0bd55e799ceadf54572ce4cae097df9381a11e9e
[mesa.git] / src / mesa / x86 / common_x86_asm.S
1 /* $Id: common_x86_asm.S,v 1.16 2003/01/09 23:43:02 brianp Exp $ */
2
3 /*
4 * Mesa 3-D graphics library
5 * Version: 5.1
6 *
7 * Copyright (C) 1999-2003 Brian Paul All Rights Reserved.
8 *
9 * Permission is hereby granted, free of charge, to any person obtaining a
10 * copy of this software and associated documentation files (the "Software"),
11 * to deal in the Software without restriction, including without limitation
12 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
13 * and/or sell copies of the Software, and to permit persons to whom the
14 * Software is furnished to do so, subject to the following conditions:
15 *
16 * The above copyright notice and this permission notice shall be included
17 * in all copies or substantial portions of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
20 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
21 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
22 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
23 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
24 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25 */
26
27 /*
28 * Check extended CPU capabilities. Now justs returns the raw CPUID
29 * feature information, allowing the higher level code to interpret the
30 * results.
31 *
32 * Written by Holger Waechtler <holger@akaflieg.extern.tu-berlin.de>
33 *
34 * Cleaned up and simplified by Gareth Hughes <gareth@valinux.com>
35 */
36
37 /*
38 * NOTE: Avoid using spaces in between '(' ')' and arguments, especially
39 * with macros like CONST, LLBL that expand to CONCAT(...). Putting spaces
40 * in there will break the build on some platforms.
41 */
42
43 #include "matypes.h"
44 #include "common_x86_features.h"
45
46
47 /* Intel vendor string
48 */
49 #define GENU 0x756e6547 /* "Genu" */
50 #define INEI 0x49656e69 /* "ineI" */
51 #define NTEL 0x6c65746e /* "ntel" */
52
53 /* AMD vendor string
54 */
55 #define AUTH 0x68747541 /* "Auth" */
56 #define ENTI 0x69746e65 /* "enti" */
57 #define CAMD 0x444d4163 /* "cAMD" */
58
59
60 SEG_DATA
61
62 /* We might want to print out some useful messages.
63 */
64 GLNAME( found_intel ): STRING( "Genuine Intel processor found\n\0" )
65 GLNAME( found_amd ): STRING( "Authentic AMD processor found\n\0" )
66
67
68 SEG_TEXT
69
70 ALIGNTEXT4
71 GLOBL GLNAME( _mesa_identify_x86_cpu_features )
72 GLNAME( _mesa_identify_x86_cpu_features ):
73
74 PUSH_L ( EBX )
75 PUSH_L ( ESI )
76
77 /* Test for the CPUID command. If the ID Flag bit in EFLAGS
78 * (bit 21) is writable, the CPUID command is present.
79 */
80 PUSHF_L
81 POP_L ( EAX )
82 MOV_L ( EAX, ECX )
83 XOR_L ( CONST(0x00200000), EAX )
84 PUSH_L ( EAX )
85 POPF_L
86 PUSHF_L
87 POP_L ( EAX )
88
89 /* Verify the ID Flag bit has been written.
90 */
91 CMP_L ( ECX, EAX )
92 JZ ( LLBL (cpuid_done) )
93
94 /* Get the CPU vendor info.
95 */
96 XOR_L ( EAX, EAX )
97 CPUID
98
99 /* Test for Intel processors. We must look for the
100 * "GenuineIntel" string in EBX, ECX and EDX.
101 */
102 CMP_L ( CONST(GENU), EBX )
103 JNE ( LLBL(cpuid_amd) )
104 CMP_L ( CONST(INEI), EDX )
105 JNE ( LLBL(cpuid_amd) )
106 CMP_L ( CONST(NTEL), ECX )
107 JNE ( LLBL(cpuid_amd) )
108
109 /* We have an Intel processor, so we can get the feature
110 * information with an CPUID input value of 1.
111 */
112 MOV_L ( CONST(0x1), EAX )
113 CPUID
114 MOV_L ( EDX, EAX )
115
116 /* Mask out highest bit, which is used by AMD for 3dnow
117 * Newer Intel have this bit set, but do not support 3dnow
118 */
119 AND_L ( CONST(0X7FFFFFFF), EAX)
120 JMP ( LLBL(cpuid_done) )
121
122 LLBL(cpuid_amd):
123
124 /* Test for AMD processors. We must look for the
125 * "AuthenticAMD" string in EBX, ECX and EDX.
126 */
127 CMP_L ( CONST(AUTH), EBX )
128 JNE ( LLBL(cpuid_other) )
129 CMP_L ( CONST(ENTI), EDX )
130 JNE ( LLBL(cpuid_other) )
131 CMP_L ( CONST(CAMD), ECX )
132 JNE ( LLBL(cpuid_other) )
133
134 /* We have an AMD processor, so we can get the feature
135 * information after we verify that the extended functions are
136 * supported.
137 */
138 /* The features we need are almost all in the extended set. The
139 * exception is SSE enable, which is in the standard set (0x1).
140 */
141 MOV_L ( CONST(0x1), EAX )
142 CPUID
143 TEST_L ( EAX, EAX )
144 JZ ( LLBL (cpuid_failed) )
145 MOV_L ( EDX, ESI )
146
147 MOV_L ( CONST(0x80000000), EAX )
148 CPUID
149 TEST_L ( EAX, EAX )
150 JZ ( LLBL (cpuid_failed) )
151
152 MOV_L ( CONST(0x80000001), EAX )
153 CPUID
154 MOV_L ( EDX, EAX )
155
156 AND_L ( CONST(0x02000000), ESI ) /* OR in the SSE bit */
157 OR_L ( ESI, EAX )
158
159 JMP ( LLBL (cpuid_done) )
160
161 LLBL(cpuid_other):
162
163 /* Test for other processors here when required.
164 */
165
166 LLBL(cpuid_failed):
167
168 /* If we can't determine the feature information, we must
169 * return zero to indicate that no platform-specific
170 * optimizations can be used.
171 */
172 MOV_L ( CONST(0), EAX )
173
174 LLBL (cpuid_done):
175
176 POP_L ( ESI )
177 POP_L ( EBX )
178 RET
179
180
181 #ifdef USE_SSE_ASM
182 /* Execute an SSE instruction to see if the operating system correctly
183 * supports SSE. A signal handler for SIGILL should have been set
184 * before calling this function, otherwise this could kill the client
185 * application.
186 */
187 ALIGNTEXT4
188 GLOBL GLNAME( _mesa_test_os_sse_support )
189 GLNAME( _mesa_test_os_sse_support ):
190
191 XORPS ( XMM0, XMM0 )
192
193 RET
194
195
196 /* Perform an SSE divide-by-zero to see if the operating system
197 * correctly supports unmasked SIMD FPU exceptions. Signal handlers for
198 * SIGILL and SIGFPE should have been set before calling this function,
199 * otherwise this could kill the client application.
200 */
201 ALIGNTEXT4
202 GLOBL GLNAME( _mesa_test_os_sse_exception_support )
203 GLNAME( _mesa_test_os_sse_exception_support ):
204
205 PUSH_L ( EBP )
206 MOV_L ( ESP, EBP )
207 SUB_L ( CONST( 8 ), ESP )
208
209 /* Save the original MXCSR register value.
210 */
211 STMXCSR ( REGOFF( -4, EBP ) )
212
213 /* Unmask the divide-by-zero exception and perform one.
214 */
215 STMXCSR ( REGOFF( -8, EBP ) )
216 AND_L ( CONST( 0xfffffdff ), REGOFF( -8, EBP ) )
217 LDMXCSR ( REGOFF( -8, EBP ) )
218
219 XORPS ( XMM0, XMM0 )
220
221 PUSH_L ( CONST( 0x3f800000 ) )
222 PUSH_L ( CONST( 0x3f800000 ) )
223 PUSH_L ( CONST( 0x3f800000 ) )
224 PUSH_L ( CONST( 0x3f800000 ) )
225
226 MOVUPS ( REGIND( ESP ), XMM1 )
227
228 DIVPS ( XMM0, XMM1 )
229
230 /* Restore the original MXCSR register value.
231 */
232 LDMXCSR ( REGOFF( -4, EBP ) )
233
234 LEAVE
235 RET
236
237 #endif