no optimizations, with gemm

no_opt_gemm

; 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"

@0 = private unnamed_addr constant [8 x i8] zeroinitializer, align 8
@1 = private unnamed_addr constant [4 x i8] c"\00\00\80\BF"
@2 = private unnamed_addr constant [4 x i8] c"\00\00\00@"
@3 = private unnamed_addr constant [4 x i8] zeroinitializer
@4 = private unnamed_addr constant [8 x i8] c"\00\00\00\00\00\00\F0?"
@5 = private unnamed_addr constant [8 x i8] c"4\EF\FC\99y\82\DA?"
@6 = private unnamed_addr constant [8 x i8] c"\FD\FF\FF\FF\FF\FF\FF\FF"
@7 = private unnamed_addr constant [8 x i8] zeroinitializer
@8 = private unnamed_addr constant [8 x i8] zeroinitializer
@9 = private unnamed_addr constant [4 x i8] c"\00\00\80\BF"
@10 = private unnamed_addr constant [4 x i8] c"\00\00\00@"
@11 = private unnamed_addr constant [4 x i8] zeroinitializer
@12 = private unnamed_addr constant [8 x i8] c"\00\00\00\00\00\00\F0?"
@13 = private unnamed_addr constant [8 x i8] c"4\EF\FC\99y\82\DA?"
@14 = private unnamed_addr constant [8 x i8] c"\FE\FF\FF\FF\FF\FF\FF\FF"
@15 = private unnamed_addr constant [8 x i8] zeroinitializer
@16 = private unnamed_addr constant [8 x i8] zeroinitializer
@17 = private unnamed_addr constant [4 x i8] c"\00\00\80\BF"
@18 = private unnamed_addr constant [4 x i8] c"\00\00\00@"
@19 = private unnamed_addr constant [4 x i8] zeroinitializer
@20 = private unnamed_addr constant [8 x i8] c"\00\00\00\00\00\00\F0?"
@21 = private unnamed_addr constant [8 x i8] c"4\EF\FC\99y\82\DA?"
@22 = private unnamed_addr constant [8 x i8] zeroinitializer
@23 = private unnamed_addr constant [8 x i8] zeroinitializer
@24 = private unnamed_addr constant [8 x i8] zeroinitializer
@25 = private unnamed_addr constant [4 x i8] c"\00\00\80\BF"
@26 = private unnamed_addr constant [4 x i8] c"\00\00\00@"
@27 = private unnamed_addr constant [4 x i8] zeroinitializer
@28 = private unnamed_addr constant [8 x i8] c"\00\00\00\00\00\00\F0?"
@29 = private unnamed_addr constant [8 x i8] c"4\EF\FC\99y\82\DA?"

; Function Attrs: uwtable
define internal void @add_F64.98(i8* %retval, i8* noalias %run_options, i8** noalias %params, i8** noalias %buffer_table, i64* noalias %prof_counters) #0 {
entry:
  %add.101 = alloca double, align 8
  %0 = getelementptr inbounds i8*, i8** %params, i64 0
  %1 = load i8*, i8** %0, !dereferenceable !0, !align !0
  %lhs.99 = bitcast i8* %1 to double*
  %2 = getelementptr inbounds i8*, i8** %params, i64 1
  %3 = load i8*, i8** %2, !dereferenceable !0, !align !0
  %rhs.100 = bitcast i8* %3 to double*
  %4 = load double, double* %lhs.99, !alias.scope !1, !noalias !4
  %5 = load double, double* %rhs.100, !alias.scope !6, !noalias !4
  %6 = fadd fast double %4, %5
  store double %6, double* %add.101, !alias.scope !4
  %load_ret_value = load double, double* %add.101
  %7 = bitcast i8* %retval to double*
  store double %load_ret_value, double* %7
  ret void
}

; Function Attrs: uwtable
define internal void @add_F64.181(i8* %retval, i8* noalias %run_options, i8** noalias %params, i8** noalias %buffer_table, i64* noalias %prof_counters) #0 {
entry:
  %add.184 = alloca double, align 8
  %0 = getelementptr inbounds i8*, i8** %params, i64 0
  %1 = load i8*, i8** %0, !dereferenceable !0, !align !0
  %lhs.182 = bitcast i8* %1 to double*
  %2 = getelementptr inbounds i8*, i8** %params, i64 1
  %3 = load i8*, i8** %2, !dereferenceable !0, !align !0
  %rhs.183 = bitcast i8* %3 to double*
  %4 = load double, double* %lhs.182, !alias.scope !8, !noalias !10
  %5 = load double, double* %rhs.183, !alias.scope !12, !noalias !10
  %6 = fadd fast double %4, %5
  store double %6, double* %add.184, !alias.scope !10
  %load_ret_value = load double, double* %add.184
  %7 = bitcast i8* %retval to double*
  store double %load_ret_value, double* %7
  ret void
}

; Function Attrs: uwtable
define internal void @add_F64.318(i8* %retval, i8* noalias %run_options, i8** noalias %params, i8** noalias %buffer_table, i64* noalias %prof_counters) #0 {
entry:
  %add.321 = alloca double, align 8
  %0 = getelementptr inbounds i8*, i8** %params, i64 0
  %1 = load i8*, i8** %0, !dereferenceable !0, !align !0
  %lhs.319 = bitcast i8* %1 to double*
  %2 = getelementptr inbounds i8*, i8** %params, i64 1
  %3 = load i8*, i8** %2, !dereferenceable !0, !align !0
  %rhs.320 = bitcast i8* %3 to double*
  %4 = load double, double* %lhs.319, !alias.scope !14, !noalias !16
  %5 = load double, double* %rhs.320, !alias.scope !18, !noalias !16
  %6 = fadd fast double %4, %5
  store double %6, double* %add.321, !alias.scope !16
  %load_ret_value = load double, double* %add.321
  %7 = bitcast i8* %retval to double*
  store double %load_ret_value, double* %7
  ret void
}

; Function Attrs: uwtable
define void @EntryModule(i8* %retval, i8* noalias %run_options, i8** noalias %params, i8** noalias %buffer_table, i64* noalias %prof_counters) #0 {
entry:
  %fusion.1.invar_address.dim.1 = alloca i64
  %fusion.1.invar_address.dim.0 = alloca i64
  %fusion.4.vectorized_inner.invar_address.reduction_dim.1 = alloca i64
  %accumulator32 = alloca <2 x double>
  %reduce.102.invar_address.dim.0 = alloca i64
  %fusion.4.invar_address.dim.2 = alloca i64
  %fusion.4.invar_address.dim.1 = alloca i64
  %fusion.4.invar_address.dim.0 = alloca i64
  %fusion.2.vectorized_inner.invar_address.reduction_dim.1 = alloca i64
  %accumulator24 = alloca <2 x double>
  %accumulator23 = alloca double
  %reduce.185.invar_address.dim.0 = alloca i64
  %fusion.2.invar_address.dim.2 = alloca i64
  %fusion.2.invar_address.dim.1 = alloca i64
  %fusion.2.invar_address.dim.0 = alloca i64
  %fusion.vectorized_inner.invar_address.reduction_dim.1 = alloca i64
  %accumulator15 = alloca <4 x double>
  %accumulator = alloca <2 x double>
  %reduce.322.invar_address.dim.0 = alloca i64
  %fusion.invar_address.dim.2 = alloca i64
  %fusion.invar_address.dim.1 = alloca i64
  %fusion.invar_address.dim.0 = alloca i64
  %accum_address10 = alloca double
  %dot.invar_address.reduction = alloca i64
  %dot.invar_address.rhs.0 = alloca i64
  %dot.invar_address.lhs.0 = alloca i64
  %fusion.5.invar_address.dim.1 = alloca i64
  %fusion.5.invar_address.dim.0 = alloca i64
  %accum_address = alloca double
  %dot.1.invar_address.reduction = alloca i64
  %dot.1.invar_address.rhs.0 = alloca i64
  %dot.1.invar_address.lhs.0 = alloca i64
  %fusion.3.invar_address.dim.1 = alloca i64
  %fusion.3.invar_address.dim.0 = alloca i64
  %0 = getelementptr inbounds i8*, i8** %buffer_table, i64 4
  %1 = load i8*, i8** %0, !invariant.load !20, !dereferenceable !21, !align !0
  %arg0.1 = bitcast i8* %1 to [3 x [1 x [1 x double]]]*
  %2 = getelementptr inbounds i8*, i8** %buffer_table, i64 9
  %3 = load i8*, i8** %2, !invariant.load !20, !dereferenceable !22, !align !0
  %arg6.7 = bitcast i8* %3 to [1 x [3 x float]]*
  %4 = getelementptr inbounds i8*, i8** %buffer_table, i64 5
  %5 = load i8*, i8** %4, !invariant.load !20, !dereferenceable !21, !align !0
  %arg7.8 = bitcast i8* %5 to [1 x [3 x double]]*
  %6 = getelementptr inbounds i8*, i8** %buffer_table, i64 2
  %7 = load i8*, i8** %6, !invariant.load !20, !dereferenceable !23, !align !0
  %arg8.9 = bitcast i8* %7 to [1 x [3 x [3 x double]]]*
  %8 = getelementptr inbounds i8*, i8** %buffer_table, i64 1
  %9 = load i8*, i8** %8, !invariant.load !20, !dereferenceable !24, !align !0
  %fusion.3 = bitcast i8* %9 to [3 x [3 x double]]*
  store i64 0, i64* %fusion.3.invar_address.dim.0
  br label %fusion.3.loop_header.dim.0

fusion.3.loop_header.dim.0:                       ; preds = %fusion.3.loop_exit.dim.1, %entry
  %fusion.3.indvar.dim.0 = load i64, i64* %fusion.3.invar_address.dim.0
  %10 = icmp uge i64 %fusion.3.indvar.dim.0, 3
  br i1 %10, label %fusion.3.loop_exit.dim.0, label %fusion.3.loop_body.dim.0

fusion.3.loop_body.dim.0:                         ; preds = %fusion.3.loop_header.dim.0
  store i64 0, i64* %fusion.3.invar_address.dim.1
  br label %fusion.3.loop_header.dim.1

fusion.3.loop_header.dim.1:                       ; preds = %fusion.3.loop_body.dim.1, %fusion.3.loop_body.dim.0
  %fusion.3.indvar.dim.1 = load i64, i64* %fusion.3.invar_address.dim.1
  %11 = icmp uge i64 %fusion.3.indvar.dim.1, 3
  br i1 %11, label %fusion.3.loop_exit.dim.1, label %fusion.3.loop_body.dim.1

fusion.3.loop_body.dim.1:                         ; preds = %fusion.3.loop_header.dim.1
  %12 = mul nuw nsw i64 %fusion.3.indvar.dim.1, 1
  %13 = add nuw nsw i64 0, %12
  %14 = udiv i64 %13, 3
  %15 = mul nuw nsw i64 %fusion.3.indvar.dim.0, 1
  %16 = add nuw nsw i64 0, %15
  %17 = udiv i64 %16, 3
  %18 = mul nuw nsw i64 %13, 1
  %19 = add nuw nsw i64 0, %18
  %20 = udiv i64 %19, 3
  %21 = mul nuw nsw i64 %16, 1
  %22 = add nuw nsw i64 0, %21
  %23 = udiv i64 %22, 3
  %24 = mul nuw nsw i64 %22, 1
  %25 = add nuw nsw i64 0, %24
  %26 = trunc i64 %25 to i32
  %27 = mul nuw nsw i64 %19, 1
  %28 = add nuw nsw i64 0, %27
  %29 = trunc i64 %28 to i32
  %30 = icmp eq i32 %26, %29
  %31 = zext i1 %30 to i8
  %32 = mul nuw nsw i64 %13, 1
  %33 = add nuw nsw i64 0, %32
  %34 = udiv i64 %33, 3
  %35 = load float, float* bitcast ([4 x i8]* @1 to float*)
  %36 = load float, float* bitcast ([4 x i8]* @2 to float*)
  %37 = getelementptr inbounds [1 x [3 x float]], [1 x [3 x float]]* %arg6.7, i64 0, i64 0, i64 %33
  %38 = load float, float* %37, !invariant.load !20, !noalias !25
  %39 = fmul fast float %36, %38
  %40 = fadd fast float %35, %39
  %41 = fcmp fast une float %40, %40
  %42 = zext i1 %41 to i8
  %43 = load float, float* bitcast ([4 x i8]* @3 to float*)
  %44 = fcmp fast one float %40, 0.000000e+00
  %45 = uitofp i1 %44 to float
  %46 = call fast float @llvm.copysign.f32(float %45, float %40)
  %47 = fcmp fast uno float %40, %40
  %48 = select fast i1 %47, float %40, float %46
  %49 = trunc i8 %42 to i1
  %50 = select fast i1 %49, float %43, float %48
  %51 = fpext float %50 to double
  %52 = load double, double* bitcast ([8 x i8]* @4 to double*)
  %53 = load double, double* bitcast ([8 x i8]* @5 to double*)
  %54 = getelementptr inbounds [1 x [3 x double]], [1 x [3 x double]]* %arg7.8, i64 0, i64 0, i64 %33
  %55 = load double, double* %54, !invariant.load !20, !noalias !25
  %56 = fmul fast double %53, %55
  %57 = fadd fast double %52, %56
  %58 = fmul fast double %51, %57
  %59 = mul nuw nsw i64 %13, 1
  %60 = add nuw nsw i64 0, %59
  %61 = udiv i64 %60, 3
  %62 = mul nuw nsw i64 %16, 1
  %63 = add nuw nsw i64 0, %62
  %64 = udiv i64 %63, 3
  %65 = load i64, i64* bitcast ([8 x i8]* @6 to i64*)
  %66 = mul nuw nsw i64 %60, 1
  %67 = add nuw nsw i64 0, %66
  %68 = mul nuw nsw i64 %63, 1
  %69 = add nuw nsw i64 0, %68
  %70 = sub i64 %67, %69
  %71 = icmp sle i64 %65, %70
  %72 = zext i1 %71 to i8
  %73 = load i64, i64* bitcast ([8 x i8]* @7 to i64*)
  %74 = icmp sle i64 %70, %73
  %75 = zext i1 %74 to i8
  %76 = and i8 %72, %75
  %77 = getelementptr inbounds [1 x [3 x [3 x double]]], [1 x [3 x [3 x double]]]* %arg8.9, i64 0, i64 0, i64 %16, i64 %13
  %78 = load double, double* %77, !invariant.load !20, !noalias !25
  %79 = load double, double* bitcast ([8 x i8]* @8 to double*)
  %80 = trunc i8 %76 to i1
  %81 = select fast i1 %80, double %78, double %79
  %82 = trunc i8 %31 to i1
  %83 = select fast i1 %82, double %58, double %81
  %84 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %fusion.3, i64 0, i64 %fusion.3.indvar.dim.0, i64 %fusion.3.indvar.dim.1
  store double %83, double* %84, !alias.scope !25, !noalias !27
  %invar.inc1 = add nuw nsw i64 %fusion.3.indvar.dim.1, 1
  store i64 %invar.inc1, i64* %fusion.3.invar_address.dim.1
  br label %fusion.3.loop_header.dim.1

fusion.3.loop_exit.dim.1:                         ; preds = %fusion.3.loop_header.dim.1
  %invar.inc = add nuw nsw i64 %fusion.3.indvar.dim.0, 1
  store i64 %invar.inc, i64* %fusion.3.invar_address.dim.0
  br label %fusion.3.loop_header.dim.0

fusion.3.loop_exit.dim.0:                         ; preds = %fusion.3.loop_header.dim.0
  %85 = getelementptr inbounds i8*, i8** %buffer_table, i64 0
  %86 = load i8*, i8** %85, !invariant.load !20, !dereferenceable !24, !align !0
  %dot.1 = bitcast i8* %86 to [3 x [3 x double]]*
  store i64 0, i64* %dot.1.invar_address.lhs.0
  br label %dot.1.loop_header.lhs.0

dot.1.loop_header.lhs.0:                          ; preds = %dot.1.loop_exit.rhs.0, %fusion.3.loop_exit.dim.0
  %dot.1.indvar.lhs.0 = load i64, i64* %dot.1.invar_address.lhs.0
  %87 = icmp uge i64 %dot.1.indvar.lhs.0, 3
  br i1 %87, label %dot.1.loop_exit.lhs.0, label %dot.1.loop_body.lhs.0

dot.1.loop_body.lhs.0:                            ; preds = %dot.1.loop_header.lhs.0
  store i64 0, i64* %dot.1.invar_address.rhs.0
  br label %dot.1.loop_header.rhs.0

dot.1.loop_header.rhs.0:                          ; preds = %dot.1.loop_exit.reduction, %dot.1.loop_body.lhs.0
  %dot.1.indvar.rhs.0 = load i64, i64* %dot.1.invar_address.rhs.0
  %88 = icmp uge i64 %dot.1.indvar.rhs.0, 3
  br i1 %88, label %dot.1.loop_exit.rhs.0, label %dot.1.loop_body.rhs.0

dot.1.loop_body.rhs.0:                            ; preds = %dot.1.loop_header.rhs.0
  store i64 0, i64* %dot.1.invar_address.reduction
  store double 0.000000e+00, double* %accum_address
  br label %dot.1.loop_header.reduction

dot.1.loop_header.reduction:                      ; preds = %dot.1.loop_body.reduction, %dot.1.loop_body.rhs.0
  %dot.1.indvar.reduction = load i64, i64* %dot.1.invar_address.reduction
  %89 = icmp uge i64 %dot.1.indvar.reduction, 3
  br i1 %89, label %dot.1.loop_exit.reduction, label %dot.1.loop_body.reduction

dot.1.loop_body.reduction:                        ; preds = %dot.1.loop_header.reduction
  %90 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %fusion.3, i64 0, i64 %dot.1.indvar.lhs.0, i64 %dot.1.indvar.reduction
  %91 = load double, double* %90, !alias.scope !25, !noalias !27
  %92 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %fusion.3, i64 0, i64 %dot.1.indvar.rhs.0, i64 %dot.1.indvar.reduction
  %93 = load double, double* %92, !alias.scope !25, !noalias !27
  %94 = load double, double* %accum_address
  %95 = fmul fast double %91, %93
  %96 = fadd fast double %94, %95
  store double %96, double* %accum_address
  %invar.inc4 = add nuw nsw i64 %dot.1.indvar.reduction, 1
  store i64 %invar.inc4, i64* %dot.1.invar_address.reduction
  br label %dot.1.loop_header.reduction, !llvm.loop !29

dot.1.loop_exit.reduction:                        ; preds = %dot.1.loop_header.reduction
  %97 = load double, double* %accum_address
  %98 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %dot.1, i64 0, i64 %dot.1.indvar.lhs.0, i64 %dot.1.indvar.rhs.0
  store double %97, double* %98, !alias.scope !27, !noalias !31
  %invar.inc3 = add nuw nsw i64 %dot.1.indvar.rhs.0, 1
  store i64 %invar.inc3, i64* %dot.1.invar_address.rhs.0
  br label %dot.1.loop_header.rhs.0

dot.1.loop_exit.rhs.0:                            ; preds = %dot.1.loop_header.rhs.0
  %invar.inc2 = add nuw nsw i64 %dot.1.indvar.lhs.0, 1
  store i64 %invar.inc2, i64* %dot.1.invar_address.lhs.0
  br label %dot.1.loop_header.lhs.0

dot.1.loop_exit.lhs.0:                            ; preds = %dot.1.loop_header.lhs.0
  %99 = getelementptr inbounds i8*, i8** %buffer_table, i64 3
  %100 = load i8*, i8** %99, !invariant.load !20, !dereferenceable !35, !align !0
  %arg1.2 = bitcast i8* %100 to [1 x [2 x [2 x double]]]*
  %101 = getelementptr inbounds i8*, i8** %buffer_table, i64 12
  %102 = load i8*, i8** %101, !invariant.load !20, !dereferenceable !0, !align !0
  %arg3.4 = bitcast i8* %102 to [1 x [2 x float]]*
  %103 = getelementptr inbounds i8*, i8** %buffer_table, i64 8
  %104 = load i8*, i8** %103, !invariant.load !20, !dereferenceable !36, !align !0
  %arg5.6 = bitcast i8* %104 to [1 x [2 x double]]*
  %105 = getelementptr inbounds i8*, i8** %buffer_table, i64 1
  %106 = load i8*, i8** %105, !invariant.load !20, !dereferenceable !24, !align !0
  %fusion.5 = bitcast i8* %106 to [2 x [2 x double]]*
  store i64 0, i64* %fusion.5.invar_address.dim.0
  br label %fusion.5.loop_header.dim.0

fusion.5.loop_header.dim.0:                       ; preds = %fusion.5.loop_exit.dim.1, %dot.1.loop_exit.lhs.0
  %fusion.5.indvar.dim.0 = load i64, i64* %fusion.5.invar_address.dim.0
  %107 = icmp uge i64 %fusion.5.indvar.dim.0, 2
  br i1 %107, label %fusion.5.loop_exit.dim.0, label %fusion.5.loop_body.dim.0

fusion.5.loop_body.dim.0:                         ; preds = %fusion.5.loop_header.dim.0
  store i64 0, i64* %fusion.5.invar_address.dim.1
  br label %fusion.5.loop_header.dim.1

fusion.5.loop_header.dim.1:                       ; preds = %fusion.5.loop_body.dim.1, %fusion.5.loop_body.dim.0
  %fusion.5.indvar.dim.1 = load i64, i64* %fusion.5.invar_address.dim.1
  %108 = icmp uge i64 %fusion.5.indvar.dim.1, 2
  br i1 %108, label %fusion.5.loop_exit.dim.1, label %fusion.5.loop_body.dim.1

fusion.5.loop_body.dim.1:                         ; preds = %fusion.5.loop_header.dim.1
  %109 = mul nuw nsw i64 %fusion.5.indvar.dim.1, 1
  %110 = add nuw nsw i64 0, %109
  %111 = udiv i64 %110, 2
  %112 = mul nuw nsw i64 %fusion.5.indvar.dim.0, 1
  %113 = add nuw nsw i64 0, %112
  %114 = udiv i64 %113, 2
  %115 = mul nuw nsw i64 %110, 1
  %116 = add nuw nsw i64 0, %115
  %117 = udiv i64 %116, 2
  %118 = mul nuw nsw i64 %113, 1
  %119 = add nuw nsw i64 0, %118
  %120 = udiv i64 %119, 2
  %121 = mul nuw nsw i64 %119, 1
  %122 = add nuw nsw i64 0, %121
  %123 = trunc i64 %122 to i32
  %124 = mul nuw nsw i64 %116, 1
  %125 = add nuw nsw i64 0, %124
  %126 = trunc i64 %125 to i32
  %127 = icmp eq i32 %123, %126
  %128 = zext i1 %127 to i8
  %129 = mul nuw nsw i64 %110, 1
  %130 = add nuw nsw i64 0, %129
  %131 = udiv i64 %130, 2
  %132 = load float, float* bitcast ([4 x i8]* @9 to float*)
  %133 = load float, float* bitcast ([4 x i8]* @10 to float*)
  %134 = getelementptr inbounds [1 x [2 x float]], [1 x [2 x float]]* %arg3.4, i64 0, i64 0, i64 %130
  %135 = load float, float* %134, !invariant.load !20, !noalias !37
  %136 = fmul fast float %133, %135
  %137 = fadd fast float %132, %136
  %138 = fcmp fast une float %137, %137
  %139 = zext i1 %138 to i8
  %140 = load float, float* bitcast ([4 x i8]* @11 to float*)
  %141 = fcmp fast one float %137, 0.000000e+00
  %142 = uitofp i1 %141 to float
  %143 = call fast float @llvm.copysign.f32(float %142, float %137)
  %144 = fcmp fast uno float %137, %137
  %145 = select fast i1 %144, float %137, float %143
  %146 = trunc i8 %139 to i1
  %147 = select fast i1 %146, float %140, float %145
  %148 = fpext float %147 to double
  %149 = load double, double* bitcast ([8 x i8]* @12 to double*)
  %150 = load double, double* bitcast ([8 x i8]* @13 to double*)
  %151 = getelementptr inbounds [1 x [2 x double]], [1 x [2 x double]]* %arg5.6, i64 0, i64 0, i64 %130
  %152 = load double, double* %151, !invariant.load !20, !noalias !37
  %153 = fmul fast double %150, %152
  %154 = fadd fast double %149, %153
  %155 = fmul fast double %148, %154
  %156 = mul nuw nsw i64 %110, 1
  %157 = add nuw nsw i64 0, %156
  %158 = udiv i64 %157, 2
  %159 = mul nuw nsw i64 %113, 1
  %160 = add nuw nsw i64 0, %159
  %161 = udiv i64 %160, 2
  %162 = load i64, i64* bitcast ([8 x i8]* @14 to i64*)
  %163 = mul nuw nsw i64 %157, 1
  %164 = add nuw nsw i64 0, %163
  %165 = mul nuw nsw i64 %160, 1
  %166 = add nuw nsw i64 0, %165
  %167 = sub i64 %164, %166
  %168 = icmp sle i64 %162, %167
  %169 = zext i1 %168 to i8
  %170 = load i64, i64* bitcast ([8 x i8]* @15 to i64*)
  %171 = icmp sle i64 %167, %170
  %172 = zext i1 %171 to i8
  %173 = and i8 %169, %172
  %174 = getelementptr inbounds [1 x [2 x [2 x double]]], [1 x [2 x [2 x double]]]* %arg1.2, i64 0, i64 0, i64 %113, i64 %110
  %175 = load double, double* %174, !invariant.load !20, !noalias !37
  %176 = load double, double* bitcast ([8 x i8]* @16 to double*)
  %177 = trunc i8 %173 to i1
  %178 = select fast i1 %177, double %175, double %176
  %179 = trunc i8 %128 to i1
  %180 = select fast i1 %179, double %155, double %178
  %181 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %fusion.5, i64 0, i64 %fusion.5.indvar.dim.0, i64 %fusion.5.indvar.dim.1
  store double %180, double* %181, !alias.scope !37, !noalias !39
  %invar.inc6 = add nuw nsw i64 %fusion.5.indvar.dim.1, 1
  store i64 %invar.inc6, i64* %fusion.5.invar_address.dim.1
  br label %fusion.5.loop_header.dim.1

fusion.5.loop_exit.dim.1:                         ; preds = %fusion.5.loop_header.dim.1
  %invar.inc5 = add nuw nsw i64 %fusion.5.indvar.dim.0, 1
  store i64 %invar.inc5, i64* %fusion.5.invar_address.dim.0
  br label %fusion.5.loop_header.dim.0

fusion.5.loop_exit.dim.0:                         ; preds = %fusion.5.loop_header.dim.0
  %182 = getelementptr inbounds i8*, i8** %buffer_table, i64 22
  %183 = load i8*, i8** %182, !invariant.load !20, !dereferenceable !40, !align !36
  %184 = getelementptr inbounds i8, i8* %183, i64 864
  %dot = bitcast i8* %184 to [2 x [2 x double]]*
  store i64 0, i64* %dot.invar_address.lhs.0
  br label %dot.loop_header.lhs.0

dot.loop_header.lhs.0:                            ; preds = %dot.loop_exit.rhs.0, %fusion.5.loop_exit.dim.0
  %dot.indvar.lhs.0 = load i64, i64* %dot.invar_address.lhs.0
  %185 = icmp uge i64 %dot.indvar.lhs.0, 2
  br i1 %185, label %dot.loop_exit.lhs.0, label %dot.loop_body.lhs.0

dot.loop_body.lhs.0:                              ; preds = %dot.loop_header.lhs.0
  store i64 0, i64* %dot.invar_address.rhs.0
  br label %dot.loop_header.rhs.0

dot.loop_header.rhs.0:                            ; preds = %dot.loop_exit.reduction, %dot.loop_body.lhs.0
  %dot.indvar.rhs.0 = load i64, i64* %dot.invar_address.rhs.0
  %186 = icmp uge i64 %dot.indvar.rhs.0, 2
  br i1 %186, label %dot.loop_exit.rhs.0, label %dot.loop_body.rhs.0

dot.loop_body.rhs.0:                              ; preds = %dot.loop_header.rhs.0
  store i64 0, i64* %dot.invar_address.reduction
  store double 0.000000e+00, double* %accum_address10
  br label %dot.loop_header.reduction

dot.loop_header.reduction:                        ; preds = %dot.loop_body.reduction, %dot.loop_body.rhs.0
  %dot.indvar.reduction = load i64, i64* %dot.invar_address.reduction
  %187 = icmp uge i64 %dot.indvar.reduction, 2
  br i1 %187, label %dot.loop_exit.reduction, label %dot.loop_body.reduction

dot.loop_body.reduction:                          ; preds = %dot.loop_header.reduction
  %188 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %fusion.5, i64 0, i64 %dot.indvar.lhs.0, i64 %dot.indvar.reduction
  %189 = load double, double* %188, !alias.scope !37, !noalias !39
  %190 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %fusion.5, i64 0, i64 %dot.indvar.rhs.0, i64 %dot.indvar.reduction
  %191 = load double, double* %190, !alias.scope !37, !noalias !39
  %192 = load double, double* %accum_address10
  %193 = fmul fast double %189, %191
  %194 = fadd fast double %192, %193
  store double %194, double* %accum_address10
  %invar.inc9 = add nuw nsw i64 %dot.indvar.reduction, 1
  store i64 %invar.inc9, i64* %dot.invar_address.reduction
  br label %dot.loop_header.reduction, !llvm.loop !41

dot.loop_exit.reduction:                          ; preds = %dot.loop_header.reduction
  %195 = load double, double* %accum_address10
  %196 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %dot, i64 0, i64 %dot.indvar.lhs.0, i64 %dot.indvar.rhs.0
  store double %195, double* %196, !alias.scope !39, !noalias !42
  %invar.inc8 = add nuw nsw i64 %dot.indvar.rhs.0, 1
  store i64 %invar.inc8, i64* %dot.invar_address.rhs.0
  br label %dot.loop_header.rhs.0

dot.loop_exit.rhs.0:                              ; preds = %dot.loop_header.rhs.0
  %invar.inc7 = add nuw nsw i64 %dot.indvar.lhs.0, 1
  store i64 %invar.inc7, i64* %dot.invar_address.lhs.0
  br label %dot.loop_header.lhs.0

dot.loop_exit.lhs.0:                              ; preds = %dot.loop_header.lhs.0
  %197 = getelementptr inbounds i8*, i8** %buffer_table, i64 10
  %198 = load i8*, i8** %197, !invariant.load !20, !dereferenceable !22, !align !0
  %arg2.3 = bitcast i8* %198 to [3 x [1 x float]]*
  %199 = getelementptr inbounds i8*, i8** %buffer_table, i64 6
  %200 = load i8*, i8** %199, !invariant.load !20, !dereferenceable !21, !align !0
  %arg4.5 = bitcast i8* %200 to [3 x [1 x double]]*
  %201 = getelementptr inbounds i8*, i8** %buffer_table, i64 22
  %202 = load i8*, i8** %201, !invariant.load !20, !dereferenceable !40, !align !36
  %fusion = bitcast i8* %202 to [3 x [6 x [6 x double]]]*
  store i64 0, i64* %fusion.invar_address.dim.0
  br label %fusion.loop_header.dim.0

fusion.loop_header.dim.0:                         ; preds = %fusion.loop_exit.dim.1, %dot.loop_exit.lhs.0
  %fusion.indvar.dim.0 = load i64, i64* %fusion.invar_address.dim.0
  %203 = icmp uge i64 %fusion.indvar.dim.0, 3
  br i1 %203, label %fusion.loop_exit.dim.0, label %fusion.loop_body.dim.0

fusion.loop_body.dim.0:                           ; preds = %fusion.loop_header.dim.0
  store i64 0, i64* %fusion.invar_address.dim.1
  br label %fusion.loop_header.dim.1

fusion.loop_header.dim.1:                         ; preds = %fusion.loop_exit.dim.2, %fusion.loop_body.dim.0
  %fusion.indvar.dim.1 = load i64, i64* %fusion.invar_address.dim.1
  %204 = icmp uge i64 %fusion.indvar.dim.1, 6
  br i1 %204, label %fusion.loop_exit.dim.1, label %fusion.loop_body.dim.1

fusion.loop_body.dim.1:                           ; preds = %fusion.loop_header.dim.1
  store i64 0, i64* %fusion.invar_address.dim.2
  br label %fusion.loop_header.dim.2

fusion.loop_header.dim.2:                         ; preds = %fusion.loop_body.dim.2, %fusion.loop_body.dim.1
  %fusion.indvar.dim.2 = load i64, i64* %fusion.invar_address.dim.2
  %205 = icmp uge i64 %fusion.indvar.dim.2, 6
  br i1 %205, label %fusion.loop_exit.dim.2, label %fusion.loop_body.dim.2

fusion.loop_body.dim.2:                           ; preds = %fusion.loop_header.dim.2
  %206 = mul nuw nsw i64 %fusion.indvar.dim.1, 1
  %207 = add nuw nsw i64 0, %206
  %208 = trunc i64 %207 to i32
  %209 = mul nuw nsw i64 %fusion.indvar.dim.2, 1
  %210 = add nuw nsw i64 0, %209
  %211 = trunc i64 %210 to i32
  %212 = icmp eq i32 %208, %211
  %213 = zext i1 %212 to i8
  %214 = mul nuw nsw i64 %fusion.indvar.dim.2, 1
  %215 = add nuw nsw i64 0, %214
  %216 = urem i64 %215, 3
  %217 = udiv i64 %215, 3
  %218 = udiv i64 %217, 2
  %219 = mul nuw nsw i64 %fusion.indvar.dim.1, 1
  %220 = add nuw nsw i64 0, %219
  %221 = urem i64 %220, 3
  %222 = udiv i64 %220, 3
  %223 = udiv i64 %222, 2
  %224 = mul nuw nsw i64 %fusion.indvar.dim.0, 1
  %225 = add nuw nsw i64 0, %224
  %226 = udiv i64 %225, 3
  %227 = mul nuw nsw i64 %217, 1
  %228 = add nuw nsw i64 0, %227
  %229 = udiv i64 %228, 2
  %230 = mul nuw nsw i64 %222, 1
  %231 = add nuw nsw i64 0, %230
  %232 = udiv i64 %231, 2
  %233 = mul nuw nsw i64 %225, 1
  %234 = add nuw nsw i64 0, %233
  %235 = udiv i64 %234, 3
  %236 = mul nuw nsw i64 %234, 1
  %237 = add nuw nsw i64 0, %236
  %238 = udiv i64 %237, 3
  %239 = load float, float* bitcast ([4 x i8]* @17 to float*)
  %240 = getelementptr inbounds [3 x [1 x float]], [3 x [1 x float]]* %arg2.3, i64 0, i64 %237, i64 0
  %241 = load float, float* %240, !invariant.load !20, !noalias !44
  %242 = load float, float* bitcast ([4 x i8]* @18 to float*)
  %243 = fmul fast float %241, %242
  %244 = fadd fast float %239, %243
  %245 = fcmp fast une float %244, %244
  %246 = zext i1 %245 to i8
  %247 = load float, float* bitcast ([4 x i8]* @19 to float*)
  %248 = fcmp fast one float %244, 0.000000e+00
  %249 = uitofp i1 %248 to float
  %250 = call fast float @llvm.copysign.f32(float %249, float %244)
  %251 = fcmp fast uno float %244, %244
  %252 = select fast i1 %251, float %244, float %250
  %253 = trunc i8 %246 to i1
  %254 = select fast i1 %253, float %247, float %252
  %255 = fpext float %254 to double
  %256 = load double, double* bitcast ([8 x i8]* @20 to double*)
  %257 = getelementptr inbounds [3 x [1 x double]], [3 x [1 x double]]* %arg4.5, i64 0, i64 %237, i64 0
  %258 = load double, double* %257, !invariant.load !20, !noalias !44
  %259 = load double, double* bitcast ([8 x i8]* @21 to double*)
  %260 = fmul fast double %258, %259
  %261 = fadd fast double %256, %260
  %262 = fmul fast double %255, %261
  %263 = fmul fast double %262, %262
  %264 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %dot, i64 0, i64 %231, i64 %228
  %265 = load double, double* %264, !alias.scope !39, !noalias !42
  %266 = fmul fast double %263, %265
  %267 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %dot.1, i64 0, i64 %221, i64 %216
  %268 = load double, double* %267, !alias.scope !27, !noalias !31
  %269 = fmul fast double %266, %268
  %270 = load double, double* bitcast ([8 x i8]* @22 to double*)
  %271 = trunc i8 %213 to i1
  %272 = select fast i1 %271, double %269, double %270
  %273 = getelementptr inbounds [3 x [6 x [6 x double]]], [3 x [6 x [6 x double]]]* %fusion, i64 0, i64 %fusion.indvar.dim.0, i64 %fusion.indvar.dim.1, i64 %fusion.indvar.dim.2
  store double %272, double* %273, !alias.scope !48, !noalias !49
  %invar.inc13 = add nuw nsw i64 %fusion.indvar.dim.2, 1
  store i64 %invar.inc13, i64* %fusion.invar_address.dim.2
  br label %fusion.loop_header.dim.2

fusion.loop_exit.dim.2:                           ; preds = %fusion.loop_header.dim.2
  %invar.inc12 = add nuw nsw i64 %fusion.indvar.dim.1, 1
  store i64 %invar.inc12, i64* %fusion.invar_address.dim.1
  br label %fusion.loop_header.dim.1

fusion.loop_exit.dim.1:                           ; preds = %fusion.loop_header.dim.1
  %invar.inc11 = add nuw nsw i64 %fusion.indvar.dim.0, 1
  store i64 %invar.inc11, i64* %fusion.invar_address.dim.0
  br label %fusion.loop_header.dim.0

fusion.loop_exit.dim.0:                           ; preds = %fusion.loop_header.dim.0
  %274 = getelementptr inbounds i8*, i8** %buffer_table, i64 1
  %275 = load i8*, i8** %274, !invariant.load !20, !dereferenceable !24, !align !0
  %reduce.322 = bitcast i8* %275 to [3 x [6 x double]]*
  store i64 0, i64* %reduce.322.invar_address.dim.0
  br label %reduce.322.loop_header.dim.0

reduce.322.loop_header.dim.0:                     ; preds = %fusion.vectorized_inner.loop_exit.reduction_dim.1, %fusion.loop_exit.dim.0
  %reduce.322.indvar.dim.0 = load i64, i64* %reduce.322.invar_address.dim.0
  %276 = icmp uge i64 %reduce.322.indvar.dim.0, 3
  br i1 %276, label %reduce.322.loop_exit.dim.0, label %reduce.322.loop_body.dim.0

reduce.322.loop_body.dim.0:                       ; preds = %reduce.322.loop_header.dim.0
  %277 = load double, double* bitcast ([8 x i8]* @0 to double*)
  %.splatinsert = insertelement <2 x double> undef, double %277, i32 0
  %.splat = shufflevector <2 x double> %.splatinsert, <2 x double> undef, <2 x i32> zeroinitializer
  store <2 x double> %.splat, <2 x double>* %accumulator, align 8
  %.splatinsert16 = insertelement <4 x double> undef, double %277, i32 0
  %.splat17 = shufflevector <4 x double> %.splatinsert16, <4 x double> undef, <4 x i32> zeroinitializer
  store <4 x double> %.splat17, <4 x double>* %accumulator15, align 8
  store i64 0, i64* %fusion.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.vectorized_inner.loop_header.reduction_dim.1

fusion.vectorized_inner.loop_header.reduction_dim.1: ; preds = %fusion.vectorized_inner.loop_body.reduction_dim.1, %reduce.322.loop_body.dim.0
  %fusion.vectorized_inner.indvar.reduction_dim.1 = load i64, i64* %fusion.vectorized_inner.invar_address.reduction_dim.1
  %278 = icmp uge i64 %fusion.vectorized_inner.indvar.reduction_dim.1, 6
  br i1 %278, label %fusion.vectorized_inner.loop_exit.reduction_dim.1, label %fusion.vectorized_inner.loop_body.reduction_dim.1

fusion.vectorized_inner.loop_body.reduction_dim.1: ; preds = %fusion.vectorized_inner.loop_header.reduction_dim.1
  %279 = getelementptr inbounds [3 x [6 x [6 x double]]], [3 x [6 x [6 x double]]]* %fusion, i64 0, i64 %reduce.322.indvar.dim.0, i64 %fusion.vectorized_inner.indvar.reduction_dim.1, i64 0
  %280 = bitcast double* %279 to i8*
  %281 = bitcast i8* %280 to <2 x double>*
  %282 = load <2 x double>, <2 x double>* %accumulator, align 8
  %283 = load <2 x double>, <2 x double>* %281, align 8, !alias.scope !48, !noalias !49
  %284 = fadd fast <2 x double> %282, %283
  store <2 x double> %284, <2 x double>* %accumulator, align 8
  %285 = getelementptr inbounds <2 x double>, <2 x double>* %281, i32 1
  %286 = bitcast <2 x double>* %285 to <4 x double>*
  %287 = load <4 x double>, <4 x double>* %accumulator15, align 8
  %288 = load <4 x double>, <4 x double>* %286, align 8, !alias.scope !48, !noalias !49
  %289 = fadd fast <4 x double> %287, %288
  store <4 x double> %289, <4 x double>* %accumulator15, align 8
  %invar.inc18 = add nuw nsw i64 %fusion.vectorized_inner.indvar.reduction_dim.1, 1
  store i64 %invar.inc18, i64* %fusion.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.vectorized_inner.loop_header.reduction_dim.1

fusion.vectorized_inner.loop_exit.reduction_dim.1: ; preds = %fusion.vectorized_inner.loop_header.reduction_dim.1
  %290 = load <2 x double>, <2 x double>* %accumulator, align 8
  %291 = load <4 x double>, <4 x double>* %accumulator15, align 8
  %292 = getelementptr inbounds [3 x [6 x double]], [3 x [6 x double]]* %reduce.322, i64 0, i64 %reduce.322.indvar.dim.0, i64 0
  %293 = bitcast double* %292 to <2 x double>*
  store <2 x double> %290, <2 x double>* %293, align 8, !alias.scope !52, !noalias !53
  %294 = getelementptr inbounds <2 x double>, <2 x double>* %293, i32 1
  %295 = bitcast <2 x double>* %294 to <4 x double>*
  store <4 x double> %291, <4 x double>* %295, align 8, !alias.scope !52, !noalias !53
  %invar.inc14 = add nuw nsw i64 %reduce.322.indvar.dim.0, 1
  store i64 %invar.inc14, i64* %reduce.322.invar_address.dim.0
  br label %reduce.322.loop_header.dim.0

reduce.322.loop_exit.dim.0:                       ; preds = %reduce.322.loop_header.dim.0
  %296 = getelementptr inbounds i8*, i8** %buffer_table, i64 22
  %297 = load i8*, i8** %296, !invariant.load !20, !dereferenceable !40, !align !36
  %fusion.2 = bitcast i8* %297 to [1 x [3 x [3 x double]]]*
  store i64 0, i64* %fusion.2.invar_address.dim.0
  br label %fusion.2.loop_header.dim.0

fusion.2.loop_header.dim.0:                       ; preds = %fusion.2.loop_exit.dim.1, %reduce.322.loop_exit.dim.0
  %fusion.2.indvar.dim.0 = load i64, i64* %fusion.2.invar_address.dim.0
  %298 = icmp uge i64 %fusion.2.indvar.dim.0, 1
  br i1 %298, label %fusion.2.loop_exit.dim.0, label %fusion.2.loop_body.dim.0

fusion.2.loop_body.dim.0:                         ; preds = %fusion.2.loop_header.dim.0
  store i64 0, i64* %fusion.2.invar_address.dim.1
  br label %fusion.2.loop_header.dim.1

fusion.2.loop_header.dim.1:                       ; preds = %fusion.2.loop_exit.dim.2, %fusion.2.loop_body.dim.0
  %fusion.2.indvar.dim.1 = load i64, i64* %fusion.2.invar_address.dim.1
  %299 = icmp uge i64 %fusion.2.indvar.dim.1, 3
  br i1 %299, label %fusion.2.loop_exit.dim.1, label %fusion.2.loop_body.dim.1

fusion.2.loop_body.dim.1:                         ; preds = %fusion.2.loop_header.dim.1
  store i64 0, i64* %fusion.2.invar_address.dim.2
  br label %fusion.2.loop_header.dim.2

fusion.2.loop_header.dim.2:                       ; preds = %fusion.2.loop_body.dim.2, %fusion.2.loop_body.dim.1
  %fusion.2.indvar.dim.2 = load i64, i64* %fusion.2.invar_address.dim.2
  %300 = icmp uge i64 %fusion.2.indvar.dim.2, 3
  br i1 %300, label %fusion.2.loop_exit.dim.2, label %fusion.2.loop_body.dim.2

fusion.2.loop_body.dim.2:                         ; preds = %fusion.2.loop_header.dim.2
  %301 = mul nuw nsw i64 %fusion.2.indvar.dim.2, 1
  %302 = add nuw nsw i64 0, %301
  %303 = udiv i64 %302, 3
  %304 = mul nuw nsw i64 %fusion.2.indvar.dim.1, 1
  %305 = add nuw nsw i64 0, %304
  %306 = udiv i64 %305, 3
  %307 = mul nuw nsw i64 %fusion.2.indvar.dim.0, 1
  %308 = add nuw nsw i64 0, %307
  %309 = mul nuw nsw i64 %305, 1
  %310 = add nuw nsw i64 0, %309
  %311 = trunc i64 %310 to i32
  %312 = mul nuw nsw i64 %302, 1
  %313 = add nuw nsw i64 0, %312
  %314 = trunc i64 %313 to i32
  %315 = icmp eq i32 %311, %314
  %316 = zext i1 %315 to i8
  %317 = getelementptr inbounds [3 x [3 x double]], [3 x [3 x double]]* %dot.1, i64 0, i64 %305, i64 %302
  %318 = load double, double* %317, !alias.scope !27, !noalias !31
  %319 = load double, double* bitcast ([8 x i8]* @23 to double*)
  %320 = trunc i8 %316 to i1
  %321 = select fast i1 %320, double %318, double %319
  %322 = getelementptr inbounds [1 x [3 x [3 x double]]], [1 x [3 x [3 x double]]]* %fusion.2, i64 0, i64 0, i64 %fusion.2.indvar.dim.1, i64 %fusion.2.indvar.dim.2
  store double %321, double* %322, !alias.scope !55, !noalias !56
  %invar.inc21 = add nuw nsw i64 %fusion.2.indvar.dim.2, 1
  store i64 %invar.inc21, i64* %fusion.2.invar_address.dim.2
  br label %fusion.2.loop_header.dim.2

fusion.2.loop_exit.dim.2:                         ; preds = %fusion.2.loop_header.dim.2
  %invar.inc20 = add nuw nsw i64 %fusion.2.indvar.dim.1, 1
  store i64 %invar.inc20, i64* %fusion.2.invar_address.dim.1
  br label %fusion.2.loop_header.dim.1

fusion.2.loop_exit.dim.1:                         ; preds = %fusion.2.loop_header.dim.1
  %invar.inc19 = add nuw nsw i64 %fusion.2.indvar.dim.0, 1
  store i64 %invar.inc19, i64* %fusion.2.invar_address.dim.0
  br label %fusion.2.loop_header.dim.0

fusion.2.loop_exit.dim.0:                         ; preds = %fusion.2.loop_header.dim.0
  %323 = getelementptr inbounds i8*, i8** %buffer_table, i64 22
  %324 = load i8*, i8** %323, !invariant.load !20, !dereferenceable !40, !align !36
  %325 = getelementptr inbounds i8, i8* %324, i64 80
  %reduce.185 = bitcast i8* %325 to [1 x [3 x double]]*
  store i64 0, i64* %reduce.185.invar_address.dim.0
  br label %reduce.185.loop_header.dim.0

reduce.185.loop_header.dim.0:                     ; preds = %fusion.2.vectorized_inner.loop_exit.reduction_dim.1, %fusion.2.loop_exit.dim.0
  %reduce.185.indvar.dim.0 = load i64, i64* %reduce.185.invar_address.dim.0
  %326 = icmp uge i64 %reduce.185.indvar.dim.0, 1
  br i1 %326, label %reduce.185.loop_exit.dim.0, label %reduce.185.loop_body.dim.0

reduce.185.loop_body.dim.0:                       ; preds = %reduce.185.loop_header.dim.0
  %327 = load double, double* bitcast ([8 x i8]* @0 to double*)
  store double %327, double* %accumulator23, align 8
  %.splatinsert25 = insertelement <2 x double> undef, double %327, i32 0
  %.splat26 = shufflevector <2 x double> %.splatinsert25, <2 x double> undef, <2 x i32> zeroinitializer
  store <2 x double> %.splat26, <2 x double>* %accumulator24, align 8
  store i64 0, i64* %fusion.2.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.2.vectorized_inner.loop_header.reduction_dim.1

fusion.2.vectorized_inner.loop_header.reduction_dim.1: ; preds = %fusion.2.vectorized_inner.loop_body.reduction_dim.1, %reduce.185.loop_body.dim.0
  %fusion.2.vectorized_inner.indvar.reduction_dim.1 = load i64, i64* %fusion.2.vectorized_inner.invar_address.reduction_dim.1
  %328 = icmp uge i64 %fusion.2.vectorized_inner.indvar.reduction_dim.1, 3
  br i1 %328, label %fusion.2.vectorized_inner.loop_exit.reduction_dim.1, label %fusion.2.vectorized_inner.loop_body.reduction_dim.1

fusion.2.vectorized_inner.loop_body.reduction_dim.1: ; preds = %fusion.2.vectorized_inner.loop_header.reduction_dim.1
  %329 = getelementptr inbounds [1 x [3 x [3 x double]]], [1 x [3 x [3 x double]]]* %fusion.2, i64 0, i64 0, i64 %fusion.2.vectorized_inner.indvar.reduction_dim.1, i64 0
  %330 = bitcast double* %329 to i8*
  %331 = bitcast i8* %330 to double*
  %332 = load double, double* %accumulator23, align 8
  %333 = load double, double* %331, align 8, !alias.scope !55, !noalias !56
  %334 = fadd fast double %332, %333
  store double %334, double* %accumulator23, align 8
  %335 = getelementptr inbounds double, double* %331, i32 1
  %336 = bitcast double* %335 to <2 x double>*
  %337 = load <2 x double>, <2 x double>* %accumulator24, align 8
  %338 = load <2 x double>, <2 x double>* %336, align 8, !alias.scope !55, !noalias !56
  %339 = fadd fast <2 x double> %337, %338
  store <2 x double> %339, <2 x double>* %accumulator24, align 8
  %invar.inc27 = add nuw nsw i64 %fusion.2.vectorized_inner.indvar.reduction_dim.1, 1
  store i64 %invar.inc27, i64* %fusion.2.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.2.vectorized_inner.loop_header.reduction_dim.1

fusion.2.vectorized_inner.loop_exit.reduction_dim.1: ; preds = %fusion.2.vectorized_inner.loop_header.reduction_dim.1
  %340 = load double, double* %accumulator23, align 8
  %341 = load <2 x double>, <2 x double>* %accumulator24, align 8
  %342 = getelementptr inbounds [1 x [3 x double]], [1 x [3 x double]]* %reduce.185, i64 0, i64 0, i64 0
  store double %340, double* %342, align 8, !alias.scope !57, !noalias !58
  %343 = getelementptr inbounds double, double* %342, i32 1
  %344 = bitcast double* %343 to <2 x double>*
  store <2 x double> %341, <2 x double>* %344, align 8, !alias.scope !57, !noalias !58
  %invar.inc22 = add nuw nsw i64 %reduce.185.indvar.dim.0, 1
  store i64 %invar.inc22, i64* %reduce.185.invar_address.dim.0
  br label %reduce.185.loop_header.dim.0

reduce.185.loop_exit.dim.0:                       ; preds = %reduce.185.loop_header.dim.0
  %345 = getelementptr inbounds i8*, i8** %buffer_table, i64 0
  %346 = load i8*, i8** %345, !invariant.load !20, !dereferenceable !24, !align !0
  %fusion.4 = bitcast i8* %346 to [1 x [2 x [2 x double]]]*
  store i64 0, i64* %fusion.4.invar_address.dim.0
  br label %fusion.4.loop_header.dim.0

fusion.4.loop_header.dim.0:                       ; preds = %fusion.4.loop_exit.dim.1, %reduce.185.loop_exit.dim.0
  %fusion.4.indvar.dim.0 = load i64, i64* %fusion.4.invar_address.dim.0
  %347 = icmp uge i64 %fusion.4.indvar.dim.0, 1
  br i1 %347, label %fusion.4.loop_exit.dim.0, label %fusion.4.loop_body.dim.0

fusion.4.loop_body.dim.0:                         ; preds = %fusion.4.loop_header.dim.0
  store i64 0, i64* %fusion.4.invar_address.dim.1
  br label %fusion.4.loop_header.dim.1

fusion.4.loop_header.dim.1:                       ; preds = %fusion.4.loop_exit.dim.2, %fusion.4.loop_body.dim.0
  %fusion.4.indvar.dim.1 = load i64, i64* %fusion.4.invar_address.dim.1
  %348 = icmp uge i64 %fusion.4.indvar.dim.1, 2
  br i1 %348, label %fusion.4.loop_exit.dim.1, label %fusion.4.loop_body.dim.1

fusion.4.loop_body.dim.1:                         ; preds = %fusion.4.loop_header.dim.1
  store i64 0, i64* %fusion.4.invar_address.dim.2
  br label %fusion.4.loop_header.dim.2

fusion.4.loop_header.dim.2:                       ; preds = %fusion.4.loop_body.dim.2, %fusion.4.loop_body.dim.1
  %fusion.4.indvar.dim.2 = load i64, i64* %fusion.4.invar_address.dim.2
  %349 = icmp uge i64 %fusion.4.indvar.dim.2, 2
  br i1 %349, label %fusion.4.loop_exit.dim.2, label %fusion.4.loop_body.dim.2

fusion.4.loop_body.dim.2:                         ; preds = %fusion.4.loop_header.dim.2
  %350 = mul nuw nsw i64 %fusion.4.indvar.dim.2, 1
  %351 = add nuw nsw i64 0, %350
  %352 = udiv i64 %351, 2
  %353 = mul nuw nsw i64 %fusion.4.indvar.dim.1, 1
  %354 = add nuw nsw i64 0, %353
  %355 = udiv i64 %354, 2
  %356 = mul nuw nsw i64 %fusion.4.indvar.dim.0, 1
  %357 = add nuw nsw i64 0, %356
  %358 = mul nuw nsw i64 %354, 1
  %359 = add nuw nsw i64 0, %358
  %360 = trunc i64 %359 to i32
  %361 = mul nuw nsw i64 %351, 1
  %362 = add nuw nsw i64 0, %361
  %363 = trunc i64 %362 to i32
  %364 = icmp eq i32 %360, %363
  %365 = zext i1 %364 to i8
  %366 = getelementptr inbounds [2 x [2 x double]], [2 x [2 x double]]* %dot, i64 0, i64 %354, i64 %351
  %367 = load double, double* %366, !alias.scope !39, !noalias !42
  %368 = load double, double* bitcast ([8 x i8]* @24 to double*)
  %369 = trunc i8 %365 to i1
  %370 = select fast i1 %369, double %367, double %368
  %371 = getelementptr inbounds [1 x [2 x [2 x double]]], [1 x [2 x [2 x double]]]* %fusion.4, i64 0, i64 0, i64 %fusion.4.indvar.dim.1, i64 %fusion.4.indvar.dim.2
  store double %370, double* %371, !alias.scope !59, !noalias !60
  %invar.inc30 = add nuw nsw i64 %fusion.4.indvar.dim.2, 1
  store i64 %invar.inc30, i64* %fusion.4.invar_address.dim.2
  br label %fusion.4.loop_header.dim.2

fusion.4.loop_exit.dim.2:                         ; preds = %fusion.4.loop_header.dim.2
  %invar.inc29 = add nuw nsw i64 %fusion.4.indvar.dim.1, 1
  store i64 %invar.inc29, i64* %fusion.4.invar_address.dim.1
  br label %fusion.4.loop_header.dim.1

fusion.4.loop_exit.dim.1:                         ; preds = %fusion.4.loop_header.dim.1
  %invar.inc28 = add nuw nsw i64 %fusion.4.indvar.dim.0, 1
  store i64 %invar.inc28, i64* %fusion.4.invar_address.dim.0
  br label %fusion.4.loop_header.dim.0

fusion.4.loop_exit.dim.0:                         ; preds = %fusion.4.loop_header.dim.0
  %372 = getelementptr inbounds i8*, i8** %buffer_table, i64 22
  %373 = load i8*, i8** %372, !invariant.load !20, !dereferenceable !40, !align !36
  %reduce.102 = bitcast i8* %373 to [1 x [2 x double]]*
  store i64 0, i64* %reduce.102.invar_address.dim.0
  br label %reduce.102.loop_header.dim.0

reduce.102.loop_header.dim.0:                     ; preds = %fusion.4.vectorized_inner.loop_exit.reduction_dim.1, %fusion.4.loop_exit.dim.0
  %reduce.102.indvar.dim.0 = load i64, i64* %reduce.102.invar_address.dim.0
  %374 = icmp uge i64 %reduce.102.indvar.dim.0, 1
  br i1 %374, label %reduce.102.loop_exit.dim.0, label %reduce.102.loop_body.dim.0

reduce.102.loop_body.dim.0:                       ; preds = %reduce.102.loop_header.dim.0
  %375 = load double, double* bitcast ([8 x i8]* @0 to double*)
  %.splatinsert33 = insertelement <2 x double> undef, double %375, i32 0
  %.splat34 = shufflevector <2 x double> %.splatinsert33, <2 x double> undef, <2 x i32> zeroinitializer
  store <2 x double> %.splat34, <2 x double>* %accumulator32, align 8
  store i64 0, i64* %fusion.4.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.4.vectorized_inner.loop_header.reduction_dim.1

fusion.4.vectorized_inner.loop_header.reduction_dim.1: ; preds = %fusion.4.vectorized_inner.loop_body.reduction_dim.1, %reduce.102.loop_body.dim.0
  %fusion.4.vectorized_inner.indvar.reduction_dim.1 = load i64, i64* %fusion.4.vectorized_inner.invar_address.reduction_dim.1
  %376 = icmp uge i64 %fusion.4.vectorized_inner.indvar.reduction_dim.1, 2
  br i1 %376, label %fusion.4.vectorized_inner.loop_exit.reduction_dim.1, label %fusion.4.vectorized_inner.loop_body.reduction_dim.1

fusion.4.vectorized_inner.loop_body.reduction_dim.1: ; preds = %fusion.4.vectorized_inner.loop_header.reduction_dim.1
  %377 = getelementptr inbounds [1 x [2 x [2 x double]]], [1 x [2 x [2 x double]]]* %fusion.4, i64 0, i64 0, i64 %fusion.4.vectorized_inner.indvar.reduction_dim.1, i64 0
  %378 = bitcast double* %377 to i8*
  %379 = bitcast i8* %378 to <2 x double>*
  %380 = load <2 x double>, <2 x double>* %accumulator32, align 8
  %381 = load <2 x double>, <2 x double>* %379, align 8, !alias.scope !59, !noalias !60
  %382 = fadd fast <2 x double> %380, %381
  store <2 x double> %382, <2 x double>* %accumulator32, align 8
  %invar.inc35 = add nuw nsw i64 %fusion.4.vectorized_inner.indvar.reduction_dim.1, 1
  store i64 %invar.inc35, i64* %fusion.4.vectorized_inner.invar_address.reduction_dim.1
  br label %fusion.4.vectorized_inner.loop_header.reduction_dim.1

fusion.4.vectorized_inner.loop_exit.reduction_dim.1: ; preds = %fusion.4.vectorized_inner.loop_header.reduction_dim.1
  %383 = load <2 x double>, <2 x double>* %accumulator32, align 8
  %384 = getelementptr inbounds [1 x [2 x double]], [1 x [2 x double]]* %reduce.102, i64 0, i64 0, i64 0
  %385 = bitcast double* %384 to <2 x double>*
  store <2 x double> %383, <2 x double>* %385, align 8, !alias.scope !61, !noalias !62
  %invar.inc31 = add nuw nsw i64 %reduce.102.indvar.dim.0, 1
  store i64 %invar.inc31, i64* %reduce.102.invar_address.dim.0
  br label %reduce.102.loop_header.dim.0

reduce.102.loop_exit.dim.0:                       ; preds = %reduce.102.loop_header.dim.0
  %386 = getelementptr inbounds i8*, i8** %buffer_table, i64 0
  %387 = load i8*, i8** %386, !invariant.load !20, !dereferenceable !24, !align !0
  %fusion.1 = bitcast i8* %387 to [3 x [6 x double]]*
  store i64 0, i64* %fusion.1.invar_address.dim.0
  br label %fusion.1.loop_header.dim.0

fusion.1.loop_header.dim.0:                       ; preds = %fusion.1.loop_exit.dim.1, %reduce.102.loop_exit.dim.0
  %fusion.1.indvar.dim.0 = load i64, i64* %fusion.1.invar_address.dim.0
  %388 = icmp uge i64 %fusion.1.indvar.dim.0, 3
  br i1 %388, label %fusion.1.loop_exit.dim.0, label %fusion.1.loop_body.dim.0

fusion.1.loop_body.dim.0:                         ; preds = %fusion.1.loop_header.dim.0
  store i64 0, i64* %fusion.1.invar_address.dim.1
  br label %fusion.1.loop_header.dim.1

fusion.1.loop_header.dim.1:                       ; preds = %fusion.1.loop_body.dim.1, %fusion.1.loop_body.dim.0
  %fusion.1.indvar.dim.1 = load i64, i64* %fusion.1.invar_address.dim.1
  %389 = icmp uge i64 %fusion.1.indvar.dim.1, 6
  br i1 %389, label %fusion.1.loop_exit.dim.1, label %fusion.1.loop_body.dim.1

fusion.1.loop_body.dim.1:                         ; preds = %fusion.1.loop_header.dim.1
  %390 = mul nuw nsw i64 %fusion.1.indvar.dim.1, 1
  %391 = add nuw nsw i64 0, %390
  %392 = urem i64 %391, 3
  %393 = udiv i64 %391, 3
  %394 = udiv i64 %393, 2
  %395 = mul nuw nsw i64 %fusion.1.indvar.dim.0, 1
  %396 = add nuw nsw i64 0, %395
  %397 = udiv i64 %396, 3
  %398 = mul nuw nsw i64 %393, 1
  %399 = add nuw nsw i64 0, %398
  %400 = udiv i64 %399, 2
  %401 = mul nuw nsw i64 %396, 1
  %402 = add nuw nsw i64 0, %401
  %403 = udiv i64 %402, 3
  %404 = mul nuw nsw i64 %402, 1
  %405 = add nuw nsw i64 0, %404
  %406 = udiv i64 %405, 3
  %407 = load float, float* bitcast ([4 x i8]* @25 to float*)
  %408 = getelementptr inbounds [3 x [1 x float]], [3 x [1 x float]]* %arg2.3, i64 0, i64 %405, i64 0
  %409 = load float, float* %408, !invariant.load !20, !noalias !44
  %410 = load float, float* bitcast ([4 x i8]* @26 to float*)
  %411 = fmul fast float %409, %410
  %412 = fadd fast float %407, %411
  %413 = fcmp fast une float %412, %412
  %414 = zext i1 %413 to i8
  %415 = load float, float* bitcast ([4 x i8]* @27 to float*)
  %416 = fcmp fast one float %412, 0.000000e+00
  %417 = uitofp i1 %416 to float
  %418 = call fast float @llvm.copysign.f32(float %417, float %412)
  %419 = fcmp fast uno float %412, %412
  %420 = select fast i1 %419, float %412, float %418
  %421 = trunc i8 %414 to i1
  %422 = select fast i1 %421, float %415, float %420
  %423 = fpext float %422 to double
  %424 = load double, double* bitcast ([8 x i8]* @28 to double*)
  %425 = getelementptr inbounds [3 x [1 x double]], [3 x [1 x double]]* %arg4.5, i64 0, i64 %405, i64 0
  %426 = load double, double* %425, !invariant.load !20, !noalias !44
  %427 = load double, double* bitcast ([8 x i8]* @29 to double*)
  %428 = fmul fast double %426, %427
  %429 = fadd fast double %424, %428
  %430 = fmul fast double %423, %429
  %431 = fmul fast double %430, %430
  %432 = mul nuw nsw i64 %399, 1
  %433 = add nuw nsw i64 0, %432
  %434 = udiv i64 %433, 2
  %435 = getelementptr inbounds [1 x [2 x double]], [1 x [2 x double]]* %reduce.102, i64 0, i64 0, i64 %433
  %436 = load double, double* %435, !alias.scope !61, !noalias !62
  %437 = fmul fast double %431, %436
  %438 = mul nuw nsw i64 %392, 1
  %439 = add nuw nsw i64 0, %438
  %440 = udiv i64 %439, 3
  %441 = getelementptr inbounds [1 x [3 x double]], [1 x [3 x double]]* %reduce.185, i64 0, i64 0, i64 %439
  %442 = load double, double* %441, !alias.scope !57, !noalias !58
  %443 = fmul fast double %437, %442
  %444 = getelementptr inbounds [3 x [6 x double]], [3 x [6 x double]]* %fusion.1, i64 0, i64 %fusion.1.indvar.dim.0, i64 %fusion.1.indvar.dim.1
  store double %443, double* %444, !alias.scope !63, !noalias !64
  %invar.inc37 = add nuw nsw i64 %fusion.1.indvar.dim.1, 1
  store i64 %invar.inc37, i64* %fusion.1.invar_address.dim.1
  br label %fusion.1.loop_header.dim.1

fusion.1.loop_exit.dim.1:                         ; preds = %fusion.1.loop_header.dim.1
  %invar.inc36 = add nuw nsw i64 %fusion.1.indvar.dim.0, 1
  store i64 %invar.inc36, i64* %fusion.1.invar_address.dim.0
  br label %fusion.1.loop_header.dim.0

fusion.1.loop_exit.dim.0:                         ; preds = %fusion.1.loop_header.dim.0
  %445 = getelementptr inbounds i8*, i8** %buffer_table, i64 7
  %446 = load i8*, i8** %445, !invariant.load !20, !dereferenceable !36, !align !0
  %tuple.325 = bitcast i8* %446 to [2 x i8*]*
  %447 = bitcast [3 x [6 x double]]* %fusion.1 to i8*
  %448 = getelementptr inbounds [2 x i8*], [2 x i8*]* %tuple.325, i64 0, i64 0
  store i8* %447, i8** %448, !alias.scope !65, !noalias !66
  %449 = bitcast [3 x [6 x double]]* %reduce.322 to i8*
  %450 = getelementptr inbounds [2 x i8*], [2 x i8*]* %tuple.325, i64 0, i64 1
  store i8* %449, i8** %450, !alias.scope !65, !noalias !66
  ret void
}

; Function Attrs: nounwind readnone speculatable willreturn
declare float @llvm.copysign.f32(float, float) #1

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

!0 = !{i64 8}
!1 = !{!2}
!2 = !{!"buffer: {index:20, offset:0, size:8}", !3}
!3 = !{!"XLA global AA domain"}
!4 = !{!5}
!5 = !{!"buffer: {index:13, offset:0, size:8}", !3}
!6 = !{!7}
!7 = !{!"buffer: {index:21, offset:0, size:8}", !3}
!8 = !{!9}
!9 = !{!"buffer: {index:17, offset:0, size:8}", !3}
!10 = !{!11}
!11 = !{!"buffer: {index:19, offset:0, size:8}", !3}
!12 = !{!13}
!13 = !{!"buffer: {index:18, offset:0, size:8}", !3}
!14 = !{!15}
!15 = !{!"buffer: {index:14, offset:0, size:8}", !3}
!16 = !{!17}
!17 = !{!"buffer: {index:16, offset:0, size:8}", !3}
!18 = !{!19}
!19 = !{!"buffer: {index:15, offset:0, size:8}", !3}
!20 = !{}
!21 = !{i64 24}
!22 = !{i64 12}
!23 = !{i64 72}
!24 = !{i64 144}
!25 = !{!26}
!26 = !{!"buffer: {index:1, offset:0, size:72}", !3}
!27 = !{!28}
!28 = !{!"buffer: {index:0, offset:0, size:72}", !3}
!29 = distinct !{!29, !30}
!30 = !{!"llvm.loop.unroll.disable"}
!31 = !{!26, !32, !33, !34}
!32 = !{!"buffer: {index:22, offset:0, size:72}", !3}
!33 = !{!"buffer: {index:22, offset:0, size:864}", !3}
!34 = !{!"buffer: {index:22, offset:864, size:32}", !3}
!35 = !{i64 32}
!36 = !{i64 16}
!37 = !{!38}
!38 = !{!"buffer: {index:1, offset:0, size:32}", !3}
!39 = !{!34}
!40 = !{i64 896}
!41 = distinct !{!41, !30}
!42 = !{!43, !28, !38, !33}
!43 = !{!"buffer: {index:0, offset:0, size:32}", !3}
!44 = !{!28, !45, !46, !33, !47, !34}
!45 = !{!"buffer: {index:0, offset:0, size:144}", !3}
!46 = !{!"buffer: {index:22, offset:0, size:16}", !3}
!47 = !{!"buffer: {index:22, offset:80, size:24}", !3}
!48 = !{!33}
!49 = !{!28, !50, !51, !34}
!50 = !{!"buffer: {index:1, offset:0, size:144}", !3}
!51 = !{!"buffer: {index:11, offset:0, size:8}", !3}
!52 = !{!50}
!53 = !{!45, !54, !51, !33}
!54 = !{!"buffer: {index:7, offset:0, size:16}", !3}
!55 = !{!32}
!56 = !{!28, !51, !47}
!57 = !{!47}
!58 = !{!45, !51, !46, !32}
!59 = !{!43}
!60 = !{!51, !46, !34}
!61 = !{!46}
!62 = !{!43, !45, !51, !47}
!63 = !{!45}
!64 = !{!50, !54, !46, !47}
!65 = !{!54}
!66 = !{!45, !50}