caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
fc712eb7aa00aabcdafda54776038efdc486d570
clang-p2996/llvm/test/Transforms/LoopVectorize/ARM/sphinx.ll
Sjoerd Meijer 9529597cf4 Recommit #2: "[LV] Induction Variable does not remain scalar under tail-folding." 
This was reverted because of a miscompilation. At closer inspection, the
problem was actually visible in a changed llvm regression test too. This
one-line follow up fix/recommit will splat the IV, which is what we are trying
to avoid if unnecessary in general, if tail-folding is requested even if all
users are scalar instructions after vectorisation. Because with tail-folding,
the splat IV will be used by the predicate of the masked loads/stores
instructions. The previous version omitted this, which caused the
miscompilation. The original commit message was:

If tail-folding of the scalar remainder loop is applied, the primary induction
variable is splat to a vector and used by the masked load/store vector
instructions, thus the IV does not remain scalar. Because we now mark
that the IV does not remain scalar for these cases, we don't emit the vector IV
if it is not used. Thus, the vectoriser produces less dead code.

Thanks to Ayal Zaks for the direction how to fix this.
2020-05-13 13:50:09 +01:00

163 lines
9.2 KiB
LLVM
Raw Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
; RUN: opt < %s -loop-vectorize -mtriple=thumbv8-unknown-unknown -mcpu=cortex-a53 -S | FileCheck %s
; This test is reduced from SPECFP 2006 482.sphinx.
; We expect vectorization with <2 x double> and <2 x float> ops.
; See https://bugs.llvm.org/show_bug.cgi?id=36280 for more details.
target datalayout = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
@a = external global i32
@v = external global i32
@mm = external global float**
@vv = external global float**
@ll = external global float*
define i32 @test(float* nocapture readonly %x) {
; CHECK-LABEL: @test(
; CHECK-NEXT: entry:
; CHECK-NEXT: [[T:%.*]] = load i32, i32* @v, align 8
; CHECK-NEXT: [[T1:%.*]] = load i32, i32* @a, align 4
; CHECK-NEXT: br label [[OUTERLOOP:%.*]]
; CHECK: outerloop:
; CHECK-NEXT: [[T2:%.*]] = phi i32 [ [[V17:%.*]], [[OUTEREND:%.*]] ], [ [[T1]], [[ENTRY:%.*]] ]
; CHECK-NEXT: [[J_0136:%.*]] = phi i32 [ [[INC144:%.*]], [[OUTEREND]] ], [ 0, [[ENTRY]] ]
; CHECK-NEXT: [[SCORE_1135:%.*]] = phi i32 [ [[CALL142:%.*]], [[OUTEREND]] ], [ -939524096, [[ENTRY]] ]
; CHECK-NEXT: [[T3:%.*]] = load float**, float*** @mm, align 4
; CHECK-NEXT: [[ARRAYIDX109:%.*]] = getelementptr inbounds float*, float** [[T3]], i32 [[T2]]
; CHECK-NEXT: [[T4:%.*]] = load float*, float** [[ARRAYIDX109]], align 4
; CHECK-NEXT: [[T5:%.*]] = load float**, float*** @vv, align 4
; CHECK-NEXT: [[ARRAYIDX111:%.*]] = getelementptr inbounds float*, float** [[T5]], i32 [[T2]]
; CHECK-NEXT: [[T6:%.*]] = load float*, float** [[ARRAYIDX111]], align 4
; CHECK-NEXT: [[T7:%.*]] = load float*, float** @ll, align 4
; CHECK-NEXT: [[ARRAYIDX113:%.*]] = getelementptr inbounds float, float* [[T7]], i32 [[T2]]
; CHECK-NEXT: [[T8:%.*]] = load float, float* [[ARRAYIDX113]], align 4
; CHECK-NEXT: [[CONV114:%.*]] = fpext float [[T8]] to double
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[T]], 2
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
; CHECK: vector.ph:
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i32 [[T]], 2
; CHECK-NEXT: [[N_VEC:%.*]] = sub i32 [[T]], [[N_MOD_VF]]
; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x double> zeroinitializer, double [[CONV114]], i32 0
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
; CHECK: vector.body:
; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x double> [ [[TMP0]], [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ]
; CHECK-NEXT: [[TMP1:%.*]] = add i32 [[INDEX]], 0
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, float* [[X:%.*]], i32 [[TMP1]]
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, float* [[TMP2]], i32 0
; CHECK-NEXT: [[TMP4:%.*]] = bitcast float* [[TMP3]] to <2 x float>*
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x float>, <2 x float>* [[TMP4]], align 4
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, float* [[T4]], i32 [[TMP1]]
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[TMP5]], i32 0
; CHECK-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <2 x float>*
; CHECK-NEXT: [[WIDE_LOAD1:%.*]] = load <2 x float>, <2 x float>* [[TMP7]], align 4
; CHECK-NEXT: [[TMP8:%.*]] = fsub fast <2 x float> [[WIDE_LOAD]], [[WIDE_LOAD1]]
; CHECK-NEXT: [[TMP9:%.*]] = fpext <2 x float> [[TMP8]] to <2 x double>
; CHECK-NEXT: [[TMP10:%.*]] = fmul fast <2 x double> [[TMP9]], [[TMP9]]
; CHECK-NEXT: [[TMP11:%.*]] = getelementptr inbounds float, float* [[T6]], i32 [[TMP1]]
; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds float, float* [[TMP11]], i32 0
; CHECK-NEXT: [[TMP13:%.*]] = bitcast float* [[TMP12]] to <2 x float>*
; CHECK-NEXT: [[WIDE_LOAD2:%.*]] = load <2 x float>, <2 x float>* [[TMP13]], align 4
; CHECK-NEXT: [[TMP14:%.*]] = fpext <2 x float> [[WIDE_LOAD2]] to <2 x double>
; CHECK-NEXT: [[TMP15:%.*]] = fmul fast <2 x double> [[TMP10]], [[TMP14]]
; CHECK-NEXT: [[TMP16]] = fsub fast <2 x double> [[VEC_PHI]], [[TMP15]]
; CHECK-NEXT: [[INDEX_NEXT]] = add i32 [[INDEX]], 2
; CHECK-NEXT: [[TMP17:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]]
; CHECK-NEXT: br i1 [[TMP17]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0
; CHECK: middle.block:
; CHECK-NEXT: [[RDX_SHUF:%.*]] = shufflevector <2 x double> [[TMP16]], <2 x double> undef, <2 x i32> <i32 1, i32 undef>
; CHECK-NEXT: [[BIN_RDX:%.*]] = fadd fast <2 x double> [[TMP16]], [[RDX_SHUF]]
; CHECK-NEXT: [[TMP18:%.*]] = extractelement <2 x double> [[BIN_RDX]], i32 0
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[T]], [[N_VEC]]
; CHECK-NEXT: br i1 [[CMP_N]], label [[OUTEREND]], label [[SCALAR_PH]]
; CHECK: scalar.ph:
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[OUTERLOOP]] ]
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi double [ [[CONV114]], [[OUTERLOOP]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: br label [[INNERLOOP:%.*]]
; CHECK: innerloop:
; CHECK-NEXT: [[I_2132:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[INC129:%.*]], [[INNERLOOP]] ]
; CHECK-NEXT: [[DVAL1_4131:%.*]] = phi double [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[SUB127:%.*]], [[INNERLOOP]] ]
; CHECK-NEXT: [[ARRAYIDX119:%.*]] = getelementptr inbounds float, float* [[X]], i32 [[I_2132]]
; CHECK-NEXT: [[T9:%.*]] = load float, float* [[ARRAYIDX119]], align 4
; CHECK-NEXT: [[ARRAYIDX120:%.*]] = getelementptr inbounds float, float* [[T4]], i32 [[I_2132]]
; CHECK-NEXT: [[T10:%.*]] = load float, float* [[ARRAYIDX120]], align 4
; CHECK-NEXT: [[SUB121:%.*]] = fsub fast float [[T9]], [[T10]]
; CHECK-NEXT: [[CONV122:%.*]] = fpext float [[SUB121]] to double
; CHECK-NEXT: [[MUL123:%.*]] = fmul fast double [[CONV122]], [[CONV122]]
; CHECK-NEXT: [[ARRAYIDX124:%.*]] = getelementptr inbounds float, float* [[T6]], i32 [[I_2132]]
; CHECK-NEXT: [[T11:%.*]] = load float, float* [[ARRAYIDX124]], align 4
; CHECK-NEXT: [[CONV125:%.*]] = fpext float [[T11]] to double
; CHECK-NEXT: [[MUL126:%.*]] = fmul fast double [[MUL123]], [[CONV125]]
; CHECK-NEXT: [[SUB127]] = fsub fast double [[DVAL1_4131]], [[MUL126]]
; CHECK-NEXT: [[INC129]] = add nuw nsw i32 [[I_2132]], 1
; CHECK-NEXT: [[EXITCOND143:%.*]] = icmp eq i32 [[INC129]], [[T]]
; CHECK-NEXT: br i1 [[EXITCOND143]], label [[OUTEREND]], label [[INNERLOOP]], !llvm.loop !2
; CHECK: outerend:
; CHECK-NEXT: [[SUB127_LCSSA:%.*]] = phi double [ [[SUB127]], [[INNERLOOP]] ], [ [[TMP18]], [[MIDDLE_BLOCK]] ]
; CHECK-NEXT: [[CONV138:%.*]] = fptosi double [[SUB127_LCSSA]] to i32
; CHECK-NEXT: [[CALL142]] = add nuw nsw i32 [[SCORE_1135]], [[CONV138]]
; CHECK-NEXT: [[INC144]] = add nuw nsw i32 [[J_0136]], 1
; CHECK-NEXT: [[ARRAYIDX102:%.*]] = getelementptr inbounds i32, i32* @a, i32 [[INC144]]
; CHECK-NEXT: [[V17]] = load i32, i32* [[ARRAYIDX102]], align 4
; CHECK-NEXT: [[CMP103:%.*]] = icmp sgt i32 [[V17]], -1
; CHECK-NEXT: br i1 [[CMP103]], label [[OUTERLOOP]], label [[EXIT:%.*]]
; CHECK: exit:
; CHECK-NEXT: ret i32 [[CALL142]]
;
entry:
%t = load i32, i32* @v, align 8
%t1 = load i32, i32* @a, align 4
br label %outerloop
outerloop:
%t2 = phi i32 [ %v17, %outerend ], [ %t1, %entry ]
%j.0136 = phi i32 [ %inc144, %outerend ], [ 0, %entry ]
%score.1135 = phi i32 [ %call142, %outerend ], [ -939524096, %entry ]
%t3 = load float**, float*** @mm, align 4
%arrayidx109 = getelementptr inbounds float*, float** %t3, i32 %t2
%t4 = load float*, float** %arrayidx109, align 4
%t5 = load float**, float*** @vv, align 4
%arrayidx111 = getelementptr inbounds float*, float** %t5, i32 %t2
%t6 = load float*, float** %arrayidx111, align 4
%t7 = load float*, float** @ll, align 4
%arrayidx113 = getelementptr inbounds float, float* %t7, i32 %t2
%t8 = load float, float* %arrayidx113, align 4
%conv114 = fpext float %t8 to double
br label %innerloop
innerloop:
%i.2132 = phi i32 [ 0, %outerloop ], [ %inc129, %innerloop ]
%dval1.4131 = phi double [ %conv114, %outerloop ], [ %sub127, %innerloop ]
%arrayidx119 = getelementptr inbounds float, float* %x, i32 %i.2132
%t9 = load float, float* %arrayidx119, align 4
%arrayidx120 = getelementptr inbounds float, float* %t4, i32 %i.2132
%t10 = load float, float* %arrayidx120, align 4
%sub121 = fsub fast float %t9, %t10
%conv122 = fpext float %sub121 to double
%mul123 = fmul fast double %conv122, %conv122
%arrayidx124 = getelementptr inbounds float, float* %t6, i32 %i.2132
%t11 = load float, float* %arrayidx124, align 4
%conv125 = fpext float %t11 to double
%mul126 = fmul fast double %mul123, %conv125
%sub127 = fsub fast double %dval1.4131, %mul126
%inc129 = add nuw nsw i32 %i.2132, 1
%exitcond143 = icmp eq i32 %inc129, %t
br i1 %exitcond143, label %outerend, label %innerloop
outerend:
%sub127.lcssa = phi double [ %sub127, %innerloop ]
%conv138 = fptosi double %sub127.lcssa to i32
%call142 = add nuw nsw i32 %score.1135, %conv138
%inc144 = add nuw nsw i32 %j.0136, 1
%arrayidx102 = getelementptr inbounds i32, i32* @a, i32 %inc144
%v17 = load i32, i32* %arrayidx102, align 4
%cmp103 = icmp sgt i32 %v17, -1
br i1 %cmp103, label %outerloop, label %exit
exit:
ret i32 %call142
}
Reference in New Issue View Git Blame Copy Permalink