[cuda] Added support for CUDA built-in variables.

Added cuda_builtin_vars.h which implements built-in CUDA variables
using __declattr(property).

Fields of built-in variables (except for warpSize) are implemented
using __declattr(property) which replaces read/write of a member field
with a call to a getter/setter member function, in this case with
appropriate NVPTX builtin.

Added a test case to check diagnostics on attempt to construct or
improperly access a built-in variable.

Differential Revision: http://reviews.llvm.org/D9064

llvm-svn: 235398

4 files changed, 196 insertions, 0 deletions

diff --git a/clang/lib/Headers/CMakeLists.txt b/clang/lib/Headers/CMakeLists.txt
index 54290925b10..75fc03540dc 100644
--- a/clang/lib/Headers/CMakeLists.txt
+++ b/clang/lib/Headers/CMakeLists.txt
@@ -13,6 +13,7 @@ set(files
   bmi2intrin.h
   bmiintrin.h
   cpuid.h
+  cuda/cuda_builtin_vars.h
   emmintrin.h
   f16cintrin.h
   float.h
diff --git a/clang/lib/Headers/cuda/cuda_builtin_vars.h b/clang/lib/Headers/cuda/cuda_builtin_vars.h
new file mode 100644
index 00000000000..901356b3d5c
--- /dev/null
+++ b/clang/lib/Headers/cuda/cuda_builtin_vars.h
@@ -0,0 +1,110 @@
+/*===---- cuda_builtin_vars.h - CUDA built-in variables ---------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CUDA_BUILTIN_VARS_H
+#define __CUDA_BUILTIN_VARS_H
+
+// The file implements built-in CUDA variables using __declspec(property).
+// https://msdn.microsoft.com/en-us/library/yhfk0thd.aspx
+// All read accesses of built-in variable fields get converted into calls to a
+// getter function which in turn would call appropriate builtin to fetch the
+// value.
+//
+// Example:
+//    int x = threadIdx.x;
+// IR output:
+//  %0 = call i32 @llvm.ptx.read.tid.x() #3
+// PTX output:
+//  mov.u32     %r2, %tid.x;
+
+#define __CUDA_DEVICE_BUILTIN(FIELD, INTRINSIC)                                \
+  __declspec(property(get = __fetch_builtin_##FIELD)) unsigned int FIELD;      \
+  static inline __attribute__((always_inline))                                 \
+      __attribute__((device)) unsigned int __fetch_builtin_##FIELD(void) {     \
+    return INTRINSIC;                                                          \
+  }
+
+#if __cplusplus >= 201103L
+#define __DELETE =delete
+#else
+#define __DELETE
+#endif
+
+// Make sure nobody can create instances of the special varible types.  nvcc
+// also disallows taking address of special variables, so we disable address-of
+// operator as well.
+#define __CUDA_DISALLOW_BUILTINVAR_ACCESS(TypeName)                            \
+  __attribute__((device)) TypeName() __DELETE;                                 \
+  __attribute__((device)) TypeName(const TypeName &) __DELETE;                 \
+  __attribute__((device)) void operator=(const TypeName &) const __DELETE;     \
+  __attribute__((device)) TypeName *operator&() const __DELETE
+
+struct __cuda_builtin_threadIdx_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_tid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_tid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_tid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_threadIdx_t);
+};
+
+struct __cuda_builtin_blockIdx_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_ctaid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_ctaid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_ctaid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockIdx_t);
+};
+
+struct __cuda_builtin_blockDim_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_ntid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_ntid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_ntid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockDim_t);
+};
+
+struct __cuda_builtin_gridDim_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_nctaid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_nctaid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_nctaid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_gridDim_t);
+};
+
+#define __CUDA_BUILTIN_VAR                                                     \
+  extern const __attribute__((device)) __attribute__((weak))
+__CUDA_BUILTIN_VAR __cuda_builtin_threadIdx_t threadIdx;
+__CUDA_BUILTIN_VAR __cuda_builtin_blockIdx_t blockIdx;
+__CUDA_BUILTIN_VAR __cuda_builtin_blockDim_t blockDim;
+__CUDA_BUILTIN_VAR __cuda_builtin_gridDim_t gridDim;
+
+// warpSize should translate to read of %WARP_SZ but there's currently no
+// builtin to do so. According to PTX v4.2 docs 'to date, all target
+// architectures have a WARP_SZ value of 32'.
+__attribute__((device)) const int warpSize = 32;
+
+#undef __CUDA_DEVICE_BUILTIN
+#undef __CUDA_BUILTIN_VAR
+#undef __CUDA_DISALLOW_BUILTINVAR_ACCESS
+
+#endif /* __CUDA_BUILTIN_VARS_H */
diff --git a/clang/test/CodeGenCUDA/cuda-builtin-vars.cu b/clang/test/CodeGenCUDA/cuda-builtin-vars.cu
new file mode 100644
index 00000000000..8d6ea24ccee
--- /dev/null
+++ b/clang/test/CodeGenCUDA/cuda-builtin-vars.cu
@@ -0,0 +1,28 @@
+// RUN: %clang_cc1 "-triple" "nvptx-nvidia-cuda" -emit-llvm -fcuda-is-device -o - %s | FileCheck %s
+
+#include "cuda/cuda_builtin_vars.h"
+
+// CHECK: define void @_Z6kernelPi(i32* %out)
+__attribute__((global))
+void kernel(int *out) {
+  int i = 0;
+  out[i++] = threadIdx.x; // CHECK: call i32 @llvm.ptx.read.tid.x()
+  out[i++] = threadIdx.y; // CHECK: call i32 @llvm.ptx.read.tid.y()
+  out[i++] = threadIdx.z; // CHECK: call i32 @llvm.ptx.read.tid.z()
+
+  out[i++] = blockIdx.x; // CHECK: call i32 @llvm.ptx.read.ctaid.x()
+  out[i++] = blockIdx.y; // CHECK: call i32 @llvm.ptx.read.ctaid.y()
+  out[i++] = blockIdx.z; // CHECK: call i32 @llvm.ptx.read.ctaid.z()
+
+  out[i++] = blockDim.x; // CHECK: call i32 @llvm.ptx.read.ntid.x()
+  out[i++] = blockDim.y; // CHECK: call i32 @llvm.ptx.read.ntid.y()
+  out[i++] = blockDim.z; // CHECK: call i32 @llvm.ptx.read.ntid.z()
+
+  out[i++] = gridDim.x; // CHECK: call i32 @llvm.ptx.read.nctaid.x()
+  out[i++] = gridDim.y; // CHECK: call i32 @llvm.ptx.read.nctaid.y()
+  out[i++] = gridDim.z; // CHECK: call i32 @llvm.ptx.read.nctaid.z()
+
+  out[i++] = warpSize; // CHECK: store i32 32,
+
+  // CHECK: ret void
+}
diff --git a/clang/test/SemaCUDA/cuda-builtin-vars.cu b/clang/test/SemaCUDA/cuda-builtin-vars.cu
new file mode 100644
index 00000000000..eda54ebd098
--- /dev/null
+++ b/clang/test/SemaCUDA/cuda-builtin-vars.cu
@@ -0,0 +1,57 @@
+// RUN: %clang_cc1 "-triple" "nvptx-nvidia-cuda" -fcuda-is-device -fsyntax-only -verify %s
+
+#include "cuda/cuda_builtin_vars.h"
+__attribute__((global))
+void kernel(int *out) {
+  int i = 0;
+  out[i++] = threadIdx.x;
+  threadIdx.x = 0; // expected-error {{no setter defined for property 'x'}}
+  out[i++] = threadIdx.y;
+  threadIdx.y = 0; // expected-error {{no setter defined for property 'y'}}
+  out[i++] = threadIdx.z;
+  threadIdx.z = 0; // expected-error {{no setter defined for property 'z'}}
+
+  out[i++] = blockIdx.x;
+  blockIdx.x = 0; // expected-error {{no setter defined for property 'x'}}
+  out[i++] = blockIdx.y;
+  blockIdx.y = 0; // expected-error {{no setter defined for property 'y'}}
+  out[i++] = blockIdx.z;
+  blockIdx.z = 0; // expected-error {{no setter defined for property 'z'}}
+
+  out[i++] = blockDim.x;
+  blockDim.x = 0; // expected-error {{no setter defined for property 'x'}}
+  out[i++] = blockDim.y;
+  blockDim.y = 0; // expected-error {{no setter defined for property 'y'}}
+  out[i++] = blockDim.z;
+  blockDim.z = 0; // expected-error {{no setter defined for property 'z'}}
+
+  out[i++] = gridDim.x;
+  gridDim.x = 0; // expected-error {{no setter defined for property 'x'}}
+  out[i++] = gridDim.y;
+  gridDim.y = 0; // expected-error {{no setter defined for property 'y'}}
+  out[i++] = gridDim.z;
+  gridDim.z = 0; // expected-error {{no setter defined for property 'z'}}
+
+  out[i++] = warpSize;
+  warpSize = 0; // expected-error {{cannot assign to variable 'warpSize' with const-qualified type 'const int'}}
+  // expected-note@cuda/cuda_builtin_vars.h:104 {{variable 'warpSize' declared const here}}
+
+  // Make sure we can't construct or assign to the special variables.
+  __cuda_builtin_threadIdx_t x; // expected-error {{calling a private constructor of class '__cuda_builtin_threadIdx_t'}}
+  // expected-note@cuda/cuda_builtin_vars.h:67 {{declared private here}}
+
+  __cuda_builtin_threadIdx_t y = threadIdx; // expected-error {{calling a private constructor of class '__cuda_builtin_threadIdx_t'}}
+  // expected-note@cuda/cuda_builtin_vars.h:67 {{declared private here}}
+
+  threadIdx = threadIdx; // expected-error {{'operator=' is a private member of '__cuda_builtin_threadIdx_t'}}
+  // expected-note@cuda/cuda_builtin_vars.h:67 {{declared private here}}
+
+  void *ptr = &threadIdx; // expected-error {{'operator&' is a private member of '__cuda_builtin_threadIdx_t'}}
+  // expected-note@cuda/cuda_builtin_vars.h:67 {{declared private here}}
+
+  // Following line should've caused an error as one is not allowed to
+  // take address of a built-in variable in CUDA. Alas there's no way
+  // to prevent getting address of a 'const int', so the line
+  // currently compiles without errors or warnings.
+  const void *wsptr = &warpSize;
+}