Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

AArch64: Implement Vector Population Count evaluator #7026

Merged
merged 1 commit into from
Jun 7, 2023
Merged

AArch64: Implement Vector Population Count evaluator #7026

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 2 additions & 0 deletions compiler/aarch64/codegen/OMRCodeGenerator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -700,6 +700,8 @@ bool OMR::ARM64::CodeGenerator::getSupportsOpCodeForAutoSIMD(TR::CPU *cpu, TR::I
case TR::vmushr:
case TR::vrol:
case TR::vmrol:
case TR::vpopcnt:
case TR::vmpopcnt:
// Float/ Double are not supported
return (et == TR::Int8 || et == TR::Int16 || et == TR::Int32 || et == TR::Int64);
case TR::vload:
Expand Down
87 changes: 74 additions & 13 deletions compiler/aarch64/codegen/OMRTreeEvaluator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3146,16 +3146,8 @@ OMR::ARM64::TreeEvaluator::inlineVectorMaskedBinaryOp(TR::Node *node, TR::CodeGe
return resReg;
}

/**
* @brief Helper functions for generating instruction sequence for masked unary operations
*
* @param[in] node: node
* @param[in] cg: CodeGenerator
* @param[in] op: unary opcode
* @return vector register containing the result
*/
static TR::Register *
inlineVectorMaskedUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op)
TR::Register *
OMR::ARM64::TreeEvaluator::inlineVectorMaskedUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op, unaryEvaluatorHelper evaluatorHelper)
{
TR::Node *firstChild = node->getFirstChild();
TR::Node *secondChild = node->getSecondChild();
Expand All @@ -3167,7 +3159,15 @@ inlineVectorMaskedUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode:
TR::Register *resReg = cg->allocateRegister(TR_VRF);
node->setRegister(resReg);

generateTrg1Src1Instruction(cg, op, node, resReg, srcReg);
TR_ASSERT_FATAL_WITH_NODE(node, (op != TR::InstOpCode::bad) || (evaluatorHelper != NULL), "If op is TR::InstOpCode::bad, evaluatorHelper must not be NULL");
if (evaluatorHelper != NULL)
{
(*evaluatorHelper)(node, resReg, srcReg, cg);
}
else
{
generateTrg1Src1Instruction(cg, op, node, resReg, srcReg);
}

bool flipMask = false;
TR::Register *maskReg = evaluateMaskNode(secondChild, flipMask, cg);
Expand Down Expand Up @@ -4115,16 +4115,77 @@ OMR::ARM64::TreeEvaluator::vmfirstNonZeroEvaluator(TR::Node *node, TR::CodeGener
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
}

/**
* @brief Helper function for vector population count operation
*
* @param[in] node: node
* @param[in] resultReg: the result register
* @param[in] srcReg: the argument register
* @param[in] cg: CodeGenerator
* @return the result register
*/
static TR::Register *
vpopcntEvaluatorHelper(TR::Node *node, TR::Register *resultReg, TR::Register *srcReg, TR::CodeGenerator *cg)
{
TR::DataType et = node->getDataType().getVectorElementType();

generateTrg1Src1Instruction(cg, TR::InstOpCode::vcnt16b, node, resultReg, srcReg);
generateTrg1Src1Instruction(cg, TR::InstOpCode::vuaddlp16b, node, resultReg, resultReg);
if (et != TR::Int16)
{
generateTrg1Src1Instruction(cg, TR::InstOpCode::vuaddlp8h, node, resultReg, resultReg);
if (et == TR::Int64)
{
generateTrg1Src1Instruction(cg, TR::InstOpCode::vuaddlp4s, node, resultReg, resultReg);
}
}
return resultReg;
}

TR::Register*
OMR::ARM64::TreeEvaluator::vpopcntEvaluator(TR::Node *node, TR::CodeGenerator *cg)
{
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
TR::InstOpCode::Mnemonic op = TR::InstOpCode::bad;
unaryEvaluatorHelper evaluationHelper = NULL;
switch (node->getDataType().getVectorElementType())
{
case TR::Int8:
op = TR::InstOpCode::vcnt16b;
break;
case TR::Int16:
case TR::Int32:
case TR::Int64:
evaluationHelper = vpopcntEvaluatorHelper;
break;

default:
TR_ASSERT_FATAL_WITH_NODE(node, false, "unrecognized vector type %s", node->getDataType().toString());
return NULL;
}
return inlineVectorUnaryOp(node, cg, op, evaluationHelper);
}

TR::Register*
OMR::ARM64::TreeEvaluator::vmpopcntEvaluator(TR::Node *node, TR::CodeGenerator *cg)
{
return TR::TreeEvaluator::unImpOpEvaluator(node, cg);
TR::InstOpCode::Mnemonic op = TR::InstOpCode::bad;
unaryEvaluatorHelper evaluationHelper = NULL;
switch (node->getDataType().getVectorElementType())
{
case TR::Int8:
op = TR::InstOpCode::vcnt16b;
break;
case TR::Int16:
case TR::Int32:
case TR::Int64:
evaluationHelper = vpopcntEvaluatorHelper;
break;

default:
TR_ASSERT_FATAL_WITH_NODE(node, false, "unrecognized vector type %s", node->getDataType().toString());
return NULL;
}
return inlineVectorMaskedUnaryOp(node, cg, op, evaluationHelper);
}

TR::Register*
Expand Down
23 changes: 23 additions & 0 deletions compiler/aarch64/codegen/OMRTreeEvaluator.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -438,6 +438,29 @@ class OMR_EXTENSIBLE TreeEvaluator: public OMR::TreeEvaluator
*/
static TR::Register *inlineVectorMaskedBinaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op, binaryEvaluatorHelper evaluatorHelper = NULL);

typedef TR::Register *(*unaryEvaluatorHelper)(TR::Node *node, TR::Register *resReg, TR::Register *srcReg, TR::CodeGenerator *cg);
/**
* @brief Helper function for generating instruction sequence for unary operations
*
* @param[in] node: node
* @param[in] cg: CodeGenerator
* @param[in] op: unary opcode
* @param[in] evaluatorHelper: optional pointer to helper function which generates instruction stream for operation
* @return vector register containing the result
*/
static TR::Register *inlineVectorUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op, unaryEvaluatorHelper evaluatorHelper = NULL);

/**
* @brief Helper function for generating instruction sequence for masked unary operations
*
* @param[in] node: node
* @param[in] cg: CodeGenerator
* @param[in] op: unary opcode
* @param[in] evaluatorHelper: optional pointer to helper function which generates instruction stream for operation
* @return vector register containing the result
*/
static TR::Register *inlineVectorMaskedUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op, unaryEvaluatorHelper evaluatorHelper = NULL);

static TR::Register *f2iuEvaluator(TR::Node *node, TR::CodeGenerator *cg);
static TR::Register *f2luEvaluator(TR::Node *node, TR::CodeGenerator *cg);
static TR::Register *f2buEvaluator(TR::Node *node, TR::CodeGenerator *cg);
Expand Down
14 changes: 11 additions & 3 deletions compiler/aarch64/codegen/UnaryEvaluator.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -116,14 +116,22 @@ TR::Register *OMR::ARM64::TreeEvaluator::lnegEvaluator(TR::Node *node, TR::CodeG
return tempReg;
}

static TR::Register *inlineVectorUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op)
TR::Register *OMR::ARM64::TreeEvaluator::inlineVectorUnaryOp(TR::Node *node, TR::CodeGenerator *cg, TR::InstOpCode::Mnemonic op, unaryEvaluatorHelper evaluatorHelper)
{
TR::Node *firstChild = node->getFirstChild();
TR::Register *srcReg = cg->evaluate(firstChild);
TR::Register *resReg = (firstChild->getReferenceCount() == 1) ? srcReg : cg->allocateRegister(TR_VRF);
knn-k marked this conversation as resolved.
Show resolved Hide resolved
TR::Register *resReg = cg->allocateRegister(TR_VRF);

node->setRegister(resReg);
generateTrg1Src1Instruction(cg, op, node, resReg, srcReg);
TR_ASSERT_FATAL_WITH_NODE(node, (op != TR::InstOpCode::bad) || (evaluatorHelper != NULL), "If op is TR::InstOpCode::bad, evaluatorHelper must not be NULL");
if (evaluatorHelper != NULL)
{
(*evaluatorHelper)(node, resReg, srcReg, cg);
}
else
{
generateTrg1Src1Instruction(cg, op, node, resReg, srcReg);
}
cg->decReferenceCount(firstChild);
return resReg;
}
Expand Down