Skip to content

[AArch64]SIMD fpcvt codegen for rounding nodes #165546

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
Lukacma merged 11 commits into from
Dec 9, 2025
+615 −0
Merged

[AArch64]SIMD fpcvt codegen for rounding nodes #165546

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
11 commits
Select commit Hold shift + click to select a range
6377946
[AArch64][GlobalISel] SIMD fpcvt codegen for rounding nodes
Lukacma Oct 29, 2025
cec5513
Merge branch 'llvm:main' into cvt-round
Lukacma Oct 29, 2025
d74cd15
Merge branch 'main' into cvt-round
Lukacma Nov 18, 2025
3821797
Added patterns for legalized rounding nodes
Lukacma Nov 20, 2025
0f90db6
split tests
Lukacma Dec 1, 2025
ed50a6a
test fix
Lukacma Dec 1, 2025
10877d2
delete s file
Lukacma Dec 1, 2025
21a0c8d
Remove globalISel work
Lukacma Dec 3, 2025
da01d0a
Merge branch 'main' into cvt-round
Lukacma Dec 5, 2025
7e1c284
Adjust tests for clarity
Lukacma Dec 5, 2025
f085f44
Merge branch 'main' into cvt-round
Lukacma Dec 9, 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
73 changes: 73 additions & 0 deletions llvm/lib/Target/AArch64/AArch64InstrInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6799,6 +6799,79 @@ defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fp_to_uint_sat_gi, ftrunc, "F
defm : FPToIntegerPats<fp_to_sint, fp_to_sint_sat, fp_to_sint_sat_gi, fround, "FCVTAS">;
defm : FPToIntegerPats<fp_to_uint, fp_to_uint_sat, fp_to_uint_sat_gi, fround, "FCVTAU">;

// For global-isel we can use register classes to determine
// which FCVT instruction to use.
let Predicates = [HasFPRCVT] in {
Copy link
Contributor

@CarolineConcatto CarolineConcatto Nov 28, 2025

Choose a reason for hiding this comment

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

I am not sure if it will help, but maybe we could split the PR into patterns only for GlobalISel and another for the standard lowering/SelectionDAG?
I imagine that the ones from lines:
6810 till 6823
and
6848 till 6861
are only for GlobalISel while the other ones are for SelectionDAG

Copy link
Contributor Author

@Lukacma Lukacma Dec 1, 2025

Choose a reason for hiding this comment

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

Yes all the patterns without bitconvert node are for GlobalISel. May I ask what would be the value in splitting the patch ? As there is only couple of GlobalISel patterns added and the tests can be shared anyway, it makes sense to me to do it in one PR, but I can do that if you insist. The previous patches were also not split and it was fine.

Copy link
Contributor

@CarolineConcatto CarolineConcatto Dec 1, 2025

Choose a reason for hiding this comment

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

The main reason was to solve the issue raised by @arsenm about the complexity to review of the patch.

def : Pat<(i64 (any_lround f32:$Rn)),
Copy link
Contributor

@CarolineConcatto CarolineConcatto Oct 31, 2025

Choose a reason for hiding this comment

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

Should we not add pattern for the GlobalISel too. It seams to me that all the tests are always falling back to the selection dag.

Copy link
Contributor Author

@Lukacma Lukacma Nov 5, 2025

Choose a reason for hiding this comment

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

I am not exactly sure, I understand your comment. If you are asking whether we should add patterns here for other type of lround, like for example f16 to i64 type, I don't think that is necessary as GlobalISel doesn't support those types anyway yet for these nodes. That's why you see it fallback to SDAG.

(FCVTASDSr f32:$Rn)>;
def : Pat<(i64 (any_llround f32:$Rn)),
(FCVTASDSr f32:$Rn)>;
}
def : Pat<(i64 (any_lround f64:$Rn)),
(FCVTASv1i64 f64:$Rn)>;
def : Pat<(i64 (any_llround f64:$Rn)),
(FCVTASv1i64 f64:$Rn)>;

let Predicates = [HasFPRCVT] in {
def : Pat<(f32 (bitconvert (i32 (any_lround f16:$Rn)))),
(FCVTASSHr f16:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_lround f16:$Rn)))),
(FCVTASDHr f16:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_llround f16:$Rn)))),
(FCVTASDHr f16:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_lround f32:$Rn)))),
(FCVTASDSr f32:$Rn)>;
def : Pat<(f32 (bitconvert (i32 (any_lround f64:$Rn)))),
(FCVTASSDr f64:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_llround f32:$Rn)))),
(FCVTASDSr f32:$Rn)>;
}
def : Pat<(f32 (bitconvert (i32 (any_lround f32:$Rn)))),
(FCVTASv1i32 f32:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_lround f64:$Rn)))),
(FCVTASv1i64 f64:$Rn)>;
def : Pat<(f64 (bitconvert (i64 (any_llround f64:$Rn)))),
(FCVTASv1i64 f64:$Rn)>;

// For global-isel we can use register classes to determine
// which FCVT instruction to use.
let Predicates = [HasFPRCVT] in {
def : Pat<(i64 (any_lrint f16:$Rn)),
(FCVTZSDHr (FRINTXHr f16:$Rn))>;
def : Pat<(i64 (any_llrint f16:$Rn)),
(FCVTZSDHr (FRINTXHr f16:$Rn))>;
def : Pat<(i64 (any_lrint f32:$Rn)),
(FCVTZSDSr (FRINTXSr f32:$Rn))>;
def : Pat<(i64 (any_llrint f32:$Rn)),
(FCVTZSDSr (FRINTXSr f32:$Rn))>;
}
def : Pat<(i64 (any_lrint f64:$Rn)),
(FCVTZSv1i64 (FRINTXDr f64:$Rn))>;
def : Pat<(i64 (any_llrint f64:$Rn)),
(FCVTZSv1i64 (FRINTXDr f64:$Rn))>;

let Predicates = [HasFPRCVT] in {
def : Pat<(f32 (bitconvert (i32 (any_lrint f16:$Rn)))),
(FCVTZSSHr (FRINTXHr f16:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_lrint f16:$Rn)))),
(FCVTZSDHr (FRINTXHr f16:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_llrint f16:$Rn)))),
(FCVTZSDHr (FRINTXHr f16:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_lrint f32:$Rn)))),
(FCVTZSDSr (FRINTXSr f32:$Rn))>;
def : Pat<(f32 (bitconvert (i32 (any_lrint f64:$Rn)))),
(FCVTZSSDr (FRINTXDr f64:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_llrint f32:$Rn)))),
(FCVTZSDSr (FRINTXSr f32:$Rn))>;
}
def : Pat<(f32 (bitconvert (i32 (any_lrint f32:$Rn)))),
(FCVTZSv1i32 (FRINTXSr f32:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_lrint f64:$Rn)))),
(FCVTZSv1i64 (FRINTXDr f64:$Rn))>;
def : Pat<(f64 (bitconvert (i64 (any_llrint f64:$Rn)))),
(FCVTZSv1i64 (FRINTXDr f64:$Rn))>;


// f16 -> s16 conversions
let Predicates = [HasFullFP16] in {
def : Pat<(i16(fp_to_sint_sat_gi f16:$Rn)), (FCVTZSv1f16 f16:$Rn)>;
Expand Down
18 changes: 5 additions & 13 deletions llvm/lib/Target/AArch64/GISel/AArch64RegisterBankInfo.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -858,7 +858,11 @@ AArch64RegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
case TargetOpcode::G_FPTOSI_SAT:
case TargetOpcode::G_FPTOUI_SAT:
case TargetOpcode::G_FPTOSI:
case TargetOpcode::G_FPTOUI: {
case TargetOpcode::G_FPTOUI:
case TargetOpcode::G_INTRINSIC_LRINT:
case TargetOpcode::G_INTRINSIC_LLRINT:
case TargetOpcode::G_LROUND:
case TargetOpcode::G_LLROUND: {
LLT DstType = MRI.getType(MI.getOperand(0).getReg());
if (DstType.isVector())
break;
Expand All @@ -879,12 +883,6 @@ AArch64RegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
OpRegBankIdx = {PMI_FirstGPR, PMI_FirstFPR};
break;
}
case TargetOpcode::G_INTRINSIC_LRINT:
case TargetOpcode::G_INTRINSIC_LLRINT:
if (MRI.getType(MI.getOperand(0).getReg()).isVector())
break;
OpRegBankIdx = {PMI_FirstGPR, PMI_FirstFPR};
break;
case TargetOpcode::G_FCMP: {
// If the result is a vector, it must use a FPR.
AArch64GenRegisterBankInfo::PartialMappingIdx Idx0 =
Expand Down Expand Up @@ -1224,12 +1222,6 @@ AArch64RegisterBankInfo::getInstrMapping(const MachineInstr &MI) const {
}
break;
}
case TargetOpcode::G_LROUND:
case TargetOpcode::G_LLROUND: {
// Source is always floating point and destination is always integer.
OpRegBankIdx = {PMI_FirstGPR, PMI_FirstFPR};
break;
}
}

// Finally construct the computed mapping.
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/AArch64/GlobalISel/regbank-llround.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ body: |
; CHECK: liveins: $d0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %fpr:fpr(s64) = COPY $d0
; CHECK-NEXT: %llround:gpr(s64) = G_LLROUND %fpr(s64)
; CHECK-NEXT: %llround:fpr(s64) = G_LLROUND %fpr(s64)
; CHECK-NEXT: $d0 = COPY %llround(s64)
; CHECK-NEXT: RET_ReallyLR implicit $s0
%fpr:_(s64) = COPY $d0
Expand All @@ -35,7 +35,7 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %gpr:gpr(s64) = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr(s64) = COPY %gpr(s64)
; CHECK-NEXT: %llround:gpr(s64) = G_LLROUND [[COPY]](s64)
; CHECK-NEXT: %llround:fpr(s64) = G_LLROUND [[COPY]](s64)
; CHECK-NEXT: $d0 = COPY %llround(s64)
; CHECK-NEXT: RET_ReallyLR implicit $s0
%gpr:_(s64) = COPY $x0
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/AArch64/GlobalISel/regbank-lround.mir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ body: |
; CHECK: liveins: $d0
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %fpr:fpr(s64) = COPY $d0
; CHECK-NEXT: %lround:gpr(s64) = G_LROUND %fpr(s64)
; CHECK-NEXT: %lround:fpr(s64) = G_LROUND %fpr(s64)
; CHECK-NEXT: $d0 = COPY %lround(s64)
; CHECK-NEXT: RET_ReallyLR implicit $s0
%fpr:_(s64) = COPY $d0
Expand All @@ -35,7 +35,7 @@ body: |
; CHECK-NEXT: {{ $}}
; CHECK-NEXT: %gpr:gpr(s64) = COPY $x0
; CHECK-NEXT: [[COPY:%[0-9]+]]:fpr(s64) = COPY %gpr(s64)
; CHECK-NEXT: %lround:gpr(s64) = G_LROUND [[COPY]](s64)
; CHECK-NEXT: %lround:fpr(s64) = G_LROUND [[COPY]](s64)
; CHECK-NEXT: $d0 = COPY %lround(s64)
; CHECK-NEXT: RET_ReallyLR implicit $s0
%gpr:_(s64) = COPY $x0
Expand Down
Loading