aboutsummaryrefslogtreecommitdiff
path: root/sysdeps/ia64/fpu/e_fmod.S
diff options
context:
space:
mode:
Diffstat (limited to 'sysdeps/ia64/fpu/e_fmod.S')
-rw-r--r--sysdeps/ia64/fpu/e_fmod.S218
1 files changed, 113 insertions, 105 deletions
diff --git a/sysdeps/ia64/fpu/e_fmod.S b/sysdeps/ia64/fpu/e_fmod.S
index 2b3ee96..dbd0a29 100644
--- a/sysdeps/ia64/fpu/e_fmod.S
+++ b/sysdeps/ia64/fpu/e_fmod.S
@@ -1,11 +1,10 @@
.file "fmod.s"
-// Copyright (C) 2000, 2001, Intel Corporation
+
+// Copyright (c) 2000 - 2003, Intel Corporation
// All rights reserved.
//
-// Contributed 2/2/2000 by John Harrison, Cristina Iordache, Ted Kubaska,
-// Bob Norin, Shane Story, and Ping Tak Peter Tang of the Computational
-// Software Lab, Intel Corporation.
+// Contributed 2000 by the Intel Numerics Group, Intel Corporation
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
@@ -21,7 +20,7 @@
// * The name of Intel Corporation may not be used to endorse or promote
// products derived from this software without specific prior written
// permission.
-//
+
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
@@ -36,38 +35,42 @@
//
// Intel Corporation is the author of this code, and requests that all
// problem reports or change requests be submitted to it directly at
-// http://developer.intel.com/opensource.
+// http://www.intel.com/software/products/opensource/libraries/num.htm.
//
// History
//====================================================================
-// 2/02/00 Initial version
-// 3/02/00 New Algorithm
-// 4/04/00 Unwind support added
-// 8/15/00 Bundle added after call to __libm_error_support to properly
+// 02/02/00 Initial version
+// 03/02/00 New Algorithm
+// 04/04/00 Unwind support added
+// 08/15/00 Bundle added after call to __libm_error_support to properly
// set [the previously overwritten] GR_Parameter_RESULT.
-//11/28/00 Set FR_Y to f9
+// 11/28/00 Set FR_Y to f9
+// 03/11/02 Fixed flags for fmod(qnan,zero)
+// 05/20/02 Cleaned up namespace and sf0 syntax
+// 02/10/03 Reordered header: .section, .global, .proc, .align
+// 04/28/03 Fix: fmod(sNaN,0) no longer sets errno
//
// API
//====================================================================
-// double fmod(double,double);
+// double fmod(double,double);
//
// Overview of operation
//====================================================================
// fmod(a,b)=a-i*b,
-// where i is an integer such that, if b!=0,
+// where i is an integer such that, if b!=0,
// |i|<|a/b| and |a/b-i|<1
//
// Algorithm
//====================================================================
// a). if |a|<|b|, return a
-// b). get quotient and reciprocal overestimates accurate to
+// b). get quotient and reciprocal overestimates accurate to
// 33 bits (q2,y2)
// c). if the exponent difference (exponent(a)-exponent(b))
// is less than 32, truncate quotient to integer and
// finish in one iteration
// d). if exponent(a)-exponent(b)>=32 (q2>=2^32)
// round quotient estimate to single precision (k=RN(q2)),
-// calculate partial remainder (a'=a-k*b),
+// calculate partial remainder (a'=a-k*b),
// get quotient estimate (a'*y2), and repeat from c).
//
// Special cases
@@ -81,14 +84,9 @@
// General registers: r2,r29,r32 (ar.pfs), r33-r39
// Floating point registers: f6-f15
-#include "libm_support.h"
-
-.section .text
-
-
GR_SAVE_B0 = r33
GR_SAVE_PFS = r34
-GR_SAVE_GP = r35
+GR_SAVE_GP = r35
GR_SAVE_SP = r36
GR_Parameter_X = r37
@@ -101,17 +99,9 @@ FR_Y = f9
FR_RESULT = f8
-.proc fmod#
-.align 32
-.global fmod#
-.align 32
+.section .text
+GLOBAL_IEEE754_ENTRY(fmod)
-fmod:
-#ifdef _LIBC
-.global __ieee754_fmod
-.type __ieee754_fmod,@function
-__ieee754_fmod:
-#endif
// inputs in f8, f9
// result in f8
@@ -133,12 +123,12 @@ __ieee754_fmod:
// (1) y0
frcpa.s1 f10,p6=f6,f7
nop.i 0
-}
+}
// Y +-NAN, +-inf, +-0? p7
{ .mfi
nop.m 999
-(p0) fclass.m.unc p7,p0 = f9, 0xe7
+ fclass.m.unc p7,p0 = f9, 0xe7
nop.i 999;;
}
@@ -149,14 +139,14 @@ __ieee754_fmod:
{ .mfi
nop.m 999
-(p0) fclass.m.unc p9,p0 = f8, 0xe3
- nop.i 999
+ fclass.m.unc p9,p0 = f8, 0xe3
+ nop.i 999
}
// |x| < |y|? Return x p8
{ .mfi
nop.m 999
-(p0) fcmp.lt.unc.s1 p8,p0 = f6,f7
+ fcmp.lt.unc.s1 p8,p0 = f6,f7
nop.i 999 ;;
}
@@ -172,33 +162,33 @@ __ieee754_fmod:
// (2) q0=a*y0
(p6) fma.s1 f13=f6,f10,f0
nop.i 0
-}
+}
{ .mfi
nop.m 0
// (3) e0 = 1 - b * y0
(p6) fnma.s1 f12=f7,f10,f1
nop.i 0;;
-}
+}
{.mfi
nop.m 0
// normalize x (if |x|<|y|)
(p8) fma.d.s0 f8=f8,f1,f0
nop.i 0
-}
+}
{.bbb
- (p9) br.cond.spnt L(FMOD_X_NAN_INF)
- (p7) br.cond.spnt L(FMOD_Y_NAN_INF_ZERO)
+ (p9) br.cond.spnt FMOD_X_NAN_INF
+ (p7) br.cond.spnt FMOD_Y_NAN_INF_ZERO
// if |x|<|y|, return
(p8) br.ret.spnt b0;;
}
- {.mfi
+ {.mfi
nop.m 0
// normalize x
fma.s0 f6=f6,f1,f0
nop.i 0
-}
+}
{.mfi
nop.m 0
// normalize y
@@ -212,45 +202,45 @@ __ieee754_fmod:
// (4) q1=q0+e0*q0
(p6) fma.s1 f13=f12,f13,f13
nop.i 0
-}
+}
{ .mfi
nop.m 0
// (5) e1 = e0 * e0 + 2^-34
(p6) fma.s1 f14=f12,f12,f11
nop.i 0;;
-}
+}
{.mlx
nop.m 0
movl r2=0x33a00000;;
-}
+}
{ .mfi
nop.m 0
// (6) y1 = y0 + e0 * y0
(p6) fma.s1 f10=f12,f10,f10
nop.i 0;;
-}
+}
{.mfi
// set f12=1.25*2^{-24}
setf.s f12=r2
// (7) q2=q1+e1*q1
(p6) fma.s1 f13=f13,f14,f13
nop.i 0;;
-}
+}
{.mfi
nop.m 0
fmerge.s f9=f8,f9
nop.i 0
-}
+}
{ .mfi
nop.m 0
// (8) y2 = y1 + e1 * y1
(p6) fma.s1 f10=f14,f10,f10
// set p6=0, p10=0
cmp.ne.and p6,p10=r0,r0;;
-}
+}
.align 32
-L(loop53):
+loop53:
{.mfi
nop.m 0
// compare q2, 2^32
@@ -280,7 +270,7 @@ L(loop53):
// normalize truncated quotient
(p8) fcvt.xf f13=f11
nop.i 0;;
-}
+}
{ .mfi
nop.m 0
// calculate remainder (assuming f13=RZ(Q))
@@ -289,7 +279,7 @@ L(loop53):
}
{.mfi
nop.m 0
- // also if exponent>32, round quotient to single precision
+ // also if exponent>32, round quotient to single precision
// and subtract 1 ulp: q=q-q*(1.25*2^{-24})
(p7) fnma.s.s1 f11=f13,f12,f13
nop.i 0;;
@@ -332,7 +322,7 @@ L(loop53):
.pred.rel "mutex",p6,p10
{.mfb
nop.m 0
- // add b to estimated remainder (to cover the case when the quotient was overestimated)
+ // add b to estimated remainder (to cover the case when the quotient was overestimated)
// also set correct sign by using f9=|b|*sgn(a), f12=sgn(a)
(p6) fma.d.s0 f8=f11,f12,f9
nop.b 0
@@ -354,97 +344,114 @@ L(loop53):
nop.m 0
// if f14 was RZ(Q), set remainder to f14
(p9) mov f6=f14
- br.cond.sptk L(loop53);;
+ br.cond.sptk loop53;;
}
-L(FMOD_X_NAN_INF):
+FMOD_X_NAN_INF:
// Y zero ?
-{.mfi
+{.mfi
+ nop.m 0
+ fclass.m p10,p0=f8,0xc3 // Test x=nan
+ nop.i 0
+}
+{.mfi
nop.m 0
fma.s1 f10=f9,f1,f0
nop.i 0;;
}
+
+{.mfi
+ nop.m 0
+ fma.s0 f8=f8,f1,f0
+ nop.i 0
+}
{.mfi
+ nop.m 0
+(p10) fclass.m p10,p0=f9,0x07 // Test x=nan, and y=zero
+ nop.i 0;;
+}
+
+{.mfb
nop.m 0
fcmp.eq.unc.s1 p11,p0=f10,f0
- nop.i 0;;
+(p10) br.ret.spnt b0;; // Exit with result=x if x=nan and y=zero
}
{.mib
nop.m 0
nop.i 0
// if Y zero
- (p11) br.cond.spnt L(FMOD_Y_ZERO);;
+ (p11) br.cond.spnt FMOD_Y_ZERO;;
}
// X infinity? Return QNAN indefinite
{ .mfi
nop.m 999
-(p0) fclass.m.unc p8,p9 = f8, 0x23
- nop.i 999;;
+ fclass.m.unc p8,p9 = f8, 0x23
+ nop.i 999;;
}
// Y NaN ?
{.mfi
- nop.m 999
+ nop.m 999
(p8) fclass.m p9,p8=f9,0xc3
- nop.i 0;;
+ nop.i 0;;
}
{.mfi
- nop.m 999
-(p8) frcpa.s0 f8,p0 = f8,f8
+ nop.m 999
+(p8) frcpa.s0 f8,p0 = f8,f8
nop.i 0
-}
+}
{ .mfi
nop.m 999
- // also set Denormal flag if necessary
+ // also set Denormal flag if necessary
(p8) fma.s0 f9=f9,f1,f0
nop.i 999 ;;
}
{ .mfb
nop.m 999
-(p8) fma.d f8=f8,f1,f0
- nop.b 999 ;;
+(p8) fma.d.s0 f8=f8,f1,f0
+ nop.b 999 ;;
}
{ .mfb
nop.m 999
-(p9) frcpa.s0 f8,p7=f8,f9
- br.ret.sptk b0 ;;
+(p9) frcpa.s0 f8,p7=f8,f9
+ br.ret.sptk b0 ;;
}
-L(FMOD_Y_NAN_INF_ZERO):
+FMOD_Y_NAN_INF_ZERO:
// Y INF
{ .mfi
nop.m 999
-(p0) fclass.m.unc p7,p0 = f9, 0x23
+ fclass.m.unc p7,p0 = f9, 0x23
nop.i 999 ;;
}
{ .mfb
nop.m 999
-(p7) fma.d f8=f8,f1,f0
-(p7) br.ret.spnt b0 ;;
+(p7) fma.d.s0 f8=f8,f1,f0
+(p7) br.ret.spnt b0 ;;
}
// Y NAN?
{ .mfi
nop.m 999
-(p0) fclass.m.unc p9,p0 = f9, 0xc3
+ fclass.m.unc p9,p0 = f9, 0xc3
nop.i 999 ;;
}
{ .mfb
nop.m 999
-(p9) fma.d f8=f9,f1,f0
-(p9) br.ret.spnt b0 ;;
+(p9) fma.d.s0 f8=f9,f1,f0
+(p9) br.ret.spnt b0 ;;
}
-L(FMOD_Y_ZERO):
+FMOD_Y_ZERO:
// Y zero? Must be zero at this point
// because it is the only choice left.
// Return QNAN indefinite
@@ -452,60 +459,57 @@ L(FMOD_Y_ZERO):
{.mfi
nop.m 0
// set Invalid
- frcpa f12,p0=f0,f0
+ frcpa.s0 f12,p0=f0,f0
nop.i 0
}
// X NAN?
{ .mfi
nop.m 999
-(p0) fclass.m.unc p9,p10 = f8, 0xc3
+ fclass.m.unc p9,p10 = f8, 0xc3
nop.i 999 ;;
}
{ .mfi
nop.m 999
-(p10) fclass.nm p9,p10 = f8, 0xff
+(p10) fclass.nm p9,p10 = f8, 0xff
nop.i 999 ;;
}
{.mfi
nop.m 999
- (p9) frcpa f11,p7=f8,f0
+ (p9) frcpa.s0 f11,p7=f8,f0
nop.i 0;;
}
{ .mfi
nop.m 999
-(p10) frcpa f11,p7 = f9,f9
-(p0) mov GR_Parameter_TAG = 121 ;;
+(p10) frcpa.s0 f11,p7 = f9,f9
+ mov GR_Parameter_TAG = 121 ;;
}
{ .mfi
nop.m 999
-(p0) fmerge.s f10 = f8, f8
+ fmerge.s f10 = f8, f8
nop.i 999
}
{ .mfb
nop.m 999
-(p0) fma.d f8=f11,f1,f0
-(p0) br.sptk __libm_error_region;;
+ fma.d.s0 f8=f11,f1,f0
+ br.sptk __libm_error_region;;
}
-.endp fmod
-ASM_SIZE_DIRECTIVE(fmod)
-ASM_SIZE_DIRECTIVE(__ieee754_fmod)
+GLOBAL_IEEE754_END(fmod)
-.proc __libm_error_region
-__libm_error_region:
+LOCAL_LIBM_ENTRY(__libm_error_region)
.prologue
{ .mfi
add GR_Parameter_Y=-32,sp // Parameter 2 value
nop.f 0
.save ar.pfs,GR_SAVE_PFS
- mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
+ mov GR_SAVE_PFS=ar.pfs // Save ar.pfs
}
{ .mfi
-.fframe 64
+.fframe 64
add sp=-64,sp // Create new stack
nop.f 0
mov GR_SAVE_GP=gp // Save gp
@@ -513,18 +517,18 @@ __libm_error_region:
{ .mmi
stfd [GR_Parameter_Y] = FR_Y,16 // Save Parameter 2 on stack
add GR_Parameter_X = 16,sp // Parameter 1 address
-.save b0, GR_SAVE_B0
- mov GR_SAVE_B0=b0 // Save b0
+.save b0, GR_SAVE_B0
+ mov GR_SAVE_B0=b0 // Save b0
};;
.body
{ .mib
- stfd [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
- add GR_Parameter_RESULT = 0,GR_Parameter_Y
- nop.b 0 // Parameter 3 address
+ stfd [GR_Parameter_X] = FR_X // Store Parameter 1 on stack
+ add GR_Parameter_RESULT = 0,GR_Parameter_Y
+ nop.b 0 // Parameter 3 address
}
{ .mib
stfd [GR_Parameter_Y] = FR_RESULT // Store Parameter 3 on stack
- add GR_Parameter_Y = -16,GR_Parameter_Y
+ add GR_Parameter_Y = -16,GR_Parameter_Y
br.call.sptk b0=__libm_error_support# // Call error handling function
};;
{ .mmi
@@ -539,13 +543,17 @@ __libm_error_region:
mov b0 = GR_SAVE_B0 // Restore return address
};;
{ .mib
- mov gp = GR_SAVE_GP // Restore gp
+ mov gp = GR_SAVE_GP // Restore gp
mov ar.pfs = GR_SAVE_PFS // Restore ar.pfs
br.ret.sptk b0 // Return
-};;
+};;
+
+LOCAL_LIBM_END(__libm_error_region)
-.endp __libm_error_region
-ASM_SIZE_DIRECTIVE(__libm_error_region)
.type __libm_error_support#,@function
.global __libm_error_support#
+
+
+
+
generated by cgit v1.1 at 2024-11-21 12:31:29 +0000