Skip to content

expandFMINIMUM_FMAXIMUM: Use FMINIMUMNUM and FMAXIMUMNUM if possible #180487

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

Draft
wzssyqa wants to merge 2 commits into
base: main
Choose a base branch
from
Draft

expandFMINIMUM_FMAXIMUM: Use FMINIMUMNUM and FMAXIMUMNUM if possible #180487

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
fc95674
expandFMINIMUM_FMAXIMUM: Use FMINIMUMNUM and FMAXIMUMNUM if possible
wzssyqa Feb 9, 2026
b829e4c
update nvptx tests
wzssyqa Feb 9, 2026
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
10 changes: 7 additions & 3 deletions llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8868,15 +8868,19 @@ SDValue TargetLowering::expandFMINIMUM_FMAXIMUM(SDNode *N,
// First, implement comparison not propagating NaN. If no native fmin or fmax
// available, use plain select with setcc instead.
SDValue MinMax;
unsigned CompOpcIeee = IsMax ? ISD::FMAXNUM_IEEE : ISD::FMINNUM_IEEE;
unsigned CompOpcIeee2008 = IsMax ? ISD::FMAXNUM_IEEE : ISD::FMINNUM_IEEE;
unsigned CompOpcIeee2019Num = IsMax ? ISD::FMAXIMUMNUM : ISD::FMINIMUMNUM;
unsigned CompOpc = IsMax ? ISD::FMAXNUM : ISD::FMINNUM;

// FIXME: We should probably define fminnum/fmaxnum variants with correct
// signed zero behavior.
bool MinMaxMustRespectOrderedZero = false;

if (isOperationLegalOrCustom(CompOpcIeee, VT)) {
MinMax = DAG.getNode(CompOpcIeee, DL, VT, LHS, RHS, Flags);
if (isOperationLegalOrCustom(CompOpcIeee2008, VT)) {
MinMax = DAG.getNode(CompOpcIeee2008, DL, VT, LHS, RHS, Flags);
MinMaxMustRespectOrderedZero = true;
} else if (isOperationLegalOrCustom(CompOpcIeee2019Num, VT)) {
MinMax = DAG.getNode(CompOpcIeee2019Num, DL, VT, LHS, RHS, Flags);
MinMaxMustRespectOrderedZero = true;
} else if (isOperationLegalOrCustom(CompOpc, VT)) {
MinMax = DAG.getNode(CompOpc, DL, VT, LHS, RHS, Flags);
Expand Down
39 changes: 39 additions & 0 deletions llvm/test/CodeGen/Hexagon/fminmax-v67.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,6 +73,45 @@ entry:
ret float %0
}

; CHECK-LABEL: t1_2019
; CHECK: dfmax
; CHECK: dfcmp.uo

define dso_local double @t1_2019(double %a, double %b) local_unnamed_addr {
entry:
%0 = tail call double @llvm.maximum.f64(double %a, double %b)
ret double %0
}

; CHECK-LABEL: t2_2019
; CHECK: dfmin
; CHECK: dfcmp.uo

define dso_local double @t2_2019(double %a, double %b) local_unnamed_addr {
entry:
%0 = tail call double @llvm.minimum.f64(double %a, double %b)
ret double %0
}

; CHECK-LABEL: t3_2019
; CHECK: sfmax
; CHECK: sfcmp.uo

define dso_local float @t3_2019(float %a, float %b) local_unnamed_addr {
entry:
%0 = tail call float @llvm.maximum.f32(float %a, float %b)
ret float %0
}

; CHECK-LABEL: t4_2019
; CHECK: sfmin
; CHECK: sfcmp.uo

define dso_local float @t4_2019(float %a, float %b) local_unnamed_addr {
entry:
%0 = tail call float @llvm.minimum.f32(float %a, float %b)
ret float %0
}

declare double @llvm.minnum.f64(double, double) #1
declare double @llvm.maxnum.f64(double, double) #1
Expand Down
18 changes: 18 additions & 0 deletions llvm/test/CodeGen/Hexagon/fminmax.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -51,6 +51,24 @@ entry:
ret float %call
}

; CHECK-LABEL: minimum
; CHECK: sfmin
; CHECK: sfcmp.uo
define float @minimum(float %x, float %y) #0 {
entry:
%call = tail call float @llvm.minimum.f32(float %x, float %y) #1
ret float %call
}

; CHECK-LABEL: maximum
; CHECK: sfmax
; CHECK: sfcmp.uo
define float @maximum(float %x, float %y) #0 {
entry:
%call = tail call float @llvm.maximum.f32(float %x, float %y) #1
ret float %call
}


declare float @fminf(float, float) #0
declare float @fmaxf(float, float) #0
Expand Down
96 changes: 36 additions & 60 deletions llvm/test/CodeGen/NVPTX/atomicrmw-sm60.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1836,30 +1836,24 @@ define float @fmax_acq_rel_float_global_cta(ptr addrspace(1) %addr, float %val)
define float @fminimum_acq_rel_float_global_cta(ptr addrspace(1) %addr, float %val) {
; SM60-LABEL: fminimum_acq_rel_float_global_cta(
; SM60: {
; SM60-NEXT: .reg .pred %p<6>;
; SM60-NEXT: .reg .b32 %r<9>;
; SM60-NEXT: .reg .pred %p<3>;
; SM60-NEXT: .reg .b32 %r<6>;
; SM60-NEXT: .reg .b64 %rd<2>;
; SM60-EMPTY:
; SM60-NEXT: // %bb.0:
; SM60-NEXT: ld.param.b32 %r2, [fminimum_acq_rel_float_global_cta_param_1];
; SM60-NEXT: ld.param.b64 %rd1, [fminimum_acq_rel_float_global_cta_param_0];
; SM60-NEXT: membar.cta;
; SM60-NEXT: ld.global.b32 %r8, [%rd1];
; SM60-NEXT: setp.eq.b32 %p3, %r2, -2147483648;
; SM60-NEXT: ld.global.b32 %r5, [%rd1];
; SM60-NEXT: $L__BB64_1: // %atomicrmw.start
; SM60-NEXT: // =>This Inner Loop Header: Depth=1
; SM60-NEXT: setp.nan.f32 %p1, %r8, %r2;
; SM60-NEXT: min.f32 %r3, %r8, %r2;
; SM60-NEXT: setp.nan.f32 %p1, %r5, %r2;
; SM60-NEXT: min.f32 %r3, %r5, %r2;
; SM60-NEXT: selp.f32 %r4, 0f7FC00000, %r3, %p1;
; SM60-NEXT: setp.eq.b32 %p2, %r8, -2147483648;
; SM60-NEXT: selp.f32 %r5, %r8, %r4, %p2;
; SM60-NEXT: selp.f32 %r6, %r2, %r5, %p3;
; SM60-NEXT: setp.eq.f32 %p4, %r4, 0f00000000;
; SM60-NEXT: selp.f32 %r7, %r6, %r4, %p4;
; SM60-NEXT: atom.cta.global.cas.b32 %r1, [%rd1], %r8, %r7;
; SM60-NEXT: setp.ne.b32 %p5, %r1, %r8;
; SM60-NEXT: mov.b32 %r8, %r1;
; SM60-NEXT: @%p5 bra $L__BB64_1;
; SM60-NEXT: atom.cta.global.cas.b32 %r1, [%rd1], %r5, %r4;
; SM60-NEXT: setp.ne.b32 %p2, %r1, %r5;
; SM60-NEXT: mov.b32 %r5, %r1;
; SM60-NEXT: @%p2 bra $L__BB64_1;
; SM60-NEXT: // %bb.2: // %atomicrmw.end
; SM60-NEXT: membar.cta;
; SM60-NEXT: st.param.b32 [func_retval0], %r1;
Expand All @@ -1871,30 +1865,24 @@ define float @fminimum_acq_rel_float_global_cta(ptr addrspace(1) %addr, float %v
define float @fmaximum_acq_rel_float_global_cta(ptr addrspace(1) %addr, float %val) {
; SM60-LABEL: fmaximum_acq_rel_float_global_cta(
; SM60: {
; SM60-NEXT: .reg .pred %p<6>;
; SM60-NEXT: .reg .b32 %r<9>;
; SM60-NEXT: .reg .pred %p<3>;
; SM60-NEXT: .reg .b32 %r<6>;
; SM60-NEXT: .reg .b64 %rd<2>;
; SM60-EMPTY:
; SM60-NEXT: // %bb.0:
; SM60-NEXT: ld.param.b32 %r2, [fmaximum_acq_rel_float_global_cta_param_1];
; SM60-NEXT: ld.param.b64 %rd1, [fmaximum_acq_rel_float_global_cta_param_0];
; SM60-NEXT: membar.cta;
; SM60-NEXT: ld.global.b32 %r8, [%rd1];
; SM60-NEXT: setp.eq.b32 %p3, %r2, 0;
; SM60-NEXT: ld.global.b32 %r5, [%rd1];
; SM60-NEXT: $L__BB65_1: // %atomicrmw.start
; SM60-NEXT: // =>This Inner Loop Header: Depth=1
; SM60-NEXT: setp.nan.f32 %p1, %r8, %r2;
; SM60-NEXT: max.f32 %r3, %r8, %r2;
; SM60-NEXT: setp.nan.f32 %p1, %r5, %r2;
; SM60-NEXT: max.f32 %r3, %r5, %r2;
; SM60-NEXT: selp.f32 %r4, 0f7FC00000, %r3, %p1;
; SM60-NEXT: setp.eq.b32 %p2, %r8, 0;
; SM60-NEXT: selp.f32 %r5, %r8, %r4, %p2;
; SM60-NEXT: selp.f32 %r6, %r2, %r5, %p3;
; SM60-NEXT: setp.eq.f32 %p4, %r4, 0f00000000;
; SM60-NEXT: selp.f32 %r7, %r6, %r4, %p4;
; SM60-NEXT: atom.cta.global.cas.b32 %r1, [%rd1], %r8, %r7;
; SM60-NEXT: setp.ne.b32 %p5, %r1, %r8;
; SM60-NEXT: mov.b32 %r8, %r1;
; SM60-NEXT: @%p5 bra $L__BB65_1;
; SM60-NEXT: atom.cta.global.cas.b32 %r1, [%rd1], %r5, %r4;
; SM60-NEXT: setp.ne.b32 %p2, %r1, %r5;
; SM60-NEXT: mov.b32 %r5, %r1;
; SM60-NEXT: @%p2 bra $L__BB65_1;
; SM60-NEXT: // %bb.2: // %atomicrmw.end
; SM60-NEXT: membar.cta;
; SM60-NEXT: st.param.b32 [func_retval0], %r1;
Expand Down Expand Up @@ -1999,29 +1987,23 @@ define double @fmax_acq_rel_double_global_cta(ptr addrspace(1) %addr, double %va
define double @fminimum_acq_rel_double_global_cta(ptr addrspace(1) %addr, double %val) {
; SM60-LABEL: fminimum_acq_rel_double_global_cta(
; SM60: {
; SM60-NEXT: .reg .pred %p<6>;
; SM60-NEXT: .reg .b64 %rd<10>;
; SM60-NEXT: .reg .pred %p<3>;
; SM60-NEXT: .reg .b64 %rd<7>;
; SM60-EMPTY:
; SM60-NEXT: // %bb.0:
; SM60-NEXT: ld.param.b64 %rd3, [fminimum_acq_rel_double_global_cta_param_1];
; SM60-NEXT: ld.param.b64 %rd2, [fminimum_acq_rel_double_global_cta_param_0];
; SM60-NEXT: membar.cta;
; SM60-NEXT: ld.global.b64 %rd9, [%rd2];
; SM60-NEXT: setp.eq.b64 %p3, %rd3, -9223372036854775808;
; SM60-NEXT: ld.global.b64 %rd6, [%rd2];
; SM60-NEXT: $L__BB70_1: // %atomicrmw.start
; SM60-NEXT: // =>This Inner Loop Header: Depth=1
; SM60-NEXT: setp.nan.f64 %p1, %rd9, %rd3;
; SM60-NEXT: min.f64 %rd4, %rd9, %rd3;
; SM60-NEXT: setp.nan.f64 %p1, %rd6, %rd3;
; SM60-NEXT: min.f64 %rd4, %rd6, %rd3;
; SM60-NEXT: selp.f64 %rd5, 0d7FF8000000000000, %rd4, %p1;
; SM60-NEXT: setp.eq.b64 %p2, %rd9, -9223372036854775808;
; SM60-NEXT: selp.f64 %rd6, %rd9, %rd5, %p2;
; SM60-NEXT: selp.f64 %rd7, %rd3, %rd6, %p3;
; SM60-NEXT: setp.eq.f64 %p4, %rd5, 0d0000000000000000;
; SM60-NEXT: selp.f64 %rd8, %rd7, %rd5, %p4;
; SM60-NEXT: atom.cta.global.cas.b64 %rd1, [%rd2], %rd9, %rd8;
; SM60-NEXT: setp.ne.b64 %p5, %rd1, %rd9;
; SM60-NEXT: mov.b64 %rd9, %rd1;
; SM60-NEXT: @%p5 bra $L__BB70_1;
; SM60-NEXT: atom.cta.global.cas.b64 %rd1, [%rd2], %rd6, %rd5;
; SM60-NEXT: setp.ne.b64 %p2, %rd1, %rd6;
; SM60-NEXT: mov.b64 %rd6, %rd1;
; SM60-NEXT: @%p2 bra $L__BB70_1;
; SM60-NEXT: // %bb.2: // %atomicrmw.end
; SM60-NEXT: membar.cta;
; SM60-NEXT: st.param.b64 [func_retval0], %rd1;
Expand All @@ -2033,29 +2015,23 @@ define double @fminimum_acq_rel_double_global_cta(ptr addrspace(1) %addr, double
define double @fmaximum_acq_rel_double_global_cta(ptr addrspace(1) %addr, double %val) {
; SM60-LABEL: fmaximum_acq_rel_double_global_cta(
; SM60: {
; SM60-NEXT: .reg .pred %p<6>;
; SM60-NEXT: .reg .b64 %rd<10>;
; SM60-NEXT: .reg .pred %p<3>;
; SM60-NEXT: .reg .b64 %rd<7>;
; SM60-EMPTY:
; SM60-NEXT: // %bb.0:
; SM60-NEXT: ld.param.b64 %rd3, [fmaximum_acq_rel_double_global_cta_param_1];
; SM60-NEXT: ld.param.b64 %rd2, [fmaximum_acq_rel_double_global_cta_param_0];
; SM60-NEXT: membar.cta;
; SM60-NEXT: ld.global.b64 %rd9, [%rd2];
; SM60-NEXT: setp.eq.b64 %p3, %rd3, 0;
; SM60-NEXT: ld.global.b64 %rd6, [%rd2];
; SM60-NEXT: $L__BB71_1: // %atomicrmw.start
; SM60-NEXT: // =>This Inner Loop Header: Depth=1
; SM60-NEXT: setp.nan.f64 %p1, %rd9, %rd3;
; SM60-NEXT: max.f64 %rd4, %rd9, %rd3;
; SM60-NEXT: setp.nan.f64 %p1, %rd6, %rd3;
; SM60-NEXT: max.f64 %rd4, %rd6, %rd3;
; SM60-NEXT: selp.f64 %rd5, 0d7FF8000000000000, %rd4, %p1;
; SM60-NEXT: setp.eq.b64 %p2, %rd9, 0;
; SM60-NEXT: selp.f64 %rd6, %rd9, %rd5, %p2;
; SM60-NEXT: selp.f64 %rd7, %rd3, %rd6, %p3;
; SM60-NEXT: setp.eq.f64 %p4, %rd5, 0d0000000000000000;
; SM60-NEXT: selp.f64 %rd8, %rd7, %rd5, %p4;
; SM60-NEXT: atom.cta.global.cas.b64 %rd1, [%rd2], %rd9, %rd8;
; SM60-NEXT: setp.ne.b64 %p5, %rd1, %rd9;
; SM60-NEXT: mov.b64 %rd9, %rd1;
; SM60-NEXT: @%p5 bra $L__BB71_1;
; SM60-NEXT: atom.cta.global.cas.b64 %rd1, [%rd2], %rd6, %rd5;
; SM60-NEXT: setp.ne.b64 %p2, %rd1, %rd6;
; SM60-NEXT: mov.b64 %rd6, %rd1;
; SM60-NEXT: @%p2 bra $L__BB71_1;
; SM60-NEXT: // %bb.2: // %atomicrmw.end
; SM60-NEXT: membar.cta;
; SM60-NEXT: st.param.b64 [func_retval0], %rd1;
Expand Down
Loading