Good input IR with optimizations

good_input_ir_opt

; ModuleID = '__compute_module'
source_filename = "__compute_module"
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-grtev4-linux-gnu"

; Function Attrs: nounwind uwtable
define void @EntryModule(i8* nocapture readnone %retval, i8* noalias %run_options, i8** noalias nocapture readnone %params, i8** noalias nocapture readonly %buffer_table, i64* noalias nocapture readnone %prof_counters) local_unnamed_addr #0 {
entry:
  %0 = getelementptr inbounds i8*, i8** %buffer_table, i64 2
  %1 = bitcast i8** %0 to i64**
  %2 = load i64*, i64** %1, align 8, !invariant.load !0, !dereferenceable !1, !align !1
  %3 = load i8*, i8** %buffer_table, align 8, !invariant.load !0, !dereferenceable !2, !align !1
  %4 = load i64, i64* %2, align 8, !invariant.load !0, !noalias !3
  %5 = insertelement <4 x i64> undef, i64 %4, i32 0
  %6 = shufflevector <4 x i64> %5, <4 x i64> undef, <4 x i32> zeroinitializer
  %7 = bitcast i8* %3 to <4 x i64>*
  store <4 x i64> %6, <4 x i64>* %7, align 8, !alias.scope !3, !noalias !6
  %8 = getelementptr inbounds i8*, i8** %buffer_table, i64 3
  %9 = bitcast i8** %8 to [2 x [2 x double]]**
  %10 = load [2 x [2 x double]]*, [2 x [2 x double]]** %9, align 8, !invariant.load !0, !dereferenceable !8, !align !1
  %11 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %10, i64 0, i64 0, i64 0
  %12 = bitcast i8* %3 to double*
  tail call void @__xla_cpu_runtime_EigenMatMulF64(i8* %run_options, double* %11, double* %12, double* %12, i64 2, i64 2, i64 2, i32 1, i32 0)
  %13 = getelementptr inbounds i8*, i8** %buffer_table, i64 1
  %14 = bitcast i8** %13 to [3 x [3 x double]]**
  %15 = load [3 x [3 x double]]*, [3 x [3 x double]]** %14, align 8, !invariant.load !0, !dereferenceable !9, !align !1
  %16 = load double, double* %11, align 8, !alias.scope !6, !noalias !10
  %17 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %15, i64 0, i64 0, i64 0
  %18 = load double, double* %17, align 8, !invariant.load !0, !noalias !12
  %19 = fmul fast double %18, %16
  %20 = bitcast i8* %3 to double*
  store double %19, double* %20, align 8, !alias.scope !13, !noalias !6
  %21 = getelementptr inbounds i8, i8* %3, i64 8
  %22 = bitcast i8* %21 to <2 x double>*
  store <2 x double> zeroinitializer, <2 x double>* %22, align 8, !alias.scope !13, !noalias !6
  %23 = getelementptr inbounds i8, i8* %3, i64 24
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(24) %23, i8 0, i64 24, i1 false)
  %24 = getelementptr inbounds i8, i8* %3, i64 48
  %25 = bitcast i8* %24 to double*
  store double 0.000000e+00, double* %25, align 8, !alias.scope !13, !noalias !6
  %26 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %15, i64 0, i64 1, i64 1
  %27 = load double, double* %26, align 8, !invariant.load !0, !noalias !12
  %28 = fmul fast double %27, %16
  %29 = getelementptr inbounds i8, i8* %3, i64 56
  %30 = bitcast i8* %29 to double*
  store double %28, double* %30, align 8, !alias.scope !13, !noalias !6
  %31 = getelementptr inbounds i8, i8* %3, i64 64
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(32) %31, i8 0, i64 32, i1 false)
  %32 = getelementptr inbounds i8, i8* %3, i64 96
  %33 = bitcast i8* %32 to <2 x double>*
  store <2 x double> zeroinitializer, <2 x double>* %33, align 8, !alias.scope !13, !noalias !6
  %34 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %15, i64 0, i64 2, i64 2
  %35 = load double, double* %34, align 8, !invariant.load !0, !noalias !12
  %36 = fmul fast double %35, %16
  %37 = getelementptr inbounds i8, i8* %3, i64 112
  %38 = bitcast i8* %37 to double*
  store double %36, double* %38, align 8, !alias.scope !13, !noalias !6
  %39 = getelementptr inbounds i8, i8* %3, i64 120
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(24) %39, i8 0, i64 24, i1 false)
  %40 = getelementptr inbounds i8, i8* %3, i64 144
  %41 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %10, i64 0, i64 1, i64 1
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(24) %40, i8 0, i64 24, i1 false)
  %42 = load double, double* %41, align 8, !alias.scope !6, !noalias !10
  %43 = fmul fast double %18, %42
  %44 = getelementptr inbounds i8, i8* %3, i64 168
  %45 = bitcast i8* %44 to double*
  store double %43, double* %45, align 8, !alias.scope !13, !noalias !6
  %46 = getelementptr inbounds i8, i8* %3, i64 176
  %47 = fmul fast double %27, %42
  %48 = getelementptr inbounds i8, i8* %3, i64 224
  %49 = bitcast i8* %48 to double*
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(48) %46, i8 0, i64 48, i1 false)
  store double %47, double* %49, align 8, !alias.scope !13, !noalias !6
  %50 = getelementptr inbounds i8, i8* %3, i64 232
  %51 = fmul fast double %35, %42
  %52 = getelementptr inbounds i8, i8* %3, i64 280
  %53 = bitcast i8* %52 to double*
  call void @llvm.memset.p0i8.i64(i8* nonnull align 8 dereferenceable(48) %50, i8 0, i64 48, i1 false)
  store double %51, double* %53, align 8, !alias.scope !13, !noalias !6
  ret void
}

; Function Attrs: argmemonly nounwind
declare void @__xla_cpu_runtime_EigenMatMulF64(i8*, double*, double*, double*, i64, i64, i64, i32, i32) local_unnamed_addr #1

; Function Attrs: argmemonly nounwind willreturn writeonly
declare void @llvm.memset.p0i8.i64(i8* nocapture writeonly, i8, i64, i1 immarg) #2

attributes #0 = { nounwind uwtable "denormal-fp-math"="preserve-sign" "no-frame-pointer-elim"="false" }
attributes #1 = { argmemonly nounwind }
attributes #2 = { argmemonly nounwind willreturn writeonly }

!0 = !{}
!1 = !{i64 8}
!2 = !{i64 288}
!3 = !{!4}
!4 = !{!"buffer: {index:0, offset:0, size:32}", !5}
!5 = !{!"XLA global AA domain"}
!6 = !{!7}
!7 = !{!"buffer: {index:3, offset:0, size:32}", !5}
!8 = !{i64 32}
!9 = !{i64 72}
!10 = !{!4, !11}
!11 = !{!"buffer: {index:0, offset:0, size:288}", !5}
!12 = !{!11, !7}
!13 = !{!11}