Skip to content

JIT: Refine assert for RMW check in crc32 codegen #108750

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
+25 −21
Open

JIT: Refine assert for RMW check in crc32 codegen #108750

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b12fae9
JIT: Refine assert for RMW check in crc32 codegen
jakobbotsch Oct 10, 2024
0050cb2
Extract a function
jakobbotsch Dec 10, 2024
0313512
Add to assert, run jit-format
jakobbotsch Dec 10, 2024
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
1 change: 1 addition & 0 deletions src/coreclr/jit/codegen.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -786,6 +786,7 @@ class CodeGen final : public CodeGenInterface
#endif
void genCodeForTreeNode(GenTree* treeNode);
void genCodeForBinary(GenTreeOp* treeNode);
bool genIsSameLocalVar(GenTree* tree1, GenTree* tree2);

#if defined(TARGET_X86)
void genCodeForLongUMod(GenTreeOp* node);
Expand Down
42 changes: 22 additions & 20 deletions src/coreclr/jit/codegenxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1143,28 +1143,11 @@ void CodeGen::genCodeForBinary(GenTreeOp* treeNode)
// In order for this operation to be correct
// we need that op is a commutative operation so
// we can convert it into reg1 = reg1 op reg2 and emit
// the same code as above
// the same code as above. Or we need both operands to
// be the same local.
else if (op2reg == targetReg)
{

#ifdef DEBUG
unsigned lclNum1 = (unsigned)-1;
unsigned lclNum2 = (unsigned)-2;

GenTree* op1Skip = op1->gtSkipReloadOrCopy();
GenTree* op2Skip = op2->gtSkipReloadOrCopy();

if (op1Skip->OperIsLocalRead())
{
lclNum1 = op1Skip->AsLclVarCommon()->GetLclNum();
}
if (op2Skip->OperIsLocalRead())
{
lclNum2 = op2Skip->AsLclVarCommon()->GetLclNum();
}

assert(GenTree::OperIsCommutative(oper) || (lclNum1 == lclNum2));
#endif
assert(GenTree::OperIsCommutative(oper) || genIsSameLocalVar(op1, op2));

dst = op2;
src = op1;
Expand Down Expand Up @@ -1232,6 +1215,25 @@ void CodeGen::genCodeForBinary(GenTreeOp* treeNode)
genProduceReg(treeNode);
}

//------------------------------------------------------------------------
// genIsSameLocalVar:
// Check if two trees represent the same scalar local value.
//
// Arguments:
// op1 - first tree
// op2 - second tree
//
// Returns:
// True if so.
//
bool CodeGen::genIsSameLocalVar(GenTree* op1, GenTree* op2)
{
GenTree* op1Skip = op1->gtSkipReloadOrCopy();
GenTree* op2Skip = op1->gtSkipReloadOrCopy();
return op1Skip->OperIs(GT_LCL_VAR) && op2Skip->OperIs(GT_LCL_VAR) &&
(op1Skip->AsLclVar()->GetLclNum() == op2Skip->AsLclVar()->GetLclNum());
}

//------------------------------------------------------------------------
// genCodeForMul: Generate code for a MUL operation.
//
Expand Down
3 changes: 2 additions & 1 deletion src/coreclr/jit/hwintrinsiccodegenxarch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2240,7 +2240,8 @@ void CodeGen::genSSE42Intrinsic(GenTreeHWIntrinsic* node, insOpts instOptions)
{
assert(instOptions == INS_OPTS_NONE);

assert(!op2->isUsedFromReg() || (op2->GetRegNum() != targetReg) || (op1Reg == targetReg));
assert(!op2->isUsedFromReg() || (op2->GetRegNum() != targetReg) || (op1Reg == targetReg) ||
genIsSameLocalVar(op1, op2));
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Something that's been nagging the back of my mind...

Do we know why LSRA is assigning a single local var to two registers simultaneously? If it's the same local and its the same instruction/use, then it really should've ensured that op1Reg == op2Reg, no?

That is, this seems like a general deficiency or issue with assignments leading to poorer codegen and higher register pressure.

emit->emitIns_Mov(INS_mov, emitTypeSize(targetType), targetReg, op1Reg, /* canSkip */ true);

if ((baseType == TYP_UBYTE) || (baseType == TYP_USHORT)) // baseType is the type of the second argument
Expand Down
Loading