Skip to content

[VP] Detect truncated and shifted EVLs during expansion #154334

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

Open
wants to merge 3 commits into
base: main
Choose a base branch
from
+119 −1
Open

[VP] Detect truncated and shifted EVLs during expansion #154334

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
de961fd
Precommit tests
lukel97 Aug 19, 2025
b13a35a
[VP] Detect truncated and shifted EVLs during expansion
lukel97 Aug 19, 2025
0a61890
Check mul/shl is nuw
lukel97 Aug 26, 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
7 changes: 7 additions & 0 deletions llvm/include/llvm/IR/PatternMatch.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2193,6 +2193,13 @@ m_TruncOrSelf(const OpTy &Op) {
return m_CombineOr(m_Trunc(Op), Op);
}

template <typename OpTy>
inline match_combine_or<NoWrapTrunc_match<OpTy, TruncInst::NoUnsignedWrap>,
OpTy>
m_NUWTruncOrSelf(const OpTy &Op) {
return m_CombineOr(m_NUWTrunc(Op), Op);
}

/// Matches SExt.
template <typename OpTy>
inline CastInst_match<OpTy, SExtInst> m_SExt(const OpTy &Op) {
Expand Down
6 changes: 5 additions & 1 deletion llvm/lib/IR/IntrinsicInst.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -613,8 +613,12 @@ bool VPIntrinsic::canIgnoreVectorLengthParam() const {
if (EC.isScalable()) {
// Compare vscale patterns
uint64_t VScaleFactor;
if (match(VLParam, m_Mul(m_VScale(), m_ConstantInt(VScaleFactor))))
if (match(VLParam, m_NUWTruncOrSelf(
m_NUWMul(m_VScale(), m_ConstantInt(VScaleFactor)))))
return VScaleFactor >= EC.getKnownMinValue();
if (match(VLParam, m_NUWTruncOrSelf(
m_NUWShl(m_VScale(), m_ConstantInt(VScaleFactor)))))
return 1 << VScaleFactor >= EC.getKnownMinValue();
return (EC.getKnownMinValue() == 1) && match(VLParam, m_VScale());
}

Expand Down
107 changes: 107 additions & 0 deletions llvm/test/Transforms/PreISelIntrinsicLowering/expand-vp-convert-evl.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,107 @@
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
; RUN: opt -passes=pre-isel-intrinsic-lowering -expandvp-override-evl-transform=Convert -expandvp-override-mask-transform=Legal -S < %s | FileCheck %s

define <vscale x 2 x i64> @unknown_evl(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m, i32 %evl) {
; CHECK-LABEL: define <vscale x 2 x i64> @unknown_evl(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]], i32 [[EVL:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i32(i32 0, i32 [[EVL]])
; CHECK-NEXT: [[TMP2:%.*]] = and <vscale x 2 x i1> [[TMP1]], [[M]]
; CHECK-NEXT: [[VSCALE:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-NEXT: [[SCALABLE_SIZE:%.*]] = mul nuw i32 [[VSCALE]], 2
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[TMP2]], i32 [[SCALABLE_SIZE]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}

define <2 x i64> @exact_evl_fixed(<2 x i64> %x, <2 x i64> %y, <2 x i1> %m) {
; CHECK-LABEL: define <2 x i64> @exact_evl_fixed(
; CHECK-SAME: <2 x i64> [[X:%.*]], <2 x i64> [[Y:%.*]], <2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[ADD:%.*]] = call <2 x i64> @llvm.vp.add.v2i64(<2 x i64> poison, <2 x i64> poison, <2 x i1> [[M]], i32 2)
; CHECK-NEXT: ret <2 x i64> [[ADD]]
;
%add = call <2 x i64> @llvm.vp.add(<2 x i64> poison, <2 x i64> poison, <2 x i1> %m, i32 2)
ret <2 x i64> %add
}

define <vscale x 2 x i64> @exact_evl_vscale_mul(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m) {
; CHECK-LABEL: define <vscale x 2 x i64> @exact_evl_vscale_mul(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[VSCALE:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-NEXT: [[EVL:%.*]] = mul nuw i32 [[VSCALE]], 2
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[M]], i32 [[EVL]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%vscale = call i32 @llvm.vscale()
%evl = mul nuw i32 %vscale, 2
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}

define <vscale x 2 x i64> @exact_evl_vscale_shl(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m) {
; CHECK-LABEL: define <vscale x 2 x i64> @exact_evl_vscale_shl(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[VSCALE:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-NEXT: [[EVL:%.*]] = shl nuw i32 [[VSCALE]], 1
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[M]], i32 [[EVL]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%vscale = call i32 @llvm.vscale()
%evl = shl nuw i32 %vscale, 1
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}

define <vscale x 2 x i64> @exact_evl_vscale_mul_trunc(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m) {
; CHECK-LABEL: define <vscale x 2 x i64> @exact_evl_vscale_mul_trunc(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[SHL:%.*]] = mul nuw i64 [[VSCALE]], 2
; CHECK-NEXT: [[EVL:%.*]] = trunc nuw i64 [[SHL]] to i32
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[M]], i32 [[EVL]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%vscale = call i64 @llvm.vscale()
%shl = mul nuw i64 %vscale, 2
%evl = trunc nuw i64 %shl to i32
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}


define <vscale x 2 x i64> @exact_evl_vscale_shl_trunc(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m) {
; CHECK-LABEL: define <vscale x 2 x i64> @exact_evl_vscale_shl_trunc(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[SHL:%.*]] = shl nuw i64 [[VSCALE]], 1
; CHECK-NEXT: [[EVL:%.*]] = trunc nuw i64 [[SHL]] to i32
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[M]], i32 [[EVL]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%vscale = call i64 @llvm.vscale()
%shl = shl nuw i64 %vscale, 1
%evl = trunc nuw i64 %shl to i32
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}

define <vscale x 2 x i64> @exact_evl_vscale_shl_trunc_no_nuw(<vscale x 2 x i64> %x, <vscale x 2 x i64> %y, <vscale x 2 x i1> %m) {
; CHECK-LABEL: define <vscale x 2 x i64> @exact_evl_vscale_shl_trunc_no_nuw(
; CHECK-SAME: <vscale x 2 x i64> [[X:%.*]], <vscale x 2 x i64> [[Y:%.*]], <vscale x 2 x i1> [[M:%.*]]) {
; CHECK-NEXT: [[VSCALE:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[SHL:%.*]] = shl nuw i64 [[VSCALE]], 1
; CHECK-NEXT: [[EVL:%.*]] = trunc i64 [[SHL]] to i32
; CHECK-NEXT: [[TMP1:%.*]] = call <vscale x 2 x i1> @llvm.get.active.lane.mask.nxv2i1.i32(i32 0, i32 [[EVL]])
; CHECK-NEXT: [[TMP2:%.*]] = and <vscale x 2 x i1> [[TMP1]], [[M]]
; CHECK-NEXT: [[VSCALE1:%.*]] = call i32 @llvm.vscale.i32()
; CHECK-NEXT: [[SCALABLE_SIZE:%.*]] = mul nuw i32 [[VSCALE1]], 2
; CHECK-NEXT: [[ADD:%.*]] = call <vscale x 2 x i64> @llvm.vp.add.nxv2i64(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> [[TMP2]], i32 [[SCALABLE_SIZE]])
; CHECK-NEXT: ret <vscale x 2 x i64> [[ADD]]
;
%vscale = call i64 @llvm.vscale()
%shl = shl nuw i64 %vscale, 1
%evl = trunc i64 %shl to i32
%add = call <vscale x 2 x i64> @llvm.vp.add(<vscale x 2 x i64> poison, <vscale x 2 x i64> poison, <vscale x 2 x i1> %m, i32 %evl)
ret <vscale x 2 x i64> %add
}
Loading