Commit Graph

61 Commits

Author SHA1 Message Date
Bob Wilson
84c5eed15b This patch combines several changes from Evan Cheng for rdar://8659675.
Making use of VFP / NEON floating point multiply-accumulate / subtraction is
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
   of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
   additional pipeline stall. So it's frequently better to single codegen
   vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
   stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
   vmla + vmla is very bad. But this isn't ideal either:
     vmul
     vadd
     vmla
   Instead, we want to expand the second vmla:
     vmla
     vmul
     vadd
   Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
   faster.

Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.

A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
   compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
   fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
   vmla / vmls will trigger one of the special hazards.

Enable these fp vmlx codegen changes for Cortex-A9.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129775 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 18:11:57 +00:00
Bob Wilson
cd70496ad1 Add -mcpu=cortex-a9-mp. It's cortex-a9 with MP extension. rdar://8648637.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129774 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 18:11:52 +00:00
Bob Wilson
5dde893c2b Avoid some 's' 16-bit instruction which partially update CPSR
(and add false dependency) when it isn't dependent on last CPSR defining
instruction. rdar://8928208

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129773 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 18:11:49 +00:00
Evan Cheng
463d358f1d Distribute (A + B) * C to (A * C) + (B * C) to make use of NEON multiplier
accumulator forwarding:
vadd d3, d0, d1
vmul d3, d3, d2
=>
vmul d3, d0, d2
vmla d3, d1, d2


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128665 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 19:38:48 +00:00
Bob Wilson
0406356cd4 Add Neon VCVT instructions for f32 <-> f16 conversions.
Clang is now providing intrinsics for these and so we need to support them
in the backend.  Radar 8068427.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@121902 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-15 22:14:12 +00:00
Evan Cheng
167be80ee7 Code clean up.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120965 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 23:03:45 +00:00
Evan Cheng
48575f6ea7 Making use of VFP / NEON floating point multiply-accumulate / subtraction is
difficult on current ARM implementations for a few reasons.
1. Even though a single vmla has latency that is one cycle shorter than a pair
   of vmul + vadd, a RAW hazard during the first (4? on Cortex-a8) can cause
   additional pipeline stall. So it's frequently better to single codegen
   vmul + vadd.
2. A vmla folowed by a vmul, vmadd, or vsub causes the second fp instruction to
   stall for 4 cycles. We need to schedule them apart.
3. A vmla followed vmla is a special case. Obvious issuing back to back RAW
   vmla + vmla is very bad. But this isn't ideal either:
     vmul
     vadd
     vmla
   Instead, we want to expand the second vmla:
     vmla
     vmul
     vadd
   Even with the 4 cycle vmul stall, the second sequence is still 2 cycles
   faster.

Up to now, isel simply avoid codegen'ing fp vmla / vmls. This works well enough
but it isn't the optimial solution. This patch attempts to make it possible to
use vmla / vmls in cases where it is profitable.

A. Add missing isel predicates which cause vmla to be codegen'ed.
B. Make sure the fmul in (fadd (fmul)) has a single use. We don't want to
   compute a fmul and a fmla.
C. Add additional isel checks for vmla, avoid cases where vmla is feeding into
   fp instructions (except for the #3 exceptional case).
D. Add ARM hazard recognizer to model the vmla / vmls hazards.
E. Add a special pre-regalloc case to expand vmla / vmls when it's likely the
   vmla / vmls will trigger one of the special hazards.

Work in progress, only A+B are enabled.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120960 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-05 22:04:16 +00:00
Evan Cheng
529916ca4a Add some missing isel predicates on def : pat patterns to avoid generating VFP vmla / vmls (they cause stalls). Disabling them in isel is properly not a right solution, I'll look into a proper solution next.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118922 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-12 20:32:20 +00:00
Evan Cheng
dfed19fe2c Fix preload instruction isel. Only v7 supports pli, and only v7 with mp extension supports pldw. Add subtarget attribute to denote mp extension support and legalize illegal ones to nothing.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118160 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-03 06:34:55 +00:00
Bob Wilson
77f42b5278 PR8359: The ARM backend may end up allocating registers D16 to D31 when
"-mattr=+vfp3" is specified. However, this will not work for hardware that
only supports 16 registers.  Add a new flag to support -"mattr=+vfp3,+d16".
Patch by Jan Voung!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116310 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 16:22:47 +00:00
Jim Grosbach
2317e40539 Nuke it from orbit. It's the only way to be sure.
(Kill the dead non-MC asm printer for the ARM target.)


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115127 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 01:57:53 +00:00
Evan Cheng
3ef1c8759a Teach if-converter to be more careful with predicating instructions that would
take multiple cycles to decode.
For the current if-converter clients (actually only ARM), the instructions that
are predicated on false are not nops. They would still take machine cycles to
decode. Micro-coded instructions such as LDM / STM can potentially take multiple
cycles to decode. If-converter should take treat them as non-micro-coded
simple instructions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113570 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-10 01:29:16 +00:00
Jim Grosbach
c5ed0134a7 80 column cleanup.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111266 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-17 18:39:16 +00:00
Chris Lattner
23e70ebf35 fix emacs language spec's, patch by Edmund Grimley-Evans!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@111241 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-17 16:20:04 +00:00
Jim Grosbach
fcba5e6b64 cortex m4 has floating point support, but only single precision.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110810 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 15:44:15 +00:00
Evan Cheng
7b4d31176e Report error if codegen tries to instantiate a ARM target when the cpu does support it. e.g. cortex-m* processors.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110798 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 07:17:46 +00:00
Evan Cheng
8d62e713ea ArchV7M implies HW division instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110797 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 07:00:16 +00:00
Evan Cheng
cb5ce6e62b ArchV6T2, V7A, and V7M implies Thumb2; Archv7A implies NEON.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110796 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 06:57:53 +00:00
Evan Cheng
d6b4632256 Add ARM Archv6M and let it implies FeatureDB (having dmb, etc.)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110795 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 06:51:54 +00:00
Evan Cheng
c7569ed4e4 Add Cortex-M0 support. It's a ARMv6m device (no ARM mode) with some 32-bit
instructions: dmb, dsb, isb, msr, and mrs.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110786 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 06:30:38 +00:00
Evan Cheng
11db068721 - Add subtarget feature -mattr=+db which determine whether an ARM cpu has the
memory and synchronization barrier dmb and dsb instructions.
- Change instruction names to something more sensible (matching name of actual
  instructions).
- Added tests for memory barrier codegen.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110785 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 06:22:01 +00:00
Evan Cheng
e44be63816 Change -prefer-32bit-thumb to attribute -mattr=+32bit instead to disable more 32-bit to 16-bit optimizations.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110584 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-09 18:35:19 +00:00
Evan Cheng
7a41599962 Add an ARM "feature". Cortex-a8 fp comparison is very slow (> 20 cycles).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@108256 91177308-0d34-0410-b5e6-96231b3b80d8
2010-07-13 19:21:50 +00:00
Jim Grosbach
29402132f3 Cleanup of ARMv7M support. Move hardware divide and Thumb2 extract/pack
instructions to subtarget features and update tests to reflect.
PR5717.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103136 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-05 23:44:43 +00:00
Jim Grosbach
b1dc393bd5 Add initial support for ARMv7M subtarget and cortex-m3 cpu. Patch by
Jordy <snhjordy@gmail.com>.

Followup patches will add some tests and adjust to use Subtarget features
for the instructions.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103119 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-05 20:44:35 +00:00
Anton Korobeynikov
2eeeff8371 Some bits of A9 scheduling: VFP
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100643 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-07 18:19:18 +00:00
Jakob Stoklund Olesen
fddb7667ca Replace TSFlagsFields and TSFlagsShifts with a simpler TSFlags field.
When a target instruction wants to set target-specific flags, it should simply
set bits in the TSFlags bit vector defined in the Instruction TableGen class.

This works well because TableGen resolves member references late:

class I : Instruction {
  AddrMode AM = AddrModeNone;
  let TSFlags{3-0} = AM.Value;
}

let AM = AddrMode4 in
def ADD : I;

TSFlags gets the expected bits from AddrMode4 in this example.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100384 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-05 03:10:20 +00:00
Jim Grosbach
1118b5e490 vml[as] are slow on 1136jf-s also.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100066 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-01 00:13:43 +00:00
Jim Grosbach
7ec7a0e96b switch the flag for using NEON for SP floating point to a subtarget 'feature'.
Re-commit. This time complete with testsuite updates.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99570 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-25 23:47:34 +00:00
Jim Grosbach
78e496e165 need to fix 'make check' tests first. revert for a moment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99569 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-25 23:34:05 +00:00
Jim Grosbach
bd17bc96bf switch the flag for using NEON for SP floating point to a subtarget 'feature'
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99568 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-25 23:32:19 +00:00
Jim Grosbach
6b2e8dc9a0 switch the use-vml[as] instructions flag to a subtarget 'feature'
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@99565 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-25 23:11:16 +00:00
Anton Korobeynikov
631379e79c Add substarget feature for FP16
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@98503 91177308-0d34-0410-b5e6-96231b3b80d8
2010-03-14 18:42:38 +00:00
David Goodwin
ebb5cb9216 Add ARMv6 itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89218 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-18 18:39:57 +00:00
Anton Korobeynikov
f95215f551 Use NEON reg-reg moves, where profitable. This reduces "domain-cross" stalls, when we used to mix vfp and neon code (the former were used for reg-reg moves)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@85764 91177308-0d34-0410-b5e6-96231b3b80d8
2009-11-02 00:10:38 +00:00
David Goodwin
9843a93e83 Remove neonfp attribute and instead set default based on CPU string. Add -arm-use-neon-fp to override the default.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@83218 91177308-0d34-0410-b5e6-96231b3b80d8
2009-10-01 22:19:57 +00:00
David Goodwin
471850ab84 Restore the -post-RA-scheduler flag as an override for the target specification. Remove -mattr for setting PostRAScheduler enable and instead use CPU string.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@83215 91177308-0d34-0410-b5e6-96231b3b80d8
2009-10-01 21:46:35 +00:00
David Goodwin
0dad89fa94 Remove -post-RA-schedule flag and add a TargetSubtarget method to enable post-register-allocation scheduling. By default it is off. For ARM, enable/disable with -mattr=+/-postrasched. Enable by default for cortex-a8.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@83122 91177308-0d34-0410-b5e6-96231b3b80d8
2009-09-30 00:10:16 +00:00
David Goodwin
127221fbdc Checkpoint NEON scheduling itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@82657 91177308-0d34-0410-b5e6-96231b3b80d8
2009-09-23 21:38:08 +00:00
David Goodwin
546952fd60 Allow a zero cycle stage to reserve/require a FU without advancing the cycle counter.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78736 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-11 22:38:43 +00:00
David Goodwin
767a952a6f Make NEON single-precision FP support the default for cortex-a8 (again).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78430 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-07 23:32:33 +00:00
David Goodwin
ce3c1f2a0e Disable NEON single-precision FP support for Cortex-A8, for now...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78209 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-05 16:40:57 +00:00
David Goodwin
1f0e404c87 By default, for cortex-a8 use NEON for single-precision FP.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78200 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-05 16:01:19 +00:00
David Goodwin
42a83f2d15 Initial support for single-precision FP using NEON. Added "neonfp" attribute to enable. Added patterns for some binary FP operations.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@78081 91177308-0d34-0410-b5e6-96231b3b80d8
2009-08-04 17:53:06 +00:00
Evan Cheng
6762d91c05 Add fake v7 itineraries for now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@76612 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-21 18:54:14 +00:00
Evan Cheng
34a0fa362d Add a Thumb2 instruction flag to that indicates whether the instruction can be transformed to 16-bit variant.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@74988 91177308-0d34-0410-b5e6-96231b3b80d8
2009-07-08 01:46:35 +00:00
Evan Cheng
8557c2bcb8 Latency information for ARM v6. It's rough and not yet hooked up. Right now we are only using branch latency to determine if-conversion limits.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73747 91177308-0d34-0410-b5e6-96231b3b80d8
2009-06-19 01:51:50 +00:00
Anton Korobeynikov
fbbf1eeccf Separate V6 from V6T2 since the latter has some extra nice instructions
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@73085 91177308-0d34-0410-b5e6-96231b3b80d8
2009-06-08 21:20:36 +00:00
Anton Korobeynikov
d4022c3fbb Add placeholder for thumb2 stuff
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72593 91177308-0d34-0410-b5e6-96231b3b80d8
2009-05-29 23:41:08 +00:00
Anton Korobeynikov
6d7d2aa38a Add ARMv7 architecture, Cortex processors and different FPU modes handling.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72337 91177308-0d34-0410-b5e6-96231b3b80d8
2009-05-23 19:51:43 +00:00