creation. These changes are still temporary but at least this pushes
knowledge of signedness out closer to where it can be determined properly
and allows signedness to be removed from VMCore.

llvm-svn: 32654

llvm-svn: 32595

used to determine whether a ZExt or SExt cast is performed. Instead, pass
an "isSigned" bool to the function and determine its value from the opcode
of the cast involved.
Also, clean up some cruft from previous patches.

llvm-svn: 32548

cast instructions. A few are bug fixes.

llvm-svn: 32544

make sure to also check that it is a zext from bool, not any other cast
operation type.

llvm-svn: 32539

The cast patch introduced the possibility that the wrong cast opcode
could be used and that this transform could trigger on different kinds
of cast operations. This patch rectifies that.

llvm-svn: 32538

so the decision of which opcode to use is pushed upward to the caller.
Adjust the callers to pass the expected opcode.

llvm-svn: 32535

obscure bugs affecting the X86 code generator.  I will reenable this
when fixed.

llvm-svn: 32524

creation calls.

llvm-svn: 32521

Rename getZeroExtend and getSignExtend to getZExt and getSExt to match
the the casting mnemonics in the rest of LLVM.

llvm-svn: 32514

llvm-svn: 32487

llvm-svn: 32486

llvm-svn: 32479

llvm-svn: 32469

llvm-svn: 32468

llvm-svn: 32426

This implements Transforms/ScalarRepl/union-packed.ll

llvm-svn: 32422

 * Add support for promoting unions with fp values in them.  This produces
   our new int<->fp bitcast instructions, implementing
   Transforms/ScalarRepl/union-fp-int.ll

As an example, this allows us to compile this:

union intfloat { int i; float f; };
float invsqrt(const float arg_x) {
    union intfloat x = { .f = arg_x };
    const float xhalf = arg_x * 0.5f;
    x.i = 0x5f3759df - (x.i >> 1);
    return x.f * (1.5f - xhalf * x.f * x.f);
}

into:

_invsqrt:
        movss 4(%esp), %xmm0
        movd %xmm0, %eax
        sarl %eax
        movl $1597463007, %ecx
        subl %eax, %ecx
        movd %ecx, %xmm1
        mulss LCPI1_0, %xmm0
        mulss %xmm1, %xmm0
        movss LCPI1_1, %xmm2
        mulss %xmm1, %xmm0
        subss %xmm0, %xmm2
        movl 8(%esp), %eax
        mulss %xmm2, %xmm1
        movss %xmm1, (%eax)
        ret

instead of:

_invsqrt:
        subl $4, %esp
        movss 8(%esp), %xmm0
        movss %xmm0, (%esp)
        movl (%esp), %eax
        movl $1597463007, %ecx
        sarl %eax
        subl %eax, %ecx
        movl %ecx, (%esp)
        mulss LCPI1_0, %xmm0
        movss (%esp), %xmm1
        mulss %xmm1, %xmm0
        mulss %xmm1, %xmm0
        movss LCPI1_1, %xmm2
        subss %xmm0, %xmm2
        mulss %xmm2, %xmm1
        movl 12(%esp), %eax
        movss %xmm1, (%eax)
        addl $4, %esp
        ret

llvm-svn: 32418

MarkAliveBlocks.

llvm-svn: 32375

llvm-svn: 32320

now cerr, cout, and NullStream resp.

llvm-svn: 32298

llvm-svn: 32288

llvm-svn: 32280

is 'unsigned'.

llvm-svn: 32279

passing false would make the asmprinter fail anyway.

llvm-svn: 32264

7.48s.  This regression is due to unforseen consequences of the cast patch.

llvm-svn: 32209

for now.

llvm-svn: 32208

llvm-svn: 32166

llvm-svn: 32113

Transforms/InstCombine/2006-12-01-BadFPVectorXform.ll

llvm-svn: 32112

llvm-svn: 32051

unsigned short swp(unsigned short a) {
       return ((a & 0xff00) >> 8 | (a & 0x00ff) << 8);
}

as an idiom for bswap.

llvm-svn: 32011

This implements InstCombine/cast.ll:test34.  It fires hundreds of times on
176.gcc.

llvm-svn: 32009

folding   seteq (bswap(x)), c -> seteq(x,bswap(c))

llvm-svn: 32006

llvm-svn: 31996

llvm-svn: 31977

producing code for "trunc to bool". This passes all tests on Linux.

llvm-svn: 31963

llvm-svn: 31941

The long awaited CAST patch. This introduces 12 new instructions into LLVM
to replace the cast instruction. Corresponding changes throughout LLVM are
provided. This passes llvm-test, llvm/test, and SPEC CPUINT2000 with the
exception of 175.vpr which fails only on a slight floating point output
difference.

llvm-svn: 31931

llvm-svn: 31923