caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
e0ad34e56590fa2e6ffdf617e044de7eadee2139
clang-p2996/llvm/test/Transforms/LoopVectorize/AArch64/sve-vector-reverse.ll
Nikita Popov a105877646 [InstCombine] Remove some of the complexity-based canonicalization (#91185) 
The idea behind this canonicalization is that it allows us to handle less
patterns, because we know that some will be canonicalized away. This is
indeed very useful to e.g. know that constants are always on the right.

However, this is only useful if the canonicalization is actually
reliable. This is the case for constants, but not for arguments: Moving
these to the right makes it look like the "more complex" expression is
guaranteed to be on the left, but this is not actually the case in
practice. It fails as soon as you replace the argument with another
instruction.

The end result is that it looks like things correctly work in tests,
while they actually don't. We use the "thwart complexity-based
canonicalization" trick to handle this in tests, but it's often a
challenge for new contributors to get this right, and based on the
regressions this PR originally exposed, we clearly don't get this right
in many cases.

For this reason, I think that it's better to remove this complexity
canonicalization. It will make it much easier to write tests for
commuted cases and make sure that they are handled.
2024-08-21 12:02:54 +02:00

215 lines
13 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; This is the loop in c++ being vectorize in this file with
;vector.reverse
; #pragma clang loop vectorize_width(8, scalable) interleave_count(2)
; for (int i = N-1; i >= 0; --i)
; a[i] = b[i] + 1.0;
; RUN: opt -passes=loop-vectorize,dce,instcombine -mtriple aarch64-linux-gnu -S \
; RUN: -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s
define void @vector_reverse_f64(i64 %N, ptr noalias %a, ptr noalias %b) #0{
; CHECK-LABEL: @vector_reverse_f64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[CMP7:%.*]] = icmp sgt i64 [[N:%.*]], 0
; CHECK-NEXT: br i1 [[CMP7]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
; CHECK: for.body.preheader:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 4
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 4
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub nsw i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT: [[TMP4:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP5:%.*]] = shl i64 [[TMP4]], 4
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP6:%.*]] = xor i64 [[INDEX]], -1
; CHECK-NEXT: [[TMP7:%.*]] = add i64 [[N]], [[TMP6]]
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds double, ptr [[B:%.*]], i64 [[TMP7]]
; CHECK-NEXT: [[TMP9:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP10:%.*]] = shl i64 [[TMP9]], 3
; CHECK-NEXT: [[TMP11:%.*]] = sub i64 1, [[TMP10]]
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds double, ptr [[TMP8]], i64 [[TMP11]]
; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP14:%.*]] = shl i64 [[TMP13]], 3
; CHECK-NEXT: [[TMP15:%.*]] = sub i64 0, [[TMP14]]
; CHECK-NEXT: [[TMP16:%.*]] = sub i64 1, [[TMP14]]
; CHECK-NEXT: [[TMP17:%.*]] = getelementptr inbounds double, ptr [[TMP8]], i64 [[TMP15]]
; CHECK-NEXT: [[TMP18:%.*]] = getelementptr inbounds double, ptr [[TMP17]], i64 [[TMP16]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x double>, ptr [[TMP12]], align 8
; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <vscale x 8 x double>, ptr [[TMP18]], align 8
; CHECK-NEXT: [[TMP19:%.*]] = fadd <vscale x 8 x double> [[WIDE_LOAD]], shufflevector (<vscale x 8 x double> insertelement (<vscale x 8 x double> poison, double 1.000000e+00, i64 0), <vscale x 8 x double> poison, <vscale x 8 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP20:%.*]] = fadd <vscale x 8 x double> [[WIDE_LOAD1]], shufflevector (<vscale x 8 x double> insertelement (<vscale x 8 x double> poison, double 1.000000e+00, i64 0), <vscale x 8 x double> poison, <vscale x 8 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds double, ptr [[A:%.*]], i64 [[TMP7]]
; CHECK-NEXT: [[TMP22:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP23:%.*]] = shl i64 [[TMP22]], 3
; CHECK-NEXT: [[TMP24:%.*]] = sub i64 1, [[TMP23]]
; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i64 [[TMP24]]
; CHECK-NEXT: [[TMP26:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP27:%.*]] = shl i64 [[TMP26]], 3
; CHECK-NEXT: [[TMP28:%.*]] = sub i64 0, [[TMP27]]
; CHECK-NEXT: [[TMP29:%.*]] = sub i64 1, [[TMP27]]
; CHECK-NEXT: [[TMP30:%.*]] = getelementptr inbounds double, ptr [[TMP21]], i64 [[TMP28]]
; CHECK-NEXT: [[TMP31:%.*]] = getelementptr inbounds double, ptr [[TMP30]], i64 [[TMP29]]
; CHECK-NEXT: store <vscale x 8 x double> [[TMP19]], ptr [[TMP25]], align 8
; CHECK-NEXT: store <vscale x 8 x double> [[TMP20]], ptr [[TMP31]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP5]]
; 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 [[N_MOD_VF]], 0
; 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_MOD_VF]], [[MIDDLE_BLOCK]] ], [ [[N]], [[FOR_BODY_PREHEADER]] ]
; 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: [[I_08_IN:%.*]] = phi i64 [ [[I_08:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
; CHECK-NEXT: [[I_08]] = add nsw i64 [[I_08_IN]], -1
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[B]], i64 [[I_08]]
; CHECK-NEXT: [[TMP33:%.*]] = load double, ptr [[ARRAYIDX]], align 8
; CHECK-NEXT: [[ADD:%.*]] = fadd double [[TMP33]], 1.000000e+00
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds double, ptr [[A]], i64 [[I_08]]
; CHECK-NEXT: store double [[ADD]], ptr [[ARRAYIDX1]], align 8
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[I_08_IN]], 1
; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP4:![0-9]+]]
;
entry:
%cmp7 = icmp sgt i64 %N, 0
br i1 %cmp7, label %for.body, label %for.cond.cleanup
for.cond.cleanup: ; preds = %for.body
ret void
for.body: ; preds = %entry, %for.body
%i.08.in = phi i64 [ %i.08, %for.body ], [ %N, %entry ]
%i.08 = add nsw i64 %i.08.in, -1
%arrayidx = getelementptr inbounds double, ptr %b, i64 %i.08
%0 = load double, ptr %arrayidx, align 8
%add = fadd double %0, 1.000000e+00
%arrayidx1 = getelementptr inbounds double, ptr %a, i64 %i.08
store double %add, ptr %arrayidx1, align 8
%cmp = icmp sgt i64 %i.08.in, 1
br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
}
define void @vector_reverse_i64(i64 %N, ptr %a, ptr %b) #0 {
; CHECK-LABEL: @vector_reverse_i64(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[A2:%.*]] = ptrtoint ptr [[A:%.*]] to i64
; CHECK-NEXT: [[B1:%.*]] = ptrtoint ptr [[B:%.*]] to i64
; CHECK-NEXT: [[CMP8:%.*]] = icmp sgt i64 [[N:%.*]], 0
; CHECK-NEXT: br i1 [[CMP8]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]]
; CHECK: for.body.preheader:
; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 4
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], [[TMP1]]
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]]
; CHECK: vector.memcheck:
; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 7
; CHECK-NEXT: [[TMP4:%.*]] = sub i64 [[B1]], [[A2]]
; CHECK-NEXT: [[DIFF_CHECK:%.*]] = icmp ult i64 [[TMP4]], [[TMP3]]
; CHECK-NEXT: br i1 [[DIFF_CHECK]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[TMP5]], 4
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP6]]
; CHECK-NEXT: [[N_VEC:%.*]] = sub nsw i64 [[N]], [[N_MOD_VF]]
; CHECK-NEXT: [[TMP7:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP8:%.*]] = shl i64 [[TMP7]], 4
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP9:%.*]] = xor i64 [[INDEX]], -1
; CHECK-NEXT: [[TMP10:%.*]] = add i64 [[N]], [[TMP9]]
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP13:%.*]] = shl i64 [[TMP12]], 3
; CHECK-NEXT: [[TMP14:%.*]] = sub i64 1, [[TMP13]]
; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i64, ptr [[TMP11]], i64 [[TMP14]]
; CHECK-NEXT: [[TMP16:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP17:%.*]] = shl i64 [[TMP16]], 3
; CHECK-NEXT: [[TMP18:%.*]] = sub i64 0, [[TMP17]]
; CHECK-NEXT: [[TMP19:%.*]] = sub i64 1, [[TMP17]]
; CHECK-NEXT: [[TMP20:%.*]] = getelementptr inbounds i64, ptr [[TMP11]], i64 [[TMP18]]
; CHECK-NEXT: [[TMP21:%.*]] = getelementptr inbounds i64, ptr [[TMP20]], i64 [[TMP19]]
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 8 x i64>, ptr [[TMP15]], align 8
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 8 x i64>, ptr [[TMP21]], align 8
; CHECK-NEXT: [[TMP22:%.*]] = add <vscale x 8 x i64> [[WIDE_LOAD]], shufflevector (<vscale x 8 x i64> insertelement (<vscale x 8 x i64> poison, i64 1, i64 0), <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP23:%.*]] = add <vscale x 8 x i64> [[WIDE_LOAD3]], shufflevector (<vscale x 8 x i64> insertelement (<vscale x 8 x i64> poison, i64 1, i64 0), <vscale x 8 x i64> poison, <vscale x 8 x i32> zeroinitializer)
; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP10]]
; CHECK-NEXT: [[TMP25:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP26:%.*]] = shl i64 [[TMP25]], 3
; CHECK-NEXT: [[TMP27:%.*]] = sub i64 1, [[TMP26]]
; CHECK-NEXT: [[TMP28:%.*]] = getelementptr inbounds i64, ptr [[TMP24]], i64 [[TMP27]]
; CHECK-NEXT: [[TMP29:%.*]] = call i64 @llvm.vscale.i64()
; CHECK-NEXT: [[TMP30:%.*]] = shl i64 [[TMP29]], 3
; CHECK-NEXT: [[TMP31:%.*]] = sub i64 0, [[TMP30]]
; CHECK-NEXT: [[TMP32:%.*]] = sub i64 1, [[TMP30]]
; CHECK-NEXT: [[TMP33:%.*]] = getelementptr inbounds i64, ptr [[TMP24]], i64 [[TMP31]]
; CHECK-NEXT: [[TMP34:%.*]] = getelementptr inbounds i64, ptr [[TMP33]], i64 [[TMP32]]
; CHECK-NEXT: store <vscale x 8 x i64> [[TMP22]], ptr [[TMP28]], align 8
; CHECK-NEXT: store <vscale x 8 x i64> [[TMP23]], ptr [[TMP34]], align 8
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP8]]
; CHECK-NEXT: [[TMP35:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP35]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]]
; CHECK: middle.block:
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
; 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_MOD_VF]], [[MIDDLE_BLOCK]] ], [ [[N]], [[FOR_BODY_PREHEADER]] ], [ [[N]], [[VECTOR_MEMCHECK]] ]
; 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: [[I_09_IN:%.*]] = phi i64 [ [[I_09:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
; CHECK-NEXT: [[I_09]] = add nsw i64 [[I_09_IN]], -1
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[B]], i64 [[I_09]]
; CHECK-NEXT: [[TMP36:%.*]] = load i64, ptr [[ARRAYIDX]], align 8
; CHECK-NEXT: [[ADD:%.*]] = add i64 [[TMP36]], 1
; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[I_09]]
; CHECK-NEXT: store i64 [[ADD]], ptr [[ARRAYIDX2]], align 8
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[I_09_IN]], 1
; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP6:![0-9]+]]
;
entry:
%cmp8 = icmp sgt i64 %N, 0
br i1 %cmp8, label %for.body, label %for.cond.cleanup
for.cond.cleanup: ; preds = %for.body
ret void
for.body: ; preds = %entry, %for.body
%i.09.in = phi i64 [ %i.09, %for.body ], [ %N, %entry ]
%i.09 = add nsw i64 %i.09.in, -1
%arrayidx = getelementptr inbounds i64, ptr %b, i64 %i.09
%0 = load i64, ptr %arrayidx, align 8
%add = add i64 %0, 1
%arrayidx2 = getelementptr inbounds i64, ptr %a, i64 %i.09
store i64 %add, ptr %arrayidx2, align 8
%cmp = icmp sgt i64 %i.09.in, 1
br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0
}
attributes #0 = { "target-cpu"="generic" "target-features"="+neon,+sve" }
!0 = distinct !{!0, !1, !2, !3, !4, !5}
!1 = !{!"llvm.loop.mustprogress"}
!2 = !{!"llvm.loop.vectorize.width", i32 8}
!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}
!4 = !{!"llvm.loop.vectorize.enable", i1 true}
!5 = !{!"llvm.loop.interleave.count", i32 2}
Reference in New Issue View Git Blame Copy Permalink