4 The current JIT implementation uses a tree matcher to generate code. We use a
5 simple algorithm to allocate registers in trees, and limit the number of used
6 temporary register to 4 when evaluating trees. So we can use 2 registers for
7 global register allocation.
9 Register Allocation for Trees
10 =============================
12 We always evaluate trees from left to right. When there are no more registers
13 available we need to spill values to memory. Here is the simplified algorithm.
16 tree_allocate_regs (tree, exclude_reg)
18 if (!tree_allocate_regs (tree->left, -1))
21 if (!tree_allocate_regs (tree->right, -1)) {
23 tree->left->spilled == TRUE;
25 free_used_regs (tree->left);
27 if (!tree_allocate_regs (tree->right, tree->left->reg))
31 free_used_regs (tree->left);
32 free_used_regs (tree->right);
34 /* try to allocate a register (reg != exclude_reg) */
35 if ((tree->reg = next_free_reg (exclude_reg)) != -1)
41 The emit routing actually spills the registers:
46 tree_emit (tree->left);
48 if (tree->left->spilled)
49 save_reg (tree->left->reg);
51 tree_emit (tree->right);
53 if (tree->left->spilled)
54 restore_reg (tree->left->reg);
61 Global Register Allocation
62 ==========================
66 Local Register Allocation
67 =========================
69 This section describes the cross-platform local register allocator which is
70 in the file mini-codegen.c.
72 The input to the allocator is a basic block which contains linear IL, ie.
73 instructions of the form:
77 where DEST, SRC1, and SRC2 are virtual registers (vregs). The job of the
78 allocator is to assign hard or physical registers (hregs) to each virtual
79 registers so the vreg references in the instructions can be replaced with their
80 assigned hreg, allowing machine code to be generated later.
82 The allocator needs information about the number and types of arguments of
83 instructions. It takes this information from the machine description files. It
84 also needs arch specific information, like the number and type of the hard
85 registers. It gets this information from arch-specific macros.
87 Currently, the vregs and hregs are partitioned into two classes: integer and
90 The allocator consists of two phases: In the first phase, a forward pass is
91 made over the instructions, collecting liveness information for vregs. In the
92 second phase, a backward pass is made over the instructions, assigning
93 registers. This backward mode of operation makes understanding the allocator
94 somewhat difficult to understand, but leads to better code in most cases.
99 The state of the allocator is stored in two arrays: iassign and isymbolic.
100 iassign maps vregs to hregs, while isymbolic is the opposite.
101 For a vreg, iassign [vreg] can contain the following values:
103 * -1 vreg has no assigned hreg
105 * hreg index (>= 0) vreg is assigned to the given hreg. This means
106 later instructions (which we have already
107 processed due to the backward direction) expect
108 the value of vreg to be found in hreg.
110 * spill slot index (< -1) vreg is spilled to the given spill slot. This
111 means later instructions expect the value of
112 vreg to be found on the stack in the given
115 Also, the allocator keeps track of which hregs are free and which are used.
116 This information is stored in a bitmask called ifree_mask.
118 There is a similar set of data structures for floating point registers.
123 When an allocator needs a free hreg, but all of them are assigned, it needs to
124 free up one of them. It does this by spilling the contents of the vreg which
125 is currently assigned to the selected hreg. Since later instructions expect
126 the vreg to be found in the selected hreg, the allocator emits a spill-load
127 instruction to load the value from the spill slot into the hreg after the
128 currently processed instruction. When the vreg which is spilled is a
129 destination in an instruction, the allocator will emit a spill-store to store
130 the value into the spill slot.
135 Some architectures, notably x86/amd64 require that the arguments/results of
136 some instructions be assigned to specific hregs. An example is the shift
137 opcodes on x86, where the second argument must be in ECX. The allocator
138 has support for this. It tries to allocate the vreg to the required hreg. If
139 thats not possible, then it will emit compensation code which moves values to
140 the correct registers before/after the instruction.
142 Fixed registers are mainly used on x86, but they are useful on more regular
143 architectures on well, for example to model that after a call instruction, the
144 return of the call is in a specific register.
146 A special case of fixed registers is two address architectures, like the x86,
147 where the instructions place their results into their first argument. This is
148 modelled in the allocator by allocating SRC1 and DEST to the same hreg.
153 Some variables might already be allocated to hardware registers during the
154 global allocation phase. In this case, SRC1, SRC2 and DEST might already be
155 a hardware register. The allocator needs to do nothing in this case, except
156 when the architecture uses fixed registers, in which case it needs to emit
162 64 bit arithmetic on 32 bit machines requires instructions whose arguments are
163 not registers, but register pairs. The allocator has support for this, both
164 for freely allocatable register pairs, and for register pairs which are
165 constrained to specific hregs (EDX:EAX on x86).
170 The x86 architecture uses a floating point register stack instead of a set of
171 fp registers. The allocator supports this by keeping track of the height of the
172 fp stack, and spilling/loading values from the stack as neccesary.
177 Calls need special handling for two reasons: first, they will clobber all
178 caller-save registers, meaning their contents will need to be spilled. Also,
179 some architectures pass arguments in registers. The registers used for
180 passing arguments are usually the same as the ones used for local allocation,
181 so the allocator needs to handle them specially. This is done as follows:
182 the MonoInst for the call instruction contains a map mapping vregs which
183 contain the argument values to hregs where the argument needs to be placed,
184 like this (on amd64):
190 When the allocator processes the call instruction, it allocates the vregs
191 in the map to their associated hregs. So the call instruction is processed as
192 if having a variable number of arguments which fixed register assignments.
200 When the call instruction is processed, R33 is assigned to RDI, and R34 is
201 assigned to RSI. Later, when the two assignment instructions are processed,
202 R33 and R34 are already assigned to a hreg, so they are replaced with the
203 associated hreg leading to the following final code:
209 Machine description files
210 =========================
212 A typical entry in the machine description files looks like this:
214 shl: dest:i src1:i src2:s clob:1 len:2
216 The allocator is only interested in the dest,src1,src2 and clob fields.
217 It understands the following values for the dest, src1, src2 fields:
221 b - base register (same as i, but the instruction does not modify the reg)
222 m - fp register, even if an fp stack is used (no fp stack tracking)
224 It understands the following values for the clob field:
226 1 - sreg1 needs to be the same as dreg
227 c - instruction clobbers the caller-save registers
229 Beside these values, an architecture can define additional values (like the 's'
230 in the example). The allocator depends on a set of arch-specific macros to
231 convert these values to information it needs during allocation.
236 These macros usually receive a value from the machine description file (like
237 the 's' in the example). The examples below are for x86.
240 * A bitmask selecting the caller-save registers (these are used for local
243 #define MONO_ARCH_CALLEE_REGS X86_CALLEE_REGS
246 * A bitmask selecting the callee-saved registers (these are usually used for
247 * global allocation).
249 #define MONO_ARCH_CALLEE_SAVED_REGS X86_CALLER_REGS
251 /* Same for the floating point registers */
252 #define MONO_ARCH_CALLEE_FREGS 0
253 #define MONO_ARCH_CALLEE_SAVED_FREGS 0
255 /* Whenever the target uses a floating point stack */
256 #define MONO_ARCH_USE_FPSTACK TRUE
258 /* The size of the floating point stack */
259 #define MONO_ARCH_FPSTACK_SIZE 6
262 * Given a descriptor value from the machine description file, return the fixed
263 * hard reg corresponding to that value.
265 #define MONO_ARCH_INST_FIXED_REG(desc) ((desc == 's') ? X86_ECX : ((desc == 'a') ? X86_EAX : ((desc == 'd') ? X86_EDX : ((desc == 'y') ? X86_EAX : ((desc == 'l') ? X86_EAX : -1)))))
268 * A bitmask selecting the hregs which can be used for allocating sreg2 for
269 * a given instruction.
271 #define MONO_ARCH_INST_SREG2_MASK(ins) (((ins [MONO_INST_CLOB] == 'a') || (ins [MONO_INST_CLOB] == 'd')) ? (1 << X86_EDX) : 0)
274 * Given a descriptor value, return whenever it denotes a register pair.
276 #define MONO_ARCH_INST_IS_REGPAIR(desc) (desc == 'l' || desc == 'L')
279 * Given a descriptor value, and the first register of a regpair, return a
280 * bitmask selecting the hregs which can be used for allocating the second
281 * register of the regpair.
283 #define MONO_ARCH_INST_REGPAIR_REG2(desc,hreg1) (desc == 'l' ? X86_EDX : -1)