Skip to content

[SPARC] Use op-then-halve instructions when we have VIS3 #135718

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 15 commits into from
May 23, 2025
Merged

[SPARC] Use op-then-halve instructions when we have VIS3 #135718

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
904fd88
[𝘀𝗽𝗿] changes to main this commit is based on
koachan Apr 15, 2025
9696d19
[𝘀𝗽𝗿] initial version
koachan Apr 15, 2025
e0f99c4
[𝘀𝗽𝗿] changes introduced through rebase
koachan Apr 15, 2025
d942a7f
Use FPImmLeaf
koachan Apr 15, 2025
31c6f5a
[𝘀𝗽𝗿] changes introduced through rebase
koachan Apr 16, 2025
561990c
Rebase & update
koachan Apr 16, 2025
71ffd3d
[𝘀𝗽𝗿] changes introduced through rebase
koachan Apr 18, 2025
7586c5f
Add vector tests
koachan Apr 18, 2025
4a968b2
[𝘀𝗽𝗿] changes introduced through rebase
koachan Apr 26, 2025
840c95d
Rebase and apply suggestions
koachan Apr 26, 2025
50422a4
Rework isFPImmLegal
koachan May 4, 2025
142bf8f
[𝘀𝗽𝗿] changes introduced through rebase
koachan May 22, 2025
d01cb06
Rebase update
koachan May 22, 2025
610572e
[𝘀𝗽𝗿] changes introduced through rebase
bcardosolopes May 23, 2025
095fcf0
Rebase update
koachan May 23, 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
18 changes: 13 additions & 5 deletions llvm/lib/Target/Sparc/SparcISelLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3605,15 +3605,23 @@ bool SparcTargetLowering::useLoadStackGuardNode(const Module &M) const {
return true;
}

bool SparcTargetLowering::isFNegFree(EVT VT) const {
if (Subtarget->isVIS3())
return VT == MVT::f32 || VT == MVT::f64;
return false;
}

bool SparcTargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT,
bool ForCodeSize) const {
bool CanLower = false;
Copy link
Contributor

@arsenm arsenm May 3, 2025

Choose a reason for hiding this comment

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

Suggested change
bool CanLower = false;

if (VT != MVT::f32 && VT != MVT::f64)
return false;
if (Imm.isZero())
return Subtarget->isVIS();
if (Imm.isExactlyValue(+0.5) || Imm.isExactlyValue(-0.5))
return Subtarget->isVIS3();
return false;
if (Subtarget->isVIS())
CanLower = CanLower || Imm.isZero();
Copy link
Contributor

@arsenm arsenm May 3, 2025

Choose a reason for hiding this comment

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

Suggested change
if (Subtarget->isVIS())
CanLower = CanLower || Imm.isZero();
if (Subtarget->isVIS() mm.isZero())
return true;

Did this mean to include -0?

Copy link
Contributor Author

@koachan koachan May 4, 2025

Choose a reason for hiding this comment

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

Yes, +0 and -0.

if (Subtarget->isVIS3())
CanLower =
CanLower || (Imm.isExactlyValue(+0.5) || Imm.isExactlyValue(-0.5));
return CanLower;
Copy link
Contributor

@arsenm arsenm May 3, 2025

Choose a reason for hiding this comment

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

Suggested change
if (Subtarget->isVIS3())
CanLower =
CanLower || (Imm.isExactlyValue(+0.5) || Imm.isExactlyValue(-0.5));
return CanLower;
if (Subtarget->isVIS3())
return Imm.isExactlyValue(+0.5) || Imm.isExactlyValue(-0.5);
return false;

Could also use getExactLog2Abs == -1?

}

bool SparcTargetLowering::isCtlzFast() const { return Subtarget->isVIS3(); }
Expand Down
2 changes: 2 additions & 0 deletions llvm/lib/Target/Sparc/SparcISelLowering.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -207,6 +207,8 @@ namespace llvm {
return VT != MVT::f128;
}

bool isFNegFree(EVT VT) const override;

bool isFPImmLegal(const APFloat &Imm, EVT VT,
bool ForCodeSize) const override;

Expand Down
7 changes: 1 addition & 6 deletions llvm/lib/Target/Sparc/SparcInstrVIS.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -335,17 +335,12 @@ def : Pat<(i64 (bitconvert f64:$src)), (MOVDTOX $src)>;
def : Pat<(f64 (bitconvert i64:$src)), (MOVXTOD $src)>;

// OP-then-neg FP operations.
// TODO handle equivalent patterns like `rs1*-rs2`.
def : Pat<(f32 (fneg (fadd f32:$rs1, f32:$rs2))), (FNADDS $rs1, $rs2)>;
def : Pat<(f64 (fneg (fadd f64:$rs1, f64:$rs2))), (FNADDD $rs1, $rs2)>;
def : Pat<(f32 (fneg (fmul f32:$rs1, f32:$rs2))), (FNMULS $rs1, $rs2)>;
def : Pat<(f32 (fmul (fneg f32:$rs1), f32:$rs2)), (FNMULS $rs1, $rs2)>;
def : Pat<(f32 (fmul f32:$rs1, (fneg f32:$rs2))), (FNMULS $rs1, $rs2)>;
def : Pat<(f64 (fneg (fmul f64:$rs1, f64:$rs2))), (FNMULD $rs1, $rs2)>;
def : Pat<(f64 (fmul (fneg f64:$rs1), f64:$rs2)), (FNMULD $rs1, $rs2)>;
def : Pat<(f64 (fmul f64:$rs1, (fneg f64:$rs2))), (FNMULD $rs1, $rs2)>;
def : Pat<(f64 (fneg (fmul (fpextend f32:$rs1), (fpextend f32:$rs2)))), (FNSMULD $rs1, $rs2)>;
def : Pat<(f64 (fmul (fneg (fpextend f32:$rs1)), (fpextend f32:$rs2))), (FNSMULD $rs1, $rs2)>;
def : Pat<(f64 (fmul (fpextend f32:$rs1), (fneg (fpextend f32:$rs2)))), (FNSMULD $rs1, $rs2)>;

// Op-then-halve FP operations.
def : Pat<(f32 (fmul (fadd f32:$rs1, f32:$rs2), fpimmhalf)), (FHADDS $rs1, $rs2)>;
Expand Down
248 changes: 12 additions & 236 deletions llvm/test/CodeGen/SPARC/float-vis3.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,8 +23,8 @@ entry:
ret double %fneg
}

define float @fnmuls1(float %a, float %b) nounwind {
; CHECK-LABEL: fnmuls1:
define float @fnmuls(float %a, float %b) nounwind {
; CHECK-LABEL: fnmuls:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuls %f1, %f3, %f0
Expand All @@ -34,30 +34,8 @@ entry:
ret float %fneg
}

define float @fnmuls2(float %a, float %b) nounwind {
; CHECK-LABEL: fnmuls2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuls %f1, %f3, %f0
entry:
%fneg = fneg float %a
%mul = fmul float %fneg, %b
ret float %mul
}

define float @fnmuls3(float %a, float %b) nounwind {
; CHECK-LABEL: fnmuls3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuls %f3, %f1, %f0
entry:
%fneg = fneg float %b
%mul = fmul float %fneg, %a
ret float %mul
}

define double @fnmuld1(double %a, double %b) nounwind {
; CHECK-LABEL: fnmuld1:
define double @fnmuld(double %a, double %b) nounwind {
; CHECK-LABEL: fnmuld:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuld %f0, %f2, %f0
Expand All @@ -67,30 +45,8 @@ entry:
ret double %fneg
}

define double @fnmuld2(double %a, double %b) nounwind {
; CHECK-LABEL: fnmuld2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuld %f0, %f2, %f0
entry:
%fneg = fneg double %a
%mul = fmul double %fneg, %b
ret double %mul
}

define double @fnmuld3(double %a, double %b) nounwind {
; CHECK-LABEL: fnmuld3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuld %f2, %f0, %f0
entry:
%fneg = fneg double %b
%mul = fmul double %fneg, %a
ret double %mul
}

define double @fnsmuld1(float %a, float %b) nounwind {
; CHECK-LABEL: fnsmuld1:
define double @fnsmuld(float %a, float %b) nounwind {
; CHECK-LABEL: fnsmuld:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnsmuld %f1, %f3, %f0
Expand All @@ -102,98 +58,6 @@ entry:
ret double %fneg
}

define double @fnsmuld2(float %a, float %b) nounwind {
; CHECK-LABEL: fnsmuld2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnsmuld %f1, %f3, %f0
entry:
%da = fpext float %a to double
%db = fpext float %b to double
%fneg = fneg double %da
%mul = fmul double %fneg, %db
ret double %mul
}

define double @fnsmuld3(float %a, float %b) nounwind {
; CHECK-LABEL: fnsmuld3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnsmuld %f3, %f1, %f0
entry:
%da = fpext float %a to double
%db = fpext float %b to double
%fneg = fneg double %db
%mul = fmul double %fneg, %da
ret double %mul
}

define float @fhadds(float %a, float %b) nounwind {
; CHECK-LABEL: fhadds:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fhadds %f1, %f3, %f0
entry:
%add = fadd float %a, %b
%div = fmul float %add, 5.000000e-01
ret float %div
}

define double @fhaddd(double %a, double %b) nounwind {
; CHECK-LABEL: fhaddd:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fhaddd %f0, %f2, %f0
entry:
%add = fadd double %a, %b
%div = fmul double %add, 5.000000e-01
ret double %div
}

define float @fhsubs(float %a, float %b) nounwind {
; CHECK-LABEL: fhsubs:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fhsubs %f1, %f3, %f0
entry:
%sub = fsub float %a, %b
%div = fmul float %sub, 5.000000e-01
ret float %div
}

define double @fhsubd(double %a, double %b) nounwind {
; CHECK-LABEL: fhsubd:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fhsubd %f0, %f2, %f0
entry:
%sub = fsub double %a, %b
%div = fmul double %sub, 5.000000e-01
ret double %div
}

define float @fnhadds(float %a, float %b) nounwind {
; CHECK-LABEL: fnhadds:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnhadds %f1, %f3, %f0
entry:
%add.i = fadd float %a, %b
%fneg = fmul float %add.i, -5.000000e-01
ret float %fneg
}

define double @fnhaddd(double %a, double %b) nounwind {
; CHECK-LABEL: fnhaddd:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: retl
; CHECK-NEXT: fnhaddd %f0, %f2, %f0
entry:
%add.i = fadd double %a, %b
%fneg = fmul double %add.i, -5.000000e-01
ret double %fneg
}

define <4 x float> @vec_fnadds(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnadds:
; CHECK: ! %bb.0: ! %entry
Expand Down Expand Up @@ -222,8 +86,8 @@ entry:
ret <4 x double> %fneg
}

define <4 x float> @vec_fnmuls1(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnmuls1:
define <4 x float> @vec_fnmuls(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnmuls:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuls %f1, %f9, %f0
; CHECK-NEXT: fnmuls %f3, %f11, %f1
Expand All @@ -236,36 +100,8 @@ entry:
ret <4 x float> %fneg
}

define <4 x float> @vec_fnmuls2(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnmuls2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuls %f1, %f9, %f0
; CHECK-NEXT: fnmuls %f3, %f11, %f1
; CHECK-NEXT: fnmuls %f5, %f13, %f2
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuls %f7, %f15, %f3
entry:
%fneg = fneg <4 x float> %a
%mul = fmul <4 x float> %fneg, %b
ret <4 x float> %mul
}

define <4 x float> @vec_fnmuls3(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnmuls3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuls %f9, %f1, %f0
; CHECK-NEXT: fnmuls %f11, %f3, %f1
; CHECK-NEXT: fnmuls %f13, %f5, %f2
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuls %f15, %f7, %f3
entry:
%fneg = fneg <4 x float> %b
%mul = fmul <4 x float> %fneg, %a
ret <4 x float> %mul
}

define <4 x double> @vec_nmuld1(<4 x double> %a, <4 x double> %b) nounwind {
; CHECK-LABEL: vec_nmuld1:
define <4 x double> @vec_fnmuld(<4 x double> %a, <4 x double> %b) nounwind {
; CHECK-LABEL: vec_fnmuld:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuld %f0, %f8, %f0
; CHECK-NEXT: fnmuld %f2, %f10, %f2
Expand All @@ -278,36 +114,8 @@ entry:
ret <4 x double> %fneg
}

define <4 x double> @vec_fnmuld2(<4 x double> %a, <4 x double> %b) nounwind {
; CHECK-LABEL: vec_fnmuld2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuld %f0, %f8, %f0
; CHECK-NEXT: fnmuld %f2, %f10, %f2
; CHECK-NEXT: fnmuld %f4, %f12, %f4
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuld %f6, %f14, %f6
entry:
%fneg = fneg <4 x double> %a
%mul = fmul <4 x double> %fneg, %b
ret <4 x double> %mul
}

define <4 x double> @vec_fnmuld3(<4 x double> %a, <4 x double> %b) nounwind {
; CHECK-LABEL: vec_fnmuld3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnmuld %f8, %f0, %f0
; CHECK-NEXT: fnmuld %f10, %f2, %f2
; CHECK-NEXT: fnmuld %f12, %f4, %f4
; CHECK-NEXT: retl
; CHECK-NEXT: fnmuld %f14, %f6, %f6
entry:
%fneg = fneg <4 x double> %b
%mul = fmul <4 x double> %fneg, %a
ret <4 x double> %mul
}

define <4 x double> @vec_fnsmuld1(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnsmuld1:
define <4 x double> @vec_fnsmuld(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnsmuld:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnsmuld %f1, %f9, %f0
; CHECK-NEXT: fnsmuld %f3, %f11, %f2
Expand All @@ -322,38 +130,6 @@ entry:
ret <4 x double> %fneg
}

define <4 x double> @vec_fnsmuld2(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnsmuld2:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnsmuld %f1, %f9, %f0
; CHECK-NEXT: fnsmuld %f3, %f11, %f2
; CHECK-NEXT: fnsmuld %f5, %f13, %f4
; CHECK-NEXT: retl
; CHECK-NEXT: fnsmuld %f7, %f15, %f6
entry:
%da = fpext <4 x float> %a to <4 x double>
%db = fpext <4 x float> %b to <4 x double>
%fneg = fneg <4 x double> %da
%mul = fmul <4 x double> %fneg, %db
ret <4 x double> %mul
}

define <4 x double> @vec_fnsmuld3(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fnsmuld3:
; CHECK: ! %bb.0: ! %entry
; CHECK-NEXT: fnsmuld %f9, %f1, %f0
; CHECK-NEXT: fnsmuld %f11, %f3, %f2
; CHECK-NEXT: fnsmuld %f13, %f5, %f4
; CHECK-NEXT: retl
; CHECK-NEXT: fnsmuld %f15, %f7, %f6
entry:
%da = fpext <4 x float> %a to <4 x double>
%db = fpext <4 x float> %b to <4 x double>
%fneg = fneg <4 x double> %db
%mul = fmul <4 x double> %fneg, %da
ret <4 x double> %mul
}

define <4 x float> @vec_fhadds(<4 x float> %a, <4 x float> %b) nounwind {
; CHECK-LABEL: vec_fhadds:
; CHECK: ! %bb.0: ! %entry
Expand Down
Loading