[OpenCL] Add cl_khr_subgroup_ballot to TableGen BIFs

[svenvh]

Summary

Add the builtin functions brought by the cl_khr_subgroup_ballot
extension to -fdeclare-opencl-builtins.

Also add placeholder comments for the other Extended Subgroup
Functions from the OpenCL Extension Specification.

I have verified (semi-manually) that this adds the same ballot functions
that are also in opencl-c.h already.

[svenvh]

Add test after rebasing on dependent commit (D95616).

[Anastasia]

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

38

Not related to this patch - I think we should have included opencl-c-base.h by default when passing -fdeclare-opencl-builtins or do you see any issues with that?

144

I appreciate we haven't addressed the standard header testing holistically yet and this functionality was added in opencl-c.h without testing the functions, but would it be better to test each function at least here?

Even though the final goal should be testing all overloads, this is outside of the scope of this work that is closing the gap between opencl-c.h and this experimental function declaration mechanism.

[svenvh]

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

38

That's something we probably want to do in the future indeed.

144

but would it be better to test each function at least here?

I think that defeats the purpose of this test. This test is meant to test the basic functionality of -fdeclare-opencl-builtins. It is not a completeness test.

A completeness test shouldn't be tied to -fdeclare-opencl-builtins alone; such a test should also be run against opencl-c.h, so this would be the wrong test to turn into a completeness test in my opinion.

[Anastasia]

LGTM! Thanks!

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

144

Ok, this test does test functionality selectively but I wouldn't be able to tell how one selects what to test. Perhaps a comment would help for the future development...

[svenvh]

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

144

There are no formal selection criteria here I think. The test aims to cover the different functional aspects of -fdeclare-opencl-builtins. The new functional aspect for -fdeclare-opencl-builtins in this patch is an extension that is not in OpenCLExtensions.def, so it makes sense to add such a case to the test.

I will add a comment to the top of the test file.

[Anastasia]

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

144

Makes sense! Thanks!

clang/lib/Sema/OpenCLBuiltins.td

 }
 
 // Extension associated to a builtin function.
 class FunctionExtension<string _Ext> : AbstractExtension<_Ext>;
 
 // FunctionExtension definitions.
 def FuncExtNone                          : FunctionExtension<"">;
 def FuncExtKhrSubgroups                  : FunctionExtension<"cl_khr_subgroups">;
+def FuncExtKhrSubgroupBallot             : FunctionExtension<"cl_khr_subgroup_ballot">;
 def FuncExtKhrGlobalInt32BaseAtomics     : FunctionExtension<"cl_khr_global_int32_base_atomics">;
 def FuncExtKhrGlobalInt32ExtendedAtomics : FunctionExtension<"cl_khr_global_int32_extended_atomics">;
 def FuncExtKhrLocalInt32BaseAtomics      : FunctionExtension<"cl_khr_local_int32_base_atomics">;
 def FuncExtKhrLocalInt32ExtendedAtomics  : FunctionExtension<"cl_khr_local_int32_extended_atomics">;
 def FuncExtKhrInt64BaseAtomics           : FunctionExtension<"cl_khr_int64_base_atomics">;
 def FuncExtKhrInt64ExtendedAtomics       : FunctionExtension<"cl_khr_int64_extended_atomics">;
 def FuncExtKhrMipmapImage                : FunctionExtension<"cl_khr_mipmap_image">;
 def FuncExtKhrMipmapImageWrites          : FunctionExtension<"cl_khr_mipmap_image_writes">;

 // uchar abs(uchar).
 def TLAllUIntsTwice : TypeList<[UChar, UChar, UShort, UShort, UInt, UInt, ULong, ULong]>;
 
 def TLAllInts       : TypeList<[Char, UChar, Short, UShort, Int, UInt, Long, ULong]>;
 
 // GenType definitions for multiple base types (e.g. all floating point types,
 // or all integer types).
 // All types
+def AGenType1              : GenericType<"AGenType1", TLAll, Vec1>;
 def AGenTypeN              : GenericType<"AGenTypeN", TLAll, VecAndScalar>;
 def AGenTypeNNoScalar      : GenericType<"AGenTypeNNoScalar", TLAll, VecNoScalar>;
 // All integer
 def AIGenType1             : GenericType<"AIGenType1", TLAllInts, Vec1>;
 def AIGenTypeN             : GenericType<"AIGenTypeN", TLAllInts, VecAndScalar>;
 def AIGenTypeNNoScalar     : GenericType<"AIGenTypeNNoScalar", TLAllInts, VecNoScalar>;
 // All integer to unsigned
 def AI2UGenTypeN           : GenericType<"AI2UGenTypeN", TLAllUIntsTwice, VecAndScalar>;
let Extension = FuncExtKhrSubgroups in {
   foreach name = ["sub_group_reduce_", "sub_group_scan_exclusive_",
                   "sub_group_scan_inclusive_"] in {
     foreach op = ["add", "min", "max"] in {
       def : Builtin<name # op, [IntLongFloatGenType1, IntLongFloatGenType1], Attr.Convergent>;
     }
   }
 }
 
+// OpenCL Extension v3.0 s38 - Extended Subgroup Functions
+
+// Section 38.4.1 - cl_khr_subgroup_extended_types
+// TODO
+
+// Section 38.5.1 - cl_khr_subgroup_non_uniform_vote
+// TODO
+
+// Section 38.6.1 - cl_khr_subgroup_ballot
+let Extension = FuncExtKhrSubgroupBallot in {
+  def : Builtin<"sub_group_non_uniform_broadcast", [AGenTypeN, AGenTypeN, UInt]>;
+  def : Builtin<"sub_group_broadcast_first", [AGenType1, AGenType1]>;
+  def : Builtin<"sub_group_ballot", [VectorType<UInt, 4>, Int]>;
+  def : Builtin<"sub_group_inverse_ballot", [Int, VectorType<UInt, 4>], Attr.Const>;
+  def : Builtin<"sub_group_ballot_bit_extract", [Int, VectorType<UInt, 4>, UInt], Attr.Const>;
+  def : Builtin<"sub_group_ballot_bit_count", [UInt, VectorType<UInt, 4>], Attr.Const>;
+  def : Builtin<"sub_group_ballot_inclusive_scan", [UInt, VectorType<UInt, 4>]>;
+  def : Builtin<"sub_group_ballot_exclusive_scan", [UInt, VectorType<UInt, 4>]>;
+  def : Builtin<"sub_group_ballot_find_lsb", [UInt, VectorType<UInt, 4>]>;
+  def : Builtin<"sub_group_ballot_find_msb", [UInt, VectorType<UInt, 4>]>;
+
+  foreach op = ["eq", "ge", "gt", "le", "lt"] in {
+    def : Builtin<"get_sub_group_" # op # "_mask", [VectorType<UInt, 4>], Attr.Const>;
+  }
+}
+
+// Section 38.7.1 - cl_khr_subgroup_non_uniform_arithmetic
+// TODO
+
+// Section 38.8.1 - cl_khr_subgroup_shuffle
+// TODO
+
+// Section 38.9.1 - cl_khr_subgroup_shuffle_relative
+// TODO
+
+// Section 38.10.1 - cl_khr_subgroup_clustered_reduce
+// TODO
+
 //--------------------------------------------------------------------
 // Arm extensions.
 let Extension = ArmIntegerDotProductInt8 in {
   foreach name = ["arm_dot"] in {
     def : Builtin<name, [UInt, VectorType<UChar, 4>, VectorType<UChar, 4>]>;
     def : Builtin<name, [Int, VectorType<Char, 4>, VectorType<Char, 4>]>;
   }
 }

clang/test/SemaOpenCL/fdeclare-opencl-builtins.cl

 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL -fdeclare-opencl-builtins -DNO_HEADER
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL -fdeclare-opencl-builtins -finclude-default-header
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL1.2 -fdeclare-opencl-builtins -DNO_HEADER
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL1.2 -fdeclare-opencl-builtins -finclude-default-header
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL2.0 -fdeclare-opencl-builtins -DNO_HEADER
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CL2.0 -fdeclare-opencl-builtins -finclude-default-header
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CLC++ -fdeclare-opencl-builtins -DNO_HEADER
 // RUN: %clang_cc1 %s -triple spir -verify -pedantic -Wconversion -Werror -fsyntax-only -cl-std=CLC++ -fdeclare-opencl-builtins -finclude-default-header
 
-// Test the -fdeclare-opencl-builtins option.
+// Test the -fdeclare-opencl-builtins option.  This is not a completeness
+// test, so it should not test for all builtins defined by OpenCL.  Instead
+// this test should cover different functional aspects of the TableGen builtin
+// function machinery.
 
 #pragma OPENCL EXTENSION cl_khr_fp16 : enable
 #if __OPENCL_C_VERSION__ < CL_VERSION_1_2
 #pragma OPENCL EXTENSION cl_khr_fp64 : enable
 #endif
 
 // Provide typedefs when invoking clang without -finclude-default-header.
 #ifdef NO_HEADER
 typedef unsigned char uchar;
 typedef unsigned int uint;
 typedef unsigned long ulong;
 typedef unsigned short ushort;
 typedef __SIZE_TYPE__ size_t;
 typedef char char2 __attribute__((ext_vector_type(2)));
 typedef char char4 __attribute__((ext_vector_type(4)));
 typedef uchar uchar4 __attribute__((ext_vector_type(4)));
 typedef float float4 __attribute__((ext_vector_type(4)));
 typedef half half4 __attribute__((ext_vector_type(4)));
 typedef int int2 __attribute__((ext_vector_type(2)));
 typedef int int4 __attribute__((ext_vector_type(4)));
 typedef uint uint4 __attribute__((ext_vector_type(4)));
 typedef long long2 __attribute__((ext_vector_type(2)));
+
+// Enable extensions that are enabled in opencl-c-base.h.
+#if (defined(__OPENCL_CPP_VERSION__) || __OPENCL_C_VERSION__ >= 200)
+#define cl_khr_subgroup_ballot 1
+#endif
 #endif
 
 kernel void test_pointers(volatile global void *global_p, global const int4 *a) {
   int i;
   unsigned int ui;
 
   prefetch(a, 2);
 

 kernel void basic_subgroup(global uint *out) {
   out[0] = get_sub_group_size();
 #if __OPENCL_C_VERSION__ <= CL_VERSION_1_2 && !defined(__OPENCL_CPP_VERSION__)
   // expected-error@-2{{implicit declaration of function 'get_sub_group_size' is invalid in OpenCL}}
   // expected-error@-3{{implicit conversion changes signedness}}
 #endif
 }
 
+kernel void extended_subgroup(global uint4 *out) {
+  out[0] = get_sub_group_eq_mask();
+#if __OPENCL_C_VERSION__ < CL_VERSION_2_0 && !defined(__OPENCL_CPP_VERSION__)
+  // expected-error@-2{{implicit declaration of function 'get_sub_group_eq_mask' is invalid in OpenCL}}
+  // expected-error@-3{{implicit conversion changes signedness}}
+#endif
+}
+
 kernel void basic_vector_data() {
 #if __OPENCL_C_VERSION__ >= CL_VERSION_2_0
   generic void *generic_p;
 #endif
   constant void *constant_p;
   local void *local_p;
   global void *global_p;
   private void *private_p;