Home | History | Annotate | only in /external/qemu/tcg
Up to higher level directory
NameDateSize
i386/03-Dec-2014
LICENSE03-Dec-2014146
optimize.c03-Dec-201435.5K
README03-Dec-201415.9K
tcg-be-ldst.h03-Dec-20143K
tcg-be-null.h03-Dec-20141.4K
tcg-op.h03-Dec-201492.3K
tcg-opc.h03-Dec-201410.1K
tcg-runtime.h03-Dec-2014825
tcg.c03-Dec-201489.5K
tcg.h03-Dec-201427.8K
TODO03-Dec-2014394

README

      1 Tiny Code Generator - Fabrice Bellard.
      2 
      3 1) Introduction
      4 
      5 TCG (Tiny Code Generator) began as a generic backend for a C
      6 compiler. It was simplified to be used in QEMU. It also has its roots
      7 in the QOP code generator written by Paul Brook. 
      8 
      9 2) Definitions
     10 
     11 The TCG "target" is the architecture for which we generate the
     12 code. It is of course not the same as the "target" of QEMU which is
     13 the emulated architecture. As TCG started as a generic C backend used
     14 for cross compiling, it is assumed that the TCG target is different
     15 from the host, although it is never the case for QEMU.
     16 
     17 In this document, we use "guest" to specify what architecture we are
     18 emulating; "target" always means the TCG target, the machine on which
     19 we are running QEMU.
     20 
     21 A TCG "function" corresponds to a QEMU Translated Block (TB).
     22 
     23 A TCG "temporary" is a variable only live in a basic
     24 block. Temporaries are allocated explicitly in each function.
     25 
     26 A TCG "local temporary" is a variable only live in a function. Local
     27 temporaries are allocated explicitly in each function.
     28 
     29 A TCG "global" is a variable which is live in all the functions
     30 (equivalent of a C global variable). They are defined before the
     31 functions defined. A TCG global can be a memory location (e.g. a QEMU
     32 CPU register), a fixed host register (e.g. the QEMU CPU state pointer)
     33 or a memory location which is stored in a register outside QEMU TBs
     34 (not implemented yet).
     35 
     36 A TCG "basic block" corresponds to a list of instructions terminated
     37 by a branch instruction. 
     38 
     39 3) Intermediate representation
     40 
     41 3.1) Introduction
     42 
     43 TCG instructions operate on variables which are temporaries, local
     44 temporaries or globals. TCG instructions and variables are strongly
     45 typed. Two types are supported: 32 bit integers and 64 bit
     46 integers. Pointers are defined as an alias to 32 bit or 64 bit
     47 integers depending on the TCG target word size.
     48 
     49 Each instruction has a fixed number of output variable operands, input
     50 variable operands and always constant operands.
     51 
     52 The notable exception is the call instruction which has a variable
     53 number of outputs and inputs.
     54 
     55 In the textual form, output operands usually come first, followed by
     56 input operands, followed by constant operands. The output type is
     57 included in the instruction name. Constants are prefixed with a '$'.
     58 
     59 add_i32 t0, t1, t2  (t0 <- t1 + t2)
     60 
     61 3.2) Assumptions
     62 
     63 * Basic blocks
     64 
     65 - Basic blocks end after branches (e.g. brcond_i32 instruction),
     66   goto_tb and exit_tb instructions.
     67 - Basic blocks start after the end of a previous basic block, or at a
     68   set_label instruction.
     69 
     70 After the end of a basic block, the content of temporaries is
     71 destroyed, but local temporaries and globals are preserved.
     72 
     73 * Floating point types are not supported yet
     74 
     75 * Pointers: depending on the TCG target, pointer size is 32 bit or 64
     76   bit. The type TCG_TYPE_PTR is an alias to TCG_TYPE_I32 or
     77   TCG_TYPE_I64.
     78 
     79 * Helpers:
     80 
     81 Using the tcg_gen_helper_x_y it is possible to call any function
     82 taking i32, i64 or pointer types. By default, before calling a helper,
     83 all globals are stored at their canonical location and it is assumed
     84 that the function can modify them. By default, the helper is allowed to
     85 modify the CPU state or raise an exception.
     86 
     87 This can be overridden using the following function modifiers:
     88 - TCG_CALL_NO_READ_GLOBALS means that the helper does not read globals,
     89   either directly or via an exception. They will not be saved to their
     90   canonical locations before calling the helper.
     91 - TCG_CALL_NO_WRITE_GLOBALS means that the helper does not modify any globals.
     92   They will only be saved to their canonical location before calling helpers,
     93   but they won't be reloaded afterwise.
     94 - TCG_CALL_NO_SIDE_EFFECTS means that the call to the function is removed if
     95   the return value is not used.
     96 
     97 Note that TCG_CALL_NO_READ_GLOBALS implies TCG_CALL_NO_WRITE_GLOBALS.
     98 
     99 On some TCG targets (e.g. x86), several calling conventions are
    100 supported.
    101 
    102 * Branches:
    103 
    104 Use the instruction 'br' to jump to a label.
    105 
    106 3.3) Code Optimizations
    107 
    108 When generating instructions, you can count on at least the following
    109 optimizations:
    110 
    111 - Single instructions are simplified, e.g.
    112 
    113    and_i32 t0, t0, $0xffffffff
    114     
    115   is suppressed.
    116 
    117 - A liveness analysis is done at the basic block level. The
    118   information is used to suppress moves from a dead variable to
    119   another one. It is also used to remove instructions which compute
    120   dead results. The later is especially useful for condition code
    121   optimization in QEMU.
    122 
    123   In the following example:
    124 
    125   add_i32 t0, t1, t2
    126   add_i32 t0, t0, $1
    127   mov_i32 t0, $1
    128 
    129   only the last instruction is kept.
    130 
    131 3.4) Instruction Reference
    132 
    133 ********* Function call
    134 
    135 * call <ret> <params> ptr
    136 
    137 call function 'ptr' (pointer type)
    138 
    139 <ret> optional 32 bit or 64 bit return value
    140 <params> optional 32 bit or 64 bit parameters
    141 
    142 ********* Jumps/Labels
    143 
    144 * set_label $label
    145 
    146 Define label 'label' at the current program point.
    147 
    148 * br $label
    149 
    150 Jump to label.
    151 
    152 * brcond_i32/i64 t0, t1, cond, label
    153 
    154 Conditional jump if t0 cond t1 is true. cond can be:
    155     TCG_COND_EQ
    156     TCG_COND_NE
    157     TCG_COND_LT /* signed */
    158     TCG_COND_GE /* signed */
    159     TCG_COND_LE /* signed */
    160     TCG_COND_GT /* signed */
    161     TCG_COND_LTU /* unsigned */
    162     TCG_COND_GEU /* unsigned */
    163     TCG_COND_LEU /* unsigned */
    164     TCG_COND_GTU /* unsigned */
    165 
    166 ********* Arithmetic
    167 
    168 * add_i32/i64 t0, t1, t2
    169 
    170 t0=t1+t2
    171 
    172 * sub_i32/i64 t0, t1, t2
    173 
    174 t0=t1-t2
    175 
    176 * neg_i32/i64 t0, t1
    177 
    178 t0=-t1 (two's complement)
    179 
    180 * mul_i32/i64 t0, t1, t2
    181 
    182 t0=t1*t2
    183 
    184 * div_i32/i64 t0, t1, t2
    185 
    186 t0=t1/t2 (signed). Undefined behavior if division by zero or overflow.
    187 
    188 * divu_i32/i64 t0, t1, t2
    189 
    190 t0=t1/t2 (unsigned). Undefined behavior if division by zero.
    191 
    192 * rem_i32/i64 t0, t1, t2
    193 
    194 t0=t1%t2 (signed). Undefined behavior if division by zero or overflow.
    195 
    196 * remu_i32/i64 t0, t1, t2
    197 
    198 t0=t1%t2 (unsigned). Undefined behavior if division by zero.
    199 
    200 ********* Logical
    201 
    202 * and_i32/i64 t0, t1, t2
    203 
    204 t0=t1&t2
    205 
    206 * or_i32/i64 t0, t1, t2
    207 
    208 t0=t1|t2
    209 
    210 * xor_i32/i64 t0, t1, t2
    211 
    212 t0=t1^t2
    213 
    214 * not_i32/i64 t0, t1
    215 
    216 t0=~t1
    217 
    218 * andc_i32/i64 t0, t1, t2
    219 
    220 t0=t1&~t2
    221 
    222 * eqv_i32/i64 t0, t1, t2
    223 
    224 t0=~(t1^t2), or equivalently, t0=t1^~t2
    225 
    226 * nand_i32/i64 t0, t1, t2
    227 
    228 t0=~(t1&t2)
    229 
    230 * nor_i32/i64 t0, t1, t2
    231 
    232 t0=~(t1|t2)
    233 
    234 * orc_i32/i64 t0, t1, t2
    235 
    236 t0=t1|~t2
    237 
    238 ********* Shifts/Rotates
    239 
    240 * shl_i32/i64 t0, t1, t2
    241 
    242 t0=t1 << t2. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
    243 
    244 * shr_i32/i64 t0, t1, t2
    245 
    246 t0=t1 >> t2 (unsigned). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
    247 
    248 * sar_i32/i64 t0, t1, t2
    249 
    250 t0=t1 >> t2 (signed). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
    251 
    252 * rotl_i32/i64 t0, t1, t2
    253 
    254 Rotation of t2 bits to the left. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
    255 
    256 * rotr_i32/i64 t0, t1, t2
    257 
    258 Rotation of t2 bits to the right. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
    259 
    260 ********* Misc
    261 
    262 * mov_i32/i64 t0, t1
    263 
    264 t0 = t1
    265 
    266 Move t1 to t0 (both operands must have the same type).
    267 
    268 * ext8s_i32/i64 t0, t1
    269 ext8u_i32/i64 t0, t1
    270 ext16s_i32/i64 t0, t1
    271 ext16u_i32/i64 t0, t1
    272 ext32s_i64 t0, t1
    273 ext32u_i64 t0, t1
    274 
    275 8, 16 or 32 bit sign/zero extension (both operands must have the same type)
    276 
    277 * bswap16_i32/i64 t0, t1
    278 
    279 16 bit byte swap on a 32/64 bit value. It assumes that the two/six high order
    280 bytes are set to zero.
    281 
    282 * bswap32_i32/i64 t0, t1
    283 
    284 32 bit byte swap on a 32/64 bit value. With a 64 bit value, it assumes that
    285 the four high order bytes are set to zero.
    286 
    287 * bswap64_i64 t0, t1
    288 
    289 64 bit byte swap
    290 
    291 * discard_i32/i64 t0
    292 
    293 Indicate that the value of t0 won't be used later. It is useful to
    294 force dead code elimination.
    295 
    296 * deposit_i32/i64 dest, t1, t2, pos, len
    297 
    298 Deposit T2 as a bitfield into T1, placing the result in DEST.
    299 The bitfield is described by POS/LEN, which are immediate values:
    300 
    301   LEN - the length of the bitfield
    302   POS - the position of the first bit, counting from the LSB
    303 
    304 For example, pos=8, len=4 indicates a 4-bit field at bit 8.
    305 This operation would be equivalent to
    306 
    307   dest = (t1 & ~0x0f00) | ((t2 << 8) & 0x0f00)
    308 
    309 
    310 ********* Conditional moves
    311 
    312 * setcond_i32/i64 dest, t1, t2, cond
    313 
    314 dest = (t1 cond t2)
    315 
    316 Set DEST to 1 if (T1 cond T2) is true, otherwise set to 0.
    317 
    318 * movcond_i32/i64 dest, c1, c2, v1, v2, cond
    319 
    320 dest = (c1 cond c2 ? v1 : v2)
    321 
    322 Set DEST to V1 if (C1 cond C2) is true, otherwise set to V2.
    323 
    324 ********* Type conversions
    325 
    326 * ext_i32_i64 t0, t1
    327 Convert t1 (32 bit) to t0 (64 bit) and does sign extension
    328 
    329 * extu_i32_i64 t0, t1
    330 Convert t1 (32 bit) to t0 (64 bit) and does zero extension
    331 
    332 * trunc_i64_i32 t0, t1
    333 Truncate t1 (64 bit) to t0 (32 bit)
    334 
    335 * concat_i32_i64 t0, t1, t2
    336 Construct t0 (64-bit) taking the low half from t1 (32 bit) and the high half
    337 from t2 (32 bit).
    338 
    339 * concat32_i64 t0, t1, t2
    340 Construct t0 (64-bit) taking the low half from t1 (64 bit) and the high half
    341 from t2 (64 bit).
    342 
    343 ********* Load/Store
    344 
    345 * ld_i32/i64 t0, t1, offset
    346 ld8s_i32/i64 t0, t1, offset
    347 ld8u_i32/i64 t0, t1, offset
    348 ld16s_i32/i64 t0, t1, offset
    349 ld16u_i32/i64 t0, t1, offset
    350 ld32s_i64 t0, t1, offset
    351 ld32u_i64 t0, t1, offset
    352 
    353 t0 = read(t1 + offset)
    354 Load 8, 16, 32 or 64 bits with or without sign extension from host memory. 
    355 offset must be a constant.
    356 
    357 * st_i32/i64 t0, t1, offset
    358 st8_i32/i64 t0, t1, offset
    359 st16_i32/i64 t0, t1, offset
    360 st32_i64 t0, t1, offset
    361 
    362 write(t0, t1 + offset)
    363 Write 8, 16, 32 or 64 bits to host memory.
    364 
    365 All this opcodes assume that the pointed host memory doesn't correspond
    366 to a global. In the latter case the behaviour is unpredictable.
    367 
    368 ********* Multiword arithmetic support
    369 
    370 * add2_i32/i64 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
    371 * sub2_i32/i64 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
    372 
    373 Similar to add/sub, except that the double-word inputs T1 and T2 are
    374 formed from two single-word arguments, and the double-word output T0
    375 is returned in two single-word outputs.
    376 
    377 * mulu2_i32/i64 t0_low, t0_high, t1, t2
    378 
    379 Similar to mul, except two unsigned inputs T1 and T2 yielding the full
    380 double-word product T0.  The later is returned in two single-word outputs.
    381 
    382 * muls2_i32/i64 t0_low, t0_high, t1, t2
    383 
    384 Similar to mulu2, except the two inputs T1 and T2 are signed.
    385 
    386 ********* 64-bit guest on 32-bit host support
    387 
    388 The following opcodes are internal to TCG.  Thus they are to be implemented by
    389 32-bit host code generators, but are not to be emitted by guest translators.
    390 They are emitted as needed by inline functions within "tcg-op.h".
    391 
    392 * brcond2_i32 t0_low, t0_high, t1_low, t1_high, cond, label
    393 
    394 Similar to brcond, except that the 64-bit values T0 and T1
    395 are formed from two 32-bit arguments.
    396 
    397 * setcond2_i32 dest, t1_low, t1_high, t2_low, t2_high, cond
    398 
    399 Similar to setcond, except that the 64-bit values T1 and T2 are
    400 formed from two 32-bit arguments.  The result is a 32-bit value.
    401 
    402 ********* QEMU specific operations
    403 
    404 * exit_tb t0
    405 
    406 Exit the current TB and return the value t0 (word type).
    407 
    408 * goto_tb index
    409 
    410 Exit the current TB and jump to the TB index 'index' (constant) if the
    411 current TB was linked to this TB. Otherwise execute the next
    412 instructions. Only indices 0 and 1 are valid and tcg_gen_goto_tb may be issued
    413 at most once with each slot index per TB.
    414 
    415 * qemu_ld_i32/i64 t0, t1, flags, memidx
    416 * qemu_st_i32/i64 t0, t1, flags, memidx
    417 
    418 Load data at the guest address t1 into t0, or store data in t0 at guest
    419 address t1.  The _i32/_i64 size applies to the size of the input/output
    420 register t0 only.  The address t1 is always sized according to the guest,
    421 and the width of the memory operation is controlled by flags.
    422 
    423 Both t0 and t1 may be split into little-endian ordered pairs of registers
    424 if dealing with 64-bit quantities on a 32-bit host.
    425 
    426 The memidx selects the qemu tlb index to use (e.g. user or kernel access).
    427 The flags are the TCGMemOp bits, selecting the sign, width, and endianness
    428 of the memory access.
    429 
    430 For a 32-bit host, qemu_ld/st_i64 is guaranteed to only be used with a
    431 64-bit memory access specified in flags.
    432 
    433 *********
    434 
    435 Note 1: Some shortcuts are defined when the last operand is known to be
    436 a constant (e.g. addi for add, movi for mov).
    437 
    438 Note 2: When using TCG, the opcodes must never be generated directly
    439 as some of them may not be available as "real" opcodes. Always use the
    440 function tcg_gen_xxx(args).
    441 
    442 4) Backend
    443 
    444 tcg-target.h contains the target specific definitions. tcg-target.c
    445 contains the target specific code.
    446 
    447 4.1) Assumptions
    448 
    449 The target word size (TCG_TARGET_REG_BITS) is expected to be 32 bit or
    450 64 bit. It is expected that the pointer has the same size as the word.
    451 
    452 On a 32 bit target, all 64 bit operations are converted to 32 bits. A
    453 few specific operations must be implemented to allow it (see add2_i32,
    454 sub2_i32, brcond2_i32).
    455 
    456 Floating point operations are not supported in this version. A
    457 previous incarnation of the code generator had full support of them,
    458 but it is better to concentrate on integer operations first.
    459 
    460 On a 64 bit target, no assumption is made in TCG about the storage of
    461 the 32 bit values in 64 bit registers.
    462 
    463 4.2) Constraints
    464 
    465 GCC like constraints are used to define the constraints of every
    466 instruction. Memory constraints are not supported in this
    467 version. Aliases are specified in the input operands as for GCC.
    468 
    469 The same register may be used for both an input and an output, even when
    470 they are not explicitly aliased.  If an op expands to multiple target
    471 instructions then care must be taken to avoid clobbering input values.
    472 GCC style "early clobber" outputs are not currently supported.
    473 
    474 A target can define specific register or constant constraints. If an
    475 operation uses a constant input constraint which does not allow all
    476 constants, it must also accept registers in order to have a fallback.
    477 
    478 The movi_i32 and movi_i64 operations must accept any constants.
    479 
    480 The mov_i32 and mov_i64 operations must accept any registers of the
    481 same type.
    482 
    483 The ld/st instructions must accept signed 32 bit constant offsets. It
    484 can be implemented by reserving a specific register to compute the
    485 address if the offset is too big.
    486 
    487 The ld/st instructions must accept any destination (ld) or source (st)
    488 register.
    489 
    490 4.3) Function call assumptions
    491 
    492 - The only supported types for parameters and return value are: 32 and
    493   64 bit integers and pointer.
    494 - The stack grows downwards.
    495 - The first N parameters are passed in registers.
    496 - The next parameters are passed on the stack by storing them as words.
    497 - Some registers are clobbered during the call. 
    498 - The function can return 0 or 1 value in registers. On a 32 bit
    499   target, functions must be able to return 2 values in registers for
    500   64 bit return type.
    501 
    502 5) Recommended coding rules for best performance
    503 
    504 - Use globals to represent the parts of the QEMU CPU state which are
    505   often modified, e.g. the integer registers and the condition
    506   codes. TCG will be able to use host registers to store them.
    507 
    508 - Avoid globals stored in fixed registers. They must be used only to
    509   store the pointer to the CPU state and possibly to store a pointer
    510   to a register window.
    511 
    512 - Use temporaries. Use local temporaries only when really needed,
    513   e.g. when you need to use a value after a jump. Local temporaries
    514   introduce a performance hit in the current TCG implementation: their
    515   content is saved to memory at end of each basic block.
    516 
    517 - Free temporaries and local temporaries when they are no longer used
    518   (tcg_temp_free). Since tcg_const_x() also creates a temporary, you
    519   should free it after it is used. Freeing temporaries does not yield
    520   a better generated code, but it reduces the memory usage of TCG and
    521   the speed of the translation.
    522 
    523 - Don't hesitate to use helpers for complicated or seldom used guest
    524   instructions. There is little performance advantage in using TCG to
    525   implement guest instructions taking more than about twenty TCG
    526   instructions. Note that this rule of thumb is more applicable to
    527   helpers doing complex logic or arithmetic, where the C compiler has
    528   scope to do a good job of optimisation; it is less relevant where
    529   the instruction is mostly doing loads and stores, and in those cases
    530   inline TCG may still be faster for longer sequences.
    531 
    532 - The hard limit on the number of TCG instructions you can generate
    533   per guest instruction is set by MAX_OP_PER_INSTR in exec-all.h --
    534   you cannot exceed this without risking a buffer overrun.
    535 
    536 - Use the 'discard' instruction if you know that TCG won't be able to
    537   prove that a given global is "dead" at a given program point. The
    538   x86 guest uses it to improve the condition codes optimisation.
    539