llvm-svn: 69967

ISD::VECTOR_SHUFFLE now stores an array of integers representing the shuffle
mask internal to the node, rather than taking a BUILD_VECTOR of ConstantSDNodes
as the shuffle mask.  A value of -1 represents UNDEF.

In addition to eliminating the creation of illegal BUILD_VECTORS just to 
represent shuffle masks, we are better about canonicalizing the shuffle mask,
resulting in substantially better code for some classes of shuffles.

A clean up of x86 shuffle code, and some canonicalizing in DAGCombiner is next.

llvm-svn: 69952

Tested by bootstrapping llvm-gcc and using that to build llvm.

llvm-svn: 68645

builds.

--- Reverse-merging (from foreign repository) r68552 into '.':
U    test/CodeGen/X86/tls8.ll
U    test/CodeGen/X86/tls10.ll
U    test/CodeGen/X86/tls2.ll
U    test/CodeGen/X86/tls6.ll
U    lib/Target/X86/X86Instr64bit.td
U    lib/Target/X86/X86InstrSSE.td
U    lib/Target/X86/X86InstrInfo.td
U    lib/Target/X86/X86RegisterInfo.cpp
U    lib/Target/X86/X86ISelLowering.cpp
U    lib/Target/X86/X86CodeEmitter.cpp
U    lib/Target/X86/X86FastISel.cpp
U    lib/Target/X86/X86InstrInfo.h
U    lib/Target/X86/X86ISelDAGToDAG.cpp
U    lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.cpp
U    lib/Target/X86/AsmPrinter/X86IntelAsmPrinter.cpp
U    lib/Target/X86/AsmPrinter/X86ATTAsmPrinter.h
U    lib/Target/X86/AsmPrinter/X86IntelAsmPrinter.h
U    lib/Target/X86/X86ISelLowering.h
U    lib/Target/X86/X86InstrInfo.cpp
U    lib/Target/X86/X86InstrBuilder.h
U    lib/Target/X86/X86RegisterInfo.td

llvm-svn: 68560

This introduces a small regression on the generated code
quality in the case we are just computing addresses, not
loading values.

Will work on it and on X86-64 support.

llvm-svn: 68552

ADDS{D|S}rr_Int and MULS{D|S}rr_Int are not commutable. The users of these intrinsics expect the high bits will not be modified.

llvm-svn: 65499

Generate better code for v16i8 shuffles on SSE2 (avoids stack)
Generate pshufb for v8i16 and v16i8 shuffles on SSSE3 where it is fewer uops.
Document the shuffle matching logic and add some FIXMEs for later further
  cleanups.
New tests that test the above.

Examples:

New:
_shuf2:
	pextrw	$7, %xmm0, %eax
	punpcklqdq	%xmm1, %xmm0
	pshuflw	$128, %xmm0, %xmm0
	pinsrw	$2, %eax, %xmm0

Old:
_shuf2:
	pextrw	$2, %xmm0, %eax
	pextrw	$7, %xmm0, %ecx
	pinsrw	$2, %ecx, %xmm0
	pinsrw	$3, %eax, %xmm0
	movd	%xmm1, %eax
	pinsrw	$4, %eax, %xmm0
	ret

=========

New:
_shuf4:
	punpcklqdq	%xmm1, %xmm0
	pshufb	LCPI1_0, %xmm0

Old:
_shuf4:
	pextrw	$3, %xmm0, %eax
	movsd	%xmm1, %xmm0
	pextrw	$3, %xmm1, %ecx
	pinsrw	$4, %ecx, %xmm0
	pinsrw	$5, %eax, %xmm0

========

New:
_shuf1:
	pushl	%ebx
	pushl	%edi
	pushl	%esi
	pextrw	$1, %xmm0, %eax
	rolw	$8, %ax
	movd	%xmm0, %ecx
	rolw	$8, %cx
	pextrw	$5, %xmm0, %edx
	pextrw	$4, %xmm0, %esi
	pextrw	$3, %xmm0, %edi
	pextrw	$2, %xmm0, %ebx
	movaps	%xmm0, %xmm1
	pinsrw	$0, %ecx, %xmm1
	pinsrw	$1, %eax, %xmm1
	rolw	$8, %bx
	pinsrw	$2, %ebx, %xmm1
	rolw	$8, %di
	pinsrw	$3, %edi, %xmm1
	rolw	$8, %si
	pinsrw	$4, %esi, %xmm1
	rolw	$8, %dx
	pinsrw	$5, %edx, %xmm1
	pextrw	$7, %xmm0, %eax
	rolw	$8, %ax
	movaps	%xmm1, %xmm0
	pinsrw	$7, %eax, %xmm0
	popl	%esi
	popl	%edi
	popl	%ebx
	ret

Old:
_shuf1:
	subl	$252, %esp
	movaps	%xmm0, (%esp)
	movaps	%xmm0, 16(%esp)
	movaps	%xmm0, 32(%esp)
	movaps	%xmm0, 48(%esp)
	movaps	%xmm0, 64(%esp)
	movaps	%xmm0, 80(%esp)
	movaps	%xmm0, 96(%esp)
	movaps	%xmm0, 224(%esp)
	movaps	%xmm0, 208(%esp)
	movaps	%xmm0, 192(%esp)
	movaps	%xmm0, 176(%esp)
	movaps	%xmm0, 160(%esp)
	movaps	%xmm0, 144(%esp)
	movaps	%xmm0, 128(%esp)
	movaps	%xmm0, 112(%esp)
	movzbl	14(%esp), %eax
	movd	%eax, %xmm1
	movzbl	22(%esp), %eax
	movd	%eax, %xmm2
	punpcklbw	%xmm1, %xmm2
	movzbl	42(%esp), %eax
	movd	%eax, %xmm1
	movzbl	50(%esp), %eax
	movd	%eax, %xmm3
	punpcklbw	%xmm1, %xmm3
	punpcklbw	%xmm2, %xmm3
	movzbl	77(%esp), %eax
	movd	%eax, %xmm1
	movzbl	84(%esp), %eax
	movd	%eax, %xmm2
	punpcklbw	%xmm1, %xmm2
	movzbl	104(%esp), %eax
	movd	%eax, %xmm1
	punpcklbw	%xmm1, %xmm0
	punpcklbw	%xmm2, %xmm0
	movaps	%xmm0, %xmm1
	punpcklbw	%xmm3, %xmm1
	movzbl	127(%esp), %eax
	movd	%eax, %xmm0
	movzbl	135(%esp), %eax
	movd	%eax, %xmm2
	punpcklbw	%xmm0, %xmm2
	movzbl	155(%esp), %eax
	movd	%eax, %xmm0
	movzbl	163(%esp), %eax
	movd	%eax, %xmm3
	punpcklbw	%xmm0, %xmm3
	punpcklbw	%xmm2, %xmm3
	movzbl	188(%esp), %eax
	movd	%eax, %xmm0
	movzbl	197(%esp), %eax
	movd	%eax, %xmm2
	punpcklbw	%xmm0, %xmm2
	movzbl	217(%esp), %eax
	movd	%eax, %xmm4
	movzbl	225(%esp), %eax
	movd	%eax, %xmm0
	punpcklbw	%xmm4, %xmm0
	punpcklbw	%xmm2, %xmm0
	punpcklbw	%xmm3, %xmm0
	punpcklbw	%xmm1, %xmm0
	addl	$252, %esp
	ret

llvm-svn: 65311

llvm-svn: 64240

llvm-svn: 63852

The memory alignment requirement on some of the mov{h|l}p{d|s} patterns are 16-byte. That is overly strict. These instructions read / write f64 memory locations without alignment requirement.

llvm-svn: 63195

the same formatting as their corresponding SSE2 instructions, for
consistency.

llvm-svn: 61971

llvm-svn: 61211

llvm-svn: 60487

foldMemoryOperand how to "fold" them, by converting them into constant-pool
loads. When they aren't folded, they use xorps/cmpeqd, but for example when
register pressure is high, they may now be folded as memory operands, which
reduces register pressure.

Also, mark V_SET0 isAsCheapAsAMove so that two-address-elimination will
remat it instead of copying zeros around (V_SETALLONES was already marked).

llvm-svn: 60461

Fix lfence and mfence encoding. These look like MRM5r and MRM6r instructions except they do not have any operands. The RegModRM byte is encoded with register number 0.

llvm-svn: 57692

an unindexed load.

llvm-svn: 57612

the predicates by extending simple predicates to create
more complex predicates instead of duplicating the logic
for the simple predicates.

This doesn't reduce much redundancy in DAGISelEmitter.cpp's
generated source yet; that will require improvements to
DAGISelEmitter.cpp's instruction sorting, to make it more
effectively group nodes with similar predicates together.

llvm-svn: 57565

the same pattern as roundpd/roundps, the Intel compiler 
builtins do not:  rounds* has an extra operand.  Fixes
gcc.target/i386/sse4_1-rounds[sd]-[1234].c

llvm-svn: 57370

Certain patterns involving the "movss" instruction were marked as requiring SSE2, when in reality movss is an SSE1 instruction.

llvm-svn: 57246

a constant vector ("{0x123, 0x456}" syntax).  The fix is to simplify the
_mm_srli_si128 macro, and  move the "* 8" from the macro into the compiler
back-end.  I can't change the existing __builtins because so many people are
using them :-(."
Patch by Stuart Hastings!

llvm-svn: 56944

Implement "punpckldq  %xmm0, $xmm0" as "pshufd  $0x50, %xmm0, %xmm" unless optimizing for code size.

llvm-svn: 56711

llvm-svn: 56697

With sse3 and when the source is a load or has multiple uses, favors movddup over shuffp*, pshufd, etc. Without sse3 or when the source is from a register, make use of movlhps

llvm-svn: 56620

llvm-svn: 56600

Fix patterns for SSE4.1 move and sign extend instructions. Also add instructions which fold VZEXT_MOVL and VZEXT_LOAD.

llvm-svn: 56594

with ConstantInt. This led to fixing a bug in TargetLowering.cpp
using getValue instead of getAPIntValue.

llvm-svn: 56159

i32>.  This is a little messy, but it works.

We should really get rid of the intrinsics, though, since they map
perfectly well to standard LLVM instructions.

llvm-svn: 55864

llvm-svn: 55466

necessary to use dyn_cast in these predicates.

llvm-svn: 55055

X86ISelLowering creates.

llvm-svn: 54544

llvm-svn: 54376

llvm-svn: 53720

llvm-svn: 53719

llvm-svn: 53386

llvm-svn: 52363

llvm-svn: 52352

wrong for volatile loads and stores.  In fact this
is almost all of them!  There are three types of
problems: (1) it is wrong to change the width of
a volatile memory access.  These may be used to
do memory mapped i/o, in which case a load can have
an effect even if the result is not used.  Consider
loading an i32 but only using the lower 8 bits.  It
is wrong to change this into a load of an i8, because
you are no longer tickling the other three bytes.  It
is also unwise to make a load/store wider.  For
example, changing an i16 load into an i32 load is
wrong no matter how aligned things are, since the
fact of loading an additional 2 bytes can have
i/o side-effects.  (2) it is wrong to change the
number of volatile load/stores: they may be counted
by the hardware.  (3) it is wrong to change a volatile
load/store that requires one memory access into one
that requires several.  For example on x86-32, you
can store a double in one processor operation, but to
store an i64 requires two (two i32 stores).  In a
multi-threaded program you may want to bitcast an i64
to a double and store as a double because that will
occur atomically, and be indivisible to other threads.
So it would be wrong to convert the store-of-double
into a store of an i64, because this will become two
i32 stores - no longer atomic.  My policy here is
to say that the number of processor operations for
an illegal operation is undefined.  So it is alright
to change a store of an i64 (requires at least two
stores; but could be validly lowered to memcpy for
example) into a store of double (one processor op).
In short, if the new store is legal and has the same
size then I say that the transform is ok.  It would
also be possible to say that transforms are always
ok if before they were illegal, whether after they
are illegal or not, but that's more awkward to do
and I doubt it buys us anything much.
However this exposed an interesting thing - on x86-32
a store of i64 is considered legal!  That is because
operations are marked legal by default, regardless of
whether the type is legal or not.  In some ways this
is clever: before type legalization this means that
operations on illegal types are considered legal;
after type legalization there are no illegal types
so now operations are only legal if they really are.
But I consider this to be too cunning for mere mortals.
Better to do things explicitly by testing AfterLegalize.
So I have changed things so that operations with illegal
types are considered illegal - indeed they can never
map to a machine operation.  However this means that
the DAG combiner is more conservative because before
it was "accidentally" performing transforms where the
type was illegal because the operation was nonetheless
marked legal.  So in a few such places I added a check
on AfterLegalize, which I suppose was actually just
forgotten before.  This causes the DAG combiner to do
slightly more than it used to, which resulted in the X86
backend blowing up because it got a slightly surprising
node it wasn't expecting, so I tweaked it.

llvm-svn: 52254

llvm-svn: 51667

llvm-svn: 51630

is a memory location

llvm-svn: 51626