1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-30 23:42:52 +01:00
Commit Graph

8140 Commits

Author SHA1 Message Date
Elena Demikhovsky
139f25ed2c AVX-512: implemented extractelement with variable index.
Added parsing of mask register and "zeroing" semantic, like {%k1} {z}.

llvm-svn: 190595
2013-09-12 08:55:00 +00:00
Hal Finkel
6164109851 PPC: Enable aggressive anti-dependency breaking
Aggressive anti-dependency breaking is enabled by default for all PPC cores.
This provides a general speedup on the P7 and other platforms (among other
factors, the instruction group formation for the non-embedded PPC cores is done
during post-RA scheduling). In order to do this safely, the incompatibility
between uses of the MFOCRF instruction and anti-dependency breaking are
resolved by marking MFOCRF with hasExtraSrcRegAllocReq. As noted in the removed
FIXME, the problem was that MFOCRF's output is sensitive to the identify of the
source register, and always paired with a shift to undo this effect. Because
anti-dependency breaking is unaware of this hidden dependency of the shift
amount on the source register of the MFOCRF instruction, changing that register
must be inhibited.

Two test cases were adjusted: The SjLj test was made more insensitive to
register choices and scheduling; the saveCR test disabled anti-dependency
breaking because part of what it is testing is proper register reuse.

llvm-svn: 190587
2013-09-12 05:24:49 +00:00
Tom Stellard
6a507da088 R600/SI: expose TBUFFER_STORE_FORMAT_* for OpenGL transform feedback
For _XYZ, the type of VDATA is v4i32, because v3i32 doesn't exist.

The ADDR64 bit is not exposed. A simpler intrinsic that doesn't take
a resource descriptor might be nicer.

The maximum number of input SGPRs is bumped to 17.

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 190575
2013-09-12 02:55:14 +00:00
Bill Wendling
911391c864 Try to fix the atom buildbots by adding an explicit 'cpu' to the 'llc' command.
llvm-svn: 190541
2013-09-11 19:06:04 +00:00
Daniel Sanders
16a6e0ac3d [mips][msa] Added test cases that were supposed to be part of r190507, r190509, r190512, and r190518.
llvm-svn: 190522
2013-09-11 12:39:25 +00:00
Daniel Sanders
e3f2e5de18 [mips][msa] Added support for matching mulv, nlzc, sll, sra, srl, and subv from normal IR (i.e. not intrinsics)
llvm-svn: 190518
2013-09-11 11:58:30 +00:00
Daniel Sanders
96466ff8b4 [mips][msa] Added support for matching fadd, fdiv, flog2, fmul, frint, fsqrt, and fsub from normal IR (i.e. not intrinsics)
llvm-svn: 190512
2013-09-11 10:51:30 +00:00
Daniel Sanders
a52c7f09dc [mips][msa] Added support for matching div_[su] from normal IR (i.e. not intrinsics)
llvm-svn: 190509
2013-09-11 10:38:58 +00:00
Daniel Sanders
f68b00e629 [mips][msa] Added support for matching addv from normal IR (i.e. not intrinsics)
The corresponding intrinsic is now lowered into equivalent IR (ISD::ADD) before instruction selection.

llvm-svn: 190507
2013-09-11 10:28:16 +00:00
Daniel Sanders
534d28aa11 [mips][msa] Corrected the definition of the dotp_[su].[hwd] intrinsics
The elements of the operands should be half the width of the elements of
the result.

llvm-svn: 190505
2013-09-11 09:59:17 +00:00
Richard Sandiford
bfcf129b8e [SystemZ] Add TM and TMY
The main complication here is that TM and TMY (the memory forms) set
CC differently from the register forms.  When the tested bits contain
some 0s and some 1s, the register forms set CC to 1 or 2 based on the
value the uppermost bit.  The memory forms instead set CC to 1
regardless of the uppermost bit.

Until now, I've tried to make it so that a branch never tests for an
impossible CC value.  E.g. NR only sets CC to 0 or 1, so branches on the
result will only test for 0 or 1.  Originally I'd tried to do the same
thing for TM and TMY by using custom matching code in ISelDAGToDAG.
That ended up being very ugly though, and would have meant duplicating
some of the chain checks that the common isel code does.

I've therefore gone for the simpler alternative of adding an extra
operand to the TM DAG opcode to say whether a memory form would be OK.
This means that the inverse of a "TM;JE" is "TM;JNE" rather than the
more precise "TM;JNLE", just like the inverse of "TMLL;JE" is "TMLL;JNE".
I suppose that's arguably less confusing though...

llvm-svn: 190400
2013-09-10 10:20:32 +00:00
Daniel Sanders
32227b7995 [mips][msa] Removed unsupported dot product instructions (dotp_[su].b)
The dotp_[su].b instructions never existed in any revision of the MSA spec.

llvm-svn: 190398
2013-09-10 09:51:43 +00:00
Bill Wendling
5e97475233 Another attempt to fix windows buildbots.
llvm-svn: 190350
2013-09-09 20:29:32 +00:00
Bill Wendling
5dadcab742 Attempt to fix buildbots by giving an explicit output to the llvm-mc command.
llvm-svn: 190349
2013-09-09 20:22:38 +00:00
Bill Wendling
10cf877e75 Expand test to make sure that we can generate compact unwind from an ASM file.
llvm-svn: 190348
2013-09-09 20:12:36 +00:00
Bill Wendling
a2bc7420c8 Expand test to make sure that we can generate compact unwind from an ASM file.
llvm-svn: 190347
2013-09-09 20:10:54 +00:00
Joey Gouly
03af45ccfe [ARMv8] Prevent generation of deprecated IT blocks on ARMv8 in Thumb mode.
IT blocks can only be one instruction lonf, and can only contain a subset of
the 16 instructions.

Patch by Artyom Skrobov!

llvm-svn: 190309
2013-09-09 14:21:49 +00:00
Robert Lytton
b73a61715b XCore handling of thread local lowering
Fix XCoreLowerThreadLocal trying to initialise globals
which have no initializer.

Add handling of const expressions containing thread local variables.
These need to be replaced with instructions, as the thread ID is
used to access the thread local variable.

llvm-svn: 190300
2013-09-09 10:42:11 +00:00
Robert Lytton
dc8d32008e XCore target: change to Sched::Source
This sidesteps a bug in PrescheduleNodesWithMultipleUses() which
does not check if callResources will be affected by the transformation.

llvm-svn: 190299
2013-09-09 10:42:05 +00:00
Robert Lytton
4a5772968b XCore target: fix weak linkage attribute handling
llvm-svn: 190298
2013-09-09 10:41:57 +00:00
Bill Wendling
2c532e9c9b Generate compact unwind encoding from CFI directives.
We used to generate the compact unwind encoding from the machine
instructions. However, this had the problem that if the user used `-save-temps'
or compiled their hand-written `.s' file (with CFI directives), we wouldn't
generate the compact unwind encoding.

Move the algorithm that generates the compact unwind encoding into the
MCAsmBackend. This way we can generate the encoding whether the code is from a
`.ll' or `.s' file.

<rdar://problem/13623355>

llvm-svn: 190290
2013-09-09 02:37:14 +00:00
Jiangning Liu
b2cc9767e4 Implement aarch64 neon instruction set AdvSIMD (3V Diff), covering the following 26 instructions,
SADDL, UADDL, SADDW, UADDW, SSUBL, USUBL, SSUBW, USUBW, ADDHN, RADDHN, SABAL, UABAL, SUBHN, RSUBHN, SABDL, UABDL, SMLAL, UMLAL, SMLSL, UMLSL, SQDMLAL, SQDMLSL, SMULL, UMULL, SQDMULL, PMULL

llvm-svn: 190288
2013-09-09 02:20:27 +00:00
Manman Ren
edc0da266c Debug Info Testing: use null instead of an empty string in context field.
llvm-svn: 190284
2013-09-09 00:12:17 +00:00
Manman Ren
fa420c3e35 Debug Info Testing: update context from empty string to null.
Context should be either null or MDNode.

llvm-svn: 190267
2013-09-08 03:11:54 +00:00
Akira Hatanaka
3fb22c57eb [mips] Fix typos.
llvm-svn: 190236
2013-09-07 01:14:42 +00:00
Akira Hatanaka
3eef445630 [mips] Enhance command line option "-mno-ldc1-sdc1" to expand base+index double
precision loads and stores as well as reg+imm double precision loads and stores.

Previously, expansion of loads and stores was done after register allocation,
but now it takes place during legalization. As a result, users will see double
precision stores and loads being emitted to spill and restore 64-bit FP registers.

llvm-svn: 190235
2013-09-07 00:52:30 +00:00
Akira Hatanaka
b84769b3d7 [mips] Set instruction itineraries of loads, stores and conditional moves.
llvm-svn: 190219
2013-09-06 23:28:24 +00:00
Manman Ren
450526b5a9 Debug Info Testing: updated to use NULL instead of "i32 0" in a few fields.
Field 2 of DIType (Context), field 9 of DIDerivedType (TypeDerivedFrom),
field 12 of DICompositeType (ContainingType), fields 2, 7, 12 of DISubprogram
(Context, Type, ContainingType).

llvm-svn: 190205
2013-09-06 21:03:58 +00:00
Aaron Watry
e4512c5eff R600: Add support for LDS atomic subtract
Signed-off-by: Aaron Watry <awatry@gmail.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 190200
2013-09-06 20:17:42 +00:00
Manman Ren
3a3457bef0 Debug Info Testing: Updated to use null instead of "i32 0" for containing-type
field of DICompositeType.

This will help the follow-on patch of using DITypeRef for containing-type field.

llvm-svn: 190187
2013-09-06 18:13:59 +00:00
Tim Northover
5e921518f7 SelectionDAG: create correct BooleanContent constants
Occasionally DAGCombiner can spot that a SETCC operation is completely
redundant and reduce it to "all true" or "all false". If this happens to a
vector, the value produced has to take account of what a normal comparison
would have produced, which may be an all-1s bitmask.

The fix in SelectionDAG.cpp is tested, however, as far as I can see the code in
TargetLowering.cpp is possibly unreachable and almost certainly irrelevant when
triggered so there are no tests. However, I believe it's still clearly the
right change and may save someone else some hassle if it suddenly becomes
reachable. So I'm doing it anyway.

llvm-svn: 190147
2013-09-06 12:38:12 +00:00
Richard Sandiford
8d6edc5218 [SystemZ] Tweak integer comparison code
The architecture has many comparison instructions, including some that
extend one of the operands.  The signed comparison instructions use sign
extensions and the unsigned comparison instructions use zero extensions.
In cases where we had a free choice between signed or unsigned comparisons,
we were trying to decide at lowering time which would best fit the available
instructions, taking things like extension type into account.  The code
to do that was getting increasingly hairy and was also making some bad
decisions.  E.g. when comparing the result of two LLCs, it is better to use
CR rather than CLR, since CR can be fused with a branch while CLR can't.

This patch removes the lowering code and instead adds an operand to
integer comparisons to say whether signed comparison is required,
whether unsigned comparison is required, or whether either is OK.
We can then leave the choice of instruction up to the normal isel code.

llvm-svn: 190138
2013-09-06 11:51:39 +00:00
Richard Sandiford
ea5b4917b9 [SystemZ] Use XC for a memset of 0
llvm-svn: 190130
2013-09-06 10:25:07 +00:00
Matt Arsenault
f658af2617 Teach CodeGenPrepare about address spaces
llvm-svn: 190112
2013-09-06 00:18:43 +00:00
Juergen Ributzka
4554c8ed29 [X86] Perform VSELECT DAG combines also before DAG type legalization.
If the DAG already has only legal types, then the second round of DAG combines
is skipped. In this case VSELECT+SETCC patterns that match a more efficient
instruction (e.g. min/max) are never recognized.

This fix allows VSELECT+SETCC combines if the types are already legal before DAG
type legalization.

Reviewer: Nadav
llvm-svn: 190105
2013-09-05 23:02:56 +00:00
Matt Arsenault
071be273be R600: Fix i64 to i32 trunc on SI
llvm-svn: 190091
2013-09-05 19:41:10 +00:00
Tom Stellard
ce0432a0c3 R600: Add support for local memory atomic add
llvm-svn: 190080
2013-09-05 18:38:09 +00:00
Tom Stellard
6c1db18560 R600: Expand SELECT nodes rather than custom lowering them
llvm-svn: 190079
2013-09-05 18:38:03 +00:00
Tom Stellard
8f7c5a681a R600: Fix incorrect LDS size calculation
GlobalAdderss nodes that appeared in more than one basic block were
being counted twice.

llvm-svn: 190078
2013-09-05 18:37:57 +00:00
Tom Stellard
d2fff2dd99 R600/SI: Don't emit S_WQM_B64 instruction for compute shaders
llvm-svn: 190077
2013-09-05 18:37:52 +00:00
Joey Gouly
071ca2ff6d [ARMv8] Implement the new DMB/DSB operands.
This removes the custom ISD Node: MEMBARRIER and replaces it
with an intrinsic.

llvm-svn: 190055
2013-09-05 15:35:24 +00:00
Tilmann Scheller
31cc184566 Reverting 190043 for now.
Solution is not sufficient to prevent 'mov pc, lr' being emitted for jump table code.
Test case doesn't trigger the added functionality.

llvm-svn: 190047
2013-09-05 11:59:43 +00:00
Tilmann Scheller
14c2ce0a1e ARM: Add GPR register class excluding LR for use with the ADR instruction.
This improves code generation for jump tables by avoiding the emission of "mov pc, lr" which could fool the processor into believing this is a return from a function causing mispredicts. The code generation logic for jump tables uses ADR to materialize the address of the jump target.

Patch by Daniel Stewart!
   

llvm-svn: 190043
2013-09-05 11:10:31 +00:00
Richard Sandiford
399318ba38 [SystemZ] Add NC, OC and XC
For now these are just used to handle scalar ANDs, ORs and XORs in which
all operands are memory.

llvm-svn: 190041
2013-09-05 10:36:45 +00:00
Venkatraman Govindaraju
b3ea970660 [Sparc] Correctly handle call to functions with ReturnsTwice attribute.
In sparc, setjmp stores only the registers %fp, %sp, %i7 and %o7. longjmp restores
the stack, and the callee-saved registers (all local/in registers: %i0-%i7, %l0-%l7)
using the stored %fp and register windows. However, this does not guarantee that the longjmp
will restore the registers, as they were when the setjmp was called. This is because these
registers may be clobbered after returning from setjmp, but before calling longjmp.

This patch prevents the registers %i0-%i5, %l0-l7 to live across the setjmp call using the register mask.  

llvm-svn: 190033
2013-09-05 05:32:16 +00:00
Andrew Trick
330821bce0 mi-sched: Force bottom up scheduling for generic targets.
Fast register pressure tracking currently only takes effect during
bottom up scheduling. Forcing this is a bit faster and simpler for
targets that don't have many scheduling constraints and don't need
top-down scheduling.

llvm-svn: 190014
2013-09-04 23:54:00 +00:00
Eric Christopher
fd11e8a82d Expand and rewrite comment.
llvm-svn: 189998
2013-09-04 21:23:23 +00:00
Arnold Schwaighofer
22610acac7 Change swift/vldm test case to be less dependent on allocation order
'Force' values in registers using the calling convention. Now, we only depend on
the calling convention and that the allocator performs copy coalescing.

llvm-svn: 189985
2013-09-04 20:51:06 +00:00
Vincent Lejeune
4fd20e35e6 R600: Use shared op optimization when checking cycle compatibility
llvm-svn: 189981
2013-09-04 19:53:54 +00:00
Vincent Lejeune
4a8c23c168 R600: Non vector only instruction can be scheduled on trans unit
llvm-svn: 189980
2013-09-04 19:53:46 +00:00