2) Definitions
+TCG receives RISC-like "TCG ops" and performs some optimizations on them,
+including liveness analysis and trivial constant expression
+evaluation. TCG ops are then implemented in the host CPU back end,
+also known as the TCG "target".
+
The TCG "target" is the architecture for which we generate the
code. It is of course not the same as the "target" of QEMU which is
the emulated architecture. As TCG started as a generic C backend used
for cross compiling, it is assumed that the TCG target is different
from the host, although it is never the case for QEMU.
+In this document, we use "guest" to specify what architecture we are
+emulating; "target" always means the TCG target, the machine on which
+we are running QEMU.
+
A TCG "function" corresponds to a QEMU Translated Block (TB).
A TCG "temporary" is a variable only live in a basic
A TCG "basic block" corresponds to a list of instructions terminated
by a branch instruction.
+An operation with "undefined behavior" may result in a crash.
+
+An operation with "unspecified behavior" shall not crash. However,
+the result may be one of several possibilities so may be considered
+an "undefined result".
+
3) Intermediate representation
3.1) Introduction
Using the tcg_gen_helper_x_y it is possible to call any function
taking i32, i64 or pointer types. By default, before calling a helper,
all globals are stored at their canonical location and it is assumed
-that the function can modify them. This can be overridden by the
-TCG_CALL_CONST function modifier. By default, the helper is allowed to
-modify the CPU state or raise an exception. This can be overridden by
-the TCG_CALL_PURE function modifier, in which case the call to the
-function is removed if the return value is not used.
+that the function can modify them. By default, the helper is allowed to
+modify the CPU state or raise an exception.
+
+This can be overridden using the following function modifiers:
+- TCG_CALL_NO_READ_GLOBALS means that the helper does not read globals,
+ either directly or via an exception. They will not be saved to their
+ canonical locations before calling the helper.
+- TCG_CALL_NO_WRITE_GLOBALS means that the helper does not modify any globals.
+ They will only be saved to their canonical location before calling helpers,
+ but they won't be reloaded afterwards.
+- TCG_CALL_NO_SIDE_EFFECTS means that the call to the function is removed if
+ the return value is not used.
+
+Note that TCG_CALL_NO_READ_GLOBALS implies TCG_CALL_NO_WRITE_GLOBALS.
On some TCG targets (e.g. x86), several calling conventions are
supported.
* Branches:
-Use the instruction 'br' to jump to a label. Use 'jmp' to jump to an
-explicit address. Conditional branches can only jump to labels.
+Use the instruction 'br' to jump to a label.
3.3) Code Optimizations
********* Jumps/Labels
-* jmp t0
-
-Absolute jump to address t0 (pointer type).
-
* set_label $label
Define label 'label' at the current program point.
Jump to label.
-* brcond_i32/i64 cond, t0, t1, label
+* brcond_i32/i64 t0, t1, cond, label
Conditional jump if t0 cond t1 is true. cond can be:
TCG_COND_EQ
t0=t1|~t2
+* clz_i32/i64 t0, t1, t2
+
+t0 = t1 ? clz(t1) : t2
+
+* ctz_i32/i64 t0, t1, t2
+
+t0 = t1 ? ctz(t1) : t2
+
+* ctpop_i32/i64 t0, t1
+
+t0 = number of bits set in t1
+With "ctpop" short for "count population", matching
+the function name used in include/qemu/host-utils.h.
+
********* Shifts/Rotates
* shl_i32/i64 t0, t1, t2
-t0=t1 << t2. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
+t0=t1 << t2. Unspecified behavior if t2 < 0 or t2 >= 32 (resp 64)
* shr_i32/i64 t0, t1, t2
-t0=t1 >> t2 (unsigned). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
+t0=t1 >> t2 (unsigned). Unspecified behavior if t2 < 0 or t2 >= 32 (resp 64)
* sar_i32/i64 t0, t1, t2
-t0=t1 >> t2 (signed). Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
+t0=t1 >> t2 (signed). Unspecified behavior if t2 < 0 or t2 >= 32 (resp 64)
* rotl_i32/i64 t0, t1, t2
-Rotation of t2 bits to the left. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
+Rotation of t2 bits to the left.
+Unspecified behavior if t2 < 0 or t2 >= 32 (resp 64)
* rotr_i32/i64 t0, t1, t2
-Rotation of t2 bits to the right. Undefined behavior if t2 < 0 or t2 >= 32 (resp 64)
+Rotation of t2 bits to the right.
+Unspecified behavior if t2 < 0 or t2 >= 32 (resp 64)
********* Misc
8, 16 or 32 bit sign/zero extension (both operands must have the same type)
-* bswap16_i32/i64 t0, t1
+* bswap16_i32/i64 t0, t1, flags
-16 bit byte swap on a 32/64 bit value. It assumes that the two/six high order
-bytes are set to zero.
+16 bit byte swap on the low bits of a 32/64 bit input.
+If flags & TCG_BSWAP_IZ, then t1 is known to be zero-extended from bit 15.
+If flags & TCG_BSWAP_OZ, then t0 will be zero-extended from bit 15.
+If flags & TCG_BSWAP_OS, then t0 will be sign-extended from bit 15.
+If neither TCG_BSWAP_OZ nor TCG_BSWAP_OS are set, then the bits of
+t0 above bit 15 may contain any value.
-* bswap32_i32/i64 t0, t1
+* bswap32_i64 t0, t1, flags
-32 bit byte swap on a 32/64 bit value. With a 64 bit value, it assumes that
-the four high order bytes are set to zero.
+32 bit byte swap on a 64-bit value. The flags are the same as for bswap16,
+except they apply from bit 31 instead of bit 15.
-* bswap64_i64 t0, t1
+* bswap32_i32 t0, t1, flags
+* bswap64_i64 t0, t1, flags
-64 bit byte swap
+32/64 bit byte swap. The flags are ignored, but still present
+for consistency with the other bswap opcodes.
* discard_i32/i64 t0
Indicate that the value of t0 won't be used later. It is useful to
force dead code elimination.
+* deposit_i32/i64 dest, t1, t2, pos, len
+
+Deposit T2 as a bitfield into T1, placing the result in DEST.
+The bitfield is described by POS/LEN, which are immediate values:
+
+ LEN - the length of the bitfield
+ POS - the position of the first bit, counting from the LSB
+
+For example, "deposit_i32 dest, t1, t2, 8, 4" indicates a 4-bit field
+at bit 8. This operation would be equivalent to
+
+ dest = (t1 & ~0x0f00) | ((t2 << 8) & 0x0f00)
+
+* extract_i32/i64 dest, t1, pos, len
+* sextract_i32/i64 dest, t1, pos, len
+
+Extract a bitfield from T1, placing the result in DEST.
+The bitfield is described by POS/LEN, which are immediate values,
+as above for deposit. For extract_*, the result will be extended
+to the left with zeros; for sextract_*, the result will be extended
+to the left with copies of the bitfield sign bit at pos + len - 1.
+
+For example, "sextract_i32 dest, t1, 8, 4" indicates a 4-bit field
+at bit 8. This operation would be equivalent to
+
+ dest = (t1 << 20) >> 28
+
+(using an arithmetic right shift).
+
+* extract2_i32/i64 dest, t1, t2, pos
+
+For N = {32,64}, extract an N-bit quantity from the concatenation
+of t2:t1, beginning at pos. The tcg_gen_extract2_{i32,i64} expander
+accepts 0 <= pos <= N as inputs. The backend code generator will
+not see either 0 or N as inputs for these opcodes.
+
+* extrl_i64_i32 t0, t1
+
+For 64-bit hosts only, extract the low 32-bits of input T1 and place it
+into 32-bit output T0. Depending on the host, this may be a simple move,
+or may require additional canonicalization.
+
+* extrh_i64_i32 t0, t1
+
+For 64-bit hosts only, extract the high 32-bits of input T1 and place it
+into 32-bit output T0. Depending on the host, this may be a simple shift,
+or may require additional canonicalization.
+
********* Conditional moves
-* setcond_i32/i64 cond, dest, t1, t2
+* setcond_i32/i64 dest, t1, t2, cond
dest = (t1 cond t2)
Set DEST to 1 if (T1 cond T2) is true, otherwise set to 0.
+* movcond_i32/i64 dest, c1, c2, v1, v2, cond
+
+dest = (c1 cond c2 ? v1 : v2)
+
+Set DEST to V1 if (C1 cond C2) is true, otherwise set to V2.
+
********* Type conversions
* ext_i32_i64 t0, t1
write(t0, t1 + offset)
Write 8, 16, 32 or 64 bits to host memory.
-********* 64-bit target on 32-bit host support
+All this opcodes assume that the pointed host memory doesn't correspond
+to a global. In the latter case the behaviour is unpredictable.
+
+********* Multiword arithmetic support
+
+* add2_i32/i64 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
+* sub2_i32/i64 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
+
+Similar to add/sub, except that the double-word inputs T1 and T2 are
+formed from two single-word arguments, and the double-word output T0
+is returned in two single-word outputs.
+
+* mulu2_i32/i64 t0_low, t0_high, t1, t2
+
+Similar to mul, except two unsigned inputs T1 and T2 yielding the full
+double-word product T0. The later is returned in two single-word outputs.
+
+* muls2_i32/i64 t0_low, t0_high, t1, t2
+
+Similar to mulu2, except the two inputs T1 and T2 are signed.
+
+* mulsh_i32/i64 t0, t1, t2
+* muluh_i32/i64 t0, t1, t2
+
+Provide the high part of a signed or unsigned multiply, respectively.
+If mulu2/muls2 are not provided by the backend, the tcg-op generator
+can obtain the same results can be obtained by emitting a pair of
+opcodes, mul+muluh/mulsh.
+
+********* Memory Barrier support
+
+* mb <$arg>
+
+Generate a target memory barrier instruction to ensure memory ordering as being
+enforced by a corresponding guest memory barrier instruction. The ordering
+enforced by the backend may be stricter than the ordering required by the guest.
+It cannot be weaker. This opcode takes a constant argument which is required to
+generate the appropriate barrier instruction. The backend should take care to
+emit the target barrier instruction only when necessary i.e., for SMP guests and
+when MTTCG is enabled.
+
+The guest translators should generate this opcode for all guest instructions
+which have ordering side effects.
+
+Please see docs/devel/atomics.rst for more information on memory barriers.
+
+********* 64-bit guest on 32-bit host support
The following opcodes are internal to TCG. Thus they are to be implemented by
32-bit host code generators, but are not to be emitted by guest translators.
They are emitted as needed by inline functions within "tcg-op.h".
-* brcond2_i32 cond, t0_low, t0_high, t1_low, t1_high, label
+* brcond2_i32 t0_low, t0_high, t1_low, t1_high, cond, label
Similar to brcond, except that the 64-bit values T0 and T1
are formed from two 32-bit arguments.
-* add2_i32 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
-* sub2_i32 t0_low, t0_high, t1_low, t1_high, t2_low, t2_high
-
-Similar to add/sub, except that the 64-bit inputs T1 and T2 are
-formed from two 32-bit arguments, and the 64-bit output T0
-is returned in two 32-bit outputs.
-
-* mulu2_i32 t0_low, t0_high, t1, t2
-
-Similar to mul, except two 32-bit (unsigned) inputs T1 and T2 yielding
-the full 64-bit product T0. The later is returned in two 32-bit outputs.
-
-* setcond2_i32 cond, dest, t1_low, t1_high, t2_low, t2_high
+* setcond2_i32 dest, t1_low, t1_high, t2_low, t2_high, cond
Similar to setcond, except that the 64-bit values T1 and T2 are
formed from two 32-bit arguments. The result is a 32-bit value.
Exit the current TB and jump to the TB index 'index' (constant) if the
current TB was linked to this TB. Otherwise execute the next
-instructions.
+instructions. Only indices 0 and 1 are valid and tcg_gen_goto_tb may be issued
+at most once with each slot index per TB.
+
+* lookup_and_goto_ptr tb_addr
-* qemu_ld8u t0, t1, flags
-qemu_ld8s t0, t1, flags
-qemu_ld16u t0, t1, flags
-qemu_ld16s t0, t1, flags
-qemu_ld32 t0, t1, flags
-qemu_ld32u t0, t1, flags
-qemu_ld32s t0, t1, flags
-qemu_ld64 t0, t1, flags
+Look up a TB address ('tb_addr') and jump to it if valid. If not valid,
+jump to the TCG epilogue to go back to the exec loop.
-Load data at the QEMU CPU address t1 into t0. t1 has the QEMU CPU address
-type. 'flags' contains the QEMU memory index (selects user or kernel access)
-for example.
+This operation is optional. If the TCG backend does not implement the
+goto_ptr opcode, emitting this op is equivalent to emitting exit_tb(0).
-Note that "qemu_ld32" implies a 32-bit result, while "qemu_ld32u" and
-"qemu_ld32s" imply a 64-bit result appropriately extended from 32 bits.
+* qemu_ld_i32/i64 t0, t1, flags, memidx
+* qemu_st_i32/i64 t0, t1, flags, memidx
+* qemu_st8_i32 t0, t1, flags, memidx
-* qemu_st8 t0, t1, flags
-qemu_st16 t0, t1, flags
-qemu_st32 t0, t1, flags
-qemu_st64 t0, t1, flags
+Load data at the guest address t1 into t0, or store data in t0 at guest
+address t1. The _i32/_i64 size applies to the size of the input/output
+register t0 only. The address t1 is always sized according to the guest,
+and the width of the memory operation is controlled by flags.
-Store the data t0 at the QEMU CPU Address t1. t1 has the QEMU CPU
-address type. 'flags' contains the QEMU memory index (selects user or
-kernel access) for example.
+Both t0 and t1 may be split into little-endian ordered pairs of registers
+if dealing with 64-bit quantities on a 32-bit host.
+
+The memidx selects the qemu tlb index to use (e.g. user or kernel access).
+The flags are the MemOp bits, selecting the sign, width, and endianness
+of the memory access.
+
+For a 32-bit host, qemu_ld/st_i64 is guaranteed to only be used with a
+64-bit memory access specified in flags.
+
+For i386, qemu_st8_i32 is exactly like qemu_st_i32, except the size of
+the memory operation is known to be 8-bit. This allows the backend to
+provide a different set of register constraints.
+
+********* Host vector operations
+
+All of the vector ops have two parameters, TCGOP_VECL & TCGOP_VECE.
+The former specifies the length of the vector in log2 64-bit units; the
+later specifies the length of the element (if applicable) in log2 8-bit units.
+E.g. VECL=1 -> 64 << 1 -> v128, and VECE=2 -> 1 << 2 -> i32.
+
+* mov_vec v0, v1
+* ld_vec v0, t1
+* st_vec v0, t1
+
+ Move, load and store.
+
+* dup_vec v0, r1
+
+ Duplicate the low N bits of R1 into VECL/VECE copies across V0.
+
+* dupi_vec v0, c
+
+ Similarly, for a constant.
+ Smaller values will be replicated to host register size by the expanders.
+
+* dup2_vec v0, r1, r2
+
+ Duplicate r2:r1 into VECL/64 copies across V0. This opcode is
+ only present for 32-bit hosts.
+
+* add_vec v0, v1, v2
+
+ v0 = v1 + v2, in elements across the vector.
+
+* sub_vec v0, v1, v2
+
+ Similarly, v0 = v1 - v2.
+
+* mul_vec v0, v1, v2
+
+ Similarly, v0 = v1 * v2.
+
+* neg_vec v0, v1
+
+ Similarly, v0 = -v1.
+
+* abs_vec v0, v1
+
+ Similarly, v0 = v1 < 0 ? -v1 : v1, in elements across the vector.
+
+* smin_vec:
+* umin_vec:
+
+ Similarly, v0 = MIN(v1, v2), for signed and unsigned element types.
+
+* smax_vec:
+* umax_vec:
+
+ Similarly, v0 = MAX(v1, v2), for signed and unsigned element types.
+
+* ssadd_vec:
+* sssub_vec:
+* usadd_vec:
+* ussub_vec:
+
+ Signed and unsigned saturating addition and subtraction. If the true
+ result is not representable within the element type, the element is
+ set to the minimum or maximum value for the type.
+
+* and_vec v0, v1, v2
+* or_vec v0, v1, v2
+* xor_vec v0, v1, v2
+* andc_vec v0, v1, v2
+* orc_vec v0, v1, v2
+* not_vec v0, v1
+
+ Similarly, logical operations with and without complement.
+ Note that VECE is unused.
+
+* shli_vec v0, v1, i2
+* shls_vec v0, v1, s2
+
+ Shift all elements from v1 by a scalar i2/s2. I.e.
+
+ for (i = 0; i < VECL/VECE; ++i) {
+ v0[i] = v1[i] << s2;
+ }
+
+* shri_vec v0, v1, i2
+* sari_vec v0, v1, i2
+* rotli_vec v0, v1, i2
+* shrs_vec v0, v1, s2
+* sars_vec v0, v1, s2
+
+ Similarly for logical and arithmetic right shift, and left rotate.
+
+* shlv_vec v0, v1, v2
+
+ Shift elements from v1 by elements from v2. I.e.
+
+ for (i = 0; i < VECL/VECE; ++i) {
+ v0[i] = v1[i] << v2[i];
+ }
+
+* shrv_vec v0, v1, v2
+* sarv_vec v0, v1, v2
+* rotlv_vec v0, v1, v2
+* rotrv_vec v0, v1, v2
+
+ Similarly for logical and arithmetic right shift, and rotates.
+
+* cmp_vec v0, v1, v2, cond
+
+ Compare vectors by element, storing -1 for true and 0 for false.
+
+* bitsel_vec v0, v1, v2, v3
+
+ Bitwise select, v0 = (v2 & v1) | (v3 & ~v1), across the entire vector.
+
+* cmpsel_vec v0, c1, c2, v3, v4, cond
+
+ Select elements based on comparison results:
+ for (i = 0; i < n; ++i) {
+ v0[i] = (c1[i] cond c2[i]) ? v3[i] : v4[i].
+ }
+
+*********
Note 1: Some shortcuts are defined when the last operand is known to be
a constant (e.g. addi for add, movi for mov).
4) Backend
-tcg-target.h contains the target specific definitions. tcg-target.c
-contains the target specific code.
+tcg-target.h contains the target specific definitions. tcg-target.c.inc
+contains the target specific code; it is #included by tcg/tcg.c, rather
+than being a standalone C file.
4.1) Assumptions
few specific operations must be implemented to allow it (see add2_i32,
sub2_i32, brcond2_i32).
+On a 64 bit target, the values are transferred between 32 and 64-bit
+registers using the following ops:
+- trunc_shr_i64_i32
+- ext_i32_i64
+- extu_i32_i64
+
+They ensure that the values are correctly truncated or extended when
+moved from a 32-bit to a 64-bit register or vice-versa. Note that the
+trunc_shr_i64_i32 is an optional op. It is not necessary to implement
+it if all the following conditions are met:
+- 64-bit registers can hold 32-bit values
+- 32-bit values in a 64-bit register do not need to stay zero or
+ sign extended
+- all 32-bit TCG ops ignore the high part of 64-bit registers
+
Floating point operations are not supported in this version. A
previous incarnation of the code generator had full support of them,
but it is better to concentrate on integer operations first.
-On a 64 bit target, no assumption is made in TCG about the storage of
-the 32 bit values in 64 bit registers.
-
4.2) Constraints
GCC like constraints are used to define the constraints of every
The same register may be used for both an input and an output, even when
they are not explicitly aliased. If an op expands to multiple target
instructions then care must be taken to avoid clobbering input values.
-GCC style "early clobber" outputs are not currently supported.
+GCC style "early clobber" outputs are supported, with '&'.
A target can define specific register or constant constraints. If an
operation uses a constant input constraint which does not allow all
constants, it must also accept registers in order to have a fallback.
+The constraint 'i' is defined generically to accept any constant.
+The constraint 'r' is not defined generically, but is consistently
+used by each backend to indicate all registers.
The movi_i32 and movi_i64 operations must accept any constants.
The mov_i32 and mov_i64 operations must accept any registers of the
same type.
-The ld/st instructions must accept signed 32 bit constant offsets. It
-can be implemented by reserving a specific register to compute the
-address if the offset is too big.
+The ld/st/sti instructions must accept signed 32 bit constant offsets.
+This can be implemented by reserving a specific register in which to
+compute the address if the offset is too big.
The ld/st instructions must accept any destination (ld) or source (st)
register.
+The sti instruction may fail if it cannot store the given constant.
+
4.3) Function call assumptions
- The only supported types for parameters and return value are: 32 and
a better generated code, but it reduces the memory usage of TCG and
the speed of the translation.
-- Don't hesitate to use helpers for complicated or seldom used target
+- Don't hesitate to use helpers for complicated or seldom used guest
instructions. There is little performance advantage in using TCG to
- implement target instructions taking more than about twenty TCG
- instructions.
+ implement guest instructions taking more than about twenty TCG
+ instructions. Note that this rule of thumb is more applicable to
+ helpers doing complex logic or arithmetic, where the C compiler has
+ scope to do a good job of optimisation; it is less relevant where
+ the instruction is mostly doing loads and stores, and in those cases
+ inline TCG may still be faster for longer sequences.
+
+- The hard limit on the number of TCG instructions you can generate
+ per guest instruction is set by MAX_OP_PER_INSTR in exec-all.h --
+ you cannot exceed this without risking a buffer overrun.
- Use the 'discard' instruction if you know that TCG won't be able to
prove that a given global is "dead" at a given program point. The
- x86 target uses it to improve the condition codes optimisation.
+ x86 guest uses it to improve the condition codes optimisation.
projects / mirror_qemu.git / blobdiff