1 Target Independent Opportunities:
3 //===---------------------------------------------------------------------===//
5 Dead argument elimination should be enhanced to handle cases when an argument is
6 dead to an externally visible function. Though the argument can't be removed
7 from the externally visible function, the caller doesn't need to pass it in.
8 For example in this testcase:
10 void foo(int X) __attribute__((noinline));
11 void foo(int X) { sideeffect(); }
12 void bar(int A) { foo(A+1); }
16 define void @bar(i32 %A) nounwind ssp {
17 %0 = add nsw i32 %A, 1 ; <i32> [#uses=1]
18 tail call void @foo(i32 %0) nounwind noinline ssp
22 The add is dead, we could pass in 'i32 undef' instead. This occurs for C++
23 templates etc, which usually have linkonce_odr/weak_odr linkage, not internal
26 //===---------------------------------------------------------------------===//
28 With the recent changes to make the implicit def/use set explicit in
29 machineinstrs, we should change the target descriptions for 'call' instructions
30 so that the .td files don't list all the call-clobbered registers as implicit
31 defs. Instead, these should be added by the code generator (e.g. on the dag).
33 This has a number of uses:
35 1. PPC32/64 and X86 32/64 can avoid having multiple copies of call instructions
36 for their different impdef sets.
37 2. Targets with multiple calling convs (e.g. x86) which have different clobber
38 sets don't need copies of call instructions.
39 3. 'Interprocedural register allocation' can be done to reduce the clobber sets
42 //===---------------------------------------------------------------------===//
44 Make the PPC branch selector target independant
46 //===---------------------------------------------------------------------===//
48 Get the C front-end to expand hypot(x,y) -> llvm.sqrt(x*x+y*y) when errno and
49 precision don't matter (ffastmath). Misc/mandel will like this. :) This isn't
50 safe in general, even on darwin. See the libm implementation of hypot for
51 examples (which special case when x/y are exactly zero to get signed zeros etc
54 //===---------------------------------------------------------------------===//
56 Solve this DAG isel folding deficiency:
74 The problem is the store's chain operand is not the load X but rather
75 a TokenFactor of the load X and load Y, which prevents the folding.
77 There are two ways to fix this:
79 1. The dag combiner can start using alias analysis to realize that y/x
80 don't alias, making the store to X not dependent on the load from Y.
81 2. The generated isel could be made smarter in the case it can't
82 disambiguate the pointers.
84 Number 1 is the preferred solution.
86 This has been "fixed" by a TableGen hack. But that is a short term workaround
87 which will be removed once the proper fix is made.
89 //===---------------------------------------------------------------------===//
91 On targets with expensive 64-bit multiply, we could LSR this:
98 for (i = ...; ++i, tmp+=tmp)
101 This would be a win on ppc32, but not x86 or ppc64.
103 //===---------------------------------------------------------------------===//
105 Shrink: (setlt (loadi32 P), 0) -> (setlt (loadi8 Phi), 0)
107 //===---------------------------------------------------------------------===//
109 Reassociate should turn things like:
111 int factorial(int X) {
112 return X*X*X*X*X*X*X*X;
115 into llvm.powi calls, allowing the code generator to produce balanced
116 multiplication trees.
118 First, the intrinsic needs to be extended to support integers, and second the
119 code generator needs to be enhanced to lower these to multiplication trees.
121 //===---------------------------------------------------------------------===//
123 Interesting? testcase for add/shift/mul reassoc:
125 int bar(int x, int y) {
126 return x*x*x+y+x*x*x*x*x*y*y*y*y;
128 int foo(int z, int n) {
129 return bar(z, n) + bar(2*z, 2*n);
132 This is blocked on not handling X*X*X -> powi(X, 3) (see note above). The issue
133 is that we end up getting t = 2*X s = t*t and don't turn this into 4*X*X,
134 which is the same number of multiplies and is canonical, because the 2*X has
135 multiple uses. Here's a simple example:
137 define i32 @test15(i32 %X1) {
138 %B = mul i32 %X1, 47 ; X1*47
144 //===---------------------------------------------------------------------===//
146 Reassociate should handle the example in GCC PR16157:
148 extern int a0, a1, a2, a3, a4; extern int b0, b1, b2, b3, b4;
149 void f () { /* this can be optimized to four additions... */
150 b4 = a4 + a3 + a2 + a1 + a0;
151 b3 = a3 + a2 + a1 + a0;
156 This requires reassociating to forms of expressions that are already available,
157 something that reassoc doesn't think about yet.
160 //===---------------------------------------------------------------------===//
162 This function: (derived from GCC PR19988)
163 double foo(double x, double y) {
164 return ((x + 0.1234 * y) * (x + -0.1234 * y));
170 mulsd LCPI1_1(%rip), %xmm1
171 mulsd LCPI1_0(%rip), %xmm2
178 Reassociate should be able to turn it into:
180 double foo(double x, double y) {
181 return ((x + 0.1234 * y) * (x - 0.1234 * y));
184 Which allows the multiply by constant to be CSE'd, producing:
187 mulsd LCPI1_0(%rip), %xmm1
194 This doesn't need -ffast-math support at all. This is particularly bad because
195 the llvm-gcc frontend is canonicalizing the later into the former, but clang
196 doesn't have this problem.
198 //===---------------------------------------------------------------------===//
200 These two functions should generate the same code on big-endian systems:
202 int g(int *j,int *l) { return memcmp(j,l,4); }
203 int h(int *j, int *l) { return *j - *l; }
205 this could be done in SelectionDAGISel.cpp, along with other special cases,
208 //===---------------------------------------------------------------------===//
210 It would be nice to revert this patch:
211 http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20060213/031986.html
213 And teach the dag combiner enough to simplify the code expanded before
214 legalize. It seems plausible that this knowledge would let it simplify other
217 //===---------------------------------------------------------------------===//
219 For vector types, TargetData.cpp::getTypeInfo() returns alignment that is equal
220 to the type size. It works but can be overly conservative as the alignment of
221 specific vector types are target dependent.
223 //===---------------------------------------------------------------------===//
225 We should produce an unaligned load from code like this:
227 v4sf example(float *P) {
228 return (v4sf){P[0], P[1], P[2], P[3] };
231 //===---------------------------------------------------------------------===//
233 Add support for conditional increments, and other related patterns. Instead
238 je LBB16_2 #cond_next
249 //===---------------------------------------------------------------------===//
251 Combine: a = sin(x), b = cos(x) into a,b = sincos(x).
253 Expand these to calls of sin/cos and stores:
254 double sincos(double x, double *sin, double *cos);
255 float sincosf(float x, float *sin, float *cos);
256 long double sincosl(long double x, long double *sin, long double *cos);
258 Doing so could allow SROA of the destination pointers. See also:
259 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17687
261 This is now easily doable with MRVs. We could even make an intrinsic for this
262 if anyone cared enough about sincos.
264 //===---------------------------------------------------------------------===//
266 Turn this into a single byte store with no load (the other 3 bytes are
269 define void @test(i32* %P) {
271 %tmp14 = or i32 %tmp, 3305111552
272 %tmp15 = and i32 %tmp14, 3321888767
273 store i32 %tmp15, i32* %P
277 //===---------------------------------------------------------------------===//
279 quantum_sigma_x in 462.libquantum contains the following loop:
281 for(i=0; i<reg->size; i++)
283 /* Flip the target bit of each basis state */
284 reg->node[i].state ^= ((MAX_UNSIGNED) 1 << target);
287 Where MAX_UNSIGNED/state is a 64-bit int. On a 32-bit platform it would be just
288 so cool to turn it into something like:
290 long long Res = ((MAX_UNSIGNED) 1 << target);
292 for(i=0; i<reg->size; i++)
293 reg->node[i].state ^= Res & 0xFFFFFFFFULL;
295 for(i=0; i<reg->size; i++)
296 reg->node[i].state ^= Res & 0xFFFFFFFF00000000ULL
299 ... which would only do one 32-bit XOR per loop iteration instead of two.
301 It would also be nice to recognize the reg->size doesn't alias reg->node[i], but
304 //===---------------------------------------------------------------------===//
306 This isn't recognized as bswap by instcombine (yes, it really is bswap):
308 unsigned long reverse(unsigned v) {
310 t = v ^ ((v << 16) | (v >> 16));
312 v = (v << 24) | (v >> 8);
316 //===---------------------------------------------------------------------===//
320 These idioms should be recognized as popcount (see PR1488):
322 unsigned countbits_slow(unsigned v) {
324 for (c = 0; v; v >>= 1)
328 unsigned countbits_fast(unsigned v){
331 v &= v - 1; // clear the least significant bit set
335 BITBOARD = unsigned long long
336 int PopCnt(register BITBOARD a) {
344 unsigned int popcount(unsigned int input) {
345 unsigned int count = 0;
346 for (unsigned int i = 0; i < 4 * 8; i++)
347 count += (input >> i) & i;
351 This is a form of idiom recognition for loops, the same thing that could be
352 useful for recognizing memset/memcpy.
354 //===---------------------------------------------------------------------===//
356 These should turn into single 16-bit (unaligned?) loads on little/big endian
359 unsigned short read_16_le(const unsigned char *adr) {
360 return adr[0] | (adr[1] << 8);
362 unsigned short read_16_be(const unsigned char *adr) {
363 return (adr[0] << 8) | adr[1];
366 //===---------------------------------------------------------------------===//
368 -instcombine should handle this transform:
369 icmp pred (sdiv X / C1 ), C2
370 when X, C1, and C2 are unsigned. Similarly for udiv and signed operands.
372 Currently InstCombine avoids this transform but will do it when the signs of
373 the operands and the sign of the divide match. See the FIXME in
374 InstructionCombining.cpp in the visitSetCondInst method after the switch case
375 for Instruction::UDiv (around line 4447) for more details.
377 The SingleSource/Benchmarks/Shootout-C++/hash and hash2 tests have examples of
380 //===---------------------------------------------------------------------===//
384 viterbi speeds up *significantly* if the various "history" related copy loops
385 are turned into memcpy calls at the source level. We need a "loops to memcpy"
388 //===---------------------------------------------------------------------===//
392 SingleSource/Benchmarks/Misc/dt.c shows several interesting optimization
393 opportunities in its double_array_divs_variable function: it needs loop
394 interchange, memory promotion (which LICM already does), vectorization and
395 variable trip count loop unrolling (since it has a constant trip count). ICC
396 apparently produces this very nice code with -ffast-math:
398 ..B1.70: # Preds ..B1.70 ..B1.69
399 mulpd %xmm0, %xmm1 #108.2
400 mulpd %xmm0, %xmm1 #108.2
401 mulpd %xmm0, %xmm1 #108.2
402 mulpd %xmm0, %xmm1 #108.2
404 cmpl $131072, %edx #108.2
405 jb ..B1.70 # Prob 99% #108.2
407 It would be better to count down to zero, but this is a lot better than what we
410 //===---------------------------------------------------------------------===//
414 typedef unsigned U32;
415 typedef unsigned long long U64;
416 int test (U32 *inst, U64 *regs) {
419 int r1 = (temp >> 20) & 0xf;
420 int b2 = (temp >> 16) & 0xf;
421 effective_addr2 = temp & 0xfff;
422 if (b2) effective_addr2 += regs[b2];
423 b2 = (temp >> 12) & 0xf;
424 if (b2) effective_addr2 += regs[b2];
425 effective_addr2 &= regs[4];
426 if ((effective_addr2 & 3) == 0)
431 Note that only the low 2 bits of effective_addr2 are used. On 32-bit systems,
432 we don't eliminate the computation of the top half of effective_addr2 because
433 we don't have whole-function selection dags. On x86, this means we use one
434 extra register for the function when effective_addr2 is declared as U64 than
435 when it is declared U32.
437 PHI Slicing could be extended to do this.
439 //===---------------------------------------------------------------------===//
441 LSR should know what GPR types a target has from TargetData. This code:
443 volatile short X, Y; // globals
447 for (i = 0; i < N; i++) { X = i; Y = i*4; }
450 produces two near identical IV's (after promotion) on PPC/ARM:
460 add r2, r2, #1 <- [0,+,1]
461 sub r0, r0, #1 <- [0,-,1]
465 LSR should reuse the "+" IV for the exit test.
467 //===---------------------------------------------------------------------===//
469 Tail call elim should be more aggressive, checking to see if the call is
470 followed by an uncond branch to an exit block.
472 ; This testcase is due to tail-duplication not wanting to copy the return
473 ; instruction into the terminating blocks because there was other code
474 ; optimized out of the function after the taildup happened.
475 ; RUN: llvm-as < %s | opt -tailcallelim | llvm-dis | not grep call
477 define i32 @t4(i32 %a) {
479 %tmp.1 = and i32 %a, 1 ; <i32> [#uses=1]
480 %tmp.2 = icmp ne i32 %tmp.1, 0 ; <i1> [#uses=1]
481 br i1 %tmp.2, label %then.0, label %else.0
483 then.0: ; preds = %entry
484 %tmp.5 = add i32 %a, -1 ; <i32> [#uses=1]
485 %tmp.3 = call i32 @t4( i32 %tmp.5 ) ; <i32> [#uses=1]
488 else.0: ; preds = %entry
489 %tmp.7 = icmp ne i32 %a, 0 ; <i1> [#uses=1]
490 br i1 %tmp.7, label %then.1, label %return
492 then.1: ; preds = %else.0
493 %tmp.11 = add i32 %a, -2 ; <i32> [#uses=1]
494 %tmp.9 = call i32 @t4( i32 %tmp.11 ) ; <i32> [#uses=1]
497 return: ; preds = %then.1, %else.0, %then.0
498 %result.0 = phi i32 [ 0, %else.0 ], [ %tmp.3, %then.0 ],
503 //===---------------------------------------------------------------------===//
505 Tail recursion elimination should handle:
510 return 2 * pow2m1 (n - 1) + 1;
513 Also, multiplies can be turned into SHL's, so they should be handled as if
514 they were associative. "return foo() << 1" can be tail recursion eliminated.
516 //===---------------------------------------------------------------------===//
518 Argument promotion should promote arguments for recursive functions, like
521 ; RUN: llvm-as < %s | opt -argpromotion | llvm-dis | grep x.val
523 define internal i32 @foo(i32* %x) {
525 %tmp = load i32* %x ; <i32> [#uses=0]
526 %tmp.foo = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
530 define i32 @bar(i32* %x) {
532 %tmp3 = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
536 //===---------------------------------------------------------------------===//
538 We should investigate an instruction sinking pass. Consider this silly
554 je LBB1_2 # cond_true
562 The PIC base computation (call+popl) is only used on one path through the
563 code, but is currently always computed in the entry block. It would be
564 better to sink the picbase computation down into the block for the
565 assertion, as it is the only one that uses it. This happens for a lot of
566 code with early outs.
568 Another example is loads of arguments, which are usually emitted into the
569 entry block on targets like x86. If not used in all paths through a
570 function, they should be sunk into the ones that do.
572 In this case, whole-function-isel would also handle this.
574 //===---------------------------------------------------------------------===//
576 Investigate lowering of sparse switch statements into perfect hash tables:
577 http://burtleburtle.net/bob/hash/perfect.html
579 //===---------------------------------------------------------------------===//
581 We should turn things like "load+fabs+store" and "load+fneg+store" into the
582 corresponding integer operations. On a yonah, this loop:
587 for (b = 0; b < 10000000; b++)
588 for (i = 0; i < 256; i++)
592 is twice as slow as this loop:
597 for (b = 0; b < 10000000; b++)
598 for (i = 0; i < 256; i++)
599 a[i] ^= (1ULL << 63);
602 and I suspect other processors are similar. On X86 in particular this is a
603 big win because doing this with integers allows the use of read/modify/write
606 //===---------------------------------------------------------------------===//
608 DAG Combiner should try to combine small loads into larger loads when
609 profitable. For example, we compile this C++ example:
611 struct THotKey { short Key; bool Control; bool Shift; bool Alt; };
612 extern THotKey m_HotKey;
613 THotKey GetHotKey () { return m_HotKey; }
615 into (-O3 -fno-exceptions -static -fomit-frame-pointer):
620 movb _m_HotKey+3, %cl
621 movb _m_HotKey+4, %dl
622 movb _m_HotKey+2, %ch
637 movzwl _m_HotKey+4, %edx
641 The LLVM IR contains the needed alignment info, so we should be able to
642 merge the loads and stores into 4-byte loads:
644 %struct.THotKey = type { i16, i8, i8, i8 }
645 define void @_Z9GetHotKeyv(%struct.THotKey* sret %agg.result) nounwind {
647 %tmp2 = load i16* getelementptr (@m_HotKey, i32 0, i32 0), align 8
648 %tmp5 = load i8* getelementptr (@m_HotKey, i32 0, i32 1), align 2
649 %tmp8 = load i8* getelementptr (@m_HotKey, i32 0, i32 2), align 1
650 %tmp11 = load i8* getelementptr (@m_HotKey, i32 0, i32 3), align 2
652 Alternatively, we should use a small amount of base-offset alias analysis
653 to make it so the scheduler doesn't need to hold all the loads in regs at
656 //===---------------------------------------------------------------------===//
658 We should add an FRINT node to the DAG to model targets that have legal
659 implementations of ceil/floor/rint.
661 //===---------------------------------------------------------------------===//
666 long long input[8] = {1,1,1,1,1,1,1,1};
670 We currently compile this into a memcpy from a global array since the
671 initializer is fairly large and not memset'able. This is good, but the memcpy
672 gets lowered to load/stores in the code generator. This is also ok, except
673 that the codegen lowering for memcpy doesn't handle the case when the source
674 is a constant global. This gives us atrocious code like this:
679 movl _C.0.1444-"L1$pb"+32(%eax), %ecx
681 movl _C.0.1444-"L1$pb"+20(%eax), %ecx
683 movl _C.0.1444-"L1$pb"+36(%eax), %ecx
685 movl _C.0.1444-"L1$pb"+44(%eax), %ecx
687 movl _C.0.1444-"L1$pb"+40(%eax), %ecx
689 movl _C.0.1444-"L1$pb"+12(%eax), %ecx
691 movl _C.0.1444-"L1$pb"+4(%eax), %ecx
703 //===---------------------------------------------------------------------===//
705 http://llvm.org/PR717:
707 The following code should compile into "ret int undef". Instead, LLVM
708 produces "ret int 0":
717 //===---------------------------------------------------------------------===//
719 The loop unroller should partially unroll loops (instead of peeling them)
720 when code growth isn't too bad and when an unroll count allows simplification
721 of some code within the loop. One trivial example is:
727 for ( nLoop = 0; nLoop < 1000; nLoop++ ) {
736 Unrolling by 2 would eliminate the '&1' in both copies, leading to a net
737 reduction in code size. The resultant code would then also be suitable for
738 exit value computation.
740 //===---------------------------------------------------------------------===//
742 We miss a bunch of rotate opportunities on various targets, including ppc, x86,
743 etc. On X86, we miss a bunch of 'rotate by variable' cases because the rotate
744 matching code in dag combine doesn't look through truncates aggressively
745 enough. Here are some testcases reduces from GCC PR17886:
747 unsigned long long f(unsigned long long x, int y) {
748 return (x << y) | (x >> 64-y);
750 unsigned f2(unsigned x, int y){
751 return (x << y) | (x >> 32-y);
753 unsigned long long f3(unsigned long long x){
755 return (x << y) | (x >> 64-y);
757 unsigned f4(unsigned x){
759 return (x << y) | (x >> 32-y);
761 unsigned long long f5(unsigned long long x, unsigned long long y) {
762 return (x << 8) | ((y >> 48) & 0xffull);
764 unsigned long long f6(unsigned long long x, unsigned long long y, int z) {
767 return (x << 8) | ((y >> 48) & 0xffull);
769 return (x << 16) | ((y >> 40) & 0xffffull);
771 return (x << 24) | ((y >> 32) & 0xffffffull);
773 return (x << 32) | ((y >> 24) & 0xffffffffull);
775 return (x << 40) | ((y >> 16) & 0xffffffffffull);
779 On X86-64, we only handle f2/f3/f4 right. On x86-32, a few of these
780 generate truly horrible code, instead of using shld and friends. On
781 ARM, we end up with calls to L___lshrdi3/L___ashldi3 in f, which is
782 badness. PPC64 misses f, f5 and f6. CellSPU aborts in isel.
784 //===---------------------------------------------------------------------===//
786 We do a number of simplifications in simplify libcalls to strength reduce
787 standard library functions, but we don't currently merge them together. For
788 example, it is useful to merge memcpy(a,b,strlen(b)) -> strcpy. This can only
789 be done safely if "b" isn't modified between the strlen and memcpy of course.
791 //===---------------------------------------------------------------------===//
793 We compile this program: (from GCC PR11680)
794 http://gcc.gnu.org/bugzilla/attachment.cgi?id=4487
796 Into code that runs the same speed in fast/slow modes, but both modes run 2x
797 slower than when compile with GCC (either 4.0 or 4.2):
799 $ llvm-g++ perf.cpp -O3 -fno-exceptions
801 1.821u 0.003s 0:01.82 100.0% 0+0k 0+0io 0pf+0w
803 $ g++ perf.cpp -O3 -fno-exceptions
805 0.821u 0.001s 0:00.82 100.0% 0+0k 0+0io 0pf+0w
807 It looks like we are making the same inlining decisions, so this may be raw
808 codegen badness or something else (haven't investigated).
810 //===---------------------------------------------------------------------===//
812 We miss some instcombines for stuff like this:
814 void foo (unsigned int a) {
815 /* This one is equivalent to a >= (3 << 2). */
820 A few other related ones are in GCC PR14753.
822 //===---------------------------------------------------------------------===//
824 Divisibility by constant can be simplified (according to GCC PR12849) from
825 being a mulhi to being a mul lo (cheaper). Testcase:
827 void bar(unsigned n) {
832 This is equivalent to the following, where 2863311531 is the multiplicative
833 inverse of 3, and 1431655766 is ((2^32)-1)/3+1:
834 void bar(unsigned n) {
835 if (n * 2863311531U < 1431655766U)
839 The same transformation can work with an even modulo with the addition of a
840 rotate: rotate the result of the multiply to the right by the number of bits
841 which need to be zero for the condition to be true, and shrink the compare RHS
842 by the same amount. Unless the target supports rotates, though, that
843 transformation probably isn't worthwhile.
845 The transformation can also easily be made to work with non-zero equality
846 comparisons: just transform, for example, "n % 3 == 1" to "(n-1) % 3 == 0".
848 //===---------------------------------------------------------------------===//
850 Better mod/ref analysis for scanf would allow us to eliminate the vtable and a
851 bunch of other stuff from this example (see PR1604):
861 std::scanf("%d", &t.val);
862 std::printf("%d\n", t.val);
865 //===---------------------------------------------------------------------===//
867 These functions perform the same computation, but produce different assembly.
869 define i8 @select(i8 %x) readnone nounwind {
870 %A = icmp ult i8 %x, 250
871 %B = select i1 %A, i8 0, i8 1
875 define i8 @addshr(i8 %x) readnone nounwind {
876 %A = zext i8 %x to i9
877 %B = add i9 %A, 6 ;; 256 - 250 == 6
879 %D = trunc i9 %C to i8
883 //===---------------------------------------------------------------------===//
887 f (unsigned long a, unsigned long b, unsigned long c)
889 return ((a & (c - 1)) != 0) || ((b & (c - 1)) != 0);
892 f (unsigned long a, unsigned long b, unsigned long c)
894 return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0);
896 Both should combine to ((a|b) & (c-1)) != 0. Currently not optimized with
897 "clang -emit-llvm-bc | opt -std-compile-opts".
899 //===---------------------------------------------------------------------===//
902 #define PMD_MASK (~((1UL << 23) - 1))
903 void clear_pmd_range(unsigned long start, unsigned long end)
905 if (!(start & ~PMD_MASK) && !(end & ~PMD_MASK))
908 The expression should optimize to something like
909 "!((start|end)&~PMD_MASK). Currently not optimized with "clang
910 -emit-llvm-bc | opt -std-compile-opts".
912 //===---------------------------------------------------------------------===//
918 return (n >= 0 ? 1 : -1);
920 Should combine to (n >> 31) | 1. Currently not optimized with "clang
921 -emit-llvm-bc | opt -std-compile-opts | llc".
923 //===---------------------------------------------------------------------===//
927 if (variable == 4 || variable == 6)
930 This should optimize to "if ((variable | 2) == 6)". Currently not
931 optimized with "clang -emit-llvm-bc | opt -std-compile-opts | llc".
933 //===---------------------------------------------------------------------===//
935 unsigned int f(unsigned int i, unsigned int n) {++i; if (i == n) ++i; return
937 unsigned int f2(unsigned int i, unsigned int n) {++i; i += i == n; return i;}
938 These should combine to the same thing. Currently, the first function
939 produces better code on X86.
941 //===---------------------------------------------------------------------===//
944 #define abs(x) x>0?x:-x
947 return (abs(x)) >= 0;
949 This should optimize to x == INT_MIN. (With -fwrapv.) Currently not
950 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
952 //===---------------------------------------------------------------------===//
956 rotate_cst (unsigned int a)
958 a = (a << 10) | (a >> 22);
963 minus_cst (unsigned int a)
972 mask_gt (unsigned int a)
974 /* This is equivalent to a > 15. */
979 rshift_gt (unsigned int a)
981 /* This is equivalent to a > 23. */
985 All should simplify to a single comparison. All of these are
986 currently not optimized with "clang -emit-llvm-bc | opt
989 //===---------------------------------------------------------------------===//
992 int c(int* x) {return (char*)x+2 == (char*)x;}
993 Should combine to 0. Currently not optimized with "clang
994 -emit-llvm-bc | opt -std-compile-opts" (although llc can optimize it).
996 //===---------------------------------------------------------------------===//
998 int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;}
999 Should be combined to "((b >> 1) | b) & 1". Currently not optimized
1000 with "clang -emit-llvm-bc | opt -std-compile-opts".
1002 //===---------------------------------------------------------------------===//
1004 unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);}
1005 Should combine to "x | (y & 3)". Currently not optimized with "clang
1006 -emit-llvm-bc | opt -std-compile-opts".
1008 //===---------------------------------------------------------------------===//
1010 int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);}
1011 Should fold to "(~a & c) | (a & b)". Currently not optimized with
1012 "clang -emit-llvm-bc | opt -std-compile-opts".
1014 //===---------------------------------------------------------------------===//
1016 int a(int a,int b) {return (~(a|b))|a;}
1017 Should fold to "a|~b". Currently not optimized with "clang
1018 -emit-llvm-bc | opt -std-compile-opts".
1020 //===---------------------------------------------------------------------===//
1022 int a(int a, int b) {return (a&&b) || (a&&!b);}
1023 Should fold to "a". Currently not optimized with "clang -emit-llvm-bc
1024 | opt -std-compile-opts".
1026 //===---------------------------------------------------------------------===//
1028 int a(int a, int b, int c) {return (a&&b) || (!a&&c);}
1029 Should fold to "a ? b : c", or at least something sane. Currently not
1030 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1032 //===---------------------------------------------------------------------===//
1034 int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);}
1035 Should fold to a && (b || c). Currently not optimized with "clang
1036 -emit-llvm-bc | opt -std-compile-opts".
1038 //===---------------------------------------------------------------------===//
1040 int a(int x) {return x | ((x & 8) ^ 8);}
1041 Should combine to x | 8. Currently not optimized with "clang
1042 -emit-llvm-bc | opt -std-compile-opts".
1044 //===---------------------------------------------------------------------===//
1046 int a(int x) {return x ^ ((x & 8) ^ 8);}
1047 Should also combine to x | 8. Currently not optimized with "clang
1048 -emit-llvm-bc | opt -std-compile-opts".
1050 //===---------------------------------------------------------------------===//
1052 int a(int x) {return (x & 8) == 0 ? -1 : -9;}
1053 Should combine to (x | -9) ^ 8. Currently not optimized with "clang
1054 -emit-llvm-bc | opt -std-compile-opts".
1056 //===---------------------------------------------------------------------===//
1058 int a(int x) {return (x & 8) == 0 ? -9 : -1;}
1059 Should combine to x | -9. Currently not optimized with "clang
1060 -emit-llvm-bc | opt -std-compile-opts".
1062 //===---------------------------------------------------------------------===//
1064 int a(int x) {return ((x | -9) ^ 8) & x;}
1065 Should combine to x & -9. Currently not optimized with "clang
1066 -emit-llvm-bc | opt -std-compile-opts".
1068 //===---------------------------------------------------------------------===//
1070 unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;}
1071 Should combine to "a * 0x88888888 >> 31". Currently not optimized
1072 with "clang -emit-llvm-bc | opt -std-compile-opts".
1074 //===---------------------------------------------------------------------===//
1076 unsigned a(char* x) {if ((*x & 32) == 0) return b();}
1077 There's an unnecessary zext in the generated code with "clang
1078 -emit-llvm-bc | opt -std-compile-opts".
1080 //===---------------------------------------------------------------------===//
1082 unsigned a(unsigned long long x) {return 40 * (x >> 1);}
1083 Should combine to "20 * (((unsigned)x) & -2)". Currently not
1084 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1086 //===---------------------------------------------------------------------===//
1088 This was noticed in the entryblock for grokdeclarator in 403.gcc:
1090 %tmp = icmp eq i32 %decl_context, 4
1091 %decl_context_addr.0 = select i1 %tmp, i32 3, i32 %decl_context
1092 %tmp1 = icmp eq i32 %decl_context_addr.0, 1
1093 %decl_context_addr.1 = select i1 %tmp1, i32 0, i32 %decl_context_addr.0
1095 tmp1 should be simplified to something like:
1096 (!tmp || decl_context == 1)
1098 This allows recursive simplifications, tmp1 is used all over the place in
1099 the function, e.g. by:
1101 %tmp23 = icmp eq i32 %decl_context_addr.1, 0 ; <i1> [#uses=1]
1102 %tmp24 = xor i1 %tmp1, true ; <i1> [#uses=1]
1103 %or.cond8 = and i1 %tmp23, %tmp24 ; <i1> [#uses=1]
1107 //===---------------------------------------------------------------------===//
1111 Store sinking: This code:
1113 void f (int n, int *cond, int *res) {
1116 for (i = 0; i < n; i++)
1118 *res ^= 234; /* (*) */
1121 On this function GVN hoists the fully redundant value of *res, but nothing
1122 moves the store out. This gives us this code:
1124 bb: ; preds = %bb2, %entry
1125 %.rle = phi i32 [ 0, %entry ], [ %.rle6, %bb2 ]
1126 %i.05 = phi i32 [ 0, %entry ], [ %indvar.next, %bb2 ]
1127 %1 = load i32* %cond, align 4
1128 %2 = icmp eq i32 %1, 0
1129 br i1 %2, label %bb2, label %bb1
1132 %3 = xor i32 %.rle, 234
1133 store i32 %3, i32* %res, align 4
1136 bb2: ; preds = %bb, %bb1
1137 %.rle6 = phi i32 [ %3, %bb1 ], [ %.rle, %bb ]
1138 %indvar.next = add i32 %i.05, 1
1139 %exitcond = icmp eq i32 %indvar.next, %n
1140 br i1 %exitcond, label %return, label %bb
1142 DSE should sink partially dead stores to get the store out of the loop.
1144 Here's another partial dead case:
1145 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12395
1147 //===---------------------------------------------------------------------===//
1149 Scalar PRE hoists the mul in the common block up to the else:
1151 int test (int a, int b, int c, int g) {
1161 It would be better to do the mul once to reduce codesize above the if.
1162 This is GCC PR38204.
1164 //===---------------------------------------------------------------------===//
1168 GCC PR37810 is an interesting case where we should sink load/store reload
1169 into the if block and outside the loop, so we don't reload/store it on the
1190 We now hoist the reload after the call (Transforms/GVN/lpre-call-wrap.ll), but
1191 we don't sink the store. We need partially dead store sinking.
1193 //===---------------------------------------------------------------------===//
1195 [LOAD PRE CRIT EDGE SPLITTING]
1197 GCC PR37166: Sinking of loads prevents SROA'ing the "g" struct on the stack
1198 leading to excess stack traffic. This could be handled by GVN with some crazy
1199 symbolic phi translation. The code we get looks like (g is on the stack):
1203 %9 = getelementptr %struct.f* %g, i32 0, i32 0
1204 store i32 %8, i32* %9, align bel %bb3
1206 bb3: ; preds = %bb1, %bb2, %bb
1207 %c_addr.0 = phi %struct.f* [ %g, %bb2 ], [ %c, %bb ], [ %c, %bb1 ]
1208 %b_addr.0 = phi %struct.f* [ %b, %bb2 ], [ %g, %bb ], [ %b, %bb1 ]
1209 %10 = getelementptr %struct.f* %c_addr.0, i32 0, i32 0
1210 %11 = load i32* %10, align 4
1212 %11 is partially redundant, an in BB2 it should have the value %8.
1214 GCC PR33344 and PR35287 are similar cases.
1217 //===---------------------------------------------------------------------===//
1221 There are many load PRE testcases in testsuite/gcc.dg/tree-ssa/loadpre* in the
1222 GCC testsuite, ones we don't get yet are (checked through loadpre25):
1224 [CRIT EDGE BREAKING]
1225 loadpre3.c predcom-4.c
1227 [PRE OF READONLY CALL]
1230 [TURN SELECT INTO BRANCH]
1231 loadpre14.c loadpre15.c
1233 actually a conditional increment: loadpre18.c loadpre19.c
1236 //===---------------------------------------------------------------------===//
1239 There are many PRE testcases in testsuite/gcc.dg/tree-ssa/ssa-pre-*.c in the
1242 //===---------------------------------------------------------------------===//
1244 There are some interesting cases in testsuite/gcc.dg/tree-ssa/pred-comm* in the
1245 GCC testsuite. For example, we get the first example in predcom-1.c, but
1246 miss the second one:
1251 __attribute__ ((noinline))
1252 void count_averages(int n) {
1254 for (i = 1; i < n; i++)
1255 avg[i] = (((unsigned long) fib[i - 1] + fib[i] + fib[i + 1]) / 3) & 0xffff;
1258 which compiles into two loads instead of one in the loop.
1260 predcom-2.c is the same as predcom-1.c
1262 predcom-3.c is very similar but needs loads feeding each other instead of
1266 //===---------------------------------------------------------------------===//
1270 Type based alias analysis:
1271 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14705
1273 We should do better analysis of posix_memalign. At the least it should
1274 no-capture its pointer argument, at best, we should know that the out-value
1275 result doesn't point to anything (like malloc). One example of this is in
1276 SingleSource/Benchmarks/Misc/dt.c
1278 //===---------------------------------------------------------------------===//
1280 A/B get pinned to the stack because we turn an if/then into a select instead
1281 of PRE'ing the load/store. This may be fixable in instcombine:
1282 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37892
1284 struct X { int i; };
1298 //===---------------------------------------------------------------------===//
1300 Interesting missed case because of control flow flattening (should be 2 loads):
1301 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26629
1302 With: llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as |
1303 opt -mem2reg -gvn -instcombine | llvm-dis
1304 we miss it because we need 1) CRIT EDGE 2) MULTIPLE DIFFERENT
1305 VALS PRODUCED BY ONE BLOCK OVER DIFFERENT PATHS
1307 //===---------------------------------------------------------------------===//
1309 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19633
1310 We could eliminate the branch condition here, loading from null is undefined:
1312 struct S { int w, x, y, z; };
1313 struct T { int r; struct S s; };
1314 void bar (struct S, int);
1315 void foo (int a, struct T b)
1323 //===---------------------------------------------------------------------===//
1325 simplifylibcalls should do several optimizations for strspn/strcspn:
1327 strcspn(x, "") -> strlen(x)
1330 strspn(x, "") -> strlen(x)
1331 strspn(x, "a") -> strchr(x, 'a')-x
1333 strcspn(x, "a") -> inlined loop for up to 3 letters (similarly for strspn):
1335 size_t __strcspn_c3 (__const char *__s, int __reject1, int __reject2,
1337 register size_t __result = 0;
1338 while (__s[__result] != '\0' && __s[__result] != __reject1 &&
1339 __s[__result] != __reject2 && __s[__result] != __reject3)
1344 This should turn into a switch on the character. See PR3253 for some notes on
1347 456.hmmer apparently uses strcspn and strspn a lot. 471.omnetpp uses strspn.
1349 //===---------------------------------------------------------------------===//
1351 "gas" uses this idiom:
1352 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
1354 else if (strchr ("<>", *intel_parser.op_string)
1356 Those should be turned into a switch.
1358 //===---------------------------------------------------------------------===//
1360 252.eon contains this interesting code:
1362 %3072 = getelementptr [100 x i8]* %tempString, i32 0, i32 0
1363 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1364 %strlen = call i32 @strlen(i8* %3072) ; uses = 1
1365 %endptr = getelementptr [100 x i8]* %tempString, i32 0, i32 %strlen
1366 call void @llvm.memcpy.i32(i8* %endptr,
1367 i8* getelementptr ([5 x i8]* @"\01LC42", i32 0, i32 0), i32 5, i32 1)
1368 %3074 = call i32 @strlen(i8* %endptr) nounwind readonly
1370 This is interesting for a couple reasons. First, in this:
1372 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1373 %strlen = call i32 @strlen(i8* %3072)
1375 The strlen could be replaced with: %strlen = sub %3072, %3073, because the
1376 strcpy call returns a pointer to the end of the string. Based on that, the
1377 endptr GEP just becomes equal to 3073, which eliminates a strlen call and GEP.
1379 Second, the memcpy+strlen strlen can be replaced with:
1381 %3074 = call i32 @strlen([5 x i8]* @"\01LC42") nounwind readonly
1383 Because the destination was just copied into the specified memory buffer. This,
1384 in turn, can be constant folded to "4".
1386 In other code, it contains:
1388 %endptr6978 = bitcast i8* %endptr69 to i32*
1389 store i32 7107374, i32* %endptr6978, align 1
1390 %3167 = call i32 @strlen(i8* %endptr69) nounwind readonly
1392 Which could also be constant folded. Whatever is producing this should probably
1393 be fixed to leave this as a memcpy from a string.
1395 Further, eon also has an interesting partially redundant strlen call:
1397 bb8: ; preds = %_ZN18eonImageCalculatorC1Ev.exit
1398 %682 = getelementptr i8** %argv, i32 6 ; <i8**> [#uses=2]
1399 %683 = load i8** %682, align 4 ; <i8*> [#uses=4]
1400 %684 = load i8* %683, align 1 ; <i8> [#uses=1]
1401 %685 = icmp eq i8 %684, 0 ; <i1> [#uses=1]
1402 br i1 %685, label %bb10, label %bb9
1405 %686 = call i32 @strlen(i8* %683) nounwind readonly
1406 %687 = icmp ugt i32 %686, 254 ; <i1> [#uses=1]
1407 br i1 %687, label %bb10, label %bb11
1409 bb10: ; preds = %bb9, %bb8
1410 %688 = call i32 @strlen(i8* %683) nounwind readonly
1412 This could be eliminated by doing the strlen once in bb8, saving code size and
1413 improving perf on the bb8->9->10 path.
1415 //===---------------------------------------------------------------------===//
1417 I see an interesting fully redundant call to strlen left in 186.crafty:InputMove
1419 %movetext11 = getelementptr [128 x i8]* %movetext, i32 0, i32 0
1422 bb62: ; preds = %bb55, %bb53
1423 %promote.0 = phi i32 [ %169, %bb55 ], [ 0, %bb53 ]
1424 %171 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1425 %172 = add i32 %171, -1 ; <i32> [#uses=1]
1426 %173 = getelementptr [128 x i8]* %movetext, i32 0, i32 %172
1429 br i1 %or.cond, label %bb65, label %bb72
1431 bb65: ; preds = %bb62
1432 store i8 0, i8* %173, align 1
1435 bb72: ; preds = %bb65, %bb62
1436 %trank.1 = phi i32 [ %176, %bb65 ], [ -1, %bb62 ]
1437 %177 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1439 Note that on the bb62->bb72 path, that the %177 strlen call is partially
1440 redundant with the %171 call. At worst, we could shove the %177 strlen call
1441 up into the bb65 block moving it out of the bb62->bb72 path. However, note
1442 that bb65 stores to the string, zeroing out the last byte. This means that on
1443 that path the value of %177 is actually just %171-1. A sub is cheaper than a
1446 This pattern repeats several times, basically doing:
1451 where it is "obvious" that B = A-1.
1453 //===---------------------------------------------------------------------===//
1455 186.crafty contains this interesting pattern:
1457 %77 = call i8* @strstr(i8* getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0),
1459 %phitmp648 = icmp eq i8* %77, getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0)
1460 br i1 %phitmp648, label %bb70, label %bb76
1462 bb70: ; preds = %OptionMatch.exit91, %bb69
1463 %78 = call i32 @strlen(i8* %30) nounwind readonly align 1 ; <i32> [#uses=1]
1467 if (strstr(cststr, P) == cststr) {
1471 The strstr call would be significantly cheaper written as:
1474 if (memcmp(P, str, strlen(P)))
1477 This is memcmp+strlen instead of strstr. This also makes the strlen fully
1480 //===---------------------------------------------------------------------===//
1482 186.crafty also contains this code:
1484 %1906 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1485 %1907 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1906
1486 %1908 = call i8* @strcpy(i8* %1907, i8* %1905) nounwind align 1
1487 %1909 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1488 %1910 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1909
1490 The last strlen is computable as 1908-@pgn_event, which means 1910=1908.
1492 //===---------------------------------------------------------------------===//
1494 186.crafty has this interesting pattern with the "out.4543" variable:
1496 call void @llvm.memcpy.i32(
1497 i8* getelementptr ([10 x i8]* @out.4543, i32 0, i32 0),
1498 i8* getelementptr ([7 x i8]* @"\01LC28700", i32 0, i32 0), i32 7, i32 1)
1499 %101 = call@printf(i8* ... @out.4543, i32 0, i32 0)) nounwind
1501 It is basically doing:
1503 memcpy(globalarray, "string");
1504 printf(..., globalarray);
1506 Anyway, by knowing that printf just reads the memory and forward substituting
1507 the string directly into the printf, this eliminates reads from globalarray.
1508 Since this pattern occurs frequently in crafty (due to the "DisplayTime" and
1509 other similar functions) there are many stores to "out". Once all the printfs
1510 stop using "out", all that is left is the memcpy's into it. This should allow
1511 globalopt to remove the "stored only" global.
1513 //===---------------------------------------------------------------------===//
1517 define inreg i32 @foo(i8* inreg %p) nounwind {
1519 %tmp1 = ashr i8 %tmp0, 5
1520 %tmp2 = sext i8 %tmp1 to i32
1524 could be dagcombine'd to a sign-extending load with a shift.
1525 For example, on x86 this currently gets this:
1531 while it could get this:
1536 //===---------------------------------------------------------------------===//
1540 int test(int x) { return 1-x == x; } // --> return false
1541 int test2(int x) { return 2-x == x; } // --> return x == 1 ?
1543 Always foldable for odd constants, what is the rule for even?
1545 //===---------------------------------------------------------------------===//
1547 PR 3381: GEP to field of size 0 inside a struct could be turned into GEP
1548 for next field in struct (which is at same address).
1550 For example: store of float into { {{}}, float } could be turned into a store to
1553 //===---------------------------------------------------------------------===//
1556 double foo(double a) { return sin(a); }
1558 This compiles into this on x86-64 Linux:
1569 //===---------------------------------------------------------------------===//
1571 The arg promotion pass should make use of nocapture to make its alias analysis
1572 stuff much more precise.
1574 //===---------------------------------------------------------------------===//
1576 The following functions should be optimized to use a select instead of a
1577 branch (from gcc PR40072):
1579 char char_int(int m) {if(m>7) return 0; return m;}
1580 int int_char(char m) {if(m>7) return 0; return m;}
1582 //===---------------------------------------------------------------------===//
1584 int func(int a, int b) { if (a & 0x80) b |= 0x80; else b &= ~0x80; return b; }
1588 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1590 %0 = and i32 %a, 128 ; <i32> [#uses=1]
1591 %1 = icmp eq i32 %0, 0 ; <i1> [#uses=1]
1592 %2 = or i32 %b, 128 ; <i32> [#uses=1]
1593 %3 = and i32 %b, -129 ; <i32> [#uses=1]
1594 %b_addr.0 = select i1 %1, i32 %3, i32 %2 ; <i32> [#uses=1]
1598 However, it's functionally equivalent to:
1600 b = (b & ~0x80) | (a & 0x80);
1602 Which generates this:
1604 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1606 %0 = and i32 %b, -129 ; <i32> [#uses=1]
1607 %1 = and i32 %a, 128 ; <i32> [#uses=1]
1608 %2 = or i32 %0, %1 ; <i32> [#uses=1]
1612 This can be generalized for other forms:
1614 b = (b & ~0x80) | (a & 0x40) << 1;
1616 //===---------------------------------------------------------------------===//
1618 These two functions produce different code. They shouldn't:
1622 uint8_t p1(uint8_t b, uint8_t a) {
1623 b = (b & ~0xc0) | (a & 0xc0);
1627 uint8_t p2(uint8_t b, uint8_t a) {
1628 b = (b & ~0x40) | (a & 0x40);
1629 b = (b & ~0x80) | (a & 0x80);
1633 define zeroext i8 @p1(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1635 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1636 %1 = and i8 %a, -64 ; <i8> [#uses=1]
1637 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1641 define zeroext i8 @p2(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1643 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1644 %.masked = and i8 %a, 64 ; <i8> [#uses=1]
1645 %1 = and i8 %a, -128 ; <i8> [#uses=1]
1646 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1647 %3 = or i8 %2, %.masked ; <i8> [#uses=1]
1651 //===---------------------------------------------------------------------===//
1653 IPSCCP does not currently propagate argument dependent constants through
1654 functions where it does not not all of the callers. This includes functions
1655 with normal external linkage as well as templates, C99 inline functions etc.
1656 Specifically, it does nothing to:
1658 define i32 @test(i32 %x, i32 %y, i32 %z) nounwind {
1660 %0 = add nsw i32 %y, %z
1663 %3 = add nsw i32 %1, %2
1667 define i32 @test2() nounwind {
1669 %0 = call i32 @test(i32 1, i32 2, i32 4) nounwind
1673 It would be interesting extend IPSCCP to be able to handle simple cases like
1674 this, where all of the arguments to a call are constant. Because IPSCCP runs
1675 before inlining, trivial templates and inline functions are not yet inlined.
1676 The results for a function + set of constant arguments should be memoized in a
1679 //===---------------------------------------------------------------------===//
1681 The libcall constant folding stuff should be moved out of SimplifyLibcalls into
1682 libanalysis' constantfolding logic. This would allow IPSCCP to be able to
1683 handle simple things like this:
1685 static int foo(const char *X) { return strlen(X); }
1686 int bar() { return foo("abcd"); }
1688 //===---------------------------------------------------------------------===//
1690 InstCombine should use SimplifyDemandedBits to remove the or instruction:
1692 define i1 @test(i8 %x, i8 %y) {
1694 %B = icmp ugt i8 %A, 3
1698 Currently instcombine calls SimplifyDemandedBits with either all bits or just
1699 the sign bit, if the comparison is obviously a sign test. In this case, we only
1700 need all but the bottom two bits from %A, and if we gave that mask to SDB it
1701 would delete the or instruction for us.
1703 //===---------------------------------------------------------------------===//
1705 functionattrs doesn't know much about memcpy/memset. This function should be
1706 marked readnone rather than readonly, since it only twiddles local memory, but
1707 functionattrs doesn't handle memset/memcpy/memmove aggressively:
1709 struct X { int *p; int *q; };
1716 p = __builtin_memcpy (&x, &y, sizeof (int *));
1720 //===---------------------------------------------------------------------===//
1722 Missed instcombine transformation:
1723 define i1 @a(i32 %x) nounwind readnone {
1725 %cmp = icmp eq i32 %x, 30
1726 %sub = add i32 %x, -30
1727 %cmp2 = icmp ugt i32 %sub, 9
1728 %or = or i1 %cmp, %cmp2
1731 This should be optimized to a single compare. Testcase derived from gcc.
1733 //===---------------------------------------------------------------------===//
1735 Missed instcombine transformation:
1737 void a(int x) { if (((1<<x)&8)==0) b(); }
1739 The shift should be optimized out. Testcase derived from gcc.
1741 //===---------------------------------------------------------------------===//
1743 Missed instcombine or reassociate transformation:
1744 int a(int a, int b) { return (a==12)&(b>47)&(b<58); }
1746 The sgt and slt should be combined into a single comparison. Testcase derived
1749 //===---------------------------------------------------------------------===//
1751 Missed instcombine transformation:
1752 define i32 @a(i32 %x) nounwind readnone {
1754 %rem = srem i32 %x, 32
1755 %shl = shl i32 1, %rem
1759 The srem can be transformed to an and because if x is negative, the shift is
1760 undefined. Testcase derived from gcc.
1762 //===---------------------------------------------------------------------===//
1764 Missed instcombine/dagcombine transformation:
1765 define i32 @a(i32 %x, i32 %y) nounwind readnone {
1767 %mul = mul i32 %y, -8
1768 %sub = sub i32 %x, %mul
1772 Should compile to something like x+y*8, but currently compiles to an
1773 inefficient result. Testcase derived from gcc.
1775 //===---------------------------------------------------------------------===//
1777 Missed instcombine/dagcombine transformation:
1778 define void @lshift_lt(i8 zeroext %a) nounwind {
1780 %conv = zext i8 %a to i32
1781 %shl = shl i32 %conv, 3
1782 %cmp = icmp ult i32 %shl, 33
1783 br i1 %cmp, label %if.then, label %if.end
1786 tail call void @bar() nounwind
1792 declare void @bar() nounwind
1794 The shift should be eliminated. Testcase derived from gcc.
1796 //===---------------------------------------------------------------------===//
1798 These compile into different code, one gets recognized as a switch and the
1799 other doesn't due to phase ordering issues (PR6212):
1801 int test1(int mainType, int subType) {
1804 else if (mainType == 9)
1806 else if (mainType == 11)
1811 int test2(int mainType, int subType) {
1821 //===---------------------------------------------------------------------===//