Skip to content

Commit b9b1f48

Browse files
Spencer-CominSpencer-Comin
committed
Add bit compress and expand IL opcdes
This commit adds opcodes for bit compress compress and expand operations (alternatively referred to as gather/scatter or extract/deposit) for byte, short, int, and long datatypes. Codegen support for Power and x86-64 is implemented with the pdepd/pextd and pdep/pext instructions, respectively. Signed-off-by: Spencer Comin <[email protected]>
1 parent c318529 commit b9b1f48

26 files changed

+860
-4
lines changed

‎compiler/aarch64/codegen/OMRTreeEvaluator.cpp

+48
Original file line numberDiff line numberDiff line change
@@ -5541,6 +5541,54 @@ OMR::ARM64::TreeEvaluator::lbitpermuteEvaluator(TR::Node *node, TR::CodeGenerato
55415541
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
55425542
}
55435543

5544+
TR::Register*
5545+
OMR::ARM64::TreeEvaluator::bcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5546+
{
5547+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5548+
}
5549+
5550+
TR::Register*
5551+
OMR::ARM64::TreeEvaluator::scompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5552+
{
5553+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5554+
}
5555+
5556+
TR::Register*
5557+
OMR::ARM64::TreeEvaluator::icompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5558+
{
5559+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5560+
}
5561+
5562+
TR::Register*
5563+
OMR::ARM64::TreeEvaluator::lcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5564+
{
5565+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5566+
}
5567+
5568+
TR::Register*
5569+
OMR::ARM64::TreeEvaluator::bexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5570+
{
5571+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5572+
}
5573+
5574+
TR::Register*
5575+
OMR::ARM64::TreeEvaluator::sexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5576+
{
5577+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5578+
}
5579+
5580+
TR::Register*
5581+
OMR::ARM64::TreeEvaluator::iexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5582+
{
5583+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5584+
}
5585+
5586+
TR::Register*
5587+
OMR::ARM64::TreeEvaluator::lexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
5588+
{
5589+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
5590+
}
5591+
55445592
TR::Instruction *loadAddressConstantInSnippet(TR::CodeGenerator *cg, TR::Node *node, intptr_t address, TR::Register *targetRegister, TR_ExternalRelocationTargetKind reloKind, bool isClassUnloadingConst, TR::Instruction *cursor)
55455593
{
55465594
TR::Compilation *comp = cg->comp();

‎compiler/aarch64/codegen/OMRTreeEvaluator.hpp

+8
Original file line numberDiff line numberDiff line change
@@ -569,6 +569,14 @@ class OMR_EXTENSIBLE TreeEvaluator: public OMR::TreeEvaluator
569569
static TR::Register *sbitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
570570
static TR::Register *ibitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
571571
static TR::Register *lbitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
572+
static TR::Register *bcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
573+
static TR::Register *scompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
574+
static TR::Register *icompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
575+
static TR::Register *lcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
576+
static TR::Register *bexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
577+
static TR::Register *sexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
578+
static TR::Register *iexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
579+
static TR::Register *lexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
572580

573581
static TR::Register *unImpOpEvaluator(TR::Node *node, TR::CodeGenerator *cg);
574582
static TR::Register *badILOpEvaluator(TR::Node *node, TR::CodeGenerator *cg);

‎compiler/arm/codegen/OMRTreeEvaluator.cpp

+48
Original file line numberDiff line numberDiff line change
@@ -2847,6 +2847,54 @@ OMR::ARM::TreeEvaluator::lbitpermuteEvaluator(TR::Node *node, TR::CodeGenerator
28472847
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
28482848
}
28492849

2850+
TR::Register*
2851+
OMR::ARM::TreeEvaluator::bcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2852+
{
2853+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2854+
}
2855+
2856+
TR::Register*
2857+
OMR::ARM::TreeEvaluator::scompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2858+
{
2859+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2860+
}
2861+
2862+
TR::Register*
2863+
OMR::ARM::TreeEvaluator::icompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2864+
{
2865+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2866+
}
2867+
2868+
TR::Register*
2869+
OMR::ARM::TreeEvaluator::lcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2870+
{
2871+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2872+
}
2873+
2874+
TR::Register*
2875+
OMR::ARM::TreeEvaluator::bexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2876+
{
2877+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2878+
}
2879+
2880+
TR::Register*
2881+
OMR::ARM::TreeEvaluator::sexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2882+
{
2883+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2884+
}
2885+
2886+
TR::Register*
2887+
OMR::ARM::TreeEvaluator::iexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2888+
{
2889+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2890+
}
2891+
2892+
TR::Register*
2893+
OMR::ARM::TreeEvaluator::lexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg)
2894+
{
2895+
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
2896+
}
2897+
28502898
TR::Register*
28512899
OMR::ARM::TreeEvaluator::PrefetchEvaluator(TR::Node *node, TR::CodeGenerator *cg)
28522900
{

‎compiler/arm/codegen/OMRTreeEvaluator.hpp

+8
Original file line numberDiff line numberDiff line change
@@ -533,6 +533,14 @@ class OMR_EXTENSIBLE TreeEvaluator: public OMR::TreeEvaluator
533533
static TR::Register *sbitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
534534
static TR::Register *ibitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
535535
static TR::Register *lbitpermuteEvaluator(TR::Node *node, TR::CodeGenerator *cg);
536+
static TR::Register *bcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
537+
static TR::Register *scompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
538+
static TR::Register *icompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
539+
static TR::Register *lcompressbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
540+
static TR::Register *bexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
541+
static TR::Register *sexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
542+
static TR::Register *iexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
543+
static TR::Register *lexpandbitsEvaluator(TR::Node *node, TR::CodeGenerator *cg);
536544
static TR::Register *PrefetchEvaluator(TR::Node *node, TR::CodeGenerator *cg);
537545

538546
static TR::Register *badILOpEvaluator(TR::Node *node, TR::CodeGenerator *cg);

‎compiler/env/OMRCPU.hpp

+4
Original file line numberDiff line numberDiff line change
@@ -130,6 +130,10 @@ class OMR_EXTENSIBLE CPU
130130
bool getSupportsHardwareRound() { return false; }
131131
bool getSupportsHardwareCopySign() { return false; }
132132
bool hasPopulationCountInstruction() { return false; }
133+
bool getSupportsHardware32bitCompress() { return false; }
134+
bool getSupportsHardware64bitCompress() { return false; }
135+
bool getSupportsHardware32bitExpand() { return false; }
136+
bool getSupportsHardware64bitExpand() { return false; }
133137
bool supportsDecimalFloatingPoint() { return false; }
134138
bool hasFPU() { return true; }
135139

‎compiler/il/OMRILOps.hpp

+12
Original file line numberDiff line numberDiff line change
@@ -1731,6 +1731,18 @@ class ILOpCode
17311731
case TR::lnolz:
17321732
vectorOperation = TR::vnolz;
17331733
break;
1734+
case TR::bcompressbits:
1735+
case TR::scompressbits:
1736+
case TR::icompressbits:
1737+
case TR::lcompressbits:
1738+
vectorOperation = TR::vcompressbits;
1739+
break;
1740+
case TR::bexpandbits:
1741+
case TR::sexpandbits:
1742+
case TR::iexpandbits:
1743+
case TR::lexpandbits:
1744+
vectorOperation = TR::vexpandbits;
1745+
break;
17341746
case TR::sbyteswap:
17351747
case TR::ibyteswap:
17361748
case TR::lbyteswap:

‎compiler/il/OMROpcodes.enum

+128
Original file line numberDiff line numberDiff line change
@@ -8229,6 +8229,134 @@ OPCODE_MACRO(\
82298229
/* .ifCompareOpCode = */ TR::BadILOp, \
82308230
/* .description = */ \
82318231
)
8232+
OPCODE_MACRO(\
8233+
/* .operation = */ bcompressbits, \
8234+
/* .name = */ "bcompressbits", \
8235+
/* .properties1 = */ 0, \
8236+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8237+
/* .properties3 = */ 0, \
8238+
/* .properties4 = */ 0, \
8239+
/* .dataType = */ TR::Int8, \
8240+
/* .typeProperties = */ ILTypeProp::Size_1 | ILTypeProp::Integer, \
8241+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int8), \
8242+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8243+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8244+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8245+
/* .ifCompareOpCode = */ TR::BadILOp, \
8246+
/* .description = compress bits from child1 selected by 1 bits in child2 into low bits of result */
8247+
)
8248+
OPCODE_MACRO(\
8249+
/* .operation = */ scompressbits, \
8250+
/* .name = */ "scompressbits", \
8251+
/* .properties1 = */ 0, \
8252+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8253+
/* .properties3 = */ 0, \
8254+
/* .properties4 = */ 0, \
8255+
/* .dataType = */ TR::Int16, \
8256+
/* .typeProperties = */ ILTypeProp::Size_2 | ILTypeProp::Integer, \
8257+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int16), \
8258+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8259+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8260+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8261+
/* .ifCompareOpCode = */ TR::BadILOp, \
8262+
/* .description = compress bits from child1 selected by 1 bits in child2 into low bits of result */
8263+
)
8264+
OPCODE_MACRO(\
8265+
/* .operation = */ icompressbits, \
8266+
/* .name = */ "icompressbits", \
8267+
/* .properties1 = */ 0, \
8268+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8269+
/* .properties3 = */ 0, \
8270+
/* .properties4 = */ 0, \
8271+
/* .dataType = */ TR::Int32, \
8272+
/* .typeProperties = */ ILTypeProp::Size_4 | ILTypeProp::Integer, \
8273+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int32), \
8274+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8275+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8276+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8277+
/* .ifCompareOpCode = */ TR::BadILOp, \
8278+
/* .description = compress bits from child1 selected by 1 bits in child2 into low bits of result */
8279+
)
8280+
OPCODE_MACRO(\
8281+
/* .operation = */ lcompressbits, \
8282+
/* .name = */ "lcompressbits", \
8283+
/* .properties1 = */ 0, \
8284+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8285+
/* .properties3 = */ 0, \
8286+
/* .properties4 = */ 0, \
8287+
/* .dataType = */ TR::Int64, \
8288+
/* .typeProperties = */ ILTypeProp::Size_8 | ILTypeProp::Integer, \
8289+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int64), \
8290+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8291+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8292+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8293+
/* .ifCompareOpCode = */ TR::BadILOp, \
8294+
/* .description = compress bits from child1 selected by 1 bits in child2 into low bits of result */
8295+
)
8296+
OPCODE_MACRO(\
8297+
/* .operation = */ bexpandbits, \
8298+
/* .name = */ "bexpandbits", \
8299+
/* .properties1 = */ 0, \
8300+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8301+
/* .properties3 = */ 0, \
8302+
/* .properties4 = */ 0, \
8303+
/* .dataType = */ TR::Int8, \
8304+
/* .typeProperties = */ ILTypeProp::Size_1 | ILTypeProp::Integer, \
8305+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int8), \
8306+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8307+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8308+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8309+
/* .ifCompareOpCode = */ TR::BadILOp, \
8310+
/* .description = starting from lowest bit, deposit bits from child1 into bits of result selected by 1 bits in child2 */
8311+
)
8312+
OPCODE_MACRO(\
8313+
/* .operation = */ sexpandbits, \
8314+
/* .name = */ "sexpandbits", \
8315+
/* .properties1 = */ 0, \
8316+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8317+
/* .properties3 = */ 0, \
8318+
/* .properties4 = */ 0, \
8319+
/* .dataType = */ TR::Int16, \
8320+
/* .typeProperties = */ ILTypeProp::Size_2 | ILTypeProp::Integer, \
8321+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int16), \
8322+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8323+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8324+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8325+
/* .ifCompareOpCode = */ TR::BadILOp, \
8326+
/* .description = starting from lowest bit, deposit bits from child1 into bits of result selected by 1 bits in child2 */
8327+
)
8328+
OPCODE_MACRO(\
8329+
/* .operation = */ iexpandbits, \
8330+
/* .name = */ "iexpandbits", \
8331+
/* .properties1 = */ 0, \
8332+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8333+
/* .properties3 = */ 0, \
8334+
/* .properties4 = */ 0, \
8335+
/* .dataType = */ TR::Int32, \
8336+
/* .typeProperties = */ ILTypeProp::Size_4 | ILTypeProp::Integer, \
8337+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int32), \
8338+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8339+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8340+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8341+
/* .ifCompareOpCode = */ TR::BadILOp, \
8342+
/* .description = starting from lowest bit, deposit bits from child1 into bits of result selected by 1 bits in child2 */
8343+
)
8344+
OPCODE_MACRO(\
8345+
/* .operation = */ lexpandbits, \
8346+
/* .name = */ "lexpandbits", \
8347+
/* .properties1 = */ 0, \
8348+
/* .properties2 = */ ILProp2::SupportedForPRE, \
8349+
/* .properties3 = */ 0, \
8350+
/* .properties4 = */ 0, \
8351+
/* .dataType = */ TR::Int64, \
8352+
/* .typeProperties = */ ILTypeProp::Size_8 | ILTypeProp::Integer, \
8353+
/* .childProperties = */ TWO_SAME_CHILD(TR::Int64), \
8354+
/* .swapChildrenOpCode = */ TR::BadILOp, \
8355+
/* .reverseBranchOpCode = */ TR::BadILOp, \
8356+
/* .booleanCompareOpCode = */ TR::BadILOp, \
8357+
/* .ifCompareOpCode = */ TR::BadILOp, \
8358+
/* .description = starting from lowest bit, deposit bits from child1 into bits of result selected by 1 bits in child2 */
8359+
)
82328360
OPCODE_MACRO(\
82338361
/* .opcode = */ sbyteswap, \
82348362
/* .name = */ "sbyteswap", \

‎compiler/optimizer/OMRSimplifierHandlers.cpp

+1
Original file line numberDiff line numberDiff line change
@@ -1652,6 +1652,7 @@ static bool reduceLongOp(TR::Node * node, TR::Block * block, TR::Simplifier * s,
16521652
case TR::land: newOp = TR::iand; break;
16531653
case TR::lor: newOp = TR::ior; break;
16541654
case TR::lxor: newOp = TR::ixor; break;
1655+
case TR::lexpandbits: newOp = TR::iexpandbits; break;
16551656
case TR::lushr:
16561657
isUnsigned = true;
16571658
case TR::lshr:

‎compiler/optimizer/OMRSimplifierTable.enum

+8
Original file line numberDiff line numberDiff line change
@@ -682,6 +682,14 @@
682682
#define sbitpermuteSimplifierHandler dftSimplifier
683683
#define ibitpermuteSimplifierHandler dftSimplifier
684684
#define lbitpermuteSimplifierHandler dftSimplifier
685+
#define lcompressbitsSimplifierHandler dftSimplifier
686+
#define icompressbitsSimplifierHandler dftSimplifier
687+
#define scompressbitsSimplifierHandler dftSimplifier
688+
#define bcompressbitsSimplifierHandler dftSimplifier
689+
#define lexpandbitsSimplifierHandler dftSimplifier
690+
#define iexpandbitsSimplifierHandler dftSimplifier
691+
#define sexpandbitsSimplifierHandler dftSimplifier
692+
#define bexpandbitsSimplifierHandler dftSimplifier
685693
#define PrefetchSimplifierHandler dftSimplifier
686694

687695
#endif

‎compiler/optimizer/ValuePropagationTable.hpp

+8
Original file line numberDiff line numberDiff line change
@@ -820,6 +820,14 @@ TR::Node * constrainLongBitCount(OMR::ValuePropagation *vp, TR::Node *node);
820820
#define sbitpermuteVPHandler constrainChildren
821821
#define ibitpermuteVPHandler constrainChildren
822822
#define lbitpermuteVPHandler constrainChildren
823+
#define lcompressbitsVPHandler constrainChildren
824+
#define icompressbitsVPHandler constrainChildren
825+
#define scompressbitsVPHandler constrainChildren
826+
#define bcompressbitsVPHandler constrainChildren
827+
#define lexpandbitsVPHandler constrainChildren
828+
#define iexpandbitsVPHandler constrainChildren
829+
#define sexpandbitsVPHandler constrainChildren
830+
#define bexpandbitsVPHandler constrainChildren
823831
#define PrefetchVPHandler constrainChildren
824832

825833
#ifdef J9_PROJECT_SPECIFIC

0 commit comments

Comments
 (0)