[llvm] [LV] Vectorize Epilogues for loops with small VF but high IC (PR #108190)
Julian Nagele via llvm-commits
llvm-commits at lists.llvm.org
Fri Sep 13 08:53:02 PDT 2024
================
@@ -0,0 +1,413 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -S < %s -passes=loop-vectorize -force-vector-interleave=4 2>&1 | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"
+target triple = "aarch64"
+
+define void @add_i8(ptr noalias nocapture noundef writeonly %A, ptr nocapture noundef readonly %B, ptr nocapture noundef readonly %C, i64 noundef %Iterations) {
+; CHECK-LABEL: @add_i8(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: [[CMP11:%.*]] = icmp sgt i64 [[ITERATIONS:%.*]], 0
+; CHECK-NEXT: br i1 [[CMP11]], label [[ITER_CHECK:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK: iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[ITERATIONS]], 8
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
+; CHECK: vector.main.loop.iter.check:
+; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[ITERATIONS]], 64
+; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK: vector.ph:
+; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[ITERATIONS]], 64
+; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF]]
+; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
+; CHECK: vector.body:
+; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 16
+; CHECK-NEXT: [[TMP2:%.*]] = add i64 [[INDEX]], 32
+; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 48
+; CHECK-NEXT: [[TMP4:%.*]] = getelementptr inbounds i8, ptr [[B:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP1]]
+; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP3]]
+; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 0
+; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 16
+; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 32
+; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i8, ptr [[TMP4]], i32 48
+; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, ptr [[TMP8]], align 1
+; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <16 x i8>, ptr [[TMP9]], align 1
+; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <16 x i8>, ptr [[TMP10]], align 1
+; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <16 x i8>, ptr [[TMP11]], align 1
+; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i8, ptr [[C:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[TMP1]]
+; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[TMP3]]
+; CHECK-NEXT: [[TMP16:%.*]] = getelementptr inbounds i8, ptr [[TMP12]], i32 0
+; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds i8, ptr [[TMP12]], i32 16
+; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds i8, ptr [[TMP12]], i32 32
+; CHECK-NEXT: [[TMP19:%.*]] = getelementptr inbounds i8, ptr [[TMP12]], i32 48
+; CHECK-NEXT: [[WIDE_LOAD5:%.*]] = load <16 x i8>, ptr [[TMP16]], align 1
+; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <16 x i8>, ptr [[TMP17]], align 1
+; CHECK-NEXT: [[WIDE_LOAD7:%.*]] = load <16 x i8>, ptr [[TMP18]], align 1
+; CHECK-NEXT: [[WIDE_LOAD8:%.*]] = load <16 x i8>, ptr [[TMP19]], align 1
+; CHECK-NEXT: [[TMP20:%.*]] = add <16 x i8> [[WIDE_LOAD5]], [[WIDE_LOAD]]
+; CHECK-NEXT: [[TMP21:%.*]] = add <16 x i8> [[WIDE_LOAD6]], [[WIDE_LOAD2]]
+; CHECK-NEXT: [[TMP22:%.*]] = add <16 x i8> [[WIDE_LOAD7]], [[WIDE_LOAD3]]
+; CHECK-NEXT: [[TMP23:%.*]] = add <16 x i8> [[WIDE_LOAD8]], [[WIDE_LOAD4]]
+; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i8, ptr [[A:%.*]], i64 [[TMP0]]
+; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP1]]
+; CHECK-NEXT: [[TMP26:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP2]]
+; CHECK-NEXT: [[TMP27:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP3]]
+; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds i8, ptr [[TMP24]], i32 0
+; CHECK-NEXT: [[TMP29:%.*]] = getelementptr inbounds i8, ptr [[TMP24]], i32 16
+; CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds i8, ptr [[TMP24]], i32 32
+; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds i8, ptr [[TMP24]], i32 48
+; CHECK-NEXT: store <16 x i8> [[TMP20]], ptr [[TMP28]], align 1
+; CHECK-NEXT: store <16 x i8> [[TMP21]], ptr [[TMP29]], align 1
+; CHECK-NEXT: store <16 x i8> [[TMP22]], ptr [[TMP30]], align 1
+; CHECK-NEXT: store <16 x i8> [[TMP23]], ptr [[TMP31]], align 1
+; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 64
+; CHECK-NEXT: [[TMP32:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[TMP32]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK: middle.block:
+; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC]]
+; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
+; CHECK: vec.epilog.iter.check:
+; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[ITERATIONS]], [[N_VEC]]
+; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 8
+; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
+; CHECK: vec.epilog.ph:
+; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
+; CHECK-NEXT: [[N_MOD_VF9:%.*]] = urem i64 [[ITERATIONS]], 8
+; CHECK-NEXT: [[N_VEC10:%.*]] = sub i64 [[ITERATIONS]], [[N_MOD_VF9]]
+; 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_NEXT15:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
+; CHECK-NEXT: [[TMP33:%.*]] = add i64 [[INDEX12]], 0
+; CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[TMP33]]
+; CHECK-NEXT: [[TMP35:%.*]] = getelementptr inbounds i8, ptr [[TMP34]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD13:%.*]] = load <8 x i8>, ptr [[TMP35]], align 1
+; CHECK-NEXT: [[TMP36:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[TMP33]]
+; CHECK-NEXT: [[TMP37:%.*]] = getelementptr inbounds i8, ptr [[TMP36]], i32 0
+; CHECK-NEXT: [[WIDE_LOAD14:%.*]] = load <8 x i8>, ptr [[TMP37]], align 1
+; CHECK-NEXT: [[TMP38:%.*]] = add <8 x i8> [[WIDE_LOAD14]], [[WIDE_LOAD13]]
+; CHECK-NEXT: [[TMP39:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[TMP33]]
+; CHECK-NEXT: [[TMP40:%.*]] = getelementptr inbounds i8, ptr [[TMP39]], i32 0
+; CHECK-NEXT: store <8 x i8> [[TMP38]], ptr [[TMP40]], align 1
+; CHECK-NEXT: [[INDEX_NEXT15]] = add nuw i64 [[INDEX12]], 8
+; CHECK-NEXT: [[TMP41:%.*]] = icmp eq i64 [[INDEX_NEXT15]], [[N_VEC10]]
+; CHECK-NEXT: br i1 [[TMP41]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+; CHECK: vec.epilog.middle.block:
+; CHECK-NEXT: [[CMP_N11:%.*]] = icmp eq i64 [[ITERATIONS]], [[N_VEC10]]
+; CHECK-NEXT: br i1 [[CMP_N11]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[VEC_EPILOG_SCALAR_PH]]
+; CHECK: vec.epilog.scalar.ph:
+; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[ITER_CHECK]] ]
+; CHECK-NEXT: br label [[FOR_BODY:%.*]]
+; CHECK: for.cond.cleanup.loopexit:
+; CHECK-NEXT: br label [[FOR_COND_CLEANUP]]
+; CHECK: for.cond.cleanup:
+; CHECK-NEXT: ret void
+; CHECK: for.body:
+; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]
+; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i8, ptr [[B]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: [[TMP42:%.*]] = load i8, ptr [[ARRAYIDX]], align 1
+; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i8, ptr [[C]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: [[TMP43:%.*]] = load i8, ptr [[ARRAYIDX2]], align 1
+; CHECK-NEXT: [[ADD:%.*]] = add i8 [[TMP43]], [[TMP42]]
+; CHECK-NEXT: [[ARRAYIDX6:%.*]] = getelementptr inbounds i8, ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-NEXT: store i8 [[ADD]], ptr [[ARRAYIDX6]], align 1
+; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[ITERATIONS]]
+; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
+;
+entry:
+ %cmp11 = icmp sgt i64 %Iterations, 0
+ br i1 %cmp11, label %for.body, label %for.cond.cleanup
----------------
juliannagele wrote:
it isn't -- removed, thanks!
https://github.com/llvm/llvm-project/pull/108190
More information about the llvm-commits
mailing list