Avinash Kethineedi
c4de6833f6
* Update copyright information and add SPDX license identifier
* Update AUTHORS
* Remove `sos_tests`
[ROCm/rocshmem commit: f6ef19f5a9]
640 строки
21 KiB
Python
640 строки
21 KiB
Python
###############################################################################
|
|
# Copyright (c) Advanced Micro Devices, Inc. All rights reserved.
|
|
#
|
|
# SPDX-License-Identifier: MIT
|
|
#
|
|
# 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.
|
|
###############################################################################
|
|
|
|
import os
|
|
|
|
types = [
|
|
("int", "int"),
|
|
("long", "long"),
|
|
("long long", "longlong"),
|
|
("unsigned int", "uint"),
|
|
("unsigned long", "ulong"),
|
|
("unsigned long long", "ulonglong"),
|
|
("int32_t", "int32"),
|
|
("int64_t", "int64"),
|
|
("uint32_t", "uint32"),
|
|
("uint64_t", "uint64"),
|
|
("size_t", "size"),
|
|
("ptrdiff_t", "ptrdiff"),
|
|
]
|
|
|
|
|
|
float_types = [
|
|
("float", "float"),
|
|
("double", "double"),
|
|
]
|
|
|
|
bitwise_types = types[3:10]
|
|
|
|
|
|
def atomic_fetch_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *source, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch(\n"
|
|
f" {T} *source, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *source, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch(\n"
|
|
f" {T} *source, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH
|
|
* @brief Atomically return the value of \p dest to the calling PE.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return The value of \p dest.
|
|
*/\n"""
|
|
|
|
for type_, tname_ in float_types:
|
|
expanded_code += atomic_fetch_api(type_, tname_)
|
|
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_fetch_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_set_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_set(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_set(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_set(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_set(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_set_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_SET
|
|
* @brief Atomically set the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
|
|
for type_, tname_ in float_types:
|
|
expanded_code += atomic_set_api(type_, tname_)
|
|
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_set_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_compare_swap_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_compare_swap(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} cond, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_compare_swap(\n"
|
|
f" {T} *dest, {T} cond, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_compare_swap(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} cond, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_compare_swap(\n"
|
|
f" {T} *dest, {T} cond, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_compare_swap_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_COMPARE_SWAP
|
|
* @brief Atomically compares if the value in \p dest with \p cond is equal
|
|
* then put \p val in \p dest. The operation returns the older value of \p dest
|
|
* to the calling PE.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] cond The value to be compare with.
|
|
* @param[in] val The value to be atomically swapped.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return The old value of \p dest.
|
|
*/\n"""
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_compare_swap_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_swap_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_swap(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_swap(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_swap(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_swap(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_swap_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_SWAP
|
|
* @brief Atomically swap the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return original value
|
|
*/\n"""
|
|
|
|
for type_, tname_ in float_types:
|
|
expanded_code += atomic_swap_api(type_, tname_)
|
|
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_swap_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_fetch_inc_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch_inc(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch_inc(\n"
|
|
f" {T} *dest, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch_inc(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch_inc(\n"
|
|
f" {T} *dest, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_inc_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH_INC
|
|
* @brief Atomically add 1 to \p dest on \p pe. The operation
|
|
* returns the older value of \p dest to the calling PE.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return The old value of \p dest before it was incremented by 1.
|
|
*/\n"""
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_fetch_inc_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_inc_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_inc(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_inc(\n"
|
|
f" {T} *dest, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_inc(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_inc(\n"
|
|
f" {T} *dest, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_inc_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_INC
|
|
* @brief Atomically add 1 to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_inc_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_fetch_add_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch_add(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch_add(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch_add(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch_add(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_add_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH_ADD
|
|
* @brief Atomically add the value \p val to \p dest on \p pe. The operation
|
|
* returns the older value of \p dest to the calling PE.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return The old value of \p dest before the \p val was added.
|
|
*/\n"""
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_fetch_add_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_add_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_add(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_add(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_add(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_add(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_add_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_ADD
|
|
* @brief Atomically add the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
for type_, tname_ in types:
|
|
expanded_code += atomic_add_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_fetch_and_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch_and(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch_and(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch_and(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch_and(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_and_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH_AND
|
|
* @brief Atomically bitwise-and the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return original value
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_fetch_and_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_and_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_and(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_and(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_and(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_and(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_and_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_AND
|
|
* @brief Atomically bitwise-and the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_and_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_fetch_or_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch_or(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch_or(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch_or(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch_or(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_or_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH_OR
|
|
* @brief Atomically bitwise-or the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return original value
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_fetch_or_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_or_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_or(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_or(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_or(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_or(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_or_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_OR
|
|
* @brief Atomically bitwise-or the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_or_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_fetch_xor_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_ctx_{TNAME}_atomic_fetch_xor(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE {T} rocshmem_{TNAME}_atomic_fetch_xor(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_ctx_{TNAME}_atomic_fetch_xor(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ {T} rocshmem_{TNAME}_atomic_fetch_xor(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_fetch_xor_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_FETCH_XOR
|
|
* @brief Atomically bitwise-xor the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return original value
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_fetch_xor_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
|
|
def atomic_xor_api(T, TNAME):
|
|
return (
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_ctx_{TNAME}_atomic_xor(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__device__ ATTR_NO_INLINE void rocshmem_{TNAME}_atomic_xor(\n"
|
|
f" {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_ctx_{TNAME}_atomic_xor(\n"
|
|
f" rocshmem_ctx_t ctx, {T} *dest, {T} value, int pe);\n"
|
|
f"__host__ void rocshmem_{TNAME}_atomic_xor(\n"
|
|
f" {T} *dest, {T} value, int pe);\n\n"
|
|
)
|
|
|
|
|
|
def generate_atomic_xor_api():
|
|
expanded_code = """
|
|
/**
|
|
* @name SHMEM_ATOMIC_XOR
|
|
* @brief Atomically bitwise-xor the value \p val to \p dest on \p pe.
|
|
*
|
|
* The operation is blocking.
|
|
*
|
|
* This function can be called from divergent control paths at per-thread
|
|
* granularity.
|
|
*
|
|
* @param[in] ctx Context with which to perform this operation.
|
|
* @param[in] dest Destination address. Must be an address on the symmetric
|
|
heap.
|
|
* @param[in] val The value to be atomically added.
|
|
* @param[in] pe PE of the remote process.
|
|
*
|
|
* @return void
|
|
*/\n"""
|
|
for type_, tname_ in bitwise_types:
|
|
expanded_code += atomic_xor_api(type_, tname_)
|
|
|
|
return expanded_code
|
|
|
|
def write_to_file(filename, content):
|
|
with open(filename, 'w') as file:
|
|
file.write(content)
|
|
|
|
|
|
def generate_AMO_header(output_dir, copyright):
|
|
expanded_code = copyright
|
|
|
|
expanded_code += """
|
|
#ifndef LIBRARY_INCLUDE_ROCSHMEM_AMO_HPP
|
|
#define LIBRARY_INCLUDE_ROCSHMEM_AMO_HPP
|
|
|
|
namespace rocshmem {
|
|
"""
|
|
|
|
expanded_code += (
|
|
generate_atomic_fetch_api() +
|
|
generate_atomic_set_api() +
|
|
generate_atomic_compare_swap_api() +
|
|
generate_atomic_swap_api() +
|
|
generate_atomic_fetch_inc_api() +
|
|
generate_atomic_inc_api() +
|
|
generate_atomic_fetch_add_api() +
|
|
generate_atomic_add_api() +
|
|
generate_atomic_fetch_and_api() +
|
|
generate_atomic_and_api() +
|
|
generate_atomic_fetch_or_api() +
|
|
generate_atomic_or_api() +
|
|
generate_atomic_fetch_xor_api() +
|
|
generate_atomic_xor_api()
|
|
)
|
|
|
|
expanded_code += """
|
|
} // namespace rocshmem
|
|
|
|
#endif // LIBRARY_INCLUDE_ROCSHMEM_AMO_HPP
|
|
"""
|
|
|
|
output_file = os.path.join(
|
|
output_dir, 'rocshmem_AMO.hpp'
|
|
)
|
|
|
|
write_to_file(output_file, expanded_code)
|