From 01e80428f7df1a91de2f16ee632e81bc41a17d40 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Thu, 13 Sep 2012 12:54:03 -0700 Subject: [PATCH] alpha: Streamline __setfpucw The convert_bit macro allows the compiler to translate the bit positions more efficiently. The assumption of only running at program startup allows eliding the __ieee_get_fp_control call. --- ports/ChangeLog.alpha | 3 + ports/sysdeps/unix/sysv/linux/alpha/setfpucw.c | 88 ++++++++++---------------- 2 files changed, 38 insertions(+), 53 deletions(-) diff --git a/ports/ChangeLog.alpha b/ports/ChangeLog.alpha index d00b1e9..ab807f7 100644 --- a/ports/ChangeLog.alpha +++ b/ports/ChangeLog.alpha @@ -1,5 +1,8 @@ 2012-12-13 Richard Henderson + * sysdeps/unix/sysv/linux/alpha/setfpucw.c (__setfpucw): Rewrite + with the assumption of being used at program startup only. + * sysdeps/unix/sysv/linux/alpha/nptl/localplt.data: Add optional entries for _OtsConvertFloatTX, _OtsCvtQUX, _OtsCvtXQ, _OtsGtrX, _OtsLeqX, _OtsNintXQ. diff --git a/ports/sysdeps/unix/sysv/linux/alpha/setfpucw.c b/ports/sysdeps/unix/sysv/linux/alpha/setfpucw.c index becc11f..99ffd06 100644 --- a/ports/sysdeps/unix/sysv/linux/alpha/setfpucw.c +++ b/ports/sysdeps/unix/sysv/linux/alpha/setfpucw.c @@ -1,5 +1,5 @@ /* Set FP exception mask and rounding mode. - Copyright (C) 1996, 1997, 1998, 2003 Free Software Foundation, Inc. + Copyright (C) 1996-2012 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 @@ -17,63 +17,45 @@ . */ #include -#include +#include -extern void __ieee_set_fp_control (unsigned long); -libc_hidden_proto(__ieee_set_fp_control) -extern unsigned long __ieee_get_fp_control (void); -libc_hidden_proto(__ieee_get_fp_control) - -static inline unsigned long -rdfpcr (void) -{ - unsigned long fpcr; - asm ("excb; mf_fpcr %0" : "=f"(fpcr)); - return fpcr; -} - - -static inline void -wrfpcr (unsigned long fpcr) -{ - asm volatile ("mt_fpcr %0; excb" : : "f"(fpcr)); -} +#define convert_bit(M, F, T) \ + ((T) < (F) \ + ? ((M) / ((F) / (T))) & (T) \ + : ((M) & (F)) * ((T) / (F))) void __setfpucw (fpu_control_t fpu_control) { - unsigned long fpcr = 0, fpcw = 0; - - if (!fpu_control) - fpu_control = _FPU_DEFAULT; - - /* first, set dynamic rounding mode: */ - - fpcr = rdfpcr(); - fpcr &= ~FPCR_DYN_MASK; - switch (fpu_control & 0xc00) - { - case _FPU_RC_NEAREST: fpcr |= FPCR_DYN_NORMAL; break; - case _FPU_RC_DOWN: fpcr |= FPCR_DYN_MINUS; break; - case _FPU_RC_UP: fpcr |= FPCR_DYN_PLUS; break; - case _FPU_RC_ZERO: fpcr |= FPCR_DYN_CHOPPED; break; - } - wrfpcr(fpcr); - - /* now tell kernel about traps that we like to hear about: */ - - fpcw = __ieee_get_fp_control(); - fpcw &= ~IEEE_TRAP_ENABLE_MASK; - - if (!(fpu_control & _FPU_MASK_IM)) fpcw |= IEEE_TRAP_ENABLE_INV; - if (!(fpu_control & _FPU_MASK_DM)) fpcw |= IEEE_TRAP_ENABLE_UNF; - if (!(fpu_control & _FPU_MASK_ZM)) fpcw |= IEEE_TRAP_ENABLE_DZE; - if (!(fpu_control & _FPU_MASK_OM)) fpcw |= IEEE_TRAP_ENABLE_OVF; - if (!(fpu_control & _FPU_MASK_PM)) fpcw |= IEEE_TRAP_ENABLE_INE; - - __fpu_control = fpu_control; /* update global copy */ - - __ieee_set_fp_control(fpcw); + unsigned long fpcr, swcr, fc = (int)fpu_control; + + /* ??? If this was a real external interface we'd want to read the current + exception state with __ieee_get_fp_control. But this is an internal + function only called at process startup, so there's no point in trying + to preserve exceptions that cannot have been raised yet. Indeed, this + entire function is likely to be one big nop unless the user overrides + the default __fpu_control variable. */ + + /* Convert the rounding mode from fpu_control.h format. */ + const unsigned long conv_rnd + = ( (FE_TOWARDZERO << (_FPU_RC_ZERO >> 8)) + | (FE_DOWNWARD << (_FPU_RC_DOWN >> 8)) + | (FE_TONEAREST << (_FPU_RC_NEAREST >> 8)) + | (FE_UPWARD << (_FPU_RC_UP >> 8))); + + fpcr = ((conv_rnd >> ((fc >> 8) & 3)) & 3) << FPCR_ROUND_SHIFT; + + /* Convert the exception mask from fpu_control.h format. */ + swcr = convert_bit (~fc, _FPU_MASK_IM, FE_INVALID >> SWCR_ENABLE_SHIFT); + swcr |= convert_bit (~fc, _FPU_MASK_DM, FE_UNDERFLOW >> SWCR_ENABLE_SHIFT); + swcr |= convert_bit (~fc, _FPU_MASK_ZM, FE_DIVBYZERO >> SWCR_ENABLE_SHIFT); + swcr |= convert_bit (~fc, _FPU_MASK_OM, FE_OVERFLOW >> SWCR_ENABLE_SHIFT); + swcr |= convert_bit (~fc, _FPU_MASK_PM, FE_INEXACT >> SWCR_ENABLE_SHIFT); + + /* Install everything. */ + __fpu_control = fc; + asm volatile ("mt_fpcr %0" : : "f"(fpcr)); + __ieee_set_fp_control(swcr); } -- 2.7.4