1 | ; $Id: bs3-fpustate-1-template.mac 98103 2023-01-17 14:15:46Z vboxsync $
|
---|
2 | ;; @file
|
---|
3 | ; BS3Kit - bs3-fpustate-1, assembly template.
|
---|
4 | ;
|
---|
5 |
|
---|
6 | ;
|
---|
7 | ; Copyright (C) 2007-2023 Oracle and/or its affiliates.
|
---|
8 | ;
|
---|
9 | ; This file is part of VirtualBox base platform packages, as
|
---|
10 | ; available from https://www.virtualbox.org.
|
---|
11 | ;
|
---|
12 | ; This program is free software; you can redistribute it and/or
|
---|
13 | ; modify it under the terms of the GNU General Public License
|
---|
14 | ; as published by the Free Software Foundation, in version 3 of the
|
---|
15 | ; License.
|
---|
16 | ;
|
---|
17 | ; This program is distributed in the hope that it will be useful, but
|
---|
18 | ; WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
19 | ; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
---|
20 | ; General Public License for more details.
|
---|
21 | ;
|
---|
22 | ; You should have received a copy of the GNU General Public License
|
---|
23 | ; along with this program; if not, see <https://www.gnu.org/licenses>.
|
---|
24 | ;
|
---|
25 | ; The contents of this file may alternatively be used under the terms
|
---|
26 | ; of the Common Development and Distribution License Version 1.0
|
---|
27 | ; (CDDL), a copy of it is provided in the "COPYING.CDDL" file included
|
---|
28 | ; in the VirtualBox distribution, in which case the provisions of the
|
---|
29 | ; CDDL are applicable instead of those of the GPL.
|
---|
30 | ;
|
---|
31 | ; You may elect to license modified versions of this file under the
|
---|
32 | ; terms and conditions of either the GPL or the CDDL or both.
|
---|
33 | ;
|
---|
34 | ; SPDX-License-Identifier: GPL-3.0-only OR CDDL-1.0
|
---|
35 | ;
|
---|
36 |
|
---|
37 |
|
---|
38 | ;*********************************************************************************************************************************
|
---|
39 | ;* Header Files *
|
---|
40 | ;*********************************************************************************************************************************
|
---|
41 | %include "bs3kit-template-header.mac" ; setup environment
|
---|
42 |
|
---|
43 |
|
---|
44 | ;*********************************************************************************************************************************
|
---|
45 | ;* External Symbols *
|
---|
46 | ;*********************************************************************************************************************************
|
---|
47 | TMPL_BEGIN_TEXT
|
---|
48 |
|
---|
49 |
|
---|
50 | ;;
|
---|
51 | ; Initializes the FPU state and saves it to pFxState.
|
---|
52 | ;
|
---|
53 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_InitState)(X86FXSTATE BS3_FAR *pFxState, void *pvMmioReg);
|
---|
54 | ;
|
---|
55 | BS3_PROC_BEGIN_MODE bs3FpuState1_InitState, BS3_PBC_NEAR
|
---|
56 | BS3_CALL_CONV_PROLOG 2
|
---|
57 | push xBP
|
---|
58 | mov xBP, xSP
|
---|
59 | push xBX
|
---|
60 | TONLY16 push ds
|
---|
61 | pushf
|
---|
62 | TONLY64 sub xSP, 20h
|
---|
63 |
|
---|
64 | ;
|
---|
65 | ; x87 state.
|
---|
66 | ;
|
---|
67 | fninit
|
---|
68 | fld dword [TMPL_DATA16_WRT(g_r32V1)]
|
---|
69 | fld qword [TMPL_DATA16_WRT(g_r64V1)]
|
---|
70 | fld tword [TMPL_DATA16_WRT(g_r80V1)]
|
---|
71 | fld qword [TMPL_DATA16_WRT(g_r64V1)]
|
---|
72 | fld dword [TMPL_DATA16_WRT(g_r32V2)]
|
---|
73 | fld dword [TMPL_DATA16_WRT(g_r80_QNaNMax)]
|
---|
74 | fld tword [TMPL_DATA16_WRT(g_r80_SNaNMax)]
|
---|
75 | fld tword [TMPL_DATA16_WRT(g_r80_ThirtyTwo)]
|
---|
76 |
|
---|
77 | ;
|
---|
78 | ; We'll later be using FMUL to test actually using the FPU in RC & R0,
|
---|
79 | ; so for everything to line up correctly with FPU CS:IP and FPU DS:DP,
|
---|
80 | ; we'll call the function here too. This has the benefitial side effect
|
---|
81 | ; of loading correct FPU DS/DS values so we can check that they don't
|
---|
82 | ; get lost either. Also, we now don't have to guess whether the CPU
|
---|
83 | ; emulation sets CS/DS or not.
|
---|
84 | ;
|
---|
85 | TONLY16 push xPRE [xBP + xCB + cbCurRetAddr + sCB + 2]
|
---|
86 | push xPRE [xBP + xCB + cbCurRetAddr + sCB]
|
---|
87 | BS3_CALL TMPL_NM(bs3FpuState1_FMul), 1
|
---|
88 | add xSP, sCB
|
---|
89 |
|
---|
90 | ;
|
---|
91 | ; SSE state
|
---|
92 | ;
|
---|
93 | movdqu xmm0, [TMPL_DATA16_WRT(g_r32_0dot1)]
|
---|
94 | movdqu xmm1, [TMPL_DATA16_WRT(g_r32_Two)]
|
---|
95 | movdqu xmm2, [TMPL_DATA16_WRT(g_r32_ThirtyTwo)]
|
---|
96 | movdqu xmm3, [TMPL_DATA16_WRT(g_r32_SNaN)]
|
---|
97 | movdqu xmm4, [TMPL_DATA16_WRT(g_r80_ThirtyTwo)]
|
---|
98 | movdqu xmm5, [TMPL_DATA16_WRT(g_r32_NegQNaN)]
|
---|
99 | movdqu xmm6, [TMPL_DATA16_WRT(g_r64_Zero)]
|
---|
100 | movdqu xmm7, [TMPL_DATA16_WRT(g_r64_Two)]
|
---|
101 | %if TMPL_BITS == 64
|
---|
102 | movdqu xmm8, [TMPL_DATA16_WRT(g_r64_Ten)]
|
---|
103 | movdqu xmm9, [TMPL_DATA16_WRT(g_r64_ThirtyTwo)]
|
---|
104 | movdqu xmm10, [TMPL_DATA16_WRT(g_r64_Max)]
|
---|
105 | movdqu xmm11, [TMPL_DATA16_WRT(g_r64_SNaN)]
|
---|
106 | movdqu xmm12, [TMPL_DATA16_WRT(g_r64_NegQNaN)]
|
---|
107 | movdqu xmm13, [TMPL_DATA16_WRT(g_r64_QNaNMax)]
|
---|
108 | movdqu xmm14, [TMPL_DATA16_WRT(g_r64_DnMax)]
|
---|
109 | movdqu xmm15, [TMPL_DATA16_WRT(g_r80_Eleven)]
|
---|
110 | %endif
|
---|
111 |
|
---|
112 | ;; @todo status regs
|
---|
113 |
|
---|
114 | ;
|
---|
115 | ; Save it. Note that DS is no longer valid in 16-bit code.
|
---|
116 | ; To be on the safe side, we load and save the state once again.
|
---|
117 | ;
|
---|
118 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
119 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
120 | cli
|
---|
121 | %if TMPL_BITS == 64
|
---|
122 | o64 fxsave [xBX]
|
---|
123 | fninit
|
---|
124 | o64 fxrstor [xBX]
|
---|
125 | o64 fxsave [xBX]
|
---|
126 | %else
|
---|
127 | fxsave [xBX]
|
---|
128 | fninit
|
---|
129 | fxrstor [xBX]
|
---|
130 | fxsave [xBX]
|
---|
131 | %endif
|
---|
132 |
|
---|
133 | .return:
|
---|
134 | TONLY64 add xSP, 20h
|
---|
135 | popf
|
---|
136 | TONLY16 pop ds
|
---|
137 | pop xBX
|
---|
138 | mov xSP, xBP
|
---|
139 | pop xBP
|
---|
140 | BS3_CALL_CONV_EPILOG 2
|
---|
141 | BS3_HYBRID_RET
|
---|
142 | BS3_PROC_END_MODE bs3FpuState1_InitState
|
---|
143 |
|
---|
144 |
|
---|
145 | ;;
|
---|
146 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_Restore)(X86FXSTATE const BS3_FAR *pFxState);
|
---|
147 | ;
|
---|
148 | BS3_PROC_BEGIN_MODE bs3FpuState1_Restore, BS3_PBC_NEAR
|
---|
149 | push xBP
|
---|
150 | mov xBP, xSP
|
---|
151 |
|
---|
152 | %if TMPL_BITS == 64
|
---|
153 | o64 fxrstor [rcx]
|
---|
154 |
|
---|
155 | %elif TMPL_BITS == 32
|
---|
156 | mov eax, [xBP + xCB*2]
|
---|
157 | fxrstor [eax]
|
---|
158 |
|
---|
159 | %elif TMPL_BITS == 16
|
---|
160 | mov ax, ds
|
---|
161 | mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
162 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
163 | fxrstor [bx]
|
---|
164 | mov ds, ax
|
---|
165 | %else
|
---|
166 | %error TMPL_BITS
|
---|
167 | %endif
|
---|
168 |
|
---|
169 | mov xSP, xBP
|
---|
170 | pop xBP
|
---|
171 | BS3_HYBRID_RET
|
---|
172 | BS3_PROC_END_MODE bs3FpuState1_Restore
|
---|
173 |
|
---|
174 | ;;
|
---|
175 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_Save)(X86FXSTATE BS3_FAR *pFxState);
|
---|
176 | ;
|
---|
177 | BS3_PROC_BEGIN_MODE bs3FpuState1_Save, BS3_PBC_NEAR
|
---|
178 | push xBP
|
---|
179 | mov xBP, xSP
|
---|
180 |
|
---|
181 | %if TMPL_BITS == 64
|
---|
182 | o64 fxsave [rcx]
|
---|
183 |
|
---|
184 | %elif TMPL_BITS == 32
|
---|
185 | mov eax, [xBP + xCB*2]
|
---|
186 | fxsave [eax]
|
---|
187 |
|
---|
188 | %elif TMPL_BITS == 16
|
---|
189 | push bx
|
---|
190 | push ds
|
---|
191 | mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
192 | mov bx, [xBP + xCB + cbCurRetAddr]
|
---|
193 | fxsave [bx]
|
---|
194 | pop ds
|
---|
195 | pop bx
|
---|
196 | %else
|
---|
197 | %error TMPL_BITS
|
---|
198 | %endif
|
---|
199 |
|
---|
200 | mov xSP, xBP
|
---|
201 | pop xBP
|
---|
202 | BS3_HYBRID_RET
|
---|
203 | BS3_PROC_END_MODE bs3FpuState1_Save
|
---|
204 |
|
---|
205 |
|
---|
206 | ;;
|
---|
207 | ; Performs a MOVDQU write on the specified memory.
|
---|
208 | ;
|
---|
209 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_MovDQU_Write)(void *pvMmioReg);
|
---|
210 | ;
|
---|
211 | BS3_PROC_BEGIN_MODE bs3FpuState1_MovDQU_Write, BS3_PBC_NEAR
|
---|
212 | BS3_CALL_CONV_PROLOG 1
|
---|
213 | push xBP
|
---|
214 | mov xBP, xSP
|
---|
215 | push xBX
|
---|
216 | TONLY16 push ds
|
---|
217 |
|
---|
218 | ; Load the register pointer.
|
---|
219 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
220 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
221 |
|
---|
222 | ; Do read.
|
---|
223 | movdqu [xBX], xmm0
|
---|
224 |
|
---|
225 | TONLY16 pop ds
|
---|
226 | pop xBX
|
---|
227 | leave
|
---|
228 | BS3_CALL_CONV_EPILOG 1
|
---|
229 | BS3_HYBRID_RET
|
---|
230 | BS3_PROC_END_MODE bs3FpuState1_MovDQU_Write
|
---|
231 |
|
---|
232 |
|
---|
233 | ;;
|
---|
234 | ; Performs a MOVDQU write to the specified memory.
|
---|
235 | ;
|
---|
236 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_MovDQU_Read)(void *pvMmioReg);
|
---|
237 | ;
|
---|
238 | BS3_PROC_BEGIN_MODE bs3FpuState1_MovDQU_Read, BS3_PBC_NEAR
|
---|
239 | BS3_CALL_CONV_PROLOG 2
|
---|
240 | push xBP
|
---|
241 | mov xBP, xSP
|
---|
242 | push xBX
|
---|
243 | TONLY16 push ds
|
---|
244 | sub xSP, 20h
|
---|
245 | %if TMPL_BITS == 16
|
---|
246 | movdqu [xBP - xCB - xCB - 2 - 18h], xmm2
|
---|
247 | %else
|
---|
248 | movdqu [xSP], xmm2
|
---|
249 | %endif
|
---|
250 |
|
---|
251 | ; Load the register pointer.
|
---|
252 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
253 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
254 |
|
---|
255 |
|
---|
256 | ; Do read.
|
---|
257 | movdqu xmm2, [xBX]
|
---|
258 |
|
---|
259 | ; Save the result.
|
---|
260 | mov xBX, [xBP + xCB + cbCurRetAddr + sCB]
|
---|
261 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + sCB + 2]
|
---|
262 | movups [xBX], xmm2
|
---|
263 |
|
---|
264 | %if TMPL_BITS == 16
|
---|
265 | movdqu xmm2, [xBP - xCB - xCB - 2 - 18h]
|
---|
266 | %else
|
---|
267 | movdqu xmm2, [xSP]
|
---|
268 | %endif
|
---|
269 | add xSP, 20h
|
---|
270 | TONLY16 pop ds
|
---|
271 | pop xBX
|
---|
272 | mov xSP, xBP
|
---|
273 | pop xBP
|
---|
274 | BS3_CALL_CONV_EPILOG 2
|
---|
275 | BS3_HYBRID_RET
|
---|
276 | BS3_PROC_END_MODE bs3FpuState1_MovDQU_Read
|
---|
277 |
|
---|
278 |
|
---|
279 | ;;
|
---|
280 | ; Performs a MOVUPS write on the specified memory.
|
---|
281 | ;
|
---|
282 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_MovUPS_Write)(void *pvMmioReg);
|
---|
283 | ;
|
---|
284 | BS3_PROC_BEGIN_MODE bs3FpuState1_MovUPS_Write, BS3_PBC_NEAR
|
---|
285 | BS3_CALL_CONV_PROLOG 1
|
---|
286 | push xBP
|
---|
287 | mov xBP, xSP
|
---|
288 | push xBX
|
---|
289 | TONLY16 push ds
|
---|
290 |
|
---|
291 | ; Load the register pointer.
|
---|
292 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
293 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
294 |
|
---|
295 | ; Do read.
|
---|
296 | movups [xBX], xmm3
|
---|
297 |
|
---|
298 | TONLY16 pop ds
|
---|
299 | pop xBX
|
---|
300 | leave
|
---|
301 | BS3_CALL_CONV_EPILOG 1
|
---|
302 | BS3_HYBRID_RET
|
---|
303 | BS3_PROC_END_MODE bs3FpuState1_MovUPS_Write
|
---|
304 |
|
---|
305 |
|
---|
306 | ;;
|
---|
307 | ; Performs a MOVUPS write to the specified memory.
|
---|
308 | ;
|
---|
309 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_MovUPS_Read)(void *pvMmioReg, void *pvResult);
|
---|
310 | ;
|
---|
311 | BS3_PROC_BEGIN_MODE bs3FpuState1_MovUPS_Read, BS3_PBC_NEAR
|
---|
312 | BS3_CALL_CONV_PROLOG 2
|
---|
313 | push xBP
|
---|
314 | mov xBP, xSP
|
---|
315 | push xBX
|
---|
316 | TONLY16 push ds
|
---|
317 | sub xSP, 20h
|
---|
318 | %if TMPL_BITS == 16
|
---|
319 | movups [xBP - xCB - xCB - 2 - 18h], xmm1
|
---|
320 | %else
|
---|
321 | movups [xSP], xmm1
|
---|
322 | %endif
|
---|
323 |
|
---|
324 | ; Load the register pointer.
|
---|
325 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
326 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
327 |
|
---|
328 |
|
---|
329 | ; Do read.
|
---|
330 | movups xmm1, [xBX]
|
---|
331 |
|
---|
332 | ; Save the result.
|
---|
333 | mov xBX, [xBP + xCB + cbCurRetAddr + sCB]
|
---|
334 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + sCB + 2]
|
---|
335 | movups [xBX], xmm1
|
---|
336 |
|
---|
337 | %if TMPL_BITS == 16
|
---|
338 | movups xmm1, [xBP - xCB - xCB - 2 - 18h]
|
---|
339 | %else
|
---|
340 | movups xmm1, [xSP]
|
---|
341 | %endif
|
---|
342 | add xSP, 20h
|
---|
343 | TONLY16 pop ds
|
---|
344 | pop xBX
|
---|
345 | mov xSP, xBP
|
---|
346 | pop xBP
|
---|
347 | BS3_CALL_CONV_EPILOG 2
|
---|
348 | BS3_HYBRID_RET
|
---|
349 | BS3_PROC_END_MODE bs3FpuState1_MovUPS_Read
|
---|
350 |
|
---|
351 |
|
---|
352 | ;;
|
---|
353 | ; Performs a FNSTENV write on the specified memory.
|
---|
354 | ;
|
---|
355 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_FNStEnv)(void *pvMmioReg);
|
---|
356 | ;
|
---|
357 | BS3_PROC_BEGIN_MODE bs3FpuState1_FNStEnv, BS3_PBC_NEAR
|
---|
358 | BS3_CALL_CONV_PROLOG 1
|
---|
359 | push xBP
|
---|
360 | mov xBP, xSP
|
---|
361 | push xBX
|
---|
362 | TONLY16 push ds
|
---|
363 |
|
---|
364 | ; Load the register pointer.
|
---|
365 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
366 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
367 |
|
---|
368 | ; Just write.
|
---|
369 | fnstenv [xBX]
|
---|
370 |
|
---|
371 | TONLY16 pop ds
|
---|
372 | pop xBX
|
---|
373 | mov xSP, xBP
|
---|
374 | pop xBP
|
---|
375 | BS3_CALL_CONV_EPILOG 1
|
---|
376 | BS3_HYBRID_RET
|
---|
377 | BS3_PROC_END_MODE bs3FpuState1_FNStEnv
|
---|
378 |
|
---|
379 |
|
---|
380 | ;;
|
---|
381 | ; Performs a FMUL on the specified memory, after writing a 64-bit value to it first.
|
---|
382 | ;
|
---|
383 | ; BS3_DECL_NEAR(void) TMPL_NM(bs3FpuState1_FMul)(void *pvMmioReg, void *pvResultIgnored);
|
---|
384 | ;
|
---|
385 | BS3_PROC_BEGIN_MODE bs3FpuState1_FMul, BS3_PBC_NEAR
|
---|
386 | BS3_CALL_CONV_PROLOG 2
|
---|
387 | push xBP
|
---|
388 | mov xBP, xSP
|
---|
389 | push xBX
|
---|
390 | TONLY16 push ds
|
---|
391 |
|
---|
392 | ; Load the value we'll be multiplying with into register(s) while ds is DATA16.
|
---|
393 | mov sAX, [TMPL_DATA16_WRT(g_r64_One)]
|
---|
394 | TNOT64 mov edx, [4 + TMPL_DATA16_WRT(g_r64_One)]
|
---|
395 |
|
---|
396 | ; Load the register pointer.
|
---|
397 | mov xBX, [xBP + xCB + cbCurRetAddr]
|
---|
398 | TONLY16 mov ds, [xBP + xCB + cbCurRetAddr + 2]
|
---|
399 |
|
---|
400 | ; Just write.
|
---|
401 | mov [xBX], sAX
|
---|
402 | TNOT64 mov [xBX + 4], edx
|
---|
403 | call .do_it
|
---|
404 |
|
---|
405 | TONLY16 pop ds
|
---|
406 | pop xBX
|
---|
407 | mov xSP, xBP
|
---|
408 | pop xBP
|
---|
409 | BS3_CALL_CONV_EPILOG 2
|
---|
410 | BS3_HYBRID_RET
|
---|
411 | .do_it:
|
---|
412 | fmul qword [xBX]
|
---|
413 | ret
|
---|
414 | BS3_PROC_END_MODE bs3FpuState1_FMul
|
---|
415 |
|
---|
416 |
|
---|
417 | %include "bs3kit-template-footer.mac" ; reset environment
|
---|
418 |
|
---|