triton-windows 3.2.0.post19__cp39-cp39-win_amd64.whl → 3.2.0.post21__cp39-cp39-win_amd64.whl

triton/backends/amd/include/hip/amd_detail/amd_warp_functions.h

@@ -1,494 +0,0 @@
-/*
-Copyright (c) 2022 - 2023 Advanced Micro Devices, Inc. All rights reserved.
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in
-all copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
-THE SOFTWARE.
-*/
-
-#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_WARP_FUNCTIONS_H
-#define HIP_INCLUDE_HIP_AMD_DETAIL_WARP_FUNCTIONS_H
-
-__device__ static inline unsigned __hip_ds_bpermute(int index, unsigned src) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = src;
-    tmp.i = __builtin_amdgcn_ds_bpermute(index, tmp.i);
-    return tmp.u;
-}
-
-__device__ static inline float __hip_ds_bpermutef(int index, float src) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = src;
-    tmp.i = __builtin_amdgcn_ds_bpermute(index, tmp.i);
-    return tmp.f;
-}
-
-__device__ static inline unsigned __hip_ds_permute(int index, unsigned src) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = src;
-    tmp.i = __builtin_amdgcn_ds_permute(index, tmp.i);
-    return tmp.u;
-}
-
-__device__ static inline float __hip_ds_permutef(int index, float src) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = src;
-    tmp.i = __builtin_amdgcn_ds_permute(index, tmp.i);
-    return tmp.f;
-}
-
-#define __hip_ds_swizzle(src, pattern)  __hip_ds_swizzle_N<(pattern)>((src))
-#define __hip_ds_swizzlef(src, pattern) __hip_ds_swizzlef_N<(pattern)>((src))
-
-template <int pattern>
-__device__ static inline unsigned __hip_ds_swizzle_N(unsigned int src) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = src;
-    tmp.i = __builtin_amdgcn_ds_swizzle(tmp.i, pattern);
-    return tmp.u;
-}
-
-template <int pattern>
-__device__ static inline float __hip_ds_swizzlef_N(float src) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = src;
-    tmp.i = __builtin_amdgcn_ds_swizzle(tmp.i, pattern);
-    return tmp.f;
-}
-
-#define __hip_move_dpp(src, dpp_ctrl, row_mask, bank_mask, bound_ctrl) \
-  __hip_move_dpp_N<(dpp_ctrl), (row_mask), (bank_mask), (bound_ctrl)>((src))
-
-template <int dpp_ctrl, int row_mask, int bank_mask, bool bound_ctrl>
-__device__ static inline int __hip_move_dpp_N(int src) {
-    return __builtin_amdgcn_mov_dpp(src, dpp_ctrl, row_mask, bank_mask,
-                                    bound_ctrl);
-}
-
-static constexpr int warpSize = __AMDGCN_WAVEFRONT_SIZE;
-
-__device__
-inline
-int __shfl(int var, int src_lane, int width = warpSize) {
-    int self = __lane_id();
-    int index = (src_lane & (width - 1)) + (self & ~(width-1));
-    return __builtin_amdgcn_ds_bpermute(index<<2, var);
-}
-__device__
-inline
-unsigned int __shfl(unsigned int var, int src_lane, int width = warpSize) {
-     union { int i; unsigned u; float f; } tmp; tmp.u = var;
-    tmp.i = __shfl(tmp.i, src_lane, width);
-    return tmp.u;
-}
-__device__
-inline
-float __shfl(float var, int src_lane, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = var;
-    tmp.i = __shfl(tmp.i, src_lane, width);
-    return tmp.f;
-}
-__device__
-inline
-double __shfl(double var, int src_lane, int width = warpSize) {
-    static_assert(sizeof(double) == 2 * sizeof(int), "");
-    static_assert(sizeof(double) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl(tmp[0], src_lane, width);
-    tmp[1] = __shfl(tmp[1], src_lane, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    double tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-long __shfl(long var, int src_lane, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl(tmp[0], src_lane, width);
-    tmp[1] = __shfl(tmp[1], src_lane, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(long) == sizeof(int), "");
-    return static_cast<long>(__shfl(static_cast<int>(var), src_lane, width));
-    #endif
-}
-__device__
-inline
-unsigned long __shfl(unsigned long var, int src_lane, int width = warpSize) {
-    #ifndef _MSC_VER
-    static_assert(sizeof(unsigned long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long) == sizeof(uint64_t), "");
-
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl(tmp[0], src_lane, width);
-    tmp[1] = __shfl(tmp[1], src_lane, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(unsigned long) == sizeof(unsigned int), "");
-    return static_cast<unsigned long>(__shfl(static_cast<unsigned int>(var), src_lane, width));
-    #endif
-}
-__device__
-inline
-long long __shfl(long long var, int src_lane, int width = warpSize)
-{
-    static_assert(sizeof(long long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long long) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl(tmp[0], src_lane, width);
-    tmp[1] = __shfl(tmp[1], src_lane, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-unsigned long long __shfl(unsigned long long var, int src_lane, int width = warpSize) {
-    static_assert(sizeof(unsigned long long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long long) == sizeof(uint64_t), "");
-
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl(tmp[0], src_lane, width);
-    tmp[1] = __shfl(tmp[1], src_lane, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-
-__device__
-inline
-int __shfl_up(int var, unsigned int lane_delta, int width = warpSize) {
-    int self = __lane_id();
-    int index = self - lane_delta;
-    index = (index < (self & ~(width-1)))?self:index;
-    return __builtin_amdgcn_ds_bpermute(index<<2, var);
-}
-__device__
-inline
-unsigned int __shfl_up(unsigned int var, unsigned int lane_delta, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = var;
-    tmp.i = __shfl_up(tmp.i, lane_delta, width);
-    return tmp.u;
-}
-__device__
-inline
-float __shfl_up(float var, unsigned int lane_delta, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = var;
-    tmp.i = __shfl_up(tmp.i, lane_delta, width);
-    return tmp.f;
-}
-__device__
-inline
-double __shfl_up(double var, unsigned int lane_delta, int width = warpSize) {
-    static_assert(sizeof(double) == 2 * sizeof(int), "");
-    static_assert(sizeof(double) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_up(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_up(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    double tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-long __shfl_up(long var, unsigned int lane_delta, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_up(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_up(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(long) == sizeof(int), "");
-    return static_cast<long>(__shfl_up(static_cast<int>(var), lane_delta, width));
-    #endif
-}
-
-__device__
-inline
-unsigned long __shfl_up(unsigned long var, unsigned int lane_delta, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(unsigned long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long) == sizeof(uint64_t), "");
-
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_up(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_up(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(unsigned long) == sizeof(unsigned int), "");
-    return static_cast<unsigned long>(__shfl_up(static_cast<unsigned int>(var), lane_delta, width));
-    #endif
-}
-
-__device__
-inline
-long long __shfl_up(long long var, unsigned int lane_delta, int width = warpSize)
-{
-    static_assert(sizeof(long long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long long) == sizeof(uint64_t), "");
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_up(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_up(tmp[1], lane_delta, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-
-__device__
-inline
-unsigned long long __shfl_up(unsigned long long var, unsigned int lane_delta, int width = warpSize)
-{
-    static_assert(sizeof(unsigned long long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long long) == sizeof(uint64_t), "");
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_up(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_up(tmp[1], lane_delta, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-
-__device__
-inline
-int __shfl_down(int var, unsigned int lane_delta, int width = warpSize) {
-    int self = __lane_id();
-    int index = self + lane_delta;
-    index = (int)((self&(width-1))+lane_delta) >= width?self:index;
-    return __builtin_amdgcn_ds_bpermute(index<<2, var);
-}
-__device__
-inline
-unsigned int __shfl_down(unsigned int var, unsigned int lane_delta, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = var;
-    tmp.i = __shfl_down(tmp.i, lane_delta, width);
-    return tmp.u;
-}
-__device__
-inline
-float __shfl_down(float var, unsigned int lane_delta, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = var;
-    tmp.i = __shfl_down(tmp.i, lane_delta, width);
-    return tmp.f;
-}
-__device__
-inline
-double __shfl_down(double var, unsigned int lane_delta, int width = warpSize) {
-    static_assert(sizeof(double) == 2 * sizeof(int), "");
-    static_assert(sizeof(double) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_down(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_down(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    double tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-long __shfl_down(long var, unsigned int lane_delta, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_down(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_down(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(long) == sizeof(int), "");
-    return static_cast<long>(__shfl_down(static_cast<int>(var), lane_delta, width));
-    #endif
-}
-__device__
-inline
-unsigned long __shfl_down(unsigned long var, unsigned int lane_delta, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(unsigned long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long) == sizeof(uint64_t), "");
-
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_down(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_down(tmp[1], lane_delta, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(unsigned long) == sizeof(unsigned int), "");
-    return static_cast<unsigned long>(__shfl_down(static_cast<unsigned int>(var), lane_delta, width));
-    #endif
-}
-__device__
-inline
-long long __shfl_down(long long var, unsigned int lane_delta, int width = warpSize)
-{
-    static_assert(sizeof(long long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long long) == sizeof(uint64_t), "");
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_down(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_down(tmp[1], lane_delta, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-unsigned long long __shfl_down(unsigned long long var, unsigned int lane_delta, int width = warpSize)
-{
-    static_assert(sizeof(unsigned long long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long long) == sizeof(uint64_t), "");
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_down(tmp[0], lane_delta, width);
-    tmp[1] = __shfl_down(tmp[1], lane_delta, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-
-__device__
-inline
-int __shfl_xor(int var, int lane_mask, int width = warpSize) {
-    int self = __lane_id();
-    int index = self^lane_mask;
-    index = index >= ((self+width)&~(width-1))?self:index;
-    return __builtin_amdgcn_ds_bpermute(index<<2, var);
-}
-__device__
-inline
-unsigned int __shfl_xor(unsigned int var, int lane_mask, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.u = var;
-    tmp.i = __shfl_xor(tmp.i, lane_mask, width);
-    return tmp.u;
-}
-__device__
-inline
-float __shfl_xor(float var, int lane_mask, int width = warpSize) {
-    union { int i; unsigned u; float f; } tmp; tmp.f = var;
-    tmp.i = __shfl_xor(tmp.i, lane_mask, width);
-    return tmp.f;
-}
-__device__
-inline
-double __shfl_xor(double var, int lane_mask, int width = warpSize) {
-    static_assert(sizeof(double) == 2 * sizeof(int), "");
-    static_assert(sizeof(double) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_xor(tmp[0], lane_mask, width);
-    tmp[1] = __shfl_xor(tmp[1], lane_mask, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    double tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-long __shfl_xor(long var, int lane_mask, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long) == sizeof(uint64_t), "");
-
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_xor(tmp[0], lane_mask, width);
-    tmp[1] = __shfl_xor(tmp[1], lane_mask, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(long) == sizeof(int), "");
-    return static_cast<long>(__shfl_xor(static_cast<int>(var), lane_mask, width));
-    #endif
-}
-__device__
-inline
-unsigned long __shfl_xor(unsigned long var, int lane_mask, int width = warpSize)
-{
-    #ifndef _MSC_VER
-    static_assert(sizeof(unsigned long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long) == sizeof(uint64_t), "");
-
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_xor(tmp[0], lane_mask, width);
-    tmp[1] = __shfl_xor(tmp[1], lane_mask, width);
-
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-    #else
-    static_assert(sizeof(unsigned long) == sizeof(unsigned int), "");
-    return static_cast<unsigned long>(__shfl_xor(static_cast<unsigned int>(var), lane_mask, width));
-    #endif
-}
-__device__
-inline
-long long __shfl_xor(long long var, int lane_mask, int width = warpSize)
-{
-    static_assert(sizeof(long long) == 2 * sizeof(int), "");
-    static_assert(sizeof(long long) == sizeof(uint64_t), "");
-    int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_xor(tmp[0], lane_mask, width);
-    tmp[1] = __shfl_xor(tmp[1], lane_mask, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-__device__
-inline
-unsigned long long __shfl_xor(unsigned long long var, int lane_mask, int width = warpSize)
-{
-    static_assert(sizeof(unsigned long long) == 2 * sizeof(unsigned int), "");
-    static_assert(sizeof(unsigned long long) == sizeof(uint64_t), "");
-    unsigned int tmp[2]; __builtin_memcpy(tmp, &var, sizeof(tmp));
-    tmp[0] = __shfl_xor(tmp[0], lane_mask, width);
-    tmp[1] = __shfl_xor(tmp[1], lane_mask, width);
-    uint64_t tmp0 = (static_cast<uint64_t>(tmp[1]) << 32ull) | static_cast<uint32_t>(tmp[0]);
-    unsigned long long tmp1;  __builtin_memcpy(&tmp1, &tmp0, sizeof(tmp0));
-    return tmp1;
-}
-
-#endif

triton/backends/amd/include/hip/amd_detail/concepts.hpp

@@ -1,30 +0,0 @@
-/*
-Copyright (c) 2015 - 2021 Advanced Micro Devices, Inc. All rights reserved.
-
-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.
-*/
-
-#pragma once
-
-namespace hip_impl  // Documentation only.
-{
-#define requires(...)
-
-#define FunctionalProcedure typename
-}  // namespace hip_impl

triton/backends/amd/include/hip/amd_detail/device_library_decls.h

@@ -1,133 +0,0 @@
-/*
-Copyright (c) 2015 - 2023 Advanced Micro Devices, Inc. All rights reserved.
-
-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.
-*/
-
-/**
- *  @file  amd_detail/device_library_decls.h
- *  @brief Contains declarations for types and functions in device library.
- *         Uses int64_t and uint64_t instead of long, long long, unsigned
- *         long and unsigned long long types for device library API
- *         declarations.
- */
-
-#ifndef HIP_INCLUDE_HIP_AMD_DETAIL_DEVICE_LIBRARY_DECLS_H
-#define HIP_INCLUDE_HIP_AMD_DETAIL_DEVICE_LIBRARY_DECLS_H
-
-#if !defined(__HIPCC_RTC__)
-#include "hip/amd_detail/host_defines.h"
-#endif
-
-typedef unsigned char uchar;
-typedef unsigned short ushort;
-typedef unsigned int uint;
-typedef unsigned long ulong;
-typedef unsigned long long ullong;
-
-extern "C" __device__ __attribute__((const)) bool __ockl_wfany_i32(int);
-extern "C" __device__ __attribute__((const)) bool __ockl_wfall_i32(int);
-extern "C" __device__ uint __ockl_activelane_u32(void);
-
-extern "C" __device__ __attribute__((const)) uint __ockl_mul24_u32(uint, uint);
-extern "C" __device__ __attribute__((const)) int __ockl_mul24_i32(int, int);
-extern "C" __device__ __attribute__((const)) uint __ockl_mul_hi_u32(uint, uint);
-extern "C" __device__ __attribute__((const)) int __ockl_mul_hi_i32(int, int);
-extern "C" __device__ __attribute__((const)) uint __ockl_sadd_u32(uint, uint, uint);
-
-extern "C" __device__ __attribute__((const)) uchar __ockl_clz_u8(uchar);
-extern "C" __device__ __attribute__((const)) ushort __ockl_clz_u16(ushort);
-extern "C" __device__ __attribute__((const)) uint __ockl_clz_u32(uint);
-extern "C" __device__ __attribute__((const)) uint64_t __ockl_clz_u64(uint64_t);
-
-extern "C" __device__ __attribute__((const)) float __ocml_floor_f32(float);
-extern "C" __device__ __attribute__((const)) float __ocml_rint_f32(float);
-extern "C" __device__ __attribute__((const)) float __ocml_ceil_f32(float);
-extern "C" __device__ __attribute__((const)) float __ocml_trunc_f32(float);
-
-extern "C" __device__ __attribute__((const)) float __ocml_fmin_f32(float, float);
-extern "C" __device__ __attribute__((const)) float __ocml_fmax_f32(float, float);
-
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtn_f32_f64(double);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtp_f32_f64(double);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtz_f32_f64(double);
-
-extern "C" __device__ __attribute__((const)) _Float16 __ocml_cvtrtn_f16_f32(float);
-extern "C" __device__ __attribute__((const)) _Float16 __ocml_cvtrtp_f16_f32(float);
-extern "C" __device__ __attribute__((const)) _Float16 __ocml_cvtrtz_f16_f32(float);
-
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtn_f32_s32(int);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtp_f32_s32(int);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtz_f32_s32(int);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtn_f32_u32(uint32_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtp_f32_u32(uint32_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtz_f32_u32(uint32_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtn_f32_s64(int64_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtp_f32_s64(int64_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtz_f32_s64(int64_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtn_f32_u64(uint64_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtp_f32_u64(uint64_t);
-extern "C" __device__ __attribute__((const)) float __ocml_cvtrtz_f32_u64(uint64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtn_f64_s64(int64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtp_f64_s64(int64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtz_f64_s64(int64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtn_f64_u64(uint64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtp_f64_u64(uint64_t);
-extern "C" __device__ __attribute__((const)) double __ocml_cvtrtz_f64_u64(uint64_t);
-
-extern "C" __device__ __attribute__((convergent)) void __ockl_gws_init(uint nwm1, uint rid);
-extern "C" __device__ __attribute__((convergent)) void __ockl_gws_barrier(uint nwm1, uint rid);
-
-extern "C" __device__ __attribute__((const)) uint32_t __ockl_lane_u32();
-extern "C" __device__ __attribute__((const)) int __ockl_grid_is_valid(void);
-extern "C" __device__ __attribute__((convergent)) void __ockl_grid_sync(void);
-extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_num_grids(void);
-extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_grid_rank(void);
-extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_size(void);
-extern "C" __device__ __attribute__((const)) uint __ockl_multi_grid_thread_rank(void);
-extern "C" __device__ __attribute__((const)) int __ockl_multi_grid_is_valid(void);
-extern "C" __device__ __attribute__((convergent)) void __ockl_multi_grid_sync(void);
-
-extern "C" __device__ void __ockl_atomic_add_noret_f32(float*, float);
-
-extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_add_i32(int a);
-extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_and_i32(int a);
-extern "C" __device__ __attribute__((convergent)) int __ockl_wgred_or_i32(int a);
-
-extern "C" __device__ uint64_t __ockl_fprintf_stderr_begin();
-extern "C" __device__ uint64_t __ockl_fprintf_append_args(uint64_t msg_desc, uint32_t num_args,
-                                                          uint64_t value0, uint64_t value1,
-                                                          uint64_t value2, uint64_t value3,
-                                                          uint64_t value4, uint64_t value5,
-                                                          uint64_t value6, uint32_t is_last);
-extern "C" __device__ uint64_t __ockl_fprintf_append_string_n(uint64_t msg_desc, const char* data,
-                                                              uint64_t length, uint32_t is_last);
-
-// Introduce local address space
-#define __local __attribute__((address_space(3)))
-
-#ifdef __HIP_DEVICE_COMPILE__
-__device__ inline static __local void* __to_local(unsigned x) { return (__local void*)x; }
-#endif //__HIP_DEVICE_COMPILE__
-
-// Using hip.amdgcn.bc - sync threads
-#define __CLK_LOCAL_MEM_FENCE    0x01
-typedef unsigned __cl_mem_fence_flags;
-
-#endif