; ModuleID = 'kernel_gemm'
source_filename = "kernel_gemm"
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"

@kernel_0 = private unnamed_addr constant [6239 x i8] c"//\0A// Generated by LLVM NVPTX Back-End\0A//\0A\0A.version 3.2\0A.target sm_30\0A.address_size 64\0A\0A\09// .globl\09kernel_0\0A\0A.visible .entry kernel_0(\0A\09.param .u64 kernel_0_param_0,\0A\09.param .u64 kernel_0_param_1,\0A\09.param .u64 kernel_0_param_2,\0A\09.param .u64 kernel_0_param_3,\0A\09.param .u64 kernel_0_param_4,\0A\09.param .u64 kernel_0_param_5,\0A\09.param .u64 kernel_0_param_6,\0A\09.param .u64 kernel_0_param_7,\0A\09.param .u64 kernel_0_param_8,\0A\09.param .u64 kernel_0_param_9\0A)\0A.maxntid 32, 16, 1\0A{\0A\09.reg .pred \09%p<19>;\0A\09.reg .b32 \09%r<5>;\0A\09.reg .b64 \09%rd<167>;\0A\0A\09ld.param.u64 \09%rd79, [kernel_0_param_9];\0A\09ld.param.u64 \09%rd78, [kernel_0_param_8];\0A\09ld.param.u64 \09%rd77, [kernel_0_param_6];\0A\09ld.param.u64 \09%rd76, [kernel_0_param_5];\0A\09ld.param.u64 \09%rd75, [kernel_0_param_4];\0A\09ld.param.u64 \09%rd74, [kernel_0_param_0];\0A\09ld.param.u64 \09%rd81, [kernel_0_param_1];\0A\09mov.u32 \09%r1, %ctaid.x;\0A\09ld.param.u64 \09%rd82, [kernel_0_param_2];\0A\09ld.param.u64 \09%rd83, [kernel_0_param_3];\0A\09mov.u32 \09%r2, %ctaid.y;\0A\09mov.u32 \09%r3, %tid.x;\0A\09cvt.u64.u32 \09%rd84, %r3;\0A\09mov.u32 \09%r4, %tid.y;\0A\09cvt.u64.u32 \09%rd85, %r4;\0A\09mul.wide.u32 \09%rd86, %r1, 32;\0A\09add.s64 \09%rd87, %rd76, -1;\0A\09sub.s64 \09%rd88, %rd87, %rd86;\0A\09shr.u64 \09%rd1, %rd88, 13;\0A\09mul.wide.u32 \09%rd89, %r2, 32;\0A\09add.s64 \09%rd90, %rd83, -1;\0A\09sub.s64 \09%rd91, %rd90, %rd89;\0A\09shr.u64 \09%rd2, %rd91, 13;\0A\09add.s64 \09%rd92, %rd75, -1;\0A\09shr.s64 \09%rd3, %rd92, 5;\0A\09add.s64 \09%rd4, %rd86, %rd84;\0A\09mul.wide.u32 \09%rd93, %r2, 2;\0A\09neg.s64 \09%rd5, %rd93;\0A\09sub.s64 \09%rd94, %rd90, %rd85;\0A\09shr.s64 \09%rd6, %rd94, 4;\0A\09add.s64 \09%rd7, %rd89, %rd85;\0A\09mul.lo.s64 \09%rd95, %rd75, %rd7;\0A\09shl.b64 \09%rd96, %rd95, 3;\0A\09add.s64 \09%rd8, %rd82, %rd96;\0A\09shl.b64 \09%rd9, %rd75, 16;\0A\09shl.b64 \09%rd10, %rd75, 7;\0A\09mul.wide.u32 \09%rd97, %r1, 256;\0A\09mul.wide.u32 \09%rd98, %r3, 8;\0A\09add.s64 \09%rd99, %rd97, %rd98;\0A\09add.s64 \09%rd145, %rd81, %rd99;\0A\09shl.b64 \09%rd12, %rd76, 8;\0A\09shl.b64 \09%rd13, %rd76, 3;\0A\09mul.lo.s64 \09%rd100, %rd77, %rd7;\0A\09add.s64 \09%rd101, %rd100, %rd86;\0A\09add.s64 \09%rd102, %rd101, %rd84;\0A\09shl.b64 \09%rd103, %rd102, 3;\0A\09add.s64 \09%rd144, %rd74, %rd103;\0A\09shl.b64 \09%rd15, %rd77, 16;\0A\09shl.b64 \09%rd16, %rd77, 7;\0A\09mov.u64 \09%rd80, 0;\0A\09setp.lt.s64 \09%p2, %rd3, 0;\0A\09setp.lt.s64 \09%p13, %rd75, 1;\0A\09mov.u64 \09%rd146, %rd80;\0ALBB0_1:\0A\09shl.b64 \09%rd105, %rd146, 13;\0A\09add.s64 \09%rd20, %rd4, %rd105;\0A\09mov.u64 \09%rd147, %rd144;\0A\09mov.u64 \09%rd148, %rd8;\0A\09mov.u64 \09%rd149, %rd80;\0ALBB0_2:\0A\09@%p2 bra \09LBB0_7;\0A\09setp.ge.s64 \09%p3, %rd20, %rd76;\0A\09shl.b64 \09%rd107, %rd149, 9;\0A\09sub.s64 \09%rd108, %rd5, %rd107;\0A\09add.s64 \09%rd109, %rd108, %rd6;\0A\09min.s64 \09%rd110, %rd109, 1;\0A\09setp.lt.s64 \09%p4, %rd110, 0;\0A\09shl.b64 \09%rd111, %rd149, 13;\0A\09add.s64 \09%rd27, %rd7, %rd111;\0A\09add.s64 \09%rd28, %rd110, -1;\0A\09or.pred  \09%p1, %p3, %p4;\0A\09mov.u64 \09%rd152, 0;\0A\09mov.u64 \09%rd150, %rd145;\0A\09mov.u64 \09%rd151, %rd148;\0ALBB0_4:\0A\09mov.u64 \09%rd34, %rd152;\0A\09@!%p1 bra \09LBB0_9;\0A\09bra.uni \09LBB0_20;\0ALBB0_9:\0A\09setp.ne.s64 \09%p5, %rd34, 0;\0A\09shl.b64 \09%rd112, %rd34, 5;\0A\09sub.s64 \09%rd113, %rd75, %rd112;\0A\09add.s64 \09%rd114, %rd113, -1;\0A\09min.s64 \09%rd38, %rd114, 31;\0A\09add.s64 \09%rd39, %rd38, -1;\0A\09@%p5 bra \09LBB0_15;\0A\09bra.uni \09LBB0_10;\0ALBB0_15:\0A\09setp.lt.s64 \09%p6, %rd38, 0;\0A\09@%p6 bra \09LBB0_20;\0A\09mov.u64 \09%rd160, 0;\0A\09mov.u64 \09%rd159, %rd151;\0ALBB0_17:\0A\09shl.b64 \09%rd117, %rd160, 4;\0A\09add.s64 \09%rd118, %rd27, %rd117;\0A\09mul.lo.s64 \09%rd119, %rd118, %rd77;\0A\09add.s64 \09%rd120, %rd119, %rd20;\0A\09shl.b64 \09%rd121, %rd120, 3;\0A\09add.s64 \09%rd56, %rd74, %rd121;\0A\09ld.global.u64 \09%rd164, [%rd56];\0A\09mov.u64 \09%rd163, -1;\0A\09mov.u64 \09%rd161, %rd150;\0A\09mov.u64 \09%rd162, %rd159;\0ALBB0_18:\0A\09ld.global.u64 \09%rd122, [%rd161];\0A\09mul.lo.s64 \09%rd123, %rd122, %rd79;\0A\09ld.global.u64 \09%rd124, [%rd162];\0A\09mul.lo.s64 \09%rd125, %rd123, %rd124;\0A\09add.s64 \09%rd164, %rd125, %rd164;\0A\09st.global.u64 \09[%rd56], %rd164;\0A\09add.s64 \09%rd163, %rd163, 1;\0A\09add.s64 \09%rd162, %rd162, 8;\0A\09add.s64 \09%rd161, %rd161, %rd13;\0A\09setp.le.s64 \09%p7, %rd163, %rd39;\0A\09@%p7 bra \09LBB0_18;\0A\09add.s64 \09%rd66, %rd160, 1;\0A\09add.s64 \09%rd159, %rd159, %rd10;\0A\09setp.gt.s64 \09%p8, %rd160, %rd28;\0A\09mov.u64 \09%rd160, %rd66;\0A\09@%p8 bra \09LBB0_20;\0A\09bra.uni \09LBB0_17;\0ALBB0_10:\0A\09mov.u64 \09%rd154, 0;\0A\09mov.u64 \09%rd153, %rd151;\0ALBB0_11:\0A\09setp.gt.s64 \09%p9, %rd38, -1;\0A\09shl.b64 \09%rd127, %rd154, 4;\0A\09add.s64 \09%rd128, %rd27, %rd127;\0A\09mul.lo.s64 \09%rd129, %rd128, %rd77;\0A\09add.s64 \09%rd130, %rd129, %rd20;\0A\09shl.b64 \09%rd131, %rd130, 3;\0A\09add.s64 \09%rd42, %rd74, %rd131;\0A\09ld.global.u64 \09%rd132, [%rd42];\0A\09mul.lo.s64 \09%rd158, %rd132, %rd78;\0A\09st.global.u64 \09[%rd42], %rd158;\0A\09@%p9 bra \09LBB0_12;\0A\09bra.uni \09LBB0_14;\0ALBB0_12:\0A\09mov.u64 \09%rd157, -1;\0A\09mov.u64 \09%rd155, %rd150;\0A\09mov.u64 \09%rd156, %rd153;\0ALBB0_13:\0A\09ld.global.u64 \09%rd134, [%rd155];\0A\09mul.lo.s64 \09%rd135, %rd134, %rd79;\0A\09ld.global.u64 \09%rd136, [%rd156];\0A\09mul.lo.s64 \09%rd137, %rd135, %rd136;\0A\09add.s64 \09%rd158, %rd137, %rd158;\0A\09st.global.u64 \09[%rd42], %rd158;\0A\09add.s64 \09%rd157, %rd157, 1;\0A\09add.s64 \09%rd156, %rd156, 8;\0A\09add.s64 \09%rd155, %rd155, %rd13;\0A\09setp.le.s64 \09%p10, %rd157, %rd39;\0A\09@%p10 bra \09LBB0_13;\0ALBB0_14:\0A\09add.s64 \09%rd52, %rd154, 1;\0A\09add.s64 \09%rd153, %rd153, %rd10;\0A\09setp.gt.s64 \09%p11, %rd154, %rd28;\0A\09mov.u64 \09%rd154, %rd52;\0A\09@%p11 bra \09LBB0_20;\0A\09bra.uni \09LBB0_11;\0ALBB0_20:\0A\09bar.sync \090;\0A\09add.s64 \09%rd152, %rd34, 1;\0A\09add.s64 \09%rd151, %rd151, 256;\0A\09add.s64 \09%rd150, %rd150, %rd12;\0A\09setp.eq.s64 \09%p12, %rd34, %rd3;\0A\09@%p12 bra \09LBB0_7;\0A\09bra.uni \09LBB0_4;\0ALBB0_7:\0A\09@%p13 bra \09LBB0_21;\0A\09bra.uni \09LBB0_8;\0ALBB0_21:\0A\09setp.lt.s64 \09%p14, %rd20, %rd76;\0A\09@%p14 bra \09LBB0_23;\0A\09bra.uni \09LBB0_22;\0ALBB0_23:\0A\09shl.b64 \09%rd138, %rd149, 9;\0A\09sub.s64 \09%rd139, %rd5, %rd138;\0A\09add.s64 \09%rd140, %rd139, %rd6;\0A\09min.s64 \09%rd68, %rd140, 1;\0A\09setp.lt.s64 \09%p15, %rd68, 0;\0A\09@%p15 bra \09LBB0_22;\0A\09add.s64 \09%rd69, %rd68, -1;\0A\09mov.u64 \09%rd165, -1;\0A\09mov.u64 \09%rd166, %rd147;\0ALBB0_25:\0A\09ld.global.u64 \09%rd142, [%rd166];\0A\09mul.lo.s64 \09%rd143, %rd142, %rd78;\0A\09st.global.u64 \09[%rd166], %rd143;\0A\09add.s64 \09%rd166, %rd166, %rd16;\0A\09add.s64 \09%rd165, %rd165, 1;\0A\09setp.gt.s64 \09%p16, %rd165, %rd69;\0A\09@%p16 bra \09LBB0_22;\0A\09bra.uni \09LBB0_25;\0ALBB0_22:\0A\09bar.sync \090;\0ALBB0_8:\0A\09add.s64 \09%rd29, %rd149, 1;\0A\09add.s64 \09%rd148, %rd148, %rd9;\0A\09add.s64 \09%rd147, %rd147, %rd15;\0A\09setp.eq.s64 \09%p17, %rd149, %rd2;\0A\09mov.u64 \09%rd149, %rd29;\0A\09@%p17 bra \09LBB0_5;\0A\09bra.uni \09LBB0_2;\0ALBB0_5:\0A\09add.s64 \09%rd21, %rd146, 1;\0A\09add.s64 \09%rd145, %rd145, 65536;\0A\09add.s64 \09%rd144, %rd144, 65536;\0A\09setp.eq.s64 \09%p18, %rd146, %rd1;\0A\09mov.u64 \09%rd146, %rd21;\0A\09@%p18 bra \09LBB0_6;\0A\09bra.uni \09LBB0_1;\0ALBB0_6:\0A\09ret;\0A}\0A\0A\0A\00"
@kernel_0_name = private unnamed_addr constant [9 x i8] c"kernel_0\00"

; Function Attrs: sspstrong
define void @julia_kernel_gemm_64400(i64, i64, i8** dereferenceable(40), i8** dereferenceable(40), i8** dereferenceable(40)) #0 !dbg !5 {
top:
  %polly_launch_0_params = alloca [10 x i8*]
  %polly_launch_0_param_0 = alloca i8*
  %polly_launch_0_param_1 = alloca i8*
  %polly_launch_0_param_2 = alloca i8*
  %polly_launch_0_param_3 = alloca i64
  %polly_launch_0_param_4 = alloca i64
  %polly_launch_0_param_5 = alloca i64
  %polly_launch_0_param_6 = alloca i64
  %polly_launch_0_param_7 = alloca void (metadata, i64, metadata, metadata)*
  %polly_launch_0_param_8 = alloca i64
  %polly_launch_0_param_9 = alloca i64
  %polly_launch_0_params_i8ptr = bitcast [10 x i8*]* %polly_launch_0_params to i8*
  br label %top.split, !dbg !27

top.split:                                        ; preds = %top
  call void @llvm.dbg.value(metadata i8** null, i64 0, metadata !18, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i8** null, i64 0, metadata !19, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i8** null, i64 0, metadata !20, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %0, i64 0, metadata !16, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %1, i64 0, metadata !17, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i8** %2, i64 0, metadata !18, metadata !28), !dbg !27
  %5 = bitcast i8** %2 to i64**
  %6 = load i64*, i64** %5, align 8, !tbaa !29
  %7 = getelementptr i8*, i8** %2, i64 3
  %8 = bitcast i8** %7 to i64*
  %9 = load i64, i64* %8, align 8, !tbaa !29
  call void @llvm.dbg.value(metadata i8** %3, i64 0, metadata !19, metadata !28), !dbg !27
  %10 = bitcast i8** %3 to i64**
  %11 = load i64*, i64** %10, align 8, !tbaa !29
  %12 = getelementptr i8*, i8** %3, i64 3
  %13 = bitcast i8** %12 to i64*
  %14 = load i64, i64* %13, align 8, !tbaa !29
  call void @llvm.dbg.value(metadata i8** %4, i64 0, metadata !20, metadata !28), !dbg !27
  %15 = bitcast i8** %4 to i64**
  %16 = load i64*, i64** %15, align 8, !tbaa !29
  %17 = getelementptr i8*, i8** %4, i64 3
  %18 = bitcast i8** %17 to i64*
  %19 = load i64, i64* %18, align 8, !tbaa !29
  %20 = getelementptr i8*, i8** %4, i64 4, !dbg !32
  %21 = bitcast i8** %20 to i64*, !dbg !32
  %22 = load i64, i64* %21, align 8, !dbg !32, !tbaa !29
  call void @llvm.dbg.value(metadata i64 %14, i64 0, metadata !24, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %14, i64 0, metadata !24, metadata !28), !dbg !27
  %23 = icmp sgt i64 %14, 0, !dbg !33
  %24 = select i1 %23, i64 %14, i64 0, !dbg !33
  %25 = icmp slt i64 %14, 1, !dbg !33
  br i1 %25, label %L65, label %if.lr.ph, !dbg !33

if.lr.ph:                                         ; preds = %top.split
  br label %polly.split_new_and_old, !dbg !33

L42.L18.loopexit_crit_edge:                       ; preds = %L42.loopexit
  br label %L18.loopexit, !dbg !34

L18.loopexit:                                     ; preds = %L42.L18.loopexit_crit_edge, %L38
  %26 = icmp eq i64 %"#temp#4.024", %24, !dbg !33
  br i1 %26, label %L18.L65_crit_edge, label %if, !dbg !33

polly.split_new_and_old:                          ; preds = %if.lr.ph
  %27 = icmp sge i64 %9, %14
  %28 = icmp sle i64 %14, 0
  %29 = icmp eq i64 %22, 9223372036854775807
  %30 = or i1 %28, %29
  %31 = sext i1 %30 to i64
  %32 = icmp eq i64 0, %31
  %33 = and i1 %27, %32
  %34 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 %19, i64 1)
  %.obit = extractvalue { i64, i1 } %34, 1
  %polly.overflow.state = or i1 false, %.obit
  %.res = extractvalue { i64, i1 } %34, 0
  %polly.access.mul. = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res, i64 %14)
  %polly.access.mul..obit = extractvalue { i64, i1 } %polly.access.mul., 1
  %polly.overflow.state25 = or i1 %polly.overflow.state, %polly.access.mul..obit
  %polly.access.mul..res = extractvalue { i64, i1 } %polly.access.mul., 0
  %polly.access.add. = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res, i64 %14)
  %polly.access.add..obit = extractvalue { i64, i1 } %polly.access.add., 1
  %polly.overflow.state26 = or i1 %polly.overflow.state25, %polly.access.add..obit
  %polly.access.add..res = extractvalue { i64, i1 } %polly.access.add., 0
  %polly.access. = getelementptr i64, i64* %11, i64 %polly.access.add..res
  %polly.access.mul.27 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 0, i64 %9)
  %polly.access.mul..obit28 = extractvalue { i64, i1 } %polly.access.mul.27, 1
  %polly.overflow.state29 = or i1 %polly.overflow.state26, %polly.access.mul..obit28
  %polly.access.mul..res30 = extractvalue { i64, i1 } %polly.access.mul.27, 0
  %polly.access.add.31 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res30, i64 0)
  %polly.access.add..obit32 = extractvalue { i64, i1 } %polly.access.add.31, 1
  %polly.overflow.state33 = or i1 %polly.overflow.state29, %polly.access.add..obit32
  %polly.access.add..res34 = extractvalue { i64, i1 } %polly.access.add.31, 0
  %polly.access.35 = getelementptr i64, i64* %6, i64 %polly.access.add..res34
  %35 = ptrtoint i64* %polly.access. to i64
  %36 = ptrtoint i64* %polly.access.35 to i64
  %37 = icmp ule i64 %35, %36
  %38 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 %22, i64 1)
  %.obit36 = extractvalue { i64, i1 } %38, 1
  %polly.overflow.state37 = or i1 %polly.overflow.state33, %.obit36
  %.res38 = extractvalue { i64, i1 } %38, 0
  %polly.access.mul.39 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res38, i64 %9)
  %polly.access.mul..obit40 = extractvalue { i64, i1 } %polly.access.mul.39, 1
  %polly.overflow.state41 = or i1 %polly.overflow.state37, %polly.access.mul..obit40
  %polly.access.mul..res42 = extractvalue { i64, i1 } %polly.access.mul.39, 0
  %polly.access.add.43 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res42, i64 %14)
  %polly.access.add..obit44 = extractvalue { i64, i1 } %polly.access.add.43, 1
  %polly.overflow.state45 = or i1 %polly.overflow.state41, %polly.access.add..obit44
  %polly.access.add..res46 = extractvalue { i64, i1 } %polly.access.add.43, 0
  %polly.access.47 = getelementptr i64, i64* %6, i64 %polly.access.add..res46
  %polly.access.mul.48 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 0, i64 %14)
  %polly.access.mul..obit49 = extractvalue { i64, i1 } %polly.access.mul.48, 1
  %polly.overflow.state50 = or i1 %polly.overflow.state45, %polly.access.mul..obit49
  %polly.access.mul..res51 = extractvalue { i64, i1 } %polly.access.mul.48, 0
  %polly.access.add.52 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res51, i64 0)
  %polly.access.add..obit53 = extractvalue { i64, i1 } %polly.access.add.52, 1
  %polly.overflow.state54 = or i1 %polly.overflow.state50, %polly.access.add..obit53
  %polly.access.add..res55 = extractvalue { i64, i1 } %polly.access.add.52, 0
  %polly.access.56 = getelementptr i64, i64* %11, i64 %polly.access.add..res55
  %39 = ptrtoint i64* %polly.access.47 to i64
  %40 = ptrtoint i64* %polly.access.56 to i64
  %41 = icmp ule i64 %39, %40
  %42 = or i1 %37, %41
  %43 = and i1 %33, %42
  %44 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 %22, i64 1)
  %.obit57 = extractvalue { i64, i1 } %44, 1
  %polly.overflow.state58 = or i1 %polly.overflow.state54, %.obit57
  %.res59 = extractvalue { i64, i1 } %44, 0
  %polly.access.mul.60 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res59, i64 %19)
  %polly.access.mul..obit61 = extractvalue { i64, i1 } %polly.access.mul.60, 1
  %polly.overflow.state62 = or i1 %polly.overflow.state58, %polly.access.mul..obit61
  %polly.access.mul..res63 = extractvalue { i64, i1 } %polly.access.mul.60, 0
  %polly.access.add.64 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res63, i64 %19)
  %polly.access.add..obit65 = extractvalue { i64, i1 } %polly.access.add.64, 1
  %polly.overflow.state66 = or i1 %polly.overflow.state62, %polly.access.add..obit65
  %polly.access.add..res67 = extractvalue { i64, i1 } %polly.access.add.64, 0
  %polly.access.68 = getelementptr i64, i64* %16, i64 %polly.access.add..res67
  %polly.access.mul.69 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 0, i64 %9)
  %polly.access.mul..obit70 = extractvalue { i64, i1 } %polly.access.mul.69, 1
  %polly.overflow.state71 = or i1 %polly.overflow.state66, %polly.access.mul..obit70
  %polly.access.mul..res72 = extractvalue { i64, i1 } %polly.access.mul.69, 0
  %polly.access.add.73 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res72, i64 0)
  %polly.access.add..obit74 = extractvalue { i64, i1 } %polly.access.add.73, 1
  %polly.overflow.state75 = or i1 %polly.overflow.state71, %polly.access.add..obit74
  %polly.access.add..res76 = extractvalue { i64, i1 } %polly.access.add.73, 0
  %polly.access.77 = getelementptr i64, i64* %6, i64 %polly.access.add..res76
  %45 = ptrtoint i64* %polly.access.68 to i64
  %46 = ptrtoint i64* %polly.access.77 to i64
  %47 = icmp ule i64 %45, %46
  %48 = call { i64, i1 } @llvm.ssub.with.overflow.i64(i64 %22, i64 1)
  %.obit78 = extractvalue { i64, i1 } %48, 1
  %polly.overflow.state79 = or i1 %polly.overflow.state75, %.obit78
  %.res80 = extractvalue { i64, i1 } %48, 0
  %polly.access.mul.81 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res80, i64 %9)
  %polly.access.mul..obit82 = extractvalue { i64, i1 } %polly.access.mul.81, 1
  %polly.overflow.state83 = or i1 %polly.overflow.state79, %polly.access.mul..obit82
  %polly.access.mul..res84 = extractvalue { i64, i1 } %polly.access.mul.81, 0
  %polly.access.add.85 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res84, i64 %14)
  %polly.access.add..obit86 = extractvalue { i64, i1 } %polly.access.add.85, 1
  %polly.overflow.state87 = or i1 %polly.overflow.state83, %polly.access.add..obit86
  %polly.access.add..res88 = extractvalue { i64, i1 } %polly.access.add.85, 0
  %polly.access.89 = getelementptr i64, i64* %6, i64 %polly.access.add..res88
  %polly.access.mul.90 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 0, i64 %19)
  %polly.access.mul..obit91 = extractvalue { i64, i1 } %polly.access.mul.90, 1
  %polly.overflow.state92 = or i1 %polly.overflow.state87, %polly.access.mul..obit91
  %polly.access.mul..res93 = extractvalue { i64, i1 } %polly.access.mul.90, 0
  %polly.access.add.94 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %polly.access.mul..res93, i64 0)
  %polly.access.add..obit95 = extractvalue { i64, i1 } %polly.access.add.94, 1
  %polly.overflow.state96 = or i1 %polly.overflow.state92, %polly.access.add..obit95
  %polly.access.add..res97 = extractvalue { i64, i1 } %polly.access.add.94, 0
  %polly.access.98 = getelementptr i64, i64* %16, i64 %polly.access.add..res97
  %49 = ptrtoint i64* %polly.access.89 to i64
  %50 = ptrtoint i64* %polly.access.98 to i64
  %51 = icmp ule i64 %49, %50
  %52 = or i1 %47, %51
  %53 = and i1 %43, %52
  %54 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 1, i64 %14)
  %.obit99 = extractvalue { i64, i1 } %54, 1
  %polly.overflow.state100 = or i1 %polly.overflow.state96, %.obit99
  %.res101 = extractvalue { i64, i1 } %54, 0
  %55 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res101, i64 %22)
  %.obit102 = extractvalue { i64, i1 } %55, 1
  %polly.overflow.state103 = or i1 %polly.overflow.state100, %.obit102
  %.res104 = extractvalue { i64, i1 } %55, 0
  %56 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 16, i64 %.res104)
  %.obit105 = extractvalue { i64, i1 } %56, 1
  %polly.overflow.state106 = or i1 %polly.overflow.state103, %.obit105
  %.res107 = extractvalue { i64, i1 } %56, 0
  %57 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 0, i64 %.res107)
  %.obit108 = extractvalue { i64, i1 } %57, 1
  %polly.overflow.state109 = or i1 %polly.overflow.state106, %.obit108
  %.res110 = extractvalue { i64, i1 } %57, 0
  %58 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 1, i64 %14)
  %.obit111 = extractvalue { i64, i1 } %58, 1
  %polly.overflow.state112 = or i1 %polly.overflow.state109, %.obit111
  %.res113 = extractvalue { i64, i1 } %58, 0
  %59 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res113, i64 %19)
  %.obit114 = extractvalue { i64, i1 } %59, 1
  %polly.overflow.state115 = or i1 %polly.overflow.state112, %.obit114
  %.res116 = extractvalue { i64, i1 } %59, 0
  %60 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 %.res116, i64 %22)
  %.obit117 = extractvalue { i64, i1 } %60, 1
  %polly.overflow.state118 = or i1 %polly.overflow.state115, %.obit117
  %.res119 = extractvalue { i64, i1 } %60, 0
  %61 = call { i64, i1 } @llvm.smul.with.overflow.i64(i64 31, i64 %.res119)
  %.obit120 = extractvalue { i64, i1 } %61, 1
  %polly.overflow.state121 = or i1 %polly.overflow.state118, %.obit120
  %.res122 = extractvalue { i64, i1 } %61, 0
  %62 = call { i64, i1 } @llvm.sadd.with.overflow.i64(i64 %.res110, i64 %.res122)
  %.obit123 = extractvalue { i64, i1 } %62, 1
  %polly.overflow.state124 = or i1 %polly.overflow.state121, %.obit123
  %.res125 = extractvalue { i64, i1 } %62, 0
  %63 = icmp sge i64 %.res125, 2621440
  %64 = and i1 %53, %63
  %polly.rtc.overflown = xor i1 %polly.overflow.state124, true
  %polly.rtc.result = and i1 %64, %polly.rtc.overflown
  br i1 %polly.rtc.result, label %polly.start, label %if.pre_entry_bb

if.pre_entry_bb:                                  ; preds = %polly.split_new_and_old
  br label %if, !dbg !33

if:                                               ; preds = %if.pre_entry_bb, %L18.loopexit
  %"#temp#4.024" = phi i64 [ %65, %L18.loopexit ], [ 1, %if.pre_entry_bb ]
  %65 = add i64 %"#temp#4.024", 1, !dbg !33
  call void @llvm.dbg.value(metadata i64 %22, i64 0, metadata !26, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %22, i64 0, metadata !26, metadata !28), !dbg !27
  %66 = icmp sgt i64 %22, 0, !dbg !35
  %67 = select i1 %66, i64 %22, i64 0, !dbg !35
  %68 = add nuw i64 %67, 1, !dbg !35
  %69 = icmp slt i64 %22, 1, !dbg !35
  br i1 %69, label %L38, label %if7.lr.ph, !dbg !35

if7.lr.ph:                                        ; preds = %if
  br label %if7, !dbg !35

L18.L65_crit_edge:                                ; preds = %L18.loopexit
  br label %polly.merge_new_and_old, !dbg !33

polly.merge_new_and_old:                          ; preds = %polly.exiting, %L18.L65_crit_edge
  br label %L65, !dbg !36

L65:                                              ; preds = %polly.merge_new_and_old, %top.split
  ret void, !dbg !36

if7:                                              ; preds = %if7.lr.ph, %if7
  %"#temp#.021" = phi i64 [ 1, %if7.lr.ph ], [ %70, %if7 ]
  %70 = add i64 %"#temp#.021", 1, !dbg !35
  call void @llvm.dbg.value(metadata i64 %"#temp#4.024", i64 0, metadata !23, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#.021", i64 0, metadata !21, metadata !28), !dbg !27
  %71 = add i64 %"#temp#4.024", -1, !dbg !37
  %72 = add i64 %"#temp#.021", -1, !dbg !37
  %73 = mul i64 %72, %9, !dbg !37
  %74 = add i64 %71, %73, !dbg !37
  %75 = getelementptr i64, i64* %6, i64 %74, !dbg !37
  %76 = load i64, i64* %75, align 8, !dbg !37, !tbaa !38
  %77 = mul i64 %76, %1, !dbg !37
  call void @llvm.dbg.value(metadata i64 %"#temp#4.024", i64 0, metadata !23, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#.021", i64 0, metadata !21, metadata !28), !dbg !27
  store i64 %77, i64* %75, align 8, !dbg !37, !tbaa !38
  %78 = icmp eq i64 %70, %68, !dbg !35
  br i1 %78, label %L27.L38_crit_edge, label %if7, !dbg !35

L27.L38_crit_edge:                                ; preds = %if7
  br label %L38, !dbg !35

L38:                                              ; preds = %L27.L38_crit_edge, %if
  call void @llvm.dbg.value(metadata i64 %19, i64 0, metadata !25, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %19, i64 0, metadata !25, metadata !28), !dbg !27
  %79 = icmp sgt i64 %19, 0, !dbg !34
  %80 = select i1 %79, i64 %19, i64 0, !dbg !34
  %81 = icmp slt i64 %19, 1, !dbg !34
  br i1 %81, label %L18.loopexit, label %if8.lr.ph, !dbg !34

if8.lr.ph:                                        ; preds = %L38
  br label %if8, !dbg !34

L50.L42.loopexit_crit_edge:                       ; preds = %if9
  br label %L42.loopexit, !dbg !34

L42.loopexit:                                     ; preds = %L50.L42.loopexit_crit_edge, %if8
  %82 = icmp eq i64 %"#temp#3.023", %80, !dbg !34
  br i1 %82, label %L42.L18.loopexit_crit_edge, label %if8, !dbg !34

if8:                                              ; preds = %if8.lr.ph, %L42.loopexit
  %"#temp#3.023" = phi i64 [ 1, %if8.lr.ph ], [ %83, %L42.loopexit ]
  %83 = add i64 %"#temp#3.023", 1, !dbg !34
  call void @llvm.dbg.value(metadata i64 %22, i64 0, metadata !26, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %22, i64 0, metadata !26, metadata !28), !dbg !27
  %84 = icmp slt i64 %22, 1, !dbg !34
  br i1 %84, label %L42.loopexit, label %if9.lr.ph, !dbg !34

if9.lr.ph:                                        ; preds = %if8
  br label %if9, !dbg !34

if9:                                              ; preds = %if9.lr.ph, %if9
  %"#temp#2.022" = phi i64 [ 1, %if9.lr.ph ], [ %85, %if9 ]
  %85 = add i64 %"#temp#2.022", 1, !dbg !34
  call void @llvm.dbg.value(metadata i64 %"#temp#4.024", i64 0, metadata !23, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#2.022", i64 0, metadata !21, metadata !28), !dbg !27
  %86 = add i64 %"#temp#4.024", -1, !dbg !36
  %87 = add i64 %"#temp#2.022", -1, !dbg !36
  %88 = mul i64 %87, %9, !dbg !36
  %89 = add i64 %86, %88, !dbg !36
  %90 = getelementptr i64, i64* %6, i64 %89, !dbg !36
  %91 = load i64, i64* %90, align 8, !dbg !36, !tbaa !38
  call void @llvm.dbg.value(metadata i64 %"#temp#4.024", i64 0, metadata !23, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#3.023", i64 0, metadata !22, metadata !28), !dbg !27
  %92 = add i64 %"#temp#3.023", -1, !dbg !36
  %93 = mul i64 %92, %14, !dbg !36
  %94 = add i64 %86, %93, !dbg !36
  %95 = getelementptr i64, i64* %11, i64 %94, !dbg !36
  %96 = load i64, i64* %95, align 8, !dbg !36, !tbaa !38
  %97 = mul i64 %96, %0, !dbg !36
  call void @llvm.dbg.value(metadata i64 %"#temp#3.023", i64 0, metadata !22, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#2.022", i64 0, metadata !21, metadata !28), !dbg !27
  %98 = mul i64 %87, %19, !dbg !36
  %99 = add i64 %92, %98, !dbg !36
  %100 = getelementptr i64, i64* %16, i64 %99, !dbg !36
  %101 = load i64, i64* %100, align 8, !dbg !36, !tbaa !38
  %102 = mul i64 %97, %101, !dbg !36
  %103 = add i64 %102, %91, !dbg !36
  call void @llvm.dbg.value(metadata i64 %"#temp#4.024", i64 0, metadata !23, metadata !28), !dbg !27
  call void @llvm.dbg.value(metadata i64 %"#temp#2.022", i64 0, metadata !21, metadata !28), !dbg !27
  store i64 %103, i64* %90, align 8, !dbg !36, !tbaa !38
  %104 = icmp eq i64 %85, %68, !dbg !34
  br i1 %104, label %L50.L42.loopexit_crit_edge, label %if9, !dbg !34

polly.start:                                      ; preds = %polly.split_new_and_old
  br label %polly.acc.initialize

polly.acc.initialize:                             ; preds = %polly.start
  %105 = call i8* @polly_initContext()
  %106 = mul nsw i64 %22, %9
  %107 = mul i64 8, %106
  %p_dev_array_MemRef0 = call i8* @polly_allocateMemoryForDevice(i64 %107)
  %108 = mul nsw i64 %19, %14
  %109 = mul i64 8, %108
  %p_dev_array_MemRef1 = call i8* @polly_allocateMemoryForDevice(i64 %109)
  %110 = mul nsw i64 %22, %19
  %111 = mul i64 8, %110
  %p_dev_array_MemRef2 = call i8* @polly_allocateMemoryForDevice(i64 %111)
  br label %polly.cond

polly.cond:                                       ; preds = %polly.acc.initialize
  %112 = icmp sge i64 %22, 1
  %113 = icmp sge i64 %14, 1
  %114 = and i1 %112, %113
  br i1 %114, label %polly.then, label %polly.else

polly.merge:                                      ; preds = %polly.else, %polly.merge136
  call void @polly_freeDeviceMemory(i8* %p_dev_array_MemRef0)
  call void @polly_freeDeviceMemory(i8* %p_dev_array_MemRef1)
  call void @polly_freeDeviceMemory(i8* %p_dev_array_MemRef2)
  call void @polly_freeContext(i8* %105)
  br label %polly.exiting

polly.exiting:                                    ; preds = %polly.merge
  br label %polly.merge_new_and_old

polly.then:                                       ; preds = %polly.cond
  br label %polly.cond126

polly.cond126:                                    ; preds = %polly.then
  %115 = icmp sge i64 %9, 1
  br i1 %115, label %polly.then128, label %polly.else129

polly.merge127:                                   ; preds = %polly.else129, %polly.then128
  br label %polly.cond130

polly.cond130:                                    ; preds = %polly.merge127
  %116 = icmp sge i64 %19, 1
  br i1 %116, label %polly.then132, label %polly.else133

polly.merge131:                                   ; preds = %polly.else133, %polly.then132
  %117 = call i8* @polly_getDevicePtr(i8* %p_dev_array_MemRef0)
  %118 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 0
  store i8* %117, i8** %polly_launch_0_param_0
  %119 = bitcast i8** %polly_launch_0_param_0 to i8*
  store i8* %119, i8** %118
  %120 = call i8* @polly_getDevicePtr(i8* %p_dev_array_MemRef1)
  %121 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 1
  store i8* %120, i8** %polly_launch_0_param_1
  %122 = bitcast i8** %polly_launch_0_param_1 to i8*
  store i8* %122, i8** %121
  %123 = call i8* @polly_getDevicePtr(i8* %p_dev_array_MemRef2)
  %124 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 2
  store i8* %123, i8** %polly_launch_0_param_2
  %125 = bitcast i8** %polly_launch_0_param_2 to i8*
  store i8* %125, i8** %124
  store i64 %22, i64* %polly_launch_0_param_3
  %126 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 3
  %127 = bitcast i64* %polly_launch_0_param_3 to i8*
  store i8* %127, i8** %126
  store i64 %19, i64* %polly_launch_0_param_4
  %128 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 4
  %129 = bitcast i64* %polly_launch_0_param_4 to i8*
  store i8* %129, i8** %128
  store i64 %14, i64* %polly_launch_0_param_5
  %130 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 5
  %131 = bitcast i64* %polly_launch_0_param_5 to i8*
  store i8* %131, i8** %130
  store i64 %9, i64* %polly_launch_0_param_6
  %132 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 6
  %133 = bitcast i64* %polly_launch_0_param_6 to i8*
  store i8* %133, i8** %132
  store void (metadata, i64, metadata, metadata)* @llvm.dbg.value, void (metadata, i64, metadata, metadata)** %polly_launch_0_param_7
  %134 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 7
  %135 = bitcast void (metadata, i64, metadata, metadata)** %polly_launch_0_param_7 to i8*
  store i8* %135, i8** %134
  store i64 %1, i64* %polly_launch_0_param_8
  %136 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 8
  %137 = bitcast i64* %polly_launch_0_param_8 to i8*
  store i8* %137, i8** %136
  store i64 %0, i64* %polly_launch_0_param_9
  %138 = getelementptr [10 x i8*], [10 x i8*]* %polly_launch_0_params, i64 0, i64 9
  %139 = bitcast i64* %polly_launch_0_param_9 to i8*
  store i8* %139, i8** %138
  %140 = call i8* @polly_getKernel(i8* getelementptr inbounds ([6239 x i8], [6239 x i8]* @kernel_0, i32 0, i32 0), i8* getelementptr inbounds ([9 x i8], [9 x i8]* @kernel_0_name, i32 0, i32 0))
  %141 = icmp sge i64 %14, 8161
  %142 = add nsw i64 %14, 31
  %polly.fdiv_q.shr = ashr i64 %142, 5
  %143 = select i1 %141, i64 256, i64 %polly.fdiv_q.shr
  %144 = trunc i64 %143 to i32
  %145 = icmp sge i64 %22, 8162
  %146 = add nsw i64 %22, 31
  %polly.fdiv_q.shr134 = ashr i64 %146, 5
  %147 = select i1 %145, i64 256, i64 %polly.fdiv_q.shr134
  %148 = trunc i64 %147 to i32
  call void @polly_launchKernel(i8* %140, i32 %144, i32 %148, i32 32, i32 16, i32 1, i8* %polly_launch_0_params_i8ptr)
  call void @polly_freeKernel(i8* %140)
  br label %polly.cond135

polly.cond135:                                    ; preds = %polly.merge131
  %149 = icmp sge i64 %9, 1
  br i1 %149, label %polly.then137, label %polly.else138

polly.merge136:                                   ; preds = %polly.else138, %polly.then137
  br label %polly.merge

polly.else:                                       ; preds = %polly.cond
  br label %polly.merge

polly.then128:                                    ; preds = %polly.cond126
  %150 = mul nsw i64 %22, %9
  %151 = mul i64 8, %150
  %152 = bitcast i64* %6 to i8*
  call void @polly_copyFromHostToDevice(i8* %152, i8* %p_dev_array_MemRef0, i64 %151)
  br label %polly.merge127

polly.else129:                                    ; preds = %polly.cond126
  br label %polly.merge127

polly.then132:                                    ; preds = %polly.cond130
  %153 = mul nsw i64 %19, %14
  %154 = mul i64 8, %153
  %155 = bitcast i64* %11 to i8*
  call void @polly_copyFromHostToDevice(i8* %155, i8* %p_dev_array_MemRef1, i64 %154)
  %156 = mul nsw i64 %22, %19
  %157 = mul i64 8, %156
  %158 = bitcast i64* %16 to i8*
  call void @polly_copyFromHostToDevice(i8* %158, i8* %p_dev_array_MemRef2, i64 %157)
  br label %polly.merge131

polly.else133:                                    ; preds = %polly.cond130
  br label %polly.merge131

polly.then137:                                    ; preds = %polly.cond135
  %159 = mul nsw i64 %22, %9
  %160 = mul i64 8, %159
  %161 = bitcast i64* %6 to i8*
  call void @polly_copyFromDeviceToHost(i8* %p_dev_array_MemRef0, i8* %161, i64 %160)
  br label %polly.merge136

polly.else138:                                    ; preds = %polly.cond135
  br label %polly.merge136
}

define i8** @jlcall_kernel_gemm_64399(i8**, i8***, i32) #1 {
top:
  br label %top.split

top.split:                                        ; preds = %top
  %3 = bitcast i8*** %1 to i64**
  %4 = load i64*, i64** %3, align 8
  %5 = load i64, i64* %4, align 8
  %6 = getelementptr i8**, i8*** %1, i64 1
  %7 = bitcast i8*** %6 to i64**
  %8 = load i64*, i64** %7, align 8
  %9 = load i64, i64* %8, align 8
  %10 = getelementptr i8**, i8*** %1, i64 2
  %11 = load i8**, i8*** %10, align 8
  %12 = getelementptr i8**, i8*** %1, i64 3
  %13 = load i8**, i8*** %12, align 8
  %14 = getelementptr i8**, i8*** %1, i64 4
  %15 = load i8**, i8*** %14, align 8
  call void @julia_kernel_gemm_64400(i64 %5, i64 %9, i8** %11, i8** %13, i8** %15)
  ret i8** inttoptr (i64 139727245705232 to i8**)
}

declare i8**** @jl_get_ptls_states()

; Function Attrs: nounwind readnone
declare void @llvm.dbg.declare(metadata, metadata, metadata) #2

; Function Attrs: argmemonly nounwind
declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture writeonly, i8* nocapture readonly, i32, i32, i1) #3

; Function Attrs: argmemonly nounwind
declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1) #3

declare void @jl_enter_handler(i8*)

; Function Attrs: returns_twice
declare i32 @__sigsetjmp(i8*, i32) #4

; Function Attrs: nounwind readnone
declare void @llvm.dbg.value(metadata, i64, metadata, metadata) #2

; Function Attrs: nounwind readnone
declare { i64, i1 } @llvm.ssub.with.overflow.i64(i64, i64) #2

; Function Attrs: nounwind readnone
declare { i64, i1 } @llvm.smul.with.overflow.i64(i64, i64) #2

; Function Attrs: nounwind readnone
declare { i64, i1 } @llvm.sadd.with.overflow.i64(i64, i64) #2

declare i8* @polly_initContext()

declare i8* @polly_allocateMemoryForDevice(i64)

declare void @polly_copyFromHostToDevice(i8*, i8*, i64)

declare i8* @polly_getDevicePtr(i8*)

declare i8* @polly_getKernel(i8*, i8*)

declare void @polly_launchKernel(i8*, i32, i32, i32, i32, i32, i8*)

declare void @polly_freeKernel(i8*)

declare void @polly_copyFromDeviceToHost(i8*, i8*, i64)

declare void @polly_freeDeviceMemory(i8*)

declare void @polly_freeContext(i8*)

attributes #0 = { sspstrong "no-frame-pointer-elim"="true" }
attributes #1 = { "no-frame-pointer-elim"="true" }
attributes #2 = { nounwind readnone }
attributes #3 = { argmemonly nounwind }
attributes #4 = { returns_twice }

!llvm.module.flags = !{!0, !1}
!llvm.dbg.cu = !{!2}

!0 = !{i32 2, !"Dwarf Version", i32 4}
!1 = !{i32 1, !"Debug Info Version", i32 3}
!2 = distinct !DICompileUnit(language: DW_LANG_C89, file: !3, producer: "julia", isOptimized: true, runtimeVersion: 0, emissionKind: FullDebug, enums: !4)
!3 = !DIFile(filename: "REPL[1]", directory: ".")
!4 = !{}
!5 = distinct !DISubprogram(name: "kernel_gemm", linkageName: "julia_kernel_gemm_64400", scope: null, file: !3, type: !6, isLocal: false, isDefinition: true, isOptimized: true, unit: !2, variables: !13)
!6 = !DISubroutineType(types: !7)
!7 = !{!8, !8, !9, !9, !9}
!8 = !DIBasicType(name: "Int64", size: 64, encoding: DW_ATE_unsigned)
!9 = !DIDerivedType(tag: DW_TAG_pointer_type, baseType: !10, size: 64, align: 64)
!10 = !DICompositeType(tag: DW_TAG_structure_type, name: "jl_value_t", file: !11, line: 71, align: 64, elements: !12)
!11 = !DIFile(filename: "julia.h", directory: "")
!12 = !{!9}
!13 = !{!14, !16, !17, !18, !19, !20, !21, !21, !22, !23, !24, !25, !26}
!14 = !DILocalVariable(name: "#self#", arg: 1, scope: !5, file: !3, line: 2, type: !15)
!15 = !DICompositeType(tag: DW_TAG_structure_type, name: "#kernel_gemm", align: 8, elements: !4, runtimeLang: DW_LANG_Julia, identifier: "#kernel_gemm_64337")
!16 = !DILocalVariable(name: "alpha", arg: 2, scope: !5, file: !3, line: 2, type: !8)
!17 = !DILocalVariable(name: "beta", arg: 3, scope: !5, file: !3, line: 2, type: !8)
!18 = !DILocalVariable(name: "C", arg: 4, scope: !5, file: !3, line: 2, type: !9)
!19 = !DILocalVariable(name: "A", arg: 5, scope: !5, file: !3, line: 2, type: !9)
!20 = !DILocalVariable(name: "B", arg: 6, scope: !5, file: !3, line: 2, type: !9)
!21 = !DILocalVariable(name: "j", scope: !5, file: !3, line: 2, type: !8)
!22 = !DILocalVariable(name: "k", scope: !5, file: !3, line: 2, type: !8)
!23 = !DILocalVariable(name: "i", scope: !5, file: !3, line: 2, type: !8)
!24 = !DILocalVariable(name: "ni", scope: !5, file: !3, line: 2, type: !8)
!25 = !DILocalVariable(name: "nk", scope: !5, file: !3, line: 2, type: !8)
!26 = !DILocalVariable(name: "nj", scope: !5, file: !3, line: 2, type: !8)
!27 = !DILocation(line: 2, scope: !5)
!28 = !DIExpression()
!29 = !{!30, !30, i64 0, i64 1}
!30 = !{!"jtbaa_const", !31, i64 0}
!31 = !{!"jtbaa"}
!32 = !DILocation(line: 3, scope: !5)
!33 = !DILocation(line: 4, scope: !5)
!34 = !DILocation(line: 8, scope: !5)
!35 = !DILocation(line: 5, scope: !5)
!36 = !DILocation(line: 9, scope: !5)
!37 = !DILocation(line: 6, scope: !5)
!38 = !{!39, !39, i64 0}
!39 = !{!"jtbaa_arraybuf", !40, i64 0}
!40 = !{!"jtbaa_data", !31, i64 0}