[llvm] [LoopVectorize] LLVM fails to vectorise loops with multi-bool varables (PR #89226)

[David Sherwood via llvm-commits]

================
@@ -0,0 +1,883 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -passes=loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -S < %s | FileCheck %s --check-prefix=CHECK
+
+
+; int multi_user_cmp(float* a, int n) {
+;  _Bool any = 0;
+;  _Bool all = 1;
+;  for (int i = 0; i < n; i++) {
+;    if (a[i] < 0.0f) {
+;      any = 1;
+;    } else {
+;      all = 0;
+;    }
+;  }
+;  return all ? 1 : any ? 2 : 3;
+;}
+define i32 @multi_user_cmp(ptr readonly %a, i32 noundef %n) {
+; CHECK-LABEL: define i32 @multi_user_cmp(
+; CHECK-SAME: ptr readonly [[A:%.*]], i32 noundef [[N:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CMP8:%.*]] = icmp sgt i32 [[N]], 0
+; CHECK-NEXT:    br i1 [[CMP8]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
+; CHECK:       for.body.preheader:
+; CHECK-NEXT:    [[WIDE_TRIP_COUNT:%.*]] = zext nneg i32 [[N]] to i64
+; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[WIDE_TRIP_COUNT]], 4
+; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
+; CHECK:       vector.ph:
+; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[WIDE_TRIP_COUNT]], 4
+; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[WIDE_TRIP_COUNT]], [[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:    [[VEC_PHI:%.*]] = phi <4 x i1> [ <i1 true, i1 true, i1 true, i1 true>, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[VEC_PHI1:%.*]] = phi <4 x i1> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP4:%.*]], [[VECTOR_BODY]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[INDEX]], 0
+; CHECK-NEXT:    [[TMP1:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP0]]
+; CHECK-NEXT:    [[TMP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0
+; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4
+; CHECK-NEXT:    [[TMP3:%.*]] = fcmp olt <4 x float> [[WIDE_LOAD]], zeroinitializer
+; CHECK-NEXT:    [[TMP4]] = select <4 x i1> [[TMP3]], <4 x i1> <i1 true, i1 true, i1 true, i1 true>, <4 x i1> [[VEC_PHI1]]
+; CHECK-NEXT:    [[TMP5]] = select <4 x i1> [[TMP3]], <4 x i1> [[VEC_PHI]], <4 x i1> zeroinitializer
+; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
+; CHECK-NEXT:    [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
+; CHECK:       middle.block:
+; CHECK-NEXT:    [[RDX_SELECT_CMP:%.*]] = icmp ne <4 x i1> [[TMP5]], <i1 true, i1 true, i1 true, i1 true>
+; CHECK-NEXT:    [[TMP7:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[RDX_SELECT_CMP]])
+; CHECK-NEXT:    [[RDX_SELECT:%.*]] = select i1 [[TMP7]], i1 false, i1 true
+; CHECK-NEXT:    [[RDX_SELECT_CMP2:%.*]] = icmp ne <4 x i1> [[TMP4]], zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = call i1 @llvm.vector.reduce.or.v4i1(<4 x i1> [[RDX_SELECT_CMP2]])
+; CHECK-NEXT:    [[RDX_SELECT3:%.*]] = select i1 [[TMP8]], i1 true, i1 false
+; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[WIDE_TRIP_COUNT]], [[N_VEC]]
+; CHECK-NEXT:    br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]]
+; CHECK:       scalar.ph:
+; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX:%.*]] = phi i1 [ true, [[FOR_BODY_PREHEADER]] ], [ [[RDX_SELECT]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[BC_MERGE_RDX4:%.*]] = phi i1 [ false, [[FOR_BODY_PREHEADER]] ], [ [[RDX_SELECT3]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    br label [[FOR_BODY:%.*]]
+; CHECK:       for.cond.cleanup.loopexit:
+; CHECK-NEXT:    [[DOTANY_0_OFF0_LCSSA:%.*]] = phi i1 [ [[DOTANY_0_OFF0:%.*]], [[FOR_BODY]] ], [ [[RDX_SELECT3]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[ALL_0_OFF0__LCSSA:%.*]] = phi i1 [ [[ALL_0_OFF0_:%.*]], [[FOR_BODY]] ], [ [[RDX_SELECT]], [[MIDDLE_BLOCK]] ]
+; CHECK-NEXT:    [[TMP9:%.*]] = select i1 [[DOTANY_0_OFF0_LCSSA]], i32 2, i32 3
+; CHECK-NEXT:    [[TMP10:%.*]] = select i1 [[ALL_0_OFF0__LCSSA]], i32 1, i32 [[TMP9]]
+; CHECK-NEXT:    br label [[FOR_COND_CLEANUP]]
+; CHECK:       for.cond.cleanup:
+; CHECK-NEXT:    [[ALL_0_OFF0_LCSSA:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ [[TMP10]], [[FOR_COND_CLEANUP_LOOPEXIT]] ]
+; CHECK-NEXT:    ret i32 [[ALL_0_OFF0_LCSSA]]
+; CHECK:       for.body:
+; CHECK-NEXT:    [[INDVARS_IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[ALL_0_OFF010:%.*]] = phi i1 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ALL_0_OFF0_]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[ANY_0_OFF09:%.*]] = phi i1 [ [[BC_MERGE_RDX4]], [[SCALAR_PH]] ], [ [[DOTANY_0_OFF0]], [[FOR_BODY]] ]
+; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDVARS_IV]]
+; CHECK-NEXT:    [[TMP11:%.*]] = load float, ptr [[ARRAYIDX]], align 4
+; CHECK-NEXT:    [[CMP1:%.*]] = fcmp olt float [[TMP11]], 0.000000e+00
+; CHECK-NEXT:    [[DOTANY_0_OFF0]] = select i1 [[CMP1]], i1 true, i1 [[ANY_0_OFF09]]
+; CHECK-NEXT:    [[ALL_0_OFF0_]] = select i1 [[CMP1]], i1 [[ALL_0_OFF010]], i1 false
+; CHECK-NEXT:    [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
+; CHECK-NEXT:    [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[WIDE_TRIP_COUNT]]
+; CHECK-NEXT:    br i1 [[EXITCOND_NOT]], label [[FOR_COND_CLEANUP_LOOPEXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]]
+;
+entry:
+  %cmp8 = icmp sgt i32 %n, 0
----------------
david-arm wrote:

In all these tests can you:

1. Remove the greater-than comparison with 0 so that the entry block unconditionally jumps to the preheader? This will remove quite a lot of code and CHECK lines. The comparison isn't essential for testing your change I think.
2. Rewrite the blocks in a more logical order, which for this function at least means putting the cleanup blocks at the end of the function.
3. Maybe instead of for.cond.cleanup you can just call it 'exit' just to make it a bit more readable?

https://github.com/llvm/llvm-project/pull/89226