1
0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 20:51:52 +01:00

17522 Commits

Author SHA1 Message Date
Matt Arsenault
4beb31bd8d AMDGPU: Use i64 scalar compare instructions
VI added eq/ne for i64, so use them.

llvm-svn: 281800
2016-09-17 02:02:19 +00:00
Tom Stellard
4715bc3836 AMDGPU/SI: Fix kernel argument ABI for HSA
Summary: i8, i16, and f16 values are not extended to 32-bit in the HSA kernel ABI.

Reviewers: arsenm

Subscribers: arsenm, kzhuravl, wdng, nhaehnle, llvm-commits, yaxunl

Differential Revision: https://reviews.llvm.org/D24621

llvm-svn: 281789
2016-09-16 22:20:24 +00:00
Matt Arsenault
ef8518e8b1 AMDGPU: Allow some control flow intrinsics to be CSEd
These clean up some unnecessary or instructions in
cases with complex loops.

In the original testcase I noticed this, the same
or with exec was repeated 5 or 6 times in a row. With
this only one is emitted or sometimes a copy.

llvm-svn: 281786
2016-09-16 22:11:18 +00:00
Tom Stellard
d9bf037744 AMDGPU: Refactor kernel argument lowering
Summary:
The main challenge in lowering kernel arguments for AMDGPU is determing the
memory type of the argument.  The generic calling convention code assumes
that only legal register types can be stored in memory, but this is not the
case for AMDGPU.

This consolidates all the logic AMDGPU uses for deducing memory types into a single
function.  This will make it much easier to support different ABIs in the future.

Reviewers: arsenm

Subscribers: arsenm, wdng, nhaehnle, llvm-commits, yaxunl

Differential Revision: https://reviews.llvm.org/D24614

llvm-svn: 281781
2016-09-16 21:53:00 +00:00
Matt Arsenault
6fa6edbc52 AMDGPU: Use SOPK compare instructions
llvm-svn: 281780
2016-09-16 21:41:16 +00:00
Tom Stellard
ca8fbd7138 AMDGPU/SI: Add support for triples with the mesa3d operating system
Summary:
mesa3d will use the same kernel calling convention as amdhsa, but it will
handle everything else like the default 'unknown' OS type.

Reviewers: arsenm

Subscribers: arsenm, llvm-commits, kzhuravl

Differential Revision: https://reviews.llvm.org/D22783

llvm-svn: 281779
2016-09-16 21:34:26 +00:00
Derek Schuff
f46cca3a86 [WebAssembly] Fix function types of CFGStackify tests
Make the function's declared type match its (lack of) return type

llvm-svn: 281773
2016-09-16 20:58:31 +00:00
Simon Pilgrim
cc0ec7ba6c [X86][SSE] Added vector sub combine tests
llvm-svn: 281769
2016-09-16 20:00:51 +00:00
Simon Pilgrim
d027b1d74d [X86][SSE] Added vector add combine tests
Some work great and others currently demonstrate the anti-vector bias prevalent in DAGCombiner

llvm-svn: 281768
2016-09-16 19:20:41 +00:00
Michael Kuperstein
528a2f9ceb Make test slightly more explicit. NFC.
llvm-svn: 281759
2016-09-16 18:20:43 +00:00
Ahmed Bougacha
f8a01e9d2c [AArch64][GlobalISel] Add default regbank mapping for int<>FP.
llvm-svn: 281739
2016-09-16 15:12:46 +00:00
Ahmed Bougacha
d60222bb60 [AArch64][GlobalISel] Add default regbank mapping for G_FCMP.
llvm-svn: 281738
2016-09-16 15:12:43 +00:00
Ahmed Bougacha
67c8014905 [AArch64][GlobalISel] Add default regbank mapping for FP ops.
These should have all their operands - even scalars - go on FPR.

llvm-svn: 281737
2016-09-16 15:12:40 +00:00
Ahmed Bougacha
33be9e7185 [AArch64][GlobalISel] Test default regbank mapping for G_ICMP.
Also relax a RegisterBankInfo verifier check that's incompatible with
1-bit mappings.

llvm-svn: 281735
2016-09-16 14:44:54 +00:00
Ahmed Bougacha
1fd329a7e8 [AArch64][GlobalISel] Add default regbank mappings for mixed-type ops.
We used to only support instructions with same-type operands.
Instead, use the per-register type information to map each
operand more accurately.

llvm-svn: 281734
2016-09-16 14:44:51 +00:00
Ahmed Bougacha
ebc58d2553 [AArch64][GlobalISel] Add tests for default RegBank mappings. NFC.
llvm-svn: 281733
2016-09-16 14:44:48 +00:00
Keith Walker
ed3a815136 Place the lowered phi instruction(s) before the DEBUG_VALUE entry
When a phi node is finally lowered to a machine instruction it is
important that the lowered "load" instruction is placed before the
associated DEBUG_VALUE entry describing the value loaded.

Renamed the existing SkipPHIsAndLabels to SkipPHIsLabelsAndDebug to
more fully describe that it also skips debug entries. Then used the
"new" function SkipPHIsAndLabels when the debug information should not
be skipped when placing the lowered "load" instructions so that it is
placed before the debug entries.

Differential Revision: https://reviews.llvm.org/D23760 

llvm-svn: 281727
2016-09-16 14:07:29 +00:00
Sjoerd Meijer
668b8b519c Reverting r281719, this is causing buildbot failures and timeouts again.
llvm-svn: 281722
2016-09-16 13:16:52 +00:00
Sjoerd Meijer
d564334693 This is an attempt to reapply r280808: [ARM] Lower UDIV+UREM to UDIV+MLS
(and the same for SREM)

This was causing buildbot failures earlier (time outs in the LNT suite).
However, we haven't been able to reproduce this and are suspecting this
was caused by another (reverted) patch.

llvm-svn: 281719
2016-09-16 12:10:09 +00:00
James Molloy
50b5b0ebc9 [ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:

      ldr r0, .CPI0
      bl printf
      bx lr
    .CPI0: &format_string
    format_string: .asciz "hello, world!\n"

We can emit:

      adr r0, .CPI0
      bl printf
      bx lr
    .CPI0: .asciz "hello, world!\n"

This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).

This recommit contains fixes for a nasty bug related to fast-isel fallback - because
fast-isel doesn't know about this optimization, if it runs and emits references to
a string that we inline (because fast-isel fell back to SDAG) we will end up
with an inlined string and also an out-of-line string, and we won't emit the
out-of-line string, causing backend failures.

It also contains fixes for emitting .text relocations which made the sanitizer
bots unhappy.

llvm-svn: 281715
2016-09-16 10:17:04 +00:00
Evandro Menezes
3204b2357d [AArch64] Support for FP FMA when -ffp-contract=fast
Currently, the machine combiner can proceed matching when -ffast-math is on.
It should also match when only -ffp-contract=fast is specified as was the
case before when DAGCombiner was doing the job.

Patch by: Abderrazek Zaafrani <a.zaafrani@samsung.com>.

Differential Revision: https://reviews.llvm.org/D24366

llvm-svn: 281649
2016-09-15 19:55:23 +00:00
Evgeniy Stepanov
eaeaf64504 Revert "[ARM] Promote small global constants to constant pools"
This reverts r281604, which adds text relocations to ARM binaries.

llvm-svn: 281645
2016-09-15 19:13:32 +00:00
James Molloy
e7d2986a37 [ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:

      ldr r0, .CPI0
      bl printf
      bx lr
    .CPI0: &format_string
    format_string: .asciz "hello, world!\n"

We can emit:

      adr r0, .CPI0
      bl printf
      bx lr
    .CPI0: .asciz "hello, world!\n"

This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).

This recommit contains fixes for a nasty bug related to fast-isel fallback - because
fast-isel doesn't know about this optimization, if it runs and emits references to
a string that we inline (because fast-isel fell back to SDAG) we will end up
with an inlined string and also an out-of-line string, and we won't emit the
out-of-line string, causing backend failures.

llvm-svn: 281604
2016-09-15 12:30:27 +00:00
Tim Northover
ed8959ddd8 GlobalISel: legalize GEP instructions with small offsets.
llvm-svn: 281602
2016-09-15 11:02:19 +00:00
Tim Northover
dd47cd5b6e GlobalISel: relax type constraints on G_ICMP to allow pointers.
llvm-svn: 281600
2016-09-15 10:40:38 +00:00
Sanjoy Das
58149dbd85 [Stackmap] Added callsite counts to emitted function information.
Summary:
It was previously not possible for tools to use solely the stackmap
information emitted to reconstruct the return addresses of callsites in
the map, which is necessary to use the information to walk a stack. This
patch adds per-function callsite counts when emitting the stackmap
section in order to resolve the problem. Note that this slightly alters
the stackmap format, so external tools parsing these maps will need to
be updated.

**Problem Details:**
Records only store their offset from the beginning of the function they
belong to. While these records and the functions are output in program
order, it is not possible to determine where the end of one function's
records are without the callsite count when processing the records to
compute return addresses.

Patch by Kavon Farvardin!

Reviewers: atrick, ributzka, sanjoy

Subscribers: nemanjai

Differential Revision: https://reviews.llvm.org/D23487

llvm-svn: 281532
2016-09-14 20:22:03 +00:00
Sanjay Patel
f29b6b5458 [x86] regenerate checks
llvm-svn: 281531
2016-09-14 20:21:28 +00:00
Sanjay Patel
8c715da81c [x86] regenerate checks
llvm-svn: 281529
2016-09-14 20:16:24 +00:00
Evgeniy Stepanov
8dd4aa2167 Revert "[ARM] Promote small global constants to constant pools"
Breaks Android tests by introducing text relocations to ARM binaries.

http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux/builds/25362/steps/run%20asan%20lit%20tests%20%5Barm%2Fbullhead-userdebug%2FMTC20F%5D/logs/stdio

llvm-svn: 281526
2016-09-14 20:02:30 +00:00
Sanjay Patel
bb5469d614 [x86] regenerate checks
llvm-svn: 281523
2016-09-14 19:42:03 +00:00
Matt Arsenault
6a290fc995 Revert "AMDGPU: Use SOPK compare instructions"
Accidentally committed

llvm-svn: 281514
2016-09-14 18:04:42 +00:00
Matt Arsenault
2196097c5a AMDGPU: Use SOPK compare instructions
llvm-svn: 281513
2016-09-14 18:03:53 +00:00
Simon Pilgrim
5f03f4e898 [X86][SSE] Improve recognition of i64 sitofp conversions that can be performed as i32 (PR29078)
Until AVX512DQ we only support i64/vXi64 sitofp conversion as scalars.

This patch sees if the sign bit extends far enough that we can truncate to a i32 type and then perform sitofp without loss of precision.

Differential Revision: https://reviews.llvm.org/D24345

llvm-svn: 281502
2016-09-14 17:15:26 +00:00
Matt Arsenault
4086e0fffa AMDGPU: Support folding FrameIndex operands
This avoids test regressions in a future commit.

llvm-svn: 281491
2016-09-14 15:51:33 +00:00
Matt Arsenault
9249507611 AMDGPU: Improve splitting 64-bit bit ops by constants
This addresses a TODO to handle operations besides and. This
also starts eliminating no-op operations with a constant that
can emerge later.

llvm-svn: 281488
2016-09-14 15:19:03 +00:00
James Molloy
30dc709e61 [ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:

      ldr r0, .CPI0
      bl printf
      bx lr
    .CPI0: &format_string
    format_string: .asciz "hello, world!\n"

We can emit:

      adr r0, .CPI0
      bl printf
      bx lr
    .CPI0: .asciz "hello, world!\n"

This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).

llvm-svn: 281484
2016-09-14 14:47:27 +00:00
Simon Pilgrim
a8f356133a [X86] Added i128 lshr+shl -> mask combine test
llvm-svn: 281480
2016-09-14 14:29:16 +00:00
Nemanja Ivanovic
05f2f7ea61 Fix code-gen crash on Power9 for insert_vector_elt with variable index (PR30189)
This patch corresponds to review:
https://reviews.llvm.org/D24021

In the initial implementation of this instruction, I forgot to account for
variable indices. This patch fixes PR30189 and should probably be merged into
3.9.1 (I'll open a bug according to the new instructions).

llvm-svn: 281479
2016-09-14 14:19:09 +00:00
Simon Pilgrim
a4ae5a2b5a [X86][SSE] Don't blend vector shifts with MOVSS/MOVSD directly, lower from generic shuffle
Shuffle lowering will correctly lower to MOVSS/MOVSD/PBLEND, improving commutation opportunities

llvm-svn: 281471
2016-09-14 14:08:18 +00:00
James Molloy
69a6bea0ca Revert "[Thumb] Teach ISel how to lower compares of AND bitmasks efficiently"
This reverts commit r281323. It caused chromium test failures and a selfhost failure.

llvm-svn: 281451
2016-09-14 09:45:28 +00:00
Tim Northover
aedd29a3b9 GlobalISel: mark pointer stores as legal on AArch64.
llvm-svn: 281448
2016-09-14 08:28:54 +00:00
Sjoerd Meijer
54f48e4444 This reapplies r281304. The issue was that I had missed
to copy the new isAdd field in the tablegen data structure.

llvm-svn: 281447
2016-09-14 08:20:03 +00:00
Elena Demikhovsky
857c15dab1 AVX-512: Fixed a bug in kortest.z intrinsic
Lowering was wrong - X86ISD::SETCC node should return i8 type.

llvm-svn: 281446
2016-09-14 08:06:54 +00:00
Igor Breger
c8f908fbec [AVX512BW] Change truncStore action (v16i16->v16i18). It can be legal only with AVX512VL.
Differential Revision: http://reviews.llvm.org/D24547

llvm-svn: 281445
2016-09-14 08:04:28 +00:00
Craig Topper
33e6517c79 [X86] Remove the VCVTSI2SD32 with rounding intrinsic. It's not used by clang and not needed since 32-bit integer to double is always exact.
llvm-svn: 281442
2016-09-14 06:27:46 +00:00
Ahmed Bougacha
842a0d0944 [AArch64] Simplify patchpoint/stackmap size test (r281301). NFC.
llvm-svn: 281407
2016-09-13 22:16:40 +00:00
Pawel Bylica
7f5873ae1e [CodeGen] Fix invalid shift in mul expansion
Summary: When expanding mul in type legalization make sure the type for shift amount can actually fit the value. This fixes PR30354 https://llvm.org/bugs/show_bug.cgi?id=30354.

Reviewers: hfinkel, majnemer, RKSimon

Subscribers: RKSimon, llvm-commits

Differential Revision: https://reviews.llvm.org/D24478

llvm-svn: 281403
2016-09-13 21:55:41 +00:00
Michael Kuperstein
0521652070 [DAG] Allow build-to-shuffle combine to combine builds from two wide vectors.
This allows us to, in some cases, create a vector_shuffle out of a build_vector, when
the inputs to the build are extract_elements from two different vectors, at least one
of which is wider than the output. (E.g. a <8 x i16> being constructed out of
elements from a <16 x i16> and a <8 x i16>).

Differential Revision: https://reviews.llvm.org/D24491

llvm-svn: 281402
2016-09-13 21:53:32 +00:00
Krzysztof Parzyszek
124f45603b [Hexagon] Better handling of HVX vector lowering
- Expand SELECT_CC and BR_CC for vector types.
- Implement TLI::isShuffleMaskLegal.

llvm-svn: 281397
2016-09-13 21:16:07 +00:00
Matthias Braun
f87d29dc88 AArch64: Cleanup tailcall CC check, enable swiftcc.
Cleanup/change the code that checks for possible tailcall conventions to
look the same as the one in the X86 target. This makes the distinction
between calling conventions that can guarnatee tailcalls and the ones
that may tailcall more obvious.

- Add Swift to the mayTailCall list
- PreserveMost seemed to be incorrectly part of the guarnteed tail call
  list, move it to the mayTailCall list.

llvm-svn: 281376
2016-09-13 19:27:38 +00:00