shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/LoopVectorize/optimal-epilog-vectorization-liveout.ll
Mel Chen 8c6658aca6
[VPlan] Sink recipes from the vector loop region in licm. (#168031) 
When a recipe can be safely sunk and all of its users are outside the
vector loop region in the same dedicated exit block, the recipe does not
need to be executed on every iteration.
This patch extends the VPlan-based LICM (Loop Invariant Code Motion) to
also sink such recipes from the vector loop region into the exit block.
This reduces redundant computation and improves cost model accuracy.

TODO: Support nested loop sinking
TODO: Support sinking `VPReplicateRecipe` (requires `replicateByVF`
fixes)
TODO: Support recipes with multiple defined values (e.g., interleaved
loads)
TODO: Clone recipes without users to all exit blocks
TODO: Support PHI node users by checking incoming value blocks
TODO: Support sinking when users are in multiple blocks
TODO: Clone recipes when users are on multiple exit paths

Co-authored-by: Luke Lau <luke@igalia.com>

---------

Co-authored-by: Luke Lau <luke@igalia.com>
Co-authored-by: Luke Lau <luke_lau@icloud.com>
2026-02-03 07:57:15 +00:00

112 lines
7.6 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; To test epilogue-vectorization we need to make sure that the vectorizer actually vectorizes the loop.
; Without a target triple this becomes difficult, unless we force vectorization through user hints.
; Currently user provided vectorization hints prevent epilogue vectorization unless the forced
; VF is the same as the epilogue vectorization VF. To make these tests target independent we'll use a
; trick where both VFs are forced to be the same value.
; RUN: opt < %s -passes='loop-vectorize' -enable-epilogue-vectorization -force-vector-width=2 -epilogue-vectorization-force-VF=2 -S | FileCheck %s --check-prefix VF-TWO-CHECK
target datalayout = "e-m:e-i64:64-n32:64"
; Some limited forms of live-outs (non-reduction, non-recurrences) are supported.
define signext i32 @f1(ptr noalias %A, ptr noalias %B, i32 signext %n) {
; VF-TWO-CHECK-LABEL: @f1(
; VF-TWO-CHECK-NEXT: entry:
; VF-TWO-CHECK-NEXT: [[CMP1:%.*]] = icmp sgt i32 [[N:%.*]], 0
; VF-TWO-CHECK-NEXT: br i1 [[CMP1]], label [[ITER_CHECK:%.*]], label [[FOR_END:%.*]]
; VF-TWO-CHECK: iter.check:
; VF-TWO-CHECK-NEXT: [[WIDE_TRIP_COUNT:%.*]] = zext i32 [[N]] to i64
; VF-TWO-CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 2
; VF-TWO-CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
; VF-TWO-CHECK: vector.main.loop.iter.check:
; VF-TWO-CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 2
; VF-TWO-CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
; VF-TWO-CHECK: vector.ph:
; VF-TWO-CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 2
; VF-TWO-CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF]]
; VF-TWO-CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; VF-TWO-CHECK: vector.body:
; VF-TWO-CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; VF-TWO-CHECK-NEXT: [[TMP1:%.*]] = getelementptr inbounds i32, ptr [[A:%.*]], i64 [[INDEX]]
; VF-TWO-CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP1]], align 4
; VF-TWO-CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds i32, ptr [[B:%.*]], i64 [[INDEX]]
; VF-TWO-CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x i32>, ptr [[TMP3]], align 4
; VF-TWO-CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 2
; VF-TWO-CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; VF-TWO-CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; VF-TWO-CHECK: middle.block:
; VF-TWO-CHECK-NEXT: [[TMP5:%.*]] = add nsw <2 x i32> [[WIDE_LOAD]], [[WIDE_LOAD2]]
; VF-TWO-CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x i32> [[TMP5]], i32 1
; VF-TWO-CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
; VF-TWO-CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
; VF-TWO-CHECK: vec.epilog.iter.check:
; VF-TWO-CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_MOD_VF]], 2
; VF-TWO-CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]
; VF-TWO-CHECK: vec.epilog.ph:
; VF-TWO-CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; VF-TWO-CHECK-NEXT: [[N_MOD_VF3:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 2
; VF-TWO-CHECK-NEXT: [[N_VEC4:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[N_MOD_VF3]]
; VF-TWO-CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
; VF-TWO-CHECK: vec.epilog.vector.body:
; VF-TWO-CHECK-NEXT: [[INDEX6:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT9:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
; VF-TWO-CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDEX6]]
; VF-TWO-CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <2 x i32>, ptr [[TMP9]], align 4
; VF-TWO-CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDEX6]]
; VF-TWO-CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <2 x i32>, ptr [[TMP11]], align 4
; VF-TWO-CHECK-NEXT: [[INDEX_NEXT9]] = add nuw i64 [[INDEX6]], 2
; VF-TWO-CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT9]], [[N_VEC4]]
; VF-TWO-CHECK-NEXT: br i1 [[TMP14]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; VF-TWO-CHECK: vec.epilog.middle.block:
; VF-TWO-CHECK-NEXT: [[TMP13:%.*]] = add nsw <2 x i32> [[WIDE_LOAD7]], [[WIDE_LOAD8]]
; VF-TWO-CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x i32> [[TMP13]], i32 1
; VF-TWO-CHECK-NEXT: [[CMP_N5:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC4]]
; VF-TWO-CHECK-NEXT: br i1 [[CMP_N5]], label [[FOR_END_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]
; VF-TWO-CHECK: vec.epilog.scalar.ph:
; VF-TWO-CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC4]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK]] ]
; VF-TWO-CHECK-NEXT: br label [[FOR_BODY:%.*]]
; VF-TWO-CHECK: for.body:
; VF-TWO-CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
; VF-TWO-CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, ptr [[A]], i64 [[INDVARS_IV]]
; VF-TWO-CHECK-NEXT: [[TMP16:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
; VF-TWO-CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i32, ptr [[B]], i64 [[INDVARS_IV]]
; VF-TWO-CHECK-NEXT: [[TMP17:%.*]] = load i32, ptr [[ARRAYIDX2]], align 4
; VF-TWO-CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP16]], [[TMP17]]
; VF-TWO-CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
; VF-TWO-CHECK-NEXT: [[EXITCOND:%.*]] = icmp ne i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
; VF-TWO-CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_BODY]], label [[FOR_END_LOOPEXIT]], !llvm.loop [[LOOP5:![0-9]+]]
; VF-TWO-CHECK: for.end.loopexit:
; VF-TWO-CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i32 [ [[ADD]], [[FOR_BODY]] ], [ [[TMP7]], [[MIDDLE_BLOCK]] ], [ [[TMP15]], [[VEC_EPILOG_MIDDLE_BLOCK]] ]
; VF-TWO-CHECK-NEXT: br label [[FOR_END]]
; VF-TWO-CHECK: for.end:
; VF-TWO-CHECK-NEXT: [[RES_0_LCSSA:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[ADD_LCSSA]], [[FOR_END_LOOPEXIT]] ]
; VF-TWO-CHECK-NEXT: ret i32 [[RES_0_LCSSA]]
;
entry:
%cmp1 = icmp sgt i32 %n, 0
br i1 %cmp1, label %for.body.preheader, label %for.end
for.body.preheader: ; preds = %entry
%wide.trip.count = zext i32 %n to i64
br label %for.body
for.body: ; preds = %for.body.preheader, %for.body
%indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
%arrayidx = getelementptr inbounds i32, ptr %A, i64 %indvars.iv
%0 = load i32, ptr %arrayidx, align 4
%arrayidx2 = getelementptr inbounds i32, ptr %B, i64 %indvars.iv
%1 = load i32, ptr %arrayidx2, align 4
%add = add nsw i32 %0, %1
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp ne i64 %indvars.iv.next, %wide.trip.count
br i1 %exitcond, label %for.body, label %for.end.loopexit
for.end.loopexit: ; preds = %for.body
%add.lcssa = phi i32 [ %add, %for.body ]
br label %for.end
for.end: ; preds = %for.end.loopexit, %entry
%res.0.lcssa = phi i32 [ 0, %entry ], [ %add.lcssa, %for.end.loopexit ]
ret i32 %res.0.lcssa
}
Reference in New Issue View Git Blame Copy Permalink