Skip to content

[JIT] ARM64 - Optimize x < 0 and x >= 0 #83176

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

Merged
merged 20 commits into from
Mar 16, 2023
Merged

[JIT] ARM64 - Optimize x < 0 and x >= 0 #83176

Show file tree
Hide file tree
Changes from 17 commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
226b6f7
Initial work that allows optimizing x < 0
TIHan Mar 9, 2023
b3da93d
Few tweaks
TIHan Mar 10, 2023
41d3ecb
Optimized x < 0 and x >= 0 for conditionals
TIHan Mar 10, 2023
767fe62
Removed old commented code. Switched tbz and tbnz
TIHan Mar 10, 2023
4a96fb7
Simplified
TIHan Mar 10, 2023
46e4cce
merged
TIHan Mar 10, 2023
c366a7a
Fix build
TIHan Mar 10, 2023
006bfea
Check for possible interference
TIHan Mar 10, 2023
72fd1d0
Reverse
TIHan Mar 10, 2023
bd286a5
Check invariant-range
TIHan Mar 10, 2023
69a49fd
Cleanup
TIHan Mar 10, 2023
359609d
Cleanup
TIHan Mar 10, 2023
a689d81
Handle new JCMP flags in gtReverseCond
TIHan Mar 13, 2023
2f6f378
Comment update
TIHan Mar 13, 2023
6e31149
Remove extra flags
TIHan Mar 14, 2023
206a8da
Simplify further
TIHan Mar 14, 2023
a95a2ba
Fix warning
TIHan Mar 14, 2023
bbabdd1
Update src/coreclr/jit/lower.cpp
TIHan Mar 14, 2023
9a0b138
Fix warning
TIHan Mar 14, 2023
887a233
Fixing assert
TIHan Mar 14, 2023
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
13 changes: 12 additions & 1 deletion src/coreclr/jit/codegenarm64.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4532,7 +4532,18 @@ void CodeGen::genCodeForCompare(GenTreeOp* tree)
if (op2->isContainedIntOrIImmed())
{
GenTreeIntConCommon* intConst = op2->AsIntConCommon();
emit->emitIns_R_I(ins, cmpSize, op1->GetRegNum(), intConst->IconValue());

regNumber op1Reg = op1->GetRegNum();

if (compiler->opts.OptimizationEnabled() && (ins == INS_cmp) && (targetReg != REG_NA) &&
tree->OperIs(GT_LT) && intConst->IsIntegralConst(0) && ((cmpSize == EA_4BYTE) || (cmpSize == EA_8BYTE)))
{
emit->emitIns_R_R_I(INS_lsr, cmpSize, targetReg, op1Reg, (int)cmpSize * 8 - 1);
genProduceReg(tree);
return;
}

emit->emitIns_R_I(ins, cmpSize, op1Reg, intConst->IconValue());
}
else
{
Expand Down
3 changes: 1 addition & 2 deletions src/coreclr/jit/gentree.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3481,8 +3481,7 @@ GenTree* Compiler::gtReverseCond(GenTree* tree)
else if (tree->OperIs(GT_JCMP))
{
// Flip the GTF_JCMP_EQ
//
// This causes switching
// On ARM64, this causes switching
// cbz <=> cbnz
// tbz <=> tbnz
tree->gtFlags ^= GTF_JCMP_EQ;
Expand Down
7 changes: 7 additions & 0 deletions src/coreclr/jit/lower.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3379,6 +3379,13 @@ GenTree* Lowering::LowerJTrue(GenTreeOp* jtrue)
flags = cond->OperIs(GT_EQ) ? GTF_JCMP_EQ : GTF_EMPTY;
useJCMP = true;
}
else if (cond->OperIs(GT_LT, GT_GE) && relopOp2->IsIntegralConst(0))
{
// Codegen will use tbnz or tbz in codegen which do not affect the flag register
flags = GTF_JCMP_TST | (cond->OperIs(GT_LT) ? GTF_EMPTY : GTF_JCMP_EQ);
useJCMP = true;
relopOp2->AsIntConCommon()->SetIntegralValue((1L << (8 * genTypeSize(genActualType(relopOp1)) - 1)));
}
else if (cond->OperIs(GT_TEST_EQ, GT_TEST_NE) && isPow2(relopOp2->AsIntCon()->IconValue()))
{
// Codegen will use tbz or tbnz in codegen which do not affect the flag register
Expand Down