1. Add new instructions for checking parity flags: JP, JNP, SETP, SETNP.
2. Set the isCommutable and isPromotableTo3Address bits on several
   instructions.

llvm-svn: 19246

llvm-svn: 19241

While we're at it, improve codegen of select instructions.  For this
testcase:

int %test(bool %C, int %A, int %B) {
  %D = select bool %C, int %A, int %B
  ret int %D
}

We used to generate this code:

_test:
        cmpwi cr0, r3, 0
        bne .LBB_test_2 ;
.LBB_test_1:    ;
        b .LBB_test_3   ;
.LBB_test_2:    ;
        or r5, r4, r4
.LBB_test_3:    ;
        or r3, r5, r5
        blr

Now we emit:

_test:
        cmpwi cr0, r3, 0
        bne .LBB_test_2 ;
.LBB_test_1:    ;
        or r4, r5, r5
.LBB_test_2:    ;
        or r3, r4, r4
        blr

-Chris

llvm-svn: 19214

In particular, instead of compiling this:

bool %test(int %A, int %B) {
  %C = setlt int %A, %B
  ret bool %C
}

to this:

test:
        save %sp, -96, %sp
        subcc %i0, %i1, %g0
        bl .LBBtest_1   !
        nop
        ba .LBBtest_2   !
        nop
.LBBtest_1:     !
        or %g0, 1, %i0
        ba .LBBtest_3   !
        nop
.LBBtest_2:     !
        or %g0, 0, %i0
        ba .LBBtest_3   !
        nop
.LBBtest_3:     !
        restore %g0, %g0, %g0
        retl
        nop

We now compile it to this:

test:
        save %sp, -96, %sp
        subcc %i0, %i1, %g0
        or %g0, 1, %i0
        bl .LBBtest_2   !
        nop
.LBBtest_1:     !
        or %g0, %g0, %i0
.LBBtest_2:     !
        restore %g0, %g0, %g0
        retl
        nop

llvm-svn: 19213

Fix testcase CodeGen/CBackend/2004-12-28-LogicalConstantExprs.ll

llvm-svn: 19176

llvm-svn: 19024

llvm-svn: 19021

llvm-svn: 19007

save small amounts of time for functions that don't call llvm.returnaddress
or llvm.frameaddress (which is almost all functions).

llvm-svn: 19006

llvm-svn: 19005

llvm-svn: 19004

llvm-svn: 19003

llvm-svn: 18994

llvm-svn: 18987

llvm-svn: 18985

of custom rules.

llvm-svn: 18984

llvm-svn: 18983

llvm-svn: 18982

llvm-svn: 18974

llvm-svn: 18932

llvm-svn: 18931

to 64-bit precision, not 80 bits.

llvm-svn: 18915

llvm-svn: 18911

don't support long double anyway, and this gives us FP results closer to
other targets.

This also speeds up 179.art from 41.4s to 18.32s, by eliminating a problem
with extra precision that causes an FP == comparison to fail (leading to
extra loop iterations).

llvm-svn: 18895

llvm-svn: 18844

when compiled with debug information.

llvm-svn: 18835

llvm-svn: 18833

llvm-svn: 18830

llvm-svn: 18826

llvm-svn: 18824

Also, fix a bug where ubyte 255 would sometimes be output as -1. This
was afflicting hbd.

llvm-svn: 18823

llvm-svn: 18820

llvm-svn: 18818

llvm-svn: 18805

llvm-svn: 18803

ctor parameters can be defaulted.

Print the transformed llvm code input to the instruction selector
when -print-machineinstrs is on, just like V9.

llvm-svn: 18794

*or g0, x      add g0, x          recognized * as a move)
 or  x, g0     add  x, g0
 or  0, x      add  0, x
 or  x, 0      add  x, 0

llvm-svn: 18793

* Don't emit the Index * ElementSize multiply if Index is a constant.
* Use a shift, not a multiply, if ElementSize is 1/2/4/8.
* If ElementSize fits in the immediate field of SMUL, then put it there.

Fix a bug where struct offsets might be truncated (ConstantSInt::get is
now used instead of ConstantInt::get).

llvm-svn: 18792

ones are failing in cbe.

llvm-svn: 18791

slipping through into the instruction selector, which can't deal with
them yet.

llvm-svn: 18758