GCC4 + IA64 = badnews

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22075 91177308-0d34-0410-b5e6-96231b3b80d8

Remove the check for the ffsll function. Its no longer needed by the
simplify-libcalls pass (pass now computes it without a call to ffsll).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22074 91177308-0d34-0410-b5e6-96231b3b80d8

Improve statement

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22073 91177308-0d34-0410-b5e6-96231b3b80d8

Don't eliminate a test case, just the form of the test case, from the
previous revision.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22072 91177308-0d34-0410-b5e6-96231b3b80d8

This does not belong

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22071 91177308-0d34-0410-b5e6-96231b3b80d8

Don't look for __builtin_ffs, we'll never see it from llvm-gcc and there's
not reason to include it for other front ends.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22070 91177308-0d34-0410-b5e6-96231b3b80d8

Add a case with non-constant argument for testing the transform:
ffs(x) -> (x == 0 ? 0 : llvm.cttz(0)+1)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22069 91177308-0d34-0410-b5e6-96231b3b80d8

Provide this optimization as well:
ffs(x) -> (x == 0 ? 0 : 1+llvm.cttz(x))

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22068 91177308-0d34-0410-b5e6-96231b3b80d8

PPC "branch and link" instructions are branches in the PPC sense, but not
in the LLVM code generator sense (they are calls).  Don't mark them as such,
which fixes the regressions on the ppc tester last night

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22065 91177308-0d34-0410-b5e6-96231b3b80d8

Fix andrews changes to fit in 80 columns

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22064 91177308-0d34-0410-b5e6-96231b3b80d8

ctlz = most signficant bits, cttz = least sig

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22061 91177308-0d34-0410-b5e6-96231b3b80d8

fix warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22060 91177308-0d34-0410-b5e6-96231b3b80d8

fix compiler warning

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22059 91177308-0d34-0410-b5e6-96231b3b80d8

Duh .. you actually have to #include Config/config.h before you can test
for one of the values that it defines!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22058 91177308-0d34-0410-b5e6-96231b3b80d8

Some cleanups for compilation with GCC 4.0.0 to remove warnings:
* Use C++ style casts, not C style casts
* Abstract base classes should have virtual destructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22057 91177308-0d34-0410-b5e6-96231b3b80d8

add a sparcv8 section, make this validate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22056 91177308-0d34-0410-b5e6-96231b3b80d8

more edits, add anchors

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22055 91177308-0d34-0410-b5e6-96231b3b80d8

make angry compilers happy again

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22054 91177308-0d34-0410-b5e6-96231b3b80d8

aCC and STLport complained about this, because they're like that

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22053 91177308-0d34-0410-b5e6-96231b3b80d8

edits to the simplify-libcalls element, move it higher in the list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22052 91177308-0d34-0410-b5e6-96231b3b80d8

add generated code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22051 91177308-0d34-0410-b5e6-96231b3b80d8

don't reserve space for tailcall arg areas.  It explicitly managed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22050 91177308-0d34-0410-b5e6-96231b3b80d8

Make note of the -simplify-libcalls pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22049 91177308-0d34-0410-b5e6-96231b3b80d8

add test to check for tail calls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22048 91177308-0d34-0410-b5e6-96231b3b80d8

Teach reginfo how to deal with ADJSTACKPTRri, allowing us to generate:
        add %ESP, 20
        jmp %EDX  # TAIL CALL

instead of:
        add %ESP, -8
        add %ESP, 28
        jmp %EDX  # TAIL CALL

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22047 91177308-0d34-0410-b5e6-96231b3b80d8

Implement proper tail calls in the X86 backend for all fastcc->fastcc
tail calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22046 91177308-0d34-0410-b5e6-96231b3b80d8

Substantial edits for the release notes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22045 91177308-0d34-0410-b5e6-96231b3b80d8

Add some simplifications for MULH[SU].  This allows us to compile this:

long %bar(long %X) {
  %Y = mul long %X, 4294967297
  ret long %Y
}

to this:

l1_bar:
        mov %EAX, DWORD PTR [%ESP + 4]
        mov %EDX, %EAX
        add %EDX, DWORD PTR [%ESP + 8]
        ret

instead of:

l1_bar:
        mov %ECX, DWORD PTR [%ESP + 4]
        mov %EDX, 1
        mov %EAX, %ECX
        mul %EDX
        add %EDX, %ECX
        add %EDX, DWORD PTR [%ESP + 8]
        mov %EAX, %ECX
        ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22044 91177308-0d34-0410-b5e6-96231b3b80d8

Add markers in the asm file for tail calls, add a new ADJSTACKPTRri
sorta-pseudo-instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22042 91177308-0d34-0410-b5e6-96231b3b80d8

When inserting callee-save register reloads, make sure to skip over any
terminator instructions before the 'ret' in case the target has a
multi-instruction return sequence.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22041 91177308-0d34-0410-b5e6-96231b3b80d8

Yes, calltarget is the operand of the day.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22040 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure this is not eligible for tail-call-elimination so that we test
the correct thing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22039 91177308-0d34-0410-b5e6-96231b3b80d8

When emitting the function epilog, check to see if there already a stack
adjustment.  If so, we merge the adjustment into the existing one.  This
allows us to generate:

caller2:
        sub %ESP, 12
        mov DWORD PTR [%ESP], 0
        mov %EAX, 1234567890
        mov %EDX, 0
        call func2
        add %ESP, 8
        ret 4

intead of:

caller2:
        sub %ESP, 12
        mov DWORD PTR [%ESP], 0
        mov %EAX, 1234567890
        mov %EDX, 0
        call func2
        sub %ESP, 4
        add %ESP, 12
        ret 4

for X86/fast-cc-merge-stack-adj.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22038 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22037 91177308-0d34-0410-b5e6-96231b3b80d8

Add some new instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22036 91177308-0d34-0410-b5e6-96231b3b80d8

add x86-64 abi

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22035 91177308-0d34-0410-b5e6-96231b3b80d8

* Truncate a long line.
* By default, doxygen output formatted for printing on letter size paper
  not A4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22034 91177308-0d34-0410-b5e6-96231b3b80d8

Forgot to commit this earlier. Its part of the simplify-libcalls
enhancement for ffs, ffsl, and ffsll optimizations. We can't do the opt
unless we also have the at least ffsll function. Notably SVR4 doesn't.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22033 91177308-0d34-0410-b5e6-96231b3b80d8

mention the C++ exception problems on alpha

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22029 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test case for testing ffs libcall conversion to constant

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22028 91177308-0d34-0410-b5e6-96231b3b80d8

Changes for ffs lib call simplification:
* Check for availability of ffsll call in configure script
* Support ffs, ffsl, and ffsll conversion to constant value if the argument
  is constant.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22027 91177308-0d34-0410-b5e6-96231b3b80d8

Fix construction of ioport intrinsics, fixing X86/io.llx and io-port.llx

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22026 91177308-0d34-0410-b5e6-96231b3b80d8

Preserve calling conv when hacking on calls

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22025 91177308-0d34-0410-b5e6-96231b3b80d8

preserve calling conventions when hacking on code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22024 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure to preserve the calling convention when changing an invoke into
a call.  This fixes Prolangs-C++/deriv2, kimwitu++, and Misc-C++/bigfib
on X86 with -enable-x86-fastcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22023 91177308-0d34-0410-b5e6-96231b3b80d8

Pass i64 values correctly split in reg/mem to fastcc calls.

This fixes fourinarow with -enable-x86-fastcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22022 91177308-0d34-0410-b5e6-96231b3b80d8

Use target-specific nodes for calls.  This allows the fastcc code to not have
to do ugly hackery to avoid emitting code like this:

   call foo
   mov vreg, EAX
   adjcallstackup ...

If foo is a fastcc call and if vreg gets spilled, we might end up with this:

   call foo
   mov [ESP+offset], EAX     ;; Offset doesn't consider the 12!
   sub ESP, 12

Which is bad.  The previous hacky code to deal with this was A) gross B) not
good enough.  In particular, it could miss cases and emit the bad code above.
Now we always emit this:

   call foo
   adjcallstackup ...
   mov vreg, EAX

directly.

This makes fastcc with callees poping the stack work much better.  Next
stop (finally!) really is tail calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22021 91177308-0d34-0410-b5e6-96231b3b80d8

allow token chain at start or end of node

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22020 91177308-0d34-0410-b5e6-96231b3b80d8

remove special case hacks for readport/readio from the binary operator
codepath

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22019 91177308-0d34-0410-b5e6-96231b3b80d8

Implement fixme's by memoizing nodes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22018 91177308-0d34-0410-b5e6-96231b3b80d8

add a new map

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22017 91177308-0d34-0410-b5e6-96231b3b80d8

Turn this into a wrapper for a simpler version of getNode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22016 91177308-0d34-0410-b5e6-96231b3b80d8

Eliminate special purpose hacks for dynamic_stack_alloc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22015 91177308-0d34-0410-b5e6-96231b3b80d8

Use the general mechanism for creating multi-value nodes instead of using
special case hacks.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22014 91177308-0d34-0410-b5e6-96231b3b80d8

Ensure these casts are done with the hardware fildll instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22013 91177308-0d34-0410-b5e6-96231b3b80d8

use a target-specific node and custom expander to lower long->FP to FILD64m.

This should fix some missing symbols problems on BSD and improve performance
of programs that use that operation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22012 91177308-0d34-0410-b5e6-96231b3b80d8

Wrap long line, actually add node to the graph.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22011 91177308-0d34-0410-b5e6-96231b3b80d8

legalize target-specific operations

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22010 91177308-0d34-0410-b5e6-96231b3b80d8

add a getNode() version that allows construction of any node type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22009 91177308-0d34-0410-b5e6-96231b3b80d8

Add a way to construct an arbitrary node, cleanly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22008 91177308-0d34-0410-b5e6-96231b3b80d8

add accessors

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22007 91177308-0d34-0410-b5e6-96231b3b80d8

Some minor adjustments for the 1.5 release.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22006 91177308-0d34-0410-b5e6-96231b3b80d8

Pass the dag into LowerOperation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22005 91177308-0d34-0410-b5e6-96231b3b80d8

LowerOperation takes a dag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22004 91177308-0d34-0410-b5e6-96231b3b80d8

add a bit about the known problems on IA64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22003 91177308-0d34-0410-b5e6-96231b3b80d8

Print the symbolic register name in a register allocator debug dump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22002 91177308-0d34-0410-b5e6-96231b3b80d8

Allow targets to have a custom int64->fp expander if desired

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22001 91177308-0d34-0410-b5e6-96231b3b80d8

this passes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@22000 91177308-0d34-0410-b5e6-96231b3b80d8

Fix this

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21999 91177308-0d34-0410-b5e6-96231b3b80d8

minor changes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21998 91177308-0d34-0410-b5e6-96231b3b80d8

user ops are not encoded in the bc format

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21997 91177308-0d34-0410-b5e6-96231b3b80d8

Document the pseudo-instruction opcodes in opcode range 56-63. These are
used to support things like volatile load/store, tail calls, and calling
conventions without reserving space for the additional information.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21996 91177308-0d34-0410-b5e6-96231b3b80d8

Make sure the start of the arg area and the end (after the RA is pushed)
is always 8-byte aligned for fastcc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21995 91177308-0d34-0410-b5e6-96231b3b80d8

fix the assertion

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21994 91177308-0d34-0410-b5e6-96231b3b80d8

Align doubles on 8-byte boundaries if possible.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21993 91177308-0d34-0410-b5e6-96231b3b80d8

print stack object alignment in -print-machineinstr dumps

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21992 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21991 91177308-0d34-0410-b5e6-96231b3b80d8

Minor fixup of typos.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21990 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21989 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the problems with callee popped argument lists

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21988 91177308-0d34-0410-b5e6-96231b3b80d8

Removed PPC Linux support, as no one seems to have used it yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21987 91177308-0d34-0410-b5e6-96231b3b80d8

Don't emit SAR X, 0  in the case of sdiv Y, 2

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21986 91177308-0d34-0410-b5e6-96231b3b80d8

Fix UnitTests/2005-05-13-SDivTwo.c

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21985 91177308-0d34-0410-b5e6-96231b3b80d8

switch to having the callee pop stack operands for fastcc.  This is currently buggy
do not use

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21984 91177308-0d34-0410-b5e6-96231b3b80d8

Tolerate instrs with extra args

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21982 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note about SPARC V8.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21981 91177308-0d34-0410-b5e6-96231b3b80d8

allow RETI

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21980 91177308-0d34-0410-b5e6-96231b3b80d8

add support for fastcc and friends

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21979 91177308-0d34-0410-b5e6-96231b3b80d8

Minor fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21978 91177308-0d34-0410-b5e6-96231b3b80d8

treat TAILCALL nodes identically to CALL nodes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21977 91177308-0d34-0410-b5e6-96231b3b80d8

Build TAILCALL nodes in LowerCallTo, treat them like normal calls everywhere.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21976 91177308-0d34-0410-b5e6-96231b3b80d8

Added some missing platforms.
Fixed some wording to be more accurate (I hope).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21975 91177308-0d34-0410-b5e6-96231b3b80d8

More typo and wording fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21974 91177308-0d34-0410-b5e6-96231b3b80d8

Add reassociation, minor cleanups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21973 91177308-0d34-0410-b5e6-96231b3b80d8

close off B< tag

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21972 91177308-0d34-0410-b5e6-96231b3b80d8

remove htmlisms

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21971 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21970 91177308-0d34-0410-b5e6-96231b3b80d8

note how to disable inlining

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21969 91177308-0d34-0410-b5e6-96231b3b80d8

fix link to llvm-ar

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21968 91177308-0d34-0410-b5e6-96231b3b80d8

these don't use passes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@21967 91177308-0d34-0410-b5e6-96231b3b80d8