This substantially shrinks the size of each machine instruction, which should
make allocation faster and the cache footprint of the machine code lighter.
Here are some timings for code generation of the larger benchmarks we have.
This are timings of code generation phases of the X86 JIT, when compiled in
debug mode:
Before After Diff
164.gzip:
InstSel 0.0878 0.0722 -21.6%
RegAlloc 0.2031 0.1757 -15.6%
TOTAL 0.5585 0.4999 -11.7%
Ptrdist-bc:
InstSel 0.0878 0.0722 -21.6%
RegAlloc 0.2070 0.1933 - 7.1%
TOTAL 0.6972 0.6464 - 7.9%
197.parser:
InstSel 0.2148 0.2148 - 0.0%
RegAlloc 0.4941 0.4277 -15.5%
TOTAL 1.3749 1.2851 - 7.0%
175.vpr:
InstSel 0.2519 0.2109 -19.4%
RegAlloc 0.5976 0.5663 - 5.5%
TOTAL 1.6933 1.6347 - 3.5%
254.gap:
InstSel 1.1328 0.9921 -14.2%
RegAlloc 2.6933 2.4804 - 8.6%
TOTAL 7.7871 7.2499 - 7.4%
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7622 91177308-0d34-0410-b5e6-96231b3b80d8
so get rid of the def/use parameters that were getting passed in.
**** This now changes the semantics of these methods to preserve the flags,
not clobber them!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7602 91177308-0d34-0410-b5e6-96231b3b80d8
* Document the MOTy namespace correctly for doxygen
* Eliminate usage of the MachineOpCode typedef, which should eventually
be eliminated entirely.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@6584 91177308-0d34-0410-b5e6-96231b3b80d8
and related functions and flags. Fixed several bugs where only
"isDef" was being checked, not "isDefAndUse".
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@6342 91177308-0d34-0410-b5e6-96231b3b80d8
* Add new isPCRelative modifier flag which should be used in place of MO_PCRelativeDisp type.
* Fix a bug in isPhysicalRegister
* Add new setOpcode and RemoveOperand methods
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@5209 91177308-0d34-0410-b5e6-96231b3b80d8
MO_MachineRegister, we no longer distinguish Virtual vs. Machine registers
externally, they're ALL registers, all equal.
Registers are only differentiated whether they are >=
MRegisterInfo::FirstVirtual or not.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4823 91177308-0d34-0410-b5e6-96231b3b80d8
reading code much easier than just seeing "true, false" especially when
default parameters default one but not both arguments.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4717 91177308-0d34-0410-b5e6-96231b3b80d8
instead record them as extra operands in the operands[] vector.
Also, move CallArgsDescriptor into this class instead of making it an
annotation on the machine instruction.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4399 91177308-0d34-0410-b5e6-96231b3b80d8
by MachineInstr.
* Add a bunch of new methods to allow incremental addition of operands to the
machine instr instance.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4356 91177308-0d34-0410-b5e6-96231b3b80d8
- Inline methods that are mostly a single line anyway
- Eliminate several methods that were never called
- Group methods a bit more consistently
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4329 91177308-0d34-0410-b5e6-96231b3b80d8
overhead: Merge 3 parallel vectors into 1, change regsUsed hash_set to be a
bitvector. Sped up LLC a little less than 10% in a debug build!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@4261 91177308-0d34-0410-b5e6-96231b3b80d8
