feat(ze): add event support & vadd demo & refactor (#5)

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

README.md

@@ -190,7 +190,8 @@ Contributions of all kinds are welcome. Some particularly impactful areas:
 
 ## License
 
-This project is licensed under the [GNU Affero General Public License v3.0](LICENSE).
+- This project is generally licensed under the [GNU Affero General Public License v3.0](LICENSE).
+- The files in [gozel](gozel) folder follows their original license, which is [MIT](https://github.com/oneapi-src/level-zero/blob/master/LICENSE).
 
 ---
 

cmd/gen/api.go

@@ -1,11 +1,14 @@
 package main
 
-import "os"
+import (
+	"os"
+	"path"
+)
 
 var apif *os.File
 
 func init() {
-	f, err := os.Create("api.go")
+	f, err := os.Create(path.Join("gozel", "api.go"))
 	if err != nil {
 		panic(err)
 	}

cmd/gen/scan.go

@@ -6,6 +6,7 @@ import (
 	"fmt"
 	"io"
 	"os"
+	"path"
 	"strconv"
 	"strings"
 	"unicode"
@@ -134,7 +135,7 @@ func scanHeader(name string, scan *bufio.Scanner) {
 			}
 			fmt.Println(infh(name), "scanning region", region)
 			k := fmt.Sprint(name, "_", region)
-			f, err := os.Create(fmt.Sprint(k, ".go"))
+			f, err := os.Create(path.Join("gozel", fmt.Sprint(k, ".go")))
 			if err != nil {
 				panic(fmt.Sprintf("%s L%d: cannot create region %s, err: %v", name, ln, region, err))
 			}

examples/vadd/main.cpp

@@ -3,7 +3,7 @@
 extern "C" SYCL_EXT_ONEAPI_FUNCTION_PROPERTY((sycl::ext::oneapi::experimental::nd_range_kernel<1>))
 void vector_add(float* a, float* b) {
     auto item = sycl::ext::oneapi::this_work_item::get_nd_item<1>();
-    int idx = item.get_global_id(0);
+    int idx = item.get_global_linear_id();
 
     a[idx] += b[idx];
 }

examples/vadd/main.go

@@ -4,6 +4,7 @@ package main
 import (
 	_ "embed"
 	"fmt"
+	"math"
 	"math/rand"
 	"os"
 	"strconv"
@@ -11,7 +12,7 @@ import (
 	"time"
 	"unsafe"
 
-	"github.com/fumiama/gozel"
+	"github.com/fumiama/gozel/gozel"
 	"github.com/fumiama/gozel/ze"
 )
 
@@ -69,8 +70,7 @@ func main() {
 	fmt.Printf("%-28s (%d, %d, %d)\n", "Max Group Count (X, Y, Z):", cprop.Maxgroupcountx, cprop.Maxgroupcounty, cprop.Maxgroupcountz)
 	fmt.Printf("%-28s %d\n", "Max Total Group Size:", cprop.Maxtotalgroupsize)
 	fmt.Printf("%-28s %d\n", "Max Shared Local Memory:", cprop.Maxsharedlocalmemory)
-	fmt.Printf("%-28s %d\n", "Num Subgroup Sizes:", cprop.Numsubgroupsizes)
+	fmt.Printf("%-28s %v\n", "Subgroup Sizes:", cprop.Subgroupsizes[:cprop.Numsubgroupsizes])
-	fmt.Printf("%-28s %v\n", "Subgroup Sizes:", cprop.Subgroupsizes[:])
 
 	var (
 		X, Y, Z    = uintptr(cprop.Maxgroupsizex), uintptr(1), uintptr(1)
@@ -84,7 +84,7 @@ func main() {
 	fmt.Printf("%-28s %d\n", "Total Elements (N):", N)
 	fmt.Printf("%-28s %d MiB\n", "Buffer Size:", bufsz/1024/1024)
 
-	q, err := ctx.CommandQueueCreate(dev)
+	q, err := ctx.CommandQueueCreate(dev, gozel.ZE_COMMAND_QUEUE_MODE_DEFAULT)
 	if err != nil {
 		panic(err)
 	}
@@ -135,11 +135,11 @@ func main() {
 	}
 	defer krn.Destroy()
 
-	err = krn.SetArgumentValue(0, unsafe.Sizeof(uintptr(0)), unsafe.Pointer(&dbufV1))
+	err = krn.SetArgumentValue(0, &dbufV1)
 	if err != nil {
 		panic(err)
 	}
-	err = krn.SetArgumentValue(1, unsafe.Sizeof(uintptr(0)), unsafe.Pointer(&dbufV2))
+	err = krn.SetArgumentValue(1, &dbufV2)
 	if err != nil {
 		panic(err)
 	}
@@ -154,16 +154,16 @@ func main() {
 	}
 	defer lstpre.Destroy()
 
-	err = lstpre.AppendMemoryCopy(dbufV1, hbufV1, bufsz)
+	err = lstpre.AppendMemoryCopy(dbufV1, hbufV1, bufsz, 0)
 	if err != nil {
 		panic(err)
 	}
-	err = lstpre.AppendMemoryCopy(dbufV2, hbufV2, bufsz)
+	err = lstpre.AppendMemoryCopy(dbufV2, hbufV2, bufsz, 0)
 	if err != nil {
 		panic(err)
 	}
 
-	err = lstpre.AppendBarrier()
+	err = lstpre.AppendBarrier(0)
 	if err != nil {
 		panic(err)
 	}
@@ -181,12 +181,12 @@ func main() {
 
 	err = lstcalc.AppendLaunchKernel(krn, &gozel.ZeGroupCount{
 		Groupcountx: uint32(groupCount), Groupcounty: 1, Groupcountz: 1,
-	})
+	}, 0)
 	if err != nil {
 		panic(err)
 	}
 
-	err = lstcalc.AppendBarrier()
+	err = lstcalc.AppendBarrier(0)
 	if err != nil {
 		panic(err)
 	}
@@ -202,7 +202,7 @@ func main() {
 	}
 	defer lstpost.Destroy()
 
-	err = lstpost.AppendMemoryCopy(hbufV1, dbufV1, bufsz)
+	err = lstpost.AppendMemoryCopy(hbufV1, dbufV1, bufsz, 0)
 	if err != nil {
 		panic(err)
 	}
@@ -217,7 +217,7 @@ func main() {
 	if err != nil {
 		panic(err)
 	}
-	err = q.Synchronize()
+	err = q.Synchronize(math.MaxUint64)
 	if err != nil {
 		panic(err)
 	}

examples/vadd/main.ll

@@ -4,18 +4,21 @@ target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:
 target triple = "spirv64-unknown-unknown"
 
 @__spirv_BuiltInGlobalInvocationId = external local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInGlobalOffset = external local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
 
 ; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: readwrite, inaccessiblemem: write)
 define spir_kernel void @__sycl_kernel_vector_add(ptr addrspace(1) noundef align 4 captures(none) %0, ptr addrspace(1) noundef readonly align 4 captures(none) %1) local_unnamed_addr #0 !kernel_arg_buffer_location !6 !sycl_fixed_targets !7 !sycl_kernel_omit_args !8 {
   %3 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32, !noalias !9
-  %4 = icmp ult i64 %3, 2147483648
+  %4 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalOffset, align 32, !noalias !16
-  tail call void @llvm.assume(i1 %4)
+  %5 = sub i64 %3, %4
-  %5 = getelementptr inbounds float, ptr addrspace(1) %1, i64 %3
+  %6 = icmp ult i64 %5, 2147483648
-  %6 = load float, ptr addrspace(1) %5, align 4, !tbaa !16
+  tail call void @llvm.assume(i1 %6)
-  %7 = getelementptr inbounds float, ptr addrspace(1) %0, i64 %3
+  %7 = getelementptr inbounds float, ptr addrspace(1) %1, i64 %5
-  %8 = load float, ptr addrspace(1) %7, align 4, !tbaa !16
+  %8 = load float, ptr addrspace(1) %7, align 4, !tbaa !23
-  %9 = fadd float %8, %6
+  %9 = getelementptr inbounds float, ptr addrspace(1) %0, i64 %5
-  store float %9, ptr addrspace(1) %7, align 4, !tbaa !16
+  %10 = load float, ptr addrspace(1) %9, align 4, !tbaa !23
+  %11 = fadd float %10, %8
+  store float %11, ptr addrspace(1) %9, align 4, !tbaa !23
   ret void
 }
 
@@ -48,7 +51,14 @@ attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessib
 !13 = distinct !{!13, !"_ZN7__spirv22initGlobalInvocationIdILi1EN4sycl3_V12idILi1EEEEET0_v"}
 !14 = distinct !{!14, !15, !"_ZNK4sycl3_V17nd_itemILi1EE13get_global_idEv: argument 0"}
 !15 = distinct !{!15, !"_ZNK4sycl3_V17nd_itemILi1EE13get_global_idEv"}
-!16 = !{!17, !17, i64 0}
+!16 = !{!17, !19, !21}
-!17 = !{!"float", !18, i64 0}
+!17 = distinct !{!17, !18, !"_ZN7__spirv23InitSizesSTGlobalOffsetILi1EN4sycl3_V12idILi1EEEE8initSizeEv: argument 0"}
-!18 = !{!"omnipotent char", !19, i64 0}
+!18 = distinct !{!18, !"_ZN7__spirv23InitSizesSTGlobalOffsetILi1EN4sycl3_V12idILi1EEEE8initSizeEv"}
-!19 = !{!"Simple C++ TBAA"}
+!19 = distinct !{!19, !20, !"_ZN7__spirv16initGlobalOffsetILi1EN4sycl3_V12idILi1EEEEET0_v: argument 0"}
+!20 = distinct !{!20, !"_ZN7__spirv16initGlobalOffsetILi1EN4sycl3_V12idILi1EEEEET0_v"}
+!21 = distinct !{!21, !22, !"_ZNK4sycl3_V17nd_itemILi1EE10get_offsetEv: argument 0"}
+!22 = distinct !{!22, !"_ZNK4sycl3_V17nd_itemILi1EE10get_offsetEv"}
+!23 = !{!24, !24, i64 0}
+!24 = !{!"float", !25, i64 0}
+!25 = !{!"omnipotent char", !26, i64 0}
+!26 = !{!"Simple C++ TBAA"}

examples/vadd_event/.gitignore

@@ -0,0 +1 @@
+/device*

examples/vadd_event/main.cpp

@@ -0,0 +1,9 @@
+#include <sycl/sycl.hpp>
+
+extern "C" SYCL_EXT_ONEAPI_FUNCTION_PROPERTY((sycl::ext::oneapi::experimental::nd_range_kernel<1>))
+void vector_add(float* a, float* b) {
+    auto item = sycl::ext::oneapi::this_work_item::get_nd_item<1>();
+    int idx = item.get_global_linear_id();
+
+    a[idx] += b[idx];
+}

examples/vadd_event/main.go

@@ -0,0 +1,231 @@
+// Package main demonstrates vector addition using the gozel Level Zero bindings.
+package main
+
+import (
+	_ "embed"
+	"fmt"
+	"math"
+	"math/rand"
+	"os"
+	"strconv"
+	"strings"
+	"time"
+	"unsafe"
+
+	"github.com/fumiama/gozel/gozel"
+	"github.com/fumiama/gozel/ze"
+)
+
+//go:generate clang++ -fsycl -fsycl-device-only -fsycl-targets=spirv64 -Xclang -emit-llvm-bc main.cpp -o device_kern.bc
+//go:generate sycl-post-link -symbols -split=auto -o device_kern.table device_kern.bc
+//go:generate clang++ -target spirv64-unknown-unknown -S -emit-llvm -x ir device_kern_0.bc -o device_kern.ll
+//go:generate llvm-spirv -o main.spv device_kern.bc
+//go:generate clang++ -target spirv64-unknown-unknown -S -emit-llvm -x ir device_kern.bc -o main.ll
+
+//go:embed main.spv
+var kernelspv []byte
+
+func main() {
+	gpus, err := ze.InitGPUDrivers()
+	if err != nil {
+		panic(err)
+	}
+	if len(gpus) == 0 {
+		panic("no gpu available")
+	}
+	gpu := gpus[0]
+
+	ctx, err := gpu.ContextCreate()
+	if err != nil {
+		panic(err)
+	}
+
+	devs, err := gpu.DeviceGet()
+	if err != nil {
+		panic(err)
+	}
+	if len(devs) == 0 {
+		panic("no device available")
+	}
+	dev := devs[0]
+
+	prop, err := dev.DeviceGetProperties()
+	if err != nil {
+		panic(err)
+	}
+
+	fmt.Println("===============  Device Basic Properties  ===============")
+	name, _, _ := strings.Cut(string(prop.Name[:]), "\x00")
+	fmt.Println(
+		"Running on device: ID =", prop.Deviceid, ", Name =", name,
+		"@", strconv.FormatFloat(float64(prop.Coreclockrate)/1024/1024/1024, 'f', 2, 64), "GHz.",
+	)
+
+	cprop, err := dev.DeviceGetComputeProperties()
+	if err != nil {
+		panic(err)
+	}
+	fmt.Println("=============== Device Compute Properties ===============")
+	fmt.Printf("%-28s (%d, %d, %d)\n", "Max Group Size (X, Y, Z):", cprop.Maxgroupsizex, cprop.Maxgroupsizey, cprop.Maxgroupsizez)
+	fmt.Printf("%-28s (%d, %d, %d)\n", "Max Group Count (X, Y, Z):", cprop.Maxgroupcountx, cprop.Maxgroupcounty, cprop.Maxgroupcountz)
+	fmt.Printf("%-28s %d\n", "Max Total Group Size:", cprop.Maxtotalgroupsize)
+	fmt.Printf("%-28s %d\n", "Max Shared Local Memory:", cprop.Maxsharedlocalmemory)
+	fmt.Printf("%-28s %d\n", "Num Subgroup Sizes:", cprop.Numsubgroupsizes)
+	fmt.Printf("%-28s %v\n", "Subgroup Sizes:", cprop.Subgroupsizes[:])
+
+	var (
+		X, Y, Z    = uintptr(cprop.Maxgroupsizex), uintptr(1), uintptr(1)
+		groupCount = uintptr(65536)
+		N          = X * groupCount
+		bufsz      = N * unsafe.Sizeof(float32(0))
+	)
+	fmt.Println("=============== Computation Configuration ===============")
+	fmt.Printf("%-28s (%d, %d, %d)\n", "Group Size (X, Y, Z):", X, Y, Z)
+	fmt.Printf("%-28s %d\n", "Group Count:", groupCount)
+	fmt.Printf("%-28s %d\n", "Total Elements (N):", N)
+	fmt.Printf("%-28s %d MiB\n", "Buffer Size:", bufsz/1024/1024)
+
+	hbufV1, err := ctx.MemAllocHost(bufsz, 1)
+	if err != nil {
+		panic(err)
+	}
+	defer ctx.MemFree(hbufV1)
+
+	hbufV2, err := ctx.MemAllocHost(bufsz, 1)
+	if err != nil {
+		panic(err)
+	}
+	defer ctx.MemFree(hbufV2)
+
+	dbufV1, err := ctx.MemAllocDevice(dev, bufsz, 1)
+	if err != nil {
+		panic(err)
+	}
+	defer ctx.MemFree(dbufV1)
+
+	dbufV2, err := ctx.MemAllocDevice(dev, bufsz, 1)
+	if err != nil {
+		panic(err)
+	}
+	defer ctx.MemFree(dbufV2)
+
+	floatbuf := make([]float32, 2*N)
+	for i := range floatbuf {
+		floatbuf[i] = rand.Float32()
+	}
+
+	zev1, zev2 := unsafe.Slice((*float32)(hbufV1), N), unsafe.Slice((*float32)(hbufV2), N)
+	copy(zev1, floatbuf[:N])
+	copy(zev2, floatbuf[N:])
+
+	mod, err := ctx.ModuleCreate(dev, kernelspv)
+	if err != nil {
+		panic(err)
+	}
+	defer mod.Destroy()
+
+	krn, err := mod.KernelCreate("__sycl_kernel_vector_add")
+	if err != nil {
+		panic(err)
+	}
+	defer krn.Destroy()
+
+	err = krn.SetArgumentValue(0, &dbufV1)
+	if err != nil {
+		panic(err)
+	}
+	err = krn.SetArgumentValue(1, &dbufV2)
+	if err != nil {
+		panic(err)
+	}
+	err = krn.SetGroupSize(uint32(X), uint32(Y), uint32(Z))
+	if err != nil {
+		panic(err)
+	}
+
+	evp, err := ctx.EventPoolCreate(3, dev)
+	if err != nil {
+		panic(err)
+	}
+	defer evp.Destroy()
+
+	evcph2dv1, err := evp.EventCreate(0, gozel.ZE_EVENT_SCOPE_FLAG_HOST, 0)
+	if err != nil {
+		panic(err)
+	}
+	defer evcph2dv1.Destroy()
+	evcph2dv2, err := evp.EventCreate(1, gozel.ZE_EVENT_SCOPE_FLAG_HOST, 0)
+	if err != nil {
+		panic(err)
+	}
+	defer evcph2dv2.Destroy()
+
+	start := time.Now()
+
+	lst, err := ctx.CommandListCreateImmediate(dev, gozel.ZE_COMMAND_QUEUE_MODE_DEFAULT)
+	if err != nil {
+		panic(err)
+	}
+	err = lst.AppendMemoryCopy(dbufV1, hbufV1, bufsz, evcph2dv1)
+	if err != nil {
+		panic(err)
+	}
+	err = lst.AppendMemoryCopy(dbufV2, hbufV2, bufsz, evcph2dv2)
+	if err != nil {
+		panic(err)
+	}
+
+	evk, err := evp.EventCreate(2, gozel.ZE_EVENT_SCOPE_FLAG_HOST, 0)
+	if err != nil {
+		panic(err)
+	}
+	defer evk.Destroy()
+
+	err = lst.AppendLaunchKernel(krn, &gozel.ZeGroupCount{
+		Groupcountx: uint32(groupCount), Groupcounty: 1, Groupcountz: 1,
+	}, evk, evcph2dv1, evcph2dv2)
+	if err != nil {
+		panic(err)
+	}
+
+	err = lst.AppendMemoryCopy(hbufV1, dbufV1, bufsz, 0, evk)
+	if err != nil {
+		panic(err)
+	}
+	err = lst.HostSynchronize(math.MaxUint64)
+	if err != nil {
+		panic(err)
+	}
+
+	elapsed := time.Since(start)
+
+	fmt.Println("===============    Calculation Results    ===============")
+	fmt.Printf("%-28s %.6f ms\n", "GPU Execution Time:", elapsed.Seconds()*1000)
+	fmt.Printf("%-28s %.2f GiB/s\n", "GPU Throughput:", float64(bufsz)/elapsed.Seconds()/1e9)
+
+	tmpbuf := make([]float32, N)
+	start = time.Now()
+	for i := range N {
+		tmpbuf[i] = floatbuf[i] + floatbuf[N+i]
+	}
+	elapsed = time.Since(start)
+
+	fmt.Println("===============    Validation Results    ===============")
+	fmt.Printf("%-28s %.6f ms\n", "CPU Execution Time:", elapsed.Seconds()*1000)
+	fmt.Printf("%-28s %.2f GiB/s\n", "CPU Throughput:", float64(bufsz)/elapsed.Seconds()/1e9)
+
+	fail := false
+	for i := range N {
+		expect := floatbuf[i] + floatbuf[N+i]
+		if zev1[i] != expect {
+			fail = true
+			fmt.Printf("[%05d] expect %f = %f + %f, got %f.\n", i, expect, floatbuf[i], floatbuf[N+i], zev1[i])
+		}
+	}
+
+	if fail {
+		os.Exit(1)
+	}
+
+	fmt.Println("Test Passed!!!")
+}

examples/vadd_event/main.ll

@@ -0,0 +1,64 @@
+; ModuleID = 'device_kern.bc'
+source_filename = "main.cpp"
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-n8:16:32:64-G1"
+target triple = "spirv64-unknown-unknown"
+
+@__spirv_BuiltInGlobalInvocationId = external local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+@__spirv_BuiltInGlobalOffset = external local_unnamed_addr addrspace(1) constant <3 x i64>, align 32
+
+; Function Attrs: mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: readwrite, inaccessiblemem: write)
+define spir_kernel void @__sycl_kernel_vector_add(ptr addrspace(1) noundef align 4 captures(none) %0, ptr addrspace(1) noundef readonly align 4 captures(none) %1) local_unnamed_addr #0 !kernel_arg_buffer_location !6 !sycl_fixed_targets !7 !sycl_kernel_omit_args !8 {
+  %3 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalInvocationId, align 32, !noalias !9
+  %4 = load i64, ptr addrspace(1) @__spirv_BuiltInGlobalOffset, align 32, !noalias !16
+  %5 = sub i64 %3, %4
+  %6 = icmp ult i64 %5, 2147483648
+  tail call void @llvm.assume(i1 %6)
+  %7 = getelementptr inbounds float, ptr addrspace(1) %1, i64 %5
+  %8 = load float, ptr addrspace(1) %7, align 4, !tbaa !23
+  %9 = getelementptr inbounds float, ptr addrspace(1) %0, i64 %5
+  %10 = load float, ptr addrspace(1) %9, align 4, !tbaa !23
+  %11 = fadd float %10, %8
+  store float %11, ptr addrspace(1) %9, align 4, !tbaa !23
+  ret void
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write)
+declare void @llvm.assume(i1 noundef) #1
+
+declare dso_local spir_func i32 @_Z18__spirv_ocl_printfPU3AS2Kcz(ptr addrspace(2), ...)
+
+attributes #0 = { mustprogress nofree norecurse nosync nounwind willreturn memory(argmem: readwrite, inaccessiblemem: write) "frame-pointer"="all" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "sycl-module-id"="main.cpp" "sycl-nd-range-kernel"="1" "sycl-optlevel"="2" "uniform-work-group-size"="true" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn memory(inaccessiblemem: write) }
+
+!llvm.module.flags = !{!0, !1, !2}
+!opencl.spir.version = !{!3}
+!spirv.Source = !{!4}
+!llvm.ident = !{!5}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, !"sycl-device", i32 1}
+!2 = !{i32 7, !"frame-pointer", i32 2}
+!3 = !{i32 1, i32 2}
+!4 = !{i32 4, i32 100000}
+!5 = !{!"clang version 21.0.0git (https://github.com/intel/llvm d5f649b706f63b5c74e1929bc95db8de91085560)"}
+!6 = !{i32 -1, i32 -1}
+!7 = !{}
+!8 = !{i1 false, i1 false}
+!9 = !{!10, !12, !14}
+!10 = distinct !{!10, !11, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi1EN4sycl3_V12idILi1EEEE8initSizeEv: argument 0"}
+!11 = distinct !{!11, !"_ZN7__spirv29InitSizesSTGlobalInvocationIdILi1EN4sycl3_V12idILi1EEEE8initSizeEv"}
+!12 = distinct !{!12, !13, !"_ZN7__spirv22initGlobalInvocationIdILi1EN4sycl3_V12idILi1EEEEET0_v: argument 0"}
+!13 = distinct !{!13, !"_ZN7__spirv22initGlobalInvocationIdILi1EN4sycl3_V12idILi1EEEEET0_v"}
+!14 = distinct !{!14, !15, !"_ZNK4sycl3_V17nd_itemILi1EE13get_global_idEv: argument 0"}
+!15 = distinct !{!15, !"_ZNK4sycl3_V17nd_itemILi1EE13get_global_idEv"}
+!16 = !{!17, !19, !21}
+!17 = distinct !{!17, !18, !"_ZN7__spirv23InitSizesSTGlobalOffsetILi1EN4sycl3_V12idILi1EEEE8initSizeEv: argument 0"}
+!18 = distinct !{!18, !"_ZN7__spirv23InitSizesSTGlobalOffsetILi1EN4sycl3_V12idILi1EEEE8initSizeEv"}
+!19 = distinct !{!19, !20, !"_ZN7__spirv16initGlobalOffsetILi1EN4sycl3_V12idILi1EEEEET0_v: argument 0"}
+!20 = distinct !{!20, !"_ZN7__spirv16initGlobalOffsetILi1EN4sycl3_V12idILi1EEEEET0_v"}
+!21 = distinct !{!21, !22, !"_ZNK4sycl3_V17nd_itemILi1EE10get_offsetEv: argument 0"}
+!22 = distinct !{!22, !"_ZNK4sycl3_V17nd_itemILi1EE10get_offsetEv"}
+!23 = !{!24, !24, i64 0}
+!24 = !{!"float", !25, i64 0}
+!25 = !{!"omnipotent char", !26, i64 0}
+!26 = !{!"Simple C++ TBAA"}