[PATCH] D33768: [Polly] Exclude Ignored Intrinsics from explicit instruction list

[Nandini Singhal via Phabricator via llvm-commits]

nandini12396 added a comment.

Sir, sorry but I still didnt understand why the polly generated code for this test case (`opt  -polly-codegen -polly-process-unprofitable`), more intrinsics are generated :

  ; ModuleID = '../llvm/tools/polly/test/Isl/CodeGen/intrinsics_misc.ll'
  source_filename = "../llvm/tools/polly/test/Isl/CodeGen/intrinsics_misc.ll"
  target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
  
  @A = common global [1024 x i32] zeroinitializer, align 16
  
  ; Function Attrs: nounwind uwtable
  define void @jd() #0 {
  entry:
    %exitcond.s2a = alloca i1
    %tmp = alloca [1024 x i32], align 16
    %tmp3 = bitcast [1024 x i32]* @A to i8*
    br label %polly.split_new_and_old
  
  polly.split_new_and_old:                          ; preds = %entry
    br i1 true, label %polly.start, label %for.cond.pre_entry_bb
  
  for.cond.pre_entry_bb:                            ; preds = %polly.split_new_and_old
    br label %for.cond
  
  for.cond:                                         ; preds = %for.cond.pre_entry_bb, %for.inc11
    %indvars.iv3 = phi i64 [ %indvars.iv.next4, %for.inc11 ], [ 0, %for.cond.pre_entry_bb ]
    %exitcond5 = icmp ne i64 %indvars.iv3, 1024
    br i1 %exitcond5, label %for.body, label %polly.merge_new_and_old
  
  for.body:                                         ; preds = %for.cond
    %lis = call {}* @llvm.invariant.start.p0i8(i64 4096, i8* %tmp3)
    br label %for.cond2
  
  for.cond2:                                        ; preds = %for.inc, %for.body
    %indvars.iv1 = phi i64 [ %indvars.iv.next2, %for.inc ], [ %indvars.iv3, %for.body ]
    %lftr.wideiv = trunc i64 %indvars.iv1 to i32
    %exitcond = icmp ne i32 %lftr.wideiv, 1024
    br i1 %exitcond, label %for.body4, label %for.end
  
  for.body4:                                        ; preds = %for.cond2
    call void @llvm.assume(i1 %exitcond)
    %0 = call i1 @llvm.expect.i1(i1 %exitcond, i1 true)
    %arrayidx = getelementptr inbounds [1024 x i32], [1024 x i32]* @A, i64 0, i64 %indvars.iv1
    %tmp6 = load i32, i32* %arrayidx, align 4
    %arrayidx6 = getelementptr inbounds [1024 x i32], [1024 x i32]* %tmp, i64 0, i64 %indvars.iv3
    call void @llvm.donothing()
    %tmp7 = load i32, i32* %arrayidx6, align 4
    %add = add nsw i32 %tmp7, %tmp6
    store i32 %add, i32* %arrayidx6, align 4
    br label %for.inc
  
  for.inc:                                          ; preds = %for.body4
    %indvars.iv.next2 = add nuw nsw i64 %indvars.iv1, 1
    br label %for.cond2
  
  for.end:                                          ; preds = %for.cond2
    %arrayidx8 = getelementptr inbounds [1024 x i32], [1024 x i32]* %tmp, i64 0, i64 %indvars.iv3
    %tmp8 = load i32, i32* %arrayidx8, align 4
    %arrayidx10 = getelementptr inbounds [1024 x i32], [1024 x i32]* @A, i64 0, i64 %indvars.iv3
    call void @llvm.invariant.end.p0i8({}* %lis, i64 4096, i8* %tmp3)
    store i32 %tmp8, i32* %arrayidx10, align 4
    br label %for.inc11
  
  for.inc11:                                        ; preds = %for.end
    %indvars.iv.next4 = add nuw nsw i64 %indvars.iv3, 1
    br label %for.cond
  
  polly.merge_new_and_old:                          ; preds = %polly.exiting, %for.cond
    br label %for.end13
  
  for.end13:                                        ; preds = %polly.merge_new_and_old
    ret void
  
  polly.start:                                      ; preds = %polly.split_new_and_old
    br label %polly.loop_preheader
  
  polly.loop_exit:                                  ; preds = %polly.stmt.for.end
    br label %polly.exiting
  
  polly.exiting:                                    ; preds = %polly.loop_exit
    br label %polly.merge_new_and_old
  
  polly.loop_header:                                ; preds = %polly.stmt.for.end, %polly.loop_preheader
    %polly.indvar = phi i64 [ 0, %polly.loop_preheader ], [ %polly.indvar_next, %polly.stmt.for.end ]
    %1 = sub nsw i64 0, %polly.indvar
    %2 = add nsw i64 %1, 1024
    br label %polly.loop_preheader2
  
  polly.loop_exit3:                                 ; preds = %polly.merge
    br label %polly.stmt.for.end
  
  polly.stmt.for.end:                               ; preds = %polly.loop_exit3
    %scevgep9 = getelementptr [1024 x i32], [1024 x i32]* %tmp, i64 0, i64 %polly.indvar
    %tmp8_p_scalar_ = load i32, i32* %scevgep9, align 4, !alias.scope !0, !noalias !2
    %scevgep10 = getelementptr [1024 x i32], [1024 x i32]* @A, i64 0, i64 %polly.indvar
    store i32 %tmp8_p_scalar_, i32* %scevgep10, align 4, !alias.scope !4, !noalias !5
    %polly.indvar_next = add nsw i64 %polly.indvar, 1
    %polly.loop_cond = icmp sle i64 %polly.indvar_next, 1023
    br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit
  
  polly.loop_preheader:                             ; preds = %polly.start
    br label %polly.loop_header
  
  polly.loop_header1:                               ; preds = %polly.merge, %polly.loop_preheader2
    %polly.indvar4 = phi i64 [ 0, %polly.loop_preheader2 ], [ %polly.indvar_next5, %polly.merge ]
    br label %polly.stmt.for.cond2
  
  polly.stmt.for.cond2:                             ; preds = %polly.loop_header1
    %3 = trunc i64 %polly.indvar4 to i32
    %4 = add i32 %7, %3
    %p_exitcond = icmp ne i32 %4, 1024
    store i1 %p_exitcond, i1* %exitcond.s2a
    br label %polly.cond
  
  polly.cond:                                       ; preds = %polly.stmt.for.cond2
    %5 = add nsw i64 %polly.indvar, %polly.indvar4
    %6 = icmp sle i64 %5, 1023
    br i1 %6, label %polly.then, label %polly.else
  
  polly.merge:                                      ; preds = %polly.else, %polly.stmt.for.body4
    %polly.indvar_next5 = add nsw i64 %polly.indvar4, 1
    %polly.loop_cond6 = icmp sle i64 %polly.indvar_next5, %2
    br i1 %polly.loop_cond6, label %polly.loop_header1, label %polly.loop_exit3
  
  polly.loop_preheader2:                            ; preds = %polly.loop_header
    %7 = trunc i64 %polly.indvar to i32
    %scevgep = getelementptr [1024 x i32], [1024 x i32]* @A, i64 0, i64 %polly.indvar
    %scevgep8 = getelementptr [1024 x i32], [1024 x i32]* %tmp, i64 0, i64 %polly.indvar
    br label %polly.loop_header1
  
  polly.then:                                       ; preds = %polly.cond
    br label %polly.stmt.for.body4
  
  polly.stmt.for.body4:                             ; preds = %polly.then
    %scevgep7 = getelementptr i32, i32* %scevgep, i64 %polly.indvar4
    %tmp6_p_scalar_ = load i32, i32* %scevgep7, align 4, !alias.scope !4, !noalias !5
    %tmp7_p_scalar_ = load i32, i32* %scevgep8, align 4, !alias.scope !0, !noalias !2
    %p_add = add nsw i32 %tmp7_p_scalar_, %tmp6_p_scalar_
    store i32 %p_add, i32* %scevgep8, align 4, !alias.scope !0, !noalias !2
    br label %polly.merge
  
  polly.else:                                       ; preds = %polly.cond
    br label %polly.merge
  }
  
  ; Function Attrs: nounwind readnone
  declare void @llvm.donothing() #1
  
  ; Function Attrs: nounwind
  declare void @llvm.assume(i1) #2
  
  ; Function Attrs: nounwind readnone
  declare i1 @llvm.expect.i1(i1, i1) #1
  
  ; Function Attrs: argmemonly nounwind
  declare {}* @llvm.invariant.start.p0i8(i64, i8* nocapture) #3
  
  ; Function Attrs: argmemonly nounwind
  declare void @llvm.invariant.end.p0i8({}*, i64, i8* nocapture) #3
  
  attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "polly-optimized" "stack-protector-buffer-size"="8" "unsafe-fp-math"="true" "use-soft-float"="false" }
  attributes #1 = { nounwind readnone }
  attributes #2 = { nounwind }
  attributes #3 = { argmemonly nounwind }
  
  !0 = distinct !{!0, !1, !"polly.alias.scope.MemRef2"}
  !1 = distinct !{!1, !"polly.alias.scope.domain"}
  !2 = !{!3, !4}
  !3 = distinct !{!3, !1, !"polly.alias.scope.MemRef0"}
  !4 = distinct !{!4, !1, !"polly.alias.scope.MemRef1"}
  !5 = !{!3, !0}


https://reviews.llvm.org/D33768