[libclc] Add OpenCL atomic_*_explicit builtins (#168318)

Implement atomic_*_explicit (e.g. atomic_store_explicit) with
memory_order plus optional memory_scope.

OpenCL memory_order maps 1:1 to Clang (e.g. OpenCL memory_order_relaxed
== Clang __ATOMIC_RELAXED), so we pass it unchanged to clc_atomic_*
function which forwards to Clang _scoped_atomic* builtins.

Other changes:
* Add __opencl_get_clang_memory_scope helper in opencl/utils.h (OpenCL
scope -> Clang scope).
* Correct atomic_compare_exchange return type to bool.
* Fix atomic_compare_exchange to return true when value stored in the
pointer equals expected value.
* Remove volatile from CLC functions so that volatile isn't present in
LLVM IR.
* Add '-fdeclare-opencl-builtins -finclude-default-header' flag to
include
declaration of memory_scope. Some constants in libclc are already
provided
  by Clang’s OpenCL header; disable those in OpenCL library build and
  enable them only for CLC library build.
This commit is contained in:
Wenju He
2025-12-05 10:12:25 +08:00
committed by GitHub
parent 7f01d72108
commit 79620d8ac0
64 changed files with 367 additions and 340 deletions

View File

@@ -477,6 +477,7 @@ foreach( t ${LIBCLC_TARGETS_TO_BUILD} )
)
list( APPEND build_flags
-Xclang -fdeclare-opencl-builtins -Xclang -finclude-default-header
-I${CMAKE_CURRENT_SOURCE_DIR}/opencl/include
)

View File

@@ -15,24 +15,23 @@
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DECLARE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, \
int MemoryScope);
ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, int MemoryScope);
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DECLARE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL void __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
int MemoryOrder, int MemoryScope);
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, int MemoryOrder, \
int MemoryScope);
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DECLARE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
__CLC_GENTYPE Value, int MemoryOrderEqual, int MemoryOrderUnequal, \
int MemoryScope);
#else
#define __CLC_DECLARE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
int MemoryOrder, int MemoryScope);
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, int MemoryOrder, \
int MemoryScope);
#endif
__CLC_DECLARE_ATOMIC(global)

View File

@@ -6,9 +6,15 @@
//
//===----------------------------------------------------------------------===//
#ifndef MAXFLOAT
#define MAXFLOAT 0x1.fffffep127f
#endif
#ifndef HUGE_VALF
#define HUGE_VALF __builtin_huge_valf()
#endif
#ifndef INFINITY
#define INFINITY __builtin_inff()
#endif
#define FLT_DIG 6
#define FLT_MANT_DIG 24
@@ -17,33 +23,67 @@
#define FLT_MIN_10_EXP -37
#define FLT_MIN_EXP -125
#define FLT_RADIX 2
#ifndef FLT_MAX
#define FLT_MAX MAXFLOAT
#endif
#define FLT_MIN 0x1.0p-126f
#define FLT_EPSILON 0x1.0p-23f
#define FLT_NAN __builtin_nanf("")
#ifndef FP_ILOGB0
#define FP_ILOGB0 (-2147483647 - 1)
#endif
#ifndef FP_ILOGBNAN
#define FP_ILOGBNAN 2147483647
#endif
#ifndef M_E_F
#define M_E_F 0x1.5bf0a8p+1f
#endif
#ifndef M_LOG2E_F
#define M_LOG2E_F 0x1.715476p+0f
#endif
#ifndef M_LOG10E_F
#define M_LOG10E_F 0x1.bcb7b2p-2f
#endif
#ifndef M_LN2_F
#define M_LN2_F 0x1.62e430p-1f
#endif
#ifndef M_LN10_F
#define M_LN10_F 0x1.26bb1cp+1f
#endif
#ifndef M_PI_F
#define M_PI_F 0x1.921fb6p+1f
#endif
#ifndef M_PI_2_F
#define M_PI_2_F 0x1.921fb6p+0f
#endif
#ifndef M_PI_4_F
#define M_PI_4_F 0x1.921fb6p-1f
#endif
#ifndef M_1_PI_F
#define M_1_PI_F 0x1.45f306p-2f
#endif
#ifndef M_2_PI_F
#define M_2_PI_F 0x1.45f306p-1f
#endif
#ifndef M_2_SQRTPI_F
#define M_2_SQRTPI_F 0x1.20dd76p+0f
#endif
#ifndef M_SQRT2_F
#define M_SQRT2_F 0x1.6a09e6p+0f
#endif
#ifndef M_SQRT1_2_F
#define M_SQRT1_2_F 0x1.6a09e6p-1f
#endif
#define M_LOG210_F 0x1.a934f0p+1f
#ifdef cl_khr_fp64
#ifndef HUGE_VAL
#define HUGE_VAL __builtin_huge_val()
#endif
#define DBL_DIG 15
#define DBL_MANT_DIG 53
@@ -82,11 +122,19 @@
#define HALF_MIN_EXP -13
#define HALF_RADIX 2
#ifndef HALF_MAX
#define HALF_MAX 0x1.ffcp15h
#endif
#ifndef HALF_MIN
#define HALF_MIN 0x1.0p-14h
#endif
#ifndef HALF_EPSILON
#define HALF_EPSILON 0x1.0p-10h
#endif
#define HALF_NAN __builtin_nanf16("")
#ifndef M_LOG2E_H
#define M_LOG2E_H 0x1.714p+0h
#endif
#endif

View File

@@ -11,15 +11,23 @@
#define CHAR_BIT 8
#define INT_MAX 2147483647
#ifndef INT_MIN
#define INT_MIN (-2147483647 - 1)
#endif
#define LONG_MAX 0x7fffffffffffffffL
#ifndef LONG_MIN
#define LONG_MIN (-0x7fffffffffffffffL - 1)
#endif
#define CHAR_MAX SCHAR_MAX
#define CHAR_MIN SCHAR_MIN
#define SCHAR_MAX 127
#ifndef SCHAR_MIN
#define SCHAR_MIN (-127 - 1)
#endif
#define SHRT_MAX 32767
#ifndef SHRT_MIN
#define SHRT_MIN (-32767 - 1)
#endif
#define UCHAR_MAX 255
#define UCHAR_MIN 0
#define USHRT_MAX 65535

View File

@@ -25,7 +25,7 @@
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_atomic_compare_exchange( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
__CLC_GENTYPE Value, int MemoryOrderEqual, int MemoryOrderUnequal, \
int MemoryScope) { \
__CLC_U_GENTYPE Comp = __CLC_AS_U_GENTYPE(Comparator); \
@@ -39,7 +39,7 @@
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __clc_atomic_compare_exchange( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Comparator, \
__CLC_GENTYPE Value, int MemoryOrderEqual, int MemoryOrderUnequal, \
int MemoryScope) { \
__scoped_atomic_compare_exchange_n(Ptr, &Comparator, Value, false, \

View File

@@ -36,32 +36,30 @@
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, \
int MemoryScope) { \
ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, int MemoryScope) { \
return __CLC_AS_RETTYPE(__CLC_IMPL_FUNCTION( \
(ADDRSPACE __CLC_CASTTYPE *)Ptr, MemoryOrder, MemoryScope)); \
}
#elif defined(__CLC_INC_DEC)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, \
int MemoryScope) { \
ADDRSPACE __CLC_GENTYPE *Ptr, int MemoryOrder, int MemoryScope) { \
return __CLC_IMPL_FUNCTION(Ptr, (__CLC_U_GENTYPE)(-1), MemoryOrder, \
MemoryScope); \
}
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF void __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
int MemoryOrder, int MemoryScope) { \
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, int MemoryOrder, \
int MemoryScope) { \
__CLC_IMPL_FUNCTION((ADDRSPACE __CLC_CASTTYPE *)Ptr, \
__CLC_AS_CASTTYPE(Value), MemoryOrder, MemoryScope); \
}
#else
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
int MemoryOrder, int MemoryScope) { \
ADDRSPACE __CLC_GENTYPE *Ptr, __CLC_GENTYPE Value, int MemoryOrder, \
int MemoryScope) { \
return __CLC_AS_RETTYPE(__CLC_IMPL_FUNCTION( \
(ADDRSPACE __CLC_CASTTYPE *)Ptr, __CLC_AS_CASTTYPE(Value), \
MemoryOrder, MemoryScope)); \

View File

@@ -29,9 +29,13 @@ function(compile_to_bc)
if( NOT ${FILE_EXT} STREQUAL ".ll" )
# Pass '-c' when not running the preprocessor
set( PP_OPTS -c )
set( EXTRA_OPTS ${ARG_EXTRA_OPTS} )
else()
set( PP_OPTS -E;-P )
set( TMP_SUFFIX .tmp )
string( REPLACE "-Xclang;-fdeclare-opencl-builtins;-Xclang;-finclude-default-header"
"" EXTRA_OPTS "${ARG_EXTRA_OPTS}"
)
endif()
set( TARGET_ARG )
@@ -48,7 +52,7 @@ function(compile_to_bc)
COMMAND ${clang_exe}
${TARGET_ARG}
${PP_OPTS}
${ARG_EXTRA_OPTS}
${EXTRA_OPTS}
-MD -MF ${ARG_OUTPUT}.d -MT ${ARG_OUTPUT}${TMP_SUFFIX}
# LLVM 13 enables standard includes by default - we don't want
# those when pre-processing IR. We disable it unconditionally.

View File

@@ -1,92 +0,0 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef __CLC_OPENCL_AS_TYPE_H__
#define __CLC_OPENCL_AS_TYPE_H__
#include <clc/opencl/opencl-base.h>
#define as_char(x) __builtin_astype(x, char)
#define as_uchar(x) __builtin_astype(x, uchar)
#define as_short(x) __builtin_astype(x, short)
#define as_ushort(x) __builtin_astype(x, ushort)
#define as_int(x) __builtin_astype(x, int)
#define as_uint(x) __builtin_astype(x, uint)
#define as_long(x) __builtin_astype(x, long)
#define as_ulong(x) __builtin_astype(x, ulong)
#define as_float(x) __builtin_astype(x, float)
#define as_char2(x) __builtin_astype(x, char2)
#define as_uchar2(x) __builtin_astype(x, uchar2)
#define as_short2(x) __builtin_astype(x, short2)
#define as_ushort2(x) __builtin_astype(x, ushort2)
#define as_int2(x) __builtin_astype(x, int2)
#define as_uint2(x) __builtin_astype(x, uint2)
#define as_long2(x) __builtin_astype(x, long2)
#define as_ulong2(x) __builtin_astype(x, ulong2)
#define as_float2(x) __builtin_astype(x, float2)
#define as_char3(x) __builtin_astype(x, char3)
#define as_uchar3(x) __builtin_astype(x, uchar3)
#define as_short3(x) __builtin_astype(x, short3)
#define as_ushort3(x) __builtin_astype(x, ushort3)
#define as_int3(x) __builtin_astype(x, int3)
#define as_uint3(x) __builtin_astype(x, uint3)
#define as_long3(x) __builtin_astype(x, long3)
#define as_ulong3(x) __builtin_astype(x, ulong3)
#define as_float3(x) __builtin_astype(x, float3)
#define as_char4(x) __builtin_astype(x, char4)
#define as_uchar4(x) __builtin_astype(x, uchar4)
#define as_short4(x) __builtin_astype(x, short4)
#define as_ushort4(x) __builtin_astype(x, ushort4)
#define as_int4(x) __builtin_astype(x, int4)
#define as_uint4(x) __builtin_astype(x, uint4)
#define as_long4(x) __builtin_astype(x, long4)
#define as_ulong4(x) __builtin_astype(x, ulong4)
#define as_float4(x) __builtin_astype(x, float4)
#define as_char8(x) __builtin_astype(x, char8)
#define as_uchar8(x) __builtin_astype(x, uchar8)
#define as_short8(x) __builtin_astype(x, short8)
#define as_ushort8(x) __builtin_astype(x, ushort8)
#define as_int8(x) __builtin_astype(x, int8)
#define as_uint8(x) __builtin_astype(x, uint8)
#define as_long8(x) __builtin_astype(x, long8)
#define as_ulong8(x) __builtin_astype(x, ulong8)
#define as_float8(x) __builtin_astype(x, float8)
#define as_char16(x) __builtin_astype(x, char16)
#define as_uchar16(x) __builtin_astype(x, uchar16)
#define as_short16(x) __builtin_astype(x, short16)
#define as_ushort16(x) __builtin_astype(x, ushort16)
#define as_int16(x) __builtin_astype(x, int16)
#define as_uint16(x) __builtin_astype(x, uint16)
#define as_long16(x) __builtin_astype(x, long16)
#define as_ulong16(x) __builtin_astype(x, ulong16)
#define as_float16(x) __builtin_astype(x, float16)
#ifdef cl_khr_fp64
#define as_double(x) __builtin_astype(x, double)
#define as_double2(x) __builtin_astype(x, double2)
#define as_double3(x) __builtin_astype(x, double3)
#define as_double4(x) __builtin_astype(x, double4)
#define as_double8(x) __builtin_astype(x, double8)
#define as_double16(x) __builtin_astype(x, double16)
#endif
#ifdef cl_khr_fp16
#define as_half(x) __builtin_astype(x, half)
#define as_half2(x) __builtin_astype(x, half2)
#define as_half3(x) __builtin_astype(x, half3)
#define as_half4(x) __builtin_astype(x, half4)
#define as_half8(x) __builtin_astype(x, half8)
#define as_half16(x) __builtin_astype(x, half16)
#endif
#endif // __CLC_OPENCL_AS_TYPE_H__

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_COMPARE_EXCHANGE_STRONG_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_COMPARE_EXCHANGE_STRONG_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_compare_exchange_strong
#define __CLC_COMPARE_EXCHANGE

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_COMPARE_EXCHANGE_WEAK_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_COMPARE_EXCHANGE_WEAK_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_compare_exchange_weak
#define __CLC_COMPARE_EXCHANGE

View File

@@ -25,32 +25,105 @@
#define __CLC_ATOMIC_GENTYPE __CLC_XCONCAT(atomic_, __CLC_GENTYPE)
#define __CLC_FUNCTION_EXPLICIT __CLC_XCONCAT(__CLC_FUNCTION, _explicit)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, memory_order Order, \
memory_scope Scope);
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL void __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order, memory_scope Scope);
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL bool __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired, \
memory_order Order, memory_scope Scope);
#else
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order, memory_scope Scope);
#endif
__CLC_DECL_ATOMIC(global)
__CLC_DECL_ATOMIC(local)
#if _CLC_GENERIC_AS_SUPPORTED
__CLC_DECL_ATOMIC()
#endif
#undef __CLC_DECL_ATOMIC
#if defined(__opencl_c_atomic_scope_device)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, memory_order Order);
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL void __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order);
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL bool __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired, \
memory_order Success, memory_order Failure);
#else
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order);
#endif
__CLC_DECL_ATOMIC(global)
__CLC_DECL_ATOMIC(local)
#if _CLC_GENERIC_AS_SUPPORTED
__CLC_DECL_ATOMIC()
#endif
#undef __CLC_DECL_ATOMIC
#endif // defined(__opencl_c_atomic_scope_device)
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr);
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL void __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value);
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL bool __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired);
#else
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
#define __CLC_DECL_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value);
#endif
__CLC_DEFINE_ATOMIC(global)
__CLC_DEFINE_ATOMIC(local)
__CLC_DECL_ATOMIC(global)
__CLC_DECL_ATOMIC(local)
#if _CLC_GENERIC_AS_SUPPORTED
__CLC_DEFINE_ATOMIC()
__CLC_DECL_ATOMIC()
#endif
#undef __CLC_DEFINE_ATOMIC
#undef __CLC_DECL_ATOMIC
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)
#endif // __CLC_HAVE_FP_ATOMIC || __CLC_HAVE_INT_ATOMIC

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_EXCHANGE_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_EXCHANGE_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_exchange
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_ADD_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_ADD_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_add
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_AND_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_AND_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_and
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_MAX_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_MAX_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_max
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_MIN_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_MIN_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_min
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_OR_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_OR_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_or
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_SUB_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_SUB_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_sub
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_XOR_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_FETCH_XOR_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_fetch_xor
#define __CLC_BODY <clc/opencl/atomic/atomic_decl.inc>

View File

@@ -10,7 +10,6 @@
#define __CLC_OPENCL_ATOMIC_ATOMIC_FLAG_CLEAR_H__
#include <clc/opencl/opencl-base.h>
#include <clc/opencl/types.h>
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)

View File

@@ -10,7 +10,6 @@
#define __CLC_OPENCL_ATOMIC_ATOMIC_FLAG_TEST_AND_SET_H__
#include <clc/opencl/opencl-base.h>
#include <clc/opencl/types.h>
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_LOAD_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_LOAD_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_load
#define __CLC_NO_VALUE_ARG

View File

@@ -9,6 +9,8 @@
#ifndef __CLC_OPENCL_ATOMIC_ATOMIC_STORE_H__
#define __CLC_OPENCL_ATOMIC_ATOMIC_STORE_H__
#include <clc/opencl/opencl-base.h>
#define __CLC_FUNCTION atomic_store
#define __CLC_RETURN_VOID

View File

@@ -10,7 +10,6 @@
#define __CLC_OPENCL_EXPLICIT_FENCE_EXPLICIT_MEMORY_FENCE_H__
#include <clc/opencl/opencl-base.h>
#include <clc/opencl/synchronization/cl_mem_fence_flags.h>
_CLC_DECL _CLC_OVERLOAD void mem_fence(cl_mem_fence_flags flags);
_CLC_DECL _CLC_OVERLOAD void read_mem_fence(cl_mem_fence_flags flags);

View File

@@ -10,7 +10,6 @@
#define __CLC_OPENCL_SYNCHRONIZATION_BARRIER_H__
#include <clc/opencl/opencl-base.h>
#include <clc/opencl/synchronization/cl_mem_fence_flags.h>
_CLC_DECL _CLC_OVERLOAD void barrier(cl_mem_fence_flags flags);

View File

@@ -1,18 +0,0 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef __CLC_OPENCL_SYNCHRONIZATION_CL_MEM_FENCE_FLAGS_H__
#define __CLC_OPENCL_SYNCHRONIZATION_CL_MEM_FENCE_FLAGS_H__
typedef uint cl_mem_fence_flags;
#define CLK_LOCAL_MEM_FENCE 1
#define CLK_GLOBAL_MEM_FENCE 2
#define CLK_IMAGE_MEM_FENCE 4
#endif // __CLC_OPENCL_SYNCHRONIZATION_CL_MEM_FENCE_FLAGS_H__

View File

@@ -11,7 +11,6 @@
#include <clc/internal/clc.h>
#include <clc/mem_fence/clc_mem_semantic.h>
#include <clc/opencl/synchronization/cl_mem_fence_flags.h>
static _CLC_INLINE int __opencl_get_memory_scope(cl_mem_fence_flags flag) {
if (flag & CLK_GLOBAL_MEM_FENCE)

View File

@@ -1,48 +0,0 @@
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#ifndef __CLC_OPENCL_TYPES_H__
#define __CLC_OPENCL_TYPES_H__
// Copied from clang/lib/Headers/opencl-c-base.h
typedef enum memory_scope {
memory_scope_work_item = __OPENCL_MEMORY_SCOPE_WORK_ITEM,
memory_scope_work_group = __OPENCL_MEMORY_SCOPE_WORK_GROUP,
memory_scope_device = __OPENCL_MEMORY_SCOPE_DEVICE,
#if defined(__opencl_c_atomic_scope_all_devices)
memory_scope_all_svm_devices = __OPENCL_MEMORY_SCOPE_ALL_SVM_DEVICES,
#if (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >= 202100)
memory_scope_all_devices = memory_scope_all_svm_devices,
#endif // (__OPENCL_C_VERSION__ >= CL_VERSION_3_0 || __OPENCL_CPP_VERSION__ >=
// 202100)
#endif // defined(__opencl_c_atomic_scope_all_devices)
/**
* Subgroups have different requirements on forward progress, so just test
* all the relevant macros.
* CL 3.0 sub-groups "they are not guaranteed to make independent forward
* progress" KHR subgroups "Subgroups within a workgroup are independent, make
* forward progress with respect to each other"
*/
#if defined(cl_intel_subgroups) || defined(cl_khr_subgroups) || \
defined(__opencl_c_subgroups)
memory_scope_sub_group = __OPENCL_MEMORY_SCOPE_SUB_GROUP
#endif
} memory_scope;
typedef enum memory_order {
memory_order_relaxed = __ATOMIC_RELAXED,
memory_order_acquire = __ATOMIC_ACQUIRE,
memory_order_release = __ATOMIC_RELEASE,
memory_order_acq_rel = __ATOMIC_ACQ_REL,
#if defined(__opencl_c_atomic_order_seq_cst)
memory_order_seq_cst = __ATOMIC_SEQ_CST
#endif
} memory_order;
#endif // __CLC_OPENCL_TYPES_H__

View File

@@ -10,9 +10,7 @@
#define __CLC_OPENCL_UTILS_H__
#include <clc/internal/clc.h>
#include <clc/opencl/types.h>
// INTEL_FEATURE_PISA
static _CLC_INLINE int __opencl_get_clang_memory_scope(memory_scope scope) {
switch (scope) {
case __OPENCL_MEMORY_SCOPE_WORK_ITEM:
@@ -30,6 +28,5 @@ static _CLC_INLINE int __opencl_get_clang_memory_scope(memory_scope scope) {
return __MEMORY_SCOPE_SYSTEM;
}
}
// end INTEL_FEATURE_PISA
#endif // __CLC_OPENCL_UTILS_H__

View File

@@ -7,7 +7,6 @@
//===----------------------------------------------------------------------===//
#include <clc/float/definitions.h>
#include <clc/opencl/as_type.h>
#include <clc/opencl/math/copysign.h>
#include <clc/opencl/math/fabs.h>
#include <clc/opencl/math/nextafter.h>

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_add(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_add(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_add(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_add(AS TYPE *p, TYPE val) { \
return atom_add((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_add(volatile AS TYPE *p, TYPE val) { \
return atom_add((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_and(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_and(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_and(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_and(AS TYPE *p, TYPE val) { \
return atom_and((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_and(volatile AS TYPE *p, TYPE val) { \
return atom_and((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_extended_atomics

View File

@@ -12,14 +12,14 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_cmpxchg(volatile AS TYPE *p, TYPE cmp, \
TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_cmpxchg(AS TYPE *p, TYPE cmp, TYPE val) { \
return __clc_atomic_compare_exchange(p, cmp, val, __ATOMIC_RELAXED, \
__ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_cmpxchg(AS TYPE *p, TYPE cmp, TYPE val) { \
return atom_cmpxchg((volatile AS TYPE *)p, cmp, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_cmpxchg(volatile AS TYPE *p, TYPE cmp, \
TYPE val) { \
return atom_cmpxchg((AS TYPE *)p, cmp, val); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,11 +12,11 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_dec(volatile AS TYPE *p) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_dec(AS TYPE *p) { \
return __clc_atomic_dec(p, __ATOMIC_RELAXED, __MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_dec(AS TYPE *p) { \
return atom_dec((volatile AS TYPE *)p); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_dec(volatile AS TYPE *p) { \
return atom_dec((AS TYPE *)p); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,11 +12,11 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_inc(volatile AS TYPE *p) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_inc(AS TYPE *p) { \
return __clc_atomic_inc(p, __ATOMIC_RELAXED, __MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_inc(AS TYPE *p) { \
return atom_inc((volatile AS TYPE *)p); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_inc(volatile AS TYPE *p) { \
return atom_inc((AS TYPE *)p); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_max(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_max(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_max(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_max(AS TYPE *p, TYPE val) { \
return atom_max((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_max(volatile AS TYPE *p, TYPE val) { \
return atom_max((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_extended_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_min(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_min(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_min(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_min(AS TYPE *p, TYPE val) { \
return atom_min((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_min(volatile AS TYPE *p, TYPE val) { \
return atom_min((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_extended_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_or(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_or(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_or(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_or(AS TYPE *p, TYPE val) { \
return atom_or((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_or(volatile AS TYPE *p, TYPE val) { \
return atom_or((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_extended_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_sub(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_sub(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_sub(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_sub(AS TYPE *p, TYPE val) { \
return atom_sub((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_sub(volatile AS TYPE *p, TYPE val) { \
return atom_sub((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xchg(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xchg(AS TYPE *p, TYPE val) { \
return __clc_atomic_exchange(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xchg(AS TYPE *p, TYPE val) { \
return atom_xchg((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xchg(volatile AS TYPE *p, TYPE val) { \
return atom_xchg((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_base_atomics

View File

@@ -12,12 +12,12 @@
// Non-volatile overloads are for backward compatibility with OpenCL 1.0.
#define __CLC_IMPL(AS, TYPE) \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xor(volatile AS TYPE *p, TYPE val) { \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xor(AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_xor(p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
} \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xor(AS TYPE *p, TYPE val) { \
return atom_xor((volatile AS TYPE *)p, val); \
_CLC_OVERLOAD _CLC_DEF TYPE atom_xor(volatile AS TYPE *p, TYPE val) { \
return atom_xor((AS TYPE *)p, val); \
}
#ifdef cl_khr_global_int32_extended_atomics

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_add(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_add(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_add((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_and(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_and(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_and((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -12,8 +12,8 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_cmpxchg(volatile AS TYPE *p, TYPE cmp, \
TYPE val) { \
return __clc_atomic_compare_exchange(p, cmp, val, __ATOMIC_RELAXED, \
__ATOMIC_RELAXED, \
return __clc_atomic_compare_exchange((AS TYPE *)p, cmp, val, \
__ATOMIC_RELAXED, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -6,13 +6,12 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_compare_exchange.h>
#include <clc/opencl/atomic/atomic_compare_exchange_strong.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_compare_exchange_strong
#define __CLC_IMPL_FUNCTION __clc_atomic_compare_exchange
#define __CLC_COMPARE_EXCHANGE
#define __CLC_BODY <atomic_def.inc>
@@ -20,6 +19,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,13 +6,12 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_compare_exchange.h>
#include <clc/opencl/atomic/atomic_compare_exchange_weak.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_compare_exchange_weak
#define __CLC_IMPL_FUNCTION __clc_atomic_compare_exchange
#define __CLC_COMPARE_EXCHANGE
#define __CLC_BODY <atomic_def.inc>
@@ -20,6 +19,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -25,41 +25,47 @@
#define __CLC_ATOMIC_GENTYPE __CLC_XCONCAT(atomic_, __CLC_GENTYPE)
#define __CLC_FUNCTION_EXPLICIT __CLC_XCONCAT(__CLC_FUNCTION, _explicit)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr) { \
return __CLC_IMPL_FUNCTION((volatile ADDRSPACE __CLC_GENTYPE *)Ptr, \
__ATOMIC_SEQ_CST, __MEMORY_SCOPE_DEVICE); \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, memory_order Order, \
memory_scope Scope) { \
return __CLC_IMPL_FUNCTION((ADDRSPACE __CLC_GENTYPE *)Ptr, Order, \
__opencl_get_clang_memory_scope(Scope)); \
}
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF void __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value) { \
__CLC_IMPL_FUNCTION((volatile ADDRSPACE __CLC_GENTYPE *)Ptr, Value, \
__ATOMIC_SEQ_CST, __MEMORY_SCOPE_DEVICE); \
_CLC_OVERLOAD _CLC_DEF void __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order, memory_scope Scope) { \
__CLC_IMPL_FUNCTION((ADDRSPACE __CLC_GENTYPE *)Ptr, Value, Order, \
__opencl_get_clang_memory_scope(Scope)); \
}
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
_CLC_OVERLOAD _CLC_DEF bool __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired) { \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired, \
memory_order Success, memory_order Failure, memory_scope Scope) { \
__CLC_GENTYPE Comparator = *Expected; \
__CLC_GENTYPE RetValue = __clc_atomic_compare_exchange( \
(volatile ADDRSPACE __CLC_GENTYPE *)Ptr, Comparator, Desired, \
__ATOMIC_SEQ_CST, __ATOMIC_RELAXED, __MEMORY_SCOPE_DEVICE); \
__CLC_GENTYPE RetValue = __CLC_IMPL_FUNCTION( \
(ADDRSPACE __CLC_GENTYPE *)Ptr, Comparator, Desired, Success, Failure, \
__opencl_get_clang_memory_scope(Scope)); \
if (Comparator != RetValue) { \
*Expected = RetValue; \
return true; \
return false; \
} \
return false; \
return true; \
}
#else
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value) { \
return __CLC_IMPL_FUNCTION((volatile ADDRSPACE __CLC_GENTYPE *)Ptr, Value, \
__ATOMIC_SEQ_CST, __MEMORY_SCOPE_DEVICE); \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order, memory_scope Scope) { \
return __CLC_IMPL_FUNCTION((ADDRSPACE __CLC_GENTYPE *)Ptr, Value, Order, \
__opencl_get_clang_memory_scope(Scope)); \
}
#endif
@@ -71,6 +77,96 @@ __CLC_DEFINE_ATOMIC()
#undef __CLC_DEFINE_ATOMIC
#if defined(__opencl_c_atomic_scope_device)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, memory_order Order) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, Order, __OPENCL_MEMORY_SCOPE_DEVICE); \
}
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF void __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order) { \
__CLC_FUNCTION_EXPLICIT(Ptr, Value, Order, __OPENCL_MEMORY_SCOPE_DEVICE); \
}
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF bool __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired, \
memory_order Success, memory_order Failure) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, Expected, Desired, Success, Failure, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#else
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION_EXPLICIT( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value, \
memory_order Order) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, Value, Order, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#endif
__CLC_DEFINE_ATOMIC(global)
__CLC_DEFINE_ATOMIC(local)
#if _CLC_GENERIC_AS_SUPPORTED
__CLC_DEFINE_ATOMIC()
#endif
#undef __CLC_DEFINE_ATOMIC
#endif // defined(__opencl_c_atomic_scope_device)
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#ifdef __CLC_NO_VALUE_ARG
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, __ATOMIC_SEQ_CST, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#elif defined(__CLC_RETURN_VOID)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF void __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value) { \
__CLC_FUNCTION_EXPLICIT(Ptr, Value, __ATOMIC_SEQ_CST, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#elif defined(__CLC_COMPARE_EXCHANGE)
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF bool __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, \
ADDRSPACE __CLC_GENTYPE *Expected, __CLC_GENTYPE Desired) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, Expected, Desired, __ATOMIC_SEQ_CST, \
__ATOMIC_SEQ_CST, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#else
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_ATOMIC_GENTYPE *Ptr, __CLC_GENTYPE Value) { \
return __CLC_FUNCTION_EXPLICIT(Ptr, Value, __ATOMIC_SEQ_CST, \
__OPENCL_MEMORY_SCOPE_DEVICE); \
}
#endif
__CLC_DEFINE_ATOMIC(global)
__CLC_DEFINE_ATOMIC(local)
#if _CLC_GENERIC_AS_SUPPORTED
__CLC_DEFINE_ATOMIC()
#endif
#undef __CLC_DEFINE_ATOMIC
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)
#endif // __CLC_HAVE_FP_ATOMIC || __CLC_HAVE_INT_ATOMIC
#undef __CLC_HAVE_INT_ATOMIC

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_exchange.h>
#include <clc/opencl/atomic/atomic_exchange.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_exchange
#define __CLC_IMPL_FUNCTION __clc_atomic_exchange
@@ -20,6 +18,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_add.h>
#include <clc/opencl/atomic/atomic_fetch_add.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_add
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_add
@@ -20,6 +18,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,17 +6,12 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_and.h>
#include <clc/opencl/atomic/atomic_fetch_and.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_and
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_and
#define __CLC_BODY <atomic_def.inc>
#include <clc/integer/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_max.h>
#include <clc/opencl/atomic/atomic_fetch_max.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_max
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_max
@@ -20,6 +18,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_min.h>
#include <clc/opencl/atomic/atomic_fetch_min.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_min
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_min
@@ -20,6 +18,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,17 +6,12 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_or.h>
#include <clc/opencl/atomic/atomic_fetch_or.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_or
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_or
#define __CLC_BODY <atomic_def.inc>
#include <clc/integer/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_sub.h>
#include <clc/opencl/atomic/atomic_fetch_sub.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_sub
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_sub
@@ -20,6 +18,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -6,17 +6,12 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_fetch_xor.h>
#include <clc/opencl/atomic/atomic_fetch_xor.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_fetch_xor
#define __CLC_IMPL_FUNCTION __clc_atomic_fetch_xor
#define __CLC_BODY <atomic_def.inc>
#include <clc/integer/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -13,7 +13,8 @@
#define __CLC_DEFINE_ATOMIC(ADDRSPACE) \
_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE __CLC_FUNCTION( \
volatile ADDRSPACE __CLC_GENTYPE *Ptr) { \
return __CLC_IMPL_FUNCTION(Ptr, __ATOMIC_RELAXED, __MEMORY_SCOPE_DEVICE); \
return __CLC_IMPL_FUNCTION((ADDRSPACE __CLC_GENTYPE *)Ptr, \
__ATOMIC_RELAXED, __MEMORY_SCOPE_DEVICE); \
}
__CLC_DEFINE_ATOMIC(global)

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_load.h>
#include <clc/opencl/atomic/atomic_load.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_load
#define __CLC_IMPL_FUNCTION __clc_atomic_load
@@ -21,6 +19,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -9,14 +9,14 @@
#include <clc/atomic/clc_atomic_fetch_max.h>
#include <clc/opencl/atomic/atomic_max.h>
#define __CLC_IMPL(TYPE, AS, OP) \
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_max(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_max(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_max((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}
__CLC_IMPL(int, global, max)
__CLC_IMPL(unsigned int, global, umax)
__CLC_IMPL(int, local, max)
__CLC_IMPL(unsigned int, local, umax)
__CLC_IMPL(int, global)
__CLC_IMPL(unsigned int, global)
__CLC_IMPL(int, local)
__CLC_IMPL(unsigned int, local)
#undef __CLC_IMPL

View File

@@ -9,14 +9,14 @@
#include <clc/atomic/clc_atomic_fetch_min.h>
#include <clc/opencl/atomic/atomic_min.h>
#define __CLC_IMPL(TYPE, AS, OP) \
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_min(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_min(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_min((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}
__CLC_IMPL(int, global, min)
__CLC_IMPL(unsigned int, global, umin)
__CLC_IMPL(int, local, min)
__CLC_IMPL(unsigned int, local, umin)
__CLC_IMPL(int, global)
__CLC_IMPL(unsigned int, global)
__CLC_IMPL(int, local)
__CLC_IMPL(unsigned int, local)
#undef __CLC_IMPL

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_or(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_or(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_or((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -6,11 +6,9 @@
//
//===----------------------------------------------------------------------===//
#if defined(__opencl_c_atomic_order_seq_cst) && \
defined(__opencl_c_atomic_scope_device)
#include <clc/atomic/clc_atomic_store.h>
#include <clc/opencl/atomic/atomic_store.h>
#include <clc/opencl/utils.h>
#define __CLC_FUNCTION atomic_store
#define __CLC_IMPL_FUNCTION __clc_atomic_store
@@ -21,6 +19,3 @@
#define __CLC_BODY <atomic_def.inc>
#include <clc/math/gentype.inc>
#endif // defined(__opencl_c_atomic_order_seq_cst) &&
// defined(__opencl_c_atomic_scope_device)

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_sub(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_sub(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_sub((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_xchg(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_exchange(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_exchange((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}

View File

@@ -11,7 +11,7 @@
#define __CLC_IMPL(TYPE, AS) \
_CLC_OVERLOAD _CLC_DEF TYPE atomic_xor(volatile AS TYPE *p, TYPE val) { \
return __clc_atomic_fetch_xor(p, val, __ATOMIC_RELAXED, \
return __clc_atomic_fetch_xor((AS TYPE *)p, val, __ATOMIC_RELAXED, \
__MEMORY_SCOPE_DEVICE); \
}