The original code does not work because the value from WorkList.end() is
invalidated once WorkList.erase() is called. To ensure proper functionality,
we must ensure that WorkList.erase() is always called before WorkList.end().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7673 91177308-0d34-0410-b5e6-96231b3b80d8
just PreSelection
* Wrapped code at 80 columns
* Added the DecomposeMultiDimRefs Pass to the JIT compilation path
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7657 91177308-0d34-0410-b5e6-96231b3b80d8
because saving i1 and i2 to their ``designated'' stack slots corrupts unknown
memory in other functions, standard libraries, and worse.
In addition, this has the benefit of improving JIT performance because we
eliminate writing out 4 instructions in CompilationCallback() and 2 loads and 2
stores.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7653 91177308-0d34-0410-b5e6-96231b3b80d8
Since we are including Makefile.test, we automatically get Makefile.common.
Furthermore, the double inclusion of Makefile.common causes the test suite to
be executed twice per invocation of the top level make.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7652 91177308-0d34-0410-b5e6-96231b3b80d8
2. Handle fp-to-uint conversions directly here instead of relying on
a pre-transformation to replace them with the 2-step conversion.
3. Use size rather than explicitly checking types when deciding what
opcodes to use, wherever possible. This is less error prone (the
bug fix above was not the first time!).
4. Float-to-pointer casts shd now work though this hasn't been tested.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7645 91177308-0d34-0410-b5e6-96231b3b80d8
this is not an optional transformation on SPARC and is now handled
directly by instruction selection.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7644 91177308-0d34-0410-b5e6-96231b3b80d8
* Doxygen-ified comments
* Added capability to make far calls (i.e., beyond 30 bits in CALL instr)
which implies that we need to delete function references that were added by
the call to addFunctionReference() because the actual call instruction is 10
instructions away (thanks to 64-bit address construction)
* Cleaned up code that generates far jumps by using an array+loop
SparcV9CodeEmitter.h:
* Explained more of the side-effects of emitFarCall()
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@7639 91177308-0d34-0410-b5e6-96231b3b80d8
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