Fix case where identical cases were not detected across case #0, because
instructions not handled would have a case value of #0, throwing things off.
This marginally shrinks the X86 asmprinter, but shrinks the sparc asmwriter
by 25 lines.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29187 91177308-0d34-0410-b5e6-96231b3b80d8

Only reuse a previous IV if it would not require a type conversion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29186 91177308-0d34-0410-b5e6-96231b3b80d8

Fix an accidentally duplicated line that caused tblgen to crash on itanium.
Add an assert that catches the real problem earlier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29185 91177308-0d34-0410-b5e6-96231b3b80d8

Maximally group commands.  When all instructions within a command set have a
series of identical commands, handle them all with one switch.  In the case
of the x86 at&t asm printer, only 3 switches are needed for all instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29184 91177308-0d34-0410-b5e6-96231b3b80d8

Change generator to remove operands as it processes them.  No change in
generated file.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29183 91177308-0d34-0410-b5e6-96231b3b80d8

Handle the last operand more intelligently.  When emitting the \n, also
return from the asmprinter to make the generated asmprinter both more
efficient and smaller.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29182 91177308-0d34-0410-b5e6-96231b3b80d8

Emit switches with 1/2 cases as unconditional code or an if/then/else for
tidyness.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29181 91177308-0d34-0410-b5e6-96231b3b80d8

Steal bits from the asm string index to use for operand information.  On both
x86 and ppc, this gets us 4 more bits to play with, since the string indices
both only use 12 bits.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29180 91177308-0d34-0410-b5e6-96231b3b80d8

Merge operand info and asmstr idx into a single 32-bit field.  No other change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29179 91177308-0d34-0410-b5e6-96231b3b80d8

Minor output changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29178 91177308-0d34-0410-b5e6-96231b3b80d8

Completely change the structure of the generated asmprinter to be more table
based and less switch-statements-with-hundreds-of-cases based.  This shrinks
the x86 asmprinters to about 1/3 their previous size.

Other improvements coming.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29177 91177308-0d34-0410-b5e6-96231b3b80d8

initial prologue and epilogue implementation. Need to define add and sub before finishing it :-)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29175 91177308-0d34-0410-b5e6-96231b3b80d8

Make the implicit def instructions look like other instrs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29174 91177308-0d34-0410-b5e6-96231b3b80d8

simple optimization: don't bother calling "exists" (which calls the syscall
"access").  Instead, just let the open call fail if the file doesn't exist.
This reduces the # syscalls executed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29173 91177308-0d34-0410-b5e6-96231b3b80d8

Remove a dead conditional: info_ can only be allocated if FD is good.  Improve
spacing

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29172 91177308-0d34-0410-b5e6-96231b3b80d8

The only entry in the stat buf this code cares about is the size.  Keep just
the size, not the whole stat buffer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29171 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce code in the error path by only allocating info_ if everything looks
good.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29170 91177308-0d34-0410-b5e6-96231b3b80d8

Unindent some code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29169 91177308-0d34-0410-b5e6-96231b3b80d8

Add functions to compute ceil(log2(N)) to match functions for floor(log2(N))

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29168 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce the size of Constants.o by 25% by generalizing specific instantiations of
std::map.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29163 91177308-0d34-0410-b5e6-96231b3b80d8

If a gccpath is specified when invoking the nightly test script we will hopefully now use that gcc for the gcc version. Also fixed some 80 char column stuff.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29162 91177308-0d34-0410-b5e6-96231b3b80d8

Use __attribute__((noinline)) only if compiled by gcc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29161 91177308-0d34-0410-b5e6-96231b3b80d8

Parameterize target node ValueType to allow more sharing of emit functions.
Also reduce the number of arguments passed to emit functions and removed a
hack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29160 91177308-0d34-0410-b5e6-96231b3b80d8

skeleton of a lowerCall implementation for ARM

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29159 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce instruction selection code size and stack frame size by factoring
code that emit target specific nodes into emit functions that are uniquified
and shared among selection routines.
e.g. This reduces X86ISelDAGToDAG.o (release) from ~2M to ~1.5M. Stack frame
size of Select_store from ~13k down to ~8k.
This is the first step. Further work to enable more sharing will follow.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29158 91177308-0d34-0410-b5e6-96231b3b80d8

On 64-bit targets like ppc64, we should use .quad to output pointer directives,
not .long.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29157 91177308-0d34-0410-b5e6-96231b3b80d8

Remove what little AIX support we have.  It has never been tested and isn't
complete.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29156 91177308-0d34-0410-b5e6-96231b3b80d8

The generated index array should be const.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29155 91177308-0d34-0410-b5e6-96231b3b80d8

Add an out-of-line virtual function to home class.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29154 91177308-0d34-0410-b5e6-96231b3b80d8

Add an out-of-line virtual method for X86DwarfWriter to give it a home.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29153 91177308-0d34-0410-b5e6-96231b3b80d8

Emit the string information for the asm writer as a single large string
and index into it, instead of emitting it like this:

  static const char * const OpStrs[] = {
    "PHINODE\n",        // PHI
    0,  // INLINEASM
    "adc ",     // ADC32mi
    "adc ",     // ADC32mi8
   ...

The old way required thousands of relocations that slows down link time and
dynamic load times.

This also cuts about 10K off each of the X86 asmprinters, and should shrink
the others as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29152 91177308-0d34-0410-b5e6-96231b3b80d8

Add two helpers for escaping and unescaping strings.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29151 91177308-0d34-0410-b5e6-96231b3b80d8

Add two helper functions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29150 91177308-0d34-0410-b5e6-96231b3b80d8

eliminate some ugly code, using ConstantExpr::getWithOperands instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29149 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new helper, simplify ConstantExpr::getWithOperandReplaced at Gabor's
request :)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29148 91177308-0d34-0410-b5e6-96231b3b80d8

Add another helper method.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29147 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed an issue where the user specified gcc was not the gcc we report to the nightly test server.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29144 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new method for bugpoint to use

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29143 91177308-0d34-0410-b5e6-96231b3b80d8

Add a new method for bugpoint to use.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29142 91177308-0d34-0410-b5e6-96231b3b80d8

Hopefully the final attempt at making IndVars preserve LCSSA.

This should fix PR 831.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29141 91177308-0d34-0410-b5e6-96231b3b80d8

Add missing PPC64 extload/truncstores

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29140 91177308-0d34-0410-b5e6-96231b3b80d8

Add a note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29139 91177308-0d34-0410-b5e6-96231b3b80d8

Bring the Xcode project up to date.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29138 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29137 91177308-0d34-0410-b5e6-96231b3b80d8

Another fix in the rotate encodings, needed when the first two operands are not
the same.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29136 91177308-0d34-0410-b5e6-96231b3b80d8

1. Simplfy bit operations.
2. Coalesce instruction cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29135 91177308-0d34-0410-b5e6-96231b3b80d8

Revert this patch temporarily until PR831 is fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29134 91177308-0d34-0410-b5e6-96231b3b80d8

Syntax error. I will run perl -c before commiting perl scripts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29133 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed some small bugs exposed when trying to get a sparc machine to run the script. Specifically some misnamed variables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29132 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed a bug handling void function types.
Requires rebuild of llvm-gcc4 (touch llvm-debug.cpp.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29131 91177308-0d34-0410-b5e6-96231b3b80d8

Print negative immediates as negative values instead of large constants
when using the immshifted addressing mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29130 91177308-0d34-0410-b5e6-96231b3b80d8

Turn an if into an else if.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29129 91177308-0d34-0410-b5e6-96231b3b80d8

Fix encoding of rotates, such as rldicl

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29128 91177308-0d34-0410-b5e6-96231b3b80d8

Handle instructions in the map, but that map to a null pointer.
This unbreaks smg2000.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29127 91177308-0d34-0410-b5e6-96231b3b80d8

IndVars now (correctly) preserves LCSSA form.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29126 91177308-0d34-0410-b5e6-96231b3b80d8

Implement PPC64 relocations types

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29125 91177308-0d34-0410-b5e6-96231b3b80d8

An overaggressive #ifdef allows a function to fall off the bottom of the
function instead of returning a value.  This sometimes allowed the ppc32 jit
to be used in 64-bit mode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29123 91177308-0d34-0410-b5e6-96231b3b80d8

Move base value of instruction to lookup table to prepare for case reduction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29122 91177308-0d34-0410-b5e6-96231b3b80d8

In addition to deleting calls, the inliner can constant fold them as well.
Handle this case, which doesn't require a new callgraph edge.  This fixes
a crash compiling MallocBench/gs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29121 91177308-0d34-0410-b5e6-96231b3b80d8

Change the callgraph representation to store the callsite along with the
target CG node.  This allows the inliner to properly update the callgraph
when using the pruning inliner.  The pruning inliner may not copy over all
call sites from a callee to a caller, so the edges corresponding to those
call sites should not be copied over either.

This fixes PR827 and Transforms/Inline/2006-07-12-InlinePruneCGUpdate.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29120 91177308-0d34-0410-b5e6-96231b3b80d8

testcase for PR827

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29119 91177308-0d34-0410-b5e6-96231b3b80d8

Add information preventing several register class constraints from working.
This implements PR828 and CodeGen/X86/2006-07-12-InlineAsmQConstraint.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29118 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for PR828.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29117 91177308-0d34-0410-b5e6-96231b3b80d8

Fix test failure on non-Apple systems.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29116 91177308-0d34-0410-b5e6-96231b3b80d8

Remove non-portable optimization that isn't worth it

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29115 91177308-0d34-0410-b5e6-96231b3b80d8

The PPC64 JIT needs register numbers to encode instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29114 91177308-0d34-0410-b5e6-96231b3b80d8

Revamp this doc to be accurate w.r.t. building llvmgcc4

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29113 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR826, testcase here: Regression/Verifier/2006-07-11-StoreStruct.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29112 91177308-0d34-0410-b5e6-96231b3b80d8

Testcase for PR826

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29111 91177308-0d34-0410-b5e6-96231b3b80d8

Emit inc / dec of registers as one byte instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29110 91177308-0d34-0410-b5e6-96231b3b80d8

Profile builds should always have debug info enabled.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29109 91177308-0d34-0410-b5e6-96231b3b80d8

Silence a warning produced in assertions-disabled mode

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29108 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29107 91177308-0d34-0410-b5e6-96231b3b80d8

It was pointed out that DEBUG() is only available with -debug.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29106 91177308-0d34-0410-b5e6-96231b3b80d8

Ensure that dump calls that are associated with asserts are removed from
non-debug build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29105 91177308-0d34-0410-b5e6-96231b3b80d8

1. Support for c++ mangled names.
2. Support for private/protected class members.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29104 91177308-0d34-0410-b5e6-96231b3b80d8

add the memri memory operand
this makes it possible for ldr instructions with non-zero immediate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29103 91177308-0d34-0410-b5e6-96231b3b80d8

Revert my indvars changes because they were breaking things.  Unfortunately this
didn't start showing up until after the recent instcombine fixes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29102 91177308-0d34-0410-b5e6-96231b3b80d8

Implement the inline asm 'A' constraint.  This implements PR825 and
CodeGen/X86/2006-07-10-InlineAsmAConstraint.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29101 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase for PR825.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29100 91177308-0d34-0410-b5e6-96231b3b80d8

Fix CodeGen/Alpha/2006-07-03-ASMFormalLowering.ll and PR818.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29099 91177308-0d34-0410-b5e6-96231b3b80d8

This is fixed

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29098 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce bloat in target libraries by removing per machine instruction assertion
from code emitter generation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29097 91177308-0d34-0410-b5e6-96231b3b80d8

In 64-bit mode, 64-bit GPRs are callee saved, not 32-bit ones.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29096 91177308-0d34-0410-b5e6-96231b3b80d8

Add a comment, and fix a typo that broke the build.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29094 91177308-0d34-0410-b5e6-96231b3b80d8

Don't indent the entire function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29093 91177308-0d34-0410-b5e6-96231b3b80d8

Update.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29092 91177308-0d34-0410-b5e6-96231b3b80d8

New entry.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29091 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed stack objects do not specify alignments, but their offsets are known.
Use that information when doing the transformation to merge multiple loads
into a 128-bit load.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29090 91177308-0d34-0410-b5e6-96231b3b80d8

Implement Regression/CodeGen/PowerPC/bswap-load-store.ll by folding bswaps
into i16/i32 load/stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29089 91177308-0d34-0410-b5e6-96231b3b80d8

New testcase for folding bswaps into i16/i32 loads and stores.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29088 91177308-0d34-0410-b5e6-96231b3b80d8

Recognize 16-bit bswaps by relaxing overconstrained pattern.
This implements Transforms/InstCombine/bswap.ll:test[34].

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29087 91177308-0d34-0410-b5e6-96231b3b80d8

Add tests for 16-bit byteswaps.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29086 91177308-0d34-0410-b5e6-96231b3b80d8

Mark internal function static

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29085 91177308-0d34-0410-b5e6-96231b3b80d8

Make instcombine not remove Phi nodes when LCSSA is live.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29083 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed some output issues where newlines were not being printed after error messages

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29082 91177308-0d34-0410-b5e6-96231b3b80d8

Fixed an issue that tried to cat externalprogramstable.txt from the testresults directory when it doesnt exist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29080 91177308-0d34-0410-b5e6-96231b3b80d8

create the raddr addressing mode that matches any register and the frame index
use raddr for the ldr instruction. This removes a dummy mov from the assembly output
remove SelectFrameIndex
remove isLoadFromStackSlot
remove isStoreToStackSlot

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29079 91177308-0d34-0410-b5e6-96231b3b80d8

Fix typo in the comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29078 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test for the LCSSA issue I just fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@29077 91177308-0d34-0410-b5e6-96231b3b80d8