Commit Graph

63 Commits

Author SHA1 Message Date
Andrew Trick
2f1d5d44c0 Cortex-A9 latency fixes (w/ -schedmodel only).
Quick review against the manual revealed a few obvious mistakes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@164361 91177308-0d34-0410-b5e6-96231b3b80d8
2012-09-21 05:06:40 +00:00
Andrew Trick
dcf31ed413 Cortex-A9 instruction-level scheduling machine model.
This models the A9 processor at the level of instruction operands, as
opposed to the itinerary, which models each operation at the level of
pipeline stages.

The two primary motivations are:

1) Allow MachineScheduler to model A9 as an out-of-order processor. It
can now distinguish between hazards that force interlocking vs.
buffered resources.

2) Reduce long-term maintenance by allowing the itinerary and target
hooks to eventually be removed. Note that almost all of the complexity
in the new model exists to model instruction variants, which the
itinerary cannot handle. Instead the scheduler previously relied on
processor-specific target hooks which are incomplete and buggy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@163921 91177308-0d34-0410-b5e6-96231b3b80d8
2012-09-14 18:31:58 +00:00
Andrew Trick
d43b5c97cf Added MispredictPenalty to SchedMachineModel.
This replaces an existing subtarget hook on ARM and allows standard
CodeGen passes to potentially use the property.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@161471 91177308-0d34-0410-b5e6-96231b3b80d8
2012-08-08 02:44:16 +00:00
Andrew Trick
2661b411cc I'm introducing a new machine model to simultaneously allow simple
subtarget CPU descriptions and support new features of
MachineScheduler.

MachineModel has three categories of data:
1) Basic properties for coarse grained instruction cost model.
2) Scheduler Read/Write resources for simple per-opcode and operand cost model (TBD).
3) Instruction itineraties for detailed per-cycle reservation tables.

These will all live side-by-side. Any subtarget can use any
combination of them. Instruction itineraries will not change in the
near term. In the long run, I expect them to only be relevant for
in-order VLIW machines that have complex contraints and require a
precise scheduling/bundling model. Once itineraries are only actively
used by VLIW-ish targets, they could be replaced by something more
appropriate for those targets.

This tablegen backend rewrite sets things up for introducing
MachineModel type #2: per opcode/operand cost model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@159891 91177308-0d34-0410-b5e6-96231b3b80d8
2012-07-07 04:00:00 +00:00
Andrew Trick
218ee74a01 Reapply "Make NumMicroOps a variable in the subtarget's instruction itinerary."
Reapplies r159406 with minor cleanup. The regressions appear to have been spurious.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@159541 91177308-0d34-0410-b5e6-96231b3b80d8
2012-07-02 18:10:42 +00:00
Andrew Trick
3e4b3b9043 Revert "Make NumMicroOps a variable in the subtarget's instruction itinerary."
This reverts commit r159406. I noticed a performance regression so I'll back out for now.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@159411 91177308-0d34-0410-b5e6-96231b3b80d8
2012-06-29 07:10:41 +00:00
Andrew Trick
0d9513c74f Make NumMicroOps a variable in the subtarget's instruction itinerary.
The TargetInstrInfo::getNumMicroOps API does not change, but soon it
will be used by MachineScheduler. Now each subtarget can specify the
number of micro-ops per itinerary class. For ARM, this is currently
always dynamic (-1), because it is used for load/store multiple which
depends on the number of register operands.

Zero is now a valid number of micro-ops. This can be used for
nop pseudo-instructions or instructions that the hardware can squash
during dispatch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@159406 91177308-0d34-0410-b5e6-96231b3b80d8
2012-06-29 03:23:18 +00:00
Andrew Trick
f94f051cf5 ARM itinerary properties.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@157980 91177308-0d34-0410-b5e6-96231b3b80d8
2012-06-05 03:44:43 +00:00
Evan Cheng
82509e5c62 Fix a number of problems with ARM fused multiply add/subtract instructions.
1. The new instruction itinerary entries are not properly described.
2. The asm parser can't handle vfms and vfnms.
3. There were no assembler, disassembler test cases.
4. HasNEON2 has the wrong assembler predicate.
rdar://10139676


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@154456 91177308-0d34-0410-b5e6-96231b3b80d8
2012-04-11 00:13:00 +00:00
Bob Wilson
11334dbd66 Improvements for the Cortex-A9 scheduling itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129770 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 18:11:36 +00:00
Evan Cheng
75b41f1540 Change A9 scheduling itineraries VLD* / VST* entries default to "aligned". That
is, it assumes addresses are 64-bit aligned (which should be the more common
case). If the alignment is found not to be aligned, then getOperandLatency()
would adjust the operand latency computation by one to compensate for it.
rdar://9294833


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129742 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-19 01:21:49 +00:00
Evan Cheng
9fe2009956 Sorry, several patches in one.
TargetInstrInfo:
Change produceSameValue() to take MachineRegisterInfo as an optional argument.
When in SSA form, targets can use it to make more aggressive equality analysis.

Machine LICM:
1. Eliminate isLoadFromConstantMemory, use MI.isInvariantLoad instead.
2. Fix a bug which prevent CSE of instructions which are not re-materializable.
3. Use improved form of produceSameValue.

ARM:
1. Teach ARM produceSameValue to look pass some PIC labels.
2. Look for operands from different loads of different constant pool entries
   which have same values.
3. Re-implement PIC GA materialization using movw + movt. Combine the pair with
   a "add pc" or "ldr [pc]" to form pseudo instructions. This makes it possible
   to re-materialize the instruction, allow machine LICM to hoist the set of
   instructions out of the loop and make it possible to CSE them. It's a bit
   hacky, but it significantly improve code quality.
4. Some minor bug fixes as well.

With the fixes, using movw + movt to materialize GAs significantly outperform the
load from constantpool method. 186.crafty and 255.vortex improved > 20%, 254.gap
and 176.gcc ~10%.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@123905 91177308-0d34-0410-b5e6-96231b3b80d8
2011-01-20 08:34:58 +00:00
Andrew Trick
6018deefe8 Fix the ARM IIC_iCMPsi itinerary and add an important assert.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122794 91177308-0d34-0410-b5e6-96231b3b80d8
2011-01-04 00:32:57 +00:00
Evan Cheng
5d9e016025 Fix an obvious cut-n-paste error.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@121307 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-08 23:01:18 +00:00
Bob Wilson
6c4c982f83 Add support for NEON VLD3-dup instructions.
The encoding for alignment in VLD4-dup instructions is still a work in progress.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120356 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-30 00:00:35 +00:00
Bob Wilson
86c6d80a7a Add support for NEON VLD3-dup instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120312 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-29 19:35:29 +00:00
Bob Wilson
2fcda63763 Fix copy-and-paste errors in VLD2-dup scheduling itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120311 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-29 19:35:23 +00:00
Bob Wilson
b1dfa7a8e0 Add support for NEON VLD2-dup instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120236 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-28 06:51:26 +00:00
Bob Wilson
2a0e97431e Add NEON VLD1-dup instructions (load 1 element to all lanes).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120194 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-27 06:35:16 +00:00
Bob Wilson
8d41294664 Fix incorrect scheduling itineraries for NEON vld1/vst1 instructions.
I added these instructions recently but I have no idea where these "1"
values in the NextCycles field came from.  As far as I can tell now,
these instruction stages are clearly intended to overlap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120193 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-27 06:35:09 +00:00
Evan Cheng
c47f7d643e Conditional moves are slightly more expensive than moves.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118985 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-13 05:14:20 +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
Evan Cheng
41957f6eb2 Modify scheduling itineraries to correct instruction latencies (not operand
latencies) of loads.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118134 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-03 00:40:22 +00:00
Bob Wilson
d0c6bc2204 Add NEON VST1-lane instructions. Partial fix for Radar 8599955.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118069 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-02 21:18:25 +00:00
Bob Wilson
b796bbb6de Add NEON VLD1-lane instructions. Partial fix for Radar 8599955.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117964 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-01 22:04:05 +00:00
Evan Cheng
e09206d4d7 Fix fpscr <-> GPR latency info.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117737 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-29 23:16:55 +00:00
Evan Cheng
7e2fe9150f Re-commit 117518 and 117519 now that ARM MC test failures are out of the way.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117531 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-28 06:47:08 +00:00
Evan Cheng
9e08ee5d16 Revert 117518 and 117519 for now. They changed scheduling and cause MC tests to fail. Ugh.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117520 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-28 02:00:25 +00:00
Evan Cheng
0104d9de04 - Assign load / store with shifter op address modes the right itinerary classes.
- For now, loads of [r, r] addressing mode is the same as the
  [r, r lsl/lsr/asr #] variants. ARMBaseInstrInfo::getOperandLatency() should
  identify the former case and reduce the output latency by 1.
- Also identify [r, r << 2] case. This special form of shifter addressing mode
  is "free".


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117519 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-28 01:49:06 +00:00
Andrew Trick
5b7a825ec5 putback r116983 and fix simple-fp-encoding.ll tests
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116992 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-21 03:40:16 +00:00
Owen Anderson
d9707e3d85 Revert r116983, which is breaking all the buildbots.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116987 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-21 03:11:16 +00:00
Evan Cheng
d6865de2d2 Add missing scheduling itineraries for transfers between core registers and VFP registers.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116983 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-21 01:12:00 +00:00
Evan Cheng
14ce175216 Limit load / store issues (at least until we have a true multi-issue aware scheduler).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116389 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 01:54:21 +00:00
Evan Cheng
08cec1ef27 More ARM scheduling itinerary fixes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116266 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-11 23:41:41 +00:00
Evan Cheng
60ff87914f Proper VST scheduling itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116251 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-11 22:03:18 +00:00
Evan Cheng
10dc63feeb Add VLD4 scheduling itineraries.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116143 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-09 04:07:58 +00:00
Evan Cheng
84f69e8436 Finish vld3 and vld4.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116140 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-09 01:45:34 +00:00
Evan Cheng
d2ca813549 Correct some load / store instruction itinerary mistakes:
1. Cortex-A8 load / store multiplies can only issue on ALU0.
2. Eliminate A8_Issue, A8_LSPipe will correctly limit the load / store issues.
3. Correctly model all vld1 and vld2 variants.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116134 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-09 01:03:04 +00:00
Evan Cheng
5a50ceeaea Model operand cycles of vldm / vstm; also fixes scheduling itineraries of vldr / vstr, etc.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115898 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-07 01:50:48 +00:00
Evan Cheng
a0792de66c - Add TargetInstrInfo::getOperandLatency() to compute operand latencies. This
allow target to correctly compute latency for cases where static scheduling
  itineraries isn't sufficient. e.g. variable_ops instructions such as
  ARM::ldm.
  This also allows target without scheduling itineraries to compute operand
  latencies. e.g. X86 can return (approximated) latencies for high latency
  instructions such as division.
- Compute operand latencies for those defined by load multiple instructions,
  e.g. ldm and those used by store multiple instructions, e.g. stm.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115755 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-06 06:27:31 +00:00
Evan Cheng
ec45f60cab Major changes to Cortex-A9 itinerary.
1. Model dual issues as two FUs.
2. Model the pipelines correctly: two symmetric ALUs, the multiplier is a
   dependent pipeline on ALU0.
The changes do not have much impact on codegen right now. But I plan to make
pre-RA scheduler multi-issue aware which should take good advantage of the
changes.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115457 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-03 02:03:59 +00:00
Evan Cheng
055028215d Fix r115332: correctly model AGU / NEON mux.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115365 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 22:52:29 +00:00
Evan Cheng
df9da6a033 Add operand cycles for vldr / vstr.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115353 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 21:40:30 +00:00
Evan Cheng
cae6a12a99 NEON scheduling info fix. vmov reg, reg are single cycle instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115344 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 20:50:58 +00:00
Evan Cheng
7c3423f413 Per Cortex-A9 pipeline diagram. AGU (core load / store issue) and NEON/FP issue are multiplexed. Model it correctly.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115332 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 19:41:46 +00:00
Evan Cheng
0e55fd61ae ARM instruction itinerary fixes:
1. Cortex-a9 8-bit and 16-bit loads / stores AGU cycles are 1 cycle longer than 32-bit ones.
2. Cortex-a9 is out-of-order so model all read cycles as cycle 1.
3. Lots of other random fixes for A8 and A9.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115121 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 01:08:25 +00:00
Evan Cheng
3881cb7a5d Model Cortex-a9 load to SUB, RSB, ADD, ADC, SBC, RSC, CMN, MVN, or CMP
pipeline forwarding path.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115098 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-29 22:42:35 +00:00
Evan Cheng
5d42c567c9 Separate itinerary classes for mvn from mov; for tst / teq from cmp / cmn.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115010 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-29 00:49:25 +00:00
Evan Cheng
7e1bf305cf Assign bitwise binary instructions different itinerary classes from ALU instructions such as add / sub.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115008 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-29 00:27:46 +00:00
Evan Cheng
63d66eed16 Add support to model pipeline bypass / forwarding.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115005 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-28 23:50:49 +00:00