Skip to content

[LLVM][DAGCombiner] fold (shl (X * vscale(C0)), C1) -> (X * vscale(C0 << C1)). #150651

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 3 commits into from
Aug 1, 2025
Merged

[LLVM][DAGCombiner] fold (shl (X * vscale(C0)), C1) -> (X * vscale(C0 << C1)). #150651

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
568aa4a
Regeneate sve-vscale-combine.ll. Add tests to show missing combine.
paulwalker-arm Jul 25, 2025
d921291
[LLVM][DAGCombiner] fold (shl (X * vscale(C0)), C1) -> (X * vscale(C0…
paulwalker-arm Jul 25, 2025
13d8d4e
Prefer NUW flags as per https://alive2.llvm.org/ce/z/_V5s8p
paulwalker-arm Jul 31, 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
9 changes: 9 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 @@ -10615,6 +10615,15 @@ SDValue DAGCombiner::visitSHL(SDNode *N) {
return DAG.getVScale(DL, VT, C0 << C1);
}

SDValue X;
APInt VS0;

// fold (shl (X * vscale(VS0)), C1) -> (X * vscale(VS0 << C1))
if (N1C && sd_match(N0, m_Mul(m_Value(X), m_VScale(m_ConstInt(VS0))))) {
SDValue VScale = DAG.getVScale(DL, VT, VS0 << N1C->getAPIntValue());
return DAG.getNode(ISD::MUL, DL, VT, X, VScale);
}

// Fold (shl step_vector(C0), C1) to (step_vector(C0 << C1)).
APInt ShlVal;
if (N0.getOpcode() == ISD::STEP_VECTOR &&
Expand Down
105 changes: 68 additions & 37 deletions llvm/test/CodeGen/AArch64/sve-vscale-combine.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,24 +1,25 @@
; RUN: llc -mtriple=aarch64--linux-gnu -mattr=+sve --asm-verbose=false < %s |FileCheck %s
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mattr=+sve < %s | FileCheck %s

declare i32 @llvm.vscale.i32()
declare i64 @llvm.vscale.i64()
target triple = "aarch64-unknown-linux-gnu"

; Fold (add (vscale * C0), (vscale * C1)) to (vscale * (C0 + C1)).
define i64 @combine_add_vscale_i64() nounwind {
; CHECK-LABEL: combine_add_vscale_i64:
; CHECK-NOT: add
; CHECK-NEXT: cntd x0
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: cntd x0
; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%add = add i64 %vscale, %vscale
ret i64 %add
}

define i32 @combine_add_vscale_i32() nounwind {
; CHECK-LABEL: combine_add_vscale_i32:
; CHECK-NOT: add
; CHECK-NEXT: cntd x0
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: cntd x0
; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%add = add i32 %vscale, %vscale
ret i32 %add
Expand All @@ -28,19 +29,20 @@ define i32 @combine_add_vscale_i32() nounwind {
; In this test, C0 = 1, C1 = 32.
define i64 @combine_mul_vscale_i64() nounwind {
; CHECK-LABEL: combine_mul_vscale_i64:
; CHECK-NOT: mul
; CHECK-NEXT: rdvl x0, #2
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x0, #2
; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%mul = mul i64 %vscale, 32
ret i64 %mul
}

define i32 @combine_mul_vscale_i32() nounwind {
; CHECK-LABEL: combine_mul_vscale_i32:
; CHECK-NOT: mul
; CHECK-NEXT: rdvl x0, #3
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x0, #3
; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%mul = mul i32 %vscale, 48
ret i32 %mul
Expand All @@ -49,23 +51,23 @@ define i32 @combine_mul_vscale_i32() nounwind {
; Canonicalize (sub X, (vscale * C)) to (add X, (vscale * -C))
define i64 @combine_sub_vscale_i64(i64 %in) nounwind {
; CHECK-LABEL: combine_sub_vscale_i64:
; CHECK-NOT: sub
; CHECK-NEXT: rdvl x8, #-1
; CHECK-NEXT: asr x8, x8, #4
; CHECK-NEXT: add x0, x0, x8
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x8, #-1
; CHECK-NEXT: asr x8, x8, #4
; CHECK-NEXT: add x0, x0, x8
; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%sub = sub i64 %in, %vscale
ret i64 %sub
}

define i32 @combine_sub_vscale_i32(i32 %in) nounwind {
; CHECK-LABEL: combine_sub_vscale_i32:
; CHECK-NOT: sub
; CHECK-NEXT: rdvl x8, #-1
; CHECK-NEXT: asr x8, x8, #4
; CHECK-NEXT: add w0, w0, w8
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x8, #-1
; CHECK-NEXT: asr x8, x8, #4
; CHECK-NEXT: add w0, w0, w8
; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%sub = sub i32 %in, %vscale
ret i32 %sub
Expand All @@ -75,12 +77,13 @@ define i32 @combine_sub_vscale_i32(i32 %in) nounwind {
; (sub X, (vscale * C)) to (add X, (vscale * -C))
define i64 @multiple_uses_sub_vscale_i64(i64 %x, i64 %y) nounwind {
; CHECK-LABEL: multiple_uses_sub_vscale_i64:
; CHECK-NEXT: rdvl x8, #1
; CHECK-NEXT: lsr x8, x8, #4
; CHECK-NEXT: sub x9, x0, x8
; CHECK-NEXT: add x8, x1, x8
; CHECK-NEXT: mul x0, x9, x8
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x8, #1
; CHECK-NEXT: lsr x8, x8, #4
; CHECK-NEXT: sub x9, x0, x8
; CHECK-NEXT: add x8, x1, x8
; CHECK-NEXT: mul x0, x9, x8
; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%sub = sub i64 %x, %vscale
%add = add i64 %y, %vscale
Expand All @@ -95,20 +98,48 @@ define i64 @multiple_uses_sub_vscale_i64(i64 %x, i64 %y) nounwind {
; Hence, the immediate for RDVL is #1.
define i64 @combine_shl_vscale_i64() nounwind {
; CHECK-LABEL: combine_shl_vscale_i64:
; CHECK-NOT: shl
; CHECK-NEXT: rdvl x0, #1
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x0, #1
; CHECK-NEXT: ret
%vscale = call i64 @llvm.vscale.i64()
%shl = shl i64 %vscale, 4
ret i64 %shl
}

define i32 @combine_shl_vscale_i32() nounwind {
; CHECK-LABEL: combine_shl_vscale_i32:
; CHECK-NOT: shl
; CHECK-NEXT: rdvl x0, #1
; CHECK-NEXT: ret
; CHECK: // %bb.0:
; CHECK-NEXT: rdvl x0, #1
; CHECK-NEXT: // kill: def $w0 killed $w0 killed $x0
; CHECK-NEXT: ret
%vscale = call i32 @llvm.vscale.i32()
%shl = shl i32 %vscale, 4
ret i32 %shl
}

define i64 @combine_shl_mul_vscale(i64 %a) nounwind {
; CHECK-LABEL: combine_shl_mul_vscale:
; CHECK: // %bb.0:
; CHECK-NEXT: cnth x8
; CHECK-NEXT: mul x0, x0, x8
; CHECK-NEXT: ret
%vscale = tail call i64 @llvm.vscale.i64()
%mul = mul i64 %a, %vscale
%shl = shl i64 %mul, 3
ret i64 %shl
}

define i64 @combine_shl_mul_vscale_commuted(i64 %a) nounwind {
; CHECK-LABEL: combine_shl_mul_vscale_commuted:
; CHECK: // %bb.0:
; CHECK-NEXT: cnth x8
; CHECK-NEXT: mul x0, x0, x8
; CHECK-NEXT: ret
%vscale = tail call i64 @llvm.vscale.i64()
%mul = mul i64 %vscale, %a
%shl = shl i64 %mul, 3
ret i64 %shl
}

declare i32 @llvm.vscale.i32()
declare i64 @llvm.vscale.i64()
8 changes: 4 additions & 4 deletions llvm/test/Transforms/LoopStrengthReduce/AArch64/vscale-fixups.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -383,14 +383,14 @@ define void @vscale_squared_offset(ptr %alloc) #0 {
; COMMON-LABEL: vscale_squared_offset:
; COMMON: // %bb.0: // %entry
; COMMON-NEXT: rdvl x9, #1
; COMMON-NEXT: rdvl x10, #4
; COMMON-NEXT: fmov z0.s, #4.00000000
; COMMON-NEXT: mov x8, xzr
; COMMON-NEXT: lsr x9, x9, #4
; COMMON-NEXT: fmov z1.s, #8.00000000
; COMMON-NEXT: cntw x10
; COMMON-NEXT: mov x8, xzr
; COMMON-NEXT: ptrue p0.s, vl1
; COMMON-NEXT: umull x9, w9, w9
; COMMON-NEXT: lsl x9, x9, #6
; COMMON-NEXT: umull x9, w9, w10
; COMMON-NEXT: cntw x10
; COMMON-NEXT: cmp x8, x10
; COMMON-NEXT: b.ge .LBB6_2
; COMMON-NEXT: .LBB6_1: // %for.body
Expand Down