Merge from 'sycl' to 'sycl-web' (#3)

clang/include/clang/Basic/DiagnosticSemaKinds.td

@@ -11197,7 +11197,7 @@ def err_sycl_function_attribute_mismatch : Error<
   "SYCL kernel without %0 attribute can't call a function with this attribute">;
 def err_sycl_x_y_z_arguments_must_be_one : Error<
   "%0 X-, Y- and Z- sizes must be 1 when %1 attribute is used with value 0">;
-def err_sycl_attibute_cannot_be_applied_here
+def err_sycl_attribute_internal_function
     : Error<"%0 attribute cannot be applied to a "
             "static function or function in an anonymous namespace">;
 def err_sycl_compiletime_property_duplication : Error<

clang/lib/Sema/SemaDeclAttr.cpp

@@ -4833,7 +4833,7 @@ static void handleOptimizeNoneAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
 static void handleSYCLDeviceAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   auto *FD = cast<FunctionDecl>(D);
   if (!FD->isExternallyVisible()) {
-    S.Diag(AL.getLoc(), diag::err_sycl_attibute_cannot_be_applied_here) << AL;
+    S.Diag(AL.getLoc(), diag::err_sycl_attribute_internal_function) << AL;
     return;
   }
 
@@ -4844,7 +4844,7 @@ static void handleSYCLDeviceIndirectlyCallableAttr(Sema &S, Decl *D,
                                                    const ParsedAttr &AL) {
   auto *FD = cast<FunctionDecl>(D);
   if (!FD->isExternallyVisible()) {
-    S.Diag(AL.getLoc(), diag::err_sycl_attibute_cannot_be_applied_here) << AL;
+    S.Diag(AL.getLoc(), diag::err_sycl_attribute_internal_function) << AL;
     return;
   }
 
@@ -4854,11 +4854,6 @@ static void handleSYCLDeviceIndirectlyCallableAttr(Sema &S, Decl *D,
 
 static void handleSYCLRegisterNumAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   auto *VD = cast<VarDecl>(D);
-  if (!VD->hasGlobalStorage()) {
-    S.Diag(AL.getLoc(), diag::err_sycl_attibute_cannot_be_applied_here)
-        << AL << 0;
-    return;
-  }
   if (!checkAttributeNumArgs(S, AL, 1))
     return;
   uint32_t RegNo = 0;

clang/lib/Sema/SemaSYCL.cpp

@@ -178,20 +178,8 @@ static bool IsSyclMathFunc(unsigned BuiltinID) {
   case Builtin::BI__builtin_truncl:
   case Builtin::BIlroundl:
   case Builtin::BI__builtin_lroundl:
-  case Builtin::BIfmax:
-  case Builtin::BI__builtin_fmax:
-  case Builtin::BIfmin:
-  case Builtin::BI__builtin_fmin:
-  case Builtin::BIfmaxf:
-  case Builtin::BI__builtin_fmaxf:
-  case Builtin::BIfminf:
-  case Builtin::BI__builtin_fminf:
   case Builtin::BIlroundf:
   case Builtin::BI__builtin_lroundf:
-  case Builtin::BI__builtin_fpclassify:
-  case Builtin::BI__builtin_isfinite:
-  case Builtin::BI__builtin_isinf:
-  case Builtin::BI__builtin_isnormal:
     return false;
   default:
     break;

clang/test/SemaSYCL/supported_math.cpp

@@ -1,4 +1,5 @@
 // RUN: %clang_cc1 -fsycl -fsycl-is-device -fsyntax-only -Wno-sycl-strict -verify %s
+// expected-no-diagnostics
 extern "C" float sinf(float);
 extern "C" float cosf(float);
 extern "C" float floorf(float);
@@ -8,6 +9,8 @@ extern "C" float rintf(float);
 extern "C" float roundf(float);
 extern "C" float truncf(float);
 extern "C" float copysignf(float, float);
+extern "C" float fminf(float, float);
+extern "C" float fmaxf(float, float);
 extern "C" double sin(double);
 extern "C" double cos(double);
 extern "C" double floor(double);
@@ -17,6 +20,8 @@ extern "C" double rint(double);
 extern "C" double round(double);
 extern "C" double trunc(double);
 extern "C" double copysign(double, double);
+extern "C" double fmin(double, double);
+extern "C" double fmax(double, double);
 template <typename name, typename Func>
 __attribute__((sycl_kernel)) void kernel(const Func &kernelFunc) {
   kernelFunc();
@@ -26,39 +31,51 @@ int main() {
   kernel<class kernel_function>([=]() {
     int acc[1] = {5};
     acc[0] *= 2;
-    acc[0] += (int)truncf(1.0f);                     // expected-no-diagnostics
-    acc[0] += (int)trunc(1.0);                       // expected-no-diagnostics
-    acc[0] += (int)roundf(1.0f);                     // expected-no-diagnostics
-    acc[0] += (int)round(1.0);                       // expected-no-diagnostics
-    acc[0] += (int)rintf(1.0f);                      // expected-no-diagnostics
-    acc[0] += (int)rint(1.0);                        // expected-no-diagnostics
-    acc[0] += (int)nearbyintf(0.5f);                 // expected-no-diagnostics
-    acc[0] += (int)nearbyint(0.5);                   // expected-no-diagnostics
-    acc[0] += (int)floorf(0.5f);                     // expected-no-diagnostics
-    acc[0] += (int)floor(0.5);                       // expected-no-diagnostics
-    acc[0] += (int)copysignf(1.0f, -0.5f);           // expected-no-diagnostics
-    acc[0] += (int)copysign(1.0, -0.5);              // expected-no-diagnostics
-    acc[0] += (int)sinf(1.0f);                       // expected-no-diagnostics
-    acc[0] += (int)sin(1.0);                         // expected-no-diagnostics
-    acc[0] += (int)__builtin_sinf(1.0f);             // expected-no-diagnostics
-    acc[0] += (int)__builtin_sin(1.0);               // expected-no-diagnostics
-    acc[0] += (int)cosf(1.0f);                       // expected-no-diagnostics
-    acc[0] += (int)cos(1.0);                         // expected-no-diagnostics
-    acc[0] += (int)__builtin_cosf(1.0f);             // expected-no-diagnostics
-    acc[0] += (int)__builtin_cos(1.0);               // expected-no-diagnostics
-    acc[0] += (int)logf(1.0f);                       // expected-no-diagnostics
-    acc[0] += (int)log(1.0);                         // expected-no-diagnostics
-    acc[0] += (int)__builtin_truncf(1.0f);           // expected-no-diagnostics
-    acc[0] += (int)__builtin_trunc(1.0);             // expected-no-diagnostics
-    acc[0] += (int)__builtin_rintf(1.0f);            // expected-no-diagnostics
-    acc[0] += (int)__builtin_rint(1.0);              // expected-no-diagnostics
-    acc[0] += (int)__builtin_nearbyintf(0.5f);       // expected-no-diagnostics
-    acc[0] += (int)__builtin_nearbyint(0.5);         // expected-no-diagnostics
-    acc[0] += (int)__builtin_floorf(0.5f);           // expected-no-diagnostics
-    acc[0] += (int)__builtin_floor(0.5);             // expected-no-diagnostics
-    acc[0] += (int)__builtin_copysignf(1.0f, -0.5f); // expected-no-diagnostics
-    acc[0] += (int)__builtin_logf(1.0f);             // expected-no-diagnostics
-    acc[0] += (int)__builtin_log(1.0);               // expected-no-diagnostics
+    acc[0] += (int)truncf(1.0f);
+    acc[0] += (int)trunc(1.0);
+    acc[0] += (int)roundf(1.0f);
+    acc[0] += (int)round(1.0);
+    acc[0] += (int)rintf(1.0f);
+    acc[0] += (int)rint(1.0);
+    acc[0] += (int)nearbyintf(0.5f);
+    acc[0] += (int)nearbyint(0.5);
+    acc[0] += (int)floorf(0.5f);
+    acc[0] += (int)floor(0.5);
+    acc[0] += (int)copysignf(1.0f, -0.5f);
+    acc[0] += (int)copysign(1.0, -0.5);
+    acc[0] += (int)fminf(1.5f, 0.5f);
+    acc[0] += (int)fmin(1.5, 0.5);
+    acc[0] += (int)fmaxf(1.5f, 0.5f);
+    acc[0] += (int)fmax(1.5, 0.5);
+    acc[0] += (int)sinf(1.0f);
+    acc[0] += (int)sin(1.0);
+    acc[0] += (int)__builtin_sinf(1.0f);
+    acc[0] += (int)__builtin_sin(1.0);
+    acc[0] += (int)cosf(1.0f);
+    acc[0] += (int)cos(1.0);
+    acc[0] += (int)__builtin_cosf(1.0f);
+    acc[0] += (int)__builtin_cos(1.0);
+    acc[0] += (int)logf(1.0f);
+    acc[0] += (int)log(1.0);
+    acc[0] += (int)__builtin_truncf(1.0f);
+    acc[0] += (int)__builtin_trunc(1.0);
+    acc[0] += (int)__builtin_rintf(1.0f);
+    acc[0] += (int)__builtin_rint(1.0);
+    acc[0] += (int)__builtin_nearbyintf(0.5f);
+    acc[0] += (int)__builtin_nearbyint(0.5);
+    acc[0] += (int)__builtin_floorf(0.5f);
+    acc[0] += (int)__builtin_floor(0.5);
+    acc[0] += (int)__builtin_copysignf(1.0f, -0.5f);
+    acc[0] += (int)__builtin_fminf(1.5f, 0.5f);
+    acc[0] += (int)__builtin_fmin(1.5, 0.5);
+    acc[0] += (int)__builtin_fmaxf(1.5f, 0.5f);
+    acc[0] += (int)__builtin_fmax(1.5, 0.5);
+    acc[0] += (int)__builtin_logf(1.0f);
+    acc[0] += (int)__builtin_log(1.0);
+    acc[0] += __builtin_isinf(1.0);
+    acc[0] += __builtin_isfinite(1.0);
+    acc[0] += __builtin_isnormal(1.0);
+    acc[0] += __builtin_fpclassify(0, 1, 4, 3, 2, 1.0);
   });
   return 0;
 }

llvm/test/tools/sycl-post-link/ir-output-only.ll

@@ -0,0 +1,36 @@
+; RUN: sycl-post-link --ir-output-only -split=auto -S %s -o %t.ll
+; RUN: FileCheck %s -input-file=%t.ll
+
+; This test checks that the --ir-output-only option writes a LLVM IR
+; file instead of a table. In comparison with other tests, this one 
+; checks that the option works OK with -split=auto.
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir64-unknown-linux-sycldevice"
+
+declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+define dso_local spir_kernel void @kernel1() #0 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @kernel2() #0 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+attributes #0 = { "sycl-module-id"="a.cpp" }
+
+!llvm.module.flags = !{!0}
+!opencl.spir.version = !{!1}
+!spirv.Source = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, i32 2}
+!2 = !{i32 0, i32 100000}
+
+; CHECK: define dso_local spir_kernel void @kernel1()
+; CHECK: define dso_local spir_kernel void @kernel2()

llvm/test/tools/sycl-post-link/sycl-esimd/basic-sycl-esimd-split.ll

@@ -0,0 +1,45 @@
+; RUN: sycl-post-link -split-esimd -S %s -o %t.table
+; RUN: FileCheck %s -input-file=%t.table
+; RUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-SYCL-IR
+; RUN: FileCheck %s -input-file=%t_esimd_0.ll --check-prefixes CHECK-ESIMD-IR
+
+; This is basic test of splitting SYCL and ESIMD kernels into separate
+; modules.
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir64-unknown-linux-sycldevice"
+
+declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+define dso_local spir_kernel void @ESIMD_kernel() #0 !sycl_explicit_simd !3{
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @SYCL_kernel() #0 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+attributes #0 = { "sycl-module-id"="a.cpp" }
+
+!llvm.module.flags = !{!0}
+!opencl.spir.version = !{!1}
+!spirv.Source = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, i32 2}
+!2 = !{i32 0, i32 100000}
+!3 = !{}
+
+; CHECK: [Code|Properties]
+; CHECK: {{.*}}_0.ll|{{.*}}_0.prop
+; CHECK: {{.*}}_esimd_0.ll|{{.*}}_esimd_0.prop
+
+; CHECK-SYCL-IR-DAG: define dso_local spir_kernel void @SYCL_kernel()
+; CHECK-SYCL-IR-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-ESIMD-IR-DAG: define dso_local spir_kernel void @ESIMD_kernel()
+; CHECK-ESIMD-IR-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()

llvm/test/tools/sycl-post-link/sycl-esimd/no-sycl-esimd-split.ll

@@ -0,0 +1,55 @@
+; RUN: sycl-post-link -split=source -S %s -o %t.table
+; RUN: FileCheck %s -input-file=%t.table
+; RUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-IR-0
+; RUN: FileCheck %s -input-file=%t_1.ll --check-prefixes CHECK-IR-1
+
+; This test checks that if no '-split-esimd' provided, ther is no 
+; splitting of SYCL and ESIMD kernels into separate modules.
+; However, the rest of the splitting still happens according to
+; the '-split=' option.
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir64-unknown-linux-sycldevice"
+
+declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+define dso_local spir_kernel void @ESIMD_kernel() #0 !sycl_explicit_simd !3{
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @SYCL_kernel1() #0 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @SYCL_kernel2() #1 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+attributes #0 = { "sycl-module-id"="a.cpp" }
+attributes #1 = { "sycl-module-id"="b.cpp" }
+
+!llvm.module.flags = !{!0}
+!opencl.spir.version = !{!1}
+!spirv.Source = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, i32 2}
+!2 = !{i32 0, i32 100000}
+!3 = !{}
+
+; CHECK: [Code|Properties]
+; CHECK: {{.*}}_0.ll|{{.*}}_0.prop
+; CHECK: {{.*}}_1.ll|{{.*}}_1.prop
+
+; CHECK-IR-0-DAG: define dso_local spir_kernel void @SYCL_kernel1()
+; CHECK-IR-0-DAG: define dso_local spir_kernel void @ESIMD_kernel()
+; CHECK-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-IR-1-DAG: define dso_local spir_kernel void @SYCL_kernel2()
+; CHECK-IR-1-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()

llvm/test/tools/sycl-post-link/sycl-esimd/sycl-esimd-split-per-kernel.ll

@@ -0,0 +1,70 @@
+; RUN: sycl-post-link -split-esimd -split=kernel -S %s -o %t.table
+; RUN: FileCheck %s -input-file=%t.table
+; RUN: FileCheck %s -input-file=%t_0.ll --check-prefixes CHECK-SYCL-IR-0
+; RUN: FileCheck %s -input-file=%t_1.ll --check-prefixes CHECK-SYCL-IR-1
+; RUN: FileCheck %s -input-file=%t_esimd_0.ll --check-prefixes CHECK-ESIMD-IR-0
+; RUN: FileCheck %s -input-file=%t_esimd_1.ll --check-prefixes CHECK-ESIMD-IR-1
+
+; This test checks that after we split SYCL and ESIMD kernels into
+; separate modules, we split those two modules further according to
+; -split option. In this case we have 2 SYCL and 2 ESIMD kernels, which
+; are split into a total of 4 separate modules.
+
+target datalayout = "e-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024"
+target triple = "spir64-unknown-linux-sycldevice"
+
+declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+define dso_local spir_kernel void @ESIMD_kernel1() #0 !sycl_explicit_simd !3{
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @ESIMD_kernel2() #0 !sycl_explicit_simd !3{
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @SYCL_kernel1() #1 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+define dso_local spir_kernel void @SYCL_kernel2() #1 {
+entry:
+  %call = tail call spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+  ret void
+}
+
+attributes #0 = { "sycl-module-id"="a.cpp" }
+attributes #1 = { "sycl-module-id"="a.cpp" }
+
+!llvm.module.flags = !{!0}
+!opencl.spir.version = !{!1}
+!spirv.Source = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 1, i32 2}
+!2 = !{i32 0, i32 100000}
+!3 = !{}
+
+; CHECK: [Code|Properties]
+; CHECK: {{.*}}_0.ll|{{.*}}_0.prop
+; CHECK: {{.*}}_1.ll|{{.*}}_1.prop
+; CHECK: {{.*}}_esimd_0.ll|{{.*}}_esimd_0.prop
+; CHECK: {{.*}}_esimd_1.ll|{{.*}}_esimd_1.prop
+
+; CHECK-SYCL-IR-0-DAG: define dso_local spir_kernel void @SYCL_kernel1()
+; CHECK-SYCL-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-SYCL-IR-1-DAG: define dso_local spir_kernel void @SYCL_kernel2()
+; CHECK-SYCL-IR-1-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-ESIMD-IR-0-DAG: define dso_local spir_kernel void @ESIMD_kernel1()
+; CHECK-ESIMD-IR-0-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()
+
+; CHECK-ESIMD-IR-1-DAG: define dso_local spir_kernel void @ESIMD_kernel2()
+; CHECK-ESIMD-IR-1-DAG: declare dso_local spir_func i64 @_Z28__spirv_GlobalInvocationId_xv()