Skip to content

[X86][GlobalISel] Use GIntrinsic during custom intrinsic selection #83498

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

Closed
wants to merge 1 commit into from
Closed

[X86][GlobalISel] Use GIntrinsic during custom intrinsic selection #83498

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: 1 addition & 1 deletion llvm/lib/Target/X86/GISel/X86InstructionSelector.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1843,7 +1843,7 @@ bool X86InstructionSelector::selectIntrinsicWSideEffects(
assert(I.getOpcode() == TargetOpcode::G_INTRINSIC_W_SIDE_EFFECTS &&
"unexpected instruction");

if (I.getOperand(0).getIntrinsicID() != Intrinsic::trap)
if (cast<GIntrinsic>(I).getIntrinsicID() != Intrinsic::trap)
Copy link
Contributor

Choose a reason for hiding this comment

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

generic intrinsics should not use G_INTRINSIC, this shouldn't have been reachable

Copy link
Contributor Author

Choose a reason for hiding this comment

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

That's interesting. It seems that llvm.trap is a special case as well as llvm.debugtrap and llvm.ubsantrap. Comparing to other intrinsics we can't lower it into combination of GlobalISel opcodes. So it is alive until instruction selection and then we custom select it.

If we take a look at SelectionDAG it lowers the intrinsic into ISD::TRAP. It seems that we can create G_TRAP/G_DEBUGTRAP/G_UBSANTRAP opcodes and lower intrinsics during irtranslator. I see that only 3 targets have a custom lowering for traps: AArch64, X86 and RISCV. So it shouldn't be a problem to migrate them all at once to proper lowering. What do you think, does it worth it?

Copy link
Contributor

Choose a reason for hiding this comment

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

Yes, that's how I would expect this to work

return false;

BuildMI(*I.getParent(), I, I.getDebugLoc(), TII.get(X86::TRAP));
Expand Down
27 changes: 27 additions & 0 deletions llvm/test/CodeGen/X86/GlobalISel/select-custom-intrinsic.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 4
# RUN: llc -mtriple=x86_64-linux-gnu -run-pass=instruction-select -global-isel-abort=2 -pass-remarks-missed='gisel*' 2>&1 %s -o - | FileCheck %s --check-prefixes=CHECK
# RUN: llc -mtriple=i386-linux-gnu -run-pass=instruction-select -global-isel-abort=2 -pass-remarks-missed='gisel*' 2>&1 %s -o - | FileCheck %s --check-prefixes=CHECK

# CHECK: remark: <unknown>:0:0: cannot select: %2:gpr(s8) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.x86.atomic.sub.cc), %0:gpr(p0), %1:gpr(s32), 4 (in function: intrinsic_returns)

...
---
name: intrinsic_returns
legalized: true
regBankSelected: true
body: |
bb.1:

; CHECK-LABEL: name: intrinsic_returns
; CHECK: [[DEF:%[0-9]+]]:gpr(p0) = IMPLICIT_DEF
; CHECK-NEXT: [[C:%[0-9]+]]:gpr(s32) = G_CONSTANT i32 1
; CHECK-NEXT: [[INT:%[0-9]+]]:gpr(s8) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.x86.atomic.sub.cc), [[DEF]](p0), [[C]](s32), 4
; CHECK-NEXT: $al = COPY [[INT]](s8)
; CHECK-NEXT: RET 0, implicit $al
%0:gpr(p0) = IMPLICIT_DEF
%1:gpr(s32) = G_CONSTANT i32 1
%2:gpr(s8) = G_INTRINSIC_W_SIDE_EFFECTS intrinsic(@llvm.x86.atomic.sub.cc), %0(p0), %1(s32), 4
$al = COPY %2(s8)
RET 0, implicit $al

...