shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity
llvm-project/llvm/test/Transforms/LoopVectorize/AArch64/interleave-with-runtime-checks.ll
Florian Hahn 77fb848894
Reapply "[LV] Simplify and unify resume value handling for epilogue vec." (#187504) 
This reverts commit cdaf29f84dd0abbd1f961982799059c92d76625b.

This version skips removeBranchOnConst when vectorizing the epilogue, as
it may trigger folds that remove the resume phi used as resume value
from the epilogue.

This fixes https://github.com/llvm/llvm-project/issues/187323.

Original message:
This patch tries to drastically simplify resume value handling for the
scalar loop when vectorizing the epilogue.

It uses a simpler, uniform approach for updating all resume values in
the scalar loop:

1. Create ResumeForEpilogue recipes for all scalar resume phis in the
main loop (the epilogue plan will have exactly the same scalar resume
phis, in exactly the same order)
2. Update ::execute for ResumeForEpilogue to set the underlying value
when executing. This is not super clean, but allows easy lookup of the
generated IR value when we update the resume phis in the epilogue. Once
we connect the 2 plans together explicitly, this can be removed.
3. Use the list of ResumeForEpilogue VPInstructions from the main loop
to update the resume/bypass values from the epilogue.

This simplifies the code quite a bit, makes it more robust (should fix
https://github.com/llvm/llvm-project/issues/179407) and also fixes a
mis-compile in the existing tests (see change in

llvm/test/Transforms/LoopVectorize/AArch64/partial-reduce-sub-epilogue-vec.ll,
where previously we would incorrectly resume using the start value when
the epilogue iteration check failed)

In some cases, we get simpler code, due to additional CSE, in some cases
the induction end value computations get moved from the epilogue
iteration check to the vector preheader. We could try to sink the
instructions as cleanup, but it is probably not worth the trouble.

Fixes https://github.com/llvm/llvm-project/issues/179407.

PR for recommit https://github.com/llvm/llvm-project/pull/188134
2026-03-23 22:09:40 +00:00

89 lines
5.1 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals none --filter-out-after "scalar.ph\:" --version 5
; RUN: opt -p loop-vectorize -S %s | FileCheck %s
target triple = "arm64-apple-macosx15.0.0"
define void @interleave_groups_separated_by_offset(ptr %A, i64 %offset) {
; CHECK-LABEL: define void @interleave_groups_separated_by_offset(
; CHECK-SAME: ptr [[A:%.*]], i64 [[OFFSET:%.*]]) {
; CHECK-NEXT: [[ITER_CHECK:.*:]]
; CHECK-NEXT: [[A_OFFSET:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET]]
; CHECK-NEXT: br i1 false, label %[[VEC_EPILOG_SCALAR_PH:.*]], label %[[VECTOR_MEMCHECK:.*]]
; CHECK: [[VECTOR_MEMCHECK]]:
; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr i8, ptr [[A]], i64 1999
; CHECK-NEXT: [[SCEVGEP1:%.*]] = getelementptr i8, ptr [[A]], i64 1
; CHECK-NEXT: [[SCEVGEP2:%.*]] = getelementptr i8, ptr [[A]], i64 2000
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET]], 1999
; CHECK-NEXT: [[SCEVGEP3:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP0]]
; CHECK-NEXT: [[BOUND0:%.*]] = icmp ult ptr [[A]], [[SCEVGEP2]]
; CHECK-NEXT: [[BOUND1:%.*]] = icmp ult ptr [[SCEVGEP1]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]]
; CHECK-NEXT: [[BOUND04:%.*]] = icmp ult ptr [[A]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND15:%.*]] = icmp ult ptr [[A_OFFSET]], [[SCEVGEP]]
; CHECK-NEXT: [[FOUND_CONFLICT6:%.*]] = and i1 [[BOUND04]], [[BOUND15]]
; CHECK-NEXT: [[CONFLICT_RDX:%.*]] = or i1 [[FOUND_CONFLICT]], [[FOUND_CONFLICT6]]
; CHECK-NEXT: [[BOUND07:%.*]] = icmp ult ptr [[SCEVGEP1]], [[SCEVGEP3]]
; CHECK-NEXT: [[BOUND18:%.*]] = icmp ult ptr [[A_OFFSET]], [[SCEVGEP2]]
; CHECK-NEXT: [[FOUND_CONFLICT9:%.*]] = and i1 [[BOUND07]], [[BOUND18]]
; CHECK-NEXT: [[CONFLICT_RDX10:%.*]] = or i1 [[CONFLICT_RDX]], [[FOUND_CONFLICT9]]
; CHECK-NEXT: br i1 [[CONFLICT_RDX10]], label %[[VEC_EPILOG_SCALAR_PH]], label %[[VECTOR_MAIN_LOOP_ITER_CHECK:.*]]
; CHECK: [[VECTOR_MAIN_LOOP_ITER_CHECK]]:
; CHECK-NEXT: br i1 false, label %[[VEC_EPILOG_PH:.*]], label %[[VECTOR_PH:.*]]
; CHECK: [[VECTOR_PH]]:
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[A]], i64 1984
; CHECK-NEXT: [[IND_END17:%.*]] = getelementptr i8, ptr [[A_OFFSET]], i64 1984
; CHECK-NEXT: br label %[[VECTOR_BODY:.*]]
; CHECK: [[VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, %[[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], %[[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[INDEX]], 2
; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[TMP1]], 32
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP1]]
; CHECK-NEXT: [[NEXT_GEP11:%.*]] = getelementptr i8, ptr [[A]], i64 [[TMP2]]
; CHECK-NEXT: store <32 x i8> zeroinitializer, ptr [[NEXT_GEP]], align 1
; CHECK-NEXT: store <32 x i8> zeroinitializer, ptr [[NEXT_GEP11]], align 1
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 32
; CHECK-NEXT: [[TMP3:%.*]] = icmp eq i64 [[INDEX_NEXT]], 992
; CHECK-NEXT: br i1 [[TMP3]], label %[[MIDDLE_BLOCK:.*]], label %[[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
; CHECK: [[MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[VEC_EPILOG_ITER_CHECK:.*]]
; CHECK: [[VEC_EPILOG_ITER_CHECK]]:
; CHECK-NEXT: br i1 true, label %[[VEC_EPILOG_SCALAR_PH]], label %[[VEC_EPILOG_PH]], !prof [[PROF3:![0-9]+]]
; CHECK: [[VEC_EPILOG_PH]]:
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ 992, %[[VEC_EPILOG_ITER_CHECK]] ], [ 0, %[[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[A]], i64 1984
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[A_OFFSET]], i64 1984
; CHECK-NEXT: br label %[[VEC_EPILOG_VECTOR_BODY:.*]]
; CHECK: [[VEC_EPILOG_VECTOR_BODY]]:
; CHECK-NEXT: [[INDEX12:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], %[[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT14:%.*]], %[[VEC_EPILOG_VECTOR_BODY]] ]
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX12]], 2
; CHECK-NEXT: [[NEXT_GEP13:%.*]] = getelementptr i8, ptr [[A]], i64 [[OFFSET_IDX]]
; CHECK-NEXT: store <16 x i8> zeroinitializer, ptr [[NEXT_GEP13]], align 1
; CHECK-NEXT: [[INDEX_NEXT14]] = add nuw i64 [[INDEX12]], 8
; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT14]], 992
; CHECK-NEXT: br i1 [[TMP6]], label %[[VEC_EPILOG_MIDDLE_BLOCK:.*]], label %[[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
; CHECK: [[VEC_EPILOG_MIDDLE_BLOCK]]:
; CHECK-NEXT: br label %[[VEC_EPILOG_SCALAR_PH]]
; CHECK: [[VEC_EPILOG_SCALAR_PH]]:
;
entry:
%A.offset = getelementptr i8, ptr %A, i64 %offset
br label %loop
loop:
%iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
%ptr.iv = phi ptr [ %A, %entry ], [ %ptr.iv.next, %loop ]
%ptr.iv.offset = phi ptr [ %A.offset, %entry ], [ %ptr.iv.offset.next, %loop ]
%ptr.iv.1 = getelementptr i8, ptr %ptr.iv, i64 1
store i8 0, ptr %ptr.iv, align 1
%ptr.iv.offset.next = getelementptr i8, ptr %ptr.iv.offset, i64 2
%1 = load i8, ptr %ptr.iv.offset, align 1
%ptr.iv.next = getelementptr i8, ptr %ptr.iv, i64 2
store i8 0, ptr %ptr.iv.1, align 1
%iv.next = add i32 %iv, 1
%ec = icmp eq i32 %iv.next, 1000
br i1 %ec, label %exit, label %loop
exit:
ret void
}
Reference in New Issue View Git Blame Copy Permalink