| Rev |
Log message |
Author |
Age |
Path |
| 1024 |
part of points-to summary, not yet enabled |
syang |
2185d 22h |
/trunk/osprey/common/com/ |
| 1016 |
1.fix the regression of 447.deaII,see bug no 313, add a recal_liveness pass
after first pass of ebo before first pass of cflow
this also fix the latest regression of IPA problem for bootstrap compiler.
2.revert 1000 and 1001's checkin, it cause 453.provay's endless running time |
tianwei |
2188d 15h |
/trunk/osprey/common/com/ |
| 1005 |
optimization about __attribute__((noreturn)) semantic |
syang |
2191d 14h |
/trunk/osprey/common/com/ |
| 1000 |
Fix the pu_size estimation of a PU. And it also fix the size update while doing inlining. We can now disable the "operator" heuristic which is aimed at making the inlining of MatrixInversion work well. |
dehao |
2193d 01h |
/trunk/osprey/common/com/ |
| 997 |
1. ipo_inline_util.cxx for fixing inliner bug #277
2. ipa_cg.cxx for fix a bug when doing IPA tracing. Some other phases (e.g. IPA_Convert_Icalls) rely on the IPA_NODE's visited flag, so clear it after printing the trace
3. fix bug 296 (x86 targ has this fix, merge it to ia64) |
fengzhou |
2194d 11h |
/trunk/osprey/common/com/ |
| 987 |
This checkin are related to three aspects:
1). Modified the TF to XF when emulate long double, WORKAROUND the build error "unable to emulate 'TF'" using gcc-4.1.
XF and TF is related to long double support, 80-bits-extented or 128-bit IEEE float. The current IMPLEMENTATION on IA64 is F10, .r.e., 80-bits-extented, not 96 nor 128. So it doesn't matter that modify like this, right? I remember Robert
has also met this problem when building opencc on X8664.
2). Fix the bug 257#, related to overwrite betweeen iput arguments of the struct of long double.
3). Fix the attribute setting of ST, bootstrap problem. |
hucheng |
2197d 13h |
/trunk/osprey/common/com/ |
| 984 |
Clear STs during initialization.
Otherwise, random paddings in ST structs may make bcmp/memcmp miscompare. |
shiyao |
2198d 00h |
/trunk/osprey/common/com/ |
| 979 |
Fix for new type merge phase.
1. Fix a merge mistake that redefine local varible "tmp" in computing hash value;
2. Fix a merge mistake that does not use the new partial match in function Validate_Recursive_Type;
3. Set anonymous array in front end (just for GNU4);
4. Add process of KIND_ARRARY in function Is_Incomplete_Or_Recursive.
Now bug #251 is fixed. In 176.gcc, I confirm that rtunion_def and rtx_def are unified in global. |
shiyao |
2198d 09h |
/trunk/osprey/common/com/ |
| 962 |
Type merge in IPA phase has the bug that cannot merge incomplete recursive structs into one TY.
The bug doesn't cause error in general, but fails in C++ devirtualization that must unify the classes in differe files.
This patch fix the bug and it should be verified by devirtualization at last. |
shiyao |
2204d 04h |
/trunk/osprey/common/com/ |
| 956 |
transfter 832 833 834 835 928 from open3.1 to new trunk, these are shuxin's implict call operand's fix, this transfer
fix the running time error for 176.gcc at -O2 -gnu4(it's a problem when return value's register r8 is reallocated by lra). it's only related to ia64, I had enclosed them by ifdef. also test it on x8664 for spec2kint at -O2. |
tianwei |
2212d 09h |
/trunk/osprey/common/com/ |
| 944 |
disable rsqrt at any optimization level because it may cause precision problems. |
fengzhou |
2221d 14h |
/trunk/osprey/common/com/ |
| 942 |
this fix is try to disable the simplification for OP (SELECT(X,c1,c2)) ->
SELECT(X,OP(c1),OP(C2)) when OP is EXTRACT_BITS. in the modified function,it only has the operator info for EXTRACT_BITS, but no other additional info,such as "bofst,bsize", they are lost. this fix the running time regression for perlbench@2006 at -gnu4 |
tianwei |
2223d 14h |
/trunk/osprey/common/com/ |
| 941 |
The fixes are about prefetch whirl node generation.
One is to enhance the condition of testing whether a whirl node is the induction variable.
The other is to increase the default value of Prefetch_iters_ahead to 8. Its previous value, 2, is too small for some cases. |
fengzhou |
2225d 10h |
/trunk/osprey/common/com/ |
| 926 |
The 'merge' branch is now our new trunk. |
ributzka |
2232d 22h |
/trunk/osprey/common/com/ |
| 924 |
Move r894 from trunk. |
ributzka |
2232d 22h |
/trunk/osprey/common/com/ |
| 881 |
Fixed bug #249, tested with cperennial and cpppernnial. |
laijx |
2243d 18h |
/trunk/osprey/common/com/ |
| 877 |
transfer r784 r790 and r814 from trunk |
syang |
2244d 19h |
/trunk/osprey/common/com/ |
| 874 |
Remove the redundant intrinsic name. |
hucheng |
2245d 12h |
/trunk/osprey/common/com/ |
| 872 |
Fix the bug "get_intrinsic_name, INTRN_TAN, intrinsic names out of sync". |
hucheng |
2246d 07h |
/trunk/osprey/common/com/ |
| 869 |
transfer 782 and 783 from trunk. the r783 is fix to the bug in r782 |
syang |
2246d 14h |
/trunk/osprey/common/com/ |