3 (declare (simple-string s))
4 (declare (optimize (speed 3) (safety 0) (debug 0)))
7 (dotimes (i (length s))
8 (when (eql (aref s i) #\1)
12 * On X86 I is represented as a tagged integer.
15 3: SLOT S!11[EDX] {SB-C::VECTOR-LENGTH 1 7} => t23[EAX]
16 4: MOVE t23[EAX] => t24[EBX]
18 --------------------------------------------------------------------------------
21 (declare (optimize (speed 3) (safety 0) (space 2) (debug 0)))
22 (declare (type (simple-array double-float 1) v))
24 (declare (type double-float s))
25 (dotimes (i (length v))
26 (setq s (+ s (aref v i))))
29 * Python does not combine + with AREF, so generates extra move and
32 * On X86 Python thinks that all FP registers are directly accessible
33 and emits costy MOVE ... => FR1.
35 --------------------------------------------------------------------------------
38 (declare (optimize (speed 3) (safety 0) (space 2))
40 (let ((v (make-list n)))
41 (setq v (make-array n))
44 * IR1 does not optimize away (MAKE-LIST N).
45 --------------------------------------------------------------------------------
48 (declare (optimize (speed 3) (safety 0) (space 2))
49 (type (simple-array base-char 1) v1 v2))
50 (dotimes (i (length v1))
51 (setf (aref v2 i) (aref v1 i))))
53 VOP DATA-VECTOR-SET/SIMPLE-STRING V2!14[EDI] t32[EAX] t30[S2]>t33[CL]
55 MOV #<TN t33[CL]>, #<TN t30[S2]>
56 MOV BYTE PTR [EDI+EAX+1], #<TN t33[CL]>
57 MOV #<TN t35[AL]>, #<TN t33[CL]>
58 MOV #<TN t34[S2]>, #<TN t35[AL]>
60 * The value of DATA-VECTOR-SET is not used, so there is no need in the
64 --------------------------------------------------------------------------------
67 (declare (optimize (speed 3) (safety 0) (debug 0)))
68 (declare (type (double-float 0d0 1d0) d))
69 (loop for i fixnum from 1 to 5
70 for x1 double-float = (sin d) ;;; !!!
71 do (loop for j fixnum from 1 to 4
72 sum x1 double-float)))
74 Without the marked declaration Python will use boxed representation for X1.
80 ;; use of X as DOUBLE-FLOAT
83 The initial binding is effectless, and without it X is of type
84 DOUBLE-FLOAT. Unhopefully, IR1 does not optimize away effectless
85 SETs/bindings, and IR2 does not perform type inference.
86 --------------------------------------------------------------------------------
87 #9 "Multi-path constant folding"
89 (if (= (cond ((irgh x) 0)
96 This code could be optimized to
102 --------------------------------------------------------------------------------
104 (inverted variant of #9)
107 (let ((y (sap-alien x c-string)))
111 It could be optimized to
113 (lambda (x) (list x x))
115 (if Y were used only once, the current compiler would optimize it)
116 --------------------------------------------------------------------------------
118 (typep (truly-the (simple-array * (*)) x) 'simple-vector)
121 --------------------------------------------------------------------------------
123 FAST-+/FIXNUM and similar should accept unboxed arguments in interests
124 of representation selection. Problem: inter-TN dependencies.
125 --------------------------------------------------------------------------------
127 The derived type of (/ (THE (DOUBLE-FLOAT (0D0)) X) (THE (DOUBLE-FLOAT
128 1D0) Y)) is (DOUBLE-FLOAT 0.0d0). While it might be reasonable, it is
129 better to derive (OR (MEMBER 0.0d0) (DOUBLE-FLOAT (0.0d0))).
130 --------------------------------------------------------------------------------
132 On the alpha, the system is reluctant to refer directly to a constant bignum,
133 preferring to load a large constant through a slow sequence of instructions,
134 then cons up a bignum for it:
137 (DECLARE (OPTIMIZE (SAFETY 1) (SPEED 3) (DEBUG 1))
138 (TYPE (INTEGER -10000 10000) A)
141 ((89 125 16) (ASH A (MIN 18 -706)))
142 (T (DPB -3 (BYTE 30 30) -1))))
143 --------------------------------------------------------------------------------
146 ((= i (the (integer 0 100) n)))
149 It is commonly expected for Python to derive (FIXNUMP I). (If ``='' is
150 replaced with ``>='', Python will do.)
151 --------------------------------------------------------------------------------
153 Type tests for (ARRAY BIT), (ARRAY T) and similar go through full
154 %TYPEP, even though it is relatively simple to establish the arrayness
155 of an object and also to obtain the element type of an array. As of
156 sbcl-0.8.12.30, this affects at least DUMP-OBJECT through
157 COMPOUND-OBJECT-P, and (LABELS MAYBE-EMIT-MAKE-LOAD-FORMS GROVEL)
158 through TYPEP UNBOXED-ARRAY, within the compiler itself.
159 --------------------------------------------------------------------------------
161 (defun-with-dx foo (x)
163 (let ((l (list nil nil)))
165 (setf (second l) (1- x))
168 (declare (dynamic-extent l))
171 Result of MAKE is not stack allocated.
172 --------------------------------------------------------------------------------
174 IR2 does not perform unused code flushing.
175 --------------------------------------------------------------------------------
177 a. Iterations on &REST lists could be rewritten with &MORE vectors.
178 b. Implement local unknown-values mv-call (useful for fast type checking).
179 --------------------------------------------------------------------------------
181 SBCL cannot derive upper bound for I and uses generic arithmetic here:
185 (dotimes (i (length l))
187 (map-foo (lambda (x) (if x (return t)))
192 (So the constraint propagator or a possible future SSA-convertor
193 should know the connection between an NLE and its CLEANUP.)
194 --------------------------------------------------------------------------------
196 Initialization of stack-allocated arrays is inefficient: we always
197 fill the vector with zeroes, even when it is not needed (as for
198 platforms with conservative GC or for arrays of unboxed objectes) and
199 is performed later explicitely.
201 (This is harder than it might look at first glance, as MAKE-ARRAY is smart
202 enough to eliminate something like ':initial-element 0'. Such an optimization
203 is valid if the vector is being allocated in the heap, but not if it is being
204 allocated on the stack. You could remove this optimization, but that makes
205 the heap-allocated case somewhat slower...)
207 To do this, extend ALLOCATE-VECTOR with ALLOW-JUNK argument, and when
208 stack allocating don't zero if it is true -- and probably ALLOW-JUNK iff
209 the vector is a specialized one (cannot have pointers.)
210 --------------------------------------------------------------------------------
212 a. Accessing raw slots in structure instances is more inefficient than
213 it could be; if we placed raw slots before the header word, we would
214 not need to do arithmetic at runtime to access them. (But beware:
215 this would complicate handling of the interior pointer).
217 b. (Also note that raw slots are currently disabled on HPPA)
219 c. Gencgc understands raw slots interleaved with tagged slots.
220 x86-64 supports this, and other backends should be made to as well.
221 --------------------------------------------------------------------------------
223 Python is overly zealous when converting high-level CL functions, such
224 as MIN/MAX, LOGBITP, and LOGTEST, to low-level CL functions. Reducing
225 Python's aggressiveness would make it easier to effect changes such as
228 * direct MIN/MAX on {SINGLE,DOUBLE}-FLOATs ({MIN,MAX}S{S,D})
231 * direct LOGBITP on word-sized integers and fixnums (BT + JC)
234 * branch-free MIN/MAX on word-sized integers and fixnums (floats could
235 be handled too, modulo safety considerations on the PPC)
238 * efficient LOGTESTs on word-sized integers and fixnums (TEST)
242 (The framework for this has been implemented as of 0.9.9.18; see the
243 vm-support-routine COMBINATION-IMPLEMENTATION-STYLE and its use in
244 src/compiler/ir1opt.lisp, IR1-OPTIMIZE-COMBINATION. The above
245 optimizations are left as an exercise for the reader.)
246 --------------------------------------------------------------------------------
248 The typecheck generated for a declaration like (integer 0 45) on x86 looks
251 ; 12B: F6C203 TEST DL, 3
253 ; 130: 8BC2 MOV EAX, EDX
254 ; 132: 83F800 CMP EAX, 0
256 ; 137: 8BC2 MOV EAX, EDX
257 ; 139: 3DB4000000 CMP EAX, 180
260 A better code sequence for this would be:
268 Doing an unsigned comparison means that, similarly to %CHECK-BOUND, we can
269 combine the <0 and >=bound tests. This sort of test is generated often
270 in SBCL and any array-based code that's serious about type-checking its
272 --------------------------------------------------------------------------------
274 The code for a vector bounds check on x86 (similarly on x86-64) where
275 the vector is in EDX and the index in EAX looks like:
277 ; 49: L0: 8B5AFD MOV EBX, [EDX-3]
278 ; 4C: 39C3 CMP EBX, EAX
281 because %CHECK-BOUND is used for bounds-checking any array dimension.
282 A more efficient specialization (%CHECK-BOUND/VECTOR) would produce:
287 Which is slightly shorter and avoids using a register.
288 --------------------------------------------------------------------------------
290 Reports from the Java camp indicate that using an SSE2-based
291 floating-point backend on x86 when possible is highly preferable to
292 using the x86 FP stack. It would be nice if SBCL included an SSE2-based
293 floating point backend with a compile-time option to switch between the
295 --------------------------------------------------------------------------------
300 (declare (type simple-vector a))
303 results in the following x86 code:
305 ; 115886E9: F7C703000000 TEST EDI, 3 ; no-arg-parsing entry point
307 ; 6F1: 8BC7 MOV EAX, EDI
308 ; 6F3: 83F800 CMP EAX, 0
310 ; 6F8: 8BC7 MOV EAX, EDI
311 ; 6FA: 3DF8FFFF7F CMP EAX, 2147483640
313 ; 701: L0: 8B057C865811 MOV EAX, [#x1158867C] ; '(MOD
315 ; 707: 0F0B0A BREAK 10 ; error trap
317 ; 70B: 1F BYTE #X1F ; OBJECT-NOT-TYPE-ERROR
318 ; 70C: FECE01 BYTE #XFE, #XCE, #X01 ; EDI
319 ; 70F: 0E BYTE #X0E ; EAX
320 ; 710: L1: 8B42FD MOV EAX, [EDX-3]
321 ; 713: 8BCF MOV ECX, EDI
322 ; 715: 39C8 CMP EAX, ECX
324 ; 719: 8B540A01 MOV EDX, [EDX+ECX+1]
326 ... plus the standard return sequence and some error blocks. The
327 `TEST EDI, 3' and associated comparisons are to ensure that `I' is a
328 positive fixnum. The associated comparisons are unnecessary, as the
329 %CHECK-BOUND VOP only requires its tested index to be a fixnum and takes
330 care of the negative fixnum case itself.
332 {HAIRY-,}DATA-VECTOR-REF are DEFKNOWN'd with EXPLICIT-CHECK, which would
333 seem to take care of this, but EXPLICIT-CHECK only seems to be used when
334 compiling calls to unknown functions or similar. Furthermore,
335 EXPLICIT-CHECK, as NJF understands it, doesn't have the right
336 semantics--it suppresses all type checking of arguments, whereas what we
337 really want is to ensure that the argument is a fixnum, but not check
339 --------------------------------------------------------------------------------
342 In #35, the CMP EAX, $foo instructions are all preceded by a MOV. They
343 appear to be unnecessary, but are necessary because in IR2, EDI is a
344 DESCRIPTOR-REG, whereas EAX is an ANY-REG--and the comparison VOPs only
345 accept ANY-REGs. Therefore, the MOVs are "necessary" to ensure that the
346 comparison VOP receives an TN of the appropriate storage class.
348 Obviously, it would be better if a) we only performed one MOV prior to
349 all three comparisons or b) eliminated the necessity of the MOV(s)
350 altogether. The former option is probably easier than the latter.
352 --------------------------------------------------------------------------------
355 (setf (subseq s1 start1 end1) (subseq s2 start2 end1))
357 could be transformed into
362 (replace s1 #:s2 :start1 start1 :end1 end1 :start2 #:start2 :end2 #:end2))
364 when the return value is unused, avoiding the need to cons up the new sequence.
366 --------------------------------------------------------------------------------
369 (let ((*foo* 42)) ...)
371 currently compiles to code that ensures the TLS index at runtime, which
372 is both a decently large chunk of code and unnecessary, as we could ensure
373 the TLS index at load-time as well.
374 [Note that x86-64 already does this.]
376 --------------------------------------------------------------------------------
379 When FTYPE is declared -- to say (function (t t t t t) t), and
380 function has a compiler-macro,
382 (apply #'foo 'x1 x2 'x3 more)
384 can be transformed into
386 (apply (lambda (x2 x4 x5) (foo 'x1 x2 'x3 x4 x5)) x2 more)
388 which allows compiler-macro-expansion for FOO. (Only constant
389 arguments can be moved inside the new lambda -- otherwise evaluation
392 --------------------------------------------------------------------------------
395 The unibyte external formats are written in a very generic way. Three
396 optimizations immediately applicable that could be automatically
399 (a) if the external format merely permutes the first 256 characters, a
400 constant-time lookup (rather than a binary search) could be
401 performed on output. This applies at least to EBCDIC, which
402 currently has a hand-rolled mapper instead.
404 (b) if there are no undefined characters corresponding to the 256
405 codes, then no error checking need be done on input.
407 (c) if there is a way to use particular bits of the exceptional
408 characters, constant-time output (rather than binary search) can
409 still be achieved as used to be done by the latin-9 external
410 format before 1.0.31.