This requires reassociating to forms of expressions that are already available,
something that reassoc doesn't think about yet.
+
+//===---------------------------------------------------------------------===//
+
+This function: (derived from GCC PR19988)
+double foo(double x, double y) {
+ return ((x + 0.1234 * y) * (x + -0.1234 * y));
+}
+
+compiles to:
+_foo:
+ movapd %xmm1, %xmm2
+ mulsd LCPI1_1(%rip), %xmm1
+ mulsd LCPI1_0(%rip), %xmm2
+ addsd %xmm0, %xmm1
+ addsd %xmm0, %xmm2
+ movapd %xmm1, %xmm0
+ mulsd %xmm2, %xmm0
+ ret
+
+Reassociate should be able to turn it into:
+
+double foo(double x, double y) {
+ return ((x + 0.1234 * y) * (x - 0.1234 * y));
+}
+
+Which allows the multiply by constant to be CSE'd, producing:
+
+_foo:
+ mulsd LCPI1_0(%rip), %xmm1
+ movapd %xmm1, %xmm2
+ addsd %xmm0, %xmm2
+ subsd %xmm1, %xmm0
+ mulsd %xmm2, %xmm0
+ ret
+
+This doesn't need -ffast-math support at all. This is particularly bad because
+the llvm-gcc frontend is canonicalizing the later into the former, but clang
+doesn't have this problem.
+
//===---------------------------------------------------------------------===//
These two functions should generate the same code on big-endian systems:
//===---------------------------------------------------------------------===//
-dag/inst combine "clz(x)>>5 -> x==0" for 32-bit x.
-
-Compile:
-
-int bar(int x)
-{
- int t = __builtin_clz(x);
- return -(t>>5);
-}
-
-to:
-
-_bar: addic r3,r3,-1
- subfe r3,r3,r3
- blr
-
-//===---------------------------------------------------------------------===//
-
quantum_sigma_x in 462.libquantum contains the following loop:
for(i=0; i<reg->size; i++)
//===---------------------------------------------------------------------===//
-This should be optimized to one 'and' and one 'or', from PR4216:
-
-define i32 @test_bitfield(i32 %bf.prev.low) nounwind ssp {
-entry:
- %bf.prev.lo.cleared10 = or i32 %bf.prev.low, 32962 ; <i32> [#uses=1]
- %0 = and i32 %bf.prev.low, -65536 ; <i32> [#uses=1]
- %1 = and i32 %bf.prev.lo.cleared10, 40186 ; <i32> [#uses=1]
- %2 = or i32 %1, %0 ; <i32> [#uses=1]
- ret i32 %2
-}
-
-//===---------------------------------------------------------------------===//
-
This isn't recognized as bswap by instcombine (yes, it really is bswap):
unsigned long reverse(unsigned v) {
//===---------------------------------------------------------------------===//
+[LOOP RECOGNITION]
+
These idioms should be recognized as popcount (see PR1488):
unsigned countbits_slow(unsigned v) {
//===---------------------------------------------------------------------===//
+[LOOP RECOGNITION]
+
viterbi speeds up *significantly* if the various "history" related copy loops
are turned into memcpy calls at the source level. We need a "loops to memcpy"
pass.
//===---------------------------------------------------------------------===//
+[LOOP OPTIMIZATION]
+
+SingleSource/Benchmarks/Misc/dt.c shows several interesting optimization
+opportunities in its double_array_divs_variable function: it needs loop
+interchange, memory promotion (which LICM already does), vectorization and
+variable trip count loop unrolling (since it has a constant trip count). ICC
+apparently produces this very nice code with -ffast-math:
+
+..B1.70: # Preds ..B1.70 ..B1.69
+ mulpd %xmm0, %xmm1 #108.2
+ mulpd %xmm0, %xmm1 #108.2
+ mulpd %xmm0, %xmm1 #108.2
+ mulpd %xmm0, %xmm1 #108.2
+ addl $8, %edx #
+ cmpl $131072, %edx #108.2
+ jb ..B1.70 # Prob 99% #108.2
+
+It would be better to count down to zero, but this is a lot better than what we
+do.
+
+//===---------------------------------------------------------------------===//
+
Consider:
typedef unsigned U32;
//===---------------------------------------------------------------------===//
-We generate a horrible libcall for llvm.powi. For example, we compile:
-
-#include <cmath>
-double f(double a) { return std::pow(a, 4); }
-
-into:
-
-__Z1fd:
- subl $12, %esp
- movsd 16(%esp), %xmm0
- movsd %xmm0, (%esp)
- movl $4, 8(%esp)
- call L___powidf2$stub
- addl $12, %esp
- ret
-
-GCC produces:
-
-__Z1fd:
- subl $12, %esp
- movsd 16(%esp), %xmm0
- mulsd %xmm0, %xmm0
- mulsd %xmm0, %xmm0
- movsd %xmm0, (%esp)
- fldl (%esp)
- addl $12, %esp
- ret
-
-//===---------------------------------------------------------------------===//
-
We compile this program: (from GCC PR11680)
http://gcc.gnu.org/bugzilla/attachment.cgi?id=4487
//===---------------------------------------------------------------------===//
+[ALIAS ANALYSIS]
+
Type based alias analysis:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14705
+We should do better analysis of posix_memalign. At the least it should
+no-capture its pointer argument, at best, we should know that the out-value
+result doesn't point to anything (like malloc). One example of this is in
+SingleSource/Benchmarks/Misc/dt.c
+
//===---------------------------------------------------------------------===//
A/B get pinned to the stack because we turn an if/then into a select instead
//===---------------------------------------------------------------------===//
-FunctionAttrs is not marking this function as readnone (just readonly):
-$ clang t.c -emit-llvm -S -o - -O0 | opt -mem2reg -S -functionattrs
-
-int t(int a, int b, int c) {
- int *p;
- if (a)
- p = &a;
- else
- p = &c;
- return *p;
-}
-
-This is because we codegen this to:
-
-define i32 @t(i32 %a, i32 %b, i32 %c) nounwind readonly ssp {
-entry:
- %a.addr = alloca i32 ; <i32*> [#uses=3]
- %c.addr = alloca i32 ; <i32*> [#uses=2]
-...
-
-if.end:
- %p.0 = phi i32* [ %a.addr, %if.then ], [ %c.addr, %if.else ]
- %tmp2 = load i32* %p.0 ; <i32> [#uses=1]
- ret i32 %tmp2
-}
-
-And functionattrs doesn't realize that the p.0 load points to function local
-memory.
-
-Also, functionattrs doesn't know about memcpy/memset. This function should be
-marked readnone, since it only twiddles local memory, but functionattrs doesn't
-handle memset/memcpy/memmove aggressively:
+functionattrs doesn't know much about memcpy/memset. This function should be
+marked readnone rather than readonly, since it only twiddles local memory, but
+functionattrs doesn't handle memset/memcpy/memmove aggressively:
struct X { int *p; int *q; };
int foo() {
//===---------------------------------------------------------------------===//
+Missed instcombine transformation:
+define i1 @a(i32 %x) nounwind readnone {
+entry:
+ %cmp = icmp eq i32 %x, 30
+ %sub = add i32 %x, -30
+ %cmp2 = icmp ugt i32 %sub, 9
+ %or = or i1 %cmp, %cmp2
+ ret i1 %or
+}
+This should be optimized to a single compare. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine transformation:
+void b();
+void a(int x) { if (((1<<x)&8)==0) b(); }
+
+The shift should be optimized out. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine or reassociate transformation:
+int a(int a, int b) { return (a==12)&(b>47)&(b<58); }
+
+The sgt and slt should be combined into a single comparison. Testcase derived
+from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine transformation:
+define i32 @a(i32 %x) nounwind readnone {
+entry:
+ %rem = srem i32 %x, 32
+ %shl = shl i32 1, %rem
+ ret i32 %shl
+}
+
+The srem can be transformed to an and because if x is negative, the shift is
+undefined. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine/dagcombine transformation:
+define i32 @a(i32 %x, i32 %y) nounwind readnone {
+entry:
+ %mul = mul i32 %y, -8
+ %sub = sub i32 %x, %mul
+ ret i32 %sub
+}
+
+Should compile to something like x+y*8, but currently compiles to an
+inefficient result. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+Missed instcombine/dagcombine transformation:
+define void @lshift_lt(i8 zeroext %a) nounwind {
+entry:
+ %conv = zext i8 %a to i32
+ %shl = shl i32 %conv, 3
+ %cmp = icmp ult i32 %shl, 33
+ br i1 %cmp, label %if.then, label %if.end
+
+if.then:
+ tail call void @bar() nounwind
+ ret void
+
+if.end:
+ ret void
+}
+declare void @bar() nounwind
+
+The shift should be eliminated. Testcase derived from gcc.
+
+//===---------------------------------------------------------------------===//
+
+These compile into different code, one gets recognized as a switch and the
+other doesn't due to phase ordering issues (PR6212):
+
+int test1(int mainType, int subType) {
+ if (mainType == 7)
+ subType = 4;
+ else if (mainType == 9)
+ subType = 6;
+ else if (mainType == 11)
+ subType = 9;
+ return subType;
+}
+
+int test2(int mainType, int subType) {
+ if (mainType == 7)
+ subType = 4;
+ if (mainType == 9)
+ subType = 6;
+ if (mainType == 11)
+ subType = 9;
+ return subType;
+}
+
+//===---------------------------------------------------------------------===//
projects / oota-llvm.git / blobdiff