2010-05-17 |
Evan Cheng | Careful with reg_sequence coalescing to not to overwrite...
|
commit | commitdiff | tree |
2010-05-17 |
Evan Cheng | Turn on -neon-reg-sequence by default.
|
commit | commitdiff | tree |
2010-05-17 |
Evan Cheng | Yes, if the redef is a copy, update the old val# with...
|
commit | commitdiff | tree |
2010-05-17 |
Evan Cheng | No reason not to run the NEON domain croassing fix...
|
commit | commitdiff | tree |
2010-05-16 |
Evan Cheng | Model vst lane instructions with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-15 |
Evan Cheng | Model 128-bit vld lane with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-15 |
Evan Cheng | v4i64 and v8i64 are only synthesizable when NEON is...
|
commit | commitdiff | tree |
2010-05-15 |
Evan Cheng | Allow TargetLowering::getRegClassFor() to be called...
|
commit | commitdiff | tree |
2010-05-15 |
Evan Cheng | Model 64-bit lane vld with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-15 |
Evan Cheng | A partial re-def instruction may be a copy.
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Teach two-address pass to do some coalescing while...
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Model VST*_UPD and VST*oddUPD pair with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Model VLD*_UPD and VLD*odd_UPD pair with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Get rid of the bit twiddling to read / set OpActions...
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Added a QQQQ register file to model 4-consecutive Q...
|
commit | commitdiff | tree |
2010-05-14 |
Evan Cheng | Fix comments.
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Adding a v8i64 512-bit vector type. This will be used...
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Eliminate use of magic numbers to access OpActions...
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Fix up LoadExtActions, TruncStoreActions, and IndexedModeAct...
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | 80 col violation.
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Add comment about the pseudo registers QQ, each of...
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Bring back VLD1q and VST1q and use them for reloading...
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Expand VMOVQQ into a pair of VMOVQ.
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | Mark some pattern-less instructions as neverHasSideEffects.
|
commit | commitdiff | tree |
2010-05-13 |
Evan Cheng | If REG_SEQUENCE source is livein, copy it first. Also...
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | Do not attempt copy coalescing if the source and dest...
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | Fix some potential issues in the pseudo instruction...
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | Remove a dead fixme.
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | vst instructions are modeled as this:
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | Teach local regalloc about virtual registers with sub...
|
commit | commitdiff | tree |
2010-05-12 |
Evan Cheng | Code clean up.
|
commit | commitdiff | tree |
2010-05-11 |
Evan Cheng | Avoid breaking vstd when reg_sequence is not used.
|
commit | commitdiff | tree |
2010-05-11 |
Evan Cheng | Select @llvm.trap to the special B with 1111 condition...
|
commit | commitdiff | tree |
2010-05-11 |
Evan Cheng | Model some vst3 and vst4 with reg_sequence.
|
commit | commitdiff | tree |
2010-05-11 |
Evan Cheng | It's not safe eliminate copies where src and dst have...
|
commit | commitdiff | tree |
2010-05-11 |
Evan Cheng | Ensure REG_SEQUENCE source operands are unique.
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Indentation.
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Model some vld3 instructions with REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | It's not safe to propagate implicit_def that defines...
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Clear RegSequences vector after eliminating REG_SEQUENCE...
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Be careful with operand promotion. For a binary operation...
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Model vld2 / vst2 with reg_sequence.
|
commit | commitdiff | tree |
2010-05-10 |
Evan Cheng | Re-defined valno is always valno even for partial re...
|
commit | commitdiff | tree |
2010-05-07 |
Evan Cheng | Use VLD2q32 / VST2q32 to reload / spill QQ (pair of...
|
commit | commitdiff | tree |
2010-05-07 |
Evan Cheng | Use VSTMD / VLDMD for spills and reloads of Q registers...
|
commit | commitdiff | tree |
2010-05-07 |
Evan Cheng | Remove VLD1q and VST1q for reloading and spilling Q...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Add argument TargetRegisterInfo to loadRegFromStackSlot...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | 80 col violation.
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Re-apply 103156 and 103157. 103156 didn't break anything...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Fixes a coalescer bug that caused llc to crash on 2009...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Fix an obvious bug in isMoveInstr. It needs to return...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Adding pseudo 256-bit registers QQ0 . . . QQ7 to represent...
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | Cosmetic changes.
|
commit | commitdiff | tree |
2010-05-06 |
Evan Cheng | storeRegToStackSlot has forgotten about QPR_8 register...
|
commit | commitdiff | tree |
2010-05-05 |
Evan Cheng | Do not pre-allocate references of D registers pairs...
|
commit | commitdiff | tree |
2010-05-05 |
Evan Cheng | Move REG_SEQUENCE removal to 2addr pass.
|
commit | commitdiff | tree |
2010-05-05 |
Evan Cheng | Model CONCAT_VECTORS of two 64-bit values as a REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-05 |
Evan Cheng | Trim include.
|
commit | commitdiff | tree |
2010-05-05 |
Evan Cheng | Teach liveintervalanalysis about virtual registers...
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | With -neon-reg-sequence, models forming a Q register...
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | Do not pre-allocate for registers which form a REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | Teach PHI elimination to remove REG_SEQUENCE instructions...
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | Rename variables for consistency.
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | Instruction selection optimizations may have moved...
|
commit | commitdiff | tree |
2010-05-04 |
Evan Cheng | Teach scheduler about REG_SEQUENCE.
|
commit | commitdiff | tree |
2010-05-01 |
Evan Cheng | Add a pseudo instruction REG_SEQUENCE that takes a...
|
commit | commitdiff | tree |
2010-04-30 |
Evan Cheng | Fix test.
|
commit | commitdiff | tree |
2010-04-30 |
Evan Cheng | Another sibcall bug. If caller and callee calling convention...
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Remove DBG_VALUE which reference dead stack slots.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Add DenseMapInfo for int.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Add comment.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Re-enable 102565 with fixes.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Load folding tail call should not use ebp / rbp after...
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Temporarily disable my changes to unbreak the build.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Do not generate duplicate dbg_value instructions for...
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Avoid emitting a dbg_value machineinstr that's not...
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Frame index can be negative.
|
commit | commitdiff | tree |
2010-04-29 |
Evan Cheng | Check Reg against zero.
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | - Really preserve dbg_value instructions when the register...
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Replace r102368 with code that's less fragile. This...
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Pretty print DBG_VALUE machine instructions.
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Enable i16 to i32 promotion by default.
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Try operation promotion only if regular dag combine...
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Unbreak the build. Only form shld / shrd after legalization.
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Update tests.
|
commit | commitdiff | tree |
2010-04-28 |
Evan Cheng | Rather than having a ton of patterns for double shift...
|
commit | commitdiff | tree |
2010-04-27 |
Evan Cheng | Fix obvious typos.
|
commit | commitdiff | tree |
2010-04-27 |
Evan Cheng | SRA promotion is also not free.
|
commit | commitdiff | tree |
2010-04-27 |
Evan Cheng | - When legal, promote a load to zextload rather than...
|
commit | commitdiff | tree |
2010-04-27 |
Evan Cheng | Do not count kill, implicit_def instructions as printed...
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Insert dbg_value instructions for function entry block...
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Promoting 16-bit cmp / test aren't free. Don't do it.
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Re-enable 102323 with fix: do not update dbg_value...
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Temporary disable spiller modifying dbg_value. It's...
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Remove a redundant comment.
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Add PPC specific emitFrameIndexDebugValue.
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | Add ARM specific emitFrameIndexDebugValue.
|
commit | commitdiff | tree |
2010-04-26 |
Evan Cheng | - Move TargetLowering::EmitTargetCodeForFrameDebugValue...
|
commit | commitdiff | tree |
2010-04-24 |
Evan Cheng | Avoid promoting a i16 node if it would eliminate a...
|
commit | commitdiff | tree |
2010-04-24 |
Evan Cheng | When a load operand is promoted to an extload, replace...
|
commit | commitdiff | tree |
next |