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 quantum_sigma_x in 462.libquantum contains the following loop:
268 for(i=0; i<reg->size; i++)
270 /* Flip the target bit of each basis state */
271 reg->node[i].state ^= ((MAX_UNSIGNED) 1 << target);
274 Where MAX_UNSIGNED/state is a 64-bit int. On a 32-bit platform it would be just
275 so cool to turn it into something like:
277 long long Res = ((MAX_UNSIGNED) 1 << target);
279 for(i=0; i<reg->size; i++)
280 reg->node[i].state ^= Res & 0xFFFFFFFFULL;
282 for(i=0; i<reg->size; i++)
283 reg->node[i].state ^= Res & 0xFFFFFFFF00000000ULL
286 ... which would only do one 32-bit XOR per loop iteration instead of two.
288 It would also be nice to recognize the reg->size doesn't alias reg->node[i], but
291 //===---------------------------------------------------------------------===//
293 This isn't recognized as bswap by instcombine (yes, it really is bswap):
295 unsigned long reverse(unsigned v) {
297 t = v ^ ((v << 16) | (v >> 16));
299 v = (v << 24) | (v >> 8);
303 //===---------------------------------------------------------------------===//
307 These idioms should be recognized as popcount (see PR1488):
309 unsigned countbits_slow(unsigned v) {
311 for (c = 0; v; v >>= 1)
315 unsigned countbits_fast(unsigned v){
318 v &= v - 1; // clear the least significant bit set
322 BITBOARD = unsigned long long
323 int PopCnt(register BITBOARD a) {
331 unsigned int popcount(unsigned int input) {
332 unsigned int count = 0;
333 for (unsigned int i = 0; i < 4 * 8; i++)
334 count += (input >> i) & i;
338 This is a form of idiom recognition for loops, the same thing that could be
339 useful for recognizing memset/memcpy.
341 //===---------------------------------------------------------------------===//
343 These should turn into single 16-bit (unaligned?) loads on little/big endian
346 unsigned short read_16_le(const unsigned char *adr) {
347 return adr[0] | (adr[1] << 8);
349 unsigned short read_16_be(const unsigned char *adr) {
350 return (adr[0] << 8) | adr[1];
353 //===---------------------------------------------------------------------===//
355 -instcombine should handle this transform:
356 icmp pred (sdiv X / C1 ), C2
357 when X, C1, and C2 are unsigned. Similarly for udiv and signed operands.
359 Currently InstCombine avoids this transform but will do it when the signs of
360 the operands and the sign of the divide match. See the FIXME in
361 InstructionCombining.cpp in the visitSetCondInst method after the switch case
362 for Instruction::UDiv (around line 4447) for more details.
364 The SingleSource/Benchmarks/Shootout-C++/hash and hash2 tests have examples of
367 //===---------------------------------------------------------------------===//
371 viterbi speeds up *significantly* if the various "history" related copy loops
372 are turned into memcpy calls at the source level. We need a "loops to memcpy"
375 //===---------------------------------------------------------------------===//
379 SingleSource/Benchmarks/Misc/dt.c shows several interesting optimization
380 opportunities in its double_array_divs_variable function: it needs loop
381 interchange, memory promotion (which LICM already does), vectorization and
382 variable trip count loop unrolling (since it has a constant trip count). ICC
383 apparently produces this very nice code with -ffast-math:
385 ..B1.70: # Preds ..B1.70 ..B1.69
386 mulpd %xmm0, %xmm1 #108.2
387 mulpd %xmm0, %xmm1 #108.2
388 mulpd %xmm0, %xmm1 #108.2
389 mulpd %xmm0, %xmm1 #108.2
391 cmpl $131072, %edx #108.2
392 jb ..B1.70 # Prob 99% #108.2
394 It would be better to count down to zero, but this is a lot better than what we
397 //===---------------------------------------------------------------------===//
401 typedef unsigned U32;
402 typedef unsigned long long U64;
403 int test (U32 *inst, U64 *regs) {
406 int r1 = (temp >> 20) & 0xf;
407 int b2 = (temp >> 16) & 0xf;
408 effective_addr2 = temp & 0xfff;
409 if (b2) effective_addr2 += regs[b2];
410 b2 = (temp >> 12) & 0xf;
411 if (b2) effective_addr2 += regs[b2];
412 effective_addr2 &= regs[4];
413 if ((effective_addr2 & 3) == 0)
418 Note that only the low 2 bits of effective_addr2 are used. On 32-bit systems,
419 we don't eliminate the computation of the top half of effective_addr2 because
420 we don't have whole-function selection dags. On x86, this means we use one
421 extra register for the function when effective_addr2 is declared as U64 than
422 when it is declared U32.
424 PHI Slicing could be extended to do this.
426 //===---------------------------------------------------------------------===//
428 LSR should know what GPR types a target has from TargetData. This code:
430 volatile short X, Y; // globals
434 for (i = 0; i < N; i++) { X = i; Y = i*4; }
437 produces two near identical IV's (after promotion) on PPC/ARM:
447 add r2, r2, #1 <- [0,+,1]
448 sub r0, r0, #1 <- [0,-,1]
452 LSR should reuse the "+" IV for the exit test.
454 //===---------------------------------------------------------------------===//
456 Tail call elim should be more aggressive, checking to see if the call is
457 followed by an uncond branch to an exit block.
459 ; This testcase is due to tail-duplication not wanting to copy the return
460 ; instruction into the terminating blocks because there was other code
461 ; optimized out of the function after the taildup happened.
462 ; RUN: llvm-as < %s | opt -tailcallelim | llvm-dis | not grep call
464 define i32 @t4(i32 %a) {
466 %tmp.1 = and i32 %a, 1 ; <i32> [#uses=1]
467 %tmp.2 = icmp ne i32 %tmp.1, 0 ; <i1> [#uses=1]
468 br i1 %tmp.2, label %then.0, label %else.0
470 then.0: ; preds = %entry
471 %tmp.5 = add i32 %a, -1 ; <i32> [#uses=1]
472 %tmp.3 = call i32 @t4( i32 %tmp.5 ) ; <i32> [#uses=1]
475 else.0: ; preds = %entry
476 %tmp.7 = icmp ne i32 %a, 0 ; <i1> [#uses=1]
477 br i1 %tmp.7, label %then.1, label %return
479 then.1: ; preds = %else.0
480 %tmp.11 = add i32 %a, -2 ; <i32> [#uses=1]
481 %tmp.9 = call i32 @t4( i32 %tmp.11 ) ; <i32> [#uses=1]
484 return: ; preds = %then.1, %else.0, %then.0
485 %result.0 = phi i32 [ 0, %else.0 ], [ %tmp.3, %then.0 ],
490 //===---------------------------------------------------------------------===//
492 Tail recursion elimination should handle:
497 return 2 * pow2m1 (n - 1) + 1;
500 Also, multiplies can be turned into SHL's, so they should be handled as if
501 they were associative. "return foo() << 1" can be tail recursion eliminated.
503 //===---------------------------------------------------------------------===//
505 Argument promotion should promote arguments for recursive functions, like
508 ; RUN: llvm-as < %s | opt -argpromotion | llvm-dis | grep x.val
510 define internal i32 @foo(i32* %x) {
512 %tmp = load i32* %x ; <i32> [#uses=0]
513 %tmp.foo = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
517 define i32 @bar(i32* %x) {
519 %tmp3 = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
523 //===---------------------------------------------------------------------===//
525 We should investigate an instruction sinking pass. Consider this silly
541 je LBB1_2 # cond_true
549 The PIC base computation (call+popl) is only used on one path through the
550 code, but is currently always computed in the entry block. It would be
551 better to sink the picbase computation down into the block for the
552 assertion, as it is the only one that uses it. This happens for a lot of
553 code with early outs.
555 Another example is loads of arguments, which are usually emitted into the
556 entry block on targets like x86. If not used in all paths through a
557 function, they should be sunk into the ones that do.
559 In this case, whole-function-isel would also handle this.
561 //===---------------------------------------------------------------------===//
563 Investigate lowering of sparse switch statements into perfect hash tables:
564 http://burtleburtle.net/bob/hash/perfect.html
566 //===---------------------------------------------------------------------===//
568 We should turn things like "load+fabs+store" and "load+fneg+store" into the
569 corresponding integer operations. On a yonah, this loop:
574 for (b = 0; b < 10000000; b++)
575 for (i = 0; i < 256; i++)
579 is twice as slow as this loop:
584 for (b = 0; b < 10000000; b++)
585 for (i = 0; i < 256; i++)
586 a[i] ^= (1ULL << 63);
589 and I suspect other processors are similar. On X86 in particular this is a
590 big win because doing this with integers allows the use of read/modify/write
593 //===---------------------------------------------------------------------===//
595 DAG Combiner should try to combine small loads into larger loads when
596 profitable. For example, we compile this C++ example:
598 struct THotKey { short Key; bool Control; bool Shift; bool Alt; };
599 extern THotKey m_HotKey;
600 THotKey GetHotKey () { return m_HotKey; }
602 into (-O3 -fno-exceptions -static -fomit-frame-pointer):
607 movb _m_HotKey+3, %cl
608 movb _m_HotKey+4, %dl
609 movb _m_HotKey+2, %ch
624 movzwl _m_HotKey+4, %edx
628 The LLVM IR contains the needed alignment info, so we should be able to
629 merge the loads and stores into 4-byte loads:
631 %struct.THotKey = type { i16, i8, i8, i8 }
632 define void @_Z9GetHotKeyv(%struct.THotKey* sret %agg.result) nounwind {
634 %tmp2 = load i16* getelementptr (@m_HotKey, i32 0, i32 0), align 8
635 %tmp5 = load i8* getelementptr (@m_HotKey, i32 0, i32 1), align 2
636 %tmp8 = load i8* getelementptr (@m_HotKey, i32 0, i32 2), align 1
637 %tmp11 = load i8* getelementptr (@m_HotKey, i32 0, i32 3), align 2
639 Alternatively, we should use a small amount of base-offset alias analysis
640 to make it so the scheduler doesn't need to hold all the loads in regs at
643 //===---------------------------------------------------------------------===//
645 We should add an FRINT node to the DAG to model targets that have legal
646 implementations of ceil/floor/rint.
648 //===---------------------------------------------------------------------===//
653 long long input[8] = {1,1,1,1,1,1,1,1};
657 We currently compile this into a memcpy from a global array since the
658 initializer is fairly large and not memset'able. This is good, but the memcpy
659 gets lowered to load/stores in the code generator. This is also ok, except
660 that the codegen lowering for memcpy doesn't handle the case when the source
661 is a constant global. This gives us atrocious code like this:
666 movl _C.0.1444-"L1$pb"+32(%eax), %ecx
668 movl _C.0.1444-"L1$pb"+20(%eax), %ecx
670 movl _C.0.1444-"L1$pb"+36(%eax), %ecx
672 movl _C.0.1444-"L1$pb"+44(%eax), %ecx
674 movl _C.0.1444-"L1$pb"+40(%eax), %ecx
676 movl _C.0.1444-"L1$pb"+12(%eax), %ecx
678 movl _C.0.1444-"L1$pb"+4(%eax), %ecx
690 //===---------------------------------------------------------------------===//
692 http://llvm.org/PR717:
694 The following code should compile into "ret int undef". Instead, LLVM
695 produces "ret int 0":
704 //===---------------------------------------------------------------------===//
706 The loop unroller should partially unroll loops (instead of peeling them)
707 when code growth isn't too bad and when an unroll count allows simplification
708 of some code within the loop. One trivial example is:
714 for ( nLoop = 0; nLoop < 1000; nLoop++ ) {
723 Unrolling by 2 would eliminate the '&1' in both copies, leading to a net
724 reduction in code size. The resultant code would then also be suitable for
725 exit value computation.
727 //===---------------------------------------------------------------------===//
729 We miss a bunch of rotate opportunities on various targets, including ppc, x86,
730 etc. On X86, we miss a bunch of 'rotate by variable' cases because the rotate
731 matching code in dag combine doesn't look through truncates aggressively
732 enough. Here are some testcases reduces from GCC PR17886:
734 unsigned long long f(unsigned long long x, int y) {
735 return (x << y) | (x >> 64-y);
737 unsigned f2(unsigned x, int y){
738 return (x << y) | (x >> 32-y);
740 unsigned long long f3(unsigned long long x){
742 return (x << y) | (x >> 64-y);
744 unsigned f4(unsigned x){
746 return (x << y) | (x >> 32-y);
748 unsigned long long f5(unsigned long long x, unsigned long long y) {
749 return (x << 8) | ((y >> 48) & 0xffull);
751 unsigned long long f6(unsigned long long x, unsigned long long y, int z) {
754 return (x << 8) | ((y >> 48) & 0xffull);
756 return (x << 16) | ((y >> 40) & 0xffffull);
758 return (x << 24) | ((y >> 32) & 0xffffffull);
760 return (x << 32) | ((y >> 24) & 0xffffffffull);
762 return (x << 40) | ((y >> 16) & 0xffffffffffull);
766 On X86-64, we only handle f2/f3/f4 right. On x86-32, a few of these
767 generate truly horrible code, instead of using shld and friends. On
768 ARM, we end up with calls to L___lshrdi3/L___ashldi3 in f, which is
769 badness. PPC64 misses f, f5 and f6. CellSPU aborts in isel.
771 //===---------------------------------------------------------------------===//
773 We do a number of simplifications in simplify libcalls to strength reduce
774 standard library functions, but we don't currently merge them together. For
775 example, it is useful to merge memcpy(a,b,strlen(b)) -> strcpy. This can only
776 be done safely if "b" isn't modified between the strlen and memcpy of course.
778 //===---------------------------------------------------------------------===//
780 We compile this program: (from GCC PR11680)
781 http://gcc.gnu.org/bugzilla/attachment.cgi?id=4487
783 Into code that runs the same speed in fast/slow modes, but both modes run 2x
784 slower than when compile with GCC (either 4.0 or 4.2):
786 $ llvm-g++ perf.cpp -O3 -fno-exceptions
788 1.821u 0.003s 0:01.82 100.0% 0+0k 0+0io 0pf+0w
790 $ g++ perf.cpp -O3 -fno-exceptions
792 0.821u 0.001s 0:00.82 100.0% 0+0k 0+0io 0pf+0w
794 It looks like we are making the same inlining decisions, so this may be raw
795 codegen badness or something else (haven't investigated).
797 //===---------------------------------------------------------------------===//
799 We miss some instcombines for stuff like this:
801 void foo (unsigned int a) {
802 /* This one is equivalent to a >= (3 << 2). */
807 A few other related ones are in GCC PR14753.
809 //===---------------------------------------------------------------------===//
811 Divisibility by constant can be simplified (according to GCC PR12849) from
812 being a mulhi to being a mul lo (cheaper). Testcase:
814 void bar(unsigned n) {
819 This is equivalent to the following, where 2863311531 is the multiplicative
820 inverse of 3, and 1431655766 is ((2^32)-1)/3+1:
821 void bar(unsigned n) {
822 if (n * 2863311531U < 1431655766U)
826 The same transformation can work with an even modulo with the addition of a
827 rotate: rotate the result of the multiply to the right by the number of bits
828 which need to be zero for the condition to be true, and shrink the compare RHS
829 by the same amount. Unless the target supports rotates, though, that
830 transformation probably isn't worthwhile.
832 The transformation can also easily be made to work with non-zero equality
833 comparisons: just transform, for example, "n % 3 == 1" to "(n-1) % 3 == 0".
835 //===---------------------------------------------------------------------===//
837 Better mod/ref analysis for scanf would allow us to eliminate the vtable and a
838 bunch of other stuff from this example (see PR1604):
848 std::scanf("%d", &t.val);
849 std::printf("%d\n", t.val);
852 //===---------------------------------------------------------------------===//
854 These functions perform the same computation, but produce different assembly.
856 define i8 @select(i8 %x) readnone nounwind {
857 %A = icmp ult i8 %x, 250
858 %B = select i1 %A, i8 0, i8 1
862 define i8 @addshr(i8 %x) readnone nounwind {
863 %A = zext i8 %x to i9
864 %B = add i9 %A, 6 ;; 256 - 250 == 6
866 %D = trunc i9 %C to i8
870 //===---------------------------------------------------------------------===//
874 f (unsigned long a, unsigned long b, unsigned long c)
876 return ((a & (c - 1)) != 0) || ((b & (c - 1)) != 0);
879 f (unsigned long a, unsigned long b, unsigned long c)
881 return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0);
883 Both should combine to ((a|b) & (c-1)) != 0. Currently not optimized with
884 "clang -emit-llvm-bc | opt -std-compile-opts".
886 //===---------------------------------------------------------------------===//
889 #define PMD_MASK (~((1UL << 23) - 1))
890 void clear_pmd_range(unsigned long start, unsigned long end)
892 if (!(start & ~PMD_MASK) && !(end & ~PMD_MASK))
895 The expression should optimize to something like
896 "!((start|end)&~PMD_MASK). Currently not optimized with "clang
897 -emit-llvm-bc | opt -std-compile-opts".
899 //===---------------------------------------------------------------------===//
905 return (n >= 0 ? 1 : -1);
907 Should combine to (n >> 31) | 1. Currently not optimized with "clang
908 -emit-llvm-bc | opt -std-compile-opts | llc".
910 //===---------------------------------------------------------------------===//
914 if (variable == 4 || variable == 6)
917 This should optimize to "if ((variable | 2) == 6)". Currently not
918 optimized with "clang -emit-llvm-bc | opt -std-compile-opts | llc".
920 //===---------------------------------------------------------------------===//
922 unsigned int f(unsigned int i, unsigned int n) {++i; if (i == n) ++i; return
924 unsigned int f2(unsigned int i, unsigned int n) {++i; i += i == n; return i;}
925 These should combine to the same thing. Currently, the first function
926 produces better code on X86.
928 //===---------------------------------------------------------------------===//
931 #define abs(x) x>0?x:-x
934 return (abs(x)) >= 0;
936 This should optimize to x == INT_MIN. (With -fwrapv.) Currently not
937 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
939 //===---------------------------------------------------------------------===//
943 rotate_cst (unsigned int a)
945 a = (a << 10) | (a >> 22);
950 minus_cst (unsigned int a)
959 mask_gt (unsigned int a)
961 /* This is equivalent to a > 15. */
966 rshift_gt (unsigned int a)
968 /* This is equivalent to a > 23. */
972 All should simplify to a single comparison. All of these are
973 currently not optimized with "clang -emit-llvm-bc | opt
976 //===---------------------------------------------------------------------===//
979 int c(int* x) {return (char*)x+2 == (char*)x;}
980 Should combine to 0. Currently not optimized with "clang
981 -emit-llvm-bc | opt -std-compile-opts" (although llc can optimize it).
983 //===---------------------------------------------------------------------===//
985 int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;}
986 Should be combined to "((b >> 1) | b) & 1". Currently not optimized
987 with "clang -emit-llvm-bc | opt -std-compile-opts".
989 //===---------------------------------------------------------------------===//
991 unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);}
992 Should combine to "x | (y & 3)". Currently not optimized with "clang
993 -emit-llvm-bc | opt -std-compile-opts".
995 //===---------------------------------------------------------------------===//
997 int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);}
998 Should fold to "(~a & c) | (a & b)". Currently not optimized with
999 "clang -emit-llvm-bc | opt -std-compile-opts".
1001 //===---------------------------------------------------------------------===//
1003 int a(int a,int b) {return (~(a|b))|a;}
1004 Should fold to "a|~b". Currently not optimized with "clang
1005 -emit-llvm-bc | opt -std-compile-opts".
1007 //===---------------------------------------------------------------------===//
1009 int a(int a, int b) {return (a&&b) || (a&&!b);}
1010 Should fold to "a". Currently not optimized with "clang -emit-llvm-bc
1011 | opt -std-compile-opts".
1013 //===---------------------------------------------------------------------===//
1015 int a(int a, int b, int c) {return (a&&b) || (!a&&c);}
1016 Should fold to "a ? b : c", or at least something sane. Currently not
1017 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1019 //===---------------------------------------------------------------------===//
1021 int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);}
1022 Should fold to a && (b || c). Currently not optimized with "clang
1023 -emit-llvm-bc | opt -std-compile-opts".
1025 //===---------------------------------------------------------------------===//
1027 int a(int x) {return x | ((x & 8) ^ 8);}
1028 Should combine to x | 8. Currently not optimized with "clang
1029 -emit-llvm-bc | opt -std-compile-opts".
1031 //===---------------------------------------------------------------------===//
1033 int a(int x) {return x ^ ((x & 8) ^ 8);}
1034 Should also combine to x | 8. Currently not optimized with "clang
1035 -emit-llvm-bc | opt -std-compile-opts".
1037 //===---------------------------------------------------------------------===//
1039 int a(int x) {return (x & 8) == 0 ? -1 : -9;}
1040 Should combine to (x | -9) ^ 8. Currently not optimized with "clang
1041 -emit-llvm-bc | opt -std-compile-opts".
1043 //===---------------------------------------------------------------------===//
1045 int a(int x) {return (x & 8) == 0 ? -9 : -1;}
1046 Should combine to x | -9. Currently not optimized with "clang
1047 -emit-llvm-bc | opt -std-compile-opts".
1049 //===---------------------------------------------------------------------===//
1051 int a(int x) {return ((x | -9) ^ 8) & x;}
1052 Should combine to x & -9. Currently not optimized with "clang
1053 -emit-llvm-bc | opt -std-compile-opts".
1055 //===---------------------------------------------------------------------===//
1057 unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;}
1058 Should combine to "a * 0x88888888 >> 31". Currently not optimized
1059 with "clang -emit-llvm-bc | opt -std-compile-opts".
1061 //===---------------------------------------------------------------------===//
1063 unsigned a(char* x) {if ((*x & 32) == 0) return b();}
1064 There's an unnecessary zext in the generated code with "clang
1065 -emit-llvm-bc | opt -std-compile-opts".
1067 //===---------------------------------------------------------------------===//
1069 unsigned a(unsigned long long x) {return 40 * (x >> 1);}
1070 Should combine to "20 * (((unsigned)x) & -2)". Currently not
1071 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1073 //===---------------------------------------------------------------------===//
1075 This was noticed in the entryblock for grokdeclarator in 403.gcc:
1077 %tmp = icmp eq i32 %decl_context, 4
1078 %decl_context_addr.0 = select i1 %tmp, i32 3, i32 %decl_context
1079 %tmp1 = icmp eq i32 %decl_context_addr.0, 1
1080 %decl_context_addr.1 = select i1 %tmp1, i32 0, i32 %decl_context_addr.0
1082 tmp1 should be simplified to something like:
1083 (!tmp || decl_context == 1)
1085 This allows recursive simplifications, tmp1 is used all over the place in
1086 the function, e.g. by:
1088 %tmp23 = icmp eq i32 %decl_context_addr.1, 0 ; <i1> [#uses=1]
1089 %tmp24 = xor i1 %tmp1, true ; <i1> [#uses=1]
1090 %or.cond8 = and i1 %tmp23, %tmp24 ; <i1> [#uses=1]
1094 //===---------------------------------------------------------------------===//
1098 Store sinking: This code:
1100 void f (int n, int *cond, int *res) {
1103 for (i = 0; i < n; i++)
1105 *res ^= 234; /* (*) */
1108 On this function GVN hoists the fully redundant value of *res, but nothing
1109 moves the store out. This gives us this code:
1111 bb: ; preds = %bb2, %entry
1112 %.rle = phi i32 [ 0, %entry ], [ %.rle6, %bb2 ]
1113 %i.05 = phi i32 [ 0, %entry ], [ %indvar.next, %bb2 ]
1114 %1 = load i32* %cond, align 4
1115 %2 = icmp eq i32 %1, 0
1116 br i1 %2, label %bb2, label %bb1
1119 %3 = xor i32 %.rle, 234
1120 store i32 %3, i32* %res, align 4
1123 bb2: ; preds = %bb, %bb1
1124 %.rle6 = phi i32 [ %3, %bb1 ], [ %.rle, %bb ]
1125 %indvar.next = add i32 %i.05, 1
1126 %exitcond = icmp eq i32 %indvar.next, %n
1127 br i1 %exitcond, label %return, label %bb
1129 DSE should sink partially dead stores to get the store out of the loop.
1131 Here's another partial dead case:
1132 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12395
1134 //===---------------------------------------------------------------------===//
1136 Scalar PRE hoists the mul in the common block up to the else:
1138 int test (int a, int b, int c, int g) {
1148 It would be better to do the mul once to reduce codesize above the if.
1149 This is GCC PR38204.
1151 //===---------------------------------------------------------------------===//
1155 GCC PR37810 is an interesting case where we should sink load/store reload
1156 into the if block and outside the loop, so we don't reload/store it on the
1177 We now hoist the reload after the call (Transforms/GVN/lpre-call-wrap.ll), but
1178 we don't sink the store. We need partially dead store sinking.
1180 //===---------------------------------------------------------------------===//
1182 [LOAD PRE CRIT EDGE SPLITTING]
1184 GCC PR37166: Sinking of loads prevents SROA'ing the "g" struct on the stack
1185 leading to excess stack traffic. This could be handled by GVN with some crazy
1186 symbolic phi translation. The code we get looks like (g is on the stack):
1190 %9 = getelementptr %struct.f* %g, i32 0, i32 0
1191 store i32 %8, i32* %9, align bel %bb3
1193 bb3: ; preds = %bb1, %bb2, %bb
1194 %c_addr.0 = phi %struct.f* [ %g, %bb2 ], [ %c, %bb ], [ %c, %bb1 ]
1195 %b_addr.0 = phi %struct.f* [ %b, %bb2 ], [ %g, %bb ], [ %b, %bb1 ]
1196 %10 = getelementptr %struct.f* %c_addr.0, i32 0, i32 0
1197 %11 = load i32* %10, align 4
1199 %11 is partially redundant, an in BB2 it should have the value %8.
1201 GCC PR33344 and PR35287 are similar cases.
1204 //===---------------------------------------------------------------------===//
1208 There are many load PRE testcases in testsuite/gcc.dg/tree-ssa/loadpre* in the
1209 GCC testsuite, ones we don't get yet are (checked through loadpre25):
1211 [CRIT EDGE BREAKING]
1212 loadpre3.c predcom-4.c
1214 [PRE OF READONLY CALL]
1217 [TURN SELECT INTO BRANCH]
1218 loadpre14.c loadpre15.c
1220 actually a conditional increment: loadpre18.c loadpre19.c
1223 //===---------------------------------------------------------------------===//
1226 There are many PRE testcases in testsuite/gcc.dg/tree-ssa/ssa-pre-*.c in the
1229 //===---------------------------------------------------------------------===//
1231 There are some interesting cases in testsuite/gcc.dg/tree-ssa/pred-comm* in the
1232 GCC testsuite. For example, we get the first example in predcom-1.c, but
1233 miss the second one:
1238 __attribute__ ((noinline))
1239 void count_averages(int n) {
1241 for (i = 1; i < n; i++)
1242 avg[i] = (((unsigned long) fib[i - 1] + fib[i] + fib[i + 1]) / 3) & 0xffff;
1245 which compiles into two loads instead of one in the loop.
1247 predcom-2.c is the same as predcom-1.c
1249 predcom-3.c is very similar but needs loads feeding each other instead of
1253 //===---------------------------------------------------------------------===//
1257 Type based alias analysis:
1258 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14705
1260 We should do better analysis of posix_memalign. At the least it should
1261 no-capture its pointer argument, at best, we should know that the out-value
1262 result doesn't point to anything (like malloc). One example of this is in
1263 SingleSource/Benchmarks/Misc/dt.c
1265 //===---------------------------------------------------------------------===//
1267 A/B get pinned to the stack because we turn an if/then into a select instead
1268 of PRE'ing the load/store. This may be fixable in instcombine:
1269 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37892
1271 struct X { int i; };
1285 //===---------------------------------------------------------------------===//
1287 Interesting missed case because of control flow flattening (should be 2 loads):
1288 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26629
1289 With: llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as |
1290 opt -mem2reg -gvn -instcombine | llvm-dis
1291 we miss it because we need 1) CRIT EDGE 2) MULTIPLE DIFFERENT
1292 VALS PRODUCED BY ONE BLOCK OVER DIFFERENT PATHS
1294 //===---------------------------------------------------------------------===//
1296 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19633
1297 We could eliminate the branch condition here, loading from null is undefined:
1299 struct S { int w, x, y, z; };
1300 struct T { int r; struct S s; };
1301 void bar (struct S, int);
1302 void foo (int a, struct T b)
1310 //===---------------------------------------------------------------------===//
1312 simplifylibcalls should do several optimizations for strspn/strcspn:
1314 strcspn(x, "") -> strlen(x)
1317 strspn(x, "") -> strlen(x)
1318 strspn(x, "a") -> strchr(x, 'a')-x
1320 strcspn(x, "a") -> inlined loop for up to 3 letters (similarly for strspn):
1322 size_t __strcspn_c3 (__const char *__s, int __reject1, int __reject2,
1324 register size_t __result = 0;
1325 while (__s[__result] != '\0' && __s[__result] != __reject1 &&
1326 __s[__result] != __reject2 && __s[__result] != __reject3)
1331 This should turn into a switch on the character. See PR3253 for some notes on
1334 456.hmmer apparently uses strcspn and strspn a lot. 471.omnetpp uses strspn.
1336 //===---------------------------------------------------------------------===//
1338 "gas" uses this idiom:
1339 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
1341 else if (strchr ("<>", *intel_parser.op_string)
1343 Those should be turned into a switch.
1345 //===---------------------------------------------------------------------===//
1347 252.eon contains this interesting code:
1349 %3072 = getelementptr [100 x i8]* %tempString, i32 0, i32 0
1350 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1351 %strlen = call i32 @strlen(i8* %3072) ; uses = 1
1352 %endptr = getelementptr [100 x i8]* %tempString, i32 0, i32 %strlen
1353 call void @llvm.memcpy.i32(i8* %endptr,
1354 i8* getelementptr ([5 x i8]* @"\01LC42", i32 0, i32 0), i32 5, i32 1)
1355 %3074 = call i32 @strlen(i8* %endptr) nounwind readonly
1357 This is interesting for a couple reasons. First, in this:
1359 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1360 %strlen = call i32 @strlen(i8* %3072)
1362 The strlen could be replaced with: %strlen = sub %3072, %3073, because the
1363 strcpy call returns a pointer to the end of the string. Based on that, the
1364 endptr GEP just becomes equal to 3073, which eliminates a strlen call and GEP.
1366 Second, the memcpy+strlen strlen can be replaced with:
1368 %3074 = call i32 @strlen([5 x i8]* @"\01LC42") nounwind readonly
1370 Because the destination was just copied into the specified memory buffer. This,
1371 in turn, can be constant folded to "4".
1373 In other code, it contains:
1375 %endptr6978 = bitcast i8* %endptr69 to i32*
1376 store i32 7107374, i32* %endptr6978, align 1
1377 %3167 = call i32 @strlen(i8* %endptr69) nounwind readonly
1379 Which could also be constant folded. Whatever is producing this should probably
1380 be fixed to leave this as a memcpy from a string.
1382 Further, eon also has an interesting partially redundant strlen call:
1384 bb8: ; preds = %_ZN18eonImageCalculatorC1Ev.exit
1385 %682 = getelementptr i8** %argv, i32 6 ; <i8**> [#uses=2]
1386 %683 = load i8** %682, align 4 ; <i8*> [#uses=4]
1387 %684 = load i8* %683, align 1 ; <i8> [#uses=1]
1388 %685 = icmp eq i8 %684, 0 ; <i1> [#uses=1]
1389 br i1 %685, label %bb10, label %bb9
1392 %686 = call i32 @strlen(i8* %683) nounwind readonly
1393 %687 = icmp ugt i32 %686, 254 ; <i1> [#uses=1]
1394 br i1 %687, label %bb10, label %bb11
1396 bb10: ; preds = %bb9, %bb8
1397 %688 = call i32 @strlen(i8* %683) nounwind readonly
1399 This could be eliminated by doing the strlen once in bb8, saving code size and
1400 improving perf on the bb8->9->10 path.
1402 //===---------------------------------------------------------------------===//
1404 I see an interesting fully redundant call to strlen left in 186.crafty:InputMove
1406 %movetext11 = getelementptr [128 x i8]* %movetext, i32 0, i32 0
1409 bb62: ; preds = %bb55, %bb53
1410 %promote.0 = phi i32 [ %169, %bb55 ], [ 0, %bb53 ]
1411 %171 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1412 %172 = add i32 %171, -1 ; <i32> [#uses=1]
1413 %173 = getelementptr [128 x i8]* %movetext, i32 0, i32 %172
1416 br i1 %or.cond, label %bb65, label %bb72
1418 bb65: ; preds = %bb62
1419 store i8 0, i8* %173, align 1
1422 bb72: ; preds = %bb65, %bb62
1423 %trank.1 = phi i32 [ %176, %bb65 ], [ -1, %bb62 ]
1424 %177 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1426 Note that on the bb62->bb72 path, that the %177 strlen call is partially
1427 redundant with the %171 call. At worst, we could shove the %177 strlen call
1428 up into the bb65 block moving it out of the bb62->bb72 path. However, note
1429 that bb65 stores to the string, zeroing out the last byte. This means that on
1430 that path the value of %177 is actually just %171-1. A sub is cheaper than a
1433 This pattern repeats several times, basically doing:
1438 where it is "obvious" that B = A-1.
1440 //===---------------------------------------------------------------------===//
1442 186.crafty contains this interesting pattern:
1444 %77 = call i8* @strstr(i8* getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0),
1446 %phitmp648 = icmp eq i8* %77, getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0)
1447 br i1 %phitmp648, label %bb70, label %bb76
1449 bb70: ; preds = %OptionMatch.exit91, %bb69
1450 %78 = call i32 @strlen(i8* %30) nounwind readonly align 1 ; <i32> [#uses=1]
1454 if (strstr(cststr, P) == cststr) {
1458 The strstr call would be significantly cheaper written as:
1461 if (memcmp(P, str, strlen(P)))
1464 This is memcmp+strlen instead of strstr. This also makes the strlen fully
1467 //===---------------------------------------------------------------------===//
1469 186.crafty also contains this code:
1471 %1906 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1472 %1907 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1906
1473 %1908 = call i8* @strcpy(i8* %1907, i8* %1905) nounwind align 1
1474 %1909 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1475 %1910 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1909
1477 The last strlen is computable as 1908-@pgn_event, which means 1910=1908.
1479 //===---------------------------------------------------------------------===//
1481 186.crafty has this interesting pattern with the "out.4543" variable:
1483 call void @llvm.memcpy.i32(
1484 i8* getelementptr ([10 x i8]* @out.4543, i32 0, i32 0),
1485 i8* getelementptr ([7 x i8]* @"\01LC28700", i32 0, i32 0), i32 7, i32 1)
1486 %101 = call@printf(i8* ... @out.4543, i32 0, i32 0)) nounwind
1488 It is basically doing:
1490 memcpy(globalarray, "string");
1491 printf(..., globalarray);
1493 Anyway, by knowing that printf just reads the memory and forward substituting
1494 the string directly into the printf, this eliminates reads from globalarray.
1495 Since this pattern occurs frequently in crafty (due to the "DisplayTime" and
1496 other similar functions) there are many stores to "out". Once all the printfs
1497 stop using "out", all that is left is the memcpy's into it. This should allow
1498 globalopt to remove the "stored only" global.
1500 //===---------------------------------------------------------------------===//
1504 define inreg i32 @foo(i8* inreg %p) nounwind {
1506 %tmp1 = ashr i8 %tmp0, 5
1507 %tmp2 = sext i8 %tmp1 to i32
1511 could be dagcombine'd to a sign-extending load with a shift.
1512 For example, on x86 this currently gets this:
1518 while it could get this:
1523 //===---------------------------------------------------------------------===//
1527 int test(int x) { return 1-x == x; } // --> return false
1528 int test2(int x) { return 2-x == x; } // --> return x == 1 ?
1530 Always foldable for odd constants, what is the rule for even?
1532 //===---------------------------------------------------------------------===//
1534 PR 3381: GEP to field of size 0 inside a struct could be turned into GEP
1535 for next field in struct (which is at same address).
1537 For example: store of float into { {{}}, float } could be turned into a store to
1540 //===---------------------------------------------------------------------===//
1543 double foo(double a) { return sin(a); }
1545 This compiles into this on x86-64 Linux:
1556 //===---------------------------------------------------------------------===//
1558 The arg promotion pass should make use of nocapture to make its alias analysis
1559 stuff much more precise.
1561 //===---------------------------------------------------------------------===//
1563 The following functions should be optimized to use a select instead of a
1564 branch (from gcc PR40072):
1566 char char_int(int m) {if(m>7) return 0; return m;}
1567 int int_char(char m) {if(m>7) return 0; return m;}
1569 //===---------------------------------------------------------------------===//
1571 int func(int a, int b) { if (a & 0x80) b |= 0x80; else b &= ~0x80; return b; }
1575 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1577 %0 = and i32 %a, 128 ; <i32> [#uses=1]
1578 %1 = icmp eq i32 %0, 0 ; <i1> [#uses=1]
1579 %2 = or i32 %b, 128 ; <i32> [#uses=1]
1580 %3 = and i32 %b, -129 ; <i32> [#uses=1]
1581 %b_addr.0 = select i1 %1, i32 %3, i32 %2 ; <i32> [#uses=1]
1585 However, it's functionally equivalent to:
1587 b = (b & ~0x80) | (a & 0x80);
1589 Which generates this:
1591 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1593 %0 = and i32 %b, -129 ; <i32> [#uses=1]
1594 %1 = and i32 %a, 128 ; <i32> [#uses=1]
1595 %2 = or i32 %0, %1 ; <i32> [#uses=1]
1599 This can be generalized for other forms:
1601 b = (b & ~0x80) | (a & 0x40) << 1;
1603 //===---------------------------------------------------------------------===//
1605 These two functions produce different code. They shouldn't:
1609 uint8_t p1(uint8_t b, uint8_t a) {
1610 b = (b & ~0xc0) | (a & 0xc0);
1614 uint8_t p2(uint8_t b, uint8_t a) {
1615 b = (b & ~0x40) | (a & 0x40);
1616 b = (b & ~0x80) | (a & 0x80);
1620 define zeroext i8 @p1(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1622 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1623 %1 = and i8 %a, -64 ; <i8> [#uses=1]
1624 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1628 define zeroext i8 @p2(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1630 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1631 %.masked = and i8 %a, 64 ; <i8> [#uses=1]
1632 %1 = and i8 %a, -128 ; <i8> [#uses=1]
1633 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1634 %3 = or i8 %2, %.masked ; <i8> [#uses=1]
1638 //===---------------------------------------------------------------------===//
1640 IPSCCP does not currently propagate argument dependent constants through
1641 functions where it does not not all of the callers. This includes functions
1642 with normal external linkage as well as templates, C99 inline functions etc.
1643 Specifically, it does nothing to:
1645 define i32 @test(i32 %x, i32 %y, i32 %z) nounwind {
1647 %0 = add nsw i32 %y, %z
1650 %3 = add nsw i32 %1, %2
1654 define i32 @test2() nounwind {
1656 %0 = call i32 @test(i32 1, i32 2, i32 4) nounwind
1660 It would be interesting extend IPSCCP to be able to handle simple cases like
1661 this, where all of the arguments to a call are constant. Because IPSCCP runs
1662 before inlining, trivial templates and inline functions are not yet inlined.
1663 The results for a function + set of constant arguments should be memoized in a
1666 //===---------------------------------------------------------------------===//
1668 The libcall constant folding stuff should be moved out of SimplifyLibcalls into
1669 libanalysis' constantfolding logic. This would allow IPSCCP to be able to
1670 handle simple things like this:
1672 static int foo(const char *X) { return strlen(X); }
1673 int bar() { return foo("abcd"); }
1675 //===---------------------------------------------------------------------===//
1677 InstCombine should use SimplifyDemandedBits to remove the or instruction:
1679 define i1 @test(i8 %x, i8 %y) {
1681 %B = icmp ugt i8 %A, 3
1685 Currently instcombine calls SimplifyDemandedBits with either all bits or just
1686 the sign bit, if the comparison is obviously a sign test. In this case, we only
1687 need all but the bottom two bits from %A, and if we gave that mask to SDB it
1688 would delete the or instruction for us.
1690 //===---------------------------------------------------------------------===//
1692 functionattrs doesn't know much about memcpy/memset. This function should be
1693 marked readnone rather than readonly, since it only twiddles local memory, but
1694 functionattrs doesn't handle memset/memcpy/memmove aggressively:
1696 struct X { int *p; int *q; };
1703 p = __builtin_memcpy (&x, &y, sizeof (int *));
1707 //===---------------------------------------------------------------------===//
1709 Missed instcombine transformation:
1710 define i1 @a(i32 %x) nounwind readnone {
1712 %cmp = icmp eq i32 %x, 30
1713 %sub = add i32 %x, -30
1714 %cmp2 = icmp ugt i32 %sub, 9
1715 %or = or i1 %cmp, %cmp2
1718 This should be optimized to a single compare. Testcase derived from gcc.
1720 //===---------------------------------------------------------------------===//
1722 Missed instcombine transformation:
1724 void a(int x) { if (((1<<x)&8)==0) b(); }
1726 The shift should be optimized out. Testcase derived from gcc.
1728 //===---------------------------------------------------------------------===//
1730 Missed instcombine or reassociate transformation:
1731 int a(int a, int b) { return (a==12)&(b>47)&(b<58); }
1733 The sgt and slt should be combined into a single comparison. Testcase derived
1736 //===---------------------------------------------------------------------===//
1738 Missed instcombine transformation:
1739 define i32 @a(i32 %x) nounwind readnone {
1741 %rem = srem i32 %x, 32
1742 %shl = shl i32 1, %rem
1746 The srem can be transformed to an and because if x is negative, the shift is
1747 undefined. Testcase derived from gcc.
1749 //===---------------------------------------------------------------------===//
1751 Missed instcombine/dagcombine transformation:
1752 define i32 @a(i32 %x, i32 %y) nounwind readnone {
1754 %mul = mul i32 %y, -8
1755 %sub = sub i32 %x, %mul
1759 Should compile to something like x+y*8, but currently compiles to an
1760 inefficient result. Testcase derived from gcc.
1762 //===---------------------------------------------------------------------===//
1764 Missed instcombine/dagcombine transformation:
1765 define void @lshift_lt(i8 zeroext %a) nounwind {
1767 %conv = zext i8 %a to i32
1768 %shl = shl i32 %conv, 3
1769 %cmp = icmp ult i32 %shl, 33
1770 br i1 %cmp, label %if.then, label %if.end
1773 tail call void @bar() nounwind
1779 declare void @bar() nounwind
1781 The shift should be eliminated. Testcase derived from gcc.
1783 //===---------------------------------------------------------------------===//
1785 These compile into different code, one gets recognized as a switch and the
1786 other doesn't due to phase ordering issues (PR6212):
1788 int test1(int mainType, int subType) {
1791 else if (mainType == 9)
1793 else if (mainType == 11)
1798 int test2(int mainType, int subType) {
1808 //===---------------------------------------------------------------------===//
1810 The following test case (from PR6576):
1812 define i32 @mul(i32 %a, i32 %b) nounwind readnone {
1814 %cond1 = icmp eq i32 %b, 0 ; <i1> [#uses=1]
1815 br i1 %cond1, label %exit, label %bb.nph
1816 bb.nph: ; preds = %entry
1817 %tmp = mul i32 %b, %a ; <i32> [#uses=1]
1819 exit: ; preds = %entry
1823 could be reduced to:
1825 define i32 @mul(i32 %a, i32 %b) nounwind readnone {
1827 %tmp = mul i32 %b, %a
1831 //===---------------------------------------------------------------------===//
1833 We should use DSE + llvm.lifetime.end to delete dead vtable pointer updates.
1836 //===---------------------------------------------------------------------===//
1841 return x > 1 ? x : 1;
1844 LLVM emits a comparison with 1 instead of 0. 0 would be equivalent
1845 and cheaper on most targets.
1847 LLVM prefers comparisons with zero over non-zero in general, but in this
1848 case it choses instead to keep the max operation obvious.
1850 //===---------------------------------------------------------------------===//