1.file "fdiml.s"
2
3
4// Copyright (c) 2001 - 2003, Intel Corporation
5// All rights reserved.
6//
7//
8// Redistribution and use in source and binary forms, with or without
9// modification, are permitted provided that the following conditions are
10// met:
11//
12// * Redistributions of source code must retain the above copyright
13// notice, this list of conditions and the following disclaimer.
14//
15// * Redistributions in binary form must reproduce the above copyright
16// notice, this list of conditions and the following disclaimer in the
17// documentation and/or other materials provided with the distribution.
18//
19// * The name of Intel Corporation may not be used to endorse or promote
20// products derived from this software without specific prior written
21// permission.
22
23// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
27// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
28// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
31// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
32// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34//
35// Intel Corporation is the author of this code, and requests that all
36// problem reports or change requests be submitted to it directly at
37// http://www.intel.com/software/products/opensource/libraries/num.htm.
38//
39// History
40//==============================================================
41// 06/08/01 Initial version
42// 08/23/01 Corrected error tag number
43// 05/20/02 Cleaned up namespace and sf0 syntax
44// 01/28/03 Improved performance; fixed parameters for call to error routine
45//
46// API
47//==============================================================
48// long double fdiml( long double x, long double y );
49// input  floating point f8, f9
50// output floating point f8
51//
52//
53// Overview of operation
54//==============================================================
55// fdiml determines the positive difference between the arguments
56//   Result = x - y   if x > y
57//          = +0      if x <= y
58//
59// Error support is called if x-y overflows for x > y
60//
61
62// Registers used
63//==============================================================
64// General purpose registers: r14, r32 - r39
65
66rExpBig               = r14
67
68// r36-39 parameters for libm_error_support
69GR_SAVE_B0            = r33
70GR_SAVE_GP            = r34
71GR_SAVE_PFS           = r35
72
73GR_Parameter_X        = r36
74GR_Parameter_Y        = r37
75GR_Parameter_RESULT   = r38
76GR_Parameter_TAG      = r39
77
78// Floating-point registers: f8 - f12
79
80f_tmp_result          = f10
81fBig                  = f11
82fNormX                = f12
83
84// Predicate registers: p6 - p10
85
86
87.section .text
88GLOBAL_LIBM_ENTRY(fdiml)
89
90{ .mfi
91      mov             rExpBig = 0x13fff  // Exponent to indicate overflow
92      fcmp.le.s1      p6,p7 = f8, f9     // Is x <= y?
93      nop.i           0
94}
95{ .mfi
96      nop.m           0
97      fnorm.s1        fNormX = f8        // Save x
98      nop.i           0
99}
100;;
101
102{ .mfi
103      setf.exp        fBig = rExpBig     // Constant to test for overflow
104      fcmp.eq.s0      p8,p0 = f8, f9     // Dummy op to set Denormal or Invalid
105      nop.i           0
106}
107;;
108
109{ .mfi
110      nop.m           0
111      fclass.m        p9,p10 = f8, 0x1e3 // Test for x natval, nan, inf
112      nop.i           0
113}
114;;
115
116{ .mfi
117      nop.m           0
118(p6)  fmerge.s        f8 = f0, f0        // Result is +0 if x <= y
119      nop.i           0
120}
121{ .mfi
122      nop.m           0
123(p7)  fms.s0          f8 = f8, f1, f9    // Result is x - y if x > y
124      nop.i           0
125}
126;;
127
128{ .mfi
129      nop.m           0
130(p10) fclass.m        p9,p10 = f9, 0x1e3 // Test for y natval, nan, inf
131      nop.i           0
132}
133;;
134
135{ .mfi
136      nop.m           0
137(p10) fcmp.ge.s1      p8,p0 = f8, fBig   // Test result for overflow
138      nop.i           0
139}
140;;
141
142{ .mbb
143(p9)  cmp.ne          p8,p0 = r0,r0      // Clear p8 if x or y natval,nan,inf
144(p8)  br.cond.spnt    FDIM_OVERFLOW      // Branch if result overflows
145      br.ret.sptk     b0                 // Normal return
146}
147;;
148
149
150// Here if result will overflow
151FDIM_OVERFLOW:
152{ .mfi
153      alloc           r32=ar.pfs,2,2,4,0
154      fms.s0          f_tmp_result = f8,f1,f9 // Normalize result force overflow
155      nop.i           0
156}
157{ .mfb
158      mov             GR_Parameter_TAG = 195  // Error code
159      nop.f           0
160      br.cond.sptk    __libm_error_region     // Branch to error code
161}
162;;
163
164GLOBAL_LIBM_END(fdiml)
165libm_alias_ldouble_other (fdim, fdim)
166
167
168LOCAL_LIBM_ENTRY(__libm_error_region)
169// Call error support to report possible range error
170.prologue
171
172{ .mfi
173        add   GR_Parameter_Y=-32,sp            // Parameter 2 value
174        nop.f 0
175.save   ar.pfs,GR_SAVE_PFS
176        mov  GR_SAVE_PFS=ar.pfs                // Save ar.pfs
177}
178{ .mfi
179.fframe 64
180        add sp=-64,sp                          // Create new stack
181        nop.f 0
182        mov GR_SAVE_GP=gp                      // Save gp
183};;
184
185{ .mmi
186        stfe [GR_Parameter_Y] = f9,16          // STORE Parameter 2 on stack
187        add GR_Parameter_X = 16,sp             // Parameter 1 address
188.save   b0, GR_SAVE_B0
189        mov GR_SAVE_B0=b0                      // Save b0
190};;
191
192.body
193{ .mib
194        stfe [GR_Parameter_X] = fNormX         // STORE Parameter 1 on stack
195        add   GR_Parameter_RESULT = 0,GR_Parameter_Y // Parameter 3 address
196        nop.b 0
197}
198{ .mib
199        stfe [GR_Parameter_Y] = f_tmp_result   // STORE Parameter 3 on stack
200        add   GR_Parameter_Y = -16,GR_Parameter_Y
201        br.call.sptk b0=__libm_error_support#  // Call error handling function
202};;
203
204{ .mmi
205        add   GR_Parameter_RESULT = 48,sp
206        nop.m 0
207        nop.i 0
208};;
209
210{ .mmi
211        ldfe  f8 = [GR_Parameter_RESULT]       // Get return result off stack
212.restore sp
213        add   sp = 64,sp                       // Restore stack pointer
214        mov   b0 = GR_SAVE_B0                  // Restore return address
215};;
216
217{ .mib
218        mov   gp = GR_SAVE_GP                  // Restore gp
219        mov   ar.pfs = GR_SAVE_PFS             // Restore ar.pfs
220        br.ret.sptk     b0                     // Return
221};;
222
223LOCAL_LIBM_END(__libm_error_region)
224
225
226.type   __libm_error_support#,@function
227.global __libm_error_support#
228