to input patterns, we can fix X86ISD::CMP and X86ISD::BT as taking
two inputs (which have to be the same type) and *returning an i32*.
This is how the SDNodes get made in the graph, but we weren't able
to model it this way due to deficiencies in the pattern language.

Now we can change things like this:

 def UCOM_FpIr80: FpI_<(outs), (ins RFP80:$lhs, RFP80:$rhs), CompareFP,
-                  [(X86cmp RFP80:$lhs, RFP80:$rhs),
-                   (implicit EFLAGS)]>; // CC = ST(0) cmp ST(i)
+                  [(set EFLAGS, (X86cmp RFP80:$lhs, RFP80:$rhs))]>;

and fix terrible crimes like this:

-def : Pat<(parallel (X86cmp GR8:$src1, 0), (implicit EFLAGS)),
+def : Pat<(X86cmp GR8:$src1, 0),
           (TEST8rr GR8:$src1, GR8:$src1)>;

This relies on matching the result of TEST8rr (which is EFLAGS, which is
an implicit def) to the result of X86cmp, an i32.

llvm-svn: 98903

llvm-svn: 98902

llvm-svn: 98901

llvm-svn: 98900

like this:

 def : Pat<(add ...),
           (FOOINST)>;

When fooinst only has a single implicit def (e.g. to R1).  This will be handled
as if written as (set R1, (FOOINST ...))

llvm-svn: 98897

script to the #! command by using bash instead of /bin/sh.  Bash
searches $PATH for its script argument, but dash, which /bin/sh
resolves to on some systems, does not.
https://bugs.kde.org/show_bug.cgi?id=231257 tracks the valgrind
problem.

llvm-svn: 98892

llvm-svn: 98889

llvm-svn: 98888

Refactor Reg-Reg copy emission routine for ARM. This makes cross-regclass copies weirdness more straightforward. Also, add GPR <-> SPR copy support.

llvm-svn: 98887

llvm-svn: 98881

were missing it on some movq instructions and were not including the appropriate
PCrel bias.

llvm-svn: 98880

llvm-svn: 98879

shouldn't change this.

llvm-svn: 98872

llvm-svn: 98871

llvm-svn: 98870

llvm-svn: 98869

instruction.  Instructions must use 'ins' and 'outs' now.

llvm-svn: 98868

llvm-svn: 98867

llvm-svn: 98866

Add checking that the input/output operand list in spelled right.

llvm-svn: 98865

llvm-svn: 98864

llvm-svn: 98862

No functional changes.

llvm-svn: 98860

llvm-svn: 98859

temporary workaround for matching inc/dec on x86_64 to the correct instruction.
 - This hack will eventually be replaced with a robust mechanism for handling
   matching instructions based on the available target features.

llvm-svn: 98858

to allow custom post-processing of matched instructions.

llvm-svn: 98857

llvm-svn: 98855

llvm-svn: 98854

llvm-svn: 98853

llvm-svn: 98852

llvm-svn: 98851

llvm-svn: 98850

RecyclingAllocator to allow client code to be simpler, and
simplify several clients.

llvm-svn: 98847

aligned storage, the capacity may be more than what is explicitly
requested.

llvm-svn: 98846

always create a new jump table.  The intention was to avoid merging jump
tables in SelectionDAGBuilder, and to wait for the branch folding pass to
merge tables.  Unfortunately, the same getJumpTableIndex() method is also
used to merge tables in branch folding, so as a result of this change
branch tables are never merged.  Worse, the branch folding code is expecting
getJumpTableIndex to always return the index of an existing table, but with
this change, it never does so.  In at least some cases, e.g., pr6543, this
creates references to non-existent tables.

I've fixed the problem by adding a new createJumpTableIndex function, which
will always create a new table, and I've changed getJumpTableIndex to only
look at existing tables.

llvm-svn: 98845

doesn't apply to the type, only to the variable, so subsequent uses
of U which expect it to be aligned weren't actually aligned.

llvm-svn: 98843

llvm-svn: 98840

files that produce special relocation types where the 
linker changes movq's into lea's.

llvm-svn: 98839

llvm-svn: 98838

llvm-svn: 98835