Commit Graph

128 Commits

Author SHA1 Message Date
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
Bill Wendling
0f63075613 Proper encoding for VLDM and VSTM instructions. The register lists for these
instructions have to distinguish between lists of single- and double-precision
registers in order for the ASM matcher to do a proper job. In all other
respects, a list of single- or double-precision registers are the same as a list
of GPR registers.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119460 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-17 04:32:08 +00:00
Bill Wendling
73c57e149c vldm and vstm are mnemonics for vldmia and vstmia resp.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119321 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-16 02:00:24 +00:00
Bill Wendling
73fe34a3ee Encode the multi-load/store instructions with their respective modes ('ia',
'db', 'ib', 'da') instead of having that mode as a separate field in the
instruction. It's more convenient for the asm parser and much more readable for
humans.
<rdar://problem/8654088>


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119310 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-16 01:16:36 +00:00
Bill Wendling
ddc918b379 Add uses of the *_ldst_multi multiclasses. These aren't used yet.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118999 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-13 10:57:02 +00:00
Bill Wendling
1f4abcfa5c Convert the modes to lower case.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118998 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-13 10:43:34 +00:00
Bill Wendling
6c470b806f Add *_ldst_mult multiclasses to the ARM back-end. These will be used in the
future to separate out the ia, ib, da, db variants of the load/store multiple
instructions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118995 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-13 09:09:38 +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
Bill Wendling
85c3f24da3 Add "write back" bit encoding.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118446 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-08 21:28:03 +00:00
Bill Wendling
2f46f1f59c Add encoding for VSTR.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118220 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-04 00:59:42 +00:00
Bill Wendling
92b5a2eb16 The MC code couldn't handle ARM LDR instructions with negative offsets:
vldr.64 d1, [r0, #-32]

The problem was with how the addressing mode 5 encodes the offsets. This change
makes sure that the way offsets are handled in addressing mode 5 is consistent
throughout the MC code. It involves re-refactoring the "getAddrModeImmOpValue"
method into an "Imm12" and "addressing mode 5" version. But not to worry! The
majority of the duplicated code has been unified.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118144 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-03 01:49:29 +00:00
Jim Grosbach
e6913600c7 Break ARM addrmode4 (load/store multiple base address) into its constituent
parts. Represent the operation mode as an optional operand instead.
rdar://8614429

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118137 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-03 01:01:43 +00:00
Bill Wendling
5df0e0a61d Rename getAddrModeImm12OpValue to getAddrModeImmOpValue and expand it to work
with immediates up to 16-bits in size. The same logic is applied to other LDR
encodings, e.g. VLDR, but which use a different immediate bit width (8-bits in
VLDR's case). Removing the "12" allows it to be more generic.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118094 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-02 22:31:46 +00:00
Bill Wendling
cd944a424c Missed reverting this bit.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117971 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-01 23:17:54 +00:00
Bill Wendling
160accad6b Minor cleanup.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117969 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-01 23:11:22 +00:00
Bill Wendling
c2bf50245f Move the machine operand MC encoding patterns to the parent classes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117956 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-01 21:17:06 +00:00
Bill Wendling
69661191ce Move instruction encoding bits into the parent class and remove the temporary
*_Encode classes. These instructions are the only ones which use those classes,
so a subclass isn't necessary.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117906 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-01 06:00:39 +00:00
Chris Lattner
39ee036f40 reapply r117858 with apparent editor malfunction fixed (somehow I
got a dulicated line).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117860 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-31 19:10:56 +00:00
Chris Lattner
8b2f0822f3 revert r117858 while I check out a failure I missed.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117859 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-31 19:05:32 +00:00
Chris Lattner
efa53760fe the asm matcher can't handle operands with modifiers (like ${foo:bar}).
Instead of silently ignoring these instructions, emit a hard error and
force the target author to either refactor the target or mark the 
instruction 'isCodeGenOnly'.

Mark a few instructions in ARM and MBlaze as isCodeGenOnly the are 
doing this.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117858 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-31 18:48:12 +00:00
Jim Grosbach
7644971232 Add FIXME.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@117787 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-30 14:54:23 +00:00
Bill Wendling
01aabdac44 Add encoding for moving a value between two ARM core registers and a doublework
extension register.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116970 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-20 23:37:40 +00:00
Bill Wendling
7d31a169af Add encodings for movement between ARM core registers and single-precision
registers.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116961 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-20 22:44:54 +00:00
Bill Wendling
2695d8edd1 Reformatting. No functionalogicality changes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116625 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-15 21:50:45 +00:00
Bill Wendling
bbbdcd453d Add support for vmov.f64/.f32 encoding. There's a bit of a hack going on
here. The f32 in FCONSTS is handled as a double instead of a float in the
code. So the encoding of the immediate into the instruction isn't exactly in
line with the documentation in that regard. But given that we know it's handled
as a double, it doesn't cause any harm.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116471 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-14 02:33:26 +00:00
Bill Wendling
946a2740a5 Add encoding for 'fmstat'.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116466 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-14 01:19:34 +00:00
Bill Wendling
88cf038436 - Add encodings for multiply add/subtract instructions in all their glory.
- Add missing patterns for some multiply add/subtract instructions.
- Add encodings for VMRS and VMSR.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116464 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-14 01:02:08 +00:00
Bill Wendling
67a704de03 Add MC encodings for VCVT* instrunctions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116431 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 20:58:46 +00:00
Bill Wendling
6932643a37 Add encodings for VNEG and VSQRT. Also add encodings for VMOV, but not a test
just yet.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116386 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 01:17:33 +00:00
Bill Wendling
54908dd72b Add encodings for VCVT instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116385 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 00:56:35 +00:00
Bill Wendling
1fc6d8837f Add VCMPZ and VABS.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116383 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 00:38:07 +00:00
Bill Wendling
cd77686254 Refactor VCMP instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116379 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-13 00:04:29 +00:00
Bill Wendling
5a1fd8cf68 Add encodings for VNMUL[SD].
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116375 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 23:47:37 +00:00
Bill Wendling
caa3d467ab Add encodings for VDIV and VMUL.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116370 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 23:22:27 +00:00
Bill Wendling
52061f83e7 Refactor some of the encoding logic into a base class. This keeps us from having
to add 10+ lines to every instruction.

It may turn out that we can move this base class into it's parent class.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116362 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 23:06:54 +00:00
Bill Wendling
dd3bc112e6 Add encoding for VSUB and VCMP.
Fear not! I'm going to try a refactoring right now. :)


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116359 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 22:55:35 +00:00
Bill Wendling
174777bb2b Encoding for VADDD. Plus a test for the VFP instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116348 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 22:08:41 +00:00
Jim Grosbach
499e886fe6 Encoding for ARM-mode VADD.F32 instruction.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116338 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-12 21:22:40 +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
Eric Christopher
5371cabfc2 Fix typo.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114931 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-28 00:35:33 +00:00
Jim Grosbach
72db182438 VFP/NEON load/store multiple instructions are addrmode4, not 5.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@113322 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-08 00:25:50 +00:00
Bob Wilson
d4bfd54ec2 Change ARM VFP VLDM/VSTM instructions to use addressing mode #4, just like
all the other LDM/STM instructions.  This fixes asm printer crashes when
compiling with -O0.  I've changed one of the NEON tests (vst3.ll) to run
with -O0 to check this in the future.

Prior to this change VLDM/VSTM used addressing mode #5, but not really.
The offset field was used to hold a count of the number of registers being
loaded or stored, and the AM5 opcode field was expanded to specify the IA
or DB mode, instead of the standard ADD/SUB specifier.  Much of the backend
was not aware of these special cases.  The crashes occured when rewriting
a frameindex caused the AM5 offset field to be changed so that it did not
have a valid submode.  I don't know exactly what changed to expose this now.
Maybe we've never done much with -O0 and NEON.  Regardless, there's no longer
any reason to keep a count of the VLDM/VSTM registers, so we can use
addressing mode #4 and clean things up in a lot of places.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112322 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-27 23:18:17 +00:00
Daniel Dunbar
3bcd9f7902 ARM: Mark some disassembler only instructions as not available for matching --
for some reason they have a very odd MCInst form where the operands overlap, but
I haven't dug in to find out why yet.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110781 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-11 04:46:13 +00:00
Nate Begeman
d1fb583128 Add support for getting & setting the FPSCR application register on ARM when VFP is enabled.
Add support for using the FPSCR in conjunction with the vcvtr instruction, for controlling fp to int rounding.
Add support for the FLT_ROUNDS_ node now that the FPSCR is exposed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@110152 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-03 21:31:55 +00:00
Jim Grosbach
18f30e6f5e Clean up 80 column violations. No functional change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@105350 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-02 21:53:11 +00:00
Evan Cheng
5fd1c9be2d Mark pattern-less mayLoad / mayStore instructions neverHasSideEffects. These do not have other un-modeled side effects.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@104111 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-19 06:07:03 +00:00
Evan Cheng
020cc1b4d0 Mark some pattern-less instructions as neverHasSideEffects.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103683 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-13 00:16:46 +00:00
Anton Korobeynikov
a31c6fb65e Define new itin classes for ARM <-> VFP reg moves to distinguish from NEON ops. Define proper scheduling itinerary for them on A9. A8 TRM does not specify latency for them at all :(
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100650 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-07 18:20:02 +00:00
Anton Korobeynikov
63401e33cb FCONST{S,D} behaves the same way as FP unary instructions. This is true for both A8 and A9.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100649 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-07 18:19:56 +00:00
Anton Korobeynikov
c492e09455 Add new itin classes for FP16 <-> FP32 conversions and make uise of them for A9.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100647 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-07 18:19:46 +00:00