Commit Graph

5925 Commits

Author SHA1 Message Date
Chad Rosier
7ae3bb83c2 [fast-isel] Add a preemptive fix for the case where we fail to materialize an
immediate in a register.  I don't believe this should ever fail, but I see no
harm in trying to make this code bullet proof.

I've added an assert to ensure my assumtion is correct.  If the assertion fires
something is wrong and we should fix it, rather then just silently fall back to
SelectionDAG isel.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@178305 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-28 23:04:47 +00:00
Michael Liao
b4f98ea121 Enhance folding of (extract_subvec (insert_subvec V1, V2, IIdx), EIdx)
- Handle the case where the result of 'insert_subvect' is bitcasted
  before 'extract_subvec'. This removes the redundant insertf128/extractf128
  pair on unaligned 256-bit vector load/store on vectors of non 64-bit integer.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177945 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-25 23:47:35 +00:00
Shuxin Yang
1cd1d02141 Disable some unsafe-fp-math DAG-combine transformation after legalization.
For instance, following transformation will be disabled:
    x + x + x => 3.0f * x;

The problem of these transformations is that it introduces a FP constant, which
following Instruction-Selection pass cannot handle.

Reviewed by Nadav, thanks a lot!

rdar://13445387


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177933 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-25 22:52:29 +00:00
Owen Anderson
7834c71433 Remove the type legality check from the SelectionDAGBuilder when it lowers @llvm.fmuladd to ISD::FMA nodes.
Performing this check unilaterally prevented us from generating FMAs when the incoming IR contained illegal vector types which would eventually be legalized to underlying types that *did* support FMA.
For example, an @llvm.fmuladd on an OpenCL float16 should become a sequence of float4 FMAs, not float4 fmul+fadd's.

NOTE: Because we still call the target-specific profitability hook, individual targets can reinstate the old behavior, if desired, by simply performing the legality check inside their callback hook.  They can also perform more sophisticated legality checks, if, for example, some illegal vector types can be productively implemented as FMAs, but not others.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177820 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-23 08:26:53 +00:00
Justin Holewinski
93c1fd49d2 Make variable name more explicit and eliminate redundant lookup in SDNodeOrdering
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177600 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 23:10:59 +00:00
Nadav Rotem
7745175be3 When computing the demanded bits of Load SDNodes, make sure that we are looking at the loaded-value operand and not the ptr result (in case of pre-inc loads).
rdar://13348420



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177596 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 22:53:44 +00:00
Christian Konig
61aba54f2c Revert "pre-RA-sched: fix TargetOpcode usage"
This reverts commit 06091513c2.

The code is obviously wrong, but the trivial fix causes
inefficient code generation on X86. Somebody with more
knowledge of the code needs to take a look here.

Signed-off-by: Christian König <christian.koenig@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177529 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 15:43:00 +00:00
Justin Holewinski
fa963a885c Move SDNode order propagation to SDNodeOrdering, which also fixes a missed
case of order propagation during isel.

Thanks Owen for the suggestion!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177525 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 14:51:01 +00:00
Christian Konig
06091513c2 pre-RA-sched: fix TargetOpcode usage
TargetOpcodes need to be treaded as Machine- and not ISD-Opcodes.

Signed-off-by: Christian König <christian.koenig@amd.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177518 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 13:49:22 +00:00
Justin Holewinski
d73dc544f5 Propagate DAG node ordering during type legalization and instruction selection
A node's ordering is only propagated during legalization if (a) the new node does
not have an ordering (is not a CSE'd node), or (b) the new node has an ordering
that is higher than the node being legalized.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@177465 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-20 00:10:32 +00:00
Bill Wendling
4cb1f5f406 Reset some of the target options which affect code generation.
This doesn't reset all of the target options within the TargetOptions
object. This is because some of those are ABI-specific and must be determined if
it's okay to change those on the fly.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176986 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-13 22:26:59 +00:00
Richard Relph
1a5c0510ec Avoid generating ISD::SELECT for vector operands to SIGN_EXTEND
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176881 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-12 18:17:18 +00:00
Nick Lewycky
33cdfe9f19 Fix a crasher newly introduced in r176659/r176649, where fast-isel tries to
lower an expect intrinsic that is a constant expression.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176830 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-11 21:44:37 +00:00
Jan Wen Voung
fa785cb22d Disable statistics on Release builds and move tests that depend on -stats.
Summary:
Statistics are still available in Release+Asserts (any +Asserts builds),
and stats can also be turned on with LLVM_ENABLE_STATS.

Move some of the FastISel stats that were moved under DEBUG()
back out of DEBUG(), since stats are disabled across the board now.

Many tests depend on grepping "-stats" output.  Move those into
a orig_dir/Stats/. so that they can be marked as unsupported
when building without statistics.

Differential Revision: http://llvm-reviews.chandlerc.com/D486

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176733 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-08 22:56:31 +00:00
Benjamin Kramer
7a58099f0a Remove default from fully covered switch.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176703 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-08 17:03:19 +00:00
Tom Stellard
03abf2f2aa LegalizeDAG: Respect the result of TLI.getBooleanContents() when expanding SETCC
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176695 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-08 15:37:02 +00:00
Tom Stellard
3ef5383b35 DAGCombiner: Use correct value type for checking legality of BR_CC v3
LegalizeDAG.cpp uses the value of the comparison operands when checking
the legality of BR_CC, so DAGCombiner should do the same.

v2:
  - Expand more BR_CC value types for NVPTX

v3:
  - Expand correct BR_CC value types for Hexagon, Mips, and XCore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176694 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-08 15:36:57 +00:00
Bill Wendling
e36b47e17b Revert r176154 in favor of a better approach.
Code generation makes some basic assumptions about the IR it's been given. In
particular, if there is only one 'invoke' in the function, then that invoke
won't be going away. However, with the advent of the `llvm.donothing' intrinsic,
those invokes may go away. If all of them go away, the landing pad no longer has
any users. This confuses the back-end, which asserts.

This happens with SjLj exceptions, because that's the model that modifies the IR
based on there being invokes, etc. in the function.

Remove any invokes of `llvm.donothing' during SjLj EH preparation. This will
give us a CFG that the back-end won't be confused about. If all of the invokes
in a function are removed, then the SjLj EH prepare pass won't insert the bogus
code the relies upon the invokes being there.
<rdar://problem/13228754&13316637>


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176677 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-08 02:21:08 +00:00
Chad Rosier
4fde76db46 [fast-isel] Seriously, add support for the expect intrinsic.
rdar://13370942

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176659 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 21:38:33 +00:00
Chad Rosier
33947b4391 [fast-isel] Add support for the expect intrinsic.
rdar://13370942

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176649 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 20:42:17 +00:00
Benjamin Kramer
3853f74aba ArrayRefize some code. No functionality change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176648 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 20:33:29 +00:00
Andrew Trick
877fcf52d1 pre-RA-sched debug-only fix
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176638 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 19:21:08 +00:00
Andrew Trick
ccd38f4efb pre-RA-sched assertion fix. This bug was exposed by r176037.
rdar:13370002 [pre-RA-sched] assertion: released too many times

I tracked this down to an earlier hack that is no longer applicable
and interfered with normal scheduler logic. With the changes in
r176037, it was causing an instruction to be scheduled multiple times.

I have an external test case that I tried hard to reduce and
failed. I can't even reproduce with llc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176636 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 19:07:57 +00:00
Nadav Rotem
4d895455fe No need to go through int64 and APInt when generating a new constant.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176615 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 06:34:49 +00:00
Jim Grosbach
34fd0d2b93 SDAG: Handle scalarizing an extend of a <1 x iN> vector.
Just scalarize the element and rebuild a vector of the result type
from that.

rdar://13281568

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176614 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-07 05:47:54 +00:00
Eli Bendersky
b804a1b178 Remove duplicate line and move another closer to its actual use
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176391 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-01 23:32:40 +00:00
Akira Hatanaka
c3c3810f51 Set properties for f128 type.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176378 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-01 21:11:44 +00:00
Chad Rosier
7590022f40 Generate an error message instead of asserting or segfaulting when we can't
handle indirect register inputs.
rdar://13322011

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176367 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-01 19:12:05 +00:00
Michael Liao
a6b20ced76 Fix PR10475
- ISD::SHL/SRL/SRA must have either both scalar or both vector operands
  but TLI.getShiftAmountTy() so far only return scalar type. As a
  result, backend logic assuming that breaks.
- Rename the original TLI.getShiftAmountTy() to
  TLI.getScalarShiftAmountTy() and re-define TLI.getShiftAmountTy() to
  return target-specificed scalar type or the same vector type as the
  1st operand.
- Fix most TICG logic assuming TLI.getShiftAmountTy() a simple scalar
  type.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176364 91177308-0d34-0410-b5e6-96231b3b80d8
2013-03-01 18:40:30 +00:00
Eli Bendersky
6437d38a0d A small refactoring + adding comments.
SelectionDAGIsel::LowerArguments needs a function, not a basic block. So it
makes sense to pass it the function instead of extracting a basic-block from
the function and then tossing it. This is also more self-documenting (functions
have arguments, BBs don't).

In addition, added comments to a couple of Select* methods.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176305 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-28 23:09:18 +00:00
Eli Bendersky
0a4da5f0fe Put some per-instruction statistics of fast isel under NDEBUG, together with
other per-instruction statistics.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176273 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-28 18:05:12 +00:00
Eric Christopher
f686cfb33b Remove unnecessary cast to void.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176222 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-27 23:49:45 +00:00
Nadav Rotem
139f50a1a8 Silence the unused variable warning.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176218 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-27 22:52:54 +00:00
Nadav Rotem
07cd932c93 The FastISEL should be fast. But when we record statistics we use atomic operations to increment the counters.
This patch disables the counters on non-debug builds. This reduces the runtime of SelectionDAGISel::SelectCodeCommon by ~5%.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176214 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-27 21:59:43 +00:00
Michael Ilseman
7dbd34b199 Reverted: r176136 - Have a way for a target to opt-out of target-independent fast isel
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176204 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-27 19:54:00 +00:00
Manman Ren
5e5974f51a SelectionDAG: If llvm.donothing has a landingpad, we should clear
CurrentCallSite to avoid an assertion failure:
assert(MMI.getCurrentCallSite() == 0 && "Overlapping call sites!");

rdar://problem/13228754


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176154 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-27 02:11:57 +00:00
Michael Ilseman
bdc98d3510 Have a way for a target to opt-out of target-independent fast isel
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176136 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-26 23:15:23 +00:00
Chad Rosier
ffa1dbadaf Fix wording.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176055 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 22:20:00 +00:00
Chad Rosier
fd3417d288 [fast-isel] Add X86FastIsel::FastLowerArguments to handle functions with 6 or
fewer scalar integer (i32 or i64) arguments. It completely eliminates the need
for SDISel for trivial functions.

Also, add the new llc -fast-isel-abort-args option, which is similar to
-fast-isel-abort option, but for formal argument lowering.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176052 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 21:59:35 +00:00
Andrew Trick
029f4fd2ff pre-RA-sched fix: only reevaluate physreg interferences when necessary.
Fixes rdar:13279013: scheduler was blowing up on select instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176037 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 19:11:48 +00:00
Matt Beaumont-Gay
50e75bfc29 'Hexadecimal' has two 'a's and only one 'i'.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176031 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 18:11:18 +00:00
Chandler Carruth
7e6ffac9ab Fix spelling noticed by Duncan.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176023 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 14:29:38 +00:00
Chandler Carruth
af23f8e403 Fix the root cause of PR15348 by correctly handling alignment 0 on
memory intrinsics in the SDAG builder.

When alignment is zero, the lang ref says that *no* alignment
assumptions can be made. This is the exact opposite of the internal API
contracts of the DAG where alignment 0 indicates that the alignment can
be made to be anything desired.

There is another, more explicit alignment that is better suited for the
role of "no alignment at all": an alignment of 1. Map the intrinsic
alignment to this early so that we don't end up generating aligned DAGs.

It is really terrifying that we've never seen this before, but we
suddenly started generating a large number of alignment 0 memcpys due to
the new code to do memcpy-based copying of POD class members. That patch
contains a bug that rounds bitfield alignments down when they are the
first field. This can in turn produce zero alignments.

This fixes weird crashes I've seen in library users of LLVM on 32-bit
hosts, etc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@176022 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-25 14:20:21 +00:00
Nadav Rotem
d99a5a3ab4 SelectionDAG compile time improvement.
One of the phases of SelectionDAG is LegalizeVectors. We don't need to sort the DAG and copy nodes around if there are no vector ops.

Speeds up the compilation time of SelectionDAG on a big scalar workload by ~8%.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175929 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-22 23:33:30 +00:00
Pete Cooper
b704ffbb46 Fix isa<> check which could never be true.
It was incorrectly checking a Function* being an IntrinsicInst* which
isn't possible.  It should always have been checking the CallInst* instead.

Added test case for x86 which ensures we only get one constant load.
It was 2 before this change.

rdar://problem/13267920

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175853 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-22 01:50:38 +00:00
Benjamin Kramer
f31bd0f7d9 DAGCombiner: Make the post-legalize vector op optimization more aggressive.
A legal BUILD_VECTOR goes in and gets constant folded into another legal
BUILD_VECTOR so we don't lose any legality here. The problematic PPC
optimization that made this check necessary was fixed recently.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175759 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-21 15:24:35 +00:00
Arnold Schwaighofer
c46e2df74c DAGCombiner: Fold pointless truncate, bitcast, buildvector series
(2xi32) (truncate ((2xi64) bitcast (buildvector i32 a, i32 x, i32 b, i32 y)))
can be folded into a (2xi32) (buildvector i32 a, i32 b).

Such a DAG would cause uneccessary vdup instructions followed by vmovn
instructions.

We generate this code on ARM NEON for a setcc olt, 2xf64, 2xf64. For example, in
the vectorized version of the code below.

double A[N];
double B[N];

void test_double_compare_to_double() {
  int i;
  for(i=0;i<N;i++)
    A[i] = (double)(A[i] < B[i]);
}

radar://13191881

Fixes bug 15283.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175670 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-20 21:33:32 +00:00
Michael Liao
eedff3547d Fix PR15267
- When extloading from a vector with non-byte-addressable element, e.g.
  <4 x i1>, the current logic breaks. Extend the current logic to
  fix the case where the element type is not byte-addressable by loading
  all bytes, bit-extracting/packing each element.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175642 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-20 18:04:21 +00:00
Benjamin Kramer
ad4da0fc32 Move the SplatByte helper to APInt and generalize it a bit.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175621 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-20 13:00:06 +00:00
Jakub Staszak
751bc8d4c9 Fix #includes, so we include only what we really need.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@175581 91177308-0d34-0410-b5e6-96231b3b80d8
2013-02-20 00:26:25 +00:00