Skip to content

[DAGCombiner] Fold umax/umin operations with vscale operands #154461

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
tclin914 merged 3 commits into from
Aug 21, 2025
Merged

[DAGCombiner] Fold umax/umin operations with vscale operands #154461

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
79ea0c1
[DAGCombiner] Fold umax/umin operations with vscale operands
tclin914 Aug 19, 2025
ee35f21
fold -> Fold
tclin914 Aug 20, 2025
737be7d
Remove `vscale_range(2,1024)`
tclin914 Aug 20, 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
10 changes: 10 additions & 0 deletions llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6070,6 +6070,16 @@ SDValue DAGCombiner::visitIMINMAX(SDNode *N) {
if (N0 == N1)
return N0;

// Fold operation with vscale operands.
Copy link
Contributor

@lukel97 lukel97 Aug 20, 2025

Choose a reason for hiding this comment

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

Suggested change
// fold operation with vscale operands.
// Fold operation with vscale operands.

if (N0.getOpcode() == ISD::VSCALE && N1.getOpcode() == ISD::VSCALE) {
uint64_t C0 = N0->getConstantOperandVal(0);
uint64_t C1 = N1->getConstantOperandVal(0);
if (Opcode == ISD::UMAX)
return C0 > C1 ? N0 : N1;
else if (Opcode == ISD::UMIN)
return C0 > C1 ? N1 : N0;
}

// canonicalize constant to RHS
if (DAG.isConstantIntBuildVectorOrConstantInt(N0) &&
!DAG.isConstantIntBuildVectorOrConstantInt(N1))
Expand Down
20 changes: 8 additions & 12 deletions llvm/test/CodeGen/RISCV/rvv/strided-vpload.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -939,21 +939,17 @@ define <vscale x 16 x i64> @zero_strided_vadd_nxv16i64(<vscale x 16 x i64> %v, p
; CHECK-RV32: # %bb.0:
; CHECK-RV32-NEXT: csrr a1, vlenb
; CHECK-RV32-NEXT: srli a2, a1, 3
; CHECK-RV32-NEXT: sub a3, a2, a1
; CHECK-RV32-NEXT: sltu a4, a2, a3
; CHECK-RV32-NEXT: addi a4, a4, -1
; CHECK-RV32-NEXT: and a3, a4, a3
; CHECK-RV32-NEXT: vsetvli zero, a3, e64, m8, ta, ma
; CHECK-RV32-NEXT: vlse64.v v24, (a0), zero
; CHECK-RV32-NEXT: bltu a2, a1, .LBB61_2
; CHECK-RV32-NEXT: # %bb.1:
; CHECK-RV32-NEXT: mv a2, a1
; CHECK-RV32-NEXT: .LBB61_2:
; CHECK-RV32-NEXT: sub a1, a2, a1
; CHECK-RV32-NEXT: sltu a3, a2, a1
; CHECK-RV32-NEXT: addi a3, a3, -1
; CHECK-RV32-NEXT: and a1, a3, a1
; CHECK-RV32-NEXT: vsetvli zero, a2, e64, m8, ta, ma
; CHECK-RV32-NEXT: vlse64.v v24, (a0), zero
; CHECK-RV32-NEXT: vsetvli zero, a1, e64, m8, ta, ma
; CHECK-RV32-NEXT: vlse64.v v0, (a0), zero
; CHECK-RV32-NEXT: vsetvli a0, zero, e64, m8, ta, ma
; CHECK-RV32-NEXT: vadd.vv v16, v16, v24
; CHECK-RV32-NEXT: vadd.vv v8, v8, v0
; CHECK-RV32-NEXT: vadd.vv v8, v8, v24
; CHECK-RV32-NEXT: vadd.vv v16, v16, v0
; CHECK-RV32-NEXT: ret
;
; CHECK-RV64-LABEL: zero_strided_vadd_nxv16i64:
Expand Down
64 changes: 39 additions & 25 deletions llvm/test/CodeGen/RISCV/rvv/vadd-vp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1393,32 +1393,46 @@ define <vscale x 32 x i32> @vadd_vi_nxv32i32_unmasked(<vscale x 32 x i32> %va, i

declare i32 @llvm.vscale.i32()

; FIXME: The upper half of the operation is doing nothing.
; FIXME: The branches comparing vscale vs. vscale should be constant-foldable.

define <vscale x 32 x i32> @vadd_vi_nxv32i32_evl_nx8(<vscale x 32 x i32> %va, <vscale x 32 x i1> %m) {
; CHECK-LABEL: vadd_vi_nxv32i32_evl_nx8:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a0, zero, e8, mf2, ta, ma
; CHECK-NEXT: vmv1r.v v24, v0
; CHECK-NEXT: csrr a0, vlenb
; CHECK-NEXT: srli a2, a0, 2
; CHECK-NEXT: slli a1, a0, 1
; CHECK-NEXT: vslidedown.vx v0, v0, a2
; CHECK-NEXT: sub a2, a0, a1
; CHECK-NEXT: sltu a3, a0, a2
; CHECK-NEXT: addi a3, a3, -1
; CHECK-NEXT: and a2, a3, a2
; CHECK-NEXT: vsetvli zero, a2, e32, m8, ta, ma
; CHECK-NEXT: vadd.vi v16, v16, -1, v0.t
; CHECK-NEXT: bltu a0, a1, .LBB120_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: mv a0, a1
; CHECK-NEXT: .LBB120_2:
; CHECK-NEXT: vmv1r.v v0, v24
; CHECK-NEXT: vsetvli zero, a0, e32, m8, ta, ma
; CHECK-NEXT: vadd.vi v8, v8, -1, v0.t
; CHECK-NEXT: ret
; RV32-LABEL: vadd_vi_nxv32i32_evl_nx8:
; RV32: # %bb.0:
; RV32-NEXT: csrr a0, vlenb
; RV32-NEXT: srli a1, a0, 2
; RV32-NEXT: vsetvli zero, a0, e32, m8, ta, ma
; RV32-NEXT: vadd.vi v8, v8, -1, v0.t
; RV32-NEXT: vsetvli a2, zero, e8, mf2, ta, ma
; RV32-NEXT: vslidedown.vx v0, v0, a1
; RV32-NEXT: slli a1, a0, 1
; RV32-NEXT: sub a1, a0, a1
; RV32-NEXT: sltu a0, a0, a1
; RV32-NEXT: addi a0, a0, -1
; RV32-NEXT: and a0, a0, a1
; RV32-NEXT: vsetvli zero, a0, e32, m8, ta, ma
; RV32-NEXT: vadd.vi v16, v16, -1, v0.t
; RV32-NEXT: ret
;
; RV64-LABEL: vadd_vi_nxv32i32_evl_nx8:
; RV64: # %bb.0:
; RV64-NEXT: vsetvli a0, zero, e8, mf2, ta, ma
; RV64-NEXT: vmv1r.v v24, v0
; RV64-NEXT: csrr a0, vlenb
; RV64-NEXT: srli a2, a0, 2
; RV64-NEXT: slli a1, a0, 1
; RV64-NEXT: vslidedown.vx v0, v0, a2
; RV64-NEXT: sub a2, a0, a1
; RV64-NEXT: sltu a3, a0, a2
; RV64-NEXT: addi a3, a3, -1
; RV64-NEXT: and a2, a3, a2
; RV64-NEXT: vsetvli zero, a2, e32, m8, ta, ma
; RV64-NEXT: vadd.vi v16, v16, -1, v0.t
; RV64-NEXT: bltu a0, a1, .LBB120_2
; RV64-NEXT: # %bb.1:
; RV64-NEXT: mv a0, a1
; RV64-NEXT: .LBB120_2:
; RV64-NEXT: vmv1r.v v0, v24
; RV64-NEXT: vsetvli zero, a0, e32, m8, ta, ma
; RV64-NEXT: vadd.vi v8, v8, -1, v0.t
; RV64-NEXT: ret
%evl = call i32 @llvm.vscale.i32()
%evl0 = mul i32 %evl, 8
%v = call <vscale x 32 x i32> @llvm.vp.add.nxv32i32(<vscale x 32 x i32> %va, <vscale x 32 x i32> splat (i32 -1), <vscale x 32 x i1> %m, i32 %evl0)
Expand Down
64 changes: 39 additions & 25 deletions llvm/test/CodeGen/RISCV/rvv/vmax-vp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1026,32 +1026,46 @@ define <vscale x 32 x i32> @vmax_vx_nxv32i32_unmasked(<vscale x 32 x i32> %va, i

declare i32 @llvm.vscale.i32()

; FIXME: The upper half of the operation is doing nothing.
; FIXME: The branches comparing vscale vs. vscale should be constant-foldable.

define <vscale x 32 x i32> @vmax_vx_nxv32i32_evl_nx8(<vscale x 32 x i32> %va, i32 %b, <vscale x 32 x i1> %m) {
; CHECK-LABEL: vmax_vx_nxv32i32_evl_nx8:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; CHECK-NEXT: vmv1r.v v24, v0
; CHECK-NEXT: csrr a1, vlenb
; CHECK-NEXT: srli a3, a1, 2
; CHECK-NEXT: slli a2, a1, 1
; CHECK-NEXT: vslidedown.vx v0, v0, a3
; CHECK-NEXT: sub a3, a1, a2
; CHECK-NEXT: sltu a4, a1, a3
; CHECK-NEXT: addi a4, a4, -1
; CHECK-NEXT: and a3, a4, a3
; CHECK-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; CHECK-NEXT: vmax.vx v16, v16, a0, v0.t
; CHECK-NEXT: bltu a1, a2, .LBB82_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: mv a1, a2
; CHECK-NEXT: .LBB82_2:
; CHECK-NEXT: vmv1r.v v0, v24
; CHECK-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; CHECK-NEXT: vmax.vx v8, v8, a0, v0.t
; CHECK-NEXT: ret
; RV32-LABEL: vmax_vx_nxv32i32_evl_nx8:
; RV32: # %bb.0:
; RV32-NEXT: csrr a1, vlenb
; RV32-NEXT: srli a2, a1, 2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmax.vx v8, v8, a0, v0.t
; RV32-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
; RV32-NEXT: vslidedown.vx v0, v0, a2
; RV32-NEXT: slli a2, a1, 1
; RV32-NEXT: sub a2, a1, a2
; RV32-NEXT: sltu a1, a1, a2
; RV32-NEXT: addi a1, a1, -1
; RV32-NEXT: and a1, a1, a2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmax.vx v16, v16, a0, v0.t
; RV32-NEXT: ret
;
; RV64-LABEL: vmax_vx_nxv32i32_evl_nx8:
; RV64: # %bb.0:
; RV64-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; RV64-NEXT: vmv1r.v v24, v0
; RV64-NEXT: csrr a1, vlenb
; RV64-NEXT: srli a3, a1, 2
; RV64-NEXT: slli a2, a1, 1
; RV64-NEXT: vslidedown.vx v0, v0, a3
; RV64-NEXT: sub a3, a1, a2
; RV64-NEXT: sltu a4, a1, a3
; RV64-NEXT: addi a4, a4, -1
; RV64-NEXT: and a3, a4, a3
; RV64-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; RV64-NEXT: vmax.vx v16, v16, a0, v0.t
; RV64-NEXT: bltu a1, a2, .LBB82_2
; RV64-NEXT: # %bb.1:
; RV64-NEXT: mv a1, a2
; RV64-NEXT: .LBB82_2:
; RV64-NEXT: vmv1r.v v0, v24
; RV64-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV64-NEXT: vmax.vx v8, v8, a0, v0.t
; RV64-NEXT: ret
%elt.head = insertelement <vscale x 32 x i32> poison, i32 %b, i32 0
%vb = shufflevector <vscale x 32 x i32> %elt.head, <vscale x 32 x i32> poison, <vscale x 32 x i32> zeroinitializer
%evl = call i32 @llvm.vscale.i32()
Expand Down
64 changes: 39 additions & 25 deletions llvm/test/CodeGen/RISCV/rvv/vmaxu-vp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1025,32 +1025,46 @@ define <vscale x 32 x i32> @vmaxu_vx_nxv32i32_unmasked(<vscale x 32 x i32> %va,

declare i32 @llvm.vscale.i32()

; FIXME: The upper half of the operation is doing nothing.
; FIXME: The branches comparing vscale vs. vscale should be constant-foldable.

define <vscale x 32 x i32> @vmaxu_vx_nxv32i32_evl_nx8(<vscale x 32 x i32> %va, i32 %b, <vscale x 32 x i1> %m) {
; CHECK-LABEL: vmaxu_vx_nxv32i32_evl_nx8:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; CHECK-NEXT: vmv1r.v v24, v0
; CHECK-NEXT: csrr a1, vlenb
; CHECK-NEXT: srli a3, a1, 2
; CHECK-NEXT: slli a2, a1, 1
; CHECK-NEXT: vslidedown.vx v0, v0, a3
; CHECK-NEXT: sub a3, a1, a2
; CHECK-NEXT: sltu a4, a1, a3
; CHECK-NEXT: addi a4, a4, -1
; CHECK-NEXT: and a3, a4, a3
; CHECK-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; CHECK-NEXT: vmaxu.vx v16, v16, a0, v0.t
; CHECK-NEXT: bltu a1, a2, .LBB82_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: mv a1, a2
; CHECK-NEXT: .LBB82_2:
; CHECK-NEXT: vmv1r.v v0, v24
; CHECK-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; CHECK-NEXT: vmaxu.vx v8, v8, a0, v0.t
; CHECK-NEXT: ret
; RV32-LABEL: vmaxu_vx_nxv32i32_evl_nx8:
; RV32: # %bb.0:
; RV32-NEXT: csrr a1, vlenb
; RV32-NEXT: srli a2, a1, 2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmaxu.vx v8, v8, a0, v0.t
; RV32-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
; RV32-NEXT: vslidedown.vx v0, v0, a2
; RV32-NEXT: slli a2, a1, 1
; RV32-NEXT: sub a2, a1, a2
; RV32-NEXT: sltu a1, a1, a2
; RV32-NEXT: addi a1, a1, -1
; RV32-NEXT: and a1, a1, a2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmaxu.vx v16, v16, a0, v0.t
; RV32-NEXT: ret
;
; RV64-LABEL: vmaxu_vx_nxv32i32_evl_nx8:
; RV64: # %bb.0:
; RV64-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; RV64-NEXT: vmv1r.v v24, v0
; RV64-NEXT: csrr a1, vlenb
; RV64-NEXT: srli a3, a1, 2
; RV64-NEXT: slli a2, a1, 1
; RV64-NEXT: vslidedown.vx v0, v0, a3
; RV64-NEXT: sub a3, a1, a2
; RV64-NEXT: sltu a4, a1, a3
; RV64-NEXT: addi a4, a4, -1
; RV64-NEXT: and a3, a4, a3
; RV64-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; RV64-NEXT: vmaxu.vx v16, v16, a0, v0.t
; RV64-NEXT: bltu a1, a2, .LBB82_2
; RV64-NEXT: # %bb.1:
; RV64-NEXT: mv a1, a2
; RV64-NEXT: .LBB82_2:
; RV64-NEXT: vmv1r.v v0, v24
; RV64-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV64-NEXT: vmaxu.vx v8, v8, a0, v0.t
; RV64-NEXT: ret
%elt.head = insertelement <vscale x 32 x i32> poison, i32 %b, i32 0
%vb = shufflevector <vscale x 32 x i32> %elt.head, <vscale x 32 x i32> poison, <vscale x 32 x i32> zeroinitializer
%evl = call i32 @llvm.vscale.i32()
Expand Down
64 changes: 39 additions & 25 deletions llvm/test/CodeGen/RISCV/rvv/vmin-vp.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1026,32 +1026,46 @@ define <vscale x 32 x i32> @vmin_vx_nxv32i32_unmasked(<vscale x 32 x i32> %va, i

declare i32 @llvm.vscale.i32()

; FIXME: The upper half of the operation is doing nothing.
; FIXME: The branches comparing vscale vs. vscale should be constant-foldable.

define <vscale x 32 x i32> @vmin_vx_nxv32i32_evl_nx8(<vscale x 32 x i32> %va, i32 %b, <vscale x 32 x i1> %m) {
; CHECK-LABEL: vmin_vx_nxv32i32_evl_nx8:
; CHECK: # %bb.0:
; CHECK-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; CHECK-NEXT: vmv1r.v v24, v0
; CHECK-NEXT: csrr a1, vlenb
; CHECK-NEXT: srli a3, a1, 2
; CHECK-NEXT: slli a2, a1, 1
; CHECK-NEXT: vslidedown.vx v0, v0, a3
; CHECK-NEXT: sub a3, a1, a2
; CHECK-NEXT: sltu a4, a1, a3
; CHECK-NEXT: addi a4, a4, -1
; CHECK-NEXT: and a3, a4, a3
; CHECK-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; CHECK-NEXT: vmin.vx v16, v16, a0, v0.t
; CHECK-NEXT: bltu a1, a2, .LBB82_2
; CHECK-NEXT: # %bb.1:
; CHECK-NEXT: mv a1, a2
; CHECK-NEXT: .LBB82_2:
; CHECK-NEXT: vmv1r.v v0, v24
; CHECK-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; CHECK-NEXT: vmin.vx v8, v8, a0, v0.t
; CHECK-NEXT: ret
; RV32-LABEL: vmin_vx_nxv32i32_evl_nx8:
; RV32: # %bb.0:
; RV32-NEXT: csrr a1, vlenb
; RV32-NEXT: srli a2, a1, 2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmin.vx v8, v8, a0, v0.t
; RV32-NEXT: vsetvli a3, zero, e8, mf2, ta, ma
; RV32-NEXT: vslidedown.vx v0, v0, a2
; RV32-NEXT: slli a2, a1, 1
; RV32-NEXT: sub a2, a1, a2
; RV32-NEXT: sltu a1, a1, a2
; RV32-NEXT: addi a1, a1, -1
; RV32-NEXT: and a1, a1, a2
; RV32-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV32-NEXT: vmin.vx v16, v16, a0, v0.t
; RV32-NEXT: ret
;
; RV64-LABEL: vmin_vx_nxv32i32_evl_nx8:
; RV64: # %bb.0:
; RV64-NEXT: vsetvli a1, zero, e8, mf2, ta, ma
; RV64-NEXT: vmv1r.v v24, v0
; RV64-NEXT: csrr a1, vlenb
; RV64-NEXT: srli a3, a1, 2
; RV64-NEXT: slli a2, a1, 1
; RV64-NEXT: vslidedown.vx v0, v0, a3
; RV64-NEXT: sub a3, a1, a2
; RV64-NEXT: sltu a4, a1, a3
; RV64-NEXT: addi a4, a4, -1
; RV64-NEXT: and a3, a4, a3
; RV64-NEXT: vsetvli zero, a3, e32, m8, ta, ma
; RV64-NEXT: vmin.vx v16, v16, a0, v0.t
; RV64-NEXT: bltu a1, a2, .LBB82_2
; RV64-NEXT: # %bb.1:
; RV64-NEXT: mv a1, a2
; RV64-NEXT: .LBB82_2:
; RV64-NEXT: vmv1r.v v0, v24
; RV64-NEXT: vsetvli zero, a1, e32, m8, ta, ma
; RV64-NEXT: vmin.vx v8, v8, a0, v0.t
; RV64-NEXT: ret
%elt.head = insertelement <vscale x 32 x i32> poison, i32 %b, i32 0
%vb = shufflevector <vscale x 32 x i32> %elt.head, <vscale x 32 x i32> poison, <vscale x 32 x i32> zeroinitializer
%evl = call i32 @llvm.vscale.i32()
Expand Down
Loading