Skip to content

[LoongArch] Use LSX for scalar FP rounding with explicit rounding mode #114766

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 5 commits into from
Nov 6, 2024
Merged

[LoongArch] Use LSX for scalar FP rounding with explicit rounding mode #114766

Show file tree
Hide file tree
Changes from 3 commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
a61e0ba
[LoongArch] Use LSX for scalar FP rounding with explicit rounding mode
zhaoqi5 Nov 2, 2024
e9cfe92
update tests
zhaoqi5 Nov 5, 2024
8f00ece
Merge branch 'main' into vector-fp-conv
zhaoqi5 Nov 5, 2024
646a96d
Merge remote-tracking branch 'origin/main' into vector-fp-conv
zhaoqi5 Nov 5, 2024
f60db3f
update tests after precommit
zhaoqi5 Nov 5, 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
4 changes: 4 additions & 0 deletions llvm/lib/Target/LoongArch/LoongArchISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,6 +286,10 @@ LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
VT, Expand);
}
setOperationAction(ISD::CTPOP, GRLenVT, Legal);
setOperationAction(ISD::FCEIL, {MVT::f32, MVT::f64}, Legal);
setOperationAction(ISD::FFLOOR, {MVT::f32, MVT::f64}, Legal);
setOperationAction(ISD::FTRUNC, {MVT::f32, MVT::f64}, Legal);
setOperationAction(ISD::FROUNDEVEN, {MVT::f32, MVT::f64}, Legal);
}

// Set operations for 'LASX' feature.
Expand Down
26 changes: 26 additions & 0 deletions llvm/lib/Target/LoongArch/LoongArchLSXInstrInfo.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2259,6 +2259,32 @@ def : Pat<(loongarch_vfrsqrte v2f64:$src),
(VFRSQRTE_D v2f64:$src)>;
}

// Vector floating-point conversion
def : Pat<(f32 (fceil FPR32:$fj)),
(f32 (EXTRACT_SUBREG (VFRINTRP_S (VREPLVEI_W
(SUBREG_TO_REG (i64 0), FPR32:$fj, sub_32), 0)), sub_32))>;
def : Pat<(f64 (fceil FPR64:$fj)),
(f64 (EXTRACT_SUBREG (VFRINTRP_D (VREPLVEI_D
(SUBREG_TO_REG (i64 0), FPR64:$fj, sub_64), 0)), sub_64))>;
def : Pat<(f32 (ffloor FPR32:$fj)),
(f32 (EXTRACT_SUBREG (VFRINTRM_S (VREPLVEI_W
(SUBREG_TO_REG (i64 0), FPR32:$fj, sub_32), 0)), sub_32))>;
def : Pat<(f64 (ffloor FPR64:$fj)),
(f64 (EXTRACT_SUBREG (VFRINTRM_D (VREPLVEI_D
(SUBREG_TO_REG (i64 0), FPR64:$fj, sub_64), 0)), sub_64))>;
def : Pat<(f32 (ftrunc FPR32:$fj)),
(f32 (EXTRACT_SUBREG (VFRINTRZ_S (VREPLVEI_W
(SUBREG_TO_REG (i64 0), FPR32:$fj, sub_32), 0)), sub_32))>;
def : Pat<(f64 (ftrunc FPR64:$fj)),
(f64 (EXTRACT_SUBREG (VFRINTRZ_D (VREPLVEI_D
(SUBREG_TO_REG (i64 0), FPR64:$fj, sub_64), 0)), sub_64))>;
def : Pat<(f32 (froundeven FPR32:$fj)),
(f32 (EXTRACT_SUBREG (VFRINTRNE_S (VREPLVEI_W
(SUBREG_TO_REG (i64 0), FPR32:$fj, sub_32), 0)), sub_32))>;
def : Pat<(f64 (froundeven FPR64:$fj)),
(f64 (EXTRACT_SUBREG (VFRINTRNE_D (VREPLVEI_D
(SUBREG_TO_REG (i64 0), FPR64:$fj, sub_64), 0)), sub_64))>;

// load
def : Pat<(int_loongarch_lsx_vld GPR:$rj, timm:$imm),
(VLD GPR:$rj, (to_valid_timm timm:$imm))>;
Expand Down
156 changes: 156 additions & 0 deletions llvm/test/CodeGen/LoongArch/fp-conversion.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,156 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc --mtriple=loongarch64 --mattr=-lsx < %s | FileCheck %s --check-prefix=NOLSX
; RUN: llc --mtriple=loongarch64 --mattr=+lsx < %s | FileCheck %s --check-prefix=LSX

;; ceilf
define float @ceil_f32(float %i) nounwind {
; NOLSX-LABEL: ceil_f32:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(ceilf)
;
; LSX-LABEL: ceil_f32:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0 killed $f0 def $vr0
; LSX-NEXT: vreplvei.w $vr0, $vr0, 0
; LSX-NEXT: vfrintrp.s $vr0, $vr0
; LSX-NEXT: # kill: def $f0 killed $f0 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call float @llvm.ceil.f32(float %i)
ret float %0
}

;; ceil
define double @ceil_f64(double %i) nounwind {
; NOLSX-LABEL: ceil_f64:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(ceil)
;
; LSX-LABEL: ceil_f64:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 def $vr0
; LSX-NEXT: vreplvei.d $vr0, $vr0, 0
; LSX-NEXT: vfrintrp.d $vr0, $vr0
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call double @llvm.ceil.f64(double %i)
ret double %0
}

;; floorf
define float @floor_f32(float %i) nounwind {
; NOLSX-LABEL: floor_f32:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(floorf)
;
; LSX-LABEL: floor_f32:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0 killed $f0 def $vr0
; LSX-NEXT: vreplvei.w $vr0, $vr0, 0
; LSX-NEXT: vfrintrm.s $vr0, $vr0
; LSX-NEXT: # kill: def $f0 killed $f0 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call float @llvm.floor.f32(float %i)
ret float %0
}

;; floor
define double @floor_f64(double %i) nounwind {
; NOLSX-LABEL: floor_f64:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(floor)
;
; LSX-LABEL: floor_f64:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 def $vr0
; LSX-NEXT: vreplvei.d $vr0, $vr0, 0
; LSX-NEXT: vfrintrm.d $vr0, $vr0
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call double @llvm.floor.f64(double %i)
ret double %0
}

;; truncf
define float @trunc_f32(float %i) nounwind {
; NOLSX-LABEL: trunc_f32:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(truncf)
;
; LSX-LABEL: trunc_f32:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0 killed $f0 def $vr0
; LSX-NEXT: vreplvei.w $vr0, $vr0, 0
; LSX-NEXT: vfrintrz.s $vr0, $vr0
; LSX-NEXT: # kill: def $f0 killed $f0 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call float @llvm.trunc.f32(float %i)
ret float %0
}

;; trunc
define double @trunc_f64(double %i) nounwind {
; NOLSX-LABEL: trunc_f64:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(trunc)
;
; LSX-LABEL: trunc_f64:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 def $vr0
; LSX-NEXT: vreplvei.d $vr0, $vr0, 0
; LSX-NEXT: vfrintrz.d $vr0, $vr0
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call double @llvm.trunc.f64(double %i)
ret double %0
}

;; roundevenf
define float @roundeven_f32(float %i) nounwind {
; NOLSX-LABEL: roundeven_f32:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(roundevenf)
;
; LSX-LABEL: roundeven_f32:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0 killed $f0 def $vr0
; LSX-NEXT: vreplvei.w $vr0, $vr0, 0
; LSX-NEXT: vfrintrne.s $vr0, $vr0
; LSX-NEXT: # kill: def $f0 killed $f0 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call float @llvm.roundeven.f32(float %i)
ret float %0
}

;; roundeven
define double @roundeven_f64(double %i) nounwind {
; NOLSX-LABEL: roundeven_f64:
; NOLSX: # %bb.0: # %entry
; NOLSX-NEXT: b %plt(roundeven)
;
; LSX-LABEL: roundeven_f64:
; LSX: # %bb.0: # %entry
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 def $vr0
; LSX-NEXT: vreplvei.d $vr0, $vr0, 0
; LSX-NEXT: vfrintrne.d $vr0, $vr0
; LSX-NEXT: # kill: def $f0_64 killed $f0_64 killed $vr0
; LSX-NEXT: ret
entry:
%0 = call double @llvm.roundeven.f64(double %i)
ret double %0
}

declare float @llvm.ceil.f32(float)
declare double @llvm.ceil.f64(double)
declare float @llvm.floor.f32(float)
declare double @llvm.floor.f64(double)
declare float @llvm.trunc.f32(float)
declare double @llvm.trunc.f64(double)
declare float @llvm.roundeven.f32(float)
declare double @llvm.roundeven.f64(double)
Loading