VirtualBox

source: vbox/trunk/src/VBox/VMM/include/IEMN8veRecompiler.h@ 104407

Last change on this file since 104407 was 104407, checked in by vboxsync, 7 months ago

VMM/IEM: Adjusted the TB exit statistics a bit more, adding a few new one, making more of the release stats that doesn't go into the TB, and organizing them to try avoid counting the same exit more than once. bugref:10376 bugref:10653

  • Property svn:eol-style set to native
  • Property svn:keywords set to Author Date Id Revision
File size: 125.5 KB
Line 
1/* $Id: IEMN8veRecompiler.h 104407 2024-04-23 23:16:04Z vboxsync $ */
2/** @file
3 * IEM - Interpreted Execution Manager - Native Recompiler Internals.
4 */
5
6/*
7 * Copyright (C) 2011-2023 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 * SPDX-License-Identifier: GPL-3.0-only
26 */
27
28#ifndef VMM_INCLUDED_SRC_include_IEMN8veRecompiler_h
29#define VMM_INCLUDED_SRC_include_IEMN8veRecompiler_h
30#ifndef RT_WITHOUT_PRAGMA_ONCE
31# pragma once
32#endif
33
34
35/** @defgroup grp_iem_n8ve_re Native Recompiler Internals.
36 * @ingroup grp_iem_int
37 * @{
38 */
39
40#include <iprt/assertcompile.h> /* for RT_IN_ASSEMBLER mode */
41
42/** @def IEMNATIVE_WITH_TB_DEBUG_INFO
43 * Enables generating internal debug info for better TB disassembly dumping. */
44#if defined(DEBUG) || defined(DOXYGEN_RUNNING)
45# define IEMNATIVE_WITH_TB_DEBUG_INFO
46#endif
47
48/** @def IEMNATIVE_WITH_LIVENESS_ANALYSIS
49 * Enables liveness analysis. */
50#if 1 || defined(DOXYGEN_RUNNING)
51# define IEMNATIVE_WITH_LIVENESS_ANALYSIS
52/*# define IEMLIVENESS_EXTENDED_LAYOUT*/
53#endif
54
55/** @def IEMNATIVE_WITH_EFLAGS_SKIPPING
56 * Enables skipping EFLAGS calculations/updating based on liveness info. */
57#if defined(IEMNATIVE_WITH_LIVENESS_ANALYSIS) || defined(DOXYGEN_RUNNING)
58# define IEMNATIVE_WITH_EFLAGS_SKIPPING
59#endif
60
61
62/** @def IEMNATIVE_STRICT_EFLAGS_SKIPPING
63 * Enables strict consistency checks around EFLAGS skipping.
64 * @note Only defined when IEMNATIVE_WITH_EFLAGS_SKIPPING is also defined. */
65#ifdef IEMNATIVE_WITH_EFLAGS_SKIPPING
66# ifdef VBOX_STRICT
67# define IEMNATIVE_STRICT_EFLAGS_SKIPPING
68# endif
69#elif defined(DOXYGEN_RUNNING)
70# define IEMNATIVE_STRICT_EFLAGS_SKIPPING
71#endif
72
73#ifdef VBOX_WITH_STATISTICS
74/** Always count instructions for now. */
75# define IEMNATIVE_WITH_INSTRUCTION_COUNTING
76#endif
77
78/** @def IEMNATIVE_WITH_RECOMPILER_PROLOGUE_SINGLETON
79 * Enables having only a single prologue for native TBs. */
80#if 1 || defined(DOXYGEN_RUNNING)
81# define IEMNATIVE_WITH_RECOMPILER_PROLOGUE_SINGLETON
82#endif
83
84
85/** @name Stack Frame Layout
86 *
87 * @{ */
88/** The size of the area for stack variables and spills and stuff.
89 * @note This limit is duplicated in the python script(s). We add 0x40 for
90 * alignment padding. */
91#define IEMNATIVE_FRAME_VAR_SIZE (0xc0 + 0x40)
92/** Number of 64-bit variable slots (0x100 / 8 = 32. */
93#define IEMNATIVE_FRAME_VAR_SLOTS (IEMNATIVE_FRAME_VAR_SIZE / 8)
94AssertCompile(IEMNATIVE_FRAME_VAR_SLOTS == 32);
95
96#ifdef RT_ARCH_AMD64
97/** An stack alignment adjustment (between non-volatile register pushes and
98 * the stack variable area, so the latter better aligned). */
99# define IEMNATIVE_FRAME_ALIGN_SIZE 8
100
101/** Number of stack arguments slots for calls made from the frame. */
102# ifdef RT_OS_WINDOWS
103# define IEMNATIVE_FRAME_STACK_ARG_COUNT 4
104# else
105# define IEMNATIVE_FRAME_STACK_ARG_COUNT 2
106# endif
107/** Number of any shadow arguments (spill area) for calls we make. */
108# ifdef RT_OS_WINDOWS
109# define IEMNATIVE_FRAME_SHADOW_ARG_COUNT 4
110# else
111# define IEMNATIVE_FRAME_SHADOW_ARG_COUNT 0
112# endif
113
114/** Frame pointer (RBP) relative offset of the last push. */
115# ifdef RT_OS_WINDOWS
116# define IEMNATIVE_FP_OFF_LAST_PUSH (7 * -8)
117# else
118# define IEMNATIVE_FP_OFF_LAST_PUSH (5 * -8)
119# endif
120/** Frame pointer (RBP) relative offset of the stack variable area (the lowest
121 * address for it). */
122# define IEMNATIVE_FP_OFF_STACK_VARS (IEMNATIVE_FP_OFF_LAST_PUSH - IEMNATIVE_FRAME_ALIGN_SIZE - IEMNATIVE_FRAME_VAR_SIZE)
123/** Frame pointer (RBP) relative offset of the first stack argument for calls. */
124# define IEMNATIVE_FP_OFF_STACK_ARG0 (IEMNATIVE_FP_OFF_STACK_VARS - IEMNATIVE_FRAME_STACK_ARG_COUNT * 8)
125/** Frame pointer (RBP) relative offset of the second stack argument for calls. */
126# define IEMNATIVE_FP_OFF_STACK_ARG1 (IEMNATIVE_FP_OFF_STACK_ARG0 + 8)
127# ifdef RT_OS_WINDOWS
128/** Frame pointer (RBP) relative offset of the third stack argument for calls. */
129# define IEMNATIVE_FP_OFF_STACK_ARG2 (IEMNATIVE_FP_OFF_STACK_ARG0 + 16)
130/** Frame pointer (RBP) relative offset of the fourth stack argument for calls. */
131# define IEMNATIVE_FP_OFF_STACK_ARG3 (IEMNATIVE_FP_OFF_STACK_ARG0 + 24)
132# endif
133
134# ifdef RT_OS_WINDOWS
135/** Frame pointer (RBP) relative offset of the first incoming shadow argument. */
136# define IEMNATIVE_FP_OFF_IN_SHADOW_ARG0 (16)
137/** Frame pointer (RBP) relative offset of the second incoming shadow argument. */
138# define IEMNATIVE_FP_OFF_IN_SHADOW_ARG1 (24)
139/** Frame pointer (RBP) relative offset of the third incoming shadow argument. */
140# define IEMNATIVE_FP_OFF_IN_SHADOW_ARG2 (32)
141/** Frame pointer (RBP) relative offset of the fourth incoming shadow argument. */
142# define IEMNATIVE_FP_OFF_IN_SHADOW_ARG3 (40)
143# endif
144
145#elif RT_ARCH_ARM64
146/** No alignment padding needed for arm64. */
147# define IEMNATIVE_FRAME_ALIGN_SIZE 0
148/** No stack argument slots, got 8 registers for arguments will suffice. */
149# define IEMNATIVE_FRAME_STACK_ARG_COUNT 0
150/** There are no argument spill area. */
151# define IEMNATIVE_FRAME_SHADOW_ARG_COUNT 0
152
153/** Number of saved registers at the top of our stack frame.
154 * This includes the return address and old frame pointer, so x19 thru x30. */
155# define IEMNATIVE_FRAME_SAVE_REG_COUNT (12)
156/** The size of the save registered (IEMNATIVE_FRAME_SAVE_REG_COUNT). */
157# define IEMNATIVE_FRAME_SAVE_REG_SIZE (IEMNATIVE_FRAME_SAVE_REG_COUNT * 8)
158
159/** Frame pointer (BP) relative offset of the last push. */
160# define IEMNATIVE_FP_OFF_LAST_PUSH (10 * -8)
161
162/** Frame pointer (BP) relative offset of the stack variable area (the lowest
163 * address for it). */
164# define IEMNATIVE_FP_OFF_STACK_VARS (IEMNATIVE_FP_OFF_LAST_PUSH - IEMNATIVE_FRAME_ALIGN_SIZE - IEMNATIVE_FRAME_VAR_SIZE)
165
166#else
167# error "port me"
168#endif
169/** @} */
170
171
172/** @name Fixed Register Allocation(s)
173 * @{ */
174/** @def IEMNATIVE_REG_FIXED_PVMCPU
175 * The number of the register holding the pVCpu pointer. */
176/** @def IEMNATIVE_REG_FIXED_PCPUMCTX
177 * The number of the register holding the &pVCpu->cpum.GstCtx pointer.
178 * @note This not available on AMD64, only ARM64. */
179/** @def IEMNATIVE_REG_FIXED_TMP0
180 * Dedicated temporary register.
181 * @todo replace this by a register allocator and content tracker. */
182/** @def IEMNATIVE_REG_FIXED_MASK
183 * Mask GPRs with fixes assignments, either by us or dictated by the CPU/OS
184 * architecture. */
185#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
186/** @def IEMNATIVE_SIMD_REG_FIXED_TMP0
187 * Mask SIMD registers with fixes assignments, either by us or dictated by the CPU/OS
188 * architecture. */
189/** @def IEMNATIVE_SIMD_REG_FIXED_TMP0
190 * Dedicated temporary SIMD register. */
191#endif
192#if defined(RT_ARCH_ARM64) || defined(DOXYGEN_RUNNING) /* arm64 goes first because of doxygen */
193# define IEMNATIVE_REG_FIXED_PVMCPU ARMV8_A64_REG_X28
194# define IEMNATIVE_REG_FIXED_PVMCPU_ASM RT_CONCAT(x,IEMNATIVE_REG_FIXED_PVMCPU)
195# define IEMNATIVE_REG_FIXED_PCPUMCTX ARMV8_A64_REG_X27
196# define IEMNATIVE_REG_FIXED_PCPUMCTX_ASM RT_CONCAT(x,IEMNATIVE_REG_FIXED_PCPUMCTX)
197# define IEMNATIVE_REG_FIXED_TMP0 ARMV8_A64_REG_X15
198# if defined(IEMNATIVE_WITH_DELAYED_PC_UPDATING) && 0 /* debug the updating with a shadow RIP. */
199# define IEMNATIVE_REG_FIXED_TMP1 ARMV8_A64_REG_X16
200# define IEMNATIVE_REG_FIXED_PC_DBG ARMV8_A64_REG_X26
201# define IEMNATIVE_REG_FIXED_MASK_ADD ( RT_BIT_32(IEMNATIVE_REG_FIXED_TMP1) \
202 | RT_BIT_32(IEMNATIVE_REG_FIXED_PC_DBG))
203# else
204# define IEMNATIVE_REG_FIXED_MASK_ADD 0
205# endif
206# define IEMNATIVE_REG_FIXED_MASK ( RT_BIT_32(ARMV8_A64_REG_SP) \
207 | RT_BIT_32(ARMV8_A64_REG_LR) \
208 | RT_BIT_32(ARMV8_A64_REG_BP) \
209 | RT_BIT_32(IEMNATIVE_REG_FIXED_PVMCPU) \
210 | RT_BIT_32(IEMNATIVE_REG_FIXED_PCPUMCTX) \
211 | RT_BIT_32(ARMV8_A64_REG_X18) \
212 | RT_BIT_32(IEMNATIVE_REG_FIXED_TMP0) \
213 | IEMNATIVE_REG_FIXED_MASK_ADD)
214
215# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
216# define IEMNATIVE_SIMD_REG_FIXED_TMP0 ARMV8_A64_REG_Q30
217# if defined(IEMNATIVE_WITH_SIMD_REG_ACCESS_ALL_REGISTERS)
218# define IEMNATIVE_SIMD_REG_FIXED_MASK RT_BIT_32(ARMV8_A64_REG_Q30)
219# else
220/** @note
221 * ARM64 has 32 registers, but they are only 128-bit wide. So, in order to
222 * support emulating 256-bit registers we pair two real registers statically to
223 * one virtual for now, leaving us with only 16 256-bit registers. We always
224 * pair v0 with v1, v2 with v3, etc. so we mark the higher register as fixed and
225 * the register allocator assumes that it will be always free when the lower is
226 * picked.
227 *
228 * Also ARM64 declares the low 64-bit of v8-v15 as callee saved, so we don't
229 * touch them in order to avoid having to save and restore them in the
230 * prologue/epilogue.
231 */
232# define IEMNATIVE_SIMD_REG_FIXED_MASK ( UINT32_C(0xff00) \
233 | RT_BIT_32(ARMV8_A64_REG_Q31) \
234 | RT_BIT_32(ARMV8_A64_REG_Q30) \
235 | RT_BIT_32(ARMV8_A64_REG_Q29) \
236 | RT_BIT_32(ARMV8_A64_REG_Q27) \
237 | RT_BIT_32(ARMV8_A64_REG_Q25) \
238 | RT_BIT_32(ARMV8_A64_REG_Q23) \
239 | RT_BIT_32(ARMV8_A64_REG_Q21) \
240 | RT_BIT_32(ARMV8_A64_REG_Q19) \
241 | RT_BIT_32(ARMV8_A64_REG_Q17) \
242 | RT_BIT_32(ARMV8_A64_REG_Q15) \
243 | RT_BIT_32(ARMV8_A64_REG_Q13) \
244 | RT_BIT_32(ARMV8_A64_REG_Q11) \
245 | RT_BIT_32(ARMV8_A64_REG_Q9) \
246 | RT_BIT_32(ARMV8_A64_REG_Q7) \
247 | RT_BIT_32(ARMV8_A64_REG_Q5) \
248 | RT_BIT_32(ARMV8_A64_REG_Q3) \
249 | RT_BIT_32(ARMV8_A64_REG_Q1))
250# endif
251# endif
252
253#elif defined(RT_ARCH_AMD64)
254# define IEMNATIVE_REG_FIXED_PVMCPU X86_GREG_xBX
255# define IEMNATIVE_REG_FIXED_PVMCPU_ASM xBX
256# define IEMNATIVE_REG_FIXED_TMP0 X86_GREG_x11
257# define IEMNATIVE_REG_FIXED_MASK ( RT_BIT_32(IEMNATIVE_REG_FIXED_PVMCPU) \
258 | RT_BIT_32(IEMNATIVE_REG_FIXED_TMP0) \
259 | RT_BIT_32(X86_GREG_xSP) \
260 | RT_BIT_32(X86_GREG_xBP) )
261
262# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
263# define IEMNATIVE_SIMD_REG_FIXED_TMP0 5 /* xmm5/ymm5 */
264# ifndef IEMNATIVE_WITH_SIMD_REG_ACCESS_ALL_REGISTERS
265# ifndef _MSC_VER
266# define IEMNATIVE_WITH_SIMD_REG_ACCESS_ALL_REGISTERS
267# endif
268# endif
269# ifdef IEMNATIVE_WITH_SIMD_REG_ACCESS_ALL_REGISTERS
270# define IEMNATIVE_SIMD_REG_FIXED_MASK (RT_BIT_32(IEMNATIVE_SIMD_REG_FIXED_TMP0))
271# else
272/** @note On Windows/AMD64 xmm6 through xmm15 are marked as callee saved. */
273# define IEMNATIVE_SIMD_REG_FIXED_MASK ( UINT32_C(0xffc0) \
274 | RT_BIT_32(IEMNATIVE_SIMD_REG_FIXED_TMP0))
275# endif
276# endif
277
278#else
279# error "port me"
280#endif
281/** @} */
282
283/** @name Call related registers.
284 * @{ */
285/** @def IEMNATIVE_CALL_RET_GREG
286 * The return value register. */
287/** @def IEMNATIVE_CALL_ARG_GREG_COUNT
288 * Number of arguments in registers. */
289/** @def IEMNATIVE_CALL_ARG0_GREG
290 * The general purpose register carrying argument \#0. */
291/** @def IEMNATIVE_CALL_ARG1_GREG
292 * The general purpose register carrying argument \#1. */
293/** @def IEMNATIVE_CALL_ARG2_GREG
294 * The general purpose register carrying argument \#2. */
295/** @def IEMNATIVE_CALL_ARG3_GREG
296 * The general purpose register carrying argument \#3. */
297/** @def IEMNATIVE_CALL_VOLATILE_GREG_MASK
298 * Mask of registers the callee will not save and may trash. */
299#ifdef RT_ARCH_AMD64
300# define IEMNATIVE_CALL_RET_GREG X86_GREG_xAX
301
302# ifdef RT_OS_WINDOWS
303# define IEMNATIVE_CALL_ARG_GREG_COUNT 4
304# define IEMNATIVE_CALL_ARG0_GREG X86_GREG_xCX
305# define IEMNATIVE_CALL_ARG1_GREG X86_GREG_xDX
306# define IEMNATIVE_CALL_ARG2_GREG X86_GREG_x8
307# define IEMNATIVE_CALL_ARG3_GREG X86_GREG_x9
308# define IEMNATIVE_CALL_ARGS_GREG_MASK ( RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \
309 | RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \
310 | RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG) \
311 | RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) )
312# define IEMNATIVE_CALL_VOLATILE_GREG_MASK ( RT_BIT_32(X86_GREG_xAX) \
313 | RT_BIT_32(X86_GREG_xCX) \
314 | RT_BIT_32(X86_GREG_xDX) \
315 | RT_BIT_32(X86_GREG_x8) \
316 | RT_BIT_32(X86_GREG_x9) \
317 | RT_BIT_32(X86_GREG_x10) \
318 | RT_BIT_32(X86_GREG_x11) )
319# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
320/* xmm0 - xmm5 are marked as volatile. */
321# define IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK (UINT32_C(0x3f))
322# endif
323
324# else /* !RT_OS_WINDOWS */
325# define IEMNATIVE_CALL_ARG_GREG_COUNT 6
326# define IEMNATIVE_CALL_ARG0_GREG X86_GREG_xDI
327# define IEMNATIVE_CALL_ARG1_GREG X86_GREG_xSI
328# define IEMNATIVE_CALL_ARG2_GREG X86_GREG_xDX
329# define IEMNATIVE_CALL_ARG3_GREG X86_GREG_xCX
330# define IEMNATIVE_CALL_ARG4_GREG X86_GREG_x8
331# define IEMNATIVE_CALL_ARG5_GREG X86_GREG_x9
332# define IEMNATIVE_CALL_ARGS_GREG_MASK ( RT_BIT_32(IEMNATIVE_CALL_ARG0_GREG) \
333 | RT_BIT_32(IEMNATIVE_CALL_ARG1_GREG) \
334 | RT_BIT_32(IEMNATIVE_CALL_ARG2_GREG) \
335 | RT_BIT_32(IEMNATIVE_CALL_ARG3_GREG) \
336 | RT_BIT_32(IEMNATIVE_CALL_ARG4_GREG) \
337 | RT_BIT_32(IEMNATIVE_CALL_ARG5_GREG) )
338# define IEMNATIVE_CALL_VOLATILE_GREG_MASK ( RT_BIT_32(X86_GREG_xAX) \
339 | RT_BIT_32(X86_GREG_xCX) \
340 | RT_BIT_32(X86_GREG_xDX) \
341 | RT_BIT_32(X86_GREG_xDI) \
342 | RT_BIT_32(X86_GREG_xSI) \
343 | RT_BIT_32(X86_GREG_x8) \
344 | RT_BIT_32(X86_GREG_x9) \
345 | RT_BIT_32(X86_GREG_x10) \
346 | RT_BIT_32(X86_GREG_x11) )
347# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
348/* xmm0 - xmm15 are marked as volatile. */
349# define IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK (UINT32_C(0xffff))
350# endif
351# endif /* !RT_OS_WINDOWS */
352
353#elif defined(RT_ARCH_ARM64)
354# define IEMNATIVE_CALL_RET_GREG ARMV8_A64_REG_X0
355# define IEMNATIVE_CALL_ARG_GREG_COUNT 8
356# define IEMNATIVE_CALL_ARG0_GREG ARMV8_A64_REG_X0
357# define IEMNATIVE_CALL_ARG1_GREG ARMV8_A64_REG_X1
358# define IEMNATIVE_CALL_ARG2_GREG ARMV8_A64_REG_X2
359# define IEMNATIVE_CALL_ARG3_GREG ARMV8_A64_REG_X3
360# define IEMNATIVE_CALL_ARG4_GREG ARMV8_A64_REG_X4
361# define IEMNATIVE_CALL_ARG5_GREG ARMV8_A64_REG_X5
362# define IEMNATIVE_CALL_ARG6_GREG ARMV8_A64_REG_X6
363# define IEMNATIVE_CALL_ARG7_GREG ARMV8_A64_REG_X7
364# define IEMNATIVE_CALL_ARGS_GREG_MASK ( RT_BIT_32(ARMV8_A64_REG_X0) \
365 | RT_BIT_32(ARMV8_A64_REG_X1) \
366 | RT_BIT_32(ARMV8_A64_REG_X2) \
367 | RT_BIT_32(ARMV8_A64_REG_X3) \
368 | RT_BIT_32(ARMV8_A64_REG_X4) \
369 | RT_BIT_32(ARMV8_A64_REG_X5) \
370 | RT_BIT_32(ARMV8_A64_REG_X6) \
371 | RT_BIT_32(ARMV8_A64_REG_X7) )
372# define IEMNATIVE_CALL_VOLATILE_GREG_MASK ( RT_BIT_32(ARMV8_A64_REG_X0) \
373 | RT_BIT_32(ARMV8_A64_REG_X1) \
374 | RT_BIT_32(ARMV8_A64_REG_X2) \
375 | RT_BIT_32(ARMV8_A64_REG_X3) \
376 | RT_BIT_32(ARMV8_A64_REG_X4) \
377 | RT_BIT_32(ARMV8_A64_REG_X5) \
378 | RT_BIT_32(ARMV8_A64_REG_X6) \
379 | RT_BIT_32(ARMV8_A64_REG_X7) \
380 | RT_BIT_32(ARMV8_A64_REG_X8) \
381 | RT_BIT_32(ARMV8_A64_REG_X9) \
382 | RT_BIT_32(ARMV8_A64_REG_X10) \
383 | RT_BIT_32(ARMV8_A64_REG_X11) \
384 | RT_BIT_32(ARMV8_A64_REG_X12) \
385 | RT_BIT_32(ARMV8_A64_REG_X13) \
386 | RT_BIT_32(ARMV8_A64_REG_X14) \
387 | RT_BIT_32(ARMV8_A64_REG_X15) \
388 | RT_BIT_32(ARMV8_A64_REG_X16) \
389 | RT_BIT_32(ARMV8_A64_REG_X17) )
390# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
391/* The low 64 bits of v8 - v15 marked as callee saved but the rest is volatile,
392 * so to simplify our life a bit we just mark everything as volatile. */
393# define IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK (UINT32_C(0xffffffff))
394# endif
395
396#endif
397
398/** This is the maximum argument count we'll ever be needing. */
399#define IEMNATIVE_CALL_MAX_ARG_COUNT 7
400#ifdef RT_OS_WINDOWS
401# ifdef VBOXSTRICTRC_STRICT_ENABLED
402# undef IEMNATIVE_CALL_MAX_ARG_COUNT
403# define IEMNATIVE_CALL_MAX_ARG_COUNT 8
404# endif
405#endif
406/** @} */
407
408
409/** @def IEMNATIVE_HST_GREG_COUNT
410 * Number of host general purpose registers we tracker. */
411/** @def IEMNATIVE_HST_GREG_MASK
412 * Mask corresponding to IEMNATIVE_HST_GREG_COUNT that can be applied to
413 * inverted register masks and such to get down to a correct set of regs. */
414#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
415/** @def IEMNATIVE_HST_SIMD_REG_COUNT
416 * Number of host SIMD registers we track. */
417/** @def IEMNATIVE_HST_SIMD_REG_MASK
418 * Mask corresponding to IEMNATIVE_HST_SIMD_REG_COUNT that can be applied to
419 * inverted register masks and such to get down to a correct set of regs. */
420#endif
421#ifdef RT_ARCH_AMD64
422# define IEMNATIVE_HST_GREG_COUNT 16
423# define IEMNATIVE_HST_GREG_MASK UINT32_C(0xffff)
424
425# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
426# define IEMNATIVE_HST_SIMD_REG_COUNT 16
427# define IEMNATIVE_HST_SIMD_REG_MASK UINT32_C(0xffff)
428# endif
429
430#elif defined(RT_ARCH_ARM64)
431# define IEMNATIVE_HST_GREG_COUNT 32
432# define IEMNATIVE_HST_GREG_MASK UINT32_MAX
433
434# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
435# define IEMNATIVE_HST_SIMD_REG_COUNT 32
436# define IEMNATIVE_HST_SIMD_REG_MASK UINT32_MAX
437# endif
438
439#else
440# error "Port me!"
441#endif
442
443
444#ifndef RT_IN_ASSEMBLER /* ASM-NOINC-START - the rest of the file */
445
446
447/** Native code generator label types. */
448typedef enum
449{
450 kIemNativeLabelType_Invalid = 0,
451 /*
452 * Labels w/o data, only once instance per TB.
453 *
454 * Note! Jumps to these requires instructions that are capable of spanning
455 * the max TB length.
456 */
457 /* Simple labels comes first for indexing reasons. RaiseXx is order by the exception's numerical value(s). */
458 kIemNativeLabelType_RaiseDe, /**< Raise (throw) X86_XCPT_DE (00h). */
459 kIemNativeLabelType_RaiseUd, /**< Raise (throw) X86_XCPT_UD (06h). */
460 kIemNativeLabelType_RaiseSseRelated, /**< Raise (throw) X86_XCPT_UD or X86_XCPT_NM according to cr0 & cr4. */
461 kIemNativeLabelType_RaiseAvxRelated, /**< Raise (throw) X86_XCPT_UD or X86_XCPT_NM according to xcr0, cr0 & cr4. */
462 kIemNativeLabelType_RaiseSseAvxFpRelated, /**< Raise (throw) X86_XCPT_UD or X86_XCPT_XF according to c4. */
463 kIemNativeLabelType_RaiseNm, /**< Raise (throw) X86_XCPT_NM (07h). */
464 kIemNativeLabelType_RaiseGp0, /**< Raise (throw) X86_XCPT_GP (0dh) w/ errcd=0. */
465 kIemNativeLabelType_RaiseMf, /**< Raise (throw) X86_XCPT_MF (10h). */
466 kIemNativeLabelType_RaiseXf, /**< Raise (throw) X86_XCPT_XF (13h). */
467 kIemNativeLabelType_ObsoleteTb,
468 kIemNativeLabelType_NeedCsLimChecking,
469 kIemNativeLabelType_CheckBranchMiss,
470 kIemNativeLabelType_LastSimple = kIemNativeLabelType_CheckBranchMiss,
471 /* Manually defined labels. */
472 kIemNativeLabelType_Return,
473 kIemNativeLabelType_ReturnBreak,
474 kIemNativeLabelType_ReturnBreakFF,
475 kIemNativeLabelType_ReturnWithFlags,
476 kIemNativeLabelType_NonZeroRetOrPassUp,
477 /** The last fixup for branches that can span almost the whole TB length. */
478 kIemNativeLabelType_LastWholeTbBranch = kIemNativeLabelType_NonZeroRetOrPassUp,
479
480 /*
481 * Labels with data, potentially multiple instances per TB:
482 *
483 * These are localized labels, so no fixed jump type restrictions here.
484 */
485 kIemNativeLabelType_FirstWithMultipleInstances,
486 kIemNativeLabelType_If = kIemNativeLabelType_FirstWithMultipleInstances,
487 kIemNativeLabelType_Else,
488 kIemNativeLabelType_Endif,
489 kIemNativeLabelType_CheckIrq,
490 kIemNativeLabelType_TlbLookup,
491 kIemNativeLabelType_TlbMiss,
492 kIemNativeLabelType_TlbDone,
493 kIemNativeLabelType_End
494} IEMNATIVELABELTYPE;
495
496/** Native code generator label definition. */
497typedef struct IEMNATIVELABEL
498{
499 /** Code offset if defined, UINT32_MAX if it needs to be generated after/in
500 * the epilog. */
501 uint32_t off;
502 /** The type of label (IEMNATIVELABELTYPE). */
503 uint16_t enmType;
504 /** Additional label data, type specific. */
505 uint16_t uData;
506} IEMNATIVELABEL;
507/** Pointer to a label. */
508typedef IEMNATIVELABEL *PIEMNATIVELABEL;
509
510
511/** Native code generator fixup types. */
512typedef enum
513{
514 kIemNativeFixupType_Invalid = 0,
515#if defined(RT_ARCH_AMD64) || defined(RT_ARCH_X86)
516 /** AMD64 fixup: PC relative 32-bit with addend in bData. */
517 kIemNativeFixupType_Rel32,
518#elif defined(RT_ARCH_ARM64)
519 /** ARM64 fixup: PC relative offset at bits 25:0 (B, BL). */
520 kIemNativeFixupType_RelImm26At0,
521 /** ARM64 fixup: PC relative offset at bits 23:5 (CBZ, CBNZ, B.CC). */
522 kIemNativeFixupType_RelImm19At5,
523 /** ARM64 fixup: PC relative offset at bits 18:5 (TBZ, TBNZ). */
524 kIemNativeFixupType_RelImm14At5,
525#endif
526 kIemNativeFixupType_End
527} IEMNATIVEFIXUPTYPE;
528
529/** Native code generator fixup. */
530typedef struct IEMNATIVEFIXUP
531{
532 /** Code offset of the fixup location. */
533 uint32_t off;
534 /** The IEMNATIVELABEL this is a fixup for. */
535 uint16_t idxLabel;
536 /** The fixup type (IEMNATIVEFIXUPTYPE). */
537 uint8_t enmType;
538 /** Addend or other data. */
539 int8_t offAddend;
540} IEMNATIVEFIXUP;
541/** Pointer to a native code generator fixup. */
542typedef IEMNATIVEFIXUP *PIEMNATIVEFIXUP;
543
544
545/**
546 * One bit of the state.
547 *
548 * Each register state takes up two bits. We keep the two bits in two separate
549 * 64-bit words to simplify applying them to the guest shadow register mask in
550 * the register allocator.
551 */
552typedef union IEMLIVENESSBIT
553{
554 uint64_t bm64;
555 RT_GCC_EXTENSION struct
556 { /* bit no */
557 uint64_t bmGprs : 16; /**< 0x00 / 0: The 16 general purpose registers. */
558 uint64_t fUnusedPc : 1; /**< 0x10 / 16: (PC in ) */
559 uint64_t fCr0 : 1; /**< 0x11 / 17: */
560 uint64_t fFcw : 1; /**< 0x12 / 18: */
561 uint64_t fFsw : 1; /**< 0x13 / 19: */
562 uint64_t bmSegBase : 6; /**< 0x14 / 20: */
563 uint64_t bmSegAttrib : 6; /**< 0x1a / 26: */
564 uint64_t bmSegLimit : 6; /**< 0x20 / 32: */
565 uint64_t bmSegSel : 6; /**< 0x26 / 38: */
566 uint64_t fCr4 : 1; /**< 0x2c / 44: */
567 uint64_t fXcr0 : 1; /**< 0x2d / 45: */
568 uint64_t fMxCsr : 1; /**< 0x2e / 46: */
569 uint64_t fEflOther : 1; /**< 0x2f / 47: Other EFLAGS bits (~X86_EFL_STATUS_BITS & X86_EFL_LIVE_MASK). First! */
570 uint64_t fEflCf : 1; /**< 0x30 / 48: Carry flag (X86_EFL_CF / 0). */
571 uint64_t fEflPf : 1; /**< 0x31 / 49: Parity flag (X86_EFL_PF / 2). */
572 uint64_t fEflAf : 1; /**< 0x32 / 50: Auxilary carry flag (X86_EFL_AF / 4). */
573 uint64_t fEflZf : 1; /**< 0x33 / 51: Zero flag (X86_EFL_ZF / 6). */
574 uint64_t fEflSf : 1; /**< 0x34 / 52: Signed flag (X86_EFL_SF / 7). */
575 uint64_t fEflOf : 1; /**< 0x35 / 53: Overflow flag (X86_EFL_OF / 12). */
576 uint64_t uUnused : 10; /* 0x36 / 54 -> 0x40/64 */
577 };
578} IEMLIVENESSBIT;
579AssertCompileSize(IEMLIVENESSBIT, 8);
580
581#define IEMLIVENESSBIT_IDX_EFL_OTHER ((unsigned)kIemNativeGstReg_EFlags + 0)
582#define IEMLIVENESSBIT_IDX_EFL_CF ((unsigned)kIemNativeGstReg_EFlags + 1)
583#define IEMLIVENESSBIT_IDX_EFL_PF ((unsigned)kIemNativeGstReg_EFlags + 2)
584#define IEMLIVENESSBIT_IDX_EFL_AF ((unsigned)kIemNativeGstReg_EFlags + 3)
585#define IEMLIVENESSBIT_IDX_EFL_ZF ((unsigned)kIemNativeGstReg_EFlags + 4)
586#define IEMLIVENESSBIT_IDX_EFL_SF ((unsigned)kIemNativeGstReg_EFlags + 5)
587#define IEMLIVENESSBIT_IDX_EFL_OF ((unsigned)kIemNativeGstReg_EFlags + 6)
588
589
590/**
591 * A liveness state entry.
592 *
593 * The first 128 bits runs parallel to kIemNativeGstReg_xxx for the most part.
594 * Once we add a SSE register shadowing, we'll add another 64-bit element for
595 * that.
596 */
597typedef union IEMLIVENESSENTRY
598{
599#ifndef IEMLIVENESS_EXTENDED_LAYOUT
600 uint64_t bm64[16 / 8];
601 uint16_t bm32[16 / 4];
602 uint16_t bm16[16 / 2];
603 uint8_t bm8[ 16 / 1];
604 IEMLIVENESSBIT aBits[2];
605#else
606 uint64_t bm64[32 / 8];
607 uint16_t bm32[32 / 4];
608 uint16_t bm16[32 / 2];
609 uint8_t bm8[ 32 / 1];
610 IEMLIVENESSBIT aBits[4];
611#endif
612 RT_GCC_EXTENSION struct
613 {
614 /** Bit \#0 of the register states. */
615 IEMLIVENESSBIT Bit0;
616 /** Bit \#1 of the register states. */
617 IEMLIVENESSBIT Bit1;
618#ifdef IEMLIVENESS_EXTENDED_LAYOUT
619 /** Bit \#2 of the register states. */
620 IEMLIVENESSBIT Bit2;
621 /** Bit \#3 of the register states. */
622 IEMLIVENESSBIT Bit3;
623#endif
624 };
625} IEMLIVENESSENTRY;
626#ifndef IEMLIVENESS_EXTENDED_LAYOUT
627AssertCompileSize(IEMLIVENESSENTRY, 16);
628#else
629AssertCompileSize(IEMLIVENESSENTRY, 32);
630#endif
631/** Pointer to a liveness state entry. */
632typedef IEMLIVENESSENTRY *PIEMLIVENESSENTRY;
633/** Pointer to a const liveness state entry. */
634typedef IEMLIVENESSENTRY const *PCIEMLIVENESSENTRY;
635
636/** @name 64-bit value masks for IEMLIVENESSENTRY.
637 * @{ */ /* 0xzzzzyyyyxxxxwwww */
638#define IEMLIVENESSBIT_MASK UINT64_C(0x003ffffffffeffff)
639
640#ifndef IEMLIVENESS_EXTENDED_LAYOUT
641# define IEMLIVENESSBIT0_XCPT_OR_CALL UINT64_C(0x0000000000000000)
642# define IEMLIVENESSBIT1_XCPT_OR_CALL IEMLIVENESSBIT_MASK
643
644# define IEMLIVENESSBIT0_ALL_UNUSED IEMLIVENESSBIT_MASK
645# define IEMLIVENESSBIT1_ALL_UNUSED UINT64_C(0x0000000000000000)
646#endif
647
648#define IEMLIVENESSBIT_ALL_EFL_MASK UINT64_C(0x003f800000000000)
649#define IEMLIVENESSBIT_STATUS_EFL_MASK UINT64_C(0x003f000000000000)
650
651#ifndef IEMLIVENESS_EXTENDED_LAYOUT
652# define IEMLIVENESSBIT0_ALL_EFL_INPUT IEMLIVENESSBIT_ALL_EFL_MASK
653# define IEMLIVENESSBIT1_ALL_EFL_INPUT IEMLIVENESSBIT_ALL_EFL_MASK
654#endif
655/** @} */
656
657
658/** @name The liveness state for a register.
659 *
660 * The state values have been picked to with state accumulation in mind (what
661 * the iemNativeLivenessFunc_xxxx functions does), as that is the most
662 * performance critical work done with the values.
663 *
664 * This is a compressed state that only requires 2 bits per register.
665 * When accumulating state, we'll be using three IEMLIVENESSENTRY copies:
666 * 1. the incoming state from the following call,
667 * 2. the outgoing state for this call,
668 * 3. mask of the entries set in the 2nd.
669 *
670 * The mask entry (3rd one above) will be used both when updating the outgoing
671 * state and when merging in incoming state for registers not touched by the
672 * current call.
673 *
674 * @{ */
675#ifndef IEMLIVENESS_EXTENDED_LAYOUT
676/** The register will be clobbered and the current value thrown away.
677 *
678 * When this is applied to the state (2) we'll simply be AND'ing it with the
679 * (old) mask (3) and adding the register to the mask. This way we'll
680 * preserve the high priority IEMLIVENESS_STATE_XCPT_OR_CALL and
681 * IEMLIVENESS_STATE_INPUT states. */
682# define IEMLIVENESS_STATE_CLOBBERED 0
683/** The register is unused in the remainder of the TB.
684 *
685 * This is an initial state and can not be set by any of the
686 * iemNativeLivenessFunc_xxxx callbacks. */
687# define IEMLIVENESS_STATE_UNUSED 1
688/** The register value is required in a potential call or exception.
689 *
690 * This means that the register value must be calculated and is best written to
691 * the state, but that any shadowing registers can be flushed thereafter as it's
692 * not used again. This state has lower priority than IEMLIVENESS_STATE_INPUT.
693 *
694 * It is typically applied across the board, but we preserve incoming
695 * IEMLIVENESS_STATE_INPUT values. This latter means we have to do some extra
696 * trickery to filter out IEMLIVENESS_STATE_UNUSED:
697 * 1. r0 = old & ~mask;
698 * 2. r0 = t1 & (t1 >> 1)'
699 * 3. state |= r0 | 0b10;
700 * 4. mask = ~0;
701 */
702# define IEMLIVENESS_STATE_XCPT_OR_CALL 2
703/** The register value is used as input.
704 *
705 * This means that the register value must be calculated and it is best to keep
706 * it in a register. It does not need to be writtent out as such. This is the
707 * highest priority state.
708 *
709 * Whether the call modifies the register or not isn't relevant to earlier
710 * calls, so that's not recorded.
711 *
712 * When applying this state we just or in the value in the outgoing state and
713 * mask. */
714# define IEMLIVENESS_STATE_INPUT 3
715/** Mask of the state bits. */
716# define IEMLIVENESS_STATE_MASK 3
717/** The number of bits per state. */
718# define IEMLIVENESS_STATE_BIT_COUNT 2
719/** Check if we're expecting read & write accesses to a register with the given (previous) liveness state. */
720# define IEMLIVENESS_STATE_IS_MODIFY_EXPECTED(a_uState) ((uint32_t)((a_uState) - 1U) >= (uint32_t)(IEMLIVENESS_STATE_INPUT - 1U))
721/** Check if we're expecting read accesses to a register with the given (previous) liveness state. */
722# define IEMLIVENESS_STATE_IS_INPUT_EXPECTED(a_uState) IEMLIVENESS_STATE_IS_MODIFY_EXPECTED(a_uState)
723/** Check if a register clobbering is expected given the (previous) liveness state.
724 * The state must be either CLOBBERED or XCPT_OR_CALL, but it may also
725 * include INPUT if the register is used in more than one place. */
726# define IEMLIVENESS_STATE_IS_CLOBBER_EXPECTED(a_uState) ((uint32_t)(a_uState) != IEMLIVENESS_STATE_UNUSED)
727
728/** Check if all status flags are going to be clobbered and doesn't need
729 * calculating in the current step.
730 * @param a_pCurEntry The current liveness entry. */
731# define IEMLIVENESS_STATE_ARE_STATUS_EFL_TO_BE_CLOBBERED(a_pCurEntry) \
732 ( (((a_pCurEntry)->Bit0.bm64 | (a_pCurEntry)->Bit1.bm64) & IEMLIVENESSBIT_STATUS_EFL_MASK) == 0 )
733
734#else /* IEMLIVENESS_EXTENDED_LAYOUT */
735/** The register is not used any more. */
736# define IEMLIVENESS_STATE_UNUSED 0
737/** Flag: The register is required in a potential exception or call. */
738# define IEMLIVENESS_STATE_POT_XCPT_OR_CALL 1
739# define IEMLIVENESS_BIT_POT_XCPT_OR_CALL 0
740/** Flag: The register is read. */
741# define IEMLIVENESS_STATE_READ 2
742# define IEMLIVENESS_BIT_READ 1
743/** Flag: The register is written. */
744# define IEMLIVENESS_STATE_WRITE 4
745# define IEMLIVENESS_BIT_WRITE 2
746/** Flag: Unconditional call (not needed, can be redefined for research). */
747# define IEMLIVENESS_STATE_CALL 8
748# define IEMLIVENESS_BIT_CALL 3
749# define IEMLIVENESS_BIT_OTHER 3 /**< More convenient name for this one. */
750# define IEMLIVENESS_STATE_IS_MODIFY_EXPECTED(a_uState) \
751 ( ((a_uState) & (IEMLIVENESS_STATE_WRITE | IEMLIVENESS_STATE_READ)) == (IEMLIVENESS_STATE_WRITE | IEMLIVENESS_STATE_READ) )
752# define IEMLIVENESS_STATE_IS_INPUT_EXPECTED(a_uState) RT_BOOL((a_uState) & IEMLIVENESS_STATE_READ)
753# define IEMLIVENESS_STATE_IS_CLOBBER_EXPECTED(a_uState) RT_BOOL((a_uState) & IEMLIVENESS_STATE_WRITE)
754
755# define IEMLIVENESS_STATE_ARE_STATUS_EFL_TO_BE_CLOBBERED(a_pCurEntry) \
756 ( ((a_pCurEntry)->aBits[IEMLIVENESS_BIT_WRITE].bm64 & IEMLIVENESSBIT_STATUS_EFL_MASK) == IEMLIVENESSBIT_STATUS_EFL_MASK \
757 && !( ((a_pCurEntry)->aBits[IEMLIVENESS_BIT_READ].bm64 | (a_pCurEntry)->aBits[IEMLIVENESS_BIT_POT_XCPT_OR_CALL].bm64) \
758 & IEMLIVENESSBIT_STATUS_EFL_MASK) )
759
760#endif /* IEMLIVENESS_EXTENDED_LAYOUT */
761/** @} */
762
763/** @name Liveness helpers for builtin functions and similar.
764 *
765 * These are not used by IEM_MC_BEGIN/END blocks, IEMAllN8veLiveness.cpp has its
766 * own set of manimulator macros for those.
767 *
768 * @{ */
769/** Initializing the state as all unused. */
770#ifndef IEMLIVENESS_EXTENDED_LAYOUT
771# define IEM_LIVENESS_RAW_INIT_AS_UNUSED(a_pOutgoing) \
772 do { \
773 (a_pOutgoing)->Bit0.bm64 = IEMLIVENESSBIT0_ALL_UNUSED; \
774 (a_pOutgoing)->Bit1.bm64 = IEMLIVENESSBIT1_ALL_UNUSED; \
775 } while (0)
776#else
777# define IEM_LIVENESS_RAW_INIT_AS_UNUSED(a_pOutgoing) \
778 do { \
779 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_POT_XCPT_OR_CALL].bm64 = 0; \
780 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ ].bm64 = 0; \
781 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_WRITE ].bm64 = 0; \
782 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_OTHER ].bm64 = 0; \
783 } while (0)
784#endif
785
786/** Initializing the outgoing state with a potential xcpt or call state.
787 * This only works when all later changes will be IEMLIVENESS_STATE_INPUT. */
788#ifndef IEMLIVENESS_EXTENDED_LAYOUT
789# define IEM_LIVENESS_RAW_INIT_WITH_XCPT_OR_CALL(a_pOutgoing, a_pIncoming) \
790 do { \
791 (a_pOutgoing)->Bit0.bm64 = (a_pIncoming)->Bit0.bm64 & (a_pIncoming)->Bit1.bm64; \
792 (a_pOutgoing)->Bit1.bm64 = IEMLIVENESSBIT1_XCPT_OR_CALL; \
793 } while (0)
794#else
795# define IEM_LIVENESS_RAW_INIT_WITH_XCPT_OR_CALL(a_pOutgoing, a_pIncoming) \
796 do { \
797 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_POT_XCPT_OR_CALL].bm64 = IEMLIVENESSBIT_MASK; \
798 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ ].bm64 = (a_pIncoming)->aBits[IEMLIVENESS_BIT_READ].bm64; \
799 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_WRITE ].bm64 = 0; \
800 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_OTHER ].bm64 = 0; \
801 } while (0)
802#endif
803
804/** Adds a segment base register as input to the outgoing state. */
805#ifndef IEMLIVENESS_EXTENDED_LAYOUT
806# define IEM_LIVENESS_RAW_SEG_BASE_INPUT(a_pOutgoing, a_iSReg) do { \
807 (a_pOutgoing)->Bit0.bmSegBase |= RT_BIT_64(a_iSReg); \
808 (a_pOutgoing)->Bit1.bmSegBase |= RT_BIT_64(a_iSReg); \
809 } while (0)
810#else
811# define IEM_LIVENESS_RAW_SEG_BASE_INPUT(a_pOutgoing, a_iSReg) do { \
812 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ].bmSegBase |= RT_BIT_64(a_iSReg); \
813 } while (0)
814#endif
815
816/** Adds a segment attribute register as input to the outgoing state. */
817#ifndef IEMLIVENESS_EXTENDED_LAYOUT
818# define IEM_LIVENESS_RAW_SEG_ATTRIB_INPUT(a_pOutgoing, a_iSReg) do { \
819 (a_pOutgoing)->Bit0.bmSegAttrib |= RT_BIT_64(a_iSReg); \
820 (a_pOutgoing)->Bit1.bmSegAttrib |= RT_BIT_64(a_iSReg); \
821 } while (0)
822#else
823# define IEM_LIVENESS_RAW_SEG_ATTRIB_INPUT(a_pOutgoing, a_iSReg) do { \
824 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ].bmSegAttrib |= RT_BIT_64(a_iSReg); \
825 } while (0)
826#endif
827
828/** Adds a segment limit register as input to the outgoing state. */
829#ifndef IEMLIVENESS_EXTENDED_LAYOUT
830# define IEM_LIVENESS_RAW_SEG_LIMIT_INPUT(a_pOutgoing, a_iSReg) do { \
831 (a_pOutgoing)->Bit0.bmSegLimit |= RT_BIT_64(a_iSReg); \
832 (a_pOutgoing)->Bit1.bmSegLimit |= RT_BIT_64(a_iSReg); \
833 } while (0)
834#else
835# define IEM_LIVENESS_RAW_SEG_LIMIT_INPUT(a_pOutgoing, a_iSReg) do { \
836 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ].bmSegLimit |= RT_BIT_64(a_iSReg); \
837 } while (0)
838#endif
839
840/** Adds a segment limit register as input to the outgoing state. */
841#ifndef IEMLIVENESS_EXTENDED_LAYOUT
842# define IEM_LIVENESS_RAW_EFLAGS_ONE_INPUT(a_pOutgoing, a_fEflMember) do { \
843 (a_pOutgoing)->Bit0.a_fEflMember |= 1; \
844 (a_pOutgoing)->Bit1.a_fEflMember |= 1; \
845 } while (0)
846#else
847# define IEM_LIVENESS_RAW_EFLAGS_ONE_INPUT(a_pOutgoing, a_fEflMember) do { \
848 (a_pOutgoing)->aBits[IEMLIVENESS_BIT_READ].a_fEflMember |= 1; \
849 } while (0)
850#endif
851/** @} */
852
853/** @def IEMNATIVE_STRICT_EFLAGS_SKIPPING_EMIT_CHECK
854 * Checks that the EFLAGS bits specified by @a a_fEflNeeded are actually
855 * calculated and up to date. This is to double check that we haven't skipped
856 * EFLAGS calculations when we actually need them. NOP in non-strict builds.
857 * @note has to be placed in
858 */
859#ifdef IEMNATIVE_STRICT_EFLAGS_SKIPPING
860# define IEMNATIVE_STRICT_EFLAGS_SKIPPING_EMIT_CHECK(a_pReNative, a_off, a_fEflNeeded) \
861 do { (a_off) = iemNativeEmitEFlagsSkippingCheck(a_pReNative, a_off, a_fEflNeeded); } while (0)
862#else
863# define IEMNATIVE_STRICT_EFLAGS_SKIPPING_EMIT_CHECK(a_pReNative, a_off, a_fEflNeeded) do { } while (0)
864#endif
865
866
867/**
868 * Guest registers that can be shadowed in GPRs.
869 *
870 * This runs parallel to the liveness state (IEMLIVENESSBIT, ++). The EFlags
871 * must be placed last, as the liveness state tracks it as 7 subcomponents and
872 * we don't want to waste space here.
873 *
874 * @note Make sure to update IEMLIVENESSBIT, IEMLIVENESSBIT_ALL_EFL_MASK and
875 * friends as well as IEMAllN8veLiveness.cpp.
876 */
877typedef enum IEMNATIVEGSTREG : uint8_t
878{
879 kIemNativeGstReg_GprFirst = 0,
880 kIemNativeGstReg_GprLast = kIemNativeGstReg_GprFirst + 15,
881 kIemNativeGstReg_Pc,
882 kIemNativeGstReg_Cr0,
883 kIemNativeGstReg_FpuFcw,
884 kIemNativeGstReg_FpuFsw,
885 kIemNativeGstReg_SegBaseFirst,
886 kIemNativeGstReg_SegBaseLast = kIemNativeGstReg_SegBaseFirst + 5,
887 kIemNativeGstReg_SegAttribFirst,
888 kIemNativeGstReg_SegAttribLast = kIemNativeGstReg_SegAttribFirst + 5,
889 kIemNativeGstReg_SegLimitFirst,
890 kIemNativeGstReg_SegLimitLast = kIemNativeGstReg_SegLimitFirst + 5,
891 kIemNativeGstReg_SegSelFirst,
892 kIemNativeGstReg_SegSelLast = kIemNativeGstReg_SegSelFirst + 5,
893 kIemNativeGstReg_Cr4,
894 kIemNativeGstReg_Xcr0,
895 kIemNativeGstReg_MxCsr,
896 kIemNativeGstReg_EFlags, /**< 32-bit, includes internal flags - last! */
897 kIemNativeGstReg_End
898} IEMNATIVEGSTREG;
899AssertCompile((int)kIemNativeGstReg_SegLimitFirst == 32);
900AssertCompile((UINT64_C(0x7f) << kIemNativeGstReg_EFlags) == IEMLIVENESSBIT_ALL_EFL_MASK);
901
902/** @name Helpers for converting register numbers to IEMNATIVEGSTREG values.
903 * @{ */
904#define IEMNATIVEGSTREG_GPR(a_iGpr) ((IEMNATIVEGSTREG)(kIemNativeGstReg_GprFirst + (a_iGpr) ))
905#define IEMNATIVEGSTREG_SEG_SEL(a_iSegReg) ((IEMNATIVEGSTREG)(kIemNativeGstReg_SegSelFirst + (a_iSegReg) ))
906#define IEMNATIVEGSTREG_SEG_BASE(a_iSegReg) ((IEMNATIVEGSTREG)(kIemNativeGstReg_SegBaseFirst + (a_iSegReg) ))
907#define IEMNATIVEGSTREG_SEG_LIMIT(a_iSegReg) ((IEMNATIVEGSTREG)(kIemNativeGstReg_SegLimitFirst + (a_iSegReg) ))
908#define IEMNATIVEGSTREG_SEG_ATTRIB(a_iSegReg) ((IEMNATIVEGSTREG)(kIemNativeGstReg_SegAttribFirst + (a_iSegReg) ))
909/** @} */
910
911#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
912
913/**
914 * Guest registers that can be shadowed in host SIMD registers.
915 *
916 * @todo r=aeichner Liveness tracking
917 * @todo r=aeichner Given that we can only track xmm/ymm here does this actually make sense?
918 */
919typedef enum IEMNATIVEGSTSIMDREG : uint8_t
920{
921 kIemNativeGstSimdReg_SimdRegFirst = 0,
922 kIemNativeGstSimdReg_SimdRegLast = kIemNativeGstSimdReg_SimdRegFirst + 15,
923 kIemNativeGstSimdReg_End
924} IEMNATIVEGSTSIMDREG;
925
926/** @name Helpers for converting register numbers to IEMNATIVEGSTSIMDREG values.
927 * @{ */
928#define IEMNATIVEGSTSIMDREG_SIMD(a_iSimdReg) ((IEMNATIVEGSTSIMDREG)(kIemNativeGstSimdReg_SimdRegFirst + (a_iSimdReg)))
929/** @} */
930
931/**
932 * The Load/store size for a SIMD guest register.
933 */
934typedef enum IEMNATIVEGSTSIMDREGLDSTSZ : uint8_t
935{
936 /** Invalid size. */
937 kIemNativeGstSimdRegLdStSz_Invalid = 0,
938 /** Loads the low 128-bit of a guest SIMD register. */
939 kIemNativeGstSimdRegLdStSz_Low128,
940 /** Loads the high 128-bit of a guest SIMD register. */
941 kIemNativeGstSimdRegLdStSz_High128,
942 /** Loads the whole 256-bits of a guest SIMD register. */
943 kIemNativeGstSimdRegLdStSz_256,
944 /** End value. */
945 kIemNativeGstSimdRegLdStSz_End
946} IEMNATIVEGSTSIMDREGLDSTSZ;
947
948#endif /* IEMNATIVE_WITH_SIMD_REG_ALLOCATOR */
949
950/**
951 * Intended use statement for iemNativeRegAllocTmpForGuestReg().
952 */
953typedef enum IEMNATIVEGSTREGUSE
954{
955 /** The usage is read-only, the register holding the guest register
956 * shadow copy will not be modified by the caller. */
957 kIemNativeGstRegUse_ReadOnly = 0,
958 /** The caller will update the guest register (think: PC += cbInstr).
959 * The guest shadow copy will follow the returned register. */
960 kIemNativeGstRegUse_ForUpdate,
961 /** The call will put an entirely new value in the guest register, so
962 * if new register is allocate it will be returned uninitialized. */
963 kIemNativeGstRegUse_ForFullWrite,
964 /** The caller will use the guest register value as input in a calculation
965 * and the host register will be modified.
966 * This means that the returned host register will not be marked as a shadow
967 * copy of the guest register. */
968 kIemNativeGstRegUse_Calculation
969} IEMNATIVEGSTREGUSE;
970
971/**
972 * Guest registers (classes) that can be referenced.
973 */
974typedef enum IEMNATIVEGSTREGREF : uint8_t
975{
976 kIemNativeGstRegRef_Invalid = 0,
977 kIemNativeGstRegRef_Gpr,
978 kIemNativeGstRegRef_GprHighByte, /**< AH, CH, DH, BH*/
979 kIemNativeGstRegRef_EFlags,
980 kIemNativeGstRegRef_MxCsr,
981 kIemNativeGstRegRef_FpuReg,
982 kIemNativeGstRegRef_MReg,
983 kIemNativeGstRegRef_XReg,
984 kIemNativeGstRegRef_X87,
985 kIemNativeGstRegRef_XState,
986 //kIemNativeGstRegRef_YReg, - doesn't work.
987 kIemNativeGstRegRef_End
988} IEMNATIVEGSTREGREF;
989
990
991/** Variable kinds. */
992typedef enum IEMNATIVEVARKIND : uint8_t
993{
994 /** Customary invalid zero value. */
995 kIemNativeVarKind_Invalid = 0,
996 /** This is either in a register or on the stack. */
997 kIemNativeVarKind_Stack,
998 /** Immediate value - loaded into register when needed, or can live on the
999 * stack if referenced (in theory). */
1000 kIemNativeVarKind_Immediate,
1001 /** Variable reference - loaded into register when needed, never stack. */
1002 kIemNativeVarKind_VarRef,
1003 /** Guest register reference - loaded into register when needed, never stack. */
1004 kIemNativeVarKind_GstRegRef,
1005 /** End of valid values. */
1006 kIemNativeVarKind_End
1007} IEMNATIVEVARKIND;
1008
1009
1010/** Variable or argument. */
1011typedef struct IEMNATIVEVAR
1012{
1013 /** The kind of variable. */
1014 IEMNATIVEVARKIND enmKind;
1015 /** The variable size in bytes. */
1016 uint8_t cbVar;
1017 /** The first stack slot (uint64_t), except for immediate and references
1018 * where it usually is UINT8_MAX. This is allocated lazily, so if a variable
1019 * has a stack slot it has been initialized and has a value. Unused variables
1020 * has neither a stack slot nor a host register assignment. */
1021 uint8_t idxStackSlot;
1022 /** The host register allocated for the variable, UINT8_MAX if not. */
1023 uint8_t idxReg;
1024 /** The argument number if argument, UINT8_MAX if regular variable. */
1025 uint8_t uArgNo;
1026 /** If referenced, the index (unpacked) of the variable referencing this one,
1027 * otherwise UINT8_MAX. A referenced variable must only be placed on the stack
1028 * and must be either kIemNativeVarKind_Stack or kIemNativeVarKind_Immediate. */
1029 uint8_t idxReferrerVar;
1030 /** Guest register being shadowed here, kIemNativeGstReg_End(/UINT8_MAX) if not.
1031 * @todo not sure what this really is for... */
1032 IEMNATIVEGSTREG enmGstReg;
1033#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1034 /** Flag whether this variable is held in a SIMD register (only supported for 128-bit and 256-bit variables),
1035 * only valid when idxReg is not UINT8_MAX. */
1036 bool fSimdReg : 1;
1037 /** Set if the registered is currently used exclusively, false if the
1038 * variable is idle and the register can be grabbed. */
1039 bool fRegAcquired : 1;
1040#else
1041 /** Set if the registered is currently used exclusively, false if the
1042 * variable is idle and the register can be grabbed. */
1043 bool fRegAcquired;
1044#endif
1045
1046 union
1047 {
1048 /** kIemNativeVarKind_Immediate: The immediate value. */
1049 uint64_t uValue;
1050 /** kIemNativeVarKind_VarRef: The index (unpacked) of the variable being referenced. */
1051 uint8_t idxRefVar;
1052 /** kIemNativeVarKind_GstRegRef: The guest register being referrenced. */
1053 struct
1054 {
1055 /** The class of register. */
1056 IEMNATIVEGSTREGREF enmClass;
1057 /** Index within the class. */
1058 uint8_t idx;
1059 } GstRegRef;
1060 } u;
1061} IEMNATIVEVAR;
1062/** Pointer to a variable or argument. */
1063typedef IEMNATIVEVAR *PIEMNATIVEVAR;
1064/** Pointer to a const variable or argument. */
1065typedef IEMNATIVEVAR const *PCIEMNATIVEVAR;
1066
1067/** What is being kept in a host register. */
1068typedef enum IEMNATIVEWHAT : uint8_t
1069{
1070 /** The traditional invalid zero value. */
1071 kIemNativeWhat_Invalid = 0,
1072 /** Mapping a variable (IEMNATIVEHSTREG::idxVar). */
1073 kIemNativeWhat_Var,
1074 /** Temporary register, this is typically freed when a MC completes. */
1075 kIemNativeWhat_Tmp,
1076 /** Call argument w/o a variable mapping. This is free (via
1077 * IEMNATIVE_CALL_VOLATILE_GREG_MASK) after the call is emitted. */
1078 kIemNativeWhat_Arg,
1079 /** Return status code.
1080 * @todo not sure if we need this... */
1081 kIemNativeWhat_rc,
1082 /** The fixed pVCpu (PVMCPUCC) register.
1083 * @todo consider offsetting this on amd64 to use negative offsets to access
1084 * more members using 8-byte disp. */
1085 kIemNativeWhat_pVCpuFixed,
1086 /** The fixed pCtx (PCPUMCTX) register.
1087 * @todo consider offsetting this on amd64 to use negative offsets to access
1088 * more members using 8-byte disp. */
1089 kIemNativeWhat_pCtxFixed,
1090 /** Fixed temporary register. */
1091 kIemNativeWhat_FixedTmp,
1092#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
1093 /** Shadow RIP for the delayed RIP updating debugging. */
1094 kIemNativeWhat_PcShadow,
1095#endif
1096 /** Register reserved by the CPU or OS architecture. */
1097 kIemNativeWhat_FixedReserved,
1098 /** End of valid values. */
1099 kIemNativeWhat_End
1100} IEMNATIVEWHAT;
1101
1102/**
1103 * Host general register entry.
1104 *
1105 * The actual allocation status is kept in IEMRECOMPILERSTATE::bmHstRegs.
1106 *
1107 * @todo Track immediate values in host registers similarlly to how we track the
1108 * guest register shadow copies. For it to be real helpful, though,
1109 * we probably need to know which will be reused and put them into
1110 * non-volatile registers, otherwise it's going to be more or less
1111 * restricted to an instruction or two.
1112 */
1113typedef struct IEMNATIVEHSTREG
1114{
1115 /** Set of guest registers this one shadows.
1116 *
1117 * Using a bitmap here so we can designate the same host register as a copy
1118 * for more than one guest register. This is expected to be useful in
1119 * situations where one value is copied to several registers in a sequence.
1120 * If the mapping is 1:1, then we'd have to pick which side of a 'MOV SRC,DST'
1121 * sequence we'd want to let this register follow to be a copy of and there
1122 * will always be places where we'd be picking the wrong one.
1123 */
1124 uint64_t fGstRegShadows;
1125 /** What is being kept in this register. */
1126 IEMNATIVEWHAT enmWhat;
1127 /** Variable index (packed) if holding a variable, otherwise UINT8_MAX. */
1128 uint8_t idxVar;
1129 /** Stack slot assigned by iemNativeVarSaveVolatileRegsPreHlpCall and freed
1130 * by iemNativeVarRestoreVolatileRegsPostHlpCall. This is not valid outside
1131 * that scope. */
1132 uint8_t idxStackSlot;
1133 /** Alignment padding. */
1134 uint8_t abAlign[5];
1135} IEMNATIVEHSTREG;
1136
1137
1138#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1139/**
1140 * Host SIMD register entry - this tracks a virtual 256-bit register split into two 128-bit
1141 * halves, on architectures where there is no 256-bit register available this entry will track
1142 * two adjacent 128-bit host registers.
1143 *
1144 * The actual allocation status is kept in IEMRECOMPILERSTATE::bmHstSimdRegs.
1145 */
1146typedef struct IEMNATIVEHSTSIMDREG
1147{
1148 /** Set of guest registers this one shadows.
1149 *
1150 * Using a bitmap here so we can designate the same host register as a copy
1151 * for more than one guest register. This is expected to be useful in
1152 * situations where one value is copied to several registers in a sequence.
1153 * If the mapping is 1:1, then we'd have to pick which side of a 'MOV SRC,DST'
1154 * sequence we'd want to let this register follow to be a copy of and there
1155 * will always be places where we'd be picking the wrong one.
1156 */
1157 uint64_t fGstRegShadows;
1158 /** What is being kept in this register. */
1159 IEMNATIVEWHAT enmWhat;
1160 /** Variable index (packed) if holding a variable, otherwise UINT8_MAX. */
1161 uint8_t idxVar;
1162 /** Flag what is currently loaded, low 128-bits, high 128-bits or complete 256-bits. */
1163 IEMNATIVEGSTSIMDREGLDSTSZ enmLoaded;
1164 /** Alignment padding. */
1165 uint8_t abAlign[5];
1166} IEMNATIVEHSTSIMDREG;
1167#endif
1168
1169
1170/**
1171 * Core state for the native recompiler, that is, things that needs careful
1172 * handling when dealing with branches.
1173 */
1174typedef struct IEMNATIVECORESTATE
1175{
1176#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
1177 /** The current instruction offset in bytes from when the guest program counter
1178 * was updated last. Used for delaying the write to the guest context program counter
1179 * as long as possible. */
1180 uint32_t offPc;
1181 /** Number of instructions where we could skip the updating. */
1182 uint32_t cInstrPcUpdateSkipped;
1183#endif
1184 /** Allocation bitmap for aHstRegs. */
1185 uint32_t bmHstRegs;
1186
1187 /** Bitmap marking which host register contains guest register shadow copies.
1188 * This is used during register allocation to try preserve copies. */
1189 uint32_t bmHstRegsWithGstShadow;
1190 /** Bitmap marking valid entries in aidxGstRegShadows. */
1191 uint64_t bmGstRegShadows;
1192#ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
1193 /** Bitmap marking the shadowed guest register as dirty and needing writeback when flushing. */
1194 uint64_t bmGstRegShadowDirty;
1195#endif
1196
1197#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1198 /** Allocation bitmap for aHstSimdRegs. */
1199 uint32_t bmHstSimdRegs;
1200
1201 /** Bitmap marking which host SIMD register contains guest SIMD register shadow copies.
1202 * This is used during register allocation to try preserve copies. */
1203 uint32_t bmHstSimdRegsWithGstShadow;
1204 /** Bitmap marking valid entries in aidxSimdGstRegShadows. */
1205 uint64_t bmGstSimdRegShadows;
1206 /** Bitmap marking whether the low 128-bit of the shadowed guest register are dirty and need writeback. */
1207 uint64_t bmGstSimdRegShadowDirtyLo128;
1208 /** Bitmap marking whether the high 128-bit of the shadowed guest register are dirty and need writeback. */
1209 uint64_t bmGstSimdRegShadowDirtyHi128;
1210#endif
1211
1212 union
1213 {
1214 /** Index of variable (unpacked) arguments, UINT8_MAX if not valid. */
1215 uint8_t aidxArgVars[8];
1216 /** For more efficient resetting. */
1217 uint64_t u64ArgVars;
1218 };
1219
1220 /** Allocation bitmap for the stack. */
1221 uint32_t bmStack;
1222 /** Allocation bitmap for aVars. */
1223 uint32_t bmVars;
1224
1225 /** Maps a guest register to a host GPR (index by IEMNATIVEGSTREG).
1226 * Entries are only valid if the corresponding bit in bmGstRegShadows is set.
1227 * (A shadow copy of a guest register can only be held in a one host register,
1228 * there are no duplicate copies or ambiguities like that). */
1229 uint8_t aidxGstRegShadows[kIemNativeGstReg_End];
1230#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1231 /** Maps a guest SIMD register to a host SIMD register (index by IEMNATIVEGSTSIMDREG).
1232 * Entries are only valid if the corresponding bit in bmGstSimdRegShadows is set.
1233 * (A shadow copy of a guest register can only be held in a one host register,
1234 * there are no duplicate copies or ambiguities like that). */
1235 uint8_t aidxGstSimdRegShadows[kIemNativeGstSimdReg_End];
1236#endif
1237
1238 /** Host register allocation tracking. */
1239 IEMNATIVEHSTREG aHstRegs[IEMNATIVE_HST_GREG_COUNT];
1240#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1241 /** Host SIMD register allocation tracking. */
1242 IEMNATIVEHSTSIMDREG aHstSimdRegs[IEMNATIVE_HST_SIMD_REG_COUNT];
1243#endif
1244
1245 /** Variables and arguments. */
1246 IEMNATIVEVAR aVars[9];
1247} IEMNATIVECORESTATE;
1248/** Pointer to core state. */
1249typedef IEMNATIVECORESTATE *PIEMNATIVECORESTATE;
1250/** Pointer to const core state. */
1251typedef IEMNATIVECORESTATE const *PCIEMNATIVECORESTATE;
1252
1253/** @def IEMNATIVE_VAR_IDX_UNPACK
1254 * @returns Index into IEMNATIVECORESTATE::aVars.
1255 * @param a_idxVar Variable index w/ magic (in strict builds).
1256 */
1257/** @def IEMNATIVE_VAR_IDX_PACK
1258 * @returns Variable index w/ magic (in strict builds).
1259 * @param a_idxVar Index into IEMNATIVECORESTATE::aVars.
1260 */
1261#ifdef VBOX_STRICT
1262# define IEMNATIVE_VAR_IDX_UNPACK(a_idxVar) ((a_idxVar) & IEMNATIVE_VAR_IDX_MASK)
1263# define IEMNATIVE_VAR_IDX_PACK(a_idxVar) ((a_idxVar) | IEMNATIVE_VAR_IDX_MAGIC)
1264# define IEMNATIVE_VAR_IDX_MAGIC UINT8_C(0xd0)
1265# define IEMNATIVE_VAR_IDX_MAGIC_MASK UINT8_C(0xf0)
1266# define IEMNATIVE_VAR_IDX_MASK UINT8_C(0x0f)
1267#else
1268# define IEMNATIVE_VAR_IDX_UNPACK(a_idxVar) (a_idxVar)
1269# define IEMNATIVE_VAR_IDX_PACK(a_idxVar) (a_idxVar)
1270#endif
1271
1272
1273#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1274/** Clear the dirty state of the given guest SIMD register. */
1275# define IEMNATIVE_SIMD_REG_STATE_CLR_DIRTY(a_pReNative, a_iSimdReg) \
1276 do { \
1277 (a_pReNative)->Core.bmGstSimdRegShadowDirtyLo128 &= ~RT_BIT_64(a_iSimdReg); \
1278 (a_pReNative)->Core.bmGstSimdRegShadowDirtyHi128 &= ~RT_BIT_64(a_iSimdReg); \
1279 } while (0)
1280
1281/** Returns whether the low 128-bits of the given guest SIMD register are dirty. */
1282# define IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_LO_U128(a_pReNative, a_iSimdReg) \
1283 RT_BOOL((a_pReNative)->Core.bmGstSimdRegShadowDirtyLo128 & RT_BIT_64(a_iSimdReg))
1284/** Returns whether the high 128-bits of the given guest SIMD register are dirty. */
1285# define IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_HI_U128(a_pReNative, a_iSimdReg) \
1286 RT_BOOL((a_pReNative)->Core.bmGstSimdRegShadowDirtyHi128 & RT_BIT_64(a_iSimdReg))
1287/** Returns whether the given guest SIMD register is dirty. */
1288# define IEMNATIVE_SIMD_REG_STATE_IS_DIRTY_U256(a_pReNative, a_iSimdReg) \
1289 RT_BOOL(((a_pReNative)->Core.bmGstSimdRegShadowDirtyLo128 | (a_pReNative)->Core.bmGstSimdRegShadowDirtyHi128) & RT_BIT_64(a_iSimdReg))
1290
1291/** Set the low 128-bits of the given guest SIMD register to the dirty state. */
1292# define IEMNATIVE_SIMD_REG_STATE_SET_DIRTY_LO_U128(a_pReNative, a_iSimdReg) \
1293 ((a_pReNative)->Core.bmGstSimdRegShadowDirtyLo128 |= RT_BIT_64(a_iSimdReg))
1294/** Set the high 128-bits of the given guest SIMD register to the dirty state. */
1295# define IEMNATIVE_SIMD_REG_STATE_SET_DIRTY_HI_U128(a_pReNative, a_iSimdReg) \
1296 ((a_pReNative)->Core.bmGstSimdRegShadowDirtyHi128 |= RT_BIT_64(a_iSimdReg))
1297
1298/** Flag for indicating that IEM_MC_MAYBE_RAISE_DEVICE_NOT_AVAILABLE() has emitted code in the current TB. */
1299# define IEMNATIVE_SIMD_RAISE_XCPT_CHECKS_EMITTED_MAYBE_DEVICE_NOT_AVAILABLE RT_BIT_32(0)
1300 /** Flag for indicating that IEM_MC_MAYBE_RAISE_DEVICE_NOT_AVAILABLE() has emitted code in the current TB. */
1301# define IEMNATIVE_SIMD_RAISE_XCPT_CHECKS_EMITTED_MAYBE_WAIT_DEVICE_NOT_AVAILABLE RT_BIT_32(1)
1302/** Flag for indicating that IEM_MC_MAYBE_RAISE_SSE_RELATED_XCPT() has emitted code in the current TB. */
1303# define IEMNATIVE_SIMD_RAISE_XCPT_CHECKS_EMITTED_MAYBE_SSE RT_BIT_32(2)
1304/** Flag for indicating that IEM_MC_MAYBE_RAISE_AVX_RELATED_XCPT() has emitted code in the current TB. */
1305# define IEMNATIVE_SIMD_RAISE_XCPT_CHECKS_EMITTED_MAYBE_AVX RT_BIT_32(3)
1306#endif
1307
1308
1309/**
1310 * Conditional stack entry.
1311 */
1312typedef struct IEMNATIVECOND
1313{
1314 /** Set if we're in the "else" part, clear if we're in the "if" before it. */
1315 bool fInElse;
1316 /** The label for the IEM_MC_ELSE. */
1317 uint32_t idxLabelElse;
1318 /** The label for the IEM_MC_ENDIF. */
1319 uint32_t idxLabelEndIf;
1320 /** The initial state snapshot as the if-block starts executing. */
1321 IEMNATIVECORESTATE InitialState;
1322 /** The state snapshot at the end of the if-block. */
1323 IEMNATIVECORESTATE IfFinalState;
1324} IEMNATIVECOND;
1325/** Pointer to a condition stack entry. */
1326typedef IEMNATIVECOND *PIEMNATIVECOND;
1327
1328
1329/**
1330 * Native recompiler state.
1331 */
1332typedef struct IEMRECOMPILERSTATE
1333{
1334 /** Size of the buffer that pbNativeRecompileBufR3 points to in
1335 * IEMNATIVEINSTR units. */
1336 uint32_t cInstrBufAlloc;
1337#ifdef VBOX_STRICT
1338 /** Strict: How far the last iemNativeInstrBufEnsure() checked. */
1339 uint32_t offInstrBufChecked;
1340#else
1341 uint32_t uPadding1; /* We don't keep track of the size here... */
1342#endif
1343 /** Fixed temporary code buffer for native recompilation. */
1344 PIEMNATIVEINSTR pInstrBuf;
1345
1346 /** Bitmaps with the label types used. */
1347 uint64_t bmLabelTypes;
1348 /** Actual number of labels in paLabels. */
1349 uint32_t cLabels;
1350 /** Max number of entries allowed in paLabels before reallocating it. */
1351 uint32_t cLabelsAlloc;
1352 /** Labels defined while recompiling (referenced by fixups). */
1353 PIEMNATIVELABEL paLabels;
1354 /** Array with indexes of unique labels (uData always 0). */
1355 uint32_t aidxUniqueLabels[kIemNativeLabelType_FirstWithMultipleInstances];
1356
1357 /** Actual number of fixups paFixups. */
1358 uint32_t cFixups;
1359 /** Max number of entries allowed in paFixups before reallocating it. */
1360 uint32_t cFixupsAlloc;
1361 /** Buffer used by the recompiler for recording fixups when generating code. */
1362 PIEMNATIVEFIXUP paFixups;
1363
1364#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
1365 /** Number of debug info entries allocated for pDbgInfo. */
1366 uint32_t cDbgInfoAlloc;
1367 uint32_t uPadding;
1368 /** Debug info. */
1369 PIEMTBDBG pDbgInfo;
1370#endif
1371
1372#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
1373 /** The current call index (liveness array and threaded calls in TB). */
1374 uint32_t idxCurCall;
1375 /** Number of liveness entries allocated. */
1376 uint32_t cLivenessEntriesAlloc;
1377 /** Liveness entries for all the calls in the TB begin recompiled.
1378 * The entry for idxCurCall contains the info for what the next call will
1379 * require wrt registers. (Which means the last entry is the initial liveness
1380 * state.) */
1381 PIEMLIVENESSENTRY paLivenessEntries;
1382#endif
1383
1384 /** The translation block being recompiled. */
1385 PCIEMTB pTbOrg;
1386 /** The VMCPU structure of the EMT. */
1387 PVMCPUCC pVCpu;
1388
1389 /** Condition sequence number (for generating unique labels). */
1390 uint16_t uCondSeqNo;
1391 /** Check IRQ seqeunce number (for generating unique labels). */
1392 uint16_t uCheckIrqSeqNo;
1393 /** TLB load sequence number (for generating unique labels). */
1394 uint16_t uTlbSeqNo;
1395 /** The current condition stack depth (aCondStack). */
1396 uint8_t cCondDepth;
1397
1398 /** The argument count + hidden regs from the IEM_MC_BEGIN_EX statement. */
1399 uint8_t cArgsX;
1400 /** The IEM_CIMPL_F_XXX flags from the IEM_MC_BEGIN statement. */
1401 uint32_t fCImpl;
1402 /** The IEM_MC_F_XXX flags from the IEM_MC_BEGIN statement. */
1403 uint32_t fMc;
1404 /** The expected IEMCPU::fExec value for the current call/instruction. */
1405 uint32_t fExec;
1406#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1407 /** IEMNATIVE_SIMD_RAISE_XCPT_CHECKS_EMITTED_XXX flags for exception flags
1408 * we only emit once per TB (or when the cr0/cr4/xcr0 register changes).
1409 *
1410 * This is an optimization because these control registers can only be changed from
1411 * by calling a C helper we can catch. Should reduce the number of instructions in a TB
1412 * consisting of multiple SIMD instructions.
1413 */
1414 uint32_t fSimdRaiseXcptChecksEmitted;
1415#endif
1416
1417 /** Core state requiring care with branches. */
1418 IEMNATIVECORESTATE Core;
1419
1420 /** The condition nesting stack. */
1421 IEMNATIVECOND aCondStack[2];
1422
1423#ifndef IEM_WITH_THROW_CATCH
1424 /** Pointer to the setjmp/longjmp buffer if we're not using C++ exceptions
1425 * for recompilation error handling. */
1426 jmp_buf JmpBuf;
1427#endif
1428} IEMRECOMPILERSTATE;
1429/** Pointer to a native recompiler state. */
1430typedef IEMRECOMPILERSTATE *PIEMRECOMPILERSTATE;
1431
1432
1433/** @def IEMNATIVE_TRY_SETJMP
1434 * Wrapper around setjmp / try, hiding all the ugly differences.
1435 *
1436 * @note Use with extreme care as this is a fragile macro.
1437 * @param a_pReNative The native recompile state.
1438 * @param a_rcTarget The variable that should receive the status code in case
1439 * of a longjmp/throw.
1440 */
1441/** @def IEMNATIVE_CATCH_LONGJMP_BEGIN
1442 * Start wrapper for catch / setjmp-else.
1443 *
1444 * This will set up a scope.
1445 *
1446 * @note Use with extreme care as this is a fragile macro.
1447 * @param a_pReNative The native recompile state.
1448 * @param a_rcTarget The variable that should receive the status code in case
1449 * of a longjmp/throw.
1450 */
1451/** @def IEMNATIVE_CATCH_LONGJMP_END
1452 * End wrapper for catch / setjmp-else.
1453 *
1454 * This will close the scope set up by IEMNATIVE_CATCH_LONGJMP_BEGIN and clean
1455 * up the state.
1456 *
1457 * @note Use with extreme care as this is a fragile macro.
1458 * @param a_pReNative The native recompile state.
1459 */
1460/** @def IEMNATIVE_DO_LONGJMP
1461 *
1462 * Wrapper around longjmp / throw.
1463 *
1464 * @param a_pReNative The native recompile state.
1465 * @param a_rc The status code jump back with / throw.
1466 */
1467#ifdef IEM_WITH_THROW_CATCH
1468# define IEMNATIVE_TRY_SETJMP(a_pReNative, a_rcTarget) \
1469 a_rcTarget = VINF_SUCCESS; \
1470 try
1471# define IEMNATIVE_CATCH_LONGJMP_BEGIN(a_pReNative, a_rcTarget) \
1472 catch (int rcThrown) \
1473 { \
1474 a_rcTarget = rcThrown
1475# define IEMNATIVE_CATCH_LONGJMP_END(a_pReNative) \
1476 } \
1477 ((void)0)
1478# define IEMNATIVE_DO_LONGJMP(a_pReNative, a_rc) throw int(a_rc)
1479#else /* !IEM_WITH_THROW_CATCH */
1480# define IEMNATIVE_TRY_SETJMP(a_pReNative, a_rcTarget) \
1481 if ((a_rcTarget = setjmp((a_pReNative)->JmpBuf)) == 0)
1482# define IEMNATIVE_CATCH_LONGJMP_BEGIN(a_pReNative, a_rcTarget) \
1483 else \
1484 { \
1485 ((void)0)
1486# define IEMNATIVE_CATCH_LONGJMP_END(a_pReNative) \
1487 }
1488# define IEMNATIVE_DO_LONGJMP(a_pReNative, a_rc) longjmp((a_pReNative)->JmpBuf, (a_rc))
1489#endif /* !IEM_WITH_THROW_CATCH */
1490
1491
1492/**
1493 * Native recompiler worker for a threaded function.
1494 *
1495 * @returns New code buffer offset; throws VBox status code in case of a failure.
1496 * @param pReNative The native recompiler state.
1497 * @param off The current code buffer offset.
1498 * @param pCallEntry The threaded call entry.
1499 *
1500 * @note This may throw/longjmp VBox status codes (int) to abort compilation, so no RT_NOEXCEPT!
1501 */
1502typedef uint32_t (VBOXCALL FNIEMNATIVERECOMPFUNC)(PIEMRECOMPILERSTATE pReNative, uint32_t off, PCIEMTHRDEDCALLENTRY pCallEntry);
1503/** Pointer to a native recompiler worker for a threaded function. */
1504typedef FNIEMNATIVERECOMPFUNC *PFNIEMNATIVERECOMPFUNC;
1505
1506/** Defines a native recompiler worker for a threaded function.
1507 * @see FNIEMNATIVERECOMPFUNC */
1508#define IEM_DECL_IEMNATIVERECOMPFUNC_DEF(a_Name) \
1509 uint32_t VBOXCALL a_Name(PIEMRECOMPILERSTATE pReNative, uint32_t off, PCIEMTHRDEDCALLENTRY pCallEntry)
1510
1511/** Prototypes a native recompiler function for a threaded function.
1512 * @see FNIEMNATIVERECOMPFUNC */
1513#define IEM_DECL_IEMNATIVERECOMPFUNC_PROTO(a_Name) FNIEMNATIVERECOMPFUNC a_Name
1514
1515
1516/**
1517 * Native recompiler liveness analysis worker for a threaded function.
1518 *
1519 * @param pCallEntry The threaded call entry.
1520 * @param pIncoming The incoming liveness state entry.
1521 * @param pOutgoing The outgoing liveness state entry.
1522 */
1523typedef DECLCALLBACKTYPE(void, FNIEMNATIVELIVENESSFUNC, (PCIEMTHRDEDCALLENTRY pCallEntry,
1524 PCIEMLIVENESSENTRY pIncoming, PIEMLIVENESSENTRY pOutgoing));
1525/** Pointer to a native recompiler liveness analysis worker for a threaded function. */
1526typedef FNIEMNATIVELIVENESSFUNC *PFNIEMNATIVELIVENESSFUNC;
1527
1528/** Defines a native recompiler liveness analysis worker for a threaded function.
1529 * @see FNIEMNATIVELIVENESSFUNC */
1530#define IEM_DECL_IEMNATIVELIVENESSFUNC_DEF(a_Name) \
1531 DECLCALLBACK(void) a_Name(PCIEMTHRDEDCALLENTRY pCallEntry, PCIEMLIVENESSENTRY pIncoming, PIEMLIVENESSENTRY pOutgoing)
1532
1533/** Prototypes a native recompiler liveness analysis function for a threaded function.
1534 * @see FNIEMNATIVELIVENESSFUNC */
1535#define IEM_DECL_IEMNATIVELIVENESSFUNC_PROTO(a_Name) FNIEMNATIVELIVENESSFUNC a_Name
1536
1537
1538/** Define a native recompiler helper function, safe to call from the TB code. */
1539#define IEM_DECL_NATIVE_HLP_DEF(a_RetType, a_Name, a_ArgList) \
1540 DECL_HIDDEN_THROW(a_RetType) VBOXCALL a_Name a_ArgList
1541/** Prototype a native recompiler helper function, safe to call from the TB code. */
1542#define IEM_DECL_NATIVE_HLP_PROTO(a_RetType, a_Name, a_ArgList) \
1543 DECL_HIDDEN_THROW(a_RetType) VBOXCALL a_Name a_ArgList
1544/** Pointer typedef a native recompiler helper function, safe to call from the TB code. */
1545#define IEM_DECL_NATIVE_HLP_PTR(a_RetType, a_Name, a_ArgList) \
1546 a_RetType (VBOXCALL *a_Name) a_ArgList
1547
1548
1549#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
1550DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddNativeOffset(PIEMRECOMPILERSTATE pReNative, uint32_t off);
1551DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddGuestRegShadowing(PIEMRECOMPILERSTATE pReNative, IEMNATIVEGSTREG enmGstReg,
1552 uint8_t idxHstReg = UINT8_MAX, uint8_t idxHstRegPrev = UINT8_MAX);
1553# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1554DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddGuestSimdRegShadowing(PIEMRECOMPILERSTATE pReNative,
1555 IEMNATIVEGSTSIMDREG enmGstSimdReg,
1556 uint8_t idxHstSimdReg = UINT8_MAX,
1557 uint8_t idxHstSimdRegPrev = UINT8_MAX);
1558# endif
1559# if defined(IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK) || defined(IEMNATIVE_WITH_SIMD_REG_ALLOCATOR)
1560DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddGuestRegDirty(PIEMRECOMPILERSTATE pReNative, bool fSimdReg,
1561 uint8_t idxGstReg, uint8_t idxHstReg);
1562DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddGuestRegWriteback(PIEMRECOMPILERSTATE pReNative, bool fSimdReg,
1563 uint64_t fGstReg);
1564# endif
1565DECL_HIDDEN_THROW(void) iemNativeDbgInfoAddDelayedPcUpdate(PIEMRECOMPILERSTATE pReNative,
1566 uint32_t offPc, uint32_t cInstrSkipped);
1567#endif /* IEMNATIVE_WITH_TB_DEBUG_INFO */
1568
1569DECL_HIDDEN_THROW(uint32_t) iemNativeLabelCreate(PIEMRECOMPILERSTATE pReNative, IEMNATIVELABELTYPE enmType,
1570 uint32_t offWhere = UINT32_MAX, uint16_t uData = 0);
1571DECL_HIDDEN_THROW(void) iemNativeLabelDefine(PIEMRECOMPILERSTATE pReNative, uint32_t idxLabel, uint32_t offWhere);
1572DECL_HIDDEN_THROW(void) iemNativeAddFixup(PIEMRECOMPILERSTATE pReNative, uint32_t offWhere, uint32_t idxLabel,
1573 IEMNATIVEFIXUPTYPE enmType, int8_t offAddend = 0);
1574DECL_HIDDEN_THROW(PIEMNATIVEINSTR) iemNativeInstrBufEnsureSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t cInstrReq);
1575
1576DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmp(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, bool fPreferVolatile = true);
1577DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmpEx(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint32_t fRegMask,
1578 bool fPreferVolatile = true);
1579DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmpImm(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint64_t uImm,
1580 bool fPreferVolatile = true);
1581DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmpForGuestReg(PIEMRECOMPILERSTATE pReNative, uint32_t *poff,
1582 IEMNATIVEGSTREG enmGstReg,
1583 IEMNATIVEGSTREGUSE enmIntendedUse = kIemNativeGstRegUse_ReadOnly,
1584 bool fNoVolatileRegs = false, bool fSkipLivenessAssert = false);
1585DECL_HIDDEN_THROW(uint8_t) iemNativeRegAllocTmpForGuestRegIfAlreadyPresent(PIEMRECOMPILERSTATE pReNative, uint32_t *poff,
1586 IEMNATIVEGSTREG enmGstReg);
1587
1588DECL_HIDDEN_THROW(uint32_t) iemNativeRegAllocArgs(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs);
1589DECL_HIDDEN_THROW(uint8_t) iemNativeRegAssignRc(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg);
1590#if (defined(IPRT_INCLUDED_x86_h) && defined(RT_ARCH_AMD64)) || (defined(IPRT_INCLUDED_armv8_h) && defined(RT_ARCH_ARM64))
1591DECL_HIDDEN_THROW(uint32_t) iemNativeRegMoveOrSpillStackVar(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxVar,
1592 uint32_t fForbiddenRegs = IEMNATIVE_CALL_VOLATILE_GREG_MASK);
1593# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1594DECL_HIDDEN_THROW(uint32_t) iemNativeSimdRegMoveOrSpillStackVar(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxVar,
1595 uint32_t fForbiddenRegs = IEMNATIVE_CALL_VOLATILE_SIMD_REG_MASK);
1596# endif
1597#endif
1598DECLHIDDEN(void) iemNativeRegFree(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT;
1599DECLHIDDEN(void) iemNativeRegFreeTmp(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT;
1600DECLHIDDEN(void) iemNativeRegFreeTmpImm(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg) RT_NOEXCEPT;
1601DECLHIDDEN(void) iemNativeRegFreeVar(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, bool fFlushShadows) RT_NOEXCEPT;
1602#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1603DECLHIDDEN(void) iemNativeSimdRegFreeVar(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg, bool fFlushShadows) RT_NOEXCEPT;
1604# ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
1605DECL_HIDDEN_THROW(uint32_t) iemNativeSimdRegFlushDirtyGuestByHostSimdRegShadow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxHstReg);
1606# endif
1607#endif
1608DECLHIDDEN(void) iemNativeRegFreeAndFlushMask(PIEMRECOMPILERSTATE pReNative, uint32_t fHstRegMask) RT_NOEXCEPT;
1609DECL_HIDDEN_THROW(uint32_t) iemNativeRegMoveAndFreeAndFlushAtCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs,
1610 uint32_t fKeepVars = 0);
1611DECLHIDDEN(void) iemNativeRegFlushGuestShadows(PIEMRECOMPILERSTATE pReNative, uint64_t fGstRegs) RT_NOEXCEPT;
1612DECLHIDDEN(void) iemNativeRegFlushGuestShadowsByHostMask(PIEMRECOMPILERSTATE pReNative, uint32_t fHstRegs) RT_NOEXCEPT;
1613DECL_HIDDEN_THROW(uint32_t) iemNativeRegRestoreGuestShadowsInVolatileRegs(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1614 uint32_t fHstRegsActiveShadows);
1615#ifdef VBOX_STRICT
1616DECLHIDDEN(void) iemNativeRegAssertSanity(PIEMRECOMPILERSTATE pReNative);
1617#endif
1618DECL_HIDDEN_THROW(uint32_t) iemNativeRegFlushPendingWritesSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint64_t fGstShwExcept,
1619 uint64_t fGstSimdShwExcept);
1620#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
1621DECL_HIDDEN_THROW(uint32_t) iemNativeEmitPcWritebackSlow(PIEMRECOMPILERSTATE pReNative, uint32_t off);
1622#endif
1623#ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
1624DECL_HIDDEN_THROW(uint32_t) iemNativeRegFlushPendingWrite(PIEMRECOMPILERSTATE pReNative, uint32_t off, IEMNATIVEGSTREG enmGstReg);
1625DECL_HIDDEN_THROW(uint32_t) iemNativeRegFlushDirtyGuest(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint64_t fFlushGstReg = UINT64_MAX);
1626DECL_HIDDEN_THROW(uint32_t) iemNativeRegFlushDirtyGuestByHostRegShadow(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxHstReg);
1627#endif
1628
1629
1630#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1631DECL_HIDDEN_THROW(uint8_t) iemNativeSimdRegAllocTmp(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, bool fPreferVolatile = true);
1632DECL_HIDDEN_THROW(uint8_t) iemNativeSimdRegAllocTmpEx(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint32_t fRegMask,
1633 bool fPreferVolatile = true);
1634DECL_HIDDEN_THROW(uint8_t) iemNativeSimdRegAllocTmpForGuestSimdReg(PIEMRECOMPILERSTATE pReNative, uint32_t *poff,
1635 IEMNATIVEGSTSIMDREG enmGstSimdReg,
1636 IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz,
1637 IEMNATIVEGSTREGUSE enmIntendedUse = kIemNativeGstRegUse_ReadOnly,
1638 bool fNoVolatileRegs = false);
1639DECLHIDDEN(void) iemNativeSimdRegFreeTmp(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg) RT_NOEXCEPT;
1640DECLHIDDEN(void) iemNativeSimdRegFlushGuestShadows(PIEMRECOMPILERSTATE pReNative, uint64_t fGstSimdRegs) RT_NOEXCEPT;
1641DECL_HIDDEN_THROW(uint32_t) iemNativeSimdRegFlushPendingWrite(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1642 IEMNATIVEGSTSIMDREG enmGstSimdReg);
1643DECL_HIDDEN_THROW(uint32_t) iemNativeEmitLoadSimdRegWithGstShadowSimdReg(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1644 uint8_t idxHstSimdReg, IEMNATIVEGSTSIMDREG enmGstSimdReg,
1645 IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz);
1646#endif
1647
1648DECL_HIDDEN_THROW(uint8_t) iemNativeArgAlloc(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t cbType);
1649DECL_HIDDEN_THROW(uint8_t) iemNativeArgAllocConst(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t cbType, uint64_t uValue);
1650DECL_HIDDEN_THROW(uint8_t) iemNativeArgAllocLocalRef(PIEMRECOMPILERSTATE pReNative, uint8_t iArgNo, uint8_t idxOtherVar);
1651DECL_HIDDEN_THROW(uint8_t) iemNativeVarAlloc(PIEMRECOMPILERSTATE pReNative, uint8_t cbType);
1652DECL_HIDDEN_THROW(uint8_t) iemNativeVarAllocConst(PIEMRECOMPILERSTATE pReNative, uint8_t cbType, uint64_t uValue);
1653DECL_HIDDEN_THROW(uint8_t) iemNativeVarAllocAssign(PIEMRECOMPILERSTATE pReNative, uint32_t *poff, uint8_t cbType, uint8_t idxVarOther);
1654DECL_HIDDEN_THROW(void) iemNativeVarSetKindToStack(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar);
1655DECL_HIDDEN_THROW(void) iemNativeVarSetKindToConst(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint64_t uValue);
1656DECL_HIDDEN_THROW(void) iemNativeVarSetKindToGstRegRef(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar,
1657 IEMNATIVEGSTREGREF enmRegClass, uint8_t idxReg);
1658DECL_HIDDEN_THROW(uint8_t) iemNativeVarGetStackSlot(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar);
1659DECL_HIDDEN_THROW(uint8_t) iemNativeVarRegisterAcquire(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint32_t *poff,
1660 bool fInitialized = false, uint8_t idxRegPref = UINT8_MAX);
1661#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1662DECL_HIDDEN_THROW(uint8_t) iemNativeVarSimdRegisterAcquire(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar, uint32_t *poff,
1663 bool fInitialized = false, uint8_t idxRegPref = UINT8_MAX);
1664#endif
1665DECL_HIDDEN_THROW(uint8_t) iemNativeVarRegisterAcquireForGuestReg(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar,
1666 IEMNATIVEGSTREG enmGstReg, uint32_t *poff);
1667DECL_HIDDEN_THROW(uint32_t) iemNativeVarSaveVolatileRegsPreHlpCall(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1668 uint32_t fHstRegsNotToSave);
1669DECL_HIDDEN_THROW(uint32_t) iemNativeVarRestoreVolatileRegsPostHlpCall(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1670 uint32_t fHstRegsNotToSave);
1671DECLHIDDEN(void) iemNativeVarFreeOneWorker(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar);
1672DECLHIDDEN(void) iemNativeVarFreeAllSlow(PIEMRECOMPILERSTATE pReNative, uint32_t bmVars);
1673
1674DECL_HIDDEN_THROW(uint32_t) iemNativeEmitLoadGprWithGstShadowReg(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1675 uint8_t idxHstReg, IEMNATIVEGSTREG enmGstReg);
1676#ifdef VBOX_STRICT
1677DECL_HIDDEN_THROW(uint32_t) iemNativeEmitTop32BitsClearCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxReg);
1678DECL_HIDDEN_THROW(uint32_t) iemNativeEmitGuestRegValueCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxReg,
1679 IEMNATIVEGSTREG enmGstReg);
1680# ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1681DECL_HIDDEN_THROW(uint32_t) iemNativeEmitGuestSimdRegValueCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxSimdReg,
1682 IEMNATIVEGSTSIMDREG enmGstSimdReg,
1683 IEMNATIVEGSTSIMDREGLDSTSZ enmLoadSz);
1684# endif
1685DECL_HIDDEN_THROW(uint32_t) iemNativeEmitExecFlagsCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fExec);
1686#endif
1687#ifdef IEMNATIVE_STRICT_EFLAGS_SKIPPING
1688DECL_HIDDEN_THROW(uint32_t) iemNativeEmitEFlagsSkippingCheck(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t fEflNeeded);
1689#endif
1690DECL_HIDDEN_THROW(uint32_t) iemNativeEmitCheckCallRetAndPassUp(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxInstr);
1691DECL_HIDDEN_THROW(uint32_t) iemNativeEmitCallCommon(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t cArgs, uint8_t cHiddenArgs, bool fFlushPendingWrites = true);
1692DECL_HIDDEN_THROW(uint32_t) iemNativeEmitCImplCall(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxInstr,
1693 uint64_t fGstShwFlush, uintptr_t pfnCImpl, uint8_t cbInstr, uint8_t cAddParams,
1694 uint64_t uParam0, uint64_t uParam1, uint64_t uParam2);
1695DECL_HIDDEN_THROW(uint32_t) iemNativeEmitThreadedCall(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1696 PCIEMTHRDEDCALLENTRY pCallEntry);
1697DECL_HIDDEN_THROW(uint32_t) iemNativeEmitCheckGprCanonicalMaybeRaiseGp0(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1698 uint8_t idxAddrReg, uint8_t idxInstr);
1699DECL_HIDDEN_THROW(uint32_t) iemNativeEmitCheckGpr32AgainstCsSegLimitMaybeRaiseGp0(PIEMRECOMPILERSTATE pReNative, uint32_t off,
1700 uint8_t idxAddrReg, uint8_t idxInstr);
1701DECL_HIDDEN_THROW(uint32_t) iemNativeEmitLeaGprByGstRegRef(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint8_t idxGprDst,
1702 IEMNATIVEGSTREGREF enmClass, uint8_t idxRegInClass);
1703
1704
1705IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecStatusCodeFiddling,(PVMCPUCC pVCpu, int rc, uint8_t idxInstr));
1706IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseGp0,(PVMCPUCC pVCpu));
1707IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseNm,(PVMCPUCC pVCpu));
1708IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseUd,(PVMCPUCC pVCpu));
1709IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseMf,(PVMCPUCC pVCpu));
1710IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseXf,(PVMCPUCC pVCpu));
1711IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpExecRaiseDe,(PVMCPUCC pVCpu));
1712IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpObsoleteTb,(PVMCPUCC pVCpu));
1713IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpNeedCsLimChecking,(PVMCPUCC pVCpu));
1714IEM_DECL_NATIVE_HLP_PROTO(int, iemNativeHlpCheckBranchMiss,(PVMCPUCC pVCpu));
1715
1716IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1717IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1718IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1719IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU8_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1720IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1721IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU16_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1722IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU16_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1723IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1724IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU32_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1725IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFetchDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg));
1726#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1727IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFetchDataU128,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PRTUINT128U pu128Dst));
1728IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFetchDataU128AlignedSse,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PRTUINT128U pu128Dst));
1729IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFetchDataU128NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PRTUINT128U pu128Dst));
1730IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFetchDataU256NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PRTUINT256U pu256Dst));
1731IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFetchDataU256AlignedAvx,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PRTUINT256U pu256Dst));
1732#endif
1733IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint8_t u8Value));
1734IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint16_t u16Value));
1735IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint32_t u32Value));
1736IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, uint64_t u64Value));
1737#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1738IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU128AlignedSse,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PCRTUINT128U pu128Src));
1739IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU128NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PCRTUINT128U pu128Src));
1740IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU256NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PCRTUINT256U pu256Src));
1741IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemStoreDataU256AlignedAvx,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t iSegReg, PCRTUINT256U pu256Src));
1742#endif
1743IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackStoreU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value));
1744IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackStoreU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value));
1745IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackStoreU32SReg,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value));
1746IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackStoreU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value));
1747IEM_DECL_NATIVE_HLP_PROTO(uint16_t, iemNativeHlpStackFetchU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1748IEM_DECL_NATIVE_HLP_PROTO(uint32_t, iemNativeHlpStackFetchU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1749IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpStackFetchU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1750
1751IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1752IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1753IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1754IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU8_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1755IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1756IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU16_Sx_U32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1757IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU16_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1758IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1759IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU32_Sx_U64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1760IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpMemFlatFetchDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1761#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1762IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatFetchDataU128,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PRTUINT128U pu128Dst));
1763IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatFetchDataU128AlignedSse,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PRTUINT128U pu128Dst));
1764IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatFetchDataU128NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PRTUINT128U pu128Dst));
1765IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatFetchDataU256NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PRTUINT256U pu256Dst));
1766IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatFetchDataU256AlignedAvx,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PRTUINT256U pu256Dst));
1767#endif
1768IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU8,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint8_t u8Value));
1769IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value));
1770IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value));
1771IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value));
1772#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1773IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU128AlignedSse,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PCRTUINT128U pu128Src));
1774IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU128NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PCRTUINT128U pu128Src));
1775IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU256NoAc,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PCRTUINT256U pu256Src));
1776IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemFlatStoreDataU256AlignedAvx,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, PCRTUINT256U pu256Src));
1777#endif
1778IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackFlatStoreU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint16_t u16Value));
1779IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackFlatStoreU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value));
1780IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackFlatStoreU32SReg,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint32_t u32Value));
1781IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpStackFlatStoreU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem, uint64_t u64Value));
1782IEM_DECL_NATIVE_HLP_PROTO(uint16_t, iemNativeHlpStackFlatFetchU16,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1783IEM_DECL_NATIVE_HLP_PROTO(uint32_t, iemNativeHlpStackFlatFetchU32,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1784IEM_DECL_NATIVE_HLP_PROTO(uint64_t, iemNativeHlpStackFlatFetchU64,(PVMCPUCC pVCpu, RTGCPTR GCPtrMem));
1785
1786IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemMapDataU8Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1787IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemMapDataU8Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1788IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemMapDataU8Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1789IEM_DECL_NATIVE_HLP_PROTO(uint8_t const *, iemNativeHlpMemMapDataU8Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1790IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemMapDataU16Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1791IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemMapDataU16Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1792IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemMapDataU16Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1793IEM_DECL_NATIVE_HLP_PROTO(uint16_t const *, iemNativeHlpMemMapDataU16Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1794IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemMapDataU32Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1795IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemMapDataU32Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1796IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemMapDataU32Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1797IEM_DECL_NATIVE_HLP_PROTO(uint32_t const *, iemNativeHlpMemMapDataU32Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1798IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemMapDataU64Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1799IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemMapDataU64Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1800IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemMapDataU64Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1801IEM_DECL_NATIVE_HLP_PROTO(uint64_t const *, iemNativeHlpMemMapDataU64Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1802IEM_DECL_NATIVE_HLP_PROTO(RTFLOAT80U *, iemNativeHlpMemMapDataR80Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1803IEM_DECL_NATIVE_HLP_PROTO(RTPBCD80U *, iemNativeHlpMemMapDataD80Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1804IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemMapDataU128Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1805IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemMapDataU128Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1806IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemMapDataU128Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1807IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U const *, iemNativeHlpMemMapDataU128Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem, uint8_t iSegReg));
1808
1809IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemFlatMapDataU8Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1810IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemFlatMapDataU8Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1811IEM_DECL_NATIVE_HLP_PROTO(uint8_t *, iemNativeHlpMemFlatMapDataU8Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1812IEM_DECL_NATIVE_HLP_PROTO(uint8_t const *, iemNativeHlpMemFlatMapDataU8Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1813IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemFlatMapDataU16Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1814IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemFlatMapDataU16Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1815IEM_DECL_NATIVE_HLP_PROTO(uint16_t *, iemNativeHlpMemFlatMapDataU16Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1816IEM_DECL_NATIVE_HLP_PROTO(uint16_t const *, iemNativeHlpMemFlatMapDataU16Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1817IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemFlatMapDataU32Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1818IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemFlatMapDataU32Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1819IEM_DECL_NATIVE_HLP_PROTO(uint32_t *, iemNativeHlpMemFlatMapDataU32Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1820IEM_DECL_NATIVE_HLP_PROTO(uint32_t const *, iemNativeHlpMemFlatMapDataU32Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1821IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemFlatMapDataU64Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1822IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemFlatMapDataU64Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1823IEM_DECL_NATIVE_HLP_PROTO(uint64_t *, iemNativeHlpMemFlatMapDataU64Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1824IEM_DECL_NATIVE_HLP_PROTO(uint64_t const *, iemNativeHlpMemFlatMapDataU64Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1825IEM_DECL_NATIVE_HLP_PROTO(RTFLOAT80U *, iemNativeHlpMemFlatMapDataR80Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1826IEM_DECL_NATIVE_HLP_PROTO(RTPBCD80U *, iemNativeHlpMemFlatMapDataD80Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1827IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemFlatMapDataU128Atomic,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1828IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemFlatMapDataU128Rw,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1829IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U *, iemNativeHlpMemFlatMapDataU128Wo,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1830IEM_DECL_NATIVE_HLP_PROTO(RTUINT128U const *, iemNativeHlpMemFlatMapDataU128Ro,(PVMCPUCC pVCpu, uint8_t *pbUnmapInfo, RTGCPTR GCPtrMem));
1831
1832IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemCommitAndUnmapAtomic,(PVMCPUCC pVCpu, uint8_t bUnmapInfo));
1833IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemCommitAndUnmapRw,(PVMCPUCC pVCpu, uint8_t bUnmapInfo));
1834IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemCommitAndUnmapWo,(PVMCPUCC pVCpu, uint8_t bUnmapInfo));
1835IEM_DECL_NATIVE_HLP_PROTO(void, iemNativeHlpMemCommitAndUnmapRo,(PVMCPUCC pVCpu, uint8_t bUnmapInfo));
1836
1837
1838/**
1839 * Info about shadowed guest register values.
1840 * @see IEMNATIVEGSTREG
1841 */
1842typedef struct IEMANTIVEGSTREGINFO
1843{
1844 /** Offset in VMCPU. */
1845 uint32_t off;
1846 /** The field size. */
1847 uint8_t cb;
1848 /** Name (for logging). */
1849 const char *pszName;
1850} IEMANTIVEGSTREGINFO;
1851extern DECL_HIDDEN_DATA(IEMANTIVEGSTREGINFO const) g_aGstShadowInfo[];
1852extern DECL_HIDDEN_DATA(const char * const) g_apszIemNativeHstRegNames[];
1853extern DECL_HIDDEN_DATA(int32_t const) g_aoffIemNativeCallStackArgBpDisp[];
1854extern DECL_HIDDEN_DATA(uint32_t const) g_afIemNativeCallRegs[];
1855extern DECL_HIDDEN_DATA(uint8_t const) g_aidxIemNativeCallRegs[];
1856
1857
1858
1859/**
1860 * Ensures that there is sufficient space in the instruction output buffer.
1861 *
1862 * This will reallocate the buffer if needed and allowed.
1863 *
1864 * @note Always use IEMNATIVE_ASSERT_INSTR_BUF_ENSURE when done to check the
1865 * allocation size.
1866 *
1867 * @returns Pointer to the instruction output buffer on success; throws VBox
1868 * status code on failure, so no need to check it.
1869 * @param pReNative The native recompile state.
1870 * @param off Current instruction offset. Works safely for UINT32_MAX
1871 * as well.
1872 * @param cInstrReq Number of instruction about to be added. It's okay to
1873 * overestimate this a bit.
1874 */
1875DECL_FORCE_INLINE_THROW(PIEMNATIVEINSTR)
1876iemNativeInstrBufEnsure(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint32_t cInstrReq)
1877{
1878 uint64_t const offChecked = off + (uint64_t)cInstrReq; /** @todo may reconsider the need for UINT32_MAX safety... */
1879 if (RT_LIKELY(offChecked <= pReNative->cInstrBufAlloc))
1880 {
1881#ifdef VBOX_STRICT
1882 pReNative->offInstrBufChecked = offChecked;
1883#endif
1884 return pReNative->pInstrBuf;
1885 }
1886 return iemNativeInstrBufEnsureSlow(pReNative, off, cInstrReq);
1887}
1888
1889/**
1890 * Checks that we didn't exceed the space requested in the last
1891 * iemNativeInstrBufEnsure() call.
1892 */
1893#define IEMNATIVE_ASSERT_INSTR_BUF_ENSURE(a_pReNative, a_off) \
1894 AssertMsg((a_off) <= (a_pReNative)->offInstrBufChecked, \
1895 ("off=%#x offInstrBufChecked=%#x\n", (a_off), (a_pReNative)->offInstrBufChecked))
1896
1897/**
1898 * Checks that a variable index is valid.
1899 */
1900#ifdef IEMNATIVE_VAR_IDX_MAGIC
1901# define IEMNATIVE_ASSERT_VAR_IDX(a_pReNative, a_idxVar) \
1902 AssertMsg( ((a_idxVar) & IEMNATIVE_VAR_IDX_MAGIC_MASK) == IEMNATIVE_VAR_IDX_MAGIC \
1903 && (unsigned)IEMNATIVE_VAR_IDX_UNPACK(a_idxVar) < RT_ELEMENTS((a_pReNative)->Core.aVars) \
1904 && ((a_pReNative)->Core.bmVars & RT_BIT_32(IEMNATIVE_VAR_IDX_UNPACK(a_idxVar))), \
1905 ("%s=%#x\n", #a_idxVar, a_idxVar))
1906#else
1907# define IEMNATIVE_ASSERT_VAR_IDX(a_pReNative, a_idxVar) \
1908 AssertMsg( (unsigned)(a_idxVar) < RT_ELEMENTS((a_pReNative)->Core.aVars) \
1909 && ((a_pReNative)->Core.bmVars & RT_BIT_32(a_idxVar)), ("%s=%d\n", #a_idxVar, a_idxVar))
1910#endif
1911
1912/**
1913 * Checks that a variable index is valid and that the variable is assigned the
1914 * correct argument number.
1915 * This also adds a RT_NOREF of a_idxVar.
1916 */
1917#ifdef IEMNATIVE_VAR_IDX_MAGIC
1918# define IEMNATIVE_ASSERT_ARG_VAR_IDX(a_pReNative, a_idxVar, a_uArgNo) do { \
1919 RT_NOREF_PV(a_idxVar); \
1920 AssertMsg( ((a_idxVar) & IEMNATIVE_VAR_IDX_MAGIC_MASK) == IEMNATIVE_VAR_IDX_MAGIC \
1921 && (unsigned)IEMNATIVE_VAR_IDX_UNPACK(a_idxVar) < RT_ELEMENTS((a_pReNative)->Core.aVars) \
1922 && ((a_pReNative)->Core.bmVars & RT_BIT_32(IEMNATIVE_VAR_IDX_UNPACK(a_idxVar))) \
1923 && (a_pReNative)->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(a_idxVar)].uArgNo == (a_uArgNo), \
1924 ("%s=%d; uArgNo=%d, expected %u\n", #a_idxVar, a_idxVar, \
1925 (a_pReNative)->Core.aVars[RT_MIN(IEMNATIVE_VAR_IDX_UNPACK(a_idxVar), \
1926 RT_ELEMENTS((a_pReNative)->Core.aVars)) - 1].uArgNo, \
1927 a_uArgNo)); \
1928 } while (0)
1929#else
1930# define IEMNATIVE_ASSERT_ARG_VAR_IDX(a_pReNative, a_idxVar, a_uArgNo) do { \
1931 RT_NOREF_PV(a_idxVar); \
1932 AssertMsg( (unsigned)(a_idxVar) < RT_ELEMENTS((a_pReNative)->Core.aVars) \
1933 && ((a_pReNative)->Core.bmVars & RT_BIT_32(a_idxVar))\
1934 && (a_pReNative)->Core.aVars[a_idxVar].uArgNo == (a_uArgNo) \
1935 , ("%s=%d; uArgNo=%d, expected %u\n", #a_idxVar, a_idxVar, \
1936 (a_pReNative)->Core.aVars[RT_MIN(a_idxVar, RT_ELEMENTS((a_pReNative)->Core.aVars)) - 1].uArgNo, a_uArgNo)); \
1937 } while (0)
1938#endif
1939
1940
1941/**
1942 * Checks that a variable has the expected size.
1943 */
1944#define IEMNATIVE_ASSERT_VAR_SIZE(a_pReNative, a_idxVar, a_cbVar) \
1945 AssertMsg((a_pReNative)->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(a_idxVar)].cbVar == (a_cbVar), \
1946 ("%s=%#x: cbVar=%#x, expected %#x!\n", #a_idxVar, a_idxVar, \
1947 (a_pReNative)->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(a_idxVar)].cbVar == (a_cbVar)))
1948
1949
1950/**
1951 * Calculates the stack address of a variable as a [r]BP displacement value.
1952 */
1953DECL_FORCE_INLINE(int32_t)
1954iemNativeStackCalcBpDisp(uint8_t idxStackSlot)
1955{
1956 Assert(idxStackSlot < IEMNATIVE_FRAME_VAR_SLOTS);
1957 return idxStackSlot * sizeof(uint64_t) + IEMNATIVE_FP_OFF_STACK_VARS;
1958}
1959
1960
1961/**
1962 * Releases the variable's register.
1963 *
1964 * The register must have been previously acquired calling
1965 * iemNativeVarRegisterAcquire(), iemNativeVarRegisterAcquireForGuestReg() or
1966 * iemNativeVarRegisterSetAndAcquire().
1967 */
1968DECL_INLINE_THROW(void) iemNativeVarRegisterRelease(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
1969{
1970 IEMNATIVE_ASSERT_VAR_IDX(pReNative, idxVar);
1971 Assert(pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].fRegAcquired);
1972 pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].fRegAcquired = false;
1973}
1974
1975
1976#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
1977DECL_INLINE_THROW(void) iemNativeVarSimdRegisterRelease(PIEMRECOMPILERSTATE pReNative, uint8_t idxVar)
1978{
1979 Assert(pReNative->Core.aVars[IEMNATIVE_VAR_IDX_UNPACK(idxVar)].fSimdReg);
1980 iemNativeVarRegisterRelease(pReNative, idxVar);
1981}
1982#endif
1983
1984
1985/**
1986 * Converts IEM_CIMPL_F_XXX flags into a guest register shadow copy flush mask.
1987 *
1988 * @returns The flush mask.
1989 * @param fCImpl The IEM_CIMPL_F_XXX flags.
1990 * @param fGstShwFlush The starting flush mask.
1991 */
1992DECL_FORCE_INLINE(uint64_t) iemNativeCImplFlagsToGuestShadowFlushMask(uint32_t fCImpl, uint64_t fGstShwFlush)
1993{
1994 if (fCImpl & IEM_CIMPL_F_BRANCH_FAR)
1995 fGstShwFlush |= RT_BIT_64(kIemNativeGstReg_SegSelFirst + X86_SREG_CS)
1996 | RT_BIT_64(kIemNativeGstReg_SegBaseFirst + X86_SREG_CS)
1997 | RT_BIT_64(kIemNativeGstReg_SegLimitFirst + X86_SREG_CS);
1998 if (fCImpl & IEM_CIMPL_F_BRANCH_STACK_FAR)
1999 fGstShwFlush |= RT_BIT_64(kIemNativeGstReg_GprFirst + X86_GREG_xSP)
2000 | RT_BIT_64(kIemNativeGstReg_SegSelFirst + X86_SREG_SS)
2001 | RT_BIT_64(kIemNativeGstReg_SegBaseFirst + X86_SREG_SS)
2002 | RT_BIT_64(kIemNativeGstReg_SegLimitFirst + X86_SREG_SS);
2003 else if (fCImpl & IEM_CIMPL_F_BRANCH_STACK)
2004 fGstShwFlush |= RT_BIT_64(kIemNativeGstReg_GprFirst + X86_GREG_xSP);
2005 if (fCImpl & (IEM_CIMPL_F_RFLAGS | IEM_CIMPL_F_STATUS_FLAGS | IEM_CIMPL_F_INHIBIT_SHADOW))
2006 fGstShwFlush |= RT_BIT_64(kIemNativeGstReg_EFlags);
2007 return fGstShwFlush;
2008}
2009
2010
2011/** Number of hidden arguments for CIMPL calls.
2012 * @note We're sufferning from the usual VBOXSTRICTRC fun on Windows. */
2013#if defined(VBOXSTRICTRC_STRICT_ENABLED) && defined(RT_OS_WINDOWS) && defined(RT_ARCH_AMD64)
2014# define IEM_CIMPL_HIDDEN_ARGS 3
2015#else
2016# define IEM_CIMPL_HIDDEN_ARGS 2
2017#endif
2018
2019
2020#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
2021/** Number of hidden arguments for SSE_AIMPL calls. */
2022# define IEM_SSE_AIMPL_HIDDEN_ARGS 1
2023/** Number of hidden arguments for AVX_AIMPL calls. */
2024# define IEM_AVX_AIMPL_HIDDEN_ARGS 1
2025#endif
2026
2027
2028#ifdef IEMNATIVE_WITH_LIVENESS_ANALYSIS
2029
2030# ifndef IEMLIVENESS_EXTENDED_LAYOUT
2031/**
2032 * Helper for iemNativeLivenessGetStateByGstReg.
2033 *
2034 * @returns IEMLIVENESS_STATE_XXX
2035 * @param fMergedStateExp2 This is the RT_BIT_32() of each sub-state
2036 * ORed together.
2037 */
2038DECL_FORCE_INLINE(uint32_t)
2039iemNativeLivenessMergeExpandedEFlagsState(uint32_t fMergedStateExp2)
2040{
2041 /* INPUT trumps anything else. */
2042 if (fMergedStateExp2 & RT_BIT_32(IEMLIVENESS_STATE_INPUT))
2043 return IEMLIVENESS_STATE_INPUT;
2044
2045 /* CLOBBERED trumps XCPT_OR_CALL and UNUSED. */
2046 if (fMergedStateExp2 & RT_BIT_32(IEMLIVENESS_STATE_CLOBBERED))
2047 {
2048 /* If not all sub-fields are clobbered they must be considered INPUT. */
2049 if (fMergedStateExp2 & (RT_BIT_32(IEMLIVENESS_STATE_UNUSED) | RT_BIT_32(IEMLIVENESS_STATE_XCPT_OR_CALL)))
2050 return IEMLIVENESS_STATE_INPUT;
2051 return IEMLIVENESS_STATE_CLOBBERED;
2052 }
2053
2054 /* XCPT_OR_CALL trumps UNUSED. */
2055 if (fMergedStateExp2 & RT_BIT_32(IEMLIVENESS_STATE_XCPT_OR_CALL))
2056 return IEMLIVENESS_STATE_XCPT_OR_CALL;
2057
2058 return IEMLIVENESS_STATE_UNUSED;
2059}
2060# endif /* !IEMLIVENESS_EXTENDED_LAYOUT */
2061
2062
2063DECL_FORCE_INLINE(uint32_t)
2064iemNativeLivenessGetStateByGstRegEx(PCIEMLIVENESSENTRY pLivenessEntry, unsigned enmGstRegEx)
2065{
2066# ifndef IEMLIVENESS_EXTENDED_LAYOUT
2067 return ((pLivenessEntry->Bit0.bm64 >> enmGstRegEx) & 1)
2068 | (((pLivenessEntry->Bit1.bm64 >> enmGstRegEx) << 1) & 2);
2069# else
2070 return ( (pLivenessEntry->Bit0.bm64 >> enmGstRegEx) & 1)
2071 | (((pLivenessEntry->Bit1.bm64 >> enmGstRegEx) << 1) & 2)
2072 | (((pLivenessEntry->Bit2.bm64 >> enmGstRegEx) << 2) & 4)
2073 | (((pLivenessEntry->Bit3.bm64 >> enmGstRegEx) << 2) & 8);
2074# endif
2075}
2076
2077
2078DECL_FORCE_INLINE(uint32_t)
2079iemNativeLivenessGetStateByGstReg(PCIEMLIVENESSENTRY pLivenessEntry, IEMNATIVEGSTREG enmGstReg)
2080{
2081 uint32_t uRet = iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, (unsigned)enmGstReg);
2082 if (enmGstReg == kIemNativeGstReg_EFlags)
2083 {
2084 /* Merge the eflags states to one. */
2085# ifndef IEMLIVENESS_EXTENDED_LAYOUT
2086 uRet = RT_BIT_32(uRet);
2087 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflCf | (pLivenessEntry->Bit1.fEflCf << 1));
2088 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflPf | (pLivenessEntry->Bit1.fEflPf << 1));
2089 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflAf | (pLivenessEntry->Bit1.fEflAf << 1));
2090 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflZf | (pLivenessEntry->Bit1.fEflZf << 1));
2091 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflSf | (pLivenessEntry->Bit1.fEflSf << 1));
2092 uRet |= RT_BIT_32(pLivenessEntry->Bit0.fEflOf | (pLivenessEntry->Bit1.fEflOf << 1));
2093 uRet = iemNativeLivenessMergeExpandedEFlagsState(uRet);
2094# else
2095 AssertCompile(IEMLIVENESSBIT_IDX_EFL_OTHER == (unsigned)kIemNativeGstReg_EFlags);
2096 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_CF);
2097 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_PF);
2098 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_AF);
2099 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_ZF);
2100 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_SF);
2101 uRet |= iemNativeLivenessGetStateByGstRegEx(pLivenessEntry, IEMLIVENESSBIT_IDX_EFL_OF);
2102# endif
2103 }
2104 return uRet;
2105}
2106
2107
2108# ifdef VBOX_STRICT
2109/** For assertions only, user checks that idxCurCall isn't zerow. */
2110DECL_FORCE_INLINE(uint32_t)
2111iemNativeLivenessGetPrevStateByGstReg(PIEMRECOMPILERSTATE pReNative, IEMNATIVEGSTREG enmGstReg)
2112{
2113 return iemNativeLivenessGetStateByGstReg(&pReNative->paLivenessEntries[pReNative->idxCurCall - 1], enmGstReg);
2114}
2115# endif /* VBOX_STRICT */
2116
2117#endif /* IEMNATIVE_WITH_LIVENESS_ANALYSIS */
2118
2119
2120/**
2121 * Gets the number of hidden arguments for an expected IEM_MC_CALL statement.
2122 */
2123DECL_FORCE_INLINE(uint8_t) iemNativeArgGetHiddenArgCount(PIEMRECOMPILERSTATE pReNative)
2124{
2125 if (pReNative->fCImpl & IEM_CIMPL_F_CALLS_CIMPL)
2126 return IEM_CIMPL_HIDDEN_ARGS;
2127 if (pReNative->fCImpl & (IEM_CIMPL_F_CALLS_AIMPL_WITH_FXSTATE | IEM_CIMPL_F_CALLS_AIMPL_WITH_XSTATE))
2128 return 1;
2129 return 0;
2130}
2131
2132
2133DECL_FORCE_INLINE(uint8_t) iemNativeRegMarkAllocated(PIEMRECOMPILERSTATE pReNative, unsigned idxReg,
2134 IEMNATIVEWHAT enmWhat, uint8_t idxVar = UINT8_MAX) RT_NOEXCEPT
2135{
2136 pReNative->Core.bmHstRegs |= RT_BIT_32(idxReg);
2137
2138 pReNative->Core.aHstRegs[idxReg].enmWhat = enmWhat;
2139 pReNative->Core.aHstRegs[idxReg].fGstRegShadows = 0;
2140 pReNative->Core.aHstRegs[idxReg].idxVar = idxVar;
2141 return (uint8_t)idxReg;
2142}
2143
2144
2145
2146/*********************************************************************************************************************************
2147* Register Allocator (GPR) *
2148*********************************************************************************************************************************/
2149
2150/**
2151 * Marks host register @a idxHstReg as containing a shadow copy of guest
2152 * register @a enmGstReg.
2153 *
2154 * ASSUMES that caller has made sure @a enmGstReg is not associated with any
2155 * host register before calling.
2156 */
2157DECL_FORCE_INLINE(void)
2158iemNativeRegMarkAsGstRegShadow(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, IEMNATIVEGSTREG enmGstReg, uint32_t off)
2159{
2160 Assert(!(pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg)));
2161 Assert(!pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows);
2162 Assert((unsigned)enmGstReg < (unsigned)kIemNativeGstReg_End);
2163
2164 pReNative->Core.aidxGstRegShadows[enmGstReg] = idxHstReg;
2165 pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = RT_BIT_64(enmGstReg); /** @todo why? not OR? */
2166 pReNative->Core.bmGstRegShadows |= RT_BIT_64(enmGstReg);
2167 pReNative->Core.bmHstRegsWithGstShadow |= RT_BIT_32(idxHstReg);
2168#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2169 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2170 iemNativeDbgInfoAddGuestRegShadowing(pReNative, enmGstReg, idxHstReg);
2171#else
2172 RT_NOREF(off);
2173#endif
2174}
2175
2176
2177/**
2178 * Clear any guest register shadow claims from @a idxHstReg.
2179 *
2180 * The register does not need to be shadowing any guest registers.
2181 */
2182DECL_FORCE_INLINE(void)
2183iemNativeRegClearGstRegShadowing(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, uint32_t off)
2184{
2185 Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows)
2186 == pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows
2187 && pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2188 Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxHstReg))
2189 == RT_BOOL(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows));
2190#ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
2191 Assert(!(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & pReNative->Core.bmGstRegShadowDirty));
2192#endif
2193
2194#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2195 uint64_t fGstRegs = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
2196 if (fGstRegs)
2197 {
2198 Assert(fGstRegs < RT_BIT_64(kIemNativeGstReg_End));
2199 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2200 while (fGstRegs)
2201 {
2202 unsigned const iGstReg = ASMBitFirstSetU64(fGstRegs) - 1;
2203 fGstRegs &= ~RT_BIT_64(iGstReg);
2204 iemNativeDbgInfoAddGuestRegShadowing(pReNative, (IEMNATIVEGSTREG)iGstReg, UINT8_MAX, idxHstReg);
2205 }
2206 }
2207#else
2208 RT_NOREF(off);
2209#endif
2210
2211 pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
2212 pReNative->Core.bmGstRegShadows &= ~pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows;
2213 pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = 0;
2214}
2215
2216
2217/**
2218 * Clear guest register shadow claim regarding @a enmGstReg from @a idxHstReg
2219 * and global overview flags.
2220 */
2221DECL_FORCE_INLINE(void)
2222iemNativeRegClearGstRegShadowingOne(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstReg, IEMNATIVEGSTREG enmGstReg, uint32_t off)
2223{
2224 Assert(pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2225 Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows)
2226 == pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows
2227 && pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2228 Assert(pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg));
2229 Assert(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & RT_BIT_64(enmGstReg));
2230 Assert(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxHstReg));
2231#ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
2232 Assert(!(pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & pReNative->Core.bmGstRegShadowDirty));
2233#endif
2234
2235#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2236 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2237 iemNativeDbgInfoAddGuestRegShadowing(pReNative, enmGstReg, UINT8_MAX, idxHstReg);
2238#else
2239 RT_NOREF(off);
2240#endif
2241
2242 uint64_t const fGstRegShadowsNew = pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows & ~RT_BIT_64(enmGstReg);
2243 pReNative->Core.aHstRegs[idxHstReg].fGstRegShadows = fGstRegShadowsNew;
2244 if (!fGstRegShadowsNew)
2245 pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxHstReg);
2246 pReNative->Core.bmGstRegShadows &= ~RT_BIT_64(enmGstReg);
2247}
2248
2249
2250#if 0 /* unused */
2251/**
2252 * Clear any guest register shadow claim for @a enmGstReg.
2253 */
2254DECL_FORCE_INLINE(void)
2255iemNativeRegClearGstRegShadowingByGstReg(PIEMRECOMPILERSTATE pReNative, IEMNATIVEGSTREG enmGstReg, uint32_t off)
2256{
2257 Assert(pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2258 if (pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg))
2259 {
2260 Assert(pReNative->Core.aidxGstRegShadows[enmGstReg] < RT_ELEMENTS(pReNative->Core.aHstRegs));
2261 iemNativeRegClearGstRegShadowingOne(pReNative, pReNative->Core.aidxGstRegShadows[enmGstReg], enmGstReg, off);
2262 }
2263}
2264#endif
2265
2266
2267/**
2268 * Clear any guest register shadow claim for @a enmGstReg and mark @a idxHstRegNew
2269 * as the new shadow of it.
2270 *
2271 * Unlike the other guest reg shadow helpers, this does the logging for you.
2272 * However, it is the liveness state is not asserted here, the caller must do
2273 * that.
2274 */
2275DECL_FORCE_INLINE(void)
2276iemNativeRegClearAndMarkAsGstRegShadow(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstRegNew,
2277 IEMNATIVEGSTREG enmGstReg, uint32_t off)
2278{
2279 Assert(pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2280 if (pReNative->Core.bmGstRegShadows & RT_BIT_64(enmGstReg))
2281 {
2282 uint8_t const idxHstRegOld = pReNative->Core.aidxGstRegShadows[enmGstReg];
2283 Assert(idxHstRegOld < RT_ELEMENTS(pReNative->Core.aHstRegs));
2284 if (idxHstRegOld == idxHstRegNew)
2285 return;
2286 Log12(("iemNativeRegClearAndMarkAsGstRegShadow: %s for guest %s (from %s)\n", g_apszIemNativeHstRegNames[idxHstRegNew],
2287 g_aGstShadowInfo[enmGstReg].pszName, g_apszIemNativeHstRegNames[idxHstRegOld]));
2288 iemNativeRegClearGstRegShadowingOne(pReNative, pReNative->Core.aidxGstRegShadows[enmGstReg], enmGstReg, off);
2289 }
2290 else
2291 Log12(("iemNativeRegClearAndMarkAsGstRegShadow: %s for guest %s\n", g_apszIemNativeHstRegNames[idxHstRegNew],
2292 g_aGstShadowInfo[enmGstReg].pszName));
2293 iemNativeRegMarkAsGstRegShadow(pReNative, idxHstRegNew, enmGstReg, off);
2294}
2295
2296
2297/**
2298 * Transfers the guest register shadow claims of @a enmGstReg from @a idxRegFrom
2299 * to @a idxRegTo.
2300 */
2301DECL_FORCE_INLINE(void)
2302iemNativeRegTransferGstRegShadowing(PIEMRECOMPILERSTATE pReNative, uint8_t idxRegFrom, uint8_t idxRegTo,
2303 IEMNATIVEGSTREG enmGstReg, uint32_t off)
2304{
2305 Assert(pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows & RT_BIT_64(enmGstReg));
2306 Assert(pReNative->Core.aidxGstRegShadows[enmGstReg] == idxRegFrom);
2307 Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows)
2308 == pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows
2309 && pReNative->Core.bmGstRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2310 Assert( (pReNative->Core.bmGstRegShadows & pReNative->Core.aHstRegs[idxRegTo].fGstRegShadows)
2311 == pReNative->Core.aHstRegs[idxRegTo].fGstRegShadows);
2312 Assert( RT_BOOL(pReNative->Core.bmHstRegsWithGstShadow & RT_BIT_32(idxRegFrom))
2313 == RT_BOOL(pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows));
2314
2315 uint64_t const fGstRegShadowsFrom = pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows & ~RT_BIT_64(enmGstReg);
2316 pReNative->Core.aHstRegs[idxRegFrom].fGstRegShadows = fGstRegShadowsFrom;
2317 if (!fGstRegShadowsFrom)
2318 pReNative->Core.bmHstRegsWithGstShadow &= ~RT_BIT_32(idxRegFrom);
2319 pReNative->Core.bmHstRegsWithGstShadow |= RT_BIT_32(idxRegTo);
2320 pReNative->Core.aHstRegs[idxRegTo].fGstRegShadows |= RT_BIT_64(enmGstReg);
2321 pReNative->Core.aidxGstRegShadows[enmGstReg] = idxRegTo;
2322#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2323 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2324 iemNativeDbgInfoAddGuestRegShadowing(pReNative, enmGstReg, idxRegTo, idxRegFrom);
2325#else
2326 RT_NOREF(off);
2327#endif
2328}
2329
2330
2331/**
2332 * Flushes any delayed guest register writes.
2333 *
2334 * This must be called prior to calling CImpl functions and any helpers that use
2335 * the guest state (like raising exceptions) and such.
2336 *
2337 * This optimization has not yet been implemented. The first target would be
2338 * RIP updates, since these are the most common ones.
2339 *
2340 * @note This function does not flush any shadowing information for guest registers. This needs to be done by
2341 * the caller if it wishes to do so.
2342 */
2343DECL_INLINE_THROW(uint32_t)
2344iemNativeRegFlushPendingWrites(PIEMRECOMPILERSTATE pReNative, uint32_t off, uint64_t fGstShwExcept = 0, uint64_t fGstSimdShwExcept = 0)
2345{
2346#ifdef IEMNATIVE_WITH_DELAYED_REGISTER_WRITEBACK
2347 uint64_t const bmGstRegShadowDirty = pReNative->Core.bmGstRegShadowDirty & ~fGstShwExcept;
2348#else
2349 uint64_t const bmGstRegShadowDirty = 0;
2350#endif
2351#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
2352 uint64_t const bmGstSimdRegShadowDirty = (pReNative->Core.bmGstSimdRegShadowDirtyLo128 | pReNative->Core.bmGstSimdRegShadowDirtyHi128)
2353 & ~fGstSimdShwExcept;
2354#else
2355 uint64_t const bmGstSimdRegShadowDirty = 0;
2356#endif
2357#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
2358 uint64_t const fWritebackPc = ~(fGstShwExcept & kIemNativeGstReg_Pc);
2359#else
2360 uint64_t const fWritebackPc = 0;
2361#endif
2362 if (bmGstRegShadowDirty | bmGstSimdRegShadowDirty | fWritebackPc)
2363 return iemNativeRegFlushPendingWritesSlow(pReNative, off, fGstShwExcept, fGstSimdShwExcept);
2364
2365 return off;
2366}
2367
2368
2369
2370/*********************************************************************************************************************************
2371* SIMD register allocator (largely code duplication of the GPR allocator for now but might diverge) *
2372*********************************************************************************************************************************/
2373
2374#ifdef IEMNATIVE_WITH_SIMD_REG_ALLOCATOR
2375
2376DECL_FORCE_INLINE(uint8_t)
2377iemNativeSimdRegMarkAllocated(PIEMRECOMPILERSTATE pReNative, uint8_t idxSimdReg,
2378 IEMNATIVEWHAT enmWhat, uint8_t idxVar = UINT8_MAX) RT_NOEXCEPT
2379{
2380 pReNative->Core.bmHstSimdRegs |= RT_BIT_32(idxSimdReg);
2381
2382 pReNative->Core.aHstSimdRegs[idxSimdReg].enmWhat = enmWhat;
2383 pReNative->Core.aHstSimdRegs[idxSimdReg].idxVar = idxVar;
2384 pReNative->Core.aHstSimdRegs[idxSimdReg].fGstRegShadows = 0;
2385 return idxSimdReg;
2386}
2387
2388
2389/**
2390 * Marks host SIMD register @a idxHstSimdReg as containing a shadow copy of guest
2391 * SIMD register @a enmGstSimdReg.
2392 *
2393 * ASSUMES that caller has made sure @a enmGstSimdReg is not associated with any
2394 * host register before calling.
2395 */
2396DECL_FORCE_INLINE(void)
2397iemNativeSimdRegMarkAsGstSimdRegShadow(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg,
2398 IEMNATIVEGSTSIMDREG enmGstSimdReg, uint32_t off)
2399{
2400 Assert(!(pReNative->Core.bmGstSimdRegShadows & RT_BIT_64(enmGstSimdReg)));
2401 Assert(!pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows);
2402 Assert((unsigned)enmGstSimdReg < (unsigned)kIemNativeGstSimdReg_End);
2403
2404 pReNative->Core.aidxGstSimdRegShadows[enmGstSimdReg] = idxHstSimdReg;
2405 pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows |= RT_BIT_64(enmGstSimdReg);
2406 pReNative->Core.bmGstSimdRegShadows |= RT_BIT_64(enmGstSimdReg);
2407 pReNative->Core.bmHstSimdRegsWithGstShadow |= RT_BIT_32(idxHstSimdReg);
2408#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2409 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2410 iemNativeDbgInfoAddGuestSimdRegShadowing(pReNative, enmGstSimdReg, idxHstSimdReg);
2411#else
2412 RT_NOREF(off);
2413#endif
2414}
2415
2416
2417/**
2418 * Transfers the guest SIMD register shadow claims of @a enmGstSimdReg from @a idxSimdRegFrom
2419 * to @a idxSimdRegTo.
2420 */
2421DECL_FORCE_INLINE(void)
2422iemNativeSimdRegTransferGstSimdRegShadowing(PIEMRECOMPILERSTATE pReNative, uint8_t idxSimdRegFrom, uint8_t idxSimdRegTo,
2423 IEMNATIVEGSTSIMDREG enmGstSimdReg, uint32_t off)
2424{
2425 Assert(pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows & RT_BIT_64(enmGstSimdReg));
2426 Assert(pReNative->Core.aidxGstSimdRegShadows[enmGstSimdReg] == idxSimdRegFrom);
2427 Assert( (pReNative->Core.bmGstSimdRegShadows & pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows)
2428 == pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows
2429 && pReNative->Core.bmGstSimdRegShadows < RT_BIT_64(kIemNativeGstReg_End));
2430 Assert( (pReNative->Core.bmGstSimdRegShadows & pReNative->Core.aHstSimdRegs[idxSimdRegTo].fGstRegShadows)
2431 == pReNative->Core.aHstSimdRegs[idxSimdRegTo].fGstRegShadows);
2432 Assert( RT_BOOL(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxSimdRegFrom))
2433 == RT_BOOL(pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows));
2434 Assert( pReNative->Core.aHstSimdRegs[idxSimdRegFrom].enmLoaded
2435 == pReNative->Core.aHstSimdRegs[idxSimdRegTo].enmLoaded);
2436
2437 uint64_t const fGstRegShadowsFrom = pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows & ~RT_BIT_64(enmGstSimdReg);
2438 pReNative->Core.aHstSimdRegs[idxSimdRegFrom].fGstRegShadows = fGstRegShadowsFrom;
2439 if (!fGstRegShadowsFrom)
2440 {
2441 pReNative->Core.bmHstSimdRegsWithGstShadow &= ~RT_BIT_32(idxSimdRegFrom);
2442 pReNative->Core.aHstSimdRegs[idxSimdRegFrom].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
2443 }
2444 pReNative->Core.bmHstSimdRegsWithGstShadow |= RT_BIT_32(idxSimdRegTo);
2445 pReNative->Core.aHstSimdRegs[idxSimdRegTo].fGstRegShadows |= RT_BIT_64(enmGstSimdReg);
2446 pReNative->Core.aidxGstSimdRegShadows[enmGstSimdReg] = idxSimdRegTo;
2447#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2448 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2449 iemNativeDbgInfoAddGuestSimdRegShadowing(pReNative, enmGstSimdReg, idxSimdRegTo, idxSimdRegFrom);
2450#else
2451 RT_NOREF(off);
2452#endif
2453}
2454
2455
2456/**
2457 * Clear any guest register shadow claims from @a idxHstSimdReg.
2458 *
2459 * The register does not need to be shadowing any guest registers.
2460 */
2461DECL_FORCE_INLINE(void)
2462iemNativeSimdRegClearGstSimdRegShadowing(PIEMRECOMPILERSTATE pReNative, uint8_t idxHstSimdReg, uint32_t off)
2463{
2464 Assert( (pReNative->Core.bmGstSimdRegShadows & pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows)
2465 == pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows
2466 && pReNative->Core.bmGstSimdRegShadows < RT_BIT_64(kIemNativeGstSimdReg_End));
2467 Assert( RT_BOOL(pReNative->Core.bmHstSimdRegsWithGstShadow & RT_BIT_32(idxHstSimdReg))
2468 == RT_BOOL(pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows));
2469 Assert( !(pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows & pReNative->Core.bmGstSimdRegShadowDirtyLo128)
2470 && !(pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows & pReNative->Core.bmGstSimdRegShadowDirtyHi128));
2471
2472#ifdef IEMNATIVE_WITH_TB_DEBUG_INFO
2473 uint64_t fGstRegs = pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows;
2474 if (fGstRegs)
2475 {
2476 Assert(fGstRegs < RT_BIT_64(kIemNativeGstSimdReg_End));
2477 iemNativeDbgInfoAddNativeOffset(pReNative, off);
2478 while (fGstRegs)
2479 {
2480 unsigned const iGstReg = ASMBitFirstSetU64(fGstRegs) - 1;
2481 fGstRegs &= ~RT_BIT_64(iGstReg);
2482 iemNativeDbgInfoAddGuestSimdRegShadowing(pReNative, (IEMNATIVEGSTSIMDREG)iGstReg, UINT8_MAX, idxHstSimdReg);
2483 }
2484 }
2485#else
2486 RT_NOREF(off);
2487#endif
2488
2489 pReNative->Core.bmHstSimdRegsWithGstShadow &= ~RT_BIT_32(idxHstSimdReg);
2490 pReNative->Core.bmGstSimdRegShadows &= ~pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows;
2491 pReNative->Core.aHstSimdRegs[idxHstSimdReg].fGstRegShadows = 0;
2492 pReNative->Core.aHstSimdRegs[idxHstSimdReg].enmLoaded = kIemNativeGstSimdRegLdStSz_Invalid;
2493}
2494
2495#endif /* IEMNATIVE_WITH_SIMD_REG_ALLOCATOR */
2496
2497
2498#ifdef IEMNATIVE_WITH_DELAYED_PC_UPDATING
2499/**
2500 * Emits code to update the guest RIP value by adding the current offset since the start of the last RIP update.
2501 */
2502DECL_INLINE_THROW(uint32_t) iemNativeEmitPcWriteback(PIEMRECOMPILERSTATE pReNative, uint32_t off)
2503{
2504 if (pReNative->Core.offPc)
2505 return iemNativeEmitPcWritebackSlow(pReNative, off);
2506 return off;
2507}
2508#endif /* IEMNATIVE_WITH_DELAYED_PC_UPDATING */
2509
2510
2511#ifdef IEMNATIVE_WITH_RECOMPILER_PROLOGUE_SINGLETON
2512/** @note iemNativeTbEntry returns VBOXSTRICTRC, but we don't declare it as
2513 * it saves us the trouble of a hidden parameter on MSC/amd64. */
2514# ifdef RT_ARCH_AMD64
2515extern "C" IEM_DECL_NATIVE_HLP_DEF(int, iemNativeTbEntry, (PVMCPUCC pVCpu, uintptr_t pfnTbBody));
2516# elif defined(RT_ARCH_ARM64)
2517extern "C" IEM_DECL_NATIVE_HLP_DEF(int, iemNativeTbEntry, (PVMCPUCC pVCpu, PCPUMCTX pCpumCtx, uintptr_t pfnTbBody));
2518# endif
2519#endif
2520
2521#endif /* !RT_IN_ASSEMBLER - ASM-NOINC-END */
2522
2523/** @} */
2524
2525#endif /* !VMM_INCLUDED_SRC_include_IEMN8veRecompiler_h */
2526
Note: See TracBrowser for help on using the repository browser.

© 2024 Oracle Support Privacy / Do Not Sell My Info Terms of Use Trademark Policy Automated Access Etiquette