llvm/lib/Target/README.txt
Evan Cheng d3864b5743 Remind us of exit value substitution
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@26862 91177308-0d34-0410-b5e6-96231b3b80d8
2006-03-19 06:09:23 +00:00

131 lines
3.6 KiB
Plaintext

Target Independent Opportunities:
===-------------------------------------------------------------------------===
FreeBench/mason contains code like this:
static p_type m0u(p_type p) {
int m[]={0, 8, 1, 2, 16, 5, 13, 7, 14, 9, 3, 4, 11, 12, 15, 10, 17, 6};
p_type pu;
pu.a = m[p.a];
pu.b = m[p.b];
pu.c = m[p.c];
return pu;
}
We currently compile this into a memcpy from a static array into 'm', then
a bunch of loads from m. It would be better to avoid the memcpy and just do
loads from the static array.
//===---------------------------------------------------------------------===//
Make the PPC branch selector target independant
//===---------------------------------------------------------------------===//
Get the C front-end to expand hypot(x,y) -> llvm.sqrt(x*x+y*y) when errno and
precision don't matter (ffastmath). Misc/mandel will like this. :)
//===---------------------------------------------------------------------===//
Solve this DAG isel folding deficiency:
int X, Y;
void fn1(void)
{
X = X | (Y << 3);
}
compiles to
fn1:
movl Y, %eax
shll $3, %eax
orl X, %eax
movl %eax, X
ret
The problem is the store's chain operand is not the load X but rather
a TokenFactor of the load X and load Y, which prevents the folding.
There are two ways to fix this:
1. The dag combiner can start using alias analysis to realize that y/x
don't alias, making the store to X not dependent on the load from Y.
2. The generated isel could be made smarter in the case it can't
disambiguate the pointers.
Number 1 is the preferred solution.
This has been "fixed" by a TableGen hack. But that is a short term workaround
which will be removed once the proper fix is made.
//===---------------------------------------------------------------------===//
Turn this into a signed shift right in instcombine:
int f(unsigned x) {
return x >> 31 ? -1 : 0;
}
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25600
http://gcc.gnu.org/ml/gcc-patches/2006-02/msg01492.html
//===---------------------------------------------------------------------===//
On targets with expensive 64-bit multiply, we could LSR this:
for (i = ...; ++i) {
x = 1ULL << i;
into:
long long tmp = 1;
for (i = ...; ++i, tmp+=tmp)
x = tmp;
This would be a win on ppc32, but not x86 or ppc64.
//===---------------------------------------------------------------------===//
Shrink: (setlt (loadi32 P), 0) -> (setlt (loadi8 Phi), 0)
//===---------------------------------------------------------------------===//
Reassociate should turn: X*X*X*X -> t=(X*X) (t*t) to eliminate a multiply.
//===---------------------------------------------------------------------===//
Interesting? testcase for add/shift/mul reassoc:
int bar(int x, int y) {
return x*x*x+y+x*x*x*x*x*y*y*y*y;
}
int foo(int z, int n) {
return bar(z, n) + bar(2*z, 2*n);
}
//===---------------------------------------------------------------------===//
These two functions should generate the same code on big-endian systems:
int g(int *j,int *l) { return memcmp(j,l,4); }
int h(int *j, int *l) { return *j - *l; }
this could be done in SelectionDAGISel.cpp, along with other special cases,
for 1,2,4,8 bytes.
//===---------------------------------------------------------------------===//
This code:
int rot(unsigned char b) { int a = ((b>>1) ^ (b<<7)) & 0xff; return a; }
Can be improved in two ways:
1. The instcombiner should eliminate the type conversions.
2. The X86 backend should turn this into a rotate by one bit.
//===---------------------------------------------------------------------===//
Add LSR exit value substitution. It'll probably be a win for Ackermann, etc.