opencl-c.h: CL3.0 generic address space

[airlied]

Summary

This is one of the easier pieces of adding CL3.0 support.

[Anastasia]

clang/lib/Headers/opencl-c.h

7341

This one is not generic address space pointer though but I think this bit is part of an #else block? If you could just re-upload as a full diff, this will make it less confusing to review: https://llvm.org/docs/Phabricator.html#requesting-a-review-via-the-web-interface

[airlied]

updated to full diff.

[airlied]

clang/lib/Headers/opencl-c.h

7341

That's a good point, I was just adding the defined() bits to the endif to match the way other parts are done, however if there is an #else in the middle does this still make sense. I find having the defined comments is helpful for matching up wayward elses. The atomic changes are much more invasive so it would be good to know what we think is correct before I rework them all.

[airlied]

I've dropped the SPIR change in the base header until things are resolved for library SPIR users to disable extension defines.

[Anastasia]

In D105526#2863386, @airlied wrote:

I've dropped the SPIR change in the base header until things are resolved for library SPIR users to disable extension defines.

Makes sense! But this is not a full diff again. Would it be possible to reupload it?

[airlied]

full diff.

[Anastasia]

LGTM! Thanks!

PS this patch is straightforward and doesn't seem to affect functionality for older standards provided that we set __opencl_c_generic_address_space correctly which is not in the scope of this change.

clang/lib/Headers/opencl-c-base.h

 #if (__OPENCL_C_VERSION__ == 300)
 // For the SPIR target all features are supported.
 #if defined(__SPIR__)
+#define __opencl_c_generic_address_space 1
 #define __opencl_c_atomic_scope_all_devices 1
 #endif // defined(__SPIR__)
 #endif // (__OPENCL_C_VERSION__ == 300)

clang/lib/Headers/opencl-c.h

  * Returns fmin(x - floor (x), 0x1.fffffep-1f ).
  * floor(x) is returned in iptr.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld fract(float x, float *iptr);
 float2 __ovld fract(float2 x, float2 *iptr);
 float3 __ovld fract(float3 x, float3 *iptr);

 half8 __ovld fract(half8 x, __private half8 *iptr);
 half16 __ovld fract(half16 x, __private half16 *iptr);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Extract mantissa and exponent from x. For each

  * magnitude in the interval [1/2, 1) or 0. Each
  * component of x equals mantissa returned * 2^exp.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld frexp(float x, int *exp);
 float2 __ovld frexp(float2 x, int2 *exp);
 float3 __ovld frexp(float3 x, int3 *exp);

 half8 __ovld frexp(half8 x, __private int8 *exp);
 half16 __ovld frexp(half16 x, __private int16 *exp);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Compute the value of the square root of x^2 + y^2

 half16 __ovld __cnfn lgamma(half16 x);
 #endif //cl_khr_fp16
 
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld lgamma_r(float x, int *signp);
 float2 __ovld lgamma_r(float2 x, int2 *signp);
 float3 __ovld lgamma_r(float3 x, int3 *signp);

 half8 __ovld lgamma_r(half8 x, __private int8 *signp);
 half16 __ovld lgamma_r(half16 x, __private int16 *signp);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Compute natural logarithm.

  * the argument. It stores the integral part in the object
  * pointed to by iptr.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld modf(float x, float *iptr);
 float2 __ovld modf(float2 x, float2 *iptr);
 float3 __ovld modf(float3 x, float3 *iptr);

 half8 __ovld modf(half8 x, __private half8 *iptr);
 half16 __ovld modf(half16 x, __private half16 *iptr);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Returns a quiet NaN. The nancode may be placed

  * sign as x/y. It stores this signed value in the object
  * pointed to by quo.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld remquo(float x, float y, int *quo);
 float2 __ovld remquo(float2 x, float2 y, int2 *quo);
 float3 __ovld remquo(float3 x, float3 y, int3 *quo);

 half8 __ovld remquo(half8 x, half8 y, __private int8 *quo);
 half16 __ovld remquo(half16 x, half16 y, __private int16 *quo);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 /**
  * Round to integral value (using round to nearest
  * even rounding mode) in floating-point format.

  * is the return value and computed cosine is returned
  * in cosval.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld sincos(float x, float *cosval);
 float2 __ovld sincos(float2 x, float2 *cosval);
 float3 __ovld sincos(float3 x, float3 *cosval);

 half8 __ovld sincos(half8 x, __private half8 *cosval);
 half16 __ovld sincos(half16 x, __private half16 *cosval);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Compute hyperbolic sine.

 half16 __ovld vload16(size_t offset, const __constant half *p);
 #endif //cl_khr_fp16
 
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 char2 __ovld vload2(size_t offset, const char *p);
 uchar2 __ovld vload2(size_t offset, const uchar *p);
 short2 __ovld vload2(size_t offset, const short *p);

 half8 __ovld vload8(size_t offset, const __private half *p);
 half16 __ovld vload16(size_t offset, const __private half *p);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 void __ovld vstore2(char2 data, size_t offset, char *p);
 void __ovld vstore2(uchar2 data, size_t offset, uchar *p);
 void __ovld vstore2(short2 data, size_t offset, short *p);

 void __ovld vstore8(half8 data, size_t offset, __private half *p);
 void __ovld vstore16(half16 data, size_t offset, __private half *p);
 #endif //cl_khr_fp16
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Read sizeof (half) bytes of data from address

  * must be 16-bit aligned.
  */
 float __ovld vload_half(size_t offset, const __constant half *p);
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld vload_half(size_t offset, const half *p);
 #else
 float __ovld vload_half(size_t offset, const __global half *p);
 float __ovld vload_half(size_t offset, const __local half *p);
 float __ovld vload_half(size_t offset, const __private half *p);
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * Read sizeof (halfn) bytes of data from address

 float4 __ovld vload_half4(size_t offset, const __constant half *p);
 float8 __ovld vload_half8(size_t offset, const __constant half *p);
 float16 __ovld vload_half16(size_t offset, const __constant half *p);
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float2 __ovld vload_half2(size_t offset, const half *p);
 float3 __ovld vload_half3(size_t offset, const half *p);
 float4 __ovld vload_half4(size_t offset, const half *p);

 float4 __ovld vload_half4(size_t offset, const __private half *p);
 float8 __ovld vload_half8(size_t offset, const __private half *p);
 float16 __ovld vload_half16(size_t offset, const __private half *p);
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * The float value given by data is first

  * The default current rounding mode is round to
  * nearest even.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 void __ovld vstore_half(float data, size_t offset, half *p);
 void __ovld vstore_half_rte(float data, size_t offset, half *p);
 void __ovld vstore_half_rtz(float data, size_t offset, half *p);

 void __ovld vstore_half_rtp(double data, size_t offset, __private half *p);
 void __ovld vstore_half_rtn(double data, size_t offset, __private half *p);
 #endif //cl_khr_fp64
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * The floatn value given by data is converted to

  * The default current rounding mode is round to
  * nearest even.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 void __ovld vstore_half2(float2 data, size_t offset, half *p);
 void __ovld vstore_half3(float3 data, size_t offset, half *p);
 void __ovld vstore_half4(float4 data, size_t offset, half *p);

 void __ovld vstore_half8_rtn(double8 data, size_t offset, __private half *p);
 void __ovld vstore_half16_rtn(double16 data, size_t offset, __private half *p);
 #endif //cl_khr_fp64
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * For n = 1, 2, 4, 8 and 16 read sizeof (halfn)

 float4 __ovld vloada_half4(size_t offset, const __constant half *p);
 float8 __ovld vloada_half8(size_t offset, const __constant half *p);
 float16 __ovld vloada_half16(size_t offset, const __constant half *p);
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 float __ovld vloada_half(size_t offset, const half *p);
 float2 __ovld vloada_half2(size_t offset, const half *p);
 float3 __ovld vloada_half3(size_t offset, const half *p);

 float4 __ovld vloada_half4(size_t offset, const __private half *p);
 float8 __ovld vloada_half8(size_t offset, const __private half *p);
 float16 __ovld vloada_half16(size_t offset, const __private half *p);
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 /**
  * The floatn value given by data is converted to

  * mode. The default current rounding mode is
  * round to nearest even.
  */
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 void __ovld vstorea_half(float data, size_t offset, half *p);
 void __ovld vstorea_half2(float2 data, size_t offset, half *p);
 void __ovld vstorea_half3(float3 data, size_t offset, half *p);

 void __ovld vstorea_half8_rtn(double8 data,size_t offset, __private half *p);
 void __ovld vstorea_half16_rtn(double16 data,size_t offset, __private half *p);
 #endif //cl_khr_fp64
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 // OpenCL v1.1 s6.11.8, v1.2 s6.12.8, v2.0 s6.13.8 - Synchronization Functions
 

 
 // OpenCL v2.0 s6.13.9 - Address Space Qualifier Functions
 
-#if defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#if defined(__opencl_c_generic_address_space)
 cl_mem_fence_flags __ovld get_fence(const void *ptr);
 cl_mem_fence_flags __ovld get_fence(void *ptr);
 

  * where gentype is builtin type or user defined type.
  */
 
-#endif //defined(__OPENCL_CPP_VERSION__) || (__OPENCL_C_VERSION__ >= CL_VERSION_2_0)
+#endif //defined(__opencl_c_generic_address_space)
 
 // OpenCL v1.1 s6.11.10, v1.2 s6.12.10, v2.0 s6.13.10 - Async Copies from Global to Local Memory, Local to Global Memory, and Prefetch