Skip to content

[NVPTX] Don't propagate ninf and nnan in lowerFREM #147125

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 2 commits into
base: main
Choose a base branch
from
Open

[NVPTX] Don't propagate ninf and nnan in lowerFREM #147125

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7484699
[NVPTX] Don't propagate `ninf` and `nnan` in `lowerFREM`
paperchalice Jul 5, 2025
9b0fec2
address comments
paperchalice Jul 7, 2025
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
14 changes: 9 additions & 5 deletions llvm/lib/Target/NVPTX/NVPTXISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2793,14 +2793,18 @@ static SDValue lowerFREM(SDValue Op, SelectionDAG &DAG,
EVT Ty = Op.getValueType();
SDNodeFlags Flags = Op->getFlags();

SDValue Div = DAG.getNode(ISD::FDIV, DL, Ty, X, Y, Flags);
SDValue Trunc = DAG.getNode(ISD::FTRUNC, DL, Ty, Div, Flags);
// fdiv can still generate inf and nan when nnan and ninf are set.
SDNodeFlags NewFlags = Flags;
NewFlags.setNoNaNs(false);
NewFlags.setNoInfs(false);
SDValue Div = DAG.getNode(ISD::FDIV, DL, Ty, X, Y, NewFlags);
SDValue Trunc = DAG.getNode(ISD::FTRUNC, DL, Ty, Div, NewFlags);
SDValue Mul = DAG.getNode(ISD::FMUL, DL, Ty, Trunc, Y,
Flags | SDNodeFlags::AllowContract);
NewFlags | SDNodeFlags::AllowContract);
SDValue Sub = DAG.getNode(ISD::FSUB, DL, Ty, X, Mul,
Flags | SDNodeFlags::AllowContract);
NewFlags | SDNodeFlags::AllowContract);

if (AllowUnsafeFPMath || Flags.hasNoInfs())
if (AllowUnsafeFPMath || (Flags.hasNoInfs() && Flags.hasApproximateFuncs()))
Copy link
Contributor

Choose a reason for hiding this comment

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

You shouldn't touch the afn thing here. I also did not mean this tail section of the function is wrong, I meant the entire implementation of this function is wrong

return Sub;

// If Y is infinite, return X
Expand Down
11 changes: 11 additions & 0 deletions llvm/test/CodeGen/NVPTX/frem-ninf-nnan.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,11 @@
; RUN: llc %s --stop-after=finalize-isel -mcpu=sm_60 -o - | FileCheck %s

target triple = "nvptx64-unknown-cuda"

define float @frem_ninf_nnan(float %a, float %b) {
; CHECK: nnan ninf FDIV32rr_prec
; CHECK-NOT: nnan ninf contract FNEGf32
; CHECK: contract FNEGf32
%r = frem ninf nnan float %a, %b
ret float %r
}
10 changes: 5 additions & 5 deletions llvm/test/CodeGen/NVPTX/frem.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -147,14 +147,14 @@ define half @frem_f16_ninf(half %a, half %b) {
; NORMAL-NEXT: ld.param.b16 %rs2, [frem_f16_ninf_param_1];
; NORMAL-NEXT: cvt.f32.f16 %r1, %rs2;
; NORMAL-NEXT: cvt.f32.f16 %r2, %rs1;
; NORMAL-NEXT: div.rn.f32 %r3, %r2, %r1;
; NORMAL-NEXT: div.approx.f32 %r3, %r2, %r1;
; NORMAL-NEXT: cvt.rzi.f32.f32 %r4, %r3;
; NORMAL-NEXT: neg.f32 %r5, %r4;
; NORMAL-NEXT: fma.rn.f32 %r6, %r5, %r1, %r2;
; NORMAL-NEXT: cvt.rn.f16.f32 %rs3, %r6;
; NORMAL-NEXT: st.param.b16 [func_retval0], %rs3;
; NORMAL-NEXT: ret;
%r = frem ninf half %a, %b
%r = frem ninf afn half %a, %b
ret half %r
}

Expand All @@ -180,13 +180,13 @@ define float @frem_f32_ninf(float %a, float %b) {
; NORMAL-NEXT: // %bb.0:
; NORMAL-NEXT: ld.param.b32 %r1, [frem_f32_ninf_param_0];
; NORMAL-NEXT: ld.param.b32 %r2, [frem_f32_ninf_param_1];
; NORMAL-NEXT: div.rn.f32 %r3, %r1, %r2;
; NORMAL-NEXT: div.approx.f32 %r3, %r1, %r2;
; NORMAL-NEXT: cvt.rzi.f32.f32 %r4, %r3;
; NORMAL-NEXT: neg.f32 %r5, %r4;
; NORMAL-NEXT: fma.rn.f32 %r6, %r5, %r2, %r1;
; NORMAL-NEXT: st.param.b32 [func_retval0], %r6;
; NORMAL-NEXT: ret;
%r = frem ninf float %a, %b
%r = frem ninf afn float %a, %b
ret float %r
}

Expand Down Expand Up @@ -218,7 +218,7 @@ define double @frem_f64_ninf(double %a, double %b) {
; NORMAL-NEXT: fma.rn.f64 %rd6, %rd5, %rd2, %rd1;
; NORMAL-NEXT: st.param.b64 [func_retval0], %rd6;
; NORMAL-NEXT: ret;
%r = frem ninf double %a, %b
%r = frem ninf afn double %a, %b
ret double %r
}

Expand Down
Loading