[SYCL] Allow gcc asm statements in kernel code.

Signed-off-by: Premanand M Rao <[email protected]>

Author: premanandrao

clang/lib/Sema/SemaStmtAsm.cpp

@@ -256,16 +256,6 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
   // The parser verifies that there is a string literal here.
   assert(AsmString->isAscii());
 
-  // Skip all the checks if we are compiling SYCL device code, but the function
-  // is not marked to be used on device, this code won't be codegen'ed anyway.
-  if (getLangOpts().SYCLIsDevice) {
-    SYCLDiagIfDeviceCode(AsmLoc, diag::err_sycl_restrict) << KernelUseAssembly;
-    return new (Context)
-        GCCAsmStmt(Context, AsmLoc, IsSimple, IsVolatile, NumOutputs, NumInputs,
-                   Names, Constraints, Exprs.data(), AsmString, NumClobbers,
-                   Clobbers, NumLabels, RParenLoc);
-  }
-
   FunctionDecl *FD = dyn_cast<FunctionDecl>(getCurLexicalContext());
   llvm::StringMap<bool> FeatureMap;
   Context.getFunctionFeatureMap(FeatureMap, FD);
@@ -902,9 +892,6 @@ StmtResult Sema::ActOnMSAsmStmt(SourceLocation AsmLoc, SourceLocation LBraceLoc,
                                 SourceLocation EndLoc) {
   bool IsSimple = (NumOutputs != 0 || NumInputs != 0);
   setFunctionHasBranchProtectedScope();
-  if (getLangOpts().SYCLIsDevice)
-    SYCLDiagIfDeviceCode(AsmLoc, diag::err_sycl_restrict)
-        << KernelUseAssembly;
   MSAsmStmt *NS =
     new (Context) MSAsmStmt(Context, AsmLoc, LBraceLoc, IsSimple,
                             /*IsVolatile*/ true, AsmToks, NumOutputs, NumInputs,

clang/test/CodeGenSYCL/inline_asm.cpp

@@ -0,0 +1,34 @@
+// RUN: %clang_cc1 -fsycl -fsycl-is-device -triple spir64-unknown-linux-sycldevice -S -emit-llvm -x c++ %s -o - | FileCheck %s
+
+inline namespace cl {
+  namespace sycl {
+    class kernel {};
+  }
+}
+
+using namespace cl::sycl;
+
+template <typename name, typename Func>
+__attribute__((sycl_kernel)) void kernel_single_task(Func kernelFunc) {
+  // CHECK: %[[ARRAY_A:[0-9a-z]+]] = alloca [100 x i32], align 4
+  // CHECK-NEXT: %0 = bitcast [100 x i32]* %[[ARRAY_A]] to i8*
+  // CHECK-NEXT: call void @llvm.lifetime.start.p0i8(i64 400, i8* nonnull %0) #2
+  // CHECK-NEXT: %[[IDX:.*]] = getelementptr inbounds [100 x i32], [100 x i32]* %[[ARRAY_A]], i64 0, i64 0
+  int a[100], i = 0;
+  // CHECK-NEXT: call void asm sideeffect 
+  // CHECK: ".decl V52 v_type=G type=d num_elts=16 align=GRF
+  // CHECK: svm_gather.4.1 (M1, 16) $0.0 V52.0
+  // CHECK: add(M1, 16) V52(0, 0)<1> V52(0, 0)<1; 1, 0> 0x1
+  // CHECK: svm_scatter.4.1 (M1, 16) $0.0 V52.0", 
+  // CHECK: "rw"(i32* nonnull %[[IDX]])
+  asm volatile(".decl V52 v_type=G type=d num_elts=16 align=GRF\n"
+               "svm_gather.4.1 (M1, 16) %0.0 V52.0\n"
+               "add(M1, 16) V52(0, 0)<1> V52(0, 0)<1; 1, 0> 0x1\n"
+               "svm_scatter.4.1 (M1, 16) %0.0 V52.0"  
+               : : "rw"(&a[i]));      
+}
+
+int main() {
+  kernel_single_task<class kernel>([]() {});
+  return 0;
+}

clang/test/SemaSYCL/inline-asm.cpp

@@ -1,6 +1,8 @@
 // RUN: %clang_cc1 -fsycl -fsycl-is-device -fsyntax-only -verify %s -DLINUX_ASM
 // RUN: %clang_cc1 -fsycl -fsycl-is-device -fsyntax-only -verify -triple x86_64-windows -fasm-blocks %s
 
+// expected-no-diagnostics
+
 void foo() {
   int a;
 #ifdef LINUX_ASM
@@ -13,26 +15,24 @@ void foo() {
 void bar() {
   int a;
 #ifdef LINUX_ASM
-   __asm__("int3");  // expected-error {{SYCL kernel cannot use inline assembly}}
+  __asm__("int3");
 #else
-   __asm int 3 // expected-error {{SYCL kernel cannot use inline assembly}}
+  __asm int 3
 #endif // LINUX_ASM
 }
 
 template <typename Name, typename Func>
 __attribute__((sycl_kernel)) void kernel_single_task(Func kernelFunc) {
-  // expected-note@+1 {{called by 'kernel_single_task<fake_kernel, (lambda}}
   kernelFunc();
 #ifdef LINUX_ASM
-   __asm__("int3");  // expected-error {{SYCL kernel cannot use inline assembly}}
+  __asm__("int3");
 #else
-   __asm int 3 // expected-error {{SYCL kernel cannot use inline assembly}}
+  __asm int 3
 #endif // LINUX_ASM
 }
 
 int main() {
   foo();
-  // expected-note@+1 {{called by 'operator()'}}
   kernel_single_task<class fake_kernel>([]() { bar(); });
   return 0;
 }