1 Target Independent Opportunities:
3 //===---------------------------------------------------------------------===//
5 With the recent changes to make the implicit def/use set explicit in
6 machineinstrs, we should change the target descriptions for 'call' instructions
7 so that the .td files don't list all the call-clobbered registers as implicit
8 defs. Instead, these should be added by the code generator (e.g. on the dag).
10 This has a number of uses:
12 1. PPC32/64 and X86 32/64 can avoid having multiple copies of call instructions
13 for their different impdef sets.
14 2. Targets with multiple calling convs (e.g. x86) which have different clobber
15 sets don't need copies of call instructions.
16 3. 'Interprocedural register allocation' can be done to reduce the clobber sets
19 //===---------------------------------------------------------------------===//
21 Make the PPC branch selector target independant
23 //===---------------------------------------------------------------------===//
25 Get the C front-end to expand hypot(x,y) -> llvm.sqrt(x*x+y*y) when errno and
26 precision don't matter (ffastmath). Misc/mandel will like this. :) This isn't
27 safe in general, even on darwin. See the libm implementation of hypot for
28 examples (which special case when x/y are exactly zero to get signed zeros etc
31 //===---------------------------------------------------------------------===//
33 Solve this DAG isel folding deficiency:
51 The problem is the store's chain operand is not the load X but rather
52 a TokenFactor of the load X and load Y, which prevents the folding.
54 There are two ways to fix this:
56 1. The dag combiner can start using alias analysis to realize that y/x
57 don't alias, making the store to X not dependent on the load from Y.
58 2. The generated isel could be made smarter in the case it can't
59 disambiguate the pointers.
61 Number 1 is the preferred solution.
63 This has been "fixed" by a TableGen hack. But that is a short term workaround
64 which will be removed once the proper fix is made.
66 //===---------------------------------------------------------------------===//
68 On targets with expensive 64-bit multiply, we could LSR this:
75 for (i = ...; ++i, tmp+=tmp)
78 This would be a win on ppc32, but not x86 or ppc64.
80 //===---------------------------------------------------------------------===//
82 Shrink: (setlt (loadi32 P), 0) -> (setlt (loadi8 Phi), 0)
84 //===---------------------------------------------------------------------===//
86 Reassociate should turn: X*X*X*X -> t=(X*X) (t*t) to eliminate a multiply.
88 //===---------------------------------------------------------------------===//
90 Interesting? testcase for add/shift/mul reassoc:
92 int bar(int x, int y) {
93 return x*x*x+y+x*x*x*x*x*y*y*y*y;
95 int foo(int z, int n) {
96 return bar(z, n) + bar(2*z, 2*n);
99 Reassociate should handle the example in GCC PR16157.
101 //===---------------------------------------------------------------------===//
103 These two functions should generate the same code on big-endian systems:
105 int g(int *j,int *l) { return memcmp(j,l,4); }
106 int h(int *j, int *l) { return *j - *l; }
108 this could be done in SelectionDAGISel.cpp, along with other special cases,
111 //===---------------------------------------------------------------------===//
113 It would be nice to revert this patch:
114 http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20060213/031986.html
116 And teach the dag combiner enough to simplify the code expanded before
117 legalize. It seems plausible that this knowledge would let it simplify other
120 //===---------------------------------------------------------------------===//
122 For vector types, TargetData.cpp::getTypeInfo() returns alignment that is equal
123 to the type size. It works but can be overly conservative as the alignment of
124 specific vector types are target dependent.
126 //===---------------------------------------------------------------------===//
128 We should produce an unaligned load from code like this:
130 v4sf example(float *P) {
131 return (v4sf){P[0], P[1], P[2], P[3] };
134 //===---------------------------------------------------------------------===//
136 Add support for conditional increments, and other related patterns. Instead
141 je LBB16_2 #cond_next
152 //===---------------------------------------------------------------------===//
154 Combine: a = sin(x), b = cos(x) into a,b = sincos(x).
156 Expand these to calls of sin/cos and stores:
157 double sincos(double x, double *sin, double *cos);
158 float sincosf(float x, float *sin, float *cos);
159 long double sincosl(long double x, long double *sin, long double *cos);
161 Doing so could allow SROA of the destination pointers. See also:
162 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=17687
164 This is now easily doable with MRVs. We could even make an intrinsic for this
165 if anyone cared enough about sincos.
167 //===---------------------------------------------------------------------===//
169 Turn this into a single byte store with no load (the other 3 bytes are
172 define void @test(i32* %P) {
174 %tmp14 = or i32 %tmp, 3305111552
175 %tmp15 = and i32 %tmp14, 3321888767
176 store i32 %tmp15, i32* %P
180 //===---------------------------------------------------------------------===//
182 dag/inst combine "clz(x)>>5 -> x==0" for 32-bit x.
188 int t = __builtin_clz(x);
198 //===---------------------------------------------------------------------===//
200 quantum_sigma_x in 462.libquantum contains the following loop:
202 for(i=0; i<reg->size; i++)
204 /* Flip the target bit of each basis state */
205 reg->node[i].state ^= ((MAX_UNSIGNED) 1 << target);
208 Where MAX_UNSIGNED/state is a 64-bit int. On a 32-bit platform it would be just
209 so cool to turn it into something like:
211 long long Res = ((MAX_UNSIGNED) 1 << target);
213 for(i=0; i<reg->size; i++)
214 reg->node[i].state ^= Res & 0xFFFFFFFFULL;
216 for(i=0; i<reg->size; i++)
217 reg->node[i].state ^= Res & 0xFFFFFFFF00000000ULL
220 ... which would only do one 32-bit XOR per loop iteration instead of two.
222 It would also be nice to recognize the reg->size doesn't alias reg->node[i], but
225 //===---------------------------------------------------------------------===//
227 This should be optimized to one 'and' and one 'or', from PR4216:
229 define i32 @test_bitfield(i32 %bf.prev.low) nounwind ssp {
231 %bf.prev.lo.cleared10 = or i32 %bf.prev.low, 32962 ; <i32> [#uses=1]
232 %0 = and i32 %bf.prev.low, -65536 ; <i32> [#uses=1]
233 %1 = and i32 %bf.prev.lo.cleared10, 40186 ; <i32> [#uses=1]
234 %2 = or i32 %1, %0 ; <i32> [#uses=1]
238 //===---------------------------------------------------------------------===//
240 This isn't recognized as bswap by instcombine (yes, it really is bswap):
242 unsigned long reverse(unsigned v) {
244 t = v ^ ((v << 16) | (v >> 16));
246 v = (v << 24) | (v >> 8);
250 //===---------------------------------------------------------------------===//
252 These idioms should be recognized as popcount (see PR1488):
254 unsigned countbits_slow(unsigned v) {
256 for (c = 0; v; v >>= 1)
260 unsigned countbits_fast(unsigned v){
263 v &= v - 1; // clear the least significant bit set
267 BITBOARD = unsigned long long
268 int PopCnt(register BITBOARD a) {
276 unsigned int popcount(unsigned int input) {
277 unsigned int count = 0;
278 for (unsigned int i = 0; i < 4 * 8; i++)
279 count += (input >> i) & i;
283 //===---------------------------------------------------------------------===//
285 These should turn into single 16-bit (unaligned?) loads on little/big endian
288 unsigned short read_16_le(const unsigned char *adr) {
289 return adr[0] | (adr[1] << 8);
291 unsigned short read_16_be(const unsigned char *adr) {
292 return (adr[0] << 8) | adr[1];
295 //===---------------------------------------------------------------------===//
297 -instcombine should handle this transform:
298 icmp pred (sdiv X / C1 ), C2
299 when X, C1, and C2 are unsigned. Similarly for udiv and signed operands.
301 Currently InstCombine avoids this transform but will do it when the signs of
302 the operands and the sign of the divide match. See the FIXME in
303 InstructionCombining.cpp in the visitSetCondInst method after the switch case
304 for Instruction::UDiv (around line 4447) for more details.
306 The SingleSource/Benchmarks/Shootout-C++/hash and hash2 tests have examples of
309 //===---------------------------------------------------------------------===//
311 viterbi speeds up *significantly* if the various "history" related copy loops
312 are turned into memcpy calls at the source level. We need a "loops to memcpy"
315 //===---------------------------------------------------------------------===//
319 typedef unsigned U32;
320 typedef unsigned long long U64;
321 int test (U32 *inst, U64 *regs) {
324 int r1 = (temp >> 20) & 0xf;
325 int b2 = (temp >> 16) & 0xf;
326 effective_addr2 = temp & 0xfff;
327 if (b2) effective_addr2 += regs[b2];
328 b2 = (temp >> 12) & 0xf;
329 if (b2) effective_addr2 += regs[b2];
330 effective_addr2 &= regs[4];
331 if ((effective_addr2 & 3) == 0)
336 Note that only the low 2 bits of effective_addr2 are used. On 32-bit systems,
337 we don't eliminate the computation of the top half of effective_addr2 because
338 we don't have whole-function selection dags. On x86, this means we use one
339 extra register for the function when effective_addr2 is declared as U64 than
340 when it is declared U32.
342 //===---------------------------------------------------------------------===//
344 LSR should know what GPR types a target has. This code:
346 volatile short X, Y; // globals
350 for (i = 0; i < N; i++) { X = i; Y = i*4; }
353 produces two near identical IV's (after promotion) on PPC/ARM:
363 add r2, r2, #1 <- [0,+,1]
364 sub r0, r0, #1 <- [0,-,1]
368 LSR should reuse the "+" IV for the exit test.
371 //===---------------------------------------------------------------------===//
373 Tail call elim should be more aggressive, checking to see if the call is
374 followed by an uncond branch to an exit block.
376 ; This testcase is due to tail-duplication not wanting to copy the return
377 ; instruction into the terminating blocks because there was other code
378 ; optimized out of the function after the taildup happened.
379 ; RUN: llvm-as < %s | opt -tailcallelim | llvm-dis | not grep call
381 define i32 @t4(i32 %a) {
383 %tmp.1 = and i32 %a, 1 ; <i32> [#uses=1]
384 %tmp.2 = icmp ne i32 %tmp.1, 0 ; <i1> [#uses=1]
385 br i1 %tmp.2, label %then.0, label %else.0
387 then.0: ; preds = %entry
388 %tmp.5 = add i32 %a, -1 ; <i32> [#uses=1]
389 %tmp.3 = call i32 @t4( i32 %tmp.5 ) ; <i32> [#uses=1]
392 else.0: ; preds = %entry
393 %tmp.7 = icmp ne i32 %a, 0 ; <i1> [#uses=1]
394 br i1 %tmp.7, label %then.1, label %return
396 then.1: ; preds = %else.0
397 %tmp.11 = add i32 %a, -2 ; <i32> [#uses=1]
398 %tmp.9 = call i32 @t4( i32 %tmp.11 ) ; <i32> [#uses=1]
401 return: ; preds = %then.1, %else.0, %then.0
402 %result.0 = phi i32 [ 0, %else.0 ], [ %tmp.3, %then.0 ],
407 //===---------------------------------------------------------------------===//
409 Tail recursion elimination should handle:
414 return 2 * pow2m1 (n - 1) + 1;
417 Also, multiplies can be turned into SHL's, so they should be handled as if
418 they were associative. "return foo() << 1" can be tail recursion eliminated.
420 //===---------------------------------------------------------------------===//
422 Argument promotion should promote arguments for recursive functions, like
425 ; RUN: llvm-as < %s | opt -argpromotion | llvm-dis | grep x.val
427 define internal i32 @foo(i32* %x) {
429 %tmp = load i32* %x ; <i32> [#uses=0]
430 %tmp.foo = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
434 define i32 @bar(i32* %x) {
436 %tmp3 = call i32 @foo( i32* %x ) ; <i32> [#uses=1]
440 //===---------------------------------------------------------------------===//
442 "basicaa" should know how to look through "or" instructions that act like add
443 instructions. For example in this code, the x*4+1 is turned into x*4 | 1, and
444 basicaa can't analyze the array subscript, leading to duplicated loads in the
447 void test(int X, int Y, int a[]) {
449 for (i=2; i<1000; i+=4) {
450 a[i+0] = a[i-1+0]*a[i-2+0];
451 a[i+1] = a[i-1+1]*a[i-2+1];
452 a[i+2] = a[i-1+2]*a[i-2+2];
453 a[i+3] = a[i-1+3]*a[i-2+3];
457 BasicAA also doesn't do this for add. It needs to know that &A[i+1] != &A[i].
459 //===---------------------------------------------------------------------===//
461 We should investigate an instruction sinking pass. Consider this silly
477 je LBB1_2 # cond_true
485 The PIC base computation (call+popl) is only used on one path through the
486 code, but is currently always computed in the entry block. It would be
487 better to sink the picbase computation down into the block for the
488 assertion, as it is the only one that uses it. This happens for a lot of
489 code with early outs.
491 Another example is loads of arguments, which are usually emitted into the
492 entry block on targets like x86. If not used in all paths through a
493 function, they should be sunk into the ones that do.
495 In this case, whole-function-isel would also handle this.
497 //===---------------------------------------------------------------------===//
499 Investigate lowering of sparse switch statements into perfect hash tables:
500 http://burtleburtle.net/bob/hash/perfect.html
502 //===---------------------------------------------------------------------===//
504 We should turn things like "load+fabs+store" and "load+fneg+store" into the
505 corresponding integer operations. On a yonah, this loop:
510 for (b = 0; b < 10000000; b++)
511 for (i = 0; i < 256; i++)
515 is twice as slow as this loop:
520 for (b = 0; b < 10000000; b++)
521 for (i = 0; i < 256; i++)
522 a[i] ^= (1ULL << 63);
525 and I suspect other processors are similar. On X86 in particular this is a
526 big win because doing this with integers allows the use of read/modify/write
529 //===---------------------------------------------------------------------===//
531 DAG Combiner should try to combine small loads into larger loads when
532 profitable. For example, we compile this C++ example:
534 struct THotKey { short Key; bool Control; bool Shift; bool Alt; };
535 extern THotKey m_HotKey;
536 THotKey GetHotKey () { return m_HotKey; }
538 into (-O3 -fno-exceptions -static -fomit-frame-pointer):
543 movb _m_HotKey+3, %cl
544 movb _m_HotKey+4, %dl
545 movb _m_HotKey+2, %ch
560 movzwl _m_HotKey+4, %edx
564 The LLVM IR contains the needed alignment info, so we should be able to
565 merge the loads and stores into 4-byte loads:
567 %struct.THotKey = type { i16, i8, i8, i8 }
568 define void @_Z9GetHotKeyv(%struct.THotKey* sret %agg.result) nounwind {
570 %tmp2 = load i16* getelementptr (@m_HotKey, i32 0, i32 0), align 8
571 %tmp5 = load i8* getelementptr (@m_HotKey, i32 0, i32 1), align 2
572 %tmp8 = load i8* getelementptr (@m_HotKey, i32 0, i32 2), align 1
573 %tmp11 = load i8* getelementptr (@m_HotKey, i32 0, i32 3), align 2
575 Alternatively, we should use a small amount of base-offset alias analysis
576 to make it so the scheduler doesn't need to hold all the loads in regs at
579 //===---------------------------------------------------------------------===//
581 We should add an FRINT node to the DAG to model targets that have legal
582 implementations of ceil/floor/rint.
584 //===---------------------------------------------------------------------===//
589 long long input[8] = {1,1,1,1,1,1,1,1};
593 We currently compile this into a memcpy from a global array since the
594 initializer is fairly large and not memset'able. This is good, but the memcpy
595 gets lowered to load/stores in the code generator. This is also ok, except
596 that the codegen lowering for memcpy doesn't handle the case when the source
597 is a constant global. This gives us atrocious code like this:
602 movl _C.0.1444-"L1$pb"+32(%eax), %ecx
604 movl _C.0.1444-"L1$pb"+20(%eax), %ecx
606 movl _C.0.1444-"L1$pb"+36(%eax), %ecx
608 movl _C.0.1444-"L1$pb"+44(%eax), %ecx
610 movl _C.0.1444-"L1$pb"+40(%eax), %ecx
612 movl _C.0.1444-"L1$pb"+12(%eax), %ecx
614 movl _C.0.1444-"L1$pb"+4(%eax), %ecx
626 //===---------------------------------------------------------------------===//
628 http://llvm.org/PR717:
630 The following code should compile into "ret int undef". Instead, LLVM
631 produces "ret int 0":
640 //===---------------------------------------------------------------------===//
642 The loop unroller should partially unroll loops (instead of peeling them)
643 when code growth isn't too bad and when an unroll count allows simplification
644 of some code within the loop. One trivial example is:
650 for ( nLoop = 0; nLoop < 1000; nLoop++ ) {
659 Unrolling by 2 would eliminate the '&1' in both copies, leading to a net
660 reduction in code size. The resultant code would then also be suitable for
661 exit value computation.
663 //===---------------------------------------------------------------------===//
665 We miss a bunch of rotate opportunities on various targets, including ppc, x86,
666 etc. On X86, we miss a bunch of 'rotate by variable' cases because the rotate
667 matching code in dag combine doesn't look through truncates aggressively
668 enough. Here are some testcases reduces from GCC PR17886:
670 unsigned long long f(unsigned long long x, int y) {
671 return (x << y) | (x >> 64-y);
673 unsigned f2(unsigned x, int y){
674 return (x << y) | (x >> 32-y);
676 unsigned long long f3(unsigned long long x){
678 return (x << y) | (x >> 64-y);
680 unsigned f4(unsigned x){
682 return (x << y) | (x >> 32-y);
684 unsigned long long f5(unsigned long long x, unsigned long long y) {
685 return (x << 8) | ((y >> 48) & 0xffull);
687 unsigned long long f6(unsigned long long x, unsigned long long y, int z) {
690 return (x << 8) | ((y >> 48) & 0xffull);
692 return (x << 16) | ((y >> 40) & 0xffffull);
694 return (x << 24) | ((y >> 32) & 0xffffffull);
696 return (x << 32) | ((y >> 24) & 0xffffffffull);
698 return (x << 40) | ((y >> 16) & 0xffffffffffull);
702 On X86-64, we only handle f2/f3/f4 right. On x86-32, a few of these
703 generate truly horrible code, instead of using shld and friends. On
704 ARM, we end up with calls to L___lshrdi3/L___ashldi3 in f, which is
705 badness. PPC64 misses f, f5 and f6. CellSPU aborts in isel.
707 //===---------------------------------------------------------------------===//
709 We do a number of simplifications in simplify libcalls to strength reduce
710 standard library functions, but we don't currently merge them together. For
711 example, it is useful to merge memcpy(a,b,strlen(b)) -> strcpy. This can only
712 be done safely if "b" isn't modified between the strlen and memcpy of course.
714 //===---------------------------------------------------------------------===//
716 Reassociate should turn things like:
718 int factorial(int X) {
719 return X*X*X*X*X*X*X*X;
722 into llvm.powi calls, allowing the code generator to produce balanced
723 multiplication trees.
725 //===---------------------------------------------------------------------===//
727 We generate a horrible libcall for llvm.powi. For example, we compile:
730 double f(double a) { return std::pow(a, 4); }
736 movsd 16(%esp), %xmm0
739 call L___powidf2$stub
747 movsd 16(%esp), %xmm0
755 //===---------------------------------------------------------------------===//
757 We compile this program: (from GCC PR11680)
758 http://gcc.gnu.org/bugzilla/attachment.cgi?id=4487
760 Into code that runs the same speed in fast/slow modes, but both modes run 2x
761 slower than when compile with GCC (either 4.0 or 4.2):
763 $ llvm-g++ perf.cpp -O3 -fno-exceptions
765 1.821u 0.003s 0:01.82 100.0% 0+0k 0+0io 0pf+0w
767 $ g++ perf.cpp -O3 -fno-exceptions
769 0.821u 0.001s 0:00.82 100.0% 0+0k 0+0io 0pf+0w
771 It looks like we are making the same inlining decisions, so this may be raw
772 codegen badness or something else (haven't investigated).
774 //===---------------------------------------------------------------------===//
776 We miss some instcombines for stuff like this:
778 void foo (unsigned int a) {
779 /* This one is equivalent to a >= (3 << 2). */
784 A few other related ones are in GCC PR14753.
786 //===---------------------------------------------------------------------===//
788 Divisibility by constant can be simplified (according to GCC PR12849) from
789 being a mulhi to being a mul lo (cheaper). Testcase:
791 void bar(unsigned n) {
796 I think this basically amounts to a dag combine to simplify comparisons against
797 multiply hi's into a comparison against the mullo.
799 //===---------------------------------------------------------------------===//
801 Better mod/ref analysis for scanf would allow us to eliminate the vtable and a
802 bunch of other stuff from this example (see PR1604):
812 std::scanf("%d", &t.val);
813 std::printf("%d\n", t.val);
816 //===---------------------------------------------------------------------===//
818 Instcombine will merge comparisons like (x >= 10) && (x < 20) by producing (x -
819 10) u< 10, but only when the comparisons have matching sign.
821 This could be converted with a similiar technique. (PR1941)
823 define i1 @test(i8 %x) {
824 %A = icmp uge i8 %x, 5
825 %B = icmp slt i8 %x, 20
830 //===---------------------------------------------------------------------===//
832 These functions perform the same computation, but produce different assembly.
834 define i8 @select(i8 %x) readnone nounwind {
835 %A = icmp ult i8 %x, 250
836 %B = select i1 %A, i8 0, i8 1
840 define i8 @addshr(i8 %x) readnone nounwind {
841 %A = zext i8 %x to i9
842 %B = add i9 %A, 6 ;; 256 - 250 == 6
844 %D = trunc i9 %C to i8
848 //===---------------------------------------------------------------------===//
852 f (unsigned long a, unsigned long b, unsigned long c)
854 return ((a & (c - 1)) != 0) || ((b & (c - 1)) != 0);
857 f (unsigned long a, unsigned long b, unsigned long c)
859 return ((a & (c - 1)) != 0) | ((b & (c - 1)) != 0);
861 Both should combine to ((a|b) & (c-1)) != 0. Currently not optimized with
862 "clang -emit-llvm-bc | opt -std-compile-opts".
864 //===---------------------------------------------------------------------===//
867 #define PMD_MASK (~((1UL << 23) - 1))
868 void clear_pmd_range(unsigned long start, unsigned long end)
870 if (!(start & ~PMD_MASK) && !(end & ~PMD_MASK))
873 The expression should optimize to something like
874 "!((start|end)&~PMD_MASK). Currently not optimized with "clang
875 -emit-llvm-bc | opt -std-compile-opts".
877 //===---------------------------------------------------------------------===//
881 foo (unsigned int a, unsigned int b)
883 if (a <= 7 && b <= 7)
886 Should combine to "(a|b) <= 7". Currently not optimized with "clang
887 -emit-llvm-bc | opt -std-compile-opts".
889 //===---------------------------------------------------------------------===//
895 return (n >= 0 ? 1 : -1);
897 Should combine to (n >> 31) | 1. Currently not optimized with "clang
898 -emit-llvm-bc | opt -std-compile-opts | llc".
900 //===---------------------------------------------------------------------===//
903 int test(int a, int b)
910 Should combine to "a <= b". Currently not optimized with "clang
911 -emit-llvm-bc | opt -std-compile-opts | llc".
913 //===---------------------------------------------------------------------===//
917 if (variable == 4 || variable == 6)
920 This should optimize to "if ((variable | 2) == 6)". Currently not
921 optimized with "clang -emit-llvm-bc | opt -std-compile-opts | llc".
923 //===---------------------------------------------------------------------===//
925 unsigned int f(unsigned int i, unsigned int n) {++i; if (i == n) ++i; return
927 unsigned int f2(unsigned int i, unsigned int n) {++i; i += i == n; return i;}
928 These should combine to the same thing. Currently, the first function
929 produces better code on X86.
931 //===---------------------------------------------------------------------===//
934 #define abs(x) x>0?x:-x
937 return (abs(x)) >= 0;
939 This should optimize to x == INT_MIN. (With -fwrapv.) Currently not
940 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
942 //===---------------------------------------------------------------------===//
946 rotate_cst (unsigned int a)
948 a = (a << 10) | (a >> 22);
953 minus_cst (unsigned int a)
962 mask_gt (unsigned int a)
964 /* This is equivalent to a > 15. */
969 rshift_gt (unsigned int a)
971 /* This is equivalent to a > 23. */
975 All should simplify to a single comparison. All of these are
976 currently not optimized with "clang -emit-llvm-bc | opt
979 //===---------------------------------------------------------------------===//
982 int c(int* x) {return (char*)x+2 == (char*)x;}
983 Should combine to 0. Currently not optimized with "clang
984 -emit-llvm-bc | opt -std-compile-opts" (although llc can optimize it).
986 //===---------------------------------------------------------------------===//
988 int a(unsigned char* b) {return *b > 99;}
989 There's an unnecessary zext in the generated code with "clang
990 -emit-llvm-bc | opt -std-compile-opts".
992 //===---------------------------------------------------------------------===//
994 int a(unsigned b) {return ((b << 31) | (b << 30)) >> 31;}
995 Should be combined to "((b >> 1) | b) & 1". Currently not optimized
996 with "clang -emit-llvm-bc | opt -std-compile-opts".
998 //===---------------------------------------------------------------------===//
1000 unsigned a(unsigned x, unsigned y) { return x | (y & 1) | (y & 2);}
1001 Should combine to "x | (y & 3)". Currently not optimized with "clang
1002 -emit-llvm-bc | opt -std-compile-opts".
1004 //===---------------------------------------------------------------------===//
1006 unsigned a(unsigned a) {return ((a | 1) & 3) | (a & -4);}
1007 Should combine to "a | 1". Currently not optimized with "clang
1008 -emit-llvm-bc | opt -std-compile-opts".
1010 //===---------------------------------------------------------------------===//
1012 int a(int a, int b, int c) {return (~a & c) | ((c|a) & b);}
1013 Should fold to "(~a & c) | (a & b)". Currently not optimized with
1014 "clang -emit-llvm-bc | opt -std-compile-opts".
1016 //===---------------------------------------------------------------------===//
1018 int a(int a,int b) {return (~(a|b))|a;}
1019 Should fold to "a|~b". Currently not optimized with "clang
1020 -emit-llvm-bc | opt -std-compile-opts".
1022 //===---------------------------------------------------------------------===//
1024 int a(int a, int b) {return (a&&b) || (a&&!b);}
1025 Should fold to "a". Currently not optimized with "clang -emit-llvm-bc
1026 | opt -std-compile-opts".
1028 //===---------------------------------------------------------------------===//
1030 int a(int a, int b, int c) {return (a&&b) || (!a&&c);}
1031 Should fold to "a ? b : c", or at least something sane. Currently not
1032 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1034 //===---------------------------------------------------------------------===//
1036 int a(int a, int b, int c) {return (a&&b) || (a&&c) || (a&&b&&c);}
1037 Should fold to a && (b || c). Currently not optimized with "clang
1038 -emit-llvm-bc | opt -std-compile-opts".
1040 //===---------------------------------------------------------------------===//
1042 int a(int x) {return x | ((x & 8) ^ 8);}
1043 Should combine to x | 8. Currently not optimized with "clang
1044 -emit-llvm-bc | opt -std-compile-opts".
1046 //===---------------------------------------------------------------------===//
1048 int a(int x) {return x ^ ((x & 8) ^ 8);}
1049 Should also combine to x | 8. Currently not optimized with "clang
1050 -emit-llvm-bc | opt -std-compile-opts".
1052 //===---------------------------------------------------------------------===//
1054 int a(int x) {return (x & 8) == 0 ? -1 : -9;}
1055 Should combine to (x | -9) ^ 8. Currently not optimized with "clang
1056 -emit-llvm-bc | opt -std-compile-opts".
1058 //===---------------------------------------------------------------------===//
1060 int a(int x) {return (x & 8) == 0 ? -9 : -1;}
1061 Should combine to x | -9. Currently not optimized with "clang
1062 -emit-llvm-bc | opt -std-compile-opts".
1064 //===---------------------------------------------------------------------===//
1066 int a(int x) {return ((x | -9) ^ 8) & x;}
1067 Should combine to x & -9. Currently not optimized with "clang
1068 -emit-llvm-bc | opt -std-compile-opts".
1070 //===---------------------------------------------------------------------===//
1072 unsigned a(unsigned a) {return a * 0x11111111 >> 28 & 1;}
1073 Should combine to "a * 0x88888888 >> 31". Currently not optimized
1074 with "clang -emit-llvm-bc | opt -std-compile-opts".
1076 //===---------------------------------------------------------------------===//
1078 unsigned a(char* x) {if ((*x & 32) == 0) return b();}
1079 There's an unnecessary zext in the generated code with "clang
1080 -emit-llvm-bc | opt -std-compile-opts".
1082 //===---------------------------------------------------------------------===//
1084 unsigned a(unsigned long long x) {return 40 * (x >> 1);}
1085 Should combine to "20 * (((unsigned)x) & -2)". Currently not
1086 optimized with "clang -emit-llvm-bc | opt -std-compile-opts".
1088 //===---------------------------------------------------------------------===//
1090 This was noticed in the entryblock for grokdeclarator in 403.gcc:
1092 %tmp = icmp eq i32 %decl_context, 4
1093 %decl_context_addr.0 = select i1 %tmp, i32 3, i32 %decl_context
1094 %tmp1 = icmp eq i32 %decl_context_addr.0, 1
1095 %decl_context_addr.1 = select i1 %tmp1, i32 0, i32 %decl_context_addr.0
1097 tmp1 should be simplified to something like:
1098 (!tmp || decl_context == 1)
1100 This allows recursive simplifications, tmp1 is used all over the place in
1101 the function, e.g. by:
1103 %tmp23 = icmp eq i32 %decl_context_addr.1, 0 ; <i1> [#uses=1]
1104 %tmp24 = xor i1 %tmp1, true ; <i1> [#uses=1]
1105 %or.cond8 = and i1 %tmp23, %tmp24 ; <i1> [#uses=1]
1109 //===---------------------------------------------------------------------===//
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 //===---------------------------------------------------------------------===//
1166 GCC PR37810 is an interesting case where we should sink load/store reload
1167 into the if block and outside the loop, so we don't reload/store it on the
1188 We now hoist the reload after the call (Transforms/GVN/lpre-call-wrap.ll), but
1189 we don't sink the store. We need partially dead store sinking.
1191 //===---------------------------------------------------------------------===//
1193 [PHI TRANSLATE GEPs]
1195 GCC PR37166: Sinking of loads prevents SROA'ing the "g" struct on the stack
1196 leading to excess stack traffic. This could be handled by GVN with some crazy
1197 symbolic phi translation. The code we get looks like (g is on the stack):
1201 %9 = getelementptr %struct.f* %g, i32 0, i32 0
1202 store i32 %8, i32* %9, align bel %bb3
1204 bb3: ; preds = %bb1, %bb2, %bb
1205 %c_addr.0 = phi %struct.f* [ %g, %bb2 ], [ %c, %bb ], [ %c, %bb1 ]
1206 %b_addr.0 = phi %struct.f* [ %b, %bb2 ], [ %g, %bb ], [ %b, %bb1 ]
1207 %10 = getelementptr %struct.f* %c_addr.0, i32 0, i32 0
1208 %11 = load i32* %10, align 4
1210 %11 is fully redundant, an in BB2 it should have the value %8.
1212 GCC PR33344 is a similar case.
1214 //===---------------------------------------------------------------------===//
1216 [PHI TRANSLATE INDEXED GEPs] PR5313
1218 Load redundancy elimination for simple loop. This loop:
1220 void append_text(const char* text,unsigned char * const io) {
1225 Compiles to have a fully redundant load in the loop (%2):
1227 define void @append_text(i8* nocapture %text, i8* nocapture %io) nounwind {
1229 %0 = load i8* %text, align 1 ; <i8> [#uses=1]
1230 %1 = icmp eq i8 %0, 0 ; <i1> [#uses=1]
1231 br i1 %1, label %return, label %bb
1233 bb: ; preds = %bb, %entry
1234 %indvar = phi i32 [ 0, %entry ], [ %tmp, %bb ] ; <i32> [#uses=2]
1235 %text_addr.04 = getelementptr i8* %text, i32 %indvar ; <i8*> [#uses=1]
1236 %2 = load i8* %text_addr.04, align 1 ; <i8> [#uses=1]
1237 store i8 %2, i8* %io, align 1
1238 %tmp = add i32 %indvar, 1 ; <i32> [#uses=2]
1239 %scevgep = getelementptr i8* %text, i32 %tmp ; <i8*> [#uses=1]
1240 %3 = load i8* %scevgep, align 1 ; <i8> [#uses=1]
1241 %4 = icmp eq i8 %3, 0 ; <i1> [#uses=1]
1242 br i1 %4, label %return, label %bb
1244 return: ; preds = %bb, %entry
1248 //===---------------------------------------------------------------------===//
1250 There are many load PRE testcases in testsuite/gcc.dg/tree-ssa/loadpre* in the
1251 GCC testsuite. There are many pre testcases as ssa-pre-*.c
1253 //===---------------------------------------------------------------------===//
1255 There are some interesting cases in testsuite/gcc.dg/tree-ssa/pred-comm* in the
1256 GCC testsuite. For example, predcom-1.c is:
1258 for (i = 2; i < 1000; i++)
1259 fib[i] = (fib[i-1] + fib[i - 2]) & 0xffff;
1261 which compiles into:
1263 bb1: ; preds = %bb1, %bb1.thread
1264 %indvar = phi i32 [ 0, %bb1.thread ], [ %0, %bb1 ]
1265 %i.0.reg2mem.0 = add i32 %indvar, 2
1266 %0 = add i32 %indvar, 1 ; <i32> [#uses=3]
1267 %1 = getelementptr [1000 x i32]* @fib, i32 0, i32 %0
1268 %2 = load i32* %1, align 4 ; <i32> [#uses=1]
1269 %3 = getelementptr [1000 x i32]* @fib, i32 0, i32 %indvar
1270 %4 = load i32* %3, align 4 ; <i32> [#uses=1]
1271 %5 = add i32 %4, %2 ; <i32> [#uses=1]
1272 %6 = and i32 %5, 65535 ; <i32> [#uses=1]
1273 %7 = getelementptr [1000 x i32]* @fib, i32 0, i32 %i.0.reg2mem.0
1274 store i32 %6, i32* %7, align 4
1275 %exitcond = icmp eq i32 %0, 998 ; <i1> [#uses=1]
1276 br i1 %exitcond, label %return, label %bb1
1283 instead of handling this as a loop or other xform, all we'd need to do is teach
1284 load PRE to phi translate the %0 add (i+1) into the predecessor as (i'+1+1) =
1285 (i'+2) (where i' is the previous iteration of i). This would find the store
1288 predcom-2.c is apparently the same as predcom-1.c
1289 predcom-3.c is very similar but needs loads feeding each other instead of
1291 predcom-4.c seems the same as the rest.
1294 //===---------------------------------------------------------------------===//
1296 Other simple load PRE cases:
1297 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35287 [LPRE crit edge splitting]
1299 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34677 (licm does this, LPRE crit edge)
1300 llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as | opt -mem2reg -simplifycfg -gvn | llvm-dis
1302 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=16799 [BITCAST PHI TRANS]
1304 //===---------------------------------------------------------------------===//
1306 Type based alias analysis:
1307 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14705
1309 //===---------------------------------------------------------------------===//
1311 A/B get pinned to the stack because we turn an if/then into a select instead
1312 of PRE'ing the load/store. This may be fixable in instcombine:
1313 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=37892
1315 struct X { int i; };
1329 //===---------------------------------------------------------------------===//
1331 Interesting missed case because of control flow flattening (should be 2 loads):
1332 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26629
1333 With: llvm-gcc t2.c -S -o - -O0 -emit-llvm | llvm-as |
1334 opt -mem2reg -gvn -instcombine | llvm-dis
1335 we miss it because we need 1) GEP PHI TRAN, 2) CRIT EDGE 3) MULTIPLE DIFFERENT
1336 VALS PRODUCED BY ONE BLOCK OVER DIFFERENT PATHS
1338 //===---------------------------------------------------------------------===//
1340 http://gcc.gnu.org/bugzilla/show_bug.cgi?id=19633
1341 We could eliminate the branch condition here, loading from null is undefined:
1343 struct S { int w, x, y, z; };
1344 struct T { int r; struct S s; };
1345 void bar (struct S, int);
1346 void foo (int a, struct T b)
1354 //===---------------------------------------------------------------------===//
1356 simplifylibcalls should do several optimizations for strspn/strcspn:
1358 strcspn(x, "") -> strlen(x)
1361 strspn(x, "") -> strlen(x)
1362 strspn(x, "a") -> strchr(x, 'a')-x
1364 strcspn(x, "a") -> inlined loop for up to 3 letters (similarly for strspn):
1366 size_t __strcspn_c3 (__const char *__s, int __reject1, int __reject2,
1368 register size_t __result = 0;
1369 while (__s[__result] != '\0' && __s[__result] != __reject1 &&
1370 __s[__result] != __reject2 && __s[__result] != __reject3)
1375 This should turn into a switch on the character. See PR3253 for some notes on
1378 456.hmmer apparently uses strcspn and strspn a lot. 471.omnetpp uses strspn.
1380 //===---------------------------------------------------------------------===//
1382 "gas" uses this idiom:
1383 else if (strchr ("+-/*%|&^:[]()~", *intel_parser.op_string))
1385 else if (strchr ("<>", *intel_parser.op_string)
1387 Those should be turned into a switch.
1389 //===---------------------------------------------------------------------===//
1391 252.eon contains this interesting code:
1393 %3072 = getelementptr [100 x i8]* %tempString, i32 0, i32 0
1394 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1395 %strlen = call i32 @strlen(i8* %3072) ; uses = 1
1396 %endptr = getelementptr [100 x i8]* %tempString, i32 0, i32 %strlen
1397 call void @llvm.memcpy.i32(i8* %endptr,
1398 i8* getelementptr ([5 x i8]* @"\01LC42", i32 0, i32 0), i32 5, i32 1)
1399 %3074 = call i32 @strlen(i8* %endptr) nounwind readonly
1401 This is interesting for a couple reasons. First, in this:
1403 %3073 = call i8* @strcpy(i8* %3072, i8* %3071) nounwind
1404 %strlen = call i32 @strlen(i8* %3072)
1406 The strlen could be replaced with: %strlen = sub %3072, %3073, because the
1407 strcpy call returns a pointer to the end of the string. Based on that, the
1408 endptr GEP just becomes equal to 3073, which eliminates a strlen call and GEP.
1410 Second, the memcpy+strlen strlen can be replaced with:
1412 %3074 = call i32 @strlen([5 x i8]* @"\01LC42") nounwind readonly
1414 Because the destination was just copied into the specified memory buffer. This,
1415 in turn, can be constant folded to "4".
1417 In other code, it contains:
1419 %endptr6978 = bitcast i8* %endptr69 to i32*
1420 store i32 7107374, i32* %endptr6978, align 1
1421 %3167 = call i32 @strlen(i8* %endptr69) nounwind readonly
1423 Which could also be constant folded. Whatever is producing this should probably
1424 be fixed to leave this as a memcpy from a string.
1426 Further, eon also has an interesting partially redundant strlen call:
1428 bb8: ; preds = %_ZN18eonImageCalculatorC1Ev.exit
1429 %682 = getelementptr i8** %argv, i32 6 ; <i8**> [#uses=2]
1430 %683 = load i8** %682, align 4 ; <i8*> [#uses=4]
1431 %684 = load i8* %683, align 1 ; <i8> [#uses=1]
1432 %685 = icmp eq i8 %684, 0 ; <i1> [#uses=1]
1433 br i1 %685, label %bb10, label %bb9
1436 %686 = call i32 @strlen(i8* %683) nounwind readonly
1437 %687 = icmp ugt i32 %686, 254 ; <i1> [#uses=1]
1438 br i1 %687, label %bb10, label %bb11
1440 bb10: ; preds = %bb9, %bb8
1441 %688 = call i32 @strlen(i8* %683) nounwind readonly
1443 This could be eliminated by doing the strlen once in bb8, saving code size and
1444 improving perf on the bb8->9->10 path.
1446 //===---------------------------------------------------------------------===//
1448 I see an interesting fully redundant call to strlen left in 186.crafty:InputMove
1450 %movetext11 = getelementptr [128 x i8]* %movetext, i32 0, i32 0
1453 bb62: ; preds = %bb55, %bb53
1454 %promote.0 = phi i32 [ %169, %bb55 ], [ 0, %bb53 ]
1455 %171 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1456 %172 = add i32 %171, -1 ; <i32> [#uses=1]
1457 %173 = getelementptr [128 x i8]* %movetext, i32 0, i32 %172
1460 br i1 %or.cond, label %bb65, label %bb72
1462 bb65: ; preds = %bb62
1463 store i8 0, i8* %173, align 1
1466 bb72: ; preds = %bb65, %bb62
1467 %trank.1 = phi i32 [ %176, %bb65 ], [ -1, %bb62 ]
1468 %177 = call i32 @strlen(i8* %movetext11) nounwind readonly align 1
1470 Note that on the bb62->bb72 path, that the %177 strlen call is partially
1471 redundant with the %171 call. At worst, we could shove the %177 strlen call
1472 up into the bb65 block moving it out of the bb62->bb72 path. However, note
1473 that bb65 stores to the string, zeroing out the last byte. This means that on
1474 that path the value of %177 is actually just %171-1. A sub is cheaper than a
1477 This pattern repeats several times, basically doing:
1482 where it is "obvious" that B = A-1.
1484 //===---------------------------------------------------------------------===//
1486 186.crafty contains this interesting pattern:
1488 %77 = call i8* @strstr(i8* getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0),
1490 %phitmp648 = icmp eq i8* %77, getelementptr ([6 x i8]* @"\01LC5", i32 0, i32 0)
1491 br i1 %phitmp648, label %bb70, label %bb76
1493 bb70: ; preds = %OptionMatch.exit91, %bb69
1494 %78 = call i32 @strlen(i8* %30) nounwind readonly align 1 ; <i32> [#uses=1]
1498 if (strstr(cststr, P) == cststr) {
1502 The strstr call would be significantly cheaper written as:
1505 if (memcmp(P, str, strlen(P)))
1508 This is memcmp+strlen instead of strstr. This also makes the strlen fully
1511 //===---------------------------------------------------------------------===//
1513 186.crafty also contains this code:
1515 %1906 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1516 %1907 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1906
1517 %1908 = call i8* @strcpy(i8* %1907, i8* %1905) nounwind align 1
1518 %1909 = call i32 @strlen(i8* getelementptr ([32 x i8]* @pgn_event, i32 0,i32 0))
1519 %1910 = getelementptr [32 x i8]* @pgn_event, i32 0, i32 %1909
1521 The last strlen is computable as 1908-@pgn_event, which means 1910=1908.
1523 //===---------------------------------------------------------------------===//
1525 186.crafty has this interesting pattern with the "out.4543" variable:
1527 call void @llvm.memcpy.i32(
1528 i8* getelementptr ([10 x i8]* @out.4543, i32 0, i32 0),
1529 i8* getelementptr ([7 x i8]* @"\01LC28700", i32 0, i32 0), i32 7, i32 1)
1530 %101 = call@printf(i8* ... @out.4543, i32 0, i32 0)) nounwind
1532 It is basically doing:
1534 memcpy(globalarray, "string");
1535 printf(..., globalarray);
1537 Anyway, by knowing that printf just reads the memory and forward substituting
1538 the string directly into the printf, this eliminates reads from globalarray.
1539 Since this pattern occurs frequently in crafty (due to the "DisplayTime" and
1540 other similar functions) there are many stores to "out". Once all the printfs
1541 stop using "out", all that is left is the memcpy's into it. This should allow
1542 globalopt to remove the "stored only" global.
1544 //===---------------------------------------------------------------------===//
1548 define inreg i32 @foo(i8* inreg %p) nounwind {
1550 %tmp1 = ashr i8 %tmp0, 5
1551 %tmp2 = sext i8 %tmp1 to i32
1555 could be dagcombine'd to a sign-extending load with a shift.
1556 For example, on x86 this currently gets this:
1562 while it could get this:
1567 //===---------------------------------------------------------------------===//
1571 int test(int x) { return 1-x == x; } // --> return false
1572 int test2(int x) { return 2-x == x; } // --> return x == 1 ?
1574 Always foldable for odd constants, what is the rule for even?
1576 //===---------------------------------------------------------------------===//
1578 PR 3381: GEP to field of size 0 inside a struct could be turned into GEP
1579 for next field in struct (which is at same address).
1581 For example: store of float into { {{}}, float } could be turned into a store to
1584 //===---------------------------------------------------------------------===//
1587 double foo(double a) { return sin(a); }
1589 This compiles into this on x86-64 Linux:
1600 //===---------------------------------------------------------------------===//
1602 The arg promotion pass should make use of nocapture to make its alias analysis
1603 stuff much more precise.
1605 //===---------------------------------------------------------------------===//
1607 The following functions should be optimized to use a select instead of a
1608 branch (from gcc PR40072):
1610 char char_int(int m) {if(m>7) return 0; return m;}
1611 int int_char(char m) {if(m>7) return 0; return m;}
1613 //===---------------------------------------------------------------------===//
1615 IPSCCP is propagating elements of first class aggregates, but is not propagating
1616 the entire aggregate itself. This leads it to miss opportunities, for example
1617 in test/Transforms/SCCP/ipsccp-basic.ll:test5b.
1619 //===---------------------------------------------------------------------===//
1621 int func(int a, int b) { if (a & 0x80) b |= 0x80; else b &= ~0x80; return b; }
1625 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1627 %0 = and i32 %a, 128 ; <i32> [#uses=1]
1628 %1 = icmp eq i32 %0, 0 ; <i1> [#uses=1]
1629 %2 = or i32 %b, 128 ; <i32> [#uses=1]
1630 %3 = and i32 %b, -129 ; <i32> [#uses=1]
1631 %b_addr.0 = select i1 %1, i32 %3, i32 %2 ; <i32> [#uses=1]
1635 However, it's functionally equivalent to:
1637 b = (b & ~0x80) | (a & 0x80);
1639 Which generates this:
1641 define i32 @func(i32 %a, i32 %b) nounwind readnone ssp {
1643 %0 = and i32 %b, -129 ; <i32> [#uses=1]
1644 %1 = and i32 %a, 128 ; <i32> [#uses=1]
1645 %2 = or i32 %0, %1 ; <i32> [#uses=1]
1649 This can be generalized for other forms:
1651 b = (b & ~0x80) | (a & 0x40) << 1;
1653 //===---------------------------------------------------------------------===//
1655 These two functions produce different code. They shouldn't:
1659 uint8_t p1(uint8_t b, uint8_t a) {
1660 b = (b & ~0xc0) | (a & 0xc0);
1664 uint8_t p2(uint8_t b, uint8_t a) {
1665 b = (b & ~0x40) | (a & 0x40);
1666 b = (b & ~0x80) | (a & 0x80);
1670 define zeroext i8 @p1(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1672 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1673 %1 = and i8 %a, -64 ; <i8> [#uses=1]
1674 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1678 define zeroext i8 @p2(i8 zeroext %b, i8 zeroext %a) nounwind readnone ssp {
1680 %0 = and i8 %b, 63 ; <i8> [#uses=1]
1681 %.masked = and i8 %a, 64 ; <i8> [#uses=1]
1682 %1 = and i8 %a, -128 ; <i8> [#uses=1]
1683 %2 = or i8 %1, %0 ; <i8> [#uses=1]
1684 %3 = or i8 %2, %.masked ; <i8> [#uses=1]
1688 //===---------------------------------------------------------------------===//