Florian Hahn 7c34848ae1
[VPlan] Hoist loads with invariant addresses using noalias metadata. (#166247)
This patch implements a transform to hoists single-scalar replicated
loads with invariant addresses out of the vector loop to the preheader
when scoped noalias metadata proves they cannot alias with any stores in
the loop.

This enables hosting of loads we can prove do not alias any stores in
the loop due to memory runtime checks added during vectorization.

PR: https://github.com/llvm/llvm-project/pull/166247
2025-11-18 09:35:48 +00:00

77 lines
4.3 KiB
LLVM

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt -passes=loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -S < %s | FileCheck %s
define void @m(ptr nocapture %p, ptr nocapture %p2, i32 %q) {
; CHECK-LABEL: @m(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[ARRAYIDX9_1:%.*]] = getelementptr inbounds i32, ptr [[P2:%.*]], i64 1
; CHECK-NEXT: [[ARRAYIDX9_2:%.*]] = getelementptr inbounds i32, ptr [[P2]], i64 2
; CHECK-NEXT: br label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[UGLYGEP:%.*]] = getelementptr i8, ptr [[P:%.*]], i64 252
; CHECK-NEXT: [[UGLYGEP1:%.*]] = getelementptr i8, ptr [[P2]], i64 12
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[P]], [[UGLYGEP1]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[P2]], [[UGLYGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX9_2]], align 4, !alias.scope [[META0:![0-9]+]]
; CHECK-NEXT: [[TMP2:%.*]] = load i32, ptr [[ARRAYIDX9_1]], align 4, !alias.scope [[META0]]
; CHECK-NEXT: [[TMP0:%.*]] = load i32, ptr [[P2]], align 4, !alias.scope [[META0]]
; CHECK-NEXT: [[TMP1:%.*]] = sub nsw i32 0, [[TMP0]]
; CHECK-NEXT: [[TMP3:%.*]] = sub nsw i32 [[TMP1]], [[TMP2]]
; CHECK-NEXT: [[TMP5:%.*]] = sub nsw i32 [[TMP3]], [[TMP4]]
; CHECK-NEXT: [[BROADCAST_SPLATINSERT4:%.*]] = insertelement <4 x i32> poison, i32 [[TMP5]], i64 0
; CHECK-NEXT: [[BROADCAST_SPLAT5:%.*]] = shufflevector <4 x i32> [[BROADCAST_SPLATINSERT4]], <4 x i32> poison, <4 x i32> zeroinitializer
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i32, ptr [[P]], i64 [[INDEX]]
; CHECK-NEXT: store <4 x i32> [[BROADCAST_SPLAT5]], ptr [[TMP7]], align 4, !alias.scope [[META3:![0-9]+]], !noalias [[META0]]
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 60
; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: br label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 60, [[MIDDLE_BLOCK]] ], [ 0, [[VECTOR_MEMCHECK]] ]
; CHECK-NEXT: br label [[FOR_COND5:%.*]]
; CHECK: for.cond5:
; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_COND5]] ]
; CHECK-NEXT: [[I3:%.*]] = load i32, ptr [[P2]], align 4
; CHECK-NEXT: [[SUB:%.*]] = sub nsw i32 0, [[I3]]
; CHECK-NEXT: [[I4:%.*]] = load i32, ptr [[ARRAYIDX9_1]], align 4
; CHECK-NEXT: [[SUB_1:%.*]] = sub nsw i32 [[SUB]], [[I4]]
; CHECK-NEXT: [[I5:%.*]] = load i32, ptr [[ARRAYIDX9_2]], align 4
; CHECK-NEXT: [[SUB_2:%.*]] = sub nsw i32 [[SUB_1]], [[I5]]
; CHECK-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds i32, ptr [[P]], i64 [[INDVARS_IV]]
; CHECK-NEXT: store i32 [[SUB_2]], ptr [[ARRAYIDX14]], align 4
; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], 63
; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END17:%.*]], label [[FOR_COND5]], !llvm.loop [[LOOP8:![0-9]+]]
; CHECK: for.end17:
; CHECK-NEXT: ret void
;
entry:
%arrayidx9.1 = getelementptr inbounds i32, ptr %p2, i64 1
%arrayidx9.2 = getelementptr inbounds i32, ptr %p2, i64 2
br label %for.cond5
for.cond5: ; preds = %entry, %for.cond5
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.cond5 ]
%i3 = load i32, ptr %p2, align 4
%sub = sub nsw i32 0, %i3
%i4 = load i32, ptr %arrayidx9.1, align 4
%sub.1 = sub nsw i32 %sub, %i4
%i5 = load i32, ptr %arrayidx9.2, align 4
%sub.2 = sub nsw i32 %sub.1, %i5
%arrayidx14 = getelementptr inbounds i32, ptr %p, i64 %indvars.iv
store i32 %sub.2, ptr %arrayidx14, align 4
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, 63
br i1 %exitcond, label %for.end17, label %for.cond5
for.end17: ; preds = %for.cond5
ret void
}