2 -- The above warning supression flag is a temporary kludge.
3 -- While working on this module you are encouraged to remove it and fix
4 -- any warnings in the module. See
5 -- http://hackage.haskell.org/trac/ghc/wiki/Commentary/CodingStyle#Warnings
8 -----------------------------------------------------------------------------
12 -- (c) The University of Glasgow 2006
14 -----------------------------------------------------------------------------
17 cmmEliminateDeadBlocks,
23 #include "HsVersions.h"
42 import Compiler.Hoopl hiding (Unique)
44 -- -----------------------------------------------------------------------------
45 -- Eliminates dead blocks
48 We repeatedly expand the set of reachable blocks until we hit a
49 fixpoint, and then prune any blocks that were not in this set. This is
50 actually a required optimization, as dead blocks can cause problems
51 for invariants in the linear register allocator (and possibly other
55 -- Deep fold over statements could probably be abstracted out, but it
56 -- might not be worth the effort since OldCmm is moribund
57 cmmEliminateDeadBlocks :: [CmmBasicBlock] -> [CmmBasicBlock]
58 cmmEliminateDeadBlocks [] = []
59 cmmEliminateDeadBlocks blocks@(BasicBlock base_id _:_) =
60 let -- Calculate what's reachable from what block
61 reachableMap = foldl' f emptyUFM blocks -- lazy in values
62 where f m (BasicBlock block_id stmts) = addToUFM m block_id (reachableFrom stmts)
63 reachableFrom stmts = foldl stmt [] stmts
66 stmt m (CmmComment _) = m
67 stmt m (CmmAssign _ e) = expr m e
68 stmt m (CmmStore e1 e2) = expr (expr m e1) e2
69 stmt m (CmmCall c _ as _ _) = f (actuals m as) c
70 where f m (CmmCallee e _) = expr m e
72 stmt m (CmmBranch b) = b:m
73 stmt m (CmmCondBranch e b) = b:(expr m e)
74 stmt m (CmmSwitch e bs) = catMaybes bs ++ expr m e
75 stmt m (CmmJump e as) = expr (actuals m as) e
76 stmt m (CmmReturn as) = actuals m as
77 actuals m as = foldl' (\m h -> expr m (hintlessCmm h)) m as
78 -- We have to do a deep fold into CmmExpr because
79 -- there may be a BlockId in the CmmBlock literal.
80 expr m (CmmLit l) = lit m l
81 expr m (CmmLoad e _) = expr m e
83 expr m (CmmMachOp _ es) = foldl' expr m es
84 expr m (CmmStackSlot _ _) = m
85 expr m (CmmRegOff _ _) = m
86 lit m (CmmBlock b) = b:m
89 reachable = go [base_id] (setEmpty :: BlockSet)
92 | setMember x m = go xs m
93 | otherwise = go (add ++ xs) (setInsert x m)
94 where add = fromMaybe (panic "cmmEliminateDeadBlocks: unknown block")
95 (lookupUFM reachableMap x)
96 in filter (\(BasicBlock block_id _) -> setMember block_id reachable) blocks
98 -- -----------------------------------------------------------------------------
102 This pass inlines assignments to temporaries. Temporaries that are
103 only used once are unconditionally inlined. Temporaries that are used
104 two or more times are only inlined if they are assigned a literal. It
107 - count uses of each temporary
108 - for each temporary:
109 - attempt to push it forward to the statement that uses it
110 - only push forward past assignments to other temporaries
111 (assumes that temporaries are single-assignment)
112 - if we reach the statement that uses it, inline the rhs
113 and delete the original assignment.
115 [N.B. In the Quick C-- compiler, this optimization is achieved by a
116 combination of two dataflow passes: forward substitution (peephole
117 optimization) and dead-assignment elimination. ---NR]
119 Possible generalisations: here is an example from factorial
124 if (_smi != 0) goto cmK;
131 jump Fac_zdwfac_info;
133 We want to inline _smi and _smn. To inline _smn:
135 - we must be able to push forward past assignments to global regs.
136 We can do this if the rhs of the assignment we are pushing
137 forward doesn't refer to the global reg being assigned to; easy
142 - It is a trivial replacement, reg for reg, but it occurs more than
144 - We can inline trivial assignments even if the temporary occurs
145 more than once, as long as we don't eliminate the original assignment
146 (this doesn't help much on its own).
147 - We need to be able to propagate the assignment forward through jumps;
148 if we did this, we would find that it can be inlined safely in all
152 countUses :: UserOfLocalRegs a => a -> UniqFM Int
153 countUses a = foldRegsUsed (\m r -> addToUFM m r (count m r + 1)) emptyUFM a
154 where count m r = lookupWithDefaultUFM m (0::Int) r
156 cmmMiniInline :: [CmmBasicBlock] -> [CmmBasicBlock]
157 cmmMiniInline blocks = map do_inline blocks
158 where do_inline (BasicBlock id stmts)
159 = BasicBlock id (cmmMiniInlineStmts (countUses blocks) stmts)
161 cmmMiniInlineStmts :: UniqFM Int -> [CmmStmt] -> [CmmStmt]
162 cmmMiniInlineStmts uses [] = []
163 cmmMiniInlineStmts uses (stmt@(CmmAssign (CmmLocal (LocalReg u _)) expr@(CmmLit _)) : stmts)
164 -- not used: just discard this assignment
165 | Nothing <- lookupUFM uses u
166 = cmmMiniInlineStmts uses stmts
168 -- used: try to inline at all the use sites
169 | Just n <- lookupUFM uses u
172 trace ("nativeGen: inlining " ++ showSDoc (pprStmt stmt)) $
174 case lookForInlineLit u expr stmts of
176 | n == m -> cmmMiniInlineStmts (delFromUFM uses u) stmts'
178 stmt : cmmMiniInlineStmts (adjustUFM (\x -> x - m) uses u) stmts'
180 cmmMiniInlineStmts uses (stmt@(CmmAssign (CmmLocal (LocalReg u _)) expr : stmts))
181 -- not used at all: just discard this assignment
182 | Nothing <- lookupUFM uses u
183 = cmmMiniInlineStmts uses stmts
185 -- used once: try to inline at the use site
186 | Just 1 <- lookupUFM uses u,
187 Just stmts' <- lookForInline u expr stmts
190 trace ("nativeGen: inlining " ++ showSDoc (pprStmt stmt)) $
192 cmmMiniInlineStmts uses stmts'
194 cmmMiniInlineStmts uses (stmt:stmts)
195 = stmt : cmmMiniInlineStmts uses stmts
197 -- | Takes a register, a 'CmmLit' expression assigned to that
198 -- register, and a list of statements. Inlines the expression at all
199 -- use sites of the register. Returns the number of substituations
200 -- made and the, possibly modified, list of statements.
201 lookForInlineLit :: Unique -> CmmExpr -> [CmmStmt] -> (Int, [CmmStmt])
202 lookForInlineLit _ _ [] = (0, [])
203 lookForInlineLit u expr stmts@(stmt : rest)
204 | Just n <- lookupUFM (countUses stmt) u
205 = case lookForInlineLit u expr rest of
206 (m, stmts) -> let z = n + m
207 in z `seq` (z, inlineStmt u expr stmt : stmts)
210 = case lookForInlineLit u expr rest of
211 (n, stmts) -> (n, stmt : stmts)
216 -- We skip over assignments to registers, unless the register
217 -- being assigned to is the one we're inlining.
218 ok_to_skip = case stmt of
219 CmmAssign (CmmLocal r@(LocalReg u' _)) _ | u' == u -> False
222 lookForInline u expr stmts = lookForInline' u expr regset stmts
223 where regset = foldRegsUsed extendRegSet emptyRegSet expr
225 lookForInline' u expr regset (stmt : rest)
226 | Just 1 <- lookupUFM (countUses stmt) u, ok_to_inline
227 = Just (inlineStmt u expr stmt : rest)
230 = case lookForInline' u expr regset rest of
232 Just stmts -> Just (stmt:stmts)
238 -- we don't inline into CmmCall if the expression refers to global
239 -- registers. This is a HACK to avoid global registers clashing with
240 -- C argument-passing registers, really the back-end ought to be able
241 -- to handle it properly, but currently neither PprC nor the NCG can
242 -- do it. See also CgForeignCall:load_args_into_temps.
243 ok_to_inline = case stmt of
244 CmmCall{} -> hasNoGlobalRegs expr
247 -- Expressions aren't side-effecting. Temporaries may or may not
248 -- be single-assignment depending on the source (the old code
249 -- generator creates single-assignment code, but hand-written Cmm
250 -- and Cmm from the new code generator is not single-assignment.)
251 -- So we do an extra check to make sure that the register being
252 -- changed is not one we were relying on. I don't know how much of a
253 -- performance hit this is (we have to create a regset for every
254 -- instruction.) -- EZY
255 ok_to_skip = case stmt of
258 CmmAssign (CmmLocal r@(LocalReg u' _)) rhs | u' /= u && not (r `elemRegSet` regset) -> True
259 CmmAssign g@(CmmGlobal _) rhs -> not (g `regUsedIn` expr)
263 inlineStmt :: Unique -> CmmExpr -> CmmStmt -> CmmStmt
264 inlineStmt u a (CmmAssign r e) = CmmAssign r (inlineExpr u a e)
265 inlineStmt u a (CmmStore e1 e2) = CmmStore (inlineExpr u a e1) (inlineExpr u a e2)
266 inlineStmt u a (CmmCall target regs es srt ret)
267 = CmmCall (infn target) regs es' srt ret
268 where infn (CmmCallee fn cconv) = CmmCallee (inlineExpr u a fn) cconv
269 infn (CmmPrim p) = CmmPrim p
270 es' = [ (CmmHinted (inlineExpr u a e) hint) | (CmmHinted e hint) <- es ]
271 inlineStmt u a (CmmCondBranch e d) = CmmCondBranch (inlineExpr u a e) d
272 inlineStmt u a (CmmSwitch e d) = CmmSwitch (inlineExpr u a e) d
273 inlineStmt u a (CmmJump e d) = CmmJump (inlineExpr u a e) d
274 inlineStmt u a other_stmt = other_stmt
276 inlineExpr :: Unique -> CmmExpr -> CmmExpr -> CmmExpr
277 inlineExpr u a e@(CmmReg (CmmLocal (LocalReg u' _)))
280 inlineExpr u a e@(CmmRegOff (CmmLocal (LocalReg u' rep)) off)
281 | u == u' = CmmMachOp (MO_Add width) [a, CmmLit (CmmInt (fromIntegral off) width)]
284 width = typeWidth rep
285 inlineExpr u a (CmmLoad e rep) = CmmLoad (inlineExpr u a e) rep
286 inlineExpr u a (CmmMachOp op es) = CmmMachOp op (map (inlineExpr u a) es)
287 inlineExpr u a other_expr = other_expr
289 -- -----------------------------------------------------------------------------
290 -- MachOp constant folder
292 -- Now, try to constant-fold the MachOps. The arguments have already
293 -- been optimized and folded.
296 :: MachOp -- The operation from an CmmMachOp
297 -> [CmmExpr] -- The optimized arguments
300 cmmMachOpFold op arg@[CmmLit (CmmInt x rep)]
302 MO_S_Neg r -> CmmLit (CmmInt (-x) rep)
303 MO_Not r -> CmmLit (CmmInt (complement x) rep)
305 -- these are interesting: we must first narrow to the
306 -- "from" type, in order to truncate to the correct size.
307 -- The final narrow/widen to the destination type
308 -- is implicit in the CmmLit.
309 MO_SF_Conv from to -> CmmLit (CmmFloat (fromInteger x) to)
310 MO_SS_Conv from to -> CmmLit (CmmInt (narrowS from x) to)
311 MO_UU_Conv from to -> CmmLit (CmmInt (narrowU from x) to)
313 _ -> panic "cmmMachOpFold: unknown unary op"
316 -- Eliminate conversion NOPs
317 cmmMachOpFold (MO_SS_Conv rep1 rep2) [x] | rep1 == rep2 = x
318 cmmMachOpFold (MO_UU_Conv rep1 rep2) [x] | rep1 == rep2 = x
320 -- Eliminate nested conversions where possible
321 cmmMachOpFold conv_outer args@[CmmMachOp conv_inner [x]]
322 | Just (rep1,rep2,signed1) <- isIntConversion conv_inner,
323 Just (_, rep3,signed2) <- isIntConversion conv_outer
325 -- widen then narrow to the same size is a nop
326 _ | rep1 < rep2 && rep1 == rep3 -> x
327 -- Widen then narrow to different size: collapse to single conversion
328 -- but remember to use the signedness from the widening, just in case
329 -- the final conversion is a widen.
330 | rep1 < rep2 && rep2 > rep3 ->
331 cmmMachOpFold (intconv signed1 rep1 rep3) [x]
332 -- Nested widenings: collapse if the signedness is the same
333 | rep1 < rep2 && rep2 < rep3 && signed1 == signed2 ->
334 cmmMachOpFold (intconv signed1 rep1 rep3) [x]
335 -- Nested narrowings: collapse
336 | rep1 > rep2 && rep2 > rep3 ->
337 cmmMachOpFold (MO_UU_Conv rep1 rep3) [x]
339 CmmMachOp conv_outer args
341 isIntConversion (MO_UU_Conv rep1 rep2)
342 = Just (rep1,rep2,False)
343 isIntConversion (MO_SS_Conv rep1 rep2)
344 = Just (rep1,rep2,True)
345 isIntConversion _ = Nothing
347 intconv True = MO_SS_Conv
348 intconv False = MO_UU_Conv
350 -- ToDo: a narrow of a load can be collapsed into a narrow load, right?
351 -- but what if the architecture only supports word-sized loads, should
352 -- we do the transformation anyway?
354 cmmMachOpFold mop args@[CmmLit (CmmInt x xrep), CmmLit (CmmInt y _)]
356 -- for comparisons: don't forget to narrow the arguments before
357 -- comparing, since they might be out of range.
358 MO_Eq r -> CmmLit (CmmInt (if x_u == y_u then 1 else 0) wordWidth)
359 MO_Ne r -> CmmLit (CmmInt (if x_u /= y_u then 1 else 0) wordWidth)
361 MO_U_Gt r -> CmmLit (CmmInt (if x_u > y_u then 1 else 0) wordWidth)
362 MO_U_Ge r -> CmmLit (CmmInt (if x_u >= y_u then 1 else 0) wordWidth)
363 MO_U_Lt r -> CmmLit (CmmInt (if x_u < y_u then 1 else 0) wordWidth)
364 MO_U_Le r -> CmmLit (CmmInt (if x_u <= y_u then 1 else 0) wordWidth)
366 MO_S_Gt r -> CmmLit (CmmInt (if x_s > y_s then 1 else 0) wordWidth)
367 MO_S_Ge r -> CmmLit (CmmInt (if x_s >= y_s then 1 else 0) wordWidth)
368 MO_S_Lt r -> CmmLit (CmmInt (if x_s < y_s then 1 else 0) wordWidth)
369 MO_S_Le r -> CmmLit (CmmInt (if x_s <= y_s then 1 else 0) wordWidth)
371 MO_Add r -> CmmLit (CmmInt (x + y) r)
372 MO_Sub r -> CmmLit (CmmInt (x - y) r)
373 MO_Mul r -> CmmLit (CmmInt (x * y) r)
374 MO_U_Quot r | y /= 0 -> CmmLit (CmmInt (x_u `quot` y_u) r)
375 MO_U_Rem r | y /= 0 -> CmmLit (CmmInt (x_u `rem` y_u) r)
376 MO_S_Quot r | y /= 0 -> CmmLit (CmmInt (x `quot` y) r)
377 MO_S_Rem r | y /= 0 -> CmmLit (CmmInt (x `rem` y) r)
379 MO_And r -> CmmLit (CmmInt (x .&. y) r)
380 MO_Or r -> CmmLit (CmmInt (x .|. y) r)
381 MO_Xor r -> CmmLit (CmmInt (x `xor` y) r)
383 MO_Shl r -> CmmLit (CmmInt (x `shiftL` fromIntegral y) r)
384 MO_U_Shr r -> CmmLit (CmmInt (x_u `shiftR` fromIntegral y) r)
385 MO_S_Shr r -> CmmLit (CmmInt (x `shiftR` fromIntegral y) r)
387 other -> CmmMachOp mop args
396 -- When possible, shift the constants to the right-hand side, so that we
397 -- can match for strength reductions. Note that the code generator will
398 -- also assume that constants have been shifted to the right when
401 cmmMachOpFold op [x@(CmmLit _), y]
402 | not (isLit y) && isCommutableMachOp op
403 = cmmMachOpFold op [y, x]
405 -- Turn (a+b)+c into a+(b+c) where possible. Because literals are
406 -- moved to the right, it is more likely that we will find
407 -- opportunities for constant folding when the expression is
410 -- ToDo: this appears to introduce a quadratic behaviour due to the
411 -- nested cmmMachOpFold. Can we fix this?
413 -- Why do we check isLit arg1? If arg1 is a lit, it means that arg2
414 -- is also a lit (otherwise arg1 would be on the right). If we
415 -- put arg1 on the left of the rearranged expression, we'll get into a
416 -- loop: (x1+x2)+x3 => x1+(x2+x3) => (x2+x3)+x1 => x2+(x3+x1) ...
418 -- Also don't do it if arg1 is PicBaseReg, so that we don't separate the
419 -- PicBaseReg from the corresponding label (or label difference).
421 cmmMachOpFold mop1 [CmmMachOp mop2 [arg1,arg2], arg3]
422 | mop2 `associates_with` mop1
423 && not (isLit arg1) && not (isPicReg arg1)
424 = cmmMachOpFold mop2 [arg1, cmmMachOpFold mop1 [arg2,arg3]]
426 MO_Add{} `associates_with` MO_Sub{} = True
427 mop1 `associates_with` mop2 =
428 mop1 == mop2 && isAssociativeMachOp mop1
430 -- special case: (a - b) + c ==> a + (c - b)
431 cmmMachOpFold mop1@(MO_Add{}) [CmmMachOp mop2@(MO_Sub{}) [arg1,arg2], arg3]
432 | not (isLit arg1) && not (isPicReg arg1)
433 = cmmMachOpFold mop1 [arg1, cmmMachOpFold mop2 [arg3,arg2]]
435 -- Make a RegOff if we can
436 cmmMachOpFold (MO_Add _) [CmmReg reg, CmmLit (CmmInt n rep)]
437 = CmmRegOff reg (fromIntegral (narrowS rep n))
438 cmmMachOpFold (MO_Add _) [CmmRegOff reg off, CmmLit (CmmInt n rep)]
439 = CmmRegOff reg (off + fromIntegral (narrowS rep n))
440 cmmMachOpFold (MO_Sub _) [CmmReg reg, CmmLit (CmmInt n rep)]
441 = CmmRegOff reg (- fromIntegral (narrowS rep n))
442 cmmMachOpFold (MO_Sub _) [CmmRegOff reg off, CmmLit (CmmInt n rep)]
443 = CmmRegOff reg (off - fromIntegral (narrowS rep n))
445 -- Fold label(+/-)offset into a CmmLit where possible
447 cmmMachOpFold (MO_Add _) [CmmLit (CmmLabel lbl), CmmLit (CmmInt i rep)]
448 = CmmLit (CmmLabelOff lbl (fromIntegral (narrowU rep i)))
449 cmmMachOpFold (MO_Add _) [CmmLit (CmmInt i rep), CmmLit (CmmLabel lbl)]
450 = CmmLit (CmmLabelOff lbl (fromIntegral (narrowU rep i)))
451 cmmMachOpFold (MO_Sub _) [CmmLit (CmmLabel lbl), CmmLit (CmmInt i rep)]
452 = CmmLit (CmmLabelOff lbl (fromIntegral (negate (narrowU rep i))))
455 -- Comparison of literal with widened operand: perform the comparison
456 -- at the smaller width, as long as the literal is within range.
458 -- We can't do the reverse trick, when the operand is narrowed:
459 -- narrowing throws away bits from the operand, there's no way to do
460 -- the same comparison at the larger size.
462 #if i386_TARGET_ARCH || x86_64_TARGET_ARCH
463 -- powerPC NCG has a TODO for I8/I16 comparisons, so don't try
465 cmmMachOpFold cmp [CmmMachOp conv [x], CmmLit (CmmInt i _)]
466 | -- if the operand is widened:
467 Just (rep, signed, narrow_fn) <- maybe_conversion conv,
468 -- and this is a comparison operation:
469 Just narrow_cmp <- maybe_comparison cmp rep signed,
470 -- and the literal fits in the smaller size:
472 -- then we can do the comparison at the smaller size
473 = cmmMachOpFold narrow_cmp [x, CmmLit (CmmInt i rep)]
475 maybe_conversion (MO_UU_Conv from to)
477 = Just (from, False, narrowU)
478 maybe_conversion (MO_SS_Conv from to)
480 = Just (from, True, narrowS)
482 -- don't attempt to apply this optimisation when the source
483 -- is a float; see #1916
484 maybe_conversion _ = Nothing
486 -- careful (#2080): if the original comparison was signed, but
487 -- we were doing an unsigned widen, then we must do an
488 -- unsigned comparison at the smaller size.
489 maybe_comparison (MO_U_Gt _) rep _ = Just (MO_U_Gt rep)
490 maybe_comparison (MO_U_Ge _) rep _ = Just (MO_U_Ge rep)
491 maybe_comparison (MO_U_Lt _) rep _ = Just (MO_U_Lt rep)
492 maybe_comparison (MO_U_Le _) rep _ = Just (MO_U_Le rep)
493 maybe_comparison (MO_Eq _) rep _ = Just (MO_Eq rep)
494 maybe_comparison (MO_S_Gt _) rep True = Just (MO_S_Gt rep)
495 maybe_comparison (MO_S_Ge _) rep True = Just (MO_S_Ge rep)
496 maybe_comparison (MO_S_Lt _) rep True = Just (MO_S_Lt rep)
497 maybe_comparison (MO_S_Le _) rep True = Just (MO_S_Le rep)
498 maybe_comparison (MO_S_Gt _) rep False = Just (MO_U_Gt rep)
499 maybe_comparison (MO_S_Ge _) rep False = Just (MO_U_Ge rep)
500 maybe_comparison (MO_S_Lt _) rep False = Just (MO_U_Lt rep)
501 maybe_comparison (MO_S_Le _) rep False = Just (MO_U_Le rep)
502 maybe_comparison _ _ _ = Nothing
506 -- We can often do something with constants of 0 and 1 ...
508 cmmMachOpFold mop args@[x, y@(CmmLit (CmmInt 0 _))]
519 MO_Ne r | isComparisonExpr x -> x
520 MO_Eq r | Just x' <- maybeInvertCmmExpr x -> x'
521 MO_U_Gt r | isComparisonExpr x -> x
522 MO_S_Gt r | isComparisonExpr x -> x
523 MO_U_Lt r | isComparisonExpr x -> CmmLit (CmmInt 0 wordWidth)
524 MO_S_Lt r | isComparisonExpr x -> CmmLit (CmmInt 0 wordWidth)
525 MO_U_Ge r | isComparisonExpr x -> CmmLit (CmmInt 1 wordWidth)
526 MO_S_Ge r | isComparisonExpr x -> CmmLit (CmmInt 1 wordWidth)
527 MO_U_Le r | Just x' <- maybeInvertCmmExpr x -> x'
528 MO_S_Le r | Just x' <- maybeInvertCmmExpr x -> x'
529 other -> CmmMachOp mop args
531 cmmMachOpFold mop args@[x, y@(CmmLit (CmmInt 1 rep))]
536 MO_S_Rem r -> CmmLit (CmmInt 0 rep)
537 MO_U_Rem r -> CmmLit (CmmInt 0 rep)
538 MO_Ne r | Just x' <- maybeInvertCmmExpr x -> x'
539 MO_Eq r | isComparisonExpr x -> x
540 MO_U_Lt r | Just x' <- maybeInvertCmmExpr x -> x'
541 MO_S_Lt r | Just x' <- maybeInvertCmmExpr x -> x'
542 MO_U_Gt r | isComparisonExpr x -> CmmLit (CmmInt 0 wordWidth)
543 MO_S_Gt r | isComparisonExpr x -> CmmLit (CmmInt 0 wordWidth)
544 MO_U_Le r | isComparisonExpr x -> CmmLit (CmmInt 1 wordWidth)
545 MO_S_Le r | isComparisonExpr x -> CmmLit (CmmInt 1 wordWidth)
546 MO_U_Ge r | isComparisonExpr x -> x
547 MO_S_Ge r | isComparisonExpr x -> x
548 other -> CmmMachOp mop args
550 -- Now look for multiplication/division by powers of 2 (integers).
552 cmmMachOpFold mop args@[x, y@(CmmLit (CmmInt n _))]
555 | Just p <- exactLog2 n ->
556 cmmMachOpFold (MO_Shl rep) [x, CmmLit (CmmInt p rep)]
558 | Just p <- exactLog2 n ->
559 cmmMachOpFold (MO_U_Shr rep) [x, CmmLit (CmmInt p rep)]
561 | Just p <- exactLog2 n,
562 CmmReg _ <- x -> -- We duplicate x below, hence require
563 -- it is a reg. FIXME: remove this restriction.
564 -- shift right is not the same as quot, because it rounds
565 -- to minus infinity, whereasq quot rounds toward zero.
566 -- To fix this up, we add one less than the divisor to the
567 -- dividend if it is a negative number.
569 -- to avoid a test/jump, we use the following sequence:
570 -- x1 = x >> word_size-1 (all 1s if -ve, all 0s if +ve)
571 -- x2 = y & (divisor-1)
572 -- result = (x+x2) >>= log2(divisor)
573 -- this could be done a bit more simply using conditional moves,
574 -- but we're processor independent here.
576 -- we optimise the divide by 2 case slightly, generating
577 -- x1 = x >> word_size-1 (unsigned)
578 -- return = (x + x1) >>= log2(divisor)
580 bits = fromIntegral (widthInBits rep) - 1
581 shr = if p == 1 then MO_U_Shr rep else MO_S_Shr rep
582 x1 = CmmMachOp shr [x, CmmLit (CmmInt bits rep)]
583 x2 = if p == 1 then x1 else
584 CmmMachOp (MO_And rep) [x1, CmmLit (CmmInt (n-1) rep)]
585 x3 = CmmMachOp (MO_Add rep) [x, x2]
587 cmmMachOpFold (MO_S_Shr rep) [x3, CmmLit (CmmInt p rep)]
591 unchanged = CmmMachOp mop args
593 -- Anything else is just too hard.
595 cmmMachOpFold mop args = CmmMachOp mop args
597 -- -----------------------------------------------------------------------------
600 -- This algorithm for determining the $\log_2$ of exact powers of 2 comes
601 -- from GCC. It requires bit manipulation primitives, and we use GHC
602 -- extensions. Tough.
604 -- Used to be in MachInstrs --SDM.
605 -- ToDo: remove use of unboxery --SDM.
607 -- Unboxery removed in favor of FastInt; but is the function supposed to fail
608 -- on inputs >= 2147483648, or was that just an implementation artifact?
609 -- And is this speed-critical, or can we just use Integer operations
610 -- (including Data.Bits)?
613 exactLog2 :: Integer -> Maybe Integer
615 = if (x_ <= 0 || x_ >= 2147483648) then
618 case iUnbox (fromInteger x_) of { x ->
619 if (x `bitAndFastInt` negateFastInt x) /=# x then
622 Just (toInteger (iBox (pow2 x)))
625 pow2 x | x ==# _ILIT(1) = _ILIT(0)
626 | otherwise = _ILIT(1) +# pow2 (x `shiftR_FastInt` _ILIT(1))
629 -- -----------------------------------------------------------------------------
633 This is a simple pass that replaces tail-recursive functions like this:
648 the latter generates better C code, because the C compiler treats it
649 like a loop, and brings full loop optimisation to bear.
651 In my measurements this makes little or no difference to anything
652 except factorial, but what the hell.
655 cmmLoopifyForC :: RawCmmTop -> RawCmmTop
656 cmmLoopifyForC p@(CmmProc info entry_lbl
657 (ListGraph blocks@(BasicBlock top_id _ : _)))
658 | null info = p -- only if there's an info table, ignore case alts
660 -- pprTrace "jump_lbl" (ppr jump_lbl <+> ppr entry_lbl) $
661 CmmProc info entry_lbl (ListGraph blocks')
662 where blocks' = [ BasicBlock id (map do_stmt stmts)
663 | BasicBlock id stmts <- blocks ]
665 do_stmt (CmmJump (CmmLit (CmmLabel lbl)) _) | lbl == jump_lbl
669 jump_lbl | tablesNextToCode = entryLblToInfoLbl entry_lbl
670 | otherwise = entry_lbl
672 cmmLoopifyForC top = top
674 -- -----------------------------------------------------------------------------
677 isLit (CmmLit _) = True
680 isComparisonExpr :: CmmExpr -> Bool
681 isComparisonExpr (CmmMachOp op _) = isComparisonMachOp op
682 isComparisonExpr _other = False
684 isPicReg (CmmReg (CmmGlobal PicBaseReg)) = True