Merge from 'sycl' to 'sycl-web' (11 commits)

Author: iclsrc

.github/workflows/sycl-linux-precommit.yml

@@ -44,6 +44,7 @@ jobs:
     uses: ./.github/workflows/sycl-detect-changes.yml
 
   build:
+    name: Self build
     needs: [detect_changes]
     if: always() && success()
     uses: ./.github/workflows/sycl-linux-build.yml

.github/workflows/sycl-post-commit.yml

@@ -34,7 +34,7 @@ jobs:
     uses: ./.github/workflows/sycl-detect-changes.yml
 
   build-lin:
-    name: Linux (Self build + no-assertions)
+    name: Linux (GCC + no-assertions)
     if: github.repository == 'intel/llvm'
     uses: ./.github/workflows/sycl-linux-build.yml
     with:

devops/dependencies-igc-dev.json

@@ -1,10 +1,10 @@
 {
   "linux": {
     "igc_dev": {
-      "github_tag": "igc-dev-e22e2a8",
-      "version": "e22e2a8",
-      "updated_at": "2025-07-31T01:11:27Z",
-      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/3654137925/zip",
+      "github_tag": "igc-dev-04d2e53",
+      "version": "04d2e53",
+      "updated_at": "2025-08-04T02:10:56Z",
+      "url": "https://api.github.com/repos/intel/intel-graphics-compiler/actions/artifacts/3677914099/zip",
       "root": "{DEPS_ROOT}/opencl/runtime/linux/oclgpu"
     }
   }

devops/scripts/benchmarks/benches/compute.py

@@ -51,7 +51,7 @@ def git_url(self) -> str:
         return "https://github.com/intel/compute-benchmarks.git"
 
     def git_hash(self) -> str:
-        return "83b9ae3ebb3563552409f3a317cdc1cf3d3ca6bd"
+        return "c9e135d4f26dd6badd83009f92f25d6285fc1e21"
 
     def setup(self) -> None:
         if options.sycl is None:
@@ -73,6 +73,8 @@ def setup(self) -> None:
             f"-DBUILD_SYCL=ON",
             f"-DSYCL_COMPILER_ROOT={options.sycl}",
             f"-DALLOW_WARNINGS=ON",
+            f"-DCMAKE_CXX_COMPILER=clang++",
+            f"-DCMAKE_C_COMPILER=clang",
         ]
 
         if options.ur_adapter == "cuda":
@@ -204,6 +206,17 @@ def benchmarks(self) -> list[Benchmark]:
 
             # Add GraphApiSubmitGraph benchmarks
             for in_order_queue in [0, 1]:
+                benches.append(
+                    GraphApiSubmitGraph(
+                        self,
+                        runtime,
+                        in_order_queue,
+                        self.submit_graph_num_kernels[-1],
+                        0,
+                        useEvents=0,
+                        useHostTasks=1,
+                    )
+                )
                 for num_kernels in self.submit_graph_num_kernels:
                     for measure_completion_time in [0, 1]:
                         for use_events in [0, 1]:
@@ -215,6 +228,7 @@ def benchmarks(self) -> list[Benchmark]:
                                     num_kernels,
                                     measure_completion_time,
                                     use_events,
+                                    useHostTasks=0,
                                 )
                             )
 
@@ -840,22 +854,25 @@ def __init__(
         numKernels,
         measureCompletionTime,
         useEvents,
+        useHostTasks,
     ):
         self.inOrderQueue = inOrderQueue
         self.numKernels = numKernels
         self.measureCompletionTime = measureCompletionTime
         self.useEvents = useEvents
+        self.useHostTasks = useHostTasks
         self.ioq_str = "in order" if self.inOrderQueue else "out of order"
         self.measure_str = (
             " with measure completion" if self.measureCompletionTime else ""
         )
         self.use_events_str = f" with events" if self.useEvents else ""
+        self.host_tasks_str = f" use host tasks" if self.useHostTasks else ""
         super().__init__(
             bench, f"graph_api_benchmark_{runtime.value}", "SubmitGraph", runtime
         )
 
     def explicit_group(self):
-        return f"SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}, {self.numKernels} kernels"
+        return f"SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}{self.host_tasks_str}, {self.numKernels} kernels"
 
     def description(self) -> str:
         return (
@@ -864,10 +881,10 @@ def description(self) -> str:
         )
 
     def name(self):
-        return f"graph_api_benchmark_{self.runtime.value} SubmitGraph{self.use_events_str} numKernels:{self.numKernels} ioq {self.inOrderQueue} measureCompletion {self.measureCompletionTime}"
+        return f"graph_api_benchmark_{self.runtime.value} SubmitGraph{self.use_events_str}{self.host_tasks_str} numKernels:{self.numKernels} ioq {self.inOrderQueue} measureCompletion {self.measureCompletionTime}"
 
     def display_name(self) -> str:
-        return f"{self.runtime.value.upper()} SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}, {self.numKernels} kernels"
+        return f"{self.runtime.value.upper()} SubmitGraph {self.ioq_str}{self.measure_str}{self.use_events_str}{self.host_tasks_str}, {self.numKernels} kernels"
 
     def get_tags(self):
         return [
@@ -888,6 +905,7 @@ def bin_args(self) -> list[str]:
             "--KernelExecutionTime=1",
             f"--UseEvents={self.useEvents}",
             "--UseExplicit=0",
+            f"--UseHostTasks={self.useHostTasks}",
         ]
 
 

devops/scripts/benchmarks/compare.py

@@ -112,13 +112,9 @@ def get_result_paths() -> list[str]:
         def validate_benchmark_result(result: BenchmarkRun) -> bool:
             """
             Returns True if result file:
-            - Was ran on the target machine/hostname specified
-            - Sanity check: ensure metadata are all expected values:
               - Date is truly before cutoff timestamp
               - Name truly matches up with specified result_name
             """
-            if result.hostname != hostname:
-                return False
             if result.name != result_name:
                 log.warning(
                     f"Result file {result_path} does not match specified result name {result.name}."

libdevice/cmath_wrapper.cpp

@@ -10,15 +10,6 @@
 
 #if defined(__SPIR__) || defined(__SPIRV__)
 
-DEVICE_EXTERN_C_INLINE
-int abs(int x) { return __devicelib_abs(x); }
-
-DEVICE_EXTERN_C_INLINE
-long int labs(long int x) { return __devicelib_labs(x); }
-
-DEVICE_EXTERN_C_INLINE
-long long int llabs(long long int x) { return __devicelib_llabs(x); }
-
 DEVICE_EXTERN_C_INLINE
 float fabsf(float x) { return __devicelib_fabsf(x); }
 
@@ -51,15 +42,6 @@ float rsqrtf(float x) { return __devicelib_rsqrtf(x); }
 DEVICE_EXTERN_C_INLINE
 float exp10f(float x) { return __devicelib_exp10f(x); }
 
-DEVICE_EXTERN_C_INLINE
-div_t div(int x, int y) { return __devicelib_div(x, y); }
-
-DEVICE_EXTERN_C_INLINE
-ldiv_t ldiv(long x, long y) { return __devicelib_ldiv(x, y); }
-
-DEVICE_EXTERN_C_INLINE
-lldiv_t lldiv(long long x, long long y) { return __devicelib_lldiv(x, y); }
-
 DEVICE_EXTERN_C_INLINE
 float roundf(float x) { return __devicelib_roundf(x); }
 

libdevice/device_math.h

@@ -14,33 +14,6 @@
     defined(__AMDGCN__)
 #include <cstdint>
 
-typedef struct {
-  int32_t quot;
-  int32_t rem;
-} __devicelib_div_t_32;
-
-typedef struct {
-  int64_t quot;
-  int64_t rem;
-} __devicelib_div_t_64;
-
-typedef __devicelib_div_t_32 div_t;
-#ifdef _WIN32
-typedef __devicelib_div_t_32 ldiv_t;
-#else
-typedef __devicelib_div_t_64 ldiv_t;
-#endif
-typedef __devicelib_div_t_64 lldiv_t;
-
-DEVICE_EXTERN_C
-int __devicelib_abs(int x);
-
-DEVICE_EXTERN_C
-long int __devicelib_labs(long int x);
-
-DEVICE_EXTERN_C
-long long int __devicelib_llabs(long long int x);
-
 DEVICE_EXTERN_C
 float __devicelib_fabsf(float x);
 
@@ -107,15 +80,6 @@ double __devicelib_exp10(double x);
 DEVICE_EXTERN_C
 float __devicelib_exp10f(float x);
 
-DEVICE_EXTERN_C
-div_t __devicelib_div(int x, int y);
-
-DEVICE_EXTERN_C
-ldiv_t __devicelib_ldiv(long int x, long int y);
-
-DEVICE_EXTERN_C
-lldiv_t __devicelib_lldiv(long long int x, long long int y);
-
 DEVICE_EXTERN_C
 double __devicelib_round(double x);
 

libdevice/fallback-cmath.cpp

@@ -16,15 +16,6 @@
 // TODO: generate the DeviceLibFuncMap in sycl-post-link.cpp automatically
 // during the build based on libdevice to avoid manually sync.
 
-DEVICE_EXTERN_C_INLINE
-int __devicelib_abs(int x) { return x < 0 ? -x : x; }
-
-DEVICE_EXTERN_C_INLINE
-long int __devicelib_labs(long int x) { return x < 0 ? -x : x; }
-
-DEVICE_EXTERN_C_INLINE
-long long int __devicelib_llabs(long long int x) { return x < 0 ? -x : x; }
-
 DEVICE_EXTERN_C_INLINE
 float __devicelib_fabsf(float x) { return x < 0 ? -x : x; }
 
@@ -62,15 +53,6 @@ float __devicelib_rsqrtf(float x) { return __spirv_ocl_rsqrt(x); }
 DEVICE_EXTERN_C_INLINE
 float __devicelib_exp10f(float x) { return __spirv_ocl_exp10(x); }
 
-DEVICE_EXTERN_C_INLINE
-div_t __devicelib_div(int x, int y) { return {x / y, x % y}; }
-
-DEVICE_EXTERN_C_INLINE
-ldiv_t __devicelib_ldiv(long x, long y) { return {x / y, x % y}; }
-
-DEVICE_EXTERN_C_INLINE
-lldiv_t __devicelib_lldiv(long long x, long long y) { return {x / y, x % y}; }
-
 DEVICE_EXTERN_C_INLINE
 float __devicelib_scalbnf(float x, int n) { return __spirv_ocl_ldexp(x, n); }
 

llvm-spirv/lib/SPIRV/SPIRVWriter.cpp

@@ -3984,6 +3984,17 @@ bool LLVMToSPIRVBase::isKnownIntrinsic(Intrinsic::ID Id) {
   case Intrinsic::masked_gather:
   case Intrinsic::masked_scatter:
   case Intrinsic::modf:
+  // INTEL_CUSTOMIZATION begin
+  // fpbuiltin intrinsics are not part of upstream LLVM, yet we have to register
+  // them as known.
+  case Intrinsic::fpbuiltin_fadd:
+  case Intrinsic::fpbuiltin_fsub:
+  case Intrinsic::fpbuiltin_fmul:
+  case Intrinsic::fpbuiltin_fdiv:
+  case Intrinsic::fpbuiltin_frem:
+  case Intrinsic::fpbuiltin_sqrt:
+  case Intrinsic::fpbuiltin_ldexp:
+    // INTEL_CUSTOMIZATION end
     return true;
   default:
     // Unknown intrinsics' declarations should always be translated

llvm-spirv/test/extensions/INTEL/SPV_INTEL_fp_max_error/IntelFPMaxError.ll

@@ -14,6 +14,10 @@
 ; RUN: llvm-spirv -r %t.spv -o %t.rev.bc
 ; RUN: llvm-dis < %t.rev.bc | FileCheck %s --check-prefix=CHECK-LLVM
 
+; INTEL_CUSTOMIZATION begin
+; CHECK-SPIRV-NOT: Name [[#]] "llvm.fpbuiltin."
+; INTEL_CUSTOMIZATION end
+
 ; CHECK-SPIRV: Capability FPMaxErrorINTEL
 ; CHECK-SPIRV: Extension "SPV_INTEL_fp_max_error"
 ; CHECK-SPIRV: ExtInstImport [[#OCLEXTID:]] "OpenCL.std"