RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-26 14:33:02 +02:00
Code Issues Projects Releases Wiki Activity
49,945 Commits 4 Branches 5 Tags 704 MiB
a565450757
Commit Graph

84 Commits

Author SHA1 Message Date
Evan Cheng
a565450757 Another coalescer bug. When a dead copy is eliminated, transfer the kill to a def of the exact register rather than a super-register. 
llvm-svn: 78376
 2009-08-07 07:14:14 +00:00
Evan Cheng
48b49cf5b9 It turns out most of the thumb2 instructions are not allowed to touch SP. The semantics of such instructions are unpredictable. We have just been lucky that tests have been passing. 
This patch takes pain to ensure all the PEI lowering code does the right thing when lowering frame indices, insert code to manipulate stack pointers, etc. It's also custom lowering dynamic stack alloc into pseudo instructions so we can insert the right instructions at scheduling time.

This fixes PR4659 and PR4682.

llvm-svn: 78361
 2009-08-07 00:34:42 +00:00
Evan Cheng
eb8c4db271 Disable stack coloring with register for now. It's not able to set kill markers. 
llvm-svn: 78179
 2009-08-05 07:26:17 +00:00
Evan Cheng
87e24bb76e Another nasty coalescer bug (is there another kind): 
After coalescing reg1027's def and kill are both at the same point:
 %reg1027,0.000000e+00 = [56,814:0)  0@70-(814)

bb5:
60   %reg1027<def> = t2MOVr %reg1027, 14, %reg0, %reg0
68   %reg1027<def> = t2LDRi12 %reg1027<kill>, 8, 14, %reg0
76   t2CMPzri %reg1038<kill,undef>, 0, 14, %reg0, %CPSR<imp-def>
84   %reg1027<def> = t2MOVr %reg1027, 14, %reg0, %reg0
96   t2Bcc mbb<bb5,0x2030910>, 1, %CPSR<kill>

Do not remove the kill marker on t2LDRi12.

llvm-svn: 78178
 2009-08-05 07:05:41 +00:00
Evan Cheng
9b8516d864 One more. Transfer kill of the larger register when lowering an EXTRACT_SUBREG. 
llvm-svn: 78145
 2009-08-05 02:25:11 +00:00
Evan Cheng
d85e73ca64 One more place where subreg lowering forgot to transfer undefness. 
llvm-svn: 78144
 2009-08-05 01:57:22 +00:00
Evan Cheng
3020badc5c If the insert_subreg source is <undef>, insert an implicit_def instead of a copy. 
llvm-svn: 78141
 2009-08-05 01:29:24 +00:00
Evan Cheng
e366789b50 Fix part 1 of pr4682. PICADD is a 16-bit instruction even in thumb2 mode. 
llvm-svn: 78126
 2009-08-04 23:47:55 +00:00
Evan Cheng
7963a20a2d Fix test. 
llvm-svn: 78113
 2009-08-04 22:22:58 +00:00
Evan Cheng
2ec9ab08d8 Enable load / store multiple pass for Thumb2. It's not using ldrd / strd yet. 
llvm-svn: 78104
 2009-08-04 21:12:13 +00:00
Evan Cheng
29fe8806d5 In thumb mode, r7 is used as frame register. This fixes pr4681. 
llvm-svn: 78086
 2009-08-04 18:46:17 +00:00
Evan Cheng
817618d570 Emit sub r, #c instead of transforming it to add r, #-c if c fits in 8-bit. This is a bit of pre-mature optimization. 8-bit variant makes it likely it will be narrowed to a 16-bit instruction. 
llvm-svn: 78030
 2009-08-04 01:41:15 +00:00
Evan Cheng
ff81990e49 Fix a coaelescer bug. If a copy val# is extended to eliminate a non-trivially coalesced copy, and the copy kills its source register. Trim the source register's live range to the last use if possible. This fixes up kill marker to make the scavenger happy. 
llvm-svn: 77967
 2009-08-03 08:41:59 +00:00
Evan Cheng
27ee95d344 Use the i12 variant of load / store opcodes if offset is zero. Now we pass all of multisource as well. 
llvm-svn: 77939
 2009-08-03 02:38:06 +00:00
Evan Cheng
3c2d9a0698 Test both darwin and linux. 
llvm-svn: 77852
 2009-08-02 02:54:34 +00:00
Eli Friedman
cd4d9ff7fa Hack to make this test work on platforms which aren't Macs. Fixing this 
myself because I'm getting tired of seeing the red buildbots, which have 
been red since 5:30PM PDT last night.

Proposed supplement to developer policy: committers should make sure to 
be around to watch for buildbot failures after committing.

llvm-svn: 77785
 2009-08-01 16:37:18 +00:00
Evan Cheng
b9b4b9aa15 Workaround a couple of Darwin assembler bugs. 
llvm-svn: 77781
 2009-08-01 06:13:52 +00:00
Evan Cheng
c165700a7f Split t2MOVCCs since some assemblers do not recognize mov shifted register alias with predicate. 
llvm-svn: 77764
 2009-08-01 01:43:45 +00:00
Evan Cheng
5ef6928dff Fix Thumb2 function call isel. Thumb1 and Thumb2 should share the same 
instructions for calls since BL and BLX are always 32-bit long and BX is always
16-bit long.

Also, we should be using BLX to call external function stubs.

llvm-svn: 77756
 2009-08-01 00:16:10 +00:00
Evan Cheng
0c9705feed Until we have a "ALIGN" pseudo instruction, have asm printer emitted a .align 
to ensure the instruction that follows a TBB (when the number of table entries
is odd) is 2-byte aligned.
Patch by Sandeep Patel.

llvm-svn: 77705
 2009-07-31 18:35:56 +00:00
Evan Cheng
c9f31ae969 When fp is not eliminated, instructions with T2_i12 modes will be changed to T2_i8 ones. Take that into consideration when determining stack size limit for reserving register scavenging slot. 
llvm-svn: 77642
 2009-07-30 23:29:25 +00:00
David Goodwin
62efc71b9f Darwin assembler now recognizes "orn", so remove workaround. 
llvm-svn: 77627
 2009-07-30 21:51:41 +00:00
David Goodwin
d29f81da16 Darwin assembler now supports "rrx", so remove workaround. 
llvm-svn: 77625
 2009-07-30 21:38:40 +00:00
David Goodwin
2eaffa79af Cleanup and include code selection for some frame index cases. 
llvm-svn: 77622
 2009-07-30 18:56:48 +00:00
Evan Cheng
31ac181755 tbb / tbh instructions only branch forward, not backwards. 
llvm-svn: 77522
 2009-07-29 23:20:20 +00:00
Evan Cheng
fc846dd401 Optimize Thumb2 jumptable to use tbb / tbh when all the offsets fit in byte / halfword. 
llvm-svn: 77422
 2009-07-29 02:18:14 +00:00
Evan Cheng
cf483eb0c0 In thumb2 mode, add pc is unpredictable. Use add + mov pc instead (that is until more optimization goes in). 
llvm-svn: 77364
 2009-07-28 20:53:24 +00:00
David Goodwin
0c9e96bf09 Remove support for ORN to workaround <rdar://problem/7096522>. 
llvm-svn: 77363
 2009-07-28 20:51:25 +00:00
David Goodwin
dbc23ece04 Add workaround for <rdar://problem/7098328>. 
llvm-svn: 77340
 2009-07-28 18:15:38 +00:00
David Goodwin
e94d490b89 Add Thumb-2 patterns for ARMsrl_flag and ARMsra_flag. 
llvm-svn: 77329
 2009-07-28 17:06:49 +00:00
Evan Cheng
b740190d2e - More refactoring. This gets rid of all of the getOpcode calls. 
- This change also makes it possible to switch between ARM / Thumb on a
  per-function basis.
- Fixed thumb2 routine which expand reg + arbitrary immediate. It was using
  using ARM so_imm logic.
- Use movw and movt to do reg + imm when profitable.
- Other code clean ups and minor optimizations.

llvm-svn: 77300
 2009-07-28 05:48:47 +00:00
David Goodwin
0bcb94eeff ORN does not require (and can not have) the ".w" suffix. "Orthogonality" is a dirty word at ARM. 
llvm-svn: 77275
 2009-07-27 23:34:12 +00:00
David Goodwin
471e9f5b8d Add ".w" suffix for wide thumb-2 instructions. 
llvm-svn: 77199
 2009-07-27 16:31:55 +00:00
Evan Cheng
d615e606c4 Change Thumb2 jumptable codegen to one that uses two level jumps: 
Before:
      adr r12, #LJTI3_0_0
      ldr pc, [r12, +r0, lsl #2]
LJTI3_0_0:
      .long    LBB3_24
      .long    LBB3_30
      .long    LBB3_31
      .long    LBB3_32

After:
      adr r12, #LJTI3_0_0
      add pc, r12, +r0, lsl #2
LJTI3_0_0:
      b.w    LBB3_24
      b.w    LBB3_30
      b.w    LBB3_31
      b.w    LBB3_32

This has several advantages.
1. This will make it easier to optimize this to a TBB / TBH instruction +
   (smaller) table.
2. This eliminate the need for ugly asm printer hack to force the address
   into thumb addresses (bit 0 is one).
3. Same codegen for pic and non-pic.
4. This eliminate the need to align the table so constantpool island pass
   won't have to over-estimate the size.

Based on my calculation, the later is probably slightly faster as well since
ldr pc with shifter address is very slow. That is, it should be a win as long
as the HW implementation can do a reasonable job of branch predict the second
branch.

llvm-svn: 77024
 2009-07-25 00:33:29 +00:00
Evan Cheng
bddff8fbe0 Remove a duplicated test. 
llvm-svn: 77020
 2009-07-25 00:24:40 +00:00
Evan Cheng
b8b61017e8 Fix these tests. 
llvm-svn: 77006
 2009-07-24 22:42:22 +00:00
Evan Cheng
bbac2397c5 Convert a test to FileCheck. 
llvm-svn: 76954
 2009-07-24 06:01:46 +00:00
Evan Cheng
90f66e1c4e Thumb2 does not allow the use of "pc" register as part of the load / store address. 
llvm-svn: 76909
 2009-07-23 23:09:51 +00:00
Evan Cheng
f6b88eae82 Fix up ARM constant island pass for Thumb2. 
Also fixed up code to fully use the SoImm field for ADR on ARM mode.

llvm-svn: 76890
 2009-07-23 18:27:47 +00:00
Evan Cheng
192fc19379 Do not select tSXTB / tSXTH in thumb2 mode. 
llvm-svn: 76600
 2009-07-21 18:15:26 +00:00
Evan Cheng
949c2404a2 Fix ARM isle code that optimize multiply by constants which are power-of-2 +/- 1. 
llvm-svn: 76520
 2009-07-21 00:31:12 +00:00
Anton Korobeynikov
dc39f4fff8 Emit cross regclass register moves for thumb2. 
Minor code duplication cleanup.

llvm-svn: 76124
 2009-07-16 23:26:06 +00:00
David Goodwin
88f38c0380 Fix detection of valid BFC immediates. 
llvm-svn: 75576
 2009-07-14 00:57:56 +00:00
Evan Cheng
20086680b0 Don't put IT instruction before conditional branches. 
llvm-svn: 75361
 2009-07-11 07:26:20 +00:00
Chris Lattner
2bdd9c7807 convert test to use FileCheck, which is much more precise and faster than 
the previous RUN lines.  Hopefully this will be an inspiration for future
tests :)

llvm-svn: 75261
 2009-07-10 18:34:47 +00:00
Evan Cheng
24f2aefc87 Add a thumb2 pass to insert IT blocks. 
llvm-svn: 75218
 2009-07-10 01:54:42 +00:00
David Goodwin
49fbd8d6b7 Use common code for both ARM and Thumb-2 instruction and register info. 
llvm-svn: 75067
 2009-07-08 23:10:31 +00:00
David Goodwin
d19a9aa207 Add rev16 test... xfail for now 
llvm-svn: 75012
 2009-07-08 16:15:06 +00:00
David Goodwin
5bdef4b3f7 Checkpoint Thumb2 Instr info work. Generalized base code so that it can be shared between ARM and Thumb2. Not yet activated because register information must be generalized first. 
llvm-svn: 75010
 2009-07-08 16:09:28 +00:00
Evan Cheng
393e38e44b Add Thumb2 movcc instructions. 
llvm-svn: 74946
 2009-07-07 20:39:03 +00:00