Annotation of /sml/trunk/src/MLRISC/x86/mltree/x86.sml

Parent Directory | Revision Log

---

Revision 1117 - (view) (download)

1 :	jhr	1117	(* x86.sml
2 :	monnier	247	*
3 :			* COPYRIGHT (c) 1998 Bell Laboratories.
4 :	george	545	*
5 :			* This is a revised version that takes into account of
6 :			* the extended x86 instruction set, and has better handling of
7 :			* non-standard types. I've factored out the integer/floating point
8 :			* comparison code, added optimizations for conditional moves.
9 :			* The latter generates SETcc and CMOVcc (Pentium Pro only) instructions.
10 :			* To avoid problems, I have tried to incorporate as much of
11 :			* Lal's original magic incantations as possible.
12 :	monnier	247	*
13 :	george	545	* Some changes:
14 :			*
15 :			* 1. REMU/REMS/REMT are now supported
16 :			* 2. COND is supported by generating SETcc and/or CMOVcc; this
17 :			* may require at least a Pentium II to work.
18 :			* 3. Division by a constant has been optimized. Division by
19 :			* a power of 2 generates SHRL or SARL.
20 :			* 4. Better addressing mode selection has been implemented. This should
21 :			* improve array indexing on SML/NJ.
22 :			* 5. Generate testl/testb instead of andl whenever appropriate. This
23 :			* is recommended by the Intel Optimization Guide and seems to improve
24 :			* boxity tests on SML/NJ.
25 :	leunga	731	*
26 :			* More changes for floating point:
27 :			* A new mode is implemented which generates pseudo 3-address instructions
28 :			* for floating point. These instructions are register allocated the
29 :			* normal way, with the virtual registers mapped onto a set of pseudo
30 :			* %fp registers. These registers are then mapped onto the %st registers
31 :			* with a new postprocessing phase.
32 :			*
33 :	george	545	* -- Allen
34 :	monnier	247	*)
35 :	george	545	local
36 :			val rewriteMemReg = true (* should we rewrite memRegs *)
37 :	leunga	731	val enableFastFPMode = true (* set this to false to disable the mode *)
38 :	george	545	in
39 :
40 :	monnier	247	functor X86
41 :			(structure X86Instr : X86INSTR
42 :	leunga	797	structure MLTreeUtils : MLTREE_UTILS
43 :	george	933	where T = X86Instr.T
44 :	george	555	structure ExtensionComp : MLTREE_EXTENSION_COMP
45 :	george	933	where I = X86Instr and T = X86Instr.T
46 :	george	984	structure MLTreeStream : MLTREE_STREAM
47 :			where T = ExtensionComp.T
48 :	george	545	datatype arch = Pentium | PentiumPro | PentiumII | PentiumIII
49 :			val arch : arch ref
50 :	leunga	593	val cvti2f :
51 :	leunga	815	{ty: X86Instr.T.ty,
52 :			src: X86Instr.operand,
53 :			(* source operand, guaranteed to be non-memory! *)
54 :			an: Annotations.annotations ref (* cluster annotations *)
55 :			} ->
56 :	leunga	593	{instrs : X86Instr.instruction list,(* the instructions *)
57 :			tempMem: X86Instr.operand, (* temporary for CVTI2F *)
58 :			cleanup: X86Instr.instruction list (* cleanup code *)
59 :			}
60 :	leunga	731	(* When the following flag is set, we allocate floating point registers
61 :			* directly on the floating point stack
62 :			*)
63 :			val fast_floating_point : bool ref
64 :	george	545	) : sig include MLTREECOMP
65 :			val rewriteMemReg : bool
66 :			end =
67 :	monnier	247	struct
68 :	leunga	775	structure I = X86Instr
69 :			structure T = I.T
70 :	george	984	structure TS = ExtensionComp.TS
71 :	george	545	structure C = I.C
72 :			structure Shuffle = Shuffle(I)
73 :	monnier	247	structure W32 = Word32
74 :	george	545	structure A = MLRiscAnnotations
75 :	george	909	structure CFG = ExtensionComp.CFG
76 :	george	889	structure CB = CellsBasis
77 :	monnier	247
78 :	george	984	type instrStream = (I.instruction,C.cellset,CFG.cfg) TS.stream
79 :			type mltreeStream = (T.stm,T.mlrisc list,CFG.cfg) TS.stream
80 :	leunga	565
81 :			datatype kind = REAL | INTEGER
82 :	george	545
83 :			structure Gen = MLTreeGen
84 :			(structure T = T
85 :	jhr	1117	structure Cells = C
86 :	george	545	val intTy = 32
87 :			val naturalWidths = [32]
88 :			datatype rep = SE | ZE | NEITHER
89 :			val rep = NEITHER
90 :			)
91 :
92 :	monnier	411	fun error msg = MLRiscErrorMsg.error("X86",msg)
93 :	monnier	247
94 :	george	545	(* Should we perform automatic MemReg translation?
95 :			* If this is on, we can avoid doing RewritePseudo phase entirely.
96 :			*)
97 :			val rewriteMemReg = rewriteMemReg
98 :	leunga	731
99 :			(* The following hardcoded *)
100 :	leunga	744	fun isMemReg r = rewriteMemReg andalso
101 :	george	889	let val r = CB.registerNum r
102 :	leunga	744	in r >= 8 andalso r < 32
103 :			end
104 :	leunga	731	fun isFMemReg r = if enableFastFPMode andalso !fast_floating_point
105 :	george	889	then let val r = CB.registerNum r
106 :	leunga	744	in r >= 8 andalso r < 32 end
107 :	leunga	731	else true
108 :	leunga	744	val isAnyFMemReg = List.exists (fn r =>
109 :	george	889	let val r = CB.registerNum r
110 :	leunga	744	in r >= 8 andalso r < 32 end
111 :			)
112 :	monnier	247
113 :	george	555	val ST0 = C.ST 0
114 :			val ST7 = C.ST 7
115 :	leunga	797	val one = T.I.int_1
116 :	george	555
117 :	leunga	797	val opcodes8 = {INC=I.INCB,DEC=I.DECB,ADD=I.ADDB,SUB=I.SUBB,
118 :			NOT=I.NOTB,NEG=I.NEGB,
119 :			SHL=I.SHLB,SHR=I.SHRB,SAR=I.SARB,
120 :			OR=I.ORB,AND=I.ANDB,XOR=I.XORB}
121 :			val opcodes16 = {INC=I.INCW,DEC=I.DECW,ADD=I.ADDW,SUB=I.SUBW,
122 :			NOT=I.NOTW,NEG=I.NEGW,
123 :			SHL=I.SHLW,SHR=I.SHRW,SAR=I.SARW,
124 :			OR=I.ORW,AND=I.ANDW,XOR=I.XORW}
125 :			val opcodes32 = {INC=I.INCL,DEC=I.DECL,ADD=I.ADDL,SUB=I.SUBL,
126 :			NOT=I.NOTL,NEG=I.NEGL,
127 :			SHL=I.SHLL,SHR=I.SHRL,SAR=I.SARL,
128 :			OR=I.ORL,AND=I.ANDL,XOR=I.XORL}
129 :
130 :	george	545	(*
131 :			* The code generator
132 :			*)
133 :	monnier	411	fun selectInstructions
134 :	george	545	(instrStream as
135 :	george	1003	TS.S.STREAM{emit=emitInstruction,defineLabel,entryLabel,pseudoOp,
136 :			annotation,getAnnotations,beginCluster,endCluster,exitBlock,comment,...}) =
137 :			let
138 :			val emit = emitInstruction o I.INSTR
139 :			exception EA
140 :	monnier	411
141 :	george	545	(* label where a trap is generated -- one per cluster *)
142 :			val trapLabel = ref (NONE: (I.instruction * Label.label) option)
143 :	monnier	247
144 :	leunga	731	(* flag floating point generation *)
145 :			val floatingPointUsed = ref false
146 :
147 :	george	545	(* effective address of an integer register *)
148 :	leunga	731	fun IntReg r = if isMemReg r then I.MemReg r else I.Direct r
149 :			and RealReg r = if isFMemReg r then I.FDirect r else I.FPR r
150 :	monnier	411
151 :	george	545	(* Add an overflow trap *)
152 :			fun trap() =
153 :			let val jmp =
154 :			case !trapLabel of
155 :	george	909	NONE => let val label = Label.label "trap" ()
156 :	george	1003	val jmp = I.jcc{cond=I.O,
157 :	leunga	775	opnd=I.ImmedLabel(T.LABEL label)}
158 :	george	545	in trapLabel := SOME(jmp, label); jmp end
159 :			| SOME(jmp, _) => jmp
160 :	george	1003	in emitInstruction jmp end
161 :	monnier	411
162 :	george	545	val newReg = C.newReg
163 :			val newFreg = C.newFreg
164 :	monnier	247
165 :	leunga	731	fun fsize 32 = I.FP32
166 :			| fsize 64 = I.FP64
167 :			| fsize 80 = I.FP80
168 :			| fsize _ = error "fsize"
169 :
170 :	george	545	(* mark an expression with a list of annotations *)
171 :	george	1009	fun mark'(i,[]) = emitInstruction(i)
172 :	george	545	| mark'(i,a::an) = mark'(I.ANNOTATION{i=i,a=a},an)
173 :	monnier	247
174 :	george	545	(* annotate an expression and emit it *)
175 :	george	1009	fun mark(i,an) = mark'(I.INSTR i,an)
176 :	monnier	247
177 :	george	1003	val emits = app emitInstruction
178 :	leunga	731
179 :	george	545	(* emit parallel copies for integers
180 :			* Translates parallel copies that involve memregs into
181 :			* individual copies.
182 :			*)
183 :			fun copy([], [], an) = ()
184 :			| copy(dst, src, an) =
185 :			let fun mvInstr{dst as I.MemReg rd, src as I.MemReg rs} =
186 :	george	889	if CB.sameColor(rd,rs) then [] else
187 :	george	545	let val tmpR = I.Direct(newReg())
188 :	george	1003	in [I.move{mvOp=I.MOVL, src=src, dst=tmpR},
189 :			I.move{mvOp=I.MOVL, src=tmpR, dst=dst}]
190 :	george	545	end
191 :			| mvInstr{dst=I.Direct rd, src=I.Direct rs} =
192 :	george	889	if CB.sameColor(rd,rs) then []
193 :	george	1009	else [I.COPY{k=CB.GP, sz=32, dst=[rd], src=[rs], tmp=NONE}]
194 :	george	1003	| mvInstr{dst, src} = [I.move{mvOp=I.MOVL, src=src, dst=dst}]
195 :	george	545	in
196 :	leunga	731	emits (Shuffle.shuffle{mvInstr=mvInstr, ea=IntReg}
197 :	leunga	744	{tmp=SOME(I.Direct(newReg())),
198 :	george	545	dst=dst, src=src})
199 :			end
200 :
201 :			(* conversions *)
202 :			val itow = Word.fromInt
203 :			val wtoi = Word.toInt
204 :	george	761	fun toInt32 i = T.I.toInt32(32, i)
205 :	george	545	val w32toi32 = Word32.toLargeIntX
206 :			val i32tow32 = Word32.fromLargeInt
207 :	monnier	247
208 :	george	545	(* One day, this is going to bite us when precision(LargeInt)>32 *)
209 :			fun wToInt32 w = Int32.fromLarge(Word32.toLargeIntX w)
210 :	monnier	247
211 :	george	545	(* some useful registers *)
212 :			val eax = I.Direct(C.eax)
213 :			val ecx = I.Direct(C.ecx)
214 :			val edx = I.Direct(C.edx)
215 :	monnier	247
216 :	leunga	775	fun immedLabel lab = I.ImmedLabel(T.LABEL lab)
217 :	george	545
218 :			(* Is the expression zero? *)
219 :	george	761	fun isZero(T.LI z) = T.I.isZero z
220 :	george	545	| isZero(T.MARK(e,a)) = isZero e
221 :			| isZero _ = false
222 :			(* Does the expression set the zero bit?
223 :			* WARNING: we assume these things are not optimized out!
224 :			*)
225 :			fun setZeroBit(T.ANDB _) = true
226 :			| setZeroBit(T.ORB _) = true
227 :			| setZeroBit(T.XORB _) = true
228 :			| setZeroBit(T.SRA _) = true
229 :			| setZeroBit(T.SRL _) = true
230 :			| setZeroBit(T.SLL _) = true
231 :	leunga	695	| setZeroBit(T.SUB _) = true
232 :			| setZeroBit(T.ADDT _) = true
233 :			| setZeroBit(T.SUBT _) = true
234 :	george	545	| setZeroBit(T.MARK(e, _)) = setZeroBit e
235 :			| setZeroBit _ = false
236 :	monnier	247
237 :	leunga	695	fun setZeroBit2(T.ANDB _) = true
238 :			| setZeroBit2(T.ORB _) = true
239 :			| setZeroBit2(T.XORB _) = true
240 :			| setZeroBit2(T.SRA _) = true
241 :			| setZeroBit2(T.SRL _) = true
242 :			| setZeroBit2(T.SLL _) = true
243 :			| setZeroBit2(T.ADD(32, _, _)) = true (* can't use leal! *)
244 :			| setZeroBit2(T.SUB _) = true
245 :			| setZeroBit2(T.ADDT _) = true
246 :			| setZeroBit2(T.SUBT _) = true
247 :			| setZeroBit2(T.MARK(e, _)) = setZeroBit2 e
248 :			| setZeroBit2 _ = false
249 :
250 :	leunga	731	(* emit parallel copies for floating point
251 :			* Normal version.
252 :			*)
253 :			fun fcopy'(fty, [], [], _) = ()
254 :			| fcopy'(fty, dst as [_], src as [_], an) =
255 :	george	1009	mark'(I.COPY{k=CB.FP, sz=fty, dst=dst,src=src,tmp=NONE}, an)
256 :	leunga	731	| fcopy'(fty, dst, src, an) =
257 :	george	1009	mark'(I.COPY{k=CB.FP, sz=fty, dst=dst,src=src,tmp=SOME(I.FDirect(newFreg()))}, an)
258 :	monnier	247
259 :	leunga	731	(* emit parallel copies for floating point.
260 :			* Fast version.
261 :			* Translates parallel copies that involve memregs into
262 :			* individual copies.
263 :			*)
264 :
265 :			fun fcopy''(fty, [], [], _) = ()
266 :			| fcopy''(fty, dst, src, an) =
267 :			if true orelse isAnyFMemReg dst orelse isAnyFMemReg src then
268 :			let val fsize = fsize fty
269 :	george	1003	fun mvInstr{dst, src} = [I.fmove{fsize=fsize, src=src, dst=dst}]
270 :	leunga	731	in
271 :			emits (Shuffle.shuffle{mvInstr=mvInstr, ea=RealReg}
272 :	leunga	744	{tmp=case dst of
273 :	leunga	731	[_] => NONE
274 :			| _ => SOME(I.FPR(newReg())),
275 :			dst=dst, src=src})
276 :			end
277 :			else
278 :	george	1009	mark'(I.COPY{k=CB.FP, sz=fty, dst=dst,
279 :			src=src,tmp=
280 :	leunga	731	case dst of
281 :			[_] => NONE
282 :			| _ => SOME(I.FPR(newFreg()))}, an)
283 :
284 :			fun fcopy x = if enableFastFPMode andalso !fast_floating_point
285 :			then fcopy'' x else fcopy' x
286 :
287 :	george	545	(* Translates MLTREE condition code to x86 condition code *)
288 :			fun cond T.LT = I.LT | cond T.LTU = I.B
289 :			| cond T.LE = I.LE | cond T.LEU = I.BE
290 :			| cond T.EQ = I.EQ | cond T.NE = I.NE
291 :			| cond T.GE = I.GE | cond T.GEU = I.AE
292 :			| cond T.GT = I.GT | cond T.GTU = I.A
293 :	monnier	247
294 :	leunga	815	fun zero dst = emit(I.BINARY{binOp=I.XORL, src=dst, dst=dst})
295 :
296 :	george	545	(* Move and annotate *)
297 :			fun move'(src as I.Direct s, dst as I.Direct d, an) =
298 :	george	889	if CB.sameColor(s,d) then ()
299 :	george	1009	else mark'(I.COPY{k=CB.GP, sz=32, dst=[d], src=[s], tmp=NONE}, an)
300 :	leunga	815	| move'(I.Immed 0, dst as I.Direct d, an) =
301 :			mark(I.BINARY{binOp=I.XORL, src=dst, dst=dst}, an)
302 :	george	545	| move'(src, dst, an) = mark(I.MOVE{mvOp=I.MOVL, src=src, dst=dst}, an)
303 :	monnier	247
304 :	george	545	(* Move only! *)
305 :			fun move(src, dst) = move'(src, dst, [])
306 :	monnier	247
307 :	george	545	val readonly = I.Region.readonly
308 :	monnier	247
309 :	george	545	(*
310 :	george	761	* Compute an effective address.
311 :	george	545	*)
312 :	george	761	fun address(ea, mem) = let
313 :	george	545	(* Keep building a bigger and bigger effective address expressions
314 :			* The input is a list of trees
315 :			* b -- base
316 :			* i -- index
317 :			* s -- scale
318 :			* d -- immed displacement
319 :			*)
320 :			fun doEA([], b, i, s, d) = makeAddressingMode(b, i, s, d)
321 :			| doEA(t::trees, b, i, s, d) =
322 :			(case t of
323 :	george	761	T.LI n => doEAImmed(trees, toInt32 n, b, i, s, d)
324 :	leunga	775	| T.CONST _ => doEALabel(trees, t, b, i, s, d)
325 :			| T.LABEL _ => doEALabel(trees, t, b, i, s, d)
326 :			| T.LABEXP le => doEALabel(trees, le, b, i, s, d)
327 :	george	545	| T.ADD(32, t1, t2 as T.REG(_,r)) =>
328 :			if isMemReg r then doEA(t2::t1::trees, b, i, s, d)
329 :			else doEA(t1::t2::trees, b, i, s, d)
330 :			| T.ADD(32, t1, t2) => doEA(t1::t2::trees, b, i, s, d)
331 :			| T.SUB(32, t1, T.LI n) =>
332 :	george	761	doEA(t1::T.LI(T.I.NEG(32,n))::trees, b, i, s, d)
333 :			| T.SLL(32, t1, T.LI n) => let
334 :			val n = T.I.toInt(32, n)
335 :			in
336 :			case n
337 :			of 0 => displace(trees, t1, b, i, s, d)
338 :			| 1 => indexed(trees, t1, t, 1, b, i, s, d)
339 :			| 2 => indexed(trees, t1, t, 2, b, i, s, d)
340 :			| 3 => indexed(trees, t1, t, 3, b, i, s, d)
341 :			| _ => displace(trees, t, b, i, s, d)
342 :			end
343 :	george	545	| t => displace(trees, t, b, i, s, d)
344 :			)
345 :	monnier	247
346 :	george	545	(* Add an immed constant *)
347 :			and doEAImmed(trees, 0, b, i, s, d) = doEA(trees, b, i, s, d)
348 :			| doEAImmed(trees, n, b, i, s, I.Immed m) =
349 :	george	761	doEA(trees, b, i, s, I.Immed(n+m))
350 :	george	545	| doEAImmed(trees, n, b, i, s, I.ImmedLabel le) =
351 :	leunga	775	doEA(trees, b, i, s,
352 :			I.ImmedLabel(T.ADD(32,le,T.LI(T.I.fromInt32(32, n)))))
353 :	george	545	| doEAImmed(trees, n, b, i, s, _) = error "doEAImmed"
354 :	monnier	247
355 :	george	545	(* Add a label expression *)
356 :			and doEALabel(trees, le, b, i, s, I.Immed 0) =
357 :			doEA(trees, b, i, s, I.ImmedLabel le)
358 :			| doEALabel(trees, le, b, i, s, I.Immed m) =
359 :			doEA(trees, b, i, s,
360 :	leunga	775	I.ImmedLabel(T.ADD(32,le,T.LI(T.I.fromInt32(32, m))))
361 :	george	545	handle Overflow => error "doEALabel: constant too large")
362 :			| doEALabel(trees, le, b, i, s, I.ImmedLabel le') =
363 :	leunga	775	doEA(trees, b, i, s, I.ImmedLabel(T.ADD(32,le,le')))
364 :	george	545	| doEALabel(trees, le, b, i, s, _) = error "doEALabel"
365 :	monnier	247
366 :	george	545	and makeAddressingMode(NONE, NONE, _, disp) = disp
367 :			| makeAddressingMode(SOME base, NONE, _, disp) =
368 :			I.Displace{base=base, disp=disp, mem=mem}
369 :			| makeAddressingMode(base, SOME index, scale, disp) =
370 :	george	761	I.Indexed{base=base, index=index, scale=scale,
371 :	george	545	disp=disp, mem=mem}
372 :	monnier	247
373 :	george	545	(* generate code for tree and ensure that it is not in %esp *)
374 :			and exprNotEsp tree =
375 :			let val r = expr tree
376 :	george	889	in if CB.sameColor(r, C.esp) then
377 :	george	545	let val tmp = newReg()
378 :			in move(I.Direct r, I.Direct tmp); tmp end
379 :			else r
380 :			end
381 :	monnier	247
382 :	george	545	(* Add a base register *)
383 :			and displace(trees, t, NONE, i, s, d) = (* no base yet *)
384 :			doEA(trees, SOME(expr t), i, s, d)
385 :			| displace(trees, t, b as SOME base, NONE, _, d) = (* no index *)
386 :			(* make t the index, but make sure that it is not %esp! *)
387 :			let val i = expr t
388 :	george	889	in if CB.sameColor(i, C.esp) then
389 :	george	545	(* swap base and index *)
390 :	george	889	if CB.sameColor(base, C.esp) then
391 :	george	545	doEA(trees, SOME i, b, 0, d)
392 :			else (* base and index = %esp! *)
393 :			let val index = newReg()
394 :			in move(I.Direct i, I.Direct index);
395 :			doEA(trees, b, SOME index, 0, d)
396 :			end
397 :			else
398 :			doEA(trees, b, SOME i, 0, d)
399 :			end
400 :			| displace(trees, t, SOME base, i, s, d) = (* base and index *)
401 :			let val b = expr(T.ADD(32,T.REG(32,base),t))
402 :			in doEA(trees, SOME b, i, s, d) end
403 :	monnier	247
404 :	george	545	(* Add an indexed register *)
405 :			and indexed(trees, t, t0, scale, b, NONE, _, d) = (* no index yet *)
406 :			doEA(trees, b, SOME(exprNotEsp t), scale, d)
407 :			| indexed(trees, _, t0, _, NONE, i, s, d) = (* no base *)
408 :			doEA(trees, SOME(expr t0), i, s, d)
409 :			| indexed(trees, _, t0, _, SOME base, i, s, d) = (*base and index*)
410 :			let val b = expr(T.ADD(32, t0, T.REG(32, base)))
411 :			in doEA(trees, SOME b, i, s, d) end
412 :
413 :			in case doEA([ea], NONE, NONE, 0, I.Immed 0) of
414 :			I.Immed _ => raise EA
415 :			| I.ImmedLabel le => I.LabelEA le
416 :			| ea => ea
417 :			end (* address *)
418 :	monnier	247
419 :	george	545	(* reduce an expression into an operand *)
420 :	george	761	and operand(T.LI i) = I.Immed(toInt32(i))
421 :	leunga	775	| operand(x as (T.CONST _ | T.LABEL _)) = I.ImmedLabel x
422 :			| operand(T.LABEXP le) = I.ImmedLabel le
423 :	george	545	| operand(T.REG(_,r)) = IntReg r
424 :			| operand(T.LOAD(32,ea,mem)) = address(ea, mem)
425 :			| operand(t) = I.Direct(expr t)
426 :	monnier	247
427 :	george	545	and moveToReg(opnd) =
428 :			let val dst = I.Direct(newReg())
429 :			in move(opnd, dst); dst
430 :			end
431 :	monnier	247
432 :	george	545	and reduceOpnd(I.Direct r) = r
433 :			| reduceOpnd opnd =
434 :			let val dst = newReg()
435 :			in move(opnd, I.Direct dst); dst
436 :			end
437 :	monnier	247
438 :	george	545	(* ensure that the operand is either an immed or register *)
439 :			and immedOrReg(opnd as I.Displace _) = moveToReg opnd
440 :			| immedOrReg(opnd as I.Indexed _) = moveToReg opnd
441 :			| immedOrReg(opnd as I.MemReg _) = moveToReg opnd
442 :			| immedOrReg(opnd as I.LabelEA _) = moveToReg opnd
443 :			| immedOrReg opnd = opnd
444 :	monnier	247
445 :	george	545	and isImmediate(I.Immed _) = true
446 :			| isImmediate(I.ImmedLabel _) = true
447 :			| isImmediate _ = false
448 :	monnier	247
449 :	george	545	and regOrMem opnd = if isImmediate opnd then moveToReg opnd else opnd
450 :
451 :			and isMemOpnd opnd =
452 :			(case opnd of
453 :			I.Displace _ => true
454 :			| I.Indexed _ => true
455 :			| I.MemReg _ => true
456 :			| I.LabelEA _ => true
457 :	george	555	| I.FDirect f => true
458 :	george	545	| _ => false
459 :			)
460 :
461 :			(*
462 :			* Compute an integer expression and put the result in
463 :			* the destination register rd.
464 :			*)
465 :	george	889	and doExpr(exp, rd : CB.cell, an) =
466 :	george	545	let val rdOpnd = IntReg rd
467 :	monnier	247
468 :	george	889	fun equalRd(I.Direct r) = CB.sameColor(r, rd)
469 :			| equalRd(I.MemReg r) = CB.sameColor(r, rd)
470 :	george	545	| equalRd _ = false
471 :	monnier	247
472 :	george	545	(* Emit a binary operator. If the destination is
473 :			* a memReg, do something smarter.
474 :			*)
475 :			fun genBinary(binOp, opnd1, opnd2) =
476 :			if isMemReg rd andalso
477 :			(isMemOpnd opnd1 orelse isMemOpnd opnd2) orelse
478 :			equalRd(opnd2)
479 :			then
480 :			let val tmpR = newReg()
481 :			val tmp = I.Direct tmpR
482 :			in move(opnd1, tmp);
483 :			mark(I.BINARY{binOp=binOp, src=opnd2, dst=tmp}, an);
484 :			move(tmp, rdOpnd)
485 :			end
486 :			else
487 :			(move(opnd1, rdOpnd);
488 :			mark(I.BINARY{binOp=binOp, src=opnd2, dst=rdOpnd}, an)
489 :			)
490 :	monnier	247
491 :	george	545	(* Generate a binary operator; it may commute *)
492 :			fun binaryComm(binOp, e1, e2) =
493 :			let val (opnd1, opnd2) =
494 :			case (operand e1, operand e2) of
495 :			(x as I.Immed _, y) => (y, x)
496 :			| (x as I.ImmedLabel _, y) => (y, x)
497 :			| (x, y as I.Direct _) => (y, x)
498 :			| (x, y) => (x, y)
499 :			in genBinary(binOp, opnd1, opnd2)
500 :			end
501 :
502 :			(* Generate a binary operator; non-commutative *)
503 :			fun binary(binOp, e1, e2) =
504 :			genBinary(binOp, operand e1, operand e2)
505 :
506 :			(* Generate a unary operator *)
507 :			fun unary(unOp, e) =
508 :			let val opnd = operand e
509 :			in if isMemReg rd andalso isMemOpnd opnd then
510 :			let val tmp = I.Direct(newReg())
511 :			in move(opnd, tmp); move(tmp, rdOpnd)
512 :			end
513 :			else move(opnd, rdOpnd);
514 :			mark(I.UNARY{unOp=unOp, opnd=rdOpnd}, an)
515 :			end
516 :
517 :			(* Generate shifts; the shift
518 :			* amount must be a constant or in %ecx *)
519 :			fun shift(opcode, e1, e2) =
520 :			let val (opnd1, opnd2) = (operand e1, operand e2)
521 :			in case opnd2 of
522 :			I.Immed _ => genBinary(opcode, opnd1, opnd2)
523 :			| _ =>
524 :			if equalRd(opnd2) then
525 :			let val tmpR = newReg()
526 :			val tmp = I.Direct tmpR
527 :			in move(opnd1, tmp);
528 :			move(opnd2, ecx);
529 :			mark(I.BINARY{binOp=opcode, src=ecx, dst=tmp},an);
530 :			move(tmp, rdOpnd)
531 :			end
532 :			else
533 :			(move(opnd1, rdOpnd);
534 :			move(opnd2, ecx);
535 :			mark(I.BINARY{binOp=opcode, src=ecx, dst=rdOpnd},an)
536 :			)
537 :			end
538 :
539 :			(* Division or remainder: divisor must be in %edx:%eax pair *)
540 :			fun divrem(signed, overflow, e1, e2, resultReg) =
541 :			let val (opnd1, opnd2) = (operand e1, operand e2)
542 :			val _ = move(opnd1, eax)
543 :	leunga	815	val oper = if signed then (emit(I.CDQ); I.IDIVL1)
544 :			else (zero edx; I.DIVL1)
545 :	george	545	in mark(I.MULTDIV{multDivOp=oper, src=regOrMem opnd2},an);
546 :			move(resultReg, rdOpnd);
547 :			if overflow then trap() else ()
548 :			end
549 :
550 :			(* Optimize the special case for division *)
551 :	george	761	fun divide(signed, overflow, e1, e2 as T.LI n') = let
552 :			val n = toInt32 n'
553 :			val w = T.I.toWord32(32, n')
554 :			fun isPowerOf2 w = W32.andb((w - 0w1), w) = 0w0
555 :	george	545	fun log2 n = (* n must be > 0!!! *)
556 :			let fun loop(0w1,pow) = pow
557 :	george	761	| loop(w,pow) = loop(W32.>>(w, 0w1),pow+1)
558 :	george	545	in loop(n,0) end
559 :			in if n > 1 andalso isPowerOf2 w then
560 :	george	761	let val pow = T.LI(T.I.fromInt(32,log2 w))
561 :	george	545	in if signed then
562 :			(* signed; simulate round towards zero *)
563 :	george	909	let val label = Label.anon()
564 :	george	545	val reg1 = expr e1
565 :			val opnd1 = I.Direct reg1
566 :			in if setZeroBit e1 then ()
567 :			else emit(I.CMPL{lsrc=opnd1, rsrc=I.Immed 0});
568 :			emit(I.JCC{cond=I.GE, opnd=immedLabel label});
569 :			emit(if n = 2 then
570 :			I.UNARY{unOp=I.INCL, opnd=opnd1}
571 :			else
572 :			I.BINARY{binOp=I.ADDL,
573 :	george	761	src=I.Immed(n - 1),
574 :	george	545	dst=opnd1});
575 :			defineLabel label;
576 :			shift(I.SARL, T.REG(32, reg1), pow)
577 :			end
578 :			else (* unsigned *)
579 :			shift(I.SHRL, e1, pow)
580 :			end
581 :			else
582 :			(* note the only way we can overflow is if
583 :			* n = 0 or n = -1
584 :			*)
585 :			divrem(signed, overflow andalso (n = ~1 orelse n = 0),
586 :			e1, e2, eax)
587 :			end
588 :			| divide(signed, overflow, e1, e2) =
589 :			divrem(signed, overflow, e1, e2, eax)
590 :	monnier	247
591 :	george	545	fun rem(signed, overflow, e1, e2) =
592 :			divrem(signed, overflow, e1, e2, edx)
593 :	leunga	815
594 :			(* Makes sure the destination must be a register *)
595 :			fun dstMustBeReg f =
596 :			if isMemReg rd then
597 :			let val tmpR = newReg()
598 :			val tmp = I.Direct(tmpR)
599 :			in f(tmpR, tmp); move(tmp, rdOpnd) end
600 :			else f(rd, rdOpnd)
601 :
602 :	george	545	(* unsigned integer multiplication *)
603 :			fun uMultiply(e1, e2) =
604 :			(* note e2 can never be (I.Direct edx) *)
605 :			(move(operand e1, eax);
606 :	leunga	815	mark(I.MULTDIV{multDivOp=I.MULL1,
607 :	george	545	src=regOrMem(operand e2)},an);
608 :			move(eax, rdOpnd)
609 :			)
610 :
611 :			(* signed integer multiplication:
612 :			* The only forms that are allowed that also sets the
613 :			* OF and CF flags are:
614 :			*
615 :	leunga	815	* (dst) (src1) (src2)
616 :	george	545	* imul r32, r32/m32, imm8
617 :	leunga	815	* (dst) (src)
618 :	george	545	* imul r32, imm8
619 :			* imul r32, imm32
620 :	leunga	815	* imul r32, r32/m32
621 :			* Note: destination must be a register!
622 :	george	545	*)
623 :			fun multiply(e1, e2) =
624 :	leunga	815	dstMustBeReg(fn (rd, rdOpnd) =>
625 :			let fun doit(i1 as I.Immed _, i2 as I.Immed _) =
626 :			(move(i1, rdOpnd);
627 :			mark(I.BINARY{binOp=I.IMULL, dst=rdOpnd, src=i2},an))
628 :			| doit(rm, i2 as I.Immed _) = doit(i2, rm)
629 :			| doit(imm as I.Immed(i), rm) =
630 :			mark(I.MUL3{dst=rd, src1=rm, src2=i},an)
631 :			| doit(r1 as I.Direct _, r2 as I.Direct _) =
632 :			(move(r1, rdOpnd);
633 :			mark(I.BINARY{binOp=I.IMULL, dst=rdOpnd, src=r2},an))
634 :			| doit(r1 as I.Direct _, rm) =
635 :			(move(r1, rdOpnd);
636 :			mark(I.BINARY{binOp=I.IMULL, dst=rdOpnd, src=rm},an))
637 :			| doit(rm, r as I.Direct _) = doit(r, rm)
638 :			| doit(rm1, rm2) =
639 :	george	545	if equalRd rm2 then
640 :			let val tmpR = newReg()
641 :			val tmp = I.Direct tmpR
642 :			in move(rm1, tmp);
643 :	leunga	815	mark(I.BINARY{binOp=I.IMULL, dst=tmp, src=rm2},an);
644 :			move(tmp, rdOpnd)
645 :	george	545	end
646 :			else
647 :	leunga	815	(move(rm1, rdOpnd);
648 :			mark(I.BINARY{binOp=I.IMULL, dst=rdOpnd, src=rm2},an)
649 :	george	545	)
650 :			val (opnd1, opnd2) = (operand e1, operand e2)
651 :	leunga	815	in doit(opnd1, opnd2)
652 :	george	545	end
653 :	leunga	815	)
654 :	monnier	247
655 :	george	545	(* Emit a load instruction; makes sure that the destination
656 :			* is a register
657 :			*)
658 :			fun genLoad(mvOp, ea, mem) =
659 :			dstMustBeReg(fn (_, dst) =>
660 :			mark(I.MOVE{mvOp=mvOp, src=address(ea, mem), dst=dst},an))
661 :
662 :			(* Generate a zero extended loads *)
663 :			fun load8(ea, mem) = genLoad(I.MOVZBL, ea, mem)
664 :			fun load16(ea, mem) = genLoad(I.MOVZWL, ea, mem)
665 :			fun load8s(ea, mem) = genLoad(I.MOVSBL, ea, mem)
666 :			fun load16s(ea, mem) = genLoad(I.MOVSWL, ea, mem)
667 :			fun load32(ea, mem) = genLoad(I.MOVL, ea, mem)
668 :
669 :			(* Generate a sign extended loads *)
670 :
671 :			(* Generate setcc instruction:
672 :			* semantics: MV(rd, COND(_, T.CMP(ty, cc, t1, t2), yes, no))
673 :	leunga	583	* Bug, if eax is either t1 or t2 then problem will occur!!!
674 :			* Note that we have to use eax as the destination of the
675 :			* setcc because it only works on the registers
676 :			* %al, %bl, %cl, %dl and %[abcd]h. The last four registers
677 :			* are inaccessible in 32 bit mode.
678 :	george	545	*)
679 :			fun setcc(ty, cc, t1, t2, yes, no) =
680 :	leunga	583	let val (cc, yes, no) =
681 :			if yes > no then (cc, yes, no)
682 :			else (T.Basis.negateCond cc, no, yes)
683 :	george	545	in (* Clear the destination first.
684 :			* This this because stupid SETcc
685 :			* only writes to the low order
686 :			* byte. That's Intel architecture, folks.
687 :			*)
688 :	leunga	695	case (yes, no, cc) of
689 :			(1, 0, T.LT) =>
690 :			let val tmp = I.Direct(expr(T.SUB(32,t1,t2)))
691 :			in move(tmp, rdOpnd);
692 :			emit(I.BINARY{binOp=I.SHRL,src=I.Immed 31,dst=rdOpnd})
693 :			end
694 :			| (1, 0, T.GT) =>
695 :			let val tmp = I.Direct(expr(T.SUB(32,t1,t2)))
696 :			in emit(I.UNARY{unOp=I.NOTL,opnd=tmp});
697 :			move(tmp, rdOpnd);
698 :			emit(I.BINARY{binOp=I.SHRL,src=I.Immed 31,dst=rdOpnd})
699 :			end
700 :			| (1, 0, _) => (* normal case *)
701 :	george	545	let val cc = cmp(true, ty, cc, t1, t2, [])
702 :	leunga	583	in mark(I.SET{cond=cond cc, opnd=eax}, an);
703 :	leunga	695	emit(I.BINARY{binOp=I.ANDL,src=I.Immed 255, dst=eax});
704 :	leunga	583	move(eax, rdOpnd)
705 :			end
706 :	leunga	695	| (C1, C2, _) =>
707 :	george	545	(* general case;
708 :	leunga	583	* from the Intel optimization guide p3-5
709 :			*)
710 :	leunga	695	let val _ = zero eax;
711 :			val cc = cmp(true, ty, cc, t1, t2, [])
712 :	leunga	583	in case C1-C2 of
713 :			D as (1 | 2 | 3 | 4 | 5 | 8 | 9) =>
714 :			let val (base,scale) =
715 :			case D of
716 :			1 => (NONE, 0)
717 :			| 2 => (NONE, 1)
718 :			| 3 => (SOME C.eax, 1)
719 :			| 4 => (NONE, 2)
720 :			| 5 => (SOME C.eax, 2)
721 :			| 8 => (NONE, 3)
722 :			| 9 => (SOME C.eax, 3)
723 :			val addr = I.Indexed{base=base,
724 :			index=C.eax,
725 :			scale=scale,
726 :			disp=I.Immed C2,
727 :	george	545	mem=readonly}
728 :	leunga	583	val tmpR = newReg()
729 :			val tmp = I.Direct tmpR
730 :			in emit(I.SET{cond=cond cc, opnd=eax});
731 :			mark(I.LEA{r32=tmpR, addr=addr}, an);
732 :			move(tmp, rdOpnd)
733 :			end
734 :			| D =>
735 :			(emit(I.SET{cond=cond(T.Basis.negateCond cc),
736 :			opnd=eax});
737 :			emit(I.UNARY{unOp=I.DECL, opnd=eax});
738 :			emit(I.BINARY{binOp=I.ANDL,
739 :			src=I.Immed D, dst=eax});
740 :			if C2 = 0 then
741 :			move(eax, rdOpnd)
742 :			else
743 :			let val tmpR = newReg()
744 :			val tmp = I.Direct tmpR
745 :			in mark(I.LEA{addr=
746 :			I.Displace{
747 :			base=C.eax,
748 :			disp=I.Immed C2,
749 :			mem=readonly},
750 :			r32=tmpR}, an);
751 :			move(tmp, rdOpnd)
752 :			end
753 :			)
754 :			end
755 :	george	545	end (* setcc *)
756 :
757 :			(* Generate cmovcc instruction.
758 :			* on Pentium Pro and Pentium II only
759 :			*)
760 :			fun cmovcc(ty, cc, t1, t2, yes, no) =
761 :			let fun genCmov(dstR, _) =
762 :			let val _ = doExpr(no, dstR, []) (* false branch *)
763 :			val cc = cmp(true, ty, cc, t1, t2, []) (* compare *)
764 :			in mark(I.CMOV{cond=cond cc, src=operand yes, dst=dstR}, an)
765 :			end
766 :			in dstMustBeReg genCmov
767 :			end
768 :
769 :			fun unknownExp exp = doExpr(Gen.compileRexp exp, rd, an)
770 :	monnier	247
771 :	leunga	606	(* Add n to rd *)
772 :			fun addN n =
773 :			let val n = operand n
774 :			val src = if isMemReg rd then immedOrReg n else n
775 :			in mark(I.BINARY{binOp=I.ADDL, src=src, dst=rdOpnd}, an) end
776 :
777 :	george	545	(* Generate addition *)
778 :			fun addition(e1, e2) =
779 :	leunga	606	case e1 of
780 :	george	889	T.REG(_,rs) => if CB.sameColor(rs,rd) then addN e2
781 :	leunga	744	else addition1(e1,e2)
782 :	leunga	606	| _ => addition1(e1,e2)
783 :			and addition1(e1, e2) =
784 :			case e2 of
785 :	george	889	T.REG(_,rs) => if CB.sameColor(rs,rd) then addN e1
786 :	leunga	744	else addition2(e1,e2)
787 :	leunga	606	| _ => addition2(e1,e2)
788 :			and addition2(e1,e2) =
789 :	george	545	(dstMustBeReg(fn (dstR, _) =>
790 :			mark(I.LEA{r32=dstR, addr=address(exp, readonly)}, an))
791 :			handle EA => binaryComm(I.ADDL, e1, e2))
792 :	monnier	247
793 :
794 :	george	545	in case exp of
795 :			T.REG(_,rs) =>
796 :			if isMemReg rs andalso isMemReg rd then
797 :			let val tmp = I.Direct(newReg())
798 :	leunga	731	in move'(I.MemReg rs, tmp, an);
799 :	george	545	move'(tmp, rdOpnd, [])
800 :			end
801 :			else move'(IntReg rs, rdOpnd, an)
802 :	george	761	| T.LI z => let
803 :			val n = toInt32 z
804 :			in
805 :			if n=0 then
806 :			(* As per Fermin's request, special optimization for rd := 0.
807 :			* Currently we don't bother with the size.
808 :			*)
809 :			if isMemReg rd then move'(I.Immed 0, rdOpnd, an)
810 :			else mark(I.BINARY{binOp=I.XORL, src=rdOpnd, dst=rdOpnd}, an)
811 :			else
812 :			move'(I.Immed(n), rdOpnd, an)
813 :			end
814 :	leunga	775	| (T.CONST _ | T.LABEL _) =>
815 :			move'(I.ImmedLabel exp, rdOpnd, an)
816 :			| T.LABEXP le => move'(I.ImmedLabel le, rdOpnd, an)
817 :	monnier	247
818 :	george	545	(* 32-bit addition *)
819 :	george	761	| T.ADD(32, e1, e2 as T.LI n) => let
820 :			val n = toInt32 n
821 :			in
822 :			case n
823 :			of 1 => unary(I.INCL, e1)
824 :			| ~1 => unary(I.DECL, e1)
825 :			| _ => addition(e1, e2)
826 :			end
827 :			| T.ADD(32, e1 as T.LI n, e2) => let
828 :			val n = toInt32 n
829 :			in
830 :			case n
831 :			of 1 => unary(I.INCL, e2)
832 :			| ~1 => unary(I.DECL, e2)
833 :			| _ => addition(e1, e2)
834 :			end
835 :	george	545	| T.ADD(32, e1, e2) => addition(e1, e2)
836 :	monnier	247
837 :	leunga	695	(* 32-bit addition but set the flag!
838 :			* This is a stupid hack for now.
839 :			*)
840 :	george	761	| T.ADD(0, e, e1 as T.LI n) => let
841 :			val n = T.I.toInt(32, n)
842 :			in
843 :			if n=1 then unary(I.INCL, e)
844 :			else if n = ~1 then unary(I.DECL, e)
845 :			else binaryComm(I.ADDL, e, e1)
846 :			end
847 :			| T.ADD(0, e1 as T.LI n, e) => let
848 :			val n = T.I.toInt(32, n)
849 :			in
850 :			if n=1 then unary(I.INCL, e)
851 :			else if n = ~1 then unary(I.DECL, e)
852 :			else binaryComm(I.ADDL, e1, e)
853 :			end
854 :			| T.ADD(0, e1, e2) => binaryComm(I.ADDL, e1, e2)
855 :
856 :	george	545	(* 32-bit subtraction *)
857 :	george	761	| T.SUB(32, e1, e2 as T.LI n) => let
858 :			val n = toInt32 n
859 :			in
860 :			case n
861 :			of 0 => doExpr(e1, rd, an)
862 :			| 1 => unary(I.DECL, e1)
863 :			| ~1 => unary(I.INCL, e1)
864 :			| _ => binary(I.SUBL, e1, e2)
865 :			end
866 :			| T.SUB(32, e1 as T.LI n, e2) =>
867 :			if T.I.isZero n then unary(I.NEGL, e2)
868 :			else binary(I.SUBL, e1, e2)
869 :	george	545	| T.SUB(32, e1, e2) => binary(I.SUBL, e1, e2)
870 :	monnier	247
871 :	george	545	| T.MULU(32, x, y) => uMultiply(x, y)
872 :			| T.DIVU(32, x, y) => divide(false, false, x, y)
873 :			| T.REMU(32, x, y) => rem(false, false, x, y)
874 :	monnier	247
875 :	george	545	| T.MULS(32, x, y) => multiply(x, y)
876 :			| T.DIVS(32, x, y) => divide(true, false, x, y)
877 :			| T.REMS(32, x, y) => rem(true, false, x, y)
878 :	monnier	247
879 :	george	545	| T.ADDT(32, x, y) => (binaryComm(I.ADDL, x, y); trap())
880 :			| T.SUBT(32, x, y) => (binary(I.SUBL, x, y); trap())
881 :			| T.MULT(32, x, y) => (multiply(x, y); trap())
882 :			| T.DIVT(32, x, y) => divide(true, true, x, y)
883 :			| T.REMT(32, x, y) => rem(true, true, x, y)
884 :	monnier	247
885 :	george	545	| T.ANDB(32, x, y) => binaryComm(I.ANDL, x, y)
886 :			| T.ORB(32, x, y) => binaryComm(I.ORL, x, y)
887 :			| T.XORB(32, x, y) => binaryComm(I.XORL, x, y)
888 :			| T.NOTB(32, x) => unary(I.NOTL, x)
889 :	monnier	247
890 :	george	545	| T.SRA(32, x, y) => shift(I.SARL, x, y)
891 :			| T.SRL(32, x, y) => shift(I.SHRL, x, y)
892 :			| T.SLL(32, x, y) => shift(I.SHLL, x, y)
893 :	monnier	247
894 :	george	545	| T.LOAD(8, ea, mem) => load8(ea, mem)
895 :			| T.LOAD(16, ea, mem) => load16(ea, mem)
896 :			| T.LOAD(32, ea, mem) => load32(ea, mem)
897 :	monnier	498
898 :	leunga	776	| T.SX(32,8,T.LOAD(8,ea,mem)) => load8s(ea, mem)
899 :			| T.SX(32,16,T.LOAD(16,ea,mem)) => load16s(ea, mem)
900 :			| T.ZX(32,8,T.LOAD(8,ea,mem)) => load8(ea, mem)
901 :	leunga	779	| T.ZX(32,16,T.LOAD(16,ea,mem)) => load16(ea, mem)
902 :	leunga	776
903 :	george	545	| T.COND(32, T.CMP(ty, cc, t1, t2), T.LI yes, T.LI no) =>
904 :	leunga	583	setcc(ty, cc, t1, t2, toInt32 yes, toInt32 no)
905 :	george	545	| T.COND(32, T.CMP(ty, cc, t1, t2), yes, no) =>
906 :			(case !arch of (* PentiumPro and higher has CMOVcc *)
907 :			Pentium => unknownExp exp
908 :			| _ => cmovcc(ty, cc, t1, t2, yes, no)
909 :			)
910 :			| T.LET(s,e) => (doStmt s; doExpr(e, rd, an))
911 :			| T.MARK(e, A.MARKREG f) => (f rd; doExpr(e, rd, an))
912 :			| T.MARK(e, a) => doExpr(e, rd, a::an)
913 :			| T.PRED(e,c) => doExpr(e, rd, A.CTRLUSE c::an)
914 :	george	555	| T.REXT e =>
915 :			ExtensionComp.compileRext (reducer()) {e=e, rd=rd, an=an}
916 :	george	545	(* simplify and try again *)
917 :			| exp => unknownExp exp
918 :			end (* doExpr *)
919 :	monnier	247
920 :	george	545	(* generate an expression and return its result register
921 :			* If rewritePseudo is on, the result is guaranteed to be in a
922 :			* non memReg register
923 :			*)
924 :			and expr(exp as T.REG(_, rd)) =
925 :			if isMemReg rd then genExpr exp else rd
926 :			| expr exp = genExpr exp
927 :	monnier	247
928 :	george	545	and genExpr exp =
929 :			let val rd = newReg() in doExpr(exp, rd, []); rd end
930 :	monnier	247
931 :	george	545	(* Compare an expression with zero.
932 :			* On the x86, TEST is superior to AND for doing the same thing,
933 :			* since it doesn't need to write out the result in a register.
934 :			*)
935 :	leunga	695	and cmpWithZero(cc as (T.EQ | T.NE), e as T.ANDB(ty, a, b), an) =
936 :	george	545	(case ty of
937 :	leunga	695	8 => test(I.TESTB, a, b, an)
938 :			| 16 => test(I.TESTW, a, b, an)
939 :			| 32 => test(I.TESTL, a, b, an)
940 :			| _ => doExpr(e, newReg(), an);
941 :			cc)
942 :			| cmpWithZero(cc, e, an) =
943 :			let val e =
944 :			case e of (* hack to disable the lea optimization XXX *)
945 :			T.ADD(_, a, b) => T.ADD(0, a, b)
946 :			| e => e
947 :			in doExpr(e, newReg(), an); cc end
948 :	monnier	247
949 :	george	545	(* Emit a test.
950 :			* The available modes are
951 :			* r/m, r
952 :			* r/m, imm
953 :			* On selecting the right instruction: TESTL/TESTW/TESTB.
954 :			* When anding an operand with a constant
955 :			* that fits within 8 (or 16) bits, it is possible to use TESTB,
956 :			* (or TESTW) instead of TESTL. Because x86 is little endian,
957 :			* this works for memory operands too. However, with TESTB, it is
958 :			* not possible to use registers other than
959 :			* AL, CL, BL, DL, and AH, CH, BH, DH. So, the best way is to
960 :			* perform register allocation first, and if the operand registers
961 :			* are one of EAX, ECX, EBX, or EDX, replace the TESTL instruction
962 :			* by TESTB.
963 :			*)
964 :	leunga	695	and test(testopcode, a, b, an) =
965 :	george	545	let val (_, opnd1, opnd2) = commuteComparison(T.EQ, true, a, b)
966 :			(* translate r, r/m => r/m, r *)
967 :			val (opnd1, opnd2) =
968 :			if isMemOpnd opnd2 then (opnd2, opnd1) else (opnd1, opnd2)
969 :	leunga	695	in mark(testopcode{lsrc=opnd1, rsrc=opnd2}, an)
970 :	george	545	end
971 :	monnier	247
972 :	leunga	815	(* %eflags <- src *)
973 :			and moveToEflags src =
974 :	george	889	if CB.sameColor(src, C.eflags) then ()
975 :	leunga	815	else (move(I.Direct src, eax); emit(I.LAHF))
976 :
977 :			(* dst <- %eflags *)
978 :			and moveFromEflags dst =
979 :	george	889	if CB.sameColor(dst, C.eflags) then ()
980 :	leunga	815	else (emit(I.SAHF); move(eax, I.Direct dst))
981 :
982 :	george	545	(* generate a condition code expression
983 :	leunga	744	* The zero is for setting the condition code!
984 :			* I have no idea why this is used.
985 :			*)
986 :			and doCCexpr(T.CMP(ty, cc, t1, t2), rd, an) =
987 :	leunga	815	(cmp(false, ty, cc, t1, t2, an);
988 :			moveFromEflags rd
989 :			)
990 :			| doCCexpr(T.CC(cond,rs), rd, an) =
991 :	george	889	if CB.sameColor(rs,C.eflags) orelse CB.sameColor(rd,C.eflags) then
992 :	leunga	815	(moveToEflags rs; moveFromEflags rd)
993 :	leunga	744	else
994 :	leunga	815	move'(I.Direct rs, I.Direct rd, an)
995 :	george	545	| doCCexpr(T.CCMARK(e,A.MARKREG f),rd,an) = (f rd; doCCexpr(e,rd,an))
996 :			| doCCexpr(T.CCMARK(e,a), rd, an) = doCCexpr(e,rd,a::an)
997 :			| doCCexpr(T.CCEXT e, cd, an) =
998 :	george	555	ExtensionComp.compileCCext (reducer()) {e=e, ccd=cd, an=an}
999 :	george	545	| doCCexpr _ = error "doCCexpr"
1000 :	monnier	247
1001 :	george	545	and ccExpr e = error "ccExpr"
1002 :	monnier	247
1003 :	george	545	(* generate a comparison and sets the condition code;
1004 :			* return the actual cc used. If the flag swapable is true,
1005 :			* we can also reorder the operands.
1006 :			*)
1007 :			and cmp(swapable, ty, cc, t1, t2, an) =
1008 :	leunga	695	(* == and <> can be always be reordered *)
1009 :			let val swapable = swapable orelse cc = T.EQ orelse cc = T.NE
1010 :			in (* Sometimes the comparison is not necessary because
1011 :			* the bits are already set!
1012 :			*)
1013 :			if isZero t1 andalso setZeroBit2 t2 then
1014 :			if swapable then
1015 :			cmpWithZero(T.Basis.swapCond cc, t2, an)
1016 :			else (* can't reorder the comparison! *)
1017 :			genCmp(ty, false, cc, t1, t2, an)
1018 :			else if isZero t2 andalso setZeroBit2 t1 then
1019 :			cmpWithZero(cc, t1, an)
1020 :			else genCmp(ty, swapable, cc, t1, t2, an)
1021 :			end
1022 :	monnier	247
1023 :	george	545	(* Give a and b which are the operands to a comparison (or test)
1024 :			* Return the appropriate condition code and operands.
1025 :			* The available modes are:
1026 :			* r/m, imm
1027 :			* r/m, r
1028 :			* r, r/m
1029 :			*)
1030 :			and commuteComparison(cc, swapable, a, b) =
1031 :			let val (opnd1, opnd2) = (operand a, operand b)
1032 :			in (* Try to fold in the operands whenever possible *)
1033 :			case (isImmediate opnd1, isImmediate opnd2) of
1034 :			(true, true) => (cc, moveToReg opnd1, opnd2)
1035 :			| (true, false) =>
1036 :			if swapable then (T.Basis.swapCond cc, opnd2, opnd1)
1037 :			else (cc, moveToReg opnd1, opnd2)
1038 :			| (false, true) => (cc, opnd1, opnd2)
1039 :			| (false, false) =>
1040 :			(case (opnd1, opnd2) of
1041 :			(_, I.Direct _) => (cc, opnd1, opnd2)
1042 :			| (I.Direct _, _) => (cc, opnd1, opnd2)
1043 :			| (_, _) => (cc, moveToReg opnd1, opnd2)
1044 :			)
1045 :			end
1046 :
1047 :			(* generate a real comparison; return the real cc used *)
1048 :			and genCmp(ty, swapable, cc, a, b, an) =
1049 :			let val (cc, opnd1, opnd2) = commuteComparison(cc, swapable, a, b)
1050 :			in mark(I.CMPL{lsrc=opnd1, rsrc=opnd2}, an); cc
1051 :			end
1052 :	monnier	247
1053 :	george	545	(* generate code for jumps *)
1054 :	leunga	775	and jmp(lexp as T.LABEL lab, labs, an) =
1055 :	george	545	mark(I.JMP(I.ImmedLabel lexp, [lab]), an)
1056 :	leunga	775	| jmp(T.LABEXP le, labs, an) = mark(I.JMP(I.ImmedLabel le, labs), an)
1057 :			| jmp(ea, labs, an) = mark(I.JMP(operand ea, labs), an)
1058 :	george	545
1059 :			(* convert mlrisc to cellset:
1060 :			*)
1061 :			and cellset mlrisc =
1062 :	jhr	900	let val addCCReg = CB.CellSet.add
1063 :	george	545	fun g([],acc) = acc
1064 :			| g(T.GPR(T.REG(_,r))::regs,acc) = g(regs,C.addReg(r,acc))
1065 :			| g(T.FPR(T.FREG(_,f))::regs,acc) = g(regs,C.addFreg(f,acc))
1066 :			| g(T.CCR(T.CC(_,cc))::regs,acc) = g(regs,addCCReg(cc,acc))
1067 :			| g(T.CCR(T.FCC(_,cc))::regs,acc) = g(regs,addCCReg(cc,acc))
1068 :			| g(_::regs, acc) = g(regs, acc)
1069 :			in g(mlrisc, C.empty) end
1070 :
1071 :			(* generate code for calls *)
1072 :	blume	839	and call(ea, flow, def, use, mem, cutsTo, an, pops) =
1073 :	leunga	815	let fun return(set, []) = set
1074 :			| return(set, a::an) =
1075 :			case #peek A.RETURN_ARG a of
1076 :	jhr	900	SOME r => return(CB.CellSet.add(r, set), an)
1077 :	leunga	815	| NONE => return(set, an)
1078 :	blume	839	in
1079 :			mark(I.CALL{opnd=operand ea,defs=cellset(def),uses=cellset(use),
1080 :			return=return(C.empty,an),cutsTo=cutsTo,mem=mem,
1081 :			pops=pops},an)
1082 :	leunga	815	end
1083 :	george	545
1084 :	leunga	815	(* generate code for integer stores; first move data to %eax
1085 :			* This is mainly because we can't allocate to registers like
1086 :			* ah, dl, dx etc.
1087 :			*)
1088 :			and genStore(mvOp, ea, d, mem, an) =
1089 :			let val src =
1090 :	george	545	case immedOrReg(operand d) of
1091 :			src as I.Direct r =>
1092 :	george	889	if CB.sameColor(r,C.eax)
1093 :	leunga	744	then src else (move(src, eax); eax)
1094 :	george	545	| src => src
1095 :	leunga	815	in mark(I.MOVE{mvOp=mvOp, src=src, dst=address(ea,mem)},an)
1096 :	george	545	end
1097 :	leunga	815
1098 :			(* generate code for 8-bit integer stores *)
1099 :			(* movb has to use %eax as source. Stupid x86! *)
1100 :			and store8(ea, d, mem, an) = genStore(I.MOVB, ea, d, mem, an)
1101 :	blume	818	and store16(ea, d, mem, an) =
1102 :			mark(I.MOVE{mvOp=I.MOVW, src=immedOrReg(operand d), dst=address(ea, mem)}, an)
1103 :	george	545	and store32(ea, d, mem, an) =
1104 :			move'(immedOrReg(operand d), address(ea, mem), an)
1105 :
1106 :			(* generate code for branching *)
1107 :			and branch(T.CMP(ty, cc, t1, t2), lab, an) =
1108 :			(* allow reordering of operands *)
1109 :			let val cc = cmp(true, ty, cc, t1, t2, [])
1110 :			in mark(I.JCC{cond=cond cc, opnd=immedLabel lab}, an) end
1111 :			| branch(T.FCMP(fty, fcc, t1, t2), lab, an) =
1112 :			fbranch(fty, fcc, t1, t2, lab, an)
1113 :			| branch(ccexp, lab, an) =
1114 :	leunga	744	(doCCexpr(ccexp, C.eflags, []);
1115 :	george	545	mark(I.JCC{cond=cond(Gen.condOf ccexp), opnd=immedLabel lab}, an)
1116 :			)
1117 :
1118 :			(* generate code for floating point compare and branch *)
1119 :			and fbranch(fty, fcc, t1, t2, lab, an) =
1120 :	leunga	731	let fun ignoreOrder (T.FREG _) = true
1121 :			| ignoreOrder (T.FLOAD _) = true
1122 :			| ignoreOrder (T.FMARK(e,_)) = ignoreOrder e
1123 :			| ignoreOrder _ = false
1124 :
1125 :			fun compare'() = (* Sethi-Ullman style *)
1126 :			(if ignoreOrder t1 orelse ignoreOrder t2 then
1127 :			(reduceFexp(fty, t2, []); reduceFexp(fty, t1, []))
1128 :			else (reduceFexp(fty, t1, []); reduceFexp(fty, t2, []);
1129 :			emit(I.FXCH{opnd=C.ST(1)}));
1130 :			emit(I.FUCOMPP);
1131 :			fcc
1132 :			)
1133 :
1134 :			fun compare''() =
1135 :			(* direct style *)
1136 :			(* Try to make lsrc the memory operand *)
1137 :			let val lsrc = foperand(fty, t1)
1138 :			val rsrc = foperand(fty, t2)
1139 :			val fsize = fsize fty
1140 :			fun cmp(lsrc, rsrc, fcc) =
1141 :			(emit(I.FCMP{fsize=fsize,lsrc=lsrc,rsrc=rsrc}); fcc)
1142 :			in case (lsrc, rsrc) of
1143 :			(I.FPR _, I.FPR _) => cmp(lsrc, rsrc, fcc)
1144 :			| (I.FPR _, mem) => cmp(mem,lsrc,T.Basis.swapFcond fcc)
1145 :			| (mem, I.FPR _) => cmp(lsrc, rsrc, fcc)
1146 :			| (lsrc, rsrc) => (* can't be both memory! *)
1147 :			let val ftmpR = newFreg()
1148 :			val ftmp = I.FPR ftmpR
1149 :			in emit(I.FMOVE{fsize=fsize,src=rsrc,dst=ftmp});
1150 :			cmp(lsrc, ftmp, fcc)
1151 :			end
1152 :			end
1153 :
1154 :			fun compare() =
1155 :			if enableFastFPMode andalso !fast_floating_point
1156 :			then compare''() else compare'()
1157 :
1158 :	george	545	fun andil i = emit(I.BINARY{binOp=I.ANDL,src=I.Immed(i),dst=eax})
1159 :	leunga	585	fun testil i = emit(I.TESTL{lsrc=eax,rsrc=I.Immed(i)})
1160 :	george	545	fun xoril i = emit(I.BINARY{binOp=I.XORL,src=I.Immed(i),dst=eax})
1161 :			fun cmpil i = emit(I.CMPL{rsrc=I.Immed(i), lsrc=eax})
1162 :			fun j(cc, lab) = mark(I.JCC{cond=cc, opnd=immedLabel lab},an)
1163 :			fun sahf() = emit(I.SAHF)
1164 :	leunga	731	fun branch(fcc) =
1165 :	george	545	case fcc
1166 :			of T.== => (andil 0x4400; xoril 0x4000; j(I.EQ, lab))
1167 :			| T.?<> => (andil 0x4400; xoril 0x4000; j(I.NE, lab))
1168 :			| T.? => (sahf(); j(I.P,lab))
1169 :			| T.<=> => (sahf(); j(I.NP,lab))
1170 :	leunga	585	| T.> => (testil 0x4500; j(I.EQ,lab))
1171 :			| T.?<= => (testil 0x4500; j(I.NE,lab))
1172 :			| T.>= => (testil 0x500; j(I.EQ,lab))
1173 :			| T.?< => (testil 0x500; j(I.NE,lab))
1174 :	george	545	| T.< => (andil 0x4500; cmpil 0x100; j(I.EQ,lab))
1175 :			| T.?>= => (andil 0x4500; cmpil 0x100; j(I.NE,lab))
1176 :			| T.<= => (andil 0x4100; cmpil 0x100; j(I.EQ,lab);
1177 :			cmpil 0x4000; j(I.EQ,lab))
1178 :	leunga	585	| T.?> => (sahf(); j(I.P,lab); testil 0x4100; j(I.EQ,lab))
1179 :			| T.<> => (testil 0x4400; j(I.EQ,lab))
1180 :			| T.?= => (testil 0x4400; j(I.NE,lab))
1181 :	george	545	| _ => error "fbranch"
1182 :			(*esac*)
1183 :	leunga	731	val fcc = compare()
1184 :			in emit I.FNSTSW;
1185 :			branch(fcc)
1186 :	monnier	411	end
1187 :	monnier	247
1188 :	leunga	731	(*========================================================
1189 :			* Floating point code generation starts here.
1190 :			* Some generic fp routines first.
1191 :			*========================================================*)
1192 :
1193 :			(* Can this tree be folded into the src operand of a floating point
1194 :			* operations?
1195 :			*)
1196 :			and foldableFexp(T.FREG _) = true
1197 :			| foldableFexp(T.FLOAD _) = true
1198 :			| foldableFexp(T.CVTI2F(_, (16 | 32), _)) = true
1199 :			| foldableFexp(T.CVTF2F(_, _, t)) = foldableFexp t
1200 :			| foldableFexp(T.FMARK(t, _)) = foldableFexp t
1201 :			| foldableFexp _ = false
1202 :
1203 :			(* Move integer e of size ty into a memory location.
1204 :			* Returns a quadruple:
1205 :			* (INTEGER,return ty,effect address of memory location,cleanup code)
1206 :			*)
1207 :			and convertIntToFloat(ty, e) =
1208 :			let val opnd = operand e
1209 :			in if isMemOpnd opnd andalso (ty = 16 orelse ty = 32)
1210 :			then (INTEGER, ty, opnd, [])
1211 :			else
1212 :	leunga	815	let val {instrs, tempMem, cleanup} =
1213 :			cvti2f{ty=ty, src=opnd, an=getAnnotations()}
1214 :	leunga	731	in emits instrs;
1215 :			(INTEGER, 32, tempMem, cleanup)
1216 :			end
1217 :			end
1218 :
1219 :			(*========================================================
1220 :			* Sethi-Ullman based floating point code generation as
1221 :			* implemented by Lal
1222 :			*========================================================*)
1223 :
1224 :	george	545	and fld(32, opnd) = I.FLDS opnd
1225 :			| fld(64, opnd) = I.FLDL opnd
1226 :	george	555	| fld(80, opnd) = I.FLDT opnd
1227 :	george	545	| fld _ = error "fld"
1228 :
1229 :	leunga	565	and fild(16, opnd) = I.FILD opnd
1230 :			| fild(32, opnd) = I.FILDL opnd
1231 :			| fild(64, opnd) = I.FILDLL opnd
1232 :			| fild _ = error "fild"
1233 :
1234 :			and fxld(INTEGER, ty, opnd) = fild(ty, opnd)
1235 :			| fxld(REAL, fty, opnd) = fld(fty, opnd)
1236 :
1237 :	george	545	and fstp(32, opnd) = I.FSTPS opnd
1238 :			| fstp(64, opnd) = I.FSTPL opnd
1239 :	george	555	| fstp(80, opnd) = I.FSTPT opnd
1240 :	george	545	| fstp _ = error "fstp"
1241 :
1242 :			(* generate code for floating point stores *)
1243 :	leunga	731	and fstore'(fty, ea, d, mem, an) =
1244 :	george	545	(case d of
1245 :			T.FREG(fty, fs) => emit(fld(fty, I.FDirect fs))
1246 :			| _ => reduceFexp(fty, d, []);
1247 :			mark(fstp(fty, address(ea, mem)), an)
1248 :			)
1249 :
1250 :	leunga	731	(* generate code for floating point loads *)
1251 :			and fload'(fty, ea, mem, fd, an) =
1252 :			let val ea = address(ea, mem)
1253 :			in mark(fld(fty, ea), an);
1254 :	george	889	if CB.sameColor(fd,ST0) then ()
1255 :	leunga	744	else emit(fstp(fty, I.FDirect fd))
1256 :	leunga	731	end
1257 :
1258 :			and fexpr' e = (reduceFexp(64, e, []); C.ST(0))
1259 :	george	545
1260 :			(* generate floating point expression and put the result in fd *)
1261 :	leunga	731	and doFexpr'(fty, T.FREG(_, fs), fd, an) =
1262 :	george	889	(if CB.sameColor(fs,fd) then ()
1263 :	george	1009	else mark'(I.COPY{k=CB.FP, sz=64, dst=[fd], src=[fs], tmp=NONE}, an)
1264 :	george	545	)
1265 :	leunga	731	| doFexpr'(_, T.FLOAD(fty, ea, mem), fd, an) =
1266 :			fload'(fty, ea, mem, fd, an)
1267 :			| doFexpr'(fty, T.FEXT fexp, fd, an) =
1268 :			(ExtensionComp.compileFext (reducer()) {e=fexp, fd=fd, an=an};
1269 :	george	889	if CB.sameColor(fd,ST0) then () else emit(fstp(fty, I.FDirect fd))
1270 :	leunga	731	)
1271 :			| doFexpr'(fty, e, fd, an) =
1272 :	george	545	(reduceFexp(fty, e, []);
1273 :	george	889	if CB.sameColor(fd,ST0) then ()
1274 :	leunga	744	else mark(fstp(fty, I.FDirect fd), an)
1275 :	george	545	)
1276 :
1277 :			(*
1278 :			* Generate floating point expression using Sethi-Ullman's scheme:
1279 :			* This function evaluates a floating point expression,
1280 :			* and put result in %ST(0).
1281 :			*)
1282 :			and reduceFexp(fty, fexp, an) =
1283 :	george	555	let val ST = I.ST(C.ST 0)
1284 :			val ST1 = I.ST(C.ST 1)
1285 :	leunga	593	val cleanupCode = ref [] : I.instruction list ref
1286 :	george	545
1287 :	leunga	565	datatype su_tree =
1288 :			LEAF of int * T.fexp * ans
1289 :			| BINARY of int * T.fty * fbinop * su_tree * su_tree * ans
1290 :			| UNARY of int * T.fty * I.funOp * su_tree * ans
1291 :			and fbinop = FADD | FSUB | FMUL | FDIV
1292 :			| FIADD | FISUB | FIMUL | FIDIV
1293 :			withtype ans = Annotations.annotations
1294 :	monnier	247
1295 :	leunga	565	fun label(LEAF(n, _, _)) = n
1296 :			| label(BINARY(n, _, _, _, _, _)) = n
1297 :			| label(UNARY(n, _, _, _, _)) = n
1298 :	george	545
1299 :	leunga	565	fun annotate(LEAF(n, x, an), a) = LEAF(n,x,a::an)
1300 :			| annotate(BINARY(n,t,b,x,y,an), a) = BINARY(n,t,b,x,y,a::an)
1301 :			| annotate(UNARY(n,t,u,x,an), a) = UNARY(n,t,u,x,a::an)
1302 :	george	545
1303 :	leunga	565	(* Generate expression tree with sethi-ullman numbers *)
1304 :			fun su(e as T.FREG _) = LEAF(1, e, [])
1305 :			| su(e as T.FLOAD _) = LEAF(1, e, [])
1306 :			| su(e as T.CVTI2F _) = LEAF(1, e, [])
1307 :			| su(T.CVTF2F(_, _, t)) = su t
1308 :			| su(T.FMARK(t, a)) = annotate(su t, a)
1309 :			| su(T.FABS(fty, t)) = suUnary(fty, I.FABS, t)
1310 :			| su(T.FNEG(fty, t)) = suUnary(fty, I.FCHS, t)
1311 :			| su(T.FSQRT(fty, t)) = suUnary(fty, I.FSQRT, t)
1312 :			| su(T.FADD(fty, t1, t2)) = suComBinary(fty,FADD,FIADD,t1,t2)
1313 :			| su(T.FMUL(fty, t1, t2)) = suComBinary(fty,FMUL,FIMUL,t1,t2)
1314 :			| su(T.FSUB(fty, t1, t2)) = suBinary(fty,FSUB,FISUB,t1,t2)
1315 :			| su(T.FDIV(fty, t1, t2)) = suBinary(fty,FDIV,FIDIV,t1,t2)
1316 :			| su _ = error "su"
1317 :
1318 :			(* Try to fold the the memory operand or integer conversion *)
1319 :			and suFold(e as T.FREG _) = (LEAF(0, e, []), false)
1320 :			| suFold(e as T.FLOAD _) = (LEAF(0, e, []), false)
1321 :			| suFold(e as T.CVTI2F(_,(16 | 32),_)) = (LEAF(0, e, []), true)
1322 :			| suFold(T.CVTF2F(_, _, t)) = suFold t
1323 :			| suFold(T.FMARK(t, a)) =
1324 :			let val (t, integer) = suFold t
1325 :			in (annotate(t, a), integer) end
1326 :			| suFold e = (su e, false)
1327 :
1328 :			(* Form unary tree *)
1329 :			and suUnary(fty, funary, t) =
1330 :			let val t = su t
1331 :			in UNARY(label t, fty, funary, t, [])
1332 :	george	545	end
1333 :	leunga	565
1334 :			(* Form binary tree *)
1335 :			and suBinary(fty, binop, ibinop, t1, t2) =
1336 :			let val t1 = su t1
1337 :			val (t2, integer) = suFold t2
1338 :			val n1 = label t1
1339 :			val n2 = label t2
1340 :			val n = if n1=n2 then n1+1 else Int.max(n1,n2)
1341 :			val myOp = if integer then ibinop else binop
1342 :			in BINARY(n, fty, myOp, t1, t2, [])
1343 :	george	545	end
1344 :	george	555
1345 :	leunga	565	(* Try to fold in the operand if possible.
1346 :			* This only applies to commutative operations.
1347 :			*)
1348 :			and suComBinary(fty, binop, ibinop, t1, t2) =
1349 :	leunga	731	let val (t1, t2) = if foldableFexp t2
1350 :			then (t1, t2) else (t2, t1)
1351 :	leunga	565	in suBinary(fty, binop, ibinop, t1, t2) end
1352 :
1353 :			and sameTree(LEAF(_, T.FREG(t1,f1), []),
1354 :	leunga	744	LEAF(_, T.FREG(t2,f2), [])) =
1355 :	george	889	t1 = t2 andalso CB.sameColor(f1,f2)
1356 :	leunga	565	| sameTree _ = false
1357 :
1358 :			(* Traverse tree and generate code *)
1359 :			fun gencode(LEAF(_, t, an)) = mark(fxld(leafEA t), an)
1360 :			| gencode(BINARY(_, _, binop, x, t2 as LEAF(0, y, a1), a2)) =
1361 :			let val _ = gencode x
1362 :			val (_, fty, src) = leafEA y
1363 :			fun gen(code) = mark(code, a1 @ a2)
1364 :			fun binary(oper32, oper64) =
1365 :			if sameTree(x, t2) then
1366 :			gen(I.FBINARY{binOp=oper64, src=ST, dst=ST})
1367 :	george	555	else
1368 :			let val oper =
1369 :	leunga	565	if isMemOpnd src then
1370 :			case fty of
1371 :			32 => oper32
1372 :			| 64 => oper64
1373 :			| _ => error "gencode: BINARY"
1374 :			else oper64
1375 :			in gen(I.FBINARY{binOp=oper, src=src, dst=ST}) end
1376 :			fun ibinary(oper16, oper32) =
1377 :			let val oper = case fty of
1378 :			16 => oper16
1379 :			| 32 => oper32
1380 :			| _ => error "gencode: IBINARY"
1381 :			in gen(I.FIBINARY{binOp=oper, src=src}) end
1382 :			in case binop of
1383 :			FADD => binary(I.FADDS, I.FADDL)
1384 :			| FSUB => binary(I.FDIVS, I.FSUBL)
1385 :			| FMUL => binary(I.FMULS, I.FMULL)
1386 :			| FDIV => binary(I.FDIVS, I.FDIVL)
1387 :			| FIADD => ibinary(I.FIADDS, I.FIADDL)
1388 :			| FISUB => ibinary(I.FIDIVS, I.FISUBL)
1389 :			| FIMUL => ibinary(I.FIMULS, I.FIMULL)
1390 :			| FIDIV => ibinary(I.FIDIVS, I.FIDIVL)
1391 :			end
1392 :			| gencode(BINARY(_, fty, binop, t1, t2, an)) =
1393 :			let fun doit(t1, t2, oper, operP, operRP) =
1394 :			let (* oper[P] => ST(1) := ST oper ST(1); [pop]
1395 :			* operR[P] => ST(1) := ST(1) oper ST; [pop]
1396 :			*)
1397 :			val n1 = label t1
1398 :			val n2 = label t2
1399 :			in if n1 < n2 andalso n1 <= 7 then
1400 :			(gencode t2;
1401 :			gencode t1;
1402 :			mark(I.FBINARY{binOp=operP, src=ST, dst=ST1}, an))
1403 :			else if n2 <= n1 andalso n2 <= 7 then
1404 :			(gencode t1;
1405 :			gencode t2;
1406 :			mark(I.FBINARY{binOp=operRP, src=ST, dst=ST1}, an))
1407 :			else
1408 :			let (* both labels > 7 *)
1409 :			val fs = I.FDirect(newFreg())
1410 :			in gencode t2;
1411 :			emit(fstp(fty, fs));
1412 :			gencode t1;
1413 :			mark(I.FBINARY{binOp=oper, src=fs, dst=ST}, an)
1414 :			end
1415 :			end
1416 :			in case binop of
1417 :			FADD => doit(t1,t2,I.FADDL,I.FADDP,I.FADDP)
1418 :			| FMUL => doit(t1,t2,I.FMULL,I.FMULP,I.FMULP)
1419 :			| FSUB => doit(t1,t2,I.FSUBL,I.FSUBP,I.FSUBRP)
1420 :			| FDIV => doit(t1,t2,I.FDIVL,I.FDIVP,I.FDIVRP)
1421 :	george	545	| _ => error "gencode.BINARY"
1422 :			end
1423 :	leunga	565	| gencode(UNARY(_, _, unaryOp, su, an)) =
1424 :			(gencode(su); mark(I.FUNARY(unaryOp),an))
1425 :
1426 :			(* Generate code for a leaf.
1427 :			* Returns the type and an effective address
1428 :			*)
1429 :			and leafEA(T.FREG(fty, f)) = (REAL, fty, I.FDirect f)
1430 :			| leafEA(T.FLOAD(fty, ea, mem)) = (REAL, fty, address(ea, mem))
1431 :	leunga	593	| leafEA(T.CVTI2F(_, 32, t)) = int2real(32, t)
1432 :			| leafEA(T.CVTI2F(_, 16, t)) = int2real(16, t)
1433 :			| leafEA(T.CVTI2F(_, 8, t)) = int2real(8, t)
1434 :	leunga	565	| leafEA _ = error "leafEA"
1435 :
1436 :	leunga	731	and int2real(ty, e) =
1437 :			let val (_, ty, ea, cleanup) = convertIntToFloat(ty, e)
1438 :			in cleanupCode := !cleanupCode @ cleanup;
1439 :			(INTEGER, ty, ea)
1440 :	george	545	end
1441 :	leunga	731
1442 :			in gencode(su fexp);
1443 :			emits(!cleanupCode)
1444 :	george	545	end (*reduceFexp*)
1445 :	leunga	731
1446 :			(*========================================================
1447 :			* This section generates 3-address style floating
1448 :			* point code.
1449 :			*========================================================*)
1450 :
1451 :			and isize 16 = I.I16
1452 :			| isize 32 = I.I32
1453 :			| isize _ = error "isize"
1454 :
1455 :			and fstore''(fty, ea, d, mem, an) =
1456 :			(floatingPointUsed := true;
1457 :			mark(I.FMOVE{fsize=fsize fty, dst=address(ea,mem),
1458 :			src=foperand(fty, d)},
1459 :			an)
1460 :			)
1461 :
1462 :			and fload''(fty, ea, mem, d, an) =
1463 :			(floatingPointUsed := true;
1464 :			mark(I.FMOVE{fsize=fsize fty, src=address(ea,mem),
1465 :			dst=RealReg d}, an)
1466 :			)
1467 :
1468 :			and fiload''(ity, ea, d, an) =
1469 :			(floatingPointUsed := true;
1470 :			mark(I.FILOAD{isize=isize ity, ea=ea, dst=RealReg d}, an)
1471 :			)
1472 :
1473 :			and fexpr''(e as T.FREG(_,f)) =
1474 :			if isFMemReg f then transFexpr e else f
1475 :			| fexpr'' e = transFexpr e
1476 :
1477 :			and transFexpr e =
1478 :			let val fd = newFreg() in doFexpr''(64, e, fd, []); fd end
1479 :
1480 :			(*
1481 :			* Process a floating point operand. Put operand in register
1482 :			* when possible. The operand should match the given fty.
1483 :			*)
1484 :			and foperand(fty, e as T.FREG(fty', f)) =
1485 :			if fty = fty' then RealReg f else I.FPR(fexpr'' e)
1486 :			| foperand(fty, T.CVTF2F(_, _, e)) =
1487 :			foperand(fty, e) (* nop on the x86 *)
1488 :			| foperand(fty, e as T.FLOAD(fty', ea, mem)) =
1489 :			(* fold operand when the precison matches *)
1490 :			if fty = fty' then address(ea, mem) else I.FPR(fexpr'' e)
1491 :			| foperand(fty, e) = I.FPR(fexpr'' e)
1492 :
1493 :			(*
1494 :			* Process a floating point operand.
1495 :			* Try to fold in a memory operand or conversion from an integer.
1496 :			*)
1497 :			and fioperand(T.FREG(fty,f)) = (REAL, fty, RealReg f, [])
1498 :			| fioperand(T.FLOAD(fty, ea, mem)) =
1499 :			(REAL, fty, address(ea, mem), [])
1500 :			| fioperand(T.CVTF2F(_, _, e)) = fioperand(e) (* nop on the x86 *)
1501 :			| fioperand(T.CVTI2F(_, ty, e)) = convertIntToFloat(ty, e)
1502 :			| fioperand(T.FMARK(e,an)) = fioperand(e) (* XXX *)
1503 :			| fioperand(e) = (REAL, 64, I.FPR(fexpr'' e), [])
1504 :
1505 :			(* Generate binary operator. Since the real binary operators
1506 :			* does not take memory as destination, we also ensure this
1507 :			* does not happen.
1508 :			*)
1509 :			and fbinop(targetFty,
1510 :			binOp, binOpR, ibinOp, ibinOpR, lsrc, rsrc, fd, an) =
1511 :			(* Put the mem operand in rsrc *)
1512 :			let val _ = floatingPointUsed := true;
1513 :			fun isMemOpnd(T.FREG(_, f)) = isFMemReg f
1514 :			| isMemOpnd(T.FLOAD _) = true
1515 :			| isMemOpnd(T.CVTI2F(_, (16 | 32), _)) = true
1516 :			| isMemOpnd(T.CVTF2F(_, _, t)) = isMemOpnd t
1517 :			| isMemOpnd(T.FMARK(t, _)) = isMemOpnd t
1518 :			| isMemOpnd _ = false
1519 :			val (binOp, ibinOp, lsrc, rsrc) =
1520 :			if isMemOpnd lsrc then (binOpR, ibinOpR, rsrc, lsrc)
1521 :			else (binOp, ibinOp, lsrc, rsrc)
1522 :			val lsrc = foperand(targetFty, lsrc)
1523 :			val (kind, fty, rsrc, code) = fioperand(rsrc)
1524 :			fun dstMustBeFreg f =
1525 :			if targetFty <> 64 then
1526 :			let val tmpR = newFreg()
1527 :			val tmp = I.FPR tmpR
1528 :			in mark(f tmp, an);
1529 :			emit(I.FMOVE{fsize=fsize targetFty,
1530 :			src=tmp, dst=RealReg fd})
1531 :			end
1532 :			else mark(f(RealReg fd), an)
1533 :			in case kind of
1534 :			REAL =>
1535 :			dstMustBeFreg(fn dst =>
1536 :			I.FBINOP{fsize=fsize fty, binOp=binOp,
1537 :			lsrc=lsrc, rsrc=rsrc, dst=dst})
1538 :			| INTEGER =>
1539 :			(dstMustBeFreg(fn dst =>
1540 :			I.FIBINOP{isize=isize fty, binOp=ibinOp,
1541 :			lsrc=lsrc, rsrc=rsrc, dst=dst});
1542 :			emits code
1543 :			)
1544 :			end
1545 :	george	545
1546 :	leunga	731	and funop(fty, unOp, src, fd, an) =
1547 :			let val src = foperand(fty, src)
1548 :			in mark(I.FUNOP{fsize=fsize fty,
1549 :			unOp=unOp, src=src, dst=RealReg fd},an)
1550 :			end
1551 :
1552 :			and doFexpr''(fty, e, fd, an) =
1553 :			case e of
1554 :	george	889	T.FREG(_,fs) => if CB.sameColor(fs,fd) then ()
1555 :	leunga	731	else fcopy''(fty, [fd], [fs], an)
1556 :			(* Stupid x86 does everything as 80-bits internally. *)
1557 :
1558 :			(* Binary operators *)
1559 :			| T.FADD(_, a, b) => fbinop(fty,
1560 :			I.FADDL, I.FADDL, I.FIADDL, I.FIADDL,
1561 :			a, b, fd, an)
1562 :			| T.FSUB(_, a, b) => fbinop(fty,
1563 :			I.FSUBL, I.FSUBRL, I.FISUBL, I.FISUBRL,
1564 :			a, b, fd, an)
1565 :			| T.FMUL(_, a, b) => fbinop(fty,
1566 :			I.FMULL, I.FMULL, I.FIMULL, I.FIMULL,
1567 :			a, b, fd, an)
1568 :			| T.FDIV(_, a, b) => fbinop(fty,
1569 :			I.FDIVL, I.FDIVRL, I.FIDIVL, I.FIDIVRL,
1570 :			a, b, fd, an)
1571 :
1572 :			(* Unary operators *)
1573 :			| T.FNEG(_, a) => funop(fty, I.FCHS, a, fd, an)
1574 :			| T.FABS(_, a) => funop(fty, I.FABS, a, fd, an)
1575 :			| T.FSQRT(_, a) => funop(fty, I.FSQRT, a, fd, an)
1576 :
1577 :			(* Load *)
1578 :			| T.FLOAD(fty,ea,mem) => fload''(fty, ea, mem, fd, an)
1579 :
1580 :			(* Type conversions *)
1581 :			| T.CVTF2F(_, _, e) => doFexpr''(fty, e, fd, an)
1582 :			| T.CVTI2F(_, ty, e) =>
1583 :			let val (_, ty, ea, cleanup) = convertIntToFloat(ty, e)
1584 :			in fiload''(ty, ea, fd, an);
1585 :			emits cleanup
1586 :			end
1587 :
1588 :			| T.FMARK(e,A.MARKREG f) => (f fd; doFexpr''(fty, e, fd, an))
1589 :			| T.FMARK(e, a) => doFexpr''(fty, e, fd, a::an)
1590 :			| T.FPRED(e, c) => doFexpr''(fty, e, fd, A.CTRLUSE c::an)
1591 :			| T.FEXT fexp =>
1592 :			ExtensionComp.compileFext (reducer()) {e=fexp, fd=fd, an=an}
1593 :			| _ => error("doFexpr''")
1594 :
1595 :			(*========================================================
1596 :			* Tie the two styles of fp code generation together
1597 :			*========================================================*)
1598 :			and fstore(fty, ea, d, mem, an) =
1599 :			if enableFastFPMode andalso !fast_floating_point
1600 :			then fstore''(fty, ea, d, mem, an)
1601 :			else fstore'(fty, ea, d, mem, an)
1602 :			and fload(fty, ea, d, mem, an) =
1603 :			if enableFastFPMode andalso !fast_floating_point
1604 :			then fload''(fty, ea, d, mem, an)
1605 :			else fload'(fty, ea, d, mem, an)
1606 :			and fexpr e =
1607 :			if enableFastFPMode andalso !fast_floating_point
1608 :			then fexpr'' e else fexpr' e
1609 :			and doFexpr(fty, e, fd, an) =
1610 :			if enableFastFPMode andalso !fast_floating_point
1611 :			then doFexpr''(fty, e, fd, an)
1612 :			else doFexpr'(fty, e, fd, an)
1613 :
1614 :	leunga	797	(*================================================================
1615 :			* Optimizations for x := x op y
1616 :			* Special optimizations:
1617 :			* Generate a binary operator, result must in memory.
1618 :			* The source must not be in memory
1619 :			*================================================================*)
1620 :			and binaryMem(binOp, src, dst, mem, an) =
1621 :			mark(I.BINARY{binOp=binOp, src=immedOrReg(operand src),
1622 :			dst=address(dst,mem)}, an)
1623 :			and unaryMem(unOp, opnd, mem, an) =
1624 :			mark(I.UNARY{unOp=unOp, opnd=address(opnd,mem)}, an)
1625 :
1626 :			and isOne(T.LI n) = n = one
1627 :			| isOne _ = false
1628 :
1629 :			(*
1630 :			* Perform optimizations based on recognizing
1631 :			* x := x op y or
1632 :			* x := y op x
1633 :			* first.
1634 :			*)
1635 :			and store(ty, ea, d, mem, an,
1636 :			{INC,DEC,ADD,SUB,NOT,NEG,SHL,SHR,SAR,OR,AND,XOR},
1637 :			doStore
1638 :			) =
1639 :			let fun default() = doStore(ea, d, mem, an)
1640 :			fun binary1(t, t', unary, binary, ea', x) =
1641 :			if t = ty andalso t' = ty then
1642 :			if MLTreeUtils.eqRexp(ea, ea') then
1643 :			if isOne x then unaryMem(unary, ea, mem, an)
1644 :			else binaryMem(binary, x, ea, mem, an)
1645 :			else default()
1646 :			else default()
1647 :			fun unary(t,unOp, ea') =
1648 :			if t = ty andalso MLTreeUtils.eqRexp(ea, ea') then
1649 :			unaryMem(unOp, ea, mem, an)
1650 :			else default()
1651 :			fun binary(t,t',binOp,ea',x) =
1652 :			if t = ty andalso t' = ty andalso
1653 :			MLTreeUtils.eqRexp(ea, ea') then
1654 :			binaryMem(binOp, x, ea, mem, an)
1655 :			else default()
1656 :
1657 :			fun binaryCom1(t,unOp,binOp,x,y) =
1658 :			if t = ty then
1659 :			let fun again() =
1660 :			case y of
1661 :			T.LOAD(ty',ea',_) =>
1662 :			if ty' = ty andalso MLTreeUtils.eqRexp(ea, ea') then
1663 :			if isOne x then unaryMem(unOp, ea, mem, an)
1664 :			else binaryMem(binOp,x,ea,mem,an)
1665 :			else default()
1666 :			| _ => default()
1667 :			in case x of
1668 :			T.LOAD(ty',ea',_) =>
1669 :			if ty' = ty andalso MLTreeUtils.eqRexp(ea, ea') then
1670 :			if isOne y then unaryMem(unOp, ea, mem, an)
1671 :			else binaryMem(binOp,y,ea,mem,an)
1672 :			else again()
1673 :			| _ => again()
1674 :			end
1675 :			else default()
1676 :
1677 :			fun binaryCom(t,binOp,x,y) =
1678 :			if t = ty then
1679 :			let fun again() =
1680 :			case y of
1681 :			T.LOAD(ty',ea',_) =>
1682 :			if ty' = ty andalso MLTreeUtils.eqRexp(ea, ea') then
1683 :			binaryMem(binOp,x,ea,mem,an)
1684 :			else default()
1685 :			| _ => default()
1686 :			in case x of
1687 :			T.LOAD(ty',ea',_) =>
1688 :			if ty' = ty andalso MLTreeUtils.eqRexp(ea, ea') then
1689 :			binaryMem(binOp,y,ea,mem,an)
1690 :			else again()
1691 :			| _ => again()
1692 :			end
1693 :			else default()
1694 :
1695 :			in case d of
1696 :			T.ADD(t,x,y) => binaryCom1(t,INC,ADD,x,y)
1697 :			| T.SUB(t,T.LOAD(t',ea',_),x) => binary1(t,t',DEC,SUB,ea',x)
1698 :			| T.ORB(t,x,y) => binaryCom(t,OR,x,y)
1699 :			| T.ANDB(t,x,y) => binaryCom(t,AND,x,y)
1700 :			| T.XORB(t,x,y) => binaryCom(t,XOR,x,y)
1701 :			| T.SLL(t,T.LOAD(t',ea',_),x) => binary(t,t',SHL,ea',x)
1702 :			| T.SRL(t,T.LOAD(t',ea',_),x) => binary(t,t',SHR,ea',x)
1703 :			| T.SRA(t,T.LOAD(t',ea',_),x) => binary(t,t',SAR,ea',x)
1704 :			| T.NEG(t,T.LOAD(t',ea',_)) => unary(t,NEG,ea')
1705 :			| T.NOTB(t,T.LOAD(t',ea',_)) => unary(t,NOT,ea')
1706 :			| _ => default()
1707 :			end (* store *)
1708 :
1709 :	george	545	(* generate code for a statement *)
1710 :			and stmt(T.MV(_, rd, e), an) = doExpr(e, rd, an)
1711 :			| stmt(T.FMV(fty, fd, e), an) = doFexpr(fty, e, fd, an)
1712 :			| stmt(T.CCMV(ccd, e), an) = doCCexpr(e, ccd, an)
1713 :			| stmt(T.COPY(_, dst, src), an) = copy(dst, src, an)
1714 :			| stmt(T.FCOPY(fty, dst, src), an) = fcopy(fty, dst, src, an)
1715 :	leunga	744	| stmt(T.JMP(e, labs), an) = jmp(e, labs, an)
1716 :	blume	839	| stmt(T.CALL{funct, targets, defs, uses, region, pops, ...}, an) =
1717 :			call(funct,targets,defs,uses,region,[],an, pops)
1718 :			| stmt(T.FLOW_TO(T.CALL{funct, targets, defs, uses, region, pops, ...},
1719 :	leunga	796	cutTo), an) =
1720 :	blume	839	call(funct,targets,defs,uses,region,cutTo,an, pops)
1721 :	george	545	| stmt(T.RET _, an) = mark(I.RET NONE, an)
1722 :	leunga	797	| stmt(T.STORE(8, ea, d, mem), an) =
1723 :			store(8, ea, d, mem, an, opcodes8, store8)
1724 :			| stmt(T.STORE(16, ea, d, mem), an) =
1725 :			store(16, ea, d, mem, an, opcodes16, store16)
1726 :			| stmt(T.STORE(32, ea, d, mem), an) =
1727 :			store(32, ea, d, mem, an, opcodes32, store32)
1728 :
1729 :	george	545	| stmt(T.FSTORE(fty, ea, d, mem), an) = fstore(fty, ea, d, mem, an)
1730 :	leunga	744	| stmt(T.BCC(cc, lab), an) = branch(cc, lab, an)
1731 :	george	545	| stmt(T.DEFINE l, _) = defineLabel l
1732 :			| stmt(T.ANNOTATION(s, a), an) = stmt(s, a::an)
1733 :	george	555	| stmt(T.EXT s, an) =
1734 :			ExtensionComp.compileSext (reducer()) {stm=s, an=an}
1735 :	george	545	| stmt(s, _) = doStmts(Gen.compileStm s)
1736 :
1737 :			and doStmt s = stmt(s, [])
1738 :			and doStmts ss = app doStmt ss
1739 :
1740 :			and beginCluster' _ =
1741 :			((* Must be cleared by the client.
1742 :			* if rewriteMemReg then memRegsUsed := 0w0 else ();
1743 :			*)
1744 :	leunga	731	floatingPointUsed := false;
1745 :			trapLabel := NONE;
1746 :			beginCluster 0
1747 :			)
1748 :	george	545	and endCluster' a =
1749 :	monnier	247	(case !trapLabel
1750 :	monnier	411	of NONE => ()
1751 :	george	545	| SOME(_, lab) => (defineLabel lab; emit(I.INTO))
1752 :	monnier	411	(*esac*);
1753 :	leunga	731	(* If floating point has been used allocate an extra
1754 :			* register just in case we didn't use any explicit register
1755 :			*)
1756 :			if !floatingPointUsed then (newFreg(); ())
1757 :			else ();
1758 :	george	545	endCluster(a)
1759 :			)
1760 :
1761 :			and reducer() =
1762 :	george	984	TS.REDUCER{reduceRexp = expr,
1763 :	george	545	reduceFexp = fexpr,
1764 :			reduceCCexp = ccExpr,
1765 :			reduceStm = stmt,
1766 :			operand = operand,
1767 :			reduceOperand = reduceOpnd,
1768 :			addressOf = fn e => address(e, I.Region.memory), (*XXX*)
1769 :	george	1009	emit = mark',
1770 :	george	545	instrStream = instrStream,
1771 :			mltreeStream = self()
1772 :			}
1773 :
1774 :			and self() =
1775 :	george	984	TS.S.STREAM
1776 :	leunga	815	{ beginCluster = beginCluster',
1777 :			endCluster = endCluster',
1778 :			emit = doStmt,
1779 :			pseudoOp = pseudoOp,
1780 :			defineLabel = defineLabel,
1781 :			entryLabel = entryLabel,
1782 :			comment = comment,
1783 :			annotation = annotation,
1784 :			getAnnotations = getAnnotations,
1785 :			exitBlock = fn mlrisc => exitBlock(cellset mlrisc)
1786 :	george	545	}
1787 :
1788 :			in self()
1789 :	monnier	247	end
1790 :
1791 :	george	545	end (* functor *)
1792 :
1793 :			end (* local *)

---

root@smlnj-gforge.cs.uchicago.edu	ViewVC Help
Powered by ViewVC 1.0.0