oota-llvm.git
11 years agoAVX/AVX2: Move the SEXT lowering code from a target specific DAGco to a lowering...
Nadav Rotem [Thu, 27 Dec 2012 22:47:16 +0000 (22:47 +0000)]
AVX/AVX2: Move the SEXT lowering code from a target specific DAGco to a lowering function.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171170 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge basic_sse12_fp_binop_p and basic_sse12_fp_binop_p_y multiclasses.
Craig Topper [Thu, 27 Dec 2012 18:51:50 +0000 (18:51 +0000)]
Merge basic_sse12_fp_binop_p and basic_sse12_fp_binop_p_y multiclasses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171166 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd support to BasicBlocks for iterating backwards over the
Chandler Carruth [Thu, 27 Dec 2012 12:00:56 +0000 (12:00 +0000)]
Add support to BasicBlocks for iterating backwards over the
instructions. This just exposes the already present reverse iterators of
the instruction ilist.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171159 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoProvide a common half-open interval map info implementation, and just
Chandler Carruth [Thu, 27 Dec 2012 11:29:17 +0000 (11:29 +0000)]
Provide a common half-open interval map info implementation, and just
re-use that for SlotIndexes. This way other users who want half-open
semantics can share the implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171158 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMake this parameter be named consistently with most other
Chandler Carruth [Thu, 27 Dec 2012 11:17:15 +0000 (11:17 +0000)]
Make this parameter be named consistently with most other
getAnalysisUsage implementations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171157 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agodocs: Add FAQ about "storing to a virtual register".
Sean Silva [Thu, 27 Dec 2012 10:23:04 +0000 (10:23 +0000)]
docs: Add FAQ about "storing to a virtual register".

This came up for the N+1'st time today in IRC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171155 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agodocs: Move link to the new "external tutorials" area.
Sean Silva [Thu, 27 Dec 2012 08:57:08 +0000 (08:57 +0000)]
docs: Move link to the new "external tutorials" area.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171154 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[ASan] Fix lifetime intrinsics handling. Now for each intrinsic we check if it descri...
Alexey Samsonov [Thu, 27 Dec 2012 08:50:58 +0000 (08:50 +0000)]
[ASan] Fix lifetime intrinsics handling. Now for each intrinsic we check if it describes one of 'interesting' allocas. Assume that allocas can go through casts and phi-nodes before apperaring as llvm.lifetime arguments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171153 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDAGCombinerInformation: add a getter that exposes the dagcombine level.
Nadav Rotem [Thu, 27 Dec 2012 08:44:35 +0000 (08:44 +0000)]
DAGCombinerInformation: add a getter that exposes the dagcombine level.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171152 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix new[]/delete mismatch in FullDependence spotted by AddressSanitizer
Alexey Samsonov [Thu, 27 Dec 2012 08:40:37 +0000 (08:40 +0000)]
Fix new[]/delete mismatch in FullDependence spotted by AddressSanitizer

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171150 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agodocs: Update the benchmark with updated perf numbers.
Nadav Rotem [Thu, 27 Dec 2012 08:32:44 +0000 (08:32 +0000)]
docs: Update the benchmark with updated perf numbers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171149 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoOn AVX/AVX2 the type v8i1 is legalized to v8i16, which is an XMM sized
Nadav Rotem [Thu, 27 Dec 2012 08:15:45 +0000 (08:15 +0000)]
On AVX/AVX2 the type v8i1 is legalized to v8i16, which is an XMM sized
register. In most cases we actually compare or select YMM-sized registers
and mixing the two types creates horrible code. This commit optimizes
some of the transition sequences.

PR14657.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171148 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAVX/AVX2: Move the code that lowers vector-trunc from a DAGCo-hook to custom lowering...
Nadav Rotem [Thu, 27 Dec 2012 07:45:10 +0000 (07:45 +0000)]
AVX/AVX2: Move the code that lowers vector-trunc from a DAGCo-hook to custom lowering hook.
The vector truncs were scalarized during LegalizeVectorOps, later vectorized again by some DAGCombine optimization
and finally, lowered by a dagcombing optimization. Now, they are properly lowered during LegalizeVectorOps.
No new testcase because the original testcases still work.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171146 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd hasSideEffects=0 to some forms of ROUND, RCP, and RSQRT.
Craig Topper [Thu, 27 Dec 2012 07:16:08 +0000 (07:16 +0000)]
Add hasSideEffects=0 to some forms of ROUND, RCP, and RSQRT.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171143 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRefactor DAGCombinerInfo. Change the different booleans that indicate if we are befor...
Nadav Rotem [Thu, 27 Dec 2012 06:47:41 +0000 (06:47 +0000)]
Refactor DAGCombinerInfo. Change the different booleans that indicate if we are before or after different runs of DAGCo, with the CombineLevel enum.
Also, added a new API for checking if we are running before or after the LegalizeVectorOps phase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171142 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMove single letter 'P' prefix out of multiclass now that tablegen allows defm to...
Craig Topper [Thu, 27 Dec 2012 06:34:54 +0000 (06:34 +0000)]
Move single letter 'P' prefix out of multiclass now that tablegen allows defm to start with #NAME. This makes instruction names more searchable again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171141 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate tablegen parser to allow defm names to start with #NAME.
Craig Topper [Thu, 27 Dec 2012 06:32:52 +0000 (06:32 +0000)]
Update tablegen parser to allow defm names to start with #NAME.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171140 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd hasSideEffects=0 to some shift and rotate instructions. None of which are current...
Craig Topper [Thu, 27 Dec 2012 03:35:44 +0000 (03:35 +0000)]
Add hasSideEffects=0 to some shift and rotate instructions. None of which are currently used by code generation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171137 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark the divide instructions as hasSideEffects=0.
Craig Topper [Thu, 27 Dec 2012 03:01:18 +0000 (03:01 +0000)]
Mark the divide instructions as hasSideEffects=0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171136 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFor the dwarf5 split debug info code split out the string section
Eric Christopher [Thu, 27 Dec 2012 02:14:01 +0000 (02:14 +0000)]
For the dwarf5 split debug info code split out the string section
per compile unit/skeleton compile unit. Update tests accordingly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171133 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFileCheck-ize.
Eric Christopher [Thu, 27 Dec 2012 02:13:58 +0000 (02:13 +0000)]
FileCheck-ize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171132 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFileCheck-ize.
Eric Christopher [Thu, 27 Dec 2012 02:13:55 +0000 (02:13 +0000)]
FileCheck-ize.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171131 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd hasSideEffects=0 to CMP*rr_REV.
Craig Topper [Thu, 27 Dec 2012 02:08:46 +0000 (02:08 +0000)]
Add hasSideEffects=0 to CMP*rr_REV.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171130 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agowhitespace
Nadav Rotem [Thu, 27 Dec 2012 02:04:12 +0000 (02:04 +0000)]
whitespace

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171129 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd mayLoad, mayStore, and hasSideEffects tags to BT/BTS/BTR/BTC instructions. Should...
Craig Topper [Thu, 27 Dec 2012 02:01:33 +0000 (02:01 +0000)]
Add mayLoad, mayStore, and hasSideEffects tags to BT/BTS/BTR/BTC instructions. Shouldn't change any functionality since they don't have patterns to select them.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171128 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRight now all of the relocations are 32-bit dwarf, and the relocation
Eric Christopher [Thu, 27 Dec 2012 01:07:07 +0000 (01:07 +0000)]
Right now all of the relocations are 32-bit dwarf, and the relocation
information doesn't return an addend for Rel relocations. Go ahead
and use this information to fix relocation handling inside dwarfdump
for 32-bit ELF REL.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171126 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoIf all of the write objects are identified then we can vectorize the loop even if...
Nadav Rotem [Wed, 26 Dec 2012 23:30:53 +0000 (23:30 +0000)]
If all of the write objects are identified then we can vectorize the loop even if the read objects are unidentified.

PR14719.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171124 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix operands and encoding form for ARPL instruction. Register form had and reversed...
Craig Topper [Wed, 26 Dec 2012 23:27:57 +0000 (23:27 +0000)]
Fix operands and encoding form for ARPL instruction. Register form had  and  reversed. Memory form writes memory, but was marked as MRMSrcMem.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171123 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd hasSideEffects=0 to some atomic instructions.
Craig Topper [Wed, 26 Dec 2012 23:08:12 +0000 (23:08 +0000)]
Add hasSideEffects=0 to some atomic instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171122 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark the AL/AX/EAX forms of the basic arithmetic operations has never having side...
Craig Topper [Wed, 26 Dec 2012 22:19:23 +0000 (22:19 +0000)]
Mark the AL/AX/EAX forms of the basic arithmetic operations has never having side effects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171121 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago80 columns. No functionality change.
Nick Lewycky [Wed, 26 Dec 2012 22:00:49 +0000 (22:00 +0000)]
80 columns. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171120 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove mid-optimizer warning. This situation should be handled differently,
Nick Lewycky [Wed, 26 Dec 2012 22:00:35 +0000 (22:00 +0000)]
Remove mid-optimizer warning. This situation should be handled differently,
such as by a compiler warning, a check in clang -fsanitizer=undefined, being
optimized to unreachable, or a combination of the above. PR14722.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171119 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark all the _REV instructions as not having side effects. They aren't really emitted...
Craig Topper [Wed, 26 Dec 2012 21:30:22 +0000 (21:30 +0000)]
Mark all the _REV instructions as not having side effects. They aren't really emitted by the backend, but it reduces the number of instructions in the output files with unmodelled side effects to make auditing easier.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171118 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove a special conditional setting of neverHasSideEffects if the instruction didn...
Craig Topper [Wed, 26 Dec 2012 21:04:30 +0000 (21:04 +0000)]
Remove a special conditional setting of neverHasSideEffects if the instruction didn't have a pattern. This was leftover from when tablegen used to complain if things were already inferred from patterns.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171117 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the docs with the new workload that was added.
Nadav Rotem [Wed, 26 Dec 2012 19:45:00 +0000 (19:45 +0000)]
Update the docs with the new workload that was added.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171115 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: Optimize the vectorization of consecutive memory access when the...
Nadav Rotem [Wed, 26 Dec 2012 19:08:17 +0000 (19:08 +0000)]
LoopVectorizer: Optimize the vectorization of consecutive memory access when the iteration step is -1

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171114 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix comment typo
Eli Bendersky [Wed, 26 Dec 2012 18:15:42 +0000 (18:15 +0000)]
Fix comment typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171113 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Raise alignment of origin stores/loads when possible.
Evgeniy Stepanov [Wed, 26 Dec 2012 11:55:09 +0000 (11:55 +0000)]
[msan] Raise alignment of origin stores/loads when possible.

Origin alignment is as high as the alignment of the corresponding application
location, but never less than 4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171110 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Expand the file comment with track-origins info.
Evgeniy Stepanov [Wed, 26 Dec 2012 10:59:00 +0000 (10:59 +0000)]
[msan] Expand the file comment with track-origins info.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171109 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix quoting in configure. Patch by Krzysztof Parzyszek!
Benjamin Kramer [Wed, 26 Dec 2012 10:48:49 +0000 (10:48 +0000)]
Fix quoting in configure. Patch by Krzysztof Parzyszek!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171108 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge still more SSE/AVX instruction definitions.
Craig Topper [Wed, 26 Dec 2012 07:54:43 +0000 (07:54 +0000)]
Merge still more SSE/AVX instruction definitions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171103 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge more SSE/AVX instruction definitions.
Craig Topper [Wed, 26 Dec 2012 07:20:35 +0000 (07:20 +0000)]
Merge more SSE/AVX instruction definitions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171102 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoTableGen/FixedLenDecoderEmitter.cpp: Fix a potential mask overflow in fieldFromInstru...
NAKAMURA Takumi [Wed, 26 Dec 2012 06:43:14 +0000 (06:43 +0000)]
TableGen/FixedLenDecoderEmitter.cpp: Fix a potential mask overflow in fieldFromInstruction().

Reported by Yang Yongyong, thanks!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171101 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agorevert an accidental commit.
Nadav Rotem [Wed, 26 Dec 2012 06:16:03 +0000 (06:16 +0000)]
revert an accidental commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171098 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix 80 column violation.
Craig Topper [Wed, 26 Dec 2012 06:15:53 +0000 (06:15 +0000)]
Fix 80 column violation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171097 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix class name in comment.
Craig Topper [Wed, 26 Dec 2012 06:15:09 +0000 (06:15 +0000)]
Fix class name in comment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171096 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge SSE/AVX PCMPEQ/PCMPGT instruction definitions.
Craig Topper [Wed, 26 Dec 2012 06:14:15 +0000 (06:14 +0000)]
Merge SSE/AVX PCMPEQ/PCMPGT instruction definitions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171095 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDoc: add fmuladd to the list of vectorizeable functions. Thanks hfinkel.
Nadav Rotem [Wed, 26 Dec 2012 06:03:35 +0000 (06:03 +0000)]
Doc: add fmuladd to the list of vectorizeable functions. Thanks hfinkel.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171094 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove 'v' from mnemonic to fix asm matching failures.
Craig Topper [Wed, 26 Dec 2012 06:02:15 +0000 (06:02 +0000)]
Remove 'v' from mnemonic to fix asm matching failures.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171093 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions...
Craig Topper [Wed, 26 Dec 2012 05:49:15 +0000 (05:49 +0000)]
Use an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions for a bunch of SSE2 integer arithmetic instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171092 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoReformat the docs.
Nadav Rotem [Wed, 26 Dec 2012 04:59:20 +0000 (04:59 +0000)]
Reformat the docs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171091 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agowhite space
Nadav Rotem [Wed, 26 Dec 2012 04:58:12 +0000 (04:58 +0000)]
white space

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171090 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions...
Craig Topper [Wed, 26 Dec 2012 04:36:03 +0000 (04:36 +0000)]
Use an additional multiclass to merge the 128/256-bit SSE/AVX instruction definitions for PAND/POR/PXOR/PANDN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171087 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMerge an AVX/SSE 256-bit and 128-bit multiclass.
Craig Topper [Wed, 26 Dec 2012 03:56:47 +0000 (03:56 +0000)]
Merge an AVX/SSE 256-bit and 128-bit multiclass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171086 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoMark VANDNPD/VANDNPDS as not commutable.
Craig Topper [Wed, 26 Dec 2012 03:48:10 +0000 (03:48 +0000)]
Mark VANDNPD/VANDNPDS as not commutable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171085 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agollvm/test/CodeGen/X86: FileCheck-ize two tests in r171083.
NAKAMURA Takumi [Wed, 26 Dec 2012 03:19:30 +0000 (03:19 +0000)]
llvm/test/CodeGen/X86: FileCheck-ize two tests in r171083.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171084 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agollvm/test/CodeGen/X86: Disable avx in two tests corresponding to r171082.
NAKAMURA Takumi [Wed, 26 Dec 2012 03:08:55 +0000 (03:08 +0000)]
llvm/test/CodeGen/X86: Disable avx in two tests corresponding to r171082.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171083 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove alignment from a bunch more VEX encoded operations in the folding tables.
Craig Topper [Wed, 26 Dec 2012 02:44:47 +0000 (02:44 +0000)]
Remove alignment from a bunch more VEX encoded operations in the folding tables.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171082 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove alignment from folding table for VMOVUPD as an unaligned instruction it should...
Craig Topper [Wed, 26 Dec 2012 02:14:19 +0000 (02:14 +0000)]
Remove alignment from folding table for VMOVUPD as an unaligned instruction it shouldn't require alignment...

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171081 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove alignment requirements from (V)EXTRACTPS. This instruction does 32-bit stores...
Craig Topper [Wed, 26 Dec 2012 01:47:12 +0000 (01:47 +0000)]
Remove alignment requirements from (V)EXTRACTPS. This instruction does 32-bit stores which aren't required to be aligned on SSE or AVX.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171080 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoBBVectorize: Use VTTI to compute costs for intrinsics vectorization
Hal Finkel [Wed, 26 Dec 2012 01:36:57 +0000 (01:36 +0000)]
BBVectorize: Use VTTI to compute costs for intrinsics vectorization

For the time being this includes only some dummy test cases. Once the
generic implementation of the intrinsics cost function does something other
than assuming scalarization in all cases, or some target specializes the
interface, some real test cases can be added.

Also, for consistency, I changed the type of IID from unsigned to Intrinsic::ID
in a few other places.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171079 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemove alignment requirement from VCVTSS2SD in folding tables. Reverting r171049...
Craig Topper [Wed, 26 Dec 2012 00:35:47 +0000 (00:35 +0000)]
Remove alignment requirement from VCVTSS2SD in folding tables. Reverting r171049. This instruction doesn't require alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171078 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: Enable vectorization of the fmuladd intrinsic
Hal Finkel [Tue, 25 Dec 2012 23:21:29 +0000 (23:21 +0000)]
LoopVectorize: Enable vectorization of the fmuladd intrinsic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171076 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoBBVectorize: Enable vectorization of the fmuladd intrinsic
Hal Finkel [Tue, 25 Dec 2012 22:36:08 +0000 (22:36 +0000)]
BBVectorize: Enable vectorization of the fmuladd intrinsic

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171075 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoosen scheduling restrictions on the PPC dcbt intrinsic
Hal Finkel [Tue, 25 Dec 2012 18:51:18 +0000 (18:51 +0000)]
Loosen scheduling restrictions on the PPC dcbt intrinsic

As with the prefetch intrinsic to which it maps, simply have dcbt
marked as reading from and writing to its arguments instead of having
unmodeled side effects. While this might cause unwanted code motion
(because aliasing checks don't really capture cache-line sharing),
it is more important that prefetches in unrolled loops don't block
the scheduler from rearranging the unrolled loop body.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171073 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoExpand PPC64 atomic load and store
Hal Finkel [Tue, 25 Dec 2012 17:22:53 +0000 (17:22 +0000)]
Expand PPC64 atomic load and store

Use of store or load with the atomic specifier on 64-bit types would
cause instruction-selection failures. As with the 32-bit case, these
can use the default expansion in terms of cmp-and-swap.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171072 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Fix handling of vectors of pointers.
Evgeniy Stepanov [Tue, 25 Dec 2012 16:04:38 +0000 (16:04 +0000)]
[msan] Fix handling of vectors of pointers.

VectorType::getInteger() can not be used with them, because pointer size
depends on the target.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171070 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[msan] Fix handling of select with vector condition.
Evgeniy Stepanov [Tue, 25 Dec 2012 14:56:21 +0000 (14:56 +0000)]
[msan] Fix handling of select with vector condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171069 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoHarden test so it's not affected by changes to compare lowering.
Benjamin Kramer [Tue, 25 Dec 2012 13:23:23 +0000 (13:23 +0000)]
Harden test so it's not affected by changes to compare lowering.

This only failed on hosts that don't have SSE41.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171066 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoX86: Shave off one shuffle from the pcmpeqq sequence for SSE2 by making use of and...
Benjamin Kramer [Tue, 25 Dec 2012 13:09:08 +0000 (13:09 +0000)]
X86: Shave off one shuffle from the pcmpeqq sequence for SSE2 by making use of and commutativity.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171064 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoX86: Custom lower <2 x i64> eq and ne when SSE41 is not available.
Benjamin Kramer [Tue, 25 Dec 2012 12:54:19 +0000 (12:54 +0000)]
X86: Custom lower <2 x i64> eq and ne when SSE41 is not available.

pcmpeqd, pshufd, pshufd, pand is cheaper than unpack + cmpq, sbbq, cmpq, sbbq + pack.
Small speedup on loop-vectorized viterbi (-march=core2).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171063 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoASan: initialize callbacks from ASan module pass in a separate function for consistency
Alexey Samsonov [Tue, 25 Dec 2012 12:28:20 +0000 (12:28 +0000)]
ASan: initialize callbacks from ASan module pass in a separate function for consistency

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171061 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoASan: move stack poisoning logic into FunctionStackPoisoner struct
Alexey Samsonov [Tue, 25 Dec 2012 12:04:36 +0000 (12:04 +0000)]
ASan: move stack poisoning logic into FunctionStackPoisoner struct

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171060 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix whitespace. No functionality change.
Nick Lewycky [Tue, 25 Dec 2012 06:13:25 +0000 (06:13 +0000)]
Fix whitespace. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171051 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoVCVTSS2SD requires a strict alignment. Thanks Elena.
Nadav Rotem [Tue, 25 Dec 2012 03:29:18 +0000 (03:29 +0000)]
VCVTSS2SD requires a strict alignment. Thanks Elena.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171049 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRename LLVMContext diagnostic handler types and functions.
Bob Wilson [Tue, 25 Dec 2012 00:07:12 +0000 (00:07 +0000)]
Rename LLVMContext diagnostic handler types and functions.

These are now generally used for all diagnostics from the backend, not just
for inline assembly, so this drops the "InlineAsm" from the names.  No
functional change.  (I've left aliases for the old names but only for long
enough to let me switch over clang to use the new ones.)

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171047 91177308-0d34-0410-b5e6-96231b3b80d8

11 years ago[CMake] AddLLVM.cmake: Tweak the corner case that "check-all" doesn't have any tests.
NAKAMURA Takumi [Mon, 24 Dec 2012 22:43:59 +0000 (22:43 +0000)]
[CMake] AddLLVM.cmake: Tweak the corner case that "check-all" doesn't have any tests.

"check-all" can be executed with 0 status, "check-all does nothing, no tools built."

LLVM_EXTERNAL_CLANG_BUILD=OFF LLVM_BUILD_TOOLS=OFF can reproduce this.
Oscar Fuentes reported this. Thank you.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171046 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoQuiet gcc's -Wparenthesis warning. No functionality change.
Nick Lewycky [Mon, 24 Dec 2012 19:58:45 +0000 (19:58 +0000)]
Quiet gcc's -Wparenthesis warning. No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171044 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix typo "Makre" -> "Make".
Nick Lewycky [Mon, 24 Dec 2012 19:55:47 +0000 (19:55 +0000)]
Fix typo "Makre" -> "Make".

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171043 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUse a std::string rather than a dynamically allocated char* buffer.
Benjamin Kramer [Mon, 24 Dec 2012 19:23:30 +0000 (19:23 +0000)]
Use a std::string rather than a dynamically allocated char* buffer.

This affords us to use std::string's allocation routines and use the destructor
for the memory management. Switching to that also means that we can use
operator==(const std::string&, const char *) to perform the string comparison
rather than resorting to libc functionality (i.e. strcmp).

Patch by Saleem Abdulrasool!

Differential Revision: http://llvm-reviews.chandlerc.com/D230

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171042 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdd LLVMContext::emitWarning methods and use them. <rdar://problem/12867368>
Bob Wilson [Mon, 24 Dec 2012 18:15:21 +0000 (18:15 +0000)]
Add LLVMContext::emitWarning methods and use them. <rdar://problem/12867368>

When the backend is used from clang, it should produce proper diagnostics
instead of just printing messages to errs(). Other clients may also want to
register their own error handlers with the LLVMContext, and the same handler
should work for warnings in the same way as the existing emitError methods.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171041 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix a typo introduced in r168577: FlAGS -> FLAGS (note the lowercase ell)
Dmitri Gribenko [Mon, 24 Dec 2012 17:52:48 +0000 (17:52 +0000)]
Fix a typo introduced in r168577: FlAGS -> FLAGS (note the lowercase ell)

Now we really pass -Wcovered-switch-default if the compiler supports it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171040 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAutoRegen.sh: update reference to documentation
Dmitri Gribenko [Mon, 24 Dec 2012 15:01:59 +0000 (15:01 +0000)]
AutoRegen.sh: update reference to documentation

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171037 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agollvm/test/CodeGen/X86/fold-vex.ll: Add explicit triple.
NAKAMURA Takumi [Mon, 24 Dec 2012 11:14:06 +0000 (11:14 +0000)]
llvm/test/CodeGen/X86/fold-vex.ll: Add explicit triple.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171029 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCostModel: We have API for checking the costs of known shuffles. This patch adds
Nadav Rotem [Mon, 24 Dec 2012 10:04:03 +0000 (10:04 +0000)]
CostModel: We have API for checking the costs of known shuffles. This patch adds
support for the insert-subvector and extract-subvector kinds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171027 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoAdded 6 more value types: v32i1, v64i1, v32i16, v32i8, v64i8, v8f64
Elena Demikhovsky [Mon, 24 Dec 2012 10:03:57 +0000 (10:03 +0000)]
Added 6 more value types: v32i1, v64i1, v32i16, v32i8, v64i8, v8f64

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171026 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoRemoved "static" from "__jit_debug_descriptor" because "static" adds C++ mangling...
Elena Demikhovsky [Mon, 24 Dec 2012 09:42:27 +0000 (09:42 +0000)]
Removed "static" from "__jit_debug_descriptor" because "static" adds C++ mangling prefix to this symbol.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171025 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoSome x86 instructions can load/store one of the operands to memory. On SSE, this...
Nadav Rotem [Mon, 24 Dec 2012 09:40:33 +0000 (09:40 +0000)]
Some x86 instructions can load/store one of the operands to memory. On SSE, this memory needs to be aligned.
When these instructions are encoded in VEX (on AVX) there is no such requirement. This changes the folding
tables and removes the alignment restrictions from VEX-encoded instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171024 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: When checking for vectorizable types, also check
Nadav Rotem [Mon, 24 Dec 2012 09:14:18 +0000 (09:14 +0000)]
LoopVectorizer: When checking for vectorizable types, also check
the StoreInst operands.

PR14705.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171023 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoChange the codegen Cost Model API for shuffeles. This patch removes the API for broad...
Nadav Rotem [Mon, 24 Dec 2012 08:57:47 +0000 (08:57 +0000)]
Change the codegen Cost Model API for shuffeles. This patch removes the API for broadcast and adds a more general API that accepts an enum of known shuffles.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171022 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoFix typo in comments
Alexey Samsonov [Mon, 24 Dec 2012 08:52:53 +0000 (08:52 +0000)]
Fix typo in comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171021 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoUpdate the docs of the cost model.
Nadav Rotem [Mon, 24 Dec 2012 05:51:12 +0000 (05:51 +0000)]
Update the docs of the cost model.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171016 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agollvm/MC/MCMachObjectWriter.h: ComputeSymbolTable(): Prune one description in the...
NAKAMURA Takumi [Mon, 24 Dec 2012 01:24:04 +0000 (01:24 +0000)]
llvm/MC/MCMachObjectWriter.h: ComputeSymbolTable(): Prune one description in the comment. [-Wdocumentation]

/// \param StringIndexMap [out] - Map from symbol names to offsets in the string table.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171010 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorizer: Fix an endless loop in the code that looks for reductions.
Nadav Rotem [Mon, 24 Dec 2012 01:22:06 +0000 (01:22 +0000)]
LoopVectorizer: Fix an endless loop in the code that looks for reductions.
The bug was in the code that detects PHIs in if-then-else block sequence.

PR14701.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171008 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoDocumentation: fix typos reported in PR13866
Dmitri Gribenko [Sun, 23 Dec 2012 18:46:11 +0000 (18:46 +0000)]
Documentation: fix typos reported in PR13866

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171006 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoCostModel: Change the default target-independent implementation for finding
Nadav Rotem [Sun, 23 Dec 2012 17:31:23 +0000 (17:31 +0000)]
CostModel: Change the default target-independent implementation for finding
the cost of arithmetic functions. We now assume that the cost of arithmetic
operations that are marked as Legal or Promote is low, but ops that are
marked as custom are higher.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171002 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: Fix accidentaly inverted condition.
Benjamin Kramer [Sun, 23 Dec 2012 13:21:41 +0000 (13:21 +0000)]
LoopVectorize: Fix accidentaly inverted condition.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171001 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoLoopVectorize: For scalars and void types there is no need to compute vector insert...
Benjamin Kramer [Sun, 23 Dec 2012 13:19:18 +0000 (13:19 +0000)]
LoopVectorize: For scalars and void types there is no need to compute vector insert/extract costs.

Fixes an assert during the build of oggenc in the test suite.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171000 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agoWe are not ready to estimate the cost of integer expansions based on the number of...
Nadav Rotem [Sun, 23 Dec 2012 09:11:07 +0000 (09:11 +0000)]
We are not ready to estimate the cost of integer expansions based on the number of parts. This test is too noisy.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170999 91177308-0d34-0410-b5e6-96231b3b80d8

11 years agodocs: Add link to external LLVM backend tutorial.
Sean Silva [Sun, 23 Dec 2012 07:34:51 +0000 (07:34 +0000)]
docs: Add link to external LLVM backend tutorial.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170998 91177308-0d34-0410-b5e6-96231b3b80d8