llvm-6502/lib/Target/R600
Quentin Colombet 2f7322b348 [ShrinkWrap] Add (a simplified version) of shrink-wrapping.
This patch introduces a new pass that computes the safe point to insert the
prologue and epilogue of the function.
The interest is to find safe points that are cheaper than the entry and exits
blocks.

As an example and to avoid regressions to be introduce, this patch also
implements the required bits to enable the shrink-wrapping pass for AArch64.


** Context **

Currently we insert the prologue and epilogue of the method/function in the
entry and exits blocks. Although this is correct, we can do a better job when
those are not immediately required and insert them at less frequently executed
places.
The job of the shrink-wrapping pass is to identify such places.


** Motivating example **

Let us consider the following function that perform a call only in one branch of
a if:
define i32 @f(i32 %a, i32 %b)  {
 %tmp = alloca i32, align 4
 %tmp2 = icmp slt i32 %a, %b
 br i1 %tmp2, label %true, label %false

true:
 store i32 %a, i32* %tmp, align 4
 %tmp4 = call i32 @doSomething(i32 0, i32* %tmp)
 br label %false

false:
 %tmp.0 = phi i32 [ %tmp4, %true ], [ %a, %0 ]
 ret i32 %tmp.0
}

On AArch64 this code generates (removing the cfi directives to ease
readabilities):
_f:                                     ; @f
; BB#0:
  stp x29, x30, [sp, #-16]!
  mov  x29, sp
  sub sp, sp, #16             ; =16
  cmp  w0, w1
  b.ge  LBB0_2
; BB#1:                                 ; %true
  stur  w0, [x29, #-4]
  sub x1, x29, #4             ; =4
  mov  w0, wzr
  bl  _doSomething
LBB0_2:                                 ; %false
  mov  sp, x29
  ldp x29, x30, [sp], #16
  ret

With shrink-wrapping we could generate:
_f:                                     ; @f
; BB#0:
  cmp  w0, w1
  b.ge  LBB0_2
; BB#1:                                 ; %true
  stp x29, x30, [sp, #-16]!
  mov  x29, sp
  sub sp, sp, #16             ; =16
  stur  w0, [x29, #-4]
  sub x1, x29, #4             ; =4
  mov  w0, wzr
  bl  _doSomething
  add sp, x29, #16            ; =16
  ldp x29, x30, [sp], #16
LBB0_2:                                 ; %false
  ret

Therefore, we would pay the overhead of setting up/destroying the frame only if
we actually do the call.


** Proposed Solution **

This patch introduces a new machine pass that perform the shrink-wrapping
analysis (See the comments at the beginning of ShrinkWrap.cpp for more details).
It then stores the safe save and restore point into the MachineFrameInfo
attached to the MachineFunction.
This information is then used by the PrologEpilogInserter (PEI) to place the
related code at the right place. This pass runs right before the PEI.

Unlike the original paper of Chow from PLDI’88, this implementation of
shrink-wrapping does not use expensive data-flow analysis and does not need hack
to properly avoid frequently executed point. Instead, it relies on dominance and
loop properties.

The pass is off by default and each target can opt-in by setting the
EnableShrinkWrap boolean to true in their derived class of TargetPassConfig.
This setting can also be overwritten on the command line by using
-enable-shrink-wrap.

Before you try out the pass for your target, make sure you properly fix your
emitProlog/emitEpilog/adjustForXXX method to cope with basic blocks that are not
necessarily the entry block.


** Design Decisions **

1. ShrinkWrap is its own pass right now. It could frankly be merged into PEI but
for debugging and clarity I thought it was best to have its own file.
2. Right now, we only support one save point and one restore point. At some
point we can expand this to several save point and restore point, the impacted
component would then be:
- The pass itself: New algorithm needed.
- MachineFrameInfo: Hold a list or set of Save/Restore point instead of one
  pointer.
- PEI: Should loop over the save point and restore point.
Anyhow, at least for this first iteration, I do not believe this is interesting
to support the complex cases. We should revisit that when we motivating
examples.

Differential Revision: http://reviews.llvm.org/D9210

<rdar://problem/3201744>


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@236507 91177308-0d34-0410-b5e6-96231b3b80d8
2015-05-05 17:38:16 +00:00
..
AsmParser R600/SI: Use a better error message for unsupported instructions in the assembler 2015-04-23 19:33:51 +00:00
InstPrinter R600/SI: Initial support for assembler and inline assembly 2015-04-08 01:09:26 +00:00
MCTargetDesc [mc] Clean up emission of byte sequences 2015-04-17 11:12:43 +00:00
TargetInfo R600/SI: Add a stub GCNTargetMachine 2015-01-06 18:00:21 +00:00
AMDGPU.h [PM] Remove a bunch of stale TTI creation method declarations. I nuked 2015-02-01 00:22:15 +00:00
AMDGPU.td R600/SI: Add a lower case alias for subtarget feature: +DumpCode 2015-04-28 17:37:00 +00:00
AMDGPUAlwaysInlinePass.cpp R600: Fix always inline pass breaking noinline functions 2015-04-22 17:10:44 +00:00
AMDGPUAsmPrinter.cpp [AsmPrinter] Make AsmPrinter's OutStreamer member a unique_ptr. 2015-04-24 19:11:51 +00:00
AMDGPUAsmPrinter.h R600/SI: Add some missing overrides 2015-04-08 02:07:05 +00:00
AMDGPUCallingConv.td Remove the target machine from CCState. Previously it was only used 2014-08-06 18:45:26 +00:00
AMDGPUFrameLowering.cpp [ShrinkWrap] Add (a simplified version) of shrink-wrapping. 2015-05-05 17:38:16 +00:00
AMDGPUFrameLowering.h [ShrinkWrap] Add (a simplified version) of shrink-wrapping. 2015-05-05 17:38:16 +00:00
AMDGPUInstrInfo.cpp Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
AMDGPUInstrInfo.h ArrayRefize memory operand folding. NFC. 2015-02-28 12:04:00 +00:00
AMDGPUInstrInfo.td Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
AMDGPUInstructions.td Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" 2015-04-28 14:05:47 +00:00
AMDGPUIntrinsicInfo.cpp
AMDGPUIntrinsicInfo.h Canonicalize header guards into a common format. 2014-08-13 16:26:38 +00:00
AMDGPUIntrinsics.td R600/SI: Add an intrinsic for S_FLBIT_I32 / V_FFBH_I32 2015-03-04 17:33:45 +00:00
AMDGPUISelDAGToDAG.cpp Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" 2015-04-28 14:05:47 +00:00
AMDGPUISelLowering.cpp Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
AMDGPUISelLowering.h Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
AMDGPUMachineFunction.cpp R600: Canonicalize access to function attributes, NFC 2015-02-14 02:45:45 +00:00
AMDGPUMachineFunction.h Reapply "R600: Add new intrinsic to read work dimensions" 2014-10-14 20:05:26 +00:00
AMDGPUMCInstLower.cpp R600/SI: Code cleanup 2015-05-04 16:45:08 +00:00
AMDGPUMCInstLower.h R600/SI: Don't shrink instructions whose e32 encoding doesn't exist 2015-01-15 18:42:51 +00:00
AMDGPUPromoteAlloca.cpp Remove superfluous .str() and replace std::string concatenation with Twine. 2015-03-27 17:51:30 +00:00
AMDGPURegisterInfo.cpp Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
AMDGPURegisterInfo.h Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
AMDGPURegisterInfo.td
AMDGPUSubtarget.cpp R600/SI: Add assembler support for all CI and VI VOP1 instructions 2015-04-23 19:33:54 +00:00
AMDGPUSubtarget.h Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
AMDGPUTargetMachine.cpp Grab a subtarget off of an AMDGPUTargetMachine rather than a 2015-03-21 03:17:25 +00:00
AMDGPUTargetMachine.h Remove the target independent TargetMachine::getSubtarget and 2015-03-21 04:22:23 +00:00
AMDGPUTargetTransformInfo.cpp DataLayout is mandatory, update the API to reflect it with references. 2015-03-10 02:37:25 +00:00
AMDGPUTargetTransformInfo.h [multiversion] Remove the function parameter from the unrolling 2015-02-01 14:31:23 +00:00
AMDILCFGStructurizer.cpp CodeGen: Use the new DebugLoc API, NFC 2015-03-30 19:14:47 +00:00
AMDKernelCodeT.h R600/SI: Emit amd_kernel_code_t header for AMDGPU environment 2014-12-02 22:00:07 +00:00
CaymanInstructions.td R600/SI: Implement correct f64 fdiv 2015-02-14 04:30:08 +00:00
CIInstructions.td R600/SI: Add assembler support for all CI and VI VOP1 instructions 2015-04-23 19:33:54 +00:00
CMakeLists.txt R600/SI: Spill VGPRs to scratch space for compute shaders 2015-01-14 15:42:31 +00:00
EvergreenInstructions.td Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
LLVMBuild.txt R600/SI: Start implementing an assembler 2014-11-14 14:08:00 +00:00
Makefile R600/SI: Start implementing an assembler 2014-11-14 14:08:00 +00:00
Processors.td R600/SI: Limit SGPRs to 80 on Tonga and Iceland 2015-03-09 15:48:09 +00:00
R600ClauseMergePass.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
R600ControlFlowFinalizer.cpp Reuse a bunch of cached subtargets and remove getSubtarget calls 2015-01-30 23:24:40 +00:00
R600Defines.h Canonicalize header guards into a common format. 2014-08-13 16:26:38 +00:00
R600EmitClauseMarkers.cpp Have MachineFunction cache a pointer to the subtarget to make lookups 2014-08-05 02:39:49 +00:00
R600ExpandSpecialInstrs.cpp Have MachineFunction cache a pointer to the subtarget to make lookups 2014-08-05 02:39:49 +00:00
R600InstrFormats.td R600/SI: Start implementing an assembler 2014-11-14 14:08:00 +00:00
R600InstrInfo.cpp Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
R600InstrInfo.h Remove unused argument to CreateTargetScheduleState and change 2014-10-09 01:59:35 +00:00
R600Instructions.td R600: Make FMIN/MAXNUM legal on all asics 2015-04-12 23:45:05 +00:00
R600Intrinsics.td
R600ISelLowering.cpp Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
R600ISelLowering.h Reinstate revisions r234755, r234759, r234760 2015-04-30 17:15:56 +00:00
R600MachineFunctionInfo.cpp
R600MachineFunctionInfo.h Canonicalize header guards into a common format. 2014-08-13 16:26:38 +00:00
R600MachineScheduler.cpp Remove a few more calls to TargetMachine::getSubtarget from the 2015-02-19 01:10:55 +00:00
R600MachineScheduler.h Canonicalize header guards into a common format. 2014-08-13 16:26:38 +00:00
R600OptimizeVectorRegisters.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
R600Packetizer.cpp Reuse a bunch of cached subtargets and remove getSubtarget calls 2015-01-30 23:24:40 +00:00
R600RegisterInfo.cpp Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
R600RegisterInfo.h Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
R600RegisterInfo.td
R600Schedule.td
R600TextureIntrinsicsReplacer.cpp Reduce dyn_cast<> to isa<> or cast<> where possible. 2015-04-10 11:24:51 +00:00
R700Instructions.td Reuse a bunch of cached subtargets and remove getSubtarget calls 2015-01-30 23:24:40 +00:00
SIAnnotateControlFlow.cpp R600/SI: Fix verifier errors from the SIAnnotateControlFlow pass 2015-05-01 03:44:08 +00:00
SIDefines.h R600/SI: Also enable WQM for image opcodes which calculate LOD v3 2015-02-06 02:51:20 +00:00
SIFixSGPRCopies.cpp R600/SI: Remove VReg_32 register class 2015-01-07 20:59:25 +00:00
SIFixSGPRLiveRanges.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
SIFoldOperands.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
SIInsertWaits.cpp Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
SIInstrFormats.td R600/SI: Initial support for assembler and inline assembly 2015-04-08 01:09:26 +00:00
SIInstrInfo.cpp R600/SI: Fix verifier error when producing v_madmk_f32 2015-04-24 01:57:58 +00:00
SIInstrInfo.h R600/SI: Special case v_mov_b32 as really rematerializable 2015-04-23 23:34:48 +00:00
SIInstrInfo.td Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" 2015-04-28 14:05:47 +00:00
SIInstructions.td R600/SI: Add VCC as an implict def of SI_KILL 2015-05-01 03:44:09 +00:00
SIIntrinsics.td
SIISelLowering.cpp Reapply r235977 "[DebugInfo] Add debug locations to constant SD nodes" 2015-04-28 14:05:47 +00:00
SIISelLowering.h R600/SI: Add some missing overrides 2015-04-08 02:07:05 +00:00
SILoadStoreOptimizer.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
SILowerControlFlow.cpp R600/SI: Fix indirect addressing with a negative constant offset 2015-04-23 20:32:01 +00:00
SILowerI1Copies.cpp R600/SI: Remove VReg_32 register class 2015-01-07 20:59:25 +00:00
SIMachineFunctionInfo.cpp Remove a few more calls to TargetMachine::getSubtarget from the 2015-02-19 01:10:55 +00:00
SIMachineFunctionInfo.h R600/SI: Add subtarget feature to enable VGPR spilling for all shader types 2015-01-20 19:33:04 +00:00
SIPrepareScratchRegs.cpp R600/SI: Fix simple-loop.ll test 2015-01-20 19:33:02 +00:00
SIRegisterInfo.cpp R600/SI: Insert more NOPs after READLANE on VI, don't use NOPs on CI 2015-03-24 13:40:38 +00:00
SIRegisterInfo.h Remove the need to cache the subtarget in the R600 TargetRegisterInfo 2015-03-11 18:43:21 +00:00
SIRegisterInfo.td R600/SI: Initial support for assembler and inline assembly 2015-04-08 01:09:26 +00:00
SISchedule.td R600/SI: Define a schedule model 2015-01-14 01:13:19 +00:00
SIShrinkInstructions.cpp Purge unused includes throughout libSupport. 2015-03-23 18:07:13 +00:00
SITypeRewriter.cpp Remove more superfluous .str() and replace std::string concatenation with Twine. 2015-03-30 15:42:36 +00:00
VIInstrFormats.td R600/SI: Rename dst encoding field to be consistent with docs 2015-02-18 02:15:37 +00:00
VIInstructions.td R600/SI: Add assembler support for all CI and VI VOP1 instructions 2015-04-23 19:33:54 +00:00