diff options
Diffstat (limited to 'sysdeps/powerpc/powerpc64')
| -rw-r--r-- | sysdeps/powerpc/powerpc64/dl-trampoline.S | 2 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc64/fpu/s_llrintl.S | 94 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc64/fpu/s_llroundl.S | 167 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc64/fpu/s_lrintl.S | 2 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc64/fpu/s_lroundl.S | 2 | ||||
| -rw-r--r-- | sysdeps/powerpc/powerpc64/memcpy.S | 3 |
6 files changed, 4 insertions, 266 deletions
diff --git a/sysdeps/powerpc/powerpc64/dl-trampoline.S b/sysdeps/powerpc/powerpc64/dl-trampoline.S index c5afe5a3fa..9ca394dda2 100644 --- a/sysdeps/powerpc/powerpc64/dl-trampoline.S +++ b/sysdeps/powerpc/powerpc64/dl-trampoline.S @@ -176,7 +176,9 @@ EALIGN(_dl_profile_resolve, 4, 0) /* Spill r30, r31 to preserve the link_map* and reloc_addr, in case we need to call _dl_call_pltexit. */ std r31,-8(r1) + cfi_offset(r31,-8) std r30,-16(r1) + cfi_offset(r30,-16) /* We need to save the registers used to pass parameters, ie. r3 thru r10; the registers are saved in a stack frame. */ stdu r1,-FRAME_SIZE(r1) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_llrintl.S b/sysdeps/powerpc/powerpc64/fpu/s_llrintl.S deleted file mode 100644 index aa487775d4..0000000000 --- a/sysdeps/powerpc/powerpc64/fpu/s_llrintl.S +++ /dev/null @@ -1,94 +0,0 @@ -/* Round long double to long int. - IBM extended format long double version. - Copyright (C) 2004,2006 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, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#include <sysdep.h> -#include <math_ldbl_opt.h> - - .section ".toc","aw" -.LC0: /* 2**52 */ - .tc FD_43300000_0[TC],0x4330000000000000 -.LC1: /* 2**63 */ - .tc FD_43E00000_0[TC],0x43e0000000000000 - .section ".text" - -/* long long int[r3] __llrintl (long double x[fp1,fp2]) */ -ENTRY (__llrintl) - lfd fp13,.LC0@toc(2) - lfd fp10,.LC1@toc(2) - fabs fp0,fp1 - fcmpu cr7,fp0,fp13 /* if (fabs(x) > TWO52) */ - fcmpu cr6,fp0,fp10 /* if (fabs(x) > TWO63) */ - beq- cr6,.L2 - fctid fp11,fp1 /* must delay this opperation to here */ - fctid fp12,fp2 /* and avoid setting "invalid operation". */ - li r0,0 - stfd fp11,-16(r1) - bgt- cr6,.L9 /* if > TWO63 return "invalid operation". */ - ble+ cr7,.L9 /* If < TWO52 only thy high double is used. */ - stfd fp12,-8(r1) - nop /* Insure the following load is in a different dispatch group */ - nop /* to avoid pipe stall on POWER4&5. */ - nop -.L8: - ld r0,-8(r1) -.L9: - ld r3,-16(r1) - add r3,r3,r0 - blr - -/* The high double is >= TWO63 so it looks like we are "out of range". - But this may be caused by rounding of the high double and the - negative low double may bring it back into range. So we need to - de-round the high double and invert the low double without changing - the effective long double value. To do this we compute a special - value (tau) that we can subtract from the high double and add to - the low double before conversion. The resulting integers can be - summed to get the total value. - - tau = floor(x_high/TWO52); - x0 = x_high - tau; - x1 = x_low + tau; */ -.L2: - fdiv fp8,fp1,fp13 /* x_high/TWO52 */ - fctidz fp0,fp8 - fcfid fp8,fp0 /* tau = floor(x_high/TWO52); */ - fsub fp3,fp1,fp8 /* x0 = x_high - tau; */ - fadd fp4,fp2,fp8 /* x1 = x_low + tau; */ - fctid fp11,fp3 - fctid fp12,fp4 - stfd fp11,-16(r1) - stfd fp12,-8(r1) - nop /* Insure the following load is in a different dispatch group */ - nop /* to avoid pipe stall on POWER4&5. */ - nop - ld r3,-16(r1) - ld r0,-8(r1) - addo. r3,r3,r0 - bnslr+ cr0 /* if the sum does not overflow, return. */ - fctid fp11,fp1 /* Otherwise we want to set "invalid operation". */ - li r0,0 - stfd fp11,-16(r1) - b .L9 - -END (__llrintl) - -strong_alias (__llrintl, __lrintl) -long_double_symbol (libm, __llrintl, llrintl) -long_double_symbol (libm, __lrintl, lrintl) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_llroundl.S b/sysdeps/powerpc/powerpc64/fpu/s_llroundl.S deleted file mode 100644 index 29eca11093..0000000000 --- a/sysdeps/powerpc/powerpc64/fpu/s_llroundl.S +++ /dev/null @@ -1,167 +0,0 @@ -/* llroundl function. - IBM extended format long double version. - Copyright (C) 2004, 2006 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, write to the Free - Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA - 02111-1307 USA. */ - -#include <sysdep.h> -#include <math_ldbl_opt.h> - - .section ".toc","aw" -.LC0: /* 0.0 */ - .tc FD_00000000_0[TC],0x0000000000000000 -.LC1: /* 0.5 */ - .tc FD_3fe00000_0[TC],0x3fe0000000000000 -.LC2: /* 2**52 */ - .tc FD_43300000_0[TC],0x4330000000000000 -.LC3: /* 2**63 */ - .tc FD_43E00000_0[TC],0x43e0000000000000 - .section ".text" - -/* long long [r3] llround (long double x [fp1,fp2]) - IEEE 1003.1 llroundl function. IEEE specifies "round to the nearest - integer value, rounding halfway cases away from zero, regardless of - the current rounding mode." However PowerPC Architecture defines - "round to Nearest" as "Choose the best approximation. In case of a - tie, choose the one that is even (least significant bit o).". - So we can't use the PowerPC "round to Nearest" mode. Instead we set - "round toward Zero" mode and round by adding +-0.5 before rounding - toward zero. The "Floating Convert To Integer Doubleword with round - toward zero" instruction handles the conversion including the - overflow cases and signalling "Invalid Operation". - - PowerPC64 long double uses the IBM extended format which is - represented two 64-floating point double values. The values are - non-overlapping giving an effective precision of 106 bits. The first - double contains the high order bits of mantisa and is always rounded - to represent a normal rounding of long double to double. Since the - long double value is sum of the high and low values, the low double - normally has the opposite sign to compensate for the this rounding. - - For long double there is 4 cases: - 1) |x| < 2**52, all the integer bits are in the high double. - Round and convert the high double to long long. - 2) 2**52 <= |x|< 2**63, Still fits but need bits from both doubles. - Round the low double, convert both, then sum the long long values. - 3) |x| == 2**63, Looks like an overflow but may not be due to rounding - of the high double. - See the description following lable L2. - 4) |x| > 2**63, This will overflow the 64-bit signed integer. - Treat like case #1. The fctidz instruction will generate the - appropriate and signal "invalid operation". - - */ - -ENTRY (__llroundl) - mffs fp7 /* Save current FPU rounding mode. */ - fabs fp0,fp1 - lfd fp13,.LC2@toc(2) /* 2**52 */ - lfd fp12,.LC3@toc(2) /* 2**63 */ - lfd fp11,.LC0@toc(2) /* 0.0 */ - lfd fp10,.LC1@toc(2) /* 0.5 */ - fabs fp9,fp2 - fcmpu cr7,fp0,fp13 /* if (fabs(x) > TWO52) */ - fcmpu cr6,fp1,fp11 /* if (x > 0.0) */ - bnl- cr7,.L2 - mtfsfi 7,1 /* Set rounding mode toward 0. */ - ble- cr6,.L1 - fadd fp9,fp1,fp10 /* x+= 0.5; */ - b .L0 -.L1: - fsub fp9,fp1,fp10 /* x-= 0.5; */ -.L0: - fctid fp0,fp9 - stfd fp0,-16(r1) - mtfsf 0x01,fp7 /* restore previous rounding mode. */ - nop /* Insure the following load is in a different dispatch group */ - nop /* to avoid pipe stall on POWER4&5. */ - nop - ld r3,-16(r1) - blr - -/* The high double is > TWO52 so we need to round the low double and - perhaps the high double. In this case we have to round the low - double and handle any adjustment to the high double that may be - caused by rounding (up). This is complicated by the fact that the - high double may already be rounded and the low double may have the - opposite sign to compensate.This gets a bit tricky so we use the - following algorithm: - - tau = trunc(x_high/TWO52); - x0 = x_high - tau; - x1 = x_low + tau; - r1 = round(x1); - y_high = x0 + r1; - y_low = x0 - y_high + r1; - return y; */ -.L2: - fcmpu cr7,fp0,fp12 /* if (|x_high| > TWO63) */ - fcmpu cr0,fp9,fp11 /* || (|x_low| == 0.0) */ - fmr fp9,fp1 - fcmpu cr5,fp2,fp11 /* if (x_low > 0.0) */ - bgt- cr7,.L0 /* return llround(x); */ - mtfsfi 7,1 /* Set rounding mode toward 0. */ - fdiv fp8,fp1,fp13 /* x_high/TWO52 */ - - bng- cr6,.L6 /* if (x > 0.0) */ - fctidz fp0,fp8 - fcfid fp8,fp0 /* tau = trunc(x_high/TWO52); */ - bng cr5,.L4 /* if (x_low > 0.0) */ - fmr fp3,fp1 - fmr fp4,fp2 - b .L5 -.L4: /* if (x_low < 0.0) */ - fsub fp3,fp1,fp8 /* x0 = x_high - tau; */ - fadd fp4,fp2,fp8 /* x1 = x_low + tau; */ -.L5: - fadd fp5,fp4,fp10 /* r1 = x1 + 0.5; */ - b .L9 -.L6: /* if (x < 0.0) */ - fctidz fp0,fp8 - fcfid fp8,fp0 /* tau = trunc(x_high/TWO52); */ - bnl cr5,.L7 /* if (x_low < 0.0) */ - fmr fp3,fp1 - fmr fp4,fp2 - b .L8 -.L7: /* if (x_low > 0.0) */ - fsub fp3,fp1,fp8 /* x0 = x_high - tau; */ - fadd fp4,fp2,fp8 /* x1 = x_low + tau; */ -.L8: - fsub fp5,fp4,fp10 /* r1 = x1 - 0.5; */ -.L9: - fctid. fp11,fp3 - fctid fp12,fp5 - stfd fp11,-16(r1) - stfd fp12,-8(r1) - mtfsf 0x01,fp7 /* restore previous rounding mode. */ - nop /* Insure the following load is in a different dispatch group */ - nop /* to avoid pipe stall on POWER4&5. */ - nop - ld r3,-16(r1) - bunlr cr1 /* if not overflow, return. */ - ld r0,-8(r1) - addo. r3,r3,r0 - bnslr cr0 - fmr fp9,fp12 - bng cr6,.L0 - fneg fp9,fp12 - b .L0 -END (__llroundl) - -strong_alias (__llroundl, __lroundl) -long_double_symbol (libm, __llroundl, llroundl) -long_double_symbol (libm, __lroundl, lroundl) diff --git a/sysdeps/powerpc/powerpc64/fpu/s_lrintl.S b/sysdeps/powerpc/powerpc64/fpu/s_lrintl.S deleted file mode 100644 index 6c82d2e222..0000000000 --- a/sysdeps/powerpc/powerpc64/fpu/s_lrintl.S +++ /dev/null @@ -1,2 +0,0 @@ -/* __lrintl is in s_llrintl.c */ -/* __lrintl is in s_llrintl.c */ diff --git a/sysdeps/powerpc/powerpc64/fpu/s_lroundl.S b/sysdeps/powerpc/powerpc64/fpu/s_lroundl.S deleted file mode 100644 index b24dfd8ded..0000000000 --- a/sysdeps/powerpc/powerpc64/fpu/s_lroundl.S +++ /dev/null @@ -1,2 +0,0 @@ -/* __lroundl is in s_llroundl.S */ -/* __lroundl is in s_llroundl.S */ diff --git a/sysdeps/powerpc/powerpc64/memcpy.S b/sysdeps/powerpc/powerpc64/memcpy.S index 9df5bb42b6..f395de9066 100644 --- a/sysdeps/powerpc/powerpc64/memcpy.S +++ b/sysdeps/powerpc/powerpc64/memcpy.S @@ -1,5 +1,5 @@ /* Optimized memcpy implementation for PowerPC64. - Copyright (C) 2003 Free Software Foundation, Inc. + Copyright (C) 2003, 2006 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 @@ -43,6 +43,7 @@ EALIGN (BP_SYM (memcpy), 5, 0) neg 0,3 std 3,-16(1) std 31,-8(1) + cfi_offset(31,-8) andi. 11,3,7 /* check alignement of dst. */ clrldi 0,0,61 /* Number of bytes until the 1st doubleword of dst. */ clrldi 10,4,61 /* check alignement of src. */ |