printf("Normal "); } if (verbose>4) printf("%lx %4lx %16lx %16lx \n", signbit, exponent, mantissa, adj_value); } /* Interpret the 16-bit value as normal/denormal/inf/NaN. */ static inline void dissect_binary16_float(uint64_t value) { unsigned long signbit; unsigned long exponent; unsigned long mantissa; unsigned long adj_value; adj_value = (value & 0xffff); signbit = ((adj_value & 0x8000) > 1); exponent = ((adj_value & HALF_EXP_MASK ) >> HALF_EXP_SHIFT) ; mantissa = (adj_value & HALF_MANTISSA_MASK); if (verbose > 4) printf(" 16 bit:"); if (signbit) printf("-"); else printf("+"); switch (exponent) { case 0x0: if (mantissa == 0) printf("zero "); else printf("denormal "); break; case HALF_EXP_MASK: if (mantissa == 0) printf("inf "); else printf("NaN "); break; default: printf("Normal "); } if (verbose > 4) printf("%lx %4lx %16lx %16lx \n", signbit, exponent>>HALF_EXP_SHIFT, mantissa, adj_value); } #define dissect_double_as_32s(vec_foo) \ printf(" "); \ dissect_binary16_float((vec_foo & 0xffffffff)); \ printf(" "); \ dissect_binary16_float((vec_foo >> 32) & 0xffffffff); #define dissect_double_as_16s(vec_foo) \ printf(" "); \ dissect_binary16_float((vec_foo&0xffff)); \ printf(" "); \ dissect_binary16_float((vec_foo>>16)&0xffff); \ printf(" "); \ dissect_binary16_float((vec_foo>>32)&0xffff); \ printf(" "); \ dissect_binary16_float((vec_foo>>48)&0xffff); /* a table of exponent values for use in the float precision tests. */ unsigned long exponent_table[] = { #ifdef EXHAUSTIVE_TESTS 0x0000, /* +/-0 or +/-DENormalized, depending on associated mantissa. */ 0x1a, /* within NORmalized for 16,32,64,128-bit. */ 0x1f, /* +/-INF or +/-NaN for 16bit, NORmalized for 32,64,128 */ 0xff, /* +/-INF or +/-NaN for 32bit, NORmalized for 64,128 */ 0x7ff, /* +/-INF or +/-NaN for 32 and 64bit, NORmalized for 128 */ 0x7fff, /* +/-INF or +/-NaN for 128bit. */ #else 0x0000, /* +/-0 or +/-DENormalized, depending on associated mantissa. */ 0xff, /* +/-INF or +/-NaN for 32bit, NORmalized for 64,128 */ 0x7ff, /* +/-INF or +/-NaN for 32 and 64bit, NORmalized for 128 */ 0x7fff, /* +/-INF or +/-NaN for 128bit. */ #endif }; #define MAX_EXPONENTS (sizeof(exponent_table) / sizeof(unsigned long)) unsigned long mantissa_table[] = { #ifdef EXHAUSTIVE_TESTS 0xbeefbeefbeef, /* NOR or DEN or NaN */ 0x000000000000, /* ZERO or INF */ 0x7fffffffffff, /* NOR or DEN or NaN */ #else 0x000000000000, /* ZERO or INF */ 0x7fffffffffff, /* NOR or DEN or NaN */ #endif }; #define MAX_MANTISSAS (sizeof(mantissa_table) / sizeof(unsigned long)) /* build in 64-bit chunks, low doubleword is zero. */ static unsigned long * float_vsxargs; static unsigned long * binary128_float_vsxargs = NULL; static unsigned long * binary64_float_vsxargs = NULL; static unsigned long * binary32_float_vsxargs = NULL; static unsigned long * binary16_float_vsxargs = NULL; unsigned long nb_float_vsxargs; #define MAX_FLOAT_VSX_ARRAY_SIZE (((MAX_EXPONENTS * MAX_MANTISSAS) * 2 + 1) * 2) void build_float_vsx_tables (void) { long i = 0; unsigned long signbit; unsigned long exponent; unsigned long mantissa;/* also referred to as FRACTION in the ISA.*/ unsigned long exponent_index; unsigned long mantissa_index; if (verbose > 2) printf("%s\n", __FUNCTION__); binary128_float_vsxargs = malloc(MAX_FLOAT_VSX_ARRAY_SIZE * sizeof(unsigned long)); float_vsxargs = binary128_float_vsxargs; binary64_float_vsxargs = malloc(MAX_FLOAT_VSX_ARRAY_SIZE * sizeof(unsigned long)); binary32_float_vsxargs = malloc(MAX_FLOAT_VSX_ARRAY_SIZE * sizeof(unsigned long)); binary16_float_vsxargs = malloc(MAX_FLOAT_VSX_ARRAY_SIZE * sizeof(unsigned long)); for (signbit = 0; signbit < 2; signbit++) { for (exponent_index = 0; exponent_index < MAX_EXPONENTS; exponent_index++) { for (mantissa_index = 0; mantissa_index < MAX_MANTISSAS; mantissa_index++) { exponent = exponent_table[exponent_index]; mantissa = mantissa_table[mantissa_index]; if (verbose > 2) { printf("signbit:%lx ", signbit); printf("exponent:%4lx ", exponent); printf("mantissa:%lx ", mantissa); printf("\n"); } binary128_float_vsxargs[i] = build_binary128_float(signbit, exponent, mantissa); binary128_float_vsxargs[i+1] = 0; binary64_float_vsxargs[i] = build_binary64_float(signbit, exponent, mantissa); binary64_float_vsxargs[i+1] = build_binary64_float(signbit, exponent, mantissa); binary32_float_vsxargs[i] = build_binary32_float(signbit, exponent, mantissa); binary32_float_vsxargs[i+1] = build_binary32_float(signbit, exponent, mantissa); binary16_float_vsxargs[i] = build_binary16_float(signbit, exponent, mantissa); binary16_float_vsxargs[i+1] = build_binary16_float(signbit, exponent, mantissa); i += 2; } } } nb_float_vsxargs = i; } /* Display entries stored in the float_vsx table. These are used as * quad/double/singles, stored as quads. */ void dump_float_vsx_table (void) { int i; printf("Float VSX Table:"); printf("128-bit (quad):\n"); for (i = 0; i < nb_float_vsxargs; i += 2) { printf("i =: %2d ", i); dissect_binary128_float(binary128_float_vsxargs[i]); } printf("64-bit (double):\n"); for (i = 0; i< nb_float_vsxargs; i += 2) { printf("i = %2d ", i); dissect_binary64_float(binary64_float_vsxargs[i]); } printf("32-bit (single):\n"); for (i = 0; i < nb_float_vsxargs; i += 2) { printf("i = %2d ", i); dissect_binary32_float(binary32_float_vsxargs[i]); } printf("16-bit (half):\n"); for (i = 0; i < nb_float_vsxargs; i += 2) { printf("i =% 2d ", i); dissect_binary16_float(binary16_float_vsxargs[i]); } printf("\n"); } static void print_dcmx_field(unsigned long local_dcmx) { /* Note - this splats out the local_dxmc field from the form used to * globally pass it, with a single set bit, into the functions that use * it. The actual DCMX field is a bit-field from 0x00 to 0x3f. If * multiple bits are ever set, this function and the way it is passed * into the users will need to be updated. This does not handle * multiple bits being set. */ printf(" DCMX=["); switch(local_dcmx) { case 0: printf("ALL"); break; case 1: printf("NaN"); break; case 2: printf("+inf"); break; case 3: printf("-inf"); break; case 4: printf("+zero"); break; case 5: printf("-zero"); break; case 6: printf("+denormal"); break; case 7: printf("-denormal"); break; default: printf("other"); break; } if (verbose > 3) printf(" %lx", local_dcmx); printf("] "); } #define MAX_CHAR_ARGS_ARRAY_SIZE 128 static unsigned char * char_args; unsigned long nb_char_args; static void build_char_table(void) { long i = 0; char ichar; char_args = memalign(32, MAX_CHAR_ARGS_ARRAY_SIZE * sizeof(char)); #ifdef EXHAUSTIVE_TESTS for (ichar = 'a'; ichar <= 'z'; ichar++) { char_args[i++] = ichar; } for (ichar = '0'; ichar <= '9'; ichar++) { char_args[i++] = ichar; } for (ichar = 'A'; ichar <= 'Z'; ichar++) { char_args[i++] = ichar; } #else for (ichar = 'a'; ichar <= 'z'; ichar+=6) { char_args[i++] = ichar; } for (ichar = '0'; ichar <= '9'; ichar+=6) { char_args[i++] = ichar; } for (ichar = 'A'; ichar <= 'Z'; ichar+=6) { char_args[i++] = ichar; } #endif char_args[i++] = ' '; char_args[i++] = '+'; char_args[i++] = '-'; char_args[i++] = '/'; char_args[i++] = '['; char_args[i++] = ']'; char_args[i++] = '`'; char_args[i++] = '_'; nb_char_args = i; } static void dump_char_table() { int i; printf("Char Table:"); for (i = 0; i= MAX_VPCV_SIZE) printf("Warning! Exceeded size of table building the vector permute control . \n"); } /* Decimal Encodings... * Packed, National, Zoned decimal content follows. * Note: Watch the conversions in and out of the * dwords / vectors for reverses with respect to * top/bottom low/high */ /* Packed Decimals: * A valid encoding of a packed decimal integer value requires the following * properties: * – Each of the 31 4-bit digits of the operand’s magnitude (bits 0:123) * must be in the range 0-9. * – The sign code (bits 124:127) must be in the range 10-15. (0xa-0xf). * Source operands with sign codes of 0b1010, 0b1100, 0b1110, and 0b1111 are * interpreted as positive values. Source operands with sign codes of * 0b1011 and 0b1101 are interpreted as negative values. * Positive and zero results are encoded with a either sign code of * 0b1100 or 0b1111, depending on the preferred sign (indicated as an * immediate operand). Negative results are encoded with a sign code * of 0b1101. * PS - This is the 'preferred sign' bit encoded in some BCD associated * instructions. */ // Note: table content is limited to values encoded, not interpreted. unsigned int packed_decimal_sign_codes[] = { /* positive operands */ 0xc, 0xf, // 0b1100, 0b1111 /* negative operands */ 0xd // 0b1101 }; #define NR_PACKED_DECIMAL_SIGNS 3 #define MAX_PACKED_DECIMAL_TABLE_SIZE 8 * 16 * 2 + 20 static unsigned long * packed_decimal_table; /* build into a pair of doubles */ unsigned long nb_packed_decimal_entries; static void dissect_packed_decimal_sign(unsigned long local_sign) { switch(local_sign) { case 0xa: /*0b1010:*/ printf("[ + ]"); break; case 0xb: /*0b1011:*/ printf("[ - ]"); break; case 0xc: /*0b1100:*/ printf("(+|0)"); break; case 0xd: /*0b1101:*/ printf("( - )"); break; case 0xe: /*0b1110:*/ printf("[ + ]"); break; case 0xf: /*0b1111:*/ printf("(+|0)"); break; default: printf("(?%02lx)", local_sign); } } int extract_packed_decimal_sign(unsigned long dword1, unsigned long dword0) { return (dword1 & 0xf); } static void dissect_packed_decimal(unsigned long dword1,unsigned long dword0) { int i; int local_sign; int nibble; local_sign = extract_packed_decimal_sign(dword1, dword0); printf("packed_decimal: ["); for (i = 60; i >= 0; i -= 4) { nibble=(dword1 >> (i)) & 0xf; printf(" %x", nibble); } for (i = 60; i >= 0; i -= 4) { nibble=(dword0 >> (i)) & 0xf; printf(" %x", nibble); } printf(" "); dissect_packed_decimal_sign(local_sign); printf(" ] "); } static void build_packed_decimal_table(void) { long sign_index; long sign_value; unsigned long i = 0; unsigned long value; #ifdef EXHAUSTIVE_TESTS int scramble; #endif if (verbose) printf("%s\n", __FUNCTION__); packed_decimal_table = malloc((MAX_PACKED_DECIMAL_TABLE_SIZE + 2) * sizeof (unsigned long)); for (sign_index = 0; sign_index < NR_PACKED_DECIMAL_SIGNS; sign_index++) { sign_value = packed_decimal_sign_codes[sign_index]; for (value = 0; value <= 9; value++) { packed_decimal_table[i] = 0x1111111111111111 * value; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x1111111111111110 * value; if (verbose>3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose>3) printf("\n"); i+=2; } #ifdef EXHAUSTIVE_TESTS for (scramble = 1; scramble <= 4; scramble++) { packed_decimal_table[i] = 0x3210321032103210 * scramble; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x0123012301230120 * scramble; if (verbose>3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose>3) printf("\n"); i+=2; } #endif /* Add some entries that will provide interesting output from * the convert TO tests. */ packed_decimal_table[i] = 0x0000000000000000; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x0000000012345670; if (verbose > 3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose>3) printf("\n"); i += 2; #ifdef EXHAUSTIVE_TESTS packed_decimal_table[i] = 0x0000000000000000; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x0000000098765430; if (verbose > 3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose > 3) printf("\n"); i += 2; packed_decimal_table[i] = 0x000000000000000b; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x0000000000000000; if (verbose > 3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose>3) printf("\n"); i += 2; #endif packed_decimal_table[i] = 0x0030000000000000; packed_decimal_table[i+1] = sign_value; packed_decimal_table[i+1] += 0x0000000000000000; if (verbose > 3) dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); if (verbose > 3) printf("\n"); i += 2; } if (verbose>2) printf("\n"); nb_packed_decimal_entries = i; } static void dump_packed_decimal_table(void) { int i; printf("packed_decimal_table:\n"); for (i = 0; i < nb_packed_decimal_entries; i += 2) { printf("i =: %2d ", i); dissect_packed_decimal(packed_decimal_table[i+1], packed_decimal_table[i]); printf("\n"); } } /* National decimals: * A valid encoding of a national decimal value requires the following. * – The contents of halfword 7 (sign code) must be * either 0x002B or 0x002D. * – The contents of halfwords 0 to 6 must be in the * range 0x0030 to 0x0039. * National decimal values having a sign code of 0x002B * are interpreted as positive values. * National decimal values having a sign code of 0x002D * are interpreted as negative values. */ unsigned int national_decimal_sign_codes[] = { /* positive */ 0x002b, /* negative */ 0x002d }; #define NR_NATIONAL_DECIMAL_SIGNS 2 unsigned int national_decimal_values[] = { #ifdef EXHAUSTIVE_TESTS 0x0030, 0x0031, 0x0032, 0x0033, 0x0034, 0x0035, 0x0036, 0x0037, 0x0038, 0x0039 #else 0x0030, 0x0031, 0x0035, 0x0039 #endif }; #define NR_NATIONAL_DECIMAL_VALUES (sizeof(national_decimal_values) / sizeof(unsigned int)) static unsigned long * national_decimal_table; #define MAX_NATIONAL_DECIMAL_TABLE_SIZE 10 * NR_NATIONAL_DECIMAL_VALUES * NR_NATIONAL_DECIMAL_SIGNS unsigned long nb_national_decimal_entries; static void dissect_national_decimal_sign(unsigned long local_sign) { switch(local_sign) { case 0x002b: printf("( + )"); break; case 0x002d: printf("( - )"); break; default: printf("unhandled sign value: %lx", local_sign); } } int extract_national_decimal_sign(unsigned long dword1, unsigned long dword0) { return (dword1 & 0x0ff); } static void dissect_national_decimal(unsigned long dword1, unsigned long dword0) { int i; int local_sign; long hword; printf("national_decimal: ["); if (verbose>4) printf("raw: [%016lx %016lx] ", dword1, dword0); for (i = 48;i >= 0; i -= 16) { hword = dword1 >> (i) & 0x00ff; /* validity of national decimal value */ /* the i>0 clause skips the validity check against the sign value. */ if (((i > 0) && (hword < 0x30)) || (hword > 0x39)) printf("!"); printf("%04lx ", hword); } for (i = 48; i >= 0; i -= 16) { hword = dword0 >> (i) & 0x00ff; if ((hword < 0x30) || (hword > 0x39)) printf("!"); printf("%04lx ", hword); } local_sign = extract_national_decimal_sign(dword1, dword0); dissect_national_decimal_sign(local_sign); printf(" ] "); } static void build_national_decimal_table(void) { long sign_index; long sign_value; unsigned long i = 0; int index; unsigned long value; if (verbose) printf("%s\n",__FUNCTION__); national_decimal_table = malloc(MAX_NATIONAL_DECIMAL_TABLE_SIZE * sizeof (unsigned long)); for (sign_index = 0; sign_index < NR_NATIONAL_DECIMAL_SIGNS; sign_index++) { sign_value = national_decimal_sign_codes[sign_index]; for (index = 0; index < NR_NATIONAL_DECIMAL_VALUES; index++) { value = national_decimal_values[index]; national_decimal_table[i] = 0x0001000100010001 * value; national_decimal_table[i+1] = 0x0001000100010000 * value; national_decimal_table[i+1] += sign_value ; if (verbose > 3) { dissect_national_decimal(national_decimal_table[i+1], national_decimal_table[i]); printf("\n"); } i += 2; } #ifdef EXHAUSTIVE_TESTS { /* a few more for fun */ national_decimal_table[i] = 0x0031003200330034; national_decimal_table[i+1] = 0x0035003600370000; national_decimal_table[i+1] += sign_value ; if (verbose > 3) { dissect_national_decimal(national_decimal_table[i+1], national_decimal_table[i]); printf("\n"); } i += 2; national_decimal_table[i] = 0x0031003200330034; national_decimal_table[i+1] = 0x0035003600370000; national_decimal_table[i+1] += sign_value ; if (verbose > 3) { dissect_national_decimal(national_decimal_table[i+1], national_decimal_table[i]); printf("\n"); } i += 2; } #endif } if (verbose > 2) printf("\n"); nb_national_decimal_entries = i; } static void dump_national_decimal_table(void) { int i; printf("national_decimal_table:\n"); for (i = 0; i < nb_national_decimal_entries; i += 2) { printf("#%2d ", i); dissect_national_decimal(national_decimal_table[i+1], national_decimal_table[i]); printf("\n"); } } /* Zoned Decimals: * * When PS=0, do the following. * A valid encoding of a zoned decimal value requires the following. * – The contents of bits 0:3 of byte 15 (sign code) can be any * value in the range 0x0 to 0xF. * – The contents of bits 0:3 of bytes 0 to 14 (zone) must * be the value 0x3. * – The contents of bits 4:7 of bytes 0 to 15 must * be a value in the range 0x0 to 0x9. * Zoned decimal values having a sign code of 0x0, 0x1, 0x2, 0x3, * 0x8, 0x9, 0xA, or 0xB are interpreted as positive values. * Zoned decimal values having a sign code of 0x4, 0x5, 0x6, 0x7, * 0xC, 0xD, 0xE, or 0xF are interpreted as negative values. :: 0,1,2,3, 8,9,a,b, are interpreted as positive. :: 4,5,6,7, c,d,e,f are interpreted as negative. * When PS=1, do the following. * A valid encoding of a zoned decimal source operand requires the following. * – The contents of bits 0:3 of byte 15 (sign code) must be a value in the * range 0xA to 0xF. * – The contents of bits 0:3 of bytes 0 to 14 (zone) must be the value 0xF. * – The contents of bits 4:7 of bytes 0 to 15 must be a value in the * range 0x0 to 0x9. * Zoned decimal source operands having a sign code of 0xA, 0xC, 0xE, * or 0xF are interpreted as positive values. * Zoned decimal source operands having a sign code of 0xB or 0xD are * interpreted as negative values. :: a, c, e,f are interpreted as positive. :: b, d, are interpreted as negative. */ /* a valid sign is anything in range 0-9,a-f, * For coverage that does not overwhelm, we have chosen to use 0,1,4,a,b,f. */ #define NM_ZONED_DECIMAL_SIGNS 6 #define NM_ZONED_VALUES 5 /* 0,2,4,6,9 */ #define NM_PS_VALUES 2 /* 0,1 */ #define NM_ZONED_ADDITIONAL_PATTERNS 4 #define MAX_ZONED_DECIMAL_TABLE_SIZE NM_ZONED_DECIMAL_SIGNS * NM_ZONED_VALUES * NM_ZONED_ADDITIONAL_PATTERNS * NM_PS_VALUES + 10 static unsigned long zoned_decimal_table_[MAX_ZONED_DECIMAL_TABLE_SIZE]; static unsigned long * zoned_decimal_table; unsigned long nb_zoned_decimal_entries; static void dissect_zoned_decimal_sign(unsigned long local_sign, int ps) { if (ps == 0) { switch(local_sign) { case 0x0: case 0x1: case 0x2: case 0x3: case 0x8: case 0x9: case 0xa: case 0xb: printf("( + )"); break; case 0x4: case 0x5: case 0x6: case 0x7: case 0xc: case 0xd: case 0xe: case 0xf: printf("( - )"); break; default: printf("zoned decimal (ps=%d). Unhandled sign value: %lx", ps, local_sign); } } if (ps == 1) { switch(local_sign) { case 0xa: case 0xc: case 0xe: case 0xf: printf("( + )"); break; case 0xb: case 0xd: printf("( - )"); break; default: printf("zoned decimal (ps=%d). Unhandled sign value: %lx", ps, local_sign); } } } /* Valid byte values within a zoned decimal are in the ranges of * 0x30..0x39 when PS==0, or 0xf0..0xff when PS==1. */ static void check_zoned_byte_validity(int byte, int ps) { if (ps == 0) { /* check the zone */ if (((byte & 0x30) != 0x30)) printf("!=30"); } else { /* ps==1 */ if (((byte & 0xf0) != 0xf0)) printf("%x !=f0 ", byte ); } /* check the numeric value */ if ((byte & 0x0f) > 0x9) printf("!(0..9)"); } int extract_zoned_decimal_sign(unsigned long dword1, unsigned long dword0) { return ((dword1 & 0xf0) >> 4); } static void dissect_zoned_decimal(unsigned long dword1, unsigned long dword0, int ps) { int i; int local_sign; int byte; printf("zoned_decimal: ["); for (i = 56; i >= 0; i -= 8) { byte = (dword1 >> (i)) & 0xff; check_zoned_byte_validity(byte, ps); printf(" %02x", byte); } for (i = 56; i >= 0; i -= 8) { byte = (dword0 >> (i)) & 0x00ff; check_zoned_byte_validity(byte, ps); if ((byte & 0xf) > 0x9) printf(" !(>9)"); printf(" %02x", byte); } local_sign = extract_zoned_decimal_sign(dword1, dword0); dissect_zoned_decimal_sign(local_sign, ps); printf(" ]"); } #ifdef EXHAUSTIVE_TESTS // Randomly chosen exhaustive coverage for k includes values: 0,2,4,7,9 # define SELECTIVE_INCREMENT_ZONED(k) \ if (k == 7) k = 9; \ else if (k == 4) k = 7; \ else if (k == 2) k = 4; \ else if (k == 0) k = 2; \ else k++; // Randomly chosen exhaustive coverage for signs includes values: 0,1,4,a,b,f # define SELECTIVE_INCREMENT_SIGNS(signs) \ if (signs == 0x0) signs = 0x1; \ else if (signs == 0x1) signs = 0x4; \ else if (signs == 0x4) signs = 0xa; \ else if (signs == 0xa) signs = 0xb; \ else if (signs == 0xb) signs = 0xf; \ else signs++; #else // Randomly chosen coverage for k includes values: 0,7,9 # define SELECTIVE_INCREMENT_ZONED(k) \ if (k == 7) k = 9; \ else if (k == 0) k = 7; \ else k++; // Randomly chosen coverage for signs includes values: 0,4,b,f # define SELECTIVE_INCREMENT_SIGNS(signs) \ if (signs == 0x0) signs = 0x4; \ else if (signs == 0x4) signs = 0xb; \ else if (signs == 0xb) signs = 0xf; \ else signs++; #endif static void build_zoned_decimal_table(void) { unsigned long signs; unsigned long i; int k; int ps; int signs_start,signs_end; if (verbose) printf("%s\n", __FUNCTION__); zoned_decimal_table = zoned_decimal_table_; i = 0; for (ps = 0; ps <= 1; ps++) { if (ps == 0) { signs_start = 0; signs_end = 0xf; } else { signs_start = 0xa; signs_end = 0xf; } for (signs = signs_start; signs <= signs_end; /* signs selectively updated below */) { if (verbose > 2) printf("ps=%d sign:%lx\n", ps, signs); for (k = 0 ; k < 9; /* k selectively updated below */) { if (ps == 0) { zoned_decimal_table[i] = 0x3030303030303030; // set bits 0:3 of bytes 0..7. zoned_decimal_table[i+1] = 0x3030303030303000; // bits 0:3 of bytes 8..14 must be 0x3 } else { zoned_decimal_table[i] = 0xf0f0f0f0f0f0f0f0; // set bits 0:3 of bytes 0..7. zoned_decimal_table[i+1] = 0xf0f0f0f0f0f0f000; // bits 0:3 of bytes 8..14 must be 0x3 } zoned_decimal_table[i] += 0x010101010101010 * k; // set bits 4..7 of bytes 0..7. zoned_decimal_table[i+1] += 0x010101010101000 * k; // bits 4:7 of bytes 8..15 must be 0..9. zoned_decimal_table[i+1] += (signs << 4); // bits 0:3 of byte 15 is the sign. if (verbose > 3) { dissect_zoned_decimal(zoned_decimal_table[i+1], zoned_decimal_table[i], ps); printf("\n"); } i += 2; SELECTIVE_INCREMENT_ZONED(k) } /* add a few more patterns outside of the k patterns. */ if (ps == 0) { zoned_decimal_table[i] = 0x3030303030303030; zoned_decimal_table[i+1] = 0x3030303030303000; } else { zoned_decimal_table[i] = 0xf0f0f0f0f0f0f0f0; zoned_decimal_table[i+1] = 0xf0f0f0f0f0f0f000; } zoned_decimal_table[i] += 0x0908070605040302; zoned_decimal_table[i+1] += 0x0102030405060700; zoned_decimal_table[i+1] += (signs<<4); // bits 0:3 of byte 15. if (verbose > 3) { dissect_zoned_decimal(zoned_decimal_table[i+1], zoned_decimal_table[i], ps); printf("\n"); } i += 2; SELECTIVE_INCREMENT_SIGNS(signs) } /* signs loop */ } /* ps loop */ nb_zoned_decimal_entries = i; } static void dump_zoned_decimal_table(void) { int i; int ps; for (ps = 0; ps <= 1; ps++) { printf("zoned_decimal_table ps=%d:\n", ps); for (i = 0; i < nb_zoned_decimal_entries; i += 2) { printf("#%2d ", i); dissect_zoned_decimal(zoned_decimal_table[i+1], zoned_decimal_table[i], ps); printf("\n"); } } } /* Build table containing shift and truncate values */ #define MAX_DECIMAL_SHIFT_TABLE_SIZE 64 static unsigned long * decimal_shift_table; unsigned long nb_decimal_shift_entries; static void build_decimal_shift_table(void) { unsigned long i = 0; unsigned long value; if (verbose) printf("%s\n",__FUNCTION__); decimal_shift_table = malloc(MAX_DECIMAL_SHIFT_TABLE_SIZE * sizeof (unsigned long)); for (value = 0; value <= 31; value++) { decimal_shift_table[i] = value; decimal_shift_table[i+1] = 0; i += 2; } if (verbose>2) printf("\n"); nb_decimal_shift_entries = i; } static void dump_decimal_shift_table(void) { int i; printf("decimal_shift_table:\n"); for (i = 0; i < nb_decimal_shift_entries; i += 2) { printf("i=:%2d ", i); printf(" 0x%2lx 0x%2lx ", decimal_shift_table[i], decimal_shift_table[i+1]); printf("\n"); } }